1 /** 2 * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project. 3 * 4 * Copyright (c) 2001-2004 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 file system. " 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 file system. 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 BUG_ON(!page_has_buffers(page)); 515 } 516 bh = head = page_buffers(page); 517 BUG_ON(!bh); 518 rl = NULL; 519 nr_bhs = 0; 520 block_start = 0; 521 m_start = kmirr - (u8*)page_address(page); 522 m_end = m_start + vol->mft_record_size; 523 do { 524 block_end = block_start + blocksize; 525 /* If the buffer is outside the mft record, skip it. */ 526 if (block_end <= m_start) 527 continue; 528 if (unlikely(block_start >= m_end)) 529 break; 530 /* Need to map the buffer if it is not mapped already. */ 531 if (unlikely(!buffer_mapped(bh))) { 532 VCN vcn; 533 LCN lcn; 534 unsigned int vcn_ofs; 535 536 /* 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 BUG_ON(!page_has_buffers(page)); 695 bh = head = page_buffers(page); 696 BUG_ON(!bh); 697 rl = NULL; 698 nr_bhs = 0; 699 block_start = 0; 700 m_start = ni->page_ofs; 701 m_end = m_start + vol->mft_record_size; 702 do { 703 block_end = block_start + blocksize; 704 /* If the buffer is outside the mft record, skip it. */ 705 if (block_end <= m_start) 706 continue; 707 if (unlikely(block_start >= m_end)) 708 break; 709 /* 710 * If this block is not the first one in the record, we ignore 711 * the buffer's dirty state because we could have raced with a 712 * parallel mark_ntfs_record_dirty(). 713 */ 714 if (block_start == m_start) { 715 /* This block is the first one in the record. */ 716 if (!buffer_dirty(bh)) { 717 BUG_ON(nr_bhs); 718 /* Clean records are not written out. */ 719 break; 720 } 721 } 722 /* Need to map the buffer if it is not mapped already. */ 723 if (unlikely(!buffer_mapped(bh))) { 724 VCN vcn; 725 LCN lcn; 726 unsigned int vcn_ofs; 727 728 /* 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 * For inode 0, i.e. $MFT itself, we cannot use ilookup5() from here or 952 * we deadlock because the inode is already locked by the kernel 953 * (fs/fs-writeback.c::__sync_single_inode()) and ilookup5() waits 954 * until the inode is unlocked before returning it and it never gets 955 * unlocked because ntfs_should_write_mft_record() never returns. )-: 956 * Fortunately, we have inode 0 pinned in icache for the duration of 957 * the mount so we can access it directly. 958 */ 959 if (!mft_no) { 960 /* Balance the below iput(). */ 961 vi = igrab(mft_vi); 962 BUG_ON(vi != mft_vi); 963 } else 964 vi = ilookup5(sb, mft_no, (test_t)ntfs_test_inode, &na); 965 if (vi) { 966 ntfs_debug("Base inode 0x%lx is in icache.", mft_no); 967 /* The inode is in icache. */ 968 ni = NTFS_I(vi); 969 /* Take a reference to the ntfs inode. */ 970 atomic_inc(&ni->count); 971 /* If the inode is dirty, do not write this record. */ 972 if (NInoDirty(ni)) { 973 ntfs_debug("Inode 0x%lx is dirty, do not write it.", 974 mft_no); 975 atomic_dec(&ni->count); 976 iput(vi); 977 return FALSE; 978 } 979 ntfs_debug("Inode 0x%lx is not dirty.", mft_no); 980 /* The inode is not dirty, try to take the mft record lock. */ 981 if (unlikely(down_trylock(&ni->mrec_lock))) { 982 ntfs_debug("Mft record 0x%lx is already locked, do " 983 "not write it.", mft_no); 984 atomic_dec(&ni->count); 985 iput(vi); 986 return FALSE; 987 } 988 ntfs_debug("Managed to lock mft record 0x%lx, write it.", 989 mft_no); 990 /* 991 * The write has to occur while we hold the mft record lock so 992 * return the locked ntfs inode. 993 */ 994 *locked_ni = ni; 995 return TRUE; 996 } 997 ntfs_debug("Inode 0x%lx is not in icache.", mft_no); 998 /* The inode is not in icache. */ 999 /* Write the record if it is not a mft record (type "FILE"). */ 1000 if (!ntfs_is_mft_record(m->magic)) { 1001 ntfs_debug("Mft record 0x%lx is not a FILE record, write it.", 1002 mft_no); 1003 return TRUE; 1004 } 1005 /* Write the mft record if it is a base inode. */ 1006 if (!m->base_mft_record) { 1007 ntfs_debug("Mft record 0x%lx is a base record, write it.", 1008 mft_no); 1009 return TRUE; 1010 } 1011 /* 1012 * This is an extent mft record. Check if the inode corresponding to 1013 * its base mft record is in icache and obtain a reference to it if it 1014 * is. 1015 */ 1016 na.mft_no = MREF_LE(m->base_mft_record); 1017 ntfs_debug("Mft record 0x%lx is an extent record. Looking for base " 1018 "inode 0x%lx in icache.", mft_no, na.mft_no); 1019 vi = ilookup5(sb, na.mft_no, (test_t)ntfs_test_inode, &na); 1020 if (!vi) { 1021 /* 1022 * The base inode is not in icache, write this extent mft 1023 * record. 1024 */ 1025 ntfs_debug("Base inode 0x%lx is not in icache, write the " 1026 "extent record.", na.mft_no); 1027 return TRUE; 1028 } 1029 ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no); 1030 /* 1031 * The base inode is in icache. Check if it has the extent inode 1032 * corresponding to this extent mft record attached. 1033 */ 1034 ni = NTFS_I(vi); 1035 down(&ni->extent_lock); 1036 if (ni->nr_extents <= 0) { 1037 /* 1038 * The base inode has no attached extent inodes, write this 1039 * extent mft record. 1040 */ 1041 up(&ni->extent_lock); 1042 iput(vi); 1043 ntfs_debug("Base inode 0x%lx has no attached extent inodes, " 1044 "write the extent record.", na.mft_no); 1045 return TRUE; 1046 } 1047 /* Iterate over the attached extent inodes. */ 1048 extent_nis = ni->ext.extent_ntfs_inos; 1049 for (eni = NULL, i = 0; i < ni->nr_extents; ++i) { 1050 if (mft_no == extent_nis[i]->mft_no) { 1051 /* 1052 * Found the extent inode corresponding to this extent 1053 * mft record. 1054 */ 1055 eni = extent_nis[i]; 1056 break; 1057 } 1058 } 1059 /* 1060 * If the extent inode was not attached to the base inode, write this 1061 * extent mft record. 1062 */ 1063 if (!eni) { 1064 up(&ni->extent_lock); 1065 iput(vi); 1066 ntfs_debug("Extent inode 0x%lx is not attached to its base " 1067 "inode 0x%lx, write the extent record.", 1068 mft_no, na.mft_no); 1069 return TRUE; 1070 } 1071 ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.", 1072 mft_no, na.mft_no); 1073 /* Take a reference to the extent ntfs inode. */ 1074 atomic_inc(&eni->count); 1075 up(&ni->extent_lock); 1076 /* 1077 * Found the extent inode coresponding to this extent mft record. 1078 * Try to take the mft record lock. 1079 */ 1080 if (unlikely(down_trylock(&eni->mrec_lock))) { 1081 atomic_dec(&eni->count); 1082 iput(vi); 1083 ntfs_debug("Extent mft record 0x%lx is already locked, do " 1084 "not write it.", mft_no); 1085 return FALSE; 1086 } 1087 ntfs_debug("Managed to lock extent mft record 0x%lx, write it.", 1088 mft_no); 1089 if (NInoTestClearDirty(eni)) 1090 ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.", 1091 mft_no); 1092 /* 1093 * The write has to occur while we hold the mft record lock so return 1094 * the locked extent ntfs inode. 1095 */ 1096 *locked_ni = eni; 1097 return TRUE; 1098 } 1099 1100 static const char *es = " Leaving inconsistent metadata. Unmount and run " 1101 "chkdsk."; 1102 1103 /** 1104 * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name 1105 * @vol: volume on which to search for a free mft record 1106 * @base_ni: open base inode if allocating an extent mft record or NULL 1107 * 1108 * Search for a free mft record in the mft bitmap attribute on the ntfs volume 1109 * @vol. 1110 * 1111 * If @base_ni is NULL start the search at the default allocator position. 1112 * 1113 * If @base_ni is not NULL start the search at the mft record after the base 1114 * mft record @base_ni. 1115 * 1116 * Return the free mft record on success and -errno on error. An error code of 1117 * -ENOSPC means that there are no free mft records in the currently 1118 * initialized mft bitmap. 1119 * 1120 * Locking: Caller must hold vol->mftbmp_lock for writing. 1121 */ 1122 static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol, 1123 ntfs_inode *base_ni) 1124 { 1125 s64 pass_end, ll, data_pos, pass_start, ofs, bit; 1126 unsigned long flags; 1127 struct address_space *mftbmp_mapping; 1128 u8 *buf, *byte; 1129 struct page *page; 1130 unsigned int page_ofs, size; 1131 u8 pass, b; 1132 1133 ntfs_debug("Searching for free mft record in the currently " 1134 "initialized mft bitmap."); 1135 mftbmp_mapping = vol->mftbmp_ino->i_mapping; 1136 /* 1137 * Set the end of the pass making sure we do not overflow the mft 1138 * bitmap. 1139 */ 1140 read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags); 1141 pass_end = NTFS_I(vol->mft_ino)->allocated_size >> 1142 vol->mft_record_size_bits; 1143 read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags); 1144 read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); 1145 ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3; 1146 read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); 1147 if (pass_end > ll) 1148 pass_end = ll; 1149 pass = 1; 1150 if (!base_ni) 1151 data_pos = vol->mft_data_pos; 1152 else 1153 data_pos = base_ni->mft_no + 1; 1154 if (data_pos < 24) 1155 data_pos = 24; 1156 if (data_pos >= pass_end) { 1157 data_pos = 24; 1158 pass = 2; 1159 /* This happens on a freshly formatted volume. */ 1160 if (data_pos >= pass_end) 1161 return -ENOSPC; 1162 } 1163 pass_start = data_pos; 1164 ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, " 1165 "pass_end 0x%llx, data_pos 0x%llx.", pass, 1166 (long long)pass_start, (long long)pass_end, 1167 (long long)data_pos); 1168 /* Loop until a free mft record is found. */ 1169 for (; pass <= 2;) { 1170 /* Cap size to pass_end. */ 1171 ofs = data_pos >> 3; 1172 page_ofs = ofs & ~PAGE_CACHE_MASK; 1173 size = PAGE_CACHE_SIZE - page_ofs; 1174 ll = ((pass_end + 7) >> 3) - ofs; 1175 if (size > ll) 1176 size = ll; 1177 size <<= 3; 1178 /* 1179 * If we are still within the active pass, search the next page 1180 * for a zero bit. 1181 */ 1182 if (size) { 1183 page = ntfs_map_page(mftbmp_mapping, 1184 ofs >> PAGE_CACHE_SHIFT); 1185 if (unlikely(IS_ERR(page))) { 1186 ntfs_error(vol->sb, "Failed to read mft " 1187 "bitmap, aborting."); 1188 return PTR_ERR(page); 1189 } 1190 buf = (u8*)page_address(page) + page_ofs; 1191 bit = data_pos & 7; 1192 data_pos &= ~7ull; 1193 ntfs_debug("Before inner for loop: size 0x%x, " 1194 "data_pos 0x%llx, bit 0x%llx", size, 1195 (long long)data_pos, (long long)bit); 1196 for (; bit < size && data_pos + bit < pass_end; 1197 bit &= ~7ull, bit += 8) { 1198 byte = buf + (bit >> 3); 1199 if (*byte == 0xff) 1200 continue; 1201 b = ffz((unsigned long)*byte); 1202 if (b < 8 && b >= (bit & 7)) { 1203 ll = data_pos + (bit & ~7ull) + b; 1204 if (unlikely(ll > (1ll << 32))) { 1205 ntfs_unmap_page(page); 1206 return -ENOSPC; 1207 } 1208 *byte |= 1 << b; 1209 flush_dcache_page(page); 1210 set_page_dirty(page); 1211 ntfs_unmap_page(page); 1212 ntfs_debug("Done. (Found and " 1213 "allocated mft record " 1214 "0x%llx.)", 1215 (long long)ll); 1216 return ll; 1217 } 1218 } 1219 ntfs_debug("After inner for loop: size 0x%x, " 1220 "data_pos 0x%llx, bit 0x%llx", size, 1221 (long long)data_pos, (long long)bit); 1222 data_pos += size; 1223 ntfs_unmap_page(page); 1224 /* 1225 * If the end of the pass has not been reached yet, 1226 * continue searching the mft bitmap for a zero bit. 1227 */ 1228 if (data_pos < pass_end) 1229 continue; 1230 } 1231 /* Do the next pass. */ 1232 if (++pass == 2) { 1233 /* 1234 * Starting the second pass, in which we scan the first 1235 * part of the zone which we omitted earlier. 1236 */ 1237 pass_end = pass_start; 1238 data_pos = pass_start = 24; 1239 ntfs_debug("pass %i, pass_start 0x%llx, pass_end " 1240 "0x%llx.", pass, (long long)pass_start, 1241 (long long)pass_end); 1242 if (data_pos >= pass_end) 1243 break; 1244 } 1245 } 1246 /* No free mft records in currently initialized mft bitmap. */ 1247 ntfs_debug("Done. (No free mft records left in currently initialized " 1248 "mft bitmap.)"); 1249 return -ENOSPC; 1250 } 1251 1252 /** 1253 * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster 1254 * @vol: volume on which to extend the mft bitmap attribute 1255 * 1256 * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster. 1257 * 1258 * Note: Only changes allocated_size, i.e. does not touch initialized_size or 1259 * data_size. 1260 * 1261 * Return 0 on success and -errno on error. 1262 * 1263 * Locking: - Caller must hold vol->mftbmp_lock for writing. 1264 * - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for 1265 * writing and releases it before returning. 1266 * - This function takes vol->lcnbmp_lock for writing and releases it 1267 * before returning. 1268 */ 1269 static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol) 1270 { 1271 LCN lcn; 1272 s64 ll; 1273 unsigned long flags; 1274 struct page *page; 1275 ntfs_inode *mft_ni, *mftbmp_ni; 1276 runlist_element *rl, *rl2 = NULL; 1277 ntfs_attr_search_ctx *ctx = NULL; 1278 MFT_RECORD *mrec; 1279 ATTR_RECORD *a = NULL; 1280 int ret, mp_size; 1281 u32 old_alen = 0; 1282 u8 *b, tb; 1283 struct { 1284 u8 added_cluster:1; 1285 u8 added_run:1; 1286 u8 mp_rebuilt:1; 1287 } status = { 0, 0, 0 }; 1288 1289 ntfs_debug("Extending mft bitmap allocation."); 1290 mft_ni = NTFS_I(vol->mft_ino); 1291 mftbmp_ni = NTFS_I(vol->mftbmp_ino); 1292 /* 1293 * Determine the last lcn of the mft bitmap. The allocated size of the 1294 * mft bitmap cannot be zero so we are ok to do this. 1295 * ntfs_find_vcn() returns the runlist locked on success. 1296 */ 1297 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 1298 ll = mftbmp_ni->allocated_size; 1299 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1300 rl = ntfs_find_vcn(mftbmp_ni, (ll - 1) >> vol->cluster_size_bits, TRUE); 1301 if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { 1302 ntfs_error(vol->sb, "Failed to determine last allocated " 1303 "cluster of mft bitmap attribute."); 1304 if (!IS_ERR(rl)) { 1305 up_write(&mftbmp_ni->runlist.lock); 1306 ret = -EIO; 1307 } else 1308 ret = PTR_ERR(rl); 1309 return ret; 1310 } 1311 lcn = rl->lcn + rl->length; 1312 ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.", 1313 (long long)lcn); 1314 /* 1315 * Attempt to get the cluster following the last allocated cluster by 1316 * hand as it may be in the MFT zone so the allocator would not give it 1317 * to us. 1318 */ 1319 ll = lcn >> 3; 1320 page = ntfs_map_page(vol->lcnbmp_ino->i_mapping, 1321 ll >> PAGE_CACHE_SHIFT); 1322 if (IS_ERR(page)) { 1323 up_write(&mftbmp_ni->runlist.lock); 1324 ntfs_error(vol->sb, "Failed to read from lcn bitmap."); 1325 return PTR_ERR(page); 1326 } 1327 b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK); 1328 tb = 1 << (lcn & 7ull); 1329 down_write(&vol->lcnbmp_lock); 1330 if (*b != 0xff && !(*b & tb)) { 1331 /* Next cluster is free, allocate it. */ 1332 *b |= tb; 1333 flush_dcache_page(page); 1334 set_page_dirty(page); 1335 up_write(&vol->lcnbmp_lock); 1336 ntfs_unmap_page(page); 1337 /* Update the mft bitmap runlist. */ 1338 rl->length++; 1339 rl[1].vcn++; 1340 status.added_cluster = 1; 1341 ntfs_debug("Appending one cluster to mft bitmap."); 1342 } else { 1343 up_write(&vol->lcnbmp_lock); 1344 ntfs_unmap_page(page); 1345 /* Allocate a cluster from the DATA_ZONE. */ 1346 rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE); 1347 if (IS_ERR(rl2)) { 1348 up_write(&mftbmp_ni->runlist.lock); 1349 ntfs_error(vol->sb, "Failed to allocate a cluster for " 1350 "the mft bitmap."); 1351 return PTR_ERR(rl2); 1352 } 1353 rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2); 1354 if (IS_ERR(rl)) { 1355 up_write(&mftbmp_ni->runlist.lock); 1356 ntfs_error(vol->sb, "Failed to merge runlists for mft " 1357 "bitmap."); 1358 if (ntfs_cluster_free_from_rl(vol, rl2)) { 1359 ntfs_error(vol->sb, "Failed to dealocate " 1360 "allocated cluster.%s", es); 1361 NVolSetErrors(vol); 1362 } 1363 ntfs_free(rl2); 1364 return PTR_ERR(rl); 1365 } 1366 mftbmp_ni->runlist.rl = rl; 1367 status.added_run = 1; 1368 ntfs_debug("Adding one run to mft bitmap."); 1369 /* Find the last run in the new runlist. */ 1370 for (; rl[1].length; rl++) 1371 ; 1372 } 1373 /* 1374 * Update the attribute record as well. Note: @rl is the last 1375 * (non-terminator) runlist element of mft bitmap. 1376 */ 1377 mrec = map_mft_record(mft_ni); 1378 if (IS_ERR(mrec)) { 1379 ntfs_error(vol->sb, "Failed to map mft record."); 1380 ret = PTR_ERR(mrec); 1381 goto undo_alloc; 1382 } 1383 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1384 if (unlikely(!ctx)) { 1385 ntfs_error(vol->sb, "Failed to get search context."); 1386 ret = -ENOMEM; 1387 goto undo_alloc; 1388 } 1389 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1390 mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, 1391 0, ctx); 1392 if (unlikely(ret)) { 1393 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1394 "mft bitmap attribute."); 1395 if (ret == -ENOENT) 1396 ret = -EIO; 1397 goto undo_alloc; 1398 } 1399 a = ctx->attr; 1400 ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); 1401 /* Search back for the previous last allocated cluster of mft bitmap. */ 1402 for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) { 1403 if (ll >= rl2->vcn) 1404 break; 1405 } 1406 BUG_ON(ll < rl2->vcn); 1407 BUG_ON(ll >= rl2->vcn + rl2->length); 1408 /* Get the size for the new mapping pairs array for this extent. */ 1409 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll); 1410 if (unlikely(mp_size <= 0)) { 1411 ntfs_error(vol->sb, "Get size for mapping pairs failed for " 1412 "mft bitmap attribute extent."); 1413 ret = mp_size; 1414 if (!ret) 1415 ret = -EIO; 1416 goto undo_alloc; 1417 } 1418 /* Expand the attribute record if necessary. */ 1419 old_alen = le32_to_cpu(a->length); 1420 ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + 1421 le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); 1422 if (unlikely(ret)) { 1423 if (ret != -ENOSPC) { 1424 ntfs_error(vol->sb, "Failed to resize attribute " 1425 "record for mft bitmap attribute."); 1426 goto undo_alloc; 1427 } 1428 // TODO: Deal with this by moving this extent to a new mft 1429 // record or by starting a new extent in a new mft record or by 1430 // moving other attributes out of this mft record. 1431 ntfs_error(vol->sb, "Not enough space in this mft record to " 1432 "accomodate extended mft bitmap attribute " 1433 "extent. Cannot handle this yet."); 1434 ret = -EOPNOTSUPP; 1435 goto undo_alloc; 1436 } 1437 status.mp_rebuilt = 1; 1438 /* Generate the mapping pairs array directly into the attr record. */ 1439 ret = ntfs_mapping_pairs_build(vol, (u8*)a + 1440 le16_to_cpu(a->data.non_resident.mapping_pairs_offset), 1441 mp_size, rl2, ll, NULL); 1442 if (unlikely(ret)) { 1443 ntfs_error(vol->sb, "Failed to build mapping pairs array for " 1444 "mft bitmap attribute."); 1445 goto undo_alloc; 1446 } 1447 /* Update the highest_vcn. */ 1448 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); 1449 /* 1450 * We now have extended the mft bitmap allocated_size by one cluster. 1451 * Reflect this in the ntfs_inode structure and the attribute record. 1452 */ 1453 if (a->data.non_resident.lowest_vcn) { 1454 /* 1455 * We are not in the first attribute extent, switch to it, but 1456 * first ensure the changes will make it to disk later. 1457 */ 1458 flush_dcache_mft_record_page(ctx->ntfs_ino); 1459 mark_mft_record_dirty(ctx->ntfs_ino); 1460 ntfs_attr_reinit_search_ctx(ctx); 1461 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1462 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 1463 0, ctx); 1464 if (unlikely(ret)) { 1465 ntfs_error(vol->sb, "Failed to find first attribute " 1466 "extent of mft bitmap attribute."); 1467 goto restore_undo_alloc; 1468 } 1469 a = ctx->attr; 1470 } 1471 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1472 mftbmp_ni->allocated_size += vol->cluster_size; 1473 a->data.non_resident.allocated_size = 1474 cpu_to_sle64(mftbmp_ni->allocated_size); 1475 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1476 /* Ensure the changes make it to disk. */ 1477 flush_dcache_mft_record_page(ctx->ntfs_ino); 1478 mark_mft_record_dirty(ctx->ntfs_ino); 1479 ntfs_attr_put_search_ctx(ctx); 1480 unmap_mft_record(mft_ni); 1481 up_write(&mftbmp_ni->runlist.lock); 1482 ntfs_debug("Done."); 1483 return 0; 1484 restore_undo_alloc: 1485 ntfs_attr_reinit_search_ctx(ctx); 1486 if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1487 mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, 1488 0, ctx)) { 1489 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1490 "mft bitmap attribute.%s", es); 1491 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1492 mftbmp_ni->allocated_size += vol->cluster_size; 1493 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1494 ntfs_attr_put_search_ctx(ctx); 1495 unmap_mft_record(mft_ni); 1496 up_write(&mftbmp_ni->runlist.lock); 1497 /* 1498 * The only thing that is now wrong is ->allocated_size of the 1499 * base attribute extent which chkdsk should be able to fix. 1500 */ 1501 NVolSetErrors(vol); 1502 return ret; 1503 } 1504 a = ctx->attr; 1505 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2); 1506 undo_alloc: 1507 if (status.added_cluster) { 1508 /* Truncate the last run in the runlist by one cluster. */ 1509 rl->length--; 1510 rl[1].vcn--; 1511 } else if (status.added_run) { 1512 lcn = rl->lcn; 1513 /* Remove the last run from the runlist. */ 1514 rl->lcn = rl[1].lcn; 1515 rl->length = 0; 1516 } 1517 /* Deallocate the cluster. */ 1518 down_write(&vol->lcnbmp_lock); 1519 if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { 1520 ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es); 1521 NVolSetErrors(vol); 1522 } 1523 up_write(&vol->lcnbmp_lock); 1524 if (status.mp_rebuilt) { 1525 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( 1526 a->data.non_resident.mapping_pairs_offset), 1527 old_alen - le16_to_cpu( 1528 a->data.non_resident.mapping_pairs_offset), 1529 rl2, ll, NULL)) { 1530 ntfs_error(vol->sb, "Failed to restore mapping pairs " 1531 "array.%s", es); 1532 NVolSetErrors(vol); 1533 } 1534 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { 1535 ntfs_error(vol->sb, "Failed to restore attribute " 1536 "record.%s", es); 1537 NVolSetErrors(vol); 1538 } 1539 flush_dcache_mft_record_page(ctx->ntfs_ino); 1540 mark_mft_record_dirty(ctx->ntfs_ino); 1541 } 1542 if (ctx) 1543 ntfs_attr_put_search_ctx(ctx); 1544 if (!IS_ERR(mrec)) 1545 unmap_mft_record(mft_ni); 1546 up_write(&mftbmp_ni->runlist.lock); 1547 return ret; 1548 } 1549 1550 /** 1551 * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data 1552 * @vol: volume on which to extend the mft bitmap attribute 1553 * 1554 * Extend the initialized portion of the mft bitmap attribute on the ntfs 1555 * volume @vol by 8 bytes. 1556 * 1557 * Note: Only changes initialized_size and data_size, i.e. requires that 1558 * allocated_size is big enough to fit the new initialized_size. 1559 * 1560 * Return 0 on success and -error on error. 1561 * 1562 * Locking: Caller must hold vol->mftbmp_lock for writing. 1563 */ 1564 static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol) 1565 { 1566 s64 old_data_size, old_initialized_size; 1567 unsigned long flags; 1568 struct inode *mftbmp_vi; 1569 ntfs_inode *mft_ni, *mftbmp_ni; 1570 ntfs_attr_search_ctx *ctx; 1571 MFT_RECORD *mrec; 1572 ATTR_RECORD *a; 1573 int ret; 1574 1575 ntfs_debug("Extending mft bitmap initiailized (and data) size."); 1576 mft_ni = NTFS_I(vol->mft_ino); 1577 mftbmp_vi = vol->mftbmp_ino; 1578 mftbmp_ni = NTFS_I(mftbmp_vi); 1579 /* Get the attribute record. */ 1580 mrec = map_mft_record(mft_ni); 1581 if (IS_ERR(mrec)) { 1582 ntfs_error(vol->sb, "Failed to map mft record."); 1583 return PTR_ERR(mrec); 1584 } 1585 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1586 if (unlikely(!ctx)) { 1587 ntfs_error(vol->sb, "Failed to get search context."); 1588 ret = -ENOMEM; 1589 goto unm_err_out; 1590 } 1591 ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1592 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx); 1593 if (unlikely(ret)) { 1594 ntfs_error(vol->sb, "Failed to find first attribute extent of " 1595 "mft bitmap attribute."); 1596 if (ret == -ENOENT) 1597 ret = -EIO; 1598 goto put_err_out; 1599 } 1600 a = ctx->attr; 1601 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1602 old_data_size = i_size_read(mftbmp_vi); 1603 old_initialized_size = mftbmp_ni->initialized_size; 1604 /* 1605 * We can simply update the initialized_size before filling the space 1606 * with zeroes because the caller is holding the mft bitmap lock for 1607 * writing which ensures that no one else is trying to access the data. 1608 */ 1609 mftbmp_ni->initialized_size += 8; 1610 a->data.non_resident.initialized_size = 1611 cpu_to_sle64(mftbmp_ni->initialized_size); 1612 if (mftbmp_ni->initialized_size > old_data_size) { 1613 i_size_write(mftbmp_vi, mftbmp_ni->initialized_size); 1614 a->data.non_resident.data_size = 1615 cpu_to_sle64(mftbmp_ni->initialized_size); 1616 } 1617 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1618 /* Ensure the changes make it to disk. */ 1619 flush_dcache_mft_record_page(ctx->ntfs_ino); 1620 mark_mft_record_dirty(ctx->ntfs_ino); 1621 ntfs_attr_put_search_ctx(ctx); 1622 unmap_mft_record(mft_ni); 1623 /* Initialize the mft bitmap attribute value with zeroes. */ 1624 ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0); 1625 if (likely(!ret)) { 1626 ntfs_debug("Done. (Wrote eight initialized bytes to mft " 1627 "bitmap."); 1628 return 0; 1629 } 1630 ntfs_error(vol->sb, "Failed to write to mft bitmap."); 1631 /* Try to recover from the error. */ 1632 mrec = map_mft_record(mft_ni); 1633 if (IS_ERR(mrec)) { 1634 ntfs_error(vol->sb, "Failed to map mft record.%s", es); 1635 NVolSetErrors(vol); 1636 return ret; 1637 } 1638 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1639 if (unlikely(!ctx)) { 1640 ntfs_error(vol->sb, "Failed to get search context.%s", es); 1641 NVolSetErrors(vol); 1642 goto unm_err_out; 1643 } 1644 if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, 1645 mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) { 1646 ntfs_error(vol->sb, "Failed to find first attribute extent of " 1647 "mft bitmap attribute.%s", es); 1648 NVolSetErrors(vol); 1649 put_err_out: 1650 ntfs_attr_put_search_ctx(ctx); 1651 unm_err_out: 1652 unmap_mft_record(mft_ni); 1653 goto err_out; 1654 } 1655 a = ctx->attr; 1656 write_lock_irqsave(&mftbmp_ni->size_lock, flags); 1657 mftbmp_ni->initialized_size = old_initialized_size; 1658 a->data.non_resident.initialized_size = 1659 cpu_to_sle64(old_initialized_size); 1660 if (i_size_read(mftbmp_vi) != old_data_size) { 1661 i_size_write(mftbmp_vi, old_data_size); 1662 a->data.non_resident.data_size = cpu_to_sle64(old_data_size); 1663 } 1664 write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1665 flush_dcache_mft_record_page(ctx->ntfs_ino); 1666 mark_mft_record_dirty(ctx->ntfs_ino); 1667 ntfs_attr_put_search_ctx(ctx); 1668 unmap_mft_record(mft_ni); 1669 #ifdef DEBUG 1670 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 1671 ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, " 1672 "data_size 0x%llx, initialized_size 0x%llx.", 1673 (long long)mftbmp_ni->allocated_size, 1674 (long long)i_size_read(mftbmp_vi), 1675 (long long)mftbmp_ni->initialized_size); 1676 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 1677 #endif /* DEBUG */ 1678 err_out: 1679 return ret; 1680 } 1681 1682 /** 1683 * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute 1684 * @vol: volume on which to extend the mft data attribute 1685 * 1686 * Extend the mft data attribute on the ntfs volume @vol by 16 mft records 1687 * worth of clusters or if not enough space for this by one mft record worth 1688 * of clusters. 1689 * 1690 * Note: Only changes allocated_size, i.e. does not touch initialized_size or 1691 * data_size. 1692 * 1693 * Return 0 on success and -errno on error. 1694 * 1695 * Locking: - Caller must hold vol->mftbmp_lock for writing. 1696 * - This function takes NTFS_I(vol->mft_ino)->runlist.lock for 1697 * writing and releases it before returning. 1698 * - This function calls functions which take vol->lcnbmp_lock for 1699 * writing and release it before returning. 1700 */ 1701 static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol) 1702 { 1703 LCN lcn; 1704 VCN old_last_vcn; 1705 s64 min_nr, nr, ll; 1706 unsigned long flags; 1707 ntfs_inode *mft_ni; 1708 runlist_element *rl, *rl2; 1709 ntfs_attr_search_ctx *ctx = NULL; 1710 MFT_RECORD *mrec; 1711 ATTR_RECORD *a = NULL; 1712 int ret, mp_size; 1713 u32 old_alen = 0; 1714 BOOL mp_rebuilt = FALSE; 1715 1716 ntfs_debug("Extending mft data allocation."); 1717 mft_ni = NTFS_I(vol->mft_ino); 1718 /* 1719 * Determine the preferred allocation location, i.e. the last lcn of 1720 * the mft data attribute. The allocated size of the mft data 1721 * attribute cannot be zero so we are ok to do this. 1722 * ntfs_find_vcn() returns the runlist locked on success. 1723 */ 1724 read_lock_irqsave(&mft_ni->size_lock, flags); 1725 ll = mft_ni->allocated_size; 1726 read_unlock_irqrestore(&mft_ni->size_lock, flags); 1727 rl = ntfs_find_vcn(mft_ni, (ll - 1) >> vol->cluster_size_bits, TRUE); 1728 if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { 1729 ntfs_error(vol->sb, "Failed to determine last allocated " 1730 "cluster of mft data attribute."); 1731 if (!IS_ERR(rl)) { 1732 up_write(&mft_ni->runlist.lock); 1733 ret = -EIO; 1734 } else 1735 ret = PTR_ERR(rl); 1736 return ret; 1737 } 1738 lcn = rl->lcn + rl->length; 1739 ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn); 1740 /* Minimum allocation is one mft record worth of clusters. */ 1741 min_nr = vol->mft_record_size >> vol->cluster_size_bits; 1742 if (!min_nr) 1743 min_nr = 1; 1744 /* Want to allocate 16 mft records worth of clusters. */ 1745 nr = vol->mft_record_size << 4 >> vol->cluster_size_bits; 1746 if (!nr) 1747 nr = min_nr; 1748 /* Ensure we do not go above 2^32-1 mft records. */ 1749 read_lock_irqsave(&mft_ni->size_lock, flags); 1750 ll = mft_ni->allocated_size; 1751 read_unlock_irqrestore(&mft_ni->size_lock, flags); 1752 if (unlikely((ll + (nr << vol->cluster_size_bits)) >> 1753 vol->mft_record_size_bits >= (1ll << 32))) { 1754 nr = min_nr; 1755 if (unlikely((ll + (nr << vol->cluster_size_bits)) >> 1756 vol->mft_record_size_bits >= (1ll << 32))) { 1757 ntfs_warning(vol->sb, "Cannot allocate mft record " 1758 "because the maximum number of inodes " 1759 "(2^32) has already been reached."); 1760 up_write(&mft_ni->runlist.lock); 1761 return -ENOSPC; 1762 } 1763 } 1764 ntfs_debug("Trying mft data allocation with %s cluster count %lli.", 1765 nr > min_nr ? "default" : "minimal", (long long)nr); 1766 old_last_vcn = rl[1].vcn; 1767 do { 1768 rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE); 1769 if (likely(!IS_ERR(rl2))) 1770 break; 1771 if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) { 1772 ntfs_error(vol->sb, "Failed to allocate the minimal " 1773 "number of clusters (%lli) for the " 1774 "mft data attribute.", (long long)nr); 1775 up_write(&mft_ni->runlist.lock); 1776 return PTR_ERR(rl2); 1777 } 1778 /* 1779 * There is not enough space to do the allocation, but there 1780 * might be enough space to do a minimal allocation so try that 1781 * before failing. 1782 */ 1783 nr = min_nr; 1784 ntfs_debug("Retrying mft data allocation with minimal cluster " 1785 "count %lli.", (long long)nr); 1786 } while (1); 1787 rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2); 1788 if (IS_ERR(rl)) { 1789 up_write(&mft_ni->runlist.lock); 1790 ntfs_error(vol->sb, "Failed to merge runlists for mft data " 1791 "attribute."); 1792 if (ntfs_cluster_free_from_rl(vol, rl2)) { 1793 ntfs_error(vol->sb, "Failed to dealocate clusters " 1794 "from the mft data attribute.%s", es); 1795 NVolSetErrors(vol); 1796 } 1797 ntfs_free(rl2); 1798 return PTR_ERR(rl); 1799 } 1800 mft_ni->runlist.rl = rl; 1801 ntfs_debug("Allocated %lli clusters.", nr); 1802 /* Find the last run in the new runlist. */ 1803 for (; rl[1].length; rl++) 1804 ; 1805 /* Update the attribute record as well. */ 1806 mrec = map_mft_record(mft_ni); 1807 if (IS_ERR(mrec)) { 1808 ntfs_error(vol->sb, "Failed to map mft record."); 1809 ret = PTR_ERR(mrec); 1810 goto undo_alloc; 1811 } 1812 ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); 1813 if (unlikely(!ctx)) { 1814 ntfs_error(vol->sb, "Failed to get search context."); 1815 ret = -ENOMEM; 1816 goto undo_alloc; 1817 } 1818 ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 1819 CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx); 1820 if (unlikely(ret)) { 1821 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1822 "mft data attribute."); 1823 if (ret == -ENOENT) 1824 ret = -EIO; 1825 goto undo_alloc; 1826 } 1827 a = ctx->attr; 1828 ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); 1829 /* Search back for the previous last allocated cluster of mft bitmap. */ 1830 for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) { 1831 if (ll >= rl2->vcn) 1832 break; 1833 } 1834 BUG_ON(ll < rl2->vcn); 1835 BUG_ON(ll >= rl2->vcn + rl2->length); 1836 /* Get the size for the new mapping pairs array for this extent. */ 1837 mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll); 1838 if (unlikely(mp_size <= 0)) { 1839 ntfs_error(vol->sb, "Get size for mapping pairs failed for " 1840 "mft data attribute extent."); 1841 ret = mp_size; 1842 if (!ret) 1843 ret = -EIO; 1844 goto undo_alloc; 1845 } 1846 /* Expand the attribute record if necessary. */ 1847 old_alen = le32_to_cpu(a->length); 1848 ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + 1849 le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); 1850 if (unlikely(ret)) { 1851 if (ret != -ENOSPC) { 1852 ntfs_error(vol->sb, "Failed to resize attribute " 1853 "record for mft data attribute."); 1854 goto undo_alloc; 1855 } 1856 // TODO: Deal with this by moving this extent to a new mft 1857 // record or by starting a new extent in a new mft record or by 1858 // moving other attributes out of this mft record. 1859 // Note: Use the special reserved mft records and ensure that 1860 // this extent is not required to find the mft record in 1861 // question. 1862 ntfs_error(vol->sb, "Not enough space in this mft record to " 1863 "accomodate extended mft data attribute " 1864 "extent. Cannot handle this yet."); 1865 ret = -EOPNOTSUPP; 1866 goto undo_alloc; 1867 } 1868 mp_rebuilt = TRUE; 1869 /* Generate the mapping pairs array directly into the attr record. */ 1870 ret = ntfs_mapping_pairs_build(vol, (u8*)a + 1871 le16_to_cpu(a->data.non_resident.mapping_pairs_offset), 1872 mp_size, rl2, ll, NULL); 1873 if (unlikely(ret)) { 1874 ntfs_error(vol->sb, "Failed to build mapping pairs array of " 1875 "mft data attribute."); 1876 goto undo_alloc; 1877 } 1878 /* Update the highest_vcn. */ 1879 a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); 1880 /* 1881 * We now have extended the mft data allocated_size by nr clusters. 1882 * Reflect this in the ntfs_inode structure and the attribute record. 1883 * @rl is the last (non-terminator) runlist element of mft data 1884 * attribute. 1885 */ 1886 if (a->data.non_resident.lowest_vcn) { 1887 /* 1888 * We are not in the first attribute extent, switch to it, but 1889 * first ensure the changes will make it to disk later. 1890 */ 1891 flush_dcache_mft_record_page(ctx->ntfs_ino); 1892 mark_mft_record_dirty(ctx->ntfs_ino); 1893 ntfs_attr_reinit_search_ctx(ctx); 1894 ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, 1895 mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, 1896 ctx); 1897 if (unlikely(ret)) { 1898 ntfs_error(vol->sb, "Failed to find first attribute " 1899 "extent of mft data attribute."); 1900 goto restore_undo_alloc; 1901 } 1902 a = ctx->attr; 1903 } 1904 write_lock_irqsave(&mft_ni->size_lock, flags); 1905 mft_ni->allocated_size += nr << vol->cluster_size_bits; 1906 a->data.non_resident.allocated_size = 1907 cpu_to_sle64(mft_ni->allocated_size); 1908 write_unlock_irqrestore(&mft_ni->size_lock, flags); 1909 /* Ensure the changes make it to disk. */ 1910 flush_dcache_mft_record_page(ctx->ntfs_ino); 1911 mark_mft_record_dirty(ctx->ntfs_ino); 1912 ntfs_attr_put_search_ctx(ctx); 1913 unmap_mft_record(mft_ni); 1914 up_write(&mft_ni->runlist.lock); 1915 ntfs_debug("Done."); 1916 return 0; 1917 restore_undo_alloc: 1918 ntfs_attr_reinit_search_ctx(ctx); 1919 if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 1920 CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) { 1921 ntfs_error(vol->sb, "Failed to find last attribute extent of " 1922 "mft data attribute.%s", es); 1923 write_lock_irqsave(&mft_ni->size_lock, flags); 1924 mft_ni->allocated_size += nr << vol->cluster_size_bits; 1925 write_unlock_irqrestore(&mft_ni->size_lock, flags); 1926 ntfs_attr_put_search_ctx(ctx); 1927 unmap_mft_record(mft_ni); 1928 up_write(&mft_ni->runlist.lock); 1929 /* 1930 * The only thing that is now wrong is ->allocated_size of the 1931 * base attribute extent which chkdsk should be able to fix. 1932 */ 1933 NVolSetErrors(vol); 1934 return ret; 1935 } 1936 a = ctx->attr; 1937 a->data.non_resident.highest_vcn = cpu_to_sle64(old_last_vcn - 1); 1938 undo_alloc: 1939 if (ntfs_cluster_free(vol->mft_ino, old_last_vcn, -1) < 0) { 1940 ntfs_error(vol->sb, "Failed to free clusters from mft data " 1941 "attribute.%s", es); 1942 NVolSetErrors(vol); 1943 } 1944 if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) { 1945 ntfs_error(vol->sb, "Failed to truncate mft data attribute " 1946 "runlist.%s", es); 1947 NVolSetErrors(vol); 1948 } 1949 if (mp_rebuilt) { 1950 if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( 1951 a->data.non_resident.mapping_pairs_offset), 1952 old_alen - le16_to_cpu( 1953 a->data.non_resident.mapping_pairs_offset), 1954 rl2, ll, NULL)) { 1955 ntfs_error(vol->sb, "Failed to restore mapping pairs " 1956 "array.%s", es); 1957 NVolSetErrors(vol); 1958 } 1959 if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { 1960 ntfs_error(vol->sb, "Failed to restore attribute " 1961 "record.%s", es); 1962 NVolSetErrors(vol); 1963 } 1964 flush_dcache_mft_record_page(ctx->ntfs_ino); 1965 mark_mft_record_dirty(ctx->ntfs_ino); 1966 } 1967 if (ctx) 1968 ntfs_attr_put_search_ctx(ctx); 1969 if (!IS_ERR(mrec)) 1970 unmap_mft_record(mft_ni); 1971 up_write(&mft_ni->runlist.lock); 1972 return ret; 1973 } 1974 1975 /** 1976 * ntfs_mft_record_layout - layout an mft record into a memory buffer 1977 * @vol: volume to which the mft record will belong 1978 * @mft_no: mft reference specifying the mft record number 1979 * @m: destination buffer of size >= @vol->mft_record_size bytes 1980 * 1981 * Layout an empty, unused mft record with the mft record number @mft_no into 1982 * the buffer @m. The volume @vol is needed because the mft record structure 1983 * was modified in NTFS 3.1 so we need to know which volume version this mft 1984 * record will be used on. 1985 * 1986 * Return 0 on success and -errno on error. 1987 */ 1988 static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no, 1989 MFT_RECORD *m) 1990 { 1991 ATTR_RECORD *a; 1992 1993 ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); 1994 if (mft_no >= (1ll << 32)) { 1995 ntfs_error(vol->sb, "Mft record number 0x%llx exceeds " 1996 "maximum of 2^32.", (long long)mft_no); 1997 return -ERANGE; 1998 } 1999 /* Start by clearing the whole mft record to gives us a clean slate. */ 2000 memset(m, 0, vol->mft_record_size); 2001 /* Aligned to 2-byte boundary. */ 2002 if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver)) 2003 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1); 2004 else { 2005 m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1); 2006 /* 2007 * Set the NTFS 3.1+ specific fields while we know that the 2008 * volume version is 3.1+. 2009 */ 2010 m->reserved = 0; 2011 m->mft_record_number = cpu_to_le32((u32)mft_no); 2012 } 2013 m->magic = magic_FILE; 2014 if (vol->mft_record_size >= NTFS_BLOCK_SIZE) 2015 m->usa_count = cpu_to_le16(vol->mft_record_size / 2016 NTFS_BLOCK_SIZE + 1); 2017 else { 2018 m->usa_count = cpu_to_le16(1); 2019 ntfs_warning(vol->sb, "Sector size is bigger than mft record " 2020 "size. Setting usa_count to 1. If chkdsk " 2021 "reports this as corruption, please email " 2022 "linux-ntfs-dev@lists.sourceforge.net stating " 2023 "that you saw this message and that the " 2024 "modified file system created was corrupt. " 2025 "Thank you."); 2026 } 2027 /* Set the update sequence number to 1. */ 2028 *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1); 2029 m->lsn = 0; 2030 m->sequence_number = cpu_to_le16(1); 2031 m->link_count = 0; 2032 /* 2033 * Place the attributes straight after the update sequence array, 2034 * aligned to 8-byte boundary. 2035 */ 2036 m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) + 2037 (le16_to_cpu(m->usa_count) << 1) + 7) & ~7); 2038 m->flags = 0; 2039 /* 2040 * Using attrs_offset plus eight bytes (for the termination attribute). 2041 * attrs_offset is already aligned to 8-byte boundary, so no need to 2042 * align again. 2043 */ 2044 m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8); 2045 m->bytes_allocated = cpu_to_le32(vol->mft_record_size); 2046 m->base_mft_record = 0; 2047 m->next_attr_instance = 0; 2048 /* Add the termination attribute. */ 2049 a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset)); 2050 a->type = AT_END; 2051 a->length = 0; 2052 ntfs_debug("Done."); 2053 return 0; 2054 } 2055 2056 /** 2057 * ntfs_mft_record_format - format an mft record on an ntfs volume 2058 * @vol: volume on which to format the mft record 2059 * @mft_no: mft record number to format 2060 * 2061 * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused 2062 * mft record into the appropriate place of the mft data attribute. This is 2063 * used when extending the mft data attribute. 2064 * 2065 * Return 0 on success and -errno on error. 2066 */ 2067 static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no) 2068 { 2069 loff_t i_size; 2070 struct inode *mft_vi = vol->mft_ino; 2071 struct page *page; 2072 MFT_RECORD *m; 2073 pgoff_t index, end_index; 2074 unsigned int ofs; 2075 int err; 2076 2077 ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); 2078 /* 2079 * The index into the page cache and the offset within the page cache 2080 * page of the wanted mft record. 2081 */ 2082 index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; 2083 ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 2084 /* The maximum valid index into the page cache for $MFT's data. */ 2085 i_size = i_size_read(mft_vi); 2086 end_index = i_size >> PAGE_CACHE_SHIFT; 2087 if (unlikely(index >= end_index)) { 2088 if (unlikely(index > end_index || ofs + vol->mft_record_size >= 2089 (i_size & ~PAGE_CACHE_MASK))) { 2090 ntfs_error(vol->sb, "Tried to format non-existing mft " 2091 "record 0x%llx.", (long long)mft_no); 2092 return -ENOENT; 2093 } 2094 } 2095 /* Read, map, and pin the page containing the mft record. */ 2096 page = ntfs_map_page(mft_vi->i_mapping, index); 2097 if (unlikely(IS_ERR(page))) { 2098 ntfs_error(vol->sb, "Failed to map page containing mft record " 2099 "to format 0x%llx.", (long long)mft_no); 2100 return PTR_ERR(page); 2101 } 2102 lock_page(page); 2103 BUG_ON(!PageUptodate(page)); 2104 ClearPageUptodate(page); 2105 m = (MFT_RECORD*)((u8*)page_address(page) + ofs); 2106 err = ntfs_mft_record_layout(vol, mft_no, m); 2107 if (unlikely(err)) { 2108 ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.", 2109 (long long)mft_no); 2110 SetPageUptodate(page); 2111 unlock_page(page); 2112 ntfs_unmap_page(page); 2113 return err; 2114 } 2115 flush_dcache_page(page); 2116 SetPageUptodate(page); 2117 unlock_page(page); 2118 /* 2119 * Make sure the mft record is written out to disk. We could use 2120 * ilookup5() to check if an inode is in icache and so on but this is 2121 * unnecessary as ntfs_writepage() will write the dirty record anyway. 2122 */ 2123 mark_ntfs_record_dirty(page, ofs); 2124 ntfs_unmap_page(page); 2125 ntfs_debug("Done."); 2126 return 0; 2127 } 2128 2129 /** 2130 * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume 2131 * @vol: [IN] volume on which to allocate the mft record 2132 * @mode: [IN] mode if want a file or directory, i.e. base inode or 0 2133 * @base_ni: [IN] open base inode if allocating an extent mft record or NULL 2134 * @mrec: [OUT] on successful return this is the mapped mft record 2135 * 2136 * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol. 2137 * 2138 * If @base_ni is NULL make the mft record a base mft record, i.e. a file or 2139 * direvctory inode, and allocate it at the default allocator position. In 2140 * this case @mode is the file mode as given to us by the caller. We in 2141 * particular use @mode to distinguish whether a file or a directory is being 2142 * created (S_IFDIR(mode) and S_IFREG(mode), respectively). 2143 * 2144 * If @base_ni is not NULL make the allocated mft record an extent record, 2145 * allocate it starting at the mft record after the base mft record and attach 2146 * the allocated and opened ntfs inode to the base inode @base_ni. In this 2147 * case @mode must be 0 as it is meaningless for extent inodes. 2148 * 2149 * You need to check the return value with IS_ERR(). If false, the function 2150 * was successful and the return value is the now opened ntfs inode of the 2151 * allocated mft record. *@mrec is then set to the allocated, mapped, pinned, 2152 * and locked mft record. If IS_ERR() is true, the function failed and the 2153 * error code is obtained from PTR_ERR(return value). *@mrec is undefined in 2154 * this case. 2155 * 2156 * Allocation strategy: 2157 * 2158 * To find a free mft record, we scan the mft bitmap for a zero bit. To 2159 * optimize this we start scanning at the place specified by @base_ni or if 2160 * @base_ni is NULL we start where we last stopped and we perform wrap around 2161 * when we reach the end. Note, we do not try to allocate mft records below 2162 * number 24 because numbers 0 to 15 are the defined system files anyway and 16 2163 * to 24 are special in that they are used for storing extension mft records 2164 * for the $DATA attribute of $MFT. This is required to avoid the possibility 2165 * of creating a runlist with a circular dependency which once written to disk 2166 * can never be read in again. Windows will only use records 16 to 24 for 2167 * normal files if the volume is completely out of space. We never use them 2168 * which means that when the volume is really out of space we cannot create any 2169 * more files while Windows can still create up to 8 small files. We can start 2170 * doing this at some later time, it does not matter much for now. 2171 * 2172 * When scanning the mft bitmap, we only search up to the last allocated mft 2173 * record. If there are no free records left in the range 24 to number of 2174 * allocated mft records, then we extend the $MFT/$DATA attribute in order to 2175 * create free mft records. We extend the allocated size of $MFT/$DATA by 16 2176 * records at a time or one cluster, if cluster size is above 16kiB. If there 2177 * is not sufficient space to do this, we try to extend by a single mft record 2178 * or one cluster, if cluster size is above the mft record size. 2179 * 2180 * No matter how many mft records we allocate, we initialize only the first 2181 * allocated mft record, incrementing mft data size and initialized size 2182 * accordingly, open an ntfs_inode for it and return it to the caller, unless 2183 * there are less than 24 mft records, in which case we allocate and initialize 2184 * mft records until we reach record 24 which we consider as the first free mft 2185 * record for use by normal files. 2186 * 2187 * If during any stage we overflow the initialized data in the mft bitmap, we 2188 * extend the initialized size (and data size) by 8 bytes, allocating another 2189 * cluster if required. The bitmap data size has to be at least equal to the 2190 * number of mft records in the mft, but it can be bigger, in which case the 2191 * superflous bits are padded with zeroes. 2192 * 2193 * Thus, when we return successfully (IS_ERR() is false), we will have: 2194 * - initialized / extended the mft bitmap if necessary, 2195 * - initialized / extended the mft data if necessary, 2196 * - set the bit corresponding to the mft record being allocated in the 2197 * mft bitmap, 2198 * - opened an ntfs_inode for the allocated mft record, and we will have 2199 * - returned the ntfs_inode as well as the allocated mapped, pinned, and 2200 * locked mft record. 2201 * 2202 * On error, the volume will be left in a consistent state and no record will 2203 * be allocated. If rolling back a partial operation fails, we may leave some 2204 * inconsistent metadata in which case we set NVolErrors() so the volume is 2205 * left dirty when unmounted. 2206 * 2207 * Note, this function cannot make use of most of the normal functions, like 2208 * for example for attribute resizing, etc, because when the run list overflows 2209 * the base mft record and an attribute list is used, it is very important that 2210 * the extension mft records used to store the $DATA attribute of $MFT can be 2211 * reached without having to read the information contained inside them, as 2212 * this would make it impossible to find them in the first place after the 2213 * volume is unmounted. $MFT/$BITMAP probably does not need to follow this 2214 * rule because the bitmap is not essential for finding the mft records, but on 2215 * the other hand, handling the bitmap in this special way would make life 2216 * easier because otherwise there might be circular invocations of functions 2217 * when reading the bitmap. 2218 */ 2219 ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode, 2220 ntfs_inode *base_ni, MFT_RECORD **mrec) 2221 { 2222 s64 ll, bit, old_data_initialized, old_data_size; 2223 unsigned long flags; 2224 struct inode *vi; 2225 struct page *page; 2226 ntfs_inode *mft_ni, *mftbmp_ni, *ni; 2227 ntfs_attr_search_ctx *ctx; 2228 MFT_RECORD *m; 2229 ATTR_RECORD *a; 2230 pgoff_t index; 2231 unsigned int ofs; 2232 int err; 2233 le16 seq_no, usn; 2234 BOOL record_formatted = FALSE; 2235 2236 if (base_ni) { 2237 ntfs_debug("Entering (allocating an extent mft record for " 2238 "base mft record 0x%llx).", 2239 (long long)base_ni->mft_no); 2240 /* @mode and @base_ni are mutually exclusive. */ 2241 BUG_ON(mode); 2242 } else 2243 ntfs_debug("Entering (allocating a base mft record)."); 2244 if (mode) { 2245 /* @mode and @base_ni are mutually exclusive. */ 2246 BUG_ON(base_ni); 2247 /* We only support creation of normal files and directories. */ 2248 if (!S_ISREG(mode) && !S_ISDIR(mode)) 2249 return ERR_PTR(-EOPNOTSUPP); 2250 } 2251 BUG_ON(!mrec); 2252 mft_ni = NTFS_I(vol->mft_ino); 2253 mftbmp_ni = NTFS_I(vol->mftbmp_ino); 2254 down_write(&vol->mftbmp_lock); 2255 bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni); 2256 if (bit >= 0) { 2257 ntfs_debug("Found and allocated free record (#1), bit 0x%llx.", 2258 (long long)bit); 2259 goto have_alloc_rec; 2260 } 2261 if (bit != -ENOSPC) { 2262 up_write(&vol->mftbmp_lock); 2263 return ERR_PTR(bit); 2264 } 2265 /* 2266 * No free mft records left. If the mft bitmap already covers more 2267 * than the currently used mft records, the next records are all free, 2268 * so we can simply allocate the first unused mft record. 2269 * Note: We also have to make sure that the mft bitmap at least covers 2270 * the first 24 mft records as they are special and whilst they may not 2271 * be in use, we do not allocate from them. 2272 */ 2273 read_lock_irqsave(&mft_ni->size_lock, flags); 2274 ll = mft_ni->initialized_size >> vol->mft_record_size_bits; 2275 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2276 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2277 old_data_initialized = mftbmp_ni->initialized_size; 2278 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2279 if (old_data_initialized << 3 > ll && old_data_initialized > 3) { 2280 bit = ll; 2281 if (bit < 24) 2282 bit = 24; 2283 if (unlikely(bit >= (1ll << 32))) 2284 goto max_err_out; 2285 ntfs_debug("Found free record (#2), bit 0x%llx.", 2286 (long long)bit); 2287 goto found_free_rec; 2288 } 2289 /* 2290 * The mft bitmap needs to be expanded until it covers the first unused 2291 * mft record that we can allocate. 2292 * Note: The smallest mft record we allocate is mft record 24. 2293 */ 2294 bit = old_data_initialized << 3; 2295 if (unlikely(bit >= (1ll << 32))) 2296 goto max_err_out; 2297 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2298 old_data_size = mftbmp_ni->allocated_size; 2299 ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, " 2300 "data_size 0x%llx, initialized_size 0x%llx.", 2301 (long long)old_data_size, 2302 (long long)i_size_read(vol->mftbmp_ino), 2303 (long long)old_data_initialized); 2304 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2305 if (old_data_initialized + 8 > old_data_size) { 2306 /* Need to extend bitmap by one more cluster. */ 2307 ntfs_debug("mftbmp: initialized_size + 8 > allocated_size."); 2308 err = ntfs_mft_bitmap_extend_allocation_nolock(vol); 2309 if (unlikely(err)) { 2310 up_write(&vol->mftbmp_lock); 2311 goto err_out; 2312 } 2313 #ifdef DEBUG 2314 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2315 ntfs_debug("Status of mftbmp after allocation extension: " 2316 "allocated_size 0x%llx, data_size 0x%llx, " 2317 "initialized_size 0x%llx.", 2318 (long long)mftbmp_ni->allocated_size, 2319 (long long)i_size_read(vol->mftbmp_ino), 2320 (long long)mftbmp_ni->initialized_size); 2321 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2322 #endif /* DEBUG */ 2323 } 2324 /* 2325 * We now have sufficient allocated space, extend the initialized_size 2326 * as well as the data_size if necessary and fill the new space with 2327 * zeroes. 2328 */ 2329 err = ntfs_mft_bitmap_extend_initialized_nolock(vol); 2330 if (unlikely(err)) { 2331 up_write(&vol->mftbmp_lock); 2332 goto err_out; 2333 } 2334 #ifdef DEBUG 2335 read_lock_irqsave(&mftbmp_ni->size_lock, flags); 2336 ntfs_debug("Status of mftbmp after initialized extention: " 2337 "allocated_size 0x%llx, data_size 0x%llx, " 2338 "initialized_size 0x%llx.", 2339 (long long)mftbmp_ni->allocated_size, 2340 (long long)i_size_read(vol->mftbmp_ino), 2341 (long long)mftbmp_ni->initialized_size); 2342 read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); 2343 #endif /* DEBUG */ 2344 ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit); 2345 found_free_rec: 2346 /* @bit is the found free mft record, allocate it in the mft bitmap. */ 2347 ntfs_debug("At found_free_rec."); 2348 err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit); 2349 if (unlikely(err)) { 2350 ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap."); 2351 up_write(&vol->mftbmp_lock); 2352 goto err_out; 2353 } 2354 ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit); 2355 have_alloc_rec: 2356 /* 2357 * The mft bitmap is now uptodate. Deal with mft data attribute now. 2358 * Note, we keep hold of the mft bitmap lock for writing until all 2359 * modifications to the mft data attribute are complete, too, as they 2360 * will impact decisions for mft bitmap and mft record allocation done 2361 * by a parallel allocation and if the lock is not maintained a 2362 * parallel allocation could allocate the same mft record as this one. 2363 */ 2364 ll = (bit + 1) << vol->mft_record_size_bits; 2365 read_lock_irqsave(&mft_ni->size_lock, flags); 2366 old_data_initialized = mft_ni->initialized_size; 2367 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2368 if (ll <= old_data_initialized) { 2369 ntfs_debug("Allocated mft record already initialized."); 2370 goto mft_rec_already_initialized; 2371 } 2372 ntfs_debug("Initializing allocated mft record."); 2373 /* 2374 * The mft record is outside the initialized data. Extend the mft data 2375 * attribute until it covers the allocated record. The loop is only 2376 * actually traversed more than once when a freshly formatted volume is 2377 * first written to so it optimizes away nicely in the common case. 2378 */ 2379 read_lock_irqsave(&mft_ni->size_lock, flags); 2380 ntfs_debug("Status of mft data before extension: " 2381 "allocated_size 0x%llx, data_size 0x%llx, " 2382 "initialized_size 0x%llx.", 2383 (long long)mft_ni->allocated_size, 2384 (long long)i_size_read(vol->mft_ino), 2385 (long long)mft_ni->initialized_size); 2386 while (ll > mft_ni->allocated_size) { 2387 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2388 err = ntfs_mft_data_extend_allocation_nolock(vol); 2389 if (unlikely(err)) { 2390 ntfs_error(vol->sb, "Failed to extend mft data " 2391 "allocation."); 2392 goto undo_mftbmp_alloc_nolock; 2393 } 2394 read_lock_irqsave(&mft_ni->size_lock, flags); 2395 ntfs_debug("Status of mft data after allocation extension: " 2396 "allocated_size 0x%llx, data_size 0x%llx, " 2397 "initialized_size 0x%llx.", 2398 (long long)mft_ni->allocated_size, 2399 (long long)i_size_read(vol->mft_ino), 2400 (long long)mft_ni->initialized_size); 2401 } 2402 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2403 /* 2404 * Extend mft data initialized size (and data size of course) to reach 2405 * the allocated mft record, formatting the mft records allong the way. 2406 * Note: We only modify the ntfs_inode structure as that is all that is 2407 * needed by ntfs_mft_record_format(). We will update the attribute 2408 * record itself in one fell swoop later on. 2409 */ 2410 write_lock_irqsave(&mft_ni->size_lock, flags); 2411 old_data_initialized = mft_ni->initialized_size; 2412 old_data_size = vol->mft_ino->i_size; 2413 while (ll > mft_ni->initialized_size) { 2414 s64 new_initialized_size, mft_no; 2415 2416 new_initialized_size = mft_ni->initialized_size + 2417 vol->mft_record_size; 2418 mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits; 2419 if (new_initialized_size > i_size_read(vol->mft_ino)) 2420 i_size_write(vol->mft_ino, new_initialized_size); 2421 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2422 ntfs_debug("Initializing mft record 0x%llx.", 2423 (long long)mft_no); 2424 err = ntfs_mft_record_format(vol, mft_no); 2425 if (unlikely(err)) { 2426 ntfs_error(vol->sb, "Failed to format mft record."); 2427 goto undo_data_init; 2428 } 2429 write_lock_irqsave(&mft_ni->size_lock, flags); 2430 mft_ni->initialized_size = new_initialized_size; 2431 } 2432 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2433 record_formatted = TRUE; 2434 /* Update the mft data attribute record to reflect the new sizes. */ 2435 m = map_mft_record(mft_ni); 2436 if (IS_ERR(m)) { 2437 ntfs_error(vol->sb, "Failed to map mft record."); 2438 err = PTR_ERR(m); 2439 goto undo_data_init; 2440 } 2441 ctx = ntfs_attr_get_search_ctx(mft_ni, m); 2442 if (unlikely(!ctx)) { 2443 ntfs_error(vol->sb, "Failed to get search context."); 2444 err = -ENOMEM; 2445 unmap_mft_record(mft_ni); 2446 goto undo_data_init; 2447 } 2448 err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, 2449 CASE_SENSITIVE, 0, NULL, 0, ctx); 2450 if (unlikely(err)) { 2451 ntfs_error(vol->sb, "Failed to find first attribute extent of " 2452 "mft data attribute."); 2453 ntfs_attr_put_search_ctx(ctx); 2454 unmap_mft_record(mft_ni); 2455 goto undo_data_init; 2456 } 2457 a = ctx->attr; 2458 read_lock_irqsave(&mft_ni->size_lock, flags); 2459 a->data.non_resident.initialized_size = 2460 cpu_to_sle64(mft_ni->initialized_size); 2461 a->data.non_resident.data_size = 2462 cpu_to_sle64(i_size_read(vol->mft_ino)); 2463 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2464 /* Ensure the changes make it to disk. */ 2465 flush_dcache_mft_record_page(ctx->ntfs_ino); 2466 mark_mft_record_dirty(ctx->ntfs_ino); 2467 ntfs_attr_put_search_ctx(ctx); 2468 unmap_mft_record(mft_ni); 2469 read_lock_irqsave(&mft_ni->size_lock, flags); 2470 ntfs_debug("Status of mft data after mft record initialization: " 2471 "allocated_size 0x%llx, data_size 0x%llx, " 2472 "initialized_size 0x%llx.", 2473 (long long)mft_ni->allocated_size, 2474 (long long)i_size_read(vol->mft_ino), 2475 (long long)mft_ni->initialized_size); 2476 BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size); 2477 BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino)); 2478 read_unlock_irqrestore(&mft_ni->size_lock, flags); 2479 mft_rec_already_initialized: 2480 /* 2481 * We can finally drop the mft bitmap lock as the mft data attribute 2482 * has been fully updated. The only disparity left is that the 2483 * allocated mft record still needs to be marked as in use to match the 2484 * set bit in the mft bitmap but this is actually not a problem since 2485 * this mft record is not referenced from anywhere yet and the fact 2486 * that it is allocated in the mft bitmap means that no-one will try to 2487 * allocate it either. 2488 */ 2489 up_write(&vol->mftbmp_lock); 2490 /* 2491 * We now have allocated and initialized the mft record. Calculate the 2492 * index of and the offset within the page cache page the record is in. 2493 */ 2494 index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; 2495 ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; 2496 /* Read, map, and pin the page containing the mft record. */ 2497 page = ntfs_map_page(vol->mft_ino->i_mapping, index); 2498 if (unlikely(IS_ERR(page))) { 2499 ntfs_error(vol->sb, "Failed to map page containing allocated " 2500 "mft record 0x%llx.", (long long)bit); 2501 err = PTR_ERR(page); 2502 goto undo_mftbmp_alloc; 2503 } 2504 lock_page(page); 2505 BUG_ON(!PageUptodate(page)); 2506 ClearPageUptodate(page); 2507 m = (MFT_RECORD*)((u8*)page_address(page) + ofs); 2508 /* If we just formatted the mft record no need to do it again. */ 2509 if (!record_formatted) { 2510 /* Sanity check that the mft record is really not in use. */ 2511 if (ntfs_is_file_record(m->magic) && 2512 (m->flags & MFT_RECORD_IN_USE)) { 2513 ntfs_error(vol->sb, "Mft record 0x%llx was marked " 2514 "free in mft bitmap but is marked " 2515 "used itself. Corrupt filesystem. " 2516 "Unmount and run chkdsk.", 2517 (long long)bit); 2518 err = -EIO; 2519 SetPageUptodate(page); 2520 unlock_page(page); 2521 ntfs_unmap_page(page); 2522 NVolSetErrors(vol); 2523 goto undo_mftbmp_alloc; 2524 } 2525 /* 2526 * We need to (re-)format the mft record, preserving the 2527 * sequence number if it is not zero as well as the update 2528 * sequence number if it is not zero or -1 (0xffff). This 2529 * means we do not need to care whether or not something went 2530 * wrong with the previous mft record. 2531 */ 2532 seq_no = m->sequence_number; 2533 usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)); 2534 err = ntfs_mft_record_layout(vol, bit, m); 2535 if (unlikely(err)) { 2536 ntfs_error(vol->sb, "Failed to layout allocated mft " 2537 "record 0x%llx.", (long long)bit); 2538 SetPageUptodate(page); 2539 unlock_page(page); 2540 ntfs_unmap_page(page); 2541 goto undo_mftbmp_alloc; 2542 } 2543 if (seq_no) 2544 m->sequence_number = seq_no; 2545 if (usn && le16_to_cpu(usn) != 0xffff) 2546 *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn; 2547 } 2548 /* Set the mft record itself in use. */ 2549 m->flags |= MFT_RECORD_IN_USE; 2550 if (S_ISDIR(mode)) 2551 m->flags |= MFT_RECORD_IS_DIRECTORY; 2552 flush_dcache_page(page); 2553 SetPageUptodate(page); 2554 if (base_ni) { 2555 /* 2556 * Setup the base mft record in the extent mft record. This 2557 * completes initialization of the allocated extent mft record 2558 * and we can simply use it with map_extent_mft_record(). 2559 */ 2560 m->base_mft_record = MK_LE_MREF(base_ni->mft_no, 2561 base_ni->seq_no); 2562 /* 2563 * Allocate an extent inode structure for the new mft record, 2564 * attach it to the base inode @base_ni and map, pin, and lock 2565 * its, i.e. the allocated, mft record. 2566 */ 2567 m = map_extent_mft_record(base_ni, bit, &ni); 2568 if (IS_ERR(m)) { 2569 ntfs_error(vol->sb, "Failed to map allocated extent " 2570 "mft record 0x%llx.", (long long)bit); 2571 err = PTR_ERR(m); 2572 /* Set the mft record itself not in use. */ 2573 m->flags &= cpu_to_le16( 2574 ~le16_to_cpu(MFT_RECORD_IN_USE)); 2575 flush_dcache_page(page); 2576 /* Make sure the mft record is written out to disk. */ 2577 mark_ntfs_record_dirty(page, ofs); 2578 unlock_page(page); 2579 ntfs_unmap_page(page); 2580 goto undo_mftbmp_alloc; 2581 } 2582 /* 2583 * Make sure the allocated mft record is written out to disk. 2584 * No need to set the inode dirty because the caller is going 2585 * to do that anyway after finishing with the new extent mft 2586 * record (e.g. at a minimum a new attribute will be added to 2587 * the mft record. 2588 */ 2589 mark_ntfs_record_dirty(page, ofs); 2590 unlock_page(page); 2591 /* 2592 * Need to unmap the page since map_extent_mft_record() mapped 2593 * it as well so we have it mapped twice at the moment. 2594 */ 2595 ntfs_unmap_page(page); 2596 } else { 2597 /* 2598 * Allocate a new VFS inode and set it up. NOTE: @vi->i_nlink 2599 * is set to 1 but the mft record->link_count is 0. The caller 2600 * needs to bear this in mind. 2601 */ 2602 vi = new_inode(vol->sb); 2603 if (unlikely(!vi)) { 2604 err = -ENOMEM; 2605 /* Set the mft record itself not in use. */ 2606 m->flags &= cpu_to_le16( 2607 ~le16_to_cpu(MFT_RECORD_IN_USE)); 2608 flush_dcache_page(page); 2609 /* Make sure the mft record is written out to disk. */ 2610 mark_ntfs_record_dirty(page, ofs); 2611 unlock_page(page); 2612 ntfs_unmap_page(page); 2613 goto undo_mftbmp_alloc; 2614 } 2615 vi->i_ino = bit; 2616 /* 2617 * This is the optimal IO size (for stat), not the fs block 2618 * size. 2619 */ 2620 vi->i_blksize = PAGE_CACHE_SIZE; 2621 /* 2622 * This is for checking whether an inode has changed w.r.t. a 2623 * file so that the file can be updated if necessary (compare 2624 * with f_version). 2625 */ 2626 vi->i_version = 1; 2627 2628 /* The owner and group come from the ntfs volume. */ 2629 vi->i_uid = vol->uid; 2630 vi->i_gid = vol->gid; 2631 2632 /* Initialize the ntfs specific part of @vi. */ 2633 ntfs_init_big_inode(vi); 2634 ni = NTFS_I(vi); 2635 /* 2636 * Set the appropriate mode, attribute type, and name. For 2637 * directories, also setup the index values to the defaults. 2638 */ 2639 if (S_ISDIR(mode)) { 2640 vi->i_mode = S_IFDIR | S_IRWXUGO; 2641 vi->i_mode &= ~vol->dmask; 2642 2643 NInoSetMstProtected(ni); 2644 ni->type = AT_INDEX_ALLOCATION; 2645 ni->name = I30; 2646 ni->name_len = 4; 2647 2648 ni->itype.index.block_size = 4096; 2649 ni->itype.index.block_size_bits = generic_ffs(4096) - 1; 2650 ni->itype.index.collation_rule = COLLATION_FILE_NAME; 2651 if (vol->cluster_size <= ni->itype.index.block_size) { 2652 ni->itype.index.vcn_size = vol->cluster_size; 2653 ni->itype.index.vcn_size_bits = 2654 vol->cluster_size_bits; 2655 } else { 2656 ni->itype.index.vcn_size = vol->sector_size; 2657 ni->itype.index.vcn_size_bits = 2658 vol->sector_size_bits; 2659 } 2660 } else { 2661 vi->i_mode = S_IFREG | S_IRWXUGO; 2662 vi->i_mode &= ~vol->fmask; 2663 2664 ni->type = AT_DATA; 2665 ni->name = NULL; 2666 ni->name_len = 0; 2667 } 2668 if (IS_RDONLY(vi)) 2669 vi->i_mode &= ~S_IWUGO; 2670 2671 /* Set the inode times to the current time. */ 2672 vi->i_atime = vi->i_mtime = vi->i_ctime = 2673 current_fs_time(vi->i_sb); 2674 /* 2675 * Set the file size to 0, the ntfs inode sizes are set to 0 by 2676 * the call to ntfs_init_big_inode() below. 2677 */ 2678 vi->i_size = 0; 2679 vi->i_blocks = 0; 2680 2681 /* Set the sequence number. */ 2682 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number); 2683 /* 2684 * Manually map, pin, and lock the mft record as we already 2685 * have its page mapped and it is very easy to do. 2686 */ 2687 atomic_inc(&ni->count); 2688 down(&ni->mrec_lock); 2689 ni->page = page; 2690 ni->page_ofs = ofs; 2691 /* 2692 * Make sure the allocated mft record is written out to disk. 2693 * NOTE: We do not set the ntfs inode dirty because this would 2694 * fail in ntfs_write_inode() because the inode does not have a 2695 * standard information attribute yet. Also, there is no need 2696 * to set the inode dirty because the caller is going to do 2697 * that anyway after finishing with the new mft record (e.g. at 2698 * a minimum some new attributes will be added to the mft 2699 * record. 2700 */ 2701 mark_ntfs_record_dirty(page, ofs); 2702 unlock_page(page); 2703 2704 /* Add the inode to the inode hash for the superblock. */ 2705 insert_inode_hash(vi); 2706 2707 /* Update the default mft allocation position. */ 2708 vol->mft_data_pos = bit + 1; 2709 } 2710 /* 2711 * Return the opened, allocated inode of the allocated mft record as 2712 * well as the mapped, pinned, and locked mft record. 2713 */ 2714 ntfs_debug("Returning opened, allocated %sinode 0x%llx.", 2715 base_ni ? "extent " : "", (long long)bit); 2716 *mrec = m; 2717 return ni; 2718 undo_data_init: 2719 write_lock_irqsave(&mft_ni->size_lock, flags); 2720 mft_ni->initialized_size = old_data_initialized; 2721 i_size_write(vol->mft_ino, old_data_size); 2722 write_unlock_irqrestore(&mft_ni->size_lock, flags); 2723 goto undo_mftbmp_alloc_nolock; 2724 undo_mftbmp_alloc: 2725 down_write(&vol->mftbmp_lock); 2726 undo_mftbmp_alloc_nolock: 2727 if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) { 2728 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); 2729 NVolSetErrors(vol); 2730 } 2731 up_write(&vol->mftbmp_lock); 2732 err_out: 2733 return ERR_PTR(err); 2734 max_err_out: 2735 ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum " 2736 "number of inodes (2^32) has already been reached."); 2737 up_write(&vol->mftbmp_lock); 2738 return ERR_PTR(-ENOSPC); 2739 } 2740 2741 /** 2742 * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume 2743 * @ni: ntfs inode of the mapped extent mft record to free 2744 * @m: mapped extent mft record of the ntfs inode @ni 2745 * 2746 * Free the mapped extent mft record @m of the extent ntfs inode @ni. 2747 * 2748 * Note that this function unmaps the mft record and closes and destroys @ni 2749 * internally and hence you cannot use either @ni nor @m any more after this 2750 * function returns success. 2751 * 2752 * On success return 0 and on error return -errno. @ni and @m are still valid 2753 * in this case and have not been freed. 2754 * 2755 * For some errors an error message is displayed and the success code 0 is 2756 * returned and the volume is then left dirty on umount. This makes sense in 2757 * case we could not rollback the changes that were already done since the 2758 * caller no longer wants to reference this mft record so it does not matter to 2759 * the caller if something is wrong with it as long as it is properly detached 2760 * from the base inode. 2761 */ 2762 int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m) 2763 { 2764 unsigned long mft_no = ni->mft_no; 2765 ntfs_volume *vol = ni->vol; 2766 ntfs_inode *base_ni; 2767 ntfs_inode **extent_nis; 2768 int i, err; 2769 le16 old_seq_no; 2770 u16 seq_no; 2771 2772 BUG_ON(NInoAttr(ni)); 2773 BUG_ON(ni->nr_extents != -1); 2774 2775 down(&ni->extent_lock); 2776 base_ni = ni->ext.base_ntfs_ino; 2777 up(&ni->extent_lock); 2778 2779 BUG_ON(base_ni->nr_extents <= 0); 2780 2781 ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n", 2782 mft_no, base_ni->mft_no); 2783 2784 down(&base_ni->extent_lock); 2785 2786 /* Make sure we are holding the only reference to the extent inode. */ 2787 if (atomic_read(&ni->count) > 2) { 2788 ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, " 2789 "not freeing.", base_ni->mft_no); 2790 up(&base_ni->extent_lock); 2791 return -EBUSY; 2792 } 2793 2794 /* Dissociate the ntfs inode from the base inode. */ 2795 extent_nis = base_ni->ext.extent_ntfs_inos; 2796 err = -ENOENT; 2797 for (i = 0; i < base_ni->nr_extents; i++) { 2798 if (ni != extent_nis[i]) 2799 continue; 2800 extent_nis += i; 2801 base_ni->nr_extents--; 2802 memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) * 2803 sizeof(ntfs_inode*)); 2804 err = 0; 2805 break; 2806 } 2807 2808 up(&base_ni->extent_lock); 2809 2810 if (unlikely(err)) { 2811 ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to " 2812 "its base inode 0x%lx.", mft_no, 2813 base_ni->mft_no); 2814 BUG(); 2815 } 2816 2817 /* 2818 * The extent inode is no longer attached to the base inode so no one 2819 * can get a reference to it any more. 2820 */ 2821 2822 /* Mark the mft record as not in use. */ 2823 m->flags &= const_cpu_to_le16(~const_le16_to_cpu(MFT_RECORD_IN_USE)); 2824 2825 /* Increment the sequence number, skipping zero, if it is not zero. */ 2826 old_seq_no = m->sequence_number; 2827 seq_no = le16_to_cpu(old_seq_no); 2828 if (seq_no == 0xffff) 2829 seq_no = 1; 2830 else if (seq_no) 2831 seq_no++; 2832 m->sequence_number = cpu_to_le16(seq_no); 2833 2834 /* 2835 * Set the ntfs inode dirty and write it out. We do not need to worry 2836 * about the base inode here since whatever caused the extent mft 2837 * record to be freed is guaranteed to do it already. 2838 */ 2839 NInoSetDirty(ni); 2840 err = write_mft_record(ni, m, 0); 2841 if (unlikely(err)) { 2842 ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not " 2843 "freeing.", mft_no); 2844 goto rollback; 2845 } 2846 rollback_error: 2847 /* Unmap and throw away the now freed extent inode. */ 2848 unmap_extent_mft_record(ni); 2849 ntfs_clear_extent_inode(ni); 2850 2851 /* Clear the bit in the $MFT/$BITMAP corresponding to this record. */ 2852 down_write(&vol->mftbmp_lock); 2853 err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no); 2854 up_write(&vol->mftbmp_lock); 2855 if (unlikely(err)) { 2856 /* 2857 * The extent inode is gone but we failed to deallocate it in 2858 * the mft bitmap. Just emit a warning and leave the volume 2859 * dirty on umount. 2860 */ 2861 ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); 2862 NVolSetErrors(vol); 2863 } 2864 return 0; 2865 rollback: 2866 /* Rollback what we did... */ 2867 down(&base_ni->extent_lock); 2868 extent_nis = base_ni->ext.extent_ntfs_inos; 2869 if (!(base_ni->nr_extents & 3)) { 2870 int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*); 2871 2872 extent_nis = (ntfs_inode**)kmalloc(new_size, GFP_NOFS); 2873 if (unlikely(!extent_nis)) { 2874 ntfs_error(vol->sb, "Failed to allocate internal " 2875 "buffer during rollback.%s", es); 2876 up(&base_ni->extent_lock); 2877 NVolSetErrors(vol); 2878 goto rollback_error; 2879 } 2880 if (base_ni->nr_extents) { 2881 BUG_ON(!base_ni->ext.extent_ntfs_inos); 2882 memcpy(extent_nis, base_ni->ext.extent_ntfs_inos, 2883 new_size - 4 * sizeof(ntfs_inode*)); 2884 kfree(base_ni->ext.extent_ntfs_inos); 2885 } 2886 base_ni->ext.extent_ntfs_inos = extent_nis; 2887 } 2888 m->flags |= MFT_RECORD_IN_USE; 2889 m->sequence_number = old_seq_no; 2890 extent_nis[base_ni->nr_extents++] = ni; 2891 up(&base_ni->extent_lock); 2892 mark_mft_record_dirty(ni); 2893 return err; 2894 } 2895 #endif /* NTFS_RW */ 2896