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 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
263 return -EFSCORRUPTED;
267 if (expected_level != 0)
274 xfs_buf_mark_corrupt(bp);
275 xfs_trans_brelse(tp, bp);
276 return -EFSCORRUPTED;
281 struct xfs_attr_list_context *context)
283 struct xfs_attr3_icleaf_hdr leafhdr;
284 struct attrlist_cursor_kern *cursor;
285 struct xfs_attr_leafblock *leaf;
286 struct xfs_da_intnode *node;
288 struct xfs_inode *dp = context->dp;
289 struct xfs_mount *mp = dp->i_mount;
292 trace_xfs_attr_node_list(context);
294 cursor = context->cursor;
298 * Do all sorts of validation on the passed-in cursor structure.
299 * If anything is amiss, ignore the cursor and look up the hashval
300 * starting from the btree root.
303 if (cursor->blkno > 0) {
304 error = xfs_da3_node_read(context->tp, dp, cursor->blkno, -1,
306 if ((error != 0) && (error != -EFSCORRUPTED))
309 struct xfs_attr_leaf_entry *entries;
312 switch (be16_to_cpu(node->hdr.info.magic)) {
313 case XFS_DA_NODE_MAGIC:
314 case XFS_DA3_NODE_MAGIC:
315 trace_xfs_attr_list_wrong_blk(context);
316 xfs_trans_brelse(context->tp, bp);
319 case XFS_ATTR_LEAF_MAGIC:
320 case XFS_ATTR3_LEAF_MAGIC:
322 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo,
324 entries = xfs_attr3_leaf_entryp(leaf);
325 if (cursor->hashval > be32_to_cpu(
326 entries[leafhdr.count - 1].hashval)) {
327 trace_xfs_attr_list_wrong_blk(context);
328 xfs_trans_brelse(context->tp, bp);
330 } else if (cursor->hashval <= be32_to_cpu(
331 entries[0].hashval)) {
332 trace_xfs_attr_list_wrong_blk(context);
333 xfs_trans_brelse(context->tp, bp);
338 trace_xfs_attr_list_wrong_blk(context);
339 xfs_trans_brelse(context->tp, bp);
346 * We did not find what we expected given the cursor's contents,
347 * so we start from the top and work down based on the hash value.
348 * Note that start of node block is same as start of leaf block.
351 error = xfs_attr_node_list_lookup(context, cursor, &bp);
358 * Roll upward through the blocks, processing each leaf block in
359 * order. As long as there is space in the result buffer, keep
360 * adding the information.
364 xfs_attr3_leaf_list_int(bp, context);
365 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
366 if (context->seen_enough || leafhdr.forw == 0)
368 cursor->blkno = leafhdr.forw;
369 xfs_trans_brelse(context->tp, bp);
370 error = xfs_attr3_leaf_read(context->tp, dp, cursor->blkno, -1, &bp);
374 xfs_trans_brelse(context->tp, bp);
379 * Copy out attribute list entries for attr_list(), for leaf attribute lists.
382 xfs_attr3_leaf_list_int(
384 struct xfs_attr_list_context *context)
386 struct attrlist_cursor_kern *cursor;
387 struct xfs_attr_leafblock *leaf;
388 struct xfs_attr3_icleaf_hdr ichdr;
389 struct xfs_attr_leaf_entry *entries;
390 struct xfs_attr_leaf_entry *entry;
392 struct xfs_mount *mp = context->dp->i_mount;
394 trace_xfs_attr_list_leaf(context);
397 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
398 entries = xfs_attr3_leaf_entryp(leaf);
400 cursor = context->cursor;
404 * Re-find our place in the leaf block if this is a new syscall.
406 if (context->resynch) {
408 for (i = 0; i < ichdr.count; entry++, i++) {
409 if (be32_to_cpu(entry->hashval) == cursor->hashval) {
410 if (cursor->offset == context->dupcnt) {
415 } else if (be32_to_cpu(entry->hashval) >
421 if (i == ichdr.count) {
422 trace_xfs_attr_list_notfound(context);
429 context->resynch = 0;
432 * We have found our place, start copying out the new attributes.
434 for (; i < ichdr.count; entry++, i++) {
436 int namelen, valuelen;
438 if (be32_to_cpu(entry->hashval) != cursor->hashval) {
439 cursor->hashval = be32_to_cpu(entry->hashval);
443 if ((entry->flags & XFS_ATTR_INCOMPLETE) &&
444 !(context->flags & ATTR_INCOMPLETE))
445 continue; /* skip incomplete entries */
447 if (entry->flags & XFS_ATTR_LOCAL) {
448 xfs_attr_leaf_name_local_t *name_loc;
450 name_loc = xfs_attr3_leaf_name_local(leaf, i);
451 name = name_loc->nameval;
452 namelen = name_loc->namelen;
453 valuelen = be16_to_cpu(name_loc->valuelen);
455 xfs_attr_leaf_name_remote_t *name_rmt;
457 name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
458 name = name_rmt->name;
459 namelen = name_rmt->namelen;
460 valuelen = be32_to_cpu(name_rmt->valuelen);
463 context->put_listent(context, entry->flags,
464 name, namelen, valuelen);
465 if (context->seen_enough)
469 trace_xfs_attr_list_leaf_end(context);
474 * Copy out attribute entries for attr_list(), for leaf attribute lists.
477 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
482 trace_xfs_attr_leaf_list(context);
484 context->cursor->blkno = 0;
485 error = xfs_attr3_leaf_read(context->tp, context->dp, 0, -1, &bp);
489 xfs_attr3_leaf_list_int(bp, context);
490 xfs_trans_brelse(context->tp, bp);
495 xfs_attr_list_int_ilocked(
496 struct xfs_attr_list_context *context)
498 struct xfs_inode *dp = context->dp;
500 ASSERT(xfs_isilocked(dp, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
503 * Decide on what work routines to call based on the inode size.
505 if (!xfs_inode_hasattr(dp))
507 else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
508 return xfs_attr_shortform_list(context);
509 else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
510 return xfs_attr_leaf_list(context);
511 return xfs_attr_node_list(context);
516 xfs_attr_list_context_t *context)
519 xfs_inode_t *dp = context->dp;
522 XFS_STATS_INC(dp->i_mount, xs_attr_list);
524 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
527 lock_mode = xfs_ilock_attr_map_shared(dp);
528 error = xfs_attr_list_int_ilocked(context);
529 xfs_iunlock(dp, lock_mode);
533 #define ATTR_ENTBASESIZE /* minimum bytes used by an attr */ \
534 (((struct attrlist_ent *) 0)->a_name - (char *) 0)
535 #define ATTR_ENTSIZE(namelen) /* actual bytes used by an attr */ \
536 ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(uint32_t)-1) \
537 & ~(sizeof(uint32_t)-1))
540 * Format an attribute and copy it out to the user's buffer.
541 * Take care to check values and protect against them changing later,
542 * we may be reading them directly out of a user buffer.
545 xfs_attr_put_listent(
546 xfs_attr_list_context_t *context,
552 struct attrlist *alist = (struct attrlist *)context->alist;
556 ASSERT(!context->seen_enough);
557 ASSERT(!(context->flags & ATTR_KERNOVAL));
558 ASSERT(context->count >= 0);
559 ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
560 ASSERT(context->firstu >= sizeof(*alist));
561 ASSERT(context->firstu <= context->bufsize);
564 * Only list entries in the right namespace.
566 if (((context->flags & ATTR_SECURE) == 0) !=
567 ((flags & XFS_ATTR_SECURE) == 0))
569 if (((context->flags & ATTR_ROOT) == 0) !=
570 ((flags & XFS_ATTR_ROOT) == 0))
573 arraytop = sizeof(*alist) +
574 context->count * sizeof(alist->al_offset[0]);
575 context->firstu -= ATTR_ENTSIZE(namelen);
576 if (context->firstu < arraytop) {
577 trace_xfs_attr_list_full(context);
579 context->seen_enough = 1;
583 aep = (attrlist_ent_t *)&context->alist[context->firstu];
584 aep->a_valuelen = valuelen;
585 memcpy(aep->a_name, name, namelen);
586 aep->a_name[namelen] = 0;
587 alist->al_offset[context->count++] = context->firstu;
588 alist->al_count = context->count;
589 trace_xfs_attr_list_add(context);
594 * Generate a list of extended attribute names and optionally
595 * also value lengths. Positive return value follows the XFS
596 * convention of being an error, zero or negative return code
597 * is the length of the buffer returned (negated), indicating
606 attrlist_cursor_kern_t *cursor)
608 xfs_attr_list_context_t context;
609 struct attrlist *alist;
613 * Validate the cursor.
615 if (cursor->pad1 || cursor->pad2)
617 if ((cursor->initted == 0) &&
618 (cursor->hashval || cursor->blkno || cursor->offset))
621 /* Only internal consumers can retrieve incomplete attrs. */
622 if (flags & ATTR_INCOMPLETE)
626 * Check for a properly aligned buffer.
628 if (((long)buffer) & (sizeof(int)-1))
630 if (flags & ATTR_KERNOVAL)
634 * Initialize the output buffer.
636 memset(&context, 0, sizeof(context));
638 context.cursor = cursor;
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;
646 alist = (struct attrlist *)context.alist;
649 alist->al_offset[0] = context.bufsize;
651 error = xfs_attr_list_int(&context);