Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /** |
| 2 | * mft.c - NTFS kernel mft record operations. Part of the Linux-NTFS project. |
| 3 | * |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 4 | * Copyright (c) 2001-2006 Anton Altaparmakov |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | { |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 48 | loff_t i_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 49 | ntfs_volume *vol = ni->vol; |
| 50 | struct inode *mft_vi = vol->mft_ino; |
| 51 | struct page *page; |
Anton Altaparmakov | 69b41e3 | 2005-10-04 14:01:14 +0100 | [diff] [blame] | 52 | unsigned long index, end_index; |
| 53 | unsigned ofs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
| 55 | BUG_ON(ni->page); |
| 56 | /* |
| 57 | * The index into the page cache and the offset within the page cache |
| 58 | * page of the wanted mft record. FIXME: We need to check for |
| 59 | * overflowing the unsigned long, but I don't think we would ever get |
| 60 | * here if the volume was that big... |
| 61 | */ |
Anton Altaparmakov | c394e45 | 2005-10-04 13:08:53 +0100 | [diff] [blame] | 62 | index = (u64)ni->mft_no << vol->mft_record_size_bits >> |
| 63 | PAGE_CACHE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | ofs = (ni->mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; |
| 65 | |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 66 | i_size = i_size_read(mft_vi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 67 | /* The maximum valid index into the page cache for $MFT's data. */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 68 | end_index = i_size >> PAGE_CACHE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | |
| 70 | /* If the wanted index is out of bounds the mft record doesn't exist. */ |
| 71 | if (unlikely(index >= end_index)) { |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 72 | if (index > end_index || (i_size & ~PAGE_CACHE_MASK) < ofs + |
| 73 | vol->mft_record_size) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | page = ERR_PTR(-ENOENT); |
| 75 | ntfs_error(vol->sb, "Attemt to read mft record 0x%lx, " |
| 76 | "which is beyond the end of the mft. " |
| 77 | "This is probably a bug in the ntfs " |
| 78 | "driver.", ni->mft_no); |
| 79 | goto err_out; |
| 80 | } |
| 81 | } |
| 82 | /* Read, map, and pin the page. */ |
| 83 | page = ntfs_map_page(mft_vi->i_mapping, index); |
| 84 | if (likely(!IS_ERR(page))) { |
| 85 | /* Catch multi sector transfer fixup errors. */ |
| 86 | if (likely(ntfs_is_mft_recordp((le32*)(page_address(page) + |
| 87 | ofs)))) { |
| 88 | ni->page = page; |
| 89 | ni->page_ofs = ofs; |
| 90 | return page_address(page) + ofs; |
| 91 | } |
| 92 | ntfs_error(vol->sb, "Mft record 0x%lx is corrupt. " |
| 93 | "Run chkdsk.", ni->mft_no); |
| 94 | ntfs_unmap_page(page); |
| 95 | page = ERR_PTR(-EIO); |
Anton Altaparmakov | f95c401 | 2006-03-23 15:59:32 +0000 | [diff] [blame^] | 96 | NVolSetErrors(vol); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | } |
| 98 | err_out: |
| 99 | ni->page = NULL; |
| 100 | ni->page_ofs = 0; |
| 101 | return (void*)page; |
| 102 | } |
| 103 | |
| 104 | /** |
| 105 | * map_mft_record - map, pin and lock an mft record |
| 106 | * @ni: ntfs inode whose MFT record to map |
| 107 | * |
| 108 | * First, take the mrec_lock semaphore. We might now be sleeping, while waiting |
| 109 | * for the semaphore if it was already locked by someone else. |
| 110 | * |
| 111 | * The page of the record is mapped using map_mft_record_page() before being |
| 112 | * returned to the caller. |
| 113 | * |
| 114 | * This in turn uses ntfs_map_page() to get the page containing the wanted mft |
| 115 | * record (it in turn calls read_cache_page() which reads it in from disk if |
| 116 | * necessary, increments the use count on the page so that it cannot disappear |
| 117 | * under us and returns a reference to the page cache page). |
| 118 | * |
| 119 | * If read_cache_page() invokes ntfs_readpage() to load the page from disk, it |
| 120 | * sets PG_locked and clears PG_uptodate on the page. Once I/O has completed |
| 121 | * and the post-read mst fixups on each mft record in the page have been |
| 122 | * performed, the page gets PG_uptodate set and PG_locked cleared (this is done |
| 123 | * in our asynchronous I/O completion handler end_buffer_read_mft_async()). |
| 124 | * ntfs_map_page() waits for PG_locked to become clear and checks if |
| 125 | * PG_uptodate is set and returns an error code if not. This provides |
| 126 | * sufficient protection against races when reading/using the page. |
| 127 | * |
| 128 | * However there is the write mapping to think about. Doing the above described |
| 129 | * checking here will be fine, because when initiating the write we will set |
| 130 | * PG_locked and clear PG_uptodate making sure nobody is touching the page |
| 131 | * contents. Doing the locking this way means that the commit to disk code in |
| 132 | * the page cache code paths is automatically sufficiently locked with us as |
| 133 | * we will not touch a page that has been locked or is not uptodate. The only |
| 134 | * locking problem then is them locking the page while we are accessing it. |
| 135 | * |
| 136 | * So that code will end up having to own the mrec_lock of all mft |
| 137 | * records/inodes present in the page before I/O can proceed. In that case we |
| 138 | * wouldn't need to bother with PG_locked and PG_uptodate as nobody will be |
| 139 | * accessing anything without owning the mrec_lock semaphore. But we do need |
| 140 | * to use them because of the read_cache_page() invocation and the code becomes |
| 141 | * so much simpler this way that it is well worth it. |
| 142 | * |
| 143 | * The mft record is now ours and we return a pointer to it. You need to check |
| 144 | * the returned pointer with IS_ERR() and if that is true, PTR_ERR() will return |
| 145 | * the error code. |
| 146 | * |
| 147 | * NOTE: Caller is responsible for setting the mft record dirty before calling |
| 148 | * unmap_mft_record(). This is obviously only necessary if the caller really |
| 149 | * modified the mft record... |
| 150 | * Q: Do we want to recycle one of the VFS inode state bits instead? |
| 151 | * A: No, the inode ones mean we want to change the mft record, not we want to |
| 152 | * write it out. |
| 153 | */ |
| 154 | MFT_RECORD *map_mft_record(ntfs_inode *ni) |
| 155 | { |
| 156 | MFT_RECORD *m; |
| 157 | |
| 158 | ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no); |
| 159 | |
| 160 | /* Make sure the ntfs inode doesn't go away. */ |
| 161 | atomic_inc(&ni->count); |
| 162 | |
| 163 | /* Serialize access to this mft record. */ |
| 164 | down(&ni->mrec_lock); |
| 165 | |
| 166 | m = map_mft_record_page(ni); |
| 167 | if (likely(!IS_ERR(m))) |
| 168 | return m; |
| 169 | |
| 170 | up(&ni->mrec_lock); |
| 171 | atomic_dec(&ni->count); |
| 172 | ntfs_error(ni->vol->sb, "Failed with error code %lu.", -PTR_ERR(m)); |
| 173 | return m; |
| 174 | } |
| 175 | |
| 176 | /** |
| 177 | * unmap_mft_record_page - unmap the page in which a specific mft record resides |
| 178 | * @ni: ntfs inode whose mft record page to unmap |
| 179 | * |
| 180 | * This unmaps the page in which the mft record of the ntfs inode @ni is |
| 181 | * situated and returns. This is a NOOP if highmem is not configured. |
| 182 | * |
| 183 | * The unmap happens via ntfs_unmap_page() which in turn decrements the use |
| 184 | * count on the page thus releasing it from the pinned state. |
| 185 | * |
| 186 | * We do not actually unmap the page from memory of course, as that will be |
| 187 | * done by the page cache code itself when memory pressure increases or |
| 188 | * whatever. |
| 189 | */ |
| 190 | static inline void unmap_mft_record_page(ntfs_inode *ni) |
| 191 | { |
| 192 | BUG_ON(!ni->page); |
| 193 | |
| 194 | // TODO: If dirty, blah... |
| 195 | ntfs_unmap_page(ni->page); |
| 196 | ni->page = NULL; |
| 197 | ni->page_ofs = 0; |
| 198 | return; |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * unmap_mft_record - release a mapped mft record |
| 203 | * @ni: ntfs inode whose MFT record to unmap |
| 204 | * |
| 205 | * We release the page mapping and the mrec_lock mutex which unmaps the mft |
| 206 | * record and releases it for others to get hold of. We also release the ntfs |
| 207 | * inode by decrementing the ntfs inode reference count. |
| 208 | * |
| 209 | * NOTE: If caller has modified the mft record, it is imperative to set the mft |
| 210 | * record dirty BEFORE calling unmap_mft_record(). |
| 211 | */ |
| 212 | void unmap_mft_record(ntfs_inode *ni) |
| 213 | { |
| 214 | struct page *page = ni->page; |
| 215 | |
| 216 | BUG_ON(!page); |
| 217 | |
| 218 | ntfs_debug("Entering for mft_no 0x%lx.", ni->mft_no); |
| 219 | |
| 220 | unmap_mft_record_page(ni); |
| 221 | up(&ni->mrec_lock); |
| 222 | atomic_dec(&ni->count); |
| 223 | /* |
| 224 | * If pure ntfs_inode, i.e. no vfs inode attached, we leave it to |
| 225 | * ntfs_clear_extent_inode() in the extent inode case, and to the |
| 226 | * caller in the non-extent, yet pure ntfs inode case, to do the actual |
| 227 | * tear down of all structures and freeing of all allocated memory. |
| 228 | */ |
| 229 | return; |
| 230 | } |
| 231 | |
| 232 | /** |
| 233 | * map_extent_mft_record - load an extent inode and attach it to its base |
| 234 | * @base_ni: base ntfs inode |
| 235 | * @mref: mft reference of the extent inode to load |
| 236 | * @ntfs_ino: on successful return, pointer to the ntfs_inode structure |
| 237 | * |
| 238 | * Load the extent mft record @mref and attach it to its base inode @base_ni. |
| 239 | * Return the mapped extent mft record if IS_ERR(result) is false. Otherwise |
| 240 | * PTR_ERR(result) gives the negative error code. |
| 241 | * |
| 242 | * On successful return, @ntfs_ino contains a pointer to the ntfs_inode |
| 243 | * structure of the mapped extent inode. |
| 244 | */ |
| 245 | MFT_RECORD *map_extent_mft_record(ntfs_inode *base_ni, MFT_REF mref, |
| 246 | ntfs_inode **ntfs_ino) |
| 247 | { |
| 248 | MFT_RECORD *m; |
| 249 | ntfs_inode *ni = NULL; |
| 250 | ntfs_inode **extent_nis = NULL; |
| 251 | int i; |
| 252 | unsigned long mft_no = MREF(mref); |
| 253 | u16 seq_no = MSEQNO(mref); |
| 254 | BOOL destroy_ni = FALSE; |
| 255 | |
| 256 | ntfs_debug("Mapping extent mft record 0x%lx (base mft record 0x%lx).", |
| 257 | mft_no, base_ni->mft_no); |
| 258 | /* Make sure the base ntfs inode doesn't go away. */ |
| 259 | atomic_inc(&base_ni->count); |
| 260 | /* |
| 261 | * Check if this extent inode has already been added to the base inode, |
| 262 | * in which case just return it. If not found, add it to the base |
| 263 | * inode before returning it. |
| 264 | */ |
| 265 | down(&base_ni->extent_lock); |
| 266 | if (base_ni->nr_extents > 0) { |
| 267 | extent_nis = base_ni->ext.extent_ntfs_inos; |
| 268 | for (i = 0; i < base_ni->nr_extents; i++) { |
| 269 | if (mft_no != extent_nis[i]->mft_no) |
| 270 | continue; |
| 271 | ni = extent_nis[i]; |
| 272 | /* Make sure the ntfs inode doesn't go away. */ |
| 273 | atomic_inc(&ni->count); |
| 274 | break; |
| 275 | } |
| 276 | } |
| 277 | if (likely(ni != NULL)) { |
| 278 | up(&base_ni->extent_lock); |
| 279 | atomic_dec(&base_ni->count); |
| 280 | /* We found the record; just have to map and return it. */ |
| 281 | m = map_mft_record(ni); |
| 282 | /* map_mft_record() has incremented this on success. */ |
| 283 | atomic_dec(&ni->count); |
| 284 | if (likely(!IS_ERR(m))) { |
| 285 | /* Verify the sequence number. */ |
| 286 | if (likely(le16_to_cpu(m->sequence_number) == seq_no)) { |
| 287 | ntfs_debug("Done 1."); |
| 288 | *ntfs_ino = ni; |
| 289 | return m; |
| 290 | } |
| 291 | unmap_mft_record(ni); |
| 292 | ntfs_error(base_ni->vol->sb, "Found stale extent mft " |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 293 | "reference! Corrupt filesystem. " |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 294 | "Run chkdsk."); |
| 295 | return ERR_PTR(-EIO); |
| 296 | } |
| 297 | map_err_out: |
| 298 | ntfs_error(base_ni->vol->sb, "Failed to map extent " |
| 299 | "mft record, error code %ld.", -PTR_ERR(m)); |
| 300 | return m; |
| 301 | } |
| 302 | /* Record wasn't there. Get a new ntfs inode and initialize it. */ |
| 303 | ni = ntfs_new_extent_inode(base_ni->vol->sb, mft_no); |
| 304 | if (unlikely(!ni)) { |
| 305 | up(&base_ni->extent_lock); |
| 306 | atomic_dec(&base_ni->count); |
| 307 | return ERR_PTR(-ENOMEM); |
| 308 | } |
| 309 | ni->vol = base_ni->vol; |
| 310 | ni->seq_no = seq_no; |
| 311 | ni->nr_extents = -1; |
| 312 | ni->ext.base_ntfs_ino = base_ni; |
| 313 | /* Now map the record. */ |
| 314 | m = map_mft_record(ni); |
| 315 | if (IS_ERR(m)) { |
| 316 | up(&base_ni->extent_lock); |
| 317 | atomic_dec(&base_ni->count); |
| 318 | ntfs_clear_extent_inode(ni); |
| 319 | goto map_err_out; |
| 320 | } |
| 321 | /* Verify the sequence number if it is present. */ |
| 322 | if (seq_no && (le16_to_cpu(m->sequence_number) != seq_no)) { |
| 323 | ntfs_error(base_ni->vol->sb, "Found stale extent mft " |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 324 | "reference! Corrupt filesystem. Run chkdsk."); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 325 | destroy_ni = TRUE; |
| 326 | m = ERR_PTR(-EIO); |
| 327 | goto unm_err_out; |
| 328 | } |
| 329 | /* Attach extent inode to base inode, reallocating memory if needed. */ |
| 330 | if (!(base_ni->nr_extents & 3)) { |
| 331 | ntfs_inode **tmp; |
| 332 | int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode *); |
| 333 | |
| 334 | tmp = (ntfs_inode **)kmalloc(new_size, GFP_NOFS); |
| 335 | if (unlikely(!tmp)) { |
| 336 | ntfs_error(base_ni->vol->sb, "Failed to allocate " |
| 337 | "internal buffer."); |
| 338 | destroy_ni = TRUE; |
| 339 | m = ERR_PTR(-ENOMEM); |
| 340 | goto unm_err_out; |
| 341 | } |
| 342 | if (base_ni->nr_extents) { |
| 343 | BUG_ON(!base_ni->ext.extent_ntfs_inos); |
| 344 | memcpy(tmp, base_ni->ext.extent_ntfs_inos, new_size - |
| 345 | 4 * sizeof(ntfs_inode *)); |
| 346 | kfree(base_ni->ext.extent_ntfs_inos); |
| 347 | } |
| 348 | base_ni->ext.extent_ntfs_inos = tmp; |
| 349 | } |
| 350 | base_ni->ext.extent_ntfs_inos[base_ni->nr_extents++] = ni; |
| 351 | up(&base_ni->extent_lock); |
| 352 | atomic_dec(&base_ni->count); |
| 353 | ntfs_debug("Done 2."); |
| 354 | *ntfs_ino = ni; |
| 355 | return m; |
| 356 | unm_err_out: |
| 357 | unmap_mft_record(ni); |
| 358 | up(&base_ni->extent_lock); |
| 359 | atomic_dec(&base_ni->count); |
| 360 | /* |
| 361 | * If the extent inode was not attached to the base inode we need to |
| 362 | * release it or we will leak memory. |
| 363 | */ |
| 364 | if (destroy_ni) |
| 365 | ntfs_clear_extent_inode(ni); |
| 366 | return m; |
| 367 | } |
| 368 | |
| 369 | #ifdef NTFS_RW |
| 370 | |
| 371 | /** |
| 372 | * __mark_mft_record_dirty - set the mft record and the page containing it dirty |
| 373 | * @ni: ntfs inode describing the mapped mft record |
| 374 | * |
| 375 | * Internal function. Users should call mark_mft_record_dirty() instead. |
| 376 | * |
| 377 | * Set the mapped (extent) mft record of the (base or extent) ntfs inode @ni, |
| 378 | * as well as the page containing the mft record, dirty. Also, mark the base |
| 379 | * vfs inode dirty. This ensures that any changes to the mft record are |
| 380 | * written out to disk. |
| 381 | * |
| 382 | * NOTE: We only set I_DIRTY_SYNC and I_DIRTY_DATASYNC (and not I_DIRTY_PAGES) |
| 383 | * on the base vfs inode, because even though file data may have been modified, |
| 384 | * it is dirty in the inode meta data rather than the data page cache of the |
| 385 | * inode, and thus there are no data pages that need writing out. Therefore, a |
| 386 | * full mark_inode_dirty() is overkill. A mark_inode_dirty_sync(), on the |
| 387 | * other hand, is not sufficient, because I_DIRTY_DATASYNC needs to be set to |
| 388 | * ensure ->write_inode is called from generic_osync_inode() and this needs to |
| 389 | * happen or the file data would not necessarily hit the device synchronously, |
| 390 | * even though the vfs inode has the O_SYNC flag set. Also, I_DIRTY_DATASYNC |
| 391 | * simply "feels" better than just I_DIRTY_SYNC, since the file data has not |
| 392 | * actually hit the block device yet, which is not what I_DIRTY_SYNC on its own |
| 393 | * would suggest. |
| 394 | */ |
| 395 | void __mark_mft_record_dirty(ntfs_inode *ni) |
| 396 | { |
| 397 | ntfs_inode *base_ni; |
| 398 | |
| 399 | ntfs_debug("Entering for inode 0x%lx.", ni->mft_no); |
| 400 | BUG_ON(NInoAttr(ni)); |
| 401 | mark_ntfs_record_dirty(ni->page, ni->page_ofs); |
| 402 | /* Determine the base vfs inode and mark it dirty, too. */ |
| 403 | down(&ni->extent_lock); |
| 404 | if (likely(ni->nr_extents >= 0)) |
| 405 | base_ni = ni; |
| 406 | else |
| 407 | base_ni = ni->ext.base_ntfs_ino; |
| 408 | up(&ni->extent_lock); |
| 409 | __mark_inode_dirty(VFS_I(base_ni), I_DIRTY_SYNC | I_DIRTY_DATASYNC); |
| 410 | } |
| 411 | |
| 412 | static const char *ntfs_please_email = "Please email " |
| 413 | "linux-ntfs-dev@lists.sourceforge.net and say that you saw " |
| 414 | "this message. Thank you."; |
| 415 | |
| 416 | /** |
| 417 | * ntfs_sync_mft_mirror_umount - synchronise an mft record to the mft mirror |
| 418 | * @vol: ntfs volume on which the mft record to synchronize resides |
| 419 | * @mft_no: mft record number of mft record to synchronize |
| 420 | * @m: mapped, mst protected (extent) mft record to synchronize |
| 421 | * |
| 422 | * Write the mapped, mst protected (extent) mft record @m with mft record |
| 423 | * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol, |
| 424 | * bypassing the page cache and the $MFTMirr inode itself. |
| 425 | * |
| 426 | * This function is only for use at umount time when the mft mirror inode has |
| 427 | * already been disposed off. We BUG() if we are called while the mft mirror |
| 428 | * inode is still attached to the volume. |
| 429 | * |
| 430 | * On success return 0. On error return -errno. |
| 431 | * |
| 432 | * NOTE: This function is not implemented yet as I am not convinced it can |
| 433 | * actually be triggered considering the sequence of commits we do in super.c:: |
| 434 | * ntfs_put_super(). But just in case we provide this place holder as the |
| 435 | * alternative would be either to BUG() or to get a NULL pointer dereference |
| 436 | * and Oops. |
| 437 | */ |
| 438 | static int ntfs_sync_mft_mirror_umount(ntfs_volume *vol, |
| 439 | const unsigned long mft_no, MFT_RECORD *m) |
| 440 | { |
| 441 | BUG_ON(vol->mftmirr_ino); |
| 442 | ntfs_error(vol->sb, "Umount time mft mirror syncing is not " |
| 443 | "implemented yet. %s", ntfs_please_email); |
| 444 | return -EOPNOTSUPP; |
| 445 | } |
| 446 | |
| 447 | /** |
| 448 | * ntfs_sync_mft_mirror - synchronize an mft record to the mft mirror |
| 449 | * @vol: ntfs volume on which the mft record to synchronize resides |
| 450 | * @mft_no: mft record number of mft record to synchronize |
| 451 | * @m: mapped, mst protected (extent) mft record to synchronize |
| 452 | * @sync: if true, wait for i/o completion |
| 453 | * |
| 454 | * Write the mapped, mst protected (extent) mft record @m with mft record |
| 455 | * number @mft_no to the mft mirror ($MFTMirr) of the ntfs volume @vol. |
| 456 | * |
| 457 | * On success return 0. On error return -errno and set the volume errors flag |
| 458 | * in the ntfs volume @vol. |
| 459 | * |
| 460 | * NOTE: We always perform synchronous i/o and ignore the @sync parameter. |
| 461 | * |
| 462 | * TODO: If @sync is false, want to do truly asynchronous i/o, i.e. just |
| 463 | * schedule i/o via ->writepage or do it via kntfsd or whatever. |
| 464 | */ |
| 465 | int ntfs_sync_mft_mirror(ntfs_volume *vol, const unsigned long mft_no, |
| 466 | MFT_RECORD *m, int sync) |
| 467 | { |
| 468 | struct page *page; |
| 469 | unsigned int blocksize = vol->sb->s_blocksize; |
| 470 | int max_bhs = vol->mft_record_size / blocksize; |
| 471 | struct buffer_head *bhs[max_bhs]; |
| 472 | struct buffer_head *bh, *head; |
| 473 | u8 *kmirr; |
| 474 | runlist_element *rl; |
| 475 | unsigned int block_start, block_end, m_start, m_end, page_ofs; |
| 476 | int i_bhs, nr_bhs, err = 0; |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 477 | unsigned char blocksize_bits = vol->sb->s_blocksize_bits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | |
| 479 | ntfs_debug("Entering for inode 0x%lx.", mft_no); |
| 480 | BUG_ON(!max_bhs); |
| 481 | if (unlikely(!vol->mftmirr_ino)) { |
| 482 | /* This could happen during umount... */ |
| 483 | err = ntfs_sync_mft_mirror_umount(vol, mft_no, m); |
| 484 | if (likely(!err)) |
| 485 | return err; |
| 486 | goto err_out; |
| 487 | } |
| 488 | /* Get the page containing the mirror copy of the mft record @m. */ |
| 489 | page = ntfs_map_page(vol->mftmirr_ino->i_mapping, mft_no >> |
| 490 | (PAGE_CACHE_SHIFT - vol->mft_record_size_bits)); |
| 491 | if (IS_ERR(page)) { |
| 492 | ntfs_error(vol->sb, "Failed to map mft mirror page."); |
| 493 | err = PTR_ERR(page); |
| 494 | goto err_out; |
| 495 | } |
| 496 | lock_page(page); |
| 497 | BUG_ON(!PageUptodate(page)); |
| 498 | ClearPageUptodate(page); |
| 499 | /* Offset of the mft mirror record inside the page. */ |
| 500 | page_ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; |
| 501 | /* The address in the page of the mirror copy of the mft record @m. */ |
| 502 | kmirr = page_address(page) + page_ofs; |
| 503 | /* Copy the mst protected mft record to the mirror. */ |
| 504 | memcpy(kmirr, m, vol->mft_record_size); |
| 505 | /* Create uptodate buffers if not present. */ |
| 506 | if (unlikely(!page_has_buffers(page))) { |
| 507 | struct buffer_head *tail; |
| 508 | |
| 509 | bh = head = alloc_page_buffers(page, blocksize, 1); |
| 510 | do { |
| 511 | set_buffer_uptodate(bh); |
| 512 | tail = bh; |
| 513 | bh = bh->b_this_page; |
| 514 | } while (bh); |
| 515 | tail->b_this_page = head; |
| 516 | attach_page_buffers(page, head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | } |
| 518 | bh = head = page_buffers(page); |
| 519 | BUG_ON(!bh); |
| 520 | rl = NULL; |
| 521 | nr_bhs = 0; |
| 522 | block_start = 0; |
| 523 | m_start = kmirr - (u8*)page_address(page); |
| 524 | m_end = m_start + vol->mft_record_size; |
| 525 | do { |
| 526 | block_end = block_start + blocksize; |
| 527 | /* If the buffer is outside the mft record, skip it. */ |
| 528 | if (block_end <= m_start) |
| 529 | continue; |
| 530 | if (unlikely(block_start >= m_end)) |
| 531 | break; |
| 532 | /* Need to map the buffer if it is not mapped already. */ |
| 533 | if (unlikely(!buffer_mapped(bh))) { |
| 534 | VCN vcn; |
| 535 | LCN lcn; |
| 536 | unsigned int vcn_ofs; |
| 537 | |
Anton Altaparmakov | e74589a | 2005-08-16 16:38:28 +0100 | [diff] [blame] | 538 | bh->b_bdev = vol->sb->s_bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | /* Obtain the vcn and offset of the current block. */ |
| 540 | vcn = ((VCN)mft_no << vol->mft_record_size_bits) + |
| 541 | (block_start - m_start); |
| 542 | vcn_ofs = vcn & vol->cluster_size_mask; |
| 543 | vcn >>= vol->cluster_size_bits; |
| 544 | if (!rl) { |
| 545 | down_read(&NTFS_I(vol->mftmirr_ino)-> |
| 546 | runlist.lock); |
| 547 | rl = NTFS_I(vol->mftmirr_ino)->runlist.rl; |
| 548 | /* |
| 549 | * $MFTMirr always has the whole of its runlist |
| 550 | * in memory. |
| 551 | */ |
| 552 | BUG_ON(!rl); |
| 553 | } |
| 554 | /* Seek to element containing target vcn. */ |
| 555 | while (rl->length && rl[1].vcn <= vcn) |
| 556 | rl++; |
| 557 | lcn = ntfs_rl_vcn_to_lcn(rl, vcn); |
| 558 | /* For $MFTMirr, only lcn >= 0 is a successful remap. */ |
| 559 | if (likely(lcn >= 0)) { |
| 560 | /* Setup buffer head to correct block. */ |
| 561 | bh->b_blocknr = ((lcn << |
| 562 | vol->cluster_size_bits) + |
| 563 | vcn_ofs) >> blocksize_bits; |
| 564 | set_buffer_mapped(bh); |
| 565 | } else { |
| 566 | bh->b_blocknr = -1; |
| 567 | ntfs_error(vol->sb, "Cannot write mft mirror " |
| 568 | "record 0x%lx because its " |
| 569 | "location on disk could not " |
| 570 | "be determined (error code " |
| 571 | "%lli).", mft_no, |
| 572 | (long long)lcn); |
| 573 | err = -EIO; |
| 574 | } |
| 575 | } |
| 576 | BUG_ON(!buffer_uptodate(bh)); |
| 577 | BUG_ON(!nr_bhs && (m_start != block_start)); |
| 578 | BUG_ON(nr_bhs >= max_bhs); |
| 579 | bhs[nr_bhs++] = bh; |
| 580 | BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end)); |
| 581 | } while (block_start = block_end, (bh = bh->b_this_page) != head); |
| 582 | if (unlikely(rl)) |
| 583 | up_read(&NTFS_I(vol->mftmirr_ino)->runlist.lock); |
| 584 | if (likely(!err)) { |
| 585 | /* Lock buffers and start synchronous write i/o on them. */ |
| 586 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { |
| 587 | struct buffer_head *tbh = bhs[i_bhs]; |
| 588 | |
| 589 | if (unlikely(test_set_buffer_locked(tbh))) |
| 590 | BUG(); |
| 591 | BUG_ON(!buffer_uptodate(tbh)); |
| 592 | clear_buffer_dirty(tbh); |
| 593 | get_bh(tbh); |
| 594 | tbh->b_end_io = end_buffer_write_sync; |
| 595 | submit_bh(WRITE, tbh); |
| 596 | } |
| 597 | /* Wait on i/o completion of buffers. */ |
| 598 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { |
| 599 | struct buffer_head *tbh = bhs[i_bhs]; |
| 600 | |
| 601 | wait_on_buffer(tbh); |
| 602 | if (unlikely(!buffer_uptodate(tbh))) { |
| 603 | err = -EIO; |
| 604 | /* |
| 605 | * Set the buffer uptodate so the page and |
| 606 | * buffer states do not become out of sync. |
| 607 | */ |
| 608 | set_buffer_uptodate(tbh); |
| 609 | } |
| 610 | } |
| 611 | } else /* if (unlikely(err)) */ { |
| 612 | /* Clean the buffers. */ |
| 613 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) |
| 614 | clear_buffer_dirty(bhs[i_bhs]); |
| 615 | } |
| 616 | /* Current state: all buffers are clean, unlocked, and uptodate. */ |
| 617 | /* Remove the mst protection fixups again. */ |
| 618 | post_write_mst_fixup((NTFS_RECORD*)kmirr); |
| 619 | flush_dcache_page(page); |
| 620 | SetPageUptodate(page); |
| 621 | unlock_page(page); |
| 622 | ntfs_unmap_page(page); |
| 623 | if (likely(!err)) { |
| 624 | ntfs_debug("Done."); |
| 625 | } else { |
| 626 | ntfs_error(vol->sb, "I/O error while writing mft mirror " |
| 627 | "record 0x%lx!", mft_no); |
| 628 | err_out: |
| 629 | ntfs_error(vol->sb, "Failed to synchronize $MFTMirr (error " |
| 630 | "code %i). Volume will be left marked dirty " |
| 631 | "on umount. Run ntfsfix on the partition " |
| 632 | "after umounting to correct this.", -err); |
| 633 | NVolSetErrors(vol); |
| 634 | } |
| 635 | return err; |
| 636 | } |
| 637 | |
| 638 | /** |
| 639 | * write_mft_record_nolock - write out a mapped (extent) mft record |
| 640 | * @ni: ntfs inode describing the mapped (extent) mft record |
| 641 | * @m: mapped (extent) mft record to write |
| 642 | * @sync: if true, wait for i/o completion |
| 643 | * |
| 644 | * Write the mapped (extent) mft record @m described by the (regular or extent) |
| 645 | * ntfs inode @ni to backing store. If the mft record @m has a counterpart in |
| 646 | * the mft mirror, that is also updated. |
| 647 | * |
| 648 | * We only write the mft record if the ntfs inode @ni is dirty and the first |
| 649 | * buffer belonging to its mft record is dirty, too. We ignore the dirty state |
| 650 | * of subsequent buffers because we could have raced with |
| 651 | * fs/ntfs/aops.c::mark_ntfs_record_dirty(). |
| 652 | * |
| 653 | * On success, clean the mft record and return 0. On error, leave the mft |
| 654 | * record dirty and return -errno. The caller should call make_bad_inode() on |
| 655 | * the base inode to ensure no more access happens to this inode. We do not do |
| 656 | * it here as the caller may want to finish writing other extent mft records |
| 657 | * first to minimize on-disk metadata inconsistencies. |
| 658 | * |
| 659 | * NOTE: We always perform synchronous i/o and ignore the @sync parameter. |
| 660 | * However, if the mft record has a counterpart in the mft mirror and @sync is |
| 661 | * true, we write the mft record, wait for i/o completion, and only then write |
| 662 | * the mft mirror copy. This ensures that if the system crashes either the mft |
| 663 | * or the mft mirror will contain a self-consistent mft record @m. If @sync is |
| 664 | * false on the other hand, we start i/o on both and then wait for completion |
| 665 | * on them. This provides a speedup but no longer guarantees that you will end |
| 666 | * up with a self-consistent mft record in the case of a crash but if you asked |
| 667 | * for asynchronous writing you probably do not care about that anyway. |
| 668 | * |
| 669 | * TODO: If @sync is false, want to do truly asynchronous i/o, i.e. just |
| 670 | * schedule i/o via ->writepage or do it via kntfsd or whatever. |
| 671 | */ |
| 672 | int write_mft_record_nolock(ntfs_inode *ni, MFT_RECORD *m, int sync) |
| 673 | { |
| 674 | ntfs_volume *vol = ni->vol; |
| 675 | struct page *page = ni->page; |
Anton Altaparmakov | 78af34f | 2006-02-24 10:32:33 +0000 | [diff] [blame] | 676 | unsigned int blocksize = vol->sb->s_blocksize; |
| 677 | unsigned char blocksize_bits = vol->sb->s_blocksize_bits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 678 | int max_bhs = vol->mft_record_size / blocksize; |
| 679 | struct buffer_head *bhs[max_bhs]; |
| 680 | struct buffer_head *bh, *head; |
| 681 | runlist_element *rl; |
| 682 | unsigned int block_start, block_end, m_start, m_end; |
| 683 | int i_bhs, nr_bhs, err = 0; |
| 684 | |
| 685 | ntfs_debug("Entering for inode 0x%lx.", ni->mft_no); |
| 686 | BUG_ON(NInoAttr(ni)); |
| 687 | BUG_ON(!max_bhs); |
| 688 | BUG_ON(!PageLocked(page)); |
| 689 | /* |
| 690 | * If the ntfs_inode is clean no need to do anything. If it is dirty, |
| 691 | * mark it as clean now so that it can be redirtied later on if needed. |
| 692 | * There is no danger of races since the caller is holding the locks |
| 693 | * for the mft record @m and the page it is in. |
| 694 | */ |
| 695 | if (!NInoTestClearDirty(ni)) |
| 696 | goto done; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 697 | bh = head = page_buffers(page); |
| 698 | BUG_ON(!bh); |
| 699 | rl = NULL; |
| 700 | nr_bhs = 0; |
| 701 | block_start = 0; |
| 702 | m_start = ni->page_ofs; |
| 703 | m_end = m_start + vol->mft_record_size; |
| 704 | do { |
| 705 | block_end = block_start + blocksize; |
| 706 | /* If the buffer is outside the mft record, skip it. */ |
| 707 | if (block_end <= m_start) |
| 708 | continue; |
| 709 | if (unlikely(block_start >= m_end)) |
| 710 | break; |
| 711 | /* |
| 712 | * If this block is not the first one in the record, we ignore |
| 713 | * the buffer's dirty state because we could have raced with a |
| 714 | * parallel mark_ntfs_record_dirty(). |
| 715 | */ |
| 716 | if (block_start == m_start) { |
| 717 | /* This block is the first one in the record. */ |
| 718 | if (!buffer_dirty(bh)) { |
| 719 | BUG_ON(nr_bhs); |
| 720 | /* Clean records are not written out. */ |
| 721 | break; |
| 722 | } |
| 723 | } |
| 724 | /* Need to map the buffer if it is not mapped already. */ |
| 725 | if (unlikely(!buffer_mapped(bh))) { |
| 726 | VCN vcn; |
| 727 | LCN lcn; |
| 728 | unsigned int vcn_ofs; |
| 729 | |
Anton Altaparmakov | e74589a | 2005-08-16 16:38:28 +0100 | [diff] [blame] | 730 | bh->b_bdev = vol->sb->s_bdev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | /* Obtain the vcn and offset of the current block. */ |
| 732 | vcn = ((VCN)ni->mft_no << vol->mft_record_size_bits) + |
| 733 | (block_start - m_start); |
| 734 | vcn_ofs = vcn & vol->cluster_size_mask; |
| 735 | vcn >>= vol->cluster_size_bits; |
| 736 | if (!rl) { |
| 737 | down_read(&NTFS_I(vol->mft_ino)->runlist.lock); |
| 738 | rl = NTFS_I(vol->mft_ino)->runlist.rl; |
| 739 | BUG_ON(!rl); |
| 740 | } |
| 741 | /* Seek to element containing target vcn. */ |
| 742 | while (rl->length && rl[1].vcn <= vcn) |
| 743 | rl++; |
| 744 | lcn = ntfs_rl_vcn_to_lcn(rl, vcn); |
| 745 | /* For $MFT, only lcn >= 0 is a successful remap. */ |
| 746 | if (likely(lcn >= 0)) { |
| 747 | /* Setup buffer head to correct block. */ |
| 748 | bh->b_blocknr = ((lcn << |
| 749 | vol->cluster_size_bits) + |
| 750 | vcn_ofs) >> blocksize_bits; |
| 751 | set_buffer_mapped(bh); |
| 752 | } else { |
| 753 | bh->b_blocknr = -1; |
| 754 | ntfs_error(vol->sb, "Cannot write mft record " |
| 755 | "0x%lx because its location " |
| 756 | "on disk could not be " |
| 757 | "determined (error code %lli).", |
| 758 | ni->mft_no, (long long)lcn); |
| 759 | err = -EIO; |
| 760 | } |
| 761 | } |
| 762 | BUG_ON(!buffer_uptodate(bh)); |
| 763 | BUG_ON(!nr_bhs && (m_start != block_start)); |
| 764 | BUG_ON(nr_bhs >= max_bhs); |
| 765 | bhs[nr_bhs++] = bh; |
| 766 | BUG_ON((nr_bhs >= max_bhs) && (m_end != block_end)); |
| 767 | } while (block_start = block_end, (bh = bh->b_this_page) != head); |
| 768 | if (unlikely(rl)) |
| 769 | up_read(&NTFS_I(vol->mft_ino)->runlist.lock); |
| 770 | if (!nr_bhs) |
| 771 | goto done; |
| 772 | if (unlikely(err)) |
| 773 | goto cleanup_out; |
| 774 | /* Apply the mst protection fixups. */ |
| 775 | err = pre_write_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size); |
| 776 | if (err) { |
| 777 | ntfs_error(vol->sb, "Failed to apply mst fixups!"); |
| 778 | goto cleanup_out; |
| 779 | } |
| 780 | flush_dcache_mft_record_page(ni); |
| 781 | /* Lock buffers and start synchronous write i/o on them. */ |
| 782 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { |
| 783 | struct buffer_head *tbh = bhs[i_bhs]; |
| 784 | |
| 785 | if (unlikely(test_set_buffer_locked(tbh))) |
| 786 | BUG(); |
| 787 | BUG_ON(!buffer_uptodate(tbh)); |
| 788 | clear_buffer_dirty(tbh); |
| 789 | get_bh(tbh); |
| 790 | tbh->b_end_io = end_buffer_write_sync; |
| 791 | submit_bh(WRITE, tbh); |
| 792 | } |
| 793 | /* Synchronize the mft mirror now if not @sync. */ |
| 794 | if (!sync && ni->mft_no < vol->mftmirr_size) |
| 795 | ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync); |
| 796 | /* Wait on i/o completion of buffers. */ |
| 797 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) { |
| 798 | struct buffer_head *tbh = bhs[i_bhs]; |
| 799 | |
| 800 | wait_on_buffer(tbh); |
| 801 | if (unlikely(!buffer_uptodate(tbh))) { |
| 802 | err = -EIO; |
| 803 | /* |
| 804 | * Set the buffer uptodate so the page and buffer |
| 805 | * states do not become out of sync. |
| 806 | */ |
| 807 | if (PageUptodate(page)) |
| 808 | set_buffer_uptodate(tbh); |
| 809 | } |
| 810 | } |
| 811 | /* If @sync, now synchronize the mft mirror. */ |
| 812 | if (sync && ni->mft_no < vol->mftmirr_size) |
| 813 | ntfs_sync_mft_mirror(vol, ni->mft_no, m, sync); |
| 814 | /* Remove the mst protection fixups again. */ |
| 815 | post_write_mst_fixup((NTFS_RECORD*)m); |
| 816 | flush_dcache_mft_record_page(ni); |
| 817 | if (unlikely(err)) { |
| 818 | /* I/O error during writing. This is really bad! */ |
| 819 | ntfs_error(vol->sb, "I/O error while writing mft record " |
| 820 | "0x%lx! Marking base inode as bad. You " |
| 821 | "should unmount the volume and run chkdsk.", |
| 822 | ni->mft_no); |
| 823 | goto err_out; |
| 824 | } |
| 825 | done: |
| 826 | ntfs_debug("Done."); |
| 827 | return 0; |
| 828 | cleanup_out: |
| 829 | /* Clean the buffers. */ |
| 830 | for (i_bhs = 0; i_bhs < nr_bhs; i_bhs++) |
| 831 | clear_buffer_dirty(bhs[i_bhs]); |
| 832 | err_out: |
| 833 | /* |
| 834 | * Current state: all buffers are clean, unlocked, and uptodate. |
| 835 | * The caller should mark the base inode as bad so that no more i/o |
| 836 | * happens. ->clear_inode() will still be invoked so all extent inodes |
| 837 | * and other allocated memory will be freed. |
| 838 | */ |
| 839 | if (err == -ENOMEM) { |
| 840 | ntfs_error(vol->sb, "Not enough memory to write mft record. " |
| 841 | "Redirtying so the write is retried later."); |
| 842 | mark_mft_record_dirty(ni); |
| 843 | err = 0; |
| 844 | } else |
| 845 | NVolSetErrors(vol); |
| 846 | return err; |
| 847 | } |
| 848 | |
| 849 | /** |
| 850 | * ntfs_may_write_mft_record - check if an mft record may be written out |
| 851 | * @vol: [IN] ntfs volume on which the mft record to check resides |
| 852 | * @mft_no: [IN] mft record number of the mft record to check |
| 853 | * @m: [IN] mapped mft record to check |
| 854 | * @locked_ni: [OUT] caller has to unlock this ntfs inode if one is returned |
| 855 | * |
| 856 | * Check if the mapped (base or extent) mft record @m with mft record number |
| 857 | * @mft_no belonging to the ntfs volume @vol may be written out. If necessary |
| 858 | * and possible the ntfs inode of the mft record is locked and the base vfs |
| 859 | * inode is pinned. The locked ntfs inode is then returned in @locked_ni. The |
| 860 | * caller is responsible for unlocking the ntfs inode and unpinning the base |
| 861 | * vfs inode. |
| 862 | * |
| 863 | * Return TRUE if the mft record may be written out and FALSE if not. |
| 864 | * |
| 865 | * The caller has locked the page and cleared the uptodate flag on it which |
| 866 | * means that we can safely write out any dirty mft records that do not have |
| 867 | * their inodes in icache as determined by ilookup5() as anyone |
| 868 | * opening/creating such an inode would block when attempting to map the mft |
| 869 | * record in read_cache_page() until we are finished with the write out. |
| 870 | * |
| 871 | * Here is a description of the tests we perform: |
| 872 | * |
| 873 | * If the inode is found in icache we know the mft record must be a base mft |
| 874 | * record. If it is dirty, we do not write it and return FALSE as the vfs |
| 875 | * inode write paths will result in the access times being updated which would |
| 876 | * cause the base mft record to be redirtied and written out again. (We know |
| 877 | * the access time update will modify the base mft record because Windows |
| 878 | * chkdsk complains if the standard information attribute is not in the base |
| 879 | * mft record.) |
| 880 | * |
| 881 | * If the inode is in icache and not dirty, we attempt to lock the mft record |
| 882 | * and if we find the lock was already taken, it is not safe to write the mft |
| 883 | * record and we return FALSE. |
| 884 | * |
| 885 | * If we manage to obtain the lock we have exclusive access to the mft record, |
| 886 | * which also allows us safe writeout of the mft record. We then set |
| 887 | * @locked_ni to the locked ntfs inode and return TRUE. |
| 888 | * |
| 889 | * Note we cannot just lock the mft record and sleep while waiting for the lock |
| 890 | * because this would deadlock due to lock reversal (normally the mft record is |
| 891 | * locked before the page is locked but we already have the page locked here |
| 892 | * when we try to lock the mft record). |
| 893 | * |
| 894 | * If the inode is not in icache we need to perform further checks. |
| 895 | * |
| 896 | * If the mft record is not a FILE record or it is a base mft record, we can |
| 897 | * safely write it and return TRUE. |
| 898 | * |
| 899 | * We now know the mft record is an extent mft record. We check if the inode |
| 900 | * corresponding to its base mft record is in icache and obtain a reference to |
| 901 | * it if it is. If it is not, we can safely write it and return TRUE. |
| 902 | * |
| 903 | * We now have the base inode for the extent mft record. We check if it has an |
| 904 | * ntfs inode for the extent mft record attached and if not it is safe to write |
| 905 | * the extent mft record and we return TRUE. |
| 906 | * |
| 907 | * The ntfs inode for the extent mft record is attached to the base inode so we |
| 908 | * attempt to lock the extent mft record and if we find the lock was already |
| 909 | * taken, it is not safe to write the extent mft record and we return FALSE. |
| 910 | * |
| 911 | * If we manage to obtain the lock we have exclusive access to the extent mft |
| 912 | * record, which also allows us safe writeout of the extent mft record. We |
| 913 | * set the ntfs inode of the extent mft record clean and then set @locked_ni to |
| 914 | * the now locked ntfs inode and return TRUE. |
| 915 | * |
| 916 | * Note, the reason for actually writing dirty mft records here and not just |
| 917 | * relying on the vfs inode dirty code paths is that we can have mft records |
| 918 | * modified without them ever having actual inodes in memory. Also we can have |
| 919 | * dirty mft records with clean ntfs inodes in memory. None of the described |
| 920 | * cases would result in the dirty mft records being written out if we only |
| 921 | * relied on the vfs inode dirty code paths. And these cases can really occur |
| 922 | * during allocation of new mft records and in particular when the |
| 923 | * initialized_size of the $MFT/$DATA attribute is extended and the new space |
| 924 | * is initialized using ntfs_mft_record_format(). The clean inode can then |
| 925 | * appear if the mft record is reused for a new inode before it got written |
| 926 | * out. |
| 927 | */ |
| 928 | BOOL ntfs_may_write_mft_record(ntfs_volume *vol, const unsigned long mft_no, |
| 929 | const MFT_RECORD *m, ntfs_inode **locked_ni) |
| 930 | { |
| 931 | struct super_block *sb = vol->sb; |
| 932 | struct inode *mft_vi = vol->mft_ino; |
| 933 | struct inode *vi; |
| 934 | ntfs_inode *ni, *eni, **extent_nis; |
| 935 | int i; |
| 936 | ntfs_attr na; |
| 937 | |
| 938 | ntfs_debug("Entering for inode 0x%lx.", mft_no); |
| 939 | /* |
| 940 | * Normally we do not return a locked inode so set @locked_ni to NULL. |
| 941 | */ |
| 942 | BUG_ON(!locked_ni); |
| 943 | *locked_ni = NULL; |
| 944 | /* |
| 945 | * Check if the inode corresponding to this mft record is in the VFS |
| 946 | * inode cache and obtain a reference to it if it is. |
| 947 | */ |
| 948 | ntfs_debug("Looking for inode 0x%lx in icache.", mft_no); |
| 949 | na.mft_no = mft_no; |
| 950 | na.name = NULL; |
| 951 | na.name_len = 0; |
| 952 | na.type = AT_UNUSED; |
| 953 | /* |
Anton Altaparmakov | ba6d237 | 2005-06-26 22:12:02 +0100 | [diff] [blame] | 954 | * Optimize inode 0, i.e. $MFT itself, since we have it in memory and |
| 955 | * we get here for it rather often. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | */ |
| 957 | if (!mft_no) { |
| 958 | /* Balance the below iput(). */ |
| 959 | vi = igrab(mft_vi); |
| 960 | BUG_ON(vi != mft_vi); |
Anton Altaparmakov | ba6d237 | 2005-06-26 22:12:02 +0100 | [diff] [blame] | 961 | } else { |
| 962 | /* |
| 963 | * Have to use ilookup5_nowait() since ilookup5() waits for the |
| 964 | * inode lock which causes ntfs to deadlock when a concurrent |
| 965 | * inode write via the inode dirty code paths and the page |
| 966 | * dirty code path of the inode dirty code path when writing |
| 967 | * $MFT occurs. |
| 968 | */ |
| 969 | vi = ilookup5_nowait(sb, mft_no, (test_t)ntfs_test_inode, &na); |
| 970 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | if (vi) { |
| 972 | ntfs_debug("Base inode 0x%lx is in icache.", mft_no); |
| 973 | /* The inode is in icache. */ |
| 974 | ni = NTFS_I(vi); |
| 975 | /* Take a reference to the ntfs inode. */ |
| 976 | atomic_inc(&ni->count); |
| 977 | /* If the inode is dirty, do not write this record. */ |
| 978 | if (NInoDirty(ni)) { |
| 979 | ntfs_debug("Inode 0x%lx is dirty, do not write it.", |
| 980 | mft_no); |
| 981 | atomic_dec(&ni->count); |
| 982 | iput(vi); |
| 983 | return FALSE; |
| 984 | } |
| 985 | ntfs_debug("Inode 0x%lx is not dirty.", mft_no); |
| 986 | /* The inode is not dirty, try to take the mft record lock. */ |
| 987 | if (unlikely(down_trylock(&ni->mrec_lock))) { |
| 988 | ntfs_debug("Mft record 0x%lx is already locked, do " |
| 989 | "not write it.", mft_no); |
| 990 | atomic_dec(&ni->count); |
| 991 | iput(vi); |
| 992 | return FALSE; |
| 993 | } |
| 994 | ntfs_debug("Managed to lock mft record 0x%lx, write it.", |
| 995 | mft_no); |
| 996 | /* |
| 997 | * The write has to occur while we hold the mft record lock so |
| 998 | * return the locked ntfs inode. |
| 999 | */ |
| 1000 | *locked_ni = ni; |
| 1001 | return TRUE; |
| 1002 | } |
| 1003 | ntfs_debug("Inode 0x%lx is not in icache.", mft_no); |
| 1004 | /* The inode is not in icache. */ |
| 1005 | /* Write the record if it is not a mft record (type "FILE"). */ |
| 1006 | if (!ntfs_is_mft_record(m->magic)) { |
| 1007 | ntfs_debug("Mft record 0x%lx is not a FILE record, write it.", |
| 1008 | mft_no); |
| 1009 | return TRUE; |
| 1010 | } |
| 1011 | /* Write the mft record if it is a base inode. */ |
| 1012 | if (!m->base_mft_record) { |
| 1013 | ntfs_debug("Mft record 0x%lx is a base record, write it.", |
| 1014 | mft_no); |
| 1015 | return TRUE; |
| 1016 | } |
| 1017 | /* |
| 1018 | * This is an extent mft record. Check if the inode corresponding to |
| 1019 | * its base mft record is in icache and obtain a reference to it if it |
| 1020 | * is. |
| 1021 | */ |
| 1022 | na.mft_no = MREF_LE(m->base_mft_record); |
| 1023 | ntfs_debug("Mft record 0x%lx is an extent record. Looking for base " |
| 1024 | "inode 0x%lx in icache.", mft_no, na.mft_no); |
Anton Altaparmakov | ba6d237 | 2005-06-26 22:12:02 +0100 | [diff] [blame] | 1025 | if (!na.mft_no) { |
| 1026 | /* Balance the below iput(). */ |
| 1027 | vi = igrab(mft_vi); |
| 1028 | BUG_ON(vi != mft_vi); |
| 1029 | } else |
| 1030 | vi = ilookup5_nowait(sb, na.mft_no, (test_t)ntfs_test_inode, |
| 1031 | &na); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | if (!vi) { |
| 1033 | /* |
| 1034 | * The base inode is not in icache, write this extent mft |
| 1035 | * record. |
| 1036 | */ |
| 1037 | ntfs_debug("Base inode 0x%lx is not in icache, write the " |
| 1038 | "extent record.", na.mft_no); |
| 1039 | return TRUE; |
| 1040 | } |
| 1041 | ntfs_debug("Base inode 0x%lx is in icache.", na.mft_no); |
| 1042 | /* |
| 1043 | * The base inode is in icache. Check if it has the extent inode |
| 1044 | * corresponding to this extent mft record attached. |
| 1045 | */ |
| 1046 | ni = NTFS_I(vi); |
| 1047 | down(&ni->extent_lock); |
| 1048 | if (ni->nr_extents <= 0) { |
| 1049 | /* |
| 1050 | * The base inode has no attached extent inodes, write this |
| 1051 | * extent mft record. |
| 1052 | */ |
| 1053 | up(&ni->extent_lock); |
| 1054 | iput(vi); |
| 1055 | ntfs_debug("Base inode 0x%lx has no attached extent inodes, " |
| 1056 | "write the extent record.", na.mft_no); |
| 1057 | return TRUE; |
| 1058 | } |
| 1059 | /* Iterate over the attached extent inodes. */ |
| 1060 | extent_nis = ni->ext.extent_ntfs_inos; |
| 1061 | for (eni = NULL, i = 0; i < ni->nr_extents; ++i) { |
| 1062 | if (mft_no == extent_nis[i]->mft_no) { |
| 1063 | /* |
| 1064 | * Found the extent inode corresponding to this extent |
| 1065 | * mft record. |
| 1066 | */ |
| 1067 | eni = extent_nis[i]; |
| 1068 | break; |
| 1069 | } |
| 1070 | } |
| 1071 | /* |
| 1072 | * If the extent inode was not attached to the base inode, write this |
| 1073 | * extent mft record. |
| 1074 | */ |
| 1075 | if (!eni) { |
| 1076 | up(&ni->extent_lock); |
| 1077 | iput(vi); |
| 1078 | ntfs_debug("Extent inode 0x%lx is not attached to its base " |
| 1079 | "inode 0x%lx, write the extent record.", |
| 1080 | mft_no, na.mft_no); |
| 1081 | return TRUE; |
| 1082 | } |
| 1083 | ntfs_debug("Extent inode 0x%lx is attached to its base inode 0x%lx.", |
| 1084 | mft_no, na.mft_no); |
| 1085 | /* Take a reference to the extent ntfs inode. */ |
| 1086 | atomic_inc(&eni->count); |
| 1087 | up(&ni->extent_lock); |
| 1088 | /* |
| 1089 | * Found the extent inode coresponding to this extent mft record. |
| 1090 | * Try to take the mft record lock. |
| 1091 | */ |
| 1092 | if (unlikely(down_trylock(&eni->mrec_lock))) { |
| 1093 | atomic_dec(&eni->count); |
| 1094 | iput(vi); |
| 1095 | ntfs_debug("Extent mft record 0x%lx is already locked, do " |
| 1096 | "not write it.", mft_no); |
| 1097 | return FALSE; |
| 1098 | } |
| 1099 | ntfs_debug("Managed to lock extent mft record 0x%lx, write it.", |
| 1100 | mft_no); |
| 1101 | if (NInoTestClearDirty(eni)) |
| 1102 | ntfs_debug("Extent inode 0x%lx is dirty, marking it clean.", |
| 1103 | mft_no); |
| 1104 | /* |
| 1105 | * The write has to occur while we hold the mft record lock so return |
| 1106 | * the locked extent ntfs inode. |
| 1107 | */ |
| 1108 | *locked_ni = eni; |
| 1109 | return TRUE; |
| 1110 | } |
| 1111 | |
| 1112 | static const char *es = " Leaving inconsistent metadata. Unmount and run " |
| 1113 | "chkdsk."; |
| 1114 | |
| 1115 | /** |
| 1116 | * ntfs_mft_bitmap_find_and_alloc_free_rec_nolock - see name |
| 1117 | * @vol: volume on which to search for a free mft record |
| 1118 | * @base_ni: open base inode if allocating an extent mft record or NULL |
| 1119 | * |
| 1120 | * Search for a free mft record in the mft bitmap attribute on the ntfs volume |
| 1121 | * @vol. |
| 1122 | * |
| 1123 | * If @base_ni is NULL start the search at the default allocator position. |
| 1124 | * |
| 1125 | * If @base_ni is not NULL start the search at the mft record after the base |
| 1126 | * mft record @base_ni. |
| 1127 | * |
| 1128 | * Return the free mft record on success and -errno on error. An error code of |
| 1129 | * -ENOSPC means that there are no free mft records in the currently |
| 1130 | * initialized mft bitmap. |
| 1131 | * |
| 1132 | * Locking: Caller must hold vol->mftbmp_lock for writing. |
| 1133 | */ |
| 1134 | static int ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(ntfs_volume *vol, |
| 1135 | ntfs_inode *base_ni) |
| 1136 | { |
| 1137 | s64 pass_end, ll, data_pos, pass_start, ofs, bit; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1138 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1139 | struct address_space *mftbmp_mapping; |
| 1140 | u8 *buf, *byte; |
| 1141 | struct page *page; |
| 1142 | unsigned int page_ofs, size; |
| 1143 | u8 pass, b; |
| 1144 | |
| 1145 | ntfs_debug("Searching for free mft record in the currently " |
| 1146 | "initialized mft bitmap."); |
| 1147 | mftbmp_mapping = vol->mftbmp_ino->i_mapping; |
| 1148 | /* |
| 1149 | * Set the end of the pass making sure we do not overflow the mft |
| 1150 | * bitmap. |
| 1151 | */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1152 | read_lock_irqsave(&NTFS_I(vol->mft_ino)->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1153 | pass_end = NTFS_I(vol->mft_ino)->allocated_size >> |
| 1154 | vol->mft_record_size_bits; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1155 | read_unlock_irqrestore(&NTFS_I(vol->mft_ino)->size_lock, flags); |
| 1156 | read_lock_irqsave(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1157 | ll = NTFS_I(vol->mftbmp_ino)->initialized_size << 3; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1158 | read_unlock_irqrestore(&NTFS_I(vol->mftbmp_ino)->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1159 | if (pass_end > ll) |
| 1160 | pass_end = ll; |
| 1161 | pass = 1; |
| 1162 | if (!base_ni) |
| 1163 | data_pos = vol->mft_data_pos; |
| 1164 | else |
| 1165 | data_pos = base_ni->mft_no + 1; |
| 1166 | if (data_pos < 24) |
| 1167 | data_pos = 24; |
| 1168 | if (data_pos >= pass_end) { |
| 1169 | data_pos = 24; |
| 1170 | pass = 2; |
| 1171 | /* This happens on a freshly formatted volume. */ |
| 1172 | if (data_pos >= pass_end) |
| 1173 | return -ENOSPC; |
| 1174 | } |
| 1175 | pass_start = data_pos; |
| 1176 | ntfs_debug("Starting bitmap search: pass %u, pass_start 0x%llx, " |
| 1177 | "pass_end 0x%llx, data_pos 0x%llx.", pass, |
| 1178 | (long long)pass_start, (long long)pass_end, |
| 1179 | (long long)data_pos); |
| 1180 | /* Loop until a free mft record is found. */ |
| 1181 | for (; pass <= 2;) { |
| 1182 | /* Cap size to pass_end. */ |
| 1183 | ofs = data_pos >> 3; |
| 1184 | page_ofs = ofs & ~PAGE_CACHE_MASK; |
| 1185 | size = PAGE_CACHE_SIZE - page_ofs; |
| 1186 | ll = ((pass_end + 7) >> 3) - ofs; |
| 1187 | if (size > ll) |
| 1188 | size = ll; |
| 1189 | size <<= 3; |
| 1190 | /* |
| 1191 | * If we are still within the active pass, search the next page |
| 1192 | * for a zero bit. |
| 1193 | */ |
| 1194 | if (size) { |
| 1195 | page = ntfs_map_page(mftbmp_mapping, |
| 1196 | ofs >> PAGE_CACHE_SHIFT); |
| 1197 | if (unlikely(IS_ERR(page))) { |
| 1198 | ntfs_error(vol->sb, "Failed to read mft " |
| 1199 | "bitmap, aborting."); |
| 1200 | return PTR_ERR(page); |
| 1201 | } |
| 1202 | buf = (u8*)page_address(page) + page_ofs; |
| 1203 | bit = data_pos & 7; |
| 1204 | data_pos &= ~7ull; |
| 1205 | ntfs_debug("Before inner for loop: size 0x%x, " |
| 1206 | "data_pos 0x%llx, bit 0x%llx", size, |
| 1207 | (long long)data_pos, (long long)bit); |
| 1208 | for (; bit < size && data_pos + bit < pass_end; |
| 1209 | bit &= ~7ull, bit += 8) { |
| 1210 | byte = buf + (bit >> 3); |
| 1211 | if (*byte == 0xff) |
| 1212 | continue; |
| 1213 | b = ffz((unsigned long)*byte); |
| 1214 | if (b < 8 && b >= (bit & 7)) { |
| 1215 | ll = data_pos + (bit & ~7ull) + b; |
| 1216 | if (unlikely(ll > (1ll << 32))) { |
| 1217 | ntfs_unmap_page(page); |
| 1218 | return -ENOSPC; |
| 1219 | } |
| 1220 | *byte |= 1 << b; |
| 1221 | flush_dcache_page(page); |
| 1222 | set_page_dirty(page); |
| 1223 | ntfs_unmap_page(page); |
| 1224 | ntfs_debug("Done. (Found and " |
| 1225 | "allocated mft record " |
| 1226 | "0x%llx.)", |
| 1227 | (long long)ll); |
| 1228 | return ll; |
| 1229 | } |
| 1230 | } |
| 1231 | ntfs_debug("After inner for loop: size 0x%x, " |
| 1232 | "data_pos 0x%llx, bit 0x%llx", size, |
| 1233 | (long long)data_pos, (long long)bit); |
| 1234 | data_pos += size; |
| 1235 | ntfs_unmap_page(page); |
| 1236 | /* |
| 1237 | * If the end of the pass has not been reached yet, |
| 1238 | * continue searching the mft bitmap for a zero bit. |
| 1239 | */ |
| 1240 | if (data_pos < pass_end) |
| 1241 | continue; |
| 1242 | } |
| 1243 | /* Do the next pass. */ |
| 1244 | if (++pass == 2) { |
| 1245 | /* |
| 1246 | * Starting the second pass, in which we scan the first |
| 1247 | * part of the zone which we omitted earlier. |
| 1248 | */ |
| 1249 | pass_end = pass_start; |
| 1250 | data_pos = pass_start = 24; |
| 1251 | ntfs_debug("pass %i, pass_start 0x%llx, pass_end " |
| 1252 | "0x%llx.", pass, (long long)pass_start, |
| 1253 | (long long)pass_end); |
| 1254 | if (data_pos >= pass_end) |
| 1255 | break; |
| 1256 | } |
| 1257 | } |
| 1258 | /* No free mft records in currently initialized mft bitmap. */ |
| 1259 | ntfs_debug("Done. (No free mft records left in currently initialized " |
| 1260 | "mft bitmap.)"); |
| 1261 | return -ENOSPC; |
| 1262 | } |
| 1263 | |
| 1264 | /** |
| 1265 | * ntfs_mft_bitmap_extend_allocation_nolock - extend mft bitmap by a cluster |
| 1266 | * @vol: volume on which to extend the mft bitmap attribute |
| 1267 | * |
| 1268 | * Extend the mft bitmap attribute on the ntfs volume @vol by one cluster. |
| 1269 | * |
| 1270 | * Note: Only changes allocated_size, i.e. does not touch initialized_size or |
| 1271 | * data_size. |
| 1272 | * |
| 1273 | * Return 0 on success and -errno on error. |
| 1274 | * |
| 1275 | * Locking: - Caller must hold vol->mftbmp_lock for writing. |
| 1276 | * - This function takes NTFS_I(vol->mftbmp_ino)->runlist.lock for |
| 1277 | * writing and releases it before returning. |
| 1278 | * - This function takes vol->lcnbmp_lock for writing and releases it |
| 1279 | * before returning. |
| 1280 | */ |
| 1281 | static int ntfs_mft_bitmap_extend_allocation_nolock(ntfs_volume *vol) |
| 1282 | { |
| 1283 | LCN lcn; |
| 1284 | s64 ll; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1285 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1286 | struct page *page; |
| 1287 | ntfs_inode *mft_ni, *mftbmp_ni; |
| 1288 | runlist_element *rl, *rl2 = NULL; |
| 1289 | ntfs_attr_search_ctx *ctx = NULL; |
| 1290 | MFT_RECORD *mrec; |
| 1291 | ATTR_RECORD *a = NULL; |
| 1292 | int ret, mp_size; |
| 1293 | u32 old_alen = 0; |
| 1294 | u8 *b, tb; |
| 1295 | struct { |
| 1296 | u8 added_cluster:1; |
| 1297 | u8 added_run:1; |
| 1298 | u8 mp_rebuilt:1; |
| 1299 | } status = { 0, 0, 0 }; |
| 1300 | |
| 1301 | ntfs_debug("Extending mft bitmap allocation."); |
| 1302 | mft_ni = NTFS_I(vol->mft_ino); |
| 1303 | mftbmp_ni = NTFS_I(vol->mftbmp_ino); |
| 1304 | /* |
| 1305 | * Determine the last lcn of the mft bitmap. The allocated size of the |
| 1306 | * mft bitmap cannot be zero so we are ok to do this. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1307 | */ |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1308 | down_write(&mftbmp_ni->runlist.lock); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1309 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
| 1310 | ll = mftbmp_ni->allocated_size; |
| 1311 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
Anton Altaparmakov | c0c1cc0 | 2005-03-07 21:43:38 +0000 | [diff] [blame] | 1312 | rl = ntfs_attr_find_vcn_nolock(mftbmp_ni, |
Anton Altaparmakov | 69b41e3 | 2005-10-04 14:01:14 +0100 | [diff] [blame] | 1313 | (ll - 1) >> vol->cluster_size_bits, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1314 | if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1315 | up_write(&mftbmp_ni->runlist.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1316 | ntfs_error(vol->sb, "Failed to determine last allocated " |
| 1317 | "cluster of mft bitmap attribute."); |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1318 | if (!IS_ERR(rl)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1319 | ret = -EIO; |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1320 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1321 | ret = PTR_ERR(rl); |
| 1322 | return ret; |
| 1323 | } |
| 1324 | lcn = rl->lcn + rl->length; |
| 1325 | ntfs_debug("Last lcn of mft bitmap attribute is 0x%llx.", |
| 1326 | (long long)lcn); |
| 1327 | /* |
| 1328 | * Attempt to get the cluster following the last allocated cluster by |
| 1329 | * hand as it may be in the MFT zone so the allocator would not give it |
| 1330 | * to us. |
| 1331 | */ |
| 1332 | ll = lcn >> 3; |
| 1333 | page = ntfs_map_page(vol->lcnbmp_ino->i_mapping, |
| 1334 | ll >> PAGE_CACHE_SHIFT); |
| 1335 | if (IS_ERR(page)) { |
| 1336 | up_write(&mftbmp_ni->runlist.lock); |
| 1337 | ntfs_error(vol->sb, "Failed to read from lcn bitmap."); |
| 1338 | return PTR_ERR(page); |
| 1339 | } |
| 1340 | b = (u8*)page_address(page) + (ll & ~PAGE_CACHE_MASK); |
| 1341 | tb = 1 << (lcn & 7ull); |
| 1342 | down_write(&vol->lcnbmp_lock); |
| 1343 | if (*b != 0xff && !(*b & tb)) { |
| 1344 | /* Next cluster is free, allocate it. */ |
| 1345 | *b |= tb; |
| 1346 | flush_dcache_page(page); |
| 1347 | set_page_dirty(page); |
| 1348 | up_write(&vol->lcnbmp_lock); |
| 1349 | ntfs_unmap_page(page); |
| 1350 | /* Update the mft bitmap runlist. */ |
| 1351 | rl->length++; |
| 1352 | rl[1].vcn++; |
| 1353 | status.added_cluster = 1; |
| 1354 | ntfs_debug("Appending one cluster to mft bitmap."); |
| 1355 | } else { |
| 1356 | up_write(&vol->lcnbmp_lock); |
| 1357 | ntfs_unmap_page(page); |
| 1358 | /* Allocate a cluster from the DATA_ZONE. */ |
Anton Altaparmakov | fc0fa7d | 2005-10-04 14:36:56 +0100 | [diff] [blame] | 1359 | rl2 = ntfs_cluster_alloc(vol, rl[1].vcn, 1, lcn, DATA_ZONE, |
| 1360 | TRUE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | if (IS_ERR(rl2)) { |
| 1362 | up_write(&mftbmp_ni->runlist.lock); |
| 1363 | ntfs_error(vol->sb, "Failed to allocate a cluster for " |
| 1364 | "the mft bitmap."); |
| 1365 | return PTR_ERR(rl2); |
| 1366 | } |
| 1367 | rl = ntfs_runlists_merge(mftbmp_ni->runlist.rl, rl2); |
| 1368 | if (IS_ERR(rl)) { |
| 1369 | up_write(&mftbmp_ni->runlist.lock); |
| 1370 | ntfs_error(vol->sb, "Failed to merge runlists for mft " |
| 1371 | "bitmap."); |
| 1372 | if (ntfs_cluster_free_from_rl(vol, rl2)) { |
| 1373 | ntfs_error(vol->sb, "Failed to dealocate " |
| 1374 | "allocated cluster.%s", es); |
| 1375 | NVolSetErrors(vol); |
| 1376 | } |
| 1377 | ntfs_free(rl2); |
| 1378 | return PTR_ERR(rl); |
| 1379 | } |
| 1380 | mftbmp_ni->runlist.rl = rl; |
| 1381 | status.added_run = 1; |
| 1382 | ntfs_debug("Adding one run to mft bitmap."); |
| 1383 | /* Find the last run in the new runlist. */ |
| 1384 | for (; rl[1].length; rl++) |
| 1385 | ; |
| 1386 | } |
| 1387 | /* |
| 1388 | * Update the attribute record as well. Note: @rl is the last |
| 1389 | * (non-terminator) runlist element of mft bitmap. |
| 1390 | */ |
| 1391 | mrec = map_mft_record(mft_ni); |
| 1392 | if (IS_ERR(mrec)) { |
| 1393 | ntfs_error(vol->sb, "Failed to map mft record."); |
| 1394 | ret = PTR_ERR(mrec); |
| 1395 | goto undo_alloc; |
| 1396 | } |
| 1397 | ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); |
| 1398 | if (unlikely(!ctx)) { |
| 1399 | ntfs_error(vol->sb, "Failed to get search context."); |
| 1400 | ret = -ENOMEM; |
| 1401 | goto undo_alloc; |
| 1402 | } |
| 1403 | ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, |
| 1404 | mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, |
| 1405 | 0, ctx); |
| 1406 | if (unlikely(ret)) { |
| 1407 | ntfs_error(vol->sb, "Failed to find last attribute extent of " |
| 1408 | "mft bitmap attribute."); |
| 1409 | if (ret == -ENOENT) |
| 1410 | ret = -EIO; |
| 1411 | goto undo_alloc; |
| 1412 | } |
| 1413 | a = ctx->attr; |
| 1414 | ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); |
| 1415 | /* Search back for the previous last allocated cluster of mft bitmap. */ |
| 1416 | for (rl2 = rl; rl2 > mftbmp_ni->runlist.rl; rl2--) { |
| 1417 | if (ll >= rl2->vcn) |
| 1418 | break; |
| 1419 | } |
| 1420 | BUG_ON(ll < rl2->vcn); |
| 1421 | BUG_ON(ll >= rl2->vcn + rl2->length); |
| 1422 | /* Get the size for the new mapping pairs array for this extent. */ |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1423 | mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1424 | if (unlikely(mp_size <= 0)) { |
| 1425 | ntfs_error(vol->sb, "Get size for mapping pairs failed for " |
| 1426 | "mft bitmap attribute extent."); |
| 1427 | ret = mp_size; |
| 1428 | if (!ret) |
| 1429 | ret = -EIO; |
| 1430 | goto undo_alloc; |
| 1431 | } |
| 1432 | /* Expand the attribute record if necessary. */ |
| 1433 | old_alen = le32_to_cpu(a->length); |
| 1434 | ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + |
| 1435 | le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); |
| 1436 | if (unlikely(ret)) { |
| 1437 | if (ret != -ENOSPC) { |
| 1438 | ntfs_error(vol->sb, "Failed to resize attribute " |
| 1439 | "record for mft bitmap attribute."); |
| 1440 | goto undo_alloc; |
| 1441 | } |
| 1442 | // TODO: Deal with this by moving this extent to a new mft |
| 1443 | // record or by starting a new extent in a new mft record or by |
| 1444 | // moving other attributes out of this mft record. |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1445 | // Note: It will need to be a special mft record and if none of |
| 1446 | // those are available it gets rather complicated... |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | ntfs_error(vol->sb, "Not enough space in this mft record to " |
| 1448 | "accomodate extended mft bitmap attribute " |
| 1449 | "extent. Cannot handle this yet."); |
| 1450 | ret = -EOPNOTSUPP; |
| 1451 | goto undo_alloc; |
| 1452 | } |
| 1453 | status.mp_rebuilt = 1; |
| 1454 | /* Generate the mapping pairs array directly into the attr record. */ |
| 1455 | ret = ntfs_mapping_pairs_build(vol, (u8*)a + |
| 1456 | le16_to_cpu(a->data.non_resident.mapping_pairs_offset), |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1457 | mp_size, rl2, ll, -1, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1458 | if (unlikely(ret)) { |
| 1459 | ntfs_error(vol->sb, "Failed to build mapping pairs array for " |
| 1460 | "mft bitmap attribute."); |
| 1461 | goto undo_alloc; |
| 1462 | } |
| 1463 | /* Update the highest_vcn. */ |
| 1464 | a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); |
| 1465 | /* |
| 1466 | * We now have extended the mft bitmap allocated_size by one cluster. |
| 1467 | * Reflect this in the ntfs_inode structure and the attribute record. |
| 1468 | */ |
| 1469 | if (a->data.non_resident.lowest_vcn) { |
| 1470 | /* |
| 1471 | * We are not in the first attribute extent, switch to it, but |
| 1472 | * first ensure the changes will make it to disk later. |
| 1473 | */ |
| 1474 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1475 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1476 | ntfs_attr_reinit_search_ctx(ctx); |
| 1477 | ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, |
| 1478 | mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, |
| 1479 | 0, ctx); |
| 1480 | if (unlikely(ret)) { |
| 1481 | ntfs_error(vol->sb, "Failed to find first attribute " |
| 1482 | "extent of mft bitmap attribute."); |
| 1483 | goto restore_undo_alloc; |
| 1484 | } |
| 1485 | a = ctx->attr; |
| 1486 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1487 | write_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | mftbmp_ni->allocated_size += vol->cluster_size; |
| 1489 | a->data.non_resident.allocated_size = |
| 1490 | cpu_to_sle64(mftbmp_ni->allocated_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1491 | write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1492 | /* Ensure the changes make it to disk. */ |
| 1493 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1494 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1495 | ntfs_attr_put_search_ctx(ctx); |
| 1496 | unmap_mft_record(mft_ni); |
| 1497 | up_write(&mftbmp_ni->runlist.lock); |
| 1498 | ntfs_debug("Done."); |
| 1499 | return 0; |
| 1500 | restore_undo_alloc: |
| 1501 | ntfs_attr_reinit_search_ctx(ctx); |
| 1502 | if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, |
| 1503 | mftbmp_ni->name_len, CASE_SENSITIVE, rl[1].vcn, NULL, |
| 1504 | 0, ctx)) { |
| 1505 | ntfs_error(vol->sb, "Failed to find last attribute extent of " |
| 1506 | "mft bitmap attribute.%s", es); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1507 | write_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | mftbmp_ni->allocated_size += vol->cluster_size; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1509 | write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | ntfs_attr_put_search_ctx(ctx); |
| 1511 | unmap_mft_record(mft_ni); |
| 1512 | up_write(&mftbmp_ni->runlist.lock); |
| 1513 | /* |
| 1514 | * The only thing that is now wrong is ->allocated_size of the |
| 1515 | * base attribute extent which chkdsk should be able to fix. |
| 1516 | */ |
| 1517 | NVolSetErrors(vol); |
| 1518 | return ret; |
| 1519 | } |
| 1520 | a = ctx->attr; |
| 1521 | a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 2); |
| 1522 | undo_alloc: |
| 1523 | if (status.added_cluster) { |
| 1524 | /* Truncate the last run in the runlist by one cluster. */ |
| 1525 | rl->length--; |
| 1526 | rl[1].vcn--; |
| 1527 | } else if (status.added_run) { |
| 1528 | lcn = rl->lcn; |
| 1529 | /* Remove the last run from the runlist. */ |
| 1530 | rl->lcn = rl[1].lcn; |
| 1531 | rl->length = 0; |
| 1532 | } |
| 1533 | /* Deallocate the cluster. */ |
| 1534 | down_write(&vol->lcnbmp_lock); |
| 1535 | if (ntfs_bitmap_clear_bit(vol->lcnbmp_ino, lcn)) { |
| 1536 | ntfs_error(vol->sb, "Failed to free allocated cluster.%s", es); |
| 1537 | NVolSetErrors(vol); |
| 1538 | } |
| 1539 | up_write(&vol->lcnbmp_lock); |
| 1540 | if (status.mp_rebuilt) { |
| 1541 | if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( |
| 1542 | a->data.non_resident.mapping_pairs_offset), |
| 1543 | old_alen - le16_to_cpu( |
| 1544 | a->data.non_resident.mapping_pairs_offset), |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1545 | rl2, ll, -1, NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1546 | ntfs_error(vol->sb, "Failed to restore mapping pairs " |
| 1547 | "array.%s", es); |
| 1548 | NVolSetErrors(vol); |
| 1549 | } |
| 1550 | if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { |
| 1551 | ntfs_error(vol->sb, "Failed to restore attribute " |
| 1552 | "record.%s", es); |
| 1553 | NVolSetErrors(vol); |
| 1554 | } |
| 1555 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1556 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1557 | } |
| 1558 | if (ctx) |
| 1559 | ntfs_attr_put_search_ctx(ctx); |
| 1560 | if (!IS_ERR(mrec)) |
| 1561 | unmap_mft_record(mft_ni); |
| 1562 | up_write(&mftbmp_ni->runlist.lock); |
| 1563 | return ret; |
| 1564 | } |
| 1565 | |
| 1566 | /** |
| 1567 | * ntfs_mft_bitmap_extend_initialized_nolock - extend mftbmp initialized data |
| 1568 | * @vol: volume on which to extend the mft bitmap attribute |
| 1569 | * |
| 1570 | * Extend the initialized portion of the mft bitmap attribute on the ntfs |
| 1571 | * volume @vol by 8 bytes. |
| 1572 | * |
| 1573 | * Note: Only changes initialized_size and data_size, i.e. requires that |
| 1574 | * allocated_size is big enough to fit the new initialized_size. |
| 1575 | * |
| 1576 | * Return 0 on success and -error on error. |
| 1577 | * |
| 1578 | * Locking: Caller must hold vol->mftbmp_lock for writing. |
| 1579 | */ |
| 1580 | static int ntfs_mft_bitmap_extend_initialized_nolock(ntfs_volume *vol) |
| 1581 | { |
| 1582 | s64 old_data_size, old_initialized_size; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1583 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1584 | struct inode *mftbmp_vi; |
| 1585 | ntfs_inode *mft_ni, *mftbmp_ni; |
| 1586 | ntfs_attr_search_ctx *ctx; |
| 1587 | MFT_RECORD *mrec; |
| 1588 | ATTR_RECORD *a; |
| 1589 | int ret; |
| 1590 | |
| 1591 | ntfs_debug("Extending mft bitmap initiailized (and data) size."); |
| 1592 | mft_ni = NTFS_I(vol->mft_ino); |
| 1593 | mftbmp_vi = vol->mftbmp_ino; |
| 1594 | mftbmp_ni = NTFS_I(mftbmp_vi); |
| 1595 | /* Get the attribute record. */ |
| 1596 | mrec = map_mft_record(mft_ni); |
| 1597 | if (IS_ERR(mrec)) { |
| 1598 | ntfs_error(vol->sb, "Failed to map mft record."); |
| 1599 | return PTR_ERR(mrec); |
| 1600 | } |
| 1601 | ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); |
| 1602 | if (unlikely(!ctx)) { |
| 1603 | ntfs_error(vol->sb, "Failed to get search context."); |
| 1604 | ret = -ENOMEM; |
| 1605 | goto unm_err_out; |
| 1606 | } |
| 1607 | ret = ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, |
| 1608 | mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 1609 | if (unlikely(ret)) { |
| 1610 | ntfs_error(vol->sb, "Failed to find first attribute extent of " |
| 1611 | "mft bitmap attribute."); |
| 1612 | if (ret == -ENOENT) |
| 1613 | ret = -EIO; |
| 1614 | goto put_err_out; |
| 1615 | } |
| 1616 | a = ctx->attr; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1617 | write_lock_irqsave(&mftbmp_ni->size_lock, flags); |
| 1618 | old_data_size = i_size_read(mftbmp_vi); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1619 | old_initialized_size = mftbmp_ni->initialized_size; |
| 1620 | /* |
| 1621 | * We can simply update the initialized_size before filling the space |
| 1622 | * with zeroes because the caller is holding the mft bitmap lock for |
| 1623 | * writing which ensures that no one else is trying to access the data. |
| 1624 | */ |
| 1625 | mftbmp_ni->initialized_size += 8; |
| 1626 | a->data.non_resident.initialized_size = |
| 1627 | cpu_to_sle64(mftbmp_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1628 | if (mftbmp_ni->initialized_size > old_data_size) { |
| 1629 | i_size_write(mftbmp_vi, mftbmp_ni->initialized_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | a->data.non_resident.data_size = |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1631 | cpu_to_sle64(mftbmp_ni->initialized_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1632 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1633 | write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1634 | /* Ensure the changes make it to disk. */ |
| 1635 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1636 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1637 | ntfs_attr_put_search_ctx(ctx); |
| 1638 | unmap_mft_record(mft_ni); |
| 1639 | /* Initialize the mft bitmap attribute value with zeroes. */ |
| 1640 | ret = ntfs_attr_set(mftbmp_ni, old_initialized_size, 8, 0); |
| 1641 | if (likely(!ret)) { |
| 1642 | ntfs_debug("Done. (Wrote eight initialized bytes to mft " |
| 1643 | "bitmap."); |
| 1644 | return 0; |
| 1645 | } |
| 1646 | ntfs_error(vol->sb, "Failed to write to mft bitmap."); |
| 1647 | /* Try to recover from the error. */ |
| 1648 | mrec = map_mft_record(mft_ni); |
| 1649 | if (IS_ERR(mrec)) { |
| 1650 | ntfs_error(vol->sb, "Failed to map mft record.%s", es); |
| 1651 | NVolSetErrors(vol); |
| 1652 | return ret; |
| 1653 | } |
| 1654 | ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); |
| 1655 | if (unlikely(!ctx)) { |
| 1656 | ntfs_error(vol->sb, "Failed to get search context.%s", es); |
| 1657 | NVolSetErrors(vol); |
| 1658 | goto unm_err_out; |
| 1659 | } |
| 1660 | if (ntfs_attr_lookup(mftbmp_ni->type, mftbmp_ni->name, |
| 1661 | mftbmp_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, ctx)) { |
| 1662 | ntfs_error(vol->sb, "Failed to find first attribute extent of " |
| 1663 | "mft bitmap attribute.%s", es); |
| 1664 | NVolSetErrors(vol); |
| 1665 | put_err_out: |
| 1666 | ntfs_attr_put_search_ctx(ctx); |
| 1667 | unm_err_out: |
| 1668 | unmap_mft_record(mft_ni); |
| 1669 | goto err_out; |
| 1670 | } |
| 1671 | a = ctx->attr; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1672 | write_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1673 | mftbmp_ni->initialized_size = old_initialized_size; |
| 1674 | a->data.non_resident.initialized_size = |
| 1675 | cpu_to_sle64(old_initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1676 | if (i_size_read(mftbmp_vi) != old_data_size) { |
| 1677 | i_size_write(mftbmp_vi, old_data_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1678 | a->data.non_resident.data_size = cpu_to_sle64(old_data_size); |
| 1679 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1680 | write_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1681 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1682 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1683 | ntfs_attr_put_search_ctx(ctx); |
| 1684 | unmap_mft_record(mft_ni); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1685 | #ifdef DEBUG |
| 1686 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | ntfs_debug("Restored status of mftbmp: allocated_size 0x%llx, " |
| 1688 | "data_size 0x%llx, initialized_size 0x%llx.", |
| 1689 | (long long)mftbmp_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1690 | (long long)i_size_read(mftbmp_vi), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1691 | (long long)mftbmp_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1692 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
| 1693 | #endif /* DEBUG */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1694 | err_out: |
| 1695 | return ret; |
| 1696 | } |
| 1697 | |
| 1698 | /** |
| 1699 | * ntfs_mft_data_extend_allocation_nolock - extend mft data attribute |
| 1700 | * @vol: volume on which to extend the mft data attribute |
| 1701 | * |
| 1702 | * Extend the mft data attribute on the ntfs volume @vol by 16 mft records |
| 1703 | * worth of clusters or if not enough space for this by one mft record worth |
| 1704 | * of clusters. |
| 1705 | * |
| 1706 | * Note: Only changes allocated_size, i.e. does not touch initialized_size or |
| 1707 | * data_size. |
| 1708 | * |
| 1709 | * Return 0 on success and -errno on error. |
| 1710 | * |
| 1711 | * Locking: - Caller must hold vol->mftbmp_lock for writing. |
| 1712 | * - This function takes NTFS_I(vol->mft_ino)->runlist.lock for |
| 1713 | * writing and releases it before returning. |
| 1714 | * - This function calls functions which take vol->lcnbmp_lock for |
| 1715 | * writing and release it before returning. |
| 1716 | */ |
| 1717 | static int ntfs_mft_data_extend_allocation_nolock(ntfs_volume *vol) |
| 1718 | { |
| 1719 | LCN lcn; |
| 1720 | VCN old_last_vcn; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1721 | s64 min_nr, nr, ll; |
| 1722 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1723 | ntfs_inode *mft_ni; |
| 1724 | runlist_element *rl, *rl2; |
| 1725 | ntfs_attr_search_ctx *ctx = NULL; |
| 1726 | MFT_RECORD *mrec; |
| 1727 | ATTR_RECORD *a = NULL; |
| 1728 | int ret, mp_size; |
| 1729 | u32 old_alen = 0; |
| 1730 | BOOL mp_rebuilt = FALSE; |
| 1731 | |
| 1732 | ntfs_debug("Extending mft data allocation."); |
| 1733 | mft_ni = NTFS_I(vol->mft_ino); |
| 1734 | /* |
| 1735 | * Determine the preferred allocation location, i.e. the last lcn of |
| 1736 | * the mft data attribute. The allocated size of the mft data |
| 1737 | * attribute cannot be zero so we are ok to do this. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1738 | */ |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1739 | down_write(&mft_ni->runlist.lock); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1740 | read_lock_irqsave(&mft_ni->size_lock, flags); |
| 1741 | ll = mft_ni->allocated_size; |
| 1742 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
Anton Altaparmakov | c0c1cc0 | 2005-03-07 21:43:38 +0000 | [diff] [blame] | 1743 | rl = ntfs_attr_find_vcn_nolock(mft_ni, |
Anton Altaparmakov | 69b41e3 | 2005-10-04 14:01:14 +0100 | [diff] [blame] | 1744 | (ll - 1) >> vol->cluster_size_bits, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1745 | if (unlikely(IS_ERR(rl) || !rl->length || rl->lcn < 0)) { |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1746 | up_write(&mft_ni->runlist.lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1747 | ntfs_error(vol->sb, "Failed to determine last allocated " |
| 1748 | "cluster of mft data attribute."); |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1749 | if (!IS_ERR(rl)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1750 | ret = -EIO; |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1751 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1752 | ret = PTR_ERR(rl); |
| 1753 | return ret; |
| 1754 | } |
| 1755 | lcn = rl->lcn + rl->length; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1756 | ntfs_debug("Last lcn of mft data attribute is 0x%llx.", (long long)lcn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | /* Minimum allocation is one mft record worth of clusters. */ |
| 1758 | min_nr = vol->mft_record_size >> vol->cluster_size_bits; |
| 1759 | if (!min_nr) |
| 1760 | min_nr = 1; |
| 1761 | /* Want to allocate 16 mft records worth of clusters. */ |
| 1762 | nr = vol->mft_record_size << 4 >> vol->cluster_size_bits; |
| 1763 | if (!nr) |
| 1764 | nr = min_nr; |
| 1765 | /* Ensure we do not go above 2^32-1 mft records. */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1766 | read_lock_irqsave(&mft_ni->size_lock, flags); |
| 1767 | ll = mft_ni->allocated_size; |
| 1768 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
| 1769 | if (unlikely((ll + (nr << vol->cluster_size_bits)) >> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1770 | vol->mft_record_size_bits >= (1ll << 32))) { |
| 1771 | nr = min_nr; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1772 | if (unlikely((ll + (nr << vol->cluster_size_bits)) >> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | vol->mft_record_size_bits >= (1ll << 32))) { |
| 1774 | ntfs_warning(vol->sb, "Cannot allocate mft record " |
| 1775 | "because the maximum number of inodes " |
| 1776 | "(2^32) has already been reached."); |
| 1777 | up_write(&mft_ni->runlist.lock); |
| 1778 | return -ENOSPC; |
| 1779 | } |
| 1780 | } |
| 1781 | ntfs_debug("Trying mft data allocation with %s cluster count %lli.", |
| 1782 | nr > min_nr ? "default" : "minimal", (long long)nr); |
| 1783 | old_last_vcn = rl[1].vcn; |
| 1784 | do { |
Anton Altaparmakov | fc0fa7d | 2005-10-04 14:36:56 +0100 | [diff] [blame] | 1785 | rl2 = ntfs_cluster_alloc(vol, old_last_vcn, nr, lcn, MFT_ZONE, |
| 1786 | TRUE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1787 | if (likely(!IS_ERR(rl2))) |
| 1788 | break; |
| 1789 | if (PTR_ERR(rl2) != -ENOSPC || nr == min_nr) { |
| 1790 | ntfs_error(vol->sb, "Failed to allocate the minimal " |
| 1791 | "number of clusters (%lli) for the " |
| 1792 | "mft data attribute.", (long long)nr); |
| 1793 | up_write(&mft_ni->runlist.lock); |
| 1794 | return PTR_ERR(rl2); |
| 1795 | } |
| 1796 | /* |
| 1797 | * There is not enough space to do the allocation, but there |
| 1798 | * might be enough space to do a minimal allocation so try that |
| 1799 | * before failing. |
| 1800 | */ |
| 1801 | nr = min_nr; |
| 1802 | ntfs_debug("Retrying mft data allocation with minimal cluster " |
| 1803 | "count %lli.", (long long)nr); |
| 1804 | } while (1); |
| 1805 | rl = ntfs_runlists_merge(mft_ni->runlist.rl, rl2); |
| 1806 | if (IS_ERR(rl)) { |
| 1807 | up_write(&mft_ni->runlist.lock); |
| 1808 | ntfs_error(vol->sb, "Failed to merge runlists for mft data " |
| 1809 | "attribute."); |
| 1810 | if (ntfs_cluster_free_from_rl(vol, rl2)) { |
| 1811 | ntfs_error(vol->sb, "Failed to dealocate clusters " |
| 1812 | "from the mft data attribute.%s", es); |
| 1813 | NVolSetErrors(vol); |
| 1814 | } |
| 1815 | ntfs_free(rl2); |
| 1816 | return PTR_ERR(rl); |
| 1817 | } |
| 1818 | mft_ni->runlist.rl = rl; |
Randy Dunlap | 8907547 | 2005-03-03 11:19:53 +0000 | [diff] [blame] | 1819 | ntfs_debug("Allocated %lli clusters.", (long long)nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | /* Find the last run in the new runlist. */ |
| 1821 | for (; rl[1].length; rl++) |
| 1822 | ; |
| 1823 | /* Update the attribute record as well. */ |
| 1824 | mrec = map_mft_record(mft_ni); |
| 1825 | if (IS_ERR(mrec)) { |
| 1826 | ntfs_error(vol->sb, "Failed to map mft record."); |
| 1827 | ret = PTR_ERR(mrec); |
| 1828 | goto undo_alloc; |
| 1829 | } |
| 1830 | ctx = ntfs_attr_get_search_ctx(mft_ni, mrec); |
| 1831 | if (unlikely(!ctx)) { |
| 1832 | ntfs_error(vol->sb, "Failed to get search context."); |
| 1833 | ret = -ENOMEM; |
| 1834 | goto undo_alloc; |
| 1835 | } |
| 1836 | ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, |
| 1837 | CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx); |
| 1838 | if (unlikely(ret)) { |
| 1839 | ntfs_error(vol->sb, "Failed to find last attribute extent of " |
| 1840 | "mft data attribute."); |
| 1841 | if (ret == -ENOENT) |
| 1842 | ret = -EIO; |
| 1843 | goto undo_alloc; |
| 1844 | } |
| 1845 | a = ctx->attr; |
| 1846 | ll = sle64_to_cpu(a->data.non_resident.lowest_vcn); |
| 1847 | /* Search back for the previous last allocated cluster of mft bitmap. */ |
| 1848 | for (rl2 = rl; rl2 > mft_ni->runlist.rl; rl2--) { |
| 1849 | if (ll >= rl2->vcn) |
| 1850 | break; |
| 1851 | } |
| 1852 | BUG_ON(ll < rl2->vcn); |
| 1853 | BUG_ON(ll >= rl2->vcn + rl2->length); |
| 1854 | /* Get the size for the new mapping pairs array for this extent. */ |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1855 | mp_size = ntfs_get_size_for_mapping_pairs(vol, rl2, ll, -1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1856 | if (unlikely(mp_size <= 0)) { |
| 1857 | ntfs_error(vol->sb, "Get size for mapping pairs failed for " |
| 1858 | "mft data attribute extent."); |
| 1859 | ret = mp_size; |
| 1860 | if (!ret) |
| 1861 | ret = -EIO; |
| 1862 | goto undo_alloc; |
| 1863 | } |
| 1864 | /* Expand the attribute record if necessary. */ |
| 1865 | old_alen = le32_to_cpu(a->length); |
| 1866 | ret = ntfs_attr_record_resize(ctx->mrec, a, mp_size + |
| 1867 | le16_to_cpu(a->data.non_resident.mapping_pairs_offset)); |
| 1868 | if (unlikely(ret)) { |
| 1869 | if (ret != -ENOSPC) { |
| 1870 | ntfs_error(vol->sb, "Failed to resize attribute " |
| 1871 | "record for mft data attribute."); |
| 1872 | goto undo_alloc; |
| 1873 | } |
| 1874 | // TODO: Deal with this by moving this extent to a new mft |
| 1875 | // record or by starting a new extent in a new mft record or by |
| 1876 | // moving other attributes out of this mft record. |
| 1877 | // Note: Use the special reserved mft records and ensure that |
| 1878 | // this extent is not required to find the mft record in |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 1879 | // question. If no free special records left we would need to |
| 1880 | // move an existing record away, insert ours in its place, and |
| 1881 | // then place the moved record into the newly allocated space |
| 1882 | // and we would then need to update all references to this mft |
| 1883 | // record appropriately. This is rather complicated... |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | ntfs_error(vol->sb, "Not enough space in this mft record to " |
| 1885 | "accomodate extended mft data attribute " |
| 1886 | "extent. Cannot handle this yet."); |
| 1887 | ret = -EOPNOTSUPP; |
| 1888 | goto undo_alloc; |
| 1889 | } |
| 1890 | mp_rebuilt = TRUE; |
| 1891 | /* Generate the mapping pairs array directly into the attr record. */ |
| 1892 | ret = ntfs_mapping_pairs_build(vol, (u8*)a + |
| 1893 | le16_to_cpu(a->data.non_resident.mapping_pairs_offset), |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1894 | mp_size, rl2, ll, -1, NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | if (unlikely(ret)) { |
| 1896 | ntfs_error(vol->sb, "Failed to build mapping pairs array of " |
| 1897 | "mft data attribute."); |
| 1898 | goto undo_alloc; |
| 1899 | } |
| 1900 | /* Update the highest_vcn. */ |
| 1901 | a->data.non_resident.highest_vcn = cpu_to_sle64(rl[1].vcn - 1); |
| 1902 | /* |
| 1903 | * We now have extended the mft data allocated_size by nr clusters. |
| 1904 | * Reflect this in the ntfs_inode structure and the attribute record. |
| 1905 | * @rl is the last (non-terminator) runlist element of mft data |
| 1906 | * attribute. |
| 1907 | */ |
| 1908 | if (a->data.non_resident.lowest_vcn) { |
| 1909 | /* |
| 1910 | * We are not in the first attribute extent, switch to it, but |
| 1911 | * first ensure the changes will make it to disk later. |
| 1912 | */ |
| 1913 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1914 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1915 | ntfs_attr_reinit_search_ctx(ctx); |
| 1916 | ret = ntfs_attr_lookup(mft_ni->type, mft_ni->name, |
| 1917 | mft_ni->name_len, CASE_SENSITIVE, 0, NULL, 0, |
| 1918 | ctx); |
| 1919 | if (unlikely(ret)) { |
| 1920 | ntfs_error(vol->sb, "Failed to find first attribute " |
| 1921 | "extent of mft data attribute."); |
| 1922 | goto restore_undo_alloc; |
| 1923 | } |
| 1924 | a = ctx->attr; |
| 1925 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1926 | write_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1927 | mft_ni->allocated_size += nr << vol->cluster_size_bits; |
| 1928 | a->data.non_resident.allocated_size = |
| 1929 | cpu_to_sle64(mft_ni->allocated_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1930 | write_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1931 | /* Ensure the changes make it to disk. */ |
| 1932 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1933 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 1934 | ntfs_attr_put_search_ctx(ctx); |
| 1935 | unmap_mft_record(mft_ni); |
| 1936 | up_write(&mft_ni->runlist.lock); |
| 1937 | ntfs_debug("Done."); |
| 1938 | return 0; |
| 1939 | restore_undo_alloc: |
| 1940 | ntfs_attr_reinit_search_ctx(ctx); |
| 1941 | if (ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, |
| 1942 | CASE_SENSITIVE, rl[1].vcn, NULL, 0, ctx)) { |
| 1943 | ntfs_error(vol->sb, "Failed to find last attribute extent of " |
| 1944 | "mft data attribute.%s", es); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1945 | write_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | mft_ni->allocated_size += nr << vol->cluster_size_bits; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 1947 | write_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1948 | ntfs_attr_put_search_ctx(ctx); |
| 1949 | unmap_mft_record(mft_ni); |
| 1950 | up_write(&mft_ni->runlist.lock); |
| 1951 | /* |
| 1952 | * The only thing that is now wrong is ->allocated_size of the |
| 1953 | * base attribute extent which chkdsk should be able to fix. |
| 1954 | */ |
| 1955 | NVolSetErrors(vol); |
| 1956 | return ret; |
| 1957 | } |
Anton Altaparmakov | 511bea5 | 2005-10-04 14:24:21 +0100 | [diff] [blame] | 1958 | ctx->attr->data.non_resident.highest_vcn = |
| 1959 | cpu_to_sle64(old_last_vcn - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1960 | undo_alloc: |
Anton Altaparmakov | 511bea5 | 2005-10-04 14:24:21 +0100 | [diff] [blame] | 1961 | if (ntfs_cluster_free(mft_ni, old_last_vcn, -1, ctx) < 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1962 | ntfs_error(vol->sb, "Failed to free clusters from mft data " |
| 1963 | "attribute.%s", es); |
| 1964 | NVolSetErrors(vol); |
| 1965 | } |
Anton Altaparmakov | 511bea5 | 2005-10-04 14:24:21 +0100 | [diff] [blame] | 1966 | a = ctx->attr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | if (ntfs_rl_truncate_nolock(vol, &mft_ni->runlist, old_last_vcn)) { |
| 1968 | ntfs_error(vol->sb, "Failed to truncate mft data attribute " |
| 1969 | "runlist.%s", es); |
| 1970 | NVolSetErrors(vol); |
| 1971 | } |
Anton Altaparmakov | 511bea5 | 2005-10-04 14:24:21 +0100 | [diff] [blame] | 1972 | if (mp_rebuilt && !IS_ERR(ctx->mrec)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1973 | if (ntfs_mapping_pairs_build(vol, (u8*)a + le16_to_cpu( |
| 1974 | a->data.non_resident.mapping_pairs_offset), |
| 1975 | old_alen - le16_to_cpu( |
| 1976 | a->data.non_resident.mapping_pairs_offset), |
Anton Altaparmakov | fa3be92 | 2005-06-25 17:15:36 +0100 | [diff] [blame] | 1977 | rl2, ll, -1, NULL)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1978 | ntfs_error(vol->sb, "Failed to restore mapping pairs " |
| 1979 | "array.%s", es); |
| 1980 | NVolSetErrors(vol); |
| 1981 | } |
| 1982 | if (ntfs_attr_record_resize(ctx->mrec, a, old_alen)) { |
| 1983 | ntfs_error(vol->sb, "Failed to restore attribute " |
| 1984 | "record.%s", es); |
| 1985 | NVolSetErrors(vol); |
| 1986 | } |
| 1987 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 1988 | mark_mft_record_dirty(ctx->ntfs_ino); |
Anton Altaparmakov | 511bea5 | 2005-10-04 14:24:21 +0100 | [diff] [blame] | 1989 | } else if (IS_ERR(ctx->mrec)) { |
| 1990 | ntfs_error(vol->sb, "Failed to restore attribute search " |
| 1991 | "context.%s", es); |
| 1992 | NVolSetErrors(vol); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1993 | } |
| 1994 | if (ctx) |
| 1995 | ntfs_attr_put_search_ctx(ctx); |
| 1996 | if (!IS_ERR(mrec)) |
| 1997 | unmap_mft_record(mft_ni); |
| 1998 | up_write(&mft_ni->runlist.lock); |
| 1999 | return ret; |
| 2000 | } |
| 2001 | |
| 2002 | /** |
| 2003 | * ntfs_mft_record_layout - layout an mft record into a memory buffer |
| 2004 | * @vol: volume to which the mft record will belong |
| 2005 | * @mft_no: mft reference specifying the mft record number |
| 2006 | * @m: destination buffer of size >= @vol->mft_record_size bytes |
| 2007 | * |
| 2008 | * Layout an empty, unused mft record with the mft record number @mft_no into |
| 2009 | * the buffer @m. The volume @vol is needed because the mft record structure |
| 2010 | * was modified in NTFS 3.1 so we need to know which volume version this mft |
| 2011 | * record will be used on. |
| 2012 | * |
| 2013 | * Return 0 on success and -errno on error. |
| 2014 | */ |
| 2015 | static int ntfs_mft_record_layout(const ntfs_volume *vol, const s64 mft_no, |
| 2016 | MFT_RECORD *m) |
| 2017 | { |
| 2018 | ATTR_RECORD *a; |
| 2019 | |
| 2020 | ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); |
| 2021 | if (mft_no >= (1ll << 32)) { |
| 2022 | ntfs_error(vol->sb, "Mft record number 0x%llx exceeds " |
| 2023 | "maximum of 2^32.", (long long)mft_no); |
| 2024 | return -ERANGE; |
| 2025 | } |
| 2026 | /* Start by clearing the whole mft record to gives us a clean slate. */ |
| 2027 | memset(m, 0, vol->mft_record_size); |
| 2028 | /* Aligned to 2-byte boundary. */ |
| 2029 | if (vol->major_ver < 3 || (vol->major_ver == 3 && !vol->minor_ver)) |
| 2030 | m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD_OLD) + 1) & ~1); |
| 2031 | else { |
| 2032 | m->usa_ofs = cpu_to_le16((sizeof(MFT_RECORD) + 1) & ~1); |
| 2033 | /* |
| 2034 | * Set the NTFS 3.1+ specific fields while we know that the |
| 2035 | * volume version is 3.1+. |
| 2036 | */ |
| 2037 | m->reserved = 0; |
| 2038 | m->mft_record_number = cpu_to_le32((u32)mft_no); |
| 2039 | } |
| 2040 | m->magic = magic_FILE; |
| 2041 | if (vol->mft_record_size >= NTFS_BLOCK_SIZE) |
| 2042 | m->usa_count = cpu_to_le16(vol->mft_record_size / |
| 2043 | NTFS_BLOCK_SIZE + 1); |
| 2044 | else { |
| 2045 | m->usa_count = cpu_to_le16(1); |
| 2046 | ntfs_warning(vol->sb, "Sector size is bigger than mft record " |
| 2047 | "size. Setting usa_count to 1. If chkdsk " |
| 2048 | "reports this as corruption, please email " |
| 2049 | "linux-ntfs-dev@lists.sourceforge.net stating " |
| 2050 | "that you saw this message and that the " |
Anton Altaparmakov | b6ad6c5 | 2005-02-15 10:08:43 +0000 | [diff] [blame] | 2051 | "modified filesystem created was corrupt. " |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | "Thank you."); |
| 2053 | } |
| 2054 | /* Set the update sequence number to 1. */ |
| 2055 | *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = cpu_to_le16(1); |
| 2056 | m->lsn = 0; |
| 2057 | m->sequence_number = cpu_to_le16(1); |
| 2058 | m->link_count = 0; |
| 2059 | /* |
| 2060 | * Place the attributes straight after the update sequence array, |
| 2061 | * aligned to 8-byte boundary. |
| 2062 | */ |
| 2063 | m->attrs_offset = cpu_to_le16((le16_to_cpu(m->usa_ofs) + |
| 2064 | (le16_to_cpu(m->usa_count) << 1) + 7) & ~7); |
| 2065 | m->flags = 0; |
| 2066 | /* |
| 2067 | * Using attrs_offset plus eight bytes (for the termination attribute). |
| 2068 | * attrs_offset is already aligned to 8-byte boundary, so no need to |
| 2069 | * align again. |
| 2070 | */ |
| 2071 | m->bytes_in_use = cpu_to_le32(le16_to_cpu(m->attrs_offset) + 8); |
| 2072 | m->bytes_allocated = cpu_to_le32(vol->mft_record_size); |
| 2073 | m->base_mft_record = 0; |
| 2074 | m->next_attr_instance = 0; |
| 2075 | /* Add the termination attribute. */ |
| 2076 | a = (ATTR_RECORD*)((u8*)m + le16_to_cpu(m->attrs_offset)); |
| 2077 | a->type = AT_END; |
| 2078 | a->length = 0; |
| 2079 | ntfs_debug("Done."); |
| 2080 | return 0; |
| 2081 | } |
| 2082 | |
| 2083 | /** |
| 2084 | * ntfs_mft_record_format - format an mft record on an ntfs volume |
| 2085 | * @vol: volume on which to format the mft record |
| 2086 | * @mft_no: mft record number to format |
| 2087 | * |
| 2088 | * Format the mft record @mft_no in $MFT/$DATA, i.e. lay out an empty, unused |
| 2089 | * mft record into the appropriate place of the mft data attribute. This is |
| 2090 | * used when extending the mft data attribute. |
| 2091 | * |
| 2092 | * Return 0 on success and -errno on error. |
| 2093 | */ |
| 2094 | static int ntfs_mft_record_format(const ntfs_volume *vol, const s64 mft_no) |
| 2095 | { |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2096 | loff_t i_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2097 | struct inode *mft_vi = vol->mft_ino; |
| 2098 | struct page *page; |
| 2099 | MFT_RECORD *m; |
| 2100 | pgoff_t index, end_index; |
| 2101 | unsigned int ofs; |
| 2102 | int err; |
| 2103 | |
| 2104 | ntfs_debug("Entering for mft record 0x%llx.", (long long)mft_no); |
| 2105 | /* |
| 2106 | * The index into the page cache and the offset within the page cache |
| 2107 | * page of the wanted mft record. |
| 2108 | */ |
| 2109 | index = mft_no << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; |
| 2110 | ofs = (mft_no << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; |
| 2111 | /* The maximum valid index into the page cache for $MFT's data. */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2112 | i_size = i_size_read(mft_vi); |
| 2113 | end_index = i_size >> PAGE_CACHE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | if (unlikely(index >= end_index)) { |
| 2115 | if (unlikely(index > end_index || ofs + vol->mft_record_size >= |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2116 | (i_size & ~PAGE_CACHE_MASK))) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2117 | ntfs_error(vol->sb, "Tried to format non-existing mft " |
| 2118 | "record 0x%llx.", (long long)mft_no); |
| 2119 | return -ENOENT; |
| 2120 | } |
| 2121 | } |
| 2122 | /* Read, map, and pin the page containing the mft record. */ |
| 2123 | page = ntfs_map_page(mft_vi->i_mapping, index); |
| 2124 | if (unlikely(IS_ERR(page))) { |
| 2125 | ntfs_error(vol->sb, "Failed to map page containing mft record " |
| 2126 | "to format 0x%llx.", (long long)mft_no); |
| 2127 | return PTR_ERR(page); |
| 2128 | } |
| 2129 | lock_page(page); |
| 2130 | BUG_ON(!PageUptodate(page)); |
| 2131 | ClearPageUptodate(page); |
| 2132 | m = (MFT_RECORD*)((u8*)page_address(page) + ofs); |
| 2133 | err = ntfs_mft_record_layout(vol, mft_no, m); |
| 2134 | if (unlikely(err)) { |
| 2135 | ntfs_error(vol->sb, "Failed to layout mft record 0x%llx.", |
| 2136 | (long long)mft_no); |
| 2137 | SetPageUptodate(page); |
| 2138 | unlock_page(page); |
| 2139 | ntfs_unmap_page(page); |
| 2140 | return err; |
| 2141 | } |
| 2142 | flush_dcache_page(page); |
| 2143 | SetPageUptodate(page); |
| 2144 | unlock_page(page); |
| 2145 | /* |
| 2146 | * Make sure the mft record is written out to disk. We could use |
| 2147 | * ilookup5() to check if an inode is in icache and so on but this is |
| 2148 | * unnecessary as ntfs_writepage() will write the dirty record anyway. |
| 2149 | */ |
| 2150 | mark_ntfs_record_dirty(page, ofs); |
| 2151 | ntfs_unmap_page(page); |
| 2152 | ntfs_debug("Done."); |
| 2153 | return 0; |
| 2154 | } |
| 2155 | |
| 2156 | /** |
| 2157 | * ntfs_mft_record_alloc - allocate an mft record on an ntfs volume |
| 2158 | * @vol: [IN] volume on which to allocate the mft record |
| 2159 | * @mode: [IN] mode if want a file or directory, i.e. base inode or 0 |
| 2160 | * @base_ni: [IN] open base inode if allocating an extent mft record or NULL |
| 2161 | * @mrec: [OUT] on successful return this is the mapped mft record |
| 2162 | * |
| 2163 | * Allocate an mft record in $MFT/$DATA of an open ntfs volume @vol. |
| 2164 | * |
| 2165 | * If @base_ni is NULL make the mft record a base mft record, i.e. a file or |
| 2166 | * direvctory inode, and allocate it at the default allocator position. In |
| 2167 | * this case @mode is the file mode as given to us by the caller. We in |
| 2168 | * particular use @mode to distinguish whether a file or a directory is being |
| 2169 | * created (S_IFDIR(mode) and S_IFREG(mode), respectively). |
| 2170 | * |
| 2171 | * If @base_ni is not NULL make the allocated mft record an extent record, |
| 2172 | * allocate it starting at the mft record after the base mft record and attach |
| 2173 | * the allocated and opened ntfs inode to the base inode @base_ni. In this |
| 2174 | * case @mode must be 0 as it is meaningless for extent inodes. |
| 2175 | * |
| 2176 | * You need to check the return value with IS_ERR(). If false, the function |
| 2177 | * was successful and the return value is the now opened ntfs inode of the |
| 2178 | * allocated mft record. *@mrec is then set to the allocated, mapped, pinned, |
| 2179 | * and locked mft record. If IS_ERR() is true, the function failed and the |
| 2180 | * error code is obtained from PTR_ERR(return value). *@mrec is undefined in |
| 2181 | * this case. |
| 2182 | * |
| 2183 | * Allocation strategy: |
| 2184 | * |
| 2185 | * To find a free mft record, we scan the mft bitmap for a zero bit. To |
| 2186 | * optimize this we start scanning at the place specified by @base_ni or if |
| 2187 | * @base_ni is NULL we start where we last stopped and we perform wrap around |
| 2188 | * when we reach the end. Note, we do not try to allocate mft records below |
| 2189 | * number 24 because numbers 0 to 15 are the defined system files anyway and 16 |
| 2190 | * to 24 are special in that they are used for storing extension mft records |
| 2191 | * for the $DATA attribute of $MFT. This is required to avoid the possibility |
| 2192 | * of creating a runlist with a circular dependency which once written to disk |
| 2193 | * can never be read in again. Windows will only use records 16 to 24 for |
| 2194 | * normal files if the volume is completely out of space. We never use them |
| 2195 | * which means that when the volume is really out of space we cannot create any |
| 2196 | * more files while Windows can still create up to 8 small files. We can start |
| 2197 | * doing this at some later time, it does not matter much for now. |
| 2198 | * |
| 2199 | * When scanning the mft bitmap, we only search up to the last allocated mft |
| 2200 | * record. If there are no free records left in the range 24 to number of |
| 2201 | * allocated mft records, then we extend the $MFT/$DATA attribute in order to |
| 2202 | * create free mft records. We extend the allocated size of $MFT/$DATA by 16 |
| 2203 | * records at a time or one cluster, if cluster size is above 16kiB. If there |
| 2204 | * is not sufficient space to do this, we try to extend by a single mft record |
| 2205 | * or one cluster, if cluster size is above the mft record size. |
| 2206 | * |
| 2207 | * No matter how many mft records we allocate, we initialize only the first |
| 2208 | * allocated mft record, incrementing mft data size and initialized size |
| 2209 | * accordingly, open an ntfs_inode for it and return it to the caller, unless |
| 2210 | * there are less than 24 mft records, in which case we allocate and initialize |
| 2211 | * mft records until we reach record 24 which we consider as the first free mft |
| 2212 | * record for use by normal files. |
| 2213 | * |
| 2214 | * If during any stage we overflow the initialized data in the mft bitmap, we |
| 2215 | * extend the initialized size (and data size) by 8 bytes, allocating another |
| 2216 | * cluster if required. The bitmap data size has to be at least equal to the |
| 2217 | * number of mft records in the mft, but it can be bigger, in which case the |
| 2218 | * superflous bits are padded with zeroes. |
| 2219 | * |
| 2220 | * Thus, when we return successfully (IS_ERR() is false), we will have: |
| 2221 | * - initialized / extended the mft bitmap if necessary, |
| 2222 | * - initialized / extended the mft data if necessary, |
| 2223 | * - set the bit corresponding to the mft record being allocated in the |
| 2224 | * mft bitmap, |
| 2225 | * - opened an ntfs_inode for the allocated mft record, and we will have |
| 2226 | * - returned the ntfs_inode as well as the allocated mapped, pinned, and |
| 2227 | * locked mft record. |
| 2228 | * |
| 2229 | * On error, the volume will be left in a consistent state and no record will |
| 2230 | * be allocated. If rolling back a partial operation fails, we may leave some |
| 2231 | * inconsistent metadata in which case we set NVolErrors() so the volume is |
| 2232 | * left dirty when unmounted. |
| 2233 | * |
| 2234 | * Note, this function cannot make use of most of the normal functions, like |
| 2235 | * for example for attribute resizing, etc, because when the run list overflows |
| 2236 | * the base mft record and an attribute list is used, it is very important that |
| 2237 | * the extension mft records used to store the $DATA attribute of $MFT can be |
| 2238 | * reached without having to read the information contained inside them, as |
| 2239 | * this would make it impossible to find them in the first place after the |
| 2240 | * volume is unmounted. $MFT/$BITMAP probably does not need to follow this |
| 2241 | * rule because the bitmap is not essential for finding the mft records, but on |
| 2242 | * the other hand, handling the bitmap in this special way would make life |
| 2243 | * easier because otherwise there might be circular invocations of functions |
| 2244 | * when reading the bitmap. |
| 2245 | */ |
| 2246 | ntfs_inode *ntfs_mft_record_alloc(ntfs_volume *vol, const int mode, |
| 2247 | ntfs_inode *base_ni, MFT_RECORD **mrec) |
| 2248 | { |
| 2249 | s64 ll, bit, old_data_initialized, old_data_size; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2250 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2251 | struct inode *vi; |
| 2252 | struct page *page; |
| 2253 | ntfs_inode *mft_ni, *mftbmp_ni, *ni; |
| 2254 | ntfs_attr_search_ctx *ctx; |
| 2255 | MFT_RECORD *m; |
| 2256 | ATTR_RECORD *a; |
| 2257 | pgoff_t index; |
| 2258 | unsigned int ofs; |
| 2259 | int err; |
| 2260 | le16 seq_no, usn; |
| 2261 | BOOL record_formatted = FALSE; |
| 2262 | |
| 2263 | if (base_ni) { |
| 2264 | ntfs_debug("Entering (allocating an extent mft record for " |
| 2265 | "base mft record 0x%llx).", |
| 2266 | (long long)base_ni->mft_no); |
| 2267 | /* @mode and @base_ni are mutually exclusive. */ |
| 2268 | BUG_ON(mode); |
| 2269 | } else |
| 2270 | ntfs_debug("Entering (allocating a base mft record)."); |
| 2271 | if (mode) { |
| 2272 | /* @mode and @base_ni are mutually exclusive. */ |
| 2273 | BUG_ON(base_ni); |
| 2274 | /* We only support creation of normal files and directories. */ |
| 2275 | if (!S_ISREG(mode) && !S_ISDIR(mode)) |
| 2276 | return ERR_PTR(-EOPNOTSUPP); |
| 2277 | } |
| 2278 | BUG_ON(!mrec); |
| 2279 | mft_ni = NTFS_I(vol->mft_ino); |
| 2280 | mftbmp_ni = NTFS_I(vol->mftbmp_ino); |
| 2281 | down_write(&vol->mftbmp_lock); |
| 2282 | bit = ntfs_mft_bitmap_find_and_alloc_free_rec_nolock(vol, base_ni); |
| 2283 | if (bit >= 0) { |
| 2284 | ntfs_debug("Found and allocated free record (#1), bit 0x%llx.", |
| 2285 | (long long)bit); |
| 2286 | goto have_alloc_rec; |
| 2287 | } |
| 2288 | if (bit != -ENOSPC) { |
| 2289 | up_write(&vol->mftbmp_lock); |
| 2290 | return ERR_PTR(bit); |
| 2291 | } |
| 2292 | /* |
| 2293 | * No free mft records left. If the mft bitmap already covers more |
| 2294 | * than the currently used mft records, the next records are all free, |
| 2295 | * so we can simply allocate the first unused mft record. |
| 2296 | * Note: We also have to make sure that the mft bitmap at least covers |
| 2297 | * the first 24 mft records as they are special and whilst they may not |
| 2298 | * be in use, we do not allocate from them. |
| 2299 | */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2300 | read_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2301 | ll = mft_ni->initialized_size >> vol->mft_record_size_bits; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2302 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
| 2303 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
| 2304 | old_data_initialized = mftbmp_ni->initialized_size; |
| 2305 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
| 2306 | if (old_data_initialized << 3 > ll && old_data_initialized > 3) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2307 | bit = ll; |
| 2308 | if (bit < 24) |
| 2309 | bit = 24; |
| 2310 | if (unlikely(bit >= (1ll << 32))) |
| 2311 | goto max_err_out; |
| 2312 | ntfs_debug("Found free record (#2), bit 0x%llx.", |
| 2313 | (long long)bit); |
| 2314 | goto found_free_rec; |
| 2315 | } |
| 2316 | /* |
| 2317 | * The mft bitmap needs to be expanded until it covers the first unused |
| 2318 | * mft record that we can allocate. |
| 2319 | * Note: The smallest mft record we allocate is mft record 24. |
| 2320 | */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2321 | bit = old_data_initialized << 3; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2322 | if (unlikely(bit >= (1ll << 32))) |
| 2323 | goto max_err_out; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2324 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
| 2325 | old_data_size = mftbmp_ni->allocated_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2326 | ntfs_debug("Status of mftbmp before extension: allocated_size 0x%llx, " |
| 2327 | "data_size 0x%llx, initialized_size 0x%llx.", |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2328 | (long long)old_data_size, |
| 2329 | (long long)i_size_read(vol->mftbmp_ino), |
| 2330 | (long long)old_data_initialized); |
| 2331 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
| 2332 | if (old_data_initialized + 8 > old_data_size) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2333 | /* Need to extend bitmap by one more cluster. */ |
| 2334 | ntfs_debug("mftbmp: initialized_size + 8 > allocated_size."); |
| 2335 | err = ntfs_mft_bitmap_extend_allocation_nolock(vol); |
| 2336 | if (unlikely(err)) { |
| 2337 | up_write(&vol->mftbmp_lock); |
| 2338 | goto err_out; |
| 2339 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2340 | #ifdef DEBUG |
| 2341 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2342 | ntfs_debug("Status of mftbmp after allocation extension: " |
| 2343 | "allocated_size 0x%llx, data_size 0x%llx, " |
| 2344 | "initialized_size 0x%llx.", |
| 2345 | (long long)mftbmp_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2346 | (long long)i_size_read(vol->mftbmp_ino), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2347 | (long long)mftbmp_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2348 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
| 2349 | #endif /* DEBUG */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2350 | } |
| 2351 | /* |
| 2352 | * We now have sufficient allocated space, extend the initialized_size |
| 2353 | * as well as the data_size if necessary and fill the new space with |
| 2354 | * zeroes. |
| 2355 | */ |
| 2356 | err = ntfs_mft_bitmap_extend_initialized_nolock(vol); |
| 2357 | if (unlikely(err)) { |
| 2358 | up_write(&vol->mftbmp_lock); |
| 2359 | goto err_out; |
| 2360 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2361 | #ifdef DEBUG |
| 2362 | read_lock_irqsave(&mftbmp_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2363 | ntfs_debug("Status of mftbmp after initialized extention: " |
| 2364 | "allocated_size 0x%llx, data_size 0x%llx, " |
| 2365 | "initialized_size 0x%llx.", |
| 2366 | (long long)mftbmp_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2367 | (long long)i_size_read(vol->mftbmp_ino), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2368 | (long long)mftbmp_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2369 | read_unlock_irqrestore(&mftbmp_ni->size_lock, flags); |
| 2370 | #endif /* DEBUG */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2371 | ntfs_debug("Found free record (#3), bit 0x%llx.", (long long)bit); |
| 2372 | found_free_rec: |
| 2373 | /* @bit is the found free mft record, allocate it in the mft bitmap. */ |
| 2374 | ntfs_debug("At found_free_rec."); |
| 2375 | err = ntfs_bitmap_set_bit(vol->mftbmp_ino, bit); |
| 2376 | if (unlikely(err)) { |
| 2377 | ntfs_error(vol->sb, "Failed to allocate bit in mft bitmap."); |
| 2378 | up_write(&vol->mftbmp_lock); |
| 2379 | goto err_out; |
| 2380 | } |
| 2381 | ntfs_debug("Set bit 0x%llx in mft bitmap.", (long long)bit); |
| 2382 | have_alloc_rec: |
| 2383 | /* |
| 2384 | * The mft bitmap is now uptodate. Deal with mft data attribute now. |
| 2385 | * Note, we keep hold of the mft bitmap lock for writing until all |
| 2386 | * modifications to the mft data attribute are complete, too, as they |
| 2387 | * will impact decisions for mft bitmap and mft record allocation done |
| 2388 | * by a parallel allocation and if the lock is not maintained a |
| 2389 | * parallel allocation could allocate the same mft record as this one. |
| 2390 | */ |
| 2391 | ll = (bit + 1) << vol->mft_record_size_bits; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2392 | read_lock_irqsave(&mft_ni->size_lock, flags); |
| 2393 | old_data_initialized = mft_ni->initialized_size; |
| 2394 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
| 2395 | if (ll <= old_data_initialized) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2396 | ntfs_debug("Allocated mft record already initialized."); |
| 2397 | goto mft_rec_already_initialized; |
| 2398 | } |
| 2399 | ntfs_debug("Initializing allocated mft record."); |
| 2400 | /* |
| 2401 | * The mft record is outside the initialized data. Extend the mft data |
| 2402 | * attribute until it covers the allocated record. The loop is only |
| 2403 | * actually traversed more than once when a freshly formatted volume is |
| 2404 | * first written to so it optimizes away nicely in the common case. |
| 2405 | */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2406 | read_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2407 | ntfs_debug("Status of mft data before extension: " |
| 2408 | "allocated_size 0x%llx, data_size 0x%llx, " |
| 2409 | "initialized_size 0x%llx.", |
Anton Altaparmakov | 3834c3f | 2005-01-13 11:04:39 +0000 | [diff] [blame] | 2410 | (long long)mft_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2411 | (long long)i_size_read(vol->mft_ino), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2412 | (long long)mft_ni->initialized_size); |
Anton Altaparmakov | 3834c3f | 2005-01-13 11:04:39 +0000 | [diff] [blame] | 2413 | while (ll > mft_ni->allocated_size) { |
| 2414 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2415 | err = ntfs_mft_data_extend_allocation_nolock(vol); |
| 2416 | if (unlikely(err)) { |
| 2417 | ntfs_error(vol->sb, "Failed to extend mft data " |
| 2418 | "allocation."); |
| 2419 | goto undo_mftbmp_alloc_nolock; |
| 2420 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2421 | read_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2422 | ntfs_debug("Status of mft data after allocation extension: " |
| 2423 | "allocated_size 0x%llx, data_size 0x%llx, " |
| 2424 | "initialized_size 0x%llx.", |
| 2425 | (long long)mft_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2426 | (long long)i_size_read(vol->mft_ino), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2427 | (long long)mft_ni->initialized_size); |
| 2428 | } |
Anton Altaparmakov | 3834c3f | 2005-01-13 11:04:39 +0000 | [diff] [blame] | 2429 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2430 | /* |
| 2431 | * Extend mft data initialized size (and data size of course) to reach |
| 2432 | * the allocated mft record, formatting the mft records allong the way. |
| 2433 | * Note: We only modify the ntfs_inode structure as that is all that is |
| 2434 | * needed by ntfs_mft_record_format(). We will update the attribute |
| 2435 | * record itself in one fell swoop later on. |
| 2436 | */ |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2437 | write_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2438 | old_data_initialized = mft_ni->initialized_size; |
| 2439 | old_data_size = vol->mft_ino->i_size; |
| 2440 | while (ll > mft_ni->initialized_size) { |
| 2441 | s64 new_initialized_size, mft_no; |
| 2442 | |
| 2443 | new_initialized_size = mft_ni->initialized_size + |
| 2444 | vol->mft_record_size; |
| 2445 | mft_no = mft_ni->initialized_size >> vol->mft_record_size_bits; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2446 | if (new_initialized_size > i_size_read(vol->mft_ino)) |
| 2447 | i_size_write(vol->mft_ino, new_initialized_size); |
| 2448 | write_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2449 | ntfs_debug("Initializing mft record 0x%llx.", |
| 2450 | (long long)mft_no); |
| 2451 | err = ntfs_mft_record_format(vol, mft_no); |
| 2452 | if (unlikely(err)) { |
| 2453 | ntfs_error(vol->sb, "Failed to format mft record."); |
| 2454 | goto undo_data_init; |
| 2455 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2456 | write_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2457 | mft_ni->initialized_size = new_initialized_size; |
| 2458 | } |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2459 | write_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2460 | record_formatted = TRUE; |
| 2461 | /* Update the mft data attribute record to reflect the new sizes. */ |
| 2462 | m = map_mft_record(mft_ni); |
| 2463 | if (IS_ERR(m)) { |
| 2464 | ntfs_error(vol->sb, "Failed to map mft record."); |
| 2465 | err = PTR_ERR(m); |
| 2466 | goto undo_data_init; |
| 2467 | } |
| 2468 | ctx = ntfs_attr_get_search_ctx(mft_ni, m); |
| 2469 | if (unlikely(!ctx)) { |
| 2470 | ntfs_error(vol->sb, "Failed to get search context."); |
| 2471 | err = -ENOMEM; |
| 2472 | unmap_mft_record(mft_ni); |
| 2473 | goto undo_data_init; |
| 2474 | } |
| 2475 | err = ntfs_attr_lookup(mft_ni->type, mft_ni->name, mft_ni->name_len, |
| 2476 | CASE_SENSITIVE, 0, NULL, 0, ctx); |
| 2477 | if (unlikely(err)) { |
| 2478 | ntfs_error(vol->sb, "Failed to find first attribute extent of " |
| 2479 | "mft data attribute."); |
| 2480 | ntfs_attr_put_search_ctx(ctx); |
| 2481 | unmap_mft_record(mft_ni); |
| 2482 | goto undo_data_init; |
| 2483 | } |
| 2484 | a = ctx->attr; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2485 | read_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2486 | a->data.non_resident.initialized_size = |
| 2487 | cpu_to_sle64(mft_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2488 | a->data.non_resident.data_size = |
| 2489 | cpu_to_sle64(i_size_read(vol->mft_ino)); |
| 2490 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2491 | /* Ensure the changes make it to disk. */ |
| 2492 | flush_dcache_mft_record_page(ctx->ntfs_ino); |
| 2493 | mark_mft_record_dirty(ctx->ntfs_ino); |
| 2494 | ntfs_attr_put_search_ctx(ctx); |
| 2495 | unmap_mft_record(mft_ni); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2496 | read_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2497 | ntfs_debug("Status of mft data after mft record initialization: " |
| 2498 | "allocated_size 0x%llx, data_size 0x%llx, " |
| 2499 | "initialized_size 0x%llx.", |
| 2500 | (long long)mft_ni->allocated_size, |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2501 | (long long)i_size_read(vol->mft_ino), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2502 | (long long)mft_ni->initialized_size); |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2503 | BUG_ON(i_size_read(vol->mft_ino) > mft_ni->allocated_size); |
| 2504 | BUG_ON(mft_ni->initialized_size > i_size_read(vol->mft_ino)); |
| 2505 | read_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2506 | mft_rec_already_initialized: |
| 2507 | /* |
| 2508 | * We can finally drop the mft bitmap lock as the mft data attribute |
| 2509 | * has been fully updated. The only disparity left is that the |
| 2510 | * allocated mft record still needs to be marked as in use to match the |
| 2511 | * set bit in the mft bitmap but this is actually not a problem since |
| 2512 | * this mft record is not referenced from anywhere yet and the fact |
| 2513 | * that it is allocated in the mft bitmap means that no-one will try to |
| 2514 | * allocate it either. |
| 2515 | */ |
| 2516 | up_write(&vol->mftbmp_lock); |
| 2517 | /* |
| 2518 | * We now have allocated and initialized the mft record. Calculate the |
| 2519 | * index of and the offset within the page cache page the record is in. |
| 2520 | */ |
| 2521 | index = bit << vol->mft_record_size_bits >> PAGE_CACHE_SHIFT; |
| 2522 | ofs = (bit << vol->mft_record_size_bits) & ~PAGE_CACHE_MASK; |
| 2523 | /* Read, map, and pin the page containing the mft record. */ |
| 2524 | page = ntfs_map_page(vol->mft_ino->i_mapping, index); |
| 2525 | if (unlikely(IS_ERR(page))) { |
| 2526 | ntfs_error(vol->sb, "Failed to map page containing allocated " |
| 2527 | "mft record 0x%llx.", (long long)bit); |
| 2528 | err = PTR_ERR(page); |
| 2529 | goto undo_mftbmp_alloc; |
| 2530 | } |
| 2531 | lock_page(page); |
| 2532 | BUG_ON(!PageUptodate(page)); |
| 2533 | ClearPageUptodate(page); |
| 2534 | m = (MFT_RECORD*)((u8*)page_address(page) + ofs); |
| 2535 | /* If we just formatted the mft record no need to do it again. */ |
| 2536 | if (!record_formatted) { |
| 2537 | /* Sanity check that the mft record is really not in use. */ |
| 2538 | if (ntfs_is_file_record(m->magic) && |
| 2539 | (m->flags & MFT_RECORD_IN_USE)) { |
| 2540 | ntfs_error(vol->sb, "Mft record 0x%llx was marked " |
| 2541 | "free in mft bitmap but is marked " |
| 2542 | "used itself. Corrupt filesystem. " |
| 2543 | "Unmount and run chkdsk.", |
| 2544 | (long long)bit); |
| 2545 | err = -EIO; |
| 2546 | SetPageUptodate(page); |
| 2547 | unlock_page(page); |
| 2548 | ntfs_unmap_page(page); |
| 2549 | NVolSetErrors(vol); |
| 2550 | goto undo_mftbmp_alloc; |
| 2551 | } |
| 2552 | /* |
| 2553 | * We need to (re-)format the mft record, preserving the |
| 2554 | * sequence number if it is not zero as well as the update |
| 2555 | * sequence number if it is not zero or -1 (0xffff). This |
| 2556 | * means we do not need to care whether or not something went |
| 2557 | * wrong with the previous mft record. |
| 2558 | */ |
| 2559 | seq_no = m->sequence_number; |
| 2560 | usn = *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)); |
| 2561 | err = ntfs_mft_record_layout(vol, bit, m); |
| 2562 | if (unlikely(err)) { |
| 2563 | ntfs_error(vol->sb, "Failed to layout allocated mft " |
| 2564 | "record 0x%llx.", (long long)bit); |
| 2565 | SetPageUptodate(page); |
| 2566 | unlock_page(page); |
| 2567 | ntfs_unmap_page(page); |
| 2568 | goto undo_mftbmp_alloc; |
| 2569 | } |
| 2570 | if (seq_no) |
| 2571 | m->sequence_number = seq_no; |
| 2572 | if (usn && le16_to_cpu(usn) != 0xffff) |
| 2573 | *(le16*)((u8*)m + le16_to_cpu(m->usa_ofs)) = usn; |
| 2574 | } |
| 2575 | /* Set the mft record itself in use. */ |
| 2576 | m->flags |= MFT_RECORD_IN_USE; |
| 2577 | if (S_ISDIR(mode)) |
| 2578 | m->flags |= MFT_RECORD_IS_DIRECTORY; |
| 2579 | flush_dcache_page(page); |
| 2580 | SetPageUptodate(page); |
| 2581 | if (base_ni) { |
| 2582 | /* |
| 2583 | * Setup the base mft record in the extent mft record. This |
| 2584 | * completes initialization of the allocated extent mft record |
| 2585 | * and we can simply use it with map_extent_mft_record(). |
| 2586 | */ |
| 2587 | m->base_mft_record = MK_LE_MREF(base_ni->mft_no, |
| 2588 | base_ni->seq_no); |
| 2589 | /* |
| 2590 | * Allocate an extent inode structure for the new mft record, |
| 2591 | * attach it to the base inode @base_ni and map, pin, and lock |
| 2592 | * its, i.e. the allocated, mft record. |
| 2593 | */ |
| 2594 | m = map_extent_mft_record(base_ni, bit, &ni); |
| 2595 | if (IS_ERR(m)) { |
| 2596 | ntfs_error(vol->sb, "Failed to map allocated extent " |
| 2597 | "mft record 0x%llx.", (long long)bit); |
| 2598 | err = PTR_ERR(m); |
| 2599 | /* Set the mft record itself not in use. */ |
| 2600 | m->flags &= cpu_to_le16( |
| 2601 | ~le16_to_cpu(MFT_RECORD_IN_USE)); |
| 2602 | flush_dcache_page(page); |
| 2603 | /* Make sure the mft record is written out to disk. */ |
| 2604 | mark_ntfs_record_dirty(page, ofs); |
| 2605 | unlock_page(page); |
| 2606 | ntfs_unmap_page(page); |
| 2607 | goto undo_mftbmp_alloc; |
| 2608 | } |
| 2609 | /* |
| 2610 | * Make sure the allocated mft record is written out to disk. |
| 2611 | * No need to set the inode dirty because the caller is going |
| 2612 | * to do that anyway after finishing with the new extent mft |
| 2613 | * record (e.g. at a minimum a new attribute will be added to |
| 2614 | * the mft record. |
| 2615 | */ |
| 2616 | mark_ntfs_record_dirty(page, ofs); |
| 2617 | unlock_page(page); |
| 2618 | /* |
| 2619 | * Need to unmap the page since map_extent_mft_record() mapped |
| 2620 | * it as well so we have it mapped twice at the moment. |
| 2621 | */ |
| 2622 | ntfs_unmap_page(page); |
| 2623 | } else { |
| 2624 | /* |
| 2625 | * Allocate a new VFS inode and set it up. NOTE: @vi->i_nlink |
| 2626 | * is set to 1 but the mft record->link_count is 0. The caller |
| 2627 | * needs to bear this in mind. |
| 2628 | */ |
| 2629 | vi = new_inode(vol->sb); |
| 2630 | if (unlikely(!vi)) { |
| 2631 | err = -ENOMEM; |
| 2632 | /* Set the mft record itself not in use. */ |
| 2633 | m->flags &= cpu_to_le16( |
| 2634 | ~le16_to_cpu(MFT_RECORD_IN_USE)); |
| 2635 | flush_dcache_page(page); |
| 2636 | /* Make sure the mft record is written out to disk. */ |
| 2637 | mark_ntfs_record_dirty(page, ofs); |
| 2638 | unlock_page(page); |
| 2639 | ntfs_unmap_page(page); |
| 2640 | goto undo_mftbmp_alloc; |
| 2641 | } |
| 2642 | vi->i_ino = bit; |
| 2643 | /* |
| 2644 | * This is the optimal IO size (for stat), not the fs block |
| 2645 | * size. |
| 2646 | */ |
| 2647 | vi->i_blksize = PAGE_CACHE_SIZE; |
| 2648 | /* |
| 2649 | * This is for checking whether an inode has changed w.r.t. a |
| 2650 | * file so that the file can be updated if necessary (compare |
| 2651 | * with f_version). |
| 2652 | */ |
| 2653 | vi->i_version = 1; |
| 2654 | |
| 2655 | /* The owner and group come from the ntfs volume. */ |
| 2656 | vi->i_uid = vol->uid; |
| 2657 | vi->i_gid = vol->gid; |
| 2658 | |
| 2659 | /* Initialize the ntfs specific part of @vi. */ |
| 2660 | ntfs_init_big_inode(vi); |
| 2661 | ni = NTFS_I(vi); |
| 2662 | /* |
| 2663 | * Set the appropriate mode, attribute type, and name. For |
| 2664 | * directories, also setup the index values to the defaults. |
| 2665 | */ |
| 2666 | if (S_ISDIR(mode)) { |
| 2667 | vi->i_mode = S_IFDIR | S_IRWXUGO; |
| 2668 | vi->i_mode &= ~vol->dmask; |
| 2669 | |
| 2670 | NInoSetMstProtected(ni); |
| 2671 | ni->type = AT_INDEX_ALLOCATION; |
| 2672 | ni->name = I30; |
| 2673 | ni->name_len = 4; |
| 2674 | |
| 2675 | ni->itype.index.block_size = 4096; |
| 2676 | ni->itype.index.block_size_bits = generic_ffs(4096) - 1; |
| 2677 | ni->itype.index.collation_rule = COLLATION_FILE_NAME; |
| 2678 | if (vol->cluster_size <= ni->itype.index.block_size) { |
| 2679 | ni->itype.index.vcn_size = vol->cluster_size; |
| 2680 | ni->itype.index.vcn_size_bits = |
| 2681 | vol->cluster_size_bits; |
| 2682 | } else { |
| 2683 | ni->itype.index.vcn_size = vol->sector_size; |
| 2684 | ni->itype.index.vcn_size_bits = |
| 2685 | vol->sector_size_bits; |
| 2686 | } |
| 2687 | } else { |
| 2688 | vi->i_mode = S_IFREG | S_IRWXUGO; |
| 2689 | vi->i_mode &= ~vol->fmask; |
| 2690 | |
| 2691 | ni->type = AT_DATA; |
| 2692 | ni->name = NULL; |
| 2693 | ni->name_len = 0; |
| 2694 | } |
| 2695 | if (IS_RDONLY(vi)) |
| 2696 | vi->i_mode &= ~S_IWUGO; |
| 2697 | |
| 2698 | /* Set the inode times to the current time. */ |
| 2699 | vi->i_atime = vi->i_mtime = vi->i_ctime = |
| 2700 | current_fs_time(vi->i_sb); |
| 2701 | /* |
| 2702 | * Set the file size to 0, the ntfs inode sizes are set to 0 by |
| 2703 | * the call to ntfs_init_big_inode() below. |
| 2704 | */ |
| 2705 | vi->i_size = 0; |
| 2706 | vi->i_blocks = 0; |
| 2707 | |
| 2708 | /* Set the sequence number. */ |
| 2709 | vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number); |
| 2710 | /* |
| 2711 | * Manually map, pin, and lock the mft record as we already |
| 2712 | * have its page mapped and it is very easy to do. |
| 2713 | */ |
| 2714 | atomic_inc(&ni->count); |
| 2715 | down(&ni->mrec_lock); |
| 2716 | ni->page = page; |
| 2717 | ni->page_ofs = ofs; |
| 2718 | /* |
| 2719 | * Make sure the allocated mft record is written out to disk. |
| 2720 | * NOTE: We do not set the ntfs inode dirty because this would |
| 2721 | * fail in ntfs_write_inode() because the inode does not have a |
| 2722 | * standard information attribute yet. Also, there is no need |
| 2723 | * to set the inode dirty because the caller is going to do |
| 2724 | * that anyway after finishing with the new mft record (e.g. at |
| 2725 | * a minimum some new attributes will be added to the mft |
| 2726 | * record. |
| 2727 | */ |
| 2728 | mark_ntfs_record_dirty(page, ofs); |
| 2729 | unlock_page(page); |
| 2730 | |
| 2731 | /* Add the inode to the inode hash for the superblock. */ |
| 2732 | insert_inode_hash(vi); |
| 2733 | |
| 2734 | /* Update the default mft allocation position. */ |
| 2735 | vol->mft_data_pos = bit + 1; |
| 2736 | } |
| 2737 | /* |
| 2738 | * Return the opened, allocated inode of the allocated mft record as |
| 2739 | * well as the mapped, pinned, and locked mft record. |
| 2740 | */ |
| 2741 | ntfs_debug("Returning opened, allocated %sinode 0x%llx.", |
| 2742 | base_ni ? "extent " : "", (long long)bit); |
| 2743 | *mrec = m; |
| 2744 | return ni; |
| 2745 | undo_data_init: |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2746 | write_lock_irqsave(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2747 | mft_ni->initialized_size = old_data_initialized; |
Anton Altaparmakov | 07a4e2d | 2005-01-12 13:08:26 +0000 | [diff] [blame] | 2748 | i_size_write(vol->mft_ino, old_data_size); |
| 2749 | write_unlock_irqrestore(&mft_ni->size_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2750 | goto undo_mftbmp_alloc_nolock; |
| 2751 | undo_mftbmp_alloc: |
| 2752 | down_write(&vol->mftbmp_lock); |
| 2753 | undo_mftbmp_alloc_nolock: |
| 2754 | if (ntfs_bitmap_clear_bit(vol->mftbmp_ino, bit)) { |
| 2755 | ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); |
| 2756 | NVolSetErrors(vol); |
| 2757 | } |
| 2758 | up_write(&vol->mftbmp_lock); |
| 2759 | err_out: |
| 2760 | return ERR_PTR(err); |
| 2761 | max_err_out: |
| 2762 | ntfs_warning(vol->sb, "Cannot allocate mft record because the maximum " |
| 2763 | "number of inodes (2^32) has already been reached."); |
| 2764 | up_write(&vol->mftbmp_lock); |
| 2765 | return ERR_PTR(-ENOSPC); |
| 2766 | } |
| 2767 | |
| 2768 | /** |
| 2769 | * ntfs_extent_mft_record_free - free an extent mft record on an ntfs volume |
| 2770 | * @ni: ntfs inode of the mapped extent mft record to free |
| 2771 | * @m: mapped extent mft record of the ntfs inode @ni |
| 2772 | * |
| 2773 | * Free the mapped extent mft record @m of the extent ntfs inode @ni. |
| 2774 | * |
| 2775 | * Note that this function unmaps the mft record and closes and destroys @ni |
| 2776 | * internally and hence you cannot use either @ni nor @m any more after this |
| 2777 | * function returns success. |
| 2778 | * |
| 2779 | * On success return 0 and on error return -errno. @ni and @m are still valid |
| 2780 | * in this case and have not been freed. |
| 2781 | * |
| 2782 | * For some errors an error message is displayed and the success code 0 is |
| 2783 | * returned and the volume is then left dirty on umount. This makes sense in |
| 2784 | * case we could not rollback the changes that were already done since the |
| 2785 | * caller no longer wants to reference this mft record so it does not matter to |
| 2786 | * the caller if something is wrong with it as long as it is properly detached |
| 2787 | * from the base inode. |
| 2788 | */ |
| 2789 | int ntfs_extent_mft_record_free(ntfs_inode *ni, MFT_RECORD *m) |
| 2790 | { |
| 2791 | unsigned long mft_no = ni->mft_no; |
| 2792 | ntfs_volume *vol = ni->vol; |
| 2793 | ntfs_inode *base_ni; |
| 2794 | ntfs_inode **extent_nis; |
| 2795 | int i, err; |
| 2796 | le16 old_seq_no; |
| 2797 | u16 seq_no; |
| 2798 | |
| 2799 | BUG_ON(NInoAttr(ni)); |
| 2800 | BUG_ON(ni->nr_extents != -1); |
| 2801 | |
| 2802 | down(&ni->extent_lock); |
| 2803 | base_ni = ni->ext.base_ntfs_ino; |
| 2804 | up(&ni->extent_lock); |
| 2805 | |
| 2806 | BUG_ON(base_ni->nr_extents <= 0); |
| 2807 | |
| 2808 | ntfs_debug("Entering for extent inode 0x%lx, base inode 0x%lx.\n", |
| 2809 | mft_no, base_ni->mft_no); |
| 2810 | |
| 2811 | down(&base_ni->extent_lock); |
| 2812 | |
| 2813 | /* Make sure we are holding the only reference to the extent inode. */ |
| 2814 | if (atomic_read(&ni->count) > 2) { |
| 2815 | ntfs_error(vol->sb, "Tried to free busy extent inode 0x%lx, " |
| 2816 | "not freeing.", base_ni->mft_no); |
| 2817 | up(&base_ni->extent_lock); |
| 2818 | return -EBUSY; |
| 2819 | } |
| 2820 | |
| 2821 | /* Dissociate the ntfs inode from the base inode. */ |
| 2822 | extent_nis = base_ni->ext.extent_ntfs_inos; |
| 2823 | err = -ENOENT; |
| 2824 | for (i = 0; i < base_ni->nr_extents; i++) { |
| 2825 | if (ni != extent_nis[i]) |
| 2826 | continue; |
| 2827 | extent_nis += i; |
| 2828 | base_ni->nr_extents--; |
| 2829 | memmove(extent_nis, extent_nis + 1, (base_ni->nr_extents - i) * |
| 2830 | sizeof(ntfs_inode*)); |
| 2831 | err = 0; |
| 2832 | break; |
| 2833 | } |
| 2834 | |
| 2835 | up(&base_ni->extent_lock); |
| 2836 | |
| 2837 | if (unlikely(err)) { |
| 2838 | ntfs_error(vol->sb, "Extent inode 0x%lx is not attached to " |
| 2839 | "its base inode 0x%lx.", mft_no, |
| 2840 | base_ni->mft_no); |
| 2841 | BUG(); |
| 2842 | } |
| 2843 | |
| 2844 | /* |
| 2845 | * The extent inode is no longer attached to the base inode so no one |
| 2846 | * can get a reference to it any more. |
| 2847 | */ |
| 2848 | |
| 2849 | /* Mark the mft record as not in use. */ |
| 2850 | m->flags &= const_cpu_to_le16(~const_le16_to_cpu(MFT_RECORD_IN_USE)); |
| 2851 | |
| 2852 | /* Increment the sequence number, skipping zero, if it is not zero. */ |
| 2853 | old_seq_no = m->sequence_number; |
| 2854 | seq_no = le16_to_cpu(old_seq_no); |
| 2855 | if (seq_no == 0xffff) |
| 2856 | seq_no = 1; |
| 2857 | else if (seq_no) |
| 2858 | seq_no++; |
| 2859 | m->sequence_number = cpu_to_le16(seq_no); |
| 2860 | |
| 2861 | /* |
| 2862 | * Set the ntfs inode dirty and write it out. We do not need to worry |
| 2863 | * about the base inode here since whatever caused the extent mft |
| 2864 | * record to be freed is guaranteed to do it already. |
| 2865 | */ |
| 2866 | NInoSetDirty(ni); |
| 2867 | err = write_mft_record(ni, m, 0); |
| 2868 | if (unlikely(err)) { |
| 2869 | ntfs_error(vol->sb, "Failed to write mft record 0x%lx, not " |
| 2870 | "freeing.", mft_no); |
| 2871 | goto rollback; |
| 2872 | } |
| 2873 | rollback_error: |
| 2874 | /* Unmap and throw away the now freed extent inode. */ |
| 2875 | unmap_extent_mft_record(ni); |
| 2876 | ntfs_clear_extent_inode(ni); |
| 2877 | |
| 2878 | /* Clear the bit in the $MFT/$BITMAP corresponding to this record. */ |
| 2879 | down_write(&vol->mftbmp_lock); |
| 2880 | err = ntfs_bitmap_clear_bit(vol->mftbmp_ino, mft_no); |
| 2881 | up_write(&vol->mftbmp_lock); |
| 2882 | if (unlikely(err)) { |
| 2883 | /* |
| 2884 | * The extent inode is gone but we failed to deallocate it in |
| 2885 | * the mft bitmap. Just emit a warning and leave the volume |
| 2886 | * dirty on umount. |
| 2887 | */ |
| 2888 | ntfs_error(vol->sb, "Failed to clear bit in mft bitmap.%s", es); |
| 2889 | NVolSetErrors(vol); |
| 2890 | } |
| 2891 | return 0; |
| 2892 | rollback: |
| 2893 | /* Rollback what we did... */ |
| 2894 | down(&base_ni->extent_lock); |
| 2895 | extent_nis = base_ni->ext.extent_ntfs_inos; |
| 2896 | if (!(base_ni->nr_extents & 3)) { |
| 2897 | int new_size = (base_ni->nr_extents + 4) * sizeof(ntfs_inode*); |
| 2898 | |
| 2899 | extent_nis = (ntfs_inode**)kmalloc(new_size, GFP_NOFS); |
| 2900 | if (unlikely(!extent_nis)) { |
| 2901 | ntfs_error(vol->sb, "Failed to allocate internal " |
| 2902 | "buffer during rollback.%s", es); |
| 2903 | up(&base_ni->extent_lock); |
| 2904 | NVolSetErrors(vol); |
| 2905 | goto rollback_error; |
| 2906 | } |
| 2907 | if (base_ni->nr_extents) { |
| 2908 | BUG_ON(!base_ni->ext.extent_ntfs_inos); |
| 2909 | memcpy(extent_nis, base_ni->ext.extent_ntfs_inos, |
| 2910 | new_size - 4 * sizeof(ntfs_inode*)); |
| 2911 | kfree(base_ni->ext.extent_ntfs_inos); |
| 2912 | } |
| 2913 | base_ni->ext.extent_ntfs_inos = extent_nis; |
| 2914 | } |
| 2915 | m->flags |= MFT_RECORD_IN_USE; |
| 2916 | m->sequence_number = old_seq_no; |
| 2917 | extent_nis[base_ni->nr_extents++] = ni; |
| 2918 | up(&base_ni->extent_lock); |
| 2919 | mark_mft_record_dirty(ni); |
| 2920 | return err; |
| 2921 | } |
| 2922 | #endif /* NTFS_RW */ |