1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * Copyright (c) 2013 Red Hat, Inc. |
5 | * All Rights Reserved. |
6 | */ |
7 | #include "xfs.h" |
8 | #include "xfs_fs.h" |
9 | #include "xfs_shared.h" |
10 | #include "xfs_format.h" |
11 | #include "xfs_log_format.h" |
12 | #include "xfs_trans_resv.h" |
13 | #include "xfs_mount.h" |
14 | #include "xfs_da_format.h" |
15 | #include "xfs_inode.h" |
16 | #include "xfs_trans.h" |
17 | #include "xfs_bmap.h" |
18 | #include "xfs_da_btree.h" |
19 | #include "xfs_attr.h" |
20 | #include "xfs_attr_sf.h" |
21 | #include "xfs_attr_leaf.h" |
22 | #include "xfs_error.h" |
23 | #include "xfs_trace.h" |
24 | #include "xfs_dir2.h" |
25 | |
26 | STATIC int |
27 | xfs_attr_shortform_compare(const void *a, const void *b) |
28 | { |
29 | xfs_attr_sf_sort_t *sa, *sb; |
30 | |
31 | sa = (xfs_attr_sf_sort_t *)a; |
32 | sb = (xfs_attr_sf_sort_t *)b; |
33 | if (sa->hash < sb->hash) { |
34 | return -1; |
35 | } else if (sa->hash > sb->hash) { |
36 | return 1; |
37 | } else { |
38 | return sa->entno - sb->entno; |
39 | } |
40 | } |
41 | |
42 | #define XFS_ISRESET_CURSOR(cursor) \ |
43 | (!((cursor)->initted) && !((cursor)->hashval) && \ |
44 | !((cursor)->blkno) && !((cursor)->offset)) |
45 | /* |
46 | * Copy out entries of shortform attribute lists for attr_list(). |
47 | * Shortform attribute lists are not stored in hashval sorted order. |
48 | * If the output buffer is not large enough to hold them all, then |
49 | * we have to calculate each entries' hashvalue and sort them before |
50 | * we can begin returning them to the user. |
51 | */ |
52 | static int |
53 | xfs_attr_shortform_list( |
54 | struct xfs_attr_list_context *context) |
55 | { |
56 | struct xfs_attrlist_cursor_kern *cursor = &context->cursor; |
57 | struct xfs_inode *dp = context->dp; |
58 | struct xfs_attr_sf_sort *sbuf, *sbp; |
59 | struct xfs_attr_shortform *sf; |
60 | struct xfs_attr_sf_entry *sfe; |
61 | int sbsize, nsbuf, count, i; |
62 | int error = 0; |
63 | |
64 | sf = (struct xfs_attr_shortform *)dp->i_af.if_u1.if_data; |
65 | ASSERT(sf != NULL); |
66 | if (!sf->hdr.count) |
67 | return 0; |
68 | |
69 | trace_xfs_attr_list_sf(ctx: context); |
70 | |
71 | /* |
72 | * If the buffer is large enough and the cursor is at the start, |
73 | * do not bother with sorting since we will return everything in |
74 | * one buffer and another call using the cursor won't need to be |
75 | * made. |
76 | * Note the generous fudge factor of 16 overhead bytes per entry. |
77 | * If bufsize is zero then put_listent must be a search function |
78 | * and can just scan through what we have. |
79 | */ |
80 | if (context->bufsize == 0 || |
81 | (XFS_ISRESET_CURSOR(cursor) && |
82 | (dp->i_af.if_bytes + sf->hdr.count * 16) < context->bufsize)) { |
83 | for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) { |
84 | if (XFS_IS_CORRUPT(context->dp->i_mount, |
85 | !xfs_attr_namecheck(sfe->nameval, |
86 | sfe->namelen))) |
87 | return -EFSCORRUPTED; |
88 | context->put_listent(context, |
89 | sfe->flags, |
90 | sfe->nameval, |
91 | (int)sfe->namelen, |
92 | (int)sfe->valuelen); |
93 | /* |
94 | * Either search callback finished early or |
95 | * didn't fit it all in the buffer after all. |
96 | */ |
97 | if (context->seen_enough) |
98 | break; |
99 | sfe = xfs_attr_sf_nextentry(sfe); |
100 | } |
101 | trace_xfs_attr_list_sf_all(ctx: context); |
102 | return 0; |
103 | } |
104 | |
105 | /* do no more for a search callback */ |
106 | if (context->bufsize == 0) |
107 | return 0; |
108 | |
109 | /* |
110 | * It didn't all fit, so we have to sort everything on hashval. |
111 | */ |
112 | sbsize = sf->hdr.count * sizeof(*sbuf); |
113 | sbp = sbuf = kmem_alloc(sbsize, KM_NOFS); |
114 | |
115 | /* |
116 | * Scan the attribute list for the rest of the entries, storing |
117 | * the relevant info from only those that match into a buffer. |
118 | */ |
119 | nsbuf = 0; |
120 | for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) { |
121 | if (unlikely( |
122 | ((char *)sfe < (char *)sf) || |
123 | ((char *)sfe >= ((char *)sf + dp->i_af.if_bytes)))) { |
124 | XFS_CORRUPTION_ERROR("xfs_attr_shortform_list" , |
125 | XFS_ERRLEVEL_LOW, |
126 | context->dp->i_mount, sfe, |
127 | sizeof(*sfe)); |
128 | kmem_free(ptr: sbuf); |
129 | return -EFSCORRUPTED; |
130 | } |
131 | |
132 | sbp->entno = i; |
133 | sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen); |
134 | sbp->name = sfe->nameval; |
135 | sbp->namelen = sfe->namelen; |
136 | /* These are bytes, and both on-disk, don't endian-flip */ |
137 | sbp->valuelen = sfe->valuelen; |
138 | sbp->flags = sfe->flags; |
139 | sfe = xfs_attr_sf_nextentry(sfe); |
140 | sbp++; |
141 | nsbuf++; |
142 | } |
143 | |
144 | /* |
145 | * Sort the entries on hash then entno. |
146 | */ |
147 | xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare); |
148 | |
149 | /* |
150 | * Re-find our place IN THE SORTED LIST. |
151 | */ |
152 | count = 0; |
153 | cursor->initted = 1; |
154 | cursor->blkno = 0; |
155 | for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) { |
156 | if (sbp->hash == cursor->hashval) { |
157 | if (cursor->offset == count) { |
158 | break; |
159 | } |
160 | count++; |
161 | } else if (sbp->hash > cursor->hashval) { |
162 | break; |
163 | } |
164 | } |
165 | if (i == nsbuf) |
166 | goto out; |
167 | |
168 | /* |
169 | * Loop putting entries into the user buffer. |
170 | */ |
171 | for ( ; i < nsbuf; i++, sbp++) { |
172 | if (cursor->hashval != sbp->hash) { |
173 | cursor->hashval = sbp->hash; |
174 | cursor->offset = 0; |
175 | } |
176 | if (XFS_IS_CORRUPT(context->dp->i_mount, |
177 | !xfs_attr_namecheck(sbp->name, |
178 | sbp->namelen))) { |
179 | error = -EFSCORRUPTED; |
180 | goto out; |
181 | } |
182 | context->put_listent(context, |
183 | sbp->flags, |
184 | sbp->name, |
185 | sbp->namelen, |
186 | sbp->valuelen); |
187 | if (context->seen_enough) |
188 | break; |
189 | cursor->offset++; |
190 | } |
191 | out: |
192 | kmem_free(ptr: sbuf); |
193 | return error; |
194 | } |
195 | |
196 | /* |
197 | * We didn't find the block & hash mentioned in the cursor state, so |
198 | * walk down the attr btree looking for the hash. |
199 | */ |
200 | STATIC int |
201 | xfs_attr_node_list_lookup( |
202 | struct xfs_attr_list_context *context, |
203 | struct xfs_attrlist_cursor_kern *cursor, |
204 | struct xfs_buf **pbp) |
205 | { |
206 | struct xfs_da3_icnode_hdr nodehdr; |
207 | struct xfs_da_intnode *node; |
208 | struct xfs_da_node_entry *btree; |
209 | struct xfs_inode *dp = context->dp; |
210 | struct xfs_mount *mp = dp->i_mount; |
211 | struct xfs_trans *tp = context->tp; |
212 | struct xfs_buf *bp; |
213 | int i; |
214 | int error = 0; |
215 | unsigned int expected_level = 0; |
216 | uint16_t magic; |
217 | |
218 | ASSERT(*pbp == NULL); |
219 | cursor->blkno = 0; |
220 | for (;;) { |
221 | error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp, |
222 | XFS_ATTR_FORK); |
223 | if (error) |
224 | return error; |
225 | node = bp->b_addr; |
226 | magic = be16_to_cpu(node->hdr.info.magic); |
227 | if (magic == XFS_ATTR_LEAF_MAGIC || |
228 | magic == XFS_ATTR3_LEAF_MAGIC) |
229 | break; |
230 | if (magic != XFS_DA_NODE_MAGIC && |
231 | magic != XFS_DA3_NODE_MAGIC) { |
232 | XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp, |
233 | node, sizeof(*node)); |
234 | goto out_corruptbuf; |
235 | } |
236 | |
237 | xfs_da3_node_hdr_from_disk(mp, &nodehdr, node); |
238 | |
239 | /* Tree taller than we can handle; bail out! */ |
240 | if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH) |
241 | goto out_corruptbuf; |
242 | |
243 | /* Check the level from the root node. */ |
244 | if (cursor->blkno == 0) |
245 | expected_level = nodehdr.level - 1; |
246 | else if (expected_level != nodehdr.level) |
247 | goto out_corruptbuf; |
248 | else |
249 | expected_level--; |
250 | |
251 | btree = nodehdr.btree; |
252 | for (i = 0; i < nodehdr.count; btree++, i++) { |
253 | if (cursor->hashval <= be32_to_cpu(btree->hashval)) { |
254 | cursor->blkno = be32_to_cpu(btree->before); |
255 | trace_xfs_attr_list_node_descend(ctx: context, |
256 | btree); |
257 | break; |
258 | } |
259 | } |
260 | xfs_trans_brelse(tp, bp); |
261 | |
262 | if (i == nodehdr.count) |
263 | return 0; |
264 | |
265 | /* We can't point back to the root. */ |
266 | if (XFS_IS_CORRUPT(mp, cursor->blkno == 0)) |
267 | return -EFSCORRUPTED; |
268 | } |
269 | |
270 | if (expected_level != 0) |
271 | goto out_corruptbuf; |
272 | |
273 | *pbp = bp; |
274 | return 0; |
275 | |
276 | out_corruptbuf: |
277 | xfs_buf_mark_corrupt(bp); |
278 | xfs_trans_brelse(tp, bp); |
279 | return -EFSCORRUPTED; |
280 | } |
281 | |
282 | STATIC int |
283 | xfs_attr_node_list( |
284 | struct xfs_attr_list_context *context) |
285 | { |
286 | struct xfs_attrlist_cursor_kern *cursor = &context->cursor; |
287 | struct xfs_attr3_icleaf_hdr leafhdr; |
288 | struct xfs_attr_leafblock *leaf; |
289 | struct xfs_da_intnode *node; |
290 | struct xfs_buf *bp; |
291 | struct xfs_inode *dp = context->dp; |
292 | struct xfs_mount *mp = dp->i_mount; |
293 | int error = 0; |
294 | |
295 | trace_xfs_attr_node_list(ctx: context); |
296 | |
297 | cursor->initted = 1; |
298 | |
299 | /* |
300 | * Do all sorts of validation on the passed-in cursor structure. |
301 | * If anything is amiss, ignore the cursor and look up the hashval |
302 | * starting from the btree root. |
303 | */ |
304 | bp = NULL; |
305 | if (cursor->blkno > 0) { |
306 | error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp, |
307 | XFS_ATTR_FORK); |
308 | if ((error != 0) && (error != -EFSCORRUPTED)) |
309 | return error; |
310 | if (bp) { |
311 | struct xfs_attr_leaf_entry *entries; |
312 | |
313 | node = bp->b_addr; |
314 | switch (be16_to_cpu(node->hdr.info.magic)) { |
315 | case XFS_DA_NODE_MAGIC: |
316 | case XFS_DA3_NODE_MAGIC: |
317 | trace_xfs_attr_list_wrong_blk(ctx: context); |
318 | xfs_trans_brelse(context->tp, bp); |
319 | bp = NULL; |
320 | break; |
321 | case XFS_ATTR_LEAF_MAGIC: |
322 | case XFS_ATTR3_LEAF_MAGIC: |
323 | leaf = bp->b_addr; |
324 | xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, |
325 | &leafhdr, leaf); |
326 | entries = xfs_attr3_leaf_entryp(leaf); |
327 | if (cursor->hashval > be32_to_cpu( |
328 | entries[leafhdr.count - 1].hashval)) { |
329 | trace_xfs_attr_list_wrong_blk(ctx: context); |
330 | xfs_trans_brelse(context->tp, bp); |
331 | bp = NULL; |
332 | } else if (cursor->hashval <= be32_to_cpu( |
333 | entries[0].hashval)) { |
334 | trace_xfs_attr_list_wrong_blk(ctx: context); |
335 | xfs_trans_brelse(context->tp, bp); |
336 | bp = NULL; |
337 | } |
338 | break; |
339 | default: |
340 | trace_xfs_attr_list_wrong_blk(ctx: context); |
341 | xfs_trans_brelse(context->tp, bp); |
342 | bp = NULL; |
343 | } |
344 | } |
345 | } |
346 | |
347 | /* |
348 | * We did not find what we expected given the cursor's contents, |
349 | * so we start from the top and work down based on the hash value. |
350 | * Note that start of node block is same as start of leaf block. |
351 | */ |
352 | if (bp == NULL) { |
353 | error = xfs_attr_node_list_lookup(context, cursor, pbp: &bp); |
354 | if (error || !bp) |
355 | return error; |
356 | } |
357 | ASSERT(bp != NULL); |
358 | |
359 | /* |
360 | * Roll upward through the blocks, processing each leaf block in |
361 | * order. As long as there is space in the result buffer, keep |
362 | * adding the information. |
363 | */ |
364 | for (;;) { |
365 | leaf = bp->b_addr; |
366 | error = xfs_attr3_leaf_list_int(bp, context); |
367 | if (error) |
368 | break; |
369 | xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf); |
370 | if (context->seen_enough || leafhdr.forw == 0) |
371 | break; |
372 | cursor->blkno = leafhdr.forw; |
373 | xfs_trans_brelse(context->tp, bp); |
374 | error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, |
375 | &bp); |
376 | if (error) |
377 | return error; |
378 | } |
379 | xfs_trans_brelse(context->tp, bp); |
380 | return error; |
381 | } |
382 | |
383 | /* |
384 | * Copy out attribute list entries for attr_list(), for leaf attribute lists. |
385 | */ |
386 | int |
387 | xfs_attr3_leaf_list_int( |
388 | struct xfs_buf *bp, |
389 | struct xfs_attr_list_context *context) |
390 | { |
391 | struct xfs_attrlist_cursor_kern *cursor = &context->cursor; |
392 | struct xfs_attr_leafblock *leaf; |
393 | struct xfs_attr3_icleaf_hdr ichdr; |
394 | struct xfs_attr_leaf_entry *entries; |
395 | struct xfs_attr_leaf_entry *entry; |
396 | int i; |
397 | struct xfs_mount *mp = context->dp->i_mount; |
398 | |
399 | trace_xfs_attr_list_leaf(ctx: context); |
400 | |
401 | leaf = bp->b_addr; |
402 | xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf); |
403 | entries = xfs_attr3_leaf_entryp(leaf); |
404 | |
405 | cursor->initted = 1; |
406 | |
407 | /* |
408 | * Re-find our place in the leaf block if this is a new syscall. |
409 | */ |
410 | if (context->resynch) { |
411 | entry = &entries[0]; |
412 | for (i = 0; i < ichdr.count; entry++, i++) { |
413 | if (be32_to_cpu(entry->hashval) == cursor->hashval) { |
414 | if (cursor->offset == context->dupcnt) { |
415 | context->dupcnt = 0; |
416 | break; |
417 | } |
418 | context->dupcnt++; |
419 | } else if (be32_to_cpu(entry->hashval) > |
420 | cursor->hashval) { |
421 | context->dupcnt = 0; |
422 | break; |
423 | } |
424 | } |
425 | if (i == ichdr.count) { |
426 | trace_xfs_attr_list_notfound(ctx: context); |
427 | return 0; |
428 | } |
429 | } else { |
430 | entry = &entries[0]; |
431 | i = 0; |
432 | } |
433 | context->resynch = 0; |
434 | |
435 | /* |
436 | * We have found our place, start copying out the new attributes. |
437 | */ |
438 | for (; i < ichdr.count; entry++, i++) { |
439 | char *name; |
440 | int namelen, valuelen; |
441 | |
442 | if (be32_to_cpu(entry->hashval) != cursor->hashval) { |
443 | cursor->hashval = be32_to_cpu(entry->hashval); |
444 | cursor->offset = 0; |
445 | } |
446 | |
447 | if ((entry->flags & XFS_ATTR_INCOMPLETE) && |
448 | !context->allow_incomplete) |
449 | continue; |
450 | |
451 | if (entry->flags & XFS_ATTR_LOCAL) { |
452 | xfs_attr_leaf_name_local_t *name_loc; |
453 | |
454 | name_loc = xfs_attr3_leaf_name_local(leaf, i); |
455 | name = name_loc->nameval; |
456 | namelen = name_loc->namelen; |
457 | valuelen = be16_to_cpu(name_loc->valuelen); |
458 | } else { |
459 | xfs_attr_leaf_name_remote_t *name_rmt; |
460 | |
461 | name_rmt = xfs_attr3_leaf_name_remote(leaf, i); |
462 | name = name_rmt->name; |
463 | namelen = name_rmt->namelen; |
464 | valuelen = be32_to_cpu(name_rmt->valuelen); |
465 | } |
466 | |
467 | if (XFS_IS_CORRUPT(context->dp->i_mount, |
468 | !xfs_attr_namecheck(name, namelen))) |
469 | return -EFSCORRUPTED; |
470 | context->put_listent(context, entry->flags, |
471 | name, namelen, valuelen); |
472 | if (context->seen_enough) |
473 | break; |
474 | cursor->offset++; |
475 | } |
476 | trace_xfs_attr_list_leaf_end(ctx: context); |
477 | return 0; |
478 | } |
479 | |
480 | /* |
481 | * Copy out attribute entries for attr_list(), for leaf attribute lists. |
482 | */ |
483 | STATIC int |
484 | xfs_attr_leaf_list( |
485 | struct xfs_attr_list_context *context) |
486 | { |
487 | struct xfs_buf *bp; |
488 | int error; |
489 | |
490 | trace_xfs_attr_leaf_list(ctx: context); |
491 | |
492 | context->cursor.blkno = 0; |
493 | error = xfs_attr3_leaf_read(context->tp, context->dp, 0, &bp); |
494 | if (error) |
495 | return error; |
496 | |
497 | error = xfs_attr3_leaf_list_int(bp, context); |
498 | xfs_trans_brelse(context->tp, bp); |
499 | return error; |
500 | } |
501 | |
502 | int |
503 | xfs_attr_list_ilocked( |
504 | struct xfs_attr_list_context *context) |
505 | { |
506 | struct xfs_inode *dp = context->dp; |
507 | |
508 | ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL)); |
509 | |
510 | /* |
511 | * Decide on what work routines to call based on the inode size. |
512 | */ |
513 | if (!xfs_inode_hasattr(dp)) |
514 | return 0; |
515 | if (dp->i_af.if_format == XFS_DINODE_FMT_LOCAL) |
516 | return xfs_attr_shortform_list(context); |
517 | if (xfs_attr_is_leaf(dp)) |
518 | return xfs_attr_leaf_list(context); |
519 | return xfs_attr_node_list(context); |
520 | } |
521 | |
522 | int |
523 | xfs_attr_list( |
524 | struct xfs_attr_list_context *context) |
525 | { |
526 | struct xfs_inode *dp = context->dp; |
527 | uint lock_mode; |
528 | int error; |
529 | |
530 | XFS_STATS_INC(dp->i_mount, xs_attr_list); |
531 | |
532 | if (xfs_is_shutdown(mp: dp->i_mount)) |
533 | return -EIO; |
534 | |
535 | lock_mode = xfs_ilock_attr_map_shared(dp); |
536 | error = xfs_attr_list_ilocked(context); |
537 | xfs_iunlock(dp, lock_mode); |
538 | return error; |
539 | } |
540 | |