xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision ae3473231e77a3f1909d48cd144cebe5e1d049b3)
1 /*
2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_bit.h"
25 #include "xfs_sb.h"
26 #include "xfs_mount.h"
27 #include "xfs_defer.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_dir2.h"
32 #include "xfs_ialloc.h"
33 #include "xfs_alloc.h"
34 #include "xfs_rtalloc.h"
35 #include "xfs_bmap.h"
36 #include "xfs_trans.h"
37 #include "xfs_trans_priv.h"
38 #include "xfs_log.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_fsops.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
44 #include "xfs_sysfs.h"
45 #include "xfs_rmap_btree.h"
46 #include "xfs_refcount_btree.h"
47 #include "xfs_reflink.h"
48 
49 
50 static DEFINE_MUTEX(xfs_uuid_table_mutex);
51 static int xfs_uuid_table_size;
52 static uuid_t *xfs_uuid_table;
53 
54 void
55 xfs_uuid_table_free(void)
56 {
57 	if (xfs_uuid_table_size == 0)
58 		return;
59 	kmem_free(xfs_uuid_table);
60 	xfs_uuid_table = NULL;
61 	xfs_uuid_table_size = 0;
62 }
63 
64 /*
65  * See if the UUID is unique among mounted XFS filesystems.
66  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
67  */
68 STATIC int
69 xfs_uuid_mount(
70 	struct xfs_mount	*mp)
71 {
72 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
73 	int			hole, i;
74 
75 	if (mp->m_flags & XFS_MOUNT_NOUUID)
76 		return 0;
77 
78 	if (uuid_is_nil(uuid)) {
79 		xfs_warn(mp, "Filesystem has nil UUID - can't mount");
80 		return -EINVAL;
81 	}
82 
83 	mutex_lock(&xfs_uuid_table_mutex);
84 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
85 		if (uuid_is_nil(&xfs_uuid_table[i])) {
86 			hole = i;
87 			continue;
88 		}
89 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
90 			goto out_duplicate;
91 	}
92 
93 	if (hole < 0) {
94 		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
95 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
96 			KM_SLEEP);
97 		hole = xfs_uuid_table_size++;
98 	}
99 	xfs_uuid_table[hole] = *uuid;
100 	mutex_unlock(&xfs_uuid_table_mutex);
101 
102 	return 0;
103 
104  out_duplicate:
105 	mutex_unlock(&xfs_uuid_table_mutex);
106 	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
107 	return -EINVAL;
108 }
109 
110 STATIC void
111 xfs_uuid_unmount(
112 	struct xfs_mount	*mp)
113 {
114 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
115 	int			i;
116 
117 	if (mp->m_flags & XFS_MOUNT_NOUUID)
118 		return;
119 
120 	mutex_lock(&xfs_uuid_table_mutex);
121 	for (i = 0; i < xfs_uuid_table_size; i++) {
122 		if (uuid_is_nil(&xfs_uuid_table[i]))
123 			continue;
124 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
125 			continue;
126 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
127 		break;
128 	}
129 	ASSERT(i < xfs_uuid_table_size);
130 	mutex_unlock(&xfs_uuid_table_mutex);
131 }
132 
133 
134 STATIC void
135 __xfs_free_perag(
136 	struct rcu_head	*head)
137 {
138 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
139 
140 	ASSERT(atomic_read(&pag->pag_ref) == 0);
141 	kmem_free(pag);
142 }
143 
144 /*
145  * Free up the per-ag resources associated with the mount structure.
146  */
147 STATIC void
148 xfs_free_perag(
149 	xfs_mount_t	*mp)
150 {
151 	xfs_agnumber_t	agno;
152 	struct xfs_perag *pag;
153 
154 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
155 		spin_lock(&mp->m_perag_lock);
156 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
157 		spin_unlock(&mp->m_perag_lock);
158 		ASSERT(pag);
159 		ASSERT(atomic_read(&pag->pag_ref) == 0);
160 		xfs_buf_hash_destroy(pag);
161 		call_rcu(&pag->rcu_head, __xfs_free_perag);
162 	}
163 }
164 
165 /*
166  * Check size of device based on the (data/realtime) block count.
167  * Note: this check is used by the growfs code as well as mount.
168  */
169 int
170 xfs_sb_validate_fsb_count(
171 	xfs_sb_t	*sbp,
172 	__uint64_t	nblocks)
173 {
174 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
175 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
176 
177 	/* Limited by ULONG_MAX of page cache index */
178 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
179 		return -EFBIG;
180 	return 0;
181 }
182 
183 int
184 xfs_initialize_perag(
185 	xfs_mount_t	*mp,
186 	xfs_agnumber_t	agcount,
187 	xfs_agnumber_t	*maxagi)
188 {
189 	xfs_agnumber_t	index;
190 	xfs_agnumber_t	first_initialised = 0;
191 	xfs_perag_t	*pag;
192 	int		error = -ENOMEM;
193 
194 	/*
195 	 * Walk the current per-ag tree so we don't try to initialise AGs
196 	 * that already exist (growfs case). Allocate and insert all the
197 	 * AGs we don't find ready for initialisation.
198 	 */
199 	for (index = 0; index < agcount; index++) {
200 		pag = xfs_perag_get(mp, index);
201 		if (pag) {
202 			xfs_perag_put(pag);
203 			continue;
204 		}
205 		if (!first_initialised)
206 			first_initialised = index;
207 
208 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
209 		if (!pag)
210 			goto out_unwind;
211 		pag->pag_agno = index;
212 		pag->pag_mount = mp;
213 		spin_lock_init(&pag->pag_ici_lock);
214 		mutex_init(&pag->pag_ici_reclaim_lock);
215 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
216 		if (xfs_buf_hash_init(pag))
217 			goto out_unwind;
218 
219 		if (radix_tree_preload(GFP_NOFS))
220 			goto out_unwind;
221 
222 		spin_lock(&mp->m_perag_lock);
223 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
224 			BUG();
225 			spin_unlock(&mp->m_perag_lock);
226 			radix_tree_preload_end();
227 			error = -EEXIST;
228 			goto out_unwind;
229 		}
230 		spin_unlock(&mp->m_perag_lock);
231 		radix_tree_preload_end();
232 	}
233 
234 	index = xfs_set_inode_alloc(mp, agcount);
235 
236 	if (maxagi)
237 		*maxagi = index;
238 
239 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
240 	return 0;
241 
242 out_unwind:
243 	xfs_buf_hash_destroy(pag);
244 	kmem_free(pag);
245 	for (; index > first_initialised; index--) {
246 		pag = radix_tree_delete(&mp->m_perag_tree, index);
247 		xfs_buf_hash_destroy(pag);
248 		kmem_free(pag);
249 	}
250 	return error;
251 }
252 
253 /*
254  * xfs_readsb
255  *
256  * Does the initial read of the superblock.
257  */
258 int
259 xfs_readsb(
260 	struct xfs_mount *mp,
261 	int		flags)
262 {
263 	unsigned int	sector_size;
264 	struct xfs_buf	*bp;
265 	struct xfs_sb	*sbp = &mp->m_sb;
266 	int		error;
267 	int		loud = !(flags & XFS_MFSI_QUIET);
268 	const struct xfs_buf_ops *buf_ops;
269 
270 	ASSERT(mp->m_sb_bp == NULL);
271 	ASSERT(mp->m_ddev_targp != NULL);
272 
273 	/*
274 	 * For the initial read, we must guess at the sector
275 	 * size based on the block device.  It's enough to
276 	 * get the sb_sectsize out of the superblock and
277 	 * then reread with the proper length.
278 	 * We don't verify it yet, because it may not be complete.
279 	 */
280 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
281 	buf_ops = NULL;
282 
283 	/*
284 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
285 	 * around at all times to optimize access to the superblock. Therefore,
286 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
287 	 * elevated.
288 	 */
289 reread:
290 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
291 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
292 				      buf_ops);
293 	if (error) {
294 		if (loud)
295 			xfs_warn(mp, "SB validate failed with error %d.", error);
296 		/* bad CRC means corrupted metadata */
297 		if (error == -EFSBADCRC)
298 			error = -EFSCORRUPTED;
299 		return error;
300 	}
301 
302 	/*
303 	 * Initialize the mount structure from the superblock.
304 	 */
305 	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
306 
307 	/*
308 	 * If we haven't validated the superblock, do so now before we try
309 	 * to check the sector size and reread the superblock appropriately.
310 	 */
311 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
312 		if (loud)
313 			xfs_warn(mp, "Invalid superblock magic number");
314 		error = -EINVAL;
315 		goto release_buf;
316 	}
317 
318 	/*
319 	 * We must be able to do sector-sized and sector-aligned IO.
320 	 */
321 	if (sector_size > sbp->sb_sectsize) {
322 		if (loud)
323 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
324 				sector_size, sbp->sb_sectsize);
325 		error = -ENOSYS;
326 		goto release_buf;
327 	}
328 
329 	if (buf_ops == NULL) {
330 		/*
331 		 * Re-read the superblock so the buffer is correctly sized,
332 		 * and properly verified.
333 		 */
334 		xfs_buf_relse(bp);
335 		sector_size = sbp->sb_sectsize;
336 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
337 		goto reread;
338 	}
339 
340 	xfs_reinit_percpu_counters(mp);
341 
342 	/* no need to be quiet anymore, so reset the buf ops */
343 	bp->b_ops = &xfs_sb_buf_ops;
344 
345 	mp->m_sb_bp = bp;
346 	xfs_buf_unlock(bp);
347 	return 0;
348 
349 release_buf:
350 	xfs_buf_relse(bp);
351 	return error;
352 }
353 
354 /*
355  * Update alignment values based on mount options and sb values
356  */
357 STATIC int
358 xfs_update_alignment(xfs_mount_t *mp)
359 {
360 	xfs_sb_t	*sbp = &(mp->m_sb);
361 
362 	if (mp->m_dalign) {
363 		/*
364 		 * If stripe unit and stripe width are not multiples
365 		 * of the fs blocksize turn off alignment.
366 		 */
367 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
368 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
369 			xfs_warn(mp,
370 		"alignment check failed: sunit/swidth vs. blocksize(%d)",
371 				sbp->sb_blocksize);
372 			return -EINVAL;
373 		} else {
374 			/*
375 			 * Convert the stripe unit and width to FSBs.
376 			 */
377 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
378 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
379 				xfs_warn(mp,
380 			"alignment check failed: sunit/swidth vs. agsize(%d)",
381 					 sbp->sb_agblocks);
382 				return -EINVAL;
383 			} else if (mp->m_dalign) {
384 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
385 			} else {
386 				xfs_warn(mp,
387 			"alignment check failed: sunit(%d) less than bsize(%d)",
388 					 mp->m_dalign, sbp->sb_blocksize);
389 				return -EINVAL;
390 			}
391 		}
392 
393 		/*
394 		 * Update superblock with new values
395 		 * and log changes
396 		 */
397 		if (xfs_sb_version_hasdalign(sbp)) {
398 			if (sbp->sb_unit != mp->m_dalign) {
399 				sbp->sb_unit = mp->m_dalign;
400 				mp->m_update_sb = true;
401 			}
402 			if (sbp->sb_width != mp->m_swidth) {
403 				sbp->sb_width = mp->m_swidth;
404 				mp->m_update_sb = true;
405 			}
406 		} else {
407 			xfs_warn(mp,
408 	"cannot change alignment: superblock does not support data alignment");
409 			return -EINVAL;
410 		}
411 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
412 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
413 			mp->m_dalign = sbp->sb_unit;
414 			mp->m_swidth = sbp->sb_width;
415 	}
416 
417 	return 0;
418 }
419 
420 /*
421  * Set the maximum inode count for this filesystem
422  */
423 STATIC void
424 xfs_set_maxicount(xfs_mount_t *mp)
425 {
426 	xfs_sb_t	*sbp = &(mp->m_sb);
427 	__uint64_t	icount;
428 
429 	if (sbp->sb_imax_pct) {
430 		/*
431 		 * Make sure the maximum inode count is a multiple
432 		 * of the units we allocate inodes in.
433 		 */
434 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
435 		do_div(icount, 100);
436 		do_div(icount, mp->m_ialloc_blks);
437 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
438 				   sbp->sb_inopblog;
439 	} else {
440 		mp->m_maxicount = 0;
441 	}
442 }
443 
444 /*
445  * Set the default minimum read and write sizes unless
446  * already specified in a mount option.
447  * We use smaller I/O sizes when the file system
448  * is being used for NFS service (wsync mount option).
449  */
450 STATIC void
451 xfs_set_rw_sizes(xfs_mount_t *mp)
452 {
453 	xfs_sb_t	*sbp = &(mp->m_sb);
454 	int		readio_log, writeio_log;
455 
456 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
457 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
458 			readio_log = XFS_WSYNC_READIO_LOG;
459 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
460 		} else {
461 			readio_log = XFS_READIO_LOG_LARGE;
462 			writeio_log = XFS_WRITEIO_LOG_LARGE;
463 		}
464 	} else {
465 		readio_log = mp->m_readio_log;
466 		writeio_log = mp->m_writeio_log;
467 	}
468 
469 	if (sbp->sb_blocklog > readio_log) {
470 		mp->m_readio_log = sbp->sb_blocklog;
471 	} else {
472 		mp->m_readio_log = readio_log;
473 	}
474 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
475 	if (sbp->sb_blocklog > writeio_log) {
476 		mp->m_writeio_log = sbp->sb_blocklog;
477 	} else {
478 		mp->m_writeio_log = writeio_log;
479 	}
480 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
481 }
482 
483 /*
484  * precalculate the low space thresholds for dynamic speculative preallocation.
485  */
486 void
487 xfs_set_low_space_thresholds(
488 	struct xfs_mount	*mp)
489 {
490 	int i;
491 
492 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
493 		__uint64_t space = mp->m_sb.sb_dblocks;
494 
495 		do_div(space, 100);
496 		mp->m_low_space[i] = space * (i + 1);
497 	}
498 }
499 
500 
501 /*
502  * Set whether we're using inode alignment.
503  */
504 STATIC void
505 xfs_set_inoalignment(xfs_mount_t *mp)
506 {
507 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
508 	    mp->m_sb.sb_inoalignmt >=
509 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
510 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
511 	else
512 		mp->m_inoalign_mask = 0;
513 	/*
514 	 * If we are using stripe alignment, check whether
515 	 * the stripe unit is a multiple of the inode alignment
516 	 */
517 	if (mp->m_dalign && mp->m_inoalign_mask &&
518 	    !(mp->m_dalign & mp->m_inoalign_mask))
519 		mp->m_sinoalign = mp->m_dalign;
520 	else
521 		mp->m_sinoalign = 0;
522 }
523 
524 /*
525  * Check that the data (and log if separate) is an ok size.
526  */
527 STATIC int
528 xfs_check_sizes(
529 	struct xfs_mount *mp)
530 {
531 	struct xfs_buf	*bp;
532 	xfs_daddr_t	d;
533 	int		error;
534 
535 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
536 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
537 		xfs_warn(mp, "filesystem size mismatch detected");
538 		return -EFBIG;
539 	}
540 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
541 					d - XFS_FSS_TO_BB(mp, 1),
542 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
543 	if (error) {
544 		xfs_warn(mp, "last sector read failed");
545 		return error;
546 	}
547 	xfs_buf_relse(bp);
548 
549 	if (mp->m_logdev_targp == mp->m_ddev_targp)
550 		return 0;
551 
552 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
553 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
554 		xfs_warn(mp, "log size mismatch detected");
555 		return -EFBIG;
556 	}
557 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
558 					d - XFS_FSB_TO_BB(mp, 1),
559 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
560 	if (error) {
561 		xfs_warn(mp, "log device read failed");
562 		return error;
563 	}
564 	xfs_buf_relse(bp);
565 	return 0;
566 }
567 
568 /*
569  * Clear the quotaflags in memory and in the superblock.
570  */
571 int
572 xfs_mount_reset_sbqflags(
573 	struct xfs_mount	*mp)
574 {
575 	mp->m_qflags = 0;
576 
577 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
578 	if (mp->m_sb.sb_qflags == 0)
579 		return 0;
580 	spin_lock(&mp->m_sb_lock);
581 	mp->m_sb.sb_qflags = 0;
582 	spin_unlock(&mp->m_sb_lock);
583 
584 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
585 		return 0;
586 
587 	return xfs_sync_sb(mp, false);
588 }
589 
590 __uint64_t
591 xfs_default_resblks(xfs_mount_t *mp)
592 {
593 	__uint64_t resblks;
594 
595 	/*
596 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
597 	 * smaller.  This is intended to cover concurrent allocation
598 	 * transactions when we initially hit enospc. These each require a 4
599 	 * block reservation. Hence by default we cover roughly 2000 concurrent
600 	 * allocation reservations.
601 	 */
602 	resblks = mp->m_sb.sb_dblocks;
603 	do_div(resblks, 20);
604 	resblks = min_t(__uint64_t, resblks, 8192);
605 	return resblks;
606 }
607 
608 /*
609  * This function does the following on an initial mount of a file system:
610  *	- reads the superblock from disk and init the mount struct
611  *	- if we're a 32-bit kernel, do a size check on the superblock
612  *		so we don't mount terabyte filesystems
613  *	- init mount struct realtime fields
614  *	- allocate inode hash table for fs
615  *	- init directory manager
616  *	- perform recovery and init the log manager
617  */
618 int
619 xfs_mountfs(
620 	struct xfs_mount	*mp)
621 {
622 	struct xfs_sb		*sbp = &(mp->m_sb);
623 	struct xfs_inode	*rip;
624 	__uint64_t		resblks;
625 	uint			quotamount = 0;
626 	uint			quotaflags = 0;
627 	int			error = 0;
628 
629 	xfs_sb_mount_common(mp, sbp);
630 
631 	/*
632 	 * Check for a mismatched features2 values.  Older kernels read & wrote
633 	 * into the wrong sb offset for sb_features2 on some platforms due to
634 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
635 	 * which made older superblock reading/writing routines swap it as a
636 	 * 64-bit value.
637 	 *
638 	 * For backwards compatibility, we make both slots equal.
639 	 *
640 	 * If we detect a mismatched field, we OR the set bits into the existing
641 	 * features2 field in case it has already been modified; we don't want
642 	 * to lose any features.  We then update the bad location with the ORed
643 	 * value so that older kernels will see any features2 flags. The
644 	 * superblock writeback code ensures the new sb_features2 is copied to
645 	 * sb_bad_features2 before it is logged or written to disk.
646 	 */
647 	if (xfs_sb_has_mismatched_features2(sbp)) {
648 		xfs_warn(mp, "correcting sb_features alignment problem");
649 		sbp->sb_features2 |= sbp->sb_bad_features2;
650 		mp->m_update_sb = true;
651 
652 		/*
653 		 * Re-check for ATTR2 in case it was found in bad_features2
654 		 * slot.
655 		 */
656 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
657 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
658 			mp->m_flags |= XFS_MOUNT_ATTR2;
659 	}
660 
661 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
662 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
663 		xfs_sb_version_removeattr2(&mp->m_sb);
664 		mp->m_update_sb = true;
665 
666 		/* update sb_versionnum for the clearing of the morebits */
667 		if (!sbp->sb_features2)
668 			mp->m_update_sb = true;
669 	}
670 
671 	/* always use v2 inodes by default now */
672 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
673 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
674 		mp->m_update_sb = true;
675 	}
676 
677 	/*
678 	 * Check if sb_agblocks is aligned at stripe boundary
679 	 * If sb_agblocks is NOT aligned turn off m_dalign since
680 	 * allocator alignment is within an ag, therefore ag has
681 	 * to be aligned at stripe boundary.
682 	 */
683 	error = xfs_update_alignment(mp);
684 	if (error)
685 		goto out;
686 
687 	xfs_alloc_compute_maxlevels(mp);
688 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
689 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
690 	xfs_ialloc_compute_maxlevels(mp);
691 	xfs_rmapbt_compute_maxlevels(mp);
692 	xfs_refcountbt_compute_maxlevels(mp);
693 
694 	xfs_set_maxicount(mp);
695 
696 	/* enable fail_at_unmount as default */
697 	mp->m_fail_unmount = 1;
698 
699 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
700 	if (error)
701 		goto out;
702 
703 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
704 			       &mp->m_kobj, "stats");
705 	if (error)
706 		goto out_remove_sysfs;
707 
708 	error = xfs_error_sysfs_init(mp);
709 	if (error)
710 		goto out_del_stats;
711 
712 
713 	error = xfs_uuid_mount(mp);
714 	if (error)
715 		goto out_remove_error_sysfs;
716 
717 	/*
718 	 * Set the minimum read and write sizes
719 	 */
720 	xfs_set_rw_sizes(mp);
721 
722 	/* set the low space thresholds for dynamic preallocation */
723 	xfs_set_low_space_thresholds(mp);
724 
725 	/*
726 	 * Set the inode cluster size.
727 	 * This may still be overridden by the file system
728 	 * block size if it is larger than the chosen cluster size.
729 	 *
730 	 * For v5 filesystems, scale the cluster size with the inode size to
731 	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
732 	 * has set the inode alignment value appropriately for larger cluster
733 	 * sizes.
734 	 */
735 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
736 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
737 		int	new_size = mp->m_inode_cluster_size;
738 
739 		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
740 		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
741 			mp->m_inode_cluster_size = new_size;
742 	}
743 
744 	/*
745 	 * If enabled, sparse inode chunk alignment is expected to match the
746 	 * cluster size. Full inode chunk alignment must match the chunk size,
747 	 * but that is checked on sb read verification...
748 	 */
749 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
750 	    mp->m_sb.sb_spino_align !=
751 			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
752 		xfs_warn(mp,
753 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
754 			 mp->m_sb.sb_spino_align,
755 			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
756 		error = -EINVAL;
757 		goto out_remove_uuid;
758 	}
759 
760 	/*
761 	 * Set inode alignment fields
762 	 */
763 	xfs_set_inoalignment(mp);
764 
765 	/*
766 	 * Check that the data (and log if separate) is an ok size.
767 	 */
768 	error = xfs_check_sizes(mp);
769 	if (error)
770 		goto out_remove_uuid;
771 
772 	/*
773 	 * Initialize realtime fields in the mount structure
774 	 */
775 	error = xfs_rtmount_init(mp);
776 	if (error) {
777 		xfs_warn(mp, "RT mount failed");
778 		goto out_remove_uuid;
779 	}
780 
781 	/*
782 	 *  Copies the low order bits of the timestamp and the randomly
783 	 *  set "sequence" number out of a UUID.
784 	 */
785 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
786 
787 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
788 
789 	error = xfs_da_mount(mp);
790 	if (error) {
791 		xfs_warn(mp, "Failed dir/attr init: %d", error);
792 		goto out_remove_uuid;
793 	}
794 
795 	/*
796 	 * Initialize the precomputed transaction reservations values.
797 	 */
798 	xfs_trans_init(mp);
799 
800 	/*
801 	 * Allocate and initialize the per-ag data.
802 	 */
803 	spin_lock_init(&mp->m_perag_lock);
804 	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
805 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
806 	if (error) {
807 		xfs_warn(mp, "Failed per-ag init: %d", error);
808 		goto out_free_dir;
809 	}
810 
811 	if (!sbp->sb_logblocks) {
812 		xfs_warn(mp, "no log defined");
813 		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
814 		error = -EFSCORRUPTED;
815 		goto out_free_perag;
816 	}
817 
818 	/*
819 	 * Log's mount-time initialization. The first part of recovery can place
820 	 * some items on the AIL, to be handled when recovery is finished or
821 	 * cancelled.
822 	 */
823 	error = xfs_log_mount(mp, mp->m_logdev_targp,
824 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
825 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
826 	if (error) {
827 		xfs_warn(mp, "log mount failed");
828 		goto out_fail_wait;
829 	}
830 
831 	/*
832 	 * Now the log is mounted, we know if it was an unclean shutdown or
833 	 * not. If it was, with the first phase of recovery has completed, we
834 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
835 	 * but they are recovered transactionally in the second recovery phase
836 	 * later.
837 	 *
838 	 * Hence we can safely re-initialise incore superblock counters from
839 	 * the per-ag data. These may not be correct if the filesystem was not
840 	 * cleanly unmounted, so we need to wait for recovery to finish before
841 	 * doing this.
842 	 *
843 	 * If the filesystem was cleanly unmounted, then we can trust the
844 	 * values in the superblock to be correct and we don't need to do
845 	 * anything here.
846 	 *
847 	 * If we are currently making the filesystem, the initialisation will
848 	 * fail as the perag data is in an undefined state.
849 	 */
850 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
851 	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
852 	     !mp->m_sb.sb_inprogress) {
853 		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
854 		if (error)
855 			goto out_log_dealloc;
856 	}
857 
858 	/*
859 	 * Get and sanity-check the root inode.
860 	 * Save the pointer to it in the mount structure.
861 	 */
862 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
863 	if (error) {
864 		xfs_warn(mp, "failed to read root inode");
865 		goto out_log_dealloc;
866 	}
867 
868 	ASSERT(rip != NULL);
869 
870 	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
871 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
872 			(unsigned long long)rip->i_ino);
873 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
874 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
875 				 mp);
876 		error = -EFSCORRUPTED;
877 		goto out_rele_rip;
878 	}
879 	mp->m_rootip = rip;	/* save it */
880 
881 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
882 
883 	/*
884 	 * Initialize realtime inode pointers in the mount structure
885 	 */
886 	error = xfs_rtmount_inodes(mp);
887 	if (error) {
888 		/*
889 		 * Free up the root inode.
890 		 */
891 		xfs_warn(mp, "failed to read RT inodes");
892 		goto out_rele_rip;
893 	}
894 
895 	/*
896 	 * If this is a read-only mount defer the superblock updates until
897 	 * the next remount into writeable mode.  Otherwise we would never
898 	 * perform the update e.g. for the root filesystem.
899 	 */
900 	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
901 		error = xfs_sync_sb(mp, false);
902 		if (error) {
903 			xfs_warn(mp, "failed to write sb changes");
904 			goto out_rtunmount;
905 		}
906 	}
907 
908 	/*
909 	 * Initialise the XFS quota management subsystem for this mount
910 	 */
911 	if (XFS_IS_QUOTA_RUNNING(mp)) {
912 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
913 		if (error)
914 			goto out_rtunmount;
915 	} else {
916 		ASSERT(!XFS_IS_QUOTA_ON(mp));
917 
918 		/*
919 		 * If a file system had quotas running earlier, but decided to
920 		 * mount without -o uquota/pquota/gquota options, revoke the
921 		 * quotachecked license.
922 		 */
923 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
924 			xfs_notice(mp, "resetting quota flags");
925 			error = xfs_mount_reset_sbqflags(mp);
926 			if (error)
927 				goto out_rtunmount;
928 		}
929 	}
930 
931 	/*
932 	 * During the second phase of log recovery, we need iget and
933 	 * iput to behave like they do for an active filesystem.
934 	 * xfs_fs_drop_inode needs to be able to prevent the deletion
935 	 * of inodes before we're done replaying log items on those
936 	 * inodes.
937 	 */
938 	mp->m_super->s_flags |= MS_ACTIVE;
939 
940 	/*
941 	 * Finish recovering the file system.  This part needed to be delayed
942 	 * until after the root and real-time bitmap inodes were consistently
943 	 * read in.
944 	 */
945 	error = xfs_log_mount_finish(mp);
946 	if (error) {
947 		xfs_warn(mp, "log mount finish failed");
948 		goto out_rtunmount;
949 	}
950 
951 	/*
952 	 * Now the log is fully replayed, we can transition to full read-only
953 	 * mode for read-only mounts. This will sync all the metadata and clean
954 	 * the log so that the recovery we just performed does not have to be
955 	 * replayed again on the next mount.
956 	 *
957 	 * We use the same quiesce mechanism as the rw->ro remount, as they are
958 	 * semantically identical operations.
959 	 */
960 	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
961 							XFS_MOUNT_RDONLY) {
962 		xfs_quiesce_attr(mp);
963 	}
964 
965 	/*
966 	 * Complete the quota initialisation, post-log-replay component.
967 	 */
968 	if (quotamount) {
969 		ASSERT(mp->m_qflags == 0);
970 		mp->m_qflags = quotaflags;
971 
972 		xfs_qm_mount_quotas(mp);
973 	}
974 
975 	/*
976 	 * Now we are mounted, reserve a small amount of unused space for
977 	 * privileged transactions. This is needed so that transaction
978 	 * space required for critical operations can dip into this pool
979 	 * when at ENOSPC. This is needed for operations like create with
980 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
981 	 * are not allowed to use this reserved space.
982 	 *
983 	 * This may drive us straight to ENOSPC on mount, but that implies
984 	 * we were already there on the last unmount. Warn if this occurs.
985 	 */
986 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
987 		resblks = xfs_default_resblks(mp);
988 		error = xfs_reserve_blocks(mp, &resblks, NULL);
989 		if (error)
990 			xfs_warn(mp,
991 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
992 
993 		/* Recover any CoW blocks that never got remapped. */
994 		error = xfs_reflink_recover_cow(mp);
995 		if (error) {
996 			xfs_err(mp,
997 	"Error %d recovering leftover CoW allocations.", error);
998 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
999 			goto out_quota;
1000 		}
1001 
1002 		/* Reserve AG blocks for future btree expansion. */
1003 		error = xfs_fs_reserve_ag_blocks(mp);
1004 		if (error && error != -ENOSPC)
1005 			goto out_agresv;
1006 	}
1007 
1008 	return 0;
1009 
1010  out_agresv:
1011 	xfs_fs_unreserve_ag_blocks(mp);
1012  out_quota:
1013 	xfs_qm_unmount_quotas(mp);
1014  out_rtunmount:
1015 	mp->m_super->s_flags &= ~MS_ACTIVE;
1016 	xfs_rtunmount_inodes(mp);
1017  out_rele_rip:
1018 	IRELE(rip);
1019 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1020 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1021  out_log_dealloc:
1022 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1023 	xfs_log_mount_cancel(mp);
1024  out_fail_wait:
1025 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1026 		xfs_wait_buftarg(mp->m_logdev_targp);
1027 	xfs_wait_buftarg(mp->m_ddev_targp);
1028  out_free_perag:
1029 	xfs_free_perag(mp);
1030  out_free_dir:
1031 	xfs_da_unmount(mp);
1032  out_remove_uuid:
1033 	xfs_uuid_unmount(mp);
1034  out_remove_error_sysfs:
1035 	xfs_error_sysfs_del(mp);
1036  out_del_stats:
1037 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1038  out_remove_sysfs:
1039 	xfs_sysfs_del(&mp->m_kobj);
1040  out:
1041 	return error;
1042 }
1043 
1044 /*
1045  * This flushes out the inodes,dquots and the superblock, unmounts the
1046  * log and makes sure that incore structures are freed.
1047  */
1048 void
1049 xfs_unmountfs(
1050 	struct xfs_mount	*mp)
1051 {
1052 	__uint64_t		resblks;
1053 	int			error;
1054 
1055 	cancel_delayed_work_sync(&mp->m_eofblocks_work);
1056 	cancel_delayed_work_sync(&mp->m_cowblocks_work);
1057 
1058 	xfs_fs_unreserve_ag_blocks(mp);
1059 	xfs_qm_unmount_quotas(mp);
1060 	xfs_rtunmount_inodes(mp);
1061 	IRELE(mp->m_rootip);
1062 
1063 	/*
1064 	 * We can potentially deadlock here if we have an inode cluster
1065 	 * that has been freed has its buffer still pinned in memory because
1066 	 * the transaction is still sitting in a iclog. The stale inodes
1067 	 * on that buffer will have their flush locks held until the
1068 	 * transaction hits the disk and the callbacks run. the inode
1069 	 * flush takes the flush lock unconditionally and with nothing to
1070 	 * push out the iclog we will never get that unlocked. hence we
1071 	 * need to force the log first.
1072 	 */
1073 	xfs_log_force(mp, XFS_LOG_SYNC);
1074 
1075 	/*
1076 	 * We now need to tell the world we are unmounting. This will allow
1077 	 * us to detect that the filesystem is going away and we should error
1078 	 * out anything that we have been retrying in the background. This will
1079 	 * prevent neverending retries in AIL pushing from hanging the unmount.
1080 	 */
1081 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1082 
1083 	/*
1084 	 * Flush all pending changes from the AIL.
1085 	 */
1086 	xfs_ail_push_all_sync(mp->m_ail);
1087 
1088 	/*
1089 	 * And reclaim all inodes.  At this point there should be no dirty
1090 	 * inodes and none should be pinned or locked, but use synchronous
1091 	 * reclaim just to be sure. We can stop background inode reclaim
1092 	 * here as well if it is still running.
1093 	 */
1094 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1095 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1096 
1097 	xfs_qm_unmount(mp);
1098 
1099 	/*
1100 	 * Unreserve any blocks we have so that when we unmount we don't account
1101 	 * the reserved free space as used. This is really only necessary for
1102 	 * lazy superblock counting because it trusts the incore superblock
1103 	 * counters to be absolutely correct on clean unmount.
1104 	 *
1105 	 * We don't bother correcting this elsewhere for lazy superblock
1106 	 * counting because on mount of an unclean filesystem we reconstruct the
1107 	 * correct counter value and this is irrelevant.
1108 	 *
1109 	 * For non-lazy counter filesystems, this doesn't matter at all because
1110 	 * we only every apply deltas to the superblock and hence the incore
1111 	 * value does not matter....
1112 	 */
1113 	resblks = 0;
1114 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1115 	if (error)
1116 		xfs_warn(mp, "Unable to free reserved block pool. "
1117 				"Freespace may not be correct on next mount.");
1118 
1119 	error = xfs_log_sbcount(mp);
1120 	if (error)
1121 		xfs_warn(mp, "Unable to update superblock counters. "
1122 				"Freespace may not be correct on next mount.");
1123 
1124 
1125 	xfs_log_unmount(mp);
1126 	xfs_da_unmount(mp);
1127 	xfs_uuid_unmount(mp);
1128 
1129 #if defined(DEBUG)
1130 	xfs_errortag_clearall(mp, 0);
1131 #endif
1132 	xfs_free_perag(mp);
1133 
1134 	xfs_error_sysfs_del(mp);
1135 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1136 	xfs_sysfs_del(&mp->m_kobj);
1137 }
1138 
1139 /*
1140  * Determine whether modifications can proceed. The caller specifies the minimum
1141  * freeze level for which modifications should not be allowed. This allows
1142  * certain operations to proceed while the freeze sequence is in progress, if
1143  * necessary.
1144  */
1145 bool
1146 xfs_fs_writable(
1147 	struct xfs_mount	*mp,
1148 	int			level)
1149 {
1150 	ASSERT(level > SB_UNFROZEN);
1151 	if ((mp->m_super->s_writers.frozen >= level) ||
1152 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1153 		return false;
1154 
1155 	return true;
1156 }
1157 
1158 /*
1159  * xfs_log_sbcount
1160  *
1161  * Sync the superblock counters to disk.
1162  *
1163  * Note this code can be called during the process of freezing, so we use the
1164  * transaction allocator that does not block when the transaction subsystem is
1165  * in its frozen state.
1166  */
1167 int
1168 xfs_log_sbcount(xfs_mount_t *mp)
1169 {
1170 	/* allow this to proceed during the freeze sequence... */
1171 	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1172 		return 0;
1173 
1174 	/*
1175 	 * we don't need to do this if we are updating the superblock
1176 	 * counters on every modification.
1177 	 */
1178 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1179 		return 0;
1180 
1181 	return xfs_sync_sb(mp, true);
1182 }
1183 
1184 /*
1185  * Deltas for the inode count are +/-64, hence we use a large batch size
1186  * of 128 so we don't need to take the counter lock on every update.
1187  */
1188 #define XFS_ICOUNT_BATCH	128
1189 int
1190 xfs_mod_icount(
1191 	struct xfs_mount	*mp,
1192 	int64_t			delta)
1193 {
1194 	__percpu_counter_add(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1195 	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1196 		ASSERT(0);
1197 		percpu_counter_add(&mp->m_icount, -delta);
1198 		return -EINVAL;
1199 	}
1200 	return 0;
1201 }
1202 
1203 int
1204 xfs_mod_ifree(
1205 	struct xfs_mount	*mp,
1206 	int64_t			delta)
1207 {
1208 	percpu_counter_add(&mp->m_ifree, delta);
1209 	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1210 		ASSERT(0);
1211 		percpu_counter_add(&mp->m_ifree, -delta);
1212 		return -EINVAL;
1213 	}
1214 	return 0;
1215 }
1216 
1217 /*
1218  * Deltas for the block count can vary from 1 to very large, but lock contention
1219  * only occurs on frequent small block count updates such as in the delayed
1220  * allocation path for buffered writes (page a time updates). Hence we set
1221  * a large batch count (1024) to minimise global counter updates except when
1222  * we get near to ENOSPC and we have to be very accurate with our updates.
1223  */
1224 #define XFS_FDBLOCKS_BATCH	1024
1225 int
1226 xfs_mod_fdblocks(
1227 	struct xfs_mount	*mp,
1228 	int64_t			delta,
1229 	bool			rsvd)
1230 {
1231 	int64_t			lcounter;
1232 	long long		res_used;
1233 	s32			batch;
1234 
1235 	if (delta > 0) {
1236 		/*
1237 		 * If the reserve pool is depleted, put blocks back into it
1238 		 * first. Most of the time the pool is full.
1239 		 */
1240 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1241 			percpu_counter_add(&mp->m_fdblocks, delta);
1242 			return 0;
1243 		}
1244 
1245 		spin_lock(&mp->m_sb_lock);
1246 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1247 
1248 		if (res_used > delta) {
1249 			mp->m_resblks_avail += delta;
1250 		} else {
1251 			delta -= res_used;
1252 			mp->m_resblks_avail = mp->m_resblks;
1253 			percpu_counter_add(&mp->m_fdblocks, delta);
1254 		}
1255 		spin_unlock(&mp->m_sb_lock);
1256 		return 0;
1257 	}
1258 
1259 	/*
1260 	 * Taking blocks away, need to be more accurate the closer we
1261 	 * are to zero.
1262 	 *
1263 	 * If the counter has a value of less than 2 * max batch size,
1264 	 * then make everything serialise as we are real close to
1265 	 * ENOSPC.
1266 	 */
1267 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1268 				     XFS_FDBLOCKS_BATCH) < 0)
1269 		batch = 1;
1270 	else
1271 		batch = XFS_FDBLOCKS_BATCH;
1272 
1273 	__percpu_counter_add(&mp->m_fdblocks, delta, batch);
1274 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1275 				     XFS_FDBLOCKS_BATCH) >= 0) {
1276 		/* we had space! */
1277 		return 0;
1278 	}
1279 
1280 	/*
1281 	 * lock up the sb for dipping into reserves before releasing the space
1282 	 * that took us to ENOSPC.
1283 	 */
1284 	spin_lock(&mp->m_sb_lock);
1285 	percpu_counter_add(&mp->m_fdblocks, -delta);
1286 	if (!rsvd)
1287 		goto fdblocks_enospc;
1288 
1289 	lcounter = (long long)mp->m_resblks_avail + delta;
1290 	if (lcounter >= 0) {
1291 		mp->m_resblks_avail = lcounter;
1292 		spin_unlock(&mp->m_sb_lock);
1293 		return 0;
1294 	}
1295 	printk_once(KERN_WARNING
1296 		"Filesystem \"%s\": reserve blocks depleted! "
1297 		"Consider increasing reserve pool size.",
1298 		mp->m_fsname);
1299 fdblocks_enospc:
1300 	spin_unlock(&mp->m_sb_lock);
1301 	return -ENOSPC;
1302 }
1303 
1304 int
1305 xfs_mod_frextents(
1306 	struct xfs_mount	*mp,
1307 	int64_t			delta)
1308 {
1309 	int64_t			lcounter;
1310 	int			ret = 0;
1311 
1312 	spin_lock(&mp->m_sb_lock);
1313 	lcounter = mp->m_sb.sb_frextents + delta;
1314 	if (lcounter < 0)
1315 		ret = -ENOSPC;
1316 	else
1317 		mp->m_sb.sb_frextents = lcounter;
1318 	spin_unlock(&mp->m_sb_lock);
1319 	return ret;
1320 }
1321 
1322 /*
1323  * xfs_getsb() is called to obtain the buffer for the superblock.
1324  * The buffer is returned locked and read in from disk.
1325  * The buffer should be released with a call to xfs_brelse().
1326  *
1327  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1328  * the superblock buffer if it can be locked without sleeping.
1329  * If it can't then we'll return NULL.
1330  */
1331 struct xfs_buf *
1332 xfs_getsb(
1333 	struct xfs_mount	*mp,
1334 	int			flags)
1335 {
1336 	struct xfs_buf		*bp = mp->m_sb_bp;
1337 
1338 	if (!xfs_buf_trylock(bp)) {
1339 		if (flags & XBF_TRYLOCK)
1340 			return NULL;
1341 		xfs_buf_lock(bp);
1342 	}
1343 
1344 	xfs_buf_hold(bp);
1345 	ASSERT(bp->b_flags & XBF_DONE);
1346 	return bp;
1347 }
1348 
1349 /*
1350  * Used to free the superblock along various error paths.
1351  */
1352 void
1353 xfs_freesb(
1354 	struct xfs_mount	*mp)
1355 {
1356 	struct xfs_buf		*bp = mp->m_sb_bp;
1357 
1358 	xfs_buf_lock(bp);
1359 	mp->m_sb_bp = NULL;
1360 	xfs_buf_relse(bp);
1361 }
1362 
1363 /*
1364  * If the underlying (data/log/rt) device is readonly, there are some
1365  * operations that cannot proceed.
1366  */
1367 int
1368 xfs_dev_is_read_only(
1369 	struct xfs_mount	*mp,
1370 	char			*message)
1371 {
1372 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1373 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1374 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1375 		xfs_notice(mp, "%s required on read-only device.", message);
1376 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1377 		return -EROFS;
1378 	}
1379 	return 0;
1380 }
1381