2 * index.c - NTFS kernel index handling. Part of the Linux-NTFS project.
4 * Copyright (c) 2004 Anton Altaparmakov
6 * This program/include file is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as published
8 * by the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program/include file is distributed in the hope that it will be
12 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
13 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program (in the main directory of the Linux-NTFS
18 * distribution in the file COPYING); if not, write to the Free Software
19 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
27 * ntfs_index_ctx_get - allocate and initialize a new index context
28 * @idx_ni: ntfs index inode with which to initialize the context
30 * Allocate a new index context, initialize it with @idx_ni and return it.
31 * Return NULL if allocation failed.
33 * Locking: Caller must hold i_sem on the index inode.
35 ntfs_index_context *ntfs_index_ctx_get(ntfs_inode *idx_ni)
37 ntfs_index_context *ictx;
39 ictx = kmem_cache_alloc(ntfs_index_ctx_cache, SLAB_NOFS);
41 ictx->idx_ni = idx_ni;
56 * ntfs_index_ctx_put - release an index context
57 * @ictx: index context to free
59 * Release the index context @ictx, releasing all associated resources.
61 * Locking: Caller must hold i_sem on the index inode.
63 void ntfs_index_ctx_put(ntfs_index_context *ictx)
66 if (ictx->is_in_root) {
68 put_attr_search_ctx(ictx->actx);
70 unmap_mft_record(ictx->base_ni);
72 struct page *page = ictx->page;
74 BUG_ON(!PageLocked(page));
76 ntfs_unmap_page(page);
80 kmem_cache_free(ntfs_index_ctx_cache, ictx);
85 * ntfs_index_lookup - find a key in an index and return its index entry
86 * @key: [IN] key for which to search in the index
87 * @key_len: [IN] length of @key in bytes
88 * @ictx: [IN/OUT] context describing the index and the returned entry
90 * Before calling ntfs_index_lookup(), @ictx must have been obtained from a
91 * call to ntfs_index_ctx_get().
93 * Look for the @key in the index specified by the index lookup context @ictx.
94 * ntfs_index_lookup() walks the contents of the index looking for the @key.
96 * If the @key is found in the index, 0 is returned and @ictx is setup to
97 * describe the index entry containing the matching @key. @ictx->entry is the
98 * index entry and @ictx->data and @ictx->data_len are the index entry data and
99 * its length in bytes, respectively.
101 * If the @key is not found in the index, -ENOENT is returned and @ictx is
102 * setup to describe the index entry whose key collates immediately after the
103 * search @key, i.e. this is the position in the index at which an index entry
104 * with a key of @key would need to be inserted.
106 * If an error occurs return the negative error code and @ictx is left
109 * When finished with the entry and its data, call ntfs_index_ctx_put() to free
110 * the context and other associated resources.
112 * If the index entry was modified, call flush_dcache_index_entry_page()
113 * immediately after the modification and either ntfs_index_entry_mark_dirty()
114 * or ntfs_index_entry_write() before the call to ntfs_index_ctx_put() to
115 * ensure that the changes are written to disk.
117 * Locking: - Caller must hold i_sem on the index inode.
118 * - Each page cache page in the index allocation mapping must be
119 * locked whilst being accessed otherwise we may find a corrupt
120 * page due to it being under ->writepage at the moment which
121 * applies the mst protection fixups before writing out and then
122 * removes them again after the write is complete after which it
125 int ntfs_index_lookup(const void *key, const int key_len,
126 ntfs_index_context *ictx)
128 ntfs_inode *idx_ni = ictx->idx_ni;
129 ntfs_volume *vol = idx_ni->vol;
130 struct super_block *sb = vol->sb;
131 ntfs_inode *base_ni = idx_ni->ext.base_ntfs_ino;
135 INDEX_ALLOCATION *ia;
137 attr_search_context *actx;
140 struct address_space *ia_mapping;
144 ntfs_debug("Entering.");
145 BUG_ON(!NInoAttr(idx_ni));
146 BUG_ON(idx_ni->type != AT_INDEX_ALLOCATION);
147 BUG_ON(idx_ni->nr_extents != -1);
150 BUG_ON(key_len <= 0);
151 if (!ntfs_is_collation_rule_supported(
152 idx_ni->itype.index.collation_rule)) {
153 ntfs_error(sb, "Index uses unsupported collation rule 0x%x. "
154 "Aborting lookup.", le32_to_cpu(
155 idx_ni->itype.index.collation_rule));
158 /* Get hold of the mft record for the index inode. */
159 m = map_mft_record(base_ni);
160 if (unlikely(IS_ERR(m))) {
161 ntfs_error(sb, "map_mft_record() failed with error code %ld.",
165 actx = get_attr_search_ctx(base_ni, m);
166 if (unlikely(!actx)) {
170 /* Find the index root attribute in the mft record. */
171 if (!lookup_attr(AT_INDEX_ROOT, idx_ni->name, idx_ni->name_len,
172 CASE_SENSITIVE, 0, NULL, 0, actx)) {
173 ntfs_error(sb, "Index root attribute missing in inode 0x%lx.",
178 /* Get to the index root value (it has been verified in read_inode). */
179 ir = (INDEX_ROOT*)((u8*)actx->attr +
180 le16_to_cpu(actx->attr->data.resident.value_offset));
181 index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
182 /* The first index entry. */
183 ie = (INDEX_ENTRY*)((u8*)&ir->index +
184 le32_to_cpu(ir->index.entries_offset));
186 * Loop until we exceed valid memory (corruption case) or until we
187 * reach the last entry.
189 for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
191 if ((u8*)ie < (u8*)actx->mrec || (u8*)ie +
192 sizeof(INDEX_ENTRY_HEADER) > index_end ||
193 (u8*)ie + le16_to_cpu(ie->length) > index_end)
196 * The last entry cannot contain a key. It can however contain
197 * a pointer to a child node in the B+tree so we just break out.
199 if (ie->flags & INDEX_ENTRY_END)
201 /* Further bounds checks. */
202 if ((u32)sizeof(INDEX_ENTRY_HEADER) +
203 le16_to_cpu(ie->key_length) >
204 le16_to_cpu(ie->data.vi.data_offset) ||
205 (u32)le16_to_cpu(ie->data.vi.data_offset) +
206 le16_to_cpu(ie->data.vi.data_length) >
207 le16_to_cpu(ie->length))
209 /* If the keys match perfectly, we setup @ictx and return 0. */
210 if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(key,
211 &ie->key, key_len)) {
213 ictx->is_in_root = TRUE;
215 ictx->base_ni = base_ni;
220 ictx->data = (u8*)ie +
221 le16_to_cpu(ie->data.vi.data_offset);
222 ictx->data_len = le16_to_cpu(ie->data.vi.data_length);
227 * Not a perfect match, need to do full blown collation so we
228 * know which way in the B+tree we have to go.
230 rc = ntfs_collate(vol, idx_ni->itype.index.collation_rule, key,
231 key_len, &ie->key, le16_to_cpu(ie->key_length));
233 * If @key collates before the key of the current entry, there
234 * is definitely no such key in this index but we might need to
235 * descend into the B+tree so we just break out of the loop.
240 * A match should never happen as the memcmp() call should have
241 * cought it, but we still treat it correctly.
245 /* The keys are not equal, continue the search. */
248 * We have finished with this index without success. Check for the
249 * presence of a child node and if not present setup @ictx and return
252 if (!(ie->flags & INDEX_ENTRY_NODE)) {
253 ntfs_debug("Entry not found.");
256 } /* Child node present, descend into it. */
257 /* Consistency check: Verify that an index allocation exists. */
258 if (!NInoIndexAllocPresent(idx_ni)) {
259 ntfs_error(sb, "No index allocation attribute but index entry "
260 "requires one. Inode 0x%lx is corrupt or "
261 "driver bug.", idx_ni->mft_no);
265 /* Get the starting vcn of the index_block holding the child node. */
266 vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8);
267 ia_mapping = VFS_I(idx_ni)->i_mapping;
269 * We are done with the index root and the mft record. Release them,
270 * otherwise we deadlock with ntfs_map_page().
272 put_attr_search_ctx(actx);
273 unmap_mft_record(base_ni);
276 descend_into_child_node:
278 * Convert vcn to index into the index allocation attribute in units
279 * of PAGE_CACHE_SIZE and map the page cache page, reading it from
282 page = ntfs_map_page(ia_mapping, vcn <<
283 idx_ni->itype.index.vcn_size_bits >> PAGE_CACHE_SHIFT);
285 ntfs_error(sb, "Failed to map index page, error %ld.",
291 kaddr = (u8*)page_address(page);
292 fast_descend_into_child_node:
293 /* Get to the index allocation block. */
294 ia = (INDEX_ALLOCATION*)(kaddr + ((vcn <<
295 idx_ni->itype.index.vcn_size_bits) & ~PAGE_CACHE_MASK));
297 if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) {
298 ntfs_error(sb, "Out of bounds check failed. Corrupt inode "
299 "0x%lx or driver bug.", idx_ni->mft_no);
303 if (sle64_to_cpu(ia->index_block_vcn) != vcn) {
304 ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is "
305 "different from expected VCN (0x%llx). Inode "
306 "0x%lx is corrupt or driver bug.",
308 sle64_to_cpu(ia->index_block_vcn),
309 (unsigned long long)vcn, idx_ni->mft_no);
313 if (le32_to_cpu(ia->index.allocated_size) + 0x18 !=
314 idx_ni->itype.index.block_size) {
315 ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx has "
316 "a size (%u) differing from the index "
317 "specified size (%u). Inode is corrupt or "
318 "driver bug.", (unsigned long long)vcn,
320 le32_to_cpu(ia->index.allocated_size) + 0x18,
321 idx_ni->itype.index.block_size);
325 index_end = (u8*)ia + idx_ni->itype.index.block_size;
326 if (index_end > kaddr + PAGE_CACHE_SIZE) {
327 ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx "
328 "crosses page boundary. Impossible! Cannot "
329 "access! This is probably a bug in the "
330 "driver.", (unsigned long long)vcn,
335 index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length);
336 if (index_end > (u8*)ia + idx_ni->itype.index.block_size) {
337 ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of inode "
338 "0x%lx exceeds maximum size.",
339 (unsigned long long)vcn, idx_ni->mft_no);
343 /* The first index entry. */
344 ie = (INDEX_ENTRY*)((u8*)&ia->index +
345 le32_to_cpu(ia->index.entries_offset));
347 * Iterate similar to above big loop but applied to index buffer, thus
348 * loop until we exceed valid memory (corruption case) or until we
349 * reach the last entry.
351 for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
353 if ((u8*)ie < (u8*)ia || (u8*)ie +
354 sizeof(INDEX_ENTRY_HEADER) > index_end ||
355 (u8*)ie + le16_to_cpu(ie->length) > index_end) {
356 ntfs_error(sb, "Index entry out of bounds in inode "
357 "0x%lx.", idx_ni->mft_no);
362 * The last entry cannot contain a ket. It can however contain
363 * a pointer to a child node in the B+tree so we just break out.
365 if (ie->flags & INDEX_ENTRY_END)
367 /* Further bounds checks. */
368 if ((u32)sizeof(INDEX_ENTRY_HEADER) +
369 le16_to_cpu(ie->key_length) >
370 le16_to_cpu(ie->data.vi.data_offset) ||
371 (u32)le16_to_cpu(ie->data.vi.data_offset) +
372 le16_to_cpu(ie->data.vi.data_length) >
373 le16_to_cpu(ie->length)) {
374 ntfs_error(sb, "Index entry out of bounds in inode "
375 "0x%lx.", idx_ni->mft_no);
379 /* If the keys match perfectly, we setup @ictx and return 0. */
380 if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(key,
381 &ie->key, key_len)) {
383 ictx->is_in_root = FALSE;
385 ictx->base_ni = NULL;
391 * Not a perfect match, need to do full blown collation so we
392 * know which way in the B+tree we have to go.
394 rc = ntfs_collate(vol, idx_ni->itype.index.collation_rule, key,
395 key_len, &ie->key, le16_to_cpu(ie->key_length));
397 * If @key collates before the key of the current entry, there
398 * is definitely no such key in this index but we might need to
399 * descend into the B+tree so we just break out of the loop.
404 * A match should never happen as the memcmp() call should have
405 * cought it, but we still treat it correctly.
409 /* The keys are not equal, continue the search. */
412 * We have finished with this index buffer without success. Check for
413 * the presence of a child node and if not present return -ENOENT.
415 if (!(ie->flags & INDEX_ENTRY_NODE)) {
416 ntfs_debug("Entry not found.");
420 if ((ia->index.flags & NODE_MASK) == LEAF_NODE) {
421 ntfs_error(sb, "Index entry with child node found in a leaf "
422 "node in inode 0x%lx.", idx_ni->mft_no);
426 /* Child node present, descend into it. */
428 vcn = sle64_to_cpup((u8*)ie + le16_to_cpu(ie->length) - 8);
431 * If vcn is in the same page cache page as old_vcn we recycle
434 if (old_vcn << vol->cluster_size_bits >>
435 PAGE_CACHE_SHIFT == vcn <<
436 vol->cluster_size_bits >>
438 goto fast_descend_into_child_node;
440 ntfs_unmap_page(page);
441 goto descend_into_child_node;
443 ntfs_error(sb, "Negative child node vcn in inode 0x%lx.",
448 ntfs_unmap_page(page);
451 put_attr_search_ctx(actx);
453 unmap_mft_record(base_ni);
456 ntfs_error(sb, "Corrupt index. Aborting lookup.");
464 * __ntfs_index_entry_mark_dirty - mark an index allocation entry dirty
465 * @ictx: ntfs index context describing the index entry
467 * NOTE: You want to use fs/ntfs/index.h::ntfs_index_entry_mark_dirty() instead!
469 * Mark the index allocation entry described by the index entry context @ictx
472 * The index entry must be in an index block belonging to the index allocation
473 * attribute. Mark the buffers belonging to the index record as well as the
474 * page cache page the index block is in dirty. This automatically marks the
475 * VFS inode of the ntfs index inode to which the index entry belongs dirty,
476 * too (I_DIRTY_PAGES) and this in turn ensures the page buffers, and hence the
477 * dirty index block, will be written out to disk later.
479 void __ntfs_index_entry_mark_dirty(ntfs_index_context *ictx)
483 struct buffer_head *bh, *head;
484 unsigned int rec_start, rec_end, bh_size, bh_start, bh_end;
486 BUG_ON(ictx->is_in_root);
489 BUG_ON(!page_has_buffers(page));
491 * If the index block is the same size as the page cache page, set all
492 * the buffers in the page, as well as the page itself, dirty.
494 if (ni->itype.index.block_size == PAGE_CACHE_SIZE) {
495 __set_page_dirty_buffers(page);
498 /* Set only the buffers in which the index block is located dirty. */
499 rec_start = (unsigned int)((u8*)ictx->ia - (u8*)page_address(page));
500 rec_end = rec_start + ni->itype.index.block_size;
501 bh_size = ni->vol->sb->s_blocksize;
503 bh = head = page_buffers(page);
505 bh_end = bh_start + bh_size;
506 if ((bh_start >= rec_start) && (bh_end <= rec_end))
507 set_buffer_dirty(bh);
509 } while ((bh = bh->b_this_page) != head);
510 /* Finally, set the page itself dirty, too. */
511 __set_page_dirty_nobuffers(page);