blob: 11fd5307d780568ce8b5a0fc8ab63cf39d1aaf02 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * index.c - NTFS kernel index handling. Part of the Linux-NTFS project.
3 *
Anton Altaparmakov442d2072005-05-27 16:42:56 +01004 * Copyright (c) 2004-2005 Anton Altaparmakov
Linus Torvalds1da177e2005-04-16 15:20:36 -07005 *
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.
10 *
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.
15 *
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
20 */
21
22#include "aops.h"
23#include "collate.h"
24#include "debug.h"
25#include "index.h"
26#include "ntfs.h"
27
28/**
29 * ntfs_index_ctx_get - allocate and initialize a new index context
30 * @idx_ni: ntfs index inode with which to initialize the context
31 *
32 * Allocate a new index context, initialize it with @idx_ni and return it.
33 * Return NULL if allocation failed.
34 *
35 * Locking: Caller must hold i_sem on the index inode.
36 */
37ntfs_index_context *ntfs_index_ctx_get(ntfs_inode *idx_ni)
38{
39 ntfs_index_context *ictx;
40
41 ictx = kmem_cache_alloc(ntfs_index_ctx_cache, SLAB_NOFS);
Anton Altaparmakov442d2072005-05-27 16:42:56 +010042 if (ictx)
43 *ictx = (ntfs_index_context){ .idx_ni = idx_ni };
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 return ictx;
45}
46
47/**
48 * ntfs_index_ctx_put - release an index context
49 * @ictx: index context to free
50 *
51 * Release the index context @ictx, releasing all associated resources.
52 *
53 * Locking: Caller must hold i_sem on the index inode.
54 */
55void ntfs_index_ctx_put(ntfs_index_context *ictx)
56{
57 if (ictx->entry) {
58 if (ictx->is_in_root) {
59 if (ictx->actx)
60 ntfs_attr_put_search_ctx(ictx->actx);
61 if (ictx->base_ni)
62 unmap_mft_record(ictx->base_ni);
63 } else {
64 struct page *page = ictx->page;
65 if (page) {
66 BUG_ON(!PageLocked(page));
67 unlock_page(page);
68 ntfs_unmap_page(page);
69 }
70 }
71 }
72 kmem_cache_free(ntfs_index_ctx_cache, ictx);
73 return;
74}
75
76/**
77 * ntfs_index_lookup - find a key in an index and return its index entry
78 * @key: [IN] key for which to search in the index
79 * @key_len: [IN] length of @key in bytes
80 * @ictx: [IN/OUT] context describing the index and the returned entry
81 *
82 * Before calling ntfs_index_lookup(), @ictx must have been obtained from a
83 * call to ntfs_index_ctx_get().
84 *
85 * Look for the @key in the index specified by the index lookup context @ictx.
86 * ntfs_index_lookup() walks the contents of the index looking for the @key.
87 *
88 * If the @key is found in the index, 0 is returned and @ictx is setup to
89 * describe the index entry containing the matching @key. @ictx->entry is the
90 * index entry and @ictx->data and @ictx->data_len are the index entry data and
91 * its length in bytes, respectively.
92 *
93 * If the @key is not found in the index, -ENOENT is returned and @ictx is
94 * setup to describe the index entry whose key collates immediately after the
95 * search @key, i.e. this is the position in the index at which an index entry
96 * with a key of @key would need to be inserted.
97 *
98 * If an error occurs return the negative error code and @ictx is left
99 * untouched.
100 *
101 * When finished with the entry and its data, call ntfs_index_ctx_put() to free
102 * the context and other associated resources.
103 *
104 * If the index entry was modified, call flush_dcache_index_entry_page()
105 * immediately after the modification and either ntfs_index_entry_mark_dirty()
106 * or ntfs_index_entry_write() before the call to ntfs_index_ctx_put() to
107 * ensure that the changes are written to disk.
108 *
109 * Locking: - Caller must hold i_sem on the index inode.
110 * - Each page cache page in the index allocation mapping must be
111 * locked whilst being accessed otherwise we may find a corrupt
112 * page due to it being under ->writepage at the moment which
113 * applies the mst protection fixups before writing out and then
114 * removes them again after the write is complete after which it
115 * unlocks the page.
116 */
117int ntfs_index_lookup(const void *key, const int key_len,
118 ntfs_index_context *ictx)
119{
120 VCN vcn, old_vcn;
121 ntfs_inode *idx_ni = ictx->idx_ni;
122 ntfs_volume *vol = idx_ni->vol;
123 struct super_block *sb = vol->sb;
124 ntfs_inode *base_ni = idx_ni->ext.base_ntfs_ino;
125 MFT_RECORD *m;
126 INDEX_ROOT *ir;
127 INDEX_ENTRY *ie;
128 INDEX_ALLOCATION *ia;
129 u8 *index_end, *kaddr;
130 ntfs_attr_search_ctx *actx;
131 struct address_space *ia_mapping;
132 struct page *page;
133 int rc, err = 0;
134
135 ntfs_debug("Entering.");
136 BUG_ON(!NInoAttr(idx_ni));
137 BUG_ON(idx_ni->type != AT_INDEX_ALLOCATION);
138 BUG_ON(idx_ni->nr_extents != -1);
139 BUG_ON(!base_ni);
140 BUG_ON(!key);
141 BUG_ON(key_len <= 0);
142 if (!ntfs_is_collation_rule_supported(
143 idx_ni->itype.index.collation_rule)) {
144 ntfs_error(sb, "Index uses unsupported collation rule 0x%x. "
145 "Aborting lookup.", le32_to_cpu(
146 idx_ni->itype.index.collation_rule));
147 return -EOPNOTSUPP;
148 }
149 /* Get hold of the mft record for the index inode. */
150 m = map_mft_record(base_ni);
151 if (IS_ERR(m)) {
152 ntfs_error(sb, "map_mft_record() failed with error code %ld.",
153 -PTR_ERR(m));
154 return PTR_ERR(m);
155 }
156 actx = ntfs_attr_get_search_ctx(base_ni, m);
157 if (unlikely(!actx)) {
158 err = -ENOMEM;
159 goto err_out;
160 }
161 /* Find the index root attribute in the mft record. */
162 err = ntfs_attr_lookup(AT_INDEX_ROOT, idx_ni->name, idx_ni->name_len,
163 CASE_SENSITIVE, 0, NULL, 0, actx);
164 if (unlikely(err)) {
165 if (err == -ENOENT) {
166 ntfs_error(sb, "Index root attribute missing in inode "
167 "0x%lx.", idx_ni->mft_no);
168 err = -EIO;
169 }
170 goto err_out;
171 }
172 /* Get to the index root value (it has been verified in read_inode). */
173 ir = (INDEX_ROOT*)((u8*)actx->attr +
174 le16_to_cpu(actx->attr->data.resident.value_offset));
175 index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
176 /* The first index entry. */
177 ie = (INDEX_ENTRY*)((u8*)&ir->index +
178 le32_to_cpu(ir->index.entries_offset));
179 /*
180 * Loop until we exceed valid memory (corruption case) or until we
181 * reach the last entry.
182 */
183 for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
184 /* Bounds checks. */
185 if ((u8*)ie < (u8*)actx->mrec || (u8*)ie +
186 sizeof(INDEX_ENTRY_HEADER) > index_end ||
187 (u8*)ie + le16_to_cpu(ie->length) > index_end)
188 goto idx_err_out;
189 /*
190 * The last entry cannot contain a key. It can however contain
191 * a pointer to a child node in the B+tree so we just break out.
192 */
193 if (ie->flags & INDEX_ENTRY_END)
194 break;
195 /* Further bounds checks. */
196 if ((u32)sizeof(INDEX_ENTRY_HEADER) +
197 le16_to_cpu(ie->key_length) >
198 le16_to_cpu(ie->data.vi.data_offset) ||
199 (u32)le16_to_cpu(ie->data.vi.data_offset) +
200 le16_to_cpu(ie->data.vi.data_length) >
201 le16_to_cpu(ie->length))
202 goto idx_err_out;
203 /* If the keys match perfectly, we setup @ictx and return 0. */
204 if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(key,
205 &ie->key, key_len)) {
206ir_done:
207 ictx->is_in_root = TRUE;
208 ictx->actx = actx;
209 ictx->base_ni = base_ni;
210 ictx->ia = NULL;
211 ictx->page = NULL;
212done:
213 ictx->entry = ie;
214 ictx->data = (u8*)ie +
215 le16_to_cpu(ie->data.vi.data_offset);
216 ictx->data_len = le16_to_cpu(ie->data.vi.data_length);
217 ntfs_debug("Done.");
218 return err;
219 }
220 /*
221 * Not a perfect match, need to do full blown collation so we
222 * know which way in the B+tree we have to go.
223 */
224 rc = ntfs_collate(vol, idx_ni->itype.index.collation_rule, key,
225 key_len, &ie->key, le16_to_cpu(ie->key_length));
226 /*
227 * If @key collates before the key of the current entry, there
228 * is definitely no such key in this index but we might need to
229 * descend into the B+tree so we just break out of the loop.
230 */
231 if (rc == -1)
232 break;
233 /*
234 * A match should never happen as the memcmp() call should have
235 * cought it, but we still treat it correctly.
236 */
237 if (!rc)
238 goto ir_done;
239 /* The keys are not equal, continue the search. */
240 }
241 /*
242 * We have finished with this index without success. Check for the
243 * presence of a child node and if not present setup @ictx and return
244 * -ENOENT.
245 */
246 if (!(ie->flags & INDEX_ENTRY_NODE)) {
247 ntfs_debug("Entry not found.");
248 err = -ENOENT;
249 goto ir_done;
250 } /* Child node present, descend into it. */
251 /* Consistency check: Verify that an index allocation exists. */
252 if (!NInoIndexAllocPresent(idx_ni)) {
253 ntfs_error(sb, "No index allocation attribute but index entry "
254 "requires one. Inode 0x%lx is corrupt or "
255 "driver bug.", idx_ni->mft_no);
256 goto err_out;
257 }
258 /* Get the starting vcn of the index_block holding the child node. */
259 vcn = sle64_to_cpup((sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8));
260 ia_mapping = VFS_I(idx_ni)->i_mapping;
261 /*
262 * We are done with the index root and the mft record. Release them,
263 * otherwise we deadlock with ntfs_map_page().
264 */
265 ntfs_attr_put_search_ctx(actx);
266 unmap_mft_record(base_ni);
267 m = NULL;
268 actx = NULL;
269descend_into_child_node:
270 /*
271 * Convert vcn to index into the index allocation attribute in units
272 * of PAGE_CACHE_SIZE and map the page cache page, reading it from
273 * disk if necessary.
274 */
275 page = ntfs_map_page(ia_mapping, vcn <<
276 idx_ni->itype.index.vcn_size_bits >> PAGE_CACHE_SHIFT);
277 if (IS_ERR(page)) {
278 ntfs_error(sb, "Failed to map index page, error %ld.",
279 -PTR_ERR(page));
280 err = PTR_ERR(page);
281 goto err_out;
282 }
283 lock_page(page);
284 kaddr = (u8*)page_address(page);
285fast_descend_into_child_node:
286 /* Get to the index allocation block. */
287 ia = (INDEX_ALLOCATION*)(kaddr + ((vcn <<
288 idx_ni->itype.index.vcn_size_bits) & ~PAGE_CACHE_MASK));
289 /* Bounds checks. */
290 if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_CACHE_SIZE) {
291 ntfs_error(sb, "Out of bounds check failed. Corrupt inode "
292 "0x%lx or driver bug.", idx_ni->mft_no);
293 goto unm_err_out;
294 }
295 /* Catch multi sector transfer fixup errors. */
296 if (unlikely(!ntfs_is_indx_record(ia->magic))) {
297 ntfs_error(sb, "Index record with vcn 0x%llx is corrupt. "
298 "Corrupt inode 0x%lx. Run chkdsk.",
299 (long long)vcn, idx_ni->mft_no);
300 goto unm_err_out;
301 }
302 if (sle64_to_cpu(ia->index_block_vcn) != vcn) {
303 ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is "
304 "different from expected VCN (0x%llx). Inode "
305 "0x%lx is corrupt or driver bug.",
306 (unsigned long long)
307 sle64_to_cpu(ia->index_block_vcn),
308 (unsigned long long)vcn, idx_ni->mft_no);
309 goto unm_err_out;
310 }
311 if (le32_to_cpu(ia->index.allocated_size) + 0x18 !=
312 idx_ni->itype.index.block_size) {
313 ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx has "
314 "a size (%u) differing from the index "
315 "specified size (%u). Inode is corrupt or "
316 "driver bug.", (unsigned long long)vcn,
317 idx_ni->mft_no,
318 le32_to_cpu(ia->index.allocated_size) + 0x18,
319 idx_ni->itype.index.block_size);
320 goto unm_err_out;
321 }
322 index_end = (u8*)ia + idx_ni->itype.index.block_size;
323 if (index_end > kaddr + PAGE_CACHE_SIZE) {
324 ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx "
325 "crosses page boundary. Impossible! Cannot "
326 "access! This is probably a bug in the "
327 "driver.", (unsigned long long)vcn,
328 idx_ni->mft_no);
329 goto unm_err_out;
330 }
331 index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length);
332 if (index_end > (u8*)ia + idx_ni->itype.index.block_size) {
333 ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of inode "
334 "0x%lx exceeds maximum size.",
335 (unsigned long long)vcn, idx_ni->mft_no);
336 goto unm_err_out;
337 }
338 /* The first index entry. */
339 ie = (INDEX_ENTRY*)((u8*)&ia->index +
340 le32_to_cpu(ia->index.entries_offset));
341 /*
342 * Iterate similar to above big loop but applied to index buffer, thus
343 * loop until we exceed valid memory (corruption case) or until we
344 * reach the last entry.
345 */
346 for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) {
347 /* Bounds checks. */
348 if ((u8*)ie < (u8*)ia || (u8*)ie +
349 sizeof(INDEX_ENTRY_HEADER) > index_end ||
350 (u8*)ie + le16_to_cpu(ie->length) > index_end) {
351 ntfs_error(sb, "Index entry out of bounds in inode "
352 "0x%lx.", idx_ni->mft_no);
353 goto unm_err_out;
354 }
355 /*
356 * The last entry cannot contain a key. It can however contain
357 * a pointer to a child node in the B+tree so we just break out.
358 */
359 if (ie->flags & INDEX_ENTRY_END)
360 break;
361 /* Further bounds checks. */
362 if ((u32)sizeof(INDEX_ENTRY_HEADER) +
363 le16_to_cpu(ie->key_length) >
364 le16_to_cpu(ie->data.vi.data_offset) ||
365 (u32)le16_to_cpu(ie->data.vi.data_offset) +
366 le16_to_cpu(ie->data.vi.data_length) >
367 le16_to_cpu(ie->length)) {
368 ntfs_error(sb, "Index entry out of bounds in inode "
369 "0x%lx.", idx_ni->mft_no);
370 goto unm_err_out;
371 }
372 /* If the keys match perfectly, we setup @ictx and return 0. */
373 if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(key,
374 &ie->key, key_len)) {
375ia_done:
376 ictx->is_in_root = FALSE;
377 ictx->actx = NULL;
378 ictx->base_ni = NULL;
379 ictx->ia = ia;
380 ictx->page = page;
381 goto done;
382 }
383 /*
384 * Not a perfect match, need to do full blown collation so we
385 * know which way in the B+tree we have to go.
386 */
387 rc = ntfs_collate(vol, idx_ni->itype.index.collation_rule, key,
388 key_len, &ie->key, le16_to_cpu(ie->key_length));
389 /*
390 * If @key collates before the key of the current entry, there
391 * is definitely no such key in this index but we might need to
392 * descend into the B+tree so we just break out of the loop.
393 */
394 if (rc == -1)
395 break;
396 /*
397 * A match should never happen as the memcmp() call should have
398 * cought it, but we still treat it correctly.
399 */
400 if (!rc)
401 goto ia_done;
402 /* The keys are not equal, continue the search. */
403 }
404 /*
405 * We have finished with this index buffer without success. Check for
406 * the presence of a child node and if not present return -ENOENT.
407 */
408 if (!(ie->flags & INDEX_ENTRY_NODE)) {
409 ntfs_debug("Entry not found.");
410 err = -ENOENT;
411 goto ia_done;
412 }
413 if ((ia->index.flags & NODE_MASK) == LEAF_NODE) {
414 ntfs_error(sb, "Index entry with child node found in a leaf "
415 "node in inode 0x%lx.", idx_ni->mft_no);
416 goto unm_err_out;
417 }
418 /* Child node present, descend into it. */
419 old_vcn = vcn;
420 vcn = sle64_to_cpup((sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8));
421 if (vcn >= 0) {
422 /*
423 * If vcn is in the same page cache page as old_vcn we recycle
424 * the mapped page.
425 */
426 if (old_vcn << vol->cluster_size_bits >>
427 PAGE_CACHE_SHIFT == vcn <<
428 vol->cluster_size_bits >>
429 PAGE_CACHE_SHIFT)
430 goto fast_descend_into_child_node;
431 unlock_page(page);
432 ntfs_unmap_page(page);
433 goto descend_into_child_node;
434 }
435 ntfs_error(sb, "Negative child node vcn in inode 0x%lx.",
436 idx_ni->mft_no);
437unm_err_out:
438 unlock_page(page);
439 ntfs_unmap_page(page);
440err_out:
441 if (!err)
442 err = -EIO;
443 if (actx)
444 ntfs_attr_put_search_ctx(actx);
445 if (m)
446 unmap_mft_record(base_ni);
447 return err;
448idx_err_out:
449 ntfs_error(sb, "Corrupt index. Aborting lookup.");
450 goto err_out;
451}