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