xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision 163b0991)
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 
36 static DEFINE_MUTEX(xfs_uuid_table_mutex);
37 static int xfs_uuid_table_size;
38 static uuid_t *xfs_uuid_table;
39 
40 void
41 xfs_uuid_table_free(void)
42 {
43 	if (xfs_uuid_table_size == 0)
44 		return;
45 	kmem_free(xfs_uuid_table);
46 	xfs_uuid_table = NULL;
47 	xfs_uuid_table_size = 0;
48 }
49 
50 /*
51  * See if the UUID is unique among mounted XFS filesystems.
52  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
53  */
54 STATIC int
55 xfs_uuid_mount(
56 	struct xfs_mount	*mp)
57 {
58 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
59 	int			hole, i;
60 
61 	/* Publish UUID in struct super_block */
62 	uuid_copy(&mp->m_super->s_uuid, uuid);
63 
64 	if (mp->m_flags & XFS_MOUNT_NOUUID)
65 		return 0;
66 
67 	if (uuid_is_null(uuid)) {
68 		xfs_warn(mp, "Filesystem has null UUID - can't mount");
69 		return -EINVAL;
70 	}
71 
72 	mutex_lock(&xfs_uuid_table_mutex);
73 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
74 		if (uuid_is_null(&xfs_uuid_table[i])) {
75 			hole = i;
76 			continue;
77 		}
78 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
79 			goto out_duplicate;
80 	}
81 
82 	if (hole < 0) {
83 		xfs_uuid_table = krealloc(xfs_uuid_table,
84 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
85 			GFP_KERNEL | __GFP_NOFAIL);
86 		hole = xfs_uuid_table_size++;
87 	}
88 	xfs_uuid_table[hole] = *uuid;
89 	mutex_unlock(&xfs_uuid_table_mutex);
90 
91 	return 0;
92 
93  out_duplicate:
94 	mutex_unlock(&xfs_uuid_table_mutex);
95 	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
96 	return -EINVAL;
97 }
98 
99 STATIC void
100 xfs_uuid_unmount(
101 	struct xfs_mount	*mp)
102 {
103 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
104 	int			i;
105 
106 	if (mp->m_flags & XFS_MOUNT_NOUUID)
107 		return;
108 
109 	mutex_lock(&xfs_uuid_table_mutex);
110 	for (i = 0; i < xfs_uuid_table_size; i++) {
111 		if (uuid_is_null(&xfs_uuid_table[i]))
112 			continue;
113 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
114 			continue;
115 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
116 		break;
117 	}
118 	ASSERT(i < xfs_uuid_table_size);
119 	mutex_unlock(&xfs_uuid_table_mutex);
120 }
121 
122 
123 STATIC void
124 __xfs_free_perag(
125 	struct rcu_head	*head)
126 {
127 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
128 
129 	ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
130 	ASSERT(atomic_read(&pag->pag_ref) == 0);
131 	kmem_free(pag);
132 }
133 
134 /*
135  * Free up the per-ag resources associated with the mount structure.
136  */
137 STATIC void
138 xfs_free_perag(
139 	xfs_mount_t	*mp)
140 {
141 	xfs_agnumber_t	agno;
142 	struct xfs_perag *pag;
143 
144 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
145 		spin_lock(&mp->m_perag_lock);
146 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
147 		spin_unlock(&mp->m_perag_lock);
148 		ASSERT(pag);
149 		ASSERT(atomic_read(&pag->pag_ref) == 0);
150 		cancel_delayed_work_sync(&pag->pag_blockgc_work);
151 		xfs_iunlink_destroy(pag);
152 		xfs_buf_hash_destroy(pag);
153 		call_rcu(&pag->rcu_head, __xfs_free_perag);
154 	}
155 }
156 
157 /*
158  * Check size of device based on the (data/realtime) block count.
159  * Note: this check is used by the growfs code as well as mount.
160  */
161 int
162 xfs_sb_validate_fsb_count(
163 	xfs_sb_t	*sbp,
164 	uint64_t	nblocks)
165 {
166 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
167 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
168 
169 	/* Limited by ULONG_MAX of page cache index */
170 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
171 		return -EFBIG;
172 	return 0;
173 }
174 
175 int
176 xfs_initialize_perag(
177 	xfs_mount_t	*mp,
178 	xfs_agnumber_t	agcount,
179 	xfs_agnumber_t	*maxagi)
180 {
181 	xfs_agnumber_t	index;
182 	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
183 	xfs_perag_t	*pag;
184 	int		error = -ENOMEM;
185 
186 	/*
187 	 * Walk the current per-ag tree so we don't try to initialise AGs
188 	 * that already exist (growfs case). Allocate and insert all the
189 	 * AGs we don't find ready for initialisation.
190 	 */
191 	for (index = 0; index < agcount; index++) {
192 		pag = xfs_perag_get(mp, index);
193 		if (pag) {
194 			xfs_perag_put(pag);
195 			continue;
196 		}
197 
198 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
199 		if (!pag) {
200 			error = -ENOMEM;
201 			goto out_unwind_new_pags;
202 		}
203 		pag->pag_agno = index;
204 		pag->pag_mount = mp;
205 		spin_lock_init(&pag->pag_ici_lock);
206 		INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
207 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
208 
209 		error = xfs_buf_hash_init(pag);
210 		if (error)
211 			goto out_free_pag;
212 		init_waitqueue_head(&pag->pagb_wait);
213 		spin_lock_init(&pag->pagb_lock);
214 		pag->pagb_count = 0;
215 		pag->pagb_tree = RB_ROOT;
216 
217 		error = radix_tree_preload(GFP_NOFS);
218 		if (error)
219 			goto out_hash_destroy;
220 
221 		spin_lock(&mp->m_perag_lock);
222 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
223 			WARN_ON_ONCE(1);
224 			spin_unlock(&mp->m_perag_lock);
225 			radix_tree_preload_end();
226 			error = -EEXIST;
227 			goto out_hash_destroy;
228 		}
229 		spin_unlock(&mp->m_perag_lock);
230 		radix_tree_preload_end();
231 		/* first new pag is fully initialized */
232 		if (first_initialised == NULLAGNUMBER)
233 			first_initialised = index;
234 		error = xfs_iunlink_init(pag);
235 		if (error)
236 			goto out_hash_destroy;
237 		spin_lock_init(&pag->pag_state_lock);
238 	}
239 
240 	index = xfs_set_inode_alloc(mp, agcount);
241 
242 	if (maxagi)
243 		*maxagi = index;
244 
245 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
246 	return 0;
247 
248 out_hash_destroy:
249 	xfs_buf_hash_destroy(pag);
250 out_free_pag:
251 	kmem_free(pag);
252 out_unwind_new_pags:
253 	/* unwind any prior newly initialized pags */
254 	for (index = first_initialised; index < agcount; index++) {
255 		pag = radix_tree_delete(&mp->m_perag_tree, index);
256 		if (!pag)
257 			break;
258 		xfs_buf_hash_destroy(pag);
259 		xfs_iunlink_destroy(pag);
260 		kmem_free(pag);
261 	}
262 	return error;
263 }
264 
265 /*
266  * xfs_readsb
267  *
268  * Does the initial read of the superblock.
269  */
270 int
271 xfs_readsb(
272 	struct xfs_mount *mp,
273 	int		flags)
274 {
275 	unsigned int	sector_size;
276 	struct xfs_buf	*bp;
277 	struct xfs_sb	*sbp = &mp->m_sb;
278 	int		error;
279 	int		loud = !(flags & XFS_MFSI_QUIET);
280 	const struct xfs_buf_ops *buf_ops;
281 
282 	ASSERT(mp->m_sb_bp == NULL);
283 	ASSERT(mp->m_ddev_targp != NULL);
284 
285 	/*
286 	 * For the initial read, we must guess at the sector
287 	 * size based on the block device.  It's enough to
288 	 * get the sb_sectsize out of the superblock and
289 	 * then reread with the proper length.
290 	 * We don't verify it yet, because it may not be complete.
291 	 */
292 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
293 	buf_ops = NULL;
294 
295 	/*
296 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
297 	 * around at all times to optimize access to the superblock. Therefore,
298 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
299 	 * elevated.
300 	 */
301 reread:
302 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
303 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
304 				      buf_ops);
305 	if (error) {
306 		if (loud)
307 			xfs_warn(mp, "SB validate failed with error %d.", error);
308 		/* bad CRC means corrupted metadata */
309 		if (error == -EFSBADCRC)
310 			error = -EFSCORRUPTED;
311 		return error;
312 	}
313 
314 	/*
315 	 * Initialize the mount structure from the superblock.
316 	 */
317 	xfs_sb_from_disk(sbp, bp->b_addr);
318 
319 	/*
320 	 * If we haven't validated the superblock, do so now before we try
321 	 * to check the sector size and reread the superblock appropriately.
322 	 */
323 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
324 		if (loud)
325 			xfs_warn(mp, "Invalid superblock magic number");
326 		error = -EINVAL;
327 		goto release_buf;
328 	}
329 
330 	/*
331 	 * We must be able to do sector-sized and sector-aligned IO.
332 	 */
333 	if (sector_size > sbp->sb_sectsize) {
334 		if (loud)
335 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
336 				sector_size, sbp->sb_sectsize);
337 		error = -ENOSYS;
338 		goto release_buf;
339 	}
340 
341 	if (buf_ops == NULL) {
342 		/*
343 		 * Re-read the superblock so the buffer is correctly sized,
344 		 * and properly verified.
345 		 */
346 		xfs_buf_relse(bp);
347 		sector_size = sbp->sb_sectsize;
348 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
349 		goto reread;
350 	}
351 
352 	xfs_reinit_percpu_counters(mp);
353 
354 	/* no need to be quiet anymore, so reset the buf ops */
355 	bp->b_ops = &xfs_sb_buf_ops;
356 
357 	mp->m_sb_bp = bp;
358 	xfs_buf_unlock(bp);
359 	return 0;
360 
361 release_buf:
362 	xfs_buf_relse(bp);
363 	return error;
364 }
365 
366 /*
367  * If the sunit/swidth change would move the precomputed root inode value, we
368  * must reject the ondisk change because repair will stumble over that.
369  * However, we allow the mount to proceed because we never rejected this
370  * combination before.  Returns true to update the sb, false otherwise.
371  */
372 static inline int
373 xfs_check_new_dalign(
374 	struct xfs_mount	*mp,
375 	int			new_dalign,
376 	bool			*update_sb)
377 {
378 	struct xfs_sb		*sbp = &mp->m_sb;
379 	xfs_ino_t		calc_ino;
380 
381 	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
382 	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
383 
384 	if (sbp->sb_rootino == calc_ino) {
385 		*update_sb = true;
386 		return 0;
387 	}
388 
389 	xfs_warn(mp,
390 "Cannot change stripe alignment; would require moving root inode.");
391 
392 	/*
393 	 * XXX: Next time we add a new incompat feature, this should start
394 	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
395 	 * that we're ignoring the administrator's instructions.
396 	 */
397 	xfs_warn(mp, "Skipping superblock stripe alignment update.");
398 	*update_sb = false;
399 	return 0;
400 }
401 
402 /*
403  * If we were provided with new sunit/swidth values as mount options, make sure
404  * that they pass basic alignment and superblock feature checks, and convert
405  * them into the same units (FSB) that everything else expects.  This step
406  * /must/ be done before computing the inode geometry.
407  */
408 STATIC int
409 xfs_validate_new_dalign(
410 	struct xfs_mount	*mp)
411 {
412 	if (mp->m_dalign == 0)
413 		return 0;
414 
415 	/*
416 	 * If stripe unit and stripe width are not multiples
417 	 * of the fs blocksize turn off alignment.
418 	 */
419 	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
420 	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
421 		xfs_warn(mp,
422 	"alignment check failed: sunit/swidth vs. blocksize(%d)",
423 			mp->m_sb.sb_blocksize);
424 		return -EINVAL;
425 	} else {
426 		/*
427 		 * Convert the stripe unit and width to FSBs.
428 		 */
429 		mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
430 		if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
431 			xfs_warn(mp,
432 		"alignment check failed: sunit/swidth vs. agsize(%d)",
433 				 mp->m_sb.sb_agblocks);
434 			return -EINVAL;
435 		} else if (mp->m_dalign) {
436 			mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
437 		} else {
438 			xfs_warn(mp,
439 		"alignment check failed: sunit(%d) less than bsize(%d)",
440 				 mp->m_dalign, mp->m_sb.sb_blocksize);
441 			return -EINVAL;
442 		}
443 	}
444 
445 	if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
446 		xfs_warn(mp,
447 "cannot change alignment: superblock does not support data alignment");
448 		return -EINVAL;
449 	}
450 
451 	return 0;
452 }
453 
454 /* Update alignment values based on mount options and sb values. */
455 STATIC int
456 xfs_update_alignment(
457 	struct xfs_mount	*mp)
458 {
459 	struct xfs_sb		*sbp = &mp->m_sb;
460 
461 	if (mp->m_dalign) {
462 		bool		update_sb;
463 		int		error;
464 
465 		if (sbp->sb_unit == mp->m_dalign &&
466 		    sbp->sb_width == mp->m_swidth)
467 			return 0;
468 
469 		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
470 		if (error || !update_sb)
471 			return error;
472 
473 		sbp->sb_unit = mp->m_dalign;
474 		sbp->sb_width = mp->m_swidth;
475 		mp->m_update_sb = true;
476 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
477 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
478 		mp->m_dalign = sbp->sb_unit;
479 		mp->m_swidth = sbp->sb_width;
480 	}
481 
482 	return 0;
483 }
484 
485 /*
486  * precalculate the low space thresholds for dynamic speculative preallocation.
487  */
488 void
489 xfs_set_low_space_thresholds(
490 	struct xfs_mount	*mp)
491 {
492 	int i;
493 
494 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
495 		uint64_t space = mp->m_sb.sb_dblocks;
496 
497 		do_div(space, 100);
498 		mp->m_low_space[i] = space * (i + 1);
499 	}
500 }
501 
502 /*
503  * Check that the data (and log if separate) is an ok size.
504  */
505 STATIC int
506 xfs_check_sizes(
507 	struct xfs_mount *mp)
508 {
509 	struct xfs_buf	*bp;
510 	xfs_daddr_t	d;
511 	int		error;
512 
513 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
514 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
515 		xfs_warn(mp, "filesystem size mismatch detected");
516 		return -EFBIG;
517 	}
518 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
519 					d - XFS_FSS_TO_BB(mp, 1),
520 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
521 	if (error) {
522 		xfs_warn(mp, "last sector read failed");
523 		return error;
524 	}
525 	xfs_buf_relse(bp);
526 
527 	if (mp->m_logdev_targp == mp->m_ddev_targp)
528 		return 0;
529 
530 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
531 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
532 		xfs_warn(mp, "log size mismatch detected");
533 		return -EFBIG;
534 	}
535 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
536 					d - XFS_FSB_TO_BB(mp, 1),
537 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
538 	if (error) {
539 		xfs_warn(mp, "log device read failed");
540 		return error;
541 	}
542 	xfs_buf_relse(bp);
543 	return 0;
544 }
545 
546 /*
547  * Clear the quotaflags in memory and in the superblock.
548  */
549 int
550 xfs_mount_reset_sbqflags(
551 	struct xfs_mount	*mp)
552 {
553 	mp->m_qflags = 0;
554 
555 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
556 	if (mp->m_sb.sb_qflags == 0)
557 		return 0;
558 	spin_lock(&mp->m_sb_lock);
559 	mp->m_sb.sb_qflags = 0;
560 	spin_unlock(&mp->m_sb_lock);
561 
562 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
563 		return 0;
564 
565 	return xfs_sync_sb(mp, false);
566 }
567 
568 uint64_t
569 xfs_default_resblks(xfs_mount_t *mp)
570 {
571 	uint64_t resblks;
572 
573 	/*
574 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
575 	 * smaller.  This is intended to cover concurrent allocation
576 	 * transactions when we initially hit enospc. These each require a 4
577 	 * block reservation. Hence by default we cover roughly 2000 concurrent
578 	 * allocation reservations.
579 	 */
580 	resblks = mp->m_sb.sb_dblocks;
581 	do_div(resblks, 20);
582 	resblks = min_t(uint64_t, resblks, 8192);
583 	return resblks;
584 }
585 
586 /* Ensure the summary counts are correct. */
587 STATIC int
588 xfs_check_summary_counts(
589 	struct xfs_mount	*mp)
590 {
591 	/*
592 	 * The AG0 superblock verifier rejects in-progress filesystems,
593 	 * so we should never see the flag set this far into mounting.
594 	 */
595 	if (mp->m_sb.sb_inprogress) {
596 		xfs_err(mp, "sb_inprogress set after log recovery??");
597 		WARN_ON(1);
598 		return -EFSCORRUPTED;
599 	}
600 
601 	/*
602 	 * Now the log is mounted, we know if it was an unclean shutdown or
603 	 * not. If it was, with the first phase of recovery has completed, we
604 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
605 	 * but they are recovered transactionally in the second recovery phase
606 	 * later.
607 	 *
608 	 * If the log was clean when we mounted, we can check the summary
609 	 * counters.  If any of them are obviously incorrect, we can recompute
610 	 * them from the AGF headers in the next step.
611 	 */
612 	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
613 	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
614 	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
615 	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
616 		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
617 
618 	/*
619 	 * We can safely re-initialise incore superblock counters from the
620 	 * per-ag data. These may not be correct if the filesystem was not
621 	 * cleanly unmounted, so we waited for recovery to finish before doing
622 	 * this.
623 	 *
624 	 * If the filesystem was cleanly unmounted or the previous check did
625 	 * not flag anything weird, then we can trust the values in the
626 	 * superblock to be correct and we don't need to do anything here.
627 	 * Otherwise, recalculate the summary counters.
628 	 */
629 	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
630 	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
631 	    !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
632 		return 0;
633 
634 	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
635 }
636 
637 /*
638  * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
639  * internal inode structures can be sitting in the CIL and AIL at this point,
640  * so we need to unpin them, write them back and/or reclaim them before unmount
641  * can proceed.
642  *
643  * An inode cluster that has been freed can have its buffer still pinned in
644  * memory because the transaction is still sitting in a iclog. The stale inodes
645  * on that buffer will be pinned to the buffer until the transaction hits the
646  * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
647  * may never see the pinned buffer, so nothing will push out the iclog and
648  * unpin the buffer.
649  *
650  * Hence we need to force the log to unpin everything first. However, log
651  * forces don't wait for the discards they issue to complete, so we have to
652  * explicitly wait for them to complete here as well.
653  *
654  * Then we can tell the world we are unmounting so that error handling knows
655  * that the filesystem is going away and we should error out anything that we
656  * have been retrying in the background.  This will prevent never-ending
657  * retries in AIL pushing from hanging the unmount.
658  *
659  * Finally, we can push the AIL to clean all the remaining dirty objects, then
660  * reclaim the remaining inodes that are still in memory at this point in time.
661  */
662 static void
663 xfs_unmount_flush_inodes(
664 	struct xfs_mount	*mp)
665 {
666 	xfs_log_force(mp, XFS_LOG_SYNC);
667 	xfs_extent_busy_wait_all(mp);
668 	flush_workqueue(xfs_discard_wq);
669 
670 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
671 
672 	xfs_ail_push_all_sync(mp->m_ail);
673 	cancel_delayed_work_sync(&mp->m_reclaim_work);
674 	xfs_reclaim_inodes(mp);
675 	xfs_health_unmount(mp);
676 }
677 
678 /*
679  * This function does the following on an initial mount of a file system:
680  *	- reads the superblock from disk and init the mount struct
681  *	- if we're a 32-bit kernel, do a size check on the superblock
682  *		so we don't mount terabyte filesystems
683  *	- init mount struct realtime fields
684  *	- allocate inode hash table for fs
685  *	- init directory manager
686  *	- perform recovery and init the log manager
687  */
688 int
689 xfs_mountfs(
690 	struct xfs_mount	*mp)
691 {
692 	struct xfs_sb		*sbp = &(mp->m_sb);
693 	struct xfs_inode	*rip;
694 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
695 	uint64_t		resblks;
696 	uint			quotamount = 0;
697 	uint			quotaflags = 0;
698 	int			error = 0;
699 
700 	xfs_sb_mount_common(mp, sbp);
701 
702 	/*
703 	 * Check for a mismatched features2 values.  Older kernels read & wrote
704 	 * into the wrong sb offset for sb_features2 on some platforms due to
705 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
706 	 * which made older superblock reading/writing routines swap it as a
707 	 * 64-bit value.
708 	 *
709 	 * For backwards compatibility, we make both slots equal.
710 	 *
711 	 * If we detect a mismatched field, we OR the set bits into the existing
712 	 * features2 field in case it has already been modified; we don't want
713 	 * to lose any features.  We then update the bad location with the ORed
714 	 * value so that older kernels will see any features2 flags. The
715 	 * superblock writeback code ensures the new sb_features2 is copied to
716 	 * sb_bad_features2 before it is logged or written to disk.
717 	 */
718 	if (xfs_sb_has_mismatched_features2(sbp)) {
719 		xfs_warn(mp, "correcting sb_features alignment problem");
720 		sbp->sb_features2 |= sbp->sb_bad_features2;
721 		mp->m_update_sb = true;
722 
723 		/*
724 		 * Re-check for ATTR2 in case it was found in bad_features2
725 		 * slot.
726 		 */
727 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
728 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
729 			mp->m_flags |= XFS_MOUNT_ATTR2;
730 	}
731 
732 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
733 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
734 		xfs_sb_version_removeattr2(&mp->m_sb);
735 		mp->m_update_sb = true;
736 
737 		/* update sb_versionnum for the clearing of the morebits */
738 		if (!sbp->sb_features2)
739 			mp->m_update_sb = true;
740 	}
741 
742 	/* always use v2 inodes by default now */
743 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
744 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
745 		mp->m_update_sb = true;
746 	}
747 
748 	/*
749 	 * If we were given new sunit/swidth options, do some basic validation
750 	 * checks and convert the incore dalign and swidth values to the
751 	 * same units (FSB) that everything else uses.  This /must/ happen
752 	 * before computing the inode geometry.
753 	 */
754 	error = xfs_validate_new_dalign(mp);
755 	if (error)
756 		goto out;
757 
758 	xfs_alloc_compute_maxlevels(mp);
759 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
760 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
761 	xfs_ialloc_setup_geometry(mp);
762 	xfs_rmapbt_compute_maxlevels(mp);
763 	xfs_refcountbt_compute_maxlevels(mp);
764 
765 	/*
766 	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
767 	 * is NOT aligned turn off m_dalign since allocator alignment is within
768 	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
769 	 * we must compute the free space and rmap btree geometry before doing
770 	 * this.
771 	 */
772 	error = xfs_update_alignment(mp);
773 	if (error)
774 		goto out;
775 
776 	/* enable fail_at_unmount as default */
777 	mp->m_fail_unmount = true;
778 
779 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
780 			       NULL, mp->m_super->s_id);
781 	if (error)
782 		goto out;
783 
784 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
785 			       &mp->m_kobj, "stats");
786 	if (error)
787 		goto out_remove_sysfs;
788 
789 	error = xfs_error_sysfs_init(mp);
790 	if (error)
791 		goto out_del_stats;
792 
793 	error = xfs_errortag_init(mp);
794 	if (error)
795 		goto out_remove_error_sysfs;
796 
797 	error = xfs_uuid_mount(mp);
798 	if (error)
799 		goto out_remove_errortag;
800 
801 	/*
802 	 * Update the preferred write size based on the information from the
803 	 * on-disk superblock.
804 	 */
805 	mp->m_allocsize_log =
806 		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
807 	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
808 
809 	/* set the low space thresholds for dynamic preallocation */
810 	xfs_set_low_space_thresholds(mp);
811 
812 	/*
813 	 * If enabled, sparse inode chunk alignment is expected to match the
814 	 * cluster size. Full inode chunk alignment must match the chunk size,
815 	 * but that is checked on sb read verification...
816 	 */
817 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
818 	    mp->m_sb.sb_spino_align !=
819 			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
820 		xfs_warn(mp,
821 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
822 			 mp->m_sb.sb_spino_align,
823 			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
824 		error = -EINVAL;
825 		goto out_remove_uuid;
826 	}
827 
828 	/*
829 	 * Check that the data (and log if separate) is an ok size.
830 	 */
831 	error = xfs_check_sizes(mp);
832 	if (error)
833 		goto out_remove_uuid;
834 
835 	/*
836 	 * Initialize realtime fields in the mount structure
837 	 */
838 	error = xfs_rtmount_init(mp);
839 	if (error) {
840 		xfs_warn(mp, "RT mount failed");
841 		goto out_remove_uuid;
842 	}
843 
844 	/*
845 	 *  Copies the low order bits of the timestamp and the randomly
846 	 *  set "sequence" number out of a UUID.
847 	 */
848 	mp->m_fixedfsid[0] =
849 		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
850 		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
851 	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
852 
853 	error = xfs_da_mount(mp);
854 	if (error) {
855 		xfs_warn(mp, "Failed dir/attr init: %d", error);
856 		goto out_remove_uuid;
857 	}
858 
859 	/*
860 	 * Initialize the precomputed transaction reservations values.
861 	 */
862 	xfs_trans_init(mp);
863 
864 	/*
865 	 * Allocate and initialize the per-ag data.
866 	 */
867 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
868 	if (error) {
869 		xfs_warn(mp, "Failed per-ag init: %d", error);
870 		goto out_free_dir;
871 	}
872 
873 	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
874 		xfs_warn(mp, "no log defined");
875 		error = -EFSCORRUPTED;
876 		goto out_free_perag;
877 	}
878 
879 	/*
880 	 * Log's mount-time initialization. The first part of recovery can place
881 	 * some items on the AIL, to be handled when recovery is finished or
882 	 * cancelled.
883 	 */
884 	error = xfs_log_mount(mp, mp->m_logdev_targp,
885 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
886 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
887 	if (error) {
888 		xfs_warn(mp, "log mount failed");
889 		goto out_fail_wait;
890 	}
891 
892 	/* Make sure the summary counts are ok. */
893 	error = xfs_check_summary_counts(mp);
894 	if (error)
895 		goto out_log_dealloc;
896 
897 	/*
898 	 * Get and sanity-check the root inode.
899 	 * Save the pointer to it in the mount structure.
900 	 */
901 	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
902 			 XFS_ILOCK_EXCL, &rip);
903 	if (error) {
904 		xfs_warn(mp,
905 			"Failed to read root inode 0x%llx, error %d",
906 			sbp->sb_rootino, -error);
907 		goto out_log_dealloc;
908 	}
909 
910 	ASSERT(rip != NULL);
911 
912 	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
913 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
914 			(unsigned long long)rip->i_ino);
915 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
916 		error = -EFSCORRUPTED;
917 		goto out_rele_rip;
918 	}
919 	mp->m_rootip = rip;	/* save it */
920 
921 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
922 
923 	/*
924 	 * Initialize realtime inode pointers in the mount structure
925 	 */
926 	error = xfs_rtmount_inodes(mp);
927 	if (error) {
928 		/*
929 		 * Free up the root inode.
930 		 */
931 		xfs_warn(mp, "failed to read RT inodes");
932 		goto out_rele_rip;
933 	}
934 
935 	/*
936 	 * If this is a read-only mount defer the superblock updates until
937 	 * the next remount into writeable mode.  Otherwise we would never
938 	 * perform the update e.g. for the root filesystem.
939 	 */
940 	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
941 		error = xfs_sync_sb(mp, false);
942 		if (error) {
943 			xfs_warn(mp, "failed to write sb changes");
944 			goto out_rtunmount;
945 		}
946 	}
947 
948 	/*
949 	 * Initialise the XFS quota management subsystem for this mount
950 	 */
951 	if (XFS_IS_QUOTA_RUNNING(mp)) {
952 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
953 		if (error)
954 			goto out_rtunmount;
955 	} else {
956 		ASSERT(!XFS_IS_QUOTA_ON(mp));
957 
958 		/*
959 		 * If a file system had quotas running earlier, but decided to
960 		 * mount without -o uquota/pquota/gquota options, revoke the
961 		 * quotachecked license.
962 		 */
963 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
964 			xfs_notice(mp, "resetting quota flags");
965 			error = xfs_mount_reset_sbqflags(mp);
966 			if (error)
967 				goto out_rtunmount;
968 		}
969 	}
970 
971 	/*
972 	 * Finish recovering the file system.  This part needed to be delayed
973 	 * until after the root and real-time bitmap inodes were consistently
974 	 * read in.
975 	 */
976 	error = xfs_log_mount_finish(mp);
977 	if (error) {
978 		xfs_warn(mp, "log mount finish failed");
979 		goto out_rtunmount;
980 	}
981 
982 	/*
983 	 * Now the log is fully replayed, we can transition to full read-only
984 	 * mode for read-only mounts. This will sync all the metadata and clean
985 	 * the log so that the recovery we just performed does not have to be
986 	 * replayed again on the next mount.
987 	 *
988 	 * We use the same quiesce mechanism as the rw->ro remount, as they are
989 	 * semantically identical operations.
990 	 */
991 	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
992 							XFS_MOUNT_RDONLY) {
993 		xfs_log_clean(mp);
994 	}
995 
996 	/*
997 	 * Complete the quota initialisation, post-log-replay component.
998 	 */
999 	if (quotamount) {
1000 		ASSERT(mp->m_qflags == 0);
1001 		mp->m_qflags = quotaflags;
1002 
1003 		xfs_qm_mount_quotas(mp);
1004 	}
1005 
1006 	/*
1007 	 * Now we are mounted, reserve a small amount of unused space for
1008 	 * privileged transactions. This is needed so that transaction
1009 	 * space required for critical operations can dip into this pool
1010 	 * when at ENOSPC. This is needed for operations like create with
1011 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1012 	 * are not allowed to use this reserved space.
1013 	 *
1014 	 * This may drive us straight to ENOSPC on mount, but that implies
1015 	 * we were already there on the last unmount. Warn if this occurs.
1016 	 */
1017 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1018 		resblks = xfs_default_resblks(mp);
1019 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1020 		if (error)
1021 			xfs_warn(mp,
1022 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
1023 
1024 		/* Recover any CoW blocks that never got remapped. */
1025 		error = xfs_reflink_recover_cow(mp);
1026 		if (error) {
1027 			xfs_err(mp,
1028 	"Error %d recovering leftover CoW allocations.", error);
1029 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1030 			goto out_quota;
1031 		}
1032 
1033 		/* Reserve AG blocks for future btree expansion. */
1034 		error = xfs_fs_reserve_ag_blocks(mp);
1035 		if (error && error != -ENOSPC)
1036 			goto out_agresv;
1037 	}
1038 
1039 	return 0;
1040 
1041  out_agresv:
1042 	xfs_fs_unreserve_ag_blocks(mp);
1043  out_quota:
1044 	xfs_qm_unmount_quotas(mp);
1045  out_rtunmount:
1046 	xfs_rtunmount_inodes(mp);
1047  out_rele_rip:
1048 	xfs_irele(rip);
1049 	/* Clean out dquots that might be in memory after quotacheck. */
1050 	xfs_qm_unmount(mp);
1051 	/*
1052 	 * Flush all inode reclamation work and flush the log.
1053 	 * We have to do this /after/ rtunmount and qm_unmount because those
1054 	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1055 	 *
1056 	 * This is slightly different from the unmountfs call sequence
1057 	 * because we could be tearing down a partially set up mount.  In
1058 	 * particular, if log_mount_finish fails we bail out without calling
1059 	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1060 	 * quota inodes.
1061 	 */
1062 	xfs_unmount_flush_inodes(mp);
1063  out_log_dealloc:
1064 	xfs_log_mount_cancel(mp);
1065  out_fail_wait:
1066 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1067 		xfs_buftarg_drain(mp->m_logdev_targp);
1068 	xfs_buftarg_drain(mp->m_ddev_targp);
1069  out_free_perag:
1070 	xfs_free_perag(mp);
1071  out_free_dir:
1072 	xfs_da_unmount(mp);
1073  out_remove_uuid:
1074 	xfs_uuid_unmount(mp);
1075  out_remove_errortag:
1076 	xfs_errortag_del(mp);
1077  out_remove_error_sysfs:
1078 	xfs_error_sysfs_del(mp);
1079  out_del_stats:
1080 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1081  out_remove_sysfs:
1082 	xfs_sysfs_del(&mp->m_kobj);
1083  out:
1084 	return error;
1085 }
1086 
1087 /*
1088  * This flushes out the inodes,dquots and the superblock, unmounts the
1089  * log and makes sure that incore structures are freed.
1090  */
1091 void
1092 xfs_unmountfs(
1093 	struct xfs_mount	*mp)
1094 {
1095 	uint64_t		resblks;
1096 	int			error;
1097 
1098 	xfs_blockgc_stop(mp);
1099 	xfs_fs_unreserve_ag_blocks(mp);
1100 	xfs_qm_unmount_quotas(mp);
1101 	xfs_rtunmount_inodes(mp);
1102 	xfs_irele(mp->m_rootip);
1103 
1104 	xfs_unmount_flush_inodes(mp);
1105 
1106 	xfs_qm_unmount(mp);
1107 
1108 	/*
1109 	 * Unreserve any blocks we have so that when we unmount we don't account
1110 	 * the reserved free space as used. This is really only necessary for
1111 	 * lazy superblock counting because it trusts the incore superblock
1112 	 * counters to be absolutely correct on clean unmount.
1113 	 *
1114 	 * We don't bother correcting this elsewhere for lazy superblock
1115 	 * counting because on mount of an unclean filesystem we reconstruct the
1116 	 * correct counter value and this is irrelevant.
1117 	 *
1118 	 * For non-lazy counter filesystems, this doesn't matter at all because
1119 	 * we only every apply deltas to the superblock and hence the incore
1120 	 * value does not matter....
1121 	 */
1122 	resblks = 0;
1123 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1124 	if (error)
1125 		xfs_warn(mp, "Unable to free reserved block pool. "
1126 				"Freespace may not be correct on next mount.");
1127 
1128 	xfs_log_unmount(mp);
1129 	xfs_da_unmount(mp);
1130 	xfs_uuid_unmount(mp);
1131 
1132 #if defined(DEBUG)
1133 	xfs_errortag_clearall(mp);
1134 #endif
1135 	xfs_free_perag(mp);
1136 
1137 	xfs_errortag_del(mp);
1138 	xfs_error_sysfs_del(mp);
1139 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1140 	xfs_sysfs_del(&mp->m_kobj);
1141 }
1142 
1143 /*
1144  * Determine whether modifications can proceed. The caller specifies the minimum
1145  * freeze level for which modifications should not be allowed. This allows
1146  * certain operations to proceed while the freeze sequence is in progress, if
1147  * necessary.
1148  */
1149 bool
1150 xfs_fs_writable(
1151 	struct xfs_mount	*mp,
1152 	int			level)
1153 {
1154 	ASSERT(level > SB_UNFROZEN);
1155 	if ((mp->m_super->s_writers.frozen >= level) ||
1156 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1157 		return false;
1158 
1159 	return true;
1160 }
1161 
1162 /*
1163  * Deltas for the block count can vary from 1 to very large, but lock contention
1164  * only occurs on frequent small block count updates such as in the delayed
1165  * allocation path for buffered writes (page a time updates). Hence we set
1166  * a large batch count (1024) to minimise global counter updates except when
1167  * we get near to ENOSPC and we have to be very accurate with our updates.
1168  */
1169 #define XFS_FDBLOCKS_BATCH	1024
1170 int
1171 xfs_mod_fdblocks(
1172 	struct xfs_mount	*mp,
1173 	int64_t			delta,
1174 	bool			rsvd)
1175 {
1176 	int64_t			lcounter;
1177 	long long		res_used;
1178 	s32			batch;
1179 
1180 	if (delta > 0) {
1181 		/*
1182 		 * If the reserve pool is depleted, put blocks back into it
1183 		 * first. Most of the time the pool is full.
1184 		 */
1185 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1186 			percpu_counter_add(&mp->m_fdblocks, delta);
1187 			return 0;
1188 		}
1189 
1190 		spin_lock(&mp->m_sb_lock);
1191 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1192 
1193 		if (res_used > delta) {
1194 			mp->m_resblks_avail += delta;
1195 		} else {
1196 			delta -= res_used;
1197 			mp->m_resblks_avail = mp->m_resblks;
1198 			percpu_counter_add(&mp->m_fdblocks, delta);
1199 		}
1200 		spin_unlock(&mp->m_sb_lock);
1201 		return 0;
1202 	}
1203 
1204 	/*
1205 	 * Taking blocks away, need to be more accurate the closer we
1206 	 * are to zero.
1207 	 *
1208 	 * If the counter has a value of less than 2 * max batch size,
1209 	 * then make everything serialise as we are real close to
1210 	 * ENOSPC.
1211 	 */
1212 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1213 				     XFS_FDBLOCKS_BATCH) < 0)
1214 		batch = 1;
1215 	else
1216 		batch = XFS_FDBLOCKS_BATCH;
1217 
1218 	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1219 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1220 				     XFS_FDBLOCKS_BATCH) >= 0) {
1221 		/* we had space! */
1222 		return 0;
1223 	}
1224 
1225 	/*
1226 	 * lock up the sb for dipping into reserves before releasing the space
1227 	 * that took us to ENOSPC.
1228 	 */
1229 	spin_lock(&mp->m_sb_lock);
1230 	percpu_counter_add(&mp->m_fdblocks, -delta);
1231 	if (!rsvd)
1232 		goto fdblocks_enospc;
1233 
1234 	lcounter = (long long)mp->m_resblks_avail + delta;
1235 	if (lcounter >= 0) {
1236 		mp->m_resblks_avail = lcounter;
1237 		spin_unlock(&mp->m_sb_lock);
1238 		return 0;
1239 	}
1240 	xfs_warn_once(mp,
1241 "Reserve blocks depleted! Consider increasing reserve pool size.");
1242 
1243 fdblocks_enospc:
1244 	spin_unlock(&mp->m_sb_lock);
1245 	return -ENOSPC;
1246 }
1247 
1248 int
1249 xfs_mod_frextents(
1250 	struct xfs_mount	*mp,
1251 	int64_t			delta)
1252 {
1253 	int64_t			lcounter;
1254 	int			ret = 0;
1255 
1256 	spin_lock(&mp->m_sb_lock);
1257 	lcounter = mp->m_sb.sb_frextents + delta;
1258 	if (lcounter < 0)
1259 		ret = -ENOSPC;
1260 	else
1261 		mp->m_sb.sb_frextents = lcounter;
1262 	spin_unlock(&mp->m_sb_lock);
1263 	return ret;
1264 }
1265 
1266 /*
1267  * Used to free the superblock along various error paths.
1268  */
1269 void
1270 xfs_freesb(
1271 	struct xfs_mount	*mp)
1272 {
1273 	struct xfs_buf		*bp = mp->m_sb_bp;
1274 
1275 	xfs_buf_lock(bp);
1276 	mp->m_sb_bp = NULL;
1277 	xfs_buf_relse(bp);
1278 }
1279 
1280 /*
1281  * If the underlying (data/log/rt) device is readonly, there are some
1282  * operations that cannot proceed.
1283  */
1284 int
1285 xfs_dev_is_read_only(
1286 	struct xfs_mount	*mp,
1287 	char			*message)
1288 {
1289 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1290 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1291 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1292 		xfs_notice(mp, "%s required on read-only device.", message);
1293 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1294 		return -EROFS;
1295 	}
1296 	return 0;
1297 }
1298 
1299 /* Force the summary counters to be recalculated at next mount. */
1300 void
1301 xfs_force_summary_recalc(
1302 	struct xfs_mount	*mp)
1303 {
1304 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1305 		return;
1306 
1307 	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1308 }
1309 
1310 /*
1311  * Update the in-core delayed block counter.
1312  *
1313  * We prefer to update the counter without having to take a spinlock for every
1314  * counter update (i.e. batching).  Each change to delayed allocation
1315  * reservations can change can easily exceed the default percpu counter
1316  * batching, so we use a larger batch factor here.
1317  *
1318  * Note that we don't currently have any callers requiring fast summation
1319  * (e.g. percpu_counter_read) so we can use a big batch value here.
1320  */
1321 #define XFS_DELALLOC_BATCH	(4096)
1322 void
1323 xfs_mod_delalloc(
1324 	struct xfs_mount	*mp,
1325 	int64_t			delta)
1326 {
1327 	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1328 			XFS_DELALLOC_BATCH);
1329 }
1330