• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_bit.h"
13 #include "xfs_sb.h"
14 #include "xfs_mount.h"
15 #include "xfs_inode.h"
16 #include "xfs_dir2.h"
17 #include "xfs_ialloc.h"
18 #include "xfs_alloc.h"
19 #include "xfs_rtalloc.h"
20 #include "xfs_bmap.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_priv.h"
23 #include "xfs_log.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_fsops.h"
27 #include "xfs_icache.h"
28 #include "xfs_sysfs.h"
29 #include "xfs_rmap_btree.h"
30 #include "xfs_refcount_btree.h"
31 #include "xfs_reflink.h"
32 #include "xfs_extent_busy.h"
33 #include "xfs_health.h"
34 #include "xfs_trace.h"
35 
36 static DEFINE_MUTEX(xfs_uuid_table_mutex);
37 static int xfs_uuid_table_size;
38 static uuid_t *xfs_uuid_table;
39 
40 void
xfs_uuid_table_free(void)41 xfs_uuid_table_free(void)
42 {
43 	if (xfs_uuid_table_size == 0)
44 		return;
45 	kmem_free(xfs_uuid_table);
46 	xfs_uuid_table = NULL;
47 	xfs_uuid_table_size = 0;
48 }
49 
50 /*
51  * See if the UUID is unique among mounted XFS filesystems.
52  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
53  */
54 STATIC int
xfs_uuid_mount(struct xfs_mount * mp)55 xfs_uuid_mount(
56 	struct xfs_mount	*mp)
57 {
58 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
59 	int			hole, i;
60 
61 	/* Publish UUID in struct super_block */
62 	uuid_copy(&mp->m_super->s_uuid, uuid);
63 
64 	if (mp->m_flags & XFS_MOUNT_NOUUID)
65 		return 0;
66 
67 	if (uuid_is_null(uuid)) {
68 		xfs_warn(mp, "Filesystem has null UUID - can't mount");
69 		return -EINVAL;
70 	}
71 
72 	mutex_lock(&xfs_uuid_table_mutex);
73 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
74 		if (uuid_is_null(&xfs_uuid_table[i])) {
75 			hole = i;
76 			continue;
77 		}
78 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
79 			goto out_duplicate;
80 	}
81 
82 	if (hole < 0) {
83 		xfs_uuid_table = krealloc(xfs_uuid_table,
84 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
85 			GFP_KERNEL | __GFP_NOFAIL);
86 		hole = xfs_uuid_table_size++;
87 	}
88 	xfs_uuid_table[hole] = *uuid;
89 	mutex_unlock(&xfs_uuid_table_mutex);
90 
91 	return 0;
92 
93  out_duplicate:
94 	mutex_unlock(&xfs_uuid_table_mutex);
95 	xfs_warn(mp, "Filesystem has duplicate UUID %pU - can't mount", uuid);
96 	return -EINVAL;
97 }
98 
99 STATIC void
xfs_uuid_unmount(struct xfs_mount * mp)100 xfs_uuid_unmount(
101 	struct xfs_mount	*mp)
102 {
103 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
104 	int			i;
105 
106 	if (mp->m_flags & XFS_MOUNT_NOUUID)
107 		return;
108 
109 	mutex_lock(&xfs_uuid_table_mutex);
110 	for (i = 0; i < xfs_uuid_table_size; i++) {
111 		if (uuid_is_null(&xfs_uuid_table[i]))
112 			continue;
113 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
114 			continue;
115 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
116 		break;
117 	}
118 	ASSERT(i < xfs_uuid_table_size);
119 	mutex_unlock(&xfs_uuid_table_mutex);
120 }
121 
122 
123 STATIC void
__xfs_free_perag(struct rcu_head * head)124 __xfs_free_perag(
125 	struct rcu_head	*head)
126 {
127 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
128 
129 	ASSERT(atomic_read(&pag->pag_ref) == 0);
130 	kmem_free(pag);
131 }
132 
133 /*
134  * Free up the per-ag resources associated with the mount structure.
135  */
136 STATIC void
xfs_free_perag(xfs_mount_t * mp)137 xfs_free_perag(
138 	xfs_mount_t	*mp)
139 {
140 	xfs_agnumber_t	agno;
141 	struct xfs_perag *pag;
142 
143 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
144 		spin_lock(&mp->m_perag_lock);
145 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
146 		spin_unlock(&mp->m_perag_lock);
147 		ASSERT(pag);
148 		ASSERT(atomic_read(&pag->pag_ref) == 0);
149 		xfs_iunlink_destroy(pag);
150 		xfs_buf_hash_destroy(pag);
151 		call_rcu(&pag->rcu_head, __xfs_free_perag);
152 	}
153 }
154 
155 /*
156  * Check size of device based on the (data/realtime) block count.
157  * Note: this check is used by the growfs code as well as mount.
158  */
159 int
xfs_sb_validate_fsb_count(xfs_sb_t * sbp,uint64_t nblocks)160 xfs_sb_validate_fsb_count(
161 	xfs_sb_t	*sbp,
162 	uint64_t	nblocks)
163 {
164 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
165 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
166 
167 	/* Limited by ULONG_MAX of page cache index */
168 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
169 		return -EFBIG;
170 	return 0;
171 }
172 
173 int
xfs_initialize_perag(xfs_mount_t * mp,xfs_agnumber_t agcount,xfs_agnumber_t * maxagi)174 xfs_initialize_perag(
175 	xfs_mount_t	*mp,
176 	xfs_agnumber_t	agcount,
177 	xfs_agnumber_t	*maxagi)
178 {
179 	xfs_agnumber_t	index;
180 	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
181 	xfs_perag_t	*pag;
182 	int		error = -ENOMEM;
183 
184 	/*
185 	 * Walk the current per-ag tree so we don't try to initialise AGs
186 	 * that already exist (growfs case). Allocate and insert all the
187 	 * AGs we don't find ready for initialisation.
188 	 */
189 	for (index = 0; index < agcount; index++) {
190 		pag = xfs_perag_get(mp, index);
191 		if (pag) {
192 			xfs_perag_put(pag);
193 			continue;
194 		}
195 
196 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
197 		if (!pag) {
198 			error = -ENOMEM;
199 			goto out_unwind_new_pags;
200 		}
201 		pag->pag_agno = index;
202 		pag->pag_mount = mp;
203 		spin_lock_init(&pag->pag_ici_lock);
204 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
205 
206 		error = xfs_buf_hash_init(pag);
207 		if (error)
208 			goto out_free_pag;
209 		init_waitqueue_head(&pag->pagb_wait);
210 		spin_lock_init(&pag->pagb_lock);
211 		pag->pagb_count = 0;
212 		pag->pagb_tree = RB_ROOT;
213 
214 		error = radix_tree_preload(GFP_NOFS);
215 		if (error)
216 			goto out_hash_destroy;
217 
218 		spin_lock(&mp->m_perag_lock);
219 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
220 			WARN_ON_ONCE(1);
221 			spin_unlock(&mp->m_perag_lock);
222 			radix_tree_preload_end();
223 			error = -EEXIST;
224 			goto out_hash_destroy;
225 		}
226 		spin_unlock(&mp->m_perag_lock);
227 		radix_tree_preload_end();
228 		/* first new pag is fully initialized */
229 		if (first_initialised == NULLAGNUMBER)
230 			first_initialised = index;
231 		error = xfs_iunlink_init(pag);
232 		if (error)
233 			goto out_hash_destroy;
234 		spin_lock_init(&pag->pag_state_lock);
235 	}
236 
237 	index = xfs_set_inode_alloc(mp, agcount);
238 
239 	if (maxagi)
240 		*maxagi = index;
241 
242 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
243 	return 0;
244 
245 out_hash_destroy:
246 	xfs_buf_hash_destroy(pag);
247 out_free_pag:
248 	kmem_free(pag);
249 out_unwind_new_pags:
250 	/* unwind any prior newly initialized pags */
251 	for (index = first_initialised; index < agcount; index++) {
252 		pag = radix_tree_delete(&mp->m_perag_tree, index);
253 		if (!pag)
254 			break;
255 		xfs_buf_hash_destroy(pag);
256 		xfs_iunlink_destroy(pag);
257 		kmem_free(pag);
258 	}
259 	return error;
260 }
261 
262 /*
263  * xfs_readsb
264  *
265  * Does the initial read of the superblock.
266  */
267 int
xfs_readsb(struct xfs_mount * mp,int flags)268 xfs_readsb(
269 	struct xfs_mount *mp,
270 	int		flags)
271 {
272 	unsigned int	sector_size;
273 	struct xfs_buf	*bp;
274 	struct xfs_sb	*sbp = &mp->m_sb;
275 	int		error;
276 	int		loud = !(flags & XFS_MFSI_QUIET);
277 	const struct xfs_buf_ops *buf_ops;
278 
279 	ASSERT(mp->m_sb_bp == NULL);
280 	ASSERT(mp->m_ddev_targp != NULL);
281 
282 	/*
283 	 * For the initial read, we must guess at the sector
284 	 * size based on the block device.  It's enough to
285 	 * get the sb_sectsize out of the superblock and
286 	 * then reread with the proper length.
287 	 * We don't verify it yet, because it may not be complete.
288 	 */
289 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
290 	buf_ops = NULL;
291 
292 	/*
293 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
294 	 * around at all times to optimize access to the superblock. Therefore,
295 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
296 	 * elevated.
297 	 */
298 reread:
299 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
300 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
301 				      buf_ops);
302 	if (error) {
303 		if (loud)
304 			xfs_warn(mp, "SB validate failed with error %d.", error);
305 		/* bad CRC means corrupted metadata */
306 		if (error == -EFSBADCRC)
307 			error = -EFSCORRUPTED;
308 		return error;
309 	}
310 
311 	/*
312 	 * Initialize the mount structure from the superblock.
313 	 */
314 	xfs_sb_from_disk(sbp, bp->b_addr);
315 
316 	/*
317 	 * If we haven't validated the superblock, do so now before we try
318 	 * to check the sector size and reread the superblock appropriately.
319 	 */
320 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
321 		if (loud)
322 			xfs_warn(mp, "Invalid superblock magic number");
323 		error = -EINVAL;
324 		goto release_buf;
325 	}
326 
327 	/*
328 	 * We must be able to do sector-sized and sector-aligned IO.
329 	 */
330 	if (sector_size > sbp->sb_sectsize) {
331 		if (loud)
332 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
333 				sector_size, sbp->sb_sectsize);
334 		error = -ENOSYS;
335 		goto release_buf;
336 	}
337 
338 	if (buf_ops == NULL) {
339 		/*
340 		 * Re-read the superblock so the buffer is correctly sized,
341 		 * and properly verified.
342 		 */
343 		xfs_buf_relse(bp);
344 		sector_size = sbp->sb_sectsize;
345 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
346 		goto reread;
347 	}
348 
349 	xfs_reinit_percpu_counters(mp);
350 
351 	/* no need to be quiet anymore, so reset the buf ops */
352 	bp->b_ops = &xfs_sb_buf_ops;
353 
354 	mp->m_sb_bp = bp;
355 	xfs_buf_unlock(bp);
356 	return 0;
357 
358 release_buf:
359 	xfs_buf_relse(bp);
360 	return error;
361 }
362 
363 /*
364  * If the sunit/swidth change would move the precomputed root inode value, we
365  * must reject the ondisk change because repair will stumble over that.
366  * However, we allow the mount to proceed because we never rejected this
367  * combination before.  Returns true to update the sb, false otherwise.
368  */
369 static inline int
xfs_check_new_dalign(struct xfs_mount * mp,int new_dalign,bool * update_sb)370 xfs_check_new_dalign(
371 	struct xfs_mount	*mp,
372 	int			new_dalign,
373 	bool			*update_sb)
374 {
375 	struct xfs_sb		*sbp = &mp->m_sb;
376 	xfs_ino_t		calc_ino;
377 
378 	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
379 	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
380 
381 	if (sbp->sb_rootino == calc_ino) {
382 		*update_sb = true;
383 		return 0;
384 	}
385 
386 	xfs_warn(mp,
387 "Cannot change stripe alignment; would require moving root inode.");
388 
389 	/*
390 	 * XXX: Next time we add a new incompat feature, this should start
391 	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
392 	 * that we're ignoring the administrator's instructions.
393 	 */
394 	xfs_warn(mp, "Skipping superblock stripe alignment update.");
395 	*update_sb = false;
396 	return 0;
397 }
398 
399 /*
400  * If we were provided with new sunit/swidth values as mount options, make sure
401  * that they pass basic alignment and superblock feature checks, and convert
402  * them into the same units (FSB) that everything else expects.  This step
403  * /must/ be done before computing the inode geometry.
404  */
405 STATIC int
xfs_validate_new_dalign(struct xfs_mount * mp)406 xfs_validate_new_dalign(
407 	struct xfs_mount	*mp)
408 {
409 	if (mp->m_dalign == 0)
410 		return 0;
411 
412 	/*
413 	 * If stripe unit and stripe width are not multiples
414 	 * of the fs blocksize turn off alignment.
415 	 */
416 	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
417 	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
418 		xfs_warn(mp,
419 	"alignment check failed: sunit/swidth vs. blocksize(%d)",
420 			mp->m_sb.sb_blocksize);
421 		return -EINVAL;
422 	} else {
423 		/*
424 		 * Convert the stripe unit and width to FSBs.
425 		 */
426 		mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
427 		if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
428 			xfs_warn(mp,
429 		"alignment check failed: sunit/swidth vs. agsize(%d)",
430 				 mp->m_sb.sb_agblocks);
431 			return -EINVAL;
432 		} else if (mp->m_dalign) {
433 			mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
434 		} else {
435 			xfs_warn(mp,
436 		"alignment check failed: sunit(%d) less than bsize(%d)",
437 				 mp->m_dalign, mp->m_sb.sb_blocksize);
438 			return -EINVAL;
439 		}
440 	}
441 
442 	if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
443 		xfs_warn(mp,
444 "cannot change alignment: superblock does not support data alignment");
445 		return -EINVAL;
446 	}
447 
448 	return 0;
449 }
450 
451 /* Update alignment values based on mount options and sb values. */
452 STATIC int
xfs_update_alignment(struct xfs_mount * mp)453 xfs_update_alignment(
454 	struct xfs_mount	*mp)
455 {
456 	struct xfs_sb		*sbp = &mp->m_sb;
457 
458 	if (mp->m_dalign) {
459 		bool		update_sb;
460 		int		error;
461 
462 		if (sbp->sb_unit == mp->m_dalign &&
463 		    sbp->sb_width == mp->m_swidth)
464 			return 0;
465 
466 		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
467 		if (error || !update_sb)
468 			return error;
469 
470 		sbp->sb_unit = mp->m_dalign;
471 		sbp->sb_width = mp->m_swidth;
472 		mp->m_update_sb = true;
473 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
474 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
475 		mp->m_dalign = sbp->sb_unit;
476 		mp->m_swidth = sbp->sb_width;
477 	}
478 
479 	return 0;
480 }
481 
482 /*
483  * precalculate the low space thresholds for dynamic speculative preallocation.
484  */
485 void
xfs_set_low_space_thresholds(struct xfs_mount * mp)486 xfs_set_low_space_thresholds(
487 	struct xfs_mount	*mp)
488 {
489 	int i;
490 
491 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
492 		uint64_t space = mp->m_sb.sb_dblocks;
493 
494 		do_div(space, 100);
495 		mp->m_low_space[i] = space * (i + 1);
496 	}
497 }
498 
499 /*
500  * Check that the data (and log if separate) is an ok size.
501  */
502 STATIC int
xfs_check_sizes(struct xfs_mount * mp)503 xfs_check_sizes(
504 	struct xfs_mount *mp)
505 {
506 	struct xfs_buf	*bp;
507 	xfs_daddr_t	d;
508 	int		error;
509 
510 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
511 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
512 		xfs_warn(mp, "filesystem size mismatch detected");
513 		return -EFBIG;
514 	}
515 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
516 					d - XFS_FSS_TO_BB(mp, 1),
517 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
518 	if (error) {
519 		xfs_warn(mp, "last sector read failed");
520 		return error;
521 	}
522 	xfs_buf_relse(bp);
523 
524 	if (mp->m_logdev_targp == mp->m_ddev_targp)
525 		return 0;
526 
527 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
528 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
529 		xfs_warn(mp, "log size mismatch detected");
530 		return -EFBIG;
531 	}
532 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
533 					d - XFS_FSB_TO_BB(mp, 1),
534 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
535 	if (error) {
536 		xfs_warn(mp, "log device read failed");
537 		return error;
538 	}
539 	xfs_buf_relse(bp);
540 	return 0;
541 }
542 
543 /*
544  * Clear the quotaflags in memory and in the superblock.
545  */
546 int
xfs_mount_reset_sbqflags(struct xfs_mount * mp)547 xfs_mount_reset_sbqflags(
548 	struct xfs_mount	*mp)
549 {
550 	mp->m_qflags = 0;
551 
552 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
553 	if (mp->m_sb.sb_qflags == 0)
554 		return 0;
555 	spin_lock(&mp->m_sb_lock);
556 	mp->m_sb.sb_qflags = 0;
557 	spin_unlock(&mp->m_sb_lock);
558 
559 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
560 		return 0;
561 
562 	return xfs_sync_sb(mp, false);
563 }
564 
565 uint64_t
xfs_default_resblks(xfs_mount_t * mp)566 xfs_default_resblks(xfs_mount_t *mp)
567 {
568 	uint64_t resblks;
569 
570 	/*
571 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
572 	 * smaller.  This is intended to cover concurrent allocation
573 	 * transactions when we initially hit enospc. These each require a 4
574 	 * block reservation. Hence by default we cover roughly 2000 concurrent
575 	 * allocation reservations.
576 	 */
577 	resblks = mp->m_sb.sb_dblocks;
578 	do_div(resblks, 20);
579 	resblks = min_t(uint64_t, resblks, 8192);
580 	return resblks;
581 }
582 
583 /* Ensure the summary counts are correct. */
584 STATIC int
xfs_check_summary_counts(struct xfs_mount * mp)585 xfs_check_summary_counts(
586 	struct xfs_mount	*mp)
587 {
588 	/*
589 	 * The AG0 superblock verifier rejects in-progress filesystems,
590 	 * so we should never see the flag set this far into mounting.
591 	 */
592 	if (mp->m_sb.sb_inprogress) {
593 		xfs_err(mp, "sb_inprogress set after log recovery??");
594 		WARN_ON(1);
595 		return -EFSCORRUPTED;
596 	}
597 
598 	/*
599 	 * Now the log is mounted, we know if it was an unclean shutdown or
600 	 * not. If it was, with the first phase of recovery has completed, we
601 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
602 	 * but they are recovered transactionally in the second recovery phase
603 	 * later.
604 	 *
605 	 * If the log was clean when we mounted, we can check the summary
606 	 * counters.  If any of them are obviously incorrect, we can recompute
607 	 * them from the AGF headers in the next step.
608 	 */
609 	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
610 	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
611 	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
612 	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
613 		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
614 
615 	/*
616 	 * We can safely re-initialise incore superblock counters from the
617 	 * per-ag data. These may not be correct if the filesystem was not
618 	 * cleanly unmounted, so we waited for recovery to finish before doing
619 	 * this.
620 	 *
621 	 * If the filesystem was cleanly unmounted or the previous check did
622 	 * not flag anything weird, then we can trust the values in the
623 	 * superblock to be correct and we don't need to do anything here.
624 	 * Otherwise, recalculate the summary counters.
625 	 */
626 	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
627 	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
628 	    !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
629 		return 0;
630 
631 	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
632 }
633 
634 /*
635  * This function does the following on an initial mount of a file system:
636  *	- reads the superblock from disk and init the mount struct
637  *	- if we're a 32-bit kernel, do a size check on the superblock
638  *		so we don't mount terabyte filesystems
639  *	- init mount struct realtime fields
640  *	- allocate inode hash table for fs
641  *	- init directory manager
642  *	- perform recovery and init the log manager
643  */
644 int
xfs_mountfs(struct xfs_mount * mp)645 xfs_mountfs(
646 	struct xfs_mount	*mp)
647 {
648 	struct xfs_sb		*sbp = &(mp->m_sb);
649 	struct xfs_inode	*rip;
650 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
651 	uint64_t		resblks;
652 	uint			quotamount = 0;
653 	uint			quotaflags = 0;
654 	int			error = 0;
655 
656 	xfs_sb_mount_common(mp, sbp);
657 
658 	/*
659 	 * Check for a mismatched features2 values.  Older kernels read & wrote
660 	 * into the wrong sb offset for sb_features2 on some platforms due to
661 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
662 	 * which made older superblock reading/writing routines swap it as a
663 	 * 64-bit value.
664 	 *
665 	 * For backwards compatibility, we make both slots equal.
666 	 *
667 	 * If we detect a mismatched field, we OR the set bits into the existing
668 	 * features2 field in case it has already been modified; we don't want
669 	 * to lose any features.  We then update the bad location with the ORed
670 	 * value so that older kernels will see any features2 flags. The
671 	 * superblock writeback code ensures the new sb_features2 is copied to
672 	 * sb_bad_features2 before it is logged or written to disk.
673 	 */
674 	if (xfs_sb_has_mismatched_features2(sbp)) {
675 		xfs_warn(mp, "correcting sb_features alignment problem");
676 		sbp->sb_features2 |= sbp->sb_bad_features2;
677 		mp->m_update_sb = true;
678 
679 		/*
680 		 * Re-check for ATTR2 in case it was found in bad_features2
681 		 * slot.
682 		 */
683 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
684 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
685 			mp->m_flags |= XFS_MOUNT_ATTR2;
686 	}
687 
688 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
689 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
690 		xfs_sb_version_removeattr2(&mp->m_sb);
691 		mp->m_update_sb = true;
692 
693 		/* update sb_versionnum for the clearing of the morebits */
694 		if (!sbp->sb_features2)
695 			mp->m_update_sb = true;
696 	}
697 
698 	/* always use v2 inodes by default now */
699 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
700 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
701 		mp->m_update_sb = true;
702 	}
703 
704 	/*
705 	 * If we were given new sunit/swidth options, do some basic validation
706 	 * checks and convert the incore dalign and swidth values to the
707 	 * same units (FSB) that everything else uses.  This /must/ happen
708 	 * before computing the inode geometry.
709 	 */
710 	error = xfs_validate_new_dalign(mp);
711 	if (error)
712 		goto out;
713 
714 	xfs_alloc_compute_maxlevels(mp);
715 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
716 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
717 	xfs_ialloc_setup_geometry(mp);
718 	xfs_rmapbt_compute_maxlevels(mp);
719 	xfs_refcountbt_compute_maxlevels(mp);
720 
721 	/*
722 	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
723 	 * is NOT aligned turn off m_dalign since allocator alignment is within
724 	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
725 	 * we must compute the free space and rmap btree geometry before doing
726 	 * this.
727 	 */
728 	error = xfs_update_alignment(mp);
729 	if (error)
730 		goto out;
731 
732 	/* enable fail_at_unmount as default */
733 	mp->m_fail_unmount = true;
734 
735 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
736 			       NULL, mp->m_super->s_id);
737 	if (error)
738 		goto out;
739 
740 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
741 			       &mp->m_kobj, "stats");
742 	if (error)
743 		goto out_remove_sysfs;
744 
745 	error = xfs_error_sysfs_init(mp);
746 	if (error)
747 		goto out_del_stats;
748 
749 	error = xfs_errortag_init(mp);
750 	if (error)
751 		goto out_remove_error_sysfs;
752 
753 	error = xfs_uuid_mount(mp);
754 	if (error)
755 		goto out_remove_errortag;
756 
757 	/*
758 	 * Update the preferred write size based on the information from the
759 	 * on-disk superblock.
760 	 */
761 	mp->m_allocsize_log =
762 		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
763 	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
764 
765 	/* set the low space thresholds for dynamic preallocation */
766 	xfs_set_low_space_thresholds(mp);
767 
768 	/*
769 	 * If enabled, sparse inode chunk alignment is expected to match the
770 	 * cluster size. Full inode chunk alignment must match the chunk size,
771 	 * but that is checked on sb read verification...
772 	 */
773 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
774 	    mp->m_sb.sb_spino_align !=
775 			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
776 		xfs_warn(mp,
777 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
778 			 mp->m_sb.sb_spino_align,
779 			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
780 		error = -EINVAL;
781 		goto out_remove_uuid;
782 	}
783 
784 	/*
785 	 * Check that the data (and log if separate) is an ok size.
786 	 */
787 	error = xfs_check_sizes(mp);
788 	if (error)
789 		goto out_remove_uuid;
790 
791 	/*
792 	 * Initialize realtime fields in the mount structure
793 	 */
794 	error = xfs_rtmount_init(mp);
795 	if (error) {
796 		xfs_warn(mp, "RT mount failed");
797 		goto out_remove_uuid;
798 	}
799 
800 	/*
801 	 *  Copies the low order bits of the timestamp and the randomly
802 	 *  set "sequence" number out of a UUID.
803 	 */
804 	mp->m_fixedfsid[0] =
805 		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
806 		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
807 	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
808 
809 	error = xfs_da_mount(mp);
810 	if (error) {
811 		xfs_warn(mp, "Failed dir/attr init: %d", error);
812 		goto out_remove_uuid;
813 	}
814 
815 	/*
816 	 * Initialize the precomputed transaction reservations values.
817 	 */
818 	xfs_trans_init(mp);
819 
820 	/*
821 	 * Allocate and initialize the per-ag data.
822 	 */
823 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
824 	if (error) {
825 		xfs_warn(mp, "Failed per-ag init: %d", error);
826 		goto out_free_dir;
827 	}
828 
829 	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
830 		xfs_warn(mp, "no log defined");
831 		error = -EFSCORRUPTED;
832 		goto out_free_perag;
833 	}
834 
835 	/*
836 	 * Log's mount-time initialization. The first part of recovery can place
837 	 * some items on the AIL, to be handled when recovery is finished or
838 	 * cancelled.
839 	 */
840 	error = xfs_log_mount(mp, mp->m_logdev_targp,
841 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
842 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
843 	if (error) {
844 		xfs_warn(mp, "log mount failed");
845 		goto out_fail_wait;
846 	}
847 
848 	/* Make sure the summary counts are ok. */
849 	error = xfs_check_summary_counts(mp);
850 	if (error)
851 		goto out_log_dealloc;
852 
853 	/*
854 	 * Get and sanity-check the root inode.
855 	 * Save the pointer to it in the mount structure.
856 	 */
857 	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
858 			 XFS_ILOCK_EXCL, &rip);
859 	if (error) {
860 		xfs_warn(mp,
861 			"Failed to read root inode 0x%llx, error %d",
862 			sbp->sb_rootino, -error);
863 		goto out_log_dealloc;
864 	}
865 
866 	ASSERT(rip != NULL);
867 
868 	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
869 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
870 			(unsigned long long)rip->i_ino);
871 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
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 	xfs_irele(rip);
1005 	/* Clean out dquots that might be in memory after quotacheck. */
1006 	xfs_qm_unmount(mp);
1007 	/*
1008 	 * Cancel all delayed reclaim work and reclaim the inodes directly.
1009 	 * We have to do this /after/ rtunmount and qm_unmount because those
1010 	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1011 	 *
1012 	 * This is slightly different from the unmountfs call sequence
1013 	 * because we could be tearing down a partially set up mount.  In
1014 	 * particular, if log_mount_finish fails we bail out without calling
1015 	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1016 	 * quota inodes.
1017 	 */
1018 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1019 	xfs_reclaim_inodes(mp);
1020 	xfs_health_unmount(mp);
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_errortag:
1035 	xfs_errortag_del(mp);
1036  out_remove_error_sysfs:
1037 	xfs_error_sysfs_del(mp);
1038  out_del_stats:
1039 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1040  out_remove_sysfs:
1041 	xfs_sysfs_del(&mp->m_kobj);
1042  out:
1043 	return error;
1044 }
1045 
1046 /*
1047  * This flushes out the inodes,dquots and the superblock, unmounts the
1048  * log and makes sure that incore structures are freed.
1049  */
1050 void
xfs_unmountfs(struct xfs_mount * mp)1051 xfs_unmountfs(
1052 	struct xfs_mount	*mp)
1053 {
1054 	uint64_t		resblks;
1055 	int			error;
1056 
1057 	xfs_stop_block_reaping(mp);
1058 	xfs_fs_unreserve_ag_blocks(mp);
1059 	xfs_qm_unmount_quotas(mp);
1060 	xfs_rtunmount_inodes(mp);
1061 	xfs_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 be pinned to the buffer until the
1068 	 * transaction hits the disk and the callbacks run. Pushing the AIL will
1069 	 * skip the stale inodes and may never see the pinned buffer, so
1070 	 * nothing will push out the iclog and unpin the buffer. Hence we
1071 	 * need to force the log here to ensure all items are flushed into the
1072 	 * AIL before we go any further.
1073 	 */
1074 	xfs_log_force(mp, XFS_LOG_SYNC);
1075 
1076 	/*
1077 	 * Wait for all busy extents to be freed, including completion of
1078 	 * any discard operation.
1079 	 */
1080 	xfs_extent_busy_wait_all(mp);
1081 	flush_workqueue(xfs_discard_wq);
1082 
1083 	/*
1084 	 * We now need to tell the world we are unmounting. This will allow
1085 	 * us to detect that the filesystem is going away and we should error
1086 	 * out anything that we have been retrying in the background. This will
1087 	 * prevent neverending retries in AIL pushing from hanging the unmount.
1088 	 */
1089 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
1090 
1091 	/*
1092 	 * Flush all pending changes from the AIL.
1093 	 */
1094 	xfs_ail_push_all_sync(mp->m_ail);
1095 
1096 	/*
1097 	 * Reclaim all inodes. At this point there should be no dirty inodes and
1098 	 * none should be pinned or locked. Stop background inode reclaim here
1099 	 * if it is still running.
1100 	 */
1101 	cancel_delayed_work_sync(&mp->m_reclaim_work);
1102 	xfs_reclaim_inodes(mp);
1103 	xfs_health_unmount(mp);
1104 
1105 	xfs_qm_unmount(mp);
1106 
1107 	/*
1108 	 * Unreserve any blocks we have so that when we unmount we don't account
1109 	 * the reserved free space as used. This is really only necessary for
1110 	 * lazy superblock counting because it trusts the incore superblock
1111 	 * counters to be absolutely correct on clean unmount.
1112 	 *
1113 	 * We don't bother correcting this elsewhere for lazy superblock
1114 	 * counting because on mount of an unclean filesystem we reconstruct the
1115 	 * correct counter value and this is irrelevant.
1116 	 *
1117 	 * For non-lazy counter filesystems, this doesn't matter at all because
1118 	 * we only every apply deltas to the superblock and hence the incore
1119 	 * value does not matter....
1120 	 */
1121 	resblks = 0;
1122 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1123 	if (error)
1124 		xfs_warn(mp, "Unable to free reserved block pool. "
1125 				"Freespace may not be correct on next mount.");
1126 
1127 	error = xfs_log_sbcount(mp);
1128 	if (error)
1129 		xfs_warn(mp, "Unable to update superblock counters. "
1130 				"Freespace may not be correct on next mount.");
1131 
1132 
1133 	xfs_log_unmount(mp);
1134 	xfs_da_unmount(mp);
1135 	xfs_uuid_unmount(mp);
1136 
1137 #if defined(DEBUG)
1138 	xfs_errortag_clearall(mp);
1139 #endif
1140 	xfs_free_perag(mp);
1141 
1142 	xfs_errortag_del(mp);
1143 	xfs_error_sysfs_del(mp);
1144 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1145 	xfs_sysfs_del(&mp->m_kobj);
1146 }
1147 
1148 /*
1149  * Determine whether modifications can proceed. The caller specifies the minimum
1150  * freeze level for which modifications should not be allowed. This allows
1151  * certain operations to proceed while the freeze sequence is in progress, if
1152  * necessary.
1153  */
1154 bool
xfs_fs_writable(struct xfs_mount * mp,int level)1155 xfs_fs_writable(
1156 	struct xfs_mount	*mp,
1157 	int			level)
1158 {
1159 	ASSERT(level > SB_UNFROZEN);
1160 	if ((mp->m_super->s_writers.frozen >= level) ||
1161 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1162 		return false;
1163 
1164 	return true;
1165 }
1166 
1167 /*
1168  * xfs_log_sbcount
1169  *
1170  * Sync the superblock counters to disk.
1171  *
1172  * Note this code can be called during the process of freezing, so we use the
1173  * transaction allocator that does not block when the transaction subsystem is
1174  * in its frozen state.
1175  */
1176 int
xfs_log_sbcount(xfs_mount_t * mp)1177 xfs_log_sbcount(xfs_mount_t *mp)
1178 {
1179 	/* allow this to proceed during the freeze sequence... */
1180 	if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE))
1181 		return 0;
1182 
1183 	/*
1184 	 * we don't need to do this if we are updating the superblock
1185 	 * counters on every modification.
1186 	 */
1187 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1188 		return 0;
1189 
1190 	return xfs_sync_sb(mp, true);
1191 }
1192 
1193 /*
1194  * Deltas for the block count can vary from 1 to very large, but lock contention
1195  * only occurs on frequent small block count updates such as in the delayed
1196  * allocation path for buffered writes (page a time updates). Hence we set
1197  * a large batch count (1024) to minimise global counter updates except when
1198  * we get near to ENOSPC and we have to be very accurate with our updates.
1199  */
1200 #define XFS_FDBLOCKS_BATCH	1024
1201 int
xfs_mod_fdblocks(struct xfs_mount * mp,int64_t delta,bool rsvd)1202 xfs_mod_fdblocks(
1203 	struct xfs_mount	*mp,
1204 	int64_t			delta,
1205 	bool			rsvd)
1206 {
1207 	int64_t			lcounter;
1208 	long long		res_used;
1209 	s32			batch;
1210 
1211 	if (delta > 0) {
1212 		/*
1213 		 * If the reserve pool is depleted, put blocks back into it
1214 		 * first. Most of the time the pool is full.
1215 		 */
1216 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1217 			percpu_counter_add(&mp->m_fdblocks, delta);
1218 			return 0;
1219 		}
1220 
1221 		spin_lock(&mp->m_sb_lock);
1222 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1223 
1224 		if (res_used > delta) {
1225 			mp->m_resblks_avail += delta;
1226 		} else {
1227 			delta -= res_used;
1228 			mp->m_resblks_avail = mp->m_resblks;
1229 			percpu_counter_add(&mp->m_fdblocks, delta);
1230 		}
1231 		spin_unlock(&mp->m_sb_lock);
1232 		return 0;
1233 	}
1234 
1235 	/*
1236 	 * Taking blocks away, need to be more accurate the closer we
1237 	 * are to zero.
1238 	 *
1239 	 * If the counter has a value of less than 2 * max batch size,
1240 	 * then make everything serialise as we are real close to
1241 	 * ENOSPC.
1242 	 */
1243 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1244 				     XFS_FDBLOCKS_BATCH) < 0)
1245 		batch = 1;
1246 	else
1247 		batch = XFS_FDBLOCKS_BATCH;
1248 
1249 	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1250 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1251 				     XFS_FDBLOCKS_BATCH) >= 0) {
1252 		/* we had space! */
1253 		return 0;
1254 	}
1255 
1256 	/*
1257 	 * lock up the sb for dipping into reserves before releasing the space
1258 	 * that took us to ENOSPC.
1259 	 */
1260 	spin_lock(&mp->m_sb_lock);
1261 	percpu_counter_add(&mp->m_fdblocks, -delta);
1262 	if (!rsvd)
1263 		goto fdblocks_enospc;
1264 
1265 	lcounter = (long long)mp->m_resblks_avail + delta;
1266 	if (lcounter >= 0) {
1267 		mp->m_resblks_avail = lcounter;
1268 		spin_unlock(&mp->m_sb_lock);
1269 		return 0;
1270 	}
1271 	xfs_warn_once(mp,
1272 "Reserve blocks depleted! Consider increasing reserve pool size.");
1273 
1274 fdblocks_enospc:
1275 	spin_unlock(&mp->m_sb_lock);
1276 	return -ENOSPC;
1277 }
1278 
1279 int
xfs_mod_frextents(struct xfs_mount * mp,int64_t delta)1280 xfs_mod_frextents(
1281 	struct xfs_mount	*mp,
1282 	int64_t			delta)
1283 {
1284 	int64_t			lcounter;
1285 	int			ret = 0;
1286 
1287 	spin_lock(&mp->m_sb_lock);
1288 	lcounter = mp->m_sb.sb_frextents + delta;
1289 	if (lcounter < 0)
1290 		ret = -ENOSPC;
1291 	else
1292 		mp->m_sb.sb_frextents = lcounter;
1293 	spin_unlock(&mp->m_sb_lock);
1294 	return ret;
1295 }
1296 
1297 /*
1298  * Used to free the superblock along various error paths.
1299  */
1300 void
xfs_freesb(struct xfs_mount * mp)1301 xfs_freesb(
1302 	struct xfs_mount	*mp)
1303 {
1304 	struct xfs_buf		*bp = mp->m_sb_bp;
1305 
1306 	xfs_buf_lock(bp);
1307 	mp->m_sb_bp = NULL;
1308 	xfs_buf_relse(bp);
1309 }
1310 
1311 /*
1312  * If the underlying (data/log/rt) device is readonly, there are some
1313  * operations that cannot proceed.
1314  */
1315 int
xfs_dev_is_read_only(struct xfs_mount * mp,char * message)1316 xfs_dev_is_read_only(
1317 	struct xfs_mount	*mp,
1318 	char			*message)
1319 {
1320 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1321 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1322 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1323 		xfs_notice(mp, "%s required on read-only device.", message);
1324 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1325 		return -EROFS;
1326 	}
1327 	return 0;
1328 }
1329 
1330 /* Force the summary counters to be recalculated at next mount. */
1331 void
xfs_force_summary_recalc(struct xfs_mount * mp)1332 xfs_force_summary_recalc(
1333 	struct xfs_mount	*mp)
1334 {
1335 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1336 		return;
1337 
1338 	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1339 }
1340 
1341 /*
1342  * Update the in-core delayed block counter.
1343  *
1344  * We prefer to update the counter without having to take a spinlock for every
1345  * counter update (i.e. batching).  Each change to delayed allocation
1346  * reservations can change can easily exceed the default percpu counter
1347  * batching, so we use a larger batch factor here.
1348  *
1349  * Note that we don't currently have any callers requiring fast summation
1350  * (e.g. percpu_counter_read) so we can use a big batch value here.
1351  */
1352 #define XFS_DELALLOC_BATCH	(4096)
1353 void
xfs_mod_delalloc(struct xfs_mount * mp,int64_t delta)1354 xfs_mod_delalloc(
1355 	struct xfs_mount	*mp,
1356 	int64_t			delta)
1357 {
1358 	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1359 			XFS_DELALLOC_BATCH);
1360 }
1361