mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / xfs / xfs_attr_list.c
blob7740c8a5e7369de93caa93cabed20745ad35d6f5
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
4 * All Rights Reserved.
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.
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.
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
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"
42 STATIC int
43 xfs_attr_shortform_compare(const void *a, const void *b)
45 xfs_attr_sf_sort_t *sa, *sb;
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;
58 #define XFS_ISRESET_CURSOR(cursor) \
59 (!((cursor)->initted) && !((cursor)->hashval) && \
60 !((cursor)->blkno) && !((cursor)->offset))
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.
68 static int
69 xfs_attr_shortform_list(xfs_attr_list_context_t *context)
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;
78 ASSERT(context != NULL);
79 dp = context->dp;
80 ASSERT(dp != NULL);
81 ASSERT(dp->i_afp != NULL);
82 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
83 ASSERT(sf != NULL);
84 if (!sf->hdr.count)
85 return 0;
86 cursor = context->cursor;
87 ASSERT(cursor != NULL);
89 trace_xfs_attr_list_sf(context);
92 * If the buffer is large enough and the cursor is at the start,
93 * do not bother with sorting since we will return everything in
94 * one buffer and another call using the cursor won't need to be
95 * made.
96 * Note the generous fudge factor of 16 overhead bytes per entry.
97 * If bufsize is zero then put_listent must be a search function
98 * and can just scan through what we have.
100 if (context->bufsize == 0 ||
101 (XFS_ISRESET_CURSOR(cursor) &&
102 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
103 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
104 context->put_listent(context,
105 sfe->flags,
106 sfe->nameval,
107 (int)sfe->namelen,
108 (int)sfe->valuelen);
110 * Either search callback finished early or
111 * didn't fit it all in the buffer after all.
113 if (context->seen_enough)
114 break;
115 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
117 trace_xfs_attr_list_sf_all(context);
118 return 0;
121 /* do no more for a search callback */
122 if (context->bufsize == 0)
123 return 0;
126 * It didn't all fit, so we have to sort everything on hashval.
128 sbsize = sf->hdr.count * sizeof(*sbuf);
129 sbp = sbuf = kmem_alloc(sbsize, KM_SLEEP | KM_NOFS);
132 * Scan the attribute list for the rest of the entries, storing
133 * the relevant info from only those that match into a buffer.
135 nsbuf = 0;
136 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
137 if (unlikely(
138 ((char *)sfe < (char *)sf) ||
139 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
140 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
141 XFS_ERRLEVEL_LOW,
142 context->dp->i_mount, sfe);
143 kmem_free(sbuf);
144 return -EFSCORRUPTED;
147 sbp->entno = i;
148 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
149 sbp->name = sfe->nameval;
150 sbp->namelen = sfe->namelen;
151 /* These are bytes, and both on-disk, don't endian-flip */
152 sbp->valuelen = sfe->valuelen;
153 sbp->flags = sfe->flags;
154 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
155 sbp++;
156 nsbuf++;
160 * Sort the entries on hash then entno.
162 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
165 * Re-find our place IN THE SORTED LIST.
167 count = 0;
168 cursor->initted = 1;
169 cursor->blkno = 0;
170 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
171 if (sbp->hash == cursor->hashval) {
172 if (cursor->offset == count) {
173 break;
175 count++;
176 } else if (sbp->hash > cursor->hashval) {
177 break;
180 if (i == nsbuf) {
181 kmem_free(sbuf);
182 return 0;
186 * Loop putting entries into the user buffer.
188 for ( ; i < nsbuf; i++, sbp++) {
189 if (cursor->hashval != sbp->hash) {
190 cursor->hashval = sbp->hash;
191 cursor->offset = 0;
193 context->put_listent(context,
194 sbp->flags,
195 sbp->name,
196 sbp->namelen,
197 sbp->valuelen);
198 if (context->seen_enough)
199 break;
200 cursor->offset++;
203 kmem_free(sbuf);
204 return 0;
207 STATIC int
208 xfs_attr_node_list(xfs_attr_list_context_t *context)
210 attrlist_cursor_kern_t *cursor;
211 xfs_attr_leafblock_t *leaf;
212 xfs_da_intnode_t *node;
213 struct xfs_attr3_icleaf_hdr leafhdr;
214 struct xfs_da3_icnode_hdr nodehdr;
215 struct xfs_da_node_entry *btree;
216 int error, i;
217 struct xfs_buf *bp;
218 struct xfs_inode *dp = context->dp;
219 struct xfs_mount *mp = dp->i_mount;
221 trace_xfs_attr_node_list(context);
223 cursor = context->cursor;
224 cursor->initted = 1;
227 * Do all sorts of validation on the passed-in cursor structure.
228 * If anything is amiss, ignore the cursor and look up the hashval
229 * starting from the btree root.
231 bp = NULL;
232 if (cursor->blkno > 0) {
233 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
234 &bp, XFS_ATTR_FORK);
235 if ((error != 0) && (error != -EFSCORRUPTED))
236 return error;
237 if (bp) {
238 struct xfs_attr_leaf_entry *entries;
240 node = bp->b_addr;
241 switch (be16_to_cpu(node->hdr.info.magic)) {
242 case XFS_DA_NODE_MAGIC:
243 case XFS_DA3_NODE_MAGIC:
244 trace_xfs_attr_list_wrong_blk(context);
245 xfs_trans_brelse(context->tp, bp);
246 bp = NULL;
247 break;
248 case XFS_ATTR_LEAF_MAGIC:
249 case XFS_ATTR3_LEAF_MAGIC:
250 leaf = bp->b_addr;
251 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
252 &leafhdr, leaf);
253 entries = xfs_attr3_leaf_entryp(leaf);
254 if (cursor->hashval > be32_to_cpu(
255 entries[leafhdr.count - 1].hashval)) {
256 trace_xfs_attr_list_wrong_blk(context);
257 xfs_trans_brelse(context->tp, bp);
258 bp = NULL;
259 } else if (cursor->hashval <= be32_to_cpu(
260 entries[0].hashval)) {
261 trace_xfs_attr_list_wrong_blk(context);
262 xfs_trans_brelse(context->tp, bp);
263 bp = NULL;
265 break;
266 default:
267 trace_xfs_attr_list_wrong_blk(context);
268 xfs_trans_brelse(context->tp, bp);
269 bp = NULL;
275 * We did not find what we expected given the cursor's contents,
276 * so we start from the top and work down based on the hash value.
277 * Note that start of node block is same as start of leaf block.
279 if (bp == NULL) {
280 cursor->blkno = 0;
281 for (;;) {
282 uint16_t magic;
284 error = xfs_da3_node_read(context->tp, dp,
285 cursor->blkno, -1, &bp,
286 XFS_ATTR_FORK);
287 if (error)
288 return error;
289 node = bp->b_addr;
290 magic = be16_to_cpu(node->hdr.info.magic);
291 if (magic == XFS_ATTR_LEAF_MAGIC ||
292 magic == XFS_ATTR3_LEAF_MAGIC)
293 break;
294 if (magic != XFS_DA_NODE_MAGIC &&
295 magic != XFS_DA3_NODE_MAGIC) {
296 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
297 XFS_ERRLEVEL_LOW,
298 context->dp->i_mount,
299 node);
300 xfs_trans_brelse(context->tp, bp);
301 return -EFSCORRUPTED;
304 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
305 btree = dp->d_ops->node_tree_p(node);
306 for (i = 0; i < nodehdr.count; btree++, i++) {
307 if (cursor->hashval
308 <= be32_to_cpu(btree->hashval)) {
309 cursor->blkno = be32_to_cpu(btree->before);
310 trace_xfs_attr_list_node_descend(context,
311 btree);
312 break;
315 if (i == nodehdr.count) {
316 xfs_trans_brelse(context->tp, bp);
317 return 0;
319 xfs_trans_brelse(context->tp, bp);
322 ASSERT(bp != NULL);
325 * Roll upward through the blocks, processing each leaf block in
326 * order. As long as there is space in the result buffer, keep
327 * adding the information.
329 for (;;) {
330 leaf = bp->b_addr;
331 xfs_attr3_leaf_list_int(bp, context);
332 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
333 if (context->seen_enough || leafhdr.forw == 0)
334 break;
335 cursor->blkno = leafhdr.forw;
336 xfs_trans_brelse(context->tp, bp);
337 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
338 if (error)
339 return error;
341 xfs_trans_brelse(context->tp, bp);
342 return 0;
346 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
348 void
349 xfs_attr3_leaf_list_int(
350 struct xfs_buf *bp,
351 struct xfs_attr_list_context *context)
353 struct attrlist_cursor_kern *cursor;
354 struct xfs_attr_leafblock *leaf;
355 struct xfs_attr3_icleaf_hdr ichdr;
356 struct xfs_attr_leaf_entry *entries;
357 struct xfs_attr_leaf_entry *entry;
358 int i;
359 struct xfs_mount *mp = context->dp->i_mount;
361 trace_xfs_attr_list_leaf(context);
363 leaf = bp->b_addr;
364 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
365 entries = xfs_attr3_leaf_entryp(leaf);
367 cursor = context->cursor;
368 cursor->initted = 1;
371 * Re-find our place in the leaf block if this is a new syscall.
373 if (context->resynch) {
374 entry = &entries[0];
375 for (i = 0; i < ichdr.count; entry++, i++) {
376 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
377 if (cursor->offset == context->dupcnt) {
378 context->dupcnt = 0;
379 break;
381 context->dupcnt++;
382 } else if (be32_to_cpu(entry->hashval) >
383 cursor->hashval) {
384 context->dupcnt = 0;
385 break;
388 if (i == ichdr.count) {
389 trace_xfs_attr_list_notfound(context);
390 return;
392 } else {
393 entry = &entries[0];
394 i = 0;
396 context->resynch = 0;
399 * We have found our place, start copying out the new attributes.
401 for (; i < ichdr.count; entry++, i++) {
402 char *name;
403 int namelen, valuelen;
405 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
406 cursor->hashval = be32_to_cpu(entry->hashval);
407 cursor->offset = 0;
410 if (entry->flags & XFS_ATTR_INCOMPLETE)
411 continue; /* skip incomplete entries */
413 if (entry->flags & XFS_ATTR_LOCAL) {
414 xfs_attr_leaf_name_local_t *name_loc;
416 name_loc = xfs_attr3_leaf_name_local(leaf, i);
417 name = name_loc->nameval;
418 namelen = name_loc->namelen;
419 valuelen = be16_to_cpu(name_loc->valuelen);
420 } else {
421 xfs_attr_leaf_name_remote_t *name_rmt;
423 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
424 name = name_rmt->name;
425 namelen = name_rmt->namelen;
426 valuelen = be32_to_cpu(name_rmt->valuelen);
429 context->put_listent(context, entry->flags,
430 name, namelen, valuelen);
431 if (context->seen_enough)
432 break;
433 cursor->offset++;
435 trace_xfs_attr_list_leaf_end(context);
436 return;
440 * Copy out attribute entries for attr_list(), for leaf attribute lists.
442 STATIC int
443 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
445 int error;
446 struct xfs_buf *bp;
448 trace_xfs_attr_leaf_list(context);
450 context->cursor->blkno = 0;
451 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
452 if (error)
453 return error;
455 xfs_attr3_leaf_list_int(bp, context);
456 xfs_trans_brelse(context->tp, bp);
457 return 0;
461 xfs_attr_list_int_ilocked(
462 struct xfs_attr_list_context *context)
464 struct xfs_inode *dp = context->dp;
466 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
469 * Decide on what work routines to call based on the inode size.
471 if (!xfs_inode_hasattr(dp))
472 return 0;
473 else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
474 return xfs_attr_shortform_list(context);
475 else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
476 return xfs_attr_leaf_list(context);
477 return xfs_attr_node_list(context);
481 xfs_attr_list_int(
482 xfs_attr_list_context_t *context)
484 int error;
485 xfs_inode_t *dp = context->dp;
486 uint lock_mode;
488 XFS_STATS_INC(dp->i_mount, xs_attr_list);
490 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
491 return -EIO;
493 lock_mode = xfs_ilock_attr_map_shared(dp);
494 error = xfs_attr_list_int_ilocked(context);
495 xfs_iunlock(dp, lock_mode);
496 return error;
499 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
500 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
501 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
502 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
503 & ~(sizeof(u_int32_t)-1))
506 * Format an attribute and copy it out to the user's buffer.
507 * Take care to check values and protect against them changing later,
508 * we may be reading them directly out of a user buffer.
510 STATIC void
511 xfs_attr_put_listent(
512 xfs_attr_list_context_t *context,
513 int flags,
514 unsigned char *name,
515 int namelen,
516 int valuelen)
518 struct attrlist *alist = (struct attrlist *)context->alist;
519 attrlist_ent_t *aep;
520 int arraytop;
522 ASSERT(!(context->flags & ATTR_KERNOVAL));
523 ASSERT(context->count >= 0);
524 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
525 ASSERT(context->firstu >= sizeof(*alist));
526 ASSERT(context->firstu <= context->bufsize);
529 * Only list entries in the right namespace.
531 if (((context->flags & ATTR_SECURE) == 0) !=
532 ((flags & XFS_ATTR_SECURE) == 0))
533 return;
534 if (((context->flags & ATTR_ROOT) == 0) !=
535 ((flags & XFS_ATTR_ROOT) == 0))
536 return;
538 arraytop = sizeof(*alist) +
539 context->count * sizeof(alist->al_offset[0]);
540 context->firstu -= ATTR_ENTSIZE(namelen);
541 if (context->firstu < arraytop) {
542 trace_xfs_attr_list_full(context);
543 alist->al_more = 1;
544 context->seen_enough = 1;
545 return;
548 aep = (attrlist_ent_t *)&context->alist[context->firstu];
549 aep->a_valuelen = valuelen;
550 memcpy(aep->a_name, name, namelen);
551 aep->a_name[namelen] = 0;
552 alist->al_offset[context->count++] = context->firstu;
553 alist->al_count = context->count;
554 trace_xfs_attr_list_add(context);
555 return;
559 * Generate a list of extended attribute names and optionally
560 * also value lengths. Positive return value follows the XFS
561 * convention of being an error, zero or negative return code
562 * is the length of the buffer returned (negated), indicating
563 * success.
566 xfs_attr_list(
567 xfs_inode_t *dp,
568 char *buffer,
569 int bufsize,
570 int flags,
571 attrlist_cursor_kern_t *cursor)
573 xfs_attr_list_context_t context;
574 struct attrlist *alist;
575 int error;
578 * Validate the cursor.
580 if (cursor->pad1 || cursor->pad2)
581 return -EINVAL;
582 if ((cursor->initted == 0) &&
583 (cursor->hashval || cursor->blkno || cursor->offset))
584 return -EINVAL;
587 * Check for a properly aligned buffer.
589 if (((long)buffer) & (sizeof(int)-1))
590 return -EFAULT;
591 if (flags & ATTR_KERNOVAL)
592 bufsize = 0;
595 * Initialize the output buffer.
597 memset(&context, 0, sizeof(context));
598 context.dp = dp;
599 context.cursor = cursor;
600 context.resynch = 1;
601 context.flags = flags;
602 context.alist = buffer;
603 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
604 context.firstu = context.bufsize;
605 context.put_listent = xfs_attr_put_listent;
607 alist = (struct attrlist *)context.alist;
608 alist->al_count = 0;
609 alist->al_more = 0;
610 alist->al_offset[0] = context.bufsize;
612 error = xfs_attr_list_int(&context);
613 ASSERT(error <= 0);
614 return error;