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
xfs_attr_shortform_compare(const void * a,const void * b)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
xfs_attr_shortform_list(struct xfs_attr_list_context * context)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(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->flags,
86 sfe->nameval,
87 sfe->namelen)))
88 return -EFSCORRUPTED;
89 context->put_listent(context,
90 sfe->flags,
91 sfe->nameval,
92 (int)sfe->namelen,
93 (int)sfe->valuelen);
94 /*
95 * Either search callback finished early or
96 * didn't fit it all in the buffer after all.
97 */
98 if (context->seen_enough)
99 break;
100 sfe = xfs_attr_sf_nextentry(sfe);
101 }
102 trace_xfs_attr_list_sf_all(context);
103 return 0;
104 }
105
106 /* do no more for a search callback */
107 if (context->bufsize == 0)
108 return 0;
109
110 /*
111 * It didn't all fit, so we have to sort everything on hashval.
112 */
113 sbsize = sf->hdr.count * sizeof(*sbuf);
114 sbp = sbuf = kmem_alloc(sbsize, KM_NOFS);
115
116 /*
117 * Scan the attribute list for the rest of the entries, storing
118 * the relevant info from only those that match into a buffer.
119 */
120 nsbuf = 0;
121 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
122 if (unlikely(
123 ((char *)sfe < (char *)sf) ||
124 ((char *)sfe >= ((char *)sf + dp->i_af.if_bytes)) ||
125 !xfs_attr_check_namespace(sfe->flags))) {
126 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
127 XFS_ERRLEVEL_LOW,
128 context->dp->i_mount, sfe,
129 sizeof(*sfe));
130 kmem_free(sbuf);
131 return -EFSCORRUPTED;
132 }
133
134 sbp->entno = i;
135 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
136 sbp->name = sfe->nameval;
137 sbp->namelen = sfe->namelen;
138 /* These are bytes, and both on-disk, don't endian-flip */
139 sbp->valuelen = sfe->valuelen;
140 sbp->flags = sfe->flags;
141 sfe = xfs_attr_sf_nextentry(sfe);
142 sbp++;
143 nsbuf++;
144 }
145
146 /*
147 * Sort the entries on hash then entno.
148 */
149 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
150
151 /*
152 * Re-find our place IN THE SORTED LIST.
153 */
154 count = 0;
155 cursor->initted = 1;
156 cursor->blkno = 0;
157 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
158 if (sbp->hash == cursor->hashval) {
159 if (cursor->offset == count) {
160 break;
161 }
162 count++;
163 } else if (sbp->hash > cursor->hashval) {
164 break;
165 }
166 }
167 if (i == nsbuf)
168 goto out;
169
170 /*
171 * Loop putting entries into the user buffer.
172 */
173 for ( ; i < nsbuf; i++, sbp++) {
174 if (cursor->hashval != sbp->hash) {
175 cursor->hashval = sbp->hash;
176 cursor->offset = 0;
177 }
178 if (XFS_IS_CORRUPT(context->dp->i_mount,
179 !xfs_attr_namecheck(sbp->flags, sbp->name,
180 sbp->namelen))) {
181 error = -EFSCORRUPTED;
182 goto out;
183 }
184 context->put_listent(context,
185 sbp->flags,
186 sbp->name,
187 sbp->namelen,
188 sbp->valuelen);
189 if (context->seen_enough)
190 break;
191 cursor->offset++;
192 }
193 out:
194 kmem_free(sbuf);
195 return error;
196 }
197
198 /*
199 * We didn't find the block & hash mentioned in the cursor state, so
200 * walk down the attr btree looking for the hash.
201 */
202 STATIC int
xfs_attr_node_list_lookup(struct xfs_attr_list_context * context,struct xfs_attrlist_cursor_kern * cursor,struct xfs_buf ** pbp)203 xfs_attr_node_list_lookup(
204 struct xfs_attr_list_context *context,
205 struct xfs_attrlist_cursor_kern *cursor,
206 struct xfs_buf **pbp)
207 {
208 struct xfs_da3_icnode_hdr nodehdr;
209 struct xfs_da_intnode *node;
210 struct xfs_da_node_entry *btree;
211 struct xfs_inode *dp = context->dp;
212 struct xfs_mount *mp = dp->i_mount;
213 struct xfs_trans *tp = context->tp;
214 struct xfs_buf *bp;
215 int i;
216 int error = 0;
217 unsigned int expected_level = 0;
218 uint16_t magic;
219
220 ASSERT(*pbp == NULL);
221 cursor->blkno = 0;
222 for (;;) {
223 error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp,
224 XFS_ATTR_FORK);
225 if (error)
226 return error;
227 node = bp->b_addr;
228 magic = be16_to_cpu(node->hdr.info.magic);
229 if (magic == XFS_ATTR_LEAF_MAGIC ||
230 magic == XFS_ATTR3_LEAF_MAGIC)
231 break;
232 if (magic != XFS_DA_NODE_MAGIC &&
233 magic != XFS_DA3_NODE_MAGIC) {
234 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
235 node, sizeof(*node));
236 goto out_corruptbuf;
237 }
238
239 xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
240
241 /* Tree taller than we can handle; bail out! */
242 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
243 goto out_corruptbuf;
244
245 /* Check the level from the root node. */
246 if (cursor->blkno == 0)
247 expected_level = nodehdr.level - 1;
248 else if (expected_level != nodehdr.level)
249 goto out_corruptbuf;
250 else
251 expected_level--;
252
253 btree = nodehdr.btree;
254 for (i = 0; i < nodehdr.count; btree++, i++) {
255 if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
256 cursor->blkno = be32_to_cpu(btree->before);
257 trace_xfs_attr_list_node_descend(context,
258 btree);
259 break;
260 }
261 }
262 xfs_trans_brelse(tp, bp);
263
264 if (i == nodehdr.count)
265 return 0;
266
267 /* We can't point back to the root. */
268 if (XFS_IS_CORRUPT(mp, cursor->blkno == 0))
269 return -EFSCORRUPTED;
270 }
271
272 if (expected_level != 0)
273 goto out_corruptbuf;
274
275 *pbp = bp;
276 return 0;
277
278 out_corruptbuf:
279 xfs_buf_mark_corrupt(bp);
280 xfs_trans_brelse(tp, bp);
281 return -EFSCORRUPTED;
282 }
283
284 STATIC int
xfs_attr_node_list(struct xfs_attr_list_context * context)285 xfs_attr_node_list(
286 struct xfs_attr_list_context *context)
287 {
288 struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
289 struct xfs_attr3_icleaf_hdr leafhdr;
290 struct xfs_attr_leafblock *leaf;
291 struct xfs_da_intnode *node;
292 struct xfs_buf *bp;
293 struct xfs_inode *dp = context->dp;
294 struct xfs_mount *mp = dp->i_mount;
295 int error = 0;
296
297 trace_xfs_attr_node_list(context);
298
299 cursor->initted = 1;
300
301 /*
302 * Do all sorts of validation on the passed-in cursor structure.
303 * If anything is amiss, ignore the cursor and look up the hashval
304 * starting from the btree root.
305 */
306 bp = NULL;
307 if (cursor->blkno > 0) {
308 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp,
309 XFS_ATTR_FORK);
310 if ((error != 0) && (error != -EFSCORRUPTED))
311 return error;
312 if (bp) {
313 struct xfs_attr_leaf_entry *entries;
314
315 node = bp->b_addr;
316 switch (be16_to_cpu(node->hdr.info.magic)) {
317 case XFS_DA_NODE_MAGIC:
318 case XFS_DA3_NODE_MAGIC:
319 trace_xfs_attr_list_wrong_blk(context);
320 xfs_trans_brelse(context->tp, bp);
321 bp = NULL;
322 break;
323 case XFS_ATTR_LEAF_MAGIC:
324 case XFS_ATTR3_LEAF_MAGIC:
325 leaf = bp->b_addr;
326 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
327 &leafhdr, leaf);
328 entries = xfs_attr3_leaf_entryp(leaf);
329 if (cursor->hashval > be32_to_cpu(
330 entries[leafhdr.count - 1].hashval)) {
331 trace_xfs_attr_list_wrong_blk(context);
332 xfs_trans_brelse(context->tp, bp);
333 bp = NULL;
334 } else if (cursor->hashval <= be32_to_cpu(
335 entries[0].hashval)) {
336 trace_xfs_attr_list_wrong_blk(context);
337 xfs_trans_brelse(context->tp, bp);
338 bp = NULL;
339 }
340 break;
341 default:
342 trace_xfs_attr_list_wrong_blk(context);
343 xfs_trans_brelse(context->tp, bp);
344 bp = NULL;
345 }
346 }
347 }
348
349 /*
350 * We did not find what we expected given the cursor's contents,
351 * so we start from the top and work down based on the hash value.
352 * Note that start of node block is same as start of leaf block.
353 */
354 if (bp == NULL) {
355 error = xfs_attr_node_list_lookup(context, cursor, &bp);
356 if (error || !bp)
357 return error;
358 }
359 ASSERT(bp != NULL);
360
361 /*
362 * Roll upward through the blocks, processing each leaf block in
363 * order. As long as there is space in the result buffer, keep
364 * adding the information.
365 */
366 for (;;) {
367 leaf = bp->b_addr;
368 error = xfs_attr3_leaf_list_int(bp, context);
369 if (error)
370 break;
371 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
372 if (context->seen_enough || leafhdr.forw == 0)
373 break;
374 cursor->blkno = leafhdr.forw;
375 xfs_trans_brelse(context->tp, bp);
376 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno,
377 &bp);
378 if (error)
379 return error;
380 }
381 xfs_trans_brelse(context->tp, bp);
382 return error;
383 }
384
385 /*
386 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
387 */
388 int
xfs_attr3_leaf_list_int(struct xfs_buf * bp,struct xfs_attr_list_context * context)389 xfs_attr3_leaf_list_int(
390 struct xfs_buf *bp,
391 struct xfs_attr_list_context *context)
392 {
393 struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
394 struct xfs_attr_leafblock *leaf;
395 struct xfs_attr3_icleaf_hdr ichdr;
396 struct xfs_attr_leaf_entry *entries;
397 struct xfs_attr_leaf_entry *entry;
398 int i;
399 struct xfs_mount *mp = context->dp->i_mount;
400
401 trace_xfs_attr_list_leaf(context);
402
403 leaf = bp->b_addr;
404 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
405 entries = xfs_attr3_leaf_entryp(leaf);
406
407 cursor->initted = 1;
408
409 /*
410 * Re-find our place in the leaf block if this is a new syscall.
411 */
412 if (context->resynch) {
413 entry = &entries[0];
414 for (i = 0; i < ichdr.count; entry++, i++) {
415 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
416 if (cursor->offset == context->dupcnt) {
417 context->dupcnt = 0;
418 break;
419 }
420 context->dupcnt++;
421 } else if (be32_to_cpu(entry->hashval) >
422 cursor->hashval) {
423 context->dupcnt = 0;
424 break;
425 }
426 }
427 if (i == ichdr.count) {
428 trace_xfs_attr_list_notfound(context);
429 return 0;
430 }
431 } else {
432 entry = &entries[0];
433 i = 0;
434 }
435 context->resynch = 0;
436
437 /*
438 * We have found our place, start copying out the new attributes.
439 */
440 for (; i < ichdr.count; entry++, i++) {
441 char *name;
442 int namelen, valuelen;
443
444 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
445 cursor->hashval = be32_to_cpu(entry->hashval);
446 cursor->offset = 0;
447 }
448
449 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
450 !context->allow_incomplete)
451 continue;
452
453 if (entry->flags & XFS_ATTR_LOCAL) {
454 xfs_attr_leaf_name_local_t *name_loc;
455
456 name_loc = xfs_attr3_leaf_name_local(leaf, i);
457 name = name_loc->nameval;
458 namelen = name_loc->namelen;
459 valuelen = be16_to_cpu(name_loc->valuelen);
460 } else {
461 xfs_attr_leaf_name_remote_t *name_rmt;
462
463 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
464 name = name_rmt->name;
465 namelen = name_rmt->namelen;
466 valuelen = be32_to_cpu(name_rmt->valuelen);
467 }
468
469 if (XFS_IS_CORRUPT(context->dp->i_mount,
470 !xfs_attr_namecheck(entry->flags, name,
471 namelen)))
472 return -EFSCORRUPTED;
473 context->put_listent(context, entry->flags,
474 name, namelen, valuelen);
475 if (context->seen_enough)
476 break;
477 cursor->offset++;
478 }
479 trace_xfs_attr_list_leaf_end(context);
480 return 0;
481 }
482
483 /*
484 * Copy out attribute entries for attr_list(), for leaf attribute lists.
485 */
486 STATIC int
xfs_attr_leaf_list(struct xfs_attr_list_context * context)487 xfs_attr_leaf_list(
488 struct xfs_attr_list_context *context)
489 {
490 struct xfs_buf *bp;
491 int error;
492
493 trace_xfs_attr_leaf_list(context);
494
495 context->cursor.blkno = 0;
496 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, &bp);
497 if (error)
498 return error;
499
500 error = xfs_attr3_leaf_list_int(bp, context);
501 xfs_trans_brelse(context->tp, bp);
502 return error;
503 }
504
505 int
xfs_attr_list_ilocked(struct xfs_attr_list_context * context)506 xfs_attr_list_ilocked(
507 struct xfs_attr_list_context *context)
508 {
509 struct xfs_inode *dp = context->dp;
510
511 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
512
513 /*
514 * Decide on what work routines to call based on the inode size.
515 */
516 if (!xfs_inode_hasattr(dp))
517 return 0;
518 if (dp->i_af.if_format == XFS_DINODE_FMT_LOCAL)
519 return xfs_attr_shortform_list(context);
520 if (xfs_attr_is_leaf(dp))
521 return xfs_attr_leaf_list(context);
522 return xfs_attr_node_list(context);
523 }
524
525 int
xfs_attr_list(struct xfs_attr_list_context * context)526 xfs_attr_list(
527 struct xfs_attr_list_context *context)
528 {
529 struct xfs_inode *dp = context->dp;
530 uint lock_mode;
531 int error;
532
533 XFS_STATS_INC(dp->i_mount, xs_attr_list);
534
535 if (xfs_is_shutdown(dp->i_mount))
536 return -EIO;
537
538 lock_mode = xfs_ilock_attr_map_shared(dp);
539 error = xfs_attr_list_ilocked(context);
540 xfs_iunlock(dp, lock_mode);
541 return error;
542 }
543