1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
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"
19 #include "xfs_attr_sf.h"
20 #include "xfs_attr_leaf.h"
21 #include "xfs_error.h"
22 #include "xfs_trace.h"
26 xfs_attr_shortform_compare(const void *a, const void *b)
28 xfs_attr_sf_sort_t *sa, *sb;
30 sa = (xfs_attr_sf_sort_t *)a;
31 sb = (xfs_attr_sf_sort_t *)b;
32 if (sa->hash < sb->hash) {
34 } else if (sa->hash > sb->hash) {
37 return sa->entno - sb->entno;
41 #define XFS_ISRESET_CURSOR(cursor) \
42 (!((cursor)->initted) && !((cursor)->hashval) && \
43 !((cursor)->blkno) && !((cursor)->offset))
45 * Copy out entries of shortform attribute lists for attr_list().
46 * Shortform attribute lists are not stored in hashval sorted order.
47 * If the output buffer is not large enough to hold them all, then we
48 * we have to calculate each entries' hashvalue and sort them before
49 * we can begin returning them to the user.
52 xfs_attr_shortform_list(
53 struct xfs_attr_list_context *context)
55 struct attrlist_cursor_kern *cursor;
56 struct xfs_attr_sf_sort *sbuf, *sbp;
57 struct xfs_attr_shortform *sf;
58 struct xfs_attr_sf_entry *sfe;
60 int sbsize, nsbuf, count, i;
63 ASSERT(context != NULL);
66 ASSERT(dp->i_afp != NULL);
67 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
71 cursor = context->cursor;
72 ASSERT(cursor != NULL);
74 trace_xfs_attr_list_sf(context);
77 * If the buffer is large enough and the cursor is at the start,
78 * do not bother with sorting since we will return everything in
79 * one buffer and another call using the cursor won't need to be
81 * Note the generous fudge factor of 16 overhead bytes per entry.
82 * If bufsize is zero then put_listent must be a search function
83 * and can just scan through what we have.
85 if (context->bufsize == 0 ||
86 (XFS_ISRESET_CURSOR(cursor) &&
87 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
88 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
89 if (!xfs_attr_namecheck(sfe->nameval, sfe->namelen)) {
90 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW,
91 context->dp->i_mount);
94 context->put_listent(context,
100 * Either search callback finished early or
101 * didn't fit it all in the buffer after all.
103 if (context->seen_enough)
105 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
107 trace_xfs_attr_list_sf_all(context);
111 /* do no more for a search callback */
112 if (context->bufsize == 0)
116 * It didn't all fit, so we have to sort everything on hashval.
118 sbsize = sf->hdr.count * sizeof(*sbuf);
119 sbp = sbuf = kmem_alloc(sbsize, KM_NOFS);
122 * Scan the attribute list for the rest of the entries, storing
123 * the relevant info from only those that match into a buffer.
126 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
128 ((char *)sfe < (char *)sf) ||
129 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
130 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
132 context->dp->i_mount, sfe,
135 return -EFSCORRUPTED;
139 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
140 sbp->name = sfe->nameval;
141 sbp->namelen = sfe->namelen;
142 /* These are bytes, and both on-disk, don't endian-flip */
143 sbp->valuelen = sfe->valuelen;
144 sbp->flags = sfe->flags;
145 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
151 * Sort the entries on hash then entno.
153 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
156 * Re-find our place IN THE SORTED LIST.
161 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
162 if (sbp->hash == cursor->hashval) {
163 if (cursor->offset == count) {
167 } else if (sbp->hash > cursor->hashval) {
175 * Loop putting entries into the user buffer.
177 for ( ; i < nsbuf; i++, sbp++) {
178 if (cursor->hashval != sbp->hash) {
179 cursor->hashval = sbp->hash;
182 if (!xfs_attr_namecheck(sbp->name, sbp->namelen)) {
183 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW,
184 context->dp->i_mount);
185 error = -EFSCORRUPTED;
188 context->put_listent(context,
193 if (context->seen_enough)
203 * We didn't find the block & hash mentioned in the cursor state, so
204 * walk down the attr btree looking for the hash.
207 xfs_attr_node_list_lookup(
208 struct xfs_attr_list_context *context,
209 struct attrlist_cursor_kern *cursor,
210 struct xfs_buf **pbp)
212 struct xfs_da3_icnode_hdr nodehdr;
213 struct xfs_da_intnode *node;
214 struct xfs_da_node_entry *btree;
215 struct xfs_inode *dp = context->dp;
216 struct xfs_mount *mp = dp->i_mount;
217 struct xfs_trans *tp = context->tp;
221 unsigned int expected_level = 0;
224 ASSERT(*pbp == NULL);
227 error = xfs_da3_node_read(tp, dp, cursor->blkno, -1, &bp,
232 magic = be16_to_cpu(node->hdr.info.magic);
233 if (magic == XFS_ATTR_LEAF_MAGIC ||
234 magic == XFS_ATTR3_LEAF_MAGIC)
236 if (magic != XFS_DA_NODE_MAGIC &&
237 magic != XFS_DA3_NODE_MAGIC) {
238 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
239 node, sizeof(*node));
243 xfs_da3_node_hdr_from_disk(mp, &nodehdr, node);
245 /* Tree taller than we can handle; bail out! */
246 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
249 /* Check the level from the root node. */
250 if (cursor->blkno == 0)
251 expected_level = nodehdr.level - 1;
252 else if (expected_level != nodehdr.level)
257 btree = dp->d_ops->node_tree_p(node);
258 for (i = 0; i < nodehdr.count; btree++, i++) {
259 if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
260 cursor->blkno = be32_to_cpu(btree->before);
261 trace_xfs_attr_list_node_descend(context,
266 xfs_trans_brelse(tp, bp);
268 if (i == nodehdr.count)
271 /* We can't point back to the root. */
272 if (cursor->blkno == 0) {
273 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
274 return -EFSCORRUPTED;
278 if (expected_level != 0)
285 xfs_buf_corruption_error(bp);
286 xfs_trans_brelse(tp, bp);
287 return -EFSCORRUPTED;
292 struct xfs_attr_list_context *context)
294 struct xfs_attr3_icleaf_hdr leafhdr;
295 struct attrlist_cursor_kern *cursor;
296 struct xfs_attr_leafblock *leaf;
297 struct xfs_da_intnode *node;
299 struct xfs_inode *dp = context->dp;
300 struct xfs_mount *mp = dp->i_mount;
303 trace_xfs_attr_node_list(context);
305 cursor = context->cursor;
309 * Do all sorts of validation on the passed-in cursor structure.
310 * If anything is amiss, ignore the cursor and look up the hashval
311 * starting from the btree root.
314 if (cursor->blkno > 0) {
315 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
317 if ((error != 0) && (error != -EFSCORRUPTED))
320 struct xfs_attr_leaf_entry *entries;
323 switch (be16_to_cpu(node->hdr.info.magic)) {
324 case XFS_DA_NODE_MAGIC:
325 case XFS_DA3_NODE_MAGIC:
326 trace_xfs_attr_list_wrong_blk(context);
327 xfs_trans_brelse(context->tp, bp);
330 case XFS_ATTR_LEAF_MAGIC:
331 case XFS_ATTR3_LEAF_MAGIC:
333 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
335 entries = xfs_attr3_leaf_entryp(leaf);
336 if (cursor->hashval > be32_to_cpu(
337 entries[leafhdr.count - 1].hashval)) {
338 trace_xfs_attr_list_wrong_blk(context);
339 xfs_trans_brelse(context->tp, bp);
341 } else if (cursor->hashval <= be32_to_cpu(
342 entries[0].hashval)) {
343 trace_xfs_attr_list_wrong_blk(context);
344 xfs_trans_brelse(context->tp, bp);
349 trace_xfs_attr_list_wrong_blk(context);
350 xfs_trans_brelse(context->tp, bp);
357 * We did not find what we expected given the cursor's contents,
358 * so we start from the top and work down based on the hash value.
359 * Note that start of node block is same as start of leaf block.
362 error = xfs_attr_node_list_lookup(context, cursor, &bp);
369 * Roll upward through the blocks, processing each leaf block in
370 * order. As long as there is space in the result buffer, keep
371 * adding the information.
375 error = xfs_attr3_leaf_list_int(bp, context);
378 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
379 if (context->seen_enough || leafhdr.forw == 0)
381 cursor->blkno = leafhdr.forw;
382 xfs_trans_brelse(context->tp, bp);
383 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
387 xfs_trans_brelse(context->tp, bp);
392 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
395 xfs_attr3_leaf_list_int(
397 struct xfs_attr_list_context *context)
399 struct attrlist_cursor_kern *cursor;
400 struct xfs_attr_leafblock *leaf;
401 struct xfs_attr3_icleaf_hdr ichdr;
402 struct xfs_attr_leaf_entry *entries;
403 struct xfs_attr_leaf_entry *entry;
405 struct xfs_mount *mp = context->dp->i_mount;
407 trace_xfs_attr_list_leaf(context);
410 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
411 entries = xfs_attr3_leaf_entryp(leaf);
413 cursor = context->cursor;
417 * Re-find our place in the leaf block if this is a new syscall.
419 if (context->resynch) {
421 for (i = 0; i < ichdr.count; entry++, i++) {
422 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
423 if (cursor->offset == context->dupcnt) {
428 } else if (be32_to_cpu(entry->hashval) >
434 if (i == ichdr.count) {
435 trace_xfs_attr_list_notfound(context);
442 context->resynch = 0;
445 * We have found our place, start copying out the new attributes.
447 for (; i < ichdr.count; entry++, i++) {
449 int namelen, valuelen;
451 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
452 cursor->hashval = be32_to_cpu(entry->hashval);
456 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
457 !(context->flags & ATTR_INCOMPLETE))
458 continue; /* skip incomplete entries */
460 if (entry->flags & XFS_ATTR_LOCAL) {
461 xfs_attr_leaf_name_local_t *name_loc;
463 name_loc = xfs_attr3_leaf_name_local(leaf, i);
464 name = name_loc->nameval;
465 namelen = name_loc->namelen;
466 valuelen = be16_to_cpu(name_loc->valuelen);
468 xfs_attr_leaf_name_remote_t *name_rmt;
470 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
471 name = name_rmt->name;
472 namelen = name_rmt->namelen;
473 valuelen = be32_to_cpu(name_rmt->valuelen);
476 if (!xfs_attr_namecheck(name, namelen)) {
477 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW,
478 context->dp->i_mount);
479 return -EFSCORRUPTED;
481 context->put_listent(context, entry->flags,
482 name, namelen, valuelen);
483 if (context->seen_enough)
487 trace_xfs_attr_list_leaf_end(context);
492 * Copy out attribute entries for attr_list(), for leaf attribute lists.
495 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
500 trace_xfs_attr_leaf_list(context);
502 context->cursor->blkno = 0;
503 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
507 error = xfs_attr3_leaf_list_int(bp, context);
508 xfs_trans_brelse(context->tp, bp);
513 xfs_attr_list_int_ilocked(
514 struct xfs_attr_list_context *context)
516 struct xfs_inode *dp = context->dp;
518 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
521 * Decide on what work routines to call based on the inode size.
523 if (!xfs_inode_hasattr(dp))
525 else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
526 return xfs_attr_shortform_list(context);
527 else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
528 return xfs_attr_leaf_list(context);
529 return xfs_attr_node_list(context);
534 xfs_attr_list_context_t *context)
537 xfs_inode_t *dp = context->dp;
540 XFS_STATS_INC(dp->i_mount, xs_attr_list);
542 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
545 lock_mode = xfs_ilock_attr_map_shared(dp);
546 error = xfs_attr_list_int_ilocked(context);
547 xfs_iunlock(dp, lock_mode);
551 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
552 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
553 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
554 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
555 & ~(sizeof(uint32_t)-1))
558 * Format an attribute and copy it out to the user's buffer.
559 * Take care to check values and protect against them changing later,
560 * we may be reading them directly out of a user buffer.
563 xfs_attr_put_listent(
564 xfs_attr_list_context_t *context,
570 struct attrlist *alist = (struct attrlist *)context->alist;
574 ASSERT(!context->seen_enough);
575 ASSERT(!(context->flags & ATTR_KERNOVAL));
576 ASSERT(context->count >= 0);
577 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
578 ASSERT(context->firstu >= sizeof(*alist));
579 ASSERT(context->firstu <= context->bufsize);
582 * Only list entries in the right namespace.
584 if (((context->flags & ATTR_SECURE) == 0) !=
585 ((flags & XFS_ATTR_SECURE) == 0))
587 if (((context->flags & ATTR_ROOT) == 0) !=
588 ((flags & XFS_ATTR_ROOT) == 0))
591 arraytop = sizeof(*alist) +
592 context->count * sizeof(alist->al_offset[0]);
593 context->firstu -= ATTR_ENTSIZE(namelen);
594 if (context->firstu < arraytop) {
595 trace_xfs_attr_list_full(context);
597 context->seen_enough = 1;
601 aep = (attrlist_ent_t *)&context->alist[context->firstu];
602 aep->a_valuelen = valuelen;
603 memcpy(aep->a_name, name, namelen);
604 aep->a_name[namelen] = 0;
605 alist->al_offset[context->count++] = context->firstu;
606 alist->al_count = context->count;
607 trace_xfs_attr_list_add(context);
612 * Generate a list of extended attribute names and optionally
613 * also value lengths. Positive return value follows the XFS
614 * convention of being an error, zero or negative return code
615 * is the length of the buffer returned (negated), indicating
624 attrlist_cursor_kern_t *cursor)
626 xfs_attr_list_context_t context;
627 struct attrlist *alist;
631 * Validate the cursor.
633 if (cursor->pad1 || cursor->pad2)
635 if ((cursor->initted == 0) &&
636 (cursor->hashval || cursor->blkno || cursor->offset))
639 /* Only internal consumers can retrieve incomplete attrs. */
640 if (flags & ATTR_INCOMPLETE)
644 * Check for a properly aligned buffer.
646 if (((long)buffer) & (sizeof(int)-1))
648 if (flags & ATTR_KERNOVAL)
652 * Initialize the output buffer.
654 memset(&context, 0, sizeof(context));
656 context.cursor = cursor;
658 context.flags = flags;
659 context.alist = buffer;
660 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
661 context.firstu = context.bufsize;
662 context.put_listent = xfs_attr_put_listent;
664 alist = (struct attrlist *)context.alist;
667 alist->al_offset[0] = context.bufsize;
669 error = xfs_attr_list_int(&context);