1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2008 Oracle. All rights reserved.
6 #include <linux/kernel.h>
7 #include <linux/slab.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>
17 #include "compression.h"
20 #include "btrfs_inode.h"
25 * Btrfs LZO compression format
27 * Regular and inlined LZO compressed data extents consist of:
30 * Fixed size. LZO_LEN (4) bytes long, LE32.
31 * Records the total size (including the header) of compressed data.
34 * Variable size. Each segment includes one segment header, followed by data
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.
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.
47 * Variable size. Size up limit should be lzo1x_worst_compress(sectorsize)
48 * which is 4419 for a 4KiB sectorsize.
50 * Example with 4K sectorsize:
52 * 0 0x2 0x4 0x6 0x8 0xa 0xc 0xe 0x10
53 * 0x0000 | Header | SegHdr 01 | Data payload 01 ... |
55 * 0x0ff0 | SegHdr N | Data payload N ... |00|
58 * 0x1000 | SegHdr N+1| Data payload N+1 ... |
61 #define WORKSPACE_BUF_LENGTH (lzo1x_worst_compress(PAGE_SIZE))
62 #define WORKSPACE_CBUF_LENGTH (lzo1x_worst_compress(PAGE_SIZE))
66 void *buf
; /* where decompressed data goes */
67 void *cbuf
; /* where compressed data goes */
68 struct list_head list
;
71 static struct workspace_manager wsm
;
73 void lzo_free_workspace(struct list_head
*ws
)
75 struct workspace
*workspace
= list_entry(ws
, struct workspace
, list
);
77 kvfree(workspace
->buf
);
78 kvfree(workspace
->cbuf
);
79 kvfree(workspace
->mem
);
83 struct list_head
*lzo_alloc_workspace(void)
85 struct workspace
*workspace
;
87 workspace
= kzalloc(sizeof(*workspace
), GFP_KERNEL
);
89 return ERR_PTR(-ENOMEM
);
91 workspace
->mem
= kvmalloc(LZO1X_MEM_COMPRESS
, GFP_KERNEL
| __GFP_NOWARN
);
92 workspace
->buf
= kvmalloc(WORKSPACE_BUF_LENGTH
, GFP_KERNEL
| __GFP_NOWARN
);
93 workspace
->cbuf
= kvmalloc(WORKSPACE_CBUF_LENGTH
, GFP_KERNEL
| __GFP_NOWARN
);
94 if (!workspace
->mem
|| !workspace
->buf
|| !workspace
->cbuf
)
97 INIT_LIST_HEAD(&workspace
->list
);
99 return &workspace
->list
;
101 lzo_free_workspace(&workspace
->list
);
102 return ERR_PTR(-ENOMEM
);
105 static inline void write_compress_length(char *buf
, size_t len
)
109 dlen
= cpu_to_le32(len
);
110 memcpy(buf
, &dlen
, LZO_LEN
);
113 static inline size_t read_compress_length(const char *buf
)
117 memcpy(&dlen
, buf
, LZO_LEN
);
118 return le32_to_cpu(dlen
);
124 * - Write a segment header into the destination
125 * - Copy the compressed buffer into the destination
126 * - Make sure we have enough space in the last sector to fit a segment header
127 * If not, we will pad at most (LZO_LEN (4)) - 1 bytes of zeros.
129 * Will allocate new pages when needed.
131 static int copy_compressed_data_to_page(char *compressed_data
,
132 size_t compressed_size
,
133 struct folio
**out_folios
,
134 unsigned long max_nr_folio
,
136 const u32 sectorsize
)
138 u32 sector_bytes_left
;
140 struct folio
*cur_folio
;
143 if ((*cur_out
/ PAGE_SIZE
) >= max_nr_folio
)
147 * We never allow a segment header crossing sector boundary, previous
148 * run should ensure we have enough space left inside the sector.
150 ASSERT((*cur_out
/ sectorsize
) == (*cur_out
+ LZO_LEN
- 1) / sectorsize
);
152 cur_folio
= out_folios
[*cur_out
/ PAGE_SIZE
];
153 /* Allocate a new page */
155 cur_folio
= btrfs_alloc_compr_folio();
158 out_folios
[*cur_out
/ PAGE_SIZE
] = cur_folio
;
161 kaddr
= kmap_local_folio(cur_folio
, 0);
162 write_compress_length(kaddr
+ offset_in_page(*cur_out
),
168 /* Copy compressed data */
169 while (*cur_out
- orig_out
< compressed_size
) {
170 u32 copy_len
= min_t(u32
, sectorsize
- *cur_out
% sectorsize
,
171 orig_out
+ compressed_size
- *cur_out
);
175 if ((*cur_out
/ PAGE_SIZE
) >= max_nr_folio
)
178 cur_folio
= out_folios
[*cur_out
/ PAGE_SIZE
];
179 /* Allocate a new page */
181 cur_folio
= btrfs_alloc_compr_folio();
184 out_folios
[*cur_out
/ PAGE_SIZE
] = cur_folio
;
186 kaddr
= kmap_local_folio(cur_folio
, 0);
188 memcpy(kaddr
+ offset_in_page(*cur_out
),
189 compressed_data
+ *cur_out
- orig_out
, copy_len
);
191 *cur_out
+= copy_len
;
195 * Check if we can fit the next segment header into the remaining space
198 sector_bytes_left
= round_up(*cur_out
, sectorsize
) - *cur_out
;
199 if (sector_bytes_left
>= LZO_LEN
|| sector_bytes_left
== 0)
202 /* The remaining size is not enough, pad it with zeros */
203 memset(kaddr
+ offset_in_page(*cur_out
), 0,
205 *cur_out
+= sector_bytes_left
;
212 int lzo_compress_folios(struct list_head
*ws
, struct address_space
*mapping
,
213 u64 start
, struct folio
**folios
, unsigned long *out_folios
,
214 unsigned long *total_in
, unsigned long *total_out
)
216 struct workspace
*workspace
= list_entry(ws
, struct workspace
, list
);
217 const u32 sectorsize
= inode_to_fs_info(mapping
->host
)->sectorsize
;
218 struct folio
*folio_in
= NULL
;
220 const unsigned long max_nr_folio
= *out_folios
;
222 /* Points to the file offset of input data */
224 /* Points to the current output byte */
226 u32 len
= *total_out
;
228 ASSERT(max_nr_folio
> 0);
234 * Skip the header for now, we will later come back and write the total
238 while (cur_in
< start
+ len
) {
240 const u32 sectorsize_mask
= sectorsize
- 1;
241 u32 sector_off
= (cur_in
- start
) & sectorsize_mask
;
245 /* Get the input page first */
247 ret
= btrfs_compress_filemap_get_folio(mapping
, cur_in
, &folio_in
);
252 /* Compress at most one sector of data each time */
253 in_len
= min_t(u32
, start
+ len
- cur_in
, sectorsize
- sector_off
);
255 data_in
= kmap_local_folio(folio_in
, 0);
256 ret
= lzo1x_1_compress(data_in
+
257 offset_in_page(cur_in
), in_len
,
258 workspace
->cbuf
, &out_len
,
260 kunmap_local(data_in
);
261 if (unlikely(ret
< 0)) {
262 /* lzo1x_1_compress never fails. */
267 ret
= copy_compressed_data_to_page(workspace
->cbuf
, out_len
,
268 folios
, max_nr_folio
,
269 &cur_out
, sectorsize
);
276 * Check if we're making it bigger after two sectors. And if
279 if (cur_in
- start
> sectorsize
* 2 && cur_in
- start
< cur_out
) {
284 /* Check if we have reached page boundary */
285 if (PAGE_ALIGNED(cur_in
)) {
291 /* Store the size of all chunks of compressed data */
292 sizes_ptr
= kmap_local_folio(folios
[0], 0);
293 write_compress_length(sizes_ptr
, cur_out
);
294 kunmap_local(sizes_ptr
);
297 *total_out
= cur_out
;
298 *total_in
= cur_in
- start
;
302 *out_folios
= DIV_ROUND_UP(cur_out
, PAGE_SIZE
);
307 * Copy the compressed segment payload into @dest.
309 * For the payload there will be no padding, just need to do page switching.
311 static void copy_compressed_segment(struct compressed_bio
*cb
,
312 char *dest
, u32 len
, u32
*cur_in
)
314 u32 orig_in
= *cur_in
;
316 while (*cur_in
< orig_in
+ len
) {
317 struct folio
*cur_folio
;
318 u32 copy_len
= min_t(u32
, PAGE_SIZE
- offset_in_page(*cur_in
),
319 orig_in
+ len
- *cur_in
);
322 cur_folio
= cb
->compressed_folios
[*cur_in
/ PAGE_SIZE
];
324 memcpy_from_folio(dest
+ *cur_in
- orig_in
, cur_folio
,
325 offset_in_folio(cur_folio
, *cur_in
), copy_len
);
331 int lzo_decompress_bio(struct list_head
*ws
, struct compressed_bio
*cb
)
333 struct workspace
*workspace
= list_entry(ws
, struct workspace
, list
);
334 const struct btrfs_fs_info
*fs_info
= cb
->bbio
.inode
->root
->fs_info
;
335 const u32 sectorsize
= fs_info
->sectorsize
;
338 /* Compressed data length, can be unaligned */
340 /* Offset inside the compressed data */
342 /* Bytes decompressed so far */
345 kaddr
= kmap_local_folio(cb
->compressed_folios
[0], 0);
346 len_in
= read_compress_length(kaddr
);
351 * LZO header length check
353 * The total length should not exceed the maximum extent length,
354 * and all sectors should be used.
355 * If this happens, it means the compressed extent is corrupted.
357 if (unlikely(len_in
> min_t(size_t, BTRFS_MAX_COMPRESSED
, cb
->compressed_len
) ||
358 round_up(len_in
, sectorsize
) < cb
->compressed_len
)) {
359 struct btrfs_inode
*inode
= cb
->bbio
.inode
;
362 "lzo header invalid, root %llu inode %llu offset %llu lzo len %u compressed len %u",
363 btrfs_root_id(inode
->root
), btrfs_ino(inode
),
364 cb
->start
, len_in
, cb
->compressed_len
);
368 /* Go through each lzo segment */
369 while (cur_in
< len_in
) {
370 struct folio
*cur_folio
;
371 /* Length of the compressed segment */
373 u32 sector_bytes_left
;
374 size_t out_len
= lzo1x_worst_compress(sectorsize
);
377 * We should always have enough space for one segment header
378 * inside current sector.
380 ASSERT(cur_in
/ sectorsize
==
381 (cur_in
+ LZO_LEN
- 1) / sectorsize
);
382 cur_folio
= cb
->compressed_folios
[cur_in
/ PAGE_SIZE
];
384 kaddr
= kmap_local_folio(cur_folio
, 0);
385 seg_len
= read_compress_length(kaddr
+ offset_in_page(cur_in
));
389 if (unlikely(seg_len
> WORKSPACE_CBUF_LENGTH
)) {
390 struct btrfs_inode
*inode
= cb
->bbio
.inode
;
393 * seg_len shouldn't be larger than we have allocated
394 * for workspace->cbuf
397 "lzo segment too big, root %llu inode %llu offset %llu len %u",
398 btrfs_root_id(inode
->root
), btrfs_ino(inode
),
403 /* Copy the compressed segment payload into workspace */
404 copy_compressed_segment(cb
, workspace
->cbuf
, seg_len
, &cur_in
);
406 /* Decompress the data */
407 ret
= lzo1x_decompress_safe(workspace
->cbuf
, seg_len
,
408 workspace
->buf
, &out_len
);
409 if (unlikely(ret
!= LZO_E_OK
)) {
410 struct btrfs_inode
*inode
= cb
->bbio
.inode
;
413 "lzo decompression failed, error %d root %llu inode %llu offset %llu",
414 ret
, btrfs_root_id(inode
->root
), btrfs_ino(inode
),
419 /* Copy the data into inode pages */
420 ret
= btrfs_decompress_buf2page(workspace
->buf
, out_len
, cb
, cur_out
);
423 /* All data read, exit */
428 /* Check if the sector has enough space for a segment header */
429 sector_bytes_left
= sectorsize
- (cur_in
% sectorsize
);
430 if (sector_bytes_left
>= LZO_LEN
)
433 /* Skip the padding zeros */
434 cur_in
+= sector_bytes_left
;
440 int lzo_decompress(struct list_head
*ws
, const u8
*data_in
,
441 struct folio
*dest_folio
, unsigned long dest_pgoff
, size_t srclen
,
444 struct workspace
*workspace
= list_entry(ws
, struct workspace
, list
);
445 struct btrfs_fs_info
*fs_info
= folio_to_fs_info(dest_folio
);
446 const u32 sectorsize
= fs_info
->sectorsize
;
449 size_t max_segment_len
= WORKSPACE_BUF_LENGTH
;
452 if (srclen
< LZO_LEN
|| srclen
> max_segment_len
+ LZO_LEN
* 2)
455 in_len
= read_compress_length(data_in
);
456 if (in_len
!= srclen
)
460 in_len
= read_compress_length(data_in
);
461 if (in_len
!= srclen
- LZO_LEN
* 2) {
467 out_len
= sectorsize
;
468 ret
= lzo1x_decompress_safe(data_in
, in_len
, workspace
->buf
, &out_len
);
469 if (unlikely(ret
!= LZO_E_OK
)) {
470 struct btrfs_inode
*inode
= folio_to_inode(dest_folio
);
473 "lzo decompression failed, error %d root %llu inode %llu offset %llu",
474 ret
, btrfs_root_id(inode
->root
), btrfs_ino(inode
),
475 folio_pos(dest_folio
));
480 ASSERT(out_len
<= sectorsize
);
481 memcpy_to_folio(dest_folio
, dest_pgoff
, workspace
->buf
, out_len
);
482 /* Early end, considered as an error. */
483 if (unlikely(out_len
< destlen
)) {
485 folio_zero_range(dest_folio
, dest_pgoff
+ out_len
, destlen
- out_len
);
491 const struct btrfs_compress_op btrfs_lzo_compress
= {
492 .workspace_manager
= &wsm
,