1 /** 2 * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project. 3 * 4 * Copyright (c) 2001-2005 Anton Altaparmakov 5 * Copyright (c) 2002 Richard Russon 6 * 7 * This program/include file is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License as published 9 * by the Free Software Foundation; either version 2 of the License, or 10 * (at your option) any later version. 11 * 12 * This program/include file is distributed in the hope that it will be 13 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty 14 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 15 * GNU General Public License for more details. 16 * 17 * You should have received a copy of the GNU General Public License 18 * along with this program (in the main directory of the Linux-NTFS 19 * distribution in the file COPYING); if not, write to the Free Software 20 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 21 */ 22 23 #include <linux/buffer_head.h> 24 #include <linux/swap.h> 25 26 #include "attrib.h" 27 #include "aops.h" 28 #include "bitmap.h" 29 #include "debug.h" 30 #include "dir.h" 31 #include "lcnalloc.h" 32 #include "malloc.h" 33 #include "mft.h" 34 #include "ntfs.h" 35 36 /** 37 * map_mft_record_page - map the page in which a specific mft record resides 38 * @ni: ntfs inode whose mft record page to map 39 * 40 * This maps the page in which the mft record of the ntfs inode @ni is situated 41 * and returns a pointer to the mft record within the mapped page. 42 * 43 * Return value needs to be checked with IS_ERR() and if that is true PTR_ERR() 44 * contains the negative error code returned. 45 */ 46 static inline MFT_RECORD *map_mft_record_page(ntfs_inode *ni) 47 { 48 loff_t i_size; 49 ntfs_volume *vol = ni->vol; 50 struct inode *mft_vi = vol->mft_ino; 51 struct page *page; 52 unsigned long index, ofs, end_index; 53 54 BUG_ON(ni->page); 55 /* 56 * The index into the page cache and the offset within the page cache 57 * page of the wanted mft record. FIXME: We need to check for 58 * overflowing the unsigned long, but I don't think we would ever get 59 * here if the volume was that big... 60 */ 61 index = ni->mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; 62 ofs = (ni->mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 63 64 i_size = i_size_read(mft_vi); 65 /* The maximum valid index into the page cache for $MFT's data. */ 66 end_index = i_size >> PAGE_CACHE_SHIFT; 67 68 /* If the wanted index is out of bounds the mft record doesn't exist. */ 69 if (unlikely(index >= end_index)) { 70 if (index > end_index || (i_size & ~PAGE_CACHE_MASK) < ofs + 71 vol->mft_record_size) { 72 page = ERR_PTR(-ENOENT); 73 ntfs_error(vol->sb, "Attemt to read mft record 0x%lx, " 74 "which is beyond the end of the mft. " 75 "This is probably a bug in the ntfs " 76 "driver.", ni->mft_no); 77 goto err_out; 78 } 79 } 80 /* Read, map, and pin the page. */ 81 page = ntfs_map_page(mft_vi->i_mapping, index); 82 if (likely(!IS_ERR(page))) { 83 /* Catch multi sector transfer fixup errors. */ 84 if (likely(ntfs_is_mft_recordp((le32*)(page_address(page) + 85 ofs)))) { 86 ni->page = page; 87 ni->page_ofs = ofs; 88 return page_address(page) + ofs; 89 } 90 ntfs_error(vol->sb, "Mft record 0x%lx is corrupt. " 91 "Run chkdsk.", ni->mft_no); 92 ntfs_unmap_page(page); 93 page = ERR_PTR(-EIO); 94 } 95 err_out: 96 ni->page = NULL; 97 ni->page_ofs = 0; 98 return (void*)page; 99 } 100 101 /** 102 * map_mft_record - map, pin and lock an mft record 103 * @ni: ntfs inode whose MFT record to map 104 * 105 * First, take the mrec_lock semaphore. We might now be sleeping, while waiting 106 * for the semaphore if it was already locked by someone else. 107 * 108 * The page of the record is mapped using map_mft_record_page() before being 109 * returned to the caller. 110 * 111 * This in turn uses ntfs_map_page() to get the page containing the wanted mft 112 * record (it in turn calls read_cache_page() which reads it in from disk if 113 * necessary, increments the use count on the page so that it cannot disappear 114 * under us and returns a reference to the page cache page). 115 * 116 * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it 117 * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed 118 * and the post-read mst fixups on each mft record in the page have been 119 * performed, the page gets PG_uptodate set and PG_locked cleared (this is done 120 * in our asynchronous I/O completion handler end_buffer_read_mft_async()). 121 * ntfs_map_page() waits for PG_locked to become clear and checks if 122 * PG_uptodate is set and returns an error code if not. This provides 123 * sufficient protection against races when reading/using the page. 124 * 125 * However there is the write mapping to think about. Doing the above described 126 * checking here will be fine, because when initiating the write we will set 127 * PG_locked and clear PG_uptodate making sure nobody is touching the page 128 * contents. Doing the locking this way means that the commit to disk code in 129 * the page cache code paths is automatically sufficiently locked with us as 130 * we will not touch a page that has been locked or is not uptodate. The only 131 * locking problem then is them locking the page while we are accessing it. 132 * 133 * So that code will end up having to own the mrec_lock of all mft 134 * records/inodes present in the page before I/O can proceed. In that case we 135 * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be 136 * accessing anything without owning the mrec_lock semaphore. But we do need 137 * to use them because of the read_cache_page() invocation and the code becomes 138 * so much simpler this way that it is well worth it. 139 * 140 * The mft record is now ours and we return a pointer to it. You need to check 141 * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return 142 * the error code. 143 * 144 * NOTE: Caller is responsible for setting the mft record dirty before calling 145 * unmap_mft_record(). This is obviously only necessary if the caller really 146 * modified the mft record... 147 * Q: Do we want to recycle one of the VFS inode state bits instead? 148 * A: No, the inode ones mean we want to change the mft record, not we want to 149 * write it out. 150 */ 151 MFT_RECORD *map_mft_record(ntfs_inode *ni) 152 { 153 MFT_RECORD *m; 154 155 ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no); 156 157 /* Make sure the ntfs inode doesn't go away. */ 158 atomic_inc(&ni->count); 159 160 /* Serialize access to this mft record. */ 161 down(&ni->mrec_lock); 162 163 m = map_mft_record_page(ni); 164 if (likely(!IS_ERR(m))) 165 return m; 166 167 up(&ni->mrec_lock); 168 atomic_dec(&ni->count); 169 ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m)); 170 return m; 171 } 172 173 /** 174 * unmap_mft_record_page - unmap the page in which a specific mft record resides 175 * @ni: ntfs inode whose mft record page to unmap 176 * 177 * This unmaps the page in which the mft record of the ntfs inode @ni is 178 * situated and returns. This is a NOOP if highmem is not configured. 179 * 180 * The unmap happens via ntfs_unmap_page() which in turn decrements the use 181 * count on the page thus releasing it from the pinned state. 182 * 183 * We do not actually unmap the page from memory of course, as that will be 184 * done by the page cache code itself when memory pressure increases or 185 * whatever. 186 */ 187 static inline void unmap_mft_record_page(ntfs_inode *ni) 188 { 189 BUG_ON(!ni->page); 190 191 // TODO: If dirty, blah... 192 ntfs_unmap_page(ni->page); 193 ni->page = NULL; 194 ni->page_ofs = 0; 195 return; 196 } 197 198 /** 199 * unmap_mft_record - release a mapped mft record 200 * @ni: ntfs inode whose MFT record to unmap 201 * 202 * We release the page mapping and the mrec_lock mutex which unmaps the mft 203 * record and releases it for others to get hold of. We also release the ntfs 204 * inode by decrementing the ntfs inode reference count. 205 * 206 * NOTE: If caller has modified the mft record, it is imperative to set the mft 207 * record dirty BEFORE calling unmap_mft_record(). 208 */ 209 void unmap_mft_record(ntfs_inode *ni) 210 { 211 struct page *page = ni->page; 212 213 BUG_ON(!page); 214 215 ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no); 216 217 unmap_mft_record_page(ni); 218 up(&ni->mrec_lock); 219 atomic_dec(&ni->count); 220 /* 221 * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to 222 * ntfs_clear_extent_inode() in the extent inode case, and to the 223 * caller in the non-extent, yet pure ntfs inode case, to do the actual 224 * tear down of all structures and freeing of all allocated memory. 225 */ 226 return; 227 } 228 229 /** 230 * map_extent_mft_record - load an extent inode and attach it to its base 231 * @base_ni: base ntfs inode 232 * @mref: mft reference of the extent inode to load 233 * @ntfs_ino: on successful return, pointer to the ntfs_inode structure 234 * 235 * Load the extent mft record @mref and attach it to its base inode @base_ni. 236 * Return the mapped extent mft record if IS_ERR(result) is false. Otherwise 237 * PTR_ERR(result) gives the negative error code. 238 * 239 * On successful return, @ntfs_ino contains a pointer to the ntfs_inode 240 * structure of the mapped extent inode. 241 */ 242 MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref, 243 ntfs_inode **ntfs_ino) 244 { 245 MFT_RECORD *m; 246 ntfs_inode *ni = NULL; 247 ntfs_inode **extent_nis = NULL; 248 int i; 249 unsigned long mft_no = MREF(mref); 250 u16 seq_no = MSEQNO(mref); 251 BOOL destroy_ni = FALSE; 252 253 ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).", 254 mft_no, base_ni->mft_no); 255 /* Make sure the base ntfs inode doesn't go away. */ 256 atomic_inc(&base_ni->count); 257 /* 258 * Check if this extent inode has already been added to the base inode, 259 * in which case just return it. If not found, add it to the base 260 * inode before returning it. 261 */ 262 down(&base_ni->extent_lock); 263 if (base_ni->nr_extents > 0) { 264 extent_nis = base_ni->ext.extent_ntfs_inos; 265 for (i = 0; i < base_ni->nr_extents; i++) { 266 if (mft_no != extent_nis[i]->mft_no) 267 continue; 268 ni = extent_nis[i]; 269 /* Make sure the ntfs inode doesn't go away. */ 270 atomic_inc(&ni->count); 271 break; 272 } 273 } 274 if (likely(ni != NULL)) { 275 up(&base_ni->extent_lock); 276 atomic_dec(&base_ni->count); 277 /* We found the record; just have to map and return it. */ 278 m = map_mft_record(ni); 279 /* map_mft_record() has incremented this on success. */ 280 atomic_dec(&ni->count); 281 if (likely(!IS_ERR(m))) { 282 /* Verify the sequence number. */ 283 if (likely(le16_to_cpu(m->sequence_number) == seq_no)) { 284 ntfs_debug("Done 1."); 285 *ntfs_ino = ni; 286 return m; 287 } 288 unmap_mft_record(ni); 289 ntfs_error(base_ni->vol->sb, "Found stale extent mft " 290 "reference! Corrupt filesystem. " 291 "Run chkdsk."); 292 return ERR_PTR(-EIO); 293 } 294 map_err_out: 295 ntfs_error(base_ni->vol->sb, "Failed to map extent " 296 "mft record, error code %ld.", -PTR_ERR(m)); 297 return m; 298 } 299 /* Record wasn't there. Get a new ntfs inode and initialize it. */ 300 ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no); 301 if (unlikely(!ni)) { 302 up(&base_ni->extent_lock); 303 atomic_dec(&base_ni->count); 304 return ERR_PTR(-ENOMEM); 305 } 306 ni->vol = base_ni->vol; 307 ni->seq_no = seq_no; 308 ni->nr_extents = -1; 309 ni->ext.base_ntfs_ino = base_ni; 310 /* Now map the record. */ 311 m = map_mft_record(ni); 312 if (IS_ERR(m)) { 313 up(&base_ni->extent_lock); 314 atomic_dec(&base_ni->count); 315 ntfs_clear_extent_inode(ni); 316 goto map_err_out; 317 } 318 /* Verify the sequence number if it is present. */ 319 if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) { 320 ntfs_error(base_ni->vol->sb, "Found stale extent mft " 321 "reference! Corrupt filesystem. Run chkdsk."); 322 destroy_ni = TRUE; 323 m = ERR_PTR(-EIO); 324 goto unm_err_out; 325 } 326 /* Attach extent inode to base inode, reallocating memory if needed. */ 327 if (!(base_ni->nr_extents & 3)) { 328 ntfs_inode **tmp; 329 int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *); 330 331 tmp = (ntfs_inode **)kmalloc(new_size, GFP_NOFS); 332 if (unlikely(!tmp)) { 333 ntfs_error(base_ni->vol->sb, "Failed to allocate " 334 "internal buffer."); 335 destroy_ni = TRUE; 336 m = ERR_PTR(-ENOMEM); 337 goto unm_err_out; 338 } 339 if (base_ni->nr_extents) { 340 BUG_ON(!base_ni->ext.extent_ntfs_inos); 341 memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size - 342 4 * sizeof(ntfs_inode *)); 343 kfree(base_ni->ext.extent_ntfs_inos); 344 } 345 base_ni->ext.extent_ntfs_inos = tmp; 346 } 347 base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni; 348 up(&base_ni->extent_lock); 349 atomic_dec(&base_ni->count); 350 ntfs_debug("Done 2."); 351 *ntfs_ino = ni; 352 return m; 353 unm_err_out: 354 unmap_mft_record(ni); 355 up(&base_ni->extent_lock); 356 atomic_dec(&base_ni->count); 357 /* 358 * If the extent inode was not attached to the base inode we need to 359 * release it or we will leak memory. 360 */ 361 if (destroy_ni) 362 ntfs_clear_extent_inode(ni); 363 return m; 364 } 365 366 #ifdef NTFS_RW 367 368 /** 369 * __mark_mft_record_dirty - set the mft record and the page containing it dirty 370 * @ni: ntfs inode describing the mapped mft record 371 * 372 * Internal function. Users should call mark_mft_record_dirty() instead. 373 * 374 * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni, 375 * as well as the page containing the mft record, dirty. Also, mark the base 376 * vfs inode dirty. This ensures that any changes to the mft record are 377 * written out to disk. 378 * 379 * NOTE: We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES) 380 * on the base vfs inode, because even though file data may have been modified, 381 * it is dirty in the inode meta data rather than the data page cache of the 382 * inode, and thus there are no data pages that need writing out. Therefore, a 383 * full mark_inode_dirty() is overkill. A mark_inode_dirty_sync(), on the 384 * other hand, is not sufficient, because I_DIRTY_DATASYNC needs to be set to 385 * ensure ->write_inode is called from generic_osync_inode() and this needs to 386 * happen or the file data would not necessarily hit the device synchronously, 387 * even though the vfs inode has the O_SYNC flag set. Also, I_DIRTY_DATASYNC 388 * simply "feels" better than just I_DIRTY_SYNC, since the file data has not 389 * actually hit the block device yet, which is not what I_DIRTY_SYNC on its own 390 * would suggest. 391 */ 392 void __mark_mft_record_dirty(ntfs_inode *ni) 393 { 394 ntfs_inode *base_ni; 395 396 ntfs_debug("Entering for inode 0x%lx.", ni->mft_no); 397 BUG_ON(NInoAttr(ni)); 398 mark_ntfs_record_dirty(ni->page, ni->page_ofs); 399 /* Determine the base vfs inode and mark it dirty, too. */ 400 down(&ni->extent_lock); 401 if (likely(ni->nr_extents >= 0)) 402 base_ni = ni; 403 else 404 base_ni = ni->ext.base_ntfs_ino; 405 up(&ni->extent_lock); 406 __mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC); 407 } 408 409 static const char *ntfs_please_email = "Please email " 410 "linux-ntfs-dev@lists.sourceforge.net and say that you saw " 411 "this message. Thank you."; 412 413 /** 414 * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror 415 * @vol: ntfs volume on which the mft record to synchronize resides 416 * @mft_no: mft record number of mft record to synchronize 417 * @m: mapped, mst protected (extent) mft record to synchronize 418 * 419 * Write the mapped, mst protected (extent) mft record @m with mft record 420 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol, 421 * bypassing the page cache and the $MFTMirr inode itself. 422 * 423 * This function is only for use at umount time when the mft mirror inode has 424 * already been disposed off. We BUG() if we are called while the mft mirror 425 * inode is still attached to the volume. 426 * 427 * On success return 0. On error return -errno. 428 * 429 * NOTE: This function is not implemented yet as I am not convinced it can 430 * actually be triggered considering the sequence of commits we do in super.c:: 431 * ntfs_put_super(). But just in case we provide this place holder as the 432 * alternative would be either to BUG() or to get a NULL pointer dereference 433 * and Oops. 434 */ 435 static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol, 436 const unsigned long mft_no, MFT_RECORD *m) 437 { 438 BUG_ON(vol->mftmirr_ino); 439 ntfs_error(vol->sb, "Umount time mft mirror syncing is not " 440 "implemented yet. %s", ntfs_please_email); 441 return -EOPNOTSUPP; 442 } 443 444 /** 445 * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror 446 * @vol: ntfs volume on which the mft record to synchronize resides 447 * @mft_no: mft record number of mft record to synchronize 448 * @m: mapped, mst protected (extent) mft record to synchronize 449 * @sync: if true, wait for i/o completion 450 * 451 * Write the mapped, mst protected (extent) mft record @m with mft record 452 * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol. 453 * 454 * On success return 0. On error return -errno and set the volume errors flag 455 * in the ntfs volume @vol. 456 * 457 * NOTE: We always perform synchronous i/o and ignore the @sync parameter. 458 * 459 * TODO: If @sync is false, want to do truly asynchronous i/o, i.e. just 460 * schedule i/o via ->writepage or do it via kntfsd or whatever. 461 */ 462 int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no, 463 MFT_RECORD *m, int sync) 464 { 465 struct page *page; 466 unsigned int blocksize = vol->sb->s_blocksize; 467 int max_bhs = vol->mft_record_size / blocksize; 468 struct buffer_head *bhs[max_bhs]; 469 struct buffer_head *bh, *head; 470 u8 *kmirr; 471 runlist_element *rl; 472 unsigned int block_start, block_end, m_start, m_end, page_ofs; 473 int i_bhs, nr_bhs, err = 0; 474 unsigned char blocksize_bits = vol->mftmirr_ino->i_blkbits; 475 476 ntfs_debug("Entering for inode 0x%lx.", mft_no); 477 BUG_ON(!max_bhs); 478 if (unlikely(!vol->mftmirr_ino)) { 479 /* This could happen during umount... */ 480 err = ntfs_sync_mft_mirror_umount(vol, mft_no, m); 481 if (likely(!err)) 482 return err; 483 goto err_out; 484 } 485 /* Get the page containing the mirror copy of the mft record @m. */ 486 page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >> 487 (PAGE_CACHE_SHIFT - vol->mft_record_size_bits)); 488 if (IS_ERR(page)) { 489 ntfs_error(vol->sb, "Failed to map mft mirror page."); 490 err = PTR_ERR(page); 491 goto err_out; 492 } 493 lock_page(page); 494 BUG_ON(!PageUptodate(page)); 495 ClearPageUptodate(page); 496 /* Offset of the mft mirror record inside the page. */ 497 page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 498 /* The address in the page of the mirror copy of the mft record @m. */ 499 kmirr = page_address(page) + page_ofs; 500 /* Copy the mst protected mft record to the mirror. */ 501 memcpy(kmirr, m, vol->mft_record_size); 502 /* Create uptodate buffers if not present. */ 503 if (unlikely(!page_has_buffers(page))) { 504 struct buffer_head *tail; 505 506 bh = head = alloc_page_buffers(page, blocksize, 1); 507 do { 508 set_buffer_uptodate(bh); 509 tail = bh; 510 bh = bh->b_this_page; 511 } while (bh); 512 tail->b_this_page = head; 513 attach_page_buffers(page, head); 514 } 515 bh = head = page_buffers(page); 516 BUG_ON(!bh); 517 rl = NULL; 518 nr_bhs = 0; 519 block_start = 0; 520 m_start = kmirr - (u8*)page_address(page); 521 m_end = m_start + vol->mft_record_size; 522 do { 523 block_end = block_start + blocksize; 524 /* If the buffer is outside the mft record, skip it. */ 525 if (block_end <= m_start) 526 continue; 527 if (unlikely(block_start >= m_end)) 528 break; 529 /* Need to map the buffer if it is not mapped already. */ 530 if (unlikely(!buffer_mapped(bh))) { 531 VCN vcn; 532 LCN lcn; 533 unsigned int vcn_ofs; 534 535 bh->b_bdev = vol->sb->s_bdev; 536 /* Obtain the vcn and offset of the current block. */ 537 vcn = ((VCN)mft_no << vol->mft_record_size_bits) + 538 (block_start - m_start); 539 vcn_ofs = vcn & vol->cluster_size_mask; 540 vcn >>= vol->cluster_size_bits; 541 if (!rl) { 542 down_read(&NTFS_I(vol->mftmirr_ino)-> 543 runlist.lock); 544 rl = NTFS_I(vol->mftmirr_ino)->runlist.rl; 545 /* 546 * $MFTMirr always has the whole of its runlist 547 * in memory. 548 */ 549 BUG_ON(!rl); 550 } 551 /* Seek to element containing target vcn. */ 552 while (rl->length && rl[1].vcn <= vcn) 553 rl++; 554 lcn = ntfs_rl_vcn_to_lcn(rl, vcn); 555 /* For $MFTMirr, only lcn >= 0 is a successful remap. */ 556 if (likely(lcn >= 0)) { 557 /* Setup buffer head to correct block. */ 558 bh->b_blocknr = ((lcn << 559 vol->cluster_size_bits) + 560 vcn_ofs) >> blocksize_bits; 561 set_buffer_mapped(bh); 562 } else { 563 bh->b_blocknr = -1; 564 ntfs_error(vol->sb, "Cannot write mft mirror " 565 "record 0x%lx because its " 566 "location on disk could not " 567 "be determined (error code " 568 "%lli).", mft_no, 569 (long long)lcn); 570 err = -EIO; 571 } 572 } 573 BUG_ON(!buffer_uptodate(bh)); 574 BUG_ON(!nr_bhs && (m_start != block_start)); 575 BUG_ON(nr_bhs >= max_bhs); 576 bhs[nr_bhs++] = bh; 577 BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end)); 578 } while (block_start = block_end, (bh = bh->b_this_page) != head); 579 if (unlikely(rl)) 580 up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock); 581 if (likely(!err)) { 582 /* Lock buffers and start synchronous write i/o on them. */ 583 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { 584 struct buffer_head *tbh = bhs[i_bhs]; 585 586 if (unlikely(test_set_buffer_locked(tbh))) 587 BUG(); 588 BUG_ON(!buffer_uptodate(tbh)); 589 clear_buffer_dirty(tbh); 590 get_bh(tbh); 591 tbh->b_end_io = end_buffer_write_sync; 592 submit_bh(WRITE, tbh); 593 } 594 /* Wait on i/o completion of buffers. */ 595 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { 596 struct buffer_head *tbh = bhs[i_bhs]; 597 598 wait_on_buffer(tbh); 599 if (unlikely(!buffer_uptodate(tbh))) { 600 err = -EIO; 601 /* 602 * Set the buffer uptodate so the page and 603 * buffer states do not become out of sync. 604 */ 605 set_buffer_uptodate(tbh); 606 } 607 } 608 } else /* if (unlikely(err)) */ { 609 /* Clean the buffers. */ 610 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) 611 clear_buffer_dirty(bhs[i_bhs]); 612 } 613 /* Current state: all buffers are clean, unlocked, and uptodate. */ 614 /* Remove the mst protection fixups again. */ 615 post_write_mst_fixup((NTFS_RECORD*)kmirr); 616 flush_dcache_page(page); 617 SetPageUptodate(page); 618 unlock_page(page); 619 ntfs_unmap_page(page); 620 if (likely(!err)) { 621 ntfs_debug("Done."); 622 } else { 623 ntfs_error(vol->sb, "I/O error while writing mft mirror " 624 "record 0x%lx!", mft_no); 625 err_out: 626 ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error " 627 "code %i). Volume will be left marked dirty " 628 "on umount. Run ntfsfix on the partition " 629 "after umounting to correct this.", -err); 630 NVolSetErrors(vol); 631 } 632 return err; 633 } 634 635 /** 636 * write_mft_record_nolock - write out a mapped (extent) mft record 637 * @ni: ntfs inode describing the mapped (extent) mft record 638 * @m: mapped (extent) mft record to write 639 * @sync: if true, wait for i/o completion 640 * 641 * Write the mapped (extent) mft record @m described by the (regular or extent) 642 * ntfs inode @ni to backing store. If the mft record @m has a counterpart in 643 * the mft mirror, that is also updated. 644 * 645 * We only write the mft record if the ntfs inode @ni is dirty and the first 646 * buffer belonging to its mft record is dirty, too. We ignore the dirty state 647 * of subsequent buffers because we could have raced with 648 * fs/ntfs/aops.c::mark_ntfs_record_dirty(). 649 * 650 * On success, clean the mft record and return 0. On error, leave the mft 651 * record dirty and return -errno. The caller should call make_bad_inode() on 652 * the base inode to ensure no more access happens to this inode. We do not do 653 * it here as the caller may want to finish writing other extent mft records 654 * first to minimize on-disk metadata inconsistencies. 655 * 656 * NOTE: We always perform synchronous i/o and ignore the @sync parameter. 657 * However, if the mft record has a counterpart in the mft mirror and @sync is 658 * true, we write the mft record, wait for i/o completion, and only then write 659 * the mft mirror copy. This ensures that if the system crashes either the mft 660 * or the mft mirror will contain a self-consistent mft record @m. If @sync is 661 * false on the other hand, we start i/o on both and then wait for completion 662 * on them. This provides a speedup but no longer guarantees that you will end 663 * up with a self-consistent mft record in the case of a crash but if you asked 664 * for asynchronous writing you probably do not care about that anyway. 665 * 666 * TODO: If @sync is false, want to do truly asynchronous i/o, i.e. just 667 * schedule i/o via ->writepage or do it via kntfsd or whatever. 668 */ 669 int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync) 670 { 671 ntfs_volume *vol = ni->vol; 672 struct page *page = ni->page; 673 unsigned char blocksize_bits = vol->mft_ino->i_blkbits; 674 unsigned int blocksize = 1 << blocksize_bits; 675 int max_bhs = vol->mft_record_size / blocksize; 676 struct buffer_head *bhs[max_bhs]; 677 struct buffer_head *bh, *head; 678 runlist_element *rl; 679 unsigned int block_start, block_end, m_start, m_end; 680 int i_bhs, nr_bhs, err = 0; 681 682 ntfs_debug("Entering for inode 0x%lx.", ni->mft_no); 683 BUG_ON(NInoAttr(ni)); 684 BUG_ON(!max_bhs); 685 BUG_ON(!PageLocked(page)); 686 /* 687 * If the ntfs_inode is clean no need to do anything. If it is dirty, 688 * mark it as clean now so that it can be redirtied later on if needed. 689 * There is no danger of races since the caller is holding the locks 690 * for the mft record @m and the page it is in. 691 */ 692 if (!NInoTestClearDirty(ni)) 693 goto done; 694 bh = head = page_buffers(page); 695 BUG_ON(!bh); 696 rl = NULL; 697 nr_bhs = 0; 698 block_start = 0; 699 m_start = ni->page_ofs; 700 m_end = m_start + vol->mft_record_size; 701 do { 702 block_end = block_start + blocksize; 703 /* If the buffer is outside the mft record, skip it. */ 704 if (block_end <= m_start) 705 continue; 706 if (unlikely(block_start >= m_end)) 707 break; 708 /* 709 * If this block is not the first one in the record, we ignore 710 * the buffer's dirty state because we could have raced with a 711 * parallel mark_ntfs_record_dirty(). 712 */ 713 if (block_start == m_start) { 714 /* This block is the first one in the record. */ 715 if (!buffer_dirty(bh)) { 716 BUG_ON(nr_bhs); 717 /* Clean records are not written out. */ 718 break; 719 } 720 } 721 /* Need to map the buffer if it is not mapped already. */ 722 if (unlikely(!buffer_mapped(bh))) { 723 VCN vcn; 724 LCN lcn; 725 unsigned int vcn_ofs; 726 727 bh->b_bdev = vol->sb->s_bdev; 728 /* Obtain the vcn and offset of the current block. */ 729 vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) + 730 (block_start - m_start); 731 vcn_ofs = vcn & vol->cluster_size_mask; 732 vcn >>= vol->cluster_size_bits; 733 if (!rl) { 734 down_read(&NTFS_I(vol->mft_ino)->runlist.lock); 735 rl = NTFS_I(vol->mft_ino)->runlist.rl; 736 BUG_ON(!rl); 737 } 738 /* Seek to element containing target vcn. */ 739 while (rl->length && rl[1].vcn <= vcn) 740 rl++; 741 lcn = ntfs_rl_vcn_to_lcn(rl, vcn); 742 /* For $MFT, only lcn >= 0 is a successful remap. */ 743 if (likely(lcn >= 0)) { 744 /* Setup buffer head to correct block. */ 745 bh->b_blocknr = ((lcn << 746 vol->cluster_size_bits) + 747 vcn_ofs) >> blocksize_bits; 748 set_buffer_mapped(bh); 749 } else { 750 bh->b_blocknr = -1; 751 ntfs_error(vol->sb, "Cannot write mft record " 752 "0x%lx because its location " 753 "on disk could not be " 754 "determined (error code %lli).", 755 ni->mft_no, (long long)lcn); 756 err = -EIO; 757 } 758 } 759 BUG_ON(!buffer_uptodate(bh)); 760 BUG_ON(!nr_bhs && (m_start != block_start)); 761 BUG_ON(nr_bhs >= max_bhs); 762 bhs[nr_bhs++] = bh; 763 BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end)); 764 } while (block_start = block_end, (bh = bh->b_this_page) != head); 765 if (unlikely(rl)) 766 up_read(&NTFS_I(vol->mft_ino)->runlist.lock); 767 if (!nr_bhs) 768 goto done; 769 if (unlikely(err)) 770 goto cleanup_out; 771 /* Apply the mst protection fixups. */ 772 err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size); 773 if (err) { 774 ntfs_error(vol->sb, "Failed to apply mst fixups!"); 775 goto cleanup_out; 776 } 777 flush_dcache_mft_record_page(ni); 778 /* Lock buffers and start synchronous write i/o on them. */ 779 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { 780 struct buffer_head *tbh = bhs[i_bhs]; 781 782 if (unlikely(test_set_buffer_locked(tbh))) 783 BUG(); 784 BUG_ON(!buffer_uptodate(tbh)); 785 clear_buffer_dirty(tbh); 786 get_bh(tbh); 787 tbh->b_end_io = end_buffer_write_sync; 788 submit_bh(WRITE, tbh); 789 } 790 /* Synchronize the mft mirror now if not @sync. */ 791 if (!sync && ni->mft_no < vol->mftmirr_size) 792 ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync); 793 /* Wait on i/o completion of buffers. */ 794 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { 795 struct buffer_head *tbh = bhs[i_bhs]; 796 797 wait_on_buffer(tbh); 798 if (unlikely(!buffer_uptodate(tbh))) { 799 err = -EIO; 800 /* 801 * Set the buffer uptodate so the page and buffer 802 * states do not become out of sync. 803 */ 804 if (PageUptodate(page)) 805 set_buffer_uptodate(tbh); 806 } 807 } 808 /* If @sync, now synchronize the mft mirror. */ 809 if (sync && ni->mft_no < vol->mftmirr_size) 810 ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync); 811 /* Remove the mst protection fixups again. */ 812 post_write_mst_fixup((NTFS_RECORD*)m); 813 flush_dcache_mft_record_page(ni); 814 if (unlikely(err)) { 815 /* I/O error during writing. This is really bad! */ 816 ntfs_error(vol->sb, "I/O error while writing mft record " 817 "0x%lx! Marking base inode as bad. You " 818 "should unmount the volume and run chkdsk.", 819 ni->mft_no); 820 goto err_out; 821 } 822 done: 823 ntfs_debug("Done."); 824 return 0; 825 cleanup_out: 826 /* Clean the buffers. */ 827 for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) 828 clear_buffer_dirty(bhs[i_bhs]); 829 err_out: 830 /* 831 * Current state: all buffers are clean, unlocked, and uptodate. 832 * The caller should mark the base inode as bad so that no more i/o 833 * happens. ->clear_inode() will still be invoked so all extent inodes 834 * and other allocated memory will be freed. 835 */ 836 if (err == -ENOMEM) { 837 ntfs_error(vol->sb, "Not enough memory to write mft record. " 838 "Redirtying so the write is retried later."); 839 mark_mft_record_dirty(ni); 840 err = 0; 841 } else 842 NVolSetErrors(vol); 843 return err; 844 } 845 846 /** 847 * ntfs_may_write_mft_record - check if an mft record may be written out 848 * @vol: [IN] ntfs volume on which the mft record to check resides 849 * @mft_no: [IN] mft record number of the mft record to check 850 * @m: [IN] mapped mft record to check 851 * @locked_ni: [OUT] caller has to unlock this ntfs inode if one is returned 852 * 853 * Check if the mapped (base or extent) mft record @m with mft record number 854 * @mft_no belonging to the ntfs volume @vol may be written out. If necessary 855 * and possible the ntfs inode of the mft record is locked and the base vfs 856 * inode is pinned. The locked ntfs inode is then returned in @locked_ni. The 857 * caller is responsible for unlocking the ntfs inode and unpinning the base 858 * vfs inode. 859 * 860 * Return TRUE if the mft record may be written out and FALSE if not. 861 * 862 * The caller has locked the page and cleared the uptodate flag on it which 863 * means that we can safely write out any dirty mft records that do not have 864 * their inodes in icache as determined by ilookup5() as anyone 865 * opening/creating such an inode would block when attempting to map the mft 866 * record in read_cache_page() until we are finished with the write out. 867 * 868 * Here is a description of the tests we perform: 869 * 870 * If the inode is found in icache we know the mft record must be a base mft 871 * record. If it is dirty, we do not write it and return FALSE as the vfs 872 * inode write paths will result in the access times being updated which would 873 * cause the base mft record to be redirtied and written out again. (We know 874 * the access time update will modify the base mft record because Windows 875 * chkdsk complains if the standard information attribute is not in the base 876 * mft record.) 877 * 878 * If the inode is in icache and not dirty, we attempt to lock the mft record 879 * and if we find the lock was already taken, it is not safe to write the mft 880 * record and we return FALSE. 881 * 882 * If we manage to obtain the lock we have exclusive access to the mft record, 883 * which also allows us safe writeout of the mft record. We then set 884 * @locked_ni to the locked ntfs inode and return TRUE. 885 * 886 * Note we cannot just lock the mft record and sleep while waiting for the lock 887 * because this would deadlock due to lock reversal (normally the mft record is 888 * locked before the page is locked but we already have the page locked here 889 * when we try to lock the mft record). 890 * 891 * If the inode is not in icache we need to perform further checks. 892 * 893 * If the mft record is not a FILE record or it is a base mft record, we can 894 * safely write it and return TRUE. 895 * 896 * We now know the mft record is an extent mft record. We check if the inode 897 * corresponding to its base mft record is in icache and obtain a reference to 898 * it if it is. If it is not, we can safely write it and return TRUE. 899 * 900 * We now have the base inode for the extent mft record. We check if it has an 901 * ntfs inode for the extent mft record attached and if not it is safe to write 902 * the extent mft record and we return TRUE. 903 * 904 * The ntfs inode for the extent mft record is attached to the base inode so we 905 * attempt to lock the extent mft record and if we find the lock was already 906 * taken, it is not safe to write the extent mft record and we return FALSE. 907 * 908 * If we manage to obtain the lock we have exclusive access to the extent mft 909 * record, which also allows us safe writeout of the extent mft record. We 910 * set the ntfs inode of the extent mft record clean and then set @locked_ni to 911 * the now locked ntfs inode and return TRUE. 912 * 913 * Note, the reason for actually writing dirty mft records here and not just 914 * relying on the vfs inode dirty code paths is that we can have mft records 915 * modified without them ever having actual inodes in memory. Also we can have 916 * dirty mft records with clean ntfs inodes in memory. None of the described 917 * cases would result in the dirty mft records being written out if we only 918 * relied on the vfs inode dirty code paths. And these cases can really occur 919 * during allocation of new mft records and in particular when the 920 * initialized_size of the $MFT/$DATA attribute is extended and the new space 921 * is initialized using ntfs_mft_record_format(). The clean inode can then 922 * appear if the mft record is reused for a new inode before it got written 923 * out. 924 */ 925 BOOL ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no, 926 const MFT_RECORD *m, ntfs_inode **locked_ni) 927 { 928 struct super_block *sb = vol->sb; 929 struct inode *mft_vi = vol->mft_ino; 930 struct inode *vi; 931 ntfs_inode *ni, *eni, **extent_nis; 932 int i; 933 ntfs_attr na; 934 935 ntfs_debug("Entering for inode 0x%lx.", mft_no); 936 /* 937 * Normally we do not return a locked inode so set @locked_ni to NULL. 938 */ 939 BUG_ON(!locked_ni); 940 *locked_ni = NULL; 941 /* 942 * Check if the inode corresponding to this mft record is in the VFS 943 * inode cache and obtain a reference to it if it is. 944 */ 945 ntfs_debug("Looking for inode 0x%lx in icache.", mft_no); 946 na.mft_no = mft_no; 947 na.name = NULL; 948 na.name_len = 0; 949 na.type = AT_UNUSED; 950 /* 951 * Optimize inode 0, i.e. $MFT itself, since we have it in memory and 952 * we get here for it rather often. 953 */ 954 if (!mft_no) { 955 /* Balance the below iput(). */ 956 vi = igrab(mft_vi); 957 BUG_ON(vi != mft_vi); 958 } else { 959 /* 960 * Have to use ilookup5_nowait() since ilookup5() waits for the 961 * inode lock which causes ntfs to deadlock when a concurrent 962 * inode write via the inode dirty code paths and the page 963 * dirty code path of the inode dirty code path when writing 964 * $MFT occurs. 965 */ 966 vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na); 967 } 968 if (vi) { 969 ntfs_debug("Base inode 0x%lx is in icache.", mft_no); 970 /* The inode is in icache. */ 971 ni = NTFS_I(vi); 972 /* Take a reference to the ntfs inode. */ 973 atomic_inc(&ni->count); 974 /* If the inode is dirty, do not write this record. */ 975 if (NInoDirty(ni)) { 976 ntfs_debug("Inode 0x%lx is dirty, do not write it.", 977 mft_no); 978 atomic_dec(&ni->count); 979 iput(vi); 980 return FALSE; 981 } 982 ntfs_debug("Inode 0x%lx is not dirty.", mft_no); 983 /* The inode is not dirty, try to take the mft record lock. */ 984 if (unlikely(down_trylock(&ni->mrec_lock))) { 985 ntfs_debug("Mft record 0x%lx is already locked, do " 986 "not write it.", mft_no); 987 atomic_dec(&ni->count); 988 iput(vi); 989 return FALSE; 990 } 991 ntfs_debug("Managed to lock mft record 0x%lx, write it.", 992 mft_no); 993 /* 994 * The write has to occur while we hold the mft record lock so 995 * return the locked ntfs inode. 996 */ 997 *locked_ni = ni; 998 return TRUE; 999 } 1000 ntfs_debug("Inode 0x%lx is not in icache.", mft_no); 1001 /* The inode is not in icache. */ 1002 /* Write the record if it is not a mft record (type "FILE"). */ 1003 if (!ntfs_is_mft_record(m->magic)) { 1004 ntfs_debug("Mft record 0x%lx is not a FILE record, write it.", 1005 mft_no); 1006 return TRUE; 1007 } 1008 /* Write the mft record if it is a base inode. */ 1009 if (!m->base_mft_record) { 1010 ntfs_debug("Mft record 0x%lx is a base record, write it.", 1011 mft_no); 1012 return TRUE; 1013 } 1014 /* 1015 * This is an extent mft record. Check if the inode corresponding to 1016 * its base mft record is in icache and obtain a reference to it if it 1017 * is. 1018 */ 1019 na.mft_no = MREF_LE(m->base_mft_record); 1020 ntfs_debug("Mft record 0x%lx is an extent record. Looking for base " 1021 "inode 0x%lx in icache.", mft_no, na.mft_no); 1022 if (!na.mft_no) { 1023 /* Balance the below iput(). */ 1024 vi = igrab(mft_vi); 1025 BUG_ON(vi != mft_vi); 1026 } else 1027 vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode, 1028 &na); 1029 if (!vi) { 1030 /* 1031 * The base inode is not in icache, write this extent mft 1032 * record. 1033 */ 1034 ntfs_debug("Base inode 0x%lx is not in icache, write the " 1035 "extent record.", na.mft_no); 1036 return TRUE; 1037 } 1038 ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no); 1039 /* 1040 * The base inode is in icache. Check if it has the extent inode 1041 * corresponding to this extent mft record attached. 1042 */ 1043 ni = NTFS_I(vi); 1044 down(&ni->extent_lock); 1045 if (ni->nr_extents <= 0) { 1046 /* 1047 * The base inode has no attached extent inodes, write this 1048 * extent mft record. 1049 */ 1050 up(&ni->extent_lock); 1051 iput(vi); 1052 ntfs_debug("Base inode 0x%lx has no attached extent inodes, " 1053 "write the extent record.", na.mft_no); 1054 return TRUE; 1055 } 1056 /* Iterate over the attached extent inodes. */ 1057 extent_nis = ni->ext.extent_ntfs_inos; 1058 for (eni = NULL, i = 0; i < ni->nr_extents; ++i) { 1059 if (mft_no == extent_nis[i]->mft_no) { 1060 /* 1061 * Found the extent inode corresponding to this extent 1062 * mft record. 1063 */ 1064 eni = extent_nis[i]; 1065 break; 1066 } 1067 } 1068 /* 1069 * If the extent inode was not attached to the base inode, write this 1070 * extent mft record. 1071 */ 1072 if (!eni) { 1073 up(&ni->extent_lock); 1074 iput(vi); 1075 ntfs_debug("Extent inode 0x%lx is not attached to its base " 1076 "inode 0x%lx, write the extent record.", 1077 mft_no, na.mft_no); 1078 return TRUE; 1079 } 1080 ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.", 1081 mft_no, na.mft_no); 1082 /* Take a reference to the extent ntfs inode. */ 1083 atomic_inc(&eni->count); 1084 up(&ni->extent_lock); 1085 /* 1086 * Found the extent inode coresponding to this extent mft record. 1087 * Try to take the mft record lock. 1088 */ 1089 if (unlikely(down_trylock(&eni->mrec_lock))) { 1090 atomic_dec(&eni->count); 1091 iput(vi); 1092 ntfs_debug("Extent mft record 0x%lx is already locked, do " 1093 "not write it.", mft_no); 1094 return FALSE; 1095 } 1096 ntfs_debug("Managed to lock extent mft record 0x%lx, write it.", 1097 mft_no); 1098 if (NInoTestClearDirty(eni)) 1099 ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.", 1100 mft_no); 1101 /* 1102 * The write has to occur while we hold the mft record lock so return 1103 * the locked extent ntfs inode. 1104 */ 1105 *locked_ni = eni; 1106 return TRUE; 1107 } 1108 1109 static const char *es = " Leaving inconsistent metadata. Unmount and run " 1110 "chkdsk."; 1111 1112 /** 1113 * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name 1114 * @vol: volume on which to search for a free mft record 1115 * @base_ni: open base inode if allocating an extent mft record or NULL 1116 * 1117 * Search for a free mft record in the mft bitmap attribute on the ntfs volume 1118 * @vol. 1119 * 1120 * If @base_ni is NULL start the search at the default allocator position. 1121 * 1122 * If @base_ni is not NULL start the search at the mft record after the base 1123 * mft record @base_ni. 1124 * 1125 * Return the free mft record on success and -errno on error. An error code of 1126 * -ENOSPC means that there are no free mft records in the currently 1127 * initialized mft bitmap. 1128 * 1129 * Locking: Caller must hold vol->mftbmp_lock for writing. 1130 */ 1131 static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol, 1132 ntfs_inode *base_ni) 1133 { 1134 s64 pass_end, ll, data_pos, pass_start, ofs, bit; 1135 unsigned long flags; 1136 struct address_space *mftbmp_mapping; 1137 u8 *buf, *byte; 1138 struct page *page; 1139 unsigned int page_ofs, size; 1140 u8 pass, b; 1141 1142 ntfs_debug("Searching for free mft record in the currently " 1143 "initialized mft bitmap."); 1144 mftbmp_mapping = vol->mftbmp_ino->i_mapping; 1145 /* 1146 * Set the end of the pass making sure we do not overflow the mft 1147 * bitmap. 1148 */ 1149 read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags); 1150 pass_end = NTFS_I(vol->mft_ino)->allocated_size >> 1151 vol->mft_record_size_bits; 1152 read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags); 1153 read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); 1154 ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3; 1155 read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); 1156 if (pass_end > ll) 1157 pass_end = ll; 1158 pass = 1; 1159 if (!base_ni) 1160 data_pos = vol->mft_data_pos; 1161 else 1162 data_pos = base_ni->mft_no + 1; 1163 if (data_pos < 24) 1164 data_pos = 24; 1165 if (data_pos >= pass_end) { 1166 data_pos = 24; 1167 pass = 2; 1168 /* This happens on a freshly formatted volume. */ 1169 if (data_pos >= pass_end) 1170 return -ENOSPC; 1171 } 1172 pass_start = data_pos; 1173 ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, " 1174 "pass_end 0x%llx, data_pos 0x%llx.", pass, 1175 (long long)pass_start, (long long)pass_end, 1176 (long long)data_pos); 1177 /* Loop until a free mft record is found. */ 1178 for (; pass <= 2;) { 1179 /* Cap size to pass_end. */ 1180 ofs = data_pos >> 3; 1181 page_ofs = ofs & ~PAGE_CACHE_MASK; 1182 size = PAGE_CACHE_SIZE - page_ofs; 1183 ll = ((pass_end + 7) >> 3) - ofs; 1184 if (size > ll) 1185 size = ll; 1186 size <<= 3; 1187 /* 1188 * If we are still within the active pass, search the next page 1189 * for a zero bit. 1190 */ 1191 if (size) { 1192 page = ntfs_map_page(mftbmp_mapping, 1193 ofs >> PAGE_CACHE_SHIFT); 1194 if (unlikely(IS_ERR(page))) { 1195 ntfs_error(vol->sb, "Failed to read mft " 1196 "bitmap, aborting."); 1197 return PTR_ERR(page); 1198 } 1199 buf = (u8*)page_address(page) + page_ofs; 1200 bit = data_pos & 7; 1201 data_pos &= ~7ull; 1202 ntfs_debug("Before inner for loop: size 0x%x, " 1203 "data_pos 0x%llx, bit 0x%llx", size, 1204 (long long)data_pos, (long long)bit); 1205 for (; bit < size && data_pos + bit < pass_end; 1206 bit &= ~7ull, bit += 8) { 1207 byte = buf + (bit >> 3); 1208 if (*byte == 0xff) 1209 continue; 1210 b = ffz((unsigned long)*byte); 1211 if (b < 8 && b >= (bit & 7)) { 1212 ll = data_pos + (bit & ~7ull) + b; 1213 if (unlikely(ll > (1ll << 32))) { 1214 ntfs_unmap_page(page); 1215 return -ENOSPC; 1216 } 1217 *byte |= 1 << b; 1218 flush_dcache_page(page); 1219 set_page_dirty(page); 1220 ntfs_unmap_page(page); 1221 ntfs_debug("Done. (Found and " 1222 "allocated mft record " 1223 "0x%llx.)", 1224 (long long)ll); 1225 return ll; 1226 } 1227 } 1228 ntfs_debug("After inner for loop: size 0x%x, " 1229 "data_pos 0x%llx, bit 0x%llx", size, 1230 (long long)data_pos, (long long)bit); 1231 data_pos += size; 1232 ntfs_unmap_page(page); 1233 /* 1234 * If the end of the pass has not been reached yet, 1235 * continue searching the mft bitmap for a zero bit. 1236 */ 1237 if (data_pos < pass_end) 1238 continue; 1239 } 1240 /* Do the next pass. */ 1241 if (++pass == 2) { 1242 /* 1243 * Starting the second pass, in which we scan the first 1244 * part of the zone which we omitted earlier. 1245 */ 1246 pass_end = pass_start; 1247 data_pos = pass_start = 24; 1248 ntfs_debug("pass %i, pass_start 0x%llx, pass_end " 1249 "0x%llx.", pass, (long long)pass_start, 1250 (long long)pass_end); 1251 if (data_pos >= pass_end) 1252 break; 1253 } 1254 } 1255 /* No free mft records in currently initialized mft bitmap. */ 1256 ntfs_debug("Done. (No free mft records left in currently initialized " 1257 "mft bitmap.)"); 1258 return -ENOSPC; 1259 } 1260 1261 /** 1262 * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster 1263 * @vol: volume on which to extend the mft bitmap attribute 1264 * 1265 * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster. 1266 * 1267 * Note: Only changes allocated_size, i.e. does not touch initialized_size or 1268 * data_size. 1269 * 1270 * Return 0 on success and -errno on error. 1271 * 1272 * Locking: - Caller must hold vol->mftbmp_lock for writing. 1273 * - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for 1274 * writing and releases it before returning. 1275 * - This function takes vol->lcnbmp_lock for writing and releases it 1276 * before returning. 1277 */ 1278 static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol) 1279 { 1280 LCN lcn; 1281 s64 ll; 1282 unsigned long flags; 1283 struct page *page; 1284 ntfs_inode *mft_ni, *mftbmp_ni; 1285 runlist_element *rl, *rl2 = NULL; 1286 ntfs_attr_search_ctx *ctx = NULL; 1287 MFT_RECORD *mrec; 1288 ATTR_RECORD *a = NULL; 1289 int ret, mp_size; 1290 u32 old_alen = 0; 1291 u8 *b, tb; 1292 struct { 1293 u8 added_cluster:1; 1294 u8 added_run:1; 1295 u8 mp_rebuilt:1; 1296 } status = { 0, 0, 0 }; 1297 1298 ntfs_debug("Extending mft bitmap allocation."); 1299 mft_ni = NTFS_I(vol->mft_ino); 1300 mftbmp_ni = NTFS_I(vol->mftbmp_ino); 1301 /* 1302 * Determine the last lcn of the mft bitmap. The allocated size of the 1303 * mft bitmap cannot be zero so we are ok to do this. 1304 */ 1305 down_write(&mftbmp_ni->runlist.lock); 1306 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 1307 ll = mftbmp_ni->allocated_size; 1308 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1309 rl = ntfs_attr_find_vcn_nolock(mftbmp_ni, 1310 (ll - 1) >> vol->cluster_size_bits, TRUE); 1311 if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { 1312 up_write(&mftbmp_ni->runlist.lock); 1313 ntfs_error(vol->sb, "Failed to determine last allocated " 1314 "cluster of mft bitmap attribute."); 1315 if (!IS_ERR(rl)) 1316 ret = -EIO; 1317 else 1318 ret = PTR_ERR(rl); 1319 return ret; 1320 } 1321 lcn = rl->lcn + rl->length; 1322 ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.", 1323 (long long)lcn); 1324 /* 1325 * Attempt to get the cluster following the last allocated cluster by 1326 * hand as it may be in the MFT zone so the allocator would not give it 1327 * to us. 1328 */ 1329 ll = lcn >> 3; 1330 page = ntfs_map_page(vol->lcnbmp_ino->i_mapping, 1331 ll >> PAGE_CACHE_SHIFT); 1332 if (IS_ERR(page)) { 1333 up_write(&mftbmp_ni->runlist.lock); 1334 ntfs_error(vol->sb, "Failed to read from lcn bitmap."); 1335 return PTR_ERR(page); 1336 } 1337 b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK); 1338 tb = 1 << (lcn & 7ull); 1339 down_write(&vol->lcnbmp_lock); 1340 if (*b != 0xff && !(*b & tb)) { 1341 /* Next cluster is free, allocate it. */ 1342 *b |= tb; 1343 flush_dcache_page(page); 1344 set_page_dirty(page); 1345 up_write(&vol->lcnbmp_lock); 1346 ntfs_unmap_page(page); 1347 /* Update the mft bitmap runlist. */ 1348 rl->length++; 1349 rl[1].vcn++; 1350 status.added_cluster = 1; 1351 ntfs_debug("Appending one cluster to mft bitmap."); 1352 } else { 1353 up_write(&vol->lcnbmp_lock); 1354 ntfs_unmap_page(page); 1355 /* Allocate a cluster from the DATA_ZONE. */ 1356 rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE); 1357 if (IS_ERR(rl2)) { 1358 up_write(&mftbmp_ni->runlist.lock); 1359 ntfs_error(vol->sb, "Failed to allocate a cluster for " 1360 "the mft bitmap."); 1361 return PTR_ERR(rl2); 1362 } 1363 rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2); 1364 if (IS_ERR(rl)) { 1365 up_write(&mftbmp_ni->runlist.lock); 1366 ntfs_error(vol->sb, "Failed to merge runlists for mft " 1367 "bitmap."); 1368 if (ntfs_cluster_free_from_rl(vol, rl2)) { 1369 ntfs_error(vol->sb, "Failed to dealocate " 1370 "allocated cluster.%s", es); 1371 NVolSetErrors(vol); 1372 } 1373 ntfs_free(rl2); 1374 return PTR_ERR(rl); 1375 } 1376 mftbmp_ni->runlist.rl = rl; 1377 status.added_run = 1; 1378 ntfs_debug("Adding one run to mft bitmap."); 1379 /* Find the last run in the new runlist. */ 1380 for (; rl[1].length; rl++) 1381 ; 1382 } 1383 /* 1384 * Update the attribute record as well. Note: @rl is the last 1385 * (non-terminator) runlist element of mft bitmap. 1386 */ 1387 mrec = map_mft_record(mft_ni); 1388 if (IS_ERR(mrec)) { 1389 ntfs_error(vol->sb, "Failed to map mft record."); 1390 ret = PTR_ERR(mrec); 1391 goto undo_alloc; 1392 } 1393 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1394 if (unlikely(!ctx)) { 1395 ntfs_error(vol->sb, "Failed to get search context."); 1396 ret = -ENOMEM; 1397 goto undo_alloc; 1398 } 1399 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1400 mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, 1401 0, ctx); 1402 if (unlikely(ret)) { 1403 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1404 "mft bitmap attribute."); 1405 if (ret == -ENOENT) 1406 ret = -EIO; 1407 goto undo_alloc; 1408 } 1409 a = ctx->attr; 1410 ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); 1411 /* Search back for the previous last allocated cluster of mft bitmap. */ 1412 for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) { 1413 if (ll >= rl2->vcn) 1414 break; 1415 } 1416 BUG_ON(ll < rl2->vcn); 1417 BUG_ON(ll >= rl2->vcn + rl2->length); 1418 /* Get the size for the new mapping pairs array for this extent. */ 1419 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1); 1420 if (unlikely(mp_size <= 0)) { 1421 ntfs_error(vol->sb, "Get size for mapping pairs failed for " 1422 "mft bitmap attribute extent."); 1423 ret = mp_size; 1424 if (!ret) 1425 ret = -EIO; 1426 goto undo_alloc; 1427 } 1428 /* Expand the attribute record if necessary. */ 1429 old_alen = le32_to_cpu(a->length); 1430 ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + 1431 le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); 1432 if (unlikely(ret)) { 1433 if (ret != -ENOSPC) { 1434 ntfs_error(vol->sb, "Failed to resize attribute " 1435 "record for mft bitmap attribute."); 1436 goto undo_alloc; 1437 } 1438 // TODO: Deal with this by moving this extent to a new mft 1439 // record or by starting a new extent in a new mft record or by 1440 // moving other attributes out of this mft record. 1441 // Note: It will need to be a special mft record and if none of 1442 // those are available it gets rather complicated... 1443 ntfs_error(vol->sb, "Not enough space in this mft record to " 1444 "accomodate extended mft bitmap attribute " 1445 "extent. Cannot handle this yet."); 1446 ret = -EOPNOTSUPP; 1447 goto undo_alloc; 1448 } 1449 status.mp_rebuilt = 1; 1450 /* Generate the mapping pairs array directly into the attr record. */ 1451 ret = ntfs_mapping_pairs_build(vol, (u8*)a + 1452 le16_to_cpu(a->data.non_resident.mapping_pairs_offset), 1453 mp_size, rl2, ll, -1, NULL); 1454 if (unlikely(ret)) { 1455 ntfs_error(vol->sb, "Failed to build mapping pairs array for " 1456 "mft bitmap attribute."); 1457 goto undo_alloc; 1458 } 1459 /* Update the highest_vcn. */ 1460 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); 1461 /* 1462 * We now have extended the mft bitmap allocated_size by one cluster. 1463 * Reflect this in the ntfs_inode structure and the attribute record. 1464 */ 1465 if (a->data.non_resident.lowest_vcn) { 1466 /* 1467 * We are not in the first attribute extent, switch to it, but 1468 * first ensure the changes will make it to disk later. 1469 */ 1470 flush_dcache_mft_record_page(ctx->ntfs_ino); 1471 mark_mft_record_dirty(ctx->ntfs_ino); 1472 ntfs_attr_reinit_search_ctx(ctx); 1473 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1474 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 1475 0, ctx); 1476 if (unlikely(ret)) { 1477 ntfs_error(vol->sb, "Failed to find first attribute " 1478 "extent of mft bitmap attribute."); 1479 goto restore_undo_alloc; 1480 } 1481 a = ctx->attr; 1482 } 1483 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1484 mftbmp_ni->allocated_size += vol->cluster_size; 1485 a->data.non_resident.allocated_size = 1486 cpu_to_sle64(mftbmp_ni->allocated_size); 1487 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1488 /* Ensure the changes make it to disk. */ 1489 flush_dcache_mft_record_page(ctx->ntfs_ino); 1490 mark_mft_record_dirty(ctx->ntfs_ino); 1491 ntfs_attr_put_search_ctx(ctx); 1492 unmap_mft_record(mft_ni); 1493 up_write(&mftbmp_ni->runlist.lock); 1494 ntfs_debug("Done."); 1495 return 0; 1496 restore_undo_alloc: 1497 ntfs_attr_reinit_search_ctx(ctx); 1498 if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1499 mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, 1500 0, ctx)) { 1501 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1502 "mft bitmap attribute.%s", es); 1503 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1504 mftbmp_ni->allocated_size += vol->cluster_size; 1505 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1506 ntfs_attr_put_search_ctx(ctx); 1507 unmap_mft_record(mft_ni); 1508 up_write(&mftbmp_ni->runlist.lock); 1509 /* 1510 * The only thing that is now wrong is ->allocated_size of the 1511 * base attribute extent which chkdsk should be able to fix. 1512 */ 1513 NVolSetErrors(vol); 1514 return ret; 1515 } 1516 a = ctx->attr; 1517 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2); 1518 undo_alloc: 1519 if (status.added_cluster) { 1520 /* Truncate the last run in the runlist by one cluster. */ 1521 rl->length--; 1522 rl[1].vcn--; 1523 } else if (status.added_run) { 1524 lcn = rl->lcn; 1525 /* Remove the last run from the runlist. */ 1526 rl->lcn = rl[1].lcn; 1527 rl->length = 0; 1528 } 1529 /* Deallocate the cluster. */ 1530 down_write(&vol->lcnbmp_lock); 1531 if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { 1532 ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es); 1533 NVolSetErrors(vol); 1534 } 1535 up_write(&vol->lcnbmp_lock); 1536 if (status.mp_rebuilt) { 1537 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( 1538 a->data.non_resident.mapping_pairs_offset), 1539 old_alen - le16_to_cpu( 1540 a->data.non_resident.mapping_pairs_offset), 1541 rl2, ll, -1, NULL)) { 1542 ntfs_error(vol->sb, "Failed to restore mapping pairs " 1543 "array.%s", es); 1544 NVolSetErrors(vol); 1545 } 1546 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { 1547 ntfs_error(vol->sb, "Failed to restore attribute " 1548 "record.%s", es); 1549 NVolSetErrors(vol); 1550 } 1551 flush_dcache_mft_record_page(ctx->ntfs_ino); 1552 mark_mft_record_dirty(ctx->ntfs_ino); 1553 } 1554 if (ctx) 1555 ntfs_attr_put_search_ctx(ctx); 1556 if (!IS_ERR(mrec)) 1557 unmap_mft_record(mft_ni); 1558 up_write(&mftbmp_ni->runlist.lock); 1559 return ret; 1560 } 1561 1562 /** 1563 * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data 1564 * @vol: volume on which to extend the mft bitmap attribute 1565 * 1566 * Extend the initialized portion of the mft bitmap attribute on the ntfs 1567 * volume @vol by 8 bytes. 1568 * 1569 * Note: Only changes initialized_size and data_size, i.e. requires that 1570 * allocated_size is big enough to fit the new initialized_size. 1571 * 1572 * Return 0 on success and -error on error. 1573 * 1574 * Locking: Caller must hold vol->mftbmp_lock for writing. 1575 */ 1576 static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol) 1577 { 1578 s64 old_data_size, old_initialized_size; 1579 unsigned long flags; 1580 struct inode *mftbmp_vi; 1581 ntfs_inode *mft_ni, *mftbmp_ni; 1582 ntfs_attr_search_ctx *ctx; 1583 MFT_RECORD *mrec; 1584 ATTR_RECORD *a; 1585 int ret; 1586 1587 ntfs_debug("Extending mft bitmap initiailized (and data) size."); 1588 mft_ni = NTFS_I(vol->mft_ino); 1589 mftbmp_vi = vol->mftbmp_ino; 1590 mftbmp_ni = NTFS_I(mftbmp_vi); 1591 /* Get the attribute record. */ 1592 mrec = map_mft_record(mft_ni); 1593 if (IS_ERR(mrec)) { 1594 ntfs_error(vol->sb, "Failed to map mft record."); 1595 return PTR_ERR(mrec); 1596 } 1597 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1598 if (unlikely(!ctx)) { 1599 ntfs_error(vol->sb, "Failed to get search context."); 1600 ret = -ENOMEM; 1601 goto unm_err_out; 1602 } 1603 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1604 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx); 1605 if (unlikely(ret)) { 1606 ntfs_error(vol->sb, "Failed to find first attribute extent of " 1607 "mft bitmap attribute."); 1608 if (ret == -ENOENT) 1609 ret = -EIO; 1610 goto put_err_out; 1611 } 1612 a = ctx->attr; 1613 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1614 old_data_size = i_size_read(mftbmp_vi); 1615 old_initialized_size = mftbmp_ni->initialized_size; 1616 /* 1617 * We can simply update the initialized_size before filling the space 1618 * with zeroes because the caller is holding the mft bitmap lock for 1619 * writing which ensures that no one else is trying to access the data. 1620 */ 1621 mftbmp_ni->initialized_size += 8; 1622 a->data.non_resident.initialized_size = 1623 cpu_to_sle64(mftbmp_ni->initialized_size); 1624 if (mftbmp_ni->initialized_size > old_data_size) { 1625 i_size_write(mftbmp_vi, mftbmp_ni->initialized_size); 1626 a->data.non_resident.data_size = 1627 cpu_to_sle64(mftbmp_ni->initialized_size); 1628 } 1629 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1630 /* Ensure the changes make it to disk. */ 1631 flush_dcache_mft_record_page(ctx->ntfs_ino); 1632 mark_mft_record_dirty(ctx->ntfs_ino); 1633 ntfs_attr_put_search_ctx(ctx); 1634 unmap_mft_record(mft_ni); 1635 /* Initialize the mft bitmap attribute value with zeroes. */ 1636 ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0); 1637 if (likely(!ret)) { 1638 ntfs_debug("Done. (Wrote eight initialized bytes to mft " 1639 "bitmap."); 1640 return 0; 1641 } 1642 ntfs_error(vol->sb, "Failed to write to mft bitmap."); 1643 /* Try to recover from the error. */ 1644 mrec = map_mft_record(mft_ni); 1645 if (IS_ERR(mrec)) { 1646 ntfs_error(vol->sb, "Failed to map mft record.%s", es); 1647 NVolSetErrors(vol); 1648 return ret; 1649 } 1650 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1651 if (unlikely(!ctx)) { 1652 ntfs_error(vol->sb, "Failed to get search context.%s", es); 1653 NVolSetErrors(vol); 1654 goto unm_err_out; 1655 } 1656 if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1657 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) { 1658 ntfs_error(vol->sb, "Failed to find first attribute extent of " 1659 "mft bitmap attribute.%s", es); 1660 NVolSetErrors(vol); 1661 put_err_out: 1662 ntfs_attr_put_search_ctx(ctx); 1663 unm_err_out: 1664 unmap_mft_record(mft_ni); 1665 goto err_out; 1666 } 1667 a = ctx->attr; 1668 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1669 mftbmp_ni->initialized_size = old_initialized_size; 1670 a->data.non_resident.initialized_size = 1671 cpu_to_sle64(old_initialized_size); 1672 if (i_size_read(mftbmp_vi) != old_data_size) { 1673 i_size_write(mftbmp_vi, old_data_size); 1674 a->data.non_resident.data_size = cpu_to_sle64(old_data_size); 1675 } 1676 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1677 flush_dcache_mft_record_page(ctx->ntfs_ino); 1678 mark_mft_record_dirty(ctx->ntfs_ino); 1679 ntfs_attr_put_search_ctx(ctx); 1680 unmap_mft_record(mft_ni); 1681 #ifdef DEBUG 1682 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 1683 ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, " 1684 "data_size 0x%llx, initialized_size 0x%llx.", 1685 (long long)mftbmp_ni->allocated_size, 1686 (long long)i_size_read(mftbmp_vi), 1687 (long long)mftbmp_ni->initialized_size); 1688 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1689 #endif /* DEBUG */ 1690 err_out: 1691 return ret; 1692 } 1693 1694 /** 1695 * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute 1696 * @vol: volume on which to extend the mft data attribute 1697 * 1698 * Extend the mft data attribute on the ntfs volume @vol by 16 mft records 1699 * worth of clusters or if not enough space for this by one mft record worth 1700 * of clusters. 1701 * 1702 * Note: Only changes allocated_size, i.e. does not touch initialized_size or 1703 * data_size. 1704 * 1705 * Return 0 on success and -errno on error. 1706 * 1707 * Locking: - Caller must hold vol->mftbmp_lock for writing. 1708 * - This function takes NTFS_I(vol->mft_ino)->runlist.lock for 1709 * writing and releases it before returning. 1710 * - This function calls functions which take vol->lcnbmp_lock for 1711 * writing and release it before returning. 1712 */ 1713 static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol) 1714 { 1715 LCN lcn; 1716 VCN old_last_vcn; 1717 s64 min_nr, nr, ll; 1718 unsigned long flags; 1719 ntfs_inode *mft_ni; 1720 runlist_element *rl, *rl2; 1721 ntfs_attr_search_ctx *ctx = NULL; 1722 MFT_RECORD *mrec; 1723 ATTR_RECORD *a = NULL; 1724 int ret, mp_size; 1725 u32 old_alen = 0; 1726 BOOL mp_rebuilt = FALSE; 1727 1728 ntfs_debug("Extending mft data allocation."); 1729 mft_ni = NTFS_I(vol->mft_ino); 1730 /* 1731 * Determine the preferred allocation location, i.e. the last lcn of 1732 * the mft data attribute. The allocated size of the mft data 1733 * attribute cannot be zero so we are ok to do this. 1734 */ 1735 down_write(&mft_ni->runlist.lock); 1736 read_lock_irqsave(&mft_ni->size_lock, flags); 1737 ll = mft_ni->allocated_size; 1738 read_unlock_irqrestore(&mft_ni->size_lock, flags); 1739 rl = ntfs_attr_find_vcn_nolock(mft_ni, 1740 (ll - 1) >> vol->cluster_size_bits, TRUE); 1741 if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { 1742 up_write(&mft_ni->runlist.lock); 1743 ntfs_error(vol->sb, "Failed to determine last allocated " 1744 "cluster of mft data attribute."); 1745 if (!IS_ERR(rl)) 1746 ret = -EIO; 1747 else 1748 ret = PTR_ERR(rl); 1749 return ret; 1750 } 1751 lcn = rl->lcn + rl->length; 1752 ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn); 1753 /* Minimum allocation is one mft record worth of clusters. */ 1754 min_nr = vol->mft_record_size >> vol->cluster_size_bits; 1755 if (!min_nr) 1756 min_nr = 1; 1757 /* Want to allocate 16 mft records worth of clusters. */ 1758 nr = vol->mft_record_size << 4 >> vol->cluster_size_bits; 1759 if (!nr) 1760 nr = min_nr; 1761 /* Ensure we do not go above 2^32-1 mft records. */ 1762 read_lock_irqsave(&mft_ni->size_lock, flags); 1763 ll = mft_ni->allocated_size; 1764 read_unlock_irqrestore(&mft_ni->size_lock, flags); 1765 if (unlikely((ll + (nr << vol->cluster_size_bits)) >> 1766 vol->mft_record_size_bits >= (1ll << 32))) { 1767 nr = min_nr; 1768 if (unlikely((ll + (nr << vol->cluster_size_bits)) >> 1769 vol->mft_record_size_bits >= (1ll << 32))) { 1770 ntfs_warning(vol->sb, "Cannot allocate mft record " 1771 "because the maximum number of inodes " 1772 "(2^32) has already been reached."); 1773 up_write(&mft_ni->runlist.lock); 1774 return -ENOSPC; 1775 } 1776 } 1777 ntfs_debug("Trying mft data allocation with %s cluster count %lli.", 1778 nr > min_nr ? "default" : "minimal", (long long)nr); 1779 old_last_vcn = rl[1].vcn; 1780 do { 1781 rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE); 1782 if (likely(!IS_ERR(rl2))) 1783 break; 1784 if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) { 1785 ntfs_error(vol->sb, "Failed to allocate the minimal " 1786 "number of clusters (%lli) for the " 1787 "mft data attribute.", (long long)nr); 1788 up_write(&mft_ni->runlist.lock); 1789 return PTR_ERR(rl2); 1790 } 1791 /* 1792 * There is not enough space to do the allocation, but there 1793 * might be enough space to do a minimal allocation so try that 1794 * before failing. 1795 */ 1796 nr = min_nr; 1797 ntfs_debug("Retrying mft data allocation with minimal cluster " 1798 "count %lli.", (long long)nr); 1799 } while (1); 1800 rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2); 1801 if (IS_ERR(rl)) { 1802 up_write(&mft_ni->runlist.lock); 1803 ntfs_error(vol->sb, "Failed to merge runlists for mft data " 1804 "attribute."); 1805 if (ntfs_cluster_free_from_rl(vol, rl2)) { 1806 ntfs_error(vol->sb, "Failed to dealocate clusters " 1807 "from the mft data attribute.%s", es); 1808 NVolSetErrors(vol); 1809 } 1810 ntfs_free(rl2); 1811 return PTR_ERR(rl); 1812 } 1813 mft_ni->runlist.rl = rl; 1814 ntfs_debug("Allocated %lli clusters.", (long long)nr); 1815 /* Find the last run in the new runlist. */ 1816 for (; rl[1].length; rl++) 1817 ; 1818 /* Update the attribute record as well. */ 1819 mrec = map_mft_record(mft_ni); 1820 if (IS_ERR(mrec)) { 1821 ntfs_error(vol->sb, "Failed to map mft record."); 1822 ret = PTR_ERR(mrec); 1823 goto undo_alloc; 1824 } 1825 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1826 if (unlikely(!ctx)) { 1827 ntfs_error(vol->sb, "Failed to get search context."); 1828 ret = -ENOMEM; 1829 goto undo_alloc; 1830 } 1831 ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 1832 CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx); 1833 if (unlikely(ret)) { 1834 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1835 "mft data attribute."); 1836 if (ret == -ENOENT) 1837 ret = -EIO; 1838 goto undo_alloc; 1839 } 1840 a = ctx->attr; 1841 ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); 1842 /* Search back for the previous last allocated cluster of mft bitmap. */ 1843 for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) { 1844 if (ll >= rl2->vcn) 1845 break; 1846 } 1847 BUG_ON(ll < rl2->vcn); 1848 BUG_ON(ll >= rl2->vcn + rl2->length); 1849 /* Get the size for the new mapping pairs array for this extent. */ 1850 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1); 1851 if (unlikely(mp_size <= 0)) { 1852 ntfs_error(vol->sb, "Get size for mapping pairs failed for " 1853 "mft data attribute extent."); 1854 ret = mp_size; 1855 if (!ret) 1856 ret = -EIO; 1857 goto undo_alloc; 1858 } 1859 /* Expand the attribute record if necessary. */ 1860 old_alen = le32_to_cpu(a->length); 1861 ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + 1862 le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); 1863 if (unlikely(ret)) { 1864 if (ret != -ENOSPC) { 1865 ntfs_error(vol->sb, "Failed to resize attribute " 1866 "record for mft data attribute."); 1867 goto undo_alloc; 1868 } 1869 // TODO: Deal with this by moving this extent to a new mft 1870 // record or by starting a new extent in a new mft record or by 1871 // moving other attributes out of this mft record. 1872 // Note: Use the special reserved mft records and ensure that 1873 // this extent is not required to find the mft record in 1874 // question. If no free special records left we would need to 1875 // move an existing record away, insert ours in its place, and 1876 // then place the moved record into the newly allocated space 1877 // and we would then need to update all references to this mft 1878 // record appropriately. This is rather complicated... 1879 ntfs_error(vol->sb, "Not enough space in this mft record to " 1880 "accomodate extended mft data attribute " 1881 "extent. Cannot handle this yet."); 1882 ret = -EOPNOTSUPP; 1883 goto undo_alloc; 1884 } 1885 mp_rebuilt = TRUE; 1886 /* Generate the mapping pairs array directly into the attr record. */ 1887 ret = ntfs_mapping_pairs_build(vol, (u8*)a + 1888 le16_to_cpu(a->data.non_resident.mapping_pairs_offset), 1889 mp_size, rl2, ll, -1, NULL); 1890 if (unlikely(ret)) { 1891 ntfs_error(vol->sb, "Failed to build mapping pairs array of " 1892 "mft data attribute."); 1893 goto undo_alloc; 1894 } 1895 /* Update the highest_vcn. */ 1896 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); 1897 /* 1898 * We now have extended the mft data allocated_size by nr clusters. 1899 * Reflect this in the ntfs_inode structure and the attribute record. 1900 * @rl is the last (non-terminator) runlist element of mft data 1901 * attribute. 1902 */ 1903 if (a->data.non_resident.lowest_vcn) { 1904 /* 1905 * We are not in the first attribute extent, switch to it, but 1906 * first ensure the changes will make it to disk later. 1907 */ 1908 flush_dcache_mft_record_page(ctx->ntfs_ino); 1909 mark_mft_record_dirty(ctx->ntfs_ino); 1910 ntfs_attr_reinit_search_ctx(ctx); 1911 ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, 1912 mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, 1913 ctx); 1914 if (unlikely(ret)) { 1915 ntfs_error(vol->sb, "Failed to find first attribute " 1916 "extent of mft data attribute."); 1917 goto restore_undo_alloc; 1918 } 1919 a = ctx->attr; 1920 } 1921 write_lock_irqsave(&mft_ni->size_lock, flags); 1922 mft_ni->allocated_size += nr << vol->cluster_size_bits; 1923 a->data.non_resident.allocated_size = 1924 cpu_to_sle64(mft_ni->allocated_size); 1925 write_unlock_irqrestore(&mft_ni->size_lock, flags); 1926 /* Ensure the changes make it to disk. */ 1927 flush_dcache_mft_record_page(ctx->ntfs_ino); 1928 mark_mft_record_dirty(ctx->ntfs_ino); 1929 ntfs_attr_put_search_ctx(ctx); 1930 unmap_mft_record(mft_ni); 1931 up_write(&mft_ni->runlist.lock); 1932 ntfs_debug("Done."); 1933 return 0; 1934 restore_undo_alloc: 1935 ntfs_attr_reinit_search_ctx(ctx); 1936 if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 1937 CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) { 1938 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1939 "mft data attribute.%s", es); 1940 write_lock_irqsave(&mft_ni->size_lock, flags); 1941 mft_ni->allocated_size += nr << vol->cluster_size_bits; 1942 write_unlock_irqrestore(&mft_ni->size_lock, flags); 1943 ntfs_attr_put_search_ctx(ctx); 1944 unmap_mft_record(mft_ni); 1945 up_write(&mft_ni->runlist.lock); 1946 /* 1947 * The only thing that is now wrong is ->allocated_size of the 1948 * base attribute extent which chkdsk should be able to fix. 1949 */ 1950 NVolSetErrors(vol); 1951 return ret; 1952 } 1953 a = ctx->attr; 1954 a->data.non_resident.highest_vcn = cpu_to_sle64(old_last_vcn - 1); 1955 undo_alloc: 1956 if (ntfs_cluster_free(vol->mft_ino, old_last_vcn, -1, TRUE) < 0) { 1957 ntfs_error(vol->sb, "Failed to free clusters from mft data " 1958 "attribute.%s", es); 1959 NVolSetErrors(vol); 1960 } 1961 if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) { 1962 ntfs_error(vol->sb, "Failed to truncate mft data attribute " 1963 "runlist.%s", es); 1964 NVolSetErrors(vol); 1965 } 1966 if (mp_rebuilt) { 1967 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( 1968 a->data.non_resident.mapping_pairs_offset), 1969 old_alen - le16_to_cpu( 1970 a->data.non_resident.mapping_pairs_offset), 1971 rl2, ll, -1, NULL)) { 1972 ntfs_error(vol->sb, "Failed to restore mapping pairs " 1973 "array.%s", es); 1974 NVolSetErrors(vol); 1975 } 1976 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { 1977 ntfs_error(vol->sb, "Failed to restore attribute " 1978 "record.%s", es); 1979 NVolSetErrors(vol); 1980 } 1981 flush_dcache_mft_record_page(ctx->ntfs_ino); 1982 mark_mft_record_dirty(ctx->ntfs_ino); 1983 } 1984 if (ctx) 1985 ntfs_attr_put_search_ctx(ctx); 1986 if (!IS_ERR(mrec)) 1987 unmap_mft_record(mft_ni); 1988 up_write(&mft_ni->runlist.lock); 1989 return ret; 1990 } 1991 1992 /** 1993 * ntfs_mft_record_layout - layout an mft record into a memory buffer 1994 * @vol: volume to which the mft record will belong 1995 * @mft_no: mft reference specifying the mft record number 1996 * @m: destination buffer of size >= @vol->mft_record_size bytes 1997 * 1998 * Layout an empty, unused mft record with the mft record number @mft_no into 1999 * the buffer @m. The volume @vol is needed because the mft record structure 2000 * was modified in NTFS 3.1 so we need to know which volume version this mft 2001 * record will be used on. 2002 * 2003 * Return 0 on success and -errno on error. 2004 */ 2005 static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no, 2006 MFT_RECORD *m) 2007 { 2008 ATTR_RECORD *a; 2009 2010 ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); 2011 if (mft_no >= (1ll << 32)) { 2012 ntfs_error(vol->sb, "Mft record number 0x%llx exceeds " 2013 "maximum of 2^32.", (long long)mft_no); 2014 return -ERANGE; 2015 } 2016 /* Start by clearing the whole mft record to gives us a clean slate. */ 2017 memset(m, 0, vol->mft_record_size); 2018 /* Aligned to 2-byte boundary. */ 2019 if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver)) 2020 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1); 2021 else { 2022 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1); 2023 /* 2024 * Set the NTFS 3.1+ specific fields while we know that the 2025 * volume version is 3.1+. 2026 */ 2027 m->reserved = 0; 2028 m->mft_record_number = cpu_to_le32((u32)mft_no); 2029 } 2030 m->magic = magic_FILE; 2031 if (vol->mft_record_size >= NTFS_BLOCK_SIZE) 2032 m->usa_count = cpu_to_le16(vol->mft_record_size / 2033 NTFS_BLOCK_SIZE + 1); 2034 else { 2035 m->usa_count = cpu_to_le16(1); 2036 ntfs_warning(vol->sb, "Sector size is bigger than mft record " 2037 "size. Setting usa_count to 1. If chkdsk " 2038 "reports this as corruption, please email " 2039 "linux-ntfs-dev@lists.sourceforge.net stating " 2040 "that you saw this message and that the " 2041 "modified filesystem created was corrupt. " 2042 "Thank you."); 2043 } 2044 /* Set the update sequence number to 1. */ 2045 *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1); 2046 m->lsn = 0; 2047 m->sequence_number = cpu_to_le16(1); 2048 m->link_count = 0; 2049 /* 2050 * Place the attributes straight after the update sequence array, 2051 * aligned to 8-byte boundary. 2052 */ 2053 m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) + 2054 (le16_to_cpu(m->usa_count) << 1) + 7) & ~7); 2055 m->flags = 0; 2056 /* 2057 * Using attrs_offset plus eight bytes (for the termination attribute). 2058 * attrs_offset is already aligned to 8-byte boundary, so no need to 2059 * align again. 2060 */ 2061 m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8); 2062 m->bytes_allocated = cpu_to_le32(vol->mft_record_size); 2063 m->base_mft_record = 0; 2064 m->next_attr_instance = 0; 2065 /* Add the termination attribute. */ 2066 a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset)); 2067 a->type = AT_END; 2068 a->length = 0; 2069 ntfs_debug("Done."); 2070 return 0; 2071 } 2072 2073 /** 2074 * ntfs_mft_record_format - format an mft record on an ntfs volume 2075 * @vol: volume on which to format the mft record 2076 * @mft_no: mft record number to format 2077 * 2078 * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused 2079 * mft record into the appropriate place of the mft data attribute. This is 2080 * used when extending the mft data attribute. 2081 * 2082 * Return 0 on success and -errno on error. 2083 */ 2084 static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no) 2085 { 2086 loff_t i_size; 2087 struct inode *mft_vi = vol->mft_ino; 2088 struct page *page; 2089 MFT_RECORD *m; 2090 pgoff_t index, end_index; 2091 unsigned int ofs; 2092 int err; 2093 2094 ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); 2095 /* 2096 * The index into the page cache and the offset within the page cache 2097 * page of the wanted mft record. 2098 */ 2099 index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; 2100 ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 2101 /* The maximum valid index into the page cache for $MFT's data. */ 2102 i_size = i_size_read(mft_vi); 2103 end_index = i_size >> PAGE_CACHE_SHIFT; 2104 if (unlikely(index >= end_index)) { 2105 if (unlikely(index > end_index || ofs + vol->mft_record_size >= 2106 (i_size & ~PAGE_CACHE_MASK))) { 2107 ntfs_error(vol->sb, "Tried to format non-existing mft " 2108 "record 0x%llx.", (long long)mft_no); 2109 return -ENOENT; 2110 } 2111 } 2112 /* Read, map, and pin the page containing the mft record. */ 2113 page = ntfs_map_page(mft_vi->i_mapping, index); 2114 if (unlikely(IS_ERR(page))) { 2115 ntfs_error(vol->sb, "Failed to map page containing mft record " 2116 "to format 0x%llx.", (long long)mft_no); 2117 return PTR_ERR(page); 2118 } 2119 lock_page(page); 2120 BUG_ON(!PageUptodate(page)); 2121 ClearPageUptodate(page); 2122 m = (MFT_RECORD*)((u8*)page_address(page) + ofs); 2123 err = ntfs_mft_record_layout(vol, mft_no, m); 2124 if (unlikely(err)) { 2125 ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.", 2126 (long long)mft_no); 2127 SetPageUptodate(page); 2128 unlock_page(page); 2129 ntfs_unmap_page(page); 2130 return err; 2131 } 2132 flush_dcache_page(page); 2133 SetPageUptodate(page); 2134 unlock_page(page); 2135 /* 2136 * Make sure the mft record is written out to disk. We could use 2137 * ilookup5() to check if an inode is in icache and so on but this is 2138 * unnecessary as ntfs_writepage() will write the dirty record anyway. 2139 */ 2140 mark_ntfs_record_dirty(page, ofs); 2141 ntfs_unmap_page(page); 2142 ntfs_debug("Done."); 2143 return 0; 2144 } 2145 2146 /** 2147 * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume 2148 * @vol: [IN] volume on which to allocate the mft record 2149 * @mode: [IN] mode if want a file or directory, i.e. base inode or 0 2150 * @base_ni: [IN] open base inode if allocating an extent mft record or NULL 2151 * @mrec: [OUT] on successful return this is the mapped mft record 2152 * 2153 * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol. 2154 * 2155 * If @base_ni is NULL make the mft record a base mft record, i.e. a file or 2156 * direvctory inode, and allocate it at the default allocator position. In 2157 * this case @mode is the file mode as given to us by the caller. We in 2158 * particular use @mode to distinguish whether a file or a directory is being 2159 * created (S_IFDIR(mode) and S_IFREG(mode), respectively). 2160 * 2161 * If @base_ni is not NULL make the allocated mft record an extent record, 2162 * allocate it starting at the mft record after the base mft record and attach 2163 * the allocated and opened ntfs inode to the base inode @base_ni. In this 2164 * case @mode must be 0 as it is meaningless for extent inodes. 2165 * 2166 * You need to check the return value with IS_ERR(). If false, the function 2167 * was successful and the return value is the now opened ntfs inode of the 2168 * allocated mft record. *@mrec is then set to the allocated, mapped, pinned, 2169 * and locked mft record. If IS_ERR() is true, the function failed and the 2170 * error code is obtained from PTR_ERR(return value). *@mrec is undefined in 2171 * this case. 2172 * 2173 * Allocation strategy: 2174 * 2175 * To find a free mft record, we scan the mft bitmap for a zero bit. To 2176 * optimize this we start scanning at the place specified by @base_ni or if 2177 * @base_ni is NULL we start where we last stopped and we perform wrap around 2178 * when we reach the end. Note, we do not try to allocate mft records below 2179 * number 24 because numbers 0 to 15 are the defined system files anyway and 16 2180 * to 24 are special in that they are used for storing extension mft records 2181 * for the $DATA attribute of $MFT. This is required to avoid the possibility 2182 * of creating a runlist with a circular dependency which once written to disk 2183 * can never be read in again. Windows will only use records 16 to 24 for 2184 * normal files if the volume is completely out of space. We never use them 2185 * which means that when the volume is really out of space we cannot create any 2186 * more files while Windows can still create up to 8 small files. We can start 2187 * doing this at some later time, it does not matter much for now. 2188 * 2189 * When scanning the mft bitmap, we only search up to the last allocated mft 2190 * record. If there are no free records left in the range 24 to number of 2191 * allocated mft records, then we extend the $MFT/$DATA attribute in order to 2192 * create free mft records. We extend the allocated size of $MFT/$DATA by 16 2193 * records at a time or one cluster, if cluster size is above 16kiB. If there 2194 * is not sufficient space to do this, we try to extend by a single mft record 2195 * or one cluster, if cluster size is above the mft record size. 2196 * 2197 * No matter how many mft records we allocate, we initialize only the first 2198 * allocated mft record, incrementing mft data size and initialized size 2199 * accordingly, open an ntfs_inode for it and return it to the caller, unless 2200 * there are less than 24 mft records, in which case we allocate and initialize 2201 * mft records until we reach record 24 which we consider as the first free mft 2202 * record for use by normal files. 2203 * 2204 * If during any stage we overflow the initialized data in the mft bitmap, we 2205 * extend the initialized size (and data size) by 8 bytes, allocating another 2206 * cluster if required. The bitmap data size has to be at least equal to the 2207 * number of mft records in the mft, but it can be bigger, in which case the 2208 * superflous bits are padded with zeroes. 2209 * 2210 * Thus, when we return successfully (IS_ERR() is false), we will have: 2211 * - initialized / extended the mft bitmap if necessary, 2212 * - initialized / extended the mft data if necessary, 2213 * - set the bit corresponding to the mft record being allocated in the 2214 * mft bitmap, 2215 * - opened an ntfs_inode for the allocated mft record, and we will have 2216 * - returned the ntfs_inode as well as the allocated mapped, pinned, and 2217 * locked mft record. 2218 * 2219 * On error, the volume will be left in a consistent state and no record will 2220 * be allocated. If rolling back a partial operation fails, we may leave some 2221 * inconsistent metadata in which case we set NVolErrors() so the volume is 2222 * left dirty when unmounted. 2223 * 2224 * Note, this function cannot make use of most of the normal functions, like 2225 * for example for attribute resizing, etc, because when the run list overflows 2226 * the base mft record and an attribute list is used, it is very important that 2227 * the extension mft records used to store the $DATA attribute of $MFT can be 2228 * reached without having to read the information contained inside them, as 2229 * this would make it impossible to find them in the first place after the 2230 * volume is unmounted. $MFT/$BITMAP probably does not need to follow this 2231 * rule because the bitmap is not essential for finding the mft records, but on 2232 * the other hand, handling the bitmap in this special way would make life 2233 * easier because otherwise there might be circular invocations of functions 2234 * when reading the bitmap. 2235 */ 2236 ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode, 2237 ntfs_inode *base_ni, MFT_RECORD **mrec) 2238 { 2239 s64 ll, bit, old_data_initialized, old_data_size; 2240 unsigned long flags; 2241 struct inode *vi; 2242 struct page *page; 2243 ntfs_inode *mft_ni, *mftbmp_ni, *ni; 2244 ntfs_attr_search_ctx *ctx; 2245 MFT_RECORD *m; 2246 ATTR_RECORD *a; 2247 pgoff_t index; 2248 unsigned int ofs; 2249 int err; 2250 le16 seq_no, usn; 2251 BOOL record_formatted = FALSE; 2252 2253 if (base_ni) { 2254 ntfs_debug("Entering (allocating an extent mft record for " 2255 "base mft record 0x%llx).", 2256 (long long)base_ni->mft_no); 2257 /* @mode and @base_ni are mutually exclusive. */ 2258 BUG_ON(mode); 2259 } else 2260 ntfs_debug("Entering (allocating a base mft record)."); 2261 if (mode) { 2262 /* @mode and @base_ni are mutually exclusive. */ 2263 BUG_ON(base_ni); 2264 /* We only support creation of normal files and directories. */ 2265 if (!S_ISREG(mode) && !S_ISDIR(mode)) 2266 return ERR_PTR(-EOPNOTSUPP); 2267 } 2268 BUG_ON(!mrec); 2269 mft_ni = NTFS_I(vol->mft_ino); 2270 mftbmp_ni = NTFS_I(vol->mftbmp_ino); 2271 down_write(&vol->mftbmp_lock); 2272 bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni); 2273 if (bit >= 0) { 2274 ntfs_debug("Found and allocated free record (#1), bit 0x%llx.", 2275 (long long)bit); 2276 goto have_alloc_rec; 2277 } 2278 if (bit != -ENOSPC) { 2279 up_write(&vol->mftbmp_lock); 2280 return ERR_PTR(bit); 2281 } 2282 /* 2283 * No free mft records left. If the mft bitmap already covers more 2284 * than the currently used mft records, the next records are all free, 2285 * so we can simply allocate the first unused mft record. 2286 * Note: We also have to make sure that the mft bitmap at least covers 2287 * the first 24 mft records as they are special and whilst they may not 2288 * be in use, we do not allocate from them. 2289 */ 2290 read_lock_irqsave(&mft_ni->size_lock, flags); 2291 ll = mft_ni->initialized_size >> vol->mft_record_size_bits; 2292 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2293 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2294 old_data_initialized = mftbmp_ni->initialized_size; 2295 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2296 if (old_data_initialized << 3 > ll && old_data_initialized > 3) { 2297 bit = ll; 2298 if (bit < 24) 2299 bit = 24; 2300 if (unlikely(bit >= (1ll << 32))) 2301 goto max_err_out; 2302 ntfs_debug("Found free record (#2), bit 0x%llx.", 2303 (long long)bit); 2304 goto found_free_rec; 2305 } 2306 /* 2307 * The mft bitmap needs to be expanded until it covers the first unused 2308 * mft record that we can allocate. 2309 * Note: The smallest mft record we allocate is mft record 24. 2310 */ 2311 bit = old_data_initialized << 3; 2312 if (unlikely(bit >= (1ll << 32))) 2313 goto max_err_out; 2314 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2315 old_data_size = mftbmp_ni->allocated_size; 2316 ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, " 2317 "data_size 0x%llx, initialized_size 0x%llx.", 2318 (long long)old_data_size, 2319 (long long)i_size_read(vol->mftbmp_ino), 2320 (long long)old_data_initialized); 2321 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2322 if (old_data_initialized + 8 > old_data_size) { 2323 /* Need to extend bitmap by one more cluster. */ 2324 ntfs_debug("mftbmp: initialized_size + 8 > allocated_size."); 2325 err = ntfs_mft_bitmap_extend_allocation_nolock(vol); 2326 if (unlikely(err)) { 2327 up_write(&vol->mftbmp_lock); 2328 goto err_out; 2329 } 2330 #ifdef DEBUG 2331 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2332 ntfs_debug("Status of mftbmp after allocation extension: " 2333 "allocated_size 0x%llx, data_size 0x%llx, " 2334 "initialized_size 0x%llx.", 2335 (long long)mftbmp_ni->allocated_size, 2336 (long long)i_size_read(vol->mftbmp_ino), 2337 (long long)mftbmp_ni->initialized_size); 2338 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2339 #endif /* DEBUG */ 2340 } 2341 /* 2342 * We now have sufficient allocated space, extend the initialized_size 2343 * as well as the data_size if necessary and fill the new space with 2344 * zeroes. 2345 */ 2346 err = ntfs_mft_bitmap_extend_initialized_nolock(vol); 2347 if (unlikely(err)) { 2348 up_write(&vol->mftbmp_lock); 2349 goto err_out; 2350 } 2351 #ifdef DEBUG 2352 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2353 ntfs_debug("Status of mftbmp after initialized extention: " 2354 "allocated_size 0x%llx, data_size 0x%llx, " 2355 "initialized_size 0x%llx.", 2356 (long long)mftbmp_ni->allocated_size, 2357 (long long)i_size_read(vol->mftbmp_ino), 2358 (long long)mftbmp_ni->initialized_size); 2359 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2360 #endif /* DEBUG */ 2361 ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit); 2362 found_free_rec: 2363 /* @bit is the found free mft record, allocate it in the mft bitmap. */ 2364 ntfs_debug("At found_free_rec."); 2365 err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit); 2366 if (unlikely(err)) { 2367 ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap."); 2368 up_write(&vol->mftbmp_lock); 2369 goto err_out; 2370 } 2371 ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit); 2372 have_alloc_rec: 2373 /* 2374 * The mft bitmap is now uptodate. Deal with mft data attribute now. 2375 * Note, we keep hold of the mft bitmap lock for writing until all 2376 * modifications to the mft data attribute are complete, too, as they 2377 * will impact decisions for mft bitmap and mft record allocation done 2378 * by a parallel allocation and if the lock is not maintained a 2379 * parallel allocation could allocate the same mft record as this one. 2380 */ 2381 ll = (bit + 1) << vol->mft_record_size_bits; 2382 read_lock_irqsave(&mft_ni->size_lock, flags); 2383 old_data_initialized = mft_ni->initialized_size; 2384 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2385 if (ll <= old_data_initialized) { 2386 ntfs_debug("Allocated mft record already initialized."); 2387 goto mft_rec_already_initialized; 2388 } 2389 ntfs_debug("Initializing allocated mft record."); 2390 /* 2391 * The mft record is outside the initialized data. Extend the mft data 2392 * attribute until it covers the allocated record. The loop is only 2393 * actually traversed more than once when a freshly formatted volume is 2394 * first written to so it optimizes away nicely in the common case. 2395 */ 2396 read_lock_irqsave(&mft_ni->size_lock, flags); 2397 ntfs_debug("Status of mft data before extension: " 2398 "allocated_size 0x%llx, data_size 0x%llx, " 2399 "initialized_size 0x%llx.", 2400 (long long)mft_ni->allocated_size, 2401 (long long)i_size_read(vol->mft_ino), 2402 (long long)mft_ni->initialized_size); 2403 while (ll > mft_ni->allocated_size) { 2404 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2405 err = ntfs_mft_data_extend_allocation_nolock(vol); 2406 if (unlikely(err)) { 2407 ntfs_error(vol->sb, "Failed to extend mft data " 2408 "allocation."); 2409 goto undo_mftbmp_alloc_nolock; 2410 } 2411 read_lock_irqsave(&mft_ni->size_lock, flags); 2412 ntfs_debug("Status of mft data after allocation extension: " 2413 "allocated_size 0x%llx, data_size 0x%llx, " 2414 "initialized_size 0x%llx.", 2415 (long long)mft_ni->allocated_size, 2416 (long long)i_size_read(vol->mft_ino), 2417 (long long)mft_ni->initialized_size); 2418 } 2419 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2420 /* 2421 * Extend mft data initialized size (and data size of course) to reach 2422 * the allocated mft record, formatting the mft records allong the way. 2423 * Note: We only modify the ntfs_inode structure as that is all that is 2424 * needed by ntfs_mft_record_format(). We will update the attribute 2425 * record itself in one fell swoop later on. 2426 */ 2427 write_lock_irqsave(&mft_ni->size_lock, flags); 2428 old_data_initialized = mft_ni->initialized_size; 2429 old_data_size = vol->mft_ino->i_size; 2430 while (ll > mft_ni->initialized_size) { 2431 s64 new_initialized_size, mft_no; 2432 2433 new_initialized_size = mft_ni->initialized_size + 2434 vol->mft_record_size; 2435 mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits; 2436 if (new_initialized_size > i_size_read(vol->mft_ino)) 2437 i_size_write(vol->mft_ino, new_initialized_size); 2438 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2439 ntfs_debug("Initializing mft record 0x%llx.", 2440 (long long)mft_no); 2441 err = ntfs_mft_record_format(vol, mft_no); 2442 if (unlikely(err)) { 2443 ntfs_error(vol->sb, "Failed to format mft record."); 2444 goto undo_data_init; 2445 } 2446 write_lock_irqsave(&mft_ni->size_lock, flags); 2447 mft_ni->initialized_size = new_initialized_size; 2448 } 2449 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2450 record_formatted = TRUE; 2451 /* Update the mft data attribute record to reflect the new sizes. */ 2452 m = map_mft_record(mft_ni); 2453 if (IS_ERR(m)) { 2454 ntfs_error(vol->sb, "Failed to map mft record."); 2455 err = PTR_ERR(m); 2456 goto undo_data_init; 2457 } 2458 ctx = ntfs_attr_get_search_ctx(mft_ni, m); 2459 if (unlikely(!ctx)) { 2460 ntfs_error(vol->sb, "Failed to get search context."); 2461 err = -ENOMEM; 2462 unmap_mft_record(mft_ni); 2463 goto undo_data_init; 2464 } 2465 err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 2466 CASE_SENSITIVE, 0, NULL, 0, ctx); 2467 if (unlikely(err)) { 2468 ntfs_error(vol->sb, "Failed to find first attribute extent of " 2469 "mft data attribute."); 2470 ntfs_attr_put_search_ctx(ctx); 2471 unmap_mft_record(mft_ni); 2472 goto undo_data_init; 2473 } 2474 a = ctx->attr; 2475 read_lock_irqsave(&mft_ni->size_lock, flags); 2476 a->data.non_resident.initialized_size = 2477 cpu_to_sle64(mft_ni->initialized_size); 2478 a->data.non_resident.data_size = 2479 cpu_to_sle64(i_size_read(vol->mft_ino)); 2480 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2481 /* Ensure the changes make it to disk. */ 2482 flush_dcache_mft_record_page(ctx->ntfs_ino); 2483 mark_mft_record_dirty(ctx->ntfs_ino); 2484 ntfs_attr_put_search_ctx(ctx); 2485 unmap_mft_record(mft_ni); 2486 read_lock_irqsave(&mft_ni->size_lock, flags); 2487 ntfs_debug("Status of mft data after mft record initialization: " 2488 "allocated_size 0x%llx, data_size 0x%llx, " 2489 "initialized_size 0x%llx.", 2490 (long long)mft_ni->allocated_size, 2491 (long long)i_size_read(vol->mft_ino), 2492 (long long)mft_ni->initialized_size); 2493 BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size); 2494 BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino)); 2495 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2496 mft_rec_already_initialized: 2497 /* 2498 * We can finally drop the mft bitmap lock as the mft data attribute 2499 * has been fully updated. The only disparity left is that the 2500 * allocated mft record still needs to be marked as in use to match the 2501 * set bit in the mft bitmap but this is actually not a problem since 2502 * this mft record is not referenced from anywhere yet and the fact 2503 * that it is allocated in the mft bitmap means that no-one will try to 2504 * allocate it either. 2505 */ 2506 up_write(&vol->mftbmp_lock); 2507 /* 2508 * We now have allocated and initialized the mft record. Calculate the 2509 * index of and the offset within the page cache page the record is in. 2510 */ 2511 index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; 2512 ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 2513 /* Read, map, and pin the page containing the mft record. */ 2514 page = ntfs_map_page(vol->mft_ino->i_mapping, index); 2515 if (unlikely(IS_ERR(page))) { 2516 ntfs_error(vol->sb, "Failed to map page containing allocated " 2517 "mft record 0x%llx.", (long long)bit); 2518 err = PTR_ERR(page); 2519 goto undo_mftbmp_alloc; 2520 } 2521 lock_page(page); 2522 BUG_ON(!PageUptodate(page)); 2523 ClearPageUptodate(page); 2524 m = (MFT_RECORD*)((u8*)page_address(page) + ofs); 2525 /* If we just formatted the mft record no need to do it again. */ 2526 if (!record_formatted) { 2527 /* Sanity check that the mft record is really not in use. */ 2528 if (ntfs_is_file_record(m->magic) && 2529 (m->flags & MFT_RECORD_IN_USE)) { 2530 ntfs_error(vol->sb, "Mft record 0x%llx was marked " 2531 "free in mft bitmap but is marked " 2532 "used itself. Corrupt filesystem. " 2533 "Unmount and run chkdsk.", 2534 (long long)bit); 2535 err = -EIO; 2536 SetPageUptodate(page); 2537 unlock_page(page); 2538 ntfs_unmap_page(page); 2539 NVolSetErrors(vol); 2540 goto undo_mftbmp_alloc; 2541 } 2542 /* 2543 * We need to (re-)format the mft record, preserving the 2544 * sequence number if it is not zero as well as the update 2545 * sequence number if it is not zero or -1 (0xffff). This 2546 * means we do not need to care whether or not something went 2547 * wrong with the previous mft record. 2548 */ 2549 seq_no = m->sequence_number; 2550 usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)); 2551 err = ntfs_mft_record_layout(vol, bit, m); 2552 if (unlikely(err)) { 2553 ntfs_error(vol->sb, "Failed to layout allocated mft " 2554 "record 0x%llx.", (long long)bit); 2555 SetPageUptodate(page); 2556 unlock_page(page); 2557 ntfs_unmap_page(page); 2558 goto undo_mftbmp_alloc; 2559 } 2560 if (seq_no) 2561 m->sequence_number = seq_no; 2562 if (usn && le16_to_cpu(usn) != 0xffff) 2563 *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn; 2564 } 2565 /* Set the mft record itself in use. */ 2566 m->flags |= MFT_RECORD_IN_USE; 2567 if (S_ISDIR(mode)) 2568 m->flags |= MFT_RECORD_IS_DIRECTORY; 2569 flush_dcache_page(page); 2570 SetPageUptodate(page); 2571 if (base_ni) { 2572 /* 2573 * Setup the base mft record in the extent mft record. This 2574 * completes initialization of the allocated extent mft record 2575 * and we can simply use it with map_extent_mft_record(). 2576 */ 2577 m->base_mft_record = MK_LE_MREF(base_ni->mft_no, 2578 base_ni->seq_no); 2579 /* 2580 * Allocate an extent inode structure for the new mft record, 2581 * attach it to the base inode @base_ni and map, pin, and lock 2582 * its, i.e. the allocated, mft record. 2583 */ 2584 m = map_extent_mft_record(base_ni, bit, &ni); 2585 if (IS_ERR(m)) { 2586 ntfs_error(vol->sb, "Failed to map allocated extent " 2587 "mft record 0x%llx.", (long long)bit); 2588 err = PTR_ERR(m); 2589 /* Set the mft record itself not in use. */ 2590 m->flags &= cpu_to_le16( 2591 ~le16_to_cpu(MFT_RECORD_IN_USE)); 2592 flush_dcache_page(page); 2593 /* Make sure the mft record is written out to disk. */ 2594 mark_ntfs_record_dirty(page, ofs); 2595 unlock_page(page); 2596 ntfs_unmap_page(page); 2597 goto undo_mftbmp_alloc; 2598 } 2599 /* 2600 * Make sure the allocated mft record is written out to disk. 2601 * No need to set the inode dirty because the caller is going 2602 * to do that anyway after finishing with the new extent mft 2603 * record (e.g. at a minimum a new attribute will be added to 2604 * the mft record. 2605 */ 2606 mark_ntfs_record_dirty(page, ofs); 2607 unlock_page(page); 2608 /* 2609 * Need to unmap the page since map_extent_mft_record() mapped 2610 * it as well so we have it mapped twice at the moment. 2611 */ 2612 ntfs_unmap_page(page); 2613 } else { 2614 /* 2615 * Allocate a new VFS inode and set it up. NOTE: @vi->i_nlink 2616 * is set to 1 but the mft record->link_count is 0. The caller 2617 * needs to bear this in mind. 2618 */ 2619 vi = new_inode(vol->sb); 2620 if (unlikely(!vi)) { 2621 err = -ENOMEM; 2622 /* Set the mft record itself not in use. */ 2623 m->flags &= cpu_to_le16( 2624 ~le16_to_cpu(MFT_RECORD_IN_USE)); 2625 flush_dcache_page(page); 2626 /* Make sure the mft record is written out to disk. */ 2627 mark_ntfs_record_dirty(page, ofs); 2628 unlock_page(page); 2629 ntfs_unmap_page(page); 2630 goto undo_mftbmp_alloc; 2631 } 2632 vi->i_ino = bit; 2633 /* 2634 * This is the optimal IO size (for stat), not the fs block 2635 * size. 2636 */ 2637 vi->i_blksize = PAGE_CACHE_SIZE; 2638 /* 2639 * This is for checking whether an inode has changed w.r.t. a 2640 * file so that the file can be updated if necessary (compare 2641 * with f_version). 2642 */ 2643 vi->i_version = 1; 2644 2645 /* The owner and group come from the ntfs volume. */ 2646 vi->i_uid = vol->uid; 2647 vi->i_gid = vol->gid; 2648 2649 /* Initialize the ntfs specific part of @vi. */ 2650 ntfs_init_big_inode(vi); 2651 ni = NTFS_I(vi); 2652 /* 2653 * Set the appropriate mode, attribute type, and name. For 2654 * directories, also setup the index values to the defaults. 2655 */ 2656 if (S_ISDIR(mode)) { 2657 vi->i_mode = S_IFDIR | S_IRWXUGO; 2658 vi->i_mode &= ~vol->dmask; 2659 2660 NInoSetMstProtected(ni); 2661 ni->type = AT_INDEX_ALLOCATION; 2662 ni->name = I30; 2663 ni->name_len = 4; 2664 2665 ni->itype.index.block_size = 4096; 2666 ni->itype.index.block_size_bits = generic_ffs(4096) - 1; 2667 ni->itype.index.collation_rule = COLLATION_FILE_NAME; 2668 if (vol->cluster_size <= ni->itype.index.block_size) { 2669 ni->itype.index.vcn_size = vol->cluster_size; 2670 ni->itype.index.vcn_size_bits = 2671 vol->cluster_size_bits; 2672 } else { 2673 ni->itype.index.vcn_size = vol->sector_size; 2674 ni->itype.index.vcn_size_bits = 2675 vol->sector_size_bits; 2676 } 2677 } else { 2678 vi->i_mode = S_IFREG | S_IRWXUGO; 2679 vi->i_mode &= ~vol->fmask; 2680 2681 ni->type = AT_DATA; 2682 ni->name = NULL; 2683 ni->name_len = 0; 2684 } 2685 if (IS_RDONLY(vi)) 2686 vi->i_mode &= ~S_IWUGO; 2687 2688 /* Set the inode times to the current time. */ 2689 vi->i_atime = vi->i_mtime = vi->i_ctime = 2690 current_fs_time(vi->i_sb); 2691 /* 2692 * Set the file size to 0, the ntfs inode sizes are set to 0 by 2693 * the call to ntfs_init_big_inode() below. 2694 */ 2695 vi->i_size = 0; 2696 vi->i_blocks = 0; 2697 2698 /* Set the sequence number. */ 2699 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number); 2700 /* 2701 * Manually map, pin, and lock the mft record as we already 2702 * have its page mapped and it is very easy to do. 2703 */ 2704 atomic_inc(&ni->count); 2705 down(&ni->mrec_lock); 2706 ni->page = page; 2707 ni->page_ofs = ofs; 2708 /* 2709 * Make sure the allocated mft record is written out to disk. 2710 * NOTE: We do not set the ntfs inode dirty because this would 2711 * fail in ntfs_write_inode() because the inode does not have a 2712 * standard information attribute yet. Also, there is no need 2713 * to set the inode dirty because the caller is going to do 2714 * that anyway after finishing with the new mft record (e.g. at 2715 * a minimum some new attributes will be added to the mft 2716 * record. 2717 */ 2718 mark_ntfs_record_dirty(page, ofs); 2719 unlock_page(page); 2720 2721 /* Add the inode to the inode hash for the superblock. */ 2722 insert_inode_hash(vi); 2723 2724 /* Update the default mft allocation position. */ 2725 vol->mft_data_pos = bit + 1; 2726 } 2727 /* 2728 * Return the opened, allocated inode of the allocated mft record as 2729 * well as the mapped, pinned, and locked mft record. 2730 */ 2731 ntfs_debug("Returning opened, allocated %sinode 0x%llx.", 2732 base_ni ? "extent " : "", (long long)bit); 2733 *mrec = m; 2734 return ni; 2735 undo_data_init: 2736 write_lock_irqsave(&mft_ni->size_lock, flags); 2737 mft_ni->initialized_size = old_data_initialized; 2738 i_size_write(vol->mft_ino, old_data_size); 2739 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2740 goto undo_mftbmp_alloc_nolock; 2741 undo_mftbmp_alloc: 2742 down_write(&vol->mftbmp_lock); 2743 undo_mftbmp_alloc_nolock: 2744 if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) { 2745 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); 2746 NVolSetErrors(vol); 2747 } 2748 up_write(&vol->mftbmp_lock); 2749 err_out: 2750 return ERR_PTR(err); 2751 max_err_out: 2752 ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum " 2753 "number of inodes (2^32) has already been reached."); 2754 up_write(&vol->mftbmp_lock); 2755 return ERR_PTR(-ENOSPC); 2756 } 2757 2758 /** 2759 * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume 2760 * @ni: ntfs inode of the mapped extent mft record to free 2761 * @m: mapped extent mft record of the ntfs inode @ni 2762 * 2763 * Free the mapped extent mft record @m of the extent ntfs inode @ni. 2764 * 2765 * Note that this function unmaps the mft record and closes and destroys @ni 2766 * internally and hence you cannot use either @ni nor @m any more after this 2767 * function returns success. 2768 * 2769 * On success return 0 and on error return -errno. @ni and @m are still valid 2770 * in this case and have not been freed. 2771 * 2772 * For some errors an error message is displayed and the success code 0 is 2773 * returned and the volume is then left dirty on umount. This makes sense in 2774 * case we could not rollback the changes that were already done since the 2775 * caller no longer wants to reference this mft record so it does not matter to 2776 * the caller if something is wrong with it as long as it is properly detached 2777 * from the base inode. 2778 */ 2779 int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m) 2780 { 2781 unsigned long mft_no = ni->mft_no; 2782 ntfs_volume *vol = ni->vol; 2783 ntfs_inode *base_ni; 2784 ntfs_inode **extent_nis; 2785 int i, err; 2786 le16 old_seq_no; 2787 u16 seq_no; 2788 2789 BUG_ON(NInoAttr(ni)); 2790 BUG_ON(ni->nr_extents != -1); 2791 2792 down(&ni->extent_lock); 2793 base_ni = ni->ext.base_ntfs_ino; 2794 up(&ni->extent_lock); 2795 2796 BUG_ON(base_ni->nr_extents <= 0); 2797 2798 ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n", 2799 mft_no, base_ni->mft_no); 2800 2801 down(&base_ni->extent_lock); 2802 2803 /* Make sure we are holding the only reference to the extent inode. */ 2804 if (atomic_read(&ni->count) > 2) { 2805 ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, " 2806 "not freeing.", base_ni->mft_no); 2807 up(&base_ni->extent_lock); 2808 return -EBUSY; 2809 } 2810 2811 /* Dissociate the ntfs inode from the base inode. */ 2812 extent_nis = base_ni->ext.extent_ntfs_inos; 2813 err = -ENOENT; 2814 for (i = 0; i < base_ni->nr_extents; i++) { 2815 if (ni != extent_nis[i]) 2816 continue; 2817 extent_nis += i; 2818 base_ni->nr_extents--; 2819 memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) * 2820 sizeof(ntfs_inode*)); 2821 err = 0; 2822 break; 2823 } 2824 2825 up(&base_ni->extent_lock); 2826 2827 if (unlikely(err)) { 2828 ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to " 2829 "its base inode 0x%lx.", mft_no, 2830 base_ni->mft_no); 2831 BUG(); 2832 } 2833 2834 /* 2835 * The extent inode is no longer attached to the base inode so no one 2836 * can get a reference to it any more. 2837 */ 2838 2839 /* Mark the mft record as not in use. */ 2840 m->flags &= const_cpu_to_le16(~const_le16_to_cpu(MFT_RECORD_IN_USE)); 2841 2842 /* Increment the sequence number, skipping zero, if it is not zero. */ 2843 old_seq_no = m->sequence_number; 2844 seq_no = le16_to_cpu(old_seq_no); 2845 if (seq_no == 0xffff) 2846 seq_no = 1; 2847 else if (seq_no) 2848 seq_no++; 2849 m->sequence_number = cpu_to_le16(seq_no); 2850 2851 /* 2852 * Set the ntfs inode dirty and write it out. We do not need to worry 2853 * about the base inode here since whatever caused the extent mft 2854 * record to be freed is guaranteed to do it already. 2855 */ 2856 NInoSetDirty(ni); 2857 err = write_mft_record(ni, m, 0); 2858 if (unlikely(err)) { 2859 ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not " 2860 "freeing.", mft_no); 2861 goto rollback; 2862 } 2863 rollback_error: 2864 /* Unmap and throw away the now freed extent inode. */ 2865 unmap_extent_mft_record(ni); 2866 ntfs_clear_extent_inode(ni); 2867 2868 /* Clear the bit in the $MFT/$BITMAP corresponding to this record. */ 2869 down_write(&vol->mftbmp_lock); 2870 err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no); 2871 up_write(&vol->mftbmp_lock); 2872 if (unlikely(err)) { 2873 /* 2874 * The extent inode is gone but we failed to deallocate it in 2875 * the mft bitmap. Just emit a warning and leave the volume 2876 * dirty on umount. 2877 */ 2878 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); 2879 NVolSetErrors(vol); 2880 } 2881 return 0; 2882 rollback: 2883 /* Rollback what we did... */ 2884 down(&base_ni->extent_lock); 2885 extent_nis = base_ni->ext.extent_ntfs_inos; 2886 if (!(base_ni->nr_extents & 3)) { 2887 int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*); 2888 2889 extent_nis = (ntfs_inode**)kmalloc(new_size, GFP_NOFS); 2890 if (unlikely(!extent_nis)) { 2891 ntfs_error(vol->sb, "Failed to allocate internal " 2892 "buffer during rollback.%s", es); 2893 up(&base_ni->extent_lock); 2894 NVolSetErrors(vol); 2895 goto rollback_error; 2896 } 2897 if (base_ni->nr_extents) { 2898 BUG_ON(!base_ni->ext.extent_ntfs_inos); 2899 memcpy(extent_nis, base_ni->ext.extent_ntfs_inos, 2900 new_size - 4 * sizeof(ntfs_inode*)); 2901 kfree(base_ni->ext.extent_ntfs_inos); 2902 } 2903 base_ni->ext.extent_ntfs_inos = extent_nis; 2904 } 2905 m->flags |= MFT_RECORD_IN_USE; 2906 m->sequence_number = old_seq_no; 2907 extent_nis[base_ni->nr_extents++] = ni; 2908 up(&base_ni->extent_lock); 2909 mark_mft_record_dirty(ni); 2910 return err; 2911 } 2912 #endif /* NTFS_RW */ 2913