1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2017-2023 Oracle. All Rights Reserved.
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_log_format.h"
13 #include "xfs_trans.h"
14 #include "xfs_inode.h"
15 #include "xfs_da_format.h"
16 #include "xfs_da_btree.h"
18 #include "xfs_attr_leaf.h"
19 #include "xfs_attr_sf.h"
20 #include "xfs_parent.h"
21 #include "scrub/scrub.h"
22 #include "scrub/common.h"
23 #include "scrub/dabtree.h"
24 #include "scrub/attr.h"
25 #include "scrub/listxattr.h"
26 #include "scrub/repair.h"
28 /* Free the buffers linked from the xattr buffer. */
30 xchk_xattr_buf_cleanup(
33 struct xchk_xattr_buf *ab = priv;
47 * Allocate the free space bitmap if we're trying harder; there are leaf blocks
48 * in the attr fork; or we can't tell if there are leaf blocks.
51 xchk_xattr_want_freemap(
54 struct xfs_ifork *ifp;
56 if (sc->flags & XCHK_TRY_HARDER)
62 ifp = xfs_ifork_ptr(sc->ip, XFS_ATTR_FORK);
66 return xfs_ifork_has_extents(ifp);
70 * Allocate enough memory to hold an attr value and attr block bitmaps,
71 * reallocating the buffer if necessary. Buffer contents are not preserved
72 * across a reallocation.
80 struct xchk_xattr_buf *ab = sc->buf;
83 bmp_sz = sizeof(long) * BITS_TO_LONGS(sc->mp->m_attr_geo->blksize);
88 ab = kvzalloc(sizeof(struct xchk_xattr_buf), XCHK_GFP_FLAGS);
92 sc->buf_cleanup = xchk_xattr_buf_cleanup;
94 ab->usedmap = kvmalloc(bmp_sz, XCHK_GFP_FLAGS);
98 if (xchk_xattr_want_freemap(sc)) {
99 ab->freemap = kvmalloc(bmp_sz, XCHK_GFP_FLAGS);
104 if (xchk_could_repair(sc)) {
105 ab->name = kvmalloc(XATTR_NAME_MAX + 1, XCHK_GFP_FLAGS);
111 if (ab->value_sz >= value_size)
120 new_val = kvmalloc(value_size, XCHK_GFP_FLAGS);
125 ab->value_sz = value_size;
129 /* Set us up to scrub an inode's extended attributes. */
132 struct xfs_scrub *sc)
136 if (xchk_could_repair(sc)) {
137 error = xrep_setup_xattr(sc);
143 * We failed to get memory while checking attrs, so this time try to
144 * get all the memory we're ever going to need. Allocate the buffer
145 * without the inode lock held, which means we can sleep.
147 if (sc->flags & XCHK_TRY_HARDER) {
148 error = xchk_setup_xattr_buf(sc, XATTR_SIZE_MAX);
153 return xchk_setup_inode_contents(sc, 0);
156 /* Extended Attributes */
159 * Check that an extended attribute key can be looked up by hash.
161 * We use the extended attribute walk helper to call this function for every
162 * attribute key in an inode. Once we're here, we load the attribute value to
163 * see if any errors happen, or if we get more or less data than we expected.
167 struct xfs_scrub *sc,
168 struct xfs_inode *ip,
169 unsigned int attr_flags,
170 const unsigned char *name,
171 unsigned int namelen,
173 unsigned int valuelen,
176 struct xfs_da_args args = {
177 .attr_filter = attr_flags & XFS_ATTR_NSP_ONDISK_MASK,
178 .geo = sc->mp->m_attr_geo,
179 .whichfork = XFS_ATTR_FORK,
184 .valuelen = valuelen,
187 struct xchk_xattr_buf *ab;
192 if (xchk_should_terminate(sc, &error))
195 if (attr_flags & ~XFS_ATTR_ONDISK_MASK) {
196 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, args.blkno);
200 if (attr_flags & XFS_ATTR_INCOMPLETE) {
201 /* Incomplete attr key, just mark the inode for preening. */
202 xchk_ino_set_preen(sc, ip->i_ino);
206 /* Does this name make sense? */
207 if (!xfs_attr_namecheck(attr_flags, name, namelen)) {
208 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, args.blkno);
212 /* Check parent pointer record. */
213 if ((attr_flags & XFS_ATTR_PARENT) &&
214 !xfs_parent_valuecheck(sc->mp, value, valuelen)) {
215 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, args.blkno);
220 * Try to allocate enough memory to extract the attr value. If that
221 * doesn't work, return -EDEADLOCK as a signal to try again with a
222 * maximally sized buffer.
224 error = xchk_setup_xattr_buf(sc, valuelen);
225 if (error == -ENOMEM)
231 * Parent pointers are matched on attr name and value, so we must
232 * supply the xfs_parent_rec here when confirming that the dabtree
233 * indexing works correctly.
235 if (attr_flags & XFS_ATTR_PARENT)
236 memcpy(ab->value, value, valuelen);
238 args.value = ab->value;
241 * Get the attr value to ensure that lookup can find this attribute
242 * through the dabtree indexing and that remote value retrieval also
245 xfs_attr_sethash(&args);
246 error = xfs_attr_get_ilocked(&args);
247 /* ENODATA means the hash lookup failed and the attr is bad */
248 if (error == -ENODATA)
249 error = -EFSCORRUPTED;
250 if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, args.blkno,
253 if (args.valuelen != valuelen)
254 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, args.blkno);
260 * Mark a range [start, start+len) in this map. Returns true if the
261 * region was free, and false if there's a conflict or a problem.
263 * Within a char, the lowest bit of the char represents the byte with
264 * the smallest address
268 struct xfs_scrub *sc,
273 unsigned int mapsize = sc->mp->m_attr_geo->blksize;
276 if (start >= mapsize)
278 if (start + len > mapsize) {
279 len = mapsize - start;
283 if (find_next_bit(map, mapsize, start) < start + len)
285 bitmap_set(map, start, len);
291 * Check the leaf freemap from the usage bitmap. Returns false if the
292 * attr freemap has problems or points to used space.
295 xchk_xattr_check_freemap(
296 struct xfs_scrub *sc,
297 struct xfs_attr3_icleaf_hdr *leafhdr)
299 struct xchk_xattr_buf *ab = sc->buf;
300 unsigned int mapsize = sc->mp->m_attr_geo->blksize;
303 /* Construct bitmap of freemap contents. */
304 bitmap_zero(ab->freemap, mapsize);
305 for (i = 0; i < XFS_ATTR_LEAF_MAPSIZE; i++) {
306 if (!xchk_xattr_set_map(sc, ab->freemap,
307 leafhdr->freemap[i].base,
308 leafhdr->freemap[i].size))
312 /* Look for bits that are set in freemap and are marked in use. */
313 return !bitmap_intersects(ab->freemap, ab->usedmap, mapsize);
317 * Check this leaf entry's relations to everything else.
318 * Returns the number of bytes used for the name/value data.
322 struct xchk_da_btree *ds,
325 struct xfs_attr_leafblock *leaf,
326 struct xfs_attr3_icleaf_hdr *leafhdr,
327 struct xfs_attr_leaf_entry *ent,
329 unsigned int *usedbytes,
332 struct xfs_mount *mp = ds->state->mp;
333 struct xchk_xattr_buf *ab = ds->sc->buf;
335 struct xfs_attr_leaf_name_local *lentry;
336 struct xfs_attr_leaf_name_remote *rentry;
337 unsigned int nameidx;
338 unsigned int namesize;
341 xchk_da_set_corrupt(ds, level);
343 /* Hash values in order? */
344 if (be32_to_cpu(ent->hashval) < *last_hashval)
345 xchk_da_set_corrupt(ds, level);
346 *last_hashval = be32_to_cpu(ent->hashval);
348 nameidx = be16_to_cpu(ent->nameidx);
349 if (nameidx < leafhdr->firstused ||
350 nameidx >= mp->m_attr_geo->blksize) {
351 xchk_da_set_corrupt(ds, level);
355 /* Check the name information. */
356 if (ent->flags & XFS_ATTR_LOCAL) {
357 lentry = xfs_attr3_leaf_name_local(leaf, idx);
358 namesize = xfs_attr_leaf_entsize_local(lentry->namelen,
359 be16_to_cpu(lentry->valuelen));
360 name_end = (char *)lentry + namesize;
361 if (lentry->namelen == 0)
362 xchk_da_set_corrupt(ds, level);
364 rentry = xfs_attr3_leaf_name_remote(leaf, idx);
365 namesize = xfs_attr_leaf_entsize_remote(rentry->namelen);
366 name_end = (char *)rentry + namesize;
367 if (rentry->namelen == 0 || rentry->valueblk == 0)
368 xchk_da_set_corrupt(ds, level);
370 if (name_end > buf_end)
371 xchk_da_set_corrupt(ds, level);
373 if (!xchk_xattr_set_map(ds->sc, ab->usedmap, nameidx, namesize))
374 xchk_da_set_corrupt(ds, level);
375 if (!(ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
376 *usedbytes += namesize;
379 /* Scrub an attribute leaf. */
382 struct xchk_da_btree *ds,
385 struct xfs_attr3_icleaf_hdr leafhdr;
386 struct xfs_mount *mp = ds->state->mp;
387 struct xfs_da_state_blk *blk = &ds->state->path.blk[level];
388 struct xfs_buf *bp = blk->bp;
389 xfs_dablk_t *last_checked = ds->private;
390 struct xfs_attr_leafblock *leaf = bp->b_addr;
391 struct xfs_attr_leaf_entry *ent;
392 struct xfs_attr_leaf_entry *entries;
393 struct xchk_xattr_buf *ab = ds->sc->buf;
396 __u32 last_hashval = 0;
397 unsigned int usedbytes = 0;
398 unsigned int hdrsize;
401 if (*last_checked == blk->blkno)
404 *last_checked = blk->blkno;
405 bitmap_zero(ab->usedmap, mp->m_attr_geo->blksize);
407 /* Check all the padding. */
408 if (xfs_has_crc(ds->sc->mp)) {
409 struct xfs_attr3_leafblock *leaf3 = bp->b_addr;
411 if (leaf3->hdr.pad1 != 0 || leaf3->hdr.pad2 != 0 ||
412 leaf3->hdr.info.hdr.pad != 0)
413 xchk_da_set_corrupt(ds, level);
415 if (leaf->hdr.pad1 != 0 || leaf->hdr.info.pad != 0)
416 xchk_da_set_corrupt(ds, level);
419 /* Check the leaf header */
420 xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &leafhdr, leaf);
421 hdrsize = xfs_attr3_leaf_hdr_size(leaf);
424 * Empty xattr leaf blocks mapped at block 0 are probably a byproduct
425 * of a race between setxattr and a log shutdown. Anywhere else in the
426 * attr fork is a corruption.
428 if (leafhdr.count == 0) {
430 xchk_da_set_preen(ds, level);
432 xchk_da_set_corrupt(ds, level);
434 if (leafhdr.usedbytes > mp->m_attr_geo->blksize)
435 xchk_da_set_corrupt(ds, level);
436 if (leafhdr.firstused > mp->m_attr_geo->blksize)
437 xchk_da_set_corrupt(ds, level);
438 if (leafhdr.firstused < hdrsize)
439 xchk_da_set_corrupt(ds, level);
440 if (!xchk_xattr_set_map(ds->sc, ab->usedmap, 0, hdrsize))
441 xchk_da_set_corrupt(ds, level);
443 xchk_da_set_preen(ds, level);
445 if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
448 entries = xfs_attr3_leaf_entryp(leaf);
449 if ((char *)&entries[leafhdr.count] > (char *)leaf + leafhdr.firstused)
450 xchk_da_set_corrupt(ds, level);
452 buf_end = (char *)bp->b_addr + mp->m_attr_geo->blksize;
453 for (i = 0, ent = entries; i < leafhdr.count; ent++, i++) {
454 /* Mark the leaf entry itself. */
455 off = (char *)ent - (char *)leaf;
456 if (!xchk_xattr_set_map(ds->sc, ab->usedmap, off,
457 sizeof(xfs_attr_leaf_entry_t))) {
458 xchk_da_set_corrupt(ds, level);
462 /* Check the entry and nameval. */
463 xchk_xattr_entry(ds, level, buf_end, leaf, &leafhdr,
464 ent, i, &usedbytes, &last_hashval);
466 if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
470 if (!xchk_xattr_check_freemap(ds->sc, &leafhdr))
471 xchk_da_set_corrupt(ds, level);
473 if (leafhdr.usedbytes != usedbytes)
474 xchk_da_set_corrupt(ds, level);
480 /* Scrub a attribute btree record. */
483 struct xchk_da_btree *ds,
486 struct xfs_mount *mp = ds->state->mp;
487 struct xfs_da_state_blk *blk = &ds->state->path.blk[level];
488 struct xfs_attr_leaf_name_local *lentry;
489 struct xfs_attr_leaf_name_remote *rentry;
491 struct xfs_attr_leaf_entry *ent;
492 xfs_dahash_t calc_hash;
498 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
500 ent = xfs_attr3_leaf_entryp(blk->bp->b_addr) + blk->index;
502 /* Check the whole block, if necessary. */
503 error = xchk_xattr_block(ds, level);
506 if (ds->sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
509 /* Check the hash of the entry. */
510 error = xchk_da_btree_hash(ds, level, &ent->hashval);
514 /* Find the attr entry's location. */
516 hdrsize = xfs_attr3_leaf_hdr_size(bp->b_addr);
517 nameidx = be16_to_cpu(ent->nameidx);
518 if (nameidx < hdrsize || nameidx >= mp->m_attr_geo->blksize) {
519 xchk_da_set_corrupt(ds, level);
523 /* Retrieve the entry and check it. */
524 hash = be32_to_cpu(ent->hashval);
525 if (ent->flags & ~XFS_ATTR_ONDISK_MASK) {
526 xchk_da_set_corrupt(ds, level);
529 if (!xfs_attr_check_namespace(ent->flags)) {
530 xchk_da_set_corrupt(ds, level);
534 if (ent->flags & XFS_ATTR_LOCAL) {
535 lentry = (struct xfs_attr_leaf_name_local *)
536 (((char *)bp->b_addr) + nameidx);
537 if (lentry->namelen <= 0) {
538 xchk_da_set_corrupt(ds, level);
541 calc_hash = xfs_attr_hashval(mp, ent->flags, lentry->nameval,
543 lentry->nameval + lentry->namelen,
544 be16_to_cpu(lentry->valuelen));
546 rentry = (struct xfs_attr_leaf_name_remote *)
547 (((char *)bp->b_addr) + nameidx);
548 if (rentry->namelen <= 0) {
549 xchk_da_set_corrupt(ds, level);
552 if (ent->flags & XFS_ATTR_PARENT) {
553 xchk_da_set_corrupt(ds, level);
556 calc_hash = xfs_attr_hashval(mp, ent->flags, rentry->name,
557 rentry->namelen, NULL,
558 be32_to_cpu(rentry->valuelen));
560 if (calc_hash != hash)
561 xchk_da_set_corrupt(ds, level);
567 /* Check space usage of shortform attrs. */
570 struct xfs_scrub *sc)
572 struct xchk_xattr_buf *ab = sc->buf;
573 struct xfs_ifork *ifp = &sc->ip->i_af;
574 struct xfs_attr_sf_hdr *sf = ifp->if_data;
575 struct xfs_attr_sf_entry *sfe = xfs_attr_sf_firstentry(sf);
576 struct xfs_attr_sf_entry *next;
577 unsigned char *end = ifp->if_data + ifp->if_bytes;
581 bitmap_zero(ab->usedmap, ifp->if_bytes);
582 xchk_xattr_set_map(sc, ab->usedmap, 0, sizeof(*sf));
584 if ((unsigned char *)sfe > end) {
585 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
589 for (i = 0; i < sf->count; i++) {
590 unsigned char *name = sfe->nameval;
591 unsigned char *value = &sfe->nameval[sfe->namelen];
593 if (xchk_should_terminate(sc, &error))
596 next = xfs_attr_sf_nextentry(sfe);
597 if ((unsigned char *)next > end) {
598 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
603 * Shortform entries do not set LOCAL or INCOMPLETE, so the
604 * only valid flag bits here are for namespaces.
606 if (sfe->flags & ~XFS_ATTR_NSP_ONDISK_MASK) {
607 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
611 if (!xchk_xattr_set_map(sc, ab->usedmap,
612 (char *)sfe - (char *)sf,
613 sizeof(struct xfs_attr_sf_entry))) {
614 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
618 if (!xchk_xattr_set_map(sc, ab->usedmap,
619 (char *)name - (char *)sf,
621 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
625 if (!xchk_xattr_set_map(sc, ab->usedmap,
626 (char *)value - (char *)sf,
628 xchk_fblock_set_corrupt(sc, XFS_ATTR_FORK, 0);
638 /* Scrub the extended attribute metadata. */
641 struct xfs_scrub *sc)
643 xfs_dablk_t last_checked = -1U;
646 if (!xfs_inode_hasattr(sc->ip))
649 /* Allocate memory for xattr checking. */
650 error = xchk_setup_xattr_buf(sc, 0);
651 if (error == -ENOMEM)
656 /* Check the physical structure of the xattr. */
657 if (sc->ip->i_af.if_format == XFS_DINODE_FMT_LOCAL)
658 error = xchk_xattr_check_sf(sc);
660 error = xchk_da_btree(sc, XFS_ATTR_FORK, xchk_xattr_rec,
665 if (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT)
669 * Look up every xattr in this file by name and hash.
671 * The VFS only locks i_rwsem when modifying attrs, so keep all
672 * three locks held because that's the only way to ensure we're
673 * the only thread poking into the da btree. We traverse the da
674 * btree while holding a leaf buffer locked for the xattr name
675 * iteration, which doesn't really follow the usual buffer
678 error = xchk_xattr_walk(sc, sc->ip, xchk_xattr_actor, NULL, NULL);
679 if (!xchk_fblock_process_error(sc, XFS_ATTR_FORK, 0, &error))