blob: 25d24106f893a691cbc7cc5010cf4b181dbdfc8a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/**
2 * compress.c - NTFS kernel compressed attributes handling.
3 * Part of the Linux-NTFS project.
4 *
5 * Copyright (c) 2001-2004 Anton Altaparmakov
6 * Copyright (c) 2002 Richard Russon
7 *
8 * This program/include file is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License as published
10 * by the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * This program/include file is distributed in the hope that it will be
14 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
15 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program (in the main directory of the Linux-NTFS
20 * distribution in the file COPYING); if not, write to the Free Software
21 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 */
23
24#include <linux/fs.h>
25#include <linux/buffer_head.h>
26#include <linux/blkdev.h>
27#include <linux/vmalloc.h>
28
29#include "attrib.h"
30#include "inode.h"
31#include "debug.h"
32#include "ntfs.h"
33
34/**
35 * ntfs_compression_constants - enum of constants used in the compression code
36 */
37typedef enum {
38 /* Token types and access mask. */
39 NTFS_SYMBOL_TOKEN = 0,
40 NTFS_PHRASE_TOKEN = 1,
41 NTFS_TOKEN_MASK = 1,
42
43 /* Compression sub-block constants. */
44 NTFS_SB_SIZE_MASK = 0x0fff,
45 NTFS_SB_SIZE = 0x1000,
46 NTFS_SB_IS_COMPRESSED = 0x8000,
47
48 /*
49 * The maximum compression block size is by definition 16 * the cluster
50 * size, with the maximum supported cluster size being 4kiB. Thus the
51 * maximum compression buffer size is 64kiB, so we use this when
52 * initializing the compression buffer.
53 */
54 NTFS_MAX_CB_SIZE = 64 * 1024,
55} ntfs_compression_constants;
56
57/**
58 * ntfs_compression_buffer - one buffer for the decompression engine
59 */
60static u8 *ntfs_compression_buffer = NULL;
61
62/**
63 * ntfs_cb_lock - spinlock which protects ntfs_compression_buffer
64 */
65static DEFINE_SPINLOCK(ntfs_cb_lock);
66
67/**
68 * allocate_compression_buffers - allocate the decompression buffers
69 *
70 * Caller has to hold the ntfs_lock semaphore.
71 *
72 * Return 0 on success or -ENOMEM if the allocations failed.
73 */
74int allocate_compression_buffers(void)
75{
76 BUG_ON(ntfs_compression_buffer);
77
78 ntfs_compression_buffer = vmalloc(NTFS_MAX_CB_SIZE);
79 if (!ntfs_compression_buffer)
80 return -ENOMEM;
81 return 0;
82}
83
84/**
85 * free_compression_buffers - free the decompression buffers
86 *
87 * Caller has to hold the ntfs_lock semaphore.
88 */
89void free_compression_buffers(void)
90{
91 BUG_ON(!ntfs_compression_buffer);
92 vfree(ntfs_compression_buffer);
93 ntfs_compression_buffer = NULL;
94}
95
96/**
97 * zero_partial_compressed_page - zero out of bounds compressed page region
98 */
Anton Altaparmakov36763672004-11-18 13:46:45 +000099static void zero_partial_compressed_page(struct page *page,
100 const s64 initialized_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
102 u8 *kp = page_address(page);
103 unsigned int kp_ofs;
104
105 ntfs_debug("Zeroing page region outside initialized size.");
Anton Altaparmakov36763672004-11-18 13:46:45 +0000106 if (((s64)page->index << PAGE_CACHE_SHIFT) >= initialized_size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 /*
108 * FIXME: Using clear_page() will become wrong when we get
109 * PAGE_CACHE_SIZE != PAGE_SIZE but for now there is no problem.
110 */
111 clear_page(kp);
112 return;
113 }
Anton Altaparmakov36763672004-11-18 13:46:45 +0000114 kp_ofs = initialized_size & ~PAGE_CACHE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 memset(kp + kp_ofs, 0, PAGE_CACHE_SIZE - kp_ofs);
116 return;
117}
118
119/**
120 * handle_bounds_compressed_page - test for&handle out of bounds compressed page
121 */
Anton Altaparmakov36763672004-11-18 13:46:45 +0000122static inline void handle_bounds_compressed_page(struct page *page,
123 const loff_t i_size, const s64 initialized_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124{
Anton Altaparmakov36763672004-11-18 13:46:45 +0000125 if ((page->index >= (initialized_size >> PAGE_CACHE_SHIFT)) &&
126 (initialized_size < i_size))
127 zero_partial_compressed_page(page, initialized_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return;
129}
130
131/**
132 * ntfs_decompress - decompress a compression block into an array of pages
133 * @dest_pages: destination array of pages
134 * @dest_index: current index into @dest_pages (IN/OUT)
135 * @dest_ofs: current offset within @dest_pages[@dest_index] (IN/OUT)
136 * @dest_max_index: maximum index into @dest_pages (IN)
137 * @dest_max_ofs: maximum offset within @dest_pages[@dest_max_index] (IN)
138 * @xpage: the target page (-1 if none) (IN)
139 * @xpage_done: set to 1 if xpage was completed successfully (IN/OUT)
140 * @cb_start: compression block to decompress (IN)
141 * @cb_size: size of compression block @cb_start in bytes (IN)
Anton Altaparmakov36763672004-11-18 13:46:45 +0000142 * @i_size: file size when we started the read (IN)
143 * @initialized_size: initialized file size when we started the read (IN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 *
145 * The caller must have disabled preemption. ntfs_decompress() reenables it when
146 * the critical section is finished.
147 *
148 * This decompresses the compression block @cb_start into the array of
149 * destination pages @dest_pages starting at index @dest_index into @dest_pages
150 * and at offset @dest_pos into the page @dest_pages[@dest_index].
151 *
152 * When the page @dest_pages[@xpage] is completed, @xpage_done is set to 1.
153 * If xpage is -1 or @xpage has not been completed, @xpage_done is not modified.
154 *
155 * @cb_start is a pointer to the compression block which needs decompressing
156 * and @cb_size is the size of @cb_start in bytes (8-64kiB).
157 *
158 * Return 0 if success or -EOVERFLOW on error in the compressed stream.
159 * @xpage_done indicates whether the target page (@dest_pages[@xpage]) was
160 * completed during the decompression of the compression block (@cb_start).
161 *
162 * Warning: This function *REQUIRES* PAGE_CACHE_SIZE >= 4096 or it will blow up
163 * unpredicatbly! You have been warned!
164 *
165 * Note to hackers: This function may not sleep until it has finished accessing
166 * the compression block @cb_start as it is a per-CPU buffer.
167 */
168static int ntfs_decompress(struct page *dest_pages[], int *dest_index,
169 int *dest_ofs, const int dest_max_index, const int dest_max_ofs,
170 const int xpage, char *xpage_done, u8 *const cb_start,
Anton Altaparmakov36763672004-11-18 13:46:45 +0000171 const u32 cb_size, const loff_t i_size,
172 const s64 initialized_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173{
174 /*
175 * Pointers into the compressed data, i.e. the compression block (cb),
176 * and the therein contained sub-blocks (sb).
177 */
178 u8 *cb_end = cb_start + cb_size; /* End of cb. */
179 u8 *cb = cb_start; /* Current position in cb. */
180 u8 *cb_sb_start = cb; /* Beginning of the current sb in the cb. */
181 u8 *cb_sb_end; /* End of current sb / beginning of next sb. */
182
183 /* Variables for uncompressed data / destination. */
184 struct page *dp; /* Current destination page being worked on. */
185 u8 *dp_addr; /* Current pointer into dp. */
186 u8 *dp_sb_start; /* Start of current sub-block in dp. */
187 u8 *dp_sb_end; /* End of current sb in dp (dp_sb_start +
188 NTFS_SB_SIZE). */
189 u16 do_sb_start; /* @dest_ofs when starting this sub-block. */
190 u16 do_sb_end; /* @dest_ofs of end of this sb (do_sb_start +
191 NTFS_SB_SIZE). */
192
193 /* Variables for tag and token parsing. */
194 u8 tag; /* Current tag. */
195 int token; /* Loop counter for the eight tokens in tag. */
196
197 /* Need this because we can't sleep, so need two stages. */
198 int completed_pages[dest_max_index - *dest_index + 1];
199 int nr_completed_pages = 0;
200
201 /* Default error code. */
202 int err = -EOVERFLOW;
203
204 ntfs_debug("Entering, cb_size = 0x%x.", cb_size);
205do_next_sb:
206 ntfs_debug("Beginning sub-block at offset = 0x%zx in the cb.",
207 cb - cb_start);
208 /*
209 * Have we reached the end of the compression block or the end of the
210 * decompressed data? The latter can happen for example if the current
211 * position in the compression block is one byte before its end so the
212 * first two checks do not detect it.
213 */
214 if (cb == cb_end || !le16_to_cpup((le16*)cb) ||
215 (*dest_index == dest_max_index &&
216 *dest_ofs == dest_max_ofs)) {
217 int i;
218
219 ntfs_debug("Completed. Returning success (0).");
220 err = 0;
221return_error:
222 /* We can sleep from now on, so we drop lock. */
223 spin_unlock(&ntfs_cb_lock);
224 /* Second stage: finalize completed pages. */
225 if (nr_completed_pages > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 for (i = 0; i < nr_completed_pages; i++) {
227 int di = completed_pages[i];
228
229 dp = dest_pages[di];
230 /*
231 * If we are outside the initialized size, zero
232 * the out of bounds page range.
233 */
Anton Altaparmakov36763672004-11-18 13:46:45 +0000234 handle_bounds_compressed_page(dp, i_size,
235 initialized_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 flush_dcache_page(dp);
237 kunmap(dp);
238 SetPageUptodate(dp);
239 unlock_page(dp);
240 if (di == xpage)
241 *xpage_done = 1;
242 else
243 page_cache_release(dp);
244 dest_pages[di] = NULL;
245 }
246 }
247 return err;
248 }
249
250 /* Setup offsets for the current sub-block destination. */
251 do_sb_start = *dest_ofs;
252 do_sb_end = do_sb_start + NTFS_SB_SIZE;
253
254 /* Check that we are still within allowed boundaries. */
255 if (*dest_index == dest_max_index && do_sb_end > dest_max_ofs)
256 goto return_overflow;
257
258 /* Does the minimum size of a compressed sb overflow valid range? */
259 if (cb + 6 > cb_end)
260 goto return_overflow;
261
262 /* Setup the current sub-block source pointers and validate range. */
263 cb_sb_start = cb;
264 cb_sb_end = cb_sb_start + (le16_to_cpup((le16*)cb) & NTFS_SB_SIZE_MASK)
265 + 3;
266 if (cb_sb_end > cb_end)
267 goto return_overflow;
268
269 /* Get the current destination page. */
270 dp = dest_pages[*dest_index];
271 if (!dp) {
272 /* No page present. Skip decompression of this sub-block. */
273 cb = cb_sb_end;
274
275 /* Advance destination position to next sub-block. */
276 *dest_ofs = (*dest_ofs + NTFS_SB_SIZE) & ~PAGE_CACHE_MASK;
277 if (!*dest_ofs && (++*dest_index > dest_max_index))
278 goto return_overflow;
279 goto do_next_sb;
280 }
281
282 /* We have a valid destination page. Setup the destination pointers. */
283 dp_addr = (u8*)page_address(dp) + do_sb_start;
284
285 /* Now, we are ready to process the current sub-block (sb). */
286 if (!(le16_to_cpup((le16*)cb) & NTFS_SB_IS_COMPRESSED)) {
287 ntfs_debug("Found uncompressed sub-block.");
288 /* This sb is not compressed, just copy it into destination. */
289
290 /* Advance source position to first data byte. */
291 cb += 2;
292
293 /* An uncompressed sb must be full size. */
294 if (cb_sb_end - cb != NTFS_SB_SIZE)
295 goto return_overflow;
296
297 /* Copy the block and advance the source position. */
298 memcpy(dp_addr, cb, NTFS_SB_SIZE);
299 cb += NTFS_SB_SIZE;
300
301 /* Advance destination position to next sub-block. */
302 *dest_ofs += NTFS_SB_SIZE;
303 if (!(*dest_ofs &= ~PAGE_CACHE_MASK)) {
304finalize_page:
305 /*
306 * First stage: add current page index to array of
307 * completed pages.
308 */
309 completed_pages[nr_completed_pages++] = *dest_index;
310 if (++*dest_index > dest_max_index)
311 goto return_overflow;
312 }
313 goto do_next_sb;
314 }
315 ntfs_debug("Found compressed sub-block.");
316 /* This sb is compressed, decompress it into destination. */
317
318 /* Setup destination pointers. */
319 dp_sb_start = dp_addr;
320 dp_sb_end = dp_sb_start + NTFS_SB_SIZE;
321
322 /* Forward to the first tag in the sub-block. */
323 cb += 2;
324do_next_tag:
325 if (cb == cb_sb_end) {
326 /* Check if the decompressed sub-block was not full-length. */
327 if (dp_addr < dp_sb_end) {
328 int nr_bytes = do_sb_end - *dest_ofs;
329
330 ntfs_debug("Filling incomplete sub-block with "
331 "zeroes.");
332 /* Zero remainder and update destination position. */
333 memset(dp_addr, 0, nr_bytes);
334 *dest_ofs += nr_bytes;
335 }
336 /* We have finished the current sub-block. */
337 if (!(*dest_ofs &= ~PAGE_CACHE_MASK))
338 goto finalize_page;
339 goto do_next_sb;
340 }
341
342 /* Check we are still in range. */
343 if (cb > cb_sb_end || dp_addr > dp_sb_end)
344 goto return_overflow;
345
346 /* Get the next tag and advance to first token. */
347 tag = *cb++;
348
349 /* Parse the eight tokens described by the tag. */
350 for (token = 0; token < 8; token++, tag >>= 1) {
351 u16 lg, pt, length, max_non_overlap;
352 register u16 i;
353 u8 *dp_back_addr;
354
355 /* Check if we are done / still in range. */
356 if (cb >= cb_sb_end || dp_addr > dp_sb_end)
357 break;
358
359 /* Determine token type and parse appropriately.*/
360 if ((tag & NTFS_TOKEN_MASK) == NTFS_SYMBOL_TOKEN) {
361 /*
362 * We have a symbol token, copy the symbol across, and
363 * advance the source and destination positions.
364 */
365 *dp_addr++ = *cb++;
366 ++*dest_ofs;
367
368 /* Continue with the next token. */
369 continue;
370 }
371
372 /*
373 * We have a phrase token. Make sure it is not the first tag in
374 * the sb as this is illegal and would confuse the code below.
375 */
376 if (dp_addr == dp_sb_start)
377 goto return_overflow;
378
379 /*
380 * Determine the number of bytes to go back (p) and the number
381 * of bytes to copy (l). We use an optimized algorithm in which
382 * we first calculate log2(current destination position in sb),
383 * which allows determination of l and p in O(1) rather than
384 * O(n). We just need an arch-optimized log2() function now.
385 */
386 lg = 0;
387 for (i = *dest_ofs - do_sb_start - 1; i >= 0x10; i >>= 1)
388 lg++;
389
390 /* Get the phrase token into i. */
391 pt = le16_to_cpup((le16*)cb);
392
393 /*
394 * Calculate starting position of the byte sequence in
395 * the destination using the fact that p = (pt >> (12 - lg)) + 1
396 * and make sure we don't go too far back.
397 */
398 dp_back_addr = dp_addr - (pt >> (12 - lg)) - 1;
399 if (dp_back_addr < dp_sb_start)
400 goto return_overflow;
401
402 /* Now calculate the length of the byte sequence. */
403 length = (pt & (0xfff >> lg)) + 3;
404
405 /* Advance destination position and verify it is in range. */
406 *dest_ofs += length;
407 if (*dest_ofs > do_sb_end)
408 goto return_overflow;
409
410 /* The number of non-overlapping bytes. */
411 max_non_overlap = dp_addr - dp_back_addr;
412
413 if (length <= max_non_overlap) {
414 /* The byte sequence doesn't overlap, just copy it. */
415 memcpy(dp_addr, dp_back_addr, length);
416
417 /* Advance destination pointer. */
418 dp_addr += length;
419 } else {
420 /*
421 * The byte sequence does overlap, copy non-overlapping
422 * part and then do a slow byte by byte copy for the
423 * overlapping part. Also, advance the destination
424 * pointer.
425 */
426 memcpy(dp_addr, dp_back_addr, max_non_overlap);
427 dp_addr += max_non_overlap;
428 dp_back_addr += max_non_overlap;
429 length -= max_non_overlap;
430 while (length--)
431 *dp_addr++ = *dp_back_addr++;
432 }
433
434 /* Advance source position and continue with the next token. */
435 cb += 2;
436 }
437
438 /* No tokens left in the current tag. Continue with the next tag. */
439 goto do_next_tag;
440
441return_overflow:
442 ntfs_error(NULL, "Failed. Returning -EOVERFLOW.");
443 goto return_error;
444}
445
446/**
447 * ntfs_read_compressed_block - read a compressed block into the page cache
448 * @page: locked page in the compression block(s) we need to read
449 *
450 * When we are called the page has already been verified to be locked and the
451 * attribute is known to be non-resident, not encrypted, but compressed.
452 *
453 * 1. Determine which compression block(s) @page is in.
454 * 2. Get hold of all pages corresponding to this/these compression block(s).
455 * 3. Read the (first) compression block.
456 * 4. Decompress it into the corresponding pages.
457 * 5. Throw the compressed data away and proceed to 3. for the next compression
458 * block or return success if no more compression blocks left.
459 *
460 * Warning: We have to be careful what we do about existing pages. They might
461 * have been written to so that we would lose data if we were to just overwrite
462 * them with the out-of-date uncompressed data.
463 *
464 * FIXME: For PAGE_CACHE_SIZE > cb_size we are not doing the Right Thing(TM) at
465 * the end of the file I think. We need to detect this case and zero the out
466 * of bounds remainder of the page in question and mark it as handled. At the
467 * moment we would just return -EIO on such a page. This bug will only become
468 * apparent if pages are above 8kiB and the NTFS volume only uses 512 byte
469 * clusters so is probably not going to be seen by anyone. Still this should
470 * be fixed. (AIA)
471 *
472 * FIXME: Again for PAGE_CACHE_SIZE > cb_size we are screwing up both in
473 * handling sparse and compressed cbs. (AIA)
474 *
475 * FIXME: At the moment we don't do any zeroing out in the case that
476 * initialized_size is less than data_size. This should be safe because of the
477 * nature of the compression algorithm used. Just in case we check and output
478 * an error message in read inode if the two sizes are not equal for a
479 * compressed file. (AIA)
480 */
481int ntfs_read_compressed_block(struct page *page)
482{
Anton Altaparmakov36763672004-11-18 13:46:45 +0000483 loff_t i_size;
484 s64 initialized_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 struct address_space *mapping = page->mapping;
486 ntfs_inode *ni = NTFS_I(mapping->host);
487 ntfs_volume *vol = ni->vol;
488 struct super_block *sb = vol->sb;
489 runlist_element *rl;
Anton Altaparmakov36763672004-11-18 13:46:45 +0000490 unsigned long flags, block_size = sb->s_blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 unsigned char block_size_bits = sb->s_blocksize_bits;
492 u8 *cb, *cb_pos, *cb_end;
493 struct buffer_head **bhs;
494 unsigned long offset, index = page->index;
495 u32 cb_size = ni->itype.compressed.block_size;
496 u64 cb_size_mask = cb_size - 1UL;
497 VCN vcn;
498 LCN lcn;
499 /* The first wanted vcn (minimum alignment is PAGE_CACHE_SIZE). */
500 VCN start_vcn = (((s64)index << PAGE_CACHE_SHIFT) & ~cb_size_mask) >>
501 vol->cluster_size_bits;
502 /*
503 * The first vcn after the last wanted vcn (minumum alignment is again
504 * PAGE_CACHE_SIZE.
505 */
506 VCN end_vcn = ((((s64)(index + 1UL) << PAGE_CACHE_SHIFT) + cb_size - 1)
507 & ~cb_size_mask) >> vol->cluster_size_bits;
508 /* Number of compression blocks (cbs) in the wanted vcn range. */
509 unsigned int nr_cbs = (end_vcn - start_vcn) << vol->cluster_size_bits
510 >> ni->itype.compressed.block_size_bits;
511 /*
512 * Number of pages required to store the uncompressed data from all
513 * compression blocks (cbs) overlapping @page. Due to alignment
514 * guarantees of start_vcn and end_vcn, no need to round up here.
515 */
516 unsigned int nr_pages = (end_vcn - start_vcn) <<
517 vol->cluster_size_bits >> PAGE_CACHE_SHIFT;
518 unsigned int xpage, max_page, cur_page, cur_ofs, i;
519 unsigned int cb_clusters, cb_max_ofs;
520 int block, max_block, cb_max_page, bhs_size, nr_bhs, err = 0;
521 struct page **pages;
522 unsigned char xpage_done = 0;
523
524 ntfs_debug("Entering, page->index = 0x%lx, cb_size = 0x%x, nr_pages = "
525 "%i.", index, cb_size, nr_pages);
526 /*
527 * Bad things happen if we get here for anything that is not an
528 * unnamed $DATA attribute.
529 */
530 BUG_ON(ni->type != AT_DATA);
531 BUG_ON(ni->name_len);
532
533 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_NOFS);
534
535 /* Allocate memory to store the buffer heads we need. */
536 bhs_size = cb_size / block_size * sizeof(struct buffer_head *);
537 bhs = kmalloc(bhs_size, GFP_NOFS);
538
539 if (unlikely(!pages || !bhs)) {
540 kfree(bhs);
541 kfree(pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 unlock_page(page);
543 ntfs_error(vol->sb, "Failed to allocate internal buffers.");
544 return -ENOMEM;
545 }
546
547 /*
548 * We have already been given one page, this is the one we must do.
549 * Once again, the alignment guarantees keep it simple.
550 */
551 offset = start_vcn << vol->cluster_size_bits >> PAGE_CACHE_SHIFT;
552 xpage = index - offset;
553 pages[xpage] = page;
554 /*
555 * The remaining pages need to be allocated and inserted into the page
556 * cache, alignment guarantees keep all the below much simpler. (-8
557 */
Anton Altaparmakov36763672004-11-18 13:46:45 +0000558 read_lock_irqsave(&ni->size_lock, flags);
559 i_size = i_size_read(VFS_I(ni));
560 initialized_size = ni->initialized_size;
561 read_unlock_irqrestore(&ni->size_lock, flags);
562 max_page = ((i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) -
563 offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 if (nr_pages < max_page)
565 max_page = nr_pages;
566 for (i = 0; i < max_page; i++, offset++) {
567 if (i != xpage)
568 pages[i] = grab_cache_page_nowait(mapping, offset);
569 page = pages[i];
570 if (page) {
571 /*
572 * We only (re)read the page if it isn't already read
573 * in and/or dirty or we would be losing data or at
574 * least wasting our time.
575 */
576 if (!PageDirty(page) && (!PageUptodate(page) ||
577 PageError(page))) {
578 ClearPageError(page);
579 kmap(page);
580 continue;
581 }
582 unlock_page(page);
583 page_cache_release(page);
584 pages[i] = NULL;
585 }
586 }
587
588 /*
589 * We have the runlist, and all the destination pages we need to fill.
590 * Now read the first compression block.
591 */
592 cur_page = 0;
593 cur_ofs = 0;
594 cb_clusters = ni->itype.compressed.block_clusters;
595do_next_cb:
596 nr_cbs--;
597 nr_bhs = 0;
598
599 /* Read all cb buffer heads one cluster at a time. */
600 rl = NULL;
601 for (vcn = start_vcn, start_vcn += cb_clusters; vcn < start_vcn;
602 vcn++) {
603 BOOL is_retry = FALSE;
604
605 if (!rl) {
606lock_retry_remap:
607 down_read(&ni->runlist.lock);
608 rl = ni->runlist.rl;
609 }
610 if (likely(rl != NULL)) {
611 /* Seek to element containing target vcn. */
612 while (rl->length && rl[1].vcn <= vcn)
613 rl++;
614 lcn = ntfs_rl_vcn_to_lcn(rl, vcn);
615 } else
616 lcn = LCN_RL_NOT_MAPPED;
617 ntfs_debug("Reading vcn = 0x%llx, lcn = 0x%llx.",
618 (unsigned long long)vcn,
619 (unsigned long long)lcn);
620 if (lcn < 0) {
621 /*
622 * When we reach the first sparse cluster we have
623 * finished with the cb.
624 */
625 if (lcn == LCN_HOLE)
626 break;
627 if (is_retry || lcn != LCN_RL_NOT_MAPPED)
628 goto rl_err;
629 is_retry = TRUE;
630 /*
631 * Attempt to map runlist, dropping lock for the
632 * duration.
633 */
634 up_read(&ni->runlist.lock);
635 if (!ntfs_map_runlist(ni, vcn))
636 goto lock_retry_remap;
637 goto map_rl_err;
638 }
639 block = lcn << vol->cluster_size_bits >> block_size_bits;
640 /* Read the lcn from device in chunks of block_size bytes. */
641 max_block = block + (vol->cluster_size >> block_size_bits);
642 do {
643 ntfs_debug("block = 0x%x.", block);
644 if (unlikely(!(bhs[nr_bhs] = sb_getblk(sb, block))))
645 goto getblk_err;
646 nr_bhs++;
647 } while (++block < max_block);
648 }
649
650 /* Release the lock if we took it. */
651 if (rl)
652 up_read(&ni->runlist.lock);
653
654 /* Setup and initiate io on all buffer heads. */
655 for (i = 0; i < nr_bhs; i++) {
656 struct buffer_head *tbh = bhs[i];
657
658 if (unlikely(test_set_buffer_locked(tbh)))
659 continue;
660 if (unlikely(buffer_uptodate(tbh))) {
661 unlock_buffer(tbh);
662 continue;
663 }
664 get_bh(tbh);
665 tbh->b_end_io = end_buffer_read_sync;
666 submit_bh(READ, tbh);
667 }
668
669 /* Wait for io completion on all buffer heads. */
670 for (i = 0; i < nr_bhs; i++) {
671 struct buffer_head *tbh = bhs[i];
672
673 if (buffer_uptodate(tbh))
674 continue;
675 wait_on_buffer(tbh);
676 /*
677 * We need an optimization barrier here, otherwise we start
678 * hitting the below fixup code when accessing a loopback
679 * mounted ntfs partition. This indicates either there is a
680 * race condition in the loop driver or, more likely, gcc
681 * overoptimises the code without the barrier and it doesn't
682 * do the Right Thing(TM).
683 */
684 barrier();
685 if (unlikely(!buffer_uptodate(tbh))) {
686 ntfs_warning(vol->sb, "Buffer is unlocked but not "
687 "uptodate! Unplugging the disk queue "
688 "and rescheduling.");
689 get_bh(tbh);
690 blk_run_address_space(mapping);
691 schedule();
692 put_bh(tbh);
693 if (unlikely(!buffer_uptodate(tbh)))
694 goto read_err;
695 ntfs_warning(vol->sb, "Buffer is now uptodate. Good.");
696 }
697 }
698
699 /*
700 * Get the compression buffer. We must not sleep any more
701 * until we are finished with it.
702 */
703 spin_lock(&ntfs_cb_lock);
704 cb = ntfs_compression_buffer;
705
706 BUG_ON(!cb);
707
708 cb_pos = cb;
709 cb_end = cb + cb_size;
710
711 /* Copy the buffer heads into the contiguous buffer. */
712 for (i = 0; i < nr_bhs; i++) {
713 memcpy(cb_pos, bhs[i]->b_data, block_size);
714 cb_pos += block_size;
715 }
716
717 /* Just a precaution. */
718 if (cb_pos + 2 <= cb + cb_size)
719 *(u16*)cb_pos = 0;
720
721 /* Reset cb_pos back to the beginning. */
722 cb_pos = cb;
723
724 /* We now have both source (if present) and destination. */
725 ntfs_debug("Successfully read the compression block.");
726
727 /* The last page and maximum offset within it for the current cb. */
728 cb_max_page = (cur_page << PAGE_CACHE_SHIFT) + cur_ofs + cb_size;
729 cb_max_ofs = cb_max_page & ~PAGE_CACHE_MASK;
730 cb_max_page >>= PAGE_CACHE_SHIFT;
731
732 /* Catch end of file inside a compression block. */
733 if (cb_max_page > max_page)
734 cb_max_page = max_page;
735
736 if (vcn == start_vcn - cb_clusters) {
737 /* Sparse cb, zero out page range overlapping the cb. */
738 ntfs_debug("Found sparse compression block.");
739 /* We can sleep from now on, so we drop lock. */
740 spin_unlock(&ntfs_cb_lock);
741 if (cb_max_ofs)
742 cb_max_page--;
743 for (; cur_page < cb_max_page; cur_page++) {
744 page = pages[cur_page];
745 if (page) {
746 /*
747 * FIXME: Using clear_page() will become wrong
748 * when we get PAGE_CACHE_SIZE != PAGE_SIZE but
749 * for now there is no problem.
750 */
751 if (likely(!cur_ofs))
752 clear_page(page_address(page));
753 else
754 memset(page_address(page) + cur_ofs, 0,
755 PAGE_CACHE_SIZE -
756 cur_ofs);
757 flush_dcache_page(page);
758 kunmap(page);
759 SetPageUptodate(page);
760 unlock_page(page);
761 if (cur_page == xpage)
762 xpage_done = 1;
763 else
764 page_cache_release(page);
765 pages[cur_page] = NULL;
766 }
767 cb_pos += PAGE_CACHE_SIZE - cur_ofs;
768 cur_ofs = 0;
769 if (cb_pos >= cb_end)
770 break;
771 }
772 /* If we have a partial final page, deal with it now. */
773 if (cb_max_ofs && cb_pos < cb_end) {
774 page = pages[cur_page];
775 if (page)
776 memset(page_address(page) + cur_ofs, 0,
777 cb_max_ofs - cur_ofs);
778 /*
779 * No need to update cb_pos at this stage:
780 * cb_pos += cb_max_ofs - cur_ofs;
781 */
782 cur_ofs = cb_max_ofs;
783 }
784 } else if (vcn == start_vcn) {
785 /* We can't sleep so we need two stages. */
786 unsigned int cur2_page = cur_page;
787 unsigned int cur_ofs2 = cur_ofs;
788 u8 *cb_pos2 = cb_pos;
789
790 ntfs_debug("Found uncompressed compression block.");
791 /* Uncompressed cb, copy it to the destination pages. */
792 /*
793 * TODO: As a big optimization, we could detect this case
794 * before we read all the pages and use block_read_full_page()
795 * on all full pages instead (we still have to treat partial
796 * pages especially but at least we are getting rid of the
797 * synchronous io for the majority of pages.
798 * Or if we choose not to do the read-ahead/-behind stuff, we
799 * could just return block_read_full_page(pages[xpage]) as long
800 * as PAGE_CACHE_SIZE <= cb_size.
801 */
802 if (cb_max_ofs)
803 cb_max_page--;
804 /* First stage: copy data into destination pages. */
805 for (; cur_page < cb_max_page; cur_page++) {
806 page = pages[cur_page];
807 if (page)
808 memcpy(page_address(page) + cur_ofs, cb_pos,
809 PAGE_CACHE_SIZE - cur_ofs);
810 cb_pos += PAGE_CACHE_SIZE - cur_ofs;
811 cur_ofs = 0;
812 if (cb_pos >= cb_end)
813 break;
814 }
815 /* If we have a partial final page, deal with it now. */
816 if (cb_max_ofs && cb_pos < cb_end) {
817 page = pages[cur_page];
818 if (page)
819 memcpy(page_address(page) + cur_ofs, cb_pos,
820 cb_max_ofs - cur_ofs);
821 cb_pos += cb_max_ofs - cur_ofs;
822 cur_ofs = cb_max_ofs;
823 }
824 /* We can sleep from now on, so drop lock. */
825 spin_unlock(&ntfs_cb_lock);
826 /* Second stage: finalize pages. */
827 for (; cur2_page < cb_max_page; cur2_page++) {
828 page = pages[cur2_page];
829 if (page) {
830 /*
831 * If we are outside the initialized size, zero
832 * the out of bounds page range.
833 */
Anton Altaparmakov36763672004-11-18 13:46:45 +0000834 handle_bounds_compressed_page(page, i_size,
835 initialized_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 flush_dcache_page(page);
837 kunmap(page);
838 SetPageUptodate(page);
839 unlock_page(page);
840 if (cur2_page == xpage)
841 xpage_done = 1;
842 else
843 page_cache_release(page);
844 pages[cur2_page] = NULL;
845 }
846 cb_pos2 += PAGE_CACHE_SIZE - cur_ofs2;
847 cur_ofs2 = 0;
848 if (cb_pos2 >= cb_end)
849 break;
850 }
851 } else {
852 /* Compressed cb, decompress it into the destination page(s). */
853 unsigned int prev_cur_page = cur_page;
854
855 ntfs_debug("Found compressed compression block.");
856 err = ntfs_decompress(pages, &cur_page, &cur_ofs,
857 cb_max_page, cb_max_ofs, xpage, &xpage_done,
Anton Altaparmakov36763672004-11-18 13:46:45 +0000858 cb_pos, cb_size - (cb_pos - cb), i_size,
859 initialized_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 /*
861 * We can sleep from now on, lock already dropped by
862 * ntfs_decompress().
863 */
864 if (err) {
865 ntfs_error(vol->sb, "ntfs_decompress() failed in inode "
866 "0x%lx with error code %i. Skipping "
867 "this compression block.",
868 ni->mft_no, -err);
869 /* Release the unfinished pages. */
870 for (; prev_cur_page < cur_page; prev_cur_page++) {
871 page = pages[prev_cur_page];
872 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 flush_dcache_page(page);
874 kunmap(page);
875 unlock_page(page);
876 if (prev_cur_page != xpage)
877 page_cache_release(page);
878 pages[prev_cur_page] = NULL;
879 }
880 }
881 }
882 }
883
884 /* Release the buffer heads. */
885 for (i = 0; i < nr_bhs; i++)
886 brelse(bhs[i]);
887
888 /* Do we have more work to do? */
889 if (nr_cbs)
890 goto do_next_cb;
891
892 /* We no longer need the list of buffer heads. */
893 kfree(bhs);
894
895 /* Clean up if we have any pages left. Should never happen. */
896 for (cur_page = 0; cur_page < max_page; cur_page++) {
897 page = pages[cur_page];
898 if (page) {
899 ntfs_error(vol->sb, "Still have pages left! "
900 "Terminating them with extreme "
901 "prejudice. Inode 0x%lx, page index "
902 "0x%lx.", ni->mft_no, page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 flush_dcache_page(page);
904 kunmap(page);
905 unlock_page(page);
906 if (cur_page != xpage)
907 page_cache_release(page);
908 pages[cur_page] = NULL;
909 }
910 }
911
912 /* We no longer need the list of pages. */
913 kfree(pages);
914
915 /* If we have completed the requested page, we return success. */
916 if (likely(xpage_done))
917 return 0;
918
919 ntfs_debug("Failed. Returning error code %s.", err == -EOVERFLOW ?
920 "EOVERFLOW" : (!err ? "EIO" : "unkown error"));
921 return err < 0 ? err : -EIO;
922
923read_err:
924 ntfs_error(vol->sb, "IO error while reading compressed data.");
925 /* Release the buffer heads. */
926 for (i = 0; i < nr_bhs; i++)
927 brelse(bhs[i]);
928 goto err_out;
929
930map_rl_err:
931 ntfs_error(vol->sb, "ntfs_map_runlist() failed. Cannot read "
932 "compression block.");
933 goto err_out;
934
935rl_err:
936 up_read(&ni->runlist.lock);
937 ntfs_error(vol->sb, "ntfs_rl_vcn_to_lcn() failed. Cannot read "
938 "compression block.");
939 goto err_out;
940
941getblk_err:
942 up_read(&ni->runlist.lock);
943 ntfs_error(vol->sb, "getblk() failed. Cannot read compression block.");
944
945err_out:
946 kfree(bhs);
947 for (i = cur_page; i < max_page; i++) {
948 page = pages[i];
949 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 flush_dcache_page(page);
951 kunmap(page);
952 unlock_page(page);
953 if (i != xpage)
954 page_cache_release(page);
955 }
956 }
957 kfree(pages);
958 return -EIO;
959}