blob: 61ede335f6c3743a09fa88f0f5fa0483b250620a [file] [log] [blame]
David Sterbac1d7c512018-04-03 19:23:33 +02001// SPDX-License-Identifier: GPL-2.0
Chris Mason6cbd5572007-06-12 09:07:21 -04002/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
Chris Mason6cbd5572007-06-12 09:07:21 -04004 */
David Sterbac1d7c512018-04-03 19:23:33 +02005
Zach Brownec6b9102007-07-11 10:00:37 -04006#include <linux/sched.h>
Ingo Molnarf361bf42017-02-03 23:47:37 +01007#include <linux/sched/signal.h>
Chris Masonedbd8d42007-12-21 16:27:24 -05008#include <linux/pagemap.h>
Chris Masonec44a352008-04-28 15:29:52 -04009#include <linux/writeback.h>
David Woodhouse21af8042008-08-12 14:13:26 +010010#include <linux/blkdev.h>
Chris Masonb7a9f292009-02-04 09:23:45 -050011#include <linux/sort.h>
Chris Mason4184ea72009-03-10 12:39:20 -040012#include <linux/rcupdate.h>
Josef Bacik817d52f2009-07-13 21:29:25 -040013#include <linux/kthread.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/slab.h>
David Sterbadff51cd2011-06-14 12:52:17 +020015#include <linux/ratelimit.h>
Josef Bacikb150a4f2013-06-19 15:00:04 -040016#include <linux/percpu_counter.h>
Josef Bacik69fe2d72017-10-19 14:15:57 -040017#include <linux/lockdep.h>
Nikolay Borisov9678c542018-01-08 11:45:05 +020018#include <linux/crc32c.h>
David Sterba602cbe92019-08-21 18:48:25 +020019#include "misc.h"
Miao Xie995946d2014-04-02 19:51:06 +080020#include "tree-log.h"
Chris Masonfec577f2007-02-26 10:40:21 -050021#include "disk-io.h"
22#include "print-tree.h"
Chris Mason0b86a832008-03-24 15:01:56 -040023#include "volumes.h"
David Woodhouse53b381b2013-01-29 18:40:14 -050024#include "raid56.h"
Chris Mason925baed2008-06-25 16:01:30 -040025#include "locking.h"
Chris Masonfa9c0d792009-04-03 09:47:43 -040026#include "free-space-cache.h"
Omar Sandoval1e144fb2015-09-29 20:50:37 -070027#include "free-space-tree.h"
Jeff Mahoney6ab0a202013-11-01 13:07:04 -040028#include "sysfs.h"
Josef Bacikfcebe452014-05-13 17:30:47 -070029#include "qgroup.h"
Josef Bacikfd708b82017-09-29 15:43:50 -040030#include "ref-verify.h"
Josef Bacik8719aaa2019-06-18 16:09:16 -040031#include "space-info.h"
Josef Bacikd12ffdd2019-06-19 13:47:17 -040032#include "block-rsv.h"
Josef Bacik86736342019-06-19 15:12:00 -040033#include "delalloc-space.h"
Josef Bacikaac00232019-06-20 15:37:44 -040034#include "block-group.h"
Dennis Zhoub0643e52019-12-13 16:22:14 -080035#include "discard.h"
Chris Masonfec577f2007-02-26 10:40:21 -050036
Arne Jansen709c0482011-09-12 12:22:57 +020037#undef SCRAMBLE_DELAYED_REFS
38
Qu Wenruo9f9b8e82018-10-24 20:24:01 +080039
Yan Zheng5d4f98a2009-06-10 10:45:14 -040040static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
Nikolay Borisove72cb922018-06-20 15:48:57 +030041 struct btrfs_delayed_ref_node *node, u64 parent,
42 u64 root_objectid, u64 owner_objectid,
43 u64 owner_offset, int refs_to_drop,
44 struct btrfs_delayed_extent_op *extra_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -040045static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
46 struct extent_buffer *leaf,
47 struct btrfs_extent_item *ei);
48static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -040049 u64 parent, u64 root_objectid,
50 u64 flags, u64 owner, u64 offset,
51 struct btrfs_key *ins, int ref_mod);
52static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +030053 struct btrfs_delayed_ref_node *node,
Nikolay Borisov21ebfbe2018-05-21 12:27:22 +030054 struct btrfs_delayed_extent_op *extent_op);
Yan Zheng11833d62009-09-11 16:11:19 -040055static int find_next_key(struct btrfs_path *path, int level,
56 struct btrfs_key *key);
Josef Bacik6a632092009-02-20 11:00:09 -050057
David Sterba32da53862019-10-29 19:20:18 +010058static int block_group_bits(struct btrfs_block_group *cache, u64 bits)
Josef Bacik0f9dd462008-09-23 13:14:11 -040059{
60 return (cache->flags & bits) == bits;
61}
62
Josef Bacik6f410d12019-06-20 15:37:49 -040063int btrfs_add_excluded_extent(struct btrfs_fs_info *fs_info,
64 u64 start, u64 num_bytes)
Josef Bacik817d52f2009-07-13 21:29:25 -040065{
Yan Zheng11833d62009-09-11 16:11:19 -040066 u64 end = start + num_bytes - 1;
Nikolay Borisovfe119a62020-01-20 16:09:18 +020067 set_extent_bits(&fs_info->excluded_extents, start, end,
68 EXTENT_UPTODATE);
Yan Zheng11833d62009-09-11 16:11:19 -040069 return 0;
Josef Bacik817d52f2009-07-13 21:29:25 -040070}
71
David Sterba32da53862019-10-29 19:20:18 +010072void btrfs_free_excluded_extents(struct btrfs_block_group *cache)
Josef Bacik817d52f2009-07-13 21:29:25 -040073{
Nikolay Borisov9e715da2018-06-20 15:49:08 +030074 struct btrfs_fs_info *fs_info = cache->fs_info;
Yan Zheng11833d62009-09-11 16:11:19 -040075 u64 start, end;
76
David Sterbab3470b52019-10-23 18:48:22 +020077 start = cache->start;
78 end = start + cache->length - 1;
Yan Zheng11833d62009-09-11 16:11:19 -040079
Nikolay Borisovfe119a62020-01-20 16:09:18 +020080 clear_extent_bits(&fs_info->excluded_extents, start, end,
81 EXTENT_UPTODATE);
Yan Zheng11833d62009-09-11 16:11:19 -040082}
83
Qu Wenruo78192442019-05-15 07:33:48 +080084static u64 generic_ref_to_space_flags(struct btrfs_ref *ref)
Omar Sandoval0d9f8242017-06-06 16:45:26 -070085{
Qu Wenruoddf30cf2019-04-04 14:45:34 +080086 if (ref->type == BTRFS_REF_METADATA) {
87 if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID)
Qu Wenruo78192442019-05-15 07:33:48 +080088 return BTRFS_BLOCK_GROUP_SYSTEM;
Omar Sandoval0d9f8242017-06-06 16:45:26 -070089 else
Qu Wenruo78192442019-05-15 07:33:48 +080090 return BTRFS_BLOCK_GROUP_METADATA;
Omar Sandoval0d9f8242017-06-06 16:45:26 -070091 }
Qu Wenruo78192442019-05-15 07:33:48 +080092 return BTRFS_BLOCK_GROUP_DATA;
93}
94
95static void add_pinned_bytes(struct btrfs_fs_info *fs_info,
96 struct btrfs_ref *ref)
97{
98 struct btrfs_space_info *space_info;
99 u64 flags = generic_ref_to_space_flags(ref);
Omar Sandoval0d9f8242017-06-06 16:45:26 -0700100
Josef Bacik280c29082019-06-18 16:09:19 -0400101 space_info = btrfs_find_space_info(fs_info, flags);
Omar Sandoval55e81962017-06-06 16:45:27 -0700102 ASSERT(space_info);
Qu Wenruo78192442019-05-15 07:33:48 +0800103 percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len,
104 BTRFS_TOTAL_BYTES_PINNED_BATCH);
105}
106
107static void sub_pinned_bytes(struct btrfs_fs_info *fs_info,
108 struct btrfs_ref *ref)
109{
110 struct btrfs_space_info *space_info;
111 u64 flags = generic_ref_to_space_flags(ref);
112
Josef Bacik280c29082019-06-18 16:09:19 -0400113 space_info = btrfs_find_space_info(fs_info, flags);
Qu Wenruo78192442019-05-15 07:33:48 +0800114 ASSERT(space_info);
115 percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len,
Ethan Liendec59fa2018-07-13 16:50:42 +0800116 BTRFS_TOTAL_BYTES_PINNED_BATCH);
Omar Sandoval0d9f8242017-06-06 16:45:26 -0700117}
118
Filipe Manana1a4ed8f2014-10-27 10:44:24 +0000119/* simple helper to search for an existing data extent at a given offset */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400120int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
Chris Masone02119d2008-09-05 16:13:11 -0400121{
122 int ret;
123 struct btrfs_key key;
Zheng Yan31840ae2008-09-23 13:14:14 -0400124 struct btrfs_path *path;
Chris Masone02119d2008-09-05 16:13:11 -0400125
Zheng Yan31840ae2008-09-23 13:14:14 -0400126 path = btrfs_alloc_path();
Mark Fashehd8926bb2011-07-13 10:38:47 -0700127 if (!path)
128 return -ENOMEM;
129
Chris Masone02119d2008-09-05 16:13:11 -0400130 key.objectid = start;
131 key.offset = len;
Josef Bacik3173a182013-03-07 14:22:04 -0500132 key.type = BTRFS_EXTENT_ITEM_KEY;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400133 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -0400134 btrfs_free_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500135 return ret;
136}
137
Chris Masond8d5f3e2007-12-11 12:42:00 -0500138/*
Josef Bacik3173a182013-03-07 14:22:04 -0500139 * helper function to lookup reference count and flags of a tree block.
Yan, Zhenga22285a2010-05-16 10:48:46 -0400140 *
141 * the head node for delayed ref is used to store the sum of all the
142 * reference count modifications queued up in the rbtree. the head
143 * node may also store the extent flags to set. This way you can check
144 * to see what the reference count and extent flags would be if all of
145 * the delayed refs are not processed.
146 */
147int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -0400148 struct btrfs_fs_info *fs_info, u64 bytenr,
Josef Bacik3173a182013-03-07 14:22:04 -0500149 u64 offset, int metadata, u64 *refs, u64 *flags)
Yan, Zhenga22285a2010-05-16 10:48:46 -0400150{
151 struct btrfs_delayed_ref_head *head;
152 struct btrfs_delayed_ref_root *delayed_refs;
153 struct btrfs_path *path;
154 struct btrfs_extent_item *ei;
155 struct extent_buffer *leaf;
156 struct btrfs_key key;
157 u32 item_size;
158 u64 num_refs;
159 u64 extent_flags;
160 int ret;
161
Josef Bacik3173a182013-03-07 14:22:04 -0500162 /*
163 * If we don't have skinny metadata, don't bother doing anything
164 * different
165 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400166 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
167 offset = fs_info->nodesize;
Josef Bacik3173a182013-03-07 14:22:04 -0500168 metadata = 0;
169 }
170
Yan, Zhenga22285a2010-05-16 10:48:46 -0400171 path = btrfs_alloc_path();
172 if (!path)
173 return -ENOMEM;
174
Yan, Zhenga22285a2010-05-16 10:48:46 -0400175 if (!trans) {
176 path->skip_locking = 1;
177 path->search_commit_root = 1;
178 }
Filipe David Borba Manana639eefc2013-12-08 00:26:29 +0000179
180search_again:
181 key.objectid = bytenr;
182 key.offset = offset;
183 if (metadata)
184 key.type = BTRFS_METADATA_ITEM_KEY;
185 else
186 key.type = BTRFS_EXTENT_ITEM_KEY;
187
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400188 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400189 if (ret < 0)
190 goto out_free;
191
Josef Bacik3173a182013-03-07 14:22:04 -0500192 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
Filipe David Borba Manana74be9512013-07-05 23:12:06 +0100193 if (path->slots[0]) {
194 path->slots[0]--;
195 btrfs_item_key_to_cpu(path->nodes[0], &key,
196 path->slots[0]);
197 if (key.objectid == bytenr &&
198 key.type == BTRFS_EXTENT_ITEM_KEY &&
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400199 key.offset == fs_info->nodesize)
Filipe David Borba Manana74be9512013-07-05 23:12:06 +0100200 ret = 0;
201 }
Josef Bacik3173a182013-03-07 14:22:04 -0500202 }
203
Yan, Zhenga22285a2010-05-16 10:48:46 -0400204 if (ret == 0) {
205 leaf = path->nodes[0];
206 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
207 if (item_size >= sizeof(*ei)) {
208 ei = btrfs_item_ptr(leaf, path->slots[0],
209 struct btrfs_extent_item);
210 num_refs = btrfs_extent_refs(leaf, ei);
211 extent_flags = btrfs_extent_flags(leaf, ei);
212 } else {
Nikolay Borisovba3c2b12018-06-26 16:57:36 +0300213 ret = -EINVAL;
214 btrfs_print_v0_err(fs_info);
215 if (trans)
216 btrfs_abort_transaction(trans, ret);
217 else
218 btrfs_handle_fs_error(fs_info, ret, NULL);
219
220 goto out_free;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400221 }
Nikolay Borisovba3c2b12018-06-26 16:57:36 +0300222
Yan, Zhenga22285a2010-05-16 10:48:46 -0400223 BUG_ON(num_refs == 0);
224 } else {
225 num_refs = 0;
226 extent_flags = 0;
227 ret = 0;
228 }
229
230 if (!trans)
231 goto out;
232
233 delayed_refs = &trans->transaction->delayed_refs;
234 spin_lock(&delayed_refs->lock);
Liu Bof72ad18e2017-01-30 12:24:37 -0800235 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400236 if (head) {
237 if (!mutex_trylock(&head->mutex)) {
Josef Bacikd2788502017-09-29 15:43:57 -0400238 refcount_inc(&head->refs);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400239 spin_unlock(&delayed_refs->lock);
240
David Sterbab3b4aa72011-04-21 01:20:15 +0200241 btrfs_release_path(path);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400242
David Sterba8cc33e52011-05-02 15:29:25 +0200243 /*
244 * Mutex was contended, block until it's released and try
245 * again
246 */
Yan, Zhenga22285a2010-05-16 10:48:46 -0400247 mutex_lock(&head->mutex);
248 mutex_unlock(&head->mutex);
Josef Bacikd2788502017-09-29 15:43:57 -0400249 btrfs_put_delayed_ref_head(head);
Filipe David Borba Manana639eefc2013-12-08 00:26:29 +0000250 goto search_again;
Yan, Zhenga22285a2010-05-16 10:48:46 -0400251 }
Josef Bacikd7df2c72014-01-23 09:21:38 -0500252 spin_lock(&head->lock);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400253 if (head->extent_op && head->extent_op->update_flags)
254 extent_flags |= head->extent_op->flags_to_set;
255 else
256 BUG_ON(num_refs == 0);
257
Josef Bacikd2788502017-09-29 15:43:57 -0400258 num_refs += head->ref_mod;
Josef Bacikd7df2c72014-01-23 09:21:38 -0500259 spin_unlock(&head->lock);
Yan, Zhenga22285a2010-05-16 10:48:46 -0400260 mutex_unlock(&head->mutex);
261 }
262 spin_unlock(&delayed_refs->lock);
263out:
264 WARN_ON(num_refs == 0);
265 if (refs)
266 *refs = num_refs;
267 if (flags)
268 *flags = extent_flags;
269out_free:
270 btrfs_free_path(path);
271 return ret;
272}
273
274/*
Chris Masond8d5f3e2007-12-11 12:42:00 -0500275 * Back reference rules. Back refs have three main goals:
276 *
277 * 1) differentiate between all holders of references to an extent so that
278 * when a reference is dropped we can make sure it was a valid reference
279 * before freeing the extent.
280 *
281 * 2) Provide enough information to quickly find the holders of an extent
282 * if we notice a given block is corrupted or bad.
283 *
284 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
285 * maintenance. This is actually the same as #2, but with a slightly
286 * different use case.
287 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400288 * There are two kinds of back refs. The implicit back refs is optimized
289 * for pointers in non-shared tree blocks. For a given pointer in a block,
290 * back refs of this kind provide information about the block's owner tree
291 * and the pointer's key. These information allow us to find the block by
292 * b-tree searching. The full back refs is for pointers in tree blocks not
293 * referenced by their owner trees. The location of tree block is recorded
294 * in the back refs. Actually the full back refs is generic, and can be
295 * used in all cases the implicit back refs is used. The major shortcoming
296 * of the full back refs is its overhead. Every time a tree block gets
297 * COWed, we have to update back refs entry for all pointers in it.
298 *
299 * For a newly allocated tree block, we use implicit back refs for
300 * pointers in it. This means most tree related operations only involve
301 * implicit back refs. For a tree block created in old transaction, the
302 * only way to drop a reference to it is COW it. So we can detect the
303 * event that tree block loses its owner tree's reference and do the
304 * back refs conversion.
305 *
Nicholas D Steeves01327612016-05-19 21:18:45 -0400306 * When a tree block is COWed through a tree, there are four cases:
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400307 *
308 * The reference count of the block is one and the tree is the block's
309 * owner tree. Nothing to do in this case.
310 *
311 * The reference count of the block is one and the tree is not the
312 * block's owner tree. In this case, full back refs is used for pointers
313 * in the block. Remove these full back refs, add implicit back refs for
314 * every pointers in the new block.
315 *
316 * The reference count of the block is greater than one and the tree is
317 * the block's owner tree. In this case, implicit back refs is used for
318 * pointers in the block. Add full back refs for every pointers in the
319 * block, increase lower level extents' reference counts. The original
320 * implicit back refs are entailed to the new block.
321 *
322 * The reference count of the block is greater than one and the tree is
323 * not the block's owner tree. Add implicit back refs for every pointer in
324 * the new block, increase lower level extents' reference count.
325 *
326 * Back Reference Key composing:
327 *
328 * The key objectid corresponds to the first byte in the extent,
329 * The key type is used to differentiate between types of back refs.
330 * There are different meanings of the key offset for different types
331 * of back refs.
332 *
Chris Masond8d5f3e2007-12-11 12:42:00 -0500333 * File extents can be referenced by:
334 *
335 * - multiple snapshots, subvolumes, or different generations in one subvol
Zheng Yan31840ae2008-09-23 13:14:14 -0400336 * - different files inside a single subvolume
Chris Masond8d5f3e2007-12-11 12:42:00 -0500337 * - different offsets inside a file (bookend extents in file.c)
338 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400339 * The extent ref structure for the implicit back refs has fields for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500340 *
341 * - Objectid of the subvolume root
Chris Masond8d5f3e2007-12-11 12:42:00 -0500342 * - objectid of the file holding the reference
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400343 * - original offset in the file
344 * - how many bookend extents
Zheng Yan31840ae2008-09-23 13:14:14 -0400345 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400346 * The key offset for the implicit back refs is hash of the first
347 * three fields.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500348 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400349 * The extent ref structure for the full back refs has field for:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500350 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400351 * - number of pointers in the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500352 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400353 * The key offset for the implicit back refs is the first byte of
354 * the tree leaf
Chris Masond8d5f3e2007-12-11 12:42:00 -0500355 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400356 * When a file extent is allocated, The implicit back refs is used.
357 * the fields are filled in:
Chris Masond8d5f3e2007-12-11 12:42:00 -0500358 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400359 * (root_key.objectid, inode objectid, offset in file, 1)
360 *
361 * When a file extent is removed file truncation, we find the
362 * corresponding implicit back refs and check the following fields:
363 *
364 * (btrfs_header_owner(leaf), inode objectid, offset in file)
Chris Masond8d5f3e2007-12-11 12:42:00 -0500365 *
366 * Btree extents can be referenced by:
367 *
368 * - Different subvolumes
Chris Masond8d5f3e2007-12-11 12:42:00 -0500369 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400370 * Both the implicit back refs and the full back refs for tree blocks
371 * only consist of key. The key offset for the implicit back refs is
372 * objectid of block's owner tree. The key offset for the full back refs
373 * is the first byte of parent block.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500374 *
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400375 * When implicit back refs is used, information about the lowest key and
376 * level of the tree block are required. These information are stored in
377 * tree block info structure.
Chris Masond8d5f3e2007-12-11 12:42:00 -0500378 */
Zheng Yan31840ae2008-09-23 13:14:14 -0400379
Liu Bo167ce952017-08-18 15:15:18 -0600380/*
381 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
Andrea Gelmini52042d82018-11-28 12:05:13 +0100382 * is_data == BTRFS_REF_TYPE_DATA, data type is requiried,
Liu Bo167ce952017-08-18 15:15:18 -0600383 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
384 */
385int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
386 struct btrfs_extent_inline_ref *iref,
387 enum btrfs_inline_ref_type is_data)
388{
389 int type = btrfs_extent_inline_ref_type(eb, iref);
Liu Bo64ecdb62017-08-18 15:15:24 -0600390 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
Liu Bo167ce952017-08-18 15:15:18 -0600391
392 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
393 type == BTRFS_SHARED_BLOCK_REF_KEY ||
394 type == BTRFS_SHARED_DATA_REF_KEY ||
395 type == BTRFS_EXTENT_DATA_REF_KEY) {
396 if (is_data == BTRFS_REF_TYPE_BLOCK) {
Liu Bo64ecdb62017-08-18 15:15:24 -0600397 if (type == BTRFS_TREE_BLOCK_REF_KEY)
Liu Bo167ce952017-08-18 15:15:18 -0600398 return type;
Liu Bo64ecdb62017-08-18 15:15:24 -0600399 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
400 ASSERT(eb->fs_info);
401 /*
402 * Every shared one has parent tree
403 * block, which must be aligned to
404 * nodesize.
405 */
406 if (offset &&
407 IS_ALIGNED(offset, eb->fs_info->nodesize))
408 return type;
409 }
Liu Bo167ce952017-08-18 15:15:18 -0600410 } else if (is_data == BTRFS_REF_TYPE_DATA) {
Liu Bo64ecdb62017-08-18 15:15:24 -0600411 if (type == BTRFS_EXTENT_DATA_REF_KEY)
Liu Bo167ce952017-08-18 15:15:18 -0600412 return type;
Liu Bo64ecdb62017-08-18 15:15:24 -0600413 if (type == BTRFS_SHARED_DATA_REF_KEY) {
414 ASSERT(eb->fs_info);
415 /*
416 * Every shared one has parent tree
417 * block, which must be aligned to
418 * nodesize.
419 */
420 if (offset &&
421 IS_ALIGNED(offset, eb->fs_info->nodesize))
422 return type;
423 }
Liu Bo167ce952017-08-18 15:15:18 -0600424 } else {
425 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
426 return type;
427 }
428 }
429
430 btrfs_print_leaf((struct extent_buffer *)eb);
431 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
432 eb->start, type);
433 WARN_ON(1);
434
435 return BTRFS_REF_TYPE_INVALID;
436}
437
Qu Wenruo0785a9a2019-08-09 09:24:24 +0800438u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400439{
440 u32 high_crc = ~(u32)0;
441 u32 low_crc = ~(u32)0;
442 __le64 lenum;
443
444 lenum = cpu_to_le64(root_objectid);
Johannes Thumshirn65019df2019-05-22 10:18:59 +0200445 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400446 lenum = cpu_to_le64(owner);
Johannes Thumshirn65019df2019-05-22 10:18:59 +0200447 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400448 lenum = cpu_to_le64(offset);
Johannes Thumshirn65019df2019-05-22 10:18:59 +0200449 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400450
451 return ((u64)high_crc << 31) ^ (u64)low_crc;
452}
453
454static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
455 struct btrfs_extent_data_ref *ref)
456{
457 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
458 btrfs_extent_data_ref_objectid(leaf, ref),
459 btrfs_extent_data_ref_offset(leaf, ref));
460}
461
462static int match_extent_data_ref(struct extent_buffer *leaf,
463 struct btrfs_extent_data_ref *ref,
464 u64 root_objectid, u64 owner, u64 offset)
465{
466 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
467 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
468 btrfs_extent_data_ref_offset(leaf, ref) != offset)
469 return 0;
470 return 1;
471}
472
473static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400474 struct btrfs_path *path,
475 u64 bytenr, u64 parent,
476 u64 root_objectid,
477 u64 owner, u64 offset)
478{
Nikolay Borisovbd1d53e2018-06-20 15:48:51 +0300479 struct btrfs_root *root = trans->fs_info->extent_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400480 struct btrfs_key key;
481 struct btrfs_extent_data_ref *ref;
482 struct extent_buffer *leaf;
483 u32 nritems;
484 int ret;
485 int recow;
486 int err = -ENOENT;
487
488 key.objectid = bytenr;
489 if (parent) {
490 key.type = BTRFS_SHARED_DATA_REF_KEY;
491 key.offset = parent;
492 } else {
493 key.type = BTRFS_EXTENT_DATA_REF_KEY;
494 key.offset = hash_extent_data_ref(root_objectid,
495 owner, offset);
496 }
497again:
498 recow = 0;
499 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
500 if (ret < 0) {
501 err = ret;
502 goto fail;
503 }
504
505 if (parent) {
506 if (!ret)
507 return 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400508 goto fail;
Zheng Yan31840ae2008-09-23 13:14:14 -0400509 }
510
511 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400512 nritems = btrfs_header_nritems(leaf);
513 while (1) {
514 if (path->slots[0] >= nritems) {
515 ret = btrfs_next_leaf(root, path);
516 if (ret < 0)
517 err = ret;
518 if (ret)
519 goto fail;
520
521 leaf = path->nodes[0];
522 nritems = btrfs_header_nritems(leaf);
523 recow = 1;
524 }
525
526 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
527 if (key.objectid != bytenr ||
528 key.type != BTRFS_EXTENT_DATA_REF_KEY)
529 goto fail;
530
531 ref = btrfs_item_ptr(leaf, path->slots[0],
532 struct btrfs_extent_data_ref);
533
534 if (match_extent_data_ref(leaf, ref, root_objectid,
535 owner, offset)) {
536 if (recow) {
David Sterbab3b4aa72011-04-21 01:20:15 +0200537 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400538 goto again;
539 }
540 err = 0;
541 break;
542 }
543 path->slots[0]++;
Zheng Yan31840ae2008-09-23 13:14:14 -0400544 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400545fail:
546 return err;
Zheng Yan31840ae2008-09-23 13:14:14 -0400547}
548
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400549static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400550 struct btrfs_path *path,
551 u64 bytenr, u64 parent,
552 u64 root_objectid, u64 owner,
553 u64 offset, int refs_to_add)
Zheng Yan31840ae2008-09-23 13:14:14 -0400554{
Nikolay Borisov62b895a2018-06-20 15:48:44 +0300555 struct btrfs_root *root = trans->fs_info->extent_root;
Zheng Yan31840ae2008-09-23 13:14:14 -0400556 struct btrfs_key key;
557 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400558 u32 size;
Zheng Yan31840ae2008-09-23 13:14:14 -0400559 u32 num_refs;
560 int ret;
561
562 key.objectid = bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400563 if (parent) {
564 key.type = BTRFS_SHARED_DATA_REF_KEY;
565 key.offset = parent;
566 size = sizeof(struct btrfs_shared_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -0400567 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400568 key.type = BTRFS_EXTENT_DATA_REF_KEY;
569 key.offset = hash_extent_data_ref(root_objectid,
570 owner, offset);
571 size = sizeof(struct btrfs_extent_data_ref);
Zheng Yan31840ae2008-09-23 13:14:14 -0400572 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400573
574 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
575 if (ret && ret != -EEXIST)
576 goto fail;
577
578 leaf = path->nodes[0];
579 if (parent) {
580 struct btrfs_shared_data_ref *ref;
581 ref = btrfs_item_ptr(leaf, path->slots[0],
582 struct btrfs_shared_data_ref);
583 if (ret == 0) {
584 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
585 } else {
586 num_refs = btrfs_shared_data_ref_count(leaf, ref);
587 num_refs += refs_to_add;
588 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
589 }
590 } else {
591 struct btrfs_extent_data_ref *ref;
592 while (ret == -EEXIST) {
593 ref = btrfs_item_ptr(leaf, path->slots[0],
594 struct btrfs_extent_data_ref);
595 if (match_extent_data_ref(leaf, ref, root_objectid,
596 owner, offset))
597 break;
David Sterbab3b4aa72011-04-21 01:20:15 +0200598 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400599 key.offset++;
600 ret = btrfs_insert_empty_item(trans, root, path, &key,
601 size);
602 if (ret && ret != -EEXIST)
603 goto fail;
604
605 leaf = path->nodes[0];
606 }
607 ref = btrfs_item_ptr(leaf, path->slots[0],
608 struct btrfs_extent_data_ref);
609 if (ret == 0) {
610 btrfs_set_extent_data_ref_root(leaf, ref,
611 root_objectid);
612 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
613 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
614 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
615 } else {
616 num_refs = btrfs_extent_data_ref_count(leaf, ref);
617 num_refs += refs_to_add;
618 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
619 }
620 }
621 btrfs_mark_buffer_dirty(leaf);
622 ret = 0;
623fail:
David Sterbab3b4aa72011-04-21 01:20:15 +0200624 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -0500625 return ret;
Chris Mason74493f72007-12-11 09:25:06 -0500626}
627
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400628static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400629 struct btrfs_path *path,
Josef Bacikfcebe452014-05-13 17:30:47 -0700630 int refs_to_drop, int *last_ref)
Zheng Yan31840ae2008-09-23 13:14:14 -0400631{
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400632 struct btrfs_key key;
633 struct btrfs_extent_data_ref *ref1 = NULL;
634 struct btrfs_shared_data_ref *ref2 = NULL;
Zheng Yan31840ae2008-09-23 13:14:14 -0400635 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400636 u32 num_refs = 0;
Zheng Yan31840ae2008-09-23 13:14:14 -0400637 int ret = 0;
638
639 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400640 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
641
642 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
643 ref1 = btrfs_item_ptr(leaf, path->slots[0],
644 struct btrfs_extent_data_ref);
645 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
646 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
647 ref2 = btrfs_item_ptr(leaf, path->slots[0],
648 struct btrfs_shared_data_ref);
649 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
David Sterba6d8ff4e2018-06-26 16:20:59 +0200650 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
Nikolay Borisovba3c2b12018-06-26 16:57:36 +0300651 btrfs_print_v0_err(trans->fs_info);
652 btrfs_abort_transaction(trans, -EINVAL);
653 return -EINVAL;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400654 } else {
655 BUG();
656 }
657
Chris Mason56bec292009-03-13 10:10:06 -0400658 BUG_ON(num_refs < refs_to_drop);
659 num_refs -= refs_to_drop;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400660
Zheng Yan31840ae2008-09-23 13:14:14 -0400661 if (num_refs == 0) {
Nikolay Borisove9f62902018-06-20 15:48:46 +0300662 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
Josef Bacikfcebe452014-05-13 17:30:47 -0700663 *last_ref = 1;
Zheng Yan31840ae2008-09-23 13:14:14 -0400664 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400665 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
666 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
667 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
668 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
Zheng Yan31840ae2008-09-23 13:14:14 -0400669 btrfs_mark_buffer_dirty(leaf);
670 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400671 return ret;
672}
673
Zhaolei9ed0dea2015-08-06 22:16:24 +0800674static noinline u32 extent_data_ref_count(struct btrfs_path *path,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400675 struct btrfs_extent_inline_ref *iref)
676{
677 struct btrfs_key key;
678 struct extent_buffer *leaf;
679 struct btrfs_extent_data_ref *ref1;
680 struct btrfs_shared_data_ref *ref2;
681 u32 num_refs = 0;
Liu Bo3de28d52017-08-18 15:15:19 -0600682 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400683
684 leaf = path->nodes[0];
685 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Nikolay Borisovba3c2b12018-06-26 16:57:36 +0300686
687 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400688 if (iref) {
Liu Bo3de28d52017-08-18 15:15:19 -0600689 /*
690 * If type is invalid, we should have bailed out earlier than
691 * this call.
692 */
693 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
694 ASSERT(type != BTRFS_REF_TYPE_INVALID);
695 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400696 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
697 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
698 } else {
699 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
700 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
701 }
702 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
703 ref1 = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_extent_data_ref);
705 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
706 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
707 ref2 = btrfs_item_ptr(leaf, path->slots[0],
708 struct btrfs_shared_data_ref);
709 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400710 } else {
711 WARN_ON(1);
712 }
713 return num_refs;
714}
715
716static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400717 struct btrfs_path *path,
718 u64 bytenr, u64 parent,
719 u64 root_objectid)
720{
Nikolay Borisovb8582ee2018-06-20 15:48:50 +0300721 struct btrfs_root *root = trans->fs_info->extent_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400722 struct btrfs_key key;
723 int ret;
724
725 key.objectid = bytenr;
726 if (parent) {
727 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
728 key.offset = parent;
729 } else {
730 key.type = BTRFS_TREE_BLOCK_REF_KEY;
731 key.offset = root_objectid;
732 }
733
734 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
735 if (ret > 0)
736 ret = -ENOENT;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400737 return ret;
738}
739
740static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400741 struct btrfs_path *path,
742 u64 bytenr, u64 parent,
743 u64 root_objectid)
744{
745 struct btrfs_key key;
746 int ret;
747
748 key.objectid = bytenr;
749 if (parent) {
750 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
751 key.offset = parent;
752 } else {
753 key.type = BTRFS_TREE_BLOCK_REF_KEY;
754 key.offset = root_objectid;
755 }
756
Nikolay Borisov10728402018-06-20 15:48:43 +0300757 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
Jeff Mahoney87bde3c2017-02-15 16:28:27 -0500758 path, &key, 0);
David Sterbab3b4aa72011-04-21 01:20:15 +0200759 btrfs_release_path(path);
Zheng Yan31840ae2008-09-23 13:14:14 -0400760 return ret;
761}
762
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400763static inline int extent_ref_type(u64 parent, u64 owner)
764{
765 int type;
766 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
767 if (parent > 0)
768 type = BTRFS_SHARED_BLOCK_REF_KEY;
769 else
770 type = BTRFS_TREE_BLOCK_REF_KEY;
771 } else {
772 if (parent > 0)
773 type = BTRFS_SHARED_DATA_REF_KEY;
774 else
775 type = BTRFS_EXTENT_DATA_REF_KEY;
776 }
777 return type;
778}
779
Yan Zheng2c47e6052009-06-27 21:07:35 -0400780static int find_next_key(struct btrfs_path *path, int level,
781 struct btrfs_key *key)
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400782
783{
Yan Zheng2c47e6052009-06-27 21:07:35 -0400784 for (; level < BTRFS_MAX_LEVEL; level++) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400785 if (!path->nodes[level])
786 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400787 if (path->slots[level] + 1 >=
788 btrfs_header_nritems(path->nodes[level]))
789 continue;
790 if (level == 0)
791 btrfs_item_key_to_cpu(path->nodes[level], key,
792 path->slots[level] + 1);
793 else
794 btrfs_node_key_to_cpu(path->nodes[level], key,
795 path->slots[level] + 1);
796 return 0;
797 }
798 return 1;
799}
800
801/*
802 * look for inline back ref. if back ref is found, *ref_ret is set
803 * to the address of inline back ref, and 0 is returned.
804 *
805 * if back ref isn't found, *ref_ret is set to the address where it
806 * should be inserted, and -ENOENT is returned.
807 *
808 * if insert is true and there are too many inline back refs, the path
809 * points to the extent item, and -EAGAIN is returned.
810 *
811 * NOTE: inline back refs are ordered in the same way that back ref
812 * items in the tree are ordered.
813 */
814static noinline_for_stack
815int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400816 struct btrfs_path *path,
817 struct btrfs_extent_inline_ref **ref_ret,
818 u64 bytenr, u64 num_bytes,
819 u64 parent, u64 root_objectid,
820 u64 owner, u64 offset, int insert)
821{
Nikolay Borisov867cc1f2018-06-20 15:48:48 +0300822 struct btrfs_fs_info *fs_info = trans->fs_info;
Jeff Mahoney87bde3c2017-02-15 16:28:27 -0500823 struct btrfs_root *root = fs_info->extent_root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400824 struct btrfs_key key;
825 struct extent_buffer *leaf;
826 struct btrfs_extent_item *ei;
827 struct btrfs_extent_inline_ref *iref;
828 u64 flags;
829 u64 item_size;
830 unsigned long ptr;
831 unsigned long end;
832 int extra_size;
833 int type;
834 int want;
835 int ret;
836 int err = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -0400837 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
Liu Bo3de28d52017-08-18 15:15:19 -0600838 int needed;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400839
840 key.objectid = bytenr;
841 key.type = BTRFS_EXTENT_ITEM_KEY;
842 key.offset = num_bytes;
843
844 want = extent_ref_type(parent, owner);
845 if (insert) {
846 extra_size = btrfs_extent_inline_ref_size(want);
Yan Zheng85d41982009-06-11 08:51:10 -0400847 path->keep_locks = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400848 } else
849 extra_size = -1;
Josef Bacik3173a182013-03-07 14:22:04 -0500850
851 /*
Nikolay Borisov16d1c062018-06-18 14:59:26 +0300852 * Owner is our level, so we can just add one to get the level for the
853 * block we are interested in.
Josef Bacik3173a182013-03-07 14:22:04 -0500854 */
855 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
856 key.type = BTRFS_METADATA_ITEM_KEY;
857 key.offset = owner;
858 }
859
860again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400861 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
862 if (ret < 0) {
863 err = ret;
864 goto out;
865 }
Josef Bacik3173a182013-03-07 14:22:04 -0500866
867 /*
868 * We may be a newly converted file system which still has the old fat
869 * extent entries for metadata, so try and see if we have one of those.
870 */
871 if (ret > 0 && skinny_metadata) {
872 skinny_metadata = false;
873 if (path->slots[0]) {
874 path->slots[0]--;
875 btrfs_item_key_to_cpu(path->nodes[0], &key,
876 path->slots[0]);
877 if (key.objectid == bytenr &&
878 key.type == BTRFS_EXTENT_ITEM_KEY &&
879 key.offset == num_bytes)
880 ret = 0;
881 }
882 if (ret) {
Filipe Manana9ce49a02014-04-24 15:15:28 +0100883 key.objectid = bytenr;
Josef Bacik3173a182013-03-07 14:22:04 -0500884 key.type = BTRFS_EXTENT_ITEM_KEY;
885 key.offset = num_bytes;
886 btrfs_release_path(path);
887 goto again;
888 }
889 }
890
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100891 if (ret && !insert) {
892 err = -ENOENT;
893 goto out;
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +0530894 } else if (WARN_ON(ret)) {
Josef Bacik492104c2013-03-08 15:41:02 -0500895 err = -EIO;
Josef Bacik492104c2013-03-08 15:41:02 -0500896 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +0100897 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400898
899 leaf = path->nodes[0];
900 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
David Sterba6d8ff4e2018-06-26 16:20:59 +0200901 if (unlikely(item_size < sizeof(*ei))) {
Nikolay Borisovba3c2b12018-06-26 16:57:36 +0300902 err = -EINVAL;
903 btrfs_print_v0_err(fs_info);
904 btrfs_abort_transaction(trans, err);
905 goto out;
906 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400907
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400908 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
909 flags = btrfs_extent_flags(leaf, ei);
910
911 ptr = (unsigned long)(ei + 1);
912 end = (unsigned long)ei + item_size;
913
Josef Bacik3173a182013-03-07 14:22:04 -0500914 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400915 ptr += sizeof(struct btrfs_tree_block_info);
916 BUG_ON(ptr > end);
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400917 }
918
Liu Bo3de28d52017-08-18 15:15:19 -0600919 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
920 needed = BTRFS_REF_TYPE_DATA;
921 else
922 needed = BTRFS_REF_TYPE_BLOCK;
923
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400924 err = -ENOENT;
925 while (1) {
926 if (ptr >= end) {
927 WARN_ON(ptr > end);
928 break;
929 }
930 iref = (struct btrfs_extent_inline_ref *)ptr;
Liu Bo3de28d52017-08-18 15:15:19 -0600931 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
932 if (type == BTRFS_REF_TYPE_INVALID) {
Su Yueaf431dc2018-06-22 16:18:01 +0800933 err = -EUCLEAN;
Liu Bo3de28d52017-08-18 15:15:19 -0600934 goto out;
935 }
936
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400937 if (want < type)
938 break;
939 if (want > type) {
940 ptr += btrfs_extent_inline_ref_size(type);
941 continue;
942 }
943
944 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
945 struct btrfs_extent_data_ref *dref;
946 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
947 if (match_extent_data_ref(leaf, dref, root_objectid,
948 owner, offset)) {
949 err = 0;
950 break;
951 }
952 if (hash_extent_data_ref_item(leaf, dref) <
953 hash_extent_data_ref(root_objectid, owner, offset))
954 break;
955 } else {
956 u64 ref_offset;
957 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
958 if (parent > 0) {
959 if (parent == ref_offset) {
960 err = 0;
961 break;
962 }
963 if (ref_offset < parent)
964 break;
965 } else {
966 if (root_objectid == ref_offset) {
967 err = 0;
968 break;
969 }
970 if (ref_offset < root_objectid)
971 break;
972 }
973 }
974 ptr += btrfs_extent_inline_ref_size(type);
975 }
976 if (err == -ENOENT && insert) {
977 if (item_size + extra_size >=
978 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
979 err = -EAGAIN;
980 goto out;
981 }
982 /*
983 * To add new inline back ref, we have to make sure
984 * there is no corresponding back ref item.
985 * For simplicity, we just do not add new inline back
986 * ref if there is any kind of item for this block
987 */
Yan Zheng2c47e6052009-06-27 21:07:35 -0400988 if (find_next_key(path, 0, &key) == 0 &&
989 key.objectid == bytenr &&
Yan Zheng85d41982009-06-11 08:51:10 -0400990 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400991 err = -EAGAIN;
992 goto out;
993 }
994 }
995 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
996out:
Yan Zheng85d41982009-06-11 08:51:10 -0400997 if (insert) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -0400998 path->keep_locks = 0;
999 btrfs_unlock_up_safe(path, 1);
1000 }
1001 return err;
1002}
1003
1004/*
1005 * helper to add new inline back ref
1006 */
1007static noinline_for_stack
Jeff Mahoney87bde3c2017-02-15 16:28:27 -05001008void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001009 struct btrfs_path *path,
1010 struct btrfs_extent_inline_ref *iref,
1011 u64 parent, u64 root_objectid,
1012 u64 owner, u64 offset, int refs_to_add,
1013 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001014{
1015 struct extent_buffer *leaf;
1016 struct btrfs_extent_item *ei;
1017 unsigned long ptr;
1018 unsigned long end;
1019 unsigned long item_offset;
1020 u64 refs;
1021 int size;
1022 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001023
1024 leaf = path->nodes[0];
1025 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1026 item_offset = (unsigned long)iref - (unsigned long)ei;
1027
1028 type = extent_ref_type(parent, owner);
1029 size = btrfs_extent_inline_ref_size(type);
1030
David Sterbac71dd882019-03-20 14:51:10 +01001031 btrfs_extend_item(path, size);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001032
1033 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1034 refs = btrfs_extent_refs(leaf, ei);
1035 refs += refs_to_add;
1036 btrfs_set_extent_refs(leaf, ei, refs);
1037 if (extent_op)
1038 __run_delayed_extent_op(extent_op, leaf, ei);
1039
1040 ptr = (unsigned long)ei + item_offset;
1041 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1042 if (ptr < end - size)
1043 memmove_extent_buffer(leaf, ptr + size, ptr,
1044 end - size - ptr);
1045
1046 iref = (struct btrfs_extent_inline_ref *)ptr;
1047 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1048 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1049 struct btrfs_extent_data_ref *dref;
1050 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1051 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1052 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1053 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1054 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1055 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1056 struct btrfs_shared_data_ref *sref;
1057 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1058 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1059 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1060 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1061 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1062 } else {
1063 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1064 }
1065 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001066}
1067
1068static int lookup_extent_backref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001069 struct btrfs_path *path,
1070 struct btrfs_extent_inline_ref **ref_ret,
1071 u64 bytenr, u64 num_bytes, u64 parent,
1072 u64 root_objectid, u64 owner, u64 offset)
1073{
1074 int ret;
1075
Nikolay Borisov867cc1f2018-06-20 15:48:48 +03001076 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1077 num_bytes, parent, root_objectid,
1078 owner, offset, 0);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001079 if (ret != -ENOENT)
1080 return ret;
1081
David Sterbab3b4aa72011-04-21 01:20:15 +02001082 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001083 *ref_ret = NULL;
1084
1085 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
Nikolay Borisovb8582ee2018-06-20 15:48:50 +03001086 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1087 root_objectid);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001088 } else {
Nikolay Borisovbd1d53e2018-06-20 15:48:51 +03001089 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1090 root_objectid, owner, offset);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001091 }
1092 return ret;
1093}
1094
1095/*
1096 * helper to update/remove inline back ref
1097 */
1098static noinline_for_stack
Nikolay Borisov61a18f12018-06-20 15:48:49 +03001099void update_inline_extent_backref(struct btrfs_path *path,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001100 struct btrfs_extent_inline_ref *iref,
1101 int refs_to_mod,
Josef Bacikfcebe452014-05-13 17:30:47 -07001102 struct btrfs_delayed_extent_op *extent_op,
1103 int *last_ref)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001104{
Nikolay Borisov61a18f12018-06-20 15:48:49 +03001105 struct extent_buffer *leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001106 struct btrfs_extent_item *ei;
1107 struct btrfs_extent_data_ref *dref = NULL;
1108 struct btrfs_shared_data_ref *sref = NULL;
1109 unsigned long ptr;
1110 unsigned long end;
1111 u32 item_size;
1112 int size;
1113 int type;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001114 u64 refs;
1115
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001116 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1117 refs = btrfs_extent_refs(leaf, ei);
1118 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1119 refs += refs_to_mod;
1120 btrfs_set_extent_refs(leaf, ei, refs);
1121 if (extent_op)
1122 __run_delayed_extent_op(extent_op, leaf, ei);
1123
Liu Bo3de28d52017-08-18 15:15:19 -06001124 /*
1125 * If type is invalid, we should have bailed out after
1126 * lookup_inline_extent_backref().
1127 */
1128 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1129 ASSERT(type != BTRFS_REF_TYPE_INVALID);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001130
1131 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1132 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1133 refs = btrfs_extent_data_ref_count(leaf, dref);
1134 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1135 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1136 refs = btrfs_shared_data_ref_count(leaf, sref);
1137 } else {
1138 refs = 1;
1139 BUG_ON(refs_to_mod != -1);
1140 }
1141
1142 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1143 refs += refs_to_mod;
1144
1145 if (refs > 0) {
1146 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1147 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1148 else
1149 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1150 } else {
Josef Bacikfcebe452014-05-13 17:30:47 -07001151 *last_ref = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001152 size = btrfs_extent_inline_ref_size(type);
1153 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1154 ptr = (unsigned long)iref;
1155 end = (unsigned long)ei + item_size;
1156 if (ptr + size < end)
1157 memmove_extent_buffer(leaf, ptr, ptr + size,
1158 end - ptr - size);
1159 item_size -= size;
David Sterba78ac4f92019-03-20 14:49:12 +01001160 btrfs_truncate_item(path, item_size, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001161 }
1162 btrfs_mark_buffer_dirty(leaf);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001163}
1164
1165static noinline_for_stack
1166int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001167 struct btrfs_path *path,
1168 u64 bytenr, u64 num_bytes, u64 parent,
1169 u64 root_objectid, u64 owner,
1170 u64 offset, int refs_to_add,
1171 struct btrfs_delayed_extent_op *extent_op)
1172{
1173 struct btrfs_extent_inline_ref *iref;
1174 int ret;
1175
Nikolay Borisov867cc1f2018-06-20 15:48:48 +03001176 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1177 num_bytes, parent, root_objectid,
1178 owner, offset, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001179 if (ret == 0) {
1180 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
Nikolay Borisov61a18f12018-06-20 15:48:49 +03001181 update_inline_extent_backref(path, iref, refs_to_add,
1182 extent_op, NULL);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001183 } else if (ret == -ENOENT) {
Nikolay Borisova639cde2018-06-20 15:49:10 +03001184 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
Jeff Mahoney143bede2012-03-01 14:56:26 +01001185 root_objectid, owner, offset,
1186 refs_to_add, extent_op);
1187 ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001188 }
1189 return ret;
1190}
1191
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001192static int remove_extent_backref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001193 struct btrfs_path *path,
1194 struct btrfs_extent_inline_ref *iref,
Josef Bacikfcebe452014-05-13 17:30:47 -07001195 int refs_to_drop, int is_data, int *last_ref)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001196{
Jeff Mahoney143bede2012-03-01 14:56:26 +01001197 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001198
1199 BUG_ON(!is_data && refs_to_drop != 1);
1200 if (iref) {
Nikolay Borisov61a18f12018-06-20 15:48:49 +03001201 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1202 last_ref);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001203 } else if (is_data) {
Nikolay Borisove9f62902018-06-20 15:48:46 +03001204 ret = remove_extent_data_ref(trans, path, refs_to_drop,
Josef Bacikfcebe452014-05-13 17:30:47 -07001205 last_ref);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001206 } else {
Josef Bacikfcebe452014-05-13 17:30:47 -07001207 *last_ref = 1;
Nikolay Borisov87cc7a82018-06-20 15:49:12 +03001208 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001209 }
1210 return ret;
1211}
1212
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001213static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1214 u64 *discarded_bytes)
Chris Mason15916de2008-11-19 21:17:22 -05001215{
Jeff Mahoney86557862015-06-15 09:41:16 -04001216 int j, ret = 0;
1217 u64 bytes_left, end;
Jeff Mahoney4d89d372015-06-15 09:41:15 -04001218 u64 aligned_start = ALIGN(start, 1 << 9);
1219
1220 if (WARN_ON(start != aligned_start)) {
1221 len -= aligned_start - start;
1222 len = round_down(len, 1 << 9);
1223 start = aligned_start;
1224 }
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001225
1226 *discarded_bytes = 0;
Jeff Mahoney86557862015-06-15 09:41:16 -04001227
1228 if (!len)
1229 return 0;
1230
1231 end = start + len;
1232 bytes_left = len;
1233
1234 /* Skip any superblocks on this device. */
1235 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1236 u64 sb_start = btrfs_sb_offset(j);
1237 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1238 u64 size = sb_start - start;
1239
1240 if (!in_range(sb_start, start, bytes_left) &&
1241 !in_range(sb_end, start, bytes_left) &&
1242 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1243 continue;
1244
1245 /*
1246 * Superblock spans beginning of range. Adjust start and
1247 * try again.
1248 */
1249 if (sb_start <= start) {
1250 start += sb_end - start;
1251 if (start > end) {
1252 bytes_left = 0;
1253 break;
1254 }
1255 bytes_left = end - start;
1256 continue;
1257 }
1258
1259 if (size) {
1260 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1261 GFP_NOFS, 0);
1262 if (!ret)
1263 *discarded_bytes += size;
1264 else if (ret != -EOPNOTSUPP)
1265 return ret;
1266 }
1267
1268 start = sb_end;
1269 if (start > end) {
1270 bytes_left = 0;
1271 break;
1272 }
1273 bytes_left = end - start;
1274 }
1275
1276 if (bytes_left) {
1277 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
Jeff Mahoney4d89d372015-06-15 09:41:15 -04001278 GFP_NOFS, 0);
1279 if (!ret)
Jeff Mahoney86557862015-06-15 09:41:16 -04001280 *discarded_bytes += bytes_left;
Jeff Mahoney4d89d372015-06-15 09:41:15 -04001281 }
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001282 return ret;
Chris Mason15916de2008-11-19 21:17:22 -05001283}
Chris Mason15916de2008-11-19 21:17:22 -05001284
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04001285int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
Filipe Manana1edb647b2014-12-08 14:01:12 +00001286 u64 num_bytes, u64 *actual_bytes)
Liu Hui1f3c79a2009-01-05 15:57:51 -05001287{
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001288 int ret = 0;
Li Dongyang5378e602011-03-24 10:24:27 +00001289 u64 discarded_bytes = 0;
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001290 u64 end = bytenr + num_bytes;
1291 u64 cur = bytenr;
Jan Schmidta1d3c472011-08-04 17:15:33 +02001292 struct btrfs_bio *bbio = NULL;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001293
Christoph Hellwige244a0a2009-10-14 09:24:59 -04001294
Filipe Manana29992412016-05-27 17:42:05 +01001295 /*
1296 * Avoid races with device replace and make sure our bbio has devices
1297 * associated to its stripes that don't go away while we are discarding.
1298 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001299 btrfs_bio_counter_inc_blocked(fs_info);
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001300 while (cur < end) {
1301 struct btrfs_bio_stripe *stripe;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001302 int i;
1303
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001304 num_bytes = end - cur;
1305 /* Tell the block device(s) that the sectors can be discarded */
1306 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, cur,
1307 &num_bytes, &bbio, 0);
1308 /*
1309 * Error can be -ENOMEM, -ENOENT (no such chunk mapping) or
1310 * -EOPNOTSUPP. For any such error, @num_bytes is not updated,
1311 * thus we can't continue anyway.
1312 */
1313 if (ret < 0)
1314 goto out;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001315
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001316 stripe = bbio->stripes;
Jan Schmidta1d3c472011-08-04 17:15:33 +02001317 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001318 u64 bytes;
Anand Jain38b5f682017-11-29 18:53:43 +08001319 struct request_queue *req_q;
1320
Filipe Manana627e0872018-01-30 18:40:22 +00001321 if (!stripe->dev->bdev) {
1322 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
1323 continue;
1324 }
Anand Jain38b5f682017-11-29 18:53:43 +08001325 req_q = bdev_get_queue(stripe->dev->bdev);
1326 if (!blk_queue_discard(req_q))
Josef Bacikd5e20032011-08-04 14:52:27 +00001327 continue;
1328
Li Dongyang5378e602011-03-24 10:24:27 +00001329 ret = btrfs_issue_discard(stripe->dev->bdev,
1330 stripe->physical,
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001331 stripe->length,
1332 &bytes);
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001333 if (!ret) {
Jeff Mahoneyd04c6b82015-06-15 09:41:14 -04001334 discarded_bytes += bytes;
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001335 } else if (ret != -EOPNOTSUPP) {
1336 /*
1337 * Logic errors or -ENOMEM, or -EIO, but
1338 * unlikely to happen.
1339 *
1340 * And since there are two loops, explicitly
1341 * go to out to avoid confusion.
1342 */
1343 btrfs_put_bbio(bbio);
1344 goto out;
1345 }
Josef Bacikd5e20032011-08-04 14:52:27 +00001346
1347 /*
1348 * Just in case we get back EOPNOTSUPP for some reason,
1349 * just ignore the return value so we don't screw up
1350 * people calling discard_extent.
1351 */
1352 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001353 }
Zhao Lei6e9606d2015-01-20 15:11:34 +08001354 btrfs_put_bbio(bbio);
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001355 cur += num_bytes;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001356 }
Qu Wenruo6b7faad2019-10-23 21:57:27 +08001357out:
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001358 btrfs_bio_counter_dec(fs_info);
Li Dongyang5378e602011-03-24 10:24:27 +00001359
1360 if (actual_bytes)
1361 *actual_bytes = discarded_bytes;
1362
Liu Hui1f3c79a2009-01-05 15:57:51 -05001363
David Woodhouse53b381b2013-01-29 18:40:14 -05001364 if (ret == -EOPNOTSUPP)
1365 ret = 0;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001366 return ret;
Liu Hui1f3c79a2009-01-05 15:57:51 -05001367}
1368
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001369/* Can return -ENOMEM */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001370int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Qu Wenruo82fa1132019-04-04 14:45:35 +08001371 struct btrfs_ref *generic_ref)
Zheng Yan31840ae2008-09-23 13:14:14 -04001372{
Qu Wenruo82fa1132019-04-04 14:45:35 +08001373 struct btrfs_fs_info *fs_info = trans->fs_info;
Omar Sandovald7eae342017-06-06 16:45:31 -07001374 int old_ref_mod, new_ref_mod;
Zheng Yan31840ae2008-09-23 13:14:14 -04001375 int ret;
Arne Jansen66d7e7f2011-09-12 15:26:38 +02001376
Qu Wenruo82fa1132019-04-04 14:45:35 +08001377 ASSERT(generic_ref->type != BTRFS_REF_NOT_SET &&
1378 generic_ref->action);
1379 BUG_ON(generic_ref->type == BTRFS_REF_METADATA &&
1380 generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID);
Zheng Yan31840ae2008-09-23 13:14:14 -04001381
Qu Wenruo82fa1132019-04-04 14:45:35 +08001382 if (generic_ref->type == BTRFS_REF_METADATA)
1383 ret = btrfs_add_delayed_tree_ref(trans, generic_ref,
Qu Wenruoed4f2552019-04-04 14:45:31 +08001384 NULL, &old_ref_mod, &new_ref_mod);
Qu Wenruo82fa1132019-04-04 14:45:35 +08001385 else
1386 ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0,
Omar Sandovald7eae342017-06-06 16:45:31 -07001387 &old_ref_mod, &new_ref_mod);
Omar Sandovald7eae342017-06-06 16:45:31 -07001388
Qu Wenruo82fa1132019-04-04 14:45:35 +08001389 btrfs_ref_tree_mod(fs_info, generic_ref);
Qu Wenruo8a5040f2019-04-04 14:45:33 +08001390
Qu Wenruoddf30cf2019-04-04 14:45:34 +08001391 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
Qu Wenruo78192442019-05-15 07:33:48 +08001392 sub_pinned_bytes(fs_info, generic_ref);
Omar Sandovald7eae342017-06-06 16:45:31 -07001393
Zheng Yan31840ae2008-09-23 13:14:14 -04001394 return ret;
1395}
1396
Nikolay Borisovbd3c6852018-06-18 14:59:25 +03001397/*
1398 * __btrfs_inc_extent_ref - insert backreference for a given extent
1399 *
1400 * @trans: Handle of transaction
1401 *
1402 * @node: The delayed ref node used to get the bytenr/length for
1403 * extent whose references are incremented.
1404 *
1405 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
1406 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
1407 * bytenr of the parent block. Since new extents are always
1408 * created with indirect references, this will only be the case
1409 * when relocating a shared extent. In that case, root_objectid
1410 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
1411 * be 0
1412 *
1413 * @root_objectid: The id of the root where this modification has originated,
1414 * this can be either one of the well-known metadata trees or
1415 * the subvolume id which references this extent.
1416 *
1417 * @owner: For data extents it is the inode number of the owning file.
1418 * For metadata extents this parameter holds the level in the
1419 * tree of the extent.
1420 *
1421 * @offset: For metadata extents the offset is ignored and is currently
1422 * always passed as 0. For data extents it is the fileoffset
1423 * this extent belongs to.
1424 *
1425 * @refs_to_add Number of references to add
1426 *
1427 * @extent_op Pointer to a structure, holding information necessary when
1428 * updating a tree block's flags
1429 *
1430 */
Chris Mason925baed2008-06-25 16:01:30 -04001431static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
Qu Wenruoc682f9b2015-03-17 16:59:47 +08001432 struct btrfs_delayed_ref_node *node,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001433 u64 parent, u64 root_objectid,
1434 u64 owner, u64 offset, int refs_to_add,
1435 struct btrfs_delayed_extent_op *extent_op)
Chris Mason56bec292009-03-13 10:10:06 -04001436{
Chris Mason5caf2a02007-04-02 11:20:42 -04001437 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001438 struct extent_buffer *leaf;
Chris Mason234b63a2007-03-13 10:46:10 -04001439 struct btrfs_extent_item *item;
Josef Bacikfcebe452014-05-13 17:30:47 -07001440 struct btrfs_key key;
Qu Wenruoc682f9b2015-03-17 16:59:47 +08001441 u64 bytenr = node->bytenr;
1442 u64 num_bytes = node->num_bytes;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001443 u64 refs;
1444 int ret;
Chris Mason037e6392007-03-07 11:50:24 -05001445
Chris Mason5caf2a02007-04-02 11:20:42 -04001446 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04001447 if (!path)
1448 return -ENOMEM;
Chris Mason26b80032007-08-08 20:17:12 -04001449
Chris Masonb9473432009-03-13 11:00:37 -04001450 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001451 /* this will setup the path even if it fails to insert the back ref */
Nikolay Borisova639cde2018-06-20 15:49:10 +03001452 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
1453 parent, root_objectid, owner,
1454 offset, refs_to_add, extent_op);
Qu Wenruo0ed47922015-04-16 16:55:08 +08001455 if ((ret < 0 && ret != -EAGAIN) || !ret)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001456 goto out;
Josef Bacikfcebe452014-05-13 17:30:47 -07001457
1458 /*
1459 * Ok we had -EAGAIN which means we didn't have space to insert and
1460 * inline extent ref, so just update the reference count and add a
1461 * normal backref.
1462 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001463 leaf = path->nodes[0];
Josef Bacikfcebe452014-05-13 17:30:47 -07001464 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001465 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1466 refs = btrfs_extent_refs(leaf, item);
1467 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
1468 if (extent_op)
1469 __run_delayed_extent_op(extent_op, leaf, item);
Zheng Yan31840ae2008-09-23 13:14:14 -04001470
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001471 btrfs_mark_buffer_dirty(leaf);
David Sterbab3b4aa72011-04-21 01:20:15 +02001472 btrfs_release_path(path);
Chris Mason7bb86312007-12-11 09:25:06 -05001473
Chris Masonb9473432009-03-13 11:00:37 -04001474 path->leave_spinning = 1;
Chris Mason56bec292009-03-13 10:10:06 -04001475 /* now insert the actual backref */
Nikolay Borisov65cd6d92020-03-04 11:33:53 +02001476 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1477 BUG_ON(refs_to_add != 1);
1478 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1479 root_objectid);
1480 } else {
1481 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1482 root_objectid, owner, offset,
1483 refs_to_add);
1484 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001485 if (ret)
Jeff Mahoney66642832016-06-10 18:19:25 -04001486 btrfs_abort_transaction(trans, ret);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001487out:
Chris Mason74493f72007-12-11 09:25:06 -05001488 btrfs_free_path(path);
Liu Bo30d133f2013-10-11 16:30:23 +08001489 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05001490}
1491
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001492static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001493 struct btrfs_delayed_ref_node *node,
1494 struct btrfs_delayed_extent_op *extent_op,
1495 int insert_reserved)
Chris Masone9d0b132007-08-10 14:06:19 -04001496{
Chris Mason56bec292009-03-13 10:10:06 -04001497 int ret = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001498 struct btrfs_delayed_data_ref *ref;
1499 struct btrfs_key ins;
1500 u64 parent = 0;
1501 u64 ref_root = 0;
1502 u64 flags = 0;
Chris Mason56bec292009-03-13 10:10:06 -04001503
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001504 ins.objectid = node->bytenr;
1505 ins.offset = node->num_bytes;
1506 ins.type = BTRFS_EXTENT_ITEM_KEY;
Chris Mason56bec292009-03-13 10:10:06 -04001507
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001508 ref = btrfs_delayed_node_to_data_ref(node);
Nikolay Borisov2bf98ef2018-06-20 15:49:00 +03001509 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
Liu Bo599c75e2013-07-16 19:03:36 +08001510
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001511 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
1512 parent = ref->parent;
Josef Bacikfcebe452014-05-13 17:30:47 -07001513 ref_root = ref->root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001514
1515 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05001516 if (extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001517 flags |= extent_op->flags_to_set;
Nikolay Borisovef89b822018-06-20 15:48:58 +03001518 ret = alloc_reserved_file_extent(trans, parent, ref_root,
1519 flags, ref->objectid,
1520 ref->offset, &ins,
1521 node->ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001522 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
Nikolay Borisov2590d0f2018-06-20 15:48:59 +03001523 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1524 ref->objectid, ref->offset,
1525 node->ref_mod, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001526 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
Nikolay Borisove72cb922018-06-20 15:48:57 +03001527 ret = __btrfs_free_extent(trans, node, parent,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001528 ref_root, ref->objectid,
1529 ref->offset, node->ref_mod,
Qu Wenruoc682f9b2015-03-17 16:59:47 +08001530 extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001531 } else {
1532 BUG();
1533 }
Chris Mason56bec292009-03-13 10:10:06 -04001534 return ret;
1535}
1536
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001537static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
1538 struct extent_buffer *leaf,
1539 struct btrfs_extent_item *ei)
1540{
1541 u64 flags = btrfs_extent_flags(leaf, ei);
1542 if (extent_op->update_flags) {
1543 flags |= extent_op->flags_to_set;
1544 btrfs_set_extent_flags(leaf, ei, flags);
1545 }
1546
1547 if (extent_op->update_key) {
1548 struct btrfs_tree_block_info *bi;
1549 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
1550 bi = (struct btrfs_tree_block_info *)(ei + 1);
1551 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
1552 }
1553}
1554
1555static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
Josef Bacikd2788502017-09-29 15:43:57 -04001556 struct btrfs_delayed_ref_head *head,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001557 struct btrfs_delayed_extent_op *extent_op)
1558{
Nikolay Borisov20b9a2d2018-06-20 15:49:01 +03001559 struct btrfs_fs_info *fs_info = trans->fs_info;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001560 struct btrfs_key key;
1561 struct btrfs_path *path;
1562 struct btrfs_extent_item *ei;
1563 struct extent_buffer *leaf;
1564 u32 item_size;
1565 int ret;
1566 int err = 0;
Josef Bacikb1c79e02013-05-09 13:49:30 -04001567 int metadata = !extent_op->is_data;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001568
David Sterbabf31f872020-02-05 17:34:34 +01001569 if (TRANS_ABORTED(trans))
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001570 return 0;
1571
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001572 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
Josef Bacik3173a182013-03-07 14:22:04 -05001573 metadata = 0;
1574
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001575 path = btrfs_alloc_path();
1576 if (!path)
1577 return -ENOMEM;
1578
Josef Bacikd2788502017-09-29 15:43:57 -04001579 key.objectid = head->bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001580
Josef Bacik3173a182013-03-07 14:22:04 -05001581 if (metadata) {
Josef Bacik3173a182013-03-07 14:22:04 -05001582 key.type = BTRFS_METADATA_ITEM_KEY;
Josef Bacikb1c79e02013-05-09 13:49:30 -04001583 key.offset = extent_op->level;
Josef Bacik3173a182013-03-07 14:22:04 -05001584 } else {
1585 key.type = BTRFS_EXTENT_ITEM_KEY;
Josef Bacikd2788502017-09-29 15:43:57 -04001586 key.offset = head->num_bytes;
Josef Bacik3173a182013-03-07 14:22:04 -05001587 }
1588
1589again:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001590 path->leave_spinning = 1;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04001591 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001592 if (ret < 0) {
1593 err = ret;
1594 goto out;
1595 }
1596 if (ret > 0) {
Josef Bacik3173a182013-03-07 14:22:04 -05001597 if (metadata) {
Filipe David Borba Manana55994882013-10-18 15:42:56 +01001598 if (path->slots[0] > 0) {
1599 path->slots[0]--;
1600 btrfs_item_key_to_cpu(path->nodes[0], &key,
1601 path->slots[0]);
Josef Bacikd2788502017-09-29 15:43:57 -04001602 if (key.objectid == head->bytenr &&
Filipe David Borba Manana55994882013-10-18 15:42:56 +01001603 key.type == BTRFS_EXTENT_ITEM_KEY &&
Josef Bacikd2788502017-09-29 15:43:57 -04001604 key.offset == head->num_bytes)
Filipe David Borba Manana55994882013-10-18 15:42:56 +01001605 ret = 0;
1606 }
1607 if (ret > 0) {
1608 btrfs_release_path(path);
1609 metadata = 0;
Josef Bacik3173a182013-03-07 14:22:04 -05001610
Josef Bacikd2788502017-09-29 15:43:57 -04001611 key.objectid = head->bytenr;
1612 key.offset = head->num_bytes;
Filipe David Borba Manana55994882013-10-18 15:42:56 +01001613 key.type = BTRFS_EXTENT_ITEM_KEY;
1614 goto again;
1615 }
1616 } else {
1617 err = -EIO;
1618 goto out;
Josef Bacik3173a182013-03-07 14:22:04 -05001619 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001620 }
1621
1622 leaf = path->nodes[0];
1623 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
Nikolay Borisovba3c2b12018-06-26 16:57:36 +03001624
David Sterba6d8ff4e2018-06-26 16:20:59 +02001625 if (unlikely(item_size < sizeof(*ei))) {
Nikolay Borisovba3c2b12018-06-26 16:57:36 +03001626 err = -EINVAL;
1627 btrfs_print_v0_err(fs_info);
1628 btrfs_abort_transaction(trans, err);
1629 goto out;
1630 }
1631
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001632 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1633 __run_delayed_extent_op(extent_op, leaf, ei);
1634
1635 btrfs_mark_buffer_dirty(leaf);
1636out:
1637 btrfs_free_path(path);
1638 return err;
1639}
1640
1641static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001642 struct btrfs_delayed_ref_node *node,
1643 struct btrfs_delayed_extent_op *extent_op,
1644 int insert_reserved)
1645{
1646 int ret = 0;
1647 struct btrfs_delayed_tree_ref *ref;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001648 u64 parent = 0;
1649 u64 ref_root = 0;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001650
1651 ref = btrfs_delayed_node_to_tree_ref(node);
Nikolay Borisovf97806f2018-06-20 15:49:04 +03001652 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
Liu Bo599c75e2013-07-16 19:03:36 +08001653
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001654 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
1655 parent = ref->parent;
Josef Bacikfcebe452014-05-13 17:30:47 -07001656 ref_root = ref->root;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001657
Liu Bo02794222016-09-14 19:19:05 -07001658 if (node->ref_mod != 1) {
Nikolay Borisovf97806f2018-06-20 15:49:04 +03001659 btrfs_err(trans->fs_info,
Liu Bo02794222016-09-14 19:19:05 -07001660 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
1661 node->bytenr, node->ref_mod, node->action, ref_root,
1662 parent);
1663 return -EIO;
1664 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001665 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
Josef Bacik3173a182013-03-07 14:22:04 -05001666 BUG_ON(!extent_op || !extent_op->update_flags);
Nikolay Borisov21ebfbe2018-05-21 12:27:22 +03001667 ret = alloc_reserved_tree_block(trans, node, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001668 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
Nikolay Borisov2590d0f2018-06-20 15:48:59 +03001669 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
1670 ref->level, 0, 1, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001671 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
Nikolay Borisove72cb922018-06-20 15:48:57 +03001672 ret = __btrfs_free_extent(trans, node, parent, ref_root,
Qu Wenruoc682f9b2015-03-17 16:59:47 +08001673 ref->level, 0, 1, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001674 } else {
1675 BUG();
1676 }
1677 return ret;
1678}
1679
Chris Mason56bec292009-03-13 10:10:06 -04001680/* helper function to actually process a single delayed ref entry */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001681static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001682 struct btrfs_delayed_ref_node *node,
1683 struct btrfs_delayed_extent_op *extent_op,
1684 int insert_reserved)
Chris Mason56bec292009-03-13 10:10:06 -04001685{
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001686 int ret = 0;
1687
David Sterbabf31f872020-02-05 17:34:34 +01001688 if (TRANS_ABORTED(trans)) {
Josef Bacik857cc2f2013-10-07 15:21:08 -04001689 if (insert_reserved)
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02001690 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001691 return 0;
Josef Bacik857cc2f2013-10-07 15:21:08 -04001692 }
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001693
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001694 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1695 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
Nikolay Borisovf97806f2018-06-20 15:49:04 +03001696 ret = run_delayed_tree_ref(trans, node, extent_op,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001697 insert_reserved);
1698 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1699 node->type == BTRFS_SHARED_DATA_REF_KEY)
Nikolay Borisov2bf98ef2018-06-20 15:49:00 +03001700 ret = run_delayed_data_ref(trans, node, extent_op,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001701 insert_reserved);
1702 else
1703 BUG();
Josef Bacik80ee54b2018-10-11 15:54:22 -04001704 if (ret && insert_reserved)
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02001705 btrfs_pin_extent(trans, node->bytenr, node->num_bytes, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04001706 return ret;
Chris Masone9d0b132007-08-10 14:06:19 -04001707}
1708
Qu Wenruoc6fc2452015-03-30 17:03:00 +08001709static inline struct btrfs_delayed_ref_node *
Chris Mason56bec292009-03-13 10:10:06 -04001710select_delayed_ref(struct btrfs_delayed_ref_head *head)
Chris Masona28ec192007-03-06 20:08:01 -05001711{
Filipe Mananacffc3372015-07-09 13:13:44 +01001712 struct btrfs_delayed_ref_node *ref;
1713
Liu Boe3d03962018-08-23 03:51:50 +08001714 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
Qu Wenruoc6fc2452015-03-30 17:03:00 +08001715 return NULL;
Josef Bacikd7df2c72014-01-23 09:21:38 -05001716
Filipe Mananacffc3372015-07-09 13:13:44 +01001717 /*
1718 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
1719 * This is to prevent a ref count from going down to zero, which deletes
1720 * the extent item from the extent tree, when there still are references
1721 * to add, which would fail because they would not find the extent item.
1722 */
Wang Xiaoguang1d57ee92016-10-26 18:07:33 +08001723 if (!list_empty(&head->ref_add_list))
1724 return list_first_entry(&head->ref_add_list,
1725 struct btrfs_delayed_ref_node, add_list);
Filipe Mananacffc3372015-07-09 13:13:44 +01001726
Liu Boe3d03962018-08-23 03:51:50 +08001727 ref = rb_entry(rb_first_cached(&head->ref_tree),
Josef Bacik0e0adbc2017-10-19 14:16:00 -04001728 struct btrfs_delayed_ref_node, ref_node);
Wang Xiaoguang1d57ee92016-10-26 18:07:33 +08001729 ASSERT(list_empty(&ref->add_list));
1730 return ref;
Chris Mason56bec292009-03-13 10:10:06 -04001731}
1732
Josef Bacik2eadaa22017-09-29 15:43:52 -04001733static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
1734 struct btrfs_delayed_ref_head *head)
1735{
1736 spin_lock(&delayed_refs->lock);
1737 head->processing = 0;
1738 delayed_refs->num_heads_ready++;
1739 spin_unlock(&delayed_refs->lock);
1740 btrfs_delayed_ref_unlock(head);
1741}
1742
Josef Bacikbedc66172018-12-03 10:20:31 -05001743static struct btrfs_delayed_extent_op *cleanup_extent_op(
1744 struct btrfs_delayed_ref_head *head)
Josef Bacikb00e6252017-09-29 15:43:53 -04001745{
1746 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
Josef Bacikbedc66172018-12-03 10:20:31 -05001747
1748 if (!extent_op)
1749 return NULL;
1750
1751 if (head->must_insert_reserved) {
1752 head->extent_op = NULL;
1753 btrfs_free_delayed_extent_op(extent_op);
1754 return NULL;
1755 }
1756 return extent_op;
1757}
1758
1759static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
1760 struct btrfs_delayed_ref_head *head)
1761{
1762 struct btrfs_delayed_extent_op *extent_op;
Josef Bacikb00e6252017-09-29 15:43:53 -04001763 int ret;
1764
Josef Bacikbedc66172018-12-03 10:20:31 -05001765 extent_op = cleanup_extent_op(head);
Josef Bacikb00e6252017-09-29 15:43:53 -04001766 if (!extent_op)
1767 return 0;
1768 head->extent_op = NULL;
Josef Bacikb00e6252017-09-29 15:43:53 -04001769 spin_unlock(&head->lock);
Nikolay Borisov20b9a2d2018-06-20 15:49:01 +03001770 ret = run_delayed_extent_op(trans, head, extent_op);
Josef Bacikb00e6252017-09-29 15:43:53 -04001771 btrfs_free_delayed_extent_op(extent_op);
1772 return ret ? ret : 1;
1773}
1774
Josef Bacik31890da2018-11-21 14:05:41 -05001775void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
1776 struct btrfs_delayed_ref_root *delayed_refs,
1777 struct btrfs_delayed_ref_head *head)
Josef Bacik07c47772018-12-03 10:20:30 -05001778{
Josef Bacikba2c4d42018-12-03 10:20:33 -05001779 int nr_items = 1; /* Dropping this ref head update. */
Josef Bacik07c47772018-12-03 10:20:30 -05001780
1781 if (head->total_ref_mod < 0) {
1782 struct btrfs_space_info *space_info;
1783 u64 flags;
1784
1785 if (head->is_data)
1786 flags = BTRFS_BLOCK_GROUP_DATA;
1787 else if (head->is_system)
1788 flags = BTRFS_BLOCK_GROUP_SYSTEM;
1789 else
1790 flags = BTRFS_BLOCK_GROUP_METADATA;
Josef Bacik280c29082019-06-18 16:09:19 -04001791 space_info = btrfs_find_space_info(fs_info, flags);
Josef Bacik07c47772018-12-03 10:20:30 -05001792 ASSERT(space_info);
1793 percpu_counter_add_batch(&space_info->total_bytes_pinned,
1794 -head->num_bytes,
1795 BTRFS_TOTAL_BYTES_PINNED_BATCH);
1796
Josef Bacikba2c4d42018-12-03 10:20:33 -05001797 /*
1798 * We had csum deletions accounted for in our delayed refs rsv,
1799 * we need to drop the csum leaves for this update from our
1800 * delayed_refs_rsv.
1801 */
Josef Bacik07c47772018-12-03 10:20:30 -05001802 if (head->is_data) {
1803 spin_lock(&delayed_refs->lock);
1804 delayed_refs->pending_csums -= head->num_bytes;
1805 spin_unlock(&delayed_refs->lock);
Josef Bacikba2c4d42018-12-03 10:20:33 -05001806 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
1807 head->num_bytes);
Josef Bacik07c47772018-12-03 10:20:30 -05001808 }
1809 }
1810
Josef Bacikba2c4d42018-12-03 10:20:33 -05001811 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
Josef Bacik07c47772018-12-03 10:20:30 -05001812}
1813
Josef Bacik194ab0b2017-09-29 15:43:54 -04001814static int cleanup_ref_head(struct btrfs_trans_handle *trans,
Josef Bacik194ab0b2017-09-29 15:43:54 -04001815 struct btrfs_delayed_ref_head *head)
1816{
Nikolay Borisovf9871ed2018-06-20 15:49:03 +03001817
1818 struct btrfs_fs_info *fs_info = trans->fs_info;
Josef Bacik194ab0b2017-09-29 15:43:54 -04001819 struct btrfs_delayed_ref_root *delayed_refs;
1820 int ret;
1821
1822 delayed_refs = &trans->transaction->delayed_refs;
1823
Josef Bacikbedc66172018-12-03 10:20:31 -05001824 ret = run_and_cleanup_extent_op(trans, head);
Josef Bacik194ab0b2017-09-29 15:43:54 -04001825 if (ret < 0) {
1826 unselect_delayed_ref_head(delayed_refs, head);
1827 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
1828 return ret;
1829 } else if (ret) {
1830 return ret;
1831 }
1832
1833 /*
1834 * Need to drop our head ref lock and re-acquire the delayed ref lock
1835 * and then re-check to make sure nobody got added.
1836 */
1837 spin_unlock(&head->lock);
1838 spin_lock(&delayed_refs->lock);
1839 spin_lock(&head->lock);
Liu Boe3d03962018-08-23 03:51:50 +08001840 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
Josef Bacik194ab0b2017-09-29 15:43:54 -04001841 spin_unlock(&head->lock);
1842 spin_unlock(&delayed_refs->lock);
1843 return 1;
1844 }
Josef Bacikd7baffd2018-12-03 10:20:29 -05001845 btrfs_delete_ref_head(delayed_refs, head);
Josef Bacikc1103f72017-09-29 15:43:56 -04001846 spin_unlock(&head->lock);
Nikolay Borisov1e7a1422018-04-11 11:21:18 +03001847 spin_unlock(&delayed_refs->lock);
Josef Bacikc1103f72017-09-29 15:43:56 -04001848
Josef Bacikc1103f72017-09-29 15:43:56 -04001849 if (head->must_insert_reserved) {
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02001850 btrfs_pin_extent(trans, head->bytenr, head->num_bytes, 1);
Josef Bacikc1103f72017-09-29 15:43:56 -04001851 if (head->is_data) {
Filipe Manana40e046a2019-12-05 16:58:30 +00001852 ret = btrfs_del_csums(trans, fs_info->csum_root,
1853 head->bytenr, head->num_bytes);
Josef Bacikc1103f72017-09-29 15:43:56 -04001854 }
1855 }
1856
Josef Bacik31890da2018-11-21 14:05:41 -05001857 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
Josef Bacik07c47772018-12-03 10:20:30 -05001858
1859 trace_run_delayed_ref_head(fs_info, head, 0);
Josef Bacikc1103f72017-09-29 15:43:56 -04001860 btrfs_delayed_ref_unlock(head);
Josef Bacikd2788502017-09-29 15:43:57 -04001861 btrfs_put_delayed_ref_head(head);
Josef Bacik194ab0b2017-09-29 15:43:54 -04001862 return 0;
1863}
1864
Nikolay Borisovb1cdbcb2018-08-15 10:39:54 +03001865static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
1866 struct btrfs_trans_handle *trans)
1867{
1868 struct btrfs_delayed_ref_root *delayed_refs =
1869 &trans->transaction->delayed_refs;
1870 struct btrfs_delayed_ref_head *head = NULL;
1871 int ret;
1872
1873 spin_lock(&delayed_refs->lock);
Lu Fengqi5637c742018-10-11 13:40:33 +08001874 head = btrfs_select_ref_head(delayed_refs);
Nikolay Borisovb1cdbcb2018-08-15 10:39:54 +03001875 if (!head) {
1876 spin_unlock(&delayed_refs->lock);
1877 return head;
1878 }
1879
1880 /*
1881 * Grab the lock that says we are going to process all the refs for
1882 * this head
1883 */
Lu Fengqi9e920a62018-10-11 13:40:34 +08001884 ret = btrfs_delayed_ref_lock(delayed_refs, head);
Nikolay Borisovb1cdbcb2018-08-15 10:39:54 +03001885 spin_unlock(&delayed_refs->lock);
1886
1887 /*
1888 * We may have dropped the spin lock to get the head mutex lock, and
1889 * that might have given someone else time to free the head. If that's
1890 * true, it has been removed from our list and we can move on.
1891 */
1892 if (ret == -EAGAIN)
1893 head = ERR_PTR(-EAGAIN);
1894
1895 return head;
1896}
1897
Nikolay Borisove7261382018-08-15 10:39:55 +03001898static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
1899 struct btrfs_delayed_ref_head *locked_ref,
1900 unsigned long *run_refs)
1901{
1902 struct btrfs_fs_info *fs_info = trans->fs_info;
1903 struct btrfs_delayed_ref_root *delayed_refs;
1904 struct btrfs_delayed_extent_op *extent_op;
1905 struct btrfs_delayed_ref_node *ref;
1906 int must_insert_reserved = 0;
1907 int ret;
1908
1909 delayed_refs = &trans->transaction->delayed_refs;
1910
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03001911 lockdep_assert_held(&locked_ref->mutex);
1912 lockdep_assert_held(&locked_ref->lock);
1913
Nikolay Borisove7261382018-08-15 10:39:55 +03001914 while ((ref = select_delayed_ref(locked_ref))) {
1915 if (ref->seq &&
1916 btrfs_check_delayed_seq(fs_info, ref->seq)) {
1917 spin_unlock(&locked_ref->lock);
1918 unselect_delayed_ref_head(delayed_refs, locked_ref);
1919 return -EAGAIN;
1920 }
1921
1922 (*run_refs)++;
1923 ref->in_tree = 0;
1924 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
1925 RB_CLEAR_NODE(&ref->ref_node);
1926 if (!list_empty(&ref->add_list))
1927 list_del(&ref->add_list);
1928 /*
1929 * When we play the delayed ref, also correct the ref_mod on
1930 * head
1931 */
1932 switch (ref->action) {
1933 case BTRFS_ADD_DELAYED_REF:
1934 case BTRFS_ADD_DELAYED_EXTENT:
1935 locked_ref->ref_mod -= ref->ref_mod;
1936 break;
1937 case BTRFS_DROP_DELAYED_REF:
1938 locked_ref->ref_mod += ref->ref_mod;
1939 break;
1940 default:
1941 WARN_ON(1);
1942 }
1943 atomic_dec(&delayed_refs->num_entries);
1944
1945 /*
1946 * Record the must_insert_reserved flag before we drop the
1947 * spin lock.
1948 */
1949 must_insert_reserved = locked_ref->must_insert_reserved;
1950 locked_ref->must_insert_reserved = 0;
1951
1952 extent_op = locked_ref->extent_op;
1953 locked_ref->extent_op = NULL;
1954 spin_unlock(&locked_ref->lock);
1955
1956 ret = run_one_delayed_ref(trans, ref, extent_op,
1957 must_insert_reserved);
1958
1959 btrfs_free_delayed_extent_op(extent_op);
1960 if (ret) {
1961 unselect_delayed_ref_head(delayed_refs, locked_ref);
1962 btrfs_put_delayed_ref(ref);
1963 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
1964 ret);
1965 return ret;
1966 }
1967
1968 btrfs_put_delayed_ref(ref);
1969 cond_resched();
1970
1971 spin_lock(&locked_ref->lock);
1972 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
1973 }
1974
1975 return 0;
1976}
1977
Jeff Mahoney79787ea2012-03-12 16:03:00 +01001978/*
1979 * Returns 0 on success or if called with an already aborted transaction.
1980 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
1981 */
Josef Bacikd7df2c72014-01-23 09:21:38 -05001982static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
Josef Bacikd7df2c72014-01-23 09:21:38 -05001983 unsigned long nr)
Chris Mason56bec292009-03-13 10:10:06 -04001984{
Nikolay Borisov0a1e4582018-03-15 16:00:27 +02001985 struct btrfs_fs_info *fs_info = trans->fs_info;
Chris Mason56bec292009-03-13 10:10:06 -04001986 struct btrfs_delayed_ref_root *delayed_refs;
Chris Mason56bec292009-03-13 10:10:06 -04001987 struct btrfs_delayed_ref_head *locked_ref = NULL;
Josef Bacik0a2b2a82014-01-23 10:54:11 -05001988 ktime_t start = ktime_get();
Chris Mason56bec292009-03-13 10:10:06 -04001989 int ret;
Josef Bacikd7df2c72014-01-23 09:21:38 -05001990 unsigned long count = 0;
Josef Bacik0a2b2a82014-01-23 10:54:11 -05001991 unsigned long actual_count = 0;
Chris Mason56bec292009-03-13 10:10:06 -04001992
1993 delayed_refs = &trans->transaction->delayed_refs;
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03001994 do {
Chris Mason56bec292009-03-13 10:10:06 -04001995 if (!locked_ref) {
Nikolay Borisovb1cdbcb2018-08-15 10:39:54 +03001996 locked_ref = btrfs_obtain_ref_head(trans);
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03001997 if (IS_ERR_OR_NULL(locked_ref)) {
1998 if (PTR_ERR(locked_ref) == -EAGAIN) {
1999 continue;
2000 } else {
2001 break;
2002 }
Chris Mason56bec292009-03-13 10:10:06 -04002003 }
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03002004 count++;
Chris Mason56bec292009-03-13 10:10:06 -04002005 }
Filipe Manana2c3cf7d2015-10-22 09:47:34 +01002006 /*
2007 * We need to try and merge add/drops of the same ref since we
2008 * can run into issues with relocate dropping the implicit ref
2009 * and then it being added back again before the drop can
2010 * finish. If we merged anything we need to re-loop so we can
2011 * get a good ref.
2012 * Or we can get node references of the same type that weren't
2013 * merged when created due to bumps in the tree mod seq, and
2014 * we need to merge them to prevent adding an inline extent
2015 * backref before dropping it (triggering a BUG_ON at
2016 * insert_inline_extent_backref()).
2017 */
Josef Bacikd7df2c72014-01-23 09:21:38 -05002018 spin_lock(&locked_ref->lock);
Nikolay Borisovbe97f132018-04-19 11:06:39 +03002019 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
Josef Bacikae1e2062012-08-07 16:00:32 -04002020
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03002021 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2022 &actual_count);
2023 if (ret < 0 && ret != -EAGAIN) {
2024 /*
2025 * Error, btrfs_run_delayed_refs_for_head already
2026 * unlocked everything so just bail out
2027 */
2028 return ret;
2029 } else if (!ret) {
2030 /*
2031 * Success, perform the usual cleanup of a processed
2032 * head
2033 */
Nikolay Borisovf9871ed2018-06-20 15:49:03 +03002034 ret = cleanup_ref_head(trans, locked_ref);
Josef Bacik194ab0b2017-09-29 15:43:54 -04002035 if (ret > 0 ) {
Josef Bacikb00e6252017-09-29 15:43:53 -04002036 /* We dropped our lock, we need to loop. */
2037 ret = 0;
Josef Bacikd7df2c72014-01-23 09:21:38 -05002038 continue;
Josef Bacik194ab0b2017-09-29 15:43:54 -04002039 } else if (ret) {
2040 return ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002041 }
Josef Bacikc1103f72017-09-29 15:43:56 -04002042 }
Chris Mason56bec292009-03-13 10:10:06 -04002043
Josef Bacikb00e6252017-09-29 15:43:53 -04002044 /*
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03002045 * Either success case or btrfs_run_delayed_refs_for_head
2046 * returned -EAGAIN, meaning we need to select another head
Josef Bacikb00e6252017-09-29 15:43:53 -04002047 */
Josef Bacikb00e6252017-09-29 15:43:53 -04002048
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03002049 locked_ref = NULL;
Chris Mason1887be62009-03-13 10:11:24 -04002050 cond_resched();
Nikolay Borisov0110a4c2018-08-15 10:39:56 +03002051 } while ((nr != -1 && count < nr) || locked_ref);
Josef Bacik0a2b2a82014-01-23 10:54:11 -05002052
2053 /*
2054 * We don't want to include ref heads since we can have empty ref heads
2055 * and those will drastically skew our runtime down since we just do
2056 * accounting, no actual extent tree updates.
2057 */
2058 if (actual_count > 0) {
2059 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2060 u64 avg;
2061
2062 /*
2063 * We weigh the current average higher than our current runtime
2064 * to avoid large swings in the average.
2065 */
2066 spin_lock(&delayed_refs->lock);
2067 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
David Sterbaf8c269d2015-01-16 17:21:12 +01002068 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
Josef Bacik0a2b2a82014-01-23 10:54:11 -05002069 spin_unlock(&delayed_refs->lock);
2070 }
Josef Bacikd7df2c72014-01-23 09:21:38 -05002071 return 0;
Chris Masonc3e69d52009-03-13 10:17:05 -04002072}
2073
Arne Jansen709c0482011-09-12 12:22:57 +02002074#ifdef SCRAMBLE_DELAYED_REFS
2075/*
2076 * Normally delayed refs get processed in ascending bytenr order. This
2077 * correlates in most cases to the order added. To expose dependencies on this
2078 * order, we start to process the tree in the middle instead of the beginning
2079 */
2080static u64 find_middle(struct rb_root *root)
2081{
2082 struct rb_node *n = root->rb_node;
2083 struct btrfs_delayed_ref_node *entry;
2084 int alt = 1;
2085 u64 middle;
2086 u64 first = 0, last = 0;
2087
2088 n = rb_first(root);
2089 if (n) {
2090 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2091 first = entry->bytenr;
2092 }
2093 n = rb_last(root);
2094 if (n) {
2095 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2096 last = entry->bytenr;
2097 }
2098 n = root->rb_node;
2099
2100 while (n) {
2101 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2102 WARN_ON(!entry->in_tree);
2103
2104 middle = entry->bytenr;
2105
2106 if (alt)
2107 n = n->rb_left;
2108 else
2109 n = n->rb_right;
2110
2111 alt = 1 - alt;
2112 }
2113 return middle;
2114}
2115#endif
2116
Josef Bacik12621332015-02-03 07:50:16 -08002117/*
2118 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2119 * would require to store the csums for that many bytes.
2120 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002121u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
Josef Bacik12621332015-02-03 07:50:16 -08002122{
2123 u64 csum_size;
2124 u64 num_csums_per_leaf;
2125 u64 num_csums;
2126
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002127 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
Josef Bacik12621332015-02-03 07:50:16 -08002128 num_csums_per_leaf = div64_u64(csum_size,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002129 (u64)btrfs_super_csum_size(fs_info->super_copy));
2130 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
Josef Bacik12621332015-02-03 07:50:16 -08002131 num_csums += num_csums_per_leaf - 1;
2132 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2133 return num_csums;
2134}
2135
Chris Masonc3e69d52009-03-13 10:17:05 -04002136/*
2137 * this starts processing the delayed reference count updates and
2138 * extent insertions we have queued up so far. count can be
2139 * 0, which means to process everything in the tree at the start
2140 * of the run (but not newly added entries), or it can be some target
2141 * number you'd like to process.
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002142 *
2143 * Returns 0 on success or if called with an aborted transaction
2144 * Returns <0 on error and aborts the transaction
Chris Masonc3e69d52009-03-13 10:17:05 -04002145 */
2146int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
Nikolay Borisovc79a70b2018-03-15 17:27:37 +02002147 unsigned long count)
Chris Masonc3e69d52009-03-13 10:17:05 -04002148{
Nikolay Borisovc79a70b2018-03-15 17:27:37 +02002149 struct btrfs_fs_info *fs_info = trans->fs_info;
Chris Masonc3e69d52009-03-13 10:17:05 -04002150 struct rb_node *node;
2151 struct btrfs_delayed_ref_root *delayed_refs;
Liu Boc46effa2013-10-14 12:59:45 +08002152 struct btrfs_delayed_ref_head *head;
Chris Masonc3e69d52009-03-13 10:17:05 -04002153 int ret;
2154 int run_all = count == (unsigned long)-1;
Chris Masonc3e69d52009-03-13 10:17:05 -04002155
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002156 /* We'll clean this up in btrfs_cleanup_transaction */
David Sterbabf31f872020-02-05 17:34:34 +01002157 if (TRANS_ABORTED(trans))
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002158 return 0;
2159
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002160 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
Chris Mason511711a2015-12-30 07:52:35 -08002161 return 0;
2162
Chris Masonc3e69d52009-03-13 10:17:05 -04002163 delayed_refs = &trans->transaction->delayed_refs;
Liu Bo26455d32014-12-17 16:14:09 +08002164 if (count == 0)
Josef Bacikd7df2c72014-01-23 09:21:38 -05002165 count = atomic_read(&delayed_refs->num_entries) * 2;
Chris Masonbb721702013-01-29 18:44:12 -05002166
Chris Masonc3e69d52009-03-13 10:17:05 -04002167again:
Arne Jansen709c0482011-09-12 12:22:57 +02002168#ifdef SCRAMBLE_DELAYED_REFS
2169 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2170#endif
Nikolay Borisov0a1e4582018-03-15 16:00:27 +02002171 ret = __btrfs_run_delayed_refs(trans, count);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002172 if (ret < 0) {
Jeff Mahoney66642832016-06-10 18:19:25 -04002173 btrfs_abort_transaction(trans, ret);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002174 return ret;
Chris Masonc3e69d52009-03-13 10:17:05 -04002175 }
2176
Chris Mason56bec292009-03-13 10:10:06 -04002177 if (run_all) {
Josef Bacik119e80d2018-11-21 14:05:42 -05002178 btrfs_create_pending_block_groups(trans);
Josef Bacikea658ba2012-09-11 16:57:25 -04002179
Josef Bacikd7df2c72014-01-23 09:21:38 -05002180 spin_lock(&delayed_refs->lock);
Liu Bo5c9d0282018-08-23 03:51:49 +08002181 node = rb_first_cached(&delayed_refs->href_root);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002182 if (!node) {
2183 spin_unlock(&delayed_refs->lock);
Chris Mason56bec292009-03-13 10:10:06 -04002184 goto out;
Josef Bacikd7df2c72014-01-23 09:21:38 -05002185 }
Josef Bacikd2788502017-09-29 15:43:57 -04002186 head = rb_entry(node, struct btrfs_delayed_ref_head,
2187 href_node);
2188 refcount_inc(&head->refs);
Chris Mason56bec292009-03-13 10:10:06 -04002189 spin_unlock(&delayed_refs->lock);
Josef Bacikd2788502017-09-29 15:43:57 -04002190
2191 /* Mutex was contended, block until it's released and retry. */
2192 mutex_lock(&head->mutex);
2193 mutex_unlock(&head->mutex);
2194
2195 btrfs_put_delayed_ref_head(head);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002196 cond_resched();
Chris Mason56bec292009-03-13 10:10:06 -04002197 goto again;
2198 }
Chris Mason54aa1f42007-06-22 14:16:25 -04002199out:
Chris Masona28ec192007-03-06 20:08:01 -05002200 return 0;
2201}
2202
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002203int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
David Sterba42c9d0b2019-03-20 11:54:13 +01002204 struct extent_buffer *eb, u64 flags,
Josef Bacikb1c79e02013-05-09 13:49:30 -04002205 int level, int is_data)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002206{
2207 struct btrfs_delayed_extent_op *extent_op;
2208 int ret;
2209
Miao Xie78a61842012-11-21 02:21:28 +00002210 extent_op = btrfs_alloc_delayed_extent_op();
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002211 if (!extent_op)
2212 return -ENOMEM;
2213
2214 extent_op->flags_to_set = flags;
David Sterba35b3ad52015-11-30 16:51:29 +01002215 extent_op->update_flags = true;
2216 extent_op->update_key = false;
2217 extent_op->is_data = is_data ? true : false;
Josef Bacikb1c79e02013-05-09 13:49:30 -04002218 extent_op->level = level;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002219
David Sterba42c9d0b2019-03-20 11:54:13 +01002220 ret = btrfs_add_delayed_extent_op(trans, eb->start, eb->len, extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002221 if (ret)
Miao Xie78a61842012-11-21 02:21:28 +00002222 btrfs_free_delayed_extent_op(extent_op);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002223 return ret;
2224}
2225
Liu Boe4c3b2d2017-01-30 12:25:28 -08002226static noinline int check_delayed_ref(struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002227 struct btrfs_path *path,
2228 u64 objectid, u64 offset, u64 bytenr)
2229{
2230 struct btrfs_delayed_ref_head *head;
2231 struct btrfs_delayed_ref_node *ref;
2232 struct btrfs_delayed_data_ref *data_ref;
2233 struct btrfs_delayed_ref_root *delayed_refs;
Liu Boe4c3b2d2017-01-30 12:25:28 -08002234 struct btrfs_transaction *cur_trans;
Josef Bacik0e0adbc2017-10-19 14:16:00 -04002235 struct rb_node *node;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002236 int ret = 0;
2237
ethanwu998ac6d2018-04-29 15:59:42 +08002238 spin_lock(&root->fs_info->trans_lock);
Liu Boe4c3b2d2017-01-30 12:25:28 -08002239 cur_trans = root->fs_info->running_transaction;
ethanwu998ac6d2018-04-29 15:59:42 +08002240 if (cur_trans)
2241 refcount_inc(&cur_trans->use_count);
2242 spin_unlock(&root->fs_info->trans_lock);
Liu Boe4c3b2d2017-01-30 12:25:28 -08002243 if (!cur_trans)
2244 return 0;
2245
2246 delayed_refs = &cur_trans->delayed_refs;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002247 spin_lock(&delayed_refs->lock);
Liu Bof72ad18e2017-01-30 12:24:37 -08002248 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002249 if (!head) {
2250 spin_unlock(&delayed_refs->lock);
ethanwu998ac6d2018-04-29 15:59:42 +08002251 btrfs_put_transaction(cur_trans);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002252 return 0;
2253 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002254
2255 if (!mutex_trylock(&head->mutex)) {
Josef Bacikd2788502017-09-29 15:43:57 -04002256 refcount_inc(&head->refs);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002257 spin_unlock(&delayed_refs->lock);
2258
David Sterbab3b4aa72011-04-21 01:20:15 +02002259 btrfs_release_path(path);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002260
David Sterba8cc33e52011-05-02 15:29:25 +02002261 /*
2262 * Mutex was contended, block until it's released and let
2263 * caller try again
2264 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002265 mutex_lock(&head->mutex);
2266 mutex_unlock(&head->mutex);
Josef Bacikd2788502017-09-29 15:43:57 -04002267 btrfs_put_delayed_ref_head(head);
ethanwu998ac6d2018-04-29 15:59:42 +08002268 btrfs_put_transaction(cur_trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002269 return -EAGAIN;
2270 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002271 spin_unlock(&delayed_refs->lock);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002272
2273 spin_lock(&head->lock);
Josef Bacik0e0adbc2017-10-19 14:16:00 -04002274 /*
2275 * XXX: We should replace this with a proper search function in the
2276 * future.
2277 */
Liu Boe3d03962018-08-23 03:51:50 +08002278 for (node = rb_first_cached(&head->ref_tree); node;
2279 node = rb_next(node)) {
Josef Bacik0e0adbc2017-10-19 14:16:00 -04002280 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
Josef Bacikd7df2c72014-01-23 09:21:38 -05002281 /* If it's a shared ref we know a cross reference exists */
2282 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2283 ret = 1;
2284 break;
2285 }
2286
2287 data_ref = btrfs_delayed_node_to_data_ref(ref);
2288
2289 /*
2290 * If our ref doesn't match the one we're currently looking at
2291 * then we have a cross reference.
2292 */
2293 if (data_ref->root != root->root_key.objectid ||
2294 data_ref->objectid != objectid ||
2295 data_ref->offset != offset) {
2296 ret = 1;
2297 break;
2298 }
2299 }
2300 spin_unlock(&head->lock);
2301 mutex_unlock(&head->mutex);
ethanwu998ac6d2018-04-29 15:59:42 +08002302 btrfs_put_transaction(cur_trans);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002303 return ret;
2304}
2305
Liu Boe4c3b2d2017-01-30 12:25:28 -08002306static noinline int check_committed_ref(struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002307 struct btrfs_path *path,
2308 u64 objectid, u64 offset, u64 bytenr)
Chris Masonbe20aa92007-12-17 20:14:01 -05002309{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002310 struct btrfs_fs_info *fs_info = root->fs_info;
2311 struct btrfs_root *extent_root = fs_info->extent_root;
Yan Zhengf321e492008-07-30 09:26:11 -04002312 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002313 struct btrfs_extent_data_ref *ref;
2314 struct btrfs_extent_inline_ref *iref;
2315 struct btrfs_extent_item *ei;
Chris Masonbe20aa92007-12-17 20:14:01 -05002316 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002317 u32 item_size;
Liu Bo3de28d52017-08-18 15:15:19 -06002318 int type;
Yan Zhengf321e492008-07-30 09:26:11 -04002319 int ret;
Chris Masonbe20aa92007-12-17 20:14:01 -05002320
Chris Masonbe20aa92007-12-17 20:14:01 -05002321 key.objectid = bytenr;
Zheng Yan31840ae2008-09-23 13:14:14 -04002322 key.offset = (u64)-1;
Yan Zhengf321e492008-07-30 09:26:11 -04002323 key.type = BTRFS_EXTENT_ITEM_KEY;
Chris Masonbe20aa92007-12-17 20:14:01 -05002324
Chris Masonbe20aa92007-12-17 20:14:01 -05002325 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2326 if (ret < 0)
2327 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002328 BUG_ON(ret == 0); /* Corruption */
Yan Zheng80ff3852008-10-30 14:20:02 -04002329
2330 ret = -ENOENT;
2331 if (path->slots[0] == 0)
Zheng Yan31840ae2008-09-23 13:14:14 -04002332 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002333
Zheng Yan31840ae2008-09-23 13:14:14 -04002334 path->slots[0]--;
Yan Zhengf321e492008-07-30 09:26:11 -04002335 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
Chris Masonbe20aa92007-12-17 20:14:01 -05002337
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002338 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
Chris Masonbe20aa92007-12-17 20:14:01 -05002339 goto out;
Chris Masonbe20aa92007-12-17 20:14:01 -05002340
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002341 ret = 1;
2342 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002343 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2344
Nikolay Borisova6bd9cd2019-08-21 10:42:57 +03002345 /* If extent item has more than 1 inline ref then it's shared */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002346 if (item_size != sizeof(*ei) +
2347 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2348 goto out;
2349
Nikolay Borisova6bd9cd2019-08-21 10:42:57 +03002350 /* If extent created before last snapshot => it's definitely shared */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002351 if (btrfs_extent_generation(leaf, ei) <=
2352 btrfs_root_last_snapshot(&root->root_item))
2353 goto out;
2354
2355 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
Liu Bo3de28d52017-08-18 15:15:19 -06002356
Nikolay Borisova6bd9cd2019-08-21 10:42:57 +03002357 /* If this extent has SHARED_DATA_REF then it's shared */
Liu Bo3de28d52017-08-18 15:15:19 -06002358 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
2359 if (type != BTRFS_EXTENT_DATA_REF_KEY)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002360 goto out;
2361
2362 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
2363 if (btrfs_extent_refs(leaf, ei) !=
2364 btrfs_extent_data_ref_count(leaf, ref) ||
2365 btrfs_extent_data_ref_root(leaf, ref) !=
2366 root->root_key.objectid ||
2367 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
2368 btrfs_extent_data_ref_offset(leaf, ref) != offset)
2369 goto out;
2370
Yan Zhengf321e492008-07-30 09:26:11 -04002371 ret = 0;
Chris Masonbe20aa92007-12-17 20:14:01 -05002372out:
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002373 return ret;
2374}
2375
Liu Boe4c3b2d2017-01-30 12:25:28 -08002376int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
2377 u64 bytenr)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002378{
2379 struct btrfs_path *path;
2380 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002381
2382 path = btrfs_alloc_path();
2383 if (!path)
Su Yue9132c4f2018-05-30 14:49:10 +08002384 return -ENOMEM;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002385
2386 do {
Liu Boe4c3b2d2017-01-30 12:25:28 -08002387 ret = check_committed_ref(root, path, objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002388 offset, bytenr);
2389 if (ret && ret != -ENOENT)
2390 goto out;
2391
Misono Tomohiro380fd062018-08-30 10:59:16 +09002392 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
2393 } while (ret == -EAGAIN);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002394
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002395out:
Yan Zhengf321e492008-07-30 09:26:11 -04002396 btrfs_free_path(path);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002397 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2398 WARN_ON(ret > 0);
Yan Zhengf321e492008-07-30 09:26:11 -04002399 return ret;
2400}
2401
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002402static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
Chris Masonb7a9f292009-02-04 09:23:45 -05002403 struct btrfs_root *root,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002404 struct extent_buffer *buf,
Josef Bacike339a6b2014-07-02 10:54:25 -07002405 int full_backref, int inc)
Zheng Yan31840ae2008-09-23 13:14:14 -04002406{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002407 struct btrfs_fs_info *fs_info = root->fs_info;
Zheng Yan31840ae2008-09-23 13:14:14 -04002408 u64 bytenr;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002409 u64 num_bytes;
2410 u64 parent;
Zheng Yan31840ae2008-09-23 13:14:14 -04002411 u64 ref_root;
Zheng Yan31840ae2008-09-23 13:14:14 -04002412 u32 nritems;
Zheng Yan31840ae2008-09-23 13:14:14 -04002413 struct btrfs_key key;
2414 struct btrfs_file_extent_item *fi;
Qu Wenruo82fa1132019-04-04 14:45:35 +08002415 struct btrfs_ref generic_ref = { 0 };
2416 bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC);
Zheng Yan31840ae2008-09-23 13:14:14 -04002417 int i;
Qu Wenruo82fa1132019-04-04 14:45:35 +08002418 int action;
Zheng Yan31840ae2008-09-23 13:14:14 -04002419 int level;
2420 int ret = 0;
David Sterbafccb84c2014-09-29 23:53:21 +02002421
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002422 if (btrfs_is_testing(fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -04002423 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +02002424
Zheng Yan31840ae2008-09-23 13:14:14 -04002425 ref_root = btrfs_header_owner(buf);
Zheng Yan31840ae2008-09-23 13:14:14 -04002426 nritems = btrfs_header_nritems(buf);
2427 level = btrfs_header_level(buf);
2428
Qu Wenruo92a7cc42020-05-15 14:01:40 +08002429 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && level == 0)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002430 return 0;
Chris Masonb7a9f292009-02-04 09:23:45 -05002431
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002432 if (full_backref)
2433 parent = buf->start;
2434 else
2435 parent = 0;
Qu Wenruo82fa1132019-04-04 14:45:35 +08002436 if (inc)
2437 action = BTRFS_ADD_DELAYED_REF;
2438 else
2439 action = BTRFS_DROP_DELAYED_REF;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002440
Zheng Yan31840ae2008-09-23 13:14:14 -04002441 for (i = 0; i < nritems; i++) {
Chris Masondb945352007-10-15 16:15:53 -04002442 if (level == 0) {
Chris Mason5f39d392007-10-15 16:14:19 -04002443 btrfs_item_key_to_cpu(buf, &key, i);
David Sterba962a2982014-06-04 18:41:45 +02002444 if (key.type != BTRFS_EXTENT_DATA_KEY)
Chris Mason54aa1f42007-06-22 14:16:25 -04002445 continue;
Chris Mason5f39d392007-10-15 16:14:19 -04002446 fi = btrfs_item_ptr(buf, i,
Chris Mason54aa1f42007-06-22 14:16:25 -04002447 struct btrfs_file_extent_item);
Chris Mason5f39d392007-10-15 16:14:19 -04002448 if (btrfs_file_extent_type(buf, fi) ==
Chris Mason54aa1f42007-06-22 14:16:25 -04002449 BTRFS_FILE_EXTENT_INLINE)
2450 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002451 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
2452 if (bytenr == 0)
Chris Mason54aa1f42007-06-22 14:16:25 -04002453 continue;
Zheng Yan31840ae2008-09-23 13:14:14 -04002454
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002455 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
2456 key.offset -= btrfs_file_extent_offset(buf, fi);
Qu Wenruo82fa1132019-04-04 14:45:35 +08002457 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2458 num_bytes, parent);
2459 generic_ref.real_root = root->root_key.objectid;
2460 btrfs_init_data_ref(&generic_ref, ref_root, key.objectid,
2461 key.offset);
2462 generic_ref.skip_qgroup = for_reloc;
Qu Wenruodd28b6a2019-04-04 14:45:30 +08002463 if (inc)
Qu Wenruo82fa1132019-04-04 14:45:35 +08002464 ret = btrfs_inc_extent_ref(trans, &generic_ref);
Qu Wenruodd28b6a2019-04-04 14:45:30 +08002465 else
Qu Wenruoffd4bb22019-04-04 14:45:36 +08002466 ret = btrfs_free_extent(trans, &generic_ref);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002467 if (ret)
2468 goto fail;
Chris Masonb7a9f292009-02-04 09:23:45 -05002469 } else {
2470 bytenr = btrfs_node_blockptr(buf, i);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002471 num_bytes = fs_info->nodesize;
Qu Wenruo82fa1132019-04-04 14:45:35 +08002472 btrfs_init_generic_ref(&generic_ref, action, bytenr,
2473 num_bytes, parent);
2474 generic_ref.real_root = root->root_key.objectid;
2475 btrfs_init_tree_ref(&generic_ref, level - 1, ref_root);
2476 generic_ref.skip_qgroup = for_reloc;
Qu Wenruodd28b6a2019-04-04 14:45:30 +08002477 if (inc)
Qu Wenruo82fa1132019-04-04 14:45:35 +08002478 ret = btrfs_inc_extent_ref(trans, &generic_ref);
Qu Wenruodd28b6a2019-04-04 14:45:30 +08002479 else
Qu Wenruoffd4bb22019-04-04 14:45:36 +08002480 ret = btrfs_free_extent(trans, &generic_ref);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002481 if (ret)
Zheng Yan31840ae2008-09-23 13:14:14 -04002482 goto fail;
Chris Mason54aa1f42007-06-22 14:16:25 -04002483 }
2484 }
Zheng Yan31840ae2008-09-23 13:14:14 -04002485 return 0;
2486fail:
Chris Mason54aa1f42007-06-22 14:16:25 -04002487 return ret;
Chris Mason02217ed2007-03-02 16:08:05 -05002488}
2489
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002490int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Josef Bacike339a6b2014-07-02 10:54:25 -07002491 struct extent_buffer *buf, int full_backref)
Zheng Yan31840ae2008-09-23 13:14:14 -04002492{
Josef Bacike339a6b2014-07-02 10:54:25 -07002493 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002494}
Zheng Yan31840ae2008-09-23 13:14:14 -04002495
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002496int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Josef Bacike339a6b2014-07-02 10:54:25 -07002497 struct extent_buffer *buf, int full_backref)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002498{
Josef Bacike339a6b2014-07-02 10:54:25 -07002499 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
Zheng Yan31840ae2008-09-23 13:14:14 -04002500}
2501
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002502int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
Yan Zhengd2fb3432008-12-11 16:30:39 -05002503{
David Sterba32da53862019-10-29 19:20:18 +01002504 struct btrfs_block_group *block_group;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002505 int readonly = 0;
2506
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002507 block_group = btrfs_lookup_block_group(fs_info, bytenr);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002508 if (!block_group || block_group->ro)
2509 readonly = 1;
2510 if (block_group)
Chris Masonfa9c0d792009-04-03 09:47:43 -04002511 btrfs_put_block_group(block_group);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002512 return readonly;
2513}
2514
Jeff Mahoney1b868262017-05-17 11:38:35 -04002515static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
Yan, Zhengb742bb822010-05-16 10:46:24 -04002516{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002517 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengb742bb822010-05-16 10:46:24 -04002518 u64 flags;
David Woodhouse53b381b2013-01-29 18:40:14 -05002519 u64 ret;
Josef Bacik6a632092009-02-20 11:00:09 -05002520
Yan, Zhengb742bb822010-05-16 10:46:24 -04002521 if (data)
2522 flags = BTRFS_BLOCK_GROUP_DATA;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002523 else if (root == fs_info->chunk_root)
Yan, Zhengb742bb822010-05-16 10:46:24 -04002524 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2525 else
2526 flags = BTRFS_BLOCK_GROUP_METADATA;
2527
Josef Bacik878d7b62019-06-20 15:38:05 -04002528 ret = btrfs_get_alloc_profile(fs_info, flags);
David Woodhouse53b381b2013-01-29 18:40:14 -05002529 return ret;
Josef Bacik6a632092009-02-20 11:00:09 -05002530}
2531
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002532static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
Chris Masona061fc82008-05-07 11:43:44 -04002533{
David Sterba32da53862019-10-29 19:20:18 +01002534 struct btrfs_block_group *cache;
Yan Zhengd2fb3432008-12-11 16:30:39 -05002535 u64 bytenr;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002536
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002537 spin_lock(&fs_info->block_group_cache_lock);
2538 bytenr = fs_info->first_logical_byte;
2539 spin_unlock(&fs_info->block_group_cache_lock);
Liu Boa1897fd2012-12-27 09:01:23 +00002540
2541 if (bytenr < (u64)-1)
2542 return bytenr;
2543
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002544 cache = btrfs_lookup_first_block_group(fs_info, search_start);
Josef Bacik0f9dd462008-09-23 13:14:11 -04002545 if (!cache)
Chris Masona061fc82008-05-07 11:43:44 -04002546 return 0;
Josef Bacik0f9dd462008-09-23 13:14:11 -04002547
David Sterbab3470b52019-10-23 18:48:22 +02002548 bytenr = cache->start;
Chris Masonfa9c0d792009-04-03 09:47:43 -04002549 btrfs_put_block_group(cache);
Yan Zhengd2fb3432008-12-11 16:30:39 -05002550
2551 return bytenr;
Chris Masona061fc82008-05-07 11:43:44 -04002552}
2553
Nikolay Borisov6690d072020-01-20 16:09:14 +02002554static int pin_down_extent(struct btrfs_trans_handle *trans,
2555 struct btrfs_block_group *cache,
Yan, Zhengf0486c62010-05-16 10:46:25 -04002556 u64 bytenr, u64 num_bytes, int reserved)
Yan324ae4d2007-11-16 14:57:08 -05002557{
David Sterbafdf08602019-03-20 12:12:32 +01002558 struct btrfs_fs_info *fs_info = cache->fs_info;
2559
Yan Zheng11833d62009-09-11 16:11:19 -04002560 spin_lock(&cache->space_info->lock);
2561 spin_lock(&cache->lock);
2562 cache->pinned += num_bytes;
Josef Bacikbb96c4e2019-06-18 16:09:21 -04002563 btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info,
2564 num_bytes);
Yan Zheng11833d62009-09-11 16:11:19 -04002565 if (reserved) {
2566 cache->reserved -= num_bytes;
2567 cache->space_info->bytes_reserved -= num_bytes;
Yan324ae4d2007-11-16 14:57:08 -05002568 }
Yan Zheng11833d62009-09-11 16:11:19 -04002569 spin_unlock(&cache->lock);
2570 spin_unlock(&cache->space_info->lock);
2571
Ethan Liendec59fa2018-07-13 16:50:42 +08002572 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
2573 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
Nikolay Borisovfe119a62020-01-20 16:09:18 +02002574 set_extent_dirty(&trans->transaction->pinned_extents, bytenr,
Yan, Zhengf0486c62010-05-16 10:46:25 -04002575 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
Yan324ae4d2007-11-16 14:57:08 -05002576 return 0;
2577}
Chris Mason9078a3e2007-04-26 16:46:15 -04002578
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02002579int btrfs_pin_extent(struct btrfs_trans_handle *trans,
Yan, Zhengf0486c62010-05-16 10:46:25 -04002580 u64 bytenr, u64 num_bytes, int reserved)
Zheng Yane8569812008-09-26 10:05:48 -04002581{
David Sterba32da53862019-10-29 19:20:18 +01002582 struct btrfs_block_group *cache;
Yan, Zhengf0486c62010-05-16 10:46:25 -04002583
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02002584 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002585 BUG_ON(!cache); /* Logic error */
Yan, Zhengf0486c62010-05-16 10:46:25 -04002586
Nikolay Borisov6690d072020-01-20 16:09:14 +02002587 pin_down_extent(trans, cache, bytenr, num_bytes, reserved);
Yan, Zhengf0486c62010-05-16 10:46:25 -04002588
2589 btrfs_put_block_group(cache);
Yan Zheng11833d62009-09-11 16:11:19 -04002590 return 0;
2591}
Zheng Yane8569812008-09-26 10:05:48 -04002592
Yan, Zhengf0486c62010-05-16 10:46:25 -04002593/*
Chris Masone688b7252011-10-31 20:52:39 -04002594 * this function must be called within transaction
Yan, Zhengf0486c62010-05-16 10:46:25 -04002595 */
Nikolay Borisov9fce5702020-01-20 16:09:13 +02002596int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
Chris Masone688b7252011-10-31 20:52:39 -04002597 u64 bytenr, u64 num_bytes)
Yan, Zhengf0486c62010-05-16 10:46:25 -04002598{
David Sterba32da53862019-10-29 19:20:18 +01002599 struct btrfs_block_group *cache;
Josef Bacikb50c6e22013-04-25 15:55:30 -04002600 int ret;
Chris Masone688b7252011-10-31 20:52:39 -04002601
Nikolay Borisovf2fb7292020-01-20 16:09:16 +02002602 btrfs_add_excluded_extent(trans->fs_info, bytenr, num_bytes);
2603
Nikolay Borisov9fce5702020-01-20 16:09:13 +02002604 cache = btrfs_lookup_block_group(trans->fs_info, bytenr);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002605 if (!cache)
2606 return -EINVAL;
Chris Masone688b7252011-10-31 20:52:39 -04002607
2608 /*
2609 * pull in the free space cache (if any) so that our pin
2610 * removes the free space from the cache. We have load_only set
2611 * to one because the slow code to read in the free extents does check
2612 * the pinned extents.
2613 */
Josef Bacik676f1f72019-06-20 15:37:48 -04002614 btrfs_cache_block_group(cache, 1);
Chris Masone688b7252011-10-31 20:52:39 -04002615
Nikolay Borisov6690d072020-01-20 16:09:14 +02002616 pin_down_extent(trans, cache, bytenr, num_bytes, 0);
Chris Masone688b7252011-10-31 20:52:39 -04002617
2618 /* remove us from the free space cache (if we're there at all) */
Josef Bacikb50c6e22013-04-25 15:55:30 -04002619 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
Chris Masone688b7252011-10-31 20:52:39 -04002620 btrfs_put_block_group(cache);
Josef Bacikb50c6e22013-04-25 15:55:30 -04002621 return ret;
Chris Masone688b7252011-10-31 20:52:39 -04002622}
2623
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002624static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
2625 u64 start, u64 num_bytes)
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002626{
2627 int ret;
David Sterba32da53862019-10-29 19:20:18 +01002628 struct btrfs_block_group *block_group;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002629 struct btrfs_caching_control *caching_ctl;
2630
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002631 block_group = btrfs_lookup_block_group(fs_info, start);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002632 if (!block_group)
2633 return -EINVAL;
2634
Josef Bacik676f1f72019-06-20 15:37:48 -04002635 btrfs_cache_block_group(block_group, 0);
Josef Bacike3cb3392019-06-20 15:37:50 -04002636 caching_ctl = btrfs_get_caching_control(block_group);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002637
2638 if (!caching_ctl) {
2639 /* Logic error */
David Sterba32da53862019-10-29 19:20:18 +01002640 BUG_ON(!btrfs_block_group_done(block_group));
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002641 ret = btrfs_remove_free_space(block_group, start, num_bytes);
2642 } else {
2643 mutex_lock(&caching_ctl->mutex);
2644
2645 if (start >= caching_ctl->progress) {
Josef Bacik6f410d12019-06-20 15:37:49 -04002646 ret = btrfs_add_excluded_extent(fs_info, start,
2647 num_bytes);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002648 } else if (start + num_bytes <= caching_ctl->progress) {
2649 ret = btrfs_remove_free_space(block_group,
2650 start, num_bytes);
2651 } else {
2652 num_bytes = caching_ctl->progress - start;
2653 ret = btrfs_remove_free_space(block_group,
2654 start, num_bytes);
2655 if (ret)
2656 goto out_lock;
2657
2658 num_bytes = (start + num_bytes) -
2659 caching_ctl->progress;
2660 start = caching_ctl->progress;
Josef Bacik6f410d12019-06-20 15:37:49 -04002661 ret = btrfs_add_excluded_extent(fs_info, start,
2662 num_bytes);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002663 }
2664out_lock:
2665 mutex_unlock(&caching_ctl->mutex);
Josef Bacike3cb3392019-06-20 15:37:50 -04002666 btrfs_put_caching_control(caching_ctl);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002667 }
2668 btrfs_put_block_group(block_group);
2669 return ret;
2670}
2671
David Sterbabcdc4282019-03-20 12:14:33 +01002672int btrfs_exclude_logged_extents(struct extent_buffer *eb)
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002673{
David Sterbabcdc4282019-03-20 12:14:33 +01002674 struct btrfs_fs_info *fs_info = eb->fs_info;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002675 struct btrfs_file_extent_item *item;
2676 struct btrfs_key key;
2677 int found_type;
2678 int i;
Gu Jinxiangb89311e2018-05-22 17:46:51 +08002679 int ret = 0;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002680
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002681 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002682 return 0;
2683
2684 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2685 btrfs_item_key_to_cpu(eb, &key, i);
2686 if (key.type != BTRFS_EXTENT_DATA_KEY)
2687 continue;
2688 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2689 found_type = btrfs_file_extent_type(eb, item);
2690 if (found_type == BTRFS_FILE_EXTENT_INLINE)
2691 continue;
2692 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
2693 continue;
2694 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
2695 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
Gu Jinxiangb89311e2018-05-22 17:46:51 +08002696 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
2697 if (ret)
2698 break;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002699 }
2700
Gu Jinxiangb89311e2018-05-22 17:46:51 +08002701 return ret;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04002702}
2703
Filipe Manana9cfa3e32016-04-26 15:39:32 +01002704static void
David Sterba32da53862019-10-29 19:20:18 +01002705btrfs_inc_block_group_reservations(struct btrfs_block_group *bg)
Filipe Manana9cfa3e32016-04-26 15:39:32 +01002706{
2707 atomic_inc(&bg->reservations);
2708}
2709
David Sterba8b74c032017-02-10 19:20:56 +01002710void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
Yan Zheng11833d62009-09-11 16:11:19 -04002711{
Yan Zheng11833d62009-09-11 16:11:19 -04002712 struct btrfs_caching_control *next;
2713 struct btrfs_caching_control *caching_ctl;
David Sterba32da53862019-10-29 19:20:18 +01002714 struct btrfs_block_group *cache;
Yan Zheng11833d62009-09-11 16:11:19 -04002715
Josef Bacik9e351cc2014-03-13 15:42:13 -04002716 down_write(&fs_info->commit_root_sem);
Yan Zheng11833d62009-09-11 16:11:19 -04002717
2718 list_for_each_entry_safe(caching_ctl, next,
2719 &fs_info->caching_block_groups, list) {
2720 cache = caching_ctl->block_group;
David Sterba32da53862019-10-29 19:20:18 +01002721 if (btrfs_block_group_done(cache)) {
Yan Zheng11833d62009-09-11 16:11:19 -04002722 cache->last_byte_to_unpin = (u64)-1;
2723 list_del_init(&caching_ctl->list);
Josef Bacike3cb3392019-06-20 15:37:50 -04002724 btrfs_put_caching_control(caching_ctl);
Yan Zheng11833d62009-09-11 16:11:19 -04002725 } else {
2726 cache->last_byte_to_unpin = caching_ctl->progress;
2727 }
2728 }
2729
Josef Bacik9e351cc2014-03-13 15:42:13 -04002730 up_write(&fs_info->commit_root_sem);
Yan, Zheng8929ecfa2010-05-16 10:49:58 -04002731
Josef Bacik67f9c222019-06-19 13:47:23 -04002732 btrfs_update_global_block_rsv(fs_info);
Yan Zheng11833d62009-09-11 16:11:19 -04002733}
2734
Josef Bacikc759c4e2015-10-02 15:25:10 -04002735/*
2736 * Returns the free cluster for the given space info and sets empty_cluster to
2737 * what it should be based on the mount options.
2738 */
2739static struct btrfs_free_cluster *
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002740fetch_cluster_info(struct btrfs_fs_info *fs_info,
2741 struct btrfs_space_info *space_info, u64 *empty_cluster)
Josef Bacikc759c4e2015-10-02 15:25:10 -04002742{
2743 struct btrfs_free_cluster *ret = NULL;
Josef Bacikc759c4e2015-10-02 15:25:10 -04002744
2745 *empty_cluster = 0;
2746 if (btrfs_mixed_space_info(space_info))
2747 return ret;
2748
Josef Bacikc759c4e2015-10-02 15:25:10 -04002749 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002750 ret = &fs_info->meta_alloc_cluster;
Hans van Kranenburg583b7232017-07-28 08:31:28 +02002751 if (btrfs_test_opt(fs_info, SSD))
2752 *empty_cluster = SZ_2M;
2753 else
Byongho Leeee221842015-12-15 01:42:10 +09002754 *empty_cluster = SZ_64K;
Hans van Kranenburg583b7232017-07-28 08:31:28 +02002755 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
2756 btrfs_test_opt(fs_info, SSD_SPREAD)) {
2757 *empty_cluster = SZ_2M;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002758 ret = &fs_info->data_alloc_cluster;
Josef Bacikc759c4e2015-10-02 15:25:10 -04002759 }
2760
2761 return ret;
2762}
2763
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002764static int unpin_extent_range(struct btrfs_fs_info *fs_info,
2765 u64 start, u64 end,
Filipe Manana678886b2014-12-07 21:31:47 +00002766 const bool return_free_space)
Yan Zheng11833d62009-09-11 16:11:19 -04002767{
David Sterba32da53862019-10-29 19:20:18 +01002768 struct btrfs_block_group *cache = NULL;
Josef Bacik7b398f82012-10-22 15:52:28 -04002769 struct btrfs_space_info *space_info;
2770 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
Josef Bacikc759c4e2015-10-02 15:25:10 -04002771 struct btrfs_free_cluster *cluster = NULL;
Yan Zheng11833d62009-09-11 16:11:19 -04002772 u64 len;
Josef Bacikc759c4e2015-10-02 15:25:10 -04002773 u64 total_unpinned = 0;
2774 u64 empty_cluster = 0;
Josef Bacik7b398f82012-10-22 15:52:28 -04002775 bool readonly;
Yan Zheng11833d62009-09-11 16:11:19 -04002776
2777 while (start <= end) {
Josef Bacik7b398f82012-10-22 15:52:28 -04002778 readonly = false;
Yan Zheng11833d62009-09-11 16:11:19 -04002779 if (!cache ||
David Sterbab3470b52019-10-23 18:48:22 +02002780 start >= cache->start + cache->length) {
Yan Zheng11833d62009-09-11 16:11:19 -04002781 if (cache)
2782 btrfs_put_block_group(cache);
Josef Bacikc759c4e2015-10-02 15:25:10 -04002783 total_unpinned = 0;
Yan Zheng11833d62009-09-11 16:11:19 -04002784 cache = btrfs_lookup_block_group(fs_info, start);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01002785 BUG_ON(!cache); /* Logic error */
Josef Bacikc759c4e2015-10-02 15:25:10 -04002786
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002787 cluster = fetch_cluster_info(fs_info,
Josef Bacikc759c4e2015-10-02 15:25:10 -04002788 cache->space_info,
2789 &empty_cluster);
2790 empty_cluster <<= 1;
Yan Zheng11833d62009-09-11 16:11:19 -04002791 }
2792
David Sterbab3470b52019-10-23 18:48:22 +02002793 len = cache->start + cache->length - start;
Yan Zheng11833d62009-09-11 16:11:19 -04002794 len = min(len, end + 1 - start);
2795
2796 if (start < cache->last_byte_to_unpin) {
2797 len = min(len, cache->last_byte_to_unpin - start);
Filipe Manana678886b2014-12-07 21:31:47 +00002798 if (return_free_space)
2799 btrfs_add_free_space(cache, start, len);
Yan Zheng11833d62009-09-11 16:11:19 -04002800 }
Josef Bacik25179202008-10-29 14:49:05 -04002801
Yan, Zhengf0486c62010-05-16 10:46:25 -04002802 start += len;
Josef Bacikc759c4e2015-10-02 15:25:10 -04002803 total_unpinned += len;
Josef Bacik7b398f82012-10-22 15:52:28 -04002804 space_info = cache->space_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04002805
Josef Bacikc759c4e2015-10-02 15:25:10 -04002806 /*
2807 * If this space cluster has been marked as fragmented and we've
2808 * unpinned enough in this block group to potentially allow a
2809 * cluster to be created inside of it go ahead and clear the
2810 * fragmented check.
2811 */
2812 if (cluster && cluster->fragmented &&
2813 total_unpinned > empty_cluster) {
2814 spin_lock(&cluster->lock);
2815 cluster->fragmented = 0;
2816 spin_unlock(&cluster->lock);
2817 }
2818
Josef Bacik7b398f82012-10-22 15:52:28 -04002819 spin_lock(&space_info->lock);
Josef Bacik25179202008-10-29 14:49:05 -04002820 spin_lock(&cache->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04002821 cache->pinned -= len;
Josef Bacikbb96c4e2019-06-18 16:09:21 -04002822 btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len);
Josef Bacik4f4db212015-09-29 11:40:47 -04002823 space_info->max_extent_size = 0;
Ethan Liendec59fa2018-07-13 16:50:42 +08002824 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2825 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
Josef Bacik7b398f82012-10-22 15:52:28 -04002826 if (cache->ro) {
2827 space_info->bytes_readonly += len;
2828 readonly = true;
2829 }
Josef Bacik25179202008-10-29 14:49:05 -04002830 spin_unlock(&cache->lock);
Josef Bacik957780e2016-05-17 13:30:55 -04002831 if (!readonly && return_free_space &&
2832 global_rsv->space_info == space_info) {
2833 u64 to_add = len;
Nikolay Borisov92ac58e2017-08-17 10:52:28 +03002834
Josef Bacik7b398f82012-10-22 15:52:28 -04002835 spin_lock(&global_rsv->lock);
2836 if (!global_rsv->full) {
Josef Bacik957780e2016-05-17 13:30:55 -04002837 to_add = min(len, global_rsv->size -
2838 global_rsv->reserved);
2839 global_rsv->reserved += to_add;
Josef Bacikbb96c4e2019-06-18 16:09:21 -04002840 btrfs_space_info_update_bytes_may_use(fs_info,
2841 space_info, to_add);
Josef Bacik7b398f82012-10-22 15:52:28 -04002842 if (global_rsv->reserved >= global_rsv->size)
2843 global_rsv->full = 1;
Josef Bacik957780e2016-05-17 13:30:55 -04002844 len -= to_add;
Josef Bacik7b398f82012-10-22 15:52:28 -04002845 }
2846 spin_unlock(&global_rsv->lock);
Josef Bacik957780e2016-05-17 13:30:55 -04002847 /* Add to any tickets we may have */
2848 if (len)
Josef Bacik18fa2282019-08-22 15:10:58 -04002849 btrfs_try_granting_tickets(fs_info,
2850 space_info);
Josef Bacik7b398f82012-10-22 15:52:28 -04002851 }
2852 spin_unlock(&space_info->lock);
Yan Zheng11833d62009-09-11 16:11:19 -04002853 }
2854
2855 if (cache)
Chris Masonfa9c0d792009-04-03 09:47:43 -04002856 btrfs_put_block_group(cache);
Chris Masonccd467d2007-06-28 15:57:36 -04002857 return 0;
2858}
2859
Nikolay Borisov5ead2dd2018-03-15 16:00:26 +02002860int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
Chris Masona28ec192007-03-06 20:08:01 -05002861{
Nikolay Borisov5ead2dd2018-03-15 16:00:26 +02002862 struct btrfs_fs_info *fs_info = trans->fs_info;
David Sterba32da53862019-10-29 19:20:18 +01002863 struct btrfs_block_group *block_group, *tmp;
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04002864 struct list_head *deleted_bgs;
Yan Zheng11833d62009-09-11 16:11:19 -04002865 struct extent_io_tree *unpin;
Chris Mason1a5bc162007-10-15 16:15:26 -04002866 u64 start;
2867 u64 end;
Chris Masona28ec192007-03-06 20:08:01 -05002868 int ret;
Chris Masona28ec192007-03-06 20:08:01 -05002869
Nikolay Borisovfe119a62020-01-20 16:09:18 +02002870 unpin = &trans->transaction->pinned_extents;
Yan Zheng11833d62009-09-11 16:11:19 -04002871
David Sterbabf31f872020-02-05 17:34:34 +01002872 while (!TRANS_ABORTED(trans)) {
Filipe Manana0e6ec382018-11-16 13:04:44 +00002873 struct extent_state *cached_state = NULL;
2874
Filipe Mananad4b450c2015-01-29 19:18:25 +00002875 mutex_lock(&fs_info->unused_bg_unpin_mutex);
Chris Mason1a5bc162007-10-15 16:15:26 -04002876 ret = find_first_extent_bit(unpin, 0, &start, &end,
Filipe Manana0e6ec382018-11-16 13:04:44 +00002877 EXTENT_DIRTY, &cached_state);
Filipe Mananad4b450c2015-01-29 19:18:25 +00002878 if (ret) {
2879 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
Chris Masona28ec192007-03-06 20:08:01 -05002880 break;
Filipe Mananad4b450c2015-01-29 19:18:25 +00002881 }
Nikolay Borisovfe119a62020-01-20 16:09:18 +02002882 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
2883 clear_extent_bits(&fs_info->excluded_extents, start,
Nikolay Borisovf2fb7292020-01-20 16:09:16 +02002884 end, EXTENT_UPTODATE);
Liu Hui1f3c79a2009-01-05 15:57:51 -05002885
Dennis Zhou46b27f52019-12-13 16:22:11 -08002886 if (btrfs_test_opt(fs_info, DISCARD_SYNC))
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002887 ret = btrfs_discard_extent(fs_info, start,
Li Dongyang5378e602011-03-24 10:24:27 +00002888 end + 1 - start, NULL);
Liu Hui1f3c79a2009-01-05 15:57:51 -05002889
Filipe Manana0e6ec382018-11-16 13:04:44 +00002890 clear_extent_dirty(unpin, start, end, &cached_state);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002891 unpin_extent_range(fs_info, start, end, true);
Filipe Mananad4b450c2015-01-29 19:18:25 +00002892 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
Filipe Manana0e6ec382018-11-16 13:04:44 +00002893 free_extent_state(cached_state);
Chris Masonb9473432009-03-13 11:00:37 -04002894 cond_resched();
Chris Masona28ec192007-03-06 20:08:01 -05002895 }
Josef Bacik817d52f2009-07-13 21:29:25 -04002896
Dennis Zhoua2309302020-01-02 16:26:35 -05002897 if (btrfs_test_opt(fs_info, DISCARD_ASYNC)) {
2898 btrfs_discard_calc_delay(&fs_info->discard_ctl);
Dennis Zhoub0643e52019-12-13 16:22:14 -08002899 btrfs_discard_schedule_work(&fs_info->discard_ctl, true);
Dennis Zhoua2309302020-01-02 16:26:35 -05002900 }
Dennis Zhoub0643e52019-12-13 16:22:14 -08002901
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04002902 /*
2903 * Transaction is finished. We don't need the lock anymore. We
2904 * do need to clean up the block groups in case of a transaction
2905 * abort.
2906 */
2907 deleted_bgs = &trans->transaction->deleted_bgs;
2908 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
2909 u64 trimmed = 0;
2910
2911 ret = -EROFS;
David Sterbabf31f872020-02-05 17:34:34 +01002912 if (!TRANS_ABORTED(trans))
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04002913 ret = btrfs_discard_extent(fs_info,
David Sterbab3470b52019-10-23 18:48:22 +02002914 block_group->start,
2915 block_group->length,
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04002916 &trimmed);
2917
2918 list_del_init(&block_group->bg_list);
Filipe Manana6b7304a2020-05-08 11:01:47 +01002919 btrfs_unfreeze_block_group(block_group);
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04002920 btrfs_put_block_group(block_group);
2921
2922 if (ret) {
2923 const char *errstr = btrfs_decode_error(ret);
2924 btrfs_warn(fs_info,
David Sterba913e1532017-07-13 15:32:18 +02002925 "discard failed while removing blockgroup: errno=%d %s",
Jeff Mahoneye33e17e2015-06-15 09:41:19 -04002926 ret, errstr);
2927 }
2928 }
2929
Chris Masone20d96d2007-03-22 12:13:20 -04002930 return 0;
2931}
2932
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002933static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
Nikolay Borisove72cb922018-06-20 15:48:57 +03002934 struct btrfs_delayed_ref_node *node, u64 parent,
2935 u64 root_objectid, u64 owner_objectid,
2936 u64 owner_offset, int refs_to_drop,
2937 struct btrfs_delayed_extent_op *extent_op)
Chris Masona28ec192007-03-06 20:08:01 -05002938{
Nikolay Borisove72cb922018-06-20 15:48:57 +03002939 struct btrfs_fs_info *info = trans->fs_info;
Chris Masone2fa7222007-03-12 16:22:34 -04002940 struct btrfs_key key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002941 struct btrfs_path *path;
Chris Mason1261ec42007-03-20 20:35:03 -04002942 struct btrfs_root *extent_root = info->extent_root;
Chris Mason5f39d392007-10-15 16:14:19 -04002943 struct extent_buffer *leaf;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002944 struct btrfs_extent_item *ei;
2945 struct btrfs_extent_inline_ref *iref;
Chris Masona28ec192007-03-06 20:08:01 -05002946 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002947 int is_data;
Chris Mason952fcca2008-02-18 16:33:44 -05002948 int extent_slot = 0;
2949 int found_extent = 0;
2950 int num_to_del = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002951 u32 item_size;
2952 u64 refs;
Qu Wenruoc682f9b2015-03-17 16:59:47 +08002953 u64 bytenr = node->bytenr;
2954 u64 num_bytes = node->num_bytes;
Josef Bacikfcebe452014-05-13 17:30:47 -07002955 int last_ref = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04002956 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
Chris Mason037e6392007-03-07 11:50:24 -05002957
Chris Mason5caf2a02007-04-02 11:20:42 -04002958 path = btrfs_alloc_path();
Chris Mason54aa1f42007-06-22 14:16:25 -04002959 if (!path)
2960 return -ENOMEM;
2961
Chris Masonb9473432009-03-13 11:00:37 -04002962 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002963
2964 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2965 BUG_ON(!is_data && refs_to_drop != 1);
2966
Josef Bacik3173a182013-03-07 14:22:04 -05002967 if (is_data)
Thomas Meyer897ca812017-10-07 16:02:21 +02002968 skinny_metadata = false;
Josef Bacik3173a182013-03-07 14:22:04 -05002969
Nikolay Borisovfbe48012018-06-20 15:48:52 +03002970 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
2971 parent, root_objectid, owner_objectid,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002972 owner_offset);
Chris Mason7bb86312007-12-11 09:25:06 -05002973 if (ret == 0) {
Chris Mason952fcca2008-02-18 16:33:44 -05002974 extent_slot = path->slots[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002975 while (extent_slot >= 0) {
2976 btrfs_item_key_to_cpu(path->nodes[0], &key,
Chris Mason952fcca2008-02-18 16:33:44 -05002977 extent_slot);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002978 if (key.objectid != bytenr)
Chris Mason952fcca2008-02-18 16:33:44 -05002979 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002980 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2981 key.offset == num_bytes) {
Chris Mason952fcca2008-02-18 16:33:44 -05002982 found_extent = 1;
2983 break;
2984 }
Josef Bacik3173a182013-03-07 14:22:04 -05002985 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2986 key.offset == owner_objectid) {
2987 found_extent = 1;
2988 break;
2989 }
Chris Mason952fcca2008-02-18 16:33:44 -05002990 if (path->slots[0] - extent_slot > 5)
2991 break;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002992 extent_slot--;
Chris Mason952fcca2008-02-18 16:33:44 -05002993 }
Nikolay Borisova79865c2018-06-21 09:45:00 +03002994
Zheng Yan31840ae2008-09-23 13:14:14 -04002995 if (!found_extent) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04002996 BUG_ON(iref);
Nikolay Borisov87cc7a82018-06-20 15:49:12 +03002997 ret = remove_extent_backref(trans, path, NULL,
Jeff Mahoney87bde3c2017-02-15 16:28:27 -05002998 refs_to_drop,
Josef Bacikfcebe452014-05-13 17:30:47 -07002999 is_data, &last_ref);
David Sterba005d6422012-09-18 07:52:32 -06003000 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003001 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003002 goto out;
3003 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003004 btrfs_release_path(path);
Chris Masonb9473432009-03-13 11:00:37 -04003005 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003006
3007 key.objectid = bytenr;
3008 key.type = BTRFS_EXTENT_ITEM_KEY;
3009 key.offset = num_bytes;
3010
Josef Bacik3173a182013-03-07 14:22:04 -05003011 if (!is_data && skinny_metadata) {
3012 key.type = BTRFS_METADATA_ITEM_KEY;
3013 key.offset = owner_objectid;
3014 }
3015
Zheng Yan31840ae2008-09-23 13:14:14 -04003016 ret = btrfs_search_slot(trans, extent_root,
3017 &key, path, -1, 1);
Josef Bacik3173a182013-03-07 14:22:04 -05003018 if (ret > 0 && skinny_metadata && path->slots[0]) {
3019 /*
3020 * Couldn't find our skinny metadata item,
3021 * see if we have ye olde extent item.
3022 */
3023 path->slots[0]--;
3024 btrfs_item_key_to_cpu(path->nodes[0], &key,
3025 path->slots[0]);
3026 if (key.objectid == bytenr &&
3027 key.type == BTRFS_EXTENT_ITEM_KEY &&
3028 key.offset == num_bytes)
3029 ret = 0;
3030 }
3031
3032 if (ret > 0 && skinny_metadata) {
3033 skinny_metadata = false;
Filipe Manana9ce49a02014-04-24 15:15:28 +01003034 key.objectid = bytenr;
Josef Bacik3173a182013-03-07 14:22:04 -05003035 key.type = BTRFS_EXTENT_ITEM_KEY;
3036 key.offset = num_bytes;
3037 btrfs_release_path(path);
3038 ret = btrfs_search_slot(trans, extent_root,
3039 &key, path, -1, 1);
3040 }
3041
Josef Bacikf3465ca2008-11-12 14:19:50 -05003042 if (ret) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003043 btrfs_err(info,
3044 "umm, got %d back from search, was looking for %llu",
3045 ret, bytenr);
Josef Bacikb783e622011-07-13 15:03:50 +00003046 if (ret > 0)
David Sterbaa4f78752017-06-29 18:37:49 +02003047 btrfs_print_leaf(path->nodes[0]);
Josef Bacikf3465ca2008-11-12 14:19:50 -05003048 }
David Sterba005d6422012-09-18 07:52:32 -06003049 if (ret < 0) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003050 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003051 goto out;
3052 }
Zheng Yan31840ae2008-09-23 13:14:14 -04003053 extent_slot = path->slots[0];
3054 }
Dulshani Gunawardhanafae7f212013-10-31 10:30:08 +05303055 } else if (WARN_ON(ret == -ENOENT)) {
David Sterbaa4f78752017-06-29 18:37:49 +02003056 btrfs_print_leaf(path->nodes[0]);
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00003057 btrfs_err(info,
3058 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02003059 bytenr, parent, root_objectid, owner_objectid,
3060 owner_offset);
Jeff Mahoney66642832016-06-10 18:19:25 -04003061 btrfs_abort_transaction(trans, ret);
Josef Bacikc4a050b2014-03-14 16:36:53 -04003062 goto out;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003063 } else {
Jeff Mahoney66642832016-06-10 18:19:25 -04003064 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003065 goto out;
Chris Mason7bb86312007-12-11 09:25:06 -05003066 }
Chris Mason5f39d392007-10-15 16:14:19 -04003067
3068 leaf = path->nodes[0];
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003069 item_size = btrfs_item_size_nr(leaf, extent_slot);
David Sterba6d8ff4e2018-06-26 16:20:59 +02003070 if (unlikely(item_size < sizeof(*ei))) {
Nikolay Borisovba3c2b12018-06-26 16:57:36 +03003071 ret = -EINVAL;
3072 btrfs_print_v0_err(info);
3073 btrfs_abort_transaction(trans, ret);
3074 goto out;
3075 }
Chris Mason952fcca2008-02-18 16:33:44 -05003076 ei = btrfs_item_ptr(leaf, extent_slot,
Chris Mason123abc82007-03-14 14:14:43 -04003077 struct btrfs_extent_item);
Josef Bacik3173a182013-03-07 14:22:04 -05003078 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
3079 key.type == BTRFS_EXTENT_ITEM_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003080 struct btrfs_tree_block_info *bi;
3081 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
3082 bi = (struct btrfs_tree_block_info *)(ei + 1);
3083 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
Chris Mason952fcca2008-02-18 16:33:44 -05003084 }
3085
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003086 refs = btrfs_extent_refs(leaf, ei);
Josef Bacik32b02532013-04-24 16:38:50 -04003087 if (refs < refs_to_drop) {
Jeff Mahoney5d163e02016-09-20 10:05:00 -04003088 btrfs_err(info,
3089 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
3090 refs_to_drop, refs, bytenr);
Josef Bacik32b02532013-04-24 16:38:50 -04003091 ret = -EINVAL;
Jeff Mahoney66642832016-06-10 18:19:25 -04003092 btrfs_abort_transaction(trans, ret);
Josef Bacik32b02532013-04-24 16:38:50 -04003093 goto out;
3094 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003095 refs -= refs_to_drop;
3096
3097 if (refs > 0) {
3098 if (extent_op)
3099 __run_delayed_extent_op(extent_op, leaf, ei);
3100 /*
3101 * In the case of inline back ref, reference count will
3102 * be updated by remove_extent_backref
3103 */
3104 if (iref) {
3105 BUG_ON(!found_extent);
3106 } else {
3107 btrfs_set_extent_refs(leaf, ei, refs);
3108 btrfs_mark_buffer_dirty(leaf);
3109 }
3110 if (found_extent) {
Nikolay Borisov87cc7a82018-06-20 15:49:12 +03003111 ret = remove_extent_backref(trans, path, iref,
3112 refs_to_drop, is_data,
3113 &last_ref);
David Sterba005d6422012-09-18 07:52:32 -06003114 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003115 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003116 goto out;
3117 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003118 }
3119 } else {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003120 if (found_extent) {
3121 BUG_ON(is_data && refs_to_drop !=
Zhaolei9ed0dea2015-08-06 22:16:24 +08003122 extent_data_ref_count(path, iref));
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003123 if (iref) {
3124 BUG_ON(path->slots[0] != extent_slot);
3125 } else {
3126 BUG_ON(path->slots[0] != extent_slot + 1);
3127 path->slots[0] = extent_slot;
3128 num_to_del = 2;
3129 }
Chris Mason78fae272007-03-25 11:35:08 -04003130 }
Chris Masonb9473432009-03-13 11:00:37 -04003131
Josef Bacikfcebe452014-05-13 17:30:47 -07003132 last_ref = 1;
Chris Mason952fcca2008-02-18 16:33:44 -05003133 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
3134 num_to_del);
David Sterba005d6422012-09-18 07:52:32 -06003135 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003136 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003137 goto out;
3138 }
David Sterbab3b4aa72011-04-21 01:20:15 +02003139 btrfs_release_path(path);
David Woodhouse21af8042008-08-12 14:13:26 +01003140
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003141 if (is_data) {
Filipe Manana40e046a2019-12-05 16:58:30 +00003142 ret = btrfs_del_csums(trans, info->csum_root, bytenr,
3143 num_bytes);
David Sterba005d6422012-09-18 07:52:32 -06003144 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003145 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003146 goto out;
3147 }
Chris Mason459931e2008-12-10 09:10:46 -05003148 }
3149
Nikolay Borisove7355e52018-05-10 15:44:55 +03003150 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
Omar Sandoval1e144fb2015-09-29 20:50:37 -07003151 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003152 btrfs_abort_transaction(trans, ret);
Omar Sandoval1e144fb2015-09-29 20:50:37 -07003153 goto out;
3154 }
3155
Josef Bacikade4b512019-06-20 15:38:01 -04003156 ret = btrfs_update_block_group(trans, bytenr, num_bytes, 0);
David Sterba005d6422012-09-18 07:52:32 -06003157 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04003158 btrfs_abort_transaction(trans, ret);
David Sterba005d6422012-09-18 07:52:32 -06003159 goto out;
3160 }
Chris Masona28ec192007-03-06 20:08:01 -05003161 }
Josef Bacikfcebe452014-05-13 17:30:47 -07003162 btrfs_release_path(path);
3163
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003164out:
Chris Mason5caf2a02007-04-02 11:20:42 -04003165 btrfs_free_path(path);
Chris Masona28ec192007-03-06 20:08:01 -05003166 return ret;
3167}
3168
3169/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04003170 * when we free an block, it is possible (and likely) that we free the last
Chris Mason1887be62009-03-13 10:11:24 -04003171 * delayed ref for that extent as well. This searches the delayed ref tree for
3172 * a given extent, and if there are no other delayed refs to be processed, it
3173 * removes it from the tree.
3174 */
3175static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003176 u64 bytenr)
Chris Mason1887be62009-03-13 10:11:24 -04003177{
3178 struct btrfs_delayed_ref_head *head;
3179 struct btrfs_delayed_ref_root *delayed_refs;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003180 int ret = 0;
Chris Mason1887be62009-03-13 10:11:24 -04003181
3182 delayed_refs = &trans->transaction->delayed_refs;
3183 spin_lock(&delayed_refs->lock);
Liu Bof72ad18e2017-01-30 12:24:37 -08003184 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
Chris Mason1887be62009-03-13 10:11:24 -04003185 if (!head)
Chris Masoncf93da72014-01-29 07:02:40 -08003186 goto out_delayed_unlock;
Chris Mason1887be62009-03-13 10:11:24 -04003187
Josef Bacikd7df2c72014-01-23 09:21:38 -05003188 spin_lock(&head->lock);
Liu Boe3d03962018-08-23 03:51:50 +08003189 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
Chris Mason1887be62009-03-13 10:11:24 -04003190 goto out;
3191
Josef Bacikbedc66172018-12-03 10:20:31 -05003192 if (cleanup_extent_op(head) != NULL)
3193 goto out;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003194
Chris Mason1887be62009-03-13 10:11:24 -04003195 /*
3196 * waiting for the lock here would deadlock. If someone else has it
3197 * locked they are already in the process of dropping it anyway
3198 */
3199 if (!mutex_trylock(&head->mutex))
3200 goto out;
3201
Josef Bacikd7baffd2018-12-03 10:20:29 -05003202 btrfs_delete_ref_head(delayed_refs, head);
Josef Bacikd7df2c72014-01-23 09:21:38 -05003203 head->processing = 0;
Josef Bacikd7baffd2018-12-03 10:20:29 -05003204
Josef Bacikd7df2c72014-01-23 09:21:38 -05003205 spin_unlock(&head->lock);
Chris Mason1887be62009-03-13 10:11:24 -04003206 spin_unlock(&delayed_refs->lock);
3207
Yan, Zhengf0486c62010-05-16 10:46:25 -04003208 BUG_ON(head->extent_op);
3209 if (head->must_insert_reserved)
3210 ret = 1;
3211
Josef Bacik31890da2018-11-21 14:05:41 -05003212 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003213 mutex_unlock(&head->mutex);
Josef Bacikd2788502017-09-29 15:43:57 -04003214 btrfs_put_delayed_ref_head(head);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003215 return ret;
Chris Mason1887be62009-03-13 10:11:24 -04003216out:
Josef Bacikd7df2c72014-01-23 09:21:38 -05003217 spin_unlock(&head->lock);
Chris Masoncf93da72014-01-29 07:02:40 -08003218
3219out_delayed_unlock:
Chris Mason1887be62009-03-13 10:11:24 -04003220 spin_unlock(&delayed_refs->lock);
3221 return 0;
3222}
3223
Yan, Zhengf0486c62010-05-16 10:46:25 -04003224void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
3225 struct btrfs_root *root,
3226 struct extent_buffer *buf,
Jan Schmidt5581a512012-05-16 17:04:52 +02003227 u64 parent, int last_ref)
Yan, Zhengf0486c62010-05-16 10:46:25 -04003228{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003229 struct btrfs_fs_info *fs_info = root->fs_info;
Qu Wenruoed4f2552019-04-04 14:45:31 +08003230 struct btrfs_ref generic_ref = { 0 };
Josef Bacikb150a4f2013-06-19 15:00:04 -04003231 int pin = 1;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003232 int ret;
3233
Qu Wenruoed4f2552019-04-04 14:45:31 +08003234 btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF,
3235 buf->start, buf->len, parent);
3236 btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf),
3237 root->root_key.objectid);
3238
Yan, Zhengf0486c62010-05-16 10:46:25 -04003239 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
Omar Sandovald7eae342017-06-06 16:45:31 -07003240 int old_ref_mod, new_ref_mod;
3241
Qu Wenruo8a5040f2019-04-04 14:45:33 +08003242 btrfs_ref_tree_mod(fs_info, &generic_ref);
Qu Wenruoed4f2552019-04-04 14:45:31 +08003243 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL,
Omar Sandovald7eae342017-06-06 16:45:31 -07003244 &old_ref_mod, &new_ref_mod);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003245 BUG_ON(ret); /* -ENOMEM */
Omar Sandovald7eae342017-06-06 16:45:31 -07003246 pin = old_ref_mod >= 0 && new_ref_mod < 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003247 }
3248
Omar Sandoval0a16c7d2017-06-06 16:45:29 -07003249 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
David Sterba32da53862019-10-29 19:20:18 +01003250 struct btrfs_block_group *cache;
Filipe Manana62198722015-01-06 20:18:45 +00003251
Yan, Zhengf0486c62010-05-16 10:46:25 -04003252 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04003253 ret = check_ref_cleanup(trans, buf->start);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003254 if (!ret)
Josef Bacik37be25b2011-08-05 10:25:38 -04003255 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003256 }
3257
Omar Sandoval4da8b762017-06-06 16:45:28 -07003258 pin = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003259 cache = btrfs_lookup_block_group(fs_info, buf->start);
Filipe Manana62198722015-01-06 20:18:45 +00003260
Yan, Zhengf0486c62010-05-16 10:46:25 -04003261 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
Nikolay Borisov6690d072020-01-20 16:09:14 +02003262 pin_down_extent(trans, cache, buf->start, buf->len, 1);
Filipe Manana62198722015-01-06 20:18:45 +00003263 btrfs_put_block_group(cache);
Josef Bacik37be25b2011-08-05 10:25:38 -04003264 goto out;
Yan, Zhengf0486c62010-05-16 10:46:25 -04003265 }
3266
3267 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
3268
3269 btrfs_add_free_space(cache, buf->start, buf->len);
Wang Xiaoguang4824f1f2016-07-25 15:51:39 +08003270 btrfs_free_reserved_bytes(cache, buf->len, 0);
Filipe Manana62198722015-01-06 20:18:45 +00003271 btrfs_put_block_group(cache);
Jeff Mahoney71ff6432016-09-06 16:00:42 -04003272 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
Yan, Zhengf0486c62010-05-16 10:46:25 -04003273 }
3274out:
Josef Bacikb150a4f2013-06-19 15:00:04 -04003275 if (pin)
Qu Wenruo78192442019-05-15 07:33:48 +08003276 add_pinned_bytes(fs_info, &generic_ref);
Josef Bacikb150a4f2013-06-19 15:00:04 -04003277
Omar Sandoval0a16c7d2017-06-06 16:45:29 -07003278 if (last_ref) {
3279 /*
3280 * Deleting the buffer, clear the corrupt flag since it doesn't
3281 * matter anymore.
3282 */
3283 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
3284 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04003285}
3286
Jeff Mahoney79787ea2012-03-12 16:03:00 +01003287/* Can return -ENOMEM */
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003288int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref)
Chris Mason925baed2008-06-25 16:01:30 -04003289{
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003290 struct btrfs_fs_info *fs_info = trans->fs_info;
Omar Sandovald7eae342017-06-06 16:45:31 -07003291 int old_ref_mod, new_ref_mod;
Chris Mason925baed2008-06-25 16:01:30 -04003292 int ret;
3293
Jeff Mahoneyf5ee5c92016-06-21 09:52:41 -04003294 if (btrfs_is_testing(fs_info))
Josef Bacikfaa2dbf2014-05-07 17:06:09 -04003295 return 0;
David Sterbafccb84c2014-09-29 23:53:21 +02003296
Chris Mason56bec292009-03-13 10:10:06 -04003297 /*
3298 * tree log blocks never actually go into the extent allocation
3299 * tree, just update pinning info and exit early.
Chris Mason56bec292009-03-13 10:10:06 -04003300 */
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003301 if ((ref->type == BTRFS_REF_METADATA &&
3302 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3303 (ref->type == BTRFS_REF_DATA &&
3304 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) {
Chris Masonb9473432009-03-13 11:00:37 -04003305 /* unlocks the pinned mutex */
Nikolay Borisovb25c36f2020-01-20 16:09:09 +02003306 btrfs_pin_extent(trans, ref->bytenr, ref->len, 1);
Omar Sandovald7eae342017-06-06 16:45:31 -07003307 old_ref_mod = new_ref_mod = 0;
Chris Mason56bec292009-03-13 10:10:06 -04003308 ret = 0;
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003309 } else if (ref->type == BTRFS_REF_METADATA) {
3310 ret = btrfs_add_delayed_tree_ref(trans, ref, NULL,
Omar Sandovald7eae342017-06-06 16:45:31 -07003311 &old_ref_mod, &new_ref_mod);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04003312 } else {
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003313 ret = btrfs_add_delayed_data_ref(trans, ref, 0,
Omar Sandovald7eae342017-06-06 16:45:31 -07003314 &old_ref_mod, &new_ref_mod);
Chris Mason56bec292009-03-13 10:10:06 -04003315 }
Omar Sandovald7eae342017-06-06 16:45:31 -07003316
Qu Wenruoffd4bb22019-04-04 14:45:36 +08003317 if (!((ref->type == BTRFS_REF_METADATA &&
3318 ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) ||
3319 (ref->type == BTRFS_REF_DATA &&
3320 ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)))
3321 btrfs_ref_tree_mod(fs_info, ref);
Qu Wenruo8a5040f2019-04-04 14:45:33 +08003322
Qu Wenruoddf30cf2019-04-04 14:45:34 +08003323 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
Qu Wenruo78192442019-05-15 07:33:48 +08003324 add_pinned_bytes(fs_info, ref);
Omar Sandovald7eae342017-06-06 16:45:31 -07003325
Chris Mason925baed2008-06-25 16:01:30 -04003326 return ret;
3327}
3328
Josef Bacik817d52f2009-07-13 21:29:25 -04003329enum btrfs_loop_type {
David Sterbaf262fa82019-06-18 20:00:08 +02003330 LOOP_CACHING_NOWAIT,
3331 LOOP_CACHING_WAIT,
3332 LOOP_ALLOC_CHUNK,
3333 LOOP_NO_EMPTY_SIZE,
Josef Bacik817d52f2009-07-13 21:29:25 -04003334};
3335
Miao Xiee570fd22014-06-19 10:42:50 +08003336static inline void
David Sterba32da53862019-10-29 19:20:18 +01003337btrfs_lock_block_group(struct btrfs_block_group *cache,
Miao Xiee570fd22014-06-19 10:42:50 +08003338 int delalloc)
3339{
3340 if (delalloc)
3341 down_read(&cache->data_rwsem);
3342}
3343
David Sterba32da53862019-10-29 19:20:18 +01003344static inline void btrfs_grab_block_group(struct btrfs_block_group *cache,
Miao Xiee570fd22014-06-19 10:42:50 +08003345 int delalloc)
3346{
3347 btrfs_get_block_group(cache);
3348 if (delalloc)
3349 down_read(&cache->data_rwsem);
3350}
3351
David Sterba32da53862019-10-29 19:20:18 +01003352static struct btrfs_block_group *btrfs_lock_cluster(
3353 struct btrfs_block_group *block_group,
Miao Xiee570fd22014-06-19 10:42:50 +08003354 struct btrfs_free_cluster *cluster,
3355 int delalloc)
Jules Irengec142c6a2020-03-31 21:46:41 +01003356 __acquires(&cluster->refill_lock)
Miao Xiee570fd22014-06-19 10:42:50 +08003357{
David Sterba32da53862019-10-29 19:20:18 +01003358 struct btrfs_block_group *used_bg = NULL;
Geert Uytterhoeven6719afd2014-06-22 14:30:09 +02003359
Miao Xiee570fd22014-06-19 10:42:50 +08003360 spin_lock(&cluster->refill_lock);
Geert Uytterhoeven6719afd2014-06-22 14:30:09 +02003361 while (1) {
3362 used_bg = cluster->block_group;
3363 if (!used_bg)
3364 return NULL;
3365
3366 if (used_bg == block_group)
3367 return used_bg;
3368
3369 btrfs_get_block_group(used_bg);
3370
3371 if (!delalloc)
3372 return used_bg;
3373
3374 if (down_read_trylock(&used_bg->data_rwsem))
3375 return used_bg;
3376
3377 spin_unlock(&cluster->refill_lock);
3378
Liu Boe321f8a2016-11-30 16:11:04 -08003379 /* We should only have one-level nested. */
3380 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
Geert Uytterhoeven6719afd2014-06-22 14:30:09 +02003381
3382 spin_lock(&cluster->refill_lock);
Miao Xiee570fd22014-06-19 10:42:50 +08003383 if (used_bg == cluster->block_group)
3384 return used_bg;
3385
3386 up_read(&used_bg->data_rwsem);
3387 btrfs_put_block_group(used_bg);
3388 }
Miao Xiee570fd22014-06-19 10:42:50 +08003389}
3390
3391static inline void
David Sterba32da53862019-10-29 19:20:18 +01003392btrfs_release_block_group(struct btrfs_block_group *cache,
Miao Xiee570fd22014-06-19 10:42:50 +08003393 int delalloc)
3394{
3395 if (delalloc)
3396 up_read(&cache->data_rwsem);
3397 btrfs_put_block_group(cache);
3398}
3399
Naohiro Aotacb2f96f2020-02-25 12:56:16 +09003400enum btrfs_extent_allocation_policy {
3401 BTRFS_EXTENT_ALLOC_CLUSTERED,
3402};
3403
Josef Bacik817d52f2009-07-13 21:29:25 -04003404/*
Qu Wenruob4bd7452018-11-02 09:39:47 +08003405 * Structure used internally for find_free_extent() function. Wraps needed
3406 * parameters.
3407 */
3408struct find_free_extent_ctl {
3409 /* Basic allocation info */
Qu Wenruob4bd7452018-11-02 09:39:47 +08003410 u64 num_bytes;
3411 u64 empty_size;
3412 u64 flags;
3413 int delalloc;
3414
3415 /* Where to start the search inside the bg */
3416 u64 search_start;
3417
3418 /* For clustered allocation */
3419 u64 empty_cluster;
Naohiro Aotac10859b2020-02-25 12:56:18 +09003420 struct btrfs_free_cluster *last_ptr;
3421 bool use_cluster;
Qu Wenruob4bd7452018-11-02 09:39:47 +08003422
3423 bool have_caching_bg;
3424 bool orig_have_caching_bg;
3425
3426 /* RAID index, converted from flags */
3427 int index;
3428
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003429 /*
3430 * Current loop number, check find_free_extent_update_loop() for details
3431 */
Qu Wenruob4bd7452018-11-02 09:39:47 +08003432 int loop;
3433
3434 /*
3435 * Whether we're refilling a cluster, if true we need to re-search
3436 * current block group but don't try to refill the cluster again.
3437 */
3438 bool retry_clustered;
3439
3440 /*
3441 * Whether we're updating free space cache, if true we need to re-search
3442 * current block group but don't try updating free space cache again.
3443 */
3444 bool retry_unclustered;
3445
3446 /* If current block group is cached */
3447 int cached;
3448
3449 /* Max contiguous hole found */
3450 u64 max_extent_size;
3451
3452 /* Total free space from free space cache, not always contiguous */
3453 u64 total_free_space;
3454
3455 /* Found result */
3456 u64 found_offset;
Naohiro Aotacb2f96f2020-02-25 12:56:16 +09003457
Naohiro Aotaea544142020-02-25 12:56:17 +09003458 /* Hint where to start looking for an empty space */
3459 u64 hint_byte;
3460
Naohiro Aotacb2f96f2020-02-25 12:56:16 +09003461 /* Allocation policy */
3462 enum btrfs_extent_allocation_policy policy;
Qu Wenruob4bd7452018-11-02 09:39:47 +08003463};
3464
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003465
3466/*
3467 * Helper function for find_free_extent().
3468 *
3469 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
3470 * Return -EAGAIN to inform caller that we need to re-search this block group
3471 * Return >0 to inform caller that we find nothing
3472 * Return 0 means we have found a location and set ffe_ctl->found_offset.
3473 */
David Sterba32da53862019-10-29 19:20:18 +01003474static int find_free_extent_clustered(struct btrfs_block_group *bg,
Naohiro Aota897cae72020-02-25 12:56:20 +09003475 struct find_free_extent_ctl *ffe_ctl,
3476 struct btrfs_block_group **cluster_bg_ret)
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003477{
David Sterba32da53862019-10-29 19:20:18 +01003478 struct btrfs_block_group *cluster_bg;
Naohiro Aota897cae72020-02-25 12:56:20 +09003479 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003480 u64 aligned_cluster;
3481 u64 offset;
3482 int ret;
3483
3484 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
3485 if (!cluster_bg)
3486 goto refill_cluster;
3487 if (cluster_bg != bg && (cluster_bg->ro ||
3488 !block_group_bits(cluster_bg, ffe_ctl->flags)))
3489 goto release_cluster;
3490
3491 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
David Sterbab3470b52019-10-23 18:48:22 +02003492 ffe_ctl->num_bytes, cluster_bg->start,
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003493 &ffe_ctl->max_extent_size);
3494 if (offset) {
3495 /* We have a block, we're done */
3496 spin_unlock(&last_ptr->refill_lock);
3497 trace_btrfs_reserve_extent_cluster(cluster_bg,
3498 ffe_ctl->search_start, ffe_ctl->num_bytes);
3499 *cluster_bg_ret = cluster_bg;
3500 ffe_ctl->found_offset = offset;
3501 return 0;
3502 }
3503 WARN_ON(last_ptr->block_group != cluster_bg);
3504
3505release_cluster:
3506 /*
3507 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
3508 * lets just skip it and let the allocator find whatever block it can
3509 * find. If we reach this point, we will have tried the cluster
3510 * allocator plenty of times and not have found anything, so we are
3511 * likely way too fragmented for the clustering stuff to find anything.
3512 *
3513 * However, if the cluster is taken from the current block group,
3514 * release the cluster first, so that we stand a better chance of
3515 * succeeding in the unclustered allocation.
3516 */
3517 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
3518 spin_unlock(&last_ptr->refill_lock);
3519 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3520 return -ENOENT;
3521 }
3522
3523 /* This cluster didn't work out, free it and start over */
3524 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3525
3526 if (cluster_bg != bg)
3527 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
3528
3529refill_cluster:
3530 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
3531 spin_unlock(&last_ptr->refill_lock);
3532 return -ENOENT;
3533 }
3534
3535 aligned_cluster = max_t(u64,
3536 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
3537 bg->full_stripe_len);
David Sterba2ceeae22019-03-20 13:53:49 +01003538 ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start,
3539 ffe_ctl->num_bytes, aligned_cluster);
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003540 if (ret == 0) {
3541 /* Now pull our allocation out of this cluster */
3542 offset = btrfs_alloc_from_cluster(bg, last_ptr,
3543 ffe_ctl->num_bytes, ffe_ctl->search_start,
3544 &ffe_ctl->max_extent_size);
3545 if (offset) {
3546 /* We found one, proceed */
3547 spin_unlock(&last_ptr->refill_lock);
3548 trace_btrfs_reserve_extent_cluster(bg,
3549 ffe_ctl->search_start,
3550 ffe_ctl->num_bytes);
3551 ffe_ctl->found_offset = offset;
3552 return 0;
3553 }
3554 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
3555 !ffe_ctl->retry_clustered) {
3556 spin_unlock(&last_ptr->refill_lock);
3557
3558 ffe_ctl->retry_clustered = true;
Josef Bacik676f1f72019-06-20 15:37:48 -04003559 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
Qu Wenruod06e3bb2018-11-02 09:39:48 +08003560 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
3561 return -EAGAIN;
3562 }
3563 /*
3564 * At this point we either didn't find a cluster or we weren't able to
3565 * allocate a block from our cluster. Free the cluster we've been
3566 * trying to use, and go to the next block group.
3567 */
3568 btrfs_return_cluster_to_free_space(NULL, last_ptr);
3569 spin_unlock(&last_ptr->refill_lock);
3570 return 1;
3571}
3572
Qu Wenruob4bd7452018-11-02 09:39:47 +08003573/*
Qu Wenruoe1a41842018-11-02 09:39:49 +08003574 * Return >0 to inform caller that we find nothing
3575 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
3576 * Return -EAGAIN to inform caller that we need to re-search this block group
3577 */
David Sterba32da53862019-10-29 19:20:18 +01003578static int find_free_extent_unclustered(struct btrfs_block_group *bg,
Naohiro Aota897cae72020-02-25 12:56:20 +09003579 struct find_free_extent_ctl *ffe_ctl)
Qu Wenruoe1a41842018-11-02 09:39:49 +08003580{
Naohiro Aota897cae72020-02-25 12:56:20 +09003581 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
Qu Wenruoe1a41842018-11-02 09:39:49 +08003582 u64 offset;
3583
3584 /*
3585 * We are doing an unclustered allocation, set the fragmented flag so
3586 * we don't bother trying to setup a cluster again until we get more
3587 * space.
3588 */
3589 if (unlikely(last_ptr)) {
3590 spin_lock(&last_ptr->lock);
3591 last_ptr->fragmented = 1;
3592 spin_unlock(&last_ptr->lock);
3593 }
3594 if (ffe_ctl->cached) {
3595 struct btrfs_free_space_ctl *free_space_ctl;
3596
3597 free_space_ctl = bg->free_space_ctl;
3598 spin_lock(&free_space_ctl->tree_lock);
3599 if (free_space_ctl->free_space <
3600 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
3601 ffe_ctl->empty_size) {
3602 ffe_ctl->total_free_space = max_t(u64,
3603 ffe_ctl->total_free_space,
3604 free_space_ctl->free_space);
3605 spin_unlock(&free_space_ctl->tree_lock);
3606 return 1;
3607 }
3608 spin_unlock(&free_space_ctl->tree_lock);
3609 }
3610
3611 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
3612 ffe_ctl->num_bytes, ffe_ctl->empty_size,
3613 &ffe_ctl->max_extent_size);
3614
3615 /*
3616 * If we didn't find a chunk, and we haven't failed on this block group
3617 * before, and this block group is in the middle of caching and we are
3618 * ok with waiting, then go ahead and wait for progress to be made, and
3619 * set @retry_unclustered to true.
3620 *
3621 * If @retry_unclustered is true then we've already waited on this
3622 * block group once and should move on to the next block group.
3623 */
3624 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
3625 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
Josef Bacik676f1f72019-06-20 15:37:48 -04003626 btrfs_wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
3627 ffe_ctl->empty_size);
Qu Wenruoe1a41842018-11-02 09:39:49 +08003628 ffe_ctl->retry_unclustered = true;
3629 return -EAGAIN;
3630 } else if (!offset) {
3631 return 1;
3632 }
3633 ffe_ctl->found_offset = offset;
3634 return 0;
3635}
3636
Naohiro Aotac6686902020-02-25 12:56:19 +09003637static int do_allocation_clustered(struct btrfs_block_group *block_group,
3638 struct find_free_extent_ctl *ffe_ctl,
3639 struct btrfs_block_group **bg_ret)
3640{
3641 int ret;
3642
3643 /* We want to try and use the cluster allocator, so lets look there */
3644 if (ffe_ctl->last_ptr && ffe_ctl->use_cluster) {
Naohiro Aota897cae72020-02-25 12:56:20 +09003645 ret = find_free_extent_clustered(block_group, ffe_ctl, bg_ret);
Naohiro Aotac6686902020-02-25 12:56:19 +09003646 if (ret >= 0 || ret == -EAGAIN)
3647 return ret;
3648 /* ret == -ENOENT case falls through */
3649 }
3650
Naohiro Aota897cae72020-02-25 12:56:20 +09003651 return find_free_extent_unclustered(block_group, ffe_ctl);
Naohiro Aotac6686902020-02-25 12:56:19 +09003652}
3653
3654static int do_allocation(struct btrfs_block_group *block_group,
3655 struct find_free_extent_ctl *ffe_ctl,
3656 struct btrfs_block_group **bg_ret)
3657{
3658 switch (ffe_ctl->policy) {
3659 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3660 return do_allocation_clustered(block_group, ffe_ctl, bg_ret);
3661 default:
3662 BUG();
3663 }
3664}
3665
Naohiro Aotababa5062020-02-25 12:56:21 +09003666static void release_block_group(struct btrfs_block_group *block_group,
3667 struct find_free_extent_ctl *ffe_ctl,
3668 int delalloc)
3669{
3670 switch (ffe_ctl->policy) {
3671 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3672 ffe_ctl->retry_clustered = false;
3673 ffe_ctl->retry_unclustered = false;
3674 break;
3675 default:
3676 BUG();
3677 }
3678
3679 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
3680 ffe_ctl->index);
3681 btrfs_release_block_group(block_group, delalloc);
3682}
3683
Naohiro Aota0ab97242020-02-25 12:56:22 +09003684static void found_extent_clustered(struct find_free_extent_ctl *ffe_ctl,
3685 struct btrfs_key *ins)
3686{
3687 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3688
3689 if (!ffe_ctl->use_cluster && last_ptr) {
3690 spin_lock(&last_ptr->lock);
3691 last_ptr->window_start = ins->objectid;
3692 spin_unlock(&last_ptr->lock);
3693 }
3694}
3695
3696static void found_extent(struct find_free_extent_ctl *ffe_ctl,
3697 struct btrfs_key *ins)
3698{
3699 switch (ffe_ctl->policy) {
3700 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3701 found_extent_clustered(ffe_ctl, ins);
3702 break;
3703 default:
3704 BUG();
3705 }
3706}
3707
Naohiro Aotac70e2132020-02-25 12:56:24 +09003708static int chunk_allocation_failed(struct find_free_extent_ctl *ffe_ctl)
3709{
3710 switch (ffe_ctl->policy) {
3711 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3712 /*
3713 * If we can't allocate a new chunk we've already looped through
3714 * at least once, move on to the NO_EMPTY_SIZE case.
3715 */
3716 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
3717 return 0;
3718 default:
3719 BUG();
3720 }
3721}
3722
Qu Wenruoe1a41842018-11-02 09:39:49 +08003723/*
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003724 * Return >0 means caller needs to re-search for free extent
3725 * Return 0 means we have the needed free extent.
3726 * Return <0 means we failed to locate any free extent.
3727 */
3728static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003729 struct btrfs_key *ins,
3730 struct find_free_extent_ctl *ffe_ctl,
Naohiro Aota15b7ee62020-02-25 12:56:23 +09003731 bool full_search)
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003732{
3733 struct btrfs_root *root = fs_info->extent_root;
3734 int ret;
3735
3736 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
3737 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
3738 ffe_ctl->orig_have_caching_bg = true;
3739
3740 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
3741 ffe_ctl->have_caching_bg)
3742 return 1;
3743
3744 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
3745 return 1;
3746
3747 if (ins->objectid) {
Naohiro Aota0ab97242020-02-25 12:56:22 +09003748 found_extent(ffe_ctl, ins);
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003749 return 0;
3750 }
3751
3752 /*
3753 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
3754 * caching kthreads as we move along
3755 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
3756 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
3757 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
3758 * again
3759 */
3760 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
3761 ffe_ctl->index = 0;
3762 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
3763 /*
3764 * We want to skip the LOOP_CACHING_WAIT step if we
3765 * don't have any uncached bgs and we've already done a
3766 * full search through.
3767 */
3768 if (ffe_ctl->orig_have_caching_bg || !full_search)
3769 ffe_ctl->loop = LOOP_CACHING_WAIT;
3770 else
3771 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
3772 } else {
3773 ffe_ctl->loop++;
3774 }
3775
3776 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
3777 struct btrfs_trans_handle *trans;
3778 int exist = 0;
3779
3780 trans = current->journal_info;
3781 if (trans)
3782 exist = 1;
3783 else
3784 trans = btrfs_join_transaction(root);
3785
3786 if (IS_ERR(trans)) {
3787 ret = PTR_ERR(trans);
3788 return ret;
3789 }
3790
Josef Bacikfc471cb2019-06-18 16:09:17 -04003791 ret = btrfs_chunk_alloc(trans, ffe_ctl->flags,
3792 CHUNK_ALLOC_FORCE);
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003793
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003794 /* Do not bail out on ENOSPC since we can do more. */
Naohiro Aotac70e2132020-02-25 12:56:24 +09003795 if (ret == -ENOSPC)
3796 ret = chunk_allocation_failed(ffe_ctl);
3797 else if (ret < 0)
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003798 btrfs_abort_transaction(trans, ret);
3799 else
3800 ret = 0;
3801 if (!exist)
3802 btrfs_end_transaction(trans);
3803 if (ret)
3804 return ret;
3805 }
3806
3807 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
Naohiro Aota45d8e032020-02-25 12:56:25 +09003808 if (ffe_ctl->policy != BTRFS_EXTENT_ALLOC_CLUSTERED)
3809 return -ENOSPC;
3810
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003811 /*
3812 * Don't loop again if we already have no empty_size and
3813 * no empty_cluster.
3814 */
3815 if (ffe_ctl->empty_size == 0 &&
3816 ffe_ctl->empty_cluster == 0)
3817 return -ENOSPC;
3818 ffe_ctl->empty_size = 0;
3819 ffe_ctl->empty_cluster = 0;
3820 }
3821 return 1;
3822 }
3823 return -ENOSPC;
3824}
3825
Naohiro Aota7e895402020-02-25 12:56:26 +09003826static int prepare_allocation_clustered(struct btrfs_fs_info *fs_info,
3827 struct find_free_extent_ctl *ffe_ctl,
3828 struct btrfs_space_info *space_info,
3829 struct btrfs_key *ins)
3830{
3831 /*
3832 * If our free space is heavily fragmented we may not be able to make
3833 * big contiguous allocations, so instead of doing the expensive search
3834 * for free space, simply return ENOSPC with our max_extent_size so we
3835 * can go ahead and search for a more manageable chunk.
3836 *
3837 * If our max_extent_size is large enough for our allocation simply
3838 * disable clustering since we will likely not be able to find enough
3839 * space to create a cluster and induce latency trying.
3840 */
3841 if (space_info->max_extent_size) {
3842 spin_lock(&space_info->lock);
3843 if (space_info->max_extent_size &&
3844 ffe_ctl->num_bytes > space_info->max_extent_size) {
3845 ins->offset = space_info->max_extent_size;
3846 spin_unlock(&space_info->lock);
3847 return -ENOSPC;
3848 } else if (space_info->max_extent_size) {
3849 ffe_ctl->use_cluster = false;
3850 }
3851 spin_unlock(&space_info->lock);
3852 }
3853
3854 ffe_ctl->last_ptr = fetch_cluster_info(fs_info, space_info,
3855 &ffe_ctl->empty_cluster);
3856 if (ffe_ctl->last_ptr) {
3857 struct btrfs_free_cluster *last_ptr = ffe_ctl->last_ptr;
3858
3859 spin_lock(&last_ptr->lock);
3860 if (last_ptr->block_group)
3861 ffe_ctl->hint_byte = last_ptr->window_start;
3862 if (last_ptr->fragmented) {
3863 /*
3864 * We still set window_start so we can keep track of the
3865 * last place we found an allocation to try and save
3866 * some time.
3867 */
3868 ffe_ctl->hint_byte = last_ptr->window_start;
3869 ffe_ctl->use_cluster = false;
3870 }
3871 spin_unlock(&last_ptr->lock);
3872 }
3873
3874 return 0;
3875}
3876
3877static int prepare_allocation(struct btrfs_fs_info *fs_info,
3878 struct find_free_extent_ctl *ffe_ctl,
3879 struct btrfs_space_info *space_info,
3880 struct btrfs_key *ins)
3881{
3882 switch (ffe_ctl->policy) {
3883 case BTRFS_EXTENT_ALLOC_CLUSTERED:
3884 return prepare_allocation_clustered(fs_info, ffe_ctl,
3885 space_info, ins);
3886 default:
3887 BUG();
3888 }
3889}
3890
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003891/*
Chris Masonfec577f2007-02-26 10:40:21 -05003892 * walks the btree of allocated extents and find a hole of a given size.
3893 * The key ins is changed to record the hole:
Miao Xiea4820392013-09-09 13:19:42 +08003894 * ins->objectid == start position
Chris Mason62e27492007-03-15 12:56:47 -04003895 * ins->flags = BTRFS_EXTENT_ITEM_KEY
Miao Xiea4820392013-09-09 13:19:42 +08003896 * ins->offset == the size of the hole.
Chris Masonfec577f2007-02-26 10:40:21 -05003897 * Any available blocks before search_start are skipped.
Miao Xiea4820392013-09-09 13:19:42 +08003898 *
3899 * If there is no suitable free space, we will record the max size of
3900 * the free space extent currently.
Qu Wenruoe72d79d2018-11-02 09:39:50 +08003901 *
3902 * The overall logic and call chain:
3903 *
3904 * find_free_extent()
3905 * |- Iterate through all block groups
3906 * | |- Get a valid block group
3907 * | |- Try to do clustered allocation in that block group
3908 * | |- Try to do unclustered allocation in that block group
3909 * | |- Check if the result is valid
3910 * | | |- If valid, then exit
3911 * | |- Jump to next block group
3912 * |
3913 * |- Push harder to find free extents
3914 * |- If not found, re-iterate all block groups
Chris Masonfec577f2007-02-26 10:40:21 -05003915 */
Jeff Mahoney87bde3c2017-02-15 16:28:27 -05003916static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
Wang Xiaoguang18513092016-07-25 15:51:40 +08003917 u64 ram_bytes, u64 num_bytes, u64 empty_size,
Naohiro Aotaea544142020-02-25 12:56:17 +09003918 u64 hint_byte_orig, struct btrfs_key *ins,
Wang Xiaoguang18513092016-07-25 15:51:40 +08003919 u64 flags, int delalloc)
Chris Masonfec577f2007-02-26 10:40:21 -05003920{
Josef Bacik80eb2342008-10-29 14:49:05 -04003921 int ret = 0;
Josef Bacikdb8fe642019-11-19 13:59:00 -05003922 int cache_block_group_error = 0;
David Sterba32da53862019-10-29 19:20:18 +01003923 struct btrfs_block_group *block_group = NULL;
Qu Wenruob4bd7452018-11-02 09:39:47 +08003924 struct find_free_extent_ctl ffe_ctl = {0};
Josef Bacik80eb2342008-10-29 14:49:05 -04003925 struct btrfs_space_info *space_info;
Josef Bacika5e681d2015-10-01 14:54:10 -04003926 bool full_search = false;
Chris Masonfec577f2007-02-26 10:40:21 -05003927
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003928 WARN_ON(num_bytes < fs_info->sectorsize);
Qu Wenruob4bd7452018-11-02 09:39:47 +08003929
Qu Wenruob4bd7452018-11-02 09:39:47 +08003930 ffe_ctl.num_bytes = num_bytes;
3931 ffe_ctl.empty_size = empty_size;
3932 ffe_ctl.flags = flags;
3933 ffe_ctl.search_start = 0;
Qu Wenruob4bd7452018-11-02 09:39:47 +08003934 ffe_ctl.delalloc = delalloc;
3935 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
3936 ffe_ctl.have_caching_bg = false;
3937 ffe_ctl.orig_have_caching_bg = false;
3938 ffe_ctl.found_offset = 0;
Naohiro Aotaea544142020-02-25 12:56:17 +09003939 ffe_ctl.hint_byte = hint_byte_orig;
Naohiro Aotacb2f96f2020-02-25 12:56:16 +09003940 ffe_ctl.policy = BTRFS_EXTENT_ALLOC_CLUSTERED;
Qu Wenruob4bd7452018-11-02 09:39:47 +08003941
Naohiro Aotac10859b2020-02-25 12:56:18 +09003942 /* For clustered allocation */
3943 ffe_ctl.retry_clustered = false;
3944 ffe_ctl.retry_unclustered = false;
3945 ffe_ctl.last_ptr = NULL;
3946 ffe_ctl.use_cluster = true;
3947
David Sterba962a2982014-06-04 18:41:45 +02003948 ins->type = BTRFS_EXTENT_ITEM_KEY;
Josef Bacik80eb2342008-10-29 14:49:05 -04003949 ins->objectid = 0;
3950 ins->offset = 0;
Chris Masonb1a4d962007-04-04 15:27:52 -04003951
Jeff Mahoney71ff6432016-09-06 16:00:42 -04003952 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
Josef Bacik3f7de032011-11-10 08:29:20 -05003953
Josef Bacik280c29082019-06-18 16:09:19 -04003954 space_info = btrfs_find_space_info(fs_info, flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00003955 if (!space_info) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04003956 btrfs_err(fs_info, "No space info for %llu", flags);
Josef Bacik1b1d1f62010-03-19 20:49:55 +00003957 return -ENOSPC;
3958 }
Josef Bacik2552d172009-04-03 10:14:19 -04003959
Naohiro Aota7e895402020-02-25 12:56:26 +09003960 ret = prepare_allocation(fs_info, &ffe_ctl, space_info, ins);
3961 if (ret < 0)
3962 return ret;
Chris Masonfa9c0d792009-04-03 09:47:43 -04003963
Qu Wenruob4bd7452018-11-02 09:39:47 +08003964 ffe_ctl.search_start = max(ffe_ctl.search_start,
3965 first_logical_byte(fs_info, 0));
Naohiro Aotaea544142020-02-25 12:56:17 +09003966 ffe_ctl.search_start = max(ffe_ctl.search_start, ffe_ctl.hint_byte);
3967 if (ffe_ctl.search_start == ffe_ctl.hint_byte) {
Qu Wenruob4bd7452018-11-02 09:39:47 +08003968 block_group = btrfs_lookup_block_group(fs_info,
3969 ffe_ctl.search_start);
Josef Bacik817d52f2009-07-13 21:29:25 -04003970 /*
3971 * we don't want to use the block group if it doesn't match our
3972 * allocation bits, or if its not cached.
Josef Bacikccf0e722009-11-10 21:23:48 -05003973 *
3974 * However if we are re-searching with an ideal block group
3975 * picked out then we don't care that the block group is cached.
Josef Bacik817d52f2009-07-13 21:29:25 -04003976 */
David Sterbab6919a52013-04-29 13:39:40 +00003977 if (block_group && block_group_bits(block_group, flags) &&
Josef Bacik285ff5a2012-01-13 15:27:45 -05003978 block_group->cached != BTRFS_CACHE_NO) {
Josef Bacik2552d172009-04-03 10:14:19 -04003979 down_read(&space_info->groups_sem);
Chris Mason44fb5512009-06-04 15:34:51 -04003980 if (list_empty(&block_group->list) ||
3981 block_group->ro) {
3982 /*
3983 * someone is removing this block group,
3984 * we can't jump into the have_block_group
3985 * target because our list pointers are not
3986 * valid
3987 */
3988 btrfs_put_block_group(block_group);
3989 up_read(&space_info->groups_sem);
Josef Bacikccf0e722009-11-10 21:23:48 -05003990 } else {
Qu Wenruob4bd7452018-11-02 09:39:47 +08003991 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
Qu Wenruo3e72ee82018-01-30 18:20:45 +08003992 block_group->flags);
Miao Xiee570fd22014-06-19 10:42:50 +08003993 btrfs_lock_block_group(block_group, delalloc);
Chris Mason44fb5512009-06-04 15:34:51 -04003994 goto have_block_group;
Josef Bacikccf0e722009-11-10 21:23:48 -05003995 }
Josef Bacik2552d172009-04-03 10:14:19 -04003996 } else if (block_group) {
Chris Masonfa9c0d792009-04-03 09:47:43 -04003997 btrfs_put_block_group(block_group);
Josef Bacik2552d172009-04-03 10:14:19 -04003998 }
Chris Mason42e70e72008-11-07 18:17:11 -05003999 }
Josef Bacik2552d172009-04-03 10:14:19 -04004000search:
Qu Wenruob4bd7452018-11-02 09:39:47 +08004001 ffe_ctl.have_caching_bg = false;
4002 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
4003 ffe_ctl.index == 0)
Josef Bacika5e681d2015-10-01 14:54:10 -04004004 full_search = true;
Josef Bacik80eb2342008-10-29 14:49:05 -04004005 down_read(&space_info->groups_sem);
Qu Wenruob4bd7452018-11-02 09:39:47 +08004006 list_for_each_entry(block_group,
4007 &space_info->block_groups[ffe_ctl.index], list) {
Naohiro Aotac6686902020-02-25 12:56:19 +09004008 struct btrfs_block_group *bg_ret;
4009
Jeff Mahoney14443932017-07-19 23:25:51 -04004010 /* If the block group is read-only, we can skip it entirely. */
4011 if (unlikely(block_group->ro))
4012 continue;
4013
Miao Xiee570fd22014-06-19 10:42:50 +08004014 btrfs_grab_block_group(block_group, delalloc);
David Sterbab3470b52019-10-23 18:48:22 +02004015 ffe_ctl.search_start = block_group->start;
Chris Mason42e70e72008-11-07 18:17:11 -05004016
Chris Mason83a50de2010-12-13 15:06:46 -05004017 /*
4018 * this can happen if we end up cycling through all the
4019 * raid types, but we want to make sure we only allocate
4020 * for the proper type.
4021 */
David Sterbab6919a52013-04-29 13:39:40 +00004022 if (!block_group_bits(block_group, flags)) {
Bart Van Asschebece2e82018-06-20 10:03:31 -07004023 u64 extra = BTRFS_BLOCK_GROUP_DUP |
David Sterbac7369b32019-05-31 15:39:31 +02004024 BTRFS_BLOCK_GROUP_RAID1_MASK |
David Sterbaa07e8a42019-05-31 16:54:26 +02004025 BTRFS_BLOCK_GROUP_RAID56_MASK |
Chris Mason83a50de2010-12-13 15:06:46 -05004026 BTRFS_BLOCK_GROUP_RAID10;
4027
4028 /*
4029 * if they asked for extra copies and this block group
4030 * doesn't provide them, bail. This does allow us to
4031 * fill raid0 from raid1.
4032 */
David Sterbab6919a52013-04-29 13:39:40 +00004033 if ((flags & extra) && !(block_group->flags & extra))
Chris Mason83a50de2010-12-13 15:06:46 -05004034 goto loop;
Qu Wenruo2a284682019-07-16 17:00:33 +08004035
4036 /*
4037 * This block group has different flags than we want.
4038 * It's possible that we have MIXED_GROUP flag but no
4039 * block group is mixed. Just skip such block group.
4040 */
4041 btrfs_release_block_group(block_group, delalloc);
4042 continue;
Chris Mason83a50de2010-12-13 15:06:46 -05004043 }
4044
Josef Bacik2552d172009-04-03 10:14:19 -04004045have_block_group:
David Sterba32da53862019-10-29 19:20:18 +01004046 ffe_ctl.cached = btrfs_block_group_done(block_group);
Qu Wenruob4bd7452018-11-02 09:39:47 +08004047 if (unlikely(!ffe_ctl.cached)) {
4048 ffe_ctl.have_caching_bg = true;
Josef Bacik676f1f72019-06-20 15:37:48 -04004049 ret = btrfs_cache_block_group(block_group, 0);
Josef Bacikdb8fe642019-11-19 13:59:00 -05004050
4051 /*
4052 * If we get ENOMEM here or something else we want to
4053 * try other block groups, because it may not be fatal.
4054 * However if we can't find anything else we need to
4055 * save our return here so that we return the actual
4056 * error that caused problems, not ENOSPC.
4057 */
4058 if (ret < 0) {
4059 if (!cache_block_group_error)
4060 cache_block_group_error = ret;
4061 ret = 0;
4062 goto loop;
4063 }
Chris Mason1d4284b2012-03-28 20:31:37 -04004064 ret = 0;
Josef Bacikea6a4782008-11-20 12:16:16 -05004065 }
4066
Josef Bacik36cce922013-08-05 11:15:21 -04004067 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
4068 goto loop;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004069
Naohiro Aotac6686902020-02-25 12:56:19 +09004070 bg_ret = NULL;
4071 ret = do_allocation(block_group, &ffe_ctl, &bg_ret);
4072 if (ret == 0) {
4073 if (bg_ret && bg_ret != block_group) {
4074 btrfs_release_block_group(block_group, delalloc);
4075 block_group = bg_ret;
Chris Masonfa9c0d792009-04-03 09:47:43 -04004076 }
Naohiro Aotac6686902020-02-25 12:56:19 +09004077 } else if (ret == -EAGAIN) {
4078 goto have_block_group;
4079 } else if (ret > 0) {
4080 goto loop;
Chris Masonfa9c0d792009-04-03 09:47:43 -04004081 }
4082
Naohiro Aotac6686902020-02-25 12:56:19 +09004083 /* Checks */
Qu Wenruob4bd7452018-11-02 09:39:47 +08004084 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
4085 fs_info->stripesize);
Chris Masone37c9e62007-05-09 20:13:14 -04004086
Josef Bacik2552d172009-04-03 10:14:19 -04004087 /* move on to the next group */
Qu Wenruob4bd7452018-11-02 09:39:47 +08004088 if (ffe_ctl.search_start + num_bytes >
David Sterbab3470b52019-10-23 18:48:22 +02004089 block_group->start + block_group->length) {
Qu Wenruob4bd7452018-11-02 09:39:47 +08004090 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4091 num_bytes);
Josef Bacik2552d172009-04-03 10:14:19 -04004092 goto loop;
Josef Bacik6226cb02009-04-03 10:14:18 -04004093 }
Josef Bacik80eb2342008-10-29 14:49:05 -04004094
Qu Wenruob4bd7452018-11-02 09:39:47 +08004095 if (ffe_ctl.found_offset < ffe_ctl.search_start)
4096 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4097 ffe_ctl.search_start - ffe_ctl.found_offset);
Josef Bacik6226cb02009-04-03 10:14:18 -04004098
Wang Xiaoguang18513092016-07-25 15:51:40 +08004099 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
4100 num_bytes, delalloc);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004101 if (ret == -EAGAIN) {
Qu Wenruob4bd7452018-11-02 09:39:47 +08004102 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
4103 num_bytes);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004104 goto loop;
4105 }
Filipe Manana9cfa3e32016-04-26 15:39:32 +01004106 btrfs_inc_block_group_reservations(block_group);
Yan Zheng11833d62009-09-11 16:11:19 -04004107
Josef Bacik2552d172009-04-03 10:14:19 -04004108 /* we are all good, lets return */
Qu Wenruob4bd7452018-11-02 09:39:47 +08004109 ins->objectid = ffe_ctl.search_start;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004110 ins->offset = num_bytes;
4111
Qu Wenruob4bd7452018-11-02 09:39:47 +08004112 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
4113 num_bytes);
Miao Xiee570fd22014-06-19 10:42:50 +08004114 btrfs_release_block_group(block_group, delalloc);
Josef Bacik2552d172009-04-03 10:14:19 -04004115 break;
4116loop:
Naohiro Aotababa5062020-02-25 12:56:21 +09004117 release_block_group(block_group, &ffe_ctl, delalloc);
Jeff Mahoney14443932017-07-19 23:25:51 -04004118 cond_resched();
Josef Bacik2552d172009-04-03 10:14:19 -04004119 }
4120 up_read(&space_info->groups_sem);
Chris Masonf5a31e12008-11-10 11:47:09 -05004121
Naohiro Aota15b7ee62020-02-25 12:56:23 +09004122 ret = find_free_extent_update_loop(fs_info, ins, &ffe_ctl, full_search);
Qu Wenruoe72d79d2018-11-02 09:39:50 +08004123 if (ret > 0)
Miao Xie60d2adb2011-09-09 17:34:35 +08004124 goto search;
4125
Josef Bacikdb8fe642019-11-19 13:59:00 -05004126 if (ret == -ENOSPC && !cache_block_group_error) {
Qu Wenruob4bd7452018-11-02 09:39:47 +08004127 /*
4128 * Use ffe_ctl->total_free_space as fallback if we can't find
4129 * any contiguous hole.
4130 */
4131 if (!ffe_ctl.max_extent_size)
4132 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
Josef Bacik4f4db212015-09-29 11:40:47 -04004133 spin_lock(&space_info->lock);
Qu Wenruob4bd7452018-11-02 09:39:47 +08004134 space_info->max_extent_size = ffe_ctl.max_extent_size;
Josef Bacik4f4db212015-09-29 11:40:47 -04004135 spin_unlock(&space_info->lock);
Qu Wenruob4bd7452018-11-02 09:39:47 +08004136 ins->offset = ffe_ctl.max_extent_size;
Josef Bacikdb8fe642019-11-19 13:59:00 -05004137 } else if (ret == -ENOSPC) {
4138 ret = cache_block_group_error;
Josef Bacik4f4db212015-09-29 11:40:47 -04004139 }
Chris Mason0f70abe2007-02-28 16:46:22 -05004140 return ret;
Chris Masonfec577f2007-02-26 10:40:21 -05004141}
Chris Masonec44a352008-04-28 15:29:52 -04004142
Nikolay Borisov6f47c702018-03-13 12:22:32 +02004143/*
4144 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
4145 * hole that is at least as big as @num_bytes.
4146 *
4147 * @root - The root that will contain this extent
4148 *
4149 * @ram_bytes - The amount of space in ram that @num_bytes take. This
4150 * is used for accounting purposes. This value differs
4151 * from @num_bytes only in the case of compressed extents.
4152 *
4153 * @num_bytes - Number of bytes to allocate on-disk.
4154 *
4155 * @min_alloc_size - Indicates the minimum amount of space that the
4156 * allocator should try to satisfy. In some cases
4157 * @num_bytes may be larger than what is required and if
4158 * the filesystem is fragmented then allocation fails.
4159 * However, the presence of @min_alloc_size gives a
4160 * chance to try and satisfy the smaller allocation.
4161 *
4162 * @empty_size - A hint that you plan on doing more COW. This is the
4163 * size in bytes the allocator should try to find free
4164 * next to the block it returns. This is just a hint and
4165 * may be ignored by the allocator.
4166 *
4167 * @hint_byte - Hint to the allocator to start searching above the byte
4168 * address passed. It might be ignored.
4169 *
4170 * @ins - This key is modified to record the found hole. It will
4171 * have the following values:
4172 * ins->objectid == start position
4173 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4174 * ins->offset == the size of the hole.
4175 *
4176 * @is_data - Boolean flag indicating whether an extent is
4177 * allocated for data (true) or metadata (false)
4178 *
4179 * @delalloc - Boolean flag indicating whether this allocation is for
4180 * delalloc or not. If 'true' data_rwsem of block groups
4181 * is going to be acquired.
4182 *
4183 *
4184 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
4185 * case -ENOSPC is returned then @ins->offset will contain the size of the
4186 * largest available hole the allocator managed to find.
4187 */
Wang Xiaoguang18513092016-07-25 15:51:40 +08004188int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
Yan Zheng11833d62009-09-11 16:11:19 -04004189 u64 num_bytes, u64 min_alloc_size,
4190 u64 empty_size, u64 hint_byte,
Miao Xiee570fd22014-06-19 10:42:50 +08004191 struct btrfs_key *ins, int is_data, int delalloc)
Chris Masonfec577f2007-02-26 10:40:21 -05004192{
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04004193 struct btrfs_fs_info *fs_info = root->fs_info;
Josef Bacik36af4e02015-09-25 16:13:11 -04004194 bool final_tried = num_bytes == min_alloc_size;
David Sterbab6919a52013-04-29 13:39:40 +00004195 u64 flags;
Chris Masonfec577f2007-02-26 10:40:21 -05004196 int ret;
Chris Mason925baed2008-06-25 16:01:30 -04004197
Jeff Mahoney1b868262017-05-17 11:38:35 -04004198 flags = get_alloc_profile_by_root(root, is_data);
Chris Mason98d20f62008-04-14 09:46:10 -04004199again:
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004200 WARN_ON(num_bytes < fs_info->sectorsize);
Jeff Mahoney87bde3c2017-02-15 16:28:27 -05004201 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
Wang Xiaoguang18513092016-07-25 15:51:40 +08004202 hint_byte, ins, flags, delalloc);
Filipe Manana9cfa3e32016-04-26 15:39:32 +01004203 if (!ret && !is_data) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04004204 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
Filipe Manana9cfa3e32016-04-26 15:39:32 +01004205 } else if (ret == -ENOSPC) {
Miao Xiea4820392013-09-09 13:19:42 +08004206 if (!final_tried && ins->offset) {
4207 num_bytes = min(num_bytes >> 1, ins->offset);
Jeff Mahoneyda170662016-06-15 09:22:56 -04004208 num_bytes = round_down(num_bytes,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004209 fs_info->sectorsize);
Miao Xie9e622d62012-01-26 15:01:12 -05004210 num_bytes = max(num_bytes, min_alloc_size);
Wang Xiaoguang18513092016-07-25 15:51:40 +08004211 ram_bytes = num_bytes;
Miao Xie9e622d62012-01-26 15:01:12 -05004212 if (num_bytes == min_alloc_size)
4213 final_tried = true;
4214 goto again;
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04004215 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
Miao Xie9e622d62012-01-26 15:01:12 -05004216 struct btrfs_space_info *sinfo;
Josef Bacik0f9dd462008-09-23 13:14:11 -04004217
Josef Bacik280c29082019-06-18 16:09:19 -04004218 sinfo = btrfs_find_space_info(fs_info, flags);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004219 btrfs_err(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04004220 "allocation failed flags %llu, wanted %llu",
4221 flags, num_bytes);
Jeff Mahoney53804282012-03-01 14:56:28 +01004222 if (sinfo)
Josef Bacik5da6afe2019-06-18 16:09:24 -04004223 btrfs_dump_space_info(fs_info, sinfo,
4224 num_bytes, 1);
Miao Xie9e622d62012-01-26 15:01:12 -05004225 }
Chris Mason925baed2008-06-25 16:01:30 -04004226 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04004227
4228 return ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004229}
4230
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004231int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
Miao Xiee570fd22014-06-19 10:42:50 +08004232 u64 start, u64 len, int delalloc)
Chris Masone688b7252011-10-31 20:52:39 -04004233{
Nikolay Borisov7ef54d52019-11-21 14:03:30 +02004234 struct btrfs_block_group *cache;
4235
4236 cache = btrfs_lookup_block_group(fs_info, start);
4237 if (!cache) {
Nikolay Borisova0fbf732019-11-21 14:03:31 +02004238 btrfs_err(fs_info, "Unable to find block group for %llu",
4239 start);
Nikolay Borisov7ef54d52019-11-21 14:03:30 +02004240 return -ENOSPC;
4241 }
4242
4243 btrfs_add_free_space(cache, start, len);
4244 btrfs_free_reserved_bytes(cache, len, delalloc);
4245 trace_btrfs_reserved_extent_free(fs_info, start, len);
4246
4247 btrfs_put_block_group(cache);
4248 return 0;
Chris Masone688b7252011-10-31 20:52:39 -04004249}
4250
Nikolay Borisov7bfc1002020-01-20 16:09:12 +02004251int btrfs_pin_reserved_extent(struct btrfs_trans_handle *trans, u64 start,
4252 u64 len)
Chris Masone688b7252011-10-31 20:52:39 -04004253{
Nikolay Borisova0fbf732019-11-21 14:03:31 +02004254 struct btrfs_block_group *cache;
4255 int ret = 0;
4256
Nikolay Borisov7bfc1002020-01-20 16:09:12 +02004257 cache = btrfs_lookup_block_group(trans->fs_info, start);
Nikolay Borisova0fbf732019-11-21 14:03:31 +02004258 if (!cache) {
Nikolay Borisov7bfc1002020-01-20 16:09:12 +02004259 btrfs_err(trans->fs_info, "unable to find block group for %llu",
4260 start);
Nikolay Borisova0fbf732019-11-21 14:03:31 +02004261 return -ENOSPC;
4262 }
4263
Nikolay Borisov6690d072020-01-20 16:09:14 +02004264 ret = pin_down_extent(trans, cache, start, len, 1);
Nikolay Borisova0fbf732019-11-21 14:03:31 +02004265 btrfs_put_block_group(cache);
4266 return ret;
Chris Masone688b7252011-10-31 20:52:39 -04004267}
4268
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004269static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004270 u64 parent, u64 root_objectid,
4271 u64 flags, u64 owner, u64 offset,
4272 struct btrfs_key *ins, int ref_mod)
Chris Masone6dcd2d2008-07-17 12:53:50 -04004273{
Nikolay Borisovef89b822018-06-20 15:48:58 +03004274 struct btrfs_fs_info *fs_info = trans->fs_info;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004275 int ret;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004276 struct btrfs_extent_item *extent_item;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004277 struct btrfs_extent_inline_ref *iref;
Chris Masone6dcd2d2008-07-17 12:53:50 -04004278 struct btrfs_path *path;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004279 struct extent_buffer *leaf;
4280 int type;
4281 u32 size;
Chris Masonf2654de2007-06-26 12:20:46 -04004282
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004283 if (parent > 0)
4284 type = BTRFS_SHARED_DATA_REF_KEY;
4285 else
4286 type = BTRFS_EXTENT_DATA_REF_KEY;
Zheng Yan31840ae2008-09-23 13:14:14 -04004287
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004288 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
Chris Mason7bb86312007-12-11 09:25:06 -05004289
4290 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00004291 if (!path)
4292 return -ENOMEM;
Chris Mason47e4bb92008-02-01 14:51:59 -05004293
Chris Masonb9473432009-03-13 11:00:37 -04004294 path->leave_spinning = 1;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004295 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
4296 ins, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004297 if (ret) {
4298 btrfs_free_path(path);
4299 return ret;
4300 }
Josef Bacik0f9dd462008-09-23 13:14:11 -04004301
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004302 leaf = path->nodes[0];
4303 extent_item = btrfs_item_ptr(leaf, path->slots[0],
Chris Mason47e4bb92008-02-01 14:51:59 -05004304 struct btrfs_extent_item);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004305 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
4306 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4307 btrfs_set_extent_flags(leaf, extent_item,
4308 flags | BTRFS_EXTENT_FLAG_DATA);
Chris Mason47e4bb92008-02-01 14:51:59 -05004309
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004310 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4311 btrfs_set_extent_inline_ref_type(leaf, iref, type);
4312 if (parent > 0) {
4313 struct btrfs_shared_data_ref *ref;
4314 ref = (struct btrfs_shared_data_ref *)(iref + 1);
4315 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
4316 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
4317 } else {
4318 struct btrfs_extent_data_ref *ref;
4319 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
4320 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
4321 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
4322 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
4323 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
4324 }
Chris Mason47e4bb92008-02-01 14:51:59 -05004325
4326 btrfs_mark_buffer_dirty(path->nodes[0]);
Chris Mason7bb86312007-12-11 09:25:06 -05004327 btrfs_free_path(path);
Chris Masonf510cfe2007-10-15 16:14:48 -04004328
Nikolay Borisov25a356d2018-05-10 15:44:54 +03004329 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
Omar Sandoval1e144fb2015-09-29 20:50:37 -07004330 if (ret)
4331 return ret;
4332
Josef Bacikade4b512019-06-20 15:38:01 -04004333 ret = btrfs_update_block_group(trans, ins->objectid, ins->offset, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004334 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004335 btrfs_err(fs_info, "update block group failed for %llu %llu",
Geert Uytterhoevenc1c9ff72013-08-20 13:20:07 +02004336 ins->objectid, ins->offset);
Chris Masonf5947062008-02-04 10:10:13 -05004337 BUG();
4338 }
Jeff Mahoney71ff6432016-09-06 16:00:42 -04004339 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004340 return ret;
4341}
4342
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004343static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004344 struct btrfs_delayed_ref_node *node,
Nikolay Borisov21ebfbe2018-05-21 12:27:22 +03004345 struct btrfs_delayed_extent_op *extent_op)
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004346{
Nikolay Borisov9dcdbe02018-05-21 12:27:20 +03004347 struct btrfs_fs_info *fs_info = trans->fs_info;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004348 int ret;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004349 struct btrfs_extent_item *extent_item;
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004350 struct btrfs_key extent_key;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004351 struct btrfs_tree_block_info *block_info;
4352 struct btrfs_extent_inline_ref *iref;
4353 struct btrfs_path *path;
4354 struct extent_buffer *leaf;
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004355 struct btrfs_delayed_tree_ref *ref;
Josef Bacik3173a182013-03-07 14:22:04 -05004356 u32 size = sizeof(*extent_item) + sizeof(*iref);
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004357 u64 num_bytes;
Nikolay Borisov21ebfbe2018-05-21 12:27:22 +03004358 u64 flags = extent_op->flags_to_set;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004359 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
Josef Bacik3173a182013-03-07 14:22:04 -05004360
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004361 ref = btrfs_delayed_node_to_tree_ref(node);
4362
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004363 extent_key.objectid = node->bytenr;
4364 if (skinny_metadata) {
4365 extent_key.offset = ref->level;
4366 extent_key.type = BTRFS_METADATA_ITEM_KEY;
4367 num_bytes = fs_info->nodesize;
4368 } else {
4369 extent_key.offset = node->num_bytes;
4370 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
Josef Bacik3173a182013-03-07 14:22:04 -05004371 size += sizeof(*block_info);
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004372 num_bytes = node->num_bytes;
4373 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004374
4375 path = btrfs_alloc_path();
Josef Bacik80ee54b2018-10-11 15:54:22 -04004376 if (!path)
Mark Fashehd8926bb2011-07-13 10:38:47 -07004377 return -ENOMEM;
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004378
4379 path->leave_spinning = 1;
4380 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004381 &extent_key, size);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004382 if (ret) {
Chris Masondd825252015-04-01 08:36:05 -07004383 btrfs_free_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004384 return ret;
4385 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004386
4387 leaf = path->nodes[0];
4388 extent_item = btrfs_item_ptr(leaf, path->slots[0],
4389 struct btrfs_extent_item);
4390 btrfs_set_extent_refs(leaf, extent_item, 1);
4391 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
4392 btrfs_set_extent_flags(leaf, extent_item,
4393 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004394
Josef Bacik3173a182013-03-07 14:22:04 -05004395 if (skinny_metadata) {
4396 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
4397 } else {
4398 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
Nikolay Borisov21ebfbe2018-05-21 12:27:22 +03004399 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004400 btrfs_set_tree_block_level(leaf, block_info, ref->level);
Josef Bacik3173a182013-03-07 14:22:04 -05004401 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
4402 }
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004403
Nikolay Borisovd4b20732018-05-21 12:27:23 +03004404 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004405 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
4406 btrfs_set_extent_inline_ref_type(leaf, iref,
4407 BTRFS_SHARED_BLOCK_REF_KEY);
Nikolay Borisovd4b20732018-05-21 12:27:23 +03004408 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004409 } else {
4410 btrfs_set_extent_inline_ref_type(leaf, iref,
4411 BTRFS_TREE_BLOCK_REF_KEY);
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004412 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004413 }
4414
4415 btrfs_mark_buffer_dirty(leaf);
4416 btrfs_free_path(path);
4417
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004418 ret = remove_from_free_space_tree(trans, extent_key.objectid,
4419 num_bytes);
Omar Sandoval1e144fb2015-09-29 20:50:37 -07004420 if (ret)
4421 return ret;
4422
Josef Bacikade4b512019-06-20 15:38:01 -04004423 ret = btrfs_update_block_group(trans, extent_key.objectid,
4424 fs_info->nodesize, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004425 if (ret) { /* -ENOENT, logic error */
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004426 btrfs_err(fs_info, "update block group failed for %llu %llu",
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004427 extent_key.objectid, extent_key.offset);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004428 BUG();
4429 }
Josef Bacik0be5dc62013-10-07 15:18:52 -04004430
Nikolay Borisov4e6bd4e2018-05-21 12:27:21 +03004431 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004432 fs_info->nodesize);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004433 return ret;
4434}
4435
4436int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
Josef Bacik84f7d8e2017-09-29 15:43:49 -04004437 struct btrfs_root *root, u64 owner,
Qu Wenruo5846a3c2015-10-26 14:11:18 +08004438 u64 offset, u64 ram_bytes,
4439 struct btrfs_key *ins)
Chris Masone6dcd2d2008-07-17 12:53:50 -04004440{
Qu Wenruo76675592019-04-04 14:45:32 +08004441 struct btrfs_ref generic_ref = { 0 };
Chris Masone6dcd2d2008-07-17 12:53:50 -04004442 int ret;
Chris Mason1c2308f82008-09-23 13:14:13 -04004443
Josef Bacik84f7d8e2017-09-29 15:43:49 -04004444 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
Chris Mason56bec292009-03-13 10:10:06 -04004445
Qu Wenruo76675592019-04-04 14:45:32 +08004446 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4447 ins->objectid, ins->offset, 0);
4448 btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset);
Qu Wenruo8a5040f2019-04-04 14:45:33 +08004449 btrfs_ref_tree_mod(root->fs_info, &generic_ref);
Qu Wenruo76675592019-04-04 14:45:32 +08004450 ret = btrfs_add_delayed_data_ref(trans, &generic_ref,
4451 ram_bytes, NULL, NULL);
Chris Masone6dcd2d2008-07-17 12:53:50 -04004452 return ret;
4453}
Chris Masone02119d2008-09-05 16:13:11 -04004454
4455/*
4456 * this is used by the tree logging recovery code. It records that
4457 * an extent has been allocated and makes sure to clear the free
4458 * space cache bits as well
4459 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004460int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
Yan Zheng5d4f98a2009-06-10 10:45:14 -04004461 u64 root_objectid, u64 owner, u64 offset,
4462 struct btrfs_key *ins)
Chris Masone02119d2008-09-05 16:13:11 -04004463{
Nikolay Borisov61da2ab2018-06-20 15:49:13 +03004464 struct btrfs_fs_info *fs_info = trans->fs_info;
Chris Masone02119d2008-09-05 16:13:11 -04004465 int ret;
David Sterba32da53862019-10-29 19:20:18 +01004466 struct btrfs_block_group *block_group;
Wang Xiaoguanged7a6942016-08-26 11:33:14 +08004467 struct btrfs_space_info *space_info;
Josef Bacik8c2a1a32013-06-06 13:19:32 -04004468
4469 /*
4470 * Mixed block groups will exclude before processing the log so we only
Nicholas D Steeves01327612016-05-19 21:18:45 -04004471 * need to do the exclude dance if this fs isn't mixed.
Josef Bacik8c2a1a32013-06-06 13:19:32 -04004472 */
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004473 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004474 ret = __exclude_logged_extent(fs_info, ins->objectid,
4475 ins->offset);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04004476 if (ret)
4477 return ret;
4478 }
Chris Masone02119d2008-09-05 16:13:11 -04004479
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004480 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
Josef Bacik8c2a1a32013-06-06 13:19:32 -04004481 if (!block_group)
4482 return -EINVAL;
Yan Zheng11833d62009-09-11 16:11:19 -04004483
Wang Xiaoguanged7a6942016-08-26 11:33:14 +08004484 space_info = block_group->space_info;
4485 spin_lock(&space_info->lock);
4486 spin_lock(&block_group->lock);
4487 space_info->bytes_reserved += ins->offset;
4488 block_group->reserved += ins->offset;
4489 spin_unlock(&block_group->lock);
4490 spin_unlock(&space_info->lock);
4491
Nikolay Borisovef89b822018-06-20 15:48:58 +03004492 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
4493 offset, ins, 1);
Josef Bacikbd727172020-02-13 10:47:30 -05004494 if (ret)
Josef Bacikab9b2c72020-02-13 10:47:30 -05004495 btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
Josef Bacikb50c6e22013-04-25 15:55:30 -04004496 btrfs_put_block_group(block_group);
Chris Masone02119d2008-09-05 16:13:11 -04004497 return ret;
4498}
4499
Eric Sandeen48a3b632013-04-25 20:41:01 +00004500static struct extent_buffer *
4501btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
Nikolay Borisovbc877d22018-06-18 14:13:19 +03004502 u64 bytenr, int level, u64 owner)
Chris Mason65b51a02008-08-01 15:11:20 -04004503{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004504 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason65b51a02008-08-01 15:11:20 -04004505 struct extent_buffer *buf;
4506
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004507 buf = btrfs_find_create_tree_block(fs_info, bytenr);
Liu Boc871b0f2016-06-06 12:01:23 -07004508 if (IS_ERR(buf))
4509 return buf;
4510
Qu Wenruob72c3ab2018-08-21 09:53:47 +08004511 /*
4512 * Extra safety check in case the extent tree is corrupted and extent
4513 * allocator chooses to use a tree block which is already used and
4514 * locked.
4515 */
4516 if (buf->lock_owner == current->pid) {
4517 btrfs_err_rl(fs_info,
4518"tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
4519 buf->start, btrfs_header_owner(buf), current->pid);
4520 free_extent_buffer(buf);
4521 return ERR_PTR(-EUCLEAN);
4522 }
4523
Chris Mason85d4e462011-07-26 16:11:19 -04004524 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
Chris Mason65b51a02008-08-01 15:11:20 -04004525 btrfs_tree_lock(buf);
David Sterba6a884d7d2019-03-20 14:30:02 +01004526 btrfs_clean_tree_block(buf);
Josef Bacik3083ee22012-03-09 16:01:49 -05004527 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
Chris Masonb4ce94d2009-02-04 09:25:08 -05004528
David Sterba8bead252018-04-04 02:03:48 +02004529 btrfs_set_lock_blocking_write(buf);
David Sterba4db8c522015-12-03 13:06:46 +01004530 set_extent_buffer_uptodate(buf);
Chris Masonb4ce94d2009-02-04 09:25:08 -05004531
Nikolay Borisovbc877d22018-06-18 14:13:19 +03004532 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
4533 btrfs_set_header_level(buf, level);
4534 btrfs_set_header_bytenr(buf, buf->start);
4535 btrfs_set_header_generation(buf, trans->transid);
4536 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
4537 btrfs_set_header_owner(buf, owner);
Nikolay Borisovde37aa52018-10-30 16:43:24 +02004538 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
Nikolay Borisovbc877d22018-06-18 14:13:19 +03004539 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
Chris Masond0c803c2008-09-11 16:17:57 -04004540 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
Filipe Manana656f30d2014-09-26 12:25:56 +01004541 buf->log_index = root->log_transid % 2;
Yan, Zheng8cef4e12009-11-12 09:33:26 +00004542 /*
4543 * we allow two log transactions at a time, use different
Andrea Gelmini52042d82018-11-28 12:05:13 +01004544 * EXTENT bit to differentiate dirty pages.
Yan, Zheng8cef4e12009-11-12 09:33:26 +00004545 */
Filipe Manana656f30d2014-09-26 12:25:56 +01004546 if (buf->log_index == 0)
Yan, Zheng8cef4e12009-11-12 09:33:26 +00004547 set_extent_dirty(&root->dirty_log_pages, buf->start,
4548 buf->start + buf->len - 1, GFP_NOFS);
4549 else
4550 set_extent_new(&root->dirty_log_pages, buf->start,
David Sterba3744dbe2016-04-26 23:54:39 +02004551 buf->start + buf->len - 1);
Chris Masond0c803c2008-09-11 16:17:57 -04004552 } else {
Filipe Manana656f30d2014-09-26 12:25:56 +01004553 buf->log_index = -1;
Chris Masond0c803c2008-09-11 16:17:57 -04004554 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
4555 buf->start + buf->len - 1, GFP_NOFS);
4556 }
Jeff Mahoney64c12922016-06-08 00:36:38 -04004557 trans->dirty = true;
Chris Masonb4ce94d2009-02-04 09:25:08 -05004558 /* this returns a buffer locked for blocking */
Chris Mason65b51a02008-08-01 15:11:20 -04004559 return buf;
4560}
4561
Chris Masonfec577f2007-02-26 10:40:21 -05004562/*
Yan, Zhengf0486c62010-05-16 10:46:25 -04004563 * finds a free extent and does all the dirty work required for allocation
Omar Sandoval67b78592015-02-24 02:47:04 -08004564 * returns the tree buffer or an ERR_PTR on error.
Chris Masonfec577f2007-02-26 10:40:21 -05004565 */
David Sterba4d75f8a2014-06-15 01:54:12 +02004566struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
Omar Sandoval310712b2017-01-17 23:24:37 -08004567 struct btrfs_root *root,
4568 u64 parent, u64 root_objectid,
4569 const struct btrfs_disk_key *key,
4570 int level, u64 hint,
4571 u64 empty_size)
Chris Masonfec577f2007-02-26 10:40:21 -05004572{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004573 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Masone2fa7222007-03-12 16:22:34 -04004574 struct btrfs_key ins;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004575 struct btrfs_block_rsv *block_rsv;
Chris Mason5f39d392007-10-15 16:14:19 -04004576 struct extent_buffer *buf;
Omar Sandoval67b78592015-02-24 02:47:04 -08004577 struct btrfs_delayed_extent_op *extent_op;
Qu Wenruoed4f2552019-04-04 14:45:31 +08004578 struct btrfs_ref generic_ref = { 0 };
Yan, Zhengf0486c62010-05-16 10:46:25 -04004579 u64 flags = 0;
4580 int ret;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004581 u32 blocksize = fs_info->nodesize;
4582 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004583
David Sterba05653ef2016-07-15 15:23:37 +02004584#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004585 if (btrfs_is_testing(fs_info)) {
Josef Bacikfaa2dbf2014-05-07 17:06:09 -04004586 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
Nikolay Borisovbc877d22018-06-18 14:13:19 +03004587 level, root_objectid);
Josef Bacikfaa2dbf2014-05-07 17:06:09 -04004588 if (!IS_ERR(buf))
4589 root->alloc_bytenr += blocksize;
4590 return buf;
4591 }
David Sterba05653ef2016-07-15 15:23:37 +02004592#endif
David Sterbafccb84c2014-09-29 23:53:21 +02004593
Josef Bacik67f9c222019-06-19 13:47:23 -04004594 block_rsv = btrfs_use_block_rsv(trans, root, blocksize);
Yan, Zhengf0486c62010-05-16 10:46:25 -04004595 if (IS_ERR(block_rsv))
4596 return ERR_CAST(block_rsv);
4597
Wang Xiaoguang18513092016-07-25 15:51:40 +08004598 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
Miao Xiee570fd22014-06-19 10:42:50 +08004599 empty_size, hint, &ins, 0, 0);
Omar Sandoval67b78592015-02-24 02:47:04 -08004600 if (ret)
4601 goto out_unuse;
Chris Mason55c69072008-01-09 15:55:33 -05004602
Nikolay Borisovbc877d22018-06-18 14:13:19 +03004603 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
4604 root_objectid);
Omar Sandoval67b78592015-02-24 02:47:04 -08004605 if (IS_ERR(buf)) {
4606 ret = PTR_ERR(buf);
4607 goto out_free_reserved;
4608 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004609
4610 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
4611 if (parent == 0)
4612 parent = ins.objectid;
4613 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
4614 } else
4615 BUG_ON(parent > 0);
4616
4617 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
Miao Xie78a61842012-11-21 02:21:28 +00004618 extent_op = btrfs_alloc_delayed_extent_op();
Omar Sandoval67b78592015-02-24 02:47:04 -08004619 if (!extent_op) {
4620 ret = -ENOMEM;
4621 goto out_free_buf;
4622 }
Yan, Zhengf0486c62010-05-16 10:46:25 -04004623 if (key)
4624 memcpy(&extent_op->key, key, sizeof(extent_op->key));
4625 else
4626 memset(&extent_op->key, 0, sizeof(extent_op->key));
4627 extent_op->flags_to_set = flags;
David Sterba35b3ad52015-11-30 16:51:29 +01004628 extent_op->update_key = skinny_metadata ? false : true;
4629 extent_op->update_flags = true;
4630 extent_op->is_data = false;
Josef Bacikb1c79e02013-05-09 13:49:30 -04004631 extent_op->level = level;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004632
Qu Wenruoed4f2552019-04-04 14:45:31 +08004633 btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
4634 ins.objectid, ins.offset, parent);
4635 generic_ref.real_root = root->root_key.objectid;
4636 btrfs_init_tree_ref(&generic_ref, level, root_objectid);
Qu Wenruo8a5040f2019-04-04 14:45:33 +08004637 btrfs_ref_tree_mod(fs_info, &generic_ref);
Qu Wenruoed4f2552019-04-04 14:45:31 +08004638 ret = btrfs_add_delayed_tree_ref(trans, &generic_ref,
Omar Sandoval7be07912017-06-06 16:45:30 -07004639 extent_op, NULL, NULL);
Omar Sandoval67b78592015-02-24 02:47:04 -08004640 if (ret)
4641 goto out_free_delayed;
Yan, Zhengf0486c62010-05-16 10:46:25 -04004642 }
Chris Masonfec577f2007-02-26 10:40:21 -05004643 return buf;
Omar Sandoval67b78592015-02-24 02:47:04 -08004644
4645out_free_delayed:
4646 btrfs_free_delayed_extent_op(extent_op);
4647out_free_buf:
4648 free_extent_buffer(buf);
4649out_free_reserved:
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004650 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
Omar Sandoval67b78592015-02-24 02:47:04 -08004651out_unuse:
Josef Bacik67f9c222019-06-19 13:47:23 -04004652 btrfs_unuse_block_rsv(fs_info, block_rsv, blocksize);
Omar Sandoval67b78592015-02-24 02:47:04 -08004653 return ERR_PTR(ret);
Chris Masonfec577f2007-02-26 10:40:21 -05004654}
Chris Masona28ec192007-03-06 20:08:01 -05004655
Yan Zheng2c47e6052009-06-27 21:07:35 -04004656struct walk_control {
4657 u64 refs[BTRFS_MAX_LEVEL];
4658 u64 flags[BTRFS_MAX_LEVEL];
4659 struct btrfs_key update_progress;
Josef Bacikaea6f022019-02-06 15:46:15 -05004660 struct btrfs_key drop_progress;
4661 int drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04004662 int stage;
4663 int level;
4664 int shared_level;
4665 int update_ref;
4666 int keep_locks;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004667 int reada_slot;
4668 int reada_count;
Josef Bacik78c52d92019-02-06 15:46:14 -05004669 int restarted;
Yan Zheng2c47e6052009-06-27 21:07:35 -04004670};
4671
4672#define DROP_REFERENCE 1
4673#define UPDATE_BACKREF 2
4674
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004675static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
4676 struct btrfs_root *root,
4677 struct walk_control *wc,
4678 struct btrfs_path *path)
4679{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004680 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004681 u64 bytenr;
4682 u64 generation;
4683 u64 refs;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004684 u64 flags;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004685 u32 nritems;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004686 struct btrfs_key key;
4687 struct extent_buffer *eb;
4688 int ret;
4689 int slot;
4690 int nread = 0;
4691
4692 if (path->slots[wc->level] < wc->reada_slot) {
4693 wc->reada_count = wc->reada_count * 2 / 3;
4694 wc->reada_count = max(wc->reada_count, 2);
4695 } else {
4696 wc->reada_count = wc->reada_count * 3 / 2;
4697 wc->reada_count = min_t(int, wc->reada_count,
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004698 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004699 }
4700
4701 eb = path->nodes[wc->level];
4702 nritems = btrfs_header_nritems(eb);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004703
4704 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
4705 if (nread >= wc->reada_count)
4706 break;
4707
4708 cond_resched();
4709 bytenr = btrfs_node_blockptr(eb, slot);
4710 generation = btrfs_node_ptr_generation(eb, slot);
4711
4712 if (slot == path->slots[wc->level])
4713 goto reada;
4714
4715 if (wc->stage == UPDATE_BACKREF &&
4716 generation <= root->root_key.offset)
4717 continue;
4718
Yan, Zheng94fcca92009-10-09 09:25:16 -04004719 /* We don't lock the tree block, it's OK to be racy here */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004720 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
Josef Bacik3173a182013-03-07 14:22:04 -05004721 wc->level - 1, 1, &refs,
4722 &flags);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004723 /* We don't care about errors in readahead. */
4724 if (ret < 0)
4725 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004726 BUG_ON(refs == 0);
4727
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004728 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004729 if (refs == 1)
4730 goto reada;
4731
Yan, Zheng94fcca92009-10-09 09:25:16 -04004732 if (wc->level == 1 &&
4733 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4734 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004735 if (!wc->update_ref ||
4736 generation <= root->root_key.offset)
4737 continue;
4738 btrfs_node_key_to_cpu(eb, &key, slot);
4739 ret = btrfs_comp_cpu_keys(&key,
4740 &wc->update_progress);
4741 if (ret < 0)
4742 continue;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004743 } else {
4744 if (wc->level == 1 &&
4745 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4746 continue;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004747 }
4748reada:
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004749 readahead_tree_block(fs_info, bytenr);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004750 nread++;
4751 }
4752 wc->reada_slot = slot;
4753}
4754
Chris Mason9aca1d52007-03-13 11:09:37 -04004755/*
Liu Bo2c016dc2012-12-26 15:32:17 +08004756 * helper to process tree block while walking down the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04004757 *
Yan Zheng2c47e6052009-06-27 21:07:35 -04004758 * when wc->stage == UPDATE_BACKREF, this function updates
4759 * back refs for pointers in the block.
4760 *
4761 * NOTE: return value 1 means we should stop walking down.
Yan Zhengf82d02d2008-10-29 14:49:05 -04004762 */
Yan Zheng2c47e6052009-06-27 21:07:35 -04004763static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
4764 struct btrfs_root *root,
4765 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04004766 struct walk_control *wc, int lookup_info)
Yan Zheng2c47e6052009-06-27 21:07:35 -04004767{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004768 struct btrfs_fs_info *fs_info = root->fs_info;
Yan Zheng2c47e6052009-06-27 21:07:35 -04004769 int level = wc->level;
4770 struct extent_buffer *eb = path->nodes[level];
Yan Zheng2c47e6052009-06-27 21:07:35 -04004771 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
4772 int ret;
4773
4774 if (wc->stage == UPDATE_BACKREF &&
4775 btrfs_header_owner(eb) != root->root_key.objectid)
4776 return 1;
4777
4778 /*
4779 * when reference count of tree block is 1, it won't increase
4780 * again. once full backref flag is set, we never clear it.
4781 */
Yan, Zheng94fcca92009-10-09 09:25:16 -04004782 if (lookup_info &&
4783 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
4784 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04004785 BUG_ON(!path->locks[level]);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004786 ret = btrfs_lookup_extent_info(trans, fs_info,
Josef Bacik3173a182013-03-07 14:22:04 -05004787 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04004788 &wc->refs[level],
4789 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004790 BUG_ON(ret == -ENOMEM);
4791 if (ret)
4792 return ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04004793 BUG_ON(wc->refs[level] == 0);
4794 }
4795
Yan Zheng2c47e6052009-06-27 21:07:35 -04004796 if (wc->stage == DROP_REFERENCE) {
4797 if (wc->refs[level] > 1)
4798 return 1;
4799
4800 if (path->locks[level] && !wc->keep_locks) {
Chris Masonbd681512011-07-16 15:23:14 -04004801 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04004802 path->locks[level] = 0;
4803 }
4804 return 0;
4805 }
4806
4807 /* wc->stage == UPDATE_BACKREF */
4808 if (!(wc->flags[level] & flag)) {
4809 BUG_ON(!path->locks[level]);
Josef Bacike339a6b2014-07-02 10:54:25 -07004810 ret = btrfs_inc_ref(trans, root, eb, 1);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004811 BUG_ON(ret); /* -ENOMEM */
Josef Bacike339a6b2014-07-02 10:54:25 -07004812 ret = btrfs_dec_ref(trans, root, eb, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004813 BUG_ON(ret); /* -ENOMEM */
David Sterba42c9d0b2019-03-20 11:54:13 +01004814 ret = btrfs_set_disk_extent_flags(trans, eb, flag,
Josef Bacikb1c79e02013-05-09 13:49:30 -04004815 btrfs_header_level(eb), 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004816 BUG_ON(ret); /* -ENOMEM */
Yan Zheng2c47e6052009-06-27 21:07:35 -04004817 wc->flags[level] |= flag;
4818 }
4819
4820 /*
4821 * the block is shared by multiple trees, so it's not good to
4822 * keep the tree lock
4823 */
4824 if (path->locks[level] && level > 0) {
Chris Masonbd681512011-07-16 15:23:14 -04004825 btrfs_tree_unlock_rw(eb, path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04004826 path->locks[level] = 0;
4827 }
4828 return 0;
4829}
4830
4831/*
Josef Bacik78c52d92019-02-06 15:46:14 -05004832 * This is used to verify a ref exists for this root to deal with a bug where we
4833 * would have a drop_progress key that hadn't been updated properly.
4834 */
4835static int check_ref_exists(struct btrfs_trans_handle *trans,
4836 struct btrfs_root *root, u64 bytenr, u64 parent,
4837 int level)
4838{
4839 struct btrfs_path *path;
4840 struct btrfs_extent_inline_ref *iref;
4841 int ret;
4842
4843 path = btrfs_alloc_path();
4844 if (!path)
4845 return -ENOMEM;
4846
4847 ret = lookup_extent_backref(trans, path, &iref, bytenr,
4848 root->fs_info->nodesize, parent,
4849 root->root_key.objectid, level, 0);
4850 btrfs_free_path(path);
4851 if (ret == -ENOENT)
4852 return 0;
4853 if (ret < 0)
4854 return ret;
4855 return 1;
4856}
4857
4858/*
Liu Bo2c016dc2012-12-26 15:32:17 +08004859 * helper to process tree block pointer.
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004860 *
4861 * when wc->stage == DROP_REFERENCE, this function checks
4862 * reference count of the block pointed to. if the block
4863 * is shared and we need update back refs for the subtree
4864 * rooted at the block, this function changes wc->stage to
4865 * UPDATE_BACKREF. if the block is shared and there is no
4866 * need to update back, this function drops the reference
4867 * to the block.
4868 *
4869 * NOTE: return value 1 means we should stop walking down.
4870 */
4871static noinline int do_walk_down(struct btrfs_trans_handle *trans,
4872 struct btrfs_root *root,
4873 struct btrfs_path *path,
Yan, Zheng94fcca92009-10-09 09:25:16 -04004874 struct walk_control *wc, int *lookup_info)
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004875{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004876 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004877 u64 bytenr;
4878 u64 generation;
4879 u64 parent;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004880 struct btrfs_key key;
Qu Wenruo581c1762018-03-29 09:08:11 +08004881 struct btrfs_key first_key;
Qu Wenruoffd4bb22019-04-04 14:45:36 +08004882 struct btrfs_ref ref = { 0 };
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004883 struct extent_buffer *next;
4884 int level = wc->level;
4885 int reada = 0;
4886 int ret = 0;
Mark Fasheh11526512014-07-17 12:39:01 -07004887 bool need_account = false;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004888
4889 generation = btrfs_node_ptr_generation(path->nodes[level],
4890 path->slots[level]);
4891 /*
4892 * if the lower level block was created before the snapshot
4893 * was created, we know there is no need to update back refs
4894 * for the subtree
4895 */
4896 if (wc->stage == UPDATE_BACKREF &&
Yan, Zheng94fcca92009-10-09 09:25:16 -04004897 generation <= root->root_key.offset) {
4898 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004899 return 1;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004900 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004901
4902 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
Qu Wenruo581c1762018-03-29 09:08:11 +08004903 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
4904 path->slots[level]);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004905
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004906 next = find_extent_buffer(fs_info, bytenr);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004907 if (!next) {
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004908 next = btrfs_find_create_tree_block(fs_info, bytenr);
Liu Boc871b0f2016-06-06 12:01:23 -07004909 if (IS_ERR(next))
4910 return PTR_ERR(next);
4911
Josef Bacikb2aaaa32013-07-05 17:05:38 -04004912 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
4913 level - 1);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004914 reada = 1;
4915 }
4916 btrfs_tree_lock(next);
David Sterba8bead252018-04-04 02:03:48 +02004917 btrfs_set_lock_blocking_write(next);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004918
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04004919 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
Yan, Zheng94fcca92009-10-09 09:25:16 -04004920 &wc->refs[level - 1],
4921 &wc->flags[level - 1]);
Josef Bacik48672682016-09-23 13:23:28 +02004922 if (ret < 0)
4923 goto out_unlock;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01004924
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004925 if (unlikely(wc->refs[level - 1] == 0)) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04004926 btrfs_err(fs_info, "Missing references.");
Josef Bacik48672682016-09-23 13:23:28 +02004927 ret = -EIO;
4928 goto out_unlock;
Simon Kirbyc2cf52e2013-03-19 22:41:23 +00004929 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04004930 *lookup_info = 0;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004931
Yan, Zheng94fcca92009-10-09 09:25:16 -04004932 if (wc->stage == DROP_REFERENCE) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004933 if (wc->refs[level - 1] > 1) {
Mark Fasheh11526512014-07-17 12:39:01 -07004934 need_account = true;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004935 if (level == 1 &&
4936 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4937 goto skip;
4938
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004939 if (!wc->update_ref ||
4940 generation <= root->root_key.offset)
4941 goto skip;
4942
4943 btrfs_node_key_to_cpu(path->nodes[level], &key,
4944 path->slots[level]);
4945 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
4946 if (ret < 0)
4947 goto skip;
4948
4949 wc->stage = UPDATE_BACKREF;
4950 wc->shared_level = level - 1;
4951 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04004952 } else {
4953 if (level == 1 &&
4954 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
4955 goto skip;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004956 }
4957
Chris Masonb9fab912012-05-06 07:23:47 -04004958 if (!btrfs_buffer_uptodate(next, generation, 0)) {
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004959 btrfs_tree_unlock(next);
4960 free_extent_buffer(next);
4961 next = NULL;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004962 *lookup_info = 1;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004963 }
4964
4965 if (!next) {
4966 if (reada && level == 1)
4967 reada_walk_down(trans, root, wc, path);
Qu Wenruo581c1762018-03-29 09:08:11 +08004968 next = read_tree_block(fs_info, bytenr, generation, level - 1,
4969 &first_key);
Liu Bo64c043d2015-05-25 17:30:15 +08004970 if (IS_ERR(next)) {
4971 return PTR_ERR(next);
4972 } else if (!extent_buffer_uptodate(next)) {
Josef Bacik416bc652013-04-23 14:17:42 -04004973 free_extent_buffer(next);
Tsutomu Itoh97d9a8a2011-03-24 06:33:21 +00004974 return -EIO;
Josef Bacik416bc652013-04-23 14:17:42 -04004975 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004976 btrfs_tree_lock(next);
David Sterba8bead252018-04-04 02:03:48 +02004977 btrfs_set_lock_blocking_write(next);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004978 }
4979
4980 level--;
Josef Bacik48672682016-09-23 13:23:28 +02004981 ASSERT(level == btrfs_header_level(next));
4982 if (level != btrfs_header_level(next)) {
4983 btrfs_err(root->fs_info, "mismatched level");
4984 ret = -EIO;
4985 goto out_unlock;
4986 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004987 path->nodes[level] = next;
4988 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04004989 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04004990 wc->level = level;
4991 if (wc->level == 1)
4992 wc->reada_slot = 0;
4993 return 0;
4994skip:
4995 wc->refs[level - 1] = 0;
4996 wc->flags[level - 1] = 0;
Yan, Zheng94fcca92009-10-09 09:25:16 -04004997 if (wc->stage == DROP_REFERENCE) {
4998 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
4999 parent = path->nodes[level]->start;
5000 } else {
Josef Bacik48672682016-09-23 13:23:28 +02005001 ASSERT(root->root_key.objectid ==
Yan, Zheng94fcca92009-10-09 09:25:16 -04005002 btrfs_header_owner(path->nodes[level]));
Josef Bacik48672682016-09-23 13:23:28 +02005003 if (root->root_key.objectid !=
5004 btrfs_header_owner(path->nodes[level])) {
5005 btrfs_err(root->fs_info,
5006 "mismatched block owner");
5007 ret = -EIO;
5008 goto out_unlock;
5009 }
Yan, Zheng94fcca92009-10-09 09:25:16 -04005010 parent = 0;
5011 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005012
Qu Wenruo2cd86d32018-09-27 14:42:33 +08005013 /*
Josef Bacik78c52d92019-02-06 15:46:14 -05005014 * If we had a drop_progress we need to verify the refs are set
5015 * as expected. If we find our ref then we know that from here
5016 * on out everything should be correct, and we can clear the
5017 * ->restarted flag.
5018 */
5019 if (wc->restarted) {
5020 ret = check_ref_exists(trans, root, bytenr, parent,
5021 level - 1);
5022 if (ret < 0)
5023 goto out_unlock;
5024 if (ret == 0)
5025 goto no_delete;
5026 ret = 0;
5027 wc->restarted = 0;
5028 }
5029
5030 /*
Qu Wenruo2cd86d32018-09-27 14:42:33 +08005031 * Reloc tree doesn't contribute to qgroup numbers, and we have
5032 * already accounted them at merge time (replace_path),
5033 * thus we could skip expensive subtree trace here.
5034 */
5035 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
5036 need_account) {
Lu Fengqideb40622018-07-18 14:45:38 +08005037 ret = btrfs_qgroup_trace_subtree(trans, next,
Qu Wenruo33d1f052016-10-18 09:31:28 +08005038 generation, level - 1);
Mark Fasheh11526512014-07-17 12:39:01 -07005039 if (ret) {
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005040 btrfs_err_rl(fs_info,
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005041 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
5042 ret);
Mark Fasheh11526512014-07-17 12:39:01 -07005043 }
5044 }
Josef Bacikaea6f022019-02-06 15:46:15 -05005045
5046 /*
5047 * We need to update the next key in our walk control so we can
5048 * update the drop_progress key accordingly. We don't care if
5049 * find_next_key doesn't find a key because that means we're at
5050 * the end and are going to clean up now.
5051 */
5052 wc->drop_level = level;
5053 find_next_key(path, level, &wc->drop_progress);
5054
Qu Wenruoffd4bb22019-04-04 14:45:36 +08005055 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
5056 fs_info->nodesize, parent);
5057 btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid);
5058 ret = btrfs_free_extent(trans, &ref);
Josef Bacik48672682016-09-23 13:23:28 +02005059 if (ret)
5060 goto out_unlock;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005061 }
Josef Bacik78c52d92019-02-06 15:46:14 -05005062no_delete:
Josef Bacik48672682016-09-23 13:23:28 +02005063 *lookup_info = 1;
5064 ret = 1;
5065
5066out_unlock:
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005067 btrfs_tree_unlock(next);
5068 free_extent_buffer(next);
Josef Bacik48672682016-09-23 13:23:28 +02005069
5070 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005071}
5072
5073/*
Liu Bo2c016dc2012-12-26 15:32:17 +08005074 * helper to process tree block while walking up the tree.
Yan Zheng2c47e6052009-06-27 21:07:35 -04005075 *
5076 * when wc->stage == DROP_REFERENCE, this function drops
5077 * reference count on the block.
5078 *
5079 * when wc->stage == UPDATE_BACKREF, this function changes
5080 * wc->stage back to DROP_REFERENCE if we changed wc->stage
5081 * to UPDATE_BACKREF previously while processing the block.
5082 *
5083 * NOTE: return value 1 means we should stop walking up.
5084 */
5085static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
5086 struct btrfs_root *root,
5087 struct btrfs_path *path,
5088 struct walk_control *wc)
5089{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005090 struct btrfs_fs_info *fs_info = root->fs_info;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005091 int ret;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005092 int level = wc->level;
5093 struct extent_buffer *eb = path->nodes[level];
5094 u64 parent = 0;
5095
5096 if (wc->stage == UPDATE_BACKREF) {
5097 BUG_ON(wc->shared_level < level);
5098 if (level < wc->shared_level)
5099 goto out;
5100
Yan Zheng2c47e6052009-06-27 21:07:35 -04005101 ret = find_next_key(path, level + 1, &wc->update_progress);
5102 if (ret > 0)
5103 wc->update_ref = 0;
5104
5105 wc->stage = DROP_REFERENCE;
5106 wc->shared_level = -1;
5107 path->slots[level] = 0;
5108
5109 /*
5110 * check reference count again if the block isn't locked.
5111 * we should start walking down the tree again if reference
5112 * count is one.
5113 */
5114 if (!path->locks[level]) {
5115 BUG_ON(level == 0);
5116 btrfs_tree_lock(eb);
David Sterba8bead252018-04-04 02:03:48 +02005117 btrfs_set_lock_blocking_write(eb);
Chris Masonbd681512011-07-16 15:23:14 -04005118 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005119
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005120 ret = btrfs_lookup_extent_info(trans, fs_info,
Josef Bacik3173a182013-03-07 14:22:04 -05005121 eb->start, level, 1,
Yan Zheng2c47e6052009-06-27 21:07:35 -04005122 &wc->refs[level],
5123 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005124 if (ret < 0) {
5125 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00005126 path->locks[level] = 0;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005127 return ret;
5128 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005129 BUG_ON(wc->refs[level] == 0);
5130 if (wc->refs[level] == 1) {
Chris Masonbd681512011-07-16 15:23:14 -04005131 btrfs_tree_unlock_rw(eb, path->locks[level]);
Liu Bo3268a242012-12-28 09:33:19 +00005132 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005133 return 1;
5134 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005135 }
5136 }
5137
5138 /* wc->stage == DROP_REFERENCE */
5139 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
5140
5141 if (wc->refs[level] == 1) {
5142 if (level == 0) {
5143 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
Josef Bacike339a6b2014-07-02 10:54:25 -07005144 ret = btrfs_dec_ref(trans, root, eb, 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005145 else
Josef Bacike339a6b2014-07-02 10:54:25 -07005146 ret = btrfs_dec_ref(trans, root, eb, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005147 BUG_ON(ret); /* -ENOMEM */
Qu Wenruoc4140cb2019-04-04 14:45:37 +08005148 if (is_fstree(root->root_key.objectid)) {
5149 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
5150 if (ret) {
5151 btrfs_err_rl(fs_info,
5152 "error %d accounting leaf items, quota is out of sync, rescan required",
Jeff Mahoney5d163e02016-09-20 10:05:00 -04005153 ret);
Qu Wenruoc4140cb2019-04-04 14:45:37 +08005154 }
Mark Fasheh11526512014-07-17 12:39:01 -07005155 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005156 }
David Sterba6a884d7d2019-03-20 14:30:02 +01005157 /* make block locked assertion in btrfs_clean_tree_block happy */
Yan Zheng2c47e6052009-06-27 21:07:35 -04005158 if (!path->locks[level] &&
5159 btrfs_header_generation(eb) == trans->transid) {
5160 btrfs_tree_lock(eb);
David Sterba8bead252018-04-04 02:03:48 +02005161 btrfs_set_lock_blocking_write(eb);
Chris Masonbd681512011-07-16 15:23:14 -04005162 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005163 }
David Sterba6a884d7d2019-03-20 14:30:02 +01005164 btrfs_clean_tree_block(eb);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005165 }
5166
5167 if (eb == root->node) {
5168 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5169 parent = eb->start;
Qu Wenruo65c6e822018-08-21 09:42:03 +08005170 else if (root->root_key.objectid != btrfs_header_owner(eb))
5171 goto owner_mismatch;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005172 } else {
5173 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
5174 parent = path->nodes[level + 1]->start;
Qu Wenruo65c6e822018-08-21 09:42:03 +08005175 else if (root->root_key.objectid !=
5176 btrfs_header_owner(path->nodes[level + 1]))
5177 goto owner_mismatch;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005178 }
5179
Jan Schmidt5581a512012-05-16 17:04:52 +02005180 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005181out:
5182 wc->refs[level] = 0;
5183 wc->flags[level] = 0;
Yan, Zhengf0486c62010-05-16 10:46:25 -04005184 return 0;
Qu Wenruo65c6e822018-08-21 09:42:03 +08005185
5186owner_mismatch:
5187 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
5188 btrfs_header_owner(eb), root->root_key.objectid);
5189 return -EUCLEAN;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005190}
5191
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005192static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
5193 struct btrfs_root *root,
Yan Zheng2c47e6052009-06-27 21:07:35 -04005194 struct btrfs_path *path,
5195 struct walk_control *wc)
Yan Zhengf82d02d2008-10-29 14:49:05 -04005196{
Yan Zheng2c47e6052009-06-27 21:07:35 -04005197 int level = wc->level;
Yan, Zheng94fcca92009-10-09 09:25:16 -04005198 int lookup_info = 1;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005199 int ret;
5200
Yan Zheng2c47e6052009-06-27 21:07:35 -04005201 while (level >= 0) {
Yan, Zheng94fcca92009-10-09 09:25:16 -04005202 ret = walk_down_proc(trans, root, path, wc, lookup_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005203 if (ret > 0)
Yan Zhengf82d02d2008-10-29 14:49:05 -04005204 break;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005205
Yan Zheng2c47e6052009-06-27 21:07:35 -04005206 if (level == 0)
5207 break;
5208
Yan, Zheng7a7965f2010-02-01 02:41:17 +00005209 if (path->slots[level] >=
5210 btrfs_header_nritems(path->nodes[level]))
5211 break;
5212
Yan, Zheng94fcca92009-10-09 09:25:16 -04005213 ret = do_walk_down(trans, root, path, wc, &lookup_info);
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005214 if (ret > 0) {
5215 path->slots[level]++;
5216 continue;
Miao Xie90d2c51d2010-03-25 12:37:12 +00005217 } else if (ret < 0)
5218 return ret;
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005219 level = wc->level;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005220 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04005221 return 0;
5222}
5223
Chris Masond3977122009-01-05 21:25:51 -05005224static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
Chris Mason98ed5172008-01-03 10:01:48 -05005225 struct btrfs_root *root,
Yan Zhengf82d02d2008-10-29 14:49:05 -04005226 struct btrfs_path *path,
Yan Zheng2c47e6052009-06-27 21:07:35 -04005227 struct walk_control *wc, int max_level)
Chris Mason20524f02007-03-10 06:35:47 -05005228{
Yan Zheng2c47e6052009-06-27 21:07:35 -04005229 int level = wc->level;
Chris Mason20524f02007-03-10 06:35:47 -05005230 int ret;
Chris Mason9f3a7422007-08-07 15:52:19 -04005231
Yan Zheng2c47e6052009-06-27 21:07:35 -04005232 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
5233 while (level < max_level && path->nodes[level]) {
5234 wc->level = level;
5235 if (path->slots[level] + 1 <
5236 btrfs_header_nritems(path->nodes[level])) {
5237 path->slots[level]++;
Chris Mason20524f02007-03-10 06:35:47 -05005238 return 0;
5239 } else {
Yan Zheng2c47e6052009-06-27 21:07:35 -04005240 ret = walk_up_proc(trans, root, path, wc);
5241 if (ret > 0)
5242 return 0;
Qu Wenruo65c6e822018-08-21 09:42:03 +08005243 if (ret < 0)
5244 return ret;
Chris Masonbd56b302009-02-04 09:27:02 -05005245
Yan Zheng2c47e6052009-06-27 21:07:35 -04005246 if (path->locks[level]) {
Chris Masonbd681512011-07-16 15:23:14 -04005247 btrfs_tree_unlock_rw(path->nodes[level],
5248 path->locks[level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005249 path->locks[level] = 0;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005250 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005251 free_extent_buffer(path->nodes[level]);
5252 path->nodes[level] = NULL;
5253 level++;
Chris Mason20524f02007-03-10 06:35:47 -05005254 }
5255 }
5256 return 1;
5257}
5258
Chris Mason9aca1d52007-03-13 11:09:37 -04005259/*
Yan Zheng2c47e6052009-06-27 21:07:35 -04005260 * drop a subvolume tree.
5261 *
5262 * this function traverses the tree freeing any blocks that only
5263 * referenced by the tree.
5264 *
5265 * when a shared tree block is found. this function decreases its
5266 * reference count by one. if update_ref is true, this function
5267 * also make sure backrefs for the shared block and all lower level
5268 * blocks are properly updated.
David Sterba9d1a2a32013-03-12 15:13:28 +00005269 *
5270 * If called with for_reloc == 0, may exit early with -EAGAIN
Chris Mason9aca1d52007-03-13 11:09:37 -04005271 */
Nikolay Borisov0078a9f2020-03-10 11:43:51 +02005272int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref, int for_reloc)
Chris Mason20524f02007-03-10 06:35:47 -05005273{
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005274 struct btrfs_fs_info *fs_info = root->fs_info;
Chris Mason5caf2a02007-04-02 11:20:42 -04005275 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005276 struct btrfs_trans_handle *trans;
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005277 struct btrfs_root *tree_root = fs_info->tree_root;
Chris Mason9f3a7422007-08-07 15:52:19 -04005278 struct btrfs_root_item *root_item = &root->root_item;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005279 struct walk_control *wc;
5280 struct btrfs_key key;
5281 int err = 0;
5282 int ret;
5283 int level;
Josef Bacikd29a9f62013-07-17 19:30:20 -04005284 bool root_dropped = false;
Chris Mason20524f02007-03-10 06:35:47 -05005285
Misono Tomohiro4fd786e2018-08-06 14:25:24 +09005286 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
Mark Fasheh11526512014-07-17 12:39:01 -07005287
Chris Mason5caf2a02007-04-02 11:20:42 -04005288 path = btrfs_alloc_path();
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00005289 if (!path) {
5290 err = -ENOMEM;
5291 goto out;
5292 }
Chris Mason20524f02007-03-10 06:35:47 -05005293
Yan Zheng2c47e6052009-06-27 21:07:35 -04005294 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Mark Fasheh38a1a912011-07-13 10:59:59 -07005295 if (!wc) {
5296 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00005297 err = -ENOMEM;
5298 goto out;
Mark Fasheh38a1a912011-07-13 10:59:59 -07005299 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005300
Qu Wenruof3e3d9c2020-07-13 09:03:20 +08005301 /*
5302 * Use join to avoid potential EINTR from transaction start. See
5303 * wait_reserve_ticket and the whole reservation callchain.
5304 */
5305 if (for_reloc)
5306 trans = btrfs_join_transaction(tree_root);
5307 else
5308 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005309 if (IS_ERR(trans)) {
5310 err = PTR_ERR(trans);
5311 goto out_free;
5312 }
Tsutomu Itoh98d5dc12011-01-20 06:19:37 +00005313
Josef Bacik0568e822018-11-30 11:52:14 -05005314 err = btrfs_run_delayed_items(trans);
5315 if (err)
5316 goto out_end_trans;
5317
Josef Bacik83354f02018-11-30 11:52:13 -05005318 /*
5319 * This will help us catch people modifying the fs tree while we're
5320 * dropping it. It is unsafe to mess with the fs tree while it's being
5321 * dropped as we unlock the root node and parent nodes as we walk down
5322 * the tree, assuming nothing will change. If something does change
5323 * then we'll have stale information and drop references to blocks we've
5324 * already dropped.
5325 */
5326 set_bit(BTRFS_ROOT_DELETING, &root->state);
Chris Mason9f3a7422007-08-07 15:52:19 -04005327 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04005328 level = btrfs_header_level(root->node);
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005329 path->nodes[level] = btrfs_lock_root_node(root);
David Sterba8bead252018-04-04 02:03:48 +02005330 btrfs_set_lock_blocking_write(path->nodes[level]);
Chris Mason9f3a7422007-08-07 15:52:19 -04005331 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04005332 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005333 memset(&wc->update_progress, 0,
5334 sizeof(wc->update_progress));
Chris Mason9f3a7422007-08-07 15:52:19 -04005335 } else {
Chris Mason9f3a7422007-08-07 15:52:19 -04005336 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005337 memcpy(&wc->update_progress, &key,
5338 sizeof(wc->update_progress));
5339
Chris Mason6702ed42007-08-07 16:15:09 -04005340 level = root_item->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005341 BUG_ON(level == 0);
Chris Mason6702ed42007-08-07 16:15:09 -04005342 path->lowest_level = level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5344 path->lowest_level = 0;
5345 if (ret < 0) {
5346 err = ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005347 goto out_end_trans;
Chris Mason9f3a7422007-08-07 15:52:19 -04005348 }
Yan, Zheng1c4850e2009-09-21 15:55:59 -04005349 WARN_ON(ret > 0);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005350
Chris Mason7d9eb122008-07-08 14:19:17 -04005351 /*
5352 * unlock our path, this is safe because only this
5353 * function is allowed to delete this snapshot
5354 */
Yan Zheng5d4f98a2009-06-10 10:45:14 -04005355 btrfs_unlock_up_safe(path, 0);
Chris Mason9aca1d52007-03-13 11:09:37 -04005356
Yan Zheng2c47e6052009-06-27 21:07:35 -04005357 level = btrfs_header_level(root->node);
5358 while (1) {
5359 btrfs_tree_lock(path->nodes[level]);
David Sterba8bead252018-04-04 02:03:48 +02005360 btrfs_set_lock_blocking_write(path->nodes[level]);
Josef Bacikfec386a2013-07-15 12:41:42 -04005361 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005362
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005363 ret = btrfs_lookup_extent_info(trans, fs_info,
Yan Zheng2c47e6052009-06-27 21:07:35 -04005364 path->nodes[level]->start,
Josef Bacik3173a182013-03-07 14:22:04 -05005365 level, 1, &wc->refs[level],
Yan Zheng2c47e6052009-06-27 21:07:35 -04005366 &wc->flags[level]);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005367 if (ret < 0) {
5368 err = ret;
5369 goto out_end_trans;
5370 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005371 BUG_ON(wc->refs[level] == 0);
5372
5373 if (level == root_item->drop_level)
5374 break;
5375
5376 btrfs_tree_unlock(path->nodes[level]);
Josef Bacikfec386a2013-07-15 12:41:42 -04005377 path->locks[level] = 0;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005378 WARN_ON(wc->refs[level] != 1);
5379 level--;
5380 }
5381 }
5382
Josef Bacik78c52d92019-02-06 15:46:14 -05005383 wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005384 wc->level = level;
5385 wc->shared_level = -1;
5386 wc->stage = DROP_REFERENCE;
5387 wc->update_ref = update_ref;
5388 wc->keep_locks = 0;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005389 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005390
5391 while (1) {
David Sterba9d1a2a32013-03-12 15:13:28 +00005392
Yan Zheng2c47e6052009-06-27 21:07:35 -04005393 ret = walk_down_tree(trans, root, path, wc);
5394 if (ret < 0) {
5395 err = ret;
Chris Masone7a84562008-06-25 16:01:31 -04005396 break;
5397 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005398
5399 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
5400 if (ret < 0) {
5401 err = ret;
5402 break;
5403 }
5404
5405 if (ret > 0) {
5406 BUG_ON(wc->stage != DROP_REFERENCE);
5407 break;
5408 }
5409
5410 if (wc->stage == DROP_REFERENCE) {
Josef Bacikaea6f022019-02-06 15:46:15 -05005411 wc->drop_level = wc->level;
5412 btrfs_node_key_to_cpu(path->nodes[wc->drop_level],
5413 &wc->drop_progress,
5414 path->slots[wc->drop_level]);
Yan Zheng2c47e6052009-06-27 21:07:35 -04005415 }
Josef Bacikaea6f022019-02-06 15:46:15 -05005416 btrfs_cpu_key_to_disk(&root_item->drop_progress,
5417 &wc->drop_progress);
5418 root_item->drop_level = wc->drop_level;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005419
5420 BUG_ON(wc->level == 0);
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005421 if (btrfs_should_end_transaction(trans) ||
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005422 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04005423 ret = btrfs_update_root(trans, tree_root,
5424 &root->root_key,
5425 root_item);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005426 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04005427 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005428 err = ret;
5429 goto out_end_trans;
5430 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005431
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005432 btrfs_end_transaction_throttle(trans);
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005433 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
Jeff Mahoneyab8d0fc2016-09-20 10:05:02 -04005434 btrfs_debug(fs_info,
5435 "drop snapshot early exit");
Josef Bacik3c8f2422013-07-15 11:57:06 -04005436 err = -EAGAIN;
5437 goto out_free;
5438 }
5439
Yan, Zhenga22285a2010-05-16 10:48:46 -04005440 trans = btrfs_start_transaction(tree_root, 0);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005441 if (IS_ERR(trans)) {
5442 err = PTR_ERR(trans);
5443 goto out_free;
5444 }
Chris Masonc3e69d52009-03-13 10:17:05 -04005445 }
Chris Mason20524f02007-03-10 06:35:47 -05005446 }
David Sterbab3b4aa72011-04-21 01:20:15 +02005447 btrfs_release_path(path);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005448 if (err)
5449 goto out_end_trans;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005450
Lu Fengqiab9ce7d2018-08-01 11:32:27 +08005451 ret = btrfs_del_root(trans, &root->root_key);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005452 if (ret) {
Jeff Mahoney66642832016-06-10 18:19:25 -04005453 btrfs_abort_transaction(trans, ret);
Jeff Mahoneye19182c2017-12-04 13:11:45 -05005454 err = ret;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005455 goto out_end_trans;
5456 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005457
Yan, Zheng76dda932009-09-21 16:00:26 -04005458 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
Miao Xiecb517ea2013-05-15 07:48:19 +00005459 ret = btrfs_find_root(tree_root, &root->root_key, path,
5460 NULL, NULL);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005461 if (ret < 0) {
Jeff Mahoney66642832016-06-10 18:19:25 -04005462 btrfs_abort_transaction(trans, ret);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005463 err = ret;
5464 goto out_end_trans;
5465 } else if (ret > 0) {
Josef Bacik84cd9482010-12-08 12:24:01 -05005466 /* if we fail to delete the orphan item this time
5467 * around, it'll get picked up the next time.
5468 *
5469 * The most common failure here is just -ENOENT.
5470 */
5471 btrfs_del_orphan_item(trans, tree_root,
5472 root->root_key.objectid);
Yan, Zheng76dda932009-09-21 16:00:26 -04005473 }
5474 }
5475
Qu Wenruoa3cf0e42020-07-14 09:12:20 +08005476 /*
5477 * This subvolume is going to be completely dropped, and won't be
5478 * recorded as dirty roots, thus pertrans meta rsv will not be freed at
5479 * commit transaction time. So free it here manually.
5480 */
5481 btrfs_qgroup_convert_reserved_meta(root, INT_MAX);
5482 btrfs_qgroup_free_meta_all_pertrans(root);
5483
Josef Bacik8c389382020-02-14 16:11:42 -05005484 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state))
Josef Bacik2b9dbef2015-09-15 10:07:04 -04005485 btrfs_add_dropped_root(trans, root);
Josef Bacik8c389382020-02-14 16:11:42 -05005486 else
Josef Bacik00246522020-01-24 09:33:01 -05005487 btrfs_put_root(root);
Josef Bacikd29a9f62013-07-17 19:30:20 -04005488 root_dropped = true;
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005489out_end_trans:
Jeff Mahoney3a45bb22016-09-09 21:39:03 -04005490 btrfs_end_transaction_throttle(trans);
Jeff Mahoney79787ea2012-03-12 16:03:00 +01005491out_free:
Yan Zheng2c47e6052009-06-27 21:07:35 -04005492 kfree(wc);
Chris Mason5caf2a02007-04-02 11:20:42 -04005493 btrfs_free_path(path);
Tsutomu Itohcb1b69f2011-08-09 07:11:13 +00005494out:
Josef Bacikd29a9f62013-07-17 19:30:20 -04005495 /*
5496 * So if we need to stop dropping the snapshot for whatever reason we
5497 * need to make sure to add it back to the dead root list so that we
5498 * keep trying to do the work later. This also cleans up roots if we
5499 * don't have it in the radix (like when we recover after a power fail
5500 * or unmount) so we don't leak memory.
5501 */
Thomas Meyer897ca812017-10-07 16:02:21 +02005502 if (!for_reloc && !root_dropped)
Josef Bacikd29a9f62013-07-17 19:30:20 -04005503 btrfs_add_dead_root(root);
Jeff Mahoney2c536792011-10-03 23:22:41 -04005504 return err;
Chris Mason20524f02007-03-10 06:35:47 -05005505}
Chris Mason9078a3e2007-04-26 16:46:15 -04005506
Yan Zheng2c47e6052009-06-27 21:07:35 -04005507/*
5508 * drop subtree rooted at tree block 'node'.
5509 *
5510 * NOTE: this function will unlock and release tree block 'node'
Arne Jansen66d7e7f2011-09-12 15:26:38 +02005511 * only used by relocation code
Yan Zheng2c47e6052009-06-27 21:07:35 -04005512 */
Yan Zhengf82d02d2008-10-29 14:49:05 -04005513int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
5514 struct btrfs_root *root,
5515 struct extent_buffer *node,
5516 struct extent_buffer *parent)
5517{
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005518 struct btrfs_fs_info *fs_info = root->fs_info;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005519 struct btrfs_path *path;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005520 struct walk_control *wc;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005521 int level;
5522 int parent_level;
5523 int ret = 0;
5524 int wret;
5525
Yan Zheng2c47e6052009-06-27 21:07:35 -04005526 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
5527
Yan Zhengf82d02d2008-10-29 14:49:05 -04005528 path = btrfs_alloc_path();
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005529 if (!path)
5530 return -ENOMEM;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005531
Yan Zheng2c47e6052009-06-27 21:07:35 -04005532 wc = kzalloc(sizeof(*wc), GFP_NOFS);
Tsutomu Itohdb5b4932011-03-23 08:14:16 +00005533 if (!wc) {
5534 btrfs_free_path(path);
5535 return -ENOMEM;
5536 }
Yan Zheng2c47e6052009-06-27 21:07:35 -04005537
Chris Masonb9447ef82009-03-09 11:45:38 -04005538 btrfs_assert_tree_locked(parent);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005539 parent_level = btrfs_header_level(parent);
David Sterba67439da2019-10-08 13:28:47 +02005540 atomic_inc(&parent->refs);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005541 path->nodes[parent_level] = parent;
5542 path->slots[parent_level] = btrfs_header_nritems(parent);
5543
Chris Masonb9447ef82009-03-09 11:45:38 -04005544 btrfs_assert_tree_locked(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005545 level = btrfs_header_level(node);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005546 path->nodes[level] = node;
5547 path->slots[level] = 0;
Chris Masonbd681512011-07-16 15:23:14 -04005548 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005549
5550 wc->refs[parent_level] = 1;
5551 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
5552 wc->level = level;
5553 wc->shared_level = -1;
5554 wc->stage = DROP_REFERENCE;
5555 wc->update_ref = 0;
5556 wc->keep_locks = 1;
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005557 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005558
5559 while (1) {
Yan Zheng2c47e6052009-06-27 21:07:35 -04005560 wret = walk_down_tree(trans, root, path, wc);
5561 if (wret < 0) {
Yan Zhengf82d02d2008-10-29 14:49:05 -04005562 ret = wret;
Yan Zhengf82d02d2008-10-29 14:49:05 -04005563 break;
Yan Zheng2c47e6052009-06-27 21:07:35 -04005564 }
Yan Zhengf82d02d2008-10-29 14:49:05 -04005565
Yan Zheng2c47e6052009-06-27 21:07:35 -04005566 wret = walk_up_tree(trans, root, path, wc, parent_level);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005567 if (wret < 0)
5568 ret = wret;
5569 if (wret != 0)
5570 break;
5571 }
5572
Yan Zheng2c47e6052009-06-27 21:07:35 -04005573 kfree(wc);
Yan Zhengf82d02d2008-10-29 14:49:05 -04005574 btrfs_free_path(path);
5575 return ret;
5576}
5577
Miao Xie6d07bce2011-01-05 10:07:31 +00005578/*
5579 * helper to account the unused space of all the readonly block group in the
Josef Bacik633c0aa2014-10-31 09:49:34 -04005580 * space_info. takes mirrors into account.
Miao Xie6d07bce2011-01-05 10:07:31 +00005581 */
Josef Bacik633c0aa2014-10-31 09:49:34 -04005582u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
Miao Xie6d07bce2011-01-05 10:07:31 +00005583{
David Sterba32da53862019-10-29 19:20:18 +01005584 struct btrfs_block_group *block_group;
Miao Xie6d07bce2011-01-05 10:07:31 +00005585 u64 free_bytes = 0;
5586 int factor;
5587
Nicholas D Steeves01327612016-05-19 21:18:45 -04005588 /* It's df, we don't care if it's racy */
Josef Bacik633c0aa2014-10-31 09:49:34 -04005589 if (list_empty(&sinfo->ro_bgs))
5590 return 0;
5591
5592 spin_lock(&sinfo->lock);
5593 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
Miao Xie6d07bce2011-01-05 10:07:31 +00005594 spin_lock(&block_group->lock);
5595
5596 if (!block_group->ro) {
5597 spin_unlock(&block_group->lock);
5598 continue;
5599 }
5600
David Sterba46df06b2018-07-13 20:46:30 +02005601 factor = btrfs_bg_type_to_factor(block_group->flags);
David Sterbab3470b52019-10-23 18:48:22 +02005602 free_bytes += (block_group->length -
David Sterbabf38be62019-10-23 18:48:11 +02005603 block_group->used) * factor;
Miao Xie6d07bce2011-01-05 10:07:31 +00005604
5605 spin_unlock(&block_group->lock);
5606 }
Miao Xie6d07bce2011-01-05 10:07:31 +00005607 spin_unlock(&sinfo->lock);
5608
5609 return free_bytes;
5610}
5611
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005612int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
5613 u64 start, u64 end)
liuboacce9522011-01-06 19:30:25 +08005614{
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005615 return unpin_extent_range(fs_info, start, end, false);
liuboacce9522011-01-06 19:30:25 +08005616}
5617
Jeff Mahoney499f3772015-06-15 09:41:17 -04005618/*
5619 * It used to be that old block groups would be left around forever.
5620 * Iterating over them would be enough to trim unused space. Since we
5621 * now automatically remove them, we also need to iterate over unallocated
5622 * space.
5623 *
5624 * We don't want a transaction for this since the discard may take a
5625 * substantial amount of time. We don't require that a transaction be
5626 * running, but we do need to take a running transaction into account
Jeff Mahoneyfee7acc2018-09-06 17:18:16 -04005627 * to ensure that we're not discarding chunks that were released or
5628 * allocated in the current transaction.
Jeff Mahoney499f3772015-06-15 09:41:17 -04005629 *
5630 * Holding the chunks lock will prevent other threads from allocating
5631 * or releasing chunks, but it won't prevent a running transaction
5632 * from committing and releasing the memory that the pending chunks
5633 * list head uses. For that, we need to take a reference to the
Jeff Mahoneyfee7acc2018-09-06 17:18:16 -04005634 * transaction and hold the commit root sem. We only need to hold
5635 * it while performing the free space search since we have already
5636 * held back allocations.
Jeff Mahoney499f3772015-06-15 09:41:17 -04005637 */
Nikolay Borisov8103d102019-06-03 13:06:00 +03005638static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed)
Jeff Mahoney499f3772015-06-15 09:41:17 -04005639{
Nikolay Borisov8103d102019-06-03 13:06:00 +03005640 u64 start = SZ_1M, len = 0, end = 0;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005641 int ret;
5642
5643 *trimmed = 0;
5644
Jeff Mahoney0be88e32018-09-06 17:18:15 -04005645 /* Discard not supported = nothing to do. */
5646 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
5647 return 0;
5648
Andrea Gelmini52042d82018-11-28 12:05:13 +01005649 /* Not writable = nothing to do. */
Anand Jainebbede42017-12-04 12:54:52 +08005650 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
Jeff Mahoney499f3772015-06-15 09:41:17 -04005651 return 0;
5652
5653 /* No free space = nothing to do. */
5654 if (device->total_bytes <= device->bytes_used)
5655 return 0;
5656
5657 ret = 0;
5658
5659 while (1) {
Jeff Mahoneyfb456252016-06-22 18:54:56 -04005660 struct btrfs_fs_info *fs_info = device->fs_info;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005661 u64 bytes;
5662
5663 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
5664 if (ret)
Jeff Mahoneyfee7acc2018-09-06 17:18:16 -04005665 break;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005666
Nikolay Borisov929be172019-03-27 14:24:18 +02005667 find_first_clear_extent_bit(&device->alloc_state, start,
5668 &start, &end,
5669 CHUNK_TRIMMED | CHUNK_ALLOCATED);
Nikolay Borisov53460a42019-06-03 13:06:01 +03005670
5671 /* Ensure we skip the reserved area in the first 1M */
5672 start = max_t(u64, start, SZ_1M);
5673
Nikolay Borisov929be172019-03-27 14:24:18 +02005674 /*
5675 * If find_first_clear_extent_bit find a range that spans the
5676 * end of the device it will set end to -1, in this case it's up
5677 * to the caller to trim the value to the size of the device.
5678 */
5679 end = min(end, device->total_bytes - 1);
Nikolay Borisov53460a42019-06-03 13:06:01 +03005680
Nikolay Borisov929be172019-03-27 14:24:18 +02005681 len = end - start + 1;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005682
Nikolay Borisov929be172019-03-27 14:24:18 +02005683 /* We didn't find any extents */
5684 if (!len) {
Jeff Mahoney499f3772015-06-15 09:41:17 -04005685 mutex_unlock(&fs_info->chunk_mutex);
Nikolay Borisov929be172019-03-27 14:24:18 +02005686 ret = 0;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005687 break;
5688 }
5689
Nikolay Borisov929be172019-03-27 14:24:18 +02005690 ret = btrfs_issue_discard(device->bdev, start, len,
5691 &bytes);
5692 if (!ret)
5693 set_extent_bits(&device->alloc_state, start,
5694 start + bytes - 1,
5695 CHUNK_TRIMMED);
Jeff Mahoney499f3772015-06-15 09:41:17 -04005696 mutex_unlock(&fs_info->chunk_mutex);
5697
5698 if (ret)
5699 break;
5700
5701 start += len;
5702 *trimmed += bytes;
5703
5704 if (fatal_signal_pending(current)) {
5705 ret = -ERESTARTSYS;
5706 break;
5707 }
5708
5709 cond_resched();
5710 }
5711
5712 return ret;
5713}
5714
Qu Wenruo93bba242018-09-07 14:16:23 +08005715/*
5716 * Trim the whole filesystem by:
5717 * 1) trimming the free space in each block group
5718 * 2) trimming the unallocated space on each device
5719 *
5720 * This will also continue trimming even if a block group or device encounters
5721 * an error. The return value will be the last error, or 0 if nothing bad
5722 * happens.
5723 */
Jeff Mahoney2ff7e612016-06-22 18:54:24 -04005724int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
Li Dongyangf7039b12011-03-24 10:24:28 +00005725{
David Sterba32da53862019-10-29 19:20:18 +01005726 struct btrfs_block_group *cache = NULL;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005727 struct btrfs_device *device;
5728 struct list_head *devices;
Li Dongyangf7039b12011-03-24 10:24:28 +00005729 u64 group_trimmed;
Qu Wenruo07301df2019-05-28 16:21:54 +08005730 u64 range_end = U64_MAX;
Li Dongyangf7039b12011-03-24 10:24:28 +00005731 u64 start;
5732 u64 end;
5733 u64 trimmed = 0;
Qu Wenruo93bba242018-09-07 14:16:23 +08005734 u64 bg_failed = 0;
5735 u64 dev_failed = 0;
5736 int bg_ret = 0;
5737 int dev_ret = 0;
Li Dongyangf7039b12011-03-24 10:24:28 +00005738 int ret = 0;
5739
Qu Wenruo07301df2019-05-28 16:21:54 +08005740 /*
5741 * Check range overflow if range->len is set.
5742 * The default range->len is U64_MAX.
5743 */
5744 if (range->len != U64_MAX &&
5745 check_add_overflow(range->start, range->len, &range_end))
5746 return -EINVAL;
5747
Qu Wenruo6ba9fc82018-09-07 14:16:24 +08005748 cache = btrfs_lookup_first_block_group(fs_info, range->start);
Josef Bacik2e405ad2019-06-20 15:37:45 -04005749 for (; cache; cache = btrfs_next_block_group(cache)) {
David Sterbab3470b52019-10-23 18:48:22 +02005750 if (cache->start >= range_end) {
Li Dongyangf7039b12011-03-24 10:24:28 +00005751 btrfs_put_block_group(cache);
5752 break;
5753 }
5754
David Sterbab3470b52019-10-23 18:48:22 +02005755 start = max(range->start, cache->start);
5756 end = min(range_end, cache->start + cache->length);
Li Dongyangf7039b12011-03-24 10:24:28 +00005757
5758 if (end - start >= range->minlen) {
David Sterba32da53862019-10-29 19:20:18 +01005759 if (!btrfs_block_group_done(cache)) {
Josef Bacik676f1f72019-06-20 15:37:48 -04005760 ret = btrfs_cache_block_group(cache, 0);
Josef Bacik1be41b72013-06-12 13:56:06 -04005761 if (ret) {
Qu Wenruo93bba242018-09-07 14:16:23 +08005762 bg_failed++;
5763 bg_ret = ret;
5764 continue;
Josef Bacik1be41b72013-06-12 13:56:06 -04005765 }
Josef Bacik676f1f72019-06-20 15:37:48 -04005766 ret = btrfs_wait_block_group_cache_done(cache);
Josef Bacik1be41b72013-06-12 13:56:06 -04005767 if (ret) {
Qu Wenruo93bba242018-09-07 14:16:23 +08005768 bg_failed++;
5769 bg_ret = ret;
5770 continue;
Josef Bacik1be41b72013-06-12 13:56:06 -04005771 }
Li Dongyangf7039b12011-03-24 10:24:28 +00005772 }
5773 ret = btrfs_trim_block_group(cache,
5774 &group_trimmed,
5775 start,
5776 end,
5777 range->minlen);
5778
5779 trimmed += group_trimmed;
5780 if (ret) {
Qu Wenruo93bba242018-09-07 14:16:23 +08005781 bg_failed++;
5782 bg_ret = ret;
5783 continue;
Li Dongyangf7039b12011-03-24 10:24:28 +00005784 }
5785 }
Li Dongyangf7039b12011-03-24 10:24:28 +00005786 }
5787
Qu Wenruo93bba242018-09-07 14:16:23 +08005788 if (bg_failed)
5789 btrfs_warn(fs_info,
5790 "failed to trim %llu block group(s), last error %d",
5791 bg_failed, bg_ret);
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005792 mutex_lock(&fs_info->fs_devices->device_list_mutex);
Jeff Mahoneyd4e329d2018-09-06 17:18:14 -04005793 devices = &fs_info->fs_devices->devices;
5794 list_for_each_entry(device, devices, dev_list) {
Nikolay Borisov8103d102019-06-03 13:06:00 +03005795 ret = btrfs_trim_free_extents(device, &group_trimmed);
Qu Wenruo93bba242018-09-07 14:16:23 +08005796 if (ret) {
5797 dev_failed++;
5798 dev_ret = ret;
Jeff Mahoney499f3772015-06-15 09:41:17 -04005799 break;
Qu Wenruo93bba242018-09-07 14:16:23 +08005800 }
Jeff Mahoney499f3772015-06-15 09:41:17 -04005801
5802 trimmed += group_trimmed;
5803 }
Jeff Mahoney0b246af2016-06-22 18:54:23 -04005804 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
Jeff Mahoney499f3772015-06-15 09:41:17 -04005805
Qu Wenruo93bba242018-09-07 14:16:23 +08005806 if (dev_failed)
5807 btrfs_warn(fs_info,
5808 "failed to trim %llu device(s), last error %d",
5809 dev_failed, dev_ret);
Li Dongyangf7039b12011-03-24 10:24:28 +00005810 range->len = trimmed;
Qu Wenruo93bba242018-09-07 14:16:23 +08005811 if (bg_ret)
5812 return bg_ret;
5813 return dev_ret;
Li Dongyangf7039b12011-03-24 10:24:28 +00005814}