xref: /openbmc/linux/fs/ntfs/index.c (revision b830f94f)
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(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(ictx->actx);
49 			if (ictx->base_ni)
50 				unmap_mft_record(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(ntfs_index_ctx_cache, 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 			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(base_ni);
139 	if (IS_ERR(m)) {
140 		ntfs_error(sb, "map_mft_record() failed with error code %ld.",
141 				-PTR_ERR(m));
142 		return PTR_ERR(m);
143 	}
144 	actx = ntfs_attr_get_search_ctx(base_ni, 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(AT_INDEX_ROOT, idx_ni->name, idx_ni->name_len,
151 			CASE_SENSITIVE, 0, NULL, 0, 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(key,
193 				&ie->key, 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, idx_ni->itype.index.collation_rule, key,
214 				key_len, &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(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((sle64*)((u8*)ie + le16_to_cpu(ie->length) - 8));
249 	ia_mapping = VFS_I(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(actx);
255 	unmap_mft_record(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(ia_mapping, vcn <<
265 			idx_ni->itype.index.vcn_size_bits >> PAGE_SHIFT);
266 	if (IS_ERR(page)) {
267 		ntfs_error(sb, "Failed to map index page, error %ld.",
268 				-PTR_ERR(page));
269 		err = PTR_ERR(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(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(key,
363 				&ie->key, 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, idx_ni->itype.index.collation_rule, key,
377 				key_len, &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((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(actx);
434 	if (m)
435 		unmap_mft_record(base_ni);
436 	return err;
437 idx_err_out:
438 	ntfs_error(sb, "Corrupt index.  Aborting lookup.");
439 	goto err_out;
440 }
441