fscache: add tracepoint when failing cookie
[linux-2.6-microblaze.git] / fs / xfs / xfs_attr_list.c
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         ASSERT(dp->i_afp != NULL);
65         sf = (struct xfs_attr_shortform *)dp->i_afp->if_u1.if_data;
66         ASSERT(sf != NULL);
67         if (!sf->hdr.count)
68                 return 0;
69
70         trace_xfs_attr_list_sf(context);
71
72         /*
73          * If the buffer is large enough and the cursor is at the start,
74          * do not bother with sorting since we will return everything in
75          * one buffer and another call using the cursor won't need to be
76          * made.
77          * Note the generous fudge factor of 16 overhead bytes per entry.
78          * If bufsize is zero then put_listent must be a search function
79          * and can just scan through what we have.
80          */
81         if (context->bufsize == 0 ||
82             (XFS_ISRESET_CURSOR(cursor) &&
83              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
84                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
85                         if (XFS_IS_CORRUPT(context->dp->i_mount,
86                                            !xfs_attr_namecheck(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_afp->if_bytes)))) {
125                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
126                                              XFS_ERRLEVEL_LOW,
127                                              context->dp->i_mount, sfe,
128                                              sizeof(*sfe));
129                         kmem_free(sbuf);
130                         return -EFSCORRUPTED;
131                 }
132
133                 sbp->entno = i;
134                 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
135                 sbp->name = sfe->nameval;
136                 sbp->namelen = sfe->namelen;
137                 /* These are bytes, and both on-disk, don't endian-flip */
138                 sbp->valuelen = sfe->valuelen;
139                 sbp->flags = sfe->flags;
140                 sfe = xfs_attr_sf_nextentry(sfe);
141                 sbp++;
142                 nsbuf++;
143         }
144
145         /*
146          * Sort the entries on hash then entno.
147          */
148         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
149
150         /*
151          * Re-find our place IN THE SORTED LIST.
152          */
153         count = 0;
154         cursor->initted = 1;
155         cursor->blkno = 0;
156         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
157                 if (sbp->hash == cursor->hashval) {
158                         if (cursor->offset == count) {
159                                 break;
160                         }
161                         count++;
162                 } else if (sbp->hash > cursor->hashval) {
163                         break;
164                 }
165         }
166         if (i == nsbuf)
167                 goto out;
168
169         /*
170          * Loop putting entries into the user buffer.
171          */
172         for ( ; i < nsbuf; i++, sbp++) {
173                 if (cursor->hashval != sbp->hash) {
174                         cursor->hashval = sbp->hash;
175                         cursor->offset = 0;
176                 }
177                 if (XFS_IS_CORRUPT(context->dp->i_mount,
178                                    !xfs_attr_namecheck(sbp->name,
179                                                        sbp->namelen))) {
180                         error = -EFSCORRUPTED;
181                         goto out;
182                 }
183                 context->put_listent(context,
184                                      sbp->flags,
185                                      sbp->name,
186                                      sbp->namelen,
187                                      sbp->valuelen);
188                 if (context->seen_enough)
189                         break;
190                 cursor->offset++;
191         }
192 out:
193         kmem_free(sbuf);
194         return error;
195 }
196
197 /*
198  * We didn't find the block & hash mentioned in the cursor state, so
199  * walk down the attr btree looking for the hash.
200  */
201 STATIC int
202 xfs_attr_node_list_lookup(
203         struct xfs_attr_list_context    *context,
204         struct xfs_attrlist_cursor_kern *cursor,
205         struct xfs_buf                  **pbp)
206 {
207         struct xfs_da3_icnode_hdr       nodehdr;
208         struct xfs_da_intnode           *node;
209         struct xfs_da_node_entry        *btree;
210         struct xfs_inode                *dp = context->dp;
211         struct xfs_mount                *mp = dp->i_mount;
212         struct xfs_trans                *tp = context->tp;
213         struct xfs_buf                  *bp;
214         int                             i;
215         int                             error = 0;
216         unsigned int                    expected_level = 0;
217         uint16_t                        magic;
218
219         ASSERT(*pbp == NULL);
220         cursor->blkno = 0;
221         for (;;) {
222                 error = xfs_da3_node_read(tp, dp, cursor->blkno, &bp,
223                                 XFS_ATTR_FORK);
224                 if (error)
225                         return error;
226                 node = bp->b_addr;
227                 magic = be16_to_cpu(node->hdr.info.magic);
228                 if (magic == XFS_ATTR_LEAF_MAGIC ||
229                     magic == XFS_ATTR3_LEAF_MAGIC)
230                         break;
231                 if (magic != XFS_DA_NODE_MAGIC &&
232                     magic != XFS_DA3_NODE_MAGIC) {
233                         XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
234                                         node, sizeof(*node));
235                         goto out_corruptbuf;
236                 }
237
238                 xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
239
240                 /* Tree taller than we can handle; bail out! */
241                 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
242                         goto out_corruptbuf;
243
244                 /* Check the level from the root node. */
245                 if (cursor->blkno == 0)
246                         expected_level = nodehdr.level - 1;
247                 else if (expected_level != nodehdr.level)
248                         goto out_corruptbuf;
249                 else
250                         expected_level--;
251
252                 btree = nodehdr.btree;
253                 for (i = 0; i < nodehdr.count; btree++, i++) {
254                         if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
255                                 cursor->blkno = be32_to_cpu(btree->before);
256                                 trace_xfs_attr_list_node_descend(context,
257                                                 btree);
258                                 break;
259                         }
260                 }
261                 xfs_trans_brelse(tp, bp);
262
263                 if (i == nodehdr.count)
264                         return 0;
265
266                 /* We can't point back to the root. */
267                 if (XFS_IS_CORRUPT(mp, cursor->blkno == 0))
268                         return -EFSCORRUPTED;
269         }
270
271         if (expected_level != 0)
272                 goto out_corruptbuf;
273
274         *pbp = bp;
275         return 0;
276
277 out_corruptbuf:
278         xfs_buf_mark_corrupt(bp);
279         xfs_trans_brelse(tp, bp);
280         return -EFSCORRUPTED;
281 }
282
283 STATIC int
284 xfs_attr_node_list(
285         struct xfs_attr_list_context    *context)
286 {
287         struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
288         struct xfs_attr3_icleaf_hdr     leafhdr;
289         struct xfs_attr_leafblock       *leaf;
290         struct xfs_da_intnode           *node;
291         struct xfs_buf                  *bp;
292         struct xfs_inode                *dp = context->dp;
293         struct xfs_mount                *mp = dp->i_mount;
294         int                             error = 0;
295
296         trace_xfs_attr_node_list(context);
297
298         cursor->initted = 1;
299
300         /*
301          * Do all sorts of validation on the passed-in cursor structure.
302          * If anything is amiss, ignore the cursor and look up the hashval
303          * starting from the btree root.
304          */
305         bp = NULL;
306         if (cursor->blkno > 0) {
307                 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, &bp,
308                                 XFS_ATTR_FORK);
309                 if ((error != 0) && (error != -EFSCORRUPTED))
310                         return error;
311                 if (bp) {
312                         struct xfs_attr_leaf_entry *entries;
313
314                         node = bp->b_addr;
315                         switch (be16_to_cpu(node->hdr.info.magic)) {
316                         case XFS_DA_NODE_MAGIC:
317                         case XFS_DA3_NODE_MAGIC:
318                                 trace_xfs_attr_list_wrong_blk(context);
319                                 xfs_trans_brelse(context->tp, bp);
320                                 bp = NULL;
321                                 break;
322                         case XFS_ATTR_LEAF_MAGIC:
323                         case XFS_ATTR3_LEAF_MAGIC:
324                                 leaf = bp->b_addr;
325                                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
326                                                              &leafhdr, leaf);
327                                 entries = xfs_attr3_leaf_entryp(leaf);
328                                 if (cursor->hashval > be32_to_cpu(
329                                                 entries[leafhdr.count - 1].hashval)) {
330                                         trace_xfs_attr_list_wrong_blk(context);
331                                         xfs_trans_brelse(context->tp, bp);
332                                         bp = NULL;
333                                 } else if (cursor->hashval <= be32_to_cpu(
334                                                 entries[0].hashval)) {
335                                         trace_xfs_attr_list_wrong_blk(context);
336                                         xfs_trans_brelse(context->tp, bp);
337                                         bp = NULL;
338                                 }
339                                 break;
340                         default:
341                                 trace_xfs_attr_list_wrong_blk(context);
342                                 xfs_trans_brelse(context->tp, bp);
343                                 bp = NULL;
344                         }
345                 }
346         }
347
348         /*
349          * We did not find what we expected given the cursor's contents,
350          * so we start from the top and work down based on the hash value.
351          * Note that start of node block is same as start of leaf block.
352          */
353         if (bp == NULL) {
354                 error = xfs_attr_node_list_lookup(context, cursor, &bp);
355                 if (error || !bp)
356                         return error;
357         }
358         ASSERT(bp != NULL);
359
360         /*
361          * Roll upward through the blocks, processing each leaf block in
362          * order.  As long as there is space in the result buffer, keep
363          * adding the information.
364          */
365         for (;;) {
366                 leaf = bp->b_addr;
367                 error = xfs_attr3_leaf_list_int(bp, context);
368                 if (error)
369                         break;
370                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
371                 if (context->seen_enough || leafhdr.forw == 0)
372                         break;
373                 cursor->blkno = leafhdr.forw;
374                 xfs_trans_brelse(context->tp, bp);
375                 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno,
376                                             &bp);
377                 if (error)
378                         return error;
379         }
380         xfs_trans_brelse(context->tp, bp);
381         return error;
382 }
383
384 /*
385  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
386  */
387 int
388 xfs_attr3_leaf_list_int(
389         struct xfs_buf                  *bp,
390         struct xfs_attr_list_context    *context)
391 {
392         struct xfs_attrlist_cursor_kern *cursor = &context->cursor;
393         struct xfs_attr_leafblock       *leaf;
394         struct xfs_attr3_icleaf_hdr     ichdr;
395         struct xfs_attr_leaf_entry      *entries;
396         struct xfs_attr_leaf_entry      *entry;
397         int                             i;
398         struct xfs_mount                *mp = context->dp->i_mount;
399
400         trace_xfs_attr_list_leaf(context);
401
402         leaf = bp->b_addr;
403         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
404         entries = xfs_attr3_leaf_entryp(leaf);
405
406         cursor->initted = 1;
407
408         /*
409          * Re-find our place in the leaf block if this is a new syscall.
410          */
411         if (context->resynch) {
412                 entry = &entries[0];
413                 for (i = 0; i < ichdr.count; entry++, i++) {
414                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
415                                 if (cursor->offset == context->dupcnt) {
416                                         context->dupcnt = 0;
417                                         break;
418                                 }
419                                 context->dupcnt++;
420                         } else if (be32_to_cpu(entry->hashval) >
421                                         cursor->hashval) {
422                                 context->dupcnt = 0;
423                                 break;
424                         }
425                 }
426                 if (i == ichdr.count) {
427                         trace_xfs_attr_list_notfound(context);
428                         return 0;
429                 }
430         } else {
431                 entry = &entries[0];
432                 i = 0;
433         }
434         context->resynch = 0;
435
436         /*
437          * We have found our place, start copying out the new attributes.
438          */
439         for (; i < ichdr.count; entry++, i++) {
440                 char *name;
441                 int namelen, valuelen;
442
443                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
444                         cursor->hashval = be32_to_cpu(entry->hashval);
445                         cursor->offset = 0;
446                 }
447
448                 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
449                     !context->allow_incomplete)
450                         continue;
451
452                 if (entry->flags & XFS_ATTR_LOCAL) {
453                         xfs_attr_leaf_name_local_t *name_loc;
454
455                         name_loc = xfs_attr3_leaf_name_local(leaf, i);
456                         name = name_loc->nameval;
457                         namelen = name_loc->namelen;
458                         valuelen = be16_to_cpu(name_loc->valuelen);
459                 } else {
460                         xfs_attr_leaf_name_remote_t *name_rmt;
461
462                         name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
463                         name = name_rmt->name;
464                         namelen = name_rmt->namelen;
465                         valuelen = be32_to_cpu(name_rmt->valuelen);
466                 }
467
468                 if (XFS_IS_CORRUPT(context->dp->i_mount,
469                                    !xfs_attr_namecheck(name, namelen)))
470                         return -EFSCORRUPTED;
471                 context->put_listent(context, entry->flags,
472                                               name, namelen, valuelen);
473                 if (context->seen_enough)
474                         break;
475                 cursor->offset++;
476         }
477         trace_xfs_attr_list_leaf_end(context);
478         return 0;
479 }
480
481 /*
482  * Copy out attribute entries for attr_list(), for leaf attribute lists.
483  */
484 STATIC int
485 xfs_attr_leaf_list(
486         struct xfs_attr_list_context    *context)
487 {
488         struct xfs_buf                  *bp;
489         int                             error;
490
491         trace_xfs_attr_leaf_list(context);
492
493         context->cursor.blkno = 0;
494         error = xfs_attr3_leaf_read(context->tp, context->dp, 0, &bp);
495         if (error)
496                 return error;
497
498         error = xfs_attr3_leaf_list_int(bp, context);
499         xfs_trans_brelse(context->tp, bp);
500         return error;
501 }
502
503 int
504 xfs_attr_list_ilocked(
505         struct xfs_attr_list_context    *context)
506 {
507         struct xfs_inode                *dp = context->dp;
508
509         ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
510
511         /*
512          * Decide on what work routines to call based on the inode size.
513          */
514         if (!xfs_inode_hasattr(dp))
515                 return 0;
516         if (dp->i_afp->if_format == XFS_DINODE_FMT_LOCAL)
517                 return xfs_attr_shortform_list(context);
518         if (xfs_attr_is_leaf(dp))
519                 return xfs_attr_leaf_list(context);
520         return xfs_attr_node_list(context);
521 }
522
523 int
524 xfs_attr_list(
525         struct xfs_attr_list_context    *context)
526 {
527         struct xfs_inode                *dp = context->dp;
528         uint                            lock_mode;
529         int                             error;
530
531         XFS_STATS_INC(dp->i_mount, xs_attr_list);
532
533         if (xfs_is_shutdown(dp->i_mount))
534                 return -EIO;
535
536         lock_mode = xfs_ilock_attr_map_shared(dp);
537         error = xfs_attr_list_ilocked(context);
538         xfs_iunlock(dp, lock_mode);
539         return error;
540 }