xfs_mount.c revision a1d86e8d
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_bit.h"
13#include "xfs_sb.h"
14#include "xfs_mount.h"
15#include "xfs_inode.h"
16#include "xfs_dir2.h"
17#include "xfs_ialloc.h"
18#include "xfs_alloc.h"
19#include "xfs_rtalloc.h"
20#include "xfs_bmap.h"
21#include "xfs_trans.h"
22#include "xfs_trans_priv.h"
23#include "xfs_log.h"
24#include "xfs_error.h"
25#include "xfs_quota.h"
26#include "xfs_fsops.h"
27#include "xfs_icache.h"
28#include "xfs_sysfs.h"
29#include "xfs_rmap_btree.h"
30#include "xfs_refcount_btree.h"
31#include "xfs_reflink.h"
32#include "xfs_extent_busy.h"
33#include "xfs_health.h"
34#include "xfs_trace.h"
35#include "xfs_ag.h"
36
37static DEFINE_MUTEX(xfs_uuid_table_mutex);
38static int xfs_uuid_table_size;
39static uuid_t *xfs_uuid_table;
40
41void
42xfs_uuid_table_free(void)
43{
44	if (xfs_uuid_table_size == 0)
45		return;
46	kmem_free(xfs_uuid_table);
47	xfs_uuid_table = NULL;
48	xfs_uuid_table_size = 0;
49}
50
51/*
52 * See if the UUID is unique among mounted XFS filesystems.
53 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
54 */
55STATIC int
56xfs_uuid_mount(
57	struct xfs_mount	*mp)
58{
59	uuid_t			*uuid = &mp->m_sb.sb_uuid;
60	int			hole, i;
61
62	/* Publish UUID in struct super_block */
63	uuid_copy(&mp->m_super->s_uuid, uuid);
64
65	if (mp->m_flags & XFS_MOUNT_NOUUID)
66		return 0;
67
68	if (uuid_is_null(uuid)) {
69		xfs_warn(mp, "Filesystem has null UUID - can't mount");
70		return -EINVAL;
71	}
72
73	mutex_lock(&xfs_uuid_table_mutex);
74	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
75		if (uuid_is_null(&xfs_uuid_table[i])) {
76			hole = i;
77			continue;
78		}
79		if (uuid_equal(uuid, &xfs_uuid_table[i]))
80			goto out_duplicate;
81	}
82
83	if (hole < 0) {
84		xfs_uuid_table = krealloc(xfs_uuid_table,
85			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
86			GFP_KERNEL | __GFP_NOFAIL);
87		hole = xfs_uuid_table_size++;
88	}
89	xfs_uuid_table[hole] = *uuid;
90	mutex_unlock(&xfs_uuid_table_mutex);
91
92	return 0;
93
94 out_duplicate:
95	mutex_unlock(&xfs_uuid_table_mutex);
96	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
97	return -EINVAL;
98}
99
100STATIC void
101xfs_uuid_unmount(
102	struct xfs_mount	*mp)
103{
104	uuid_t			*uuid = &mp->m_sb.sb_uuid;
105	int			i;
106
107	if (mp->m_flags & XFS_MOUNT_NOUUID)
108		return;
109
110	mutex_lock(&xfs_uuid_table_mutex);
111	for (i = 0; i < xfs_uuid_table_size; i++) {
112		if (uuid_is_null(&xfs_uuid_table[i]))
113			continue;
114		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
115			continue;
116		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
117		break;
118	}
119	ASSERT(i < xfs_uuid_table_size);
120	mutex_unlock(&xfs_uuid_table_mutex);
121}
122
123/*
124 * Check size of device based on the (data/realtime) block count.
125 * Note: this check is used by the growfs code as well as mount.
126 */
127int
128xfs_sb_validate_fsb_count(
129	xfs_sb_t	*sbp,
130	uint64_t	nblocks)
131{
132	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
133	ASSERT(sbp->sb_blocklog >= BBSHIFT);
134
135	/* Limited by ULONG_MAX of page cache index */
136	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
137		return -EFBIG;
138	return 0;
139}
140
141/*
142 * xfs_readsb
143 *
144 * Does the initial read of the superblock.
145 */
146int
147xfs_readsb(
148	struct xfs_mount *mp,
149	int		flags)
150{
151	unsigned int	sector_size;
152	struct xfs_buf	*bp;
153	struct xfs_sb	*sbp = &mp->m_sb;
154	int		error;
155	int		loud = !(flags & XFS_MFSI_QUIET);
156	const struct xfs_buf_ops *buf_ops;
157
158	ASSERT(mp->m_sb_bp == NULL);
159	ASSERT(mp->m_ddev_targp != NULL);
160
161	/*
162	 * For the initial read, we must guess at the sector
163	 * size based on the block device.  It's enough to
164	 * get the sb_sectsize out of the superblock and
165	 * then reread with the proper length.
166	 * We don't verify it yet, because it may not be complete.
167	 */
168	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
169	buf_ops = NULL;
170
171	/*
172	 * Allocate a (locked) buffer to hold the superblock. This will be kept
173	 * around at all times to optimize access to the superblock. Therefore,
174	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
175	 * elevated.
176	 */
177reread:
178	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
179				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
180				      buf_ops);
181	if (error) {
182		if (loud)
183			xfs_warn(mp, "SB validate failed with error %d.", error);
184		/* bad CRC means corrupted metadata */
185		if (error == -EFSBADCRC)
186			error = -EFSCORRUPTED;
187		return error;
188	}
189
190	/*
191	 * Initialize the mount structure from the superblock.
192	 */
193	xfs_sb_from_disk(sbp, bp->b_addr);
194
195	/*
196	 * If we haven't validated the superblock, do so now before we try
197	 * to check the sector size and reread the superblock appropriately.
198	 */
199	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
200		if (loud)
201			xfs_warn(mp, "Invalid superblock magic number");
202		error = -EINVAL;
203		goto release_buf;
204	}
205
206	/*
207	 * We must be able to do sector-sized and sector-aligned IO.
208	 */
209	if (sector_size > sbp->sb_sectsize) {
210		if (loud)
211			xfs_warn(mp, "device supports %u byte sectors (not %u)",
212				sector_size, sbp->sb_sectsize);
213		error = -ENOSYS;
214		goto release_buf;
215	}
216
217	if (buf_ops == NULL) {
218		/*
219		 * Re-read the superblock so the buffer is correctly sized,
220		 * and properly verified.
221		 */
222		xfs_buf_relse(bp);
223		sector_size = sbp->sb_sectsize;
224		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
225		goto reread;
226	}
227
228	mp->m_features |= xfs_sb_version_to_features(sbp);
229	xfs_reinit_percpu_counters(mp);
230
231	/* no need to be quiet anymore, so reset the buf ops */
232	bp->b_ops = &xfs_sb_buf_ops;
233
234	mp->m_sb_bp = bp;
235	xfs_buf_unlock(bp);
236	return 0;
237
238release_buf:
239	xfs_buf_relse(bp);
240	return error;
241}
242
243/*
244 * If the sunit/swidth change would move the precomputed root inode value, we
245 * must reject the ondisk change because repair will stumble over that.
246 * However, we allow the mount to proceed because we never rejected this
247 * combination before.  Returns true to update the sb, false otherwise.
248 */
249static inline int
250xfs_check_new_dalign(
251	struct xfs_mount	*mp,
252	int			new_dalign,
253	bool			*update_sb)
254{
255	struct xfs_sb		*sbp = &mp->m_sb;
256	xfs_ino_t		calc_ino;
257
258	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
259	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
260
261	if (sbp->sb_rootino == calc_ino) {
262		*update_sb = true;
263		return 0;
264	}
265
266	xfs_warn(mp,
267"Cannot change stripe alignment; would require moving root inode.");
268
269	/*
270	 * XXX: Next time we add a new incompat feature, this should start
271	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
272	 * that we're ignoring the administrator's instructions.
273	 */
274	xfs_warn(mp, "Skipping superblock stripe alignment update.");
275	*update_sb = false;
276	return 0;
277}
278
279/*
280 * If we were provided with new sunit/swidth values as mount options, make sure
281 * that they pass basic alignment and superblock feature checks, and convert
282 * them into the same units (FSB) that everything else expects.  This step
283 * /must/ be done before computing the inode geometry.
284 */
285STATIC int
286xfs_validate_new_dalign(
287	struct xfs_mount	*mp)
288{
289	if (mp->m_dalign == 0)
290		return 0;
291
292	/*
293	 * If stripe unit and stripe width are not multiples
294	 * of the fs blocksize turn off alignment.
295	 */
296	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
297	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
298		xfs_warn(mp,
299	"alignment check failed: sunit/swidth vs. blocksize(%d)",
300			mp->m_sb.sb_blocksize);
301		return -EINVAL;
302	} else {
303		/*
304		 * Convert the stripe unit and width to FSBs.
305		 */
306		mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
307		if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
308			xfs_warn(mp,
309		"alignment check failed: sunit/swidth vs. agsize(%d)",
310				 mp->m_sb.sb_agblocks);
311			return -EINVAL;
312		} else if (mp->m_dalign) {
313			mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
314		} else {
315			xfs_warn(mp,
316		"alignment check failed: sunit(%d) less than bsize(%d)",
317				 mp->m_dalign, mp->m_sb.sb_blocksize);
318			return -EINVAL;
319		}
320	}
321
322	if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
323		xfs_warn(mp,
324"cannot change alignment: superblock does not support data alignment");
325		return -EINVAL;
326	}
327
328	return 0;
329}
330
331/* Update alignment values based on mount options and sb values. */
332STATIC int
333xfs_update_alignment(
334	struct xfs_mount	*mp)
335{
336	struct xfs_sb		*sbp = &mp->m_sb;
337
338	if (mp->m_dalign) {
339		bool		update_sb;
340		int		error;
341
342		if (sbp->sb_unit == mp->m_dalign &&
343		    sbp->sb_width == mp->m_swidth)
344			return 0;
345
346		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
347		if (error || !update_sb)
348			return error;
349
350		sbp->sb_unit = mp->m_dalign;
351		sbp->sb_width = mp->m_swidth;
352		mp->m_update_sb = true;
353	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
354		    xfs_sb_version_hasdalign(&mp->m_sb)) {
355		mp->m_dalign = sbp->sb_unit;
356		mp->m_swidth = sbp->sb_width;
357	}
358
359	return 0;
360}
361
362/*
363 * precalculate the low space thresholds for dynamic speculative preallocation.
364 */
365void
366xfs_set_low_space_thresholds(
367	struct xfs_mount	*mp)
368{
369	uint64_t		dblocks = mp->m_sb.sb_dblocks;
370	uint64_t		rtexts = mp->m_sb.sb_rextents;
371	int			i;
372
373	do_div(dblocks, 100);
374	do_div(rtexts, 100);
375
376	for (i = 0; i < XFS_LOWSP_MAX; i++) {
377		mp->m_low_space[i] = dblocks * (i + 1);
378		mp->m_low_rtexts[i] = rtexts * (i + 1);
379	}
380}
381
382/*
383 * Check that the data (and log if separate) is an ok size.
384 */
385STATIC int
386xfs_check_sizes(
387	struct xfs_mount *mp)
388{
389	struct xfs_buf	*bp;
390	xfs_daddr_t	d;
391	int		error;
392
393	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
394	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
395		xfs_warn(mp, "filesystem size mismatch detected");
396		return -EFBIG;
397	}
398	error = xfs_buf_read_uncached(mp->m_ddev_targp,
399					d - XFS_FSS_TO_BB(mp, 1),
400					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
401	if (error) {
402		xfs_warn(mp, "last sector read failed");
403		return error;
404	}
405	xfs_buf_relse(bp);
406
407	if (mp->m_logdev_targp == mp->m_ddev_targp)
408		return 0;
409
410	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
411	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
412		xfs_warn(mp, "log size mismatch detected");
413		return -EFBIG;
414	}
415	error = xfs_buf_read_uncached(mp->m_logdev_targp,
416					d - XFS_FSB_TO_BB(mp, 1),
417					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
418	if (error) {
419		xfs_warn(mp, "log device read failed");
420		return error;
421	}
422	xfs_buf_relse(bp);
423	return 0;
424}
425
426/*
427 * Clear the quotaflags in memory and in the superblock.
428 */
429int
430xfs_mount_reset_sbqflags(
431	struct xfs_mount	*mp)
432{
433	mp->m_qflags = 0;
434
435	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
436	if (mp->m_sb.sb_qflags == 0)
437		return 0;
438	spin_lock(&mp->m_sb_lock);
439	mp->m_sb.sb_qflags = 0;
440	spin_unlock(&mp->m_sb_lock);
441
442	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
443		return 0;
444
445	return xfs_sync_sb(mp, false);
446}
447
448uint64_t
449xfs_default_resblks(xfs_mount_t *mp)
450{
451	uint64_t resblks;
452
453	/*
454	 * We default to 5% or 8192 fsbs of space reserved, whichever is
455	 * smaller.  This is intended to cover concurrent allocation
456	 * transactions when we initially hit enospc. These each require a 4
457	 * block reservation. Hence by default we cover roughly 2000 concurrent
458	 * allocation reservations.
459	 */
460	resblks = mp->m_sb.sb_dblocks;
461	do_div(resblks, 20);
462	resblks = min_t(uint64_t, resblks, 8192);
463	return resblks;
464}
465
466/* Ensure the summary counts are correct. */
467STATIC int
468xfs_check_summary_counts(
469	struct xfs_mount	*mp)
470{
471	/*
472	 * The AG0 superblock verifier rejects in-progress filesystems,
473	 * so we should never see the flag set this far into mounting.
474	 */
475	if (mp->m_sb.sb_inprogress) {
476		xfs_err(mp, "sb_inprogress set after log recovery??");
477		WARN_ON(1);
478		return -EFSCORRUPTED;
479	}
480
481	/*
482	 * Now the log is mounted, we know if it was an unclean shutdown or
483	 * not. If it was, with the first phase of recovery has completed, we
484	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
485	 * but they are recovered transactionally in the second recovery phase
486	 * later.
487	 *
488	 * If the log was clean when we mounted, we can check the summary
489	 * counters.  If any of them are obviously incorrect, we can recompute
490	 * them from the AGF headers in the next step.
491	 */
492	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
493	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
494	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
495	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
496		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
497
498	/*
499	 * We can safely re-initialise incore superblock counters from the
500	 * per-ag data. These may not be correct if the filesystem was not
501	 * cleanly unmounted, so we waited for recovery to finish before doing
502	 * this.
503	 *
504	 * If the filesystem was cleanly unmounted or the previous check did
505	 * not flag anything weird, then we can trust the values in the
506	 * superblock to be correct and we don't need to do anything here.
507	 * Otherwise, recalculate the summary counters.
508	 */
509	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
510	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
511	    !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
512		return 0;
513
514	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
515}
516
517/*
518 * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
519 * internal inode structures can be sitting in the CIL and AIL at this point,
520 * so we need to unpin them, write them back and/or reclaim them before unmount
521 * can proceed.  In other words, callers are required to have inactivated all
522 * inodes.
523 *
524 * An inode cluster that has been freed can have its buffer still pinned in
525 * memory because the transaction is still sitting in a iclog. The stale inodes
526 * on that buffer will be pinned to the buffer until the transaction hits the
527 * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
528 * may never see the pinned buffer, so nothing will push out the iclog and
529 * unpin the buffer.
530 *
531 * Hence we need to force the log to unpin everything first. However, log
532 * forces don't wait for the discards they issue to complete, so we have to
533 * explicitly wait for them to complete here as well.
534 *
535 * Then we can tell the world we are unmounting so that error handling knows
536 * that the filesystem is going away and we should error out anything that we
537 * have been retrying in the background.  This will prevent never-ending
538 * retries in AIL pushing from hanging the unmount.
539 *
540 * Finally, we can push the AIL to clean all the remaining dirty objects, then
541 * reclaim the remaining inodes that are still in memory at this point in time.
542 */
543static void
544xfs_unmount_flush_inodes(
545	struct xfs_mount	*mp)
546{
547	xfs_log_force(mp, XFS_LOG_SYNC);
548	xfs_extent_busy_wait_all(mp);
549	flush_workqueue(xfs_discard_wq);
550
551	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
552
553	xfs_ail_push_all_sync(mp->m_ail);
554	xfs_inodegc_stop(mp);
555	cancel_delayed_work_sync(&mp->m_reclaim_work);
556	xfs_reclaim_inodes(mp);
557	xfs_health_unmount(mp);
558}
559
560static void
561xfs_mount_setup_inode_geom(
562	struct xfs_mount	*mp)
563{
564	struct xfs_ino_geometry *igeo = M_IGEO(mp);
565
566	igeo->attr_fork_offset = xfs_bmap_compute_attr_offset(mp);
567	ASSERT(igeo->attr_fork_offset < XFS_LITINO(mp));
568
569	xfs_ialloc_setup_geometry(mp);
570}
571
572/*
573 * This function does the following on an initial mount of a file system:
574 *	- reads the superblock from disk and init the mount struct
575 *	- if we're a 32-bit kernel, do a size check on the superblock
576 *		so we don't mount terabyte filesystems
577 *	- init mount struct realtime fields
578 *	- allocate inode hash table for fs
579 *	- init directory manager
580 *	- perform recovery and init the log manager
581 */
582int
583xfs_mountfs(
584	struct xfs_mount	*mp)
585{
586	struct xfs_sb		*sbp = &(mp->m_sb);
587	struct xfs_inode	*rip;
588	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
589	uint64_t		resblks;
590	uint			quotamount = 0;
591	uint			quotaflags = 0;
592	int			error = 0;
593
594	xfs_sb_mount_common(mp, sbp);
595
596	/*
597	 * Check for a mismatched features2 values.  Older kernels read & wrote
598	 * into the wrong sb offset for sb_features2 on some platforms due to
599	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
600	 * which made older superblock reading/writing routines swap it as a
601	 * 64-bit value.
602	 *
603	 * For backwards compatibility, we make both slots equal.
604	 *
605	 * If we detect a mismatched field, we OR the set bits into the existing
606	 * features2 field in case it has already been modified; we don't want
607	 * to lose any features.  We then update the bad location with the ORed
608	 * value so that older kernels will see any features2 flags. The
609	 * superblock writeback code ensures the new sb_features2 is copied to
610	 * sb_bad_features2 before it is logged or written to disk.
611	 */
612	if (xfs_sb_has_mismatched_features2(sbp)) {
613		xfs_warn(mp, "correcting sb_features alignment problem");
614		sbp->sb_features2 |= sbp->sb_bad_features2;
615		mp->m_update_sb = true;
616	}
617
618
619	/* always use v2 inodes by default now */
620	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
621		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
622		mp->m_update_sb = true;
623	}
624
625	/*
626	 * If we were given new sunit/swidth options, do some basic validation
627	 * checks and convert the incore dalign and swidth values to the
628	 * same units (FSB) that everything else uses.  This /must/ happen
629	 * before computing the inode geometry.
630	 */
631	error = xfs_validate_new_dalign(mp);
632	if (error)
633		goto out;
634
635	xfs_alloc_compute_maxlevels(mp);
636	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
637	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
638	xfs_mount_setup_inode_geom(mp);
639	xfs_rmapbt_compute_maxlevels(mp);
640	xfs_refcountbt_compute_maxlevels(mp);
641
642	/*
643	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
644	 * is NOT aligned turn off m_dalign since allocator alignment is within
645	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
646	 * we must compute the free space and rmap btree geometry before doing
647	 * this.
648	 */
649	error = xfs_update_alignment(mp);
650	if (error)
651		goto out;
652
653	/* enable fail_at_unmount as default */
654	mp->m_fail_unmount = true;
655
656	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
657			       NULL, mp->m_super->s_id);
658	if (error)
659		goto out;
660
661	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
662			       &mp->m_kobj, "stats");
663	if (error)
664		goto out_remove_sysfs;
665
666	error = xfs_error_sysfs_init(mp);
667	if (error)
668		goto out_del_stats;
669
670	error = xfs_errortag_init(mp);
671	if (error)
672		goto out_remove_error_sysfs;
673
674	error = xfs_uuid_mount(mp);
675	if (error)
676		goto out_remove_errortag;
677
678	/*
679	 * Update the preferred write size based on the information from the
680	 * on-disk superblock.
681	 */
682	mp->m_allocsize_log =
683		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
684	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
685
686	/* set the low space thresholds for dynamic preallocation */
687	xfs_set_low_space_thresholds(mp);
688
689	/*
690	 * If enabled, sparse inode chunk alignment is expected to match the
691	 * cluster size. Full inode chunk alignment must match the chunk size,
692	 * but that is checked on sb read verification...
693	 */
694	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
695	    mp->m_sb.sb_spino_align !=
696			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
697		xfs_warn(mp,
698	"Sparse inode block alignment (%u) must match cluster size (%llu).",
699			 mp->m_sb.sb_spino_align,
700			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
701		error = -EINVAL;
702		goto out_remove_uuid;
703	}
704
705	/*
706	 * Check that the data (and log if separate) is an ok size.
707	 */
708	error = xfs_check_sizes(mp);
709	if (error)
710		goto out_remove_uuid;
711
712	/*
713	 * Initialize realtime fields in the mount structure
714	 */
715	error = xfs_rtmount_init(mp);
716	if (error) {
717		xfs_warn(mp, "RT mount failed");
718		goto out_remove_uuid;
719	}
720
721	/*
722	 *  Copies the low order bits of the timestamp and the randomly
723	 *  set "sequence" number out of a UUID.
724	 */
725	mp->m_fixedfsid[0] =
726		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
727		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
728	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
729
730	error = xfs_da_mount(mp);
731	if (error) {
732		xfs_warn(mp, "Failed dir/attr init: %d", error);
733		goto out_remove_uuid;
734	}
735
736	/*
737	 * Initialize the precomputed transaction reservations values.
738	 */
739	xfs_trans_init(mp);
740
741	/*
742	 * Allocate and initialize the per-ag data.
743	 */
744	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
745	if (error) {
746		xfs_warn(mp, "Failed per-ag init: %d", error);
747		goto out_free_dir;
748	}
749
750	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
751		xfs_warn(mp, "no log defined");
752		error = -EFSCORRUPTED;
753		goto out_free_perag;
754	}
755
756	error = xfs_inodegc_register_shrinker(mp);
757	if (error)
758		goto out_fail_wait;
759
760	/*
761	 * Log's mount-time initialization. The first part of recovery can place
762	 * some items on the AIL, to be handled when recovery is finished or
763	 * cancelled.
764	 */
765	error = xfs_log_mount(mp, mp->m_logdev_targp,
766			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
767			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
768	if (error) {
769		xfs_warn(mp, "log mount failed");
770		goto out_inodegc_shrinker;
771	}
772
773	/* Make sure the summary counts are ok. */
774	error = xfs_check_summary_counts(mp);
775	if (error)
776		goto out_log_dealloc;
777
778	/* Enable background inode inactivation workers. */
779	xfs_inodegc_start(mp);
780	xfs_blockgc_start(mp);
781
782	/*
783	 * Now that we've recovered any pending superblock feature bit
784	 * additions, we can finish setting up the attr2 behaviour for the
785	 * mount. If no attr2 mount options were specified, the we use the
786	 * behaviour specified by the superblock feature bit.
787	 */
788	if (!(mp->m_flags & (XFS_MOUNT_ATTR2|XFS_MOUNT_NOATTR2)) &&
789	    xfs_sb_version_hasattr2(&mp->m_sb))
790		mp->m_flags |= XFS_MOUNT_ATTR2;
791
792	/*
793	 * Get and sanity-check the root inode.
794	 * Save the pointer to it in the mount structure.
795	 */
796	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
797			 XFS_ILOCK_EXCL, &rip);
798	if (error) {
799		xfs_warn(mp,
800			"Failed to read root inode 0x%llx, error %d",
801			sbp->sb_rootino, -error);
802		goto out_log_dealloc;
803	}
804
805	ASSERT(rip != NULL);
806
807	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
808		xfs_warn(mp, "corrupted root inode %llu: not a directory",
809			(unsigned long long)rip->i_ino);
810		xfs_iunlock(rip, XFS_ILOCK_EXCL);
811		error = -EFSCORRUPTED;
812		goto out_rele_rip;
813	}
814	mp->m_rootip = rip;	/* save it */
815
816	xfs_iunlock(rip, XFS_ILOCK_EXCL);
817
818	/*
819	 * Initialize realtime inode pointers in the mount structure
820	 */
821	error = xfs_rtmount_inodes(mp);
822	if (error) {
823		/*
824		 * Free up the root inode.
825		 */
826		xfs_warn(mp, "failed to read RT inodes");
827		goto out_rele_rip;
828	}
829
830	/*
831	 * If this is a read-only mount defer the superblock updates until
832	 * the next remount into writeable mode.  Otherwise we would never
833	 * perform the update e.g. for the root filesystem.
834	 */
835	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
836		error = xfs_sync_sb(mp, false);
837		if (error) {
838			xfs_warn(mp, "failed to write sb changes");
839			goto out_rtunmount;
840		}
841	}
842
843	/*
844	 * Initialise the XFS quota management subsystem for this mount
845	 */
846	if (XFS_IS_QUOTA_ON(mp)) {
847		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
848		if (error)
849			goto out_rtunmount;
850	} else {
851		/*
852		 * If a file system had quotas running earlier, but decided to
853		 * mount without -o uquota/pquota/gquota options, revoke the
854		 * quotachecked license.
855		 */
856		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
857			xfs_notice(mp, "resetting quota flags");
858			error = xfs_mount_reset_sbqflags(mp);
859			if (error)
860				goto out_rtunmount;
861		}
862	}
863
864	/*
865	 * Finish recovering the file system.  This part needed to be delayed
866	 * until after the root and real-time bitmap inodes were consistently
867	 * read in.  Temporarily create per-AG space reservations for metadata
868	 * btree shape changes because space freeing transactions (for inode
869	 * inactivation) require the per-AG reservation in lieu of reserving
870	 * blocks.
871	 */
872	error = xfs_fs_reserve_ag_blocks(mp);
873	if (error && error == -ENOSPC)
874		xfs_warn(mp,
875	"ENOSPC reserving per-AG metadata pool, log recovery may fail.");
876	error = xfs_log_mount_finish(mp);
877	xfs_fs_unreserve_ag_blocks(mp);
878	if (error) {
879		xfs_warn(mp, "log mount finish failed");
880		goto out_rtunmount;
881	}
882
883	/*
884	 * Now the log is fully replayed, we can transition to full read-only
885	 * mode for read-only mounts. This will sync all the metadata and clean
886	 * the log so that the recovery we just performed does not have to be
887	 * replayed again on the next mount.
888	 *
889	 * We use the same quiesce mechanism as the rw->ro remount, as they are
890	 * semantically identical operations.
891	 */
892	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
893							XFS_MOUNT_RDONLY) {
894		xfs_log_clean(mp);
895	}
896
897	/*
898	 * Complete the quota initialisation, post-log-replay component.
899	 */
900	if (quotamount) {
901		ASSERT(mp->m_qflags == 0);
902		mp->m_qflags = quotaflags;
903
904		xfs_qm_mount_quotas(mp);
905	}
906
907	/*
908	 * Now we are mounted, reserve a small amount of unused space for
909	 * privileged transactions. This is needed so that transaction
910	 * space required for critical operations can dip into this pool
911	 * when at ENOSPC. This is needed for operations like create with
912	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
913	 * are not allowed to use this reserved space.
914	 *
915	 * This may drive us straight to ENOSPC on mount, but that implies
916	 * we were already there on the last unmount. Warn if this occurs.
917	 */
918	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
919		resblks = xfs_default_resblks(mp);
920		error = xfs_reserve_blocks(mp, &resblks, NULL);
921		if (error)
922			xfs_warn(mp,
923	"Unable to allocate reserve blocks. Continuing without reserve pool.");
924
925		/* Recover any CoW blocks that never got remapped. */
926		error = xfs_reflink_recover_cow(mp);
927		if (error) {
928			xfs_err(mp,
929	"Error %d recovering leftover CoW allocations.", error);
930			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
931			goto out_quota;
932		}
933
934		/* Reserve AG blocks for future btree expansion. */
935		error = xfs_fs_reserve_ag_blocks(mp);
936		if (error && error != -ENOSPC)
937			goto out_agresv;
938	}
939
940	return 0;
941
942 out_agresv:
943	xfs_fs_unreserve_ag_blocks(mp);
944 out_quota:
945	xfs_qm_unmount_quotas(mp);
946 out_rtunmount:
947	xfs_rtunmount_inodes(mp);
948 out_rele_rip:
949	xfs_irele(rip);
950	/* Clean out dquots that might be in memory after quotacheck. */
951	xfs_qm_unmount(mp);
952
953	/*
954	 * Inactivate all inodes that might still be in memory after a log
955	 * intent recovery failure so that reclaim can free them.  Metadata
956	 * inodes and the root directory shouldn't need inactivation, but the
957	 * mount failed for some reason, so pull down all the state and flee.
958	 */
959	xfs_inodegc_flush(mp);
960
961	/*
962	 * Flush all inode reclamation work and flush the log.
963	 * We have to do this /after/ rtunmount and qm_unmount because those
964	 * two will have scheduled delayed reclaim for the rt/quota inodes.
965	 *
966	 * This is slightly different from the unmountfs call sequence
967	 * because we could be tearing down a partially set up mount.  In
968	 * particular, if log_mount_finish fails we bail out without calling
969	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
970	 * quota inodes.
971	 */
972	xfs_unmount_flush_inodes(mp);
973 out_log_dealloc:
974	xfs_log_mount_cancel(mp);
975 out_inodegc_shrinker:
976	unregister_shrinker(&mp->m_inodegc_shrinker);
977 out_fail_wait:
978	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
979		xfs_buftarg_drain(mp->m_logdev_targp);
980	xfs_buftarg_drain(mp->m_ddev_targp);
981 out_free_perag:
982	xfs_free_perag(mp);
983 out_free_dir:
984	xfs_da_unmount(mp);
985 out_remove_uuid:
986	xfs_uuid_unmount(mp);
987 out_remove_errortag:
988	xfs_errortag_del(mp);
989 out_remove_error_sysfs:
990	xfs_error_sysfs_del(mp);
991 out_del_stats:
992	xfs_sysfs_del(&mp->m_stats.xs_kobj);
993 out_remove_sysfs:
994	xfs_sysfs_del(&mp->m_kobj);
995 out:
996	return error;
997}
998
999/*
1000 * This flushes out the inodes,dquots and the superblock, unmounts the
1001 * log and makes sure that incore structures are freed.
1002 */
1003void
1004xfs_unmountfs(
1005	struct xfs_mount	*mp)
1006{
1007	uint64_t		resblks;
1008	int			error;
1009
1010	/*
1011	 * Perform all on-disk metadata updates required to inactivate inodes
1012	 * that the VFS evicted earlier in the unmount process.  Freeing inodes
1013	 * and discarding CoW fork preallocations can cause shape changes to
1014	 * the free inode and refcount btrees, respectively, so we must finish
1015	 * this before we discard the metadata space reservations.  Metadata
1016	 * inodes and the root directory do not require inactivation.
1017	 */
1018	xfs_inodegc_flush(mp);
1019
1020	xfs_blockgc_stop(mp);
1021	xfs_fs_unreserve_ag_blocks(mp);
1022	xfs_qm_unmount_quotas(mp);
1023	xfs_rtunmount_inodes(mp);
1024	xfs_irele(mp->m_rootip);
1025
1026	xfs_unmount_flush_inodes(mp);
1027
1028	xfs_qm_unmount(mp);
1029
1030	/*
1031	 * Unreserve any blocks we have so that when we unmount we don't account
1032	 * the reserved free space as used. This is really only necessary for
1033	 * lazy superblock counting because it trusts the incore superblock
1034	 * counters to be absolutely correct on clean unmount.
1035	 *
1036	 * We don't bother correcting this elsewhere for lazy superblock
1037	 * counting because on mount of an unclean filesystem we reconstruct the
1038	 * correct counter value and this is irrelevant.
1039	 *
1040	 * For non-lazy counter filesystems, this doesn't matter at all because
1041	 * we only every apply deltas to the superblock and hence the incore
1042	 * value does not matter....
1043	 */
1044	resblks = 0;
1045	error = xfs_reserve_blocks(mp, &resblks, NULL);
1046	if (error)
1047		xfs_warn(mp, "Unable to free reserved block pool. "
1048				"Freespace may not be correct on next mount.");
1049
1050	xfs_log_unmount(mp);
1051	xfs_da_unmount(mp);
1052	xfs_uuid_unmount(mp);
1053
1054#if defined(DEBUG)
1055	xfs_errortag_clearall(mp);
1056#endif
1057	unregister_shrinker(&mp->m_inodegc_shrinker);
1058	xfs_free_perag(mp);
1059
1060	xfs_errortag_del(mp);
1061	xfs_error_sysfs_del(mp);
1062	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1063	xfs_sysfs_del(&mp->m_kobj);
1064}
1065
1066/*
1067 * Determine whether modifications can proceed. The caller specifies the minimum
1068 * freeze level for which modifications should not be allowed. This allows
1069 * certain operations to proceed while the freeze sequence is in progress, if
1070 * necessary.
1071 */
1072bool
1073xfs_fs_writable(
1074	struct xfs_mount	*mp,
1075	int			level)
1076{
1077	ASSERT(level > SB_UNFROZEN);
1078	if ((mp->m_super->s_writers.frozen >= level) ||
1079	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1080		return false;
1081
1082	return true;
1083}
1084
1085int
1086xfs_mod_fdblocks(
1087	struct xfs_mount	*mp,
1088	int64_t			delta,
1089	bool			rsvd)
1090{
1091	int64_t			lcounter;
1092	long long		res_used;
1093	s32			batch;
1094	uint64_t		set_aside;
1095
1096	if (delta > 0) {
1097		/*
1098		 * If the reserve pool is depleted, put blocks back into it
1099		 * first. Most of the time the pool is full.
1100		 */
1101		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1102			percpu_counter_add(&mp->m_fdblocks, delta);
1103			return 0;
1104		}
1105
1106		spin_lock(&mp->m_sb_lock);
1107		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1108
1109		if (res_used > delta) {
1110			mp->m_resblks_avail += delta;
1111		} else {
1112			delta -= res_used;
1113			mp->m_resblks_avail = mp->m_resblks;
1114			percpu_counter_add(&mp->m_fdblocks, delta);
1115		}
1116		spin_unlock(&mp->m_sb_lock);
1117		return 0;
1118	}
1119
1120	/*
1121	 * Taking blocks away, need to be more accurate the closer we
1122	 * are to zero.
1123	 *
1124	 * If the counter has a value of less than 2 * max batch size,
1125	 * then make everything serialise as we are real close to
1126	 * ENOSPC.
1127	 */
1128	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1129				     XFS_FDBLOCKS_BATCH) < 0)
1130		batch = 1;
1131	else
1132		batch = XFS_FDBLOCKS_BATCH;
1133
1134	/*
1135	 * Set aside allocbt blocks because these blocks are tracked as free
1136	 * space but not available for allocation. Technically this means that a
1137	 * single reservation cannot consume all remaining free space, but the
1138	 * ratio of allocbt blocks to usable free blocks should be rather small.
1139	 * The tradeoff without this is that filesystems that maintain high
1140	 * perag block reservations can over reserve physical block availability
1141	 * and fail physical allocation, which leads to much more serious
1142	 * problems (i.e. transaction abort, pagecache discards, etc.) than
1143	 * slightly premature -ENOSPC.
1144	 */
1145	set_aside = mp->m_alloc_set_aside + atomic64_read(&mp->m_allocbt_blks);
1146	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1147	if (__percpu_counter_compare(&mp->m_fdblocks, set_aside,
1148				     XFS_FDBLOCKS_BATCH) >= 0) {
1149		/* we had space! */
1150		return 0;
1151	}
1152
1153	/*
1154	 * lock up the sb for dipping into reserves before releasing the space
1155	 * that took us to ENOSPC.
1156	 */
1157	spin_lock(&mp->m_sb_lock);
1158	percpu_counter_add(&mp->m_fdblocks, -delta);
1159	if (!rsvd)
1160		goto fdblocks_enospc;
1161
1162	lcounter = (long long)mp->m_resblks_avail + delta;
1163	if (lcounter >= 0) {
1164		mp->m_resblks_avail = lcounter;
1165		spin_unlock(&mp->m_sb_lock);
1166		return 0;
1167	}
1168	xfs_warn_once(mp,
1169"Reserve blocks depleted! Consider increasing reserve pool size.");
1170
1171fdblocks_enospc:
1172	spin_unlock(&mp->m_sb_lock);
1173	return -ENOSPC;
1174}
1175
1176int
1177xfs_mod_frextents(
1178	struct xfs_mount	*mp,
1179	int64_t			delta)
1180{
1181	int64_t			lcounter;
1182	int			ret = 0;
1183
1184	spin_lock(&mp->m_sb_lock);
1185	lcounter = mp->m_sb.sb_frextents + delta;
1186	if (lcounter < 0)
1187		ret = -ENOSPC;
1188	else
1189		mp->m_sb.sb_frextents = lcounter;
1190	spin_unlock(&mp->m_sb_lock);
1191	return ret;
1192}
1193
1194/*
1195 * Used to free the superblock along various error paths.
1196 */
1197void
1198xfs_freesb(
1199	struct xfs_mount	*mp)
1200{
1201	struct xfs_buf		*bp = mp->m_sb_bp;
1202
1203	xfs_buf_lock(bp);
1204	mp->m_sb_bp = NULL;
1205	xfs_buf_relse(bp);
1206}
1207
1208/*
1209 * If the underlying (data/log/rt) device is readonly, there are some
1210 * operations that cannot proceed.
1211 */
1212int
1213xfs_dev_is_read_only(
1214	struct xfs_mount	*mp,
1215	char			*message)
1216{
1217	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1218	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1219	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1220		xfs_notice(mp, "%s required on read-only device.", message);
1221		xfs_notice(mp, "write access unavailable, cannot proceed.");
1222		return -EROFS;
1223	}
1224	return 0;
1225}
1226
1227/* Force the summary counters to be recalculated at next mount. */
1228void
1229xfs_force_summary_recalc(
1230	struct xfs_mount	*mp)
1231{
1232	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1233		return;
1234
1235	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1236}
1237
1238/*
1239 * Enable a log incompat feature flag in the primary superblock.  The caller
1240 * cannot have any other transactions in progress.
1241 */
1242int
1243xfs_add_incompat_log_feature(
1244	struct xfs_mount	*mp,
1245	uint32_t		feature)
1246{
1247	struct xfs_dsb		*dsb;
1248	int			error;
1249
1250	ASSERT(hweight32(feature) == 1);
1251	ASSERT(!(feature & XFS_SB_FEAT_INCOMPAT_LOG_UNKNOWN));
1252
1253	/*
1254	 * Force the log to disk and kick the background AIL thread to reduce
1255	 * the chances that the bwrite will stall waiting for the AIL to unpin
1256	 * the primary superblock buffer.  This isn't a data integrity
1257	 * operation, so we don't need a synchronous push.
1258	 */
1259	error = xfs_log_force(mp, XFS_LOG_SYNC);
1260	if (error)
1261		return error;
1262	xfs_ail_push_all(mp->m_ail);
1263
1264	/*
1265	 * Lock the primary superblock buffer to serialize all callers that
1266	 * are trying to set feature bits.
1267	 */
1268	xfs_buf_lock(mp->m_sb_bp);
1269	xfs_buf_hold(mp->m_sb_bp);
1270
1271	if (XFS_FORCED_SHUTDOWN(mp)) {
1272		error = -EIO;
1273		goto rele;
1274	}
1275
1276	if (xfs_sb_has_incompat_log_feature(&mp->m_sb, feature))
1277		goto rele;
1278
1279	/*
1280	 * Write the primary superblock to disk immediately, because we need
1281	 * the log_incompat bit to be set in the primary super now to protect
1282	 * the log items that we're going to commit later.
1283	 */
1284	dsb = mp->m_sb_bp->b_addr;
1285	xfs_sb_to_disk(dsb, &mp->m_sb);
1286	dsb->sb_features_log_incompat |= cpu_to_be32(feature);
1287	error = xfs_bwrite(mp->m_sb_bp);
1288	if (error)
1289		goto shutdown;
1290
1291	/*
1292	 * Add the feature bits to the incore superblock before we unlock the
1293	 * buffer.
1294	 */
1295	xfs_sb_add_incompat_log_features(&mp->m_sb, feature);
1296	xfs_buf_relse(mp->m_sb_bp);
1297
1298	/* Log the superblock to disk. */
1299	return xfs_sync_sb(mp, false);
1300shutdown:
1301	xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1302rele:
1303	xfs_buf_relse(mp->m_sb_bp);
1304	return error;
1305}
1306
1307/*
1308 * Clear all the log incompat flags from the superblock.
1309 *
1310 * The caller cannot be in a transaction, must ensure that the log does not
1311 * contain any log items protected by any log incompat bit, and must ensure
1312 * that there are no other threads that depend on the state of the log incompat
1313 * feature flags in the primary super.
1314 *
1315 * Returns true if the superblock is dirty.
1316 */
1317bool
1318xfs_clear_incompat_log_features(
1319	struct xfs_mount	*mp)
1320{
1321	bool			ret = false;
1322
1323	if (!xfs_sb_version_hascrc(&mp->m_sb) ||
1324	    !xfs_sb_has_incompat_log_feature(&mp->m_sb,
1325				XFS_SB_FEAT_INCOMPAT_LOG_ALL) ||
1326	    XFS_FORCED_SHUTDOWN(mp))
1327		return false;
1328
1329	/*
1330	 * Update the incore superblock.  We synchronize on the primary super
1331	 * buffer lock to be consistent with the add function, though at least
1332	 * in theory this shouldn't be necessary.
1333	 */
1334	xfs_buf_lock(mp->m_sb_bp);
1335	xfs_buf_hold(mp->m_sb_bp);
1336
1337	if (xfs_sb_has_incompat_log_feature(&mp->m_sb,
1338				XFS_SB_FEAT_INCOMPAT_LOG_ALL)) {
1339		xfs_info(mp, "Clearing log incompat feature flags.");
1340		xfs_sb_remove_incompat_log_features(&mp->m_sb);
1341		ret = true;
1342	}
1343
1344	xfs_buf_relse(mp->m_sb_bp);
1345	return ret;
1346}
1347
1348/*
1349 * Update the in-core delayed block counter.
1350 *
1351 * We prefer to update the counter without having to take a spinlock for every
1352 * counter update (i.e. batching).  Each change to delayed allocation
1353 * reservations can change can easily exceed the default percpu counter
1354 * batching, so we use a larger batch factor here.
1355 *
1356 * Note that we don't currently have any callers requiring fast summation
1357 * (e.g. percpu_counter_read) so we can use a big batch value here.
1358 */
1359#define XFS_DELALLOC_BATCH	(4096)
1360void
1361xfs_mod_delalloc(
1362	struct xfs_mount	*mp,
1363	int64_t			delta)
1364{
1365	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1366			XFS_DELALLOC_BATCH);
1367}
1368