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(xfs_attr_list_context_t *context)
54 attrlist_cursor_kern_t *cursor;
55 xfs_attr_sf_sort_t *sbuf, *sbp;
56 xfs_attr_shortform_t *sf;
57 xfs_attr_sf_entry_t *sfe;
59 int sbsize, nsbuf, count, i;
61 ASSERT(context != NULL);
64 ASSERT(dp->i_afp != NULL);
65 sf = (xfs_attr_shortform_t *)dp->i_afp->if_u1.if_data;
69 cursor = context->cursor;
70 ASSERT(cursor != NULL);
72 trace_xfs_attr_list_sf(context);
75 * If the buffer is large enough and the cursor is at the start,
76 * do not bother with sorting since we will return everything in
77 * one buffer and another call using the cursor won't need to be
79 * Note the generous fudge factor of 16 overhead bytes per entry.
80 * If bufsize is zero then put_listent must be a search function
81 * and can just scan through what we have.
83 if (context->bufsize == 0 ||
84 (XFS_ISRESET_CURSOR(cursor) &&
85 (dp->i_afp->if_bytes + sf->hdr.count * 16) < context->bufsize)) {
86 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
87 context->put_listent(context,
93 * Either search callback finished early or
94 * didn't fit it all in the buffer after all.
96 if (context->seen_enough)
98 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
100 trace_xfs_attr_list_sf_all(context);
104 /* do no more for a search callback */
105 if (context->bufsize == 0)
109 * It didn't all fit, so we have to sort everything on hashval.
111 sbsize = sf->hdr.count * sizeof(*sbuf);
112 sbp = sbuf = kmem_alloc(sbsize, KM_NOFS);
115 * Scan the attribute list for the rest of the entries, storing
116 * the relevant info from only those that match into a buffer.
119 for (i = 0, sfe = &sf->list[0]; i < sf->hdr.count; i++) {
121 ((char *)sfe < (char *)sf) ||
122 ((char *)sfe >= ((char *)sf + dp->i_afp->if_bytes)))) {
123 XFS_CORRUPTION_ERROR("xfs_attr_shortform_list",
125 context->dp->i_mount, sfe,
128 return -EFSCORRUPTED;
132 sbp->hash = xfs_da_hashname(sfe->nameval, sfe->namelen);
133 sbp->name = sfe->nameval;
134 sbp->namelen = sfe->namelen;
135 /* These are bytes, and both on-disk, don't endian-flip */
136 sbp->valuelen = sfe->valuelen;
137 sbp->flags = sfe->flags;
138 sfe = XFS_ATTR_SF_NEXTENTRY(sfe);
144 * Sort the entries on hash then entno.
146 xfs_sort(sbuf, nsbuf, sizeof(*sbuf), xfs_attr_shortform_compare);
149 * Re-find our place IN THE SORTED LIST.
154 for (sbp = sbuf, i = 0; i < nsbuf; i++, sbp++) {
155 if (sbp->hash == cursor->hashval) {
156 if (cursor->offset == count) {
160 } else if (sbp->hash > cursor->hashval) {
170 * Loop putting entries into the user buffer.
172 for ( ; i < nsbuf; i++, sbp++) {
173 if (cursor->hashval != sbp->hash) {
174 cursor->hashval = sbp->hash;
177 context->put_listent(context,
182 if (context->seen_enough)
192 * We didn't find the block & hash mentioned in the cursor state, so
193 * walk down the attr btree looking for the hash.
196 xfs_attr_node_list_lookup(
197 struct xfs_attr_list_context *context,
198 struct attrlist_cursor_kern *cursor,
199 struct xfs_buf **pbp)
201 struct xfs_da3_icnode_hdr nodehdr;
202 struct xfs_da_intnode *node;
203 struct xfs_da_node_entry *btree;
204 struct xfs_inode *dp = context->dp;
205 struct xfs_mount *mp = dp->i_mount;
206 struct xfs_trans *tp = context->tp;
210 unsigned int expected_level = 0;
213 ASSERT(*pbp == NULL);
216 error = xfs_da3_node_read(tp, dp, cursor->blkno, -1, &bp,
221 magic = be16_to_cpu(node->hdr.info.magic);
222 if (magic == XFS_ATTR_LEAF_MAGIC ||
223 magic == XFS_ATTR3_LEAF_MAGIC)
225 if (magic != XFS_DA_NODE_MAGIC &&
226 magic != XFS_DA3_NODE_MAGIC) {
227 XFS_CORRUPTION_ERROR(__func__, XFS_ERRLEVEL_LOW, mp,
228 node, sizeof(*node));
232 dp->d_ops->node_hdr_from_disk(&nodehdr, node);
234 /* Tree taller than we can handle; bail out! */
235 if (nodehdr.level >= XFS_DA_NODE_MAXDEPTH)
238 /* Check the level from the root node. */
239 if (cursor->blkno == 0)
240 expected_level = nodehdr.level - 1;
241 else if (expected_level != nodehdr.level)
246 btree = dp->d_ops->node_tree_p(node);
247 for (i = 0; i < nodehdr.count; btree++, i++) {
248 if (cursor->hashval <= be32_to_cpu(btree->hashval)) {
249 cursor->blkno = be32_to_cpu(btree->before);
250 trace_xfs_attr_list_node_descend(context,
255 xfs_trans_brelse(tp, bp);
257 if (i == nodehdr.count)
260 /* We can't point back to the root. */
261 if (cursor->blkno == 0)
262 return -EFSCORRUPTED;
265 if (expected_level != 0)
272 xfs_trans_brelse(tp, bp);
273 return -EFSCORRUPTED;
278 struct xfs_attr_list_context *context)
280 struct xfs_attr3_icleaf_hdr leafhdr;
281 struct attrlist_cursor_kern *cursor;
282 struct xfs_attr_leafblock *leaf;
283 struct xfs_da_intnode *node;
285 struct xfs_inode *dp = context->dp;
286 struct xfs_mount *mp = dp->i_mount;
289 trace_xfs_attr_node_list(context);
291 cursor = context->cursor;
295 * Do all sorts of validation on the passed-in cursor structure.
296 * If anything is amiss, ignore the cursor and look up the hashval
297 * starting from the btree root.
300 if (cursor->blkno > 0) {
301 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
303 if ((error != 0) && (error != -EFSCORRUPTED))
306 struct xfs_attr_leaf_entry *entries;
309 switch (be16_to_cpu(node->hdr.info.magic)) {
310 case XFS_DA_NODE_MAGIC:
311 case XFS_DA3_NODE_MAGIC:
312 trace_xfs_attr_list_wrong_blk(context);
313 xfs_trans_brelse(context->tp, bp);
316 case XFS_ATTR_LEAF_MAGIC:
317 case XFS_ATTR3_LEAF_MAGIC:
319 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
321 entries = xfs_attr3_leaf_entryp(leaf);
322 if (cursor->hashval > be32_to_cpu(
323 entries[leafhdr.count - 1].hashval)) {
324 trace_xfs_attr_list_wrong_blk(context);
325 xfs_trans_brelse(context->tp, bp);
327 } else if (cursor->hashval <= be32_to_cpu(
328 entries[0].hashval)) {
329 trace_xfs_attr_list_wrong_blk(context);
330 xfs_trans_brelse(context->tp, bp);
335 trace_xfs_attr_list_wrong_blk(context);
336 xfs_trans_brelse(context->tp, bp);
343 * We did not find what we expected given the cursor's contents,
344 * so we start from the top and work down based on the hash value.
345 * Note that start of node block is same as start of leaf block.
348 error = xfs_attr_node_list_lookup(context, cursor, &bp);
355 * Roll upward through the blocks, processing each leaf block in
356 * order. As long as there is space in the result buffer, keep
357 * adding the information.
361 xfs_attr3_leaf_list_int(bp, context);
362 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
363 if (context->seen_enough || leafhdr.forw == 0)
365 cursor->blkno = leafhdr.forw;
366 xfs_trans_brelse(context->tp, bp);
367 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
371 xfs_trans_brelse(context->tp, bp);
376 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
379 xfs_attr3_leaf_list_int(
381 struct xfs_attr_list_context *context)
383 struct attrlist_cursor_kern *cursor;
384 struct xfs_attr_leafblock *leaf;
385 struct xfs_attr3_icleaf_hdr ichdr;
386 struct xfs_attr_leaf_entry *entries;
387 struct xfs_attr_leaf_entry *entry;
389 struct xfs_mount *mp = context->dp->i_mount;
391 trace_xfs_attr_list_leaf(context);
394 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
395 entries = xfs_attr3_leaf_entryp(leaf);
397 cursor = context->cursor;
401 * Re-find our place in the leaf block if this is a new syscall.
403 if (context->resynch) {
405 for (i = 0; i < ichdr.count; entry++, i++) {
406 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
407 if (cursor->offset == context->dupcnt) {
412 } else if (be32_to_cpu(entry->hashval) >
418 if (i == ichdr.count) {
419 trace_xfs_attr_list_notfound(context);
426 context->resynch = 0;
429 * We have found our place, start copying out the new attributes.
431 for (; i < ichdr.count; entry++, i++) {
433 int namelen, valuelen;
435 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
436 cursor->hashval = be32_to_cpu(entry->hashval);
440 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
441 !(context->flags & ATTR_INCOMPLETE))
442 continue; /* skip incomplete entries */
444 if (entry->flags & XFS_ATTR_LOCAL) {
445 xfs_attr_leaf_name_local_t *name_loc;
447 name_loc = xfs_attr3_leaf_name_local(leaf, i);
448 name = name_loc->nameval;
449 namelen = name_loc->namelen;
450 valuelen = be16_to_cpu(name_loc->valuelen);
452 xfs_attr_leaf_name_remote_t *name_rmt;
454 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
455 name = name_rmt->name;
456 namelen = name_rmt->namelen;
457 valuelen = be32_to_cpu(name_rmt->valuelen);
460 context->put_listent(context, entry->flags,
461 name, namelen, valuelen);
462 if (context->seen_enough)
466 trace_xfs_attr_list_leaf_end(context);
471 * Copy out attribute entries for attr_list(), for leaf attribute lists.
474 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
479 trace_xfs_attr_leaf_list(context);
481 context->cursor->blkno = 0;
482 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
486 xfs_attr3_leaf_list_int(bp, context);
487 xfs_trans_brelse(context->tp, bp);
492 xfs_attr_list_int_ilocked(
493 struct xfs_attr_list_context *context)
495 struct xfs_inode *dp = context->dp;
497 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
500 * Decide on what work routines to call based on the inode size.
502 if (!xfs_inode_hasattr(dp))
504 else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
505 return xfs_attr_shortform_list(context);
506 else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
507 return xfs_attr_leaf_list(context);
508 return xfs_attr_node_list(context);
513 xfs_attr_list_context_t *context)
516 xfs_inode_t *dp = context->dp;
519 XFS_STATS_INC(dp->i_mount, xs_attr_list);
521 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
524 lock_mode = xfs_ilock_attr_map_shared(dp);
525 error = xfs_attr_list_int_ilocked(context);
526 xfs_iunlock(dp, lock_mode);
530 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
531 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
532 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
533 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
534 & ~(sizeof(uint32_t)-1))
537 * Format an attribute and copy it out to the user's buffer.
538 * Take care to check values and protect against them changing later,
539 * we may be reading them directly out of a user buffer.
542 xfs_attr_put_listent(
543 xfs_attr_list_context_t *context,
549 struct attrlist *alist = (struct attrlist *)context->alist;
553 ASSERT(!context->seen_enough);
554 ASSERT(!(context->flags & ATTR_KERNOVAL));
555 ASSERT(context->count >= 0);
556 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
557 ASSERT(context->firstu >= sizeof(*alist));
558 ASSERT(context->firstu <= context->bufsize);
561 * Only list entries in the right namespace.
563 if (((context->flags & ATTR_SECURE) == 0) !=
564 ((flags & XFS_ATTR_SECURE) == 0))
566 if (((context->flags & ATTR_ROOT) == 0) !=
567 ((flags & XFS_ATTR_ROOT) == 0))
570 arraytop = sizeof(*alist) +
571 context->count * sizeof(alist->al_offset[0]);
572 context->firstu -= ATTR_ENTSIZE(namelen);
573 if (context->firstu < arraytop) {
574 trace_xfs_attr_list_full(context);
576 context->seen_enough = 1;
580 aep = (attrlist_ent_t *)&context->alist[context->firstu];
581 aep->a_valuelen = valuelen;
582 memcpy(aep->a_name, name, namelen);
583 aep->a_name[namelen] = 0;
584 alist->al_offset[context->count++] = context->firstu;
585 alist->al_count = context->count;
586 trace_xfs_attr_list_add(context);
591 * Generate a list of extended attribute names and optionally
592 * also value lengths. Positive return value follows the XFS
593 * convention of being an error, zero or negative return code
594 * is the length of the buffer returned (negated), indicating
603 attrlist_cursor_kern_t *cursor)
605 xfs_attr_list_context_t context;
606 struct attrlist *alist;
610 * Validate the cursor.
612 if (cursor->pad1 || cursor->pad2)
614 if ((cursor->initted == 0) &&
615 (cursor->hashval || cursor->blkno || cursor->offset))
618 /* Only internal consumers can retrieve incomplete attrs. */
619 if (flags & ATTR_INCOMPLETE)
623 * Check for a properly aligned buffer.
625 if (((long)buffer) & (sizeof(int)-1))
627 if (flags & ATTR_KERNOVAL)
631 * Initialize the output buffer.
633 memset(&context, 0, sizeof(context));
635 context.cursor = cursor;
637 context.flags = flags;
638 context.alist = buffer;
639 context.bufsize = (bufsize & ~(sizeof(int)-1)); /* align */
640 context.firstu = context.bufsize;
641 context.put_listent = xfs_attr_put_listent;
643 alist = (struct attrlist *)context.alist;
646 alist->al_offset[0] = context.bufsize;
648 error = xfs_attr_list_int(&context);