blob: c25dfd1a8a54a0b901e89140252b4df4b5c0e299 [file] [log] [blame]
David Sterbac1d7c512018-04-03 19:23:33 +02001// SPDX-License-Identifier: GPL-2.0
Li Zefana6fa6fa2010-10-25 15:12:26 +08002/*
3 * Copyright (C) 2008 Oracle. All rights reserved.
Li Zefana6fa6fa2010-10-25 15:12:26 +08004 */
5
6#include <linux/kernel.h>
7#include <linux/slab.h>
David Sterba6acafd12017-05-31 17:21:15 +02008#include <linux/mm.h>
Li Zefana6fa6fa2010-10-25 15:12:26 +08009#include <linux/init.h>
10#include <linux/err.h>
11#include <linux/sched.h>
12#include <linux/pagemap.h>
13#include <linux/bio.h>
14#include <linux/lzo.h>
Anand Jaine1ddce72017-05-26 15:44:59 +080015#include <linux/refcount.h>
Li Zefana6fa6fa2010-10-25 15:12:26 +080016#include "compression.h"
Qu Wenruoa6e66e62021-07-26 14:34:55 +080017#include "ctree.h"
Li Zefana6fa6fa2010-10-25 15:12:26 +080018
19#define LZO_LEN 4
20
Qu Wenruo2a1f7c02018-05-17 13:10:01 +080021/*
22 * Btrfs LZO compression format
23 *
24 * Regular and inlined LZO compressed data extents consist of:
25 *
26 * 1. Header
27 * Fixed size. LZO_LEN (4) bytes long, LE32.
28 * Records the total size (including the header) of compressed data.
29 *
30 * 2. Segment(s)
Andrea Gelmini52042d82018-11-28 12:05:13 +010031 * Variable size. Each segment includes one segment header, followed by data
Qu Wenruo2a1f7c02018-05-17 13:10:01 +080032 * payload.
33 * One regular LZO compressed extent can have one or more segments.
34 * For inlined LZO compressed extent, only one segment is allowed.
35 * One segment represents at most one page of uncompressed data.
36 *
37 * 2.1 Segment header
38 * Fixed size. LZO_LEN (4) bytes long, LE32.
39 * Records the total size of the segment (not including the header).
40 * Segment header never crosses page boundary, thus it's possible to
41 * have at most 3 padding zeros at the end of the page.
42 *
43 * 2.2 Data Payload
44 * Variable size. Size up limit should be lzo1x_worst_compress(PAGE_SIZE)
45 * which is 4419 for a 4KiB page.
46 *
47 * Example:
48 * Page 1:
49 * 0 0x2 0x4 0x6 0x8 0xa 0xc 0xe 0x10
50 * 0x0000 | Header | SegHdr 01 | Data payload 01 ... |
51 * ...
52 * 0x0ff0 | SegHdr N | Data payload N ... |00|
53 * ^^ padding zeros
54 * Page 2:
55 * 0x1000 | SegHdr N+1| Data payload N+1 ... |
56 */
57
Li Zefana6fa6fa2010-10-25 15:12:26 +080058struct workspace {
59 void *mem;
Jie Liu3fb40372013-06-06 13:38:50 +000060 void *buf; /* where decompressed data goes */
61 void *cbuf; /* where compressed data goes */
Li Zefana6fa6fa2010-10-25 15:12:26 +080062 struct list_head list;
63};
64
Dennis Zhou92ee55302019-02-04 15:20:03 -050065static struct workspace_manager wsm;
66
David Sterbad20f3952019-10-04 02:21:48 +020067void lzo_free_workspace(struct list_head *ws)
Li Zefana6fa6fa2010-10-25 15:12:26 +080068{
69 struct workspace *workspace = list_entry(ws, struct workspace, list);
70
David Sterba6acafd12017-05-31 17:21:15 +020071 kvfree(workspace->buf);
72 kvfree(workspace->cbuf);
73 kvfree(workspace->mem);
Li Zefana6fa6fa2010-10-25 15:12:26 +080074 kfree(workspace);
75}
76
David Sterbad20f3952019-10-04 02:21:48 +020077struct list_head *lzo_alloc_workspace(unsigned int level)
Li Zefana6fa6fa2010-10-25 15:12:26 +080078{
79 struct workspace *workspace;
80
David Sterba389a6cf2017-05-31 17:21:15 +020081 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
Li Zefana6fa6fa2010-10-25 15:12:26 +080082 if (!workspace)
83 return ERR_PTR(-ENOMEM);
84
David Sterba6acafd12017-05-31 17:21:15 +020085 workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
86 workspace->buf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL);
87 workspace->cbuf = kvmalloc(lzo1x_worst_compress(PAGE_SIZE), GFP_KERNEL);
Li Zefana6fa6fa2010-10-25 15:12:26 +080088 if (!workspace->mem || !workspace->buf || !workspace->cbuf)
89 goto fail;
90
91 INIT_LIST_HEAD(&workspace->list);
92
93 return &workspace->list;
94fail:
95 lzo_free_workspace(&workspace->list);
96 return ERR_PTR(-ENOMEM);
97}
98
99static inline void write_compress_length(char *buf, size_t len)
100{
101 __le32 dlen;
102
103 dlen = cpu_to_le32(len);
104 memcpy(buf, &dlen, LZO_LEN);
105}
106
David Sterba14a33572017-02-14 17:58:04 +0100107static inline size_t read_compress_length(const char *buf)
Li Zefana6fa6fa2010-10-25 15:12:26 +0800108{
109 __le32 dlen;
110
111 memcpy(&dlen, buf, LZO_LEN);
112 return le32_to_cpu(dlen);
113}
114
David Sterbac4bf6652019-10-01 22:38:34 +0200115int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
116 u64 start, struct page **pages, unsigned long *out_pages,
117 unsigned long *total_in, unsigned long *total_out)
Li Zefana6fa6fa2010-10-25 15:12:26 +0800118{
119 struct workspace *workspace = list_entry(ws, struct workspace, list);
120 int ret = 0;
121 char *data_in;
Ira Weiny58c1a352021-02-16 18:48:23 -0800122 char *cpage_out, *sizes_ptr;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800123 int nr_pages = 0;
124 struct page *in_page = NULL;
125 struct page *out_page = NULL;
126 unsigned long bytes_left;
David Sterba38c31462017-02-14 19:04:07 +0100127 unsigned long len = *total_out;
David Sterba4d3a8002017-02-14 19:04:07 +0100128 unsigned long nr_dest_pages = *out_pages;
David Sterbae5d74902017-02-14 19:45:05 +0100129 const unsigned long max_out = nr_dest_pages * PAGE_SIZE;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800130 size_t in_len;
131 size_t out_len;
132 char *buf;
133 unsigned long tot_in = 0;
134 unsigned long tot_out = 0;
135 unsigned long pg_bytes_left;
136 unsigned long out_offset;
137 unsigned long bytes;
138
139 *out_pages = 0;
140 *total_out = 0;
141 *total_in = 0;
142
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300143 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
David Sterba8c945d32021-06-14 22:25:53 +0200144 data_in = page_address(in_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800145
146 /*
147 * store the size of all chunks of compressed data in
148 * the first 4 bytes
149 */
David Sterbab0ee5e12021-06-14 22:22:22 +0200150 out_page = alloc_page(GFP_NOFS);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800151 if (out_page == NULL) {
152 ret = -ENOMEM;
153 goto out;
154 }
David Sterba8c945d32021-06-14 22:25:53 +0200155 cpage_out = page_address(out_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800156 out_offset = LZO_LEN;
157 tot_out = LZO_LEN;
158 pages[0] = out_page;
159 nr_pages = 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300160 pg_bytes_left = PAGE_SIZE - LZO_LEN;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800161
162 /* compress at most one page of data each time */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300163 in_len = min(len, PAGE_SIZE);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800164 while (tot_in < len) {
165 ret = lzo1x_1_compress(data_in, in_len, workspace->cbuf,
166 &out_len, workspace->mem);
167 if (ret != LZO_E_OK) {
Timofey Titovets036b0212017-05-25 21:12:19 +0300168 pr_debug("BTRFS: lzo in loop returned %d\n",
Li Zefana6fa6fa2010-10-25 15:12:26 +0800169 ret);
Zach Brown60e19752014-05-09 17:15:08 -0400170 ret = -EIO;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800171 goto out;
172 }
173
174 /* store the size of this chunk of compressed data */
175 write_compress_length(cpage_out + out_offset, out_len);
176 tot_out += LZO_LEN;
177 out_offset += LZO_LEN;
178 pg_bytes_left -= LZO_LEN;
179
180 tot_in += in_len;
181 tot_out += out_len;
182
183 /* copy bytes from the working buffer into the pages */
184 buf = workspace->cbuf;
185 while (out_len) {
186 bytes = min_t(unsigned long, pg_bytes_left, out_len);
187
188 memcpy(cpage_out + out_offset, buf, bytes);
189
190 out_len -= bytes;
191 pg_bytes_left -= bytes;
192 buf += bytes;
193 out_offset += bytes;
194
195 /*
196 * we need another page for writing out.
197 *
198 * Note if there's less than 4 bytes left, we just
199 * skip to a new page.
200 */
201 if ((out_len == 0 && pg_bytes_left < LZO_LEN) ||
202 pg_bytes_left == 0) {
203 if (pg_bytes_left) {
204 memset(cpage_out + out_offset, 0,
205 pg_bytes_left);
206 tot_out += pg_bytes_left;
207 }
208
209 /* we're done, don't allocate new page */
210 if (out_len == 0 && tot_in >= len)
211 break;
212
Li Zefana6fa6fa2010-10-25 15:12:26 +0800213 if (nr_pages == nr_dest_pages) {
214 out_page = NULL;
Zach Brown60e19752014-05-09 17:15:08 -0400215 ret = -E2BIG;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800216 goto out;
217 }
218
David Sterbab0ee5e12021-06-14 22:22:22 +0200219 out_page = alloc_page(GFP_NOFS);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800220 if (out_page == NULL) {
221 ret = -ENOMEM;
222 goto out;
223 }
David Sterba8c945d32021-06-14 22:25:53 +0200224 cpage_out = page_address(out_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800225 pages[nr_pages++] = out_page;
226
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300227 pg_bytes_left = PAGE_SIZE;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800228 out_offset = 0;
229 }
230 }
231
232 /* we're making it bigger, give up */
Stefan Agner59516f62013-07-01 20:33:39 +0200233 if (tot_in > 8192 && tot_in < tot_out) {
Zach Brown60e19752014-05-09 17:15:08 -0400234 ret = -E2BIG;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800235 goto out;
Stefan Agner59516f62013-07-01 20:33:39 +0200236 }
Li Zefana6fa6fa2010-10-25 15:12:26 +0800237
238 /* we're all done */
239 if (tot_in >= len)
240 break;
241
242 if (tot_out > max_out)
243 break;
244
245 bytes_left = len - tot_in;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300246 put_page(in_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800247
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300248 start += PAGE_SIZE;
249 in_page = find_get_page(mapping, start >> PAGE_SHIFT);
David Sterba8c945d32021-06-14 22:25:53 +0200250 data_in = page_address(in_page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300251 in_len = min(bytes_left, PAGE_SIZE);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800252 }
253
Timofey Titovets1e9d7292017-05-30 02:18:04 +0300254 if (tot_out >= tot_in) {
255 ret = -E2BIG;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800256 goto out;
Timofey Titovets1e9d7292017-05-30 02:18:04 +0300257 }
Li Zefana6fa6fa2010-10-25 15:12:26 +0800258
259 /* store the size of all chunks of compressed data */
David Sterba8c945d32021-06-14 22:25:53 +0200260 sizes_ptr = page_address(pages[0]);
Ira Weiny58c1a352021-02-16 18:48:23 -0800261 write_compress_length(sizes_ptr, tot_out);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800262
263 ret = 0;
264 *total_out = tot_out;
265 *total_in = tot_in;
266out:
267 *out_pages = nr_pages;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800268
David Sterba8c945d32021-06-14 22:25:53 +0200269 if (in_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300270 put_page(in_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800271
272 return ret;
273}
274
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800275/*
276 * Copy the compressed segment payload into @dest.
277 *
278 * For the payload there will be no padding, just need to do page switching.
279 */
280static void copy_compressed_segment(struct compressed_bio *cb,
281 char *dest, u32 len, u32 *cur_in)
282{
283 u32 orig_in = *cur_in;
284
285 while (*cur_in < orig_in + len) {
286 struct page *cur_page;
287 u32 copy_len = min_t(u32, PAGE_SIZE - offset_in_page(*cur_in),
288 orig_in + len - *cur_in);
289
290 ASSERT(copy_len);
291 cur_page = cb->compressed_pages[*cur_in / PAGE_SIZE];
292
293 memcpy(dest + *cur_in - orig_in,
294 page_address(cur_page) + offset_in_page(*cur_in),
295 copy_len);
296
297 *cur_in += copy_len;
298 }
299}
300
David Sterbac4bf6652019-10-01 22:38:34 +0200301int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
Li Zefana6fa6fa2010-10-25 15:12:26 +0800302{
303 struct workspace *workspace = list_entry(ws, struct workspace, list);
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800304 const struct btrfs_fs_info *fs_info = btrfs_sb(cb->inode->i_sb);
305 const u32 sectorsize = fs_info->sectorsize;
306 int ret;
307 /* Compressed data length, can be unaligned */
308 u32 len_in;
309 /* Offset inside the compressed data */
310 u32 cur_in = 0;
311 /* Bytes decompressed so far */
312 u32 cur_out = 0;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800313
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800314 len_in = read_compress_length(page_address(cb->compressed_pages[0]));
315 cur_in += LZO_LEN;
316
Qu Wenruo314bfa42018-05-15 14:57:51 +0800317 /*
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800318 * LZO header length check
Qu Wenruo314bfa42018-05-15 14:57:51 +0800319 *
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800320 * The total length should not exceed the maximum extent length,
321 * and all sectors should be used.
322 * If this happens, it means the compressed extent is corrupted.
Qu Wenruo314bfa42018-05-15 14:57:51 +0800323 */
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800324 if (len_in > min_t(size_t, BTRFS_MAX_COMPRESSED, cb->compressed_len) ||
325 round_up(len_in, sectorsize) < cb->compressed_len) {
326 btrfs_err(fs_info,
327 "invalid lzo header, lzo len %u compressed len %u",
328 len_in, cb->compressed_len);
329 return -EUCLEAN;
Qu Wenruo314bfa42018-05-15 14:57:51 +0800330 }
Li Zefana6fa6fa2010-10-25 15:12:26 +0800331
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800332 /* Go through each lzo segment */
333 while (cur_in < len_in) {
334 struct page *cur_page;
335 /* Length of the compressed segment */
336 u32 seg_len;
337 u32 sector_bytes_left;
338 size_t out_len = lzo1x_worst_compress(sectorsize);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800339
Qu Wenruo314bfa42018-05-15 14:57:51 +0800340 /*
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800341 * We should always have enough space for one segment header
342 * inside current sector.
Qu Wenruo314bfa42018-05-15 14:57:51 +0800343 */
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800344 ASSERT(cur_in / sectorsize ==
345 (cur_in + LZO_LEN - 1) / sectorsize);
346 cur_page = cb->compressed_pages[cur_in / PAGE_SIZE];
347 ASSERT(cur_page);
348 seg_len = read_compress_length(page_address(cur_page) +
349 offset_in_page(cur_in));
350 cur_in += LZO_LEN;
Qu Wenruo314bfa42018-05-15 14:57:51 +0800351
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800352 /* Copy the compressed segment payload into workspace */
353 copy_compressed_segment(cb, workspace->cbuf, seg_len, &cur_in);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800354
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800355 /* Decompress the data */
356 ret = lzo1x_decompress_safe(workspace->cbuf, seg_len,
357 workspace->buf, &out_len);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800358 if (ret != LZO_E_OK) {
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800359 btrfs_err(fs_info, "failed to decompress");
Zach Brown60e19752014-05-09 17:15:08 -0400360 ret = -EIO;
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800361 goto out;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800362 }
363
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800364 /* Copy the data into inode pages */
365 ret = btrfs_decompress_buf2page(workspace->buf, out_len, cb, cur_out);
366 cur_out += out_len;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800367
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800368 /* All data read, exit */
369 if (ret == 0)
370 goto out;
371 ret = 0;
372
373 /* Check if the sector has enough space for a segment header */
374 sector_bytes_left = sectorsize - (cur_in % sectorsize);
375 if (sector_bytes_left >= LZO_LEN)
376 continue;
377
378 /* Skip the padding zeros */
379 cur_in += sector_bytes_left;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800380 }
Qu Wenruoa6e66e62021-07-26 14:34:55 +0800381out:
Chris Mason2f19cad2014-11-30 08:56:33 -0500382 if (!ret)
Qu Wenruo1c3dc172021-07-05 10:00:58 +0800383 zero_fill_bio(cb->orig_bio);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800384 return ret;
385}
386
David Sterbac4bf6652019-10-01 22:38:34 +0200387int lzo_decompress(struct list_head *ws, unsigned char *data_in,
388 struct page *dest_page, unsigned long start_byte, size_t srclen,
389 size_t destlen)
Li Zefana6fa6fa2010-10-25 15:12:26 +0800390{
391 struct workspace *workspace = list_entry(ws, struct workspace, list);
392 size_t in_len;
393 size_t out_len;
Qu Wenruode885e32018-05-17 14:10:29 +0800394 size_t max_segment_len = lzo1x_worst_compress(PAGE_SIZE);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800395 int ret = 0;
396 char *kaddr;
397 unsigned long bytes;
398
Qu Wenruode885e32018-05-17 14:10:29 +0800399 if (srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2)
400 return -EUCLEAN;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800401
Qu Wenruode885e32018-05-17 14:10:29 +0800402 in_len = read_compress_length(data_in);
403 if (in_len != srclen)
404 return -EUCLEAN;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800405 data_in += LZO_LEN;
406
407 in_len = read_compress_length(data_in);
Qu Wenruode885e32018-05-17 14:10:29 +0800408 if (in_len != srclen - LZO_LEN * 2) {
409 ret = -EUCLEAN;
410 goto out;
411 }
Li Zefana6fa6fa2010-10-25 15:12:26 +0800412 data_in += LZO_LEN;
413
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300414 out_len = PAGE_SIZE;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800415 ret = lzo1x_decompress_safe(data_in, in_len, workspace->buf, &out_len);
416 if (ret != LZO_E_OK) {
Jeff Mahoney62e85572016-09-20 10:05:01 -0400417 pr_warn("BTRFS: decompress failed!\n");
Zach Brown60e19752014-05-09 17:15:08 -0400418 ret = -EIO;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800419 goto out;
420 }
421
422 if (out_len < start_byte) {
Zach Brown60e19752014-05-09 17:15:08 -0400423 ret = -EIO;
Li Zefana6fa6fa2010-10-25 15:12:26 +0800424 goto out;
425 }
426
Chris Mason2f19cad2014-11-30 08:56:33 -0500427 /*
428 * the caller is already checking against PAGE_SIZE, but lets
429 * move this check closer to the memcpy/memset
430 */
431 destlen = min_t(unsigned long, destlen, PAGE_SIZE);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800432 bytes = min_t(unsigned long, destlen, out_len - start_byte);
433
David Sterba8c945d32021-06-14 22:25:53 +0200434 kaddr = page_address(dest_page);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800435 memcpy(kaddr, workspace->buf + start_byte, bytes);
Chris Mason2f19cad2014-11-30 08:56:33 -0500436
437 /*
438 * btrfs_getblock is doing a zero on the tail of the page too,
439 * but this will cover anything missing from the decompressed
440 * data.
441 */
442 if (bytes < destlen)
443 memset(kaddr+bytes, 0, destlen-bytes);
Li Zefana6fa6fa2010-10-25 15:12:26 +0800444out:
445 return ret;
446}
447
David Sterbae8c9f182015-01-02 18:23:10 +0100448const struct btrfs_compress_op btrfs_lzo_compress = {
David Sterbabe9510452019-10-02 00:53:31 +0200449 .workspace_manager = &wsm,
David Sterbae18333a2019-08-09 16:25:34 +0200450 .max_level = 1,
451 .default_level = 1,
Li Zefana6fa6fa2010-10-25 15:12:26 +0800452};