1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) 2017-2023 Oracle. All Rights Reserved.
4 * Author: Darrick J. Wong <djwong@kernel.org>
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_quota.h"
17 #include "xfs_scrub.h"
18 #include "xfs_buf_mem.h"
20 #include "xfs_exchrange.h"
21 #include "xfs_exchmaps.h"
23 #include "xfs_parent.h"
24 #include "xfs_icache.h"
25 #include "scrub/scrub.h"
26 #include "scrub/common.h"
27 #include "scrub/trace.h"
28 #include "scrub/repair.h"
29 #include "scrub/health.h"
30 #include "scrub/stats.h"
31 #include "scrub/xfile.h"
32 #include "scrub/tempfile.h"
33 #include "scrub/orphanage.h"
36 * Online Scrub and Repair
38 * Traditionally, XFS (the kernel driver) did not know how to check or
39 * repair on-disk data structures. That task was left to the xfs_check
40 * and xfs_repair tools, both of which require taking the filesystem
41 * offline for a thorough but time consuming examination. Online
42 * scrub & repair, on the other hand, enables us to check the metadata
43 * for obvious errors while carefully stepping around the filesystem's
44 * ongoing operations, locking rules, etc.
46 * Given that most XFS metadata consist of records stored in a btree,
47 * most of the checking functions iterate the btree blocks themselves
48 * looking for irregularities. When a record block is encountered, each
49 * record can be checked for obviously bad values. Record values can
50 * also be cross-referenced against other btrees to look for potential
51 * misunderstandings between pieces of metadata.
53 * It is expected that the checkers responsible for per-AG metadata
54 * structures will lock the AG headers (AGI, AGF, AGFL), iterate the
55 * metadata structure, and perform any relevant cross-referencing before
56 * unlocking the AG and returning the results to userspace. These
57 * scrubbers must not keep an AG locked for too long to avoid tying up
58 * the block and inode allocators.
60 * Block maps and b-trees rooted in an inode present a special challenge
61 * because they can involve extents from any AG. The general scrubber
62 * structure of lock -> check -> xref -> unlock still holds, but AG
63 * locking order rules /must/ be obeyed to avoid deadlocks. The
64 * ordering rule, of course, is that we must lock in increasing AG
65 * order. Helper functions are provided to track which AG headers we've
66 * already locked. If we detect an imminent locking order violation, we
67 * can signal a potential deadlock, in which case the scrubber can jump
68 * out to the top level, lock all the AGs in order, and retry the scrub.
70 * For file data (directories, extended attributes, symlinks) scrub, we
71 * can simply lock the inode and walk the data. For btree data
72 * (directories and attributes) we follow the same btree-scrubbing
73 * strategy outlined previously to check the records.
75 * We use a bit of trickery with transactions to avoid buffer deadlocks
76 * if there is a cycle in the metadata. The basic problem is that
77 * travelling down a btree involves locking the current buffer at each
78 * tree level. If a pointer should somehow point back to a buffer that
79 * we've already examined, we will deadlock due to the second buffer
80 * locking attempt. Note however that grabbing a buffer in transaction
81 * context links the locked buffer to the transaction. If we try to
82 * re-grab the buffer in the context of the same transaction, we avoid
83 * the second lock attempt and continue. Between the verifier and the
84 * scrubber, something will notice that something is amiss and report
85 * the corruption. Therefore, each scrubber will allocate an empty
86 * transaction, attach buffers to it, and cancel the transaction at the
87 * end of the scrub run. Cancelling a non-dirty transaction simply
88 * unlocks the buffers.
90 * There are four pieces of data that scrub can communicate to
91 * userspace. The first is the error code (errno), which can be used to
92 * communicate operational errors in performing the scrub. There are
93 * also three flags that can be set in the scrub context. If the data
94 * structure itself is corrupt, the CORRUPT flag will be set. If
95 * the metadata is correct but otherwise suboptimal, the PREEN flag
98 * We perform secondary validation of filesystem metadata by
99 * cross-referencing every record with all other available metadata.
100 * For example, for block mapping extents, we verify that there are no
101 * records in the free space and inode btrees corresponding to that
102 * space extent and that there is a corresponding entry in the reverse
103 * mapping btree. Inconsistent metadata is noted by setting the
104 * XCORRUPT flag; btree query function errors are noted by setting the
105 * XFAIL flag and deleting the cursor to prevent further attempts to
106 * cross-reference with a defective btree.
108 * If a piece of metadata proves corrupt or suboptimal, the userspace
109 * program can ask the kernel to apply some tender loving care (TLC) to
110 * the metadata object by setting the REPAIR flag and re-calling the
111 * scrub ioctl. "Corruption" is defined by metadata violating the
112 * on-disk specification; operations cannot continue if the violation is
113 * left untreated. It is possible for XFS to continue if an object is
114 * "suboptimal", however performance may be degraded. Repairs are
115 * usually performed by rebuilding the metadata entirely out of
116 * redundant metadata. Optimizing, on the other hand, can sometimes be
117 * done without rebuilding entire structures.
119 * Generally speaking, the repair code has the following code structure:
120 * Lock -> scrub -> repair -> commit -> re-lock -> re-scrub -> unlock.
121 * The first check helps us figure out if we need to rebuild or simply
122 * optimize the structure so that the rebuild knows what to do. The
123 * second check evaluates the completeness of the repair; that is what
124 * is reported to userspace.
126 * A quick note on symbol prefixes:
127 * - "xfs_" are general XFS symbols.
128 * - "xchk_" are symbols related to metadata checking.
129 * - "xrep_" are symbols related to metadata repair.
130 * - "xfs_scrub_" are symbols that tie online fsck to the rest of XFS.
134 * Scrub probe -- userspace uses this to probe if we're willing to scrub
135 * or repair a given mountpoint. This will be used by xfs_scrub to
136 * probe the kernel's abilities to scrub (and repair) the metadata. We
137 * do this by validating the ioctl inputs from userspace, preparing the
138 * filesystem for a scrub (or a repair) operation, and immediately
139 * returning to userspace. Userspace can use the returned errno and
140 * structure state to decide (in broad terms) if scrub/repair are
141 * supported by the running kernel.
145 struct xfs_scrub *sc)
149 if (xchk_should_terminate(sc, &error))
155 /* Scrub setup and teardown */
158 xchk_fsgates_disable(
159 struct xfs_scrub *sc)
161 if (!(sc->flags & XCHK_FSGATES_ALL))
164 trace_xchk_fsgates_disable(sc, sc->flags & XCHK_FSGATES_ALL);
166 if (sc->flags & XCHK_FSGATES_DRAIN)
167 xfs_drain_wait_disable();
169 if (sc->flags & XCHK_FSGATES_QUOTA)
170 xfs_dqtrx_hook_disable();
172 if (sc->flags & XCHK_FSGATES_DIRENTS)
173 xfs_dir_hook_disable();
175 if (sc->flags & XCHK_FSGATES_RMAP)
176 xfs_rmap_hook_disable();
178 sc->flags &= ~XCHK_FSGATES_ALL;
181 /* Free the resources associated with a scrub subtype. */
183 xchk_scrub_free_subord(
184 struct xfs_scrub_subord *sub)
186 struct xfs_scrub *sc = sub->parent_sc;
188 ASSERT(sc->ip == sub->sc.ip);
189 ASSERT(sc->orphanage == sub->sc.orphanage);
190 ASSERT(sc->tempip == sub->sc.tempip);
192 sc->sm->sm_type = sub->old_smtype;
193 sc->sm->sm_flags = sub->old_smflags |
194 (sc->sm->sm_flags & XFS_SCRUB_FLAGS_OUT);
198 if (sub->sc.buf_cleanup)
199 sub->sc.buf_cleanup(sub->sc.buf);
203 xmbuf_free(sub->sc.xmbtp);
205 xfile_destroy(sub->sc.xfile);
207 sc->ilock_flags = sub->sc.ilock_flags;
208 sc->orphanage_ilock_flags = sub->sc.orphanage_ilock_flags;
209 sc->temp_ilock_flags = sub->sc.temp_ilock_flags;
214 /* Free all the resources and finish the transactions. */
217 struct xfs_scrub *sc,
220 xchk_ag_free(sc, &sc->sa);
222 if (error == 0 && (sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
223 error = xfs_trans_commit(sc->tp);
225 xfs_trans_cancel(sc->tp);
230 xchk_iunlock(sc, sc->ilock_flags);
231 xchk_irele(sc, sc->ip);
234 if (sc->flags & XCHK_HAVE_FREEZE_PROT) {
235 sc->flags &= ~XCHK_HAVE_FREEZE_PROT;
236 mnt_drop_write_file(sc->file);
239 xmbuf_free(sc->xmbtp);
243 xfile_destroy(sc->xfile);
248 sc->buf_cleanup(sc->buf);
250 sc->buf_cleanup = NULL;
254 xrep_tempfile_rele(sc);
255 xrep_orphanage_rele(sc);
256 xchk_fsgates_disable(sc);
260 /* Scrubbing dispatch. */
262 static const struct xchk_meta_ops meta_scrub_ops[] = {
263 [XFS_SCRUB_TYPE_PROBE] = { /* ioctl presence test */
265 .setup = xchk_setup_fs,
267 .repair = xrep_probe,
269 [XFS_SCRUB_TYPE_SB] = { /* superblock */
271 .setup = xchk_setup_agheader,
272 .scrub = xchk_superblock,
273 .repair = xrep_superblock,
275 [XFS_SCRUB_TYPE_AGF] = { /* agf */
277 .setup = xchk_setup_agheader,
281 [XFS_SCRUB_TYPE_AGFL]= { /* agfl */
283 .setup = xchk_setup_agheader,
287 [XFS_SCRUB_TYPE_AGI] = { /* agi */
289 .setup = xchk_setup_agheader,
293 [XFS_SCRUB_TYPE_BNOBT] = { /* bnobt */
295 .setup = xchk_setup_ag_allocbt,
296 .scrub = xchk_allocbt,
297 .repair = xrep_allocbt,
298 .repair_eval = xrep_revalidate_allocbt,
300 [XFS_SCRUB_TYPE_CNTBT] = { /* cntbt */
302 .setup = xchk_setup_ag_allocbt,
303 .scrub = xchk_allocbt,
304 .repair = xrep_allocbt,
305 .repair_eval = xrep_revalidate_allocbt,
307 [XFS_SCRUB_TYPE_INOBT] = { /* inobt */
309 .setup = xchk_setup_ag_iallocbt,
310 .scrub = xchk_iallocbt,
311 .repair = xrep_iallocbt,
312 .repair_eval = xrep_revalidate_iallocbt,
314 [XFS_SCRUB_TYPE_FINOBT] = { /* finobt */
316 .setup = xchk_setup_ag_iallocbt,
317 .scrub = xchk_iallocbt,
318 .has = xfs_has_finobt,
319 .repair = xrep_iallocbt,
320 .repair_eval = xrep_revalidate_iallocbt,
322 [XFS_SCRUB_TYPE_RMAPBT] = { /* rmapbt */
324 .setup = xchk_setup_ag_rmapbt,
325 .scrub = xchk_rmapbt,
326 .has = xfs_has_rmapbt,
327 .repair = xrep_rmapbt,
329 [XFS_SCRUB_TYPE_REFCNTBT] = { /* refcountbt */
331 .setup = xchk_setup_ag_refcountbt,
332 .scrub = xchk_refcountbt,
333 .has = xfs_has_reflink,
334 .repair = xrep_refcountbt,
336 [XFS_SCRUB_TYPE_INODE] = { /* inode record */
338 .setup = xchk_setup_inode,
340 .repair = xrep_inode,
342 [XFS_SCRUB_TYPE_BMBTD] = { /* inode data fork */
344 .setup = xchk_setup_inode_bmap,
345 .scrub = xchk_bmap_data,
346 .repair = xrep_bmap_data,
348 [XFS_SCRUB_TYPE_BMBTA] = { /* inode attr fork */
350 .setup = xchk_setup_inode_bmap,
351 .scrub = xchk_bmap_attr,
352 .repair = xrep_bmap_attr,
354 [XFS_SCRUB_TYPE_BMBTC] = { /* inode CoW fork */
356 .setup = xchk_setup_inode_bmap,
357 .scrub = xchk_bmap_cow,
358 .repair = xrep_bmap_cow,
360 [XFS_SCRUB_TYPE_DIR] = { /* directory */
362 .setup = xchk_setup_directory,
363 .scrub = xchk_directory,
364 .repair = xrep_directory,
366 [XFS_SCRUB_TYPE_XATTR] = { /* extended attributes */
368 .setup = xchk_setup_xattr,
370 .repair = xrep_xattr,
372 [XFS_SCRUB_TYPE_SYMLINK] = { /* symbolic link */
374 .setup = xchk_setup_symlink,
375 .scrub = xchk_symlink,
376 .repair = xrep_symlink,
378 [XFS_SCRUB_TYPE_PARENT] = { /* parent pointers */
380 .setup = xchk_setup_parent,
381 .scrub = xchk_parent,
382 .repair = xrep_parent,
384 [XFS_SCRUB_TYPE_RTBITMAP] = { /* realtime bitmap */
386 .setup = xchk_setup_rtbitmap,
387 .scrub = xchk_rtbitmap,
388 .repair = xrep_rtbitmap,
390 [XFS_SCRUB_TYPE_RTSUM] = { /* realtime summary */
392 .setup = xchk_setup_rtsummary,
393 .scrub = xchk_rtsummary,
394 .repair = xrep_rtsummary,
396 [XFS_SCRUB_TYPE_UQUOTA] = { /* user quota */
398 .setup = xchk_setup_quota,
400 .repair = xrep_quota,
402 [XFS_SCRUB_TYPE_GQUOTA] = { /* group quota */
404 .setup = xchk_setup_quota,
406 .repair = xrep_quota,
408 [XFS_SCRUB_TYPE_PQUOTA] = { /* project quota */
410 .setup = xchk_setup_quota,
412 .repair = xrep_quota,
414 [XFS_SCRUB_TYPE_FSCOUNTERS] = { /* fs summary counters */
416 .setup = xchk_setup_fscounters,
417 .scrub = xchk_fscounters,
418 .repair = xrep_fscounters,
420 [XFS_SCRUB_TYPE_QUOTACHECK] = { /* quota counters */
422 .setup = xchk_setup_quotacheck,
423 .scrub = xchk_quotacheck,
424 .repair = xrep_quotacheck,
426 [XFS_SCRUB_TYPE_NLINKS] = { /* inode link counts */
428 .setup = xchk_setup_nlinks,
429 .scrub = xchk_nlinks,
430 .repair = xrep_nlinks,
432 [XFS_SCRUB_TYPE_HEALTHY] = { /* fs healthy; clean all reminders */
434 .setup = xchk_setup_fs,
435 .scrub = xchk_health_record,
436 .repair = xrep_notsupported,
438 [XFS_SCRUB_TYPE_DIRTREE] = { /* directory tree structure */
440 .setup = xchk_setup_dirtree,
441 .scrub = xchk_dirtree,
442 .has = xfs_has_parent,
443 .repair = xrep_dirtree,
448 xchk_validate_inputs(
449 struct xfs_mount *mp,
450 struct xfs_scrub_metadata *sm)
453 const struct xchk_meta_ops *ops;
456 /* Check our inputs. */
457 sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
458 if (sm->sm_flags & ~XFS_SCRUB_FLAGS_IN)
460 /* sm_reserved[] must be zero */
461 if (memchr_inv(sm->sm_reserved, 0, sizeof(sm->sm_reserved)))
465 /* Do we know about this type of metadata? */
466 if (sm->sm_type >= XFS_SCRUB_TYPE_NR)
468 ops = &meta_scrub_ops[sm->sm_type];
469 if (ops->setup == NULL || ops->scrub == NULL)
471 /* Does this fs even support this type of metadata? */
472 if (ops->has && !ops->has(mp))
476 /* restricting fields must be appropriate for type */
480 if (sm->sm_ino || sm->sm_gen || sm->sm_agno)
484 if (sm->sm_ino || sm->sm_gen ||
485 sm->sm_agno >= mp->m_sb.sb_agcount)
489 if (sm->sm_agno || (sm->sm_gen && !sm->sm_ino))
496 /* No rebuild without repair. */
497 if ((sm->sm_flags & XFS_SCRUB_IFLAG_FORCE_REBUILD) &&
498 !(sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR))
502 * We only want to repair read-write v5+ filesystems. Defer the check
503 * for ops->repair until after our scrub confirms that we need to
504 * perform repairs so that we avoid failing due to not supporting
505 * repairing an object that doesn't need repairs.
507 if (sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) {
509 if (!xfs_has_crc(mp))
513 if (xfs_is_readonly(mp))
522 #ifdef CONFIG_XFS_ONLINE_REPAIR
523 static inline void xchk_postmortem(struct xfs_scrub *sc)
526 * Userspace asked us to repair something, we repaired it, rescanned
527 * it, and the rescan says it's still broken. Scream about this in
530 if ((sc->sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) &&
531 (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
532 XFS_SCRUB_OFLAG_XCORRUPT)))
533 xrep_failure(sc->mp);
536 static inline void xchk_postmortem(struct xfs_scrub *sc)
539 * Userspace asked us to scrub something, it's broken, and we have no
540 * way of fixing it. Scream in the logs.
542 if (sc->sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
543 XFS_SCRUB_OFLAG_XCORRUPT))
544 xfs_alert_ratelimited(sc->mp,
545 "Corruption detected during scrub.");
547 #endif /* CONFIG_XFS_ONLINE_REPAIR */
550 * Create a new scrub context from an existing one, but with a different scrub
553 struct xfs_scrub_subord *
554 xchk_scrub_create_subord(
555 struct xfs_scrub *sc,
556 unsigned int subtype)
558 struct xfs_scrub_subord *sub;
560 sub = kzalloc(sizeof(*sub), XCHK_GFP_FLAGS);
562 return ERR_PTR(-ENOMEM);
564 sub->old_smtype = sc->sm->sm_type;
565 sub->old_smflags = sc->sm->sm_flags;
567 memcpy(&sub->sc, sc, sizeof(struct xfs_scrub));
568 sub->sc.ops = &meta_scrub_ops[subtype];
569 sub->sc.sm->sm_type = subtype;
570 sub->sc.sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
572 sub->sc.buf_cleanup = NULL;
573 sub->sc.xfile = NULL;
574 sub->sc.xmbtp = NULL;
579 /* Dispatch metadata scrubbing. */
583 struct xfs_scrub_metadata *sm)
585 struct xchk_stats_run run = { };
586 struct xfs_scrub *sc;
587 struct xfs_mount *mp = XFS_I(file_inode(file))->i_mount;
591 BUILD_BUG_ON(sizeof(meta_scrub_ops) !=
592 (sizeof(struct xchk_meta_ops) * XFS_SCRUB_TYPE_NR));
594 trace_xchk_start(XFS_I(file_inode(file)), sm, error);
596 /* Forbidden if we are shut down or mounted norecovery. */
598 if (xfs_is_shutdown(mp))
600 error = -ENOTRECOVERABLE;
601 if (xfs_has_norecovery(mp))
604 error = xchk_validate_inputs(mp, sm);
608 xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SCRUB,
609 "EXPERIMENTAL online scrub feature in use. Use at your own risk!");
611 sc = kzalloc(sizeof(struct xfs_scrub), XCHK_GFP_FLAGS);
620 sc->ops = &meta_scrub_ops[sm->sm_type];
621 sc->sick_mask = xchk_health_mask_for_scrub_type(sm->sm_type);
622 sc->relax = INIT_XCHK_RELAX;
625 * When repairs are allowed, prevent freezing or readonly remount while
626 * scrub is running with a real transaction.
628 if (sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR) {
629 error = mnt_want_write_file(sc->file);
633 sc->flags |= XCHK_HAVE_FREEZE_PROT;
636 /* Set up for the operation. */
637 error = sc->ops->setup(sc);
638 if (error == -EDEADLOCK && !(sc->flags & XCHK_TRY_HARDER))
640 if (error == -ECHRNG && !(sc->flags & XCHK_NEED_DRAIN))
645 /* Scrub for errors. */
646 check_start = xchk_stats_now();
647 if ((sc->flags & XREP_ALREADY_FIXED) && sc->ops->repair_eval != NULL)
648 error = sc->ops->repair_eval(sc);
650 error = sc->ops->scrub(sc);
651 run.scrub_ns += xchk_stats_elapsed_ns(check_start);
652 if (error == -EDEADLOCK && !(sc->flags & XCHK_TRY_HARDER))
654 if (error == -ECHRNG && !(sc->flags & XCHK_NEED_DRAIN))
656 if (error || (sm->sm_flags & XFS_SCRUB_OFLAG_INCOMPLETE))
659 xchk_update_health(sc);
661 if (xchk_could_repair(sc)) {
663 * If userspace asked for a repair but it wasn't necessary,
664 * report that back to userspace.
666 if (!xrep_will_attempt(sc)) {
667 sc->sm->sm_flags |= XFS_SCRUB_OFLAG_NO_REPAIR_NEEDED;
672 * If it's broken, userspace wants us to fix it, and we haven't
673 * already tried to fix it, then attempt a repair.
675 error = xrep_attempt(sc, &run);
676 if (error == -EAGAIN) {
678 * Either the repair function succeeded or it couldn't
679 * get all the resources it needs; either way, we go
680 * back to the beginning and call the scrub function.
682 error = xchk_teardown(sc, 0);
694 error = xchk_teardown(sc, error);
696 if (error != -ENOENT)
697 xchk_stats_merge(mp, sm, &run);
700 trace_xchk_done(XFS_I(file_inode(file)), sm, error);
701 if (error == -EFSCORRUPTED || error == -EFSBADCRC) {
702 sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
707 error = xchk_teardown(sc, 0);
710 sc->flags |= XCHK_NEED_DRAIN;
715 * Scrubbers return -EDEADLOCK to mean 'try harder'. Tear down
716 * everything we hold, then set up again with preparation for
717 * worst-case scenarios.
719 error = xchk_teardown(sc, 0);
722 sc->flags |= XCHK_TRY_HARDER;
727 /* Scrub one aspect of one piece of metadata. */
729 xfs_ioc_scrub_metadata(
733 struct xfs_scrub_metadata scrub;
736 if (!capable(CAP_SYS_ADMIN))
739 if (copy_from_user(&scrub, arg, sizeof(scrub)))
742 error = xfs_scrub_metadata(file, &scrub);
746 if (copy_to_user(arg, &scrub, sizeof(scrub)))
752 /* Decide if there have been any scrub failures up to this point. */
754 xfs_scrubv_check_barrier(
755 struct xfs_mount *mp,
756 const struct xfs_scrub_vec *vectors,
757 const struct xfs_scrub_vec *stop_vec)
759 const struct xfs_scrub_vec *v;
762 failmask = stop_vec->sv_flags & XFS_SCRUB_FLAGS_OUT;
764 for (v = vectors; v < stop_vec; v++) {
765 if (v->sv_type == XFS_SCRUB_TYPE_BARRIER)
769 * Runtime errors count as a previous failure, except the ones
770 * used to ask userspace to retry.
783 * If any of the out-flags on the scrub vector match the mask
784 * that was set on the barrier vector, that's a previous fail.
786 if (v->sv_flags & failmask)
794 * If the caller provided us with a nonzero inode number that isn't the ioctl
795 * file, try to grab a reference to it to eliminate all further untrusted inode
796 * lookups. If we can't get the inode, let each scrub function try again.
798 STATIC struct xfs_inode *
799 xchk_scrubv_open_by_handle(
800 struct xfs_mount *mp,
801 const struct xfs_scrub_vec_head *head)
803 struct xfs_trans *tp;
804 struct xfs_inode *ip;
807 error = xfs_trans_alloc_empty(mp, &tp);
811 error = xfs_iget(mp, tp, head->svh_ino, XCHK_IGET_FLAGS, 0, &ip);
812 xfs_trans_cancel(tp);
816 if (VFS_I(ip)->i_generation != head->svh_gen) {
824 /* Vectored scrub implementation to reduce ioctl calls. */
826 xfs_ioc_scrubv_metadata(
830 struct xfs_scrub_vec_head head;
831 struct xfs_scrub_vec_head __user *uhead = arg;
832 struct xfs_scrub_vec *vectors;
833 struct xfs_scrub_vec __user *uvectors;
834 struct xfs_inode *ip_in = XFS_I(file_inode(file));
835 struct xfs_mount *mp = ip_in->i_mount;
836 struct xfs_inode *handle_ip = NULL;
837 struct xfs_scrub_vec *v;
842 if (!capable(CAP_SYS_ADMIN))
845 if (copy_from_user(&head, uhead, sizeof(head)))
848 if (head.svh_reserved)
850 if (head.svh_flags & ~XFS_SCRUB_VEC_FLAGS_ALL)
852 if (head.svh_nr == 0)
855 vec_bytes = array_size(head.svh_nr, sizeof(struct xfs_scrub_vec));
856 if (vec_bytes > PAGE_SIZE)
859 uvectors = u64_to_user_ptr(head.svh_vectors);
860 vectors = memdup_user(uvectors, vec_bytes);
862 return PTR_ERR(vectors);
864 trace_xchk_scrubv_start(ip_in, &head);
866 for (i = 0, v = vectors; i < head.svh_nr; i++, v++) {
867 if (v->sv_reserved) {
872 if (v->sv_type == XFS_SCRUB_TYPE_BARRIER &&
873 (v->sv_flags & ~XFS_SCRUB_FLAGS_OUT)) {
878 trace_xchk_scrubv_item(mp, &head, i, v);
882 * If the caller wants us to do a scrub-by-handle and the file used to
883 * call the ioctl is not the same file, load the incore inode and pin
884 * it across all the scrubv actions to avoid repeated UNTRUSTED
885 * lookups. The reference is not passed to deeper layers of scrub
886 * because each scrubber gets to decide its own strategy and return
887 * values for getting an inode.
889 if (head.svh_ino && head.svh_ino != ip_in->i_ino)
890 handle_ip = xchk_scrubv_open_by_handle(mp, &head);
892 /* Run all the scrubbers. */
893 for (i = 0, v = vectors; i < head.svh_nr; i++, v++) {
894 struct xfs_scrub_metadata sm = {
895 .sm_type = v->sv_type,
896 .sm_flags = v->sv_flags,
897 .sm_ino = head.svh_ino,
898 .sm_gen = head.svh_gen,
899 .sm_agno = head.svh_agno,
902 if (v->sv_type == XFS_SCRUB_TYPE_BARRIER) {
903 v->sv_ret = xfs_scrubv_check_barrier(mp, vectors, v);
905 trace_xchk_scrubv_barrier_fail(mp, &head, i, v);
912 v->sv_ret = xfs_scrub_metadata(file, &sm);
913 v->sv_flags = sm.sm_flags;
915 trace_xchk_scrubv_outcome(mp, &head, i, v);
917 if (head.svh_rest_us) {
920 expires = ktime_add_ns(ktime_get(),
921 head.svh_rest_us * 1000);
922 set_current_state(TASK_KILLABLE);
923 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
926 if (fatal_signal_pending(current)) {
932 if (copy_to_user(uvectors, vectors, vec_bytes) ||
933 copy_to_user(uhead, &head, sizeof(head))) {
940 xfs_irele(handle_ip);