Upgrade to 4.4.50-rt62
[kvmfornfv.git] / kernel / fs / xfs / xfs_attr_list.c
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * Copyright (c) 2013 Red Hat, Inc.
4  * All Rights Reserved.
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License as
8  * published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it would be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write the Free Software Foundation,
17  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
18  */
19 #include "xfs.h"
20 #include "xfs_fs.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
29 #include "xfs_trans.h"
30 #include "xfs_inode_item.h"
31 #include "xfs_bmap.h"
32 #include "xfs_attr.h"
33 #include "xfs_attr_sf.h"
34 #include "xfs_attr_remote.h"
35 #include "xfs_attr_leaf.h"
36 #include "xfs_error.h"
37 #include "xfs_trace.h"
38 #include "xfs_buf_item.h"
39 #include "xfs_cksum.h"
40 #include "xfs_dir2.h"
41
42 STATIC int
43 xfs_attr_shortform_compare(const void *a, const void *b)
44 {
45         xfs_attr_sf_sort_t *sa, *sb;
46
47         sa = (xfs_attr_sf_sort_t *)a;
48         sb = (xfs_attr_sf_sort_t *)b;
49         if (sa->hash < sb->hash) {
50                 return -1;
51         } else if (sa->hash > sb->hash) {
52                 return 1;
53         } else {
54                 return sa->entno - sb->entno;
55         }
56 }
57
58 #define XFS_ISRESET_CURSOR(cursor) \
59         (!((cursor)->initted) && !((cursor)->hashval) && \
60          !((cursor)->blkno) && !((cursor)->offset))
61 /*
62  * Copy out entries of shortform attribute lists for attr_list().
63  * Shortform attribute lists are not stored in hashval sorted order.
64  * If the output buffer is not large enough to hold them all, then we
65  * we have to calculate each entries' hashvalue and sort them before
66  * we can begin returning them to the user.
67  */
68 int
69 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
70 {
71         attrlist_cursor_kern_t *cursor;
72         xfs_attr_sf_sort_t *sbuf, *sbp;
73         xfs_attr_shortform_t *sf;
74         xfs_attr_sf_entry_t *sfe;
75         xfs_inode_t *dp;
76         int sbsize, nsbuf, count, i;
77         int error;
78
79         ASSERT(context != NULL);
80         dp = context->dp;
81         ASSERT(dp != NULL);
82         ASSERT(dp->i_afp != NULL);
83         sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
84         ASSERT(sf != NULL);
85         if (!sf->hdr.count)
86                 return 0;
87         cursor = context->cursor;
88         ASSERT(cursor != NULL);
89
90         trace_xfs_attr_list_sf(context);
91
92         /*
93          * If the buffer is large enough and the cursor is at the start,
94          * do not bother with sorting since we will return everything in
95          * one buffer and another call using the cursor won't need to be
96          * made.
97          * Note the generous fudge factor of 16 overhead bytes per entry.
98          * If bufsize is zero then put_listent must be a search function
99          * and can just scan through what we have.
100          */
101         if (context->bufsize == 0 ||
102             (XFS_ISRESET_CURSOR(cursor) &&
103              (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
104                 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
105                         error = context->put_listent(context,
106                                            sfe->flags,
107                                            sfe->nameval,
108                                            (int)sfe->namelen,
109                                            (int)sfe->valuelen,
110                                            &sfe->nameval[sfe->namelen]);
111
112                         /*
113                          * Either search callback finished early or
114                          * didn't fit it all in the buffer after all.
115                          */
116                         if (context->seen_enough)
117                                 break;
118
119                         if (error)
120                                 return error;
121                         sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
122                 }
123                 trace_xfs_attr_list_sf_all(context);
124                 return 0;
125         }
126
127         /* do no more for a search callback */
128         if (context->bufsize == 0)
129                 return 0;
130
131         /*
132          * It didn't all fit, so we have to sort everything on hashval.
133          */
134         sbsize = sf->hdr.count * sizeof(*sbuf);
135         sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
136
137         /*
138          * Scan the attribute list for the rest of the entries, storing
139          * the relevant info from only those that match into a buffer.
140          */
141         nsbuf = 0;
142         for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
143                 if (unlikely(
144                     ((char *)sfe < (char *)sf) ||
145                     ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
146                         XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
147                                              XFS_ERRLEVEL_LOW,
148                                              context->dp->i_mount, sfe);
149                         kmem_free(sbuf);
150                         return -EFSCORRUPTED;
151                 }
152
153                 sbp->entno = i;
154                 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
155                 sbp->name = sfe->nameval;
156                 sbp->namelen = sfe->namelen;
157                 /* These are bytes, and both on-disk, don't endian-flip */
158                 sbp->valuelen = sfe->valuelen;
159                 sbp->flags = sfe->flags;
160                 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
161                 sbp++;
162                 nsbuf++;
163         }
164
165         /*
166          * Sort the entries on hash then entno.
167          */
168         xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
169
170         /*
171          * Re-find our place IN THE SORTED LIST.
172          */
173         count = 0;
174         cursor->initted = 1;
175         cursor->blkno = 0;
176         for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
177                 if (sbp->hash == cursor->hashval) {
178                         if (cursor->offset == count) {
179                                 break;
180                         }
181                         count++;
182                 } else if (sbp->hash > cursor->hashval) {
183                         break;
184                 }
185         }
186         if (i == nsbuf) {
187                 kmem_free(sbuf);
188                 return 0;
189         }
190
191         /*
192          * Loop putting entries into the user buffer.
193          */
194         for ( ; i < nsbuf; i++, sbp++) {
195                 if (cursor->hashval != sbp->hash) {
196                         cursor->hashval = sbp->hash;
197                         cursor->offset = 0;
198                 }
199                 error = context->put_listent(context,
200                                         sbp->flags,
201                                         sbp->name,
202                                         sbp->namelen,
203                                         sbp->valuelen,
204                                         &sbp->name[sbp->namelen]);
205                 if (error) {
206                         kmem_free(sbuf);
207                         return error;
208                 }
209                 if (context->seen_enough)
210                         break;
211                 cursor->offset++;
212         }
213
214         kmem_free(sbuf);
215         return 0;
216 }
217
218 STATIC int
219 xfs_attr_node_list(xfs_attr_list_context_t *context)
220 {
221         attrlist_cursor_kern_t *cursor;
222         xfs_attr_leafblock_t *leaf;
223         xfs_da_intnode_t *node;
224         struct xfs_attr3_icleaf_hdr leafhdr;
225         struct xfs_da3_icnode_hdr nodehdr;
226         struct xfs_da_node_entry *btree;
227         int error, i;
228         struct xfs_buf *bp;
229         struct xfs_inode        *dp = context->dp;
230         struct xfs_mount        *mp = dp->i_mount;
231
232         trace_xfs_attr_node_list(context);
233
234         cursor = context->cursor;
235         cursor->initted = 1;
236
237         /*
238          * Do all sorts of validation on the passed-in cursor structure.
239          * If anything is amiss, ignore the cursor and look up the hashval
240          * starting from the btree root.
241          */
242         bp = NULL;
243         if (cursor->blkno > 0) {
244                 error = xfs_da3_node_read(NULL, dp, cursor->blkno, -1,
245                                               &bp, XFS_ATTR_FORK);
246                 if ((error != 0) && (error != -EFSCORRUPTED))
247                         return error;
248                 if (bp) {
249                         struct xfs_attr_leaf_entry *entries;
250
251                         node = bp->b_addr;
252                         switch (be16_to_cpu(node->hdr.info.magic)) {
253                         case XFS_DA_NODE_MAGIC:
254                         case XFS_DA3_NODE_MAGIC:
255                                 trace_xfs_attr_list_wrong_blk(context);
256                                 xfs_trans_brelse(NULL, bp);
257                                 bp = NULL;
258                                 break;
259                         case XFS_ATTR_LEAF_MAGIC:
260                         case XFS_ATTR3_LEAF_MAGIC:
261                                 leaf = bp->b_addr;
262                                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
263                                                              &leafhdr, leaf);
264                                 entries = xfs_attr3_leaf_entryp(leaf);
265                                 if (cursor->hashval > be32_to_cpu(
266                                                 entries[leafhdr.count - 1].hashval)) {
267                                         trace_xfs_attr_list_wrong_blk(context);
268                                         xfs_trans_brelse(NULL, bp);
269                                         bp = NULL;
270                                 } else if (cursor->hashval <= be32_to_cpu(
271                                                 entries[0].hashval)) {
272                                         trace_xfs_attr_list_wrong_blk(context);
273                                         xfs_trans_brelse(NULL, bp);
274                                         bp = NULL;
275                                 }
276                                 break;
277                         default:
278                                 trace_xfs_attr_list_wrong_blk(context);
279                                 xfs_trans_brelse(NULL, bp);
280                                 bp = NULL;
281                         }
282                 }
283         }
284
285         /*
286          * We did not find what we expected given the cursor's contents,
287          * so we start from the top and work down based on the hash value.
288          * Note that start of node block is same as start of leaf block.
289          */
290         if (bp == NULL) {
291                 cursor->blkno = 0;
292                 for (;;) {
293                         __uint16_t magic;
294
295                         error = xfs_da3_node_read(NULL, dp,
296                                                       cursor->blkno, -1, &bp,
297                                                       XFS_ATTR_FORK);
298                         if (error)
299                                 return error;
300                         node = bp->b_addr;
301                         magic = be16_to_cpu(node->hdr.info.magic);
302                         if (magic == XFS_ATTR_LEAF_MAGIC ||
303                             magic == XFS_ATTR3_LEAF_MAGIC)
304                                 break;
305                         if (magic != XFS_DA_NODE_MAGIC &&
306                             magic != XFS_DA3_NODE_MAGIC) {
307                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
308                                                      XFS_ERRLEVEL_LOW,
309                                                      context->dp->i_mount,
310                                                      node);
311                                 xfs_trans_brelse(NULL, bp);
312                                 return -EFSCORRUPTED;
313                         }
314
315                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
316                         btree = dp->d_ops->node_tree_p(node);
317                         for (i = 0; i < nodehdr.count; btree++, i++) {
318                                 if (cursor->hashval
319                                                 <= be32_to_cpu(btree->hashval)) {
320                                         cursor->blkno = be32_to_cpu(btree->before);
321                                         trace_xfs_attr_list_node_descend(context,
322                                                                          btree);
323                                         break;
324                                 }
325                         }
326                         if (i == nodehdr.count) {
327                                 xfs_trans_brelse(NULL, bp);
328                                 return 0;
329                         }
330                         xfs_trans_brelse(NULL, bp);
331                 }
332         }
333         ASSERT(bp != NULL);
334
335         /*
336          * Roll upward through the blocks, processing each leaf block in
337          * order.  As long as there is space in the result buffer, keep
338          * adding the information.
339          */
340         for (;;) {
341                 leaf = bp->b_addr;
342                 error = xfs_attr3_leaf_list_int(bp, context);
343                 if (error) {
344                         xfs_trans_brelse(NULL, bp);
345                         return error;
346                 }
347                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
348                 if (context->seen_enough || leafhdr.forw == 0)
349                         break;
350                 cursor->blkno = leafhdr.forw;
351                 xfs_trans_brelse(NULL, bp);
352                 error = xfs_attr3_leaf_read(NULL, dp, cursor->blkno, -1, &bp);
353                 if (error)
354                         return error;
355         }
356         xfs_trans_brelse(NULL, bp);
357         return 0;
358 }
359
360 /*
361  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
362  */
363 int
364 xfs_attr3_leaf_list_int(
365         struct xfs_buf                  *bp,
366         struct xfs_attr_list_context    *context)
367 {
368         struct attrlist_cursor_kern     *cursor;
369         struct xfs_attr_leafblock       *leaf;
370         struct xfs_attr3_icleaf_hdr     ichdr;
371         struct xfs_attr_leaf_entry      *entries;
372         struct xfs_attr_leaf_entry      *entry;
373         int                             retval;
374         int                             i;
375         struct xfs_mount                *mp = context->dp->i_mount;
376
377         trace_xfs_attr_list_leaf(context);
378
379         leaf = bp->b_addr;
380         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
381         entries = xfs_attr3_leaf_entryp(leaf);
382
383         cursor = context->cursor;
384         cursor->initted = 1;
385
386         /*
387          * Re-find our place in the leaf block if this is a new syscall.
388          */
389         if (context->resynch) {
390                 entry = &entries[0];
391                 for (i = 0; i < ichdr.count; entry++, i++) {
392                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
393                                 if (cursor->offset == context->dupcnt) {
394                                         context->dupcnt = 0;
395                                         break;
396                                 }
397                                 context->dupcnt++;
398                         } else if (be32_to_cpu(entry->hashval) >
399                                         cursor->hashval) {
400                                 context->dupcnt = 0;
401                                 break;
402                         }
403                 }
404                 if (i == ichdr.count) {
405                         trace_xfs_attr_list_notfound(context);
406                         return 0;
407                 }
408         } else {
409                 entry = &entries[0];
410                 i = 0;
411         }
412         context->resynch = 0;
413
414         /*
415          * We have found our place, start copying out the new attributes.
416          */
417         retval = 0;
418         for (; i < ichdr.count; entry++, i++) {
419                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
420                         cursor->hashval = be32_to_cpu(entry->hashval);
421                         cursor->offset = 0;
422                 }
423
424                 if (entry->flags & XFS_ATTR_INCOMPLETE)
425                         continue;               /* skip incomplete entries */
426
427                 if (entry->flags & XFS_ATTR_LOCAL) {
428                         xfs_attr_leaf_name_local_t *name_loc =
429                                 xfs_attr3_leaf_name_local(leaf, i);
430
431                         retval = context->put_listent(context,
432                                                 entry->flags,
433                                                 name_loc->nameval,
434                                                 (int)name_loc->namelen,
435                                                 be16_to_cpu(name_loc->valuelen),
436                                                 &name_loc->nameval[name_loc->namelen]);
437                         if (retval)
438                                 return retval;
439                 } else {
440                         xfs_attr_leaf_name_remote_t *name_rmt =
441                                 xfs_attr3_leaf_name_remote(leaf, i);
442
443                         int valuelen = be32_to_cpu(name_rmt->valuelen);
444
445                         if (context->put_value) {
446                                 xfs_da_args_t args;
447
448                                 memset((char *)&args, 0, sizeof(args));
449                                 args.geo = context->dp->i_mount->m_attr_geo;
450                                 args.dp = context->dp;
451                                 args.whichfork = XFS_ATTR_FORK;
452                                 args.valuelen = valuelen;
453                                 args.rmtvaluelen = valuelen;
454                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
455                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
456                                 args.rmtblkcnt = xfs_attr3_rmt_blocks(
457                                                         args.dp->i_mount, valuelen);
458                                 retval = xfs_attr_rmtval_get(&args);
459                                 if (!retval)
460                                         retval = context->put_listent(context,
461                                                         entry->flags,
462                                                         name_rmt->name,
463                                                         (int)name_rmt->namelen,
464                                                         valuelen,
465                                                         args.value);
466                                 kmem_free(args.value);
467                         } else {
468                                 retval = context->put_listent(context,
469                                                 entry->flags,
470                                                 name_rmt->name,
471                                                 (int)name_rmt->namelen,
472                                                 valuelen,
473                                                 NULL);
474                         }
475                         if (retval)
476                                 return retval;
477                 }
478                 if (context->seen_enough)
479                         break;
480                 cursor->offset++;
481         }
482         trace_xfs_attr_list_leaf_end(context);
483         return retval;
484 }
485
486 /*
487  * Copy out attribute entries for attr_list(), for leaf attribute lists.
488  */
489 STATIC int
490 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
491 {
492         int error;
493         struct xfs_buf *bp;
494
495         trace_xfs_attr_leaf_list(context);
496
497         context->cursor->blkno = 0;
498         error = xfs_attr3_leaf_read(NULL, context->dp, 0, -1, &bp);
499         if (error)
500                 return error;
501
502         error = xfs_attr3_leaf_list_int(bp, context);
503         xfs_trans_brelse(NULL, bp);
504         return error;
505 }
506
507 int
508 xfs_attr_list_int(
509         xfs_attr_list_context_t *context)
510 {
511         int error;
512         xfs_inode_t *dp = context->dp;
513         uint            lock_mode;
514
515         XFS_STATS_INC(dp->i_mount, xs_attr_list);
516
517         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
518                 return -EIO;
519
520         /*
521          * Decide on what work routines to call based on the inode size.
522          */
523         lock_mode = xfs_ilock_attr_map_shared(dp);
524         if (!xfs_inode_hasattr(dp)) {
525                 error = 0;
526         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
527                 error = xfs_attr_shortform_list(context);
528         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
529                 error = xfs_attr_leaf_list(context);
530         } else {
531                 error = xfs_attr_node_list(context);
532         }
533         xfs_iunlock(dp, lock_mode);
534         return error;
535 }
536
537 #define ATTR_ENTBASESIZE                /* minimum bytes used by an attr */ \
538         (((struct attrlist_ent *) 0)->a_name - (char *) 0)
539 #define ATTR_ENTSIZE(namelen)           /* actual bytes used by an attr */ \
540         ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
541          & ~(sizeof(u_int32_t)-1))
542
543 /*
544  * Format an attribute and copy it out to the user's buffer.
545  * Take care to check values and protect against them changing later,
546  * we may be reading them directly out of a user buffer.
547  */
548 STATIC int
549 xfs_attr_put_listent(
550         xfs_attr_list_context_t *context,
551         int             flags,
552         unsigned char   *name,
553         int             namelen,
554         int             valuelen,
555         unsigned char   *value)
556 {
557         struct attrlist *alist = (struct attrlist *)context->alist;
558         attrlist_ent_t *aep;
559         int arraytop;
560
561         ASSERT(!(context->flags & ATTR_KERNOVAL));
562         ASSERT(context->count >= 0);
563         ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
564         ASSERT(context->firstu >= sizeof(*alist));
565         ASSERT(context->firstu <= context->bufsize);
566
567         /*
568          * Only list entries in the right namespace.
569          */
570         if (((context->flags & ATTR_SECURE) == 0) !=
571             ((flags & XFS_ATTR_SECURE) == 0))
572                 return 0;
573         if (((context->flags & ATTR_ROOT) == 0) !=
574             ((flags & XFS_ATTR_ROOT) == 0))
575                 return 0;
576
577         arraytop = sizeof(*alist) +
578                         context->count * sizeof(alist->al_offset[0]);
579         context->firstu -= ATTR_ENTSIZE(namelen);
580         if (context->firstu < arraytop) {
581                 trace_xfs_attr_list_full(context);
582                 alist->al_more = 1;
583                 context->seen_enough = 1;
584                 return 1;
585         }
586
587         aep = (attrlist_ent_t *)&context->alist[context->firstu];
588         aep->a_valuelen = valuelen;
589         memcpy(aep->a_name, name, namelen);
590         aep->a_name[namelen] = 0;
591         alist->al_offset[context->count++] = context->firstu;
592         alist->al_count = context->count;
593         trace_xfs_attr_list_add(context);
594         return 0;
595 }
596
597 /*
598  * Generate a list of extended attribute names and optionally
599  * also value lengths.  Positive return value follows the XFS
600  * convention of being an error, zero or negative return code
601  * is the length of the buffer returned (negated), indicating
602  * success.
603  */
604 int
605 xfs_attr_list(
606         xfs_inode_t     *dp,
607         char            *buffer,
608         int             bufsize,
609         int             flags,
610         attrlist_cursor_kern_t *cursor)
611 {
612         xfs_attr_list_context_t context;
613         struct attrlist *alist;
614         int error;
615
616         /*
617          * Validate the cursor.
618          */
619         if (cursor->pad1 || cursor->pad2)
620                 return -EINVAL;
621         if ((cursor->initted == 0) &&
622             (cursor->hashval || cursor->blkno || cursor->offset))
623                 return -EINVAL;
624
625         /*
626          * Check for a properly aligned buffer.
627          */
628         if (((long)buffer) & (sizeof(int)-1))
629                 return -EFAULT;
630         if (flags & ATTR_KERNOVAL)
631                 bufsize = 0;
632
633         /*
634          * Initialize the output buffer.
635          */
636         memset(&context, 0, sizeof(context));
637         context.dp = dp;
638         context.cursor = cursor;
639         context.resynch = 1;
640         context.flags = flags;
641         context.alist = buffer;
642         context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
643         context.firstu = context.bufsize;
644         context.put_listent = xfs_attr_put_listent;
645
646         alist = (struct attrlist *)context.alist;
647         alist->al_count = 0;
648         alist->al_more = 0;
649         alist->al_offset[0] = context.bufsize;
650
651         error = xfs_attr_list_int(&context);
652         ASSERT(error <= 0);
653         return error;
654 }