Add the rt linux 4.1.3-rt3 as base
[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                         return error;
207                 if (context->seen_enough)
208                         break;
209                 cursor->offset++;
210         }
211
212         kmem_free(sbuf);
213         return 0;
214 }
215
216 STATIC int
217 xfs_attr_node_list(xfs_attr_list_context_t *context)
218 {
219         attrlist_cursor_kern_t *cursor;
220         xfs_attr_leafblock_t *leaf;
221         xfs_da_intnode_t *node;
222         struct xfs_attr3_icleaf_hdr leafhdr;
223         struct xfs_da3_icnode_hdr nodehdr;
224         struct xfs_da_node_entry *btree;
225         int error, i;
226         struct xfs_buf *bp;
227         struct xfs_inode        *dp = context->dp;
228         struct xfs_mount        *mp = dp->i_mount;
229
230         trace_xfs_attr_node_list(context);
231
232         cursor = context->cursor;
233         cursor->initted = 1;
234
235         /*
236          * Do all sorts of validation on the passed-in cursor structure.
237          * If anything is amiss, ignore the cursor and look up the hashval
238          * starting from the btree root.
239          */
240         bp = NULL;
241         if (cursor->blkno > 0) {
242                 error = xfs_da3_node_read(NULL, dp, cursor->blkno, -1,
243                                               &bp, XFS_ATTR_FORK);
244                 if ((error != 0) && (error != -EFSCORRUPTED))
245                         return error;
246                 if (bp) {
247                         struct xfs_attr_leaf_entry *entries;
248
249                         node = bp->b_addr;
250                         switch (be16_to_cpu(node->hdr.info.magic)) {
251                         case XFS_DA_NODE_MAGIC:
252                         case XFS_DA3_NODE_MAGIC:
253                                 trace_xfs_attr_list_wrong_blk(context);
254                                 xfs_trans_brelse(NULL, bp);
255                                 bp = NULL;
256                                 break;
257                         case XFS_ATTR_LEAF_MAGIC:
258                         case XFS_ATTR3_LEAF_MAGIC:
259                                 leaf = bp->b_addr;
260                                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
261                                                              &leafhdr, leaf);
262                                 entries = xfs_attr3_leaf_entryp(leaf);
263                                 if (cursor->hashval > be32_to_cpu(
264                                                 entries[leafhdr.count - 1].hashval)) {
265                                         trace_xfs_attr_list_wrong_blk(context);
266                                         xfs_trans_brelse(NULL, bp);
267                                         bp = NULL;
268                                 } else if (cursor->hashval <= be32_to_cpu(
269                                                 entries[0].hashval)) {
270                                         trace_xfs_attr_list_wrong_blk(context);
271                                         xfs_trans_brelse(NULL, bp);
272                                         bp = NULL;
273                                 }
274                                 break;
275                         default:
276                                 trace_xfs_attr_list_wrong_blk(context);
277                                 xfs_trans_brelse(NULL, bp);
278                                 bp = NULL;
279                         }
280                 }
281         }
282
283         /*
284          * We did not find what we expected given the cursor's contents,
285          * so we start from the top and work down based on the hash value.
286          * Note that start of node block is same as start of leaf block.
287          */
288         if (bp == NULL) {
289                 cursor->blkno = 0;
290                 for (;;) {
291                         __uint16_t magic;
292
293                         error = xfs_da3_node_read(NULL, dp,
294                                                       cursor->blkno, -1, &bp,
295                                                       XFS_ATTR_FORK);
296                         if (error)
297                                 return error;
298                         node = bp->b_addr;
299                         magic = be16_to_cpu(node->hdr.info.magic);
300                         if (magic == XFS_ATTR_LEAF_MAGIC ||
301                             magic == XFS_ATTR3_LEAF_MAGIC)
302                                 break;
303                         if (magic != XFS_DA_NODE_MAGIC &&
304                             magic != XFS_DA3_NODE_MAGIC) {
305                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
306                                                      XFS_ERRLEVEL_LOW,
307                                                      context->dp->i_mount,
308                                                      node);
309                                 xfs_trans_brelse(NULL, bp);
310                                 return -EFSCORRUPTED;
311                         }
312
313                         dp->d_ops->node_hdr_from_disk(&nodehdr, node);
314                         btree = dp->d_ops->node_tree_p(node);
315                         for (i = 0; i < nodehdr.count; btree++, i++) {
316                                 if (cursor->hashval
317                                                 <= be32_to_cpu(btree->hashval)) {
318                                         cursor->blkno = be32_to_cpu(btree->before);
319                                         trace_xfs_attr_list_node_descend(context,
320                                                                          btree);
321                                         break;
322                                 }
323                         }
324                         if (i == nodehdr.count) {
325                                 xfs_trans_brelse(NULL, bp);
326                                 return 0;
327                         }
328                         xfs_trans_brelse(NULL, bp);
329                 }
330         }
331         ASSERT(bp != NULL);
332
333         /*
334          * Roll upward through the blocks, processing each leaf block in
335          * order.  As long as there is space in the result buffer, keep
336          * adding the information.
337          */
338         for (;;) {
339                 leaf = bp->b_addr;
340                 error = xfs_attr3_leaf_list_int(bp, context);
341                 if (error) {
342                         xfs_trans_brelse(NULL, bp);
343                         return error;
344                 }
345                 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
346                 if (context->seen_enough || leafhdr.forw == 0)
347                         break;
348                 cursor->blkno = leafhdr.forw;
349                 xfs_trans_brelse(NULL, bp);
350                 error = xfs_attr3_leaf_read(NULL, dp, cursor->blkno, -1, &bp);
351                 if (error)
352                         return error;
353         }
354         xfs_trans_brelse(NULL, bp);
355         return 0;
356 }
357
358 /*
359  * Copy out attribute list entries for attr_list(), for leaf attribute lists.
360  */
361 int
362 xfs_attr3_leaf_list_int(
363         struct xfs_buf                  *bp,
364         struct xfs_attr_list_context    *context)
365 {
366         struct attrlist_cursor_kern     *cursor;
367         struct xfs_attr_leafblock       *leaf;
368         struct xfs_attr3_icleaf_hdr     ichdr;
369         struct xfs_attr_leaf_entry      *entries;
370         struct xfs_attr_leaf_entry      *entry;
371         int                             retval;
372         int                             i;
373         struct xfs_mount                *mp = context->dp->i_mount;
374
375         trace_xfs_attr_list_leaf(context);
376
377         leaf = bp->b_addr;
378         xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
379         entries = xfs_attr3_leaf_entryp(leaf);
380
381         cursor = context->cursor;
382         cursor->initted = 1;
383
384         /*
385          * Re-find our place in the leaf block if this is a new syscall.
386          */
387         if (context->resynch) {
388                 entry = &entries[0];
389                 for (i = 0; i < ichdr.count; entry++, i++) {
390                         if (be32_to_cpu(entry->hashval) == cursor->hashval) {
391                                 if (cursor->offset == context->dupcnt) {
392                                         context->dupcnt = 0;
393                                         break;
394                                 }
395                                 context->dupcnt++;
396                         } else if (be32_to_cpu(entry->hashval) >
397                                         cursor->hashval) {
398                                 context->dupcnt = 0;
399                                 break;
400                         }
401                 }
402                 if (i == ichdr.count) {
403                         trace_xfs_attr_list_notfound(context);
404                         return 0;
405                 }
406         } else {
407                 entry = &entries[0];
408                 i = 0;
409         }
410         context->resynch = 0;
411
412         /*
413          * We have found our place, start copying out the new attributes.
414          */
415         retval = 0;
416         for (; i < ichdr.count; entry++, i++) {
417                 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
418                         cursor->hashval = be32_to_cpu(entry->hashval);
419                         cursor->offset = 0;
420                 }
421
422                 if (entry->flags & XFS_ATTR_INCOMPLETE)
423                         continue;               /* skip incomplete entries */
424
425                 if (entry->flags & XFS_ATTR_LOCAL) {
426                         xfs_attr_leaf_name_local_t *name_loc =
427                                 xfs_attr3_leaf_name_local(leaf, i);
428
429                         retval = context->put_listent(context,
430                                                 entry->flags,
431                                                 name_loc->nameval,
432                                                 (int)name_loc->namelen,
433                                                 be16_to_cpu(name_loc->valuelen),
434                                                 &name_loc->nameval[name_loc->namelen]);
435                         if (retval)
436                                 return retval;
437                 } else {
438                         xfs_attr_leaf_name_remote_t *name_rmt =
439                                 xfs_attr3_leaf_name_remote(leaf, i);
440
441                         int valuelen = be32_to_cpu(name_rmt->valuelen);
442
443                         if (context->put_value) {
444                                 xfs_da_args_t args;
445
446                                 memset((char *)&args, 0, sizeof(args));
447                                 args.geo = context->dp->i_mount->m_attr_geo;
448                                 args.dp = context->dp;
449                                 args.whichfork = XFS_ATTR_FORK;
450                                 args.valuelen = valuelen;
451                                 args.rmtvaluelen = valuelen;
452                                 args.value = kmem_alloc(valuelen, KM_SLEEP | KM_NOFS);
453                                 args.rmtblkno = be32_to_cpu(name_rmt->valueblk);
454                                 args.rmtblkcnt = xfs_attr3_rmt_blocks(
455                                                         args.dp->i_mount, valuelen);
456                                 retval = xfs_attr_rmtval_get(&args);
457                                 if (retval)
458                                         return retval;
459                                 retval = context->put_listent(context,
460                                                 entry->flags,
461                                                 name_rmt->name,
462                                                 (int)name_rmt->namelen,
463                                                 valuelen,
464                                                 args.value);
465                                 kmem_free(args.value);
466                         } else {
467                                 retval = context->put_listent(context,
468                                                 entry->flags,
469                                                 name_rmt->name,
470                                                 (int)name_rmt->namelen,
471                                                 valuelen,
472                                                 NULL);
473                         }
474                         if (retval)
475                                 return retval;
476                 }
477                 if (context->seen_enough)
478                         break;
479                 cursor->offset++;
480         }
481         trace_xfs_attr_list_leaf_end(context);
482         return retval;
483 }
484
485 /*
486  * Copy out attribute entries for attr_list(), for leaf attribute lists.
487  */
488 STATIC int
489 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
490 {
491         int error;
492         struct xfs_buf *bp;
493
494         trace_xfs_attr_leaf_list(context);
495
496         context->cursor->blkno = 0;
497         error = xfs_attr3_leaf_read(NULL, context->dp, 0, -1, &bp);
498         if (error)
499                 return error;
500
501         error = xfs_attr3_leaf_list_int(bp, context);
502         xfs_trans_brelse(NULL, bp);
503         return error;
504 }
505
506 int
507 xfs_attr_list_int(
508         xfs_attr_list_context_t *context)
509 {
510         int error;
511         xfs_inode_t *dp = context->dp;
512         uint            lock_mode;
513
514         XFS_STATS_INC(xs_attr_list);
515
516         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
517                 return -EIO;
518
519         /*
520          * Decide on what work routines to call based on the inode size.
521          */
522         lock_mode = xfs_ilock_attr_map_shared(dp);
523         if (!xfs_inode_hasattr(dp)) {
524                 error = 0;
525         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
526                 error = xfs_attr_shortform_list(context);
527         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
528                 error = xfs_attr_leaf_list(context);
529         } else {
530                 error = xfs_attr_node_list(context);
531         }
532         xfs_iunlock(dp, lock_mode);
533         return error;
534 }
535
536 #define ATTR_ENTBASESIZE                /* minimum bytes used by an attr */ \
537         (((struct attrlist_ent *) 0)->a_name - (char *) 0)
538 #define ATTR_ENTSIZE(namelen)           /* actual bytes used by an attr */ \
539         ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
540          & ~(sizeof(u_int32_t)-1))
541
542 /*
543  * Format an attribute and copy it out to the user's buffer.
544  * Take care to check values and protect against them changing later,
545  * we may be reading them directly out of a user buffer.
546  */
547 STATIC int
548 xfs_attr_put_listent(
549         xfs_attr_list_context_t *context,
550         int             flags,
551         unsigned char   *name,
552         int             namelen,
553         int             valuelen,
554         unsigned char   *value)
555 {
556         struct attrlist *alist = (struct attrlist *)context->alist;
557         attrlist_ent_t *aep;
558         int arraytop;
559
560         ASSERT(!(context->flags & ATTR_KERNOVAL));
561         ASSERT(context->count >= 0);
562         ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
563         ASSERT(context->firstu >= sizeof(*alist));
564         ASSERT(context->firstu <= context->bufsize);
565
566         /*
567          * Only list entries in the right namespace.
568          */
569         if (((context->flags & ATTR_SECURE) == 0) !=
570             ((flags & XFS_ATTR_SECURE) == 0))
571                 return 0;
572         if (((context->flags & ATTR_ROOT) == 0) !=
573             ((flags & XFS_ATTR_ROOT) == 0))
574                 return 0;
575
576         arraytop = sizeof(*alist) +
577                         context->count * sizeof(alist->al_offset[0]);
578         context->firstu -= ATTR_ENTSIZE(namelen);
579         if (context->firstu < arraytop) {
580                 trace_xfs_attr_list_full(context);
581                 alist->al_more = 1;
582                 context->seen_enough = 1;
583                 return 1;
584         }
585
586         aep = (attrlist_ent_t *)&context->alist[context->firstu];
587         aep->a_valuelen = valuelen;
588         memcpy(aep->a_name, name, namelen);
589         aep->a_name[namelen] = 0;
590         alist->al_offset[context->count++] = context->firstu;
591         alist->al_count = context->count;
592         trace_xfs_attr_list_add(context);
593         return 0;
594 }
595
596 /*
597  * Generate a list of extended attribute names and optionally
598  * also value lengths.  Positive return value follows the XFS
599  * convention of being an error, zero or negative return code
600  * is the length of the buffer returned (negated), indicating
601  * success.
602  */
603 int
604 xfs_attr_list(
605         xfs_inode_t     *dp,
606         char            *buffer,
607         int             bufsize,
608         int             flags,
609         attrlist_cursor_kern_t *cursor)
610 {
611         xfs_attr_list_context_t context;
612         struct attrlist *alist;
613         int error;
614
615         /*
616          * Validate the cursor.
617          */
618         if (cursor->pad1 || cursor->pad2)
619                 return -EINVAL;
620         if ((cursor->initted == 0) &&
621             (cursor->hashval || cursor->blkno || cursor->offset))
622                 return -EINVAL;
623
624         /*
625          * Check for a properly aligned buffer.
626          */
627         if (((long)buffer) & (sizeof(int)-1))
628                 return -EFAULT;
629         if (flags & ATTR_KERNOVAL)
630                 bufsize = 0;
631
632         /*
633          * Initialize the output buffer.
634          */
635         memset(&context, 0, sizeof(context));
636         context.dp = dp;
637         context.cursor = cursor;
638         context.resynch = 1;
639         context.flags = flags;
640         context.alist = buffer;
641         context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
642         context.firstu = context.bufsize;
643         context.put_listent = xfs_attr_put_listent;
644
645         alist = (struct attrlist *)context.alist;
646         alist->al_count = 0;
647         alist->al_more = 0;
648         alist->al_offset[0] = context.bufsize;
649
650         error = xfs_attr_list_int(&context);
651         ASSERT(error <= 0);
652         return error;
653 }