blob: 44e59881a1f0b5d6d7f18082d36acc53f0d8233e [file] [log] [blame]
Theodore Ts'of5166762017-12-17 22:00:59 -05001// SPDX-License-Identifier: GPL-2.0
Alex Tomasa86c6182006-10-11 01:21:03 -07002/*
3 * Copyright (c) 2003-2006, Cluster File Systems, Inc, info@clusterfs.com
4 * Written by Alex Tomas <alex@clusterfs.com>
Alex Tomasa86c6182006-10-11 01:21:03 -07005 */
6
Christoph Hellwig3dcf5452008-04-29 18:13:32 -04007#ifndef _EXT4_EXTENTS
8#define _EXT4_EXTENTS
Alex Tomasa86c6182006-10-11 01:21:03 -07009
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040010#include "ext4.h"
Alex Tomasa86c6182006-10-11 01:21:03 -070011
12/*
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +010013 * With AGGRESSIVE_TEST defined, the capacity of index/leaf blocks
Randy Dunlapd0d856e2006-10-11 01:21:07 -070014 * becomes very small, so index split, in-depth growing and
15 * other hard changes happen much more often.
16 * This is for debug purposes only.
Alex Tomasa86c6182006-10-11 01:21:03 -070017 */
Robert P. J. Daybbf2f9f2007-02-17 19:20:16 +010018#define AGGRESSIVE_TEST_
Alex Tomasa86c6182006-10-11 01:21:03 -070019
20/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070021 * With EXTENTS_STATS defined, the number of blocks and extents
22 * are collected in the truncate path. They'll be shown at
23 * umount time.
Alex Tomasa86c6182006-10-11 01:21:03 -070024 */
25#define EXTENTS_STATS__
26
27/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070028 * If CHECK_BINSEARCH is defined, then the results of the binary search
29 * will also be checked by linear search.
Alex Tomasa86c6182006-10-11 01:21:03 -070030 */
31#define CHECK_BINSEARCH__
32
33/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070034 * If EXT_STATS is defined then stats numbers are collected.
35 * These number will be displayed at umount time.
Alex Tomasa86c6182006-10-11 01:21:03 -070036 */
37#define EXT_STATS_
38
39
40/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070041 * ext4_inode has i_block array (60 bytes total).
42 * The first 12 bytes store ext4_extent_header;
43 * the remainder stores an array of ext4_extent.
Darrick J. Wonge6153912012-04-29 18:23:10 -040044 * For non-inode extent blocks, ext4_extent_tail
45 * follows the array.
Alex Tomasa86c6182006-10-11 01:21:03 -070046 */
47
48/*
Darrick J. Wonge6153912012-04-29 18:23:10 -040049 * This is the extent tail on-disk structure.
50 * All other extent structures are 12 bytes long. It turns out that
51 * block_size % 12 >= 4 for at least all powers of 2 greater than 512, which
52 * covers all valid ext4 block sizes. Therefore, this tail structure can be
53 * crammed into the end of the block without having to rebalance the tree.
54 */
55struct ext4_extent_tail {
56 __le32 et_checksum; /* crc32c(uuid+inum+extent_block) */
57};
58
59/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070060 * This is the extent on-disk structure.
61 * It's used at the bottom of the tree.
Alex Tomasa86c6182006-10-11 01:21:03 -070062 */
63struct ext4_extent {
64 __le32 ee_block; /* first logical block extent covers */
65 __le16 ee_len; /* number of blocks covered by extent */
66 __le16 ee_start_hi; /* high 16 bits of physical block */
Aneesh Kumar K.Vb3776112007-10-16 18:38:25 -040067 __le32 ee_start_lo; /* low 32 bits of physical block */
Alex Tomasa86c6182006-10-11 01:21:03 -070068};
69
70/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070071 * This is index on-disk structure.
72 * It's used at all the levels except the bottom.
Alex Tomasa86c6182006-10-11 01:21:03 -070073 */
74struct ext4_extent_idx {
75 __le32 ei_block; /* index covers logical blocks from 'block' */
Aneesh Kumar K.Vd8dd0b42007-10-16 18:38:25 -040076 __le32 ei_leaf_lo; /* pointer to the physical block of the next *
Randy Dunlapd0d856e2006-10-11 01:21:07 -070077 * level. leaf or next index could be there */
Alex Tomasa86c6182006-10-11 01:21:03 -070078 __le16 ei_leaf_hi; /* high 16 bits of physical block */
79 __u16 ei_unused;
80};
81
82/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -070083 * Each block (leaves and indexes), even inode-stored has header.
Alex Tomasa86c6182006-10-11 01:21:03 -070084 */
85struct ext4_extent_header {
86 __le16 eh_magic; /* probably will support different formats */
87 __le16 eh_entries; /* number of valid entries */
88 __le16 eh_max; /* capacity of store in entries */
Randy Dunlapd0d856e2006-10-11 01:21:07 -070089 __le16 eh_depth; /* has tree real underlying blocks? */
Alex Tomasa86c6182006-10-11 01:21:03 -070090 __le32 eh_generation; /* generation of the tree */
91};
92
93#define EXT4_EXT_MAGIC cpu_to_le16(0xf30a)
Theodore Ts'obc890a62018-06-14 12:55:10 -040094#define EXT4_MAX_EXTENT_DEPTH 5
Alex Tomasa86c6182006-10-11 01:21:03 -070095
Darrick J. Wong7ac59902012-04-29 18:37:10 -040096#define EXT4_EXTENT_TAIL_OFFSET(hdr) \
97 (sizeof(struct ext4_extent_header) + \
98 (sizeof(struct ext4_extent) * le16_to_cpu((hdr)->eh_max)))
99
100static inline struct ext4_extent_tail *
101find_ext4_extent_tail(struct ext4_extent_header *eh)
102{
103 return (struct ext4_extent_tail *)(((void *)eh) +
104 EXT4_EXTENT_TAIL_OFFSET(eh));
105}
106
Alex Tomasa86c6182006-10-11 01:21:03 -0700107/*
Randy Dunlapd0d856e2006-10-11 01:21:07 -0700108 * Array of ext4_ext_path contains path to some extent.
109 * Creation/lookup routines use it for traversal/splitting/etc.
110 * Truncate uses it to simulate recursive walking.
Alex Tomasa86c6182006-10-11 01:21:03 -0700111 */
112struct ext4_ext_path {
Alex Tomasf65e6fb2006-10-11 01:21:05 -0700113 ext4_fsblk_t p_block;
Alex Tomasa86c6182006-10-11 01:21:03 -0700114 __u16 p_depth;
Theodore Ts'o10809df82014-09-01 14:40:09 -0400115 __u16 p_maxdepth;
Alex Tomasa86c6182006-10-11 01:21:03 -0700116 struct ext4_extent *p_ext;
117 struct ext4_extent_idx *p_idx;
118 struct ext4_extent_header *p_hdr;
119 struct buffer_head *p_bh;
120};
121
122/*
Eric Whitney9fe67142018-10-01 14:25:08 -0400123 * Used to record a portion of a cluster found at the beginning or end
124 * of an extent while traversing the extent tree during space removal.
125 * A partial cluster may be removed if it does not contain blocks shared
126 * with extents that aren't being deleted (tofree state). Otherwise,
127 * it cannot be removed (nofree state).
128 */
129struct partial_cluster {
130 ext4_fsblk_t pclu; /* physical cluster number */
131 ext4_lblk_t lblk; /* logical block number within logical cluster */
132 enum {initial, tofree, nofree} state;
133};
134
135/*
Alex Tomasa86c6182006-10-11 01:21:03 -0700136 * structure for external API
137 */
138
Eric Sandeen6873fa02008-10-07 00:46:36 -0400139/*
Amit Arora749269f2007-07-18 09:02:56 -0400140 * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
141 * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
142 * MSB of ee_len field in the extent datastructure to signify if this
Lukas Czerner556615d2014-04-20 23:45:47 -0400143 * particular extent is an initialized extent or an unwritten (i.e.
Amit Arora749269f2007-07-18 09:02:56 -0400144 * preallocated).
Lukas Czerner556615d2014-04-20 23:45:47 -0400145 * EXT_UNWRITTEN_MAX_LEN is the maximum number of blocks we can have in an
146 * unwritten extent.
Amit Arora749269f2007-07-18 09:02:56 -0400147 * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
Lukas Czerner556615d2014-04-20 23:45:47 -0400148 * unwritten one. In other words, if MSB of ee_len is set, it is an
149 * unwritten extent with only one special scenario when ee_len = 0x8000.
150 * In this case we can not have an unwritten extent of zero length and
Amit Arora749269f2007-07-18 09:02:56 -0400151 * thus we make it as a special case of initialized extent with 0x8000 length.
152 * This way we get better extent-to-group alignment for initialized extents.
153 * Hence, the maximum number of blocks we can have in an *initialized*
Lukas Czerner556615d2014-04-20 23:45:47 -0400154 * extent is 2^15 (32768) and in an *unwritten* extent is 2^15-1 (32767).
Amit Arora749269f2007-07-18 09:02:56 -0400155 */
156#define EXT_INIT_MAX_LEN (1UL << 15)
Lukas Czerner556615d2014-04-20 23:45:47 -0400157#define EXT_UNWRITTEN_MAX_LEN (EXT_INIT_MAX_LEN - 1)
Suparna Bhattacharya471d4012006-10-11 01:21:06 -0700158
Alex Tomasa86c6182006-10-11 01:21:03 -0700159
160#define EXT_FIRST_EXTENT(__hdr__) \
161 ((struct ext4_extent *) (((char *) (__hdr__)) + \
162 sizeof(struct ext4_extent_header)))
163#define EXT_FIRST_INDEX(__hdr__) \
164 ((struct ext4_extent_idx *) (((char *) (__hdr__)) + \
165 sizeof(struct ext4_extent_header)))
166#define EXT_HAS_FREE_INDEX(__path__) \
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400167 (le16_to_cpu((__path__)->p_hdr->eh_entries) \
168 < le16_to_cpu((__path__)->p_hdr->eh_max))
Alex Tomasa86c6182006-10-11 01:21:03 -0700169#define EXT_LAST_EXTENT(__hdr__) \
170 (EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
171#define EXT_LAST_INDEX(__hdr__) \
172 (EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_entries) - 1)
Harshad Shirwadkarc36a71b2020-04-20 19:39:59 -0700173#define EXT_MAX_EXTENT(__hdr__) \
174 ((le16_to_cpu((__hdr__)->eh_max)) ? \
175 ((EXT_FIRST_EXTENT((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) \
176 : 0)
Alex Tomasa86c6182006-10-11 01:21:03 -0700177#define EXT_MAX_INDEX(__hdr__) \
Harshad Shirwadkarc36a71b2020-04-20 19:39:59 -0700178 ((le16_to_cpu((__hdr__)->eh_max)) ? \
179 ((EXT_FIRST_INDEX((__hdr__)) + le16_to_cpu((__hdr__)->eh_max) - 1)) : 0)
Alex Tomasa86c6182006-10-11 01:21:03 -0700180
181static inline struct ext4_extent_header *ext_inode_hdr(struct inode *inode)
182{
183 return (struct ext4_extent_header *) EXT4_I(inode)->i_data;
184}
185
186static inline struct ext4_extent_header *ext_block_hdr(struct buffer_head *bh)
187{
188 return (struct ext4_extent_header *) bh->b_data;
189}
190
191static inline unsigned short ext_depth(struct inode *inode)
192{
193 return le16_to_cpu(ext_inode_hdr(inode)->eh_depth);
194}
195
Lukas Czerner556615d2014-04-20 23:45:47 -0400196static inline void ext4_ext_mark_unwritten(struct ext4_extent *ext)
Amit Aroraa2df2a62007-07-17 21:42:41 -0400197{
Lukas Czerner556615d2014-04-20 23:45:47 -0400198 /* We can not have an unwritten extent of zero length! */
Amit Arora749269f2007-07-18 09:02:56 -0400199 BUG_ON((le16_to_cpu(ext->ee_len) & ~EXT_INIT_MAX_LEN) == 0);
200 ext->ee_len |= cpu_to_le16(EXT_INIT_MAX_LEN);
Amit Aroraa2df2a62007-07-17 21:42:41 -0400201}
202
Lukas Czerner556615d2014-04-20 23:45:47 -0400203static inline int ext4_ext_is_unwritten(struct ext4_extent *ext)
Amit Aroraa2df2a62007-07-17 21:42:41 -0400204{
Amit Arora749269f2007-07-18 09:02:56 -0400205 /* Extent with ee_len of 0x8000 is treated as an initialized extent */
206 return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
Amit Aroraa2df2a62007-07-17 21:42:41 -0400207}
208
209static inline int ext4_ext_get_actual_len(struct ext4_extent *ext)
210{
Amit Arora749269f2007-07-18 09:02:56 -0400211 return (le16_to_cpu(ext->ee_len) <= EXT_INIT_MAX_LEN ?
212 le16_to_cpu(ext->ee_len) :
213 (le16_to_cpu(ext->ee_len) - EXT_INIT_MAX_LEN));
Amit Aroraa2df2a62007-07-17 21:42:41 -0400214}
215
Mingming Cao00314622009-09-28 15:49:08 -0400216static inline void ext4_ext_mark_initialized(struct ext4_extent *ext)
217{
218 ext->ee_len = cpu_to_le16(ext4_ext_get_actual_len(ext));
219}
220
Theodore Ts'obf89d162010-10-27 21:30:14 -0400221/*
222 * ext4_ext_pblock:
223 * combine low and high parts of physical block number into ext4_fsblk_t
224 */
225static inline ext4_fsblk_t ext4_ext_pblock(struct ext4_extent *ex)
226{
227 ext4_fsblk_t block;
228
229 block = le32_to_cpu(ex->ee_start_lo);
230 block |= ((ext4_fsblk_t) le16_to_cpu(ex->ee_start_hi) << 31) << 1;
231 return block;
232}
233
234/*
235 * ext4_idx_pblock:
236 * combine low and high parts of a leaf physical block number into ext4_fsblk_t
237 */
238static inline ext4_fsblk_t ext4_idx_pblock(struct ext4_extent_idx *ix)
239{
240 ext4_fsblk_t block;
241
242 block = le32_to_cpu(ix->ei_leaf_lo);
243 block |= ((ext4_fsblk_t) le16_to_cpu(ix->ei_leaf_hi) << 31) << 1;
244 return block;
245}
246
247/*
248 * ext4_ext_store_pblock:
249 * stores a large physical block number into an extent struct,
250 * breaking it into parts
251 */
252static inline void ext4_ext_store_pblock(struct ext4_extent *ex,
253 ext4_fsblk_t pb)
254{
255 ex->ee_start_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
256 ex->ee_start_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
257 0xffff);
258}
259
260/*
261 * ext4_idx_store_pblock:
262 * stores a large physical block number into an index struct,
263 * breaking it into parts
264 */
265static inline void ext4_idx_store_pblock(struct ext4_extent_idx *ix,
266 ext4_fsblk_t pb)
267{
268 ix->ei_leaf_lo = cpu_to_le32((unsigned long) (pb & 0xffffffff));
269 ix->ei_leaf_hi = cpu_to_le16((unsigned long) ((pb >> 31) >> 1) &
270 0xffff);
271}
272
Christoph Hellwig3dcf5452008-04-29 18:13:32 -0400273#endif /* _EXT4_EXTENTS */
Alex Tomasa86c6182006-10-11 01:21:03 -0700274