1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2008 Oracle. All rights reserved.
4 */
5
6#include <linux/kernel.h>
7#include <linux/slab.h>
8#include <linux/mm.h>
9#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>
15#include <linux/refcount.h>
16#include "messages.h"
17#include "compression.h"
18#include "ctree.h"
19#include "super.h"
20#include "btrfs_inode.h"
21
22#define LZO_LEN 4
23
24/*
25 * Btrfs LZO compression format
26 *
27 * Regular and inlined LZO compressed data extents consist of:
28 *
29 * 1. Header
30 * Fixed size. LZO_LEN (4) bytes long, LE32.
31 * Records the total size (including the header) of compressed data.
32 *
33 * 2. Segment(s)
34 * Variable size. Each segment includes one segment header, followed by data
35 * payload.
36 * One regular LZO compressed extent can have one or more segments.
37 * For inlined LZO compressed extent, only one segment is allowed.
38 * One segment represents at most one sector of uncompressed data.
39 *
40 * 2.1 Segment header
41 * Fixed size. LZO_LEN (4) bytes long, LE32.
42 * Records the total size of the segment (not including the header).
43 * Segment header never crosses sector boundary, thus it's possible to
44 * have at most 3 padding zeros at the end of the sector.
45 *
46 * 2.2 Data Payload
47 * Variable size. Size up limit should be lzo1x_worst_compress(sectorsize)
48 * which is 4419 for a 4KiB sectorsize.
49 *
50 * Example with 4K sectorsize:
51 * Page 1:
52 * 0 0x2 0x4 0x6 0x8 0xa 0xc 0xe 0x10
53 * 0x0000 | Header | SegHdr 01 | Data payload 01 ... |
54 * ...
55 * 0x0ff0 | SegHdr N | Data payload N ... |00|
56 * ^^ padding zeros
57 * Page 2:
58 * 0x1000 | SegHdr N+1| Data payload N+1 ... |
59 */
60
61struct workspace {
62 void *mem;
63 void *buf; /* where decompressed data goes */
64 void *cbuf; /* where compressed data goes */
65 struct list_head list;
66};
67
68static u32 workspace_buf_length(const struct btrfs_fs_info *fs_info)
69{
70 return lzo1x_worst_compress(fs_info->sectorsize);
71}
72static u32 workspace_cbuf_length(const struct btrfs_fs_info *fs_info)
73{
74 return lzo1x_worst_compress(fs_info->sectorsize);
75}
76
77void lzo_free_workspace(struct list_head *ws)
78{
79 struct workspace *workspace = list_entry(ws, struct workspace, list);
80
81 kvfree(addr: workspace->buf);
82 kvfree(addr: workspace->cbuf);
83 kvfree(addr: workspace->mem);
84 kfree(objp: workspace);
85}
86
87struct list_head *lzo_alloc_workspace(struct btrfs_fs_info *fs_info)
88{
89 struct workspace *workspace;
90
91 workspace = kzalloc(sizeof(*workspace), GFP_KERNEL);
92 if (!workspace)
93 return ERR_PTR(error: -ENOMEM);
94
95 workspace->mem = kvmalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL | __GFP_NOWARN);
96 workspace->buf = kvmalloc(workspace_buf_length(fs_info), GFP_KERNEL | __GFP_NOWARN);
97 workspace->cbuf = kvmalloc(workspace_cbuf_length(fs_info), GFP_KERNEL | __GFP_NOWARN);
98 if (!workspace->mem || !workspace->buf || !workspace->cbuf)
99 goto fail;
100
101 INIT_LIST_HEAD(list: &workspace->list);
102
103 return &workspace->list;
104fail:
105 lzo_free_workspace(ws: &workspace->list);
106 return ERR_PTR(error: -ENOMEM);
107}
108
109static inline void write_compress_length(char *buf, size_t len)
110{
111 __le32 dlen;
112
113 dlen = cpu_to_le32(len);
114 memcpy(buf, &dlen, LZO_LEN);
115}
116
117static inline size_t read_compress_length(const char *buf)
118{
119 __le32 dlen;
120
121 memcpy(&dlen, buf, LZO_LEN);
122 return le32_to_cpu(dlen);
123}
124
125/*
126 * Will do:
127 *
128 * - Write a segment header into the destination
129 * - Copy the compressed buffer into the destination
130 * - Make sure we have enough space in the last sector to fit a segment header
131 * If not, we will pad at most (LZO_LEN (4)) - 1 bytes of zeros.
132 *
133 * Will allocate new pages when needed.
134 */
135static int copy_compressed_data_to_page(struct btrfs_fs_info *fs_info,
136 char *compressed_data,
137 size_t compressed_size,
138 struct folio **out_folios,
139 unsigned long max_nr_folio,
140 u32 *cur_out)
141{
142 const u32 sectorsize = fs_info->sectorsize;
143 const u32 min_folio_shift = PAGE_SHIFT + fs_info->block_min_order;
144 u32 sector_bytes_left;
145 u32 orig_out;
146 struct folio *cur_folio;
147 char *kaddr;
148
149 if ((*cur_out >> min_folio_shift) >= max_nr_folio)
150 return -E2BIG;
151
152 /*
153 * We never allow a segment header crossing sector boundary, previous
154 * run should ensure we have enough space left inside the sector.
155 */
156 ASSERT((*cur_out / sectorsize) == (*cur_out + LZO_LEN - 1) / sectorsize);
157
158 cur_folio = out_folios[*cur_out >> min_folio_shift];
159 /* Allocate a new page */
160 if (!cur_folio) {
161 cur_folio = btrfs_alloc_compr_folio(fs_info);
162 if (!cur_folio)
163 return -ENOMEM;
164 out_folios[*cur_out >> min_folio_shift] = cur_folio;
165 }
166
167 kaddr = kmap_local_folio(folio: cur_folio, offset_in_folio(cur_folio, *cur_out));
168 write_compress_length(buf: kaddr, len: compressed_size);
169 *cur_out += LZO_LEN;
170
171 orig_out = *cur_out;
172
173 /* Copy compressed data */
174 while (*cur_out - orig_out < compressed_size) {
175 u32 copy_len = min_t(u32, sectorsize - *cur_out % sectorsize,
176 orig_out + compressed_size - *cur_out);
177
178 kunmap_local(kaddr);
179
180 if ((*cur_out >> min_folio_shift) >= max_nr_folio)
181 return -E2BIG;
182
183 cur_folio = out_folios[*cur_out >> min_folio_shift];
184 /* Allocate a new page */
185 if (!cur_folio) {
186 cur_folio = btrfs_alloc_compr_folio(fs_info);
187 if (!cur_folio)
188 return -ENOMEM;
189 out_folios[*cur_out >> min_folio_shift] = cur_folio;
190 }
191 kaddr = kmap_local_folio(folio: cur_folio, offset: 0);
192
193 memcpy(kaddr + offset_in_folio(cur_folio, *cur_out),
194 compressed_data + *cur_out - orig_out, copy_len);
195
196 *cur_out += copy_len;
197 }
198
199 /*
200 * Check if we can fit the next segment header into the remaining space
201 * of the sector.
202 */
203 sector_bytes_left = round_up(*cur_out, sectorsize) - *cur_out;
204 if (sector_bytes_left >= LZO_LEN || sector_bytes_left == 0)
205 goto out;
206
207 /* The remaining size is not enough, pad it with zeros */
208 memset(kaddr + offset_in_page(*cur_out), 0,
209 sector_bytes_left);
210 *cur_out += sector_bytes_left;
211
212out:
213 kunmap_local(kaddr);
214 return 0;
215}
216
217int lzo_compress_folios(struct list_head *ws, struct btrfs_inode *inode,
218 u64 start, struct folio **folios, unsigned long *out_folios,
219 unsigned long *total_in, unsigned long *total_out)
220{
221 struct btrfs_fs_info *fs_info = inode->root->fs_info;
222 struct workspace *workspace = list_entry(ws, struct workspace, list);
223 const u32 sectorsize = fs_info->sectorsize;
224 const u32 min_folio_size = btrfs_min_folio_size(fs_info);
225 struct address_space *mapping = inode->vfs_inode.i_mapping;
226 struct folio *folio_in = NULL;
227 char *sizes_ptr;
228 const unsigned long max_nr_folio = *out_folios;
229 int ret = 0;
230 /* Points to the file offset of input data */
231 u64 cur_in = start;
232 /* Points to the current output byte */
233 u32 cur_out = 0;
234 u32 len = *total_out;
235
236 ASSERT(max_nr_folio > 0);
237 *out_folios = 0;
238 *total_out = 0;
239 *total_in = 0;
240
241 /*
242 * Skip the header for now, we will later come back and write the total
243 * compressed size
244 */
245 cur_out += LZO_LEN;
246 while (cur_in < start + len) {
247 char *data_in;
248 const u32 sectorsize_mask = sectorsize - 1;
249 u32 sector_off = (cur_in - start) & sectorsize_mask;
250 u32 in_len;
251 size_t out_len;
252
253 /* Get the input page first */
254 if (!folio_in) {
255 ret = btrfs_compress_filemap_get_folio(mapping, start: cur_in, in_folio_ret: &folio_in);
256 if (ret < 0)
257 goto out;
258 }
259
260 /* Compress at most one sector of data each time */
261 in_len = min_t(u32, start + len - cur_in, sectorsize - sector_off);
262 ASSERT(in_len);
263 data_in = kmap_local_folio(folio: folio_in, offset_in_folio(folio_in, cur_in));
264 ret = lzo1x_1_compress(src: data_in, src_len: in_len,
265 dst: workspace->cbuf, dst_len: &out_len,
266 wrkmem: workspace->mem);
267 kunmap_local(data_in);
268 if (unlikely(ret < 0)) {
269 /* lzo1x_1_compress never fails. */
270 ret = -EIO;
271 goto out;
272 }
273
274 ret = copy_compressed_data_to_page(fs_info, compressed_data: workspace->cbuf, compressed_size: out_len,
275 out_folios: folios, max_nr_folio,
276 cur_out: &cur_out);
277 if (ret < 0)
278 goto out;
279
280 cur_in += in_len;
281
282 /*
283 * Check if we're making it bigger after two sectors. And if
284 * it is so, give up.
285 */
286 if (cur_in - start > sectorsize * 2 && cur_in - start < cur_out) {
287 ret = -E2BIG;
288 goto out;
289 }
290
291 /* Check if we have reached folio boundary. */
292 if (IS_ALIGNED(cur_in, min_folio_size)) {
293 folio_put(folio: folio_in);
294 folio_in = NULL;
295 }
296 }
297
298 /* Store the size of all chunks of compressed data */
299 sizes_ptr = kmap_local_folio(folio: folios[0], offset: 0);
300 write_compress_length(buf: sizes_ptr, len: cur_out);
301 kunmap_local(sizes_ptr);
302
303 ret = 0;
304 *total_out = cur_out;
305 *total_in = cur_in - start;
306out:
307 if (folio_in)
308 folio_put(folio: folio_in);
309 *out_folios = DIV_ROUND_UP(cur_out, min_folio_size);
310 return ret;
311}
312
313/*
314 * Copy the compressed segment payload into @dest.
315 *
316 * For the payload there will be no padding, just need to do page switching.
317 */
318static void copy_compressed_segment(struct compressed_bio *cb,
319 char *dest, u32 len, u32 *cur_in)
320{
321 struct btrfs_fs_info *fs_info = cb_to_fs_info(cb);
322 const u32 min_folio_shift = PAGE_SHIFT + fs_info->block_min_order;
323 u32 orig_in = *cur_in;
324
325 while (*cur_in < orig_in + len) {
326 struct folio *cur_folio = cb->compressed_folios[*cur_in >> min_folio_shift];
327 u32 copy_len = min_t(u32, orig_in + len - *cur_in,
328 folio_size(cur_folio) - offset_in_folio(cur_folio, *cur_in));
329
330 ASSERT(copy_len);
331
332 memcpy_from_folio(to: dest + *cur_in - orig_in, folio: cur_folio,
333 offset_in_folio(cur_folio, *cur_in), len: copy_len);
334
335 *cur_in += copy_len;
336 }
337}
338
339int lzo_decompress_bio(struct list_head *ws, struct compressed_bio *cb)
340{
341 struct workspace *workspace = list_entry(ws, struct workspace, list);
342 const struct btrfs_fs_info *fs_info = cb->bbio.inode->root->fs_info;
343 const u32 sectorsize = fs_info->sectorsize;
344 const u32 min_folio_shift = PAGE_SHIFT + fs_info->block_min_order;
345 char *kaddr;
346 int ret;
347 /* Compressed data length, can be unaligned */
348 u32 len_in;
349 /* Offset inside the compressed data */
350 u32 cur_in = 0;
351 /* Bytes decompressed so far */
352 u32 cur_out = 0;
353
354 kaddr = kmap_local_folio(folio: cb->compressed_folios[0], offset: 0);
355 len_in = read_compress_length(buf: kaddr);
356 kunmap_local(kaddr);
357 cur_in += LZO_LEN;
358
359 /*
360 * LZO header length check
361 *
362 * The total length should not exceed the maximum extent length,
363 * and all sectors should be used.
364 * If this happens, it means the compressed extent is corrupted.
365 */
366 if (unlikely(len_in > min_t(size_t, BTRFS_MAX_COMPRESSED, cb->compressed_len) ||
367 round_up(len_in, sectorsize) < cb->compressed_len)) {
368 struct btrfs_inode *inode = cb->bbio.inode;
369
370 btrfs_err(fs_info,
371"lzo header invalid, root %llu inode %llu offset %llu lzo len %u compressed len %u",
372 btrfs_root_id(inode->root), btrfs_ino(inode),
373 cb->start, len_in, cb->compressed_len);
374 return -EUCLEAN;
375 }
376
377 /* Go through each lzo segment */
378 while (cur_in < len_in) {
379 struct folio *cur_folio;
380 /* Length of the compressed segment */
381 u32 seg_len;
382 u32 sector_bytes_left;
383 size_t out_len = lzo1x_worst_compress(sectorsize);
384
385 /*
386 * We should always have enough space for one segment header
387 * inside current sector.
388 */
389 ASSERT(cur_in / sectorsize ==
390 (cur_in + LZO_LEN - 1) / sectorsize);
391 cur_folio = cb->compressed_folios[cur_in >> min_folio_shift];
392 ASSERT(cur_folio);
393 kaddr = kmap_local_folio(folio: cur_folio, offset: 0);
394 seg_len = read_compress_length(buf: kaddr + offset_in_folio(cur_folio, cur_in));
395 kunmap_local(kaddr);
396 cur_in += LZO_LEN;
397
398 if (unlikely(seg_len > workspace_cbuf_length(fs_info))) {
399 struct btrfs_inode *inode = cb->bbio.inode;
400
401 /*
402 * seg_len shouldn't be larger than we have allocated
403 * for workspace->cbuf
404 */
405 btrfs_err(fs_info,
406 "lzo segment too big, root %llu inode %llu offset %llu len %u",
407 btrfs_root_id(inode->root), btrfs_ino(inode),
408 cb->start, seg_len);
409 return -EIO;
410 }
411
412 /* Copy the compressed segment payload into workspace */
413 copy_compressed_segment(cb, dest: workspace->cbuf, len: seg_len, cur_in: &cur_in);
414
415 /* Decompress the data */
416 ret = lzo1x_decompress_safe(src: workspace->cbuf, src_len: seg_len,
417 dst: workspace->buf, dst_len: &out_len);
418 if (unlikely(ret != LZO_E_OK)) {
419 struct btrfs_inode *inode = cb->bbio.inode;
420
421 btrfs_err(fs_info,
422 "lzo decompression failed, error %d root %llu inode %llu offset %llu",
423 ret, btrfs_root_id(inode->root), btrfs_ino(inode),
424 cb->start);
425 return -EIO;
426 }
427
428 /* Copy the data into inode pages */
429 ret = btrfs_decompress_buf2page(buf: workspace->buf, buf_len: out_len, cb, decompressed: cur_out);
430 cur_out += out_len;
431
432 /* All data read, exit */
433 if (ret == 0)
434 return 0;
435 ret = 0;
436
437 /* Check if the sector has enough space for a segment header */
438 sector_bytes_left = sectorsize - (cur_in % sectorsize);
439 if (sector_bytes_left >= LZO_LEN)
440 continue;
441
442 /* Skip the padding zeros */
443 cur_in += sector_bytes_left;
444 }
445
446 return 0;
447}
448
449int lzo_decompress(struct list_head *ws, const u8 *data_in,
450 struct folio *dest_folio, unsigned long dest_pgoff, size_t srclen,
451 size_t destlen)
452{
453 struct workspace *workspace = list_entry(ws, struct workspace, list);
454 struct btrfs_fs_info *fs_info = folio_to_fs_info(dest_folio);
455 const u32 sectorsize = fs_info->sectorsize;
456 size_t in_len;
457 size_t out_len;
458 size_t max_segment_len = workspace_buf_length(fs_info);
459 int ret = 0;
460
461 if (unlikely(srclen < LZO_LEN || srclen > max_segment_len + LZO_LEN * 2))
462 return -EUCLEAN;
463
464 in_len = read_compress_length(buf: data_in);
465 if (unlikely(in_len != srclen))
466 return -EUCLEAN;
467 data_in += LZO_LEN;
468
469 in_len = read_compress_length(buf: data_in);
470 if (unlikely(in_len != srclen - LZO_LEN * 2)) {
471 ret = -EUCLEAN;
472 goto out;
473 }
474 data_in += LZO_LEN;
475
476 out_len = sectorsize;
477 ret = lzo1x_decompress_safe(src: data_in, src_len: in_len, dst: workspace->buf, dst_len: &out_len);
478 if (unlikely(ret != LZO_E_OK)) {
479 struct btrfs_inode *inode = folio_to_inode(dest_folio);
480
481 btrfs_err(fs_info,
482 "lzo decompression failed, error %d root %llu inode %llu offset %llu",
483 ret, btrfs_root_id(inode->root), btrfs_ino(inode),
484 folio_pos(dest_folio));
485 ret = -EIO;
486 goto out;
487 }
488
489 ASSERT(out_len <= sectorsize);
490 memcpy_to_folio(folio: dest_folio, offset: dest_pgoff, from: workspace->buf, len: out_len);
491 /* Early end, considered as an error. */
492 if (unlikely(out_len < destlen)) {
493 ret = -EIO;
494 folio_zero_range(folio: dest_folio, start: dest_pgoff + out_len, length: destlen - out_len);
495 }
496out:
497 return ret;
498}
499
500const struct btrfs_compress_levels btrfs_lzo_compress = {
501 .max_level = 1,
502 .default_level = 1,
503};
504

source code of linux/fs/btrfs/lzo.c