Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 1 | /** |
| 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 | */ |
| 37 | typedef 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 | */ |
| 60 | static u8 *ntfs_compression_buffer = NULL; |
| 61 | |
| 62 | /** |
| 63 | * ntfs_cb_lock - spinlock which protects ntfs_compression_buffer |
| 64 | */ |
| 65 | static 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 | */ |
| 74 | int 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 | */ |
| 89 | void 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 | */ |
| 99 | static void zero_partial_compressed_page(ntfs_inode *ni, struct page *page) |
| 100 | { |
| 101 | u8 *kp = page_address(page); |
| 102 | unsigned int kp_ofs; |
| 103 | |
| 104 | ntfs_debug("Zeroing page region outside initialized size."); |
| 105 | if (((s64)page->index << PAGE_CACHE_SHIFT) >= ni->initialized_size) { |
| 106 | /* |
| 107 | * FIXME: Using clear_page() will become wrong when we get |
| 108 | * PAGE_CACHE_SIZE != PAGE_SIZE but for now there is no problem. |
| 109 | */ |
| 110 | clear_page(kp); |
| 111 | return; |
| 112 | } |
| 113 | kp_ofs = ni->initialized_size & ~PAGE_CACHE_MASK; |
| 114 | memset(kp + kp_ofs, 0, PAGE_CACHE_SIZE - kp_ofs); |
| 115 | return; |
| 116 | } |
| 117 | |
| 118 | /** |
| 119 | * handle_bounds_compressed_page - test for&handle out of bounds compressed page |
| 120 | */ |
| 121 | static inline void handle_bounds_compressed_page(ntfs_inode *ni, |
| 122 | struct page *page) |
| 123 | { |
| 124 | if ((page->index >= (ni->initialized_size >> PAGE_CACHE_SHIFT)) && |
| 125 | (ni->initialized_size < VFS_I(ni)->i_size)) |
| 126 | zero_partial_compressed_page(ni, page); |
| 127 | return; |
| 128 | } |
| 129 | |
| 130 | /** |
| 131 | * ntfs_decompress - decompress a compression block into an array of pages |
| 132 | * @dest_pages: destination array of pages |
| 133 | * @dest_index: current index into @dest_pages (IN/OUT) |
| 134 | * @dest_ofs: current offset within @dest_pages[@dest_index] (IN/OUT) |
| 135 | * @dest_max_index: maximum index into @dest_pages (IN) |
| 136 | * @dest_max_ofs: maximum offset within @dest_pages[@dest_max_index] (IN) |
| 137 | * @xpage: the target page (-1 if none) (IN) |
| 138 | * @xpage_done: set to 1 if xpage was completed successfully (IN/OUT) |
| 139 | * @cb_start: compression block to decompress (IN) |
| 140 | * @cb_size: size of compression block @cb_start in bytes (IN) |
| 141 | * |
| 142 | * The caller must have disabled preemption. ntfs_decompress() reenables it when |
| 143 | * the critical section is finished. |
| 144 | * |
| 145 | * This decompresses the compression block @cb_start into the array of |
| 146 | * destination pages @dest_pages starting at index @dest_index into @dest_pages |
| 147 | * and at offset @dest_pos into the page @dest_pages[@dest_index]. |
| 148 | * |
| 149 | * When the page @dest_pages[@xpage] is completed, @xpage_done is set to 1. |
| 150 | * If xpage is -1 or @xpage has not been completed, @xpage_done is not modified. |
| 151 | * |
| 152 | * @cb_start is a pointer to the compression block which needs decompressing |
| 153 | * and @cb_size is the size of @cb_start in bytes (8-64kiB). |
| 154 | * |
| 155 | * Return 0 if success or -EOVERFLOW on error in the compressed stream. |
| 156 | * @xpage_done indicates whether the target page (@dest_pages[@xpage]) was |
| 157 | * completed during the decompression of the compression block (@cb_start). |
| 158 | * |
| 159 | * Warning: This function *REQUIRES* PAGE_CACHE_SIZE >= 4096 or it will blow up |
| 160 | * unpredicatbly! You have been warned! |
| 161 | * |
| 162 | * Note to hackers: This function may not sleep until it has finished accessing |
| 163 | * the compression block @cb_start as it is a per-CPU buffer. |
| 164 | */ |
| 165 | static int ntfs_decompress(struct page *dest_pages[], int *dest_index, |
| 166 | int *dest_ofs, const int dest_max_index, const int dest_max_ofs, |
| 167 | const int xpage, char *xpage_done, u8 *const cb_start, |
| 168 | const u32 cb_size) |
| 169 | { |
| 170 | /* |
| 171 | * Pointers into the compressed data, i.e. the compression block (cb), |
| 172 | * and the therein contained sub-blocks (sb). |
| 173 | */ |
| 174 | u8 *cb_end = cb_start + cb_size; /* End of cb. */ |
| 175 | u8 *cb = cb_start; /* Current position in cb. */ |
| 176 | u8 *cb_sb_start = cb; /* Beginning of the current sb in the cb. */ |
| 177 | u8 *cb_sb_end; /* End of current sb / beginning of next sb. */ |
| 178 | |
| 179 | /* Variables for uncompressed data / destination. */ |
| 180 | struct page *dp; /* Current destination page being worked on. */ |
| 181 | u8 *dp_addr; /* Current pointer into dp. */ |
| 182 | u8 *dp_sb_start; /* Start of current sub-block in dp. */ |
| 183 | u8 *dp_sb_end; /* End of current sb in dp (dp_sb_start + |
| 184 | NTFS_SB_SIZE). */ |
| 185 | u16 do_sb_start; /* @dest_ofs when starting this sub-block. */ |
| 186 | u16 do_sb_end; /* @dest_ofs of end of this sb (do_sb_start + |
| 187 | NTFS_SB_SIZE). */ |
| 188 | |
| 189 | /* Variables for tag and token parsing. */ |
| 190 | u8 tag; /* Current tag. */ |
| 191 | int token; /* Loop counter for the eight tokens in tag. */ |
| 192 | |
| 193 | /* Need this because we can't sleep, so need two stages. */ |
| 194 | int completed_pages[dest_max_index - *dest_index + 1]; |
| 195 | int nr_completed_pages = 0; |
| 196 | |
| 197 | /* Default error code. */ |
| 198 | int err = -EOVERFLOW; |
| 199 | |
| 200 | ntfs_debug("Entering, cb_size = 0x%x.", cb_size); |
| 201 | do_next_sb: |
| 202 | ntfs_debug("Beginning sub-block at offset = 0x%zx in the cb.", |
| 203 | cb - cb_start); |
| 204 | /* |
| 205 | * Have we reached the end of the compression block or the end of the |
| 206 | * decompressed data? The latter can happen for example if the current |
| 207 | * position in the compression block is one byte before its end so the |
| 208 | * first two checks do not detect it. |
| 209 | */ |
| 210 | if (cb == cb_end || !le16_to_cpup((le16*)cb) || |
| 211 | (*dest_index == dest_max_index && |
| 212 | *dest_ofs == dest_max_ofs)) { |
| 213 | int i; |
| 214 | |
| 215 | ntfs_debug("Completed. Returning success (0)."); |
| 216 | err = 0; |
| 217 | return_error: |
| 218 | /* We can sleep from now on, so we drop lock. */ |
| 219 | spin_unlock(&ntfs_cb_lock); |
| 220 | /* Second stage: finalize completed pages. */ |
| 221 | if (nr_completed_pages > 0) { |
| 222 | struct page *page = dest_pages[completed_pages[0]]; |
| 223 | ntfs_inode *ni = NTFS_I(page->mapping->host); |
| 224 | |
| 225 | for (i = 0; i < nr_completed_pages; i++) { |
| 226 | int di = completed_pages[i]; |
| 227 | |
| 228 | dp = dest_pages[di]; |
| 229 | /* |
| 230 | * If we are outside the initialized size, zero |
| 231 | * the out of bounds page range. |
| 232 | */ |
| 233 | handle_bounds_compressed_page(ni, dp); |
| 234 | flush_dcache_page(dp); |
| 235 | kunmap(dp); |
| 236 | SetPageUptodate(dp); |
| 237 | unlock_page(dp); |
| 238 | if (di == xpage) |
| 239 | *xpage_done = 1; |
| 240 | else |
| 241 | page_cache_release(dp); |
| 242 | dest_pages[di] = NULL; |
| 243 | } |
| 244 | } |
| 245 | return err; |
| 246 | } |
| 247 | |
| 248 | /* Setup offsets for the current sub-block destination. */ |
| 249 | do_sb_start = *dest_ofs; |
| 250 | do_sb_end = do_sb_start + NTFS_SB_SIZE; |
| 251 | |
| 252 | /* Check that we are still within allowed boundaries. */ |
| 253 | if (*dest_index == dest_max_index && do_sb_end > dest_max_ofs) |
| 254 | goto return_overflow; |
| 255 | |
| 256 | /* Does the minimum size of a compressed sb overflow valid range? */ |
| 257 | if (cb + 6 > cb_end) |
| 258 | goto return_overflow; |
| 259 | |
| 260 | /* Setup the current sub-block source pointers and validate range. */ |
| 261 | cb_sb_start = cb; |
| 262 | cb_sb_end = cb_sb_start + (le16_to_cpup((le16*)cb) & NTFS_SB_SIZE_MASK) |
| 263 | + 3; |
| 264 | if (cb_sb_end > cb_end) |
| 265 | goto return_overflow; |
| 266 | |
| 267 | /* Get the current destination page. */ |
| 268 | dp = dest_pages[*dest_index]; |
| 269 | if (!dp) { |
| 270 | /* No page present. Skip decompression of this sub-block. */ |
| 271 | cb = cb_sb_end; |
| 272 | |
| 273 | /* Advance destination position to next sub-block. */ |
| 274 | *dest_ofs = (*dest_ofs + NTFS_SB_SIZE) & ~PAGE_CACHE_MASK; |
| 275 | if (!*dest_ofs && (++*dest_index > dest_max_index)) |
| 276 | goto return_overflow; |
| 277 | goto do_next_sb; |
| 278 | } |
| 279 | |
| 280 | /* We have a valid destination page. Setup the destination pointers. */ |
| 281 | dp_addr = (u8*)page_address(dp) + do_sb_start; |
| 282 | |
| 283 | /* Now, we are ready to process the current sub-block (sb). */ |
| 284 | if (!(le16_to_cpup((le16*)cb) & NTFS_SB_IS_COMPRESSED)) { |
| 285 | ntfs_debug("Found uncompressed sub-block."); |
| 286 | /* This sb is not compressed, just copy it into destination. */ |
| 287 | |
| 288 | /* Advance source position to first data byte. */ |
| 289 | cb += 2; |
| 290 | |
| 291 | /* An uncompressed sb must be full size. */ |
| 292 | if (cb_sb_end - cb != NTFS_SB_SIZE) |
| 293 | goto return_overflow; |
| 294 | |
| 295 | /* Copy the block and advance the source position. */ |
| 296 | memcpy(dp_addr, cb, NTFS_SB_SIZE); |
| 297 | cb += NTFS_SB_SIZE; |
| 298 | |
| 299 | /* Advance destination position to next sub-block. */ |
| 300 | *dest_ofs += NTFS_SB_SIZE; |
| 301 | if (!(*dest_ofs &= ~PAGE_CACHE_MASK)) { |
| 302 | finalize_page: |
| 303 | /* |
| 304 | * First stage: add current page index to array of |
| 305 | * completed pages. |
| 306 | */ |
| 307 | completed_pages[nr_completed_pages++] = *dest_index; |
| 308 | if (++*dest_index > dest_max_index) |
| 309 | goto return_overflow; |
| 310 | } |
| 311 | goto do_next_sb; |
| 312 | } |
| 313 | ntfs_debug("Found compressed sub-block."); |
| 314 | /* This sb is compressed, decompress it into destination. */ |
| 315 | |
| 316 | /* Setup destination pointers. */ |
| 317 | dp_sb_start = dp_addr; |
| 318 | dp_sb_end = dp_sb_start + NTFS_SB_SIZE; |
| 319 | |
| 320 | /* Forward to the first tag in the sub-block. */ |
| 321 | cb += 2; |
| 322 | do_next_tag: |
| 323 | if (cb == cb_sb_end) { |
| 324 | /* Check if the decompressed sub-block was not full-length. */ |
| 325 | if (dp_addr < dp_sb_end) { |
| 326 | int nr_bytes = do_sb_end - *dest_ofs; |
| 327 | |
| 328 | ntfs_debug("Filling incomplete sub-block with " |
| 329 | "zeroes."); |
| 330 | /* Zero remainder and update destination position. */ |
| 331 | memset(dp_addr, 0, nr_bytes); |
| 332 | *dest_ofs += nr_bytes; |
| 333 | } |
| 334 | /* We have finished the current sub-block. */ |
| 335 | if (!(*dest_ofs &= ~PAGE_CACHE_MASK)) |
| 336 | goto finalize_page; |
| 337 | goto do_next_sb; |
| 338 | } |
| 339 | |
| 340 | /* Check we are still in range. */ |
| 341 | if (cb > cb_sb_end || dp_addr > dp_sb_end) |
| 342 | goto return_overflow; |
| 343 | |
| 344 | /* Get the next tag and advance to first token. */ |
| 345 | tag = *cb++; |
| 346 | |
| 347 | /* Parse the eight tokens described by the tag. */ |
| 348 | for (token = 0; token < 8; token++, tag >>= 1) { |
| 349 | u16 lg, pt, length, max_non_overlap; |
| 350 | register u16 i; |
| 351 | u8 *dp_back_addr; |
| 352 | |
| 353 | /* Check if we are done / still in range. */ |
| 354 | if (cb >= cb_sb_end || dp_addr > dp_sb_end) |
| 355 | break; |
| 356 | |
| 357 | /* Determine token type and parse appropriately.*/ |
| 358 | if ((tag & NTFS_TOKEN_MASK) == NTFS_SYMBOL_TOKEN) { |
| 359 | /* |
| 360 | * We have a symbol token, copy the symbol across, and |
| 361 | * advance the source and destination positions. |
| 362 | */ |
| 363 | *dp_addr++ = *cb++; |
| 364 | ++*dest_ofs; |
| 365 | |
| 366 | /* Continue with the next token. */ |
| 367 | continue; |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * We have a phrase token. Make sure it is not the first tag in |
| 372 | * the sb as this is illegal and would confuse the code below. |
| 373 | */ |
| 374 | if (dp_addr == dp_sb_start) |
| 375 | goto return_overflow; |
| 376 | |
| 377 | /* |
| 378 | * Determine the number of bytes to go back (p) and the number |
| 379 | * of bytes to copy (l). We use an optimized algorithm in which |
| 380 | * we first calculate log2(current destination position in sb), |
| 381 | * which allows determination of l and p in O(1) rather than |
| 382 | * O(n). We just need an arch-optimized log2() function now. |
| 383 | */ |
| 384 | lg = 0; |
| 385 | for (i = *dest_ofs - do_sb_start - 1; i >= 0x10; i >>= 1) |
| 386 | lg++; |
| 387 | |
| 388 | /* Get the phrase token into i. */ |
| 389 | pt = le16_to_cpup((le16*)cb); |
| 390 | |
| 391 | /* |
| 392 | * Calculate starting position of the byte sequence in |
| 393 | * the destination using the fact that p = (pt >> (12 - lg)) + 1 |
| 394 | * and make sure we don't go too far back. |
| 395 | */ |
| 396 | dp_back_addr = dp_addr - (pt >> (12 - lg)) - 1; |
| 397 | if (dp_back_addr < dp_sb_start) |
| 398 | goto return_overflow; |
| 399 | |
| 400 | /* Now calculate the length of the byte sequence. */ |
| 401 | length = (pt & (0xfff >> lg)) + 3; |
| 402 | |
| 403 | /* Advance destination position and verify it is in range. */ |
| 404 | *dest_ofs += length; |
| 405 | if (*dest_ofs > do_sb_end) |
| 406 | goto return_overflow; |
| 407 | |
| 408 | /* The number of non-overlapping bytes. */ |
| 409 | max_non_overlap = dp_addr - dp_back_addr; |
| 410 | |
| 411 | if (length <= max_non_overlap) { |
| 412 | /* The byte sequence doesn't overlap, just copy it. */ |
| 413 | memcpy(dp_addr, dp_back_addr, length); |
| 414 | |
| 415 | /* Advance destination pointer. */ |
| 416 | dp_addr += length; |
| 417 | } else { |
| 418 | /* |
| 419 | * The byte sequence does overlap, copy non-overlapping |
| 420 | * part and then do a slow byte by byte copy for the |
| 421 | * overlapping part. Also, advance the destination |
| 422 | * pointer. |
| 423 | */ |
| 424 | memcpy(dp_addr, dp_back_addr, max_non_overlap); |
| 425 | dp_addr += max_non_overlap; |
| 426 | dp_back_addr += max_non_overlap; |
| 427 | length -= max_non_overlap; |
| 428 | while (length--) |
| 429 | *dp_addr++ = *dp_back_addr++; |
| 430 | } |
| 431 | |
| 432 | /* Advance source position and continue with the next token. */ |
| 433 | cb += 2; |
| 434 | } |
| 435 | |
| 436 | /* No tokens left in the current tag. Continue with the next tag. */ |
| 437 | goto do_next_tag; |
| 438 | |
| 439 | return_overflow: |
| 440 | ntfs_error(NULL, "Failed. Returning -EOVERFLOW."); |
| 441 | goto return_error; |
| 442 | } |
| 443 | |
| 444 | /** |
| 445 | * ntfs_read_compressed_block - read a compressed block into the page cache |
| 446 | * @page: locked page in the compression block(s) we need to read |
| 447 | * |
| 448 | * When we are called the page has already been verified to be locked and the |
| 449 | * attribute is known to be non-resident, not encrypted, but compressed. |
| 450 | * |
| 451 | * 1. Determine which compression block(s) @page is in. |
| 452 | * 2. Get hold of all pages corresponding to this/these compression block(s). |
| 453 | * 3. Read the (first) compression block. |
| 454 | * 4. Decompress it into the corresponding pages. |
| 455 | * 5. Throw the compressed data away and proceed to 3. for the next compression |
| 456 | * block or return success if no more compression blocks left. |
| 457 | * |
| 458 | * Warning: We have to be careful what we do about existing pages. They might |
| 459 | * have been written to so that we would lose data if we were to just overwrite |
| 460 | * them with the out-of-date uncompressed data. |
| 461 | * |
| 462 | * FIXME: For PAGE_CACHE_SIZE > cb_size we are not doing the Right Thing(TM) at |
| 463 | * the end of the file I think. We need to detect this case and zero the out |
| 464 | * of bounds remainder of the page in question and mark it as handled. At the |
| 465 | * moment we would just return -EIO on such a page. This bug will only become |
| 466 | * apparent if pages are above 8kiB and the NTFS volume only uses 512 byte |
| 467 | * clusters so is probably not going to be seen by anyone. Still this should |
| 468 | * be fixed. (AIA) |
| 469 | * |
| 470 | * FIXME: Again for PAGE_CACHE_SIZE > cb_size we are screwing up both in |
| 471 | * handling sparse and compressed cbs. (AIA) |
| 472 | * |
| 473 | * FIXME: At the moment we don't do any zeroing out in the case that |
| 474 | * initialized_size is less than data_size. This should be safe because of the |
| 475 | * nature of the compression algorithm used. Just in case we check and output |
| 476 | * an error message in read inode if the two sizes are not equal for a |
| 477 | * compressed file. (AIA) |
| 478 | */ |
| 479 | int ntfs_read_compressed_block(struct page *page) |
| 480 | { |
| 481 | struct address_space *mapping = page->mapping; |
| 482 | ntfs_inode *ni = NTFS_I(mapping->host); |
| 483 | ntfs_volume *vol = ni->vol; |
| 484 | struct super_block *sb = vol->sb; |
| 485 | runlist_element *rl; |
| 486 | unsigned long block_size = sb->s_blocksize; |
| 487 | unsigned char block_size_bits = sb->s_blocksize_bits; |
| 488 | u8 *cb, *cb_pos, *cb_end; |
| 489 | struct buffer_head **bhs; |
| 490 | unsigned long offset, index = page->index; |
| 491 | u32 cb_size = ni->itype.compressed.block_size; |
| 492 | u64 cb_size_mask = cb_size - 1UL; |
| 493 | VCN vcn; |
| 494 | LCN lcn; |
| 495 | /* The first wanted vcn (minimum alignment is PAGE_CACHE_SIZE). */ |
| 496 | VCN start_vcn = (((s64)index << PAGE_CACHE_SHIFT) & ~cb_size_mask) >> |
| 497 | vol->cluster_size_bits; |
| 498 | /* |
| 499 | * The first vcn after the last wanted vcn (minumum alignment is again |
| 500 | * PAGE_CACHE_SIZE. |
| 501 | */ |
| 502 | VCN end_vcn = ((((s64)(index + 1UL) << PAGE_CACHE_SHIFT) + cb_size - 1) |
| 503 | & ~cb_size_mask) >> vol->cluster_size_bits; |
| 504 | /* Number of compression blocks (cbs) in the wanted vcn range. */ |
| 505 | unsigned int nr_cbs = (end_vcn - start_vcn) << vol->cluster_size_bits |
| 506 | >> ni->itype.compressed.block_size_bits; |
| 507 | /* |
| 508 | * Number of pages required to store the uncompressed data from all |
| 509 | * compression blocks (cbs) overlapping @page. Due to alignment |
| 510 | * guarantees of start_vcn and end_vcn, no need to round up here. |
| 511 | */ |
| 512 | unsigned int nr_pages = (end_vcn - start_vcn) << |
| 513 | vol->cluster_size_bits >> PAGE_CACHE_SHIFT; |
| 514 | unsigned int xpage, max_page, cur_page, cur_ofs, i; |
| 515 | unsigned int cb_clusters, cb_max_ofs; |
| 516 | int block, max_block, cb_max_page, bhs_size, nr_bhs, err = 0; |
| 517 | struct page **pages; |
| 518 | unsigned char xpage_done = 0; |
| 519 | |
| 520 | ntfs_debug("Entering, page->index = 0x%lx, cb_size = 0x%x, nr_pages = " |
| 521 | "%i.", index, cb_size, nr_pages); |
| 522 | /* |
| 523 | * Bad things happen if we get here for anything that is not an |
| 524 | * unnamed $DATA attribute. |
| 525 | */ |
| 526 | BUG_ON(ni->type != AT_DATA); |
| 527 | BUG_ON(ni->name_len); |
| 528 | |
| 529 | pages = kmalloc(nr_pages * sizeof(struct page *), GFP_NOFS); |
| 530 | |
| 531 | /* Allocate memory to store the buffer heads we need. */ |
| 532 | bhs_size = cb_size / block_size * sizeof(struct buffer_head *); |
| 533 | bhs = kmalloc(bhs_size, GFP_NOFS); |
| 534 | |
| 535 | if (unlikely(!pages || !bhs)) { |
| 536 | kfree(bhs); |
| 537 | kfree(pages); |
| 538 | SetPageError(page); |
| 539 | unlock_page(page); |
| 540 | ntfs_error(vol->sb, "Failed to allocate internal buffers."); |
| 541 | return -ENOMEM; |
| 542 | } |
| 543 | |
| 544 | /* |
| 545 | * We have already been given one page, this is the one we must do. |
| 546 | * Once again, the alignment guarantees keep it simple. |
| 547 | */ |
| 548 | offset = start_vcn << vol->cluster_size_bits >> PAGE_CACHE_SHIFT; |
| 549 | xpage = index - offset; |
| 550 | pages[xpage] = page; |
| 551 | /* |
| 552 | * The remaining pages need to be allocated and inserted into the page |
| 553 | * cache, alignment guarantees keep all the below much simpler. (-8 |
| 554 | */ |
| 555 | max_page = ((VFS_I(ni)->i_size + PAGE_CACHE_SIZE - 1) >> |
| 556 | PAGE_CACHE_SHIFT) - offset; |
| 557 | if (nr_pages < max_page) |
| 558 | max_page = nr_pages; |
| 559 | for (i = 0; i < max_page; i++, offset++) { |
| 560 | if (i != xpage) |
| 561 | pages[i] = grab_cache_page_nowait(mapping, offset); |
| 562 | page = pages[i]; |
| 563 | if (page) { |
| 564 | /* |
| 565 | * We only (re)read the page if it isn't already read |
| 566 | * in and/or dirty or we would be losing data or at |
| 567 | * least wasting our time. |
| 568 | */ |
| 569 | if (!PageDirty(page) && (!PageUptodate(page) || |
| 570 | PageError(page))) { |
| 571 | ClearPageError(page); |
| 572 | kmap(page); |
| 573 | continue; |
| 574 | } |
| 575 | unlock_page(page); |
| 576 | page_cache_release(page); |
| 577 | pages[i] = NULL; |
| 578 | } |
| 579 | } |
| 580 | |
| 581 | /* |
| 582 | * We have the runlist, and all the destination pages we need to fill. |
| 583 | * Now read the first compression block. |
| 584 | */ |
| 585 | cur_page = 0; |
| 586 | cur_ofs = 0; |
| 587 | cb_clusters = ni->itype.compressed.block_clusters; |
| 588 | do_next_cb: |
| 589 | nr_cbs--; |
| 590 | nr_bhs = 0; |
| 591 | |
| 592 | /* Read all cb buffer heads one cluster at a time. */ |
| 593 | rl = NULL; |
| 594 | for (vcn = start_vcn, start_vcn += cb_clusters; vcn < start_vcn; |
| 595 | vcn++) { |
| 596 | BOOL is_retry = FALSE; |
| 597 | |
| 598 | if (!rl) { |
| 599 | lock_retry_remap: |
| 600 | down_read(&ni->runlist.lock); |
| 601 | rl = ni->runlist.rl; |
| 602 | } |
| 603 | if (likely(rl != NULL)) { |
| 604 | /* Seek to element containing target vcn. */ |
| 605 | while (rl->length && rl[1].vcn <= vcn) |
| 606 | rl++; |
| 607 | lcn = ntfs_rl_vcn_to_lcn(rl, vcn); |
| 608 | } else |
| 609 | lcn = LCN_RL_NOT_MAPPED; |
| 610 | ntfs_debug("Reading vcn = 0x%llx, lcn = 0x%llx.", |
| 611 | (unsigned long long)vcn, |
| 612 | (unsigned long long)lcn); |
| 613 | if (lcn < 0) { |
| 614 | /* |
| 615 | * When we reach the first sparse cluster we have |
| 616 | * finished with the cb. |
| 617 | */ |
| 618 | if (lcn == LCN_HOLE) |
| 619 | break; |
| 620 | if (is_retry || lcn != LCN_RL_NOT_MAPPED) |
| 621 | goto rl_err; |
| 622 | is_retry = TRUE; |
| 623 | /* |
| 624 | * Attempt to map runlist, dropping lock for the |
| 625 | * duration. |
| 626 | */ |
| 627 | up_read(&ni->runlist.lock); |
| 628 | if (!ntfs_map_runlist(ni, vcn)) |
| 629 | goto lock_retry_remap; |
| 630 | goto map_rl_err; |
| 631 | } |
| 632 | block = lcn << vol->cluster_size_bits >> block_size_bits; |
| 633 | /* Read the lcn from device in chunks of block_size bytes. */ |
| 634 | max_block = block + (vol->cluster_size >> block_size_bits); |
| 635 | do { |
| 636 | ntfs_debug("block = 0x%x.", block); |
| 637 | if (unlikely(!(bhs[nr_bhs] = sb_getblk(sb, block)))) |
| 638 | goto getblk_err; |
| 639 | nr_bhs++; |
| 640 | } while (++block < max_block); |
| 641 | } |
| 642 | |
| 643 | /* Release the lock if we took it. */ |
| 644 | if (rl) |
| 645 | up_read(&ni->runlist.lock); |
| 646 | |
| 647 | /* Setup and initiate io on all buffer heads. */ |
| 648 | for (i = 0; i < nr_bhs; i++) { |
| 649 | struct buffer_head *tbh = bhs[i]; |
| 650 | |
| 651 | if (unlikely(test_set_buffer_locked(tbh))) |
| 652 | continue; |
| 653 | if (unlikely(buffer_uptodate(tbh))) { |
| 654 | unlock_buffer(tbh); |
| 655 | continue; |
| 656 | } |
| 657 | get_bh(tbh); |
| 658 | tbh->b_end_io = end_buffer_read_sync; |
| 659 | submit_bh(READ, tbh); |
| 660 | } |
| 661 | |
| 662 | /* Wait for io completion on all buffer heads. */ |
| 663 | for (i = 0; i < nr_bhs; i++) { |
| 664 | struct buffer_head *tbh = bhs[i]; |
| 665 | |
| 666 | if (buffer_uptodate(tbh)) |
| 667 | continue; |
| 668 | wait_on_buffer(tbh); |
| 669 | /* |
| 670 | * We need an optimization barrier here, otherwise we start |
| 671 | * hitting the below fixup code when accessing a loopback |
| 672 | * mounted ntfs partition. This indicates either there is a |
| 673 | * race condition in the loop driver or, more likely, gcc |
| 674 | * overoptimises the code without the barrier and it doesn't |
| 675 | * do the Right Thing(TM). |
| 676 | */ |
| 677 | barrier(); |
| 678 | if (unlikely(!buffer_uptodate(tbh))) { |
| 679 | ntfs_warning(vol->sb, "Buffer is unlocked but not " |
| 680 | "uptodate! Unplugging the disk queue " |
| 681 | "and rescheduling."); |
| 682 | get_bh(tbh); |
| 683 | blk_run_address_space(mapping); |
| 684 | schedule(); |
| 685 | put_bh(tbh); |
| 686 | if (unlikely(!buffer_uptodate(tbh))) |
| 687 | goto read_err; |
| 688 | ntfs_warning(vol->sb, "Buffer is now uptodate. Good."); |
| 689 | } |
| 690 | } |
| 691 | |
| 692 | /* |
| 693 | * Get the compression buffer. We must not sleep any more |
| 694 | * until we are finished with it. |
| 695 | */ |
| 696 | spin_lock(&ntfs_cb_lock); |
| 697 | cb = ntfs_compression_buffer; |
| 698 | |
| 699 | BUG_ON(!cb); |
| 700 | |
| 701 | cb_pos = cb; |
| 702 | cb_end = cb + cb_size; |
| 703 | |
| 704 | /* Copy the buffer heads into the contiguous buffer. */ |
| 705 | for (i = 0; i < nr_bhs; i++) { |
| 706 | memcpy(cb_pos, bhs[i]->b_data, block_size); |
| 707 | cb_pos += block_size; |
| 708 | } |
| 709 | |
| 710 | /* Just a precaution. */ |
| 711 | if (cb_pos + 2 <= cb + cb_size) |
| 712 | *(u16*)cb_pos = 0; |
| 713 | |
| 714 | /* Reset cb_pos back to the beginning. */ |
| 715 | cb_pos = cb; |
| 716 | |
| 717 | /* We now have both source (if present) and destination. */ |
| 718 | ntfs_debug("Successfully read the compression block."); |
| 719 | |
| 720 | /* The last page and maximum offset within it for the current cb. */ |
| 721 | cb_max_page = (cur_page << PAGE_CACHE_SHIFT) + cur_ofs + cb_size; |
| 722 | cb_max_ofs = cb_max_page & ~PAGE_CACHE_MASK; |
| 723 | cb_max_page >>= PAGE_CACHE_SHIFT; |
| 724 | |
| 725 | /* Catch end of file inside a compression block. */ |
| 726 | if (cb_max_page > max_page) |
| 727 | cb_max_page = max_page; |
| 728 | |
| 729 | if (vcn == start_vcn - cb_clusters) { |
| 730 | /* Sparse cb, zero out page range overlapping the cb. */ |
| 731 | ntfs_debug("Found sparse compression block."); |
| 732 | /* We can sleep from now on, so we drop lock. */ |
| 733 | spin_unlock(&ntfs_cb_lock); |
| 734 | if (cb_max_ofs) |
| 735 | cb_max_page--; |
| 736 | for (; cur_page < cb_max_page; cur_page++) { |
| 737 | page = pages[cur_page]; |
| 738 | if (page) { |
| 739 | /* |
| 740 | * FIXME: Using clear_page() will become wrong |
| 741 | * when we get PAGE_CACHE_SIZE != PAGE_SIZE but |
| 742 | * for now there is no problem. |
| 743 | */ |
| 744 | if (likely(!cur_ofs)) |
| 745 | clear_page(page_address(page)); |
| 746 | else |
| 747 | memset(page_address(page) + cur_ofs, 0, |
| 748 | PAGE_CACHE_SIZE - |
| 749 | cur_ofs); |
| 750 | flush_dcache_page(page); |
| 751 | kunmap(page); |
| 752 | SetPageUptodate(page); |
| 753 | unlock_page(page); |
| 754 | if (cur_page == xpage) |
| 755 | xpage_done = 1; |
| 756 | else |
| 757 | page_cache_release(page); |
| 758 | pages[cur_page] = NULL; |
| 759 | } |
| 760 | cb_pos += PAGE_CACHE_SIZE - cur_ofs; |
| 761 | cur_ofs = 0; |
| 762 | if (cb_pos >= cb_end) |
| 763 | break; |
| 764 | } |
| 765 | /* If we have a partial final page, deal with it now. */ |
| 766 | if (cb_max_ofs && cb_pos < cb_end) { |
| 767 | page = pages[cur_page]; |
| 768 | if (page) |
| 769 | memset(page_address(page) + cur_ofs, 0, |
| 770 | cb_max_ofs - cur_ofs); |
| 771 | /* |
| 772 | * No need to update cb_pos at this stage: |
| 773 | * cb_pos += cb_max_ofs - cur_ofs; |
| 774 | */ |
| 775 | cur_ofs = cb_max_ofs; |
| 776 | } |
| 777 | } else if (vcn == start_vcn) { |
| 778 | /* We can't sleep so we need two stages. */ |
| 779 | unsigned int cur2_page = cur_page; |
| 780 | unsigned int cur_ofs2 = cur_ofs; |
| 781 | u8 *cb_pos2 = cb_pos; |
| 782 | |
| 783 | ntfs_debug("Found uncompressed compression block."); |
| 784 | /* Uncompressed cb, copy it to the destination pages. */ |
| 785 | /* |
| 786 | * TODO: As a big optimization, we could detect this case |
| 787 | * before we read all the pages and use block_read_full_page() |
| 788 | * on all full pages instead (we still have to treat partial |
| 789 | * pages especially but at least we are getting rid of the |
| 790 | * synchronous io for the majority of pages. |
| 791 | * Or if we choose not to do the read-ahead/-behind stuff, we |
| 792 | * could just return block_read_full_page(pages[xpage]) as long |
| 793 | * as PAGE_CACHE_SIZE <= cb_size. |
| 794 | */ |
| 795 | if (cb_max_ofs) |
| 796 | cb_max_page--; |
| 797 | /* First stage: copy data into destination pages. */ |
| 798 | for (; cur_page < cb_max_page; cur_page++) { |
| 799 | page = pages[cur_page]; |
| 800 | if (page) |
| 801 | memcpy(page_address(page) + cur_ofs, cb_pos, |
| 802 | PAGE_CACHE_SIZE - cur_ofs); |
| 803 | cb_pos += PAGE_CACHE_SIZE - cur_ofs; |
| 804 | cur_ofs = 0; |
| 805 | if (cb_pos >= cb_end) |
| 806 | break; |
| 807 | } |
| 808 | /* If we have a partial final page, deal with it now. */ |
| 809 | if (cb_max_ofs && cb_pos < cb_end) { |
| 810 | page = pages[cur_page]; |
| 811 | if (page) |
| 812 | memcpy(page_address(page) + cur_ofs, cb_pos, |
| 813 | cb_max_ofs - cur_ofs); |
| 814 | cb_pos += cb_max_ofs - cur_ofs; |
| 815 | cur_ofs = cb_max_ofs; |
| 816 | } |
| 817 | /* We can sleep from now on, so drop lock. */ |
| 818 | spin_unlock(&ntfs_cb_lock); |
| 819 | /* Second stage: finalize pages. */ |
| 820 | for (; cur2_page < cb_max_page; cur2_page++) { |
| 821 | page = pages[cur2_page]; |
| 822 | if (page) { |
| 823 | /* |
| 824 | * If we are outside the initialized size, zero |
| 825 | * the out of bounds page range. |
| 826 | */ |
| 827 | handle_bounds_compressed_page(ni, page); |
| 828 | flush_dcache_page(page); |
| 829 | kunmap(page); |
| 830 | SetPageUptodate(page); |
| 831 | unlock_page(page); |
| 832 | if (cur2_page == xpage) |
| 833 | xpage_done = 1; |
| 834 | else |
| 835 | page_cache_release(page); |
| 836 | pages[cur2_page] = NULL; |
| 837 | } |
| 838 | cb_pos2 += PAGE_CACHE_SIZE - cur_ofs2; |
| 839 | cur_ofs2 = 0; |
| 840 | if (cb_pos2 >= cb_end) |
| 841 | break; |
| 842 | } |
| 843 | } else { |
| 844 | /* Compressed cb, decompress it into the destination page(s). */ |
| 845 | unsigned int prev_cur_page = cur_page; |
| 846 | |
| 847 | ntfs_debug("Found compressed compression block."); |
| 848 | err = ntfs_decompress(pages, &cur_page, &cur_ofs, |
| 849 | cb_max_page, cb_max_ofs, xpage, &xpage_done, |
| 850 | cb_pos, cb_size - (cb_pos - cb)); |
| 851 | /* |
| 852 | * We can sleep from now on, lock already dropped by |
| 853 | * ntfs_decompress(). |
| 854 | */ |
| 855 | if (err) { |
| 856 | ntfs_error(vol->sb, "ntfs_decompress() failed in inode " |
| 857 | "0x%lx with error code %i. Skipping " |
| 858 | "this compression block.", |
| 859 | ni->mft_no, -err); |
| 860 | /* Release the unfinished pages. */ |
| 861 | for (; prev_cur_page < cur_page; prev_cur_page++) { |
| 862 | page = pages[prev_cur_page]; |
| 863 | if (page) { |
| 864 | if (prev_cur_page == xpage && |
| 865 | !xpage_done) |
| 866 | SetPageError(page); |
| 867 | flush_dcache_page(page); |
| 868 | kunmap(page); |
| 869 | unlock_page(page); |
| 870 | if (prev_cur_page != xpage) |
| 871 | page_cache_release(page); |
| 872 | pages[prev_cur_page] = NULL; |
| 873 | } |
| 874 | } |
| 875 | } |
| 876 | } |
| 877 | |
| 878 | /* Release the buffer heads. */ |
| 879 | for (i = 0; i < nr_bhs; i++) |
| 880 | brelse(bhs[i]); |
| 881 | |
| 882 | /* Do we have more work to do? */ |
| 883 | if (nr_cbs) |
| 884 | goto do_next_cb; |
| 885 | |
| 886 | /* We no longer need the list of buffer heads. */ |
| 887 | kfree(bhs); |
| 888 | |
| 889 | /* Clean up if we have any pages left. Should never happen. */ |
| 890 | for (cur_page = 0; cur_page < max_page; cur_page++) { |
| 891 | page = pages[cur_page]; |
| 892 | if (page) { |
| 893 | ntfs_error(vol->sb, "Still have pages left! " |
| 894 | "Terminating them with extreme " |
| 895 | "prejudice. Inode 0x%lx, page index " |
| 896 | "0x%lx.", ni->mft_no, page->index); |
| 897 | if (cur_page == xpage && !xpage_done) |
| 898 | SetPageError(page); |
| 899 | flush_dcache_page(page); |
| 900 | kunmap(page); |
| 901 | unlock_page(page); |
| 902 | if (cur_page != xpage) |
| 903 | page_cache_release(page); |
| 904 | pages[cur_page] = NULL; |
| 905 | } |
| 906 | } |
| 907 | |
| 908 | /* We no longer need the list of pages. */ |
| 909 | kfree(pages); |
| 910 | |
| 911 | /* If we have completed the requested page, we return success. */ |
| 912 | if (likely(xpage_done)) |
| 913 | return 0; |
| 914 | |
| 915 | ntfs_debug("Failed. Returning error code %s.", err == -EOVERFLOW ? |
| 916 | "EOVERFLOW" : (!err ? "EIO" : "unkown error")); |
| 917 | return err < 0 ? err : -EIO; |
| 918 | |
| 919 | read_err: |
| 920 | ntfs_error(vol->sb, "IO error while reading compressed data."); |
| 921 | /* Release the buffer heads. */ |
| 922 | for (i = 0; i < nr_bhs; i++) |
| 923 | brelse(bhs[i]); |
| 924 | goto err_out; |
| 925 | |
| 926 | map_rl_err: |
| 927 | ntfs_error(vol->sb, "ntfs_map_runlist() failed. Cannot read " |
| 928 | "compression block."); |
| 929 | goto err_out; |
| 930 | |
| 931 | rl_err: |
| 932 | up_read(&ni->runlist.lock); |
| 933 | ntfs_error(vol->sb, "ntfs_rl_vcn_to_lcn() failed. Cannot read " |
| 934 | "compression block."); |
| 935 | goto err_out; |
| 936 | |
| 937 | getblk_err: |
| 938 | up_read(&ni->runlist.lock); |
| 939 | ntfs_error(vol->sb, "getblk() failed. Cannot read compression block."); |
| 940 | |
| 941 | err_out: |
| 942 | kfree(bhs); |
| 943 | for (i = cur_page; i < max_page; i++) { |
| 944 | page = pages[i]; |
| 945 | if (page) { |
| 946 | if (i == xpage && !xpage_done) |
| 947 | SetPageError(page); |
| 948 | flush_dcache_page(page); |
| 949 | kunmap(page); |
| 950 | unlock_page(page); |
| 951 | if (i != xpage) |
| 952 | page_cache_release(page); |
| 953 | } |
| 954 | } |
| 955 | kfree(pages); |
| 956 | return -EIO; |
| 957 | } |