Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Squashfs - a compressed read only filesystem for Linux
4 *
5 * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
6 * Phillip Lougher <phillip@squashfs.org.uk>
7 *
8 * file.c
9 */
10
11/*
12 * This file contains code for handling regular files. A regular file
13 * consists of a sequence of contiguous compressed blocks, and/or a
14 * compressed fragment block (tail-end packed block). The compressed size
15 * of each datablock is stored in a block list contained within the
16 * file inode (itself stored in one or more compressed metadata blocks).
17 *
18 * To speed up access to datablocks when reading 'large' files (256 Mbytes or
19 * larger), the code implements an index cache that caches the mapping from
20 * block index to datablock location on disk.
21 *
22 * The index cache allows Squashfs to handle large files (up to 1.75 TiB) while
23 * retaining a simple and space-efficient block list on disk. The cache
24 * is split into slots, caching up to eight 224 GiB files (128 KiB blocks).
25 * Larger files use multiple slots, with 1.75 TiB files using all 8 slots.
26 * The index cache is designed to be memory efficient, and by default uses
27 * 16 KiB.
28 */
29
30#include <linux/fs.h>
31#include <linux/vfs.h>
32#include <linux/kernel.h>
33#include <linux/slab.h>
34#include <linux/string.h>
35#include <linux/pagemap.h>
36#include <linux/mutex.h>
37
38#include "squashfs_fs.h"
39#include "squashfs_fs_sb.h"
40#include "squashfs_fs_i.h"
41#include "squashfs.h"
42#include "page_actor.h"
43
44/*
45 * Locate cache slot in range [offset, index] for specified inode. If
46 * there's more than one return the slot closest to index.
47 */
48static struct meta_index *locate_meta_index(struct inode *inode, int offset,
49 int index)
50{
51 struct meta_index *meta = NULL;
52 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
53 int i;
54
55 mutex_lock(&msblk->meta_index_mutex);
56
57 TRACE("locate_meta_index: index %d, offset %d\n", index, offset);
58
59 if (msblk->meta_index == NULL)
60 goto not_allocated;
61
62 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
63 if (msblk->meta_index[i].inode_number == inode->i_ino &&
64 msblk->meta_index[i].offset >= offset &&
65 msblk->meta_index[i].offset <= index &&
66 msblk->meta_index[i].locked == 0) {
67 TRACE("locate_meta_index: entry %d, offset %d\n", i,
68 msblk->meta_index[i].offset);
69 meta = &msblk->meta_index[i];
70 offset = meta->offset;
71 }
72 }
73
74 if (meta)
75 meta->locked = 1;
76
77not_allocated:
78 mutex_unlock(&msblk->meta_index_mutex);
79
80 return meta;
81}
82
83
84/*
85 * Find and initialise an empty cache slot for index offset.
86 */
87static struct meta_index *empty_meta_index(struct inode *inode, int offset,
88 int skip)
89{
90 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
91 struct meta_index *meta = NULL;
92 int i;
93
94 mutex_lock(&msblk->meta_index_mutex);
95
96 TRACE("empty_meta_index: offset %d, skip %d\n", offset, skip);
97
98 if (msblk->meta_index == NULL) {
99 /*
100 * First time cache index has been used, allocate and
101 * initialise. The cache index could be allocated at
102 * mount time but doing it here means it is allocated only
103 * if a 'large' file is read.
104 */
105 msblk->meta_index = kcalloc(SQUASHFS_META_SLOTS,
106 sizeof(*(msblk->meta_index)), GFP_KERNEL);
107 if (msblk->meta_index == NULL) {
108 ERROR("Failed to allocate meta_index\n");
109 goto failed;
110 }
111 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
112 msblk->meta_index[i].inode_number = 0;
113 msblk->meta_index[i].locked = 0;
114 }
115 msblk->next_meta_index = 0;
116 }
117
118 for (i = SQUASHFS_META_SLOTS; i &&
119 msblk->meta_index[msblk->next_meta_index].locked; i--)
120 msblk->next_meta_index = (msblk->next_meta_index + 1) %
121 SQUASHFS_META_SLOTS;
122
123 if (i == 0) {
124 TRACE("empty_meta_index: failed!\n");
125 goto failed;
126 }
127
128 TRACE("empty_meta_index: returned meta entry %d, %p\n",
129 msblk->next_meta_index,
130 &msblk->meta_index[msblk->next_meta_index]);
131
132 meta = &msblk->meta_index[msblk->next_meta_index];
133 msblk->next_meta_index = (msblk->next_meta_index + 1) %
134 SQUASHFS_META_SLOTS;
135
136 meta->inode_number = inode->i_ino;
137 meta->offset = offset;
138 meta->skip = skip;
139 meta->entries = 0;
140 meta->locked = 1;
141
142failed:
143 mutex_unlock(&msblk->meta_index_mutex);
144 return meta;
145}
146
147
148static void release_meta_index(struct inode *inode, struct meta_index *meta)
149{
150 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
151 mutex_lock(&msblk->meta_index_mutex);
152 meta->locked = 0;
153 mutex_unlock(&msblk->meta_index_mutex);
154}
155
156
157/*
158 * Read the next n blocks from the block list, starting from
159 * metadata block <start_block, offset>.
160 */
161static long long read_indexes(struct super_block *sb, int n,
162 u64 *start_block, int *offset)
163{
164 int err, i;
165 long long block = 0;
166 __le32 *blist = kmalloc(PAGE_SIZE, GFP_KERNEL);
167
168 if (blist == NULL) {
169 ERROR("read_indexes: Failed to allocate block_list\n");
170 return -ENOMEM;
171 }
172
173 while (n) {
174 int blocks = min_t(int, n, PAGE_SIZE >> 2);
175
176 err = squashfs_read_metadata(sb, blist, start_block,
177 offset, blocks << 2);
178 if (err < 0) {
179 ERROR("read_indexes: reading block [%llx:%x]\n",
180 *start_block, *offset);
181 goto failure;
182 }
183
184 for (i = 0; i < blocks; i++) {
185 int size = squashfs_block_size(blist[i]);
186 if (size < 0) {
187 err = size;
188 goto failure;
189 }
190 block += SQUASHFS_COMPRESSED_SIZE_BLOCK(size);
191 }
192 n -= blocks;
193 }
194
195 kfree(blist);
196 return block;
197
198failure:
199 kfree(blist);
200 return err;
201}
202
203
204/*
205 * Each cache index slot has SQUASHFS_META_ENTRIES, each of which
206 * can cache one index -> datablock/blocklist-block mapping. We wish
207 * to distribute these over the length of the file, entry[0] maps index x,
208 * entry[1] maps index x + skip, entry[2] maps index x + 2 * skip, and so on.
209 * The larger the file, the greater the skip factor. The skip factor is
210 * limited to the size of the metadata cache (SQUASHFS_CACHED_BLKS) to ensure
211 * the number of metadata blocks that need to be read fits into the cache.
212 * If the skip factor is limited in this way then the file will use multiple
213 * slots.
214 */
215static inline int calculate_skip(u64 blocks)
216{
217 u64 skip = blocks / ((SQUASHFS_META_ENTRIES + 1)
218 * SQUASHFS_META_INDEXES);
219 return min((u64) SQUASHFS_CACHED_BLKS - 1, skip + 1);
220}
221
222
223/*
224 * Search and grow the index cache for the specified inode, returning the
225 * on-disk locations of the datablock and block list metadata block
226 * <index_block, index_offset> for index (scaled to nearest cache index).
227 */
228static int fill_meta_index(struct inode *inode, int index,
229 u64 *index_block, int *index_offset, u64 *data_block)
230{
231 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
232 int skip = calculate_skip(i_size_read(inode) >> msblk->block_log);
233 int offset = 0;
234 struct meta_index *meta;
235 struct meta_entry *meta_entry;
236 u64 cur_index_block = squashfs_i(inode)->block_list_start;
237 int cur_offset = squashfs_i(inode)->offset;
238 u64 cur_data_block = squashfs_i(inode)->start;
239 int err, i;
240
241 /*
242 * Scale index to cache index (cache slot entry)
243 */
244 index /= SQUASHFS_META_INDEXES * skip;
245
246 while (offset < index) {
247 meta = locate_meta_index(inode, offset + 1, index);
248
249 if (meta == NULL) {
250 meta = empty_meta_index(inode, offset + 1, skip);
251 if (meta == NULL)
252 goto all_done;
253 } else {
254 offset = index < meta->offset + meta->entries ? index :
255 meta->offset + meta->entries - 1;
256 meta_entry = &meta->meta_entry[offset - meta->offset];
257 cur_index_block = meta_entry->index_block +
258 msblk->inode_table;
259 cur_offset = meta_entry->offset;
260 cur_data_block = meta_entry->data_block;
261 TRACE("get_meta_index: offset %d, meta->offset %d, "
262 "meta->entries %d\n", offset, meta->offset,
263 meta->entries);
264 TRACE("get_meta_index: index_block 0x%llx, offset 0x%x"
265 " data_block 0x%llx\n", cur_index_block,
266 cur_offset, cur_data_block);
267 }
268
269 /*
270 * If necessary grow cache slot by reading block list. Cache
271 * slot is extended up to index or to the end of the slot, in
272 * which case further slots will be used.
273 */
274 for (i = meta->offset + meta->entries; i <= index &&
275 i < meta->offset + SQUASHFS_META_ENTRIES; i++) {
276 int blocks = skip * SQUASHFS_META_INDEXES;
277 long long res = read_indexes(inode->i_sb, blocks,
278 &cur_index_block, &cur_offset);
279
280 if (res < 0) {
281 if (meta->entries == 0)
282 /*
283 * Don't leave an empty slot on read
284 * error allocated to this inode...
285 */
286 meta->inode_number = 0;
287 err = res;
288 goto failed;
289 }
290
291 cur_data_block += res;
292 meta_entry = &meta->meta_entry[i - meta->offset];
293 meta_entry->index_block = cur_index_block -
294 msblk->inode_table;
295 meta_entry->offset = cur_offset;
296 meta_entry->data_block = cur_data_block;
297 meta->entries++;
298 offset++;
299 }
300
301 TRACE("get_meta_index: meta->offset %d, meta->entries %d\n",
302 meta->offset, meta->entries);
303
304 release_meta_index(inode, meta);
305 }
306
307all_done:
308 *index_block = cur_index_block;
309 *index_offset = cur_offset;
310 *data_block = cur_data_block;
311
312 /*
313 * Scale cache index (cache slot entry) to index
314 */
315 return offset * SQUASHFS_META_INDEXES * skip;
316
317failed:
318 release_meta_index(inode, meta);
319 return err;
320}
321
322
323/*
324 * Get the on-disk location and compressed size of the datablock
325 * specified by index. Fill_meta_index() does most of the work.
326 */
327static int read_blocklist(struct inode *inode, int index, u64 *block)
328{
329 u64 start;
330 long long blks;
331 int offset;
332 __le32 size;
333 int res = fill_meta_index(inode, index, &start, &offset, block);
334
335 TRACE("read_blocklist: res %d, index %d, start 0x%llx, offset"
336 " 0x%x, block 0x%llx\n", res, index, start, offset,
337 *block);
338
339 if (res < 0)
340 return res;
341
342 /*
343 * res contains the index of the mapping returned by fill_meta_index(),
344 * this will likely be less than the desired index (because the
345 * meta_index cache works at a higher granularity). Read any
346 * extra block indexes needed.
347 */
348 if (res < index) {
349 blks = read_indexes(inode->i_sb, index - res, &start, &offset);
350 if (blks < 0)
351 return (int) blks;
352 *block += blks;
353 }
354
355 /*
356 * Read length of block specified by index.
357 */
358 res = squashfs_read_metadata(inode->i_sb, &size, &start, &offset,
359 sizeof(size));
360 if (res < 0)
361 return res;
362 return squashfs_block_size(size);
363}
364
365void squashfs_fill_page(struct page *page, struct squashfs_cache_entry *buffer, int offset, int avail)
366{
367 int copied;
368 void *pageaddr;
369
370 pageaddr = kmap_atomic(page);
371 copied = squashfs_copy_data(pageaddr, buffer, offset, avail);
372 memset(pageaddr + copied, 0, PAGE_SIZE - copied);
373 kunmap_atomic(pageaddr);
374
375 flush_dcache_page(page);
376 if (copied == avail)
377 SetPageUptodate(page);
378 else
379 SetPageError(page);
380}
381
382/* Copy data into page cache */
383void squashfs_copy_cache(struct page *page, struct squashfs_cache_entry *buffer,
384 int bytes, int offset)
385{
386 struct inode *inode = page->mapping->host;
387 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
388 int i, mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1;
389 int start_index = page->index & ~mask, end_index = start_index | mask;
390
391 /*
392 * Loop copying datablock into pages. As the datablock likely covers
393 * many PAGE_SIZE pages (default block size is 128 KiB) explicitly
394 * grab the pages from the page cache, except for the page that we've
395 * been called to fill.
396 */
397 for (i = start_index; i <= end_index && bytes > 0; i++,
398 bytes -= PAGE_SIZE, offset += PAGE_SIZE) {
399 struct page *push_page;
400 int avail = buffer ? min_t(int, bytes, PAGE_SIZE) : 0;
401
402 TRACE("bytes %d, i %d, available_bytes %d\n", bytes, i, avail);
403
404 push_page = (i == page->index) ? page :
405 grab_cache_page_nowait(page->mapping, i);
406
407 if (!push_page)
408 continue;
409
410 if (PageUptodate(push_page))
411 goto skip_page;
412
413 squashfs_fill_page(push_page, buffer, offset, avail);
414skip_page:
415 unlock_page(push_page);
416 if (i != page->index)
417 put_page(push_page);
418 }
419}
420
421/* Read datablock stored packed inside a fragment (tail-end packed block) */
422static int squashfs_readpage_fragment(struct page *page, int expected)
423{
424 struct inode *inode = page->mapping->host;
425 struct squashfs_cache_entry *buffer = squashfs_get_fragment(inode->i_sb,
426 squashfs_i(inode)->fragment_block,
427 squashfs_i(inode)->fragment_size);
428 int res = buffer->error;
429
430 if (res)
431 ERROR("Unable to read page, block %llx, size %x\n",
432 squashfs_i(inode)->fragment_block,
433 squashfs_i(inode)->fragment_size);
434 else
435 squashfs_copy_cache(page, buffer, expected,
436 squashfs_i(inode)->fragment_offset);
437
438 squashfs_cache_put(buffer);
439 return res;
440}
441
442static int squashfs_readpage_sparse(struct page *page, int expected)
443{
444 squashfs_copy_cache(page, NULL, expected, 0);
445 return 0;
446}
447
448static int squashfs_read_folio(struct file *file, struct folio *folio)
449{
450 struct page *page = &folio->page;
451 struct inode *inode = page->mapping->host;
452 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
453 int index = page->index >> (msblk->block_log - PAGE_SHIFT);
454 int file_end = i_size_read(inode) >> msblk->block_log;
455 int expected = index == file_end ?
456 (i_size_read(inode) & (msblk->block_size - 1)) :
457 msblk->block_size;
458 int res = 0;
459 void *pageaddr;
460
461 TRACE("Entered squashfs_readpage, page index %lx, start block %llx\n",
462 page->index, squashfs_i(inode)->start);
463
464 if (page->index >= ((i_size_read(inode) + PAGE_SIZE - 1) >>
465 PAGE_SHIFT))
466 goto out;
467
468 if (index < file_end || squashfs_i(inode)->fragment_block ==
469 SQUASHFS_INVALID_BLK) {
470 u64 block = 0;
471
472 res = read_blocklist(inode, index, &block);
473 if (res < 0)
474 goto error_out;
475
476 if (res == 0)
477 res = squashfs_readpage_sparse(page, expected);
478 else
479 res = squashfs_readpage_block(page, block, res, expected);
480 } else
481 res = squashfs_readpage_fragment(page, expected);
482
483 if (!res)
484 return 0;
485
486error_out:
487 SetPageError(page);
488out:
489 pageaddr = kmap_atomic(page);
490 memset(pageaddr, 0, PAGE_SIZE);
491 kunmap_atomic(pageaddr);
492 flush_dcache_page(page);
493 if (res == 0)
494 SetPageUptodate(page);
495 unlock_page(page);
496
497 return res;
498}
499
500static int squashfs_readahead_fragment(struct page **page,
501 unsigned int pages, unsigned int expected)
502{
503 struct inode *inode = page[0]->mapping->host;
504 struct squashfs_cache_entry *buffer = squashfs_get_fragment(inode->i_sb,
505 squashfs_i(inode)->fragment_block,
506 squashfs_i(inode)->fragment_size);
507 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
508 unsigned int n, mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1;
509 int error = buffer->error;
510
511 if (error)
512 goto out;
513
514 expected += squashfs_i(inode)->fragment_offset;
515
516 for (n = 0; n < pages; n++) {
517 unsigned int base = (page[n]->index & mask) << PAGE_SHIFT;
518 unsigned int offset = base + squashfs_i(inode)->fragment_offset;
519
520 if (expected > offset) {
521 unsigned int avail = min_t(unsigned int, expected -
522 offset, PAGE_SIZE);
523
524 squashfs_fill_page(page[n], buffer, offset, avail);
525 }
526
527 unlock_page(page[n]);
528 put_page(page[n]);
529 }
530
531out:
532 squashfs_cache_put(buffer);
533 return error;
534}
535
536static void squashfs_readahead(struct readahead_control *ractl)
537{
538 struct inode *inode = ractl->mapping->host;
539 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
540 size_t mask = (1UL << msblk->block_log) - 1;
541 unsigned short shift = msblk->block_log - PAGE_SHIFT;
542 loff_t start = readahead_pos(ractl) & ~mask;
543 size_t len = readahead_length(ractl) + readahead_pos(ractl) - start;
544 struct squashfs_page_actor *actor;
545 unsigned int nr_pages = 0;
546 struct page **pages;
547 int i, file_end = i_size_read(inode) >> msblk->block_log;
548 unsigned int max_pages = 1UL << shift;
549
550 readahead_expand(ractl, start, (len | mask) + 1);
551
552 pages = kmalloc_array(max_pages, sizeof(void *), GFP_KERNEL);
553 if (!pages)
554 return;
555
556 for (;;) {
557 pgoff_t index;
558 int res, bsize;
559 u64 block = 0;
560 unsigned int expected;
561 struct page *last_page;
562
563 expected = start >> msblk->block_log == file_end ?
564 (i_size_read(inode) & (msblk->block_size - 1)) :
565 msblk->block_size;
566
567 max_pages = (expected + PAGE_SIZE - 1) >> PAGE_SHIFT;
568
569 nr_pages = __readahead_batch(ractl, pages, max_pages);
570 if (!nr_pages)
571 break;
572
573 if (readahead_pos(ractl) >= i_size_read(inode))
574 goto skip_pages;
575
576 index = pages[0]->index >> shift;
577
578 if ((pages[nr_pages - 1]->index >> shift) != index)
579 goto skip_pages;
580
581 if (index == file_end && squashfs_i(inode)->fragment_block !=
582 SQUASHFS_INVALID_BLK) {
583 res = squashfs_readahead_fragment(pages, nr_pages,
584 expected);
585 if (res)
586 goto skip_pages;
587 continue;
588 }
589
590 bsize = read_blocklist(inode, index, &block);
591 if (bsize == 0)
592 goto skip_pages;
593
594 actor = squashfs_page_actor_init_special(msblk, pages, nr_pages,
595 expected);
596 if (!actor)
597 goto skip_pages;
598
599 res = squashfs_read_data(inode->i_sb, block, bsize, NULL, actor);
600
601 last_page = squashfs_page_actor_free(actor);
602
603 if (res == expected) {
604 int bytes;
605
606 /* Last page (if present) may have trailing bytes not filled */
607 bytes = res % PAGE_SIZE;
608 if (index == file_end && bytes && last_page)
609 memzero_page(last_page, bytes,
610 PAGE_SIZE - bytes);
611
612 for (i = 0; i < nr_pages; i++) {
613 flush_dcache_page(pages[i]);
614 SetPageUptodate(pages[i]);
615 }
616 }
617
618 for (i = 0; i < nr_pages; i++) {
619 unlock_page(pages[i]);
620 put_page(pages[i]);
621 }
622 }
623
624 kfree(pages);
625 return;
626
627skip_pages:
628 for (i = 0; i < nr_pages; i++) {
629 unlock_page(pages[i]);
630 put_page(pages[i]);
631 }
632 kfree(pages);
633}
634
635const struct address_space_operations squashfs_aops = {
636 .read_folio = squashfs_read_folio,
637 .readahead = squashfs_readahead
638};
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Squashfs - a compressed read only filesystem for Linux
4 *
5 * Copyright (c) 2002, 2003, 2004, 2005, 2006, 2007, 2008
6 * Phillip Lougher <phillip@squashfs.org.uk>
7 *
8 * file.c
9 */
10
11/*
12 * This file contains code for handling regular files. A regular file
13 * consists of a sequence of contiguous compressed blocks, and/or a
14 * compressed fragment block (tail-end packed block). The compressed size
15 * of each datablock is stored in a block list contained within the
16 * file inode (itself stored in one or more compressed metadata blocks).
17 *
18 * To speed up access to datablocks when reading 'large' files (256 Mbytes or
19 * larger), the code implements an index cache that caches the mapping from
20 * block index to datablock location on disk.
21 *
22 * The index cache allows Squashfs to handle large files (up to 1.75 TiB) while
23 * retaining a simple and space-efficient block list on disk. The cache
24 * is split into slots, caching up to eight 224 GiB files (128 KiB blocks).
25 * Larger files use multiple slots, with 1.75 TiB files using all 8 slots.
26 * The index cache is designed to be memory efficient, and by default uses
27 * 16 KiB.
28 */
29
30#include <linux/fs.h>
31#include <linux/vfs.h>
32#include <linux/kernel.h>
33#include <linux/slab.h>
34#include <linux/string.h>
35#include <linux/pagemap.h>
36#include <linux/mutex.h>
37
38#include "squashfs_fs.h"
39#include "squashfs_fs_sb.h"
40#include "squashfs_fs_i.h"
41#include "squashfs.h"
42
43/*
44 * Locate cache slot in range [offset, index] for specified inode. If
45 * there's more than one return the slot closest to index.
46 */
47static struct meta_index *locate_meta_index(struct inode *inode, int offset,
48 int index)
49{
50 struct meta_index *meta = NULL;
51 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
52 int i;
53
54 mutex_lock(&msblk->meta_index_mutex);
55
56 TRACE("locate_meta_index: index %d, offset %d\n", index, offset);
57
58 if (msblk->meta_index == NULL)
59 goto not_allocated;
60
61 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
62 if (msblk->meta_index[i].inode_number == inode->i_ino &&
63 msblk->meta_index[i].offset >= offset &&
64 msblk->meta_index[i].offset <= index &&
65 msblk->meta_index[i].locked == 0) {
66 TRACE("locate_meta_index: entry %d, offset %d\n", i,
67 msblk->meta_index[i].offset);
68 meta = &msblk->meta_index[i];
69 offset = meta->offset;
70 }
71 }
72
73 if (meta)
74 meta->locked = 1;
75
76not_allocated:
77 mutex_unlock(&msblk->meta_index_mutex);
78
79 return meta;
80}
81
82
83/*
84 * Find and initialise an empty cache slot for index offset.
85 */
86static struct meta_index *empty_meta_index(struct inode *inode, int offset,
87 int skip)
88{
89 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
90 struct meta_index *meta = NULL;
91 int i;
92
93 mutex_lock(&msblk->meta_index_mutex);
94
95 TRACE("empty_meta_index: offset %d, skip %d\n", offset, skip);
96
97 if (msblk->meta_index == NULL) {
98 /*
99 * First time cache index has been used, allocate and
100 * initialise. The cache index could be allocated at
101 * mount time but doing it here means it is allocated only
102 * if a 'large' file is read.
103 */
104 msblk->meta_index = kcalloc(SQUASHFS_META_SLOTS,
105 sizeof(*(msblk->meta_index)), GFP_KERNEL);
106 if (msblk->meta_index == NULL) {
107 ERROR("Failed to allocate meta_index\n");
108 goto failed;
109 }
110 for (i = 0; i < SQUASHFS_META_SLOTS; i++) {
111 msblk->meta_index[i].inode_number = 0;
112 msblk->meta_index[i].locked = 0;
113 }
114 msblk->next_meta_index = 0;
115 }
116
117 for (i = SQUASHFS_META_SLOTS; i &&
118 msblk->meta_index[msblk->next_meta_index].locked; i--)
119 msblk->next_meta_index = (msblk->next_meta_index + 1) %
120 SQUASHFS_META_SLOTS;
121
122 if (i == 0) {
123 TRACE("empty_meta_index: failed!\n");
124 goto failed;
125 }
126
127 TRACE("empty_meta_index: returned meta entry %d, %p\n",
128 msblk->next_meta_index,
129 &msblk->meta_index[msblk->next_meta_index]);
130
131 meta = &msblk->meta_index[msblk->next_meta_index];
132 msblk->next_meta_index = (msblk->next_meta_index + 1) %
133 SQUASHFS_META_SLOTS;
134
135 meta->inode_number = inode->i_ino;
136 meta->offset = offset;
137 meta->skip = skip;
138 meta->entries = 0;
139 meta->locked = 1;
140
141failed:
142 mutex_unlock(&msblk->meta_index_mutex);
143 return meta;
144}
145
146
147static void release_meta_index(struct inode *inode, struct meta_index *meta)
148{
149 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
150 mutex_lock(&msblk->meta_index_mutex);
151 meta->locked = 0;
152 mutex_unlock(&msblk->meta_index_mutex);
153}
154
155
156/*
157 * Read the next n blocks from the block list, starting from
158 * metadata block <start_block, offset>.
159 */
160static long long read_indexes(struct super_block *sb, int n,
161 u64 *start_block, int *offset)
162{
163 int err, i;
164 long long block = 0;
165 __le32 *blist = kmalloc(PAGE_SIZE, GFP_KERNEL);
166
167 if (blist == NULL) {
168 ERROR("read_indexes: Failed to allocate block_list\n");
169 return -ENOMEM;
170 }
171
172 while (n) {
173 int blocks = min_t(int, n, PAGE_SIZE >> 2);
174
175 err = squashfs_read_metadata(sb, blist, start_block,
176 offset, blocks << 2);
177 if (err < 0) {
178 ERROR("read_indexes: reading block [%llx:%x]\n",
179 *start_block, *offset);
180 goto failure;
181 }
182
183 for (i = 0; i < blocks; i++) {
184 int size = squashfs_block_size(blist[i]);
185 if (size < 0) {
186 err = size;
187 goto failure;
188 }
189 block += SQUASHFS_COMPRESSED_SIZE_BLOCK(size);
190 }
191 n -= blocks;
192 }
193
194 kfree(blist);
195 return block;
196
197failure:
198 kfree(blist);
199 return err;
200}
201
202
203/*
204 * Each cache index slot has SQUASHFS_META_ENTRIES, each of which
205 * can cache one index -> datablock/blocklist-block mapping. We wish
206 * to distribute these over the length of the file, entry[0] maps index x,
207 * entry[1] maps index x + skip, entry[2] maps index x + 2 * skip, and so on.
208 * The larger the file, the greater the skip factor. The skip factor is
209 * limited to the size of the metadata cache (SQUASHFS_CACHED_BLKS) to ensure
210 * the number of metadata blocks that need to be read fits into the cache.
211 * If the skip factor is limited in this way then the file will use multiple
212 * slots.
213 */
214static inline int calculate_skip(int blocks)
215{
216 int skip = blocks / ((SQUASHFS_META_ENTRIES + 1)
217 * SQUASHFS_META_INDEXES);
218 return min(SQUASHFS_CACHED_BLKS - 1, skip + 1);
219}
220
221
222/*
223 * Search and grow the index cache for the specified inode, returning the
224 * on-disk locations of the datablock and block list metadata block
225 * <index_block, index_offset> for index (scaled to nearest cache index).
226 */
227static int fill_meta_index(struct inode *inode, int index,
228 u64 *index_block, int *index_offset, u64 *data_block)
229{
230 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
231 int skip = calculate_skip(i_size_read(inode) >> msblk->block_log);
232 int offset = 0;
233 struct meta_index *meta;
234 struct meta_entry *meta_entry;
235 u64 cur_index_block = squashfs_i(inode)->block_list_start;
236 int cur_offset = squashfs_i(inode)->offset;
237 u64 cur_data_block = squashfs_i(inode)->start;
238 int err, i;
239
240 /*
241 * Scale index to cache index (cache slot entry)
242 */
243 index /= SQUASHFS_META_INDEXES * skip;
244
245 while (offset < index) {
246 meta = locate_meta_index(inode, offset + 1, index);
247
248 if (meta == NULL) {
249 meta = empty_meta_index(inode, offset + 1, skip);
250 if (meta == NULL)
251 goto all_done;
252 } else {
253 offset = index < meta->offset + meta->entries ? index :
254 meta->offset + meta->entries - 1;
255 meta_entry = &meta->meta_entry[offset - meta->offset];
256 cur_index_block = meta_entry->index_block +
257 msblk->inode_table;
258 cur_offset = meta_entry->offset;
259 cur_data_block = meta_entry->data_block;
260 TRACE("get_meta_index: offset %d, meta->offset %d, "
261 "meta->entries %d\n", offset, meta->offset,
262 meta->entries);
263 TRACE("get_meta_index: index_block 0x%llx, offset 0x%x"
264 " data_block 0x%llx\n", cur_index_block,
265 cur_offset, cur_data_block);
266 }
267
268 /*
269 * If necessary grow cache slot by reading block list. Cache
270 * slot is extended up to index or to the end of the slot, in
271 * which case further slots will be used.
272 */
273 for (i = meta->offset + meta->entries; i <= index &&
274 i < meta->offset + SQUASHFS_META_ENTRIES; i++) {
275 int blocks = skip * SQUASHFS_META_INDEXES;
276 long long res = read_indexes(inode->i_sb, blocks,
277 &cur_index_block, &cur_offset);
278
279 if (res < 0) {
280 if (meta->entries == 0)
281 /*
282 * Don't leave an empty slot on read
283 * error allocated to this inode...
284 */
285 meta->inode_number = 0;
286 err = res;
287 goto failed;
288 }
289
290 cur_data_block += res;
291 meta_entry = &meta->meta_entry[i - meta->offset];
292 meta_entry->index_block = cur_index_block -
293 msblk->inode_table;
294 meta_entry->offset = cur_offset;
295 meta_entry->data_block = cur_data_block;
296 meta->entries++;
297 offset++;
298 }
299
300 TRACE("get_meta_index: meta->offset %d, meta->entries %d\n",
301 meta->offset, meta->entries);
302
303 release_meta_index(inode, meta);
304 }
305
306all_done:
307 *index_block = cur_index_block;
308 *index_offset = cur_offset;
309 *data_block = cur_data_block;
310
311 /*
312 * Scale cache index (cache slot entry) to index
313 */
314 return offset * SQUASHFS_META_INDEXES * skip;
315
316failed:
317 release_meta_index(inode, meta);
318 return err;
319}
320
321
322/*
323 * Get the on-disk location and compressed size of the datablock
324 * specified by index. Fill_meta_index() does most of the work.
325 */
326static int read_blocklist(struct inode *inode, int index, u64 *block)
327{
328 u64 start;
329 long long blks;
330 int offset;
331 __le32 size;
332 int res = fill_meta_index(inode, index, &start, &offset, block);
333
334 TRACE("read_blocklist: res %d, index %d, start 0x%llx, offset"
335 " 0x%x, block 0x%llx\n", res, index, start, offset,
336 *block);
337
338 if (res < 0)
339 return res;
340
341 /*
342 * res contains the index of the mapping returned by fill_meta_index(),
343 * this will likely be less than the desired index (because the
344 * meta_index cache works at a higher granularity). Read any
345 * extra block indexes needed.
346 */
347 if (res < index) {
348 blks = read_indexes(inode->i_sb, index - res, &start, &offset);
349 if (blks < 0)
350 return (int) blks;
351 *block += blks;
352 }
353
354 /*
355 * Read length of block specified by index.
356 */
357 res = squashfs_read_metadata(inode->i_sb, &size, &start, &offset,
358 sizeof(size));
359 if (res < 0)
360 return res;
361 return squashfs_block_size(size);
362}
363
364void squashfs_fill_page(struct page *page, struct squashfs_cache_entry *buffer, int offset, int avail)
365{
366 int copied;
367 void *pageaddr;
368
369 pageaddr = kmap_atomic(page);
370 copied = squashfs_copy_data(pageaddr, buffer, offset, avail);
371 memset(pageaddr + copied, 0, PAGE_SIZE - copied);
372 kunmap_atomic(pageaddr);
373
374 flush_dcache_page(page);
375 if (copied == avail)
376 SetPageUptodate(page);
377 else
378 SetPageError(page);
379}
380
381/* Copy data into page cache */
382void squashfs_copy_cache(struct page *page, struct squashfs_cache_entry *buffer,
383 int bytes, int offset)
384{
385 struct inode *inode = page->mapping->host;
386 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
387 int i, mask = (1 << (msblk->block_log - PAGE_SHIFT)) - 1;
388 int start_index = page->index & ~mask, end_index = start_index | mask;
389
390 /*
391 * Loop copying datablock into pages. As the datablock likely covers
392 * many PAGE_SIZE pages (default block size is 128 KiB) explicitly
393 * grab the pages from the page cache, except for the page that we've
394 * been called to fill.
395 */
396 for (i = start_index; i <= end_index && bytes > 0; i++,
397 bytes -= PAGE_SIZE, offset += PAGE_SIZE) {
398 struct page *push_page;
399 int avail = buffer ? min_t(int, bytes, PAGE_SIZE) : 0;
400
401 TRACE("bytes %d, i %d, available_bytes %d\n", bytes, i, avail);
402
403 push_page = (i == page->index) ? page :
404 grab_cache_page_nowait(page->mapping, i);
405
406 if (!push_page)
407 continue;
408
409 if (PageUptodate(push_page))
410 goto skip_page;
411
412 squashfs_fill_page(push_page, buffer, offset, avail);
413skip_page:
414 unlock_page(push_page);
415 if (i != page->index)
416 put_page(push_page);
417 }
418}
419
420/* Read datablock stored packed inside a fragment (tail-end packed block) */
421static int squashfs_readpage_fragment(struct page *page, int expected)
422{
423 struct inode *inode = page->mapping->host;
424 struct squashfs_cache_entry *buffer = squashfs_get_fragment(inode->i_sb,
425 squashfs_i(inode)->fragment_block,
426 squashfs_i(inode)->fragment_size);
427 int res = buffer->error;
428
429 if (res)
430 ERROR("Unable to read page, block %llx, size %x\n",
431 squashfs_i(inode)->fragment_block,
432 squashfs_i(inode)->fragment_size);
433 else
434 squashfs_copy_cache(page, buffer, expected,
435 squashfs_i(inode)->fragment_offset);
436
437 squashfs_cache_put(buffer);
438 return res;
439}
440
441static int squashfs_readpage_sparse(struct page *page, int expected)
442{
443 squashfs_copy_cache(page, NULL, expected, 0);
444 return 0;
445}
446
447static int squashfs_readpage(struct file *file, struct page *page)
448{
449 struct inode *inode = page->mapping->host;
450 struct squashfs_sb_info *msblk = inode->i_sb->s_fs_info;
451 int index = page->index >> (msblk->block_log - PAGE_SHIFT);
452 int file_end = i_size_read(inode) >> msblk->block_log;
453 int expected = index == file_end ?
454 (i_size_read(inode) & (msblk->block_size - 1)) :
455 msblk->block_size;
456 int res;
457 void *pageaddr;
458
459 TRACE("Entered squashfs_readpage, page index %lx, start block %llx\n",
460 page->index, squashfs_i(inode)->start);
461
462 if (page->index >= ((i_size_read(inode) + PAGE_SIZE - 1) >>
463 PAGE_SHIFT))
464 goto out;
465
466 if (index < file_end || squashfs_i(inode)->fragment_block ==
467 SQUASHFS_INVALID_BLK) {
468 u64 block = 0;
469 int bsize = read_blocklist(inode, index, &block);
470 if (bsize < 0)
471 goto error_out;
472
473 if (bsize == 0)
474 res = squashfs_readpage_sparse(page, expected);
475 else
476 res = squashfs_readpage_block(page, block, bsize, expected);
477 } else
478 res = squashfs_readpage_fragment(page, expected);
479
480 if (!res)
481 return 0;
482
483error_out:
484 SetPageError(page);
485out:
486 pageaddr = kmap_atomic(page);
487 memset(pageaddr, 0, PAGE_SIZE);
488 kunmap_atomic(pageaddr);
489 flush_dcache_page(page);
490 if (!PageError(page))
491 SetPageUptodate(page);
492 unlock_page(page);
493
494 return 0;
495}
496
497
498const struct address_space_operations squashfs_aops = {
499 .readpage = squashfs_readpage
500};