1 | // SPDX-License-Identifier: GPL-2.0-or-later |
2 | /* |
3 | * index.c - NTFS kernel index handling. Part of the Linux-NTFS project. |
4 | * |
5 | * Copyright (c) 2004-2005 Anton Altaparmakov |
6 | */ |
7 | |
8 | #include <linux/slab.h> |
9 | |
10 | #include "aops.h" |
11 | #include "collate.h" |
12 | #include "debug.h" |
13 | #include "index.h" |
14 | #include "ntfs.h" |
15 | |
16 | /** |
17 | * ntfs_index_ctx_get - allocate and initialize a new index context |
18 | * @idx_ni: ntfs index inode with which to initialize the context |
19 | * |
20 | * Allocate a new index context, initialize it with @idx_ni and return it. |
21 | * Return NULL if allocation failed. |
22 | * |
23 | * Locking: Caller must hold i_mutex on the index inode. |
24 | */ |
25 | ntfs_index_context *ntfs_index_ctx_get(ntfs_inode *idx_ni) |
26 | { |
27 | ntfs_index_context *ictx; |
28 | |
29 | ictx = kmem_cache_alloc(cachep: ntfs_index_ctx_cache, GFP_NOFS); |
30 | if (ictx) |
31 | *ictx = (ntfs_index_context){ .idx_ni = idx_ni }; |
32 | return ictx; |
33 | } |
34 | |
35 | /** |
36 | * ntfs_index_ctx_put - release an index context |
37 | * @ictx: index context to free |
38 | * |
39 | * Release the index context @ictx, releasing all associated resources. |
40 | * |
41 | * Locking: Caller must hold i_mutex on the index inode. |
42 | */ |
43 | void ntfs_index_ctx_put(ntfs_index_context *ictx) |
44 | { |
45 | if (ictx->entry) { |
46 | if (ictx->is_in_root) { |
47 | if (ictx->actx) |
48 | ntfs_attr_put_search_ctx(ctx: ictx->actx); |
49 | if (ictx->base_ni) |
50 | unmap_mft_record(ni: ictx->base_ni); |
51 | } else { |
52 | struct page *page = ictx->page; |
53 | if (page) { |
54 | BUG_ON(!PageLocked(page)); |
55 | unlock_page(page); |
56 | ntfs_unmap_page(page); |
57 | } |
58 | } |
59 | } |
60 | kmem_cache_free(s: ntfs_index_ctx_cache, objp: ictx); |
61 | return; |
62 | } |
63 | |
64 | /** |
65 | * ntfs_index_lookup - find a key in an index and return its index entry |
66 | * @key: [IN] key for which to search in the index |
67 | * @key_len: [IN] length of @key in bytes |
68 | * @ictx: [IN/OUT] context describing the index and the returned entry |
69 | * |
70 | * Before calling ntfs_index_lookup(), @ictx must have been obtained from a |
71 | * call to ntfs_index_ctx_get(). |
72 | * |
73 | * Look for the @key in the index specified by the index lookup context @ictx. |
74 | * ntfs_index_lookup() walks the contents of the index looking for the @key. |
75 | * |
76 | * If the @key is found in the index, 0 is returned and @ictx is setup to |
77 | * describe the index entry containing the matching @key. @ictx->entry is the |
78 | * index entry and @ictx->data and @ictx->data_len are the index entry data and |
79 | * its length in bytes, respectively. |
80 | * |
81 | * If the @key is not found in the index, -ENOENT is returned and @ictx is |
82 | * setup to describe the index entry whose key collates immediately after the |
83 | * search @key, i.e. this is the position in the index at which an index entry |
84 | * with a key of @key would need to be inserted. |
85 | * |
86 | * If an error occurs return the negative error code and @ictx is left |
87 | * untouched. |
88 | * |
89 | * When finished with the entry and its data, call ntfs_index_ctx_put() to free |
90 | * the context and other associated resources. |
91 | * |
92 | * If the index entry was modified, call flush_dcache_index_entry_page() |
93 | * immediately after the modification and either ntfs_index_entry_mark_dirty() |
94 | * or ntfs_index_entry_write() before the call to ntfs_index_ctx_put() to |
95 | * ensure that the changes are written to disk. |
96 | * |
97 | * Locking: - Caller must hold i_mutex on the index inode. |
98 | * - Each page cache page in the index allocation mapping must be |
99 | * locked whilst being accessed otherwise we may find a corrupt |
100 | * page due to it being under ->writepage at the moment which |
101 | * applies the mst protection fixups before writing out and then |
102 | * removes them again after the write is complete after which it |
103 | * unlocks the page. |
104 | */ |
105 | int ntfs_index_lookup(const void *key, const int key_len, |
106 | ntfs_index_context *ictx) |
107 | { |
108 | VCN vcn, old_vcn; |
109 | ntfs_inode *idx_ni = ictx->idx_ni; |
110 | ntfs_volume *vol = idx_ni->vol; |
111 | struct super_block *sb = vol->sb; |
112 | ntfs_inode *base_ni = idx_ni->ext.base_ntfs_ino; |
113 | MFT_RECORD *m; |
114 | INDEX_ROOT *ir; |
115 | INDEX_ENTRY *ie; |
116 | INDEX_ALLOCATION *ia; |
117 | u8 *index_end, *kaddr; |
118 | ntfs_attr_search_ctx *actx; |
119 | struct address_space *ia_mapping; |
120 | struct page *page; |
121 | int rc, err = 0; |
122 | |
123 | ntfs_debug("Entering." ); |
124 | BUG_ON(!NInoAttr(idx_ni)); |
125 | BUG_ON(idx_ni->type != AT_INDEX_ALLOCATION); |
126 | BUG_ON(idx_ni->nr_extents != -1); |
127 | BUG_ON(!base_ni); |
128 | BUG_ON(!key); |
129 | BUG_ON(key_len <= 0); |
130 | if (!ntfs_is_collation_rule_supported( |
131 | cr: idx_ni->itype.index.collation_rule)) { |
132 | ntfs_error(sb, "Index uses unsupported collation rule 0x%x. " |
133 | "Aborting lookup." , le32_to_cpu( |
134 | idx_ni->itype.index.collation_rule)); |
135 | return -EOPNOTSUPP; |
136 | } |
137 | /* Get hold of the mft record for the index inode. */ |
138 | m = map_mft_record(ni: base_ni); |
139 | if (IS_ERR(ptr: m)) { |
140 | ntfs_error(sb, "map_mft_record() failed with error code %ld." , |
141 | -PTR_ERR(m)); |
142 | return PTR_ERR(ptr: m); |
143 | } |
144 | actx = ntfs_attr_get_search_ctx(ni: base_ni, mrec: m); |
145 | if (unlikely(!actx)) { |
146 | err = -ENOMEM; |
147 | goto err_out; |
148 | } |
149 | /* Find the index root attribute in the mft record. */ |
150 | err = ntfs_attr_lookup(type: AT_INDEX_ROOT, name: idx_ni->name, name_len: idx_ni->name_len, |
151 | ic: CASE_SENSITIVE, lowest_vcn: 0, NULL, val_len: 0, ctx: actx); |
152 | if (unlikely(err)) { |
153 | if (err == -ENOENT) { |
154 | ntfs_error(sb, "Index root attribute missing in inode " |
155 | "0x%lx." , idx_ni->mft_no); |
156 | err = -EIO; |
157 | } |
158 | goto err_out; |
159 | } |
160 | /* Get to the index root value (it has been verified in read_inode). */ |
161 | ir = (INDEX_ROOT*)((u8*)actx->attr + |
162 | le16_to_cpu(actx->attr->data.resident.value_offset)); |
163 | index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length); |
164 | /* The first index entry. */ |
165 | ie = (INDEX_ENTRY*)((u8*)&ir->index + |
166 | le32_to_cpu(ir->index.entries_offset)); |
167 | /* |
168 | * Loop until we exceed valid memory (corruption case) or until we |
169 | * reach the last entry. |
170 | */ |
171 | for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) { |
172 | /* Bounds checks. */ |
173 | if ((u8*)ie < (u8*)actx->mrec || (u8*)ie + |
174 | sizeof(INDEX_ENTRY_HEADER) > index_end || |
175 | (u8*)ie + le16_to_cpu(ie->length) > index_end) |
176 | goto idx_err_out; |
177 | /* |
178 | * The last entry cannot contain a key. It can however contain |
179 | * a pointer to a child node in the B+tree so we just break out. |
180 | */ |
181 | if (ie->flags & INDEX_ENTRY_END) |
182 | break; |
183 | /* Further bounds checks. */ |
184 | if ((u32)sizeof(INDEX_ENTRY_HEADER) + |
185 | le16_to_cpu(ie->key_length) > |
186 | le16_to_cpu(ie->data.vi.data_offset) || |
187 | (u32)le16_to_cpu(ie->data.vi.data_offset) + |
188 | le16_to_cpu(ie->data.vi.data_length) > |
189 | le16_to_cpu(ie->length)) |
190 | goto idx_err_out; |
191 | /* If the keys match perfectly, we setup @ictx and return 0. */ |
192 | if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(p: key, |
193 | q: &ie->key, size: key_len)) { |
194 | ir_done: |
195 | ictx->is_in_root = true; |
196 | ictx->ir = ir; |
197 | ictx->actx = actx; |
198 | ictx->base_ni = base_ni; |
199 | ictx->ia = NULL; |
200 | ictx->page = NULL; |
201 | done: |
202 | ictx->entry = ie; |
203 | ictx->data = (u8*)ie + |
204 | le16_to_cpu(ie->data.vi.data_offset); |
205 | ictx->data_len = le16_to_cpu(ie->data.vi.data_length); |
206 | ntfs_debug("Done." ); |
207 | return err; |
208 | } |
209 | /* |
210 | * Not a perfect match, need to do full blown collation so we |
211 | * know which way in the B+tree we have to go. |
212 | */ |
213 | rc = ntfs_collate(vol, cr: idx_ni->itype.index.collation_rule, data1: key, |
214 | data1_len: key_len, data2: &ie->key, le16_to_cpu(ie->key_length)); |
215 | /* |
216 | * If @key collates before the key of the current entry, there |
217 | * is definitely no such key in this index but we might need to |
218 | * descend into the B+tree so we just break out of the loop. |
219 | */ |
220 | if (rc == -1) |
221 | break; |
222 | /* |
223 | * A match should never happen as the memcmp() call should have |
224 | * cought it, but we still treat it correctly. |
225 | */ |
226 | if (!rc) |
227 | goto ir_done; |
228 | /* The keys are not equal, continue the search. */ |
229 | } |
230 | /* |
231 | * We have finished with this index without success. Check for the |
232 | * presence of a child node and if not present setup @ictx and return |
233 | * -ENOENT. |
234 | */ |
235 | if (!(ie->flags & INDEX_ENTRY_NODE)) { |
236 | ntfs_debug("Entry not found." ); |
237 | err = -ENOENT; |
238 | goto ir_done; |
239 | } /* Child node present, descend into it. */ |
240 | /* Consistency check: Verify that an index allocation exists. */ |
241 | if (!NInoIndexAllocPresent(ni: idx_ni)) { |
242 | ntfs_error(sb, "No index allocation attribute but index entry " |
243 | "requires one. Inode 0x%lx is corrupt or " |
244 | "driver bug." , idx_ni->mft_no); |
245 | goto err_out; |
246 | } |
247 | /* Get the starting vcn of the index_block holding the child node. */ |
248 | vcn = sle64_to_cpup(x: (sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8)); |
249 | ia_mapping = VFS_I(ni: idx_ni)->i_mapping; |
250 | /* |
251 | * We are done with the index root and the mft record. Release them, |
252 | * otherwise we deadlock with ntfs_map_page(). |
253 | */ |
254 | ntfs_attr_put_search_ctx(ctx: actx); |
255 | unmap_mft_record(ni: base_ni); |
256 | m = NULL; |
257 | actx = NULL; |
258 | descend_into_child_node: |
259 | /* |
260 | * Convert vcn to index into the index allocation attribute in units |
261 | * of PAGE_SIZE and map the page cache page, reading it from |
262 | * disk if necessary. |
263 | */ |
264 | page = ntfs_map_page(mapping: ia_mapping, index: vcn << |
265 | idx_ni->itype.index.vcn_size_bits >> PAGE_SHIFT); |
266 | if (IS_ERR(ptr: page)) { |
267 | ntfs_error(sb, "Failed to map index page, error %ld." , |
268 | -PTR_ERR(page)); |
269 | err = PTR_ERR(ptr: page); |
270 | goto err_out; |
271 | } |
272 | lock_page(page); |
273 | kaddr = (u8*)page_address(page); |
274 | fast_descend_into_child_node: |
275 | /* Get to the index allocation block. */ |
276 | ia = (INDEX_ALLOCATION*)(kaddr + ((vcn << |
277 | idx_ni->itype.index.vcn_size_bits) & ~PAGE_MASK)); |
278 | /* Bounds checks. */ |
279 | if ((u8*)ia < kaddr || (u8*)ia > kaddr + PAGE_SIZE) { |
280 | ntfs_error(sb, "Out of bounds check failed. Corrupt inode " |
281 | "0x%lx or driver bug." , idx_ni->mft_no); |
282 | goto unm_err_out; |
283 | } |
284 | /* Catch multi sector transfer fixup errors. */ |
285 | if (unlikely(!ntfs_is_indx_record(ia->magic))) { |
286 | ntfs_error(sb, "Index record with vcn 0x%llx is corrupt. " |
287 | "Corrupt inode 0x%lx. Run chkdsk." , |
288 | (long long)vcn, idx_ni->mft_no); |
289 | goto unm_err_out; |
290 | } |
291 | if (sle64_to_cpu(x: ia->index_block_vcn) != vcn) { |
292 | ntfs_error(sb, "Actual VCN (0x%llx) of index buffer is " |
293 | "different from expected VCN (0x%llx). Inode " |
294 | "0x%lx is corrupt or driver bug." , |
295 | (unsigned long long) |
296 | sle64_to_cpu(ia->index_block_vcn), |
297 | (unsigned long long)vcn, idx_ni->mft_no); |
298 | goto unm_err_out; |
299 | } |
300 | if (le32_to_cpu(ia->index.allocated_size) + 0x18 != |
301 | idx_ni->itype.index.block_size) { |
302 | ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx has " |
303 | "a size (%u) differing from the index " |
304 | "specified size (%u). Inode is corrupt or " |
305 | "driver bug." , (unsigned long long)vcn, |
306 | idx_ni->mft_no, |
307 | le32_to_cpu(ia->index.allocated_size) + 0x18, |
308 | idx_ni->itype.index.block_size); |
309 | goto unm_err_out; |
310 | } |
311 | index_end = (u8*)ia + idx_ni->itype.index.block_size; |
312 | if (index_end > kaddr + PAGE_SIZE) { |
313 | ntfs_error(sb, "Index buffer (VCN 0x%llx) of inode 0x%lx " |
314 | "crosses page boundary. Impossible! Cannot " |
315 | "access! This is probably a bug in the " |
316 | "driver." , (unsigned long long)vcn, |
317 | idx_ni->mft_no); |
318 | goto unm_err_out; |
319 | } |
320 | index_end = (u8*)&ia->index + le32_to_cpu(ia->index.index_length); |
321 | if (index_end > (u8*)ia + idx_ni->itype.index.block_size) { |
322 | ntfs_error(sb, "Size of index buffer (VCN 0x%llx) of inode " |
323 | "0x%lx exceeds maximum size." , |
324 | (unsigned long long)vcn, idx_ni->mft_no); |
325 | goto unm_err_out; |
326 | } |
327 | /* The first index entry. */ |
328 | ie = (INDEX_ENTRY*)((u8*)&ia->index + |
329 | le32_to_cpu(ia->index.entries_offset)); |
330 | /* |
331 | * Iterate similar to above big loop but applied to index buffer, thus |
332 | * loop until we exceed valid memory (corruption case) or until we |
333 | * reach the last entry. |
334 | */ |
335 | for (;; ie = (INDEX_ENTRY*)((u8*)ie + le16_to_cpu(ie->length))) { |
336 | /* Bounds checks. */ |
337 | if ((u8*)ie < (u8*)ia || (u8*)ie + |
338 | sizeof(INDEX_ENTRY_HEADER) > index_end || |
339 | (u8*)ie + le16_to_cpu(ie->length) > index_end) { |
340 | ntfs_error(sb, "Index entry out of bounds in inode " |
341 | "0x%lx." , idx_ni->mft_no); |
342 | goto unm_err_out; |
343 | } |
344 | /* |
345 | * The last entry cannot contain a key. It can however contain |
346 | * a pointer to a child node in the B+tree so we just break out. |
347 | */ |
348 | if (ie->flags & INDEX_ENTRY_END) |
349 | break; |
350 | /* Further bounds checks. */ |
351 | if ((u32)sizeof(INDEX_ENTRY_HEADER) + |
352 | le16_to_cpu(ie->key_length) > |
353 | le16_to_cpu(ie->data.vi.data_offset) || |
354 | (u32)le16_to_cpu(ie->data.vi.data_offset) + |
355 | le16_to_cpu(ie->data.vi.data_length) > |
356 | le16_to_cpu(ie->length)) { |
357 | ntfs_error(sb, "Index entry out of bounds in inode " |
358 | "0x%lx." , idx_ni->mft_no); |
359 | goto unm_err_out; |
360 | } |
361 | /* If the keys match perfectly, we setup @ictx and return 0. */ |
362 | if ((key_len == le16_to_cpu(ie->key_length)) && !memcmp(p: key, |
363 | q: &ie->key, size: key_len)) { |
364 | ia_done: |
365 | ictx->is_in_root = false; |
366 | ictx->actx = NULL; |
367 | ictx->base_ni = NULL; |
368 | ictx->ia = ia; |
369 | ictx->page = page; |
370 | goto done; |
371 | } |
372 | /* |
373 | * Not a perfect match, need to do full blown collation so we |
374 | * know which way in the B+tree we have to go. |
375 | */ |
376 | rc = ntfs_collate(vol, cr: idx_ni->itype.index.collation_rule, data1: key, |
377 | data1_len: key_len, data2: &ie->key, le16_to_cpu(ie->key_length)); |
378 | /* |
379 | * If @key collates before the key of the current entry, there |
380 | * is definitely no such key in this index but we might need to |
381 | * descend into the B+tree so we just break out of the loop. |
382 | */ |
383 | if (rc == -1) |
384 | break; |
385 | /* |
386 | * A match should never happen as the memcmp() call should have |
387 | * cought it, but we still treat it correctly. |
388 | */ |
389 | if (!rc) |
390 | goto ia_done; |
391 | /* The keys are not equal, continue the search. */ |
392 | } |
393 | /* |
394 | * We have finished with this index buffer without success. Check for |
395 | * the presence of a child node and if not present return -ENOENT. |
396 | */ |
397 | if (!(ie->flags & INDEX_ENTRY_NODE)) { |
398 | ntfs_debug("Entry not found." ); |
399 | err = -ENOENT; |
400 | goto ia_done; |
401 | } |
402 | if ((ia->index.flags & NODE_MASK) == LEAF_NODE) { |
403 | ntfs_error(sb, "Index entry with child node found in a leaf " |
404 | "node in inode 0x%lx." , idx_ni->mft_no); |
405 | goto unm_err_out; |
406 | } |
407 | /* Child node present, descend into it. */ |
408 | old_vcn = vcn; |
409 | vcn = sle64_to_cpup(x: (sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8)); |
410 | if (vcn >= 0) { |
411 | /* |
412 | * If vcn is in the same page cache page as old_vcn we recycle |
413 | * the mapped page. |
414 | */ |
415 | if (old_vcn << vol->cluster_size_bits >> |
416 | PAGE_SHIFT == vcn << |
417 | vol->cluster_size_bits >> |
418 | PAGE_SHIFT) |
419 | goto fast_descend_into_child_node; |
420 | unlock_page(page); |
421 | ntfs_unmap_page(page); |
422 | goto descend_into_child_node; |
423 | } |
424 | ntfs_error(sb, "Negative child node vcn in inode 0x%lx." , |
425 | idx_ni->mft_no); |
426 | unm_err_out: |
427 | unlock_page(page); |
428 | ntfs_unmap_page(page); |
429 | err_out: |
430 | if (!err) |
431 | err = -EIO; |
432 | if (actx) |
433 | ntfs_attr_put_search_ctx(ctx: actx); |
434 | if (m) |
435 | unmap_mft_record(ni: base_ni); |
436 | return err; |
437 | idx_err_out: |
438 | ntfs_error(sb, "Corrupt index. Aborting lookup." ); |
439 | goto err_out; |
440 | } |
441 | |