xref: /linux/fs/xfs/xfs_mount.c (revision d060296cc0300ae8ed08004ebd3994bf325fa257)
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 #include "xfs_extent_busy.h"
49 
50 
51 static DEFINE_MUTEX(xfs_uuid_table_mutex);
52 static int xfs_uuid_table_size;
53 static uuid_t *xfs_uuid_table;
54 
55 void
56 xfs_uuid_table_free(void)
57 {
58 	if (xfs_uuid_table_size == 0)
59 		return;
60 	kmem_free(xfs_uuid_table);
61 	xfs_uuid_table = NULL;
62 	xfs_uuid_table_size = 0;
63 }
64 
65 /*
66  * See if the UUID is unique among mounted XFS filesystems.
67  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
68  */
69 STATIC int
70 xfs_uuid_mount(
71 	struct xfs_mount	*mp)
72 {
73 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
74 	int			hole, i;
75 
76 	/* Publish UUID in struct super_block */
77 	uuid_copy(&mp->m_super->s_uuid, uuid);
78 
79 	if (mp->m_flags & XFS_MOUNT_NOUUID)
80 		return 0;
81 
82 	if (uuid_is_null(uuid)) {
83 		xfs_warn(mp, "Filesystem has null UUID - can't mount");
84 		return -EINVAL;
85 	}
86 
87 	mutex_lock(&xfs_uuid_table_mutex);
88 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
89 		if (uuid_is_null(&xfs_uuid_table[i])) {
90 			hole = i;
91 			continue;
92 		}
93 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
94 			goto out_duplicate;
95 	}
96 
97 	if (hole < 0) {
98 		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
99 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
100 			KM_SLEEP);
101 		hole = xfs_uuid_table_size++;
102 	}
103 	xfs_uuid_table[hole] = *uuid;
104 	mutex_unlock(&xfs_uuid_table_mutex);
105 
106 	return 0;
107 
108  out_duplicate:
109 	mutex_unlock(&xfs_uuid_table_mutex);
110 	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
111 	return -EINVAL;
112 }
113 
114 STATIC void
115 xfs_uuid_unmount(
116 	struct xfs_mount	*mp)
117 {
118 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
119 	int			i;
120 
121 	if (mp->m_flags & XFS_MOUNT_NOUUID)
122 		return;
123 
124 	mutex_lock(&xfs_uuid_table_mutex);
125 	for (i = 0; i < xfs_uuid_table_size; i++) {
126 		if (uuid_is_null(&xfs_uuid_table[i]))
127 			continue;
128 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
129 			continue;
130 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
131 		break;
132 	}
133 	ASSERT(i < xfs_uuid_table_size);
134 	mutex_unlock(&xfs_uuid_table_mutex);
135 }
136 
137 
138 STATIC void
139 __xfs_free_perag(
140 	struct rcu_head	*head)
141 {
142 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
143 
144 	ASSERT(atomic_read(&pag->pag_ref) == 0);
145 	kmem_free(pag);
146 }
147 
148 /*
149  * Free up the per-ag resources associated with the mount structure.
150  */
151 STATIC void
152 xfs_free_perag(
153 	xfs_mount_t	*mp)
154 {
155 	xfs_agnumber_t	agno;
156 	struct xfs_perag *pag;
157 
158 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
159 		spin_lock(&mp->m_perag_lock);
160 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
161 		spin_unlock(&mp->m_perag_lock);
162 		ASSERT(pag);
163 		ASSERT(atomic_read(&pag->pag_ref) == 0);
164 		xfs_buf_hash_destroy(pag);
165 		call_rcu(&pag->rcu_head, __xfs_free_perag);
166 	}
167 }
168 
169 /*
170  * Check size of device based on the (data/realtime) block count.
171  * Note: this check is used by the growfs code as well as mount.
172  */
173 int
174 xfs_sb_validate_fsb_count(
175 	xfs_sb_t	*sbp,
176 	uint64_t	nblocks)
177 {
178 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
179 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
180 
181 	/* Limited by ULONG_MAX of page cache index */
182 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
183 		return -EFBIG;
184 	return 0;
185 }
186 
187 int
188 xfs_initialize_perag(
189 	xfs_mount_t	*mp,
190 	xfs_agnumber_t	agcount,
191 	xfs_agnumber_t	*maxagi)
192 {
193 	xfs_agnumber_t	index;
194 	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
195 	xfs_perag_t	*pag;
196 	int		error = -ENOMEM;
197 
198 	/*
199 	 * Walk the current per-ag tree so we don't try to initialise AGs
200 	 * that already exist (growfs case). Allocate and insert all the
201 	 * AGs we don't find ready for initialisation.
202 	 */
203 	for (index = 0; index < agcount; index++) {
204 		pag = xfs_perag_get(mp, index);
205 		if (pag) {
206 			xfs_perag_put(pag);
207 			continue;
208 		}
209 
210 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
211 		if (!pag)
212 			goto out_unwind_new_pags;
213 		pag->pag_agno = index;
214 		pag->pag_mount = mp;
215 		spin_lock_init(&pag->pag_ici_lock);
216 		mutex_init(&pag->pag_ici_reclaim_lock);
217 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
218 		if (xfs_buf_hash_init(pag))
219 			goto out_free_pag;
220 		init_waitqueue_head(&pag->pagb_wait);
221 
222 		if (radix_tree_preload(GFP_NOFS))
223 			goto out_hash_destroy;
224 
225 		spin_lock(&mp->m_perag_lock);
226 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
227 			BUG();
228 			spin_unlock(&mp->m_perag_lock);
229 			radix_tree_preload_end();
230 			error = -EEXIST;
231 			goto out_hash_destroy;
232 		}
233 		spin_unlock(&mp->m_perag_lock);
234 		radix_tree_preload_end();
235 		/* first new pag is fully initialized */
236 		if (first_initialised == NULLAGNUMBER)
237 			first_initialised = index;
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 		kmem_free(pag);
260 	}
261 	return error;
262 }
263 
264 /*
265  * xfs_readsb
266  *
267  * Does the initial read of the superblock.
268  */
269 int
270 xfs_readsb(
271 	struct xfs_mount *mp,
272 	int		flags)
273 {
274 	unsigned int	sector_size;
275 	struct xfs_buf	*bp;
276 	struct xfs_sb	*sbp = &mp->m_sb;
277 	int		error;
278 	int		loud = !(flags & XFS_MFSI_QUIET);
279 	const struct xfs_buf_ops *buf_ops;
280 
281 	ASSERT(mp->m_sb_bp == NULL);
282 	ASSERT(mp->m_ddev_targp != NULL);
283 
284 	/*
285 	 * For the initial read, we must guess at the sector
286 	 * size based on the block device.  It's enough to
287 	 * get the sb_sectsize out of the superblock and
288 	 * then reread with the proper length.
289 	 * We don't verify it yet, because it may not be complete.
290 	 */
291 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
292 	buf_ops = NULL;
293 
294 	/*
295 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
296 	 * around at all times to optimize access to the superblock. Therefore,
297 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
298 	 * elevated.
299 	 */
300 reread:
301 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
302 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
303 				      buf_ops);
304 	if (error) {
305 		if (loud)
306 			xfs_warn(mp, "SB validate failed with error %d.", error);
307 		/* bad CRC means corrupted metadata */
308 		if (error == -EFSBADCRC)
309 			error = -EFSCORRUPTED;
310 		return error;
311 	}
312 
313 	/*
314 	 * Initialize the mount structure from the superblock.
315 	 */
316 	xfs_sb_from_disk(sbp, XFS_BUF_TO_SBP(bp));
317 
318 	/*
319 	 * If we haven't validated the superblock, do so now before we try
320 	 * to check the sector size and reread the superblock appropriately.
321 	 */
322 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
323 		if (loud)
324 			xfs_warn(mp, "Invalid superblock magic number");
325 		error = -EINVAL;
326 		goto release_buf;
327 	}
328 
329 	/*
330 	 * We must be able to do sector-sized and sector-aligned IO.
331 	 */
332 	if (sector_size > sbp->sb_sectsize) {
333 		if (loud)
334 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
335 				sector_size, sbp->sb_sectsize);
336 		error = -ENOSYS;
337 		goto release_buf;
338 	}
339 
340 	if (buf_ops == NULL) {
341 		/*
342 		 * Re-read the superblock so the buffer is correctly sized,
343 		 * and properly verified.
344 		 */
345 		xfs_buf_relse(bp);
346 		sector_size = sbp->sb_sectsize;
347 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
348 		goto reread;
349 	}
350 
351 	xfs_reinit_percpu_counters(mp);
352 
353 	/* no need to be quiet anymore, so reset the buf ops */
354 	bp->b_ops = &xfs_sb_buf_ops;
355 
356 	mp->m_sb_bp = bp;
357 	xfs_buf_unlock(bp);
358 	return 0;
359 
360 release_buf:
361 	xfs_buf_relse(bp);
362 	return error;
363 }
364 
365 /*
366  * Update alignment values based on mount options and sb values
367  */
368 STATIC int
369 xfs_update_alignment(xfs_mount_t *mp)
370 {
371 	xfs_sb_t	*sbp = &(mp->m_sb);
372 
373 	if (mp->m_dalign) {
374 		/*
375 		 * If stripe unit and stripe width are not multiples
376 		 * of the fs blocksize turn off alignment.
377 		 */
378 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
379 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
380 			xfs_warn(mp,
381 		"alignment check failed: sunit/swidth vs. blocksize(%d)",
382 				sbp->sb_blocksize);
383 			return -EINVAL;
384 		} else {
385 			/*
386 			 * Convert the stripe unit and width to FSBs.
387 			 */
388 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
389 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
390 				xfs_warn(mp,
391 			"alignment check failed: sunit/swidth vs. agsize(%d)",
392 					 sbp->sb_agblocks);
393 				return -EINVAL;
394 			} else if (mp->m_dalign) {
395 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
396 			} else {
397 				xfs_warn(mp,
398 			"alignment check failed: sunit(%d) less than bsize(%d)",
399 					 mp->m_dalign, sbp->sb_blocksize);
400 				return -EINVAL;
401 			}
402 		}
403 
404 		/*
405 		 * Update superblock with new values
406 		 * and log changes
407 		 */
408 		if (xfs_sb_version_hasdalign(sbp)) {
409 			if (sbp->sb_unit != mp->m_dalign) {
410 				sbp->sb_unit = mp->m_dalign;
411 				mp->m_update_sb = true;
412 			}
413 			if (sbp->sb_width != mp->m_swidth) {
414 				sbp->sb_width = mp->m_swidth;
415 				mp->m_update_sb = true;
416 			}
417 		} else {
418 			xfs_warn(mp,
419 	"cannot change alignment: superblock does not support data alignment");
420 			return -EINVAL;
421 		}
422 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
423 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
424 			mp->m_dalign = sbp->sb_unit;
425 			mp->m_swidth = sbp->sb_width;
426 	}
427 
428 	return 0;
429 }
430 
431 /*
432  * Set the maximum inode count for this filesystem
433  */
434 STATIC void
435 xfs_set_maxicount(xfs_mount_t *mp)
436 {
437 	xfs_sb_t	*sbp = &(mp->m_sb);
438 	uint64_t	icount;
439 
440 	if (sbp->sb_imax_pct) {
441 		/*
442 		 * Make sure the maximum inode count is a multiple
443 		 * of the units we allocate inodes in.
444 		 */
445 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
446 		do_div(icount, 100);
447 		do_div(icount, mp->m_ialloc_blks);
448 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
449 				   sbp->sb_inopblog;
450 	} else {
451 		mp->m_maxicount = 0;
452 	}
453 }
454 
455 /*
456  * Set the default minimum read and write sizes unless
457  * already specified in a mount option.
458  * We use smaller I/O sizes when the file system
459  * is being used for NFS service (wsync mount option).
460  */
461 STATIC void
462 xfs_set_rw_sizes(xfs_mount_t *mp)
463 {
464 	xfs_sb_t	*sbp = &(mp->m_sb);
465 	int		readio_log, writeio_log;
466 
467 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
468 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
469 			readio_log = XFS_WSYNC_READIO_LOG;
470 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
471 		} else {
472 			readio_log = XFS_READIO_LOG_LARGE;
473 			writeio_log = XFS_WRITEIO_LOG_LARGE;
474 		}
475 	} else {
476 		readio_log = mp->m_readio_log;
477 		writeio_log = mp->m_writeio_log;
478 	}
479 
480 	if (sbp->sb_blocklog > readio_log) {
481 		mp->m_readio_log = sbp->sb_blocklog;
482 	} else {
483 		mp->m_readio_log = readio_log;
484 	}
485 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
486 	if (sbp->sb_blocklog > writeio_log) {
487 		mp->m_writeio_log = sbp->sb_blocklog;
488 	} else {
489 		mp->m_writeio_log = writeio_log;
490 	}
491 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
492 }
493 
494 /*
495  * precalculate the low space thresholds for dynamic speculative preallocation.
496  */
497 void
498 xfs_set_low_space_thresholds(
499 	struct xfs_mount	*mp)
500 {
501 	int i;
502 
503 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
504 		uint64_t space = mp->m_sb.sb_dblocks;
505 
506 		do_div(space, 100);
507 		mp->m_low_space[i] = space * (i + 1);
508 	}
509 }
510 
511 
512 /*
513  * Set whether we're using inode alignment.
514  */
515 STATIC void
516 xfs_set_inoalignment(xfs_mount_t *mp)
517 {
518 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
519 		mp->m_sb.sb_inoalignmt >= xfs_icluster_size_fsb(mp))
520 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
521 	else
522 		mp->m_inoalign_mask = 0;
523 	/*
524 	 * If we are using stripe alignment, check whether
525 	 * the stripe unit is a multiple of the inode alignment
526 	 */
527 	if (mp->m_dalign && mp->m_inoalign_mask &&
528 	    !(mp->m_dalign & mp->m_inoalign_mask))
529 		mp->m_sinoalign = mp->m_dalign;
530 	else
531 		mp->m_sinoalign = 0;
532 }
533 
534 /*
535  * Check that the data (and log if separate) is an ok size.
536  */
537 STATIC int
538 xfs_check_sizes(
539 	struct xfs_mount *mp)
540 {
541 	struct xfs_buf	*bp;
542 	xfs_daddr_t	d;
543 	int		error;
544 
545 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
546 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
547 		xfs_warn(mp, "filesystem size mismatch detected");
548 		return -EFBIG;
549 	}
550 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
551 					d - XFS_FSS_TO_BB(mp, 1),
552 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
553 	if (error) {
554 		xfs_warn(mp, "last sector read failed");
555 		return error;
556 	}
557 	xfs_buf_relse(bp);
558 
559 	if (mp->m_logdev_targp == mp->m_ddev_targp)
560 		return 0;
561 
562 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
563 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
564 		xfs_warn(mp, "log size mismatch detected");
565 		return -EFBIG;
566 	}
567 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
568 					d - XFS_FSB_TO_BB(mp, 1),
569 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
570 	if (error) {
571 		xfs_warn(mp, "log device read failed");
572 		return error;
573 	}
574 	xfs_buf_relse(bp);
575 	return 0;
576 }
577 
578 /*
579  * Clear the quotaflags in memory and in the superblock.
580  */
581 int
582 xfs_mount_reset_sbqflags(
583 	struct xfs_mount	*mp)
584 {
585 	mp->m_qflags = 0;
586 
587 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
588 	if (mp->m_sb.sb_qflags == 0)
589 		return 0;
590 	spin_lock(&mp->m_sb_lock);
591 	mp->m_sb.sb_qflags = 0;
592 	spin_unlock(&mp->m_sb_lock);
593 
594 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
595 		return 0;
596 
597 	return xfs_sync_sb(mp, false);
598 }
599 
600 uint64_t
601 xfs_default_resblks(xfs_mount_t *mp)
602 {
603 	uint64_t resblks;
604 
605 	/*
606 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
607 	 * smaller.  This is intended to cover concurrent allocation
608 	 * transactions when we initially hit enospc. These each require a 4
609 	 * block reservation. Hence by default we cover roughly 2000 concurrent
610 	 * allocation reservations.
611 	 */
612 	resblks = mp->m_sb.sb_dblocks;
613 	do_div(resblks, 20);
614 	resblks = min_t(uint64_t, resblks, 8192);
615 	return resblks;
616 }
617 
618 /*
619  * This function does the following on an initial mount of a file system:
620  *	- reads the superblock from disk and init the mount struct
621  *	- if we're a 32-bit kernel, do a size check on the superblock
622  *		so we don't mount terabyte filesystems
623  *	- init mount struct realtime fields
624  *	- allocate inode hash table for fs
625  *	- init directory manager
626  *	- perform recovery and init the log manager
627  */
628 int
629 xfs_mountfs(
630 	struct xfs_mount	*mp)
631 {
632 	struct xfs_sb		*sbp = &(mp->m_sb);
633 	struct xfs_inode	*rip;
634 	uint64_t		resblks;
635 	uint			quotamount = 0;
636 	uint			quotaflags = 0;
637 	int			error = 0;
638 
639 	xfs_sb_mount_common(mp, sbp);
640 
641 	/*
642 	 * Check for a mismatched features2 values.  Older kernels read & wrote
643 	 * into the wrong sb offset for sb_features2 on some platforms due to
644 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
645 	 * which made older superblock reading/writing routines swap it as a
646 	 * 64-bit value.
647 	 *
648 	 * For backwards compatibility, we make both slots equal.
649 	 *
650 	 * If we detect a mismatched field, we OR the set bits into the existing
651 	 * features2 field in case it has already been modified; we don't want
652 	 * to lose any features.  We then update the bad location with the ORed
653 	 * value so that older kernels will see any features2 flags. The
654 	 * superblock writeback code ensures the new sb_features2 is copied to
655 	 * sb_bad_features2 before it is logged or written to disk.
656 	 */
657 	if (xfs_sb_has_mismatched_features2(sbp)) {
658 		xfs_warn(mp, "correcting sb_features alignment problem");
659 		sbp->sb_features2 |= sbp->sb_bad_features2;
660 		mp->m_update_sb = true;
661 
662 		/*
663 		 * Re-check for ATTR2 in case it was found in bad_features2
664 		 * slot.
665 		 */
666 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
667 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
668 			mp->m_flags |= XFS_MOUNT_ATTR2;
669 	}
670 
671 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
672 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
673 		xfs_sb_version_removeattr2(&mp->m_sb);
674 		mp->m_update_sb = true;
675 
676 		/* update sb_versionnum for the clearing of the morebits */
677 		if (!sbp->sb_features2)
678 			mp->m_update_sb = true;
679 	}
680 
681 	/* always use v2 inodes by default now */
682 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
683 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
684 		mp->m_update_sb = true;
685 	}
686 
687 	/*
688 	 * Check if sb_agblocks is aligned at stripe boundary
689 	 * If sb_agblocks is NOT aligned turn off m_dalign since
690 	 * allocator alignment is within an ag, therefore ag has
691 	 * to be aligned at stripe boundary.
692 	 */
693 	error = xfs_update_alignment(mp);
694 	if (error)
695 		goto out;
696 
697 	xfs_alloc_compute_maxlevels(mp);
698 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
699 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
700 	xfs_ialloc_compute_maxlevels(mp);
701 	xfs_rmapbt_compute_maxlevels(mp);
702 	xfs_refcountbt_compute_maxlevels(mp);
703 
704 	xfs_set_maxicount(mp);
705 
706 	/* enable fail_at_unmount as default */
707 	mp->m_fail_unmount = 1;
708 
709 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype, NULL, mp->m_fsname);
710 	if (error)
711 		goto out;
712 
713 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
714 			       &mp->m_kobj, "stats");
715 	if (error)
716 		goto out_remove_sysfs;
717 
718 	error = xfs_error_sysfs_init(mp);
719 	if (error)
720 		goto out_del_stats;
721 
722 	error = xfs_errortag_init(mp);
723 	if (error)
724 		goto out_remove_error_sysfs;
725 
726 	error = xfs_uuid_mount(mp);
727 	if (error)
728 		goto out_remove_errortag;
729 
730 	/*
731 	 * Set the minimum read and write sizes
732 	 */
733 	xfs_set_rw_sizes(mp);
734 
735 	/* set the low space thresholds for dynamic preallocation */
736 	xfs_set_low_space_thresholds(mp);
737 
738 	/*
739 	 * Set the inode cluster size.
740 	 * This may still be overridden by the file system
741 	 * block size if it is larger than the chosen cluster size.
742 	 *
743 	 * For v5 filesystems, scale the cluster size with the inode size to
744 	 * keep a constant ratio of inode per cluster buffer, but only if mkfs
745 	 * has set the inode alignment value appropriately for larger cluster
746 	 * sizes.
747 	 */
748 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
749 	if (xfs_sb_version_hascrc(&mp->m_sb)) {
750 		int	new_size = mp->m_inode_cluster_size;
751 
752 		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
753 		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
754 			mp->m_inode_cluster_size = new_size;
755 	}
756 
757 	/*
758 	 * If enabled, sparse inode chunk alignment is expected to match the
759 	 * cluster size. Full inode chunk alignment must match the chunk size,
760 	 * but that is checked on sb read verification...
761 	 */
762 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
763 	    mp->m_sb.sb_spino_align !=
764 			XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size)) {
765 		xfs_warn(mp,
766 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
767 			 mp->m_sb.sb_spino_align,
768 			 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size));
769 		error = -EINVAL;
770 		goto out_remove_uuid;
771 	}
772 
773 	/*
774 	 * Set inode alignment fields
775 	 */
776 	xfs_set_inoalignment(mp);
777 
778 	/*
779 	 * Check that the data (and log if separate) is an ok size.
780 	 */
781 	error = xfs_check_sizes(mp);
782 	if (error)
783 		goto out_remove_uuid;
784 
785 	/*
786 	 * Initialize realtime fields in the mount structure
787 	 */
788 	error = xfs_rtmount_init(mp);
789 	if (error) {
790 		xfs_warn(mp, "RT mount failed");
791 		goto out_remove_uuid;
792 	}
793 
794 	/*
795 	 *  Copies the low order bits of the timestamp and the randomly
796 	 *  set "sequence" number out of a UUID.
797 	 */
798 	mp->m_fixedfsid[0] =
799 		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
800 		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
801 	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
802 
803 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
804 
805 	error = xfs_da_mount(mp);
806 	if (error) {
807 		xfs_warn(mp, "Failed dir/attr init: %d", error);
808 		goto out_remove_uuid;
809 	}
810 
811 	/*
812 	 * Initialize the precomputed transaction reservations values.
813 	 */
814 	xfs_trans_init(mp);
815 
816 	/*
817 	 * Allocate and initialize the per-ag data.
818 	 */
819 	spin_lock_init(&mp->m_perag_lock);
820 	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
821 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
822 	if (error) {
823 		xfs_warn(mp, "Failed per-ag init: %d", error);
824 		goto out_free_dir;
825 	}
826 
827 	if (!sbp->sb_logblocks) {
828 		xfs_warn(mp, "no log defined");
829 		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
830 		error = -EFSCORRUPTED;
831 		goto out_free_perag;
832 	}
833 
834 	/*
835 	 * Log's mount-time initialization. The first part of recovery can place
836 	 * some items on the AIL, to be handled when recovery is finished or
837 	 * cancelled.
838 	 */
839 	error = xfs_log_mount(mp, mp->m_logdev_targp,
840 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
841 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
842 	if (error) {
843 		xfs_warn(mp, "log mount failed");
844 		goto out_fail_wait;
845 	}
846 
847 	/*
848 	 * Now the log is mounted, we know if it was an unclean shutdown or
849 	 * not. If it was, with the first phase of recovery has completed, we
850 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
851 	 * but they are recovered transactionally in the second recovery phase
852 	 * later.
853 	 *
854 	 * Hence we can safely re-initialise incore superblock counters from
855 	 * the per-ag data. These may not be correct if the filesystem was not
856 	 * cleanly unmounted, so we need to wait for recovery to finish before
857 	 * doing this.
858 	 *
859 	 * If the filesystem was cleanly unmounted, then we can trust the
860 	 * values in the superblock to be correct and we don't need to do
861 	 * anything here.
862 	 *
863 	 * If we are currently making the filesystem, the initialisation will
864 	 * fail as the perag data is in an undefined state.
865 	 */
866 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
867 	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
868 	     !mp->m_sb.sb_inprogress) {
869 		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
870 		if (error)
871 			goto out_log_dealloc;
872 	}
873 
874 	/*
875 	 * Get and sanity-check the root inode.
876 	 * Save the pointer to it in the mount structure.
877 	 */
878 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
879 	if (error) {
880 		xfs_warn(mp, "failed to read root inode");
881 		goto out_log_dealloc;
882 	}
883 
884 	ASSERT(rip != NULL);
885 
886 	if (unlikely(!S_ISDIR(VFS_I(rip)->i_mode))) {
887 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
888 			(unsigned long long)rip->i_ino);
889 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
890 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
891 				 mp);
892 		error = -EFSCORRUPTED;
893 		goto out_rele_rip;
894 	}
895 	mp->m_rootip = rip;	/* save it */
896 
897 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
898 
899 	/*
900 	 * Initialize realtime inode pointers in the mount structure
901 	 */
902 	error = xfs_rtmount_inodes(mp);
903 	if (error) {
904 		/*
905 		 * Free up the root inode.
906 		 */
907 		xfs_warn(mp, "failed to read RT inodes");
908 		goto out_rele_rip;
909 	}
910 
911 	/*
912 	 * If this is a read-only mount defer the superblock updates until
913 	 * the next remount into writeable mode.  Otherwise we would never
914 	 * perform the update e.g. for the root filesystem.
915 	 */
916 	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
917 		error = xfs_sync_sb(mp, false);
918 		if (error) {
919 			xfs_warn(mp, "failed to write sb changes");
920 			goto out_rtunmount;
921 		}
922 	}
923 
924 	/*
925 	 * Initialise the XFS quota management subsystem for this mount
926 	 */
927 	if (XFS_IS_QUOTA_RUNNING(mp)) {
928 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
929 		if (error)
930 			goto out_rtunmount;
931 	} else {
932 		ASSERT(!XFS_IS_QUOTA_ON(mp));
933 
934 		/*
935 		 * If a file system had quotas running earlier, but decided to
936 		 * mount without -o uquota/pquota/gquota options, revoke the
937 		 * quotachecked license.
938 		 */
939 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
940 			xfs_notice(mp, "resetting quota flags");
941 			error = xfs_mount_reset_sbqflags(mp);
942 			if (error)
943 				goto out_rtunmount;
944 		}
945 	}
946 
947 	/*
948 	 * During the second phase of log recovery, we need iget and
949 	 * iput to behave like they do for an active filesystem.
950 	 * xfs_fs_drop_inode needs to be able to prevent the deletion
951 	 * of inodes before we're done replaying log items on those
952 	 * inodes.
953 	 */
954 	mp->m_super->s_flags |= MS_ACTIVE;
955 
956 	/*
957 	 * Finish recovering the file system.  This part needed to be delayed
958 	 * until after the root and real-time bitmap inodes were consistently
959 	 * read in.
960 	 */
961 	error = xfs_log_mount_finish(mp);
962 	if (error) {
963 		xfs_warn(mp, "log mount finish failed");
964 		goto out_rtunmount;
965 	}
966 
967 	/*
968 	 * Now the log is fully replayed, we can transition to full read-only
969 	 * mode for read-only mounts. This will sync all the metadata and clean
970 	 * the log so that the recovery we just performed does not have to be
971 	 * replayed again on the next mount.
972 	 *
973 	 * We use the same quiesce mechanism as the rw->ro remount, as they are
974 	 * semantically identical operations.
975 	 */
976 	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
977 							XFS_MOUNT_RDONLY) {
978 		xfs_quiesce_attr(mp);
979 	}
980 
981 	/*
982 	 * Complete the quota initialisation, post-log-replay component.
983 	 */
984 	if (quotamount) {
985 		ASSERT(mp->m_qflags == 0);
986 		mp->m_qflags = quotaflags;
987 
988 		xfs_qm_mount_quotas(mp);
989 	}
990 
991 	/*
992 	 * Now we are mounted, reserve a small amount of unused space for
993 	 * privileged transactions. This is needed so that transaction
994 	 * space required for critical operations can dip into this pool
995 	 * when at ENOSPC. This is needed for operations like create with
996 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
997 	 * are not allowed to use this reserved space.
998 	 *
999 	 * This may drive us straight to ENOSPC on mount, but that implies
1000 	 * we were already there on the last unmount. Warn if this occurs.
1001 	 */
1002 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1003 		resblks = xfs_default_resblks(mp);
1004 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1005 		if (error)
1006 			xfs_warn(mp,
1007 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
1008 
1009 		/* Recover any CoW blocks that never got remapped. */
1010 		error = xfs_reflink_recover_cow(mp);
1011 		if (error) {
1012 			xfs_err(mp,
1013 	"Error %d recovering leftover CoW allocations.", error);
1014 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1015 			goto out_quota;
1016 		}
1017 
1018 		/* Reserve AG blocks for future btree expansion. */
1019 		error = xfs_fs_reserve_ag_blocks(mp);
1020 		if (error && error != -ENOSPC)
1021 			goto out_agresv;
1022 	}
1023 
1024 	return 0;
1025 
1026  out_agresv:
1027 	xfs_fs_unreserve_ag_blocks(mp);
1028  out_quota:
1029 	xfs_qm_unmount_quotas(mp);
1030  out_rtunmount:
1031 	mp->m_super->s_flags &= ~MS_ACTIVE;
1032 	xfs_rtunmount_inodes(mp);
1033  out_rele_rip:
1034 	IRELE(rip);
1035 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1036 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1037  out_log_dealloc:
1038 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1039 	xfs_log_mount_cancel(mp);
1040  out_fail_wait:
1041 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1042 		xfs_wait_buftarg(mp->m_logdev_targp);
1043 	xfs_wait_buftarg(mp->m_ddev_targp);
1044  out_free_perag:
1045 	xfs_free_perag(mp);
1046  out_free_dir:
1047 	xfs_da_unmount(mp);
1048  out_remove_uuid:
1049 	xfs_uuid_unmount(mp);
1050  out_remove_errortag:
1051 	xfs_errortag_del(mp);
1052  out_remove_error_sysfs:
1053 	xfs_error_sysfs_del(mp);
1054  out_del_stats:
1055 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1056  out_remove_sysfs:
1057 	xfs_sysfs_del(&mp->m_kobj);
1058  out:
1059 	return error;
1060 }
1061 
1062 /*
1063  * This flushes out the inodes,dquots and the superblock, unmounts the
1064  * log and makes sure that incore structures are freed.
1065  */
1066 void
1067 xfs_unmountfs(
1068 	struct xfs_mount	*mp)
1069 {
1070 	uint64_t		resblks;
1071 	int			error;
1072 
1073 	cancel_delayed_work_sync(&mp->m_eofblocks_work);
1074 	cancel_delayed_work_sync(&mp->m_cowblocks_work);
1075 
1076 	xfs_fs_unreserve_ag_blocks(mp);
1077 	xfs_qm_unmount_quotas(mp);
1078 	xfs_rtunmount_inodes(mp);
1079 	IRELE(mp->m_rootip);
1080 
1081 	/*
1082 	 * We can potentially deadlock here if we have an inode cluster
1083 	 * that has been freed has its buffer still pinned in memory because
1084 	 * the transaction is still sitting in a iclog. The stale inodes
1085 	 * on that buffer will have their flush locks held until the
1086 	 * transaction hits the disk and the callbacks run. the inode
1087 	 * flush takes the flush lock unconditionally and with nothing to
1088 	 * push out the iclog we will never get that unlocked. hence we
1089 	 * need to force the log first.
1090 	 */
1091 	xfs_log_force(mp, XFS_LOG_SYNC);
1092 
1093 	/*
1094 	 * Wait for all busy extents to be freed, including completion of
1095 	 * any discard operation.
1096 	 */
1097 	xfs_extent_busy_wait_all(mp);
1098 	flush_workqueue(xfs_discard_wq);
1099 
1100 	/*
1101 	 * We now need to tell the world we are unmounting. This will allow
1102 	 * us to detect that the filesystem is going away and we should error
1103 	 * out anything that we have been retrying in the background. This will
1104 	 * prevent neverending retries in AIL pushing from hanging the unmount.
1105 	 */
1106 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1107 
1108 	/*
1109 	 * Flush all pending changes from the AIL.
1110 	 */
1111 	xfs_ail_push_all_sync(mp->m_ail);
1112 
1113 	/*
1114 	 * And reclaim all inodes.  At this point there should be no dirty
1115 	 * inodes and none should be pinned or locked, but use synchronous
1116 	 * reclaim just to be sure. We can stop background inode reclaim
1117 	 * here as well if it is still running.
1118 	 */
1119 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1120 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1121 
1122 	xfs_qm_unmount(mp);
1123 
1124 	/*
1125 	 * Unreserve any blocks we have so that when we unmount we don't account
1126 	 * the reserved free space as used. This is really only necessary for
1127 	 * lazy superblock counting because it trusts the incore superblock
1128 	 * counters to be absolutely correct on clean unmount.
1129 	 *
1130 	 * We don't bother correcting this elsewhere for lazy superblock
1131 	 * counting because on mount of an unclean filesystem we reconstruct the
1132 	 * correct counter value and this is irrelevant.
1133 	 *
1134 	 * For non-lazy counter filesystems, this doesn't matter at all because
1135 	 * we only every apply deltas to the superblock and hence the incore
1136 	 * value does not matter....
1137 	 */
1138 	resblks = 0;
1139 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1140 	if (error)
1141 		xfs_warn(mp, "Unable to free reserved block pool. "
1142 				"Freespace may not be correct on next mount.");
1143 
1144 	error = xfs_log_sbcount(mp);
1145 	if (error)
1146 		xfs_warn(mp, "Unable to update superblock counters. "
1147 				"Freespace may not be correct on next mount.");
1148 
1149 
1150 	xfs_log_unmount(mp);
1151 	xfs_da_unmount(mp);
1152 	xfs_uuid_unmount(mp);
1153 
1154 #if defined(DEBUG)
1155 	xfs_errortag_clearall(mp);
1156 #endif
1157 	xfs_free_perag(mp);
1158 
1159 	xfs_errortag_del(mp);
1160 	xfs_error_sysfs_del(mp);
1161 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1162 	xfs_sysfs_del(&mp->m_kobj);
1163 }
1164 
1165 /*
1166  * Determine whether modifications can proceed. The caller specifies the minimum
1167  * freeze level for which modifications should not be allowed. This allows
1168  * certain operations to proceed while the freeze sequence is in progress, if
1169  * necessary.
1170  */
1171 bool
1172 xfs_fs_writable(
1173 	struct xfs_mount	*mp,
1174 	int			level)
1175 {
1176 	ASSERT(level > SB_UNFROZEN);
1177 	if ((mp->m_super->s_writers.frozen >= level) ||
1178 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1179 		return false;
1180 
1181 	return true;
1182 }
1183 
1184 /*
1185  * xfs_log_sbcount
1186  *
1187  * Sync the superblock counters to disk.
1188  *
1189  * Note this code can be called during the process of freezing, so we use the
1190  * transaction allocator that does not block when the transaction subsystem is
1191  * in its frozen state.
1192  */
1193 int
1194 xfs_log_sbcount(xfs_mount_t *mp)
1195 {
1196 	/* allow this to proceed during the freeze sequence... */
1197 	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1198 		return 0;
1199 
1200 	/*
1201 	 * we don't need to do this if we are updating the superblock
1202 	 * counters on every modification.
1203 	 */
1204 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1205 		return 0;
1206 
1207 	return xfs_sync_sb(mp, true);
1208 }
1209 
1210 /*
1211  * Deltas for the inode count are +/-64, hence we use a large batch size
1212  * of 128 so we don't need to take the counter lock on every update.
1213  */
1214 #define XFS_ICOUNT_BATCH	128
1215 int
1216 xfs_mod_icount(
1217 	struct xfs_mount	*mp,
1218 	int64_t			delta)
1219 {
1220 	percpu_counter_add_batch(&mp->m_icount, delta, XFS_ICOUNT_BATCH);
1221 	if (__percpu_counter_compare(&mp->m_icount, 0, XFS_ICOUNT_BATCH) < 0) {
1222 		ASSERT(0);
1223 		percpu_counter_add(&mp->m_icount, -delta);
1224 		return -EINVAL;
1225 	}
1226 	return 0;
1227 }
1228 
1229 int
1230 xfs_mod_ifree(
1231 	struct xfs_mount	*mp,
1232 	int64_t			delta)
1233 {
1234 	percpu_counter_add(&mp->m_ifree, delta);
1235 	if (percpu_counter_compare(&mp->m_ifree, 0) < 0) {
1236 		ASSERT(0);
1237 		percpu_counter_add(&mp->m_ifree, -delta);
1238 		return -EINVAL;
1239 	}
1240 	return 0;
1241 }
1242 
1243 /*
1244  * Deltas for the block count can vary from 1 to very large, but lock contention
1245  * only occurs on frequent small block count updates such as in the delayed
1246  * allocation path for buffered writes (page a time updates). Hence we set
1247  * a large batch count (1024) to minimise global counter updates except when
1248  * we get near to ENOSPC and we have to be very accurate with our updates.
1249  */
1250 #define XFS_FDBLOCKS_BATCH	1024
1251 int
1252 xfs_mod_fdblocks(
1253 	struct xfs_mount	*mp,
1254 	int64_t			delta,
1255 	bool			rsvd)
1256 {
1257 	int64_t			lcounter;
1258 	long long		res_used;
1259 	s32			batch;
1260 
1261 	if (delta > 0) {
1262 		/*
1263 		 * If the reserve pool is depleted, put blocks back into it
1264 		 * first. Most of the time the pool is full.
1265 		 */
1266 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1267 			percpu_counter_add(&mp->m_fdblocks, delta);
1268 			return 0;
1269 		}
1270 
1271 		spin_lock(&mp->m_sb_lock);
1272 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1273 
1274 		if (res_used > delta) {
1275 			mp->m_resblks_avail += delta;
1276 		} else {
1277 			delta -= res_used;
1278 			mp->m_resblks_avail = mp->m_resblks;
1279 			percpu_counter_add(&mp->m_fdblocks, delta);
1280 		}
1281 		spin_unlock(&mp->m_sb_lock);
1282 		return 0;
1283 	}
1284 
1285 	/*
1286 	 * Taking blocks away, need to be more accurate the closer we
1287 	 * are to zero.
1288 	 *
1289 	 * If the counter has a value of less than 2 * max batch size,
1290 	 * then make everything serialise as we are real close to
1291 	 * ENOSPC.
1292 	 */
1293 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1294 				     XFS_FDBLOCKS_BATCH) < 0)
1295 		batch = 1;
1296 	else
1297 		batch = XFS_FDBLOCKS_BATCH;
1298 
1299 	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1300 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1301 				     XFS_FDBLOCKS_BATCH) >= 0) {
1302 		/* we had space! */
1303 		return 0;
1304 	}
1305 
1306 	/*
1307 	 * lock up the sb for dipping into reserves before releasing the space
1308 	 * that took us to ENOSPC.
1309 	 */
1310 	spin_lock(&mp->m_sb_lock);
1311 	percpu_counter_add(&mp->m_fdblocks, -delta);
1312 	if (!rsvd)
1313 		goto fdblocks_enospc;
1314 
1315 	lcounter = (long long)mp->m_resblks_avail + delta;
1316 	if (lcounter >= 0) {
1317 		mp->m_resblks_avail = lcounter;
1318 		spin_unlock(&mp->m_sb_lock);
1319 		return 0;
1320 	}
1321 	printk_once(KERN_WARNING
1322 		"Filesystem \"%s\": reserve blocks depleted! "
1323 		"Consider increasing reserve pool size.",
1324 		mp->m_fsname);
1325 fdblocks_enospc:
1326 	spin_unlock(&mp->m_sb_lock);
1327 	return -ENOSPC;
1328 }
1329 
1330 int
1331 xfs_mod_frextents(
1332 	struct xfs_mount	*mp,
1333 	int64_t			delta)
1334 {
1335 	int64_t			lcounter;
1336 	int			ret = 0;
1337 
1338 	spin_lock(&mp->m_sb_lock);
1339 	lcounter = mp->m_sb.sb_frextents + delta;
1340 	if (lcounter < 0)
1341 		ret = -ENOSPC;
1342 	else
1343 		mp->m_sb.sb_frextents = lcounter;
1344 	spin_unlock(&mp->m_sb_lock);
1345 	return ret;
1346 }
1347 
1348 /*
1349  * xfs_getsb() is called to obtain the buffer for the superblock.
1350  * The buffer is returned locked and read in from disk.
1351  * The buffer should be released with a call to xfs_brelse().
1352  *
1353  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1354  * the superblock buffer if it can be locked without sleeping.
1355  * If it can't then we'll return NULL.
1356  */
1357 struct xfs_buf *
1358 xfs_getsb(
1359 	struct xfs_mount	*mp,
1360 	int			flags)
1361 {
1362 	struct xfs_buf		*bp = mp->m_sb_bp;
1363 
1364 	if (!xfs_buf_trylock(bp)) {
1365 		if (flags & XBF_TRYLOCK)
1366 			return NULL;
1367 		xfs_buf_lock(bp);
1368 	}
1369 
1370 	xfs_buf_hold(bp);
1371 	ASSERT(bp->b_flags & XBF_DONE);
1372 	return bp;
1373 }
1374 
1375 /*
1376  * Used to free the superblock along various error paths.
1377  */
1378 void
1379 xfs_freesb(
1380 	struct xfs_mount	*mp)
1381 {
1382 	struct xfs_buf		*bp = mp->m_sb_bp;
1383 
1384 	xfs_buf_lock(bp);
1385 	mp->m_sb_bp = NULL;
1386 	xfs_buf_relse(bp);
1387 }
1388 
1389 /*
1390  * If the underlying (data/log/rt) device is readonly, there are some
1391  * operations that cannot proceed.
1392  */
1393 int
1394 xfs_dev_is_read_only(
1395 	struct xfs_mount	*mp,
1396 	char			*message)
1397 {
1398 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1399 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1400 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1401 		xfs_notice(mp, "%s required on read-only device.", message);
1402 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1403 		return -EROFS;
1404 	}
1405 	return 0;
1406 }
1407