• 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 	kmem_free(pag);
130 }
131 
132 /*
133  * Free up the per-ag resources associated with the mount structure.
134  */
135 STATIC void
xfs_free_perag(xfs_mount_t * mp)136 xfs_free_perag(
137 	xfs_mount_t	*mp)
138 {
139 	xfs_agnumber_t	agno;
140 	struct xfs_perag *pag;
141 
142 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
143 		spin_lock(&mp->m_perag_lock);
144 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
145 		spin_unlock(&mp->m_perag_lock);
146 		ASSERT(pag);
147 		XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0);
148 		xfs_iunlink_destroy(pag);
149 		xfs_buf_hash_destroy(pag);
150 		call_rcu(&pag->rcu_head, __xfs_free_perag);
151 	}
152 }
153 
154 /*
155  * Check size of device based on the (data/realtime) block count.
156  * Note: this check is used by the growfs code as well as mount.
157  */
158 int
xfs_sb_validate_fsb_count(xfs_sb_t * sbp,uint64_t nblocks)159 xfs_sb_validate_fsb_count(
160 	xfs_sb_t	*sbp,
161 	uint64_t	nblocks)
162 {
163 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
164 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
165 
166 	/* Limited by ULONG_MAX of page cache index */
167 	if (nblocks >> (PAGE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
168 		return -EFBIG;
169 	return 0;
170 }
171 
172 int
xfs_initialize_perag(xfs_mount_t * mp,xfs_agnumber_t agcount,xfs_agnumber_t * maxagi)173 xfs_initialize_perag(
174 	xfs_mount_t	*mp,
175 	xfs_agnumber_t	agcount,
176 	xfs_agnumber_t	*maxagi)
177 {
178 	xfs_agnumber_t	index;
179 	xfs_agnumber_t	first_initialised = NULLAGNUMBER;
180 	xfs_perag_t	*pag;
181 	int		error = -ENOMEM;
182 
183 	/*
184 	 * Walk the current per-ag tree so we don't try to initialise AGs
185 	 * that already exist (growfs case). Allocate and insert all the
186 	 * AGs we don't find ready for initialisation.
187 	 */
188 	for (index = 0; index < agcount; index++) {
189 		pag = xfs_perag_get(mp, index);
190 		if (pag) {
191 			xfs_perag_put(pag);
192 			continue;
193 		}
194 
195 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
196 		if (!pag) {
197 			error = -ENOMEM;
198 			goto out_unwind_new_pags;
199 		}
200 		pag->pag_agno = index;
201 		pag->pag_mount = mp;
202 		spin_lock_init(&pag->pag_ici_lock);
203 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
204 
205 		error = xfs_buf_hash_init(pag);
206 		if (error)
207 			goto out_free_pag;
208 		init_waitqueue_head(&pag->pagb_wait);
209 		spin_lock_init(&pag->pagb_lock);
210 		pag->pagb_count = 0;
211 		pag->pagb_tree = RB_ROOT;
212 
213 		error = radix_tree_preload(GFP_NOFS);
214 		if (error)
215 			goto out_hash_destroy;
216 
217 		spin_lock(&mp->m_perag_lock);
218 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
219 			WARN_ON_ONCE(1);
220 			spin_unlock(&mp->m_perag_lock);
221 			radix_tree_preload_end();
222 			error = -EEXIST;
223 			goto out_hash_destroy;
224 		}
225 		spin_unlock(&mp->m_perag_lock);
226 		radix_tree_preload_end();
227 		/* first new pag is fully initialized */
228 		if (first_initialised == NULLAGNUMBER)
229 			first_initialised = index;
230 		error = xfs_iunlink_init(pag);
231 		if (error)
232 			goto out_hash_destroy;
233 		spin_lock_init(&pag->pag_state_lock);
234 	}
235 
236 	index = xfs_set_inode_alloc(mp, agcount);
237 
238 	if (maxagi)
239 		*maxagi = index;
240 
241 	mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
242 	return 0;
243 
244 out_hash_destroy:
245 	xfs_buf_hash_destroy(pag);
246 out_free_pag:
247 	kmem_free(pag);
248 out_unwind_new_pags:
249 	/* unwind any prior newly initialized pags */
250 	for (index = first_initialised; index < agcount; index++) {
251 		pag = radix_tree_delete(&mp->m_perag_tree, index);
252 		if (!pag)
253 			break;
254 		xfs_buf_hash_destroy(pag);
255 		xfs_iunlink_destroy(pag);
256 		kmem_free(pag);
257 	}
258 	return error;
259 }
260 
261 /*
262  * xfs_readsb
263  *
264  * Does the initial read of the superblock.
265  */
266 int
xfs_readsb(struct xfs_mount * mp,int flags)267 xfs_readsb(
268 	struct xfs_mount *mp,
269 	int		flags)
270 {
271 	unsigned int	sector_size;
272 	struct xfs_buf	*bp;
273 	struct xfs_sb	*sbp = &mp->m_sb;
274 	int		error;
275 	int		loud = !(flags & XFS_MFSI_QUIET);
276 	const struct xfs_buf_ops *buf_ops;
277 
278 	ASSERT(mp->m_sb_bp == NULL);
279 	ASSERT(mp->m_ddev_targp != NULL);
280 
281 	/*
282 	 * For the initial read, we must guess at the sector
283 	 * size based on the block device.  It's enough to
284 	 * get the sb_sectsize out of the superblock and
285 	 * then reread with the proper length.
286 	 * We don't verify it yet, because it may not be complete.
287 	 */
288 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
289 	buf_ops = NULL;
290 
291 	/*
292 	 * Allocate a (locked) buffer to hold the superblock. This will be kept
293 	 * around at all times to optimize access to the superblock. Therefore,
294 	 * set XBF_NO_IOACCT to make sure it doesn't hold the buftarg count
295 	 * elevated.
296 	 */
297 reread:
298 	error = xfs_buf_read_uncached(mp->m_ddev_targp, XFS_SB_DADDR,
299 				      BTOBB(sector_size), XBF_NO_IOACCT, &bp,
300 				      buf_ops);
301 	if (error) {
302 		if (loud)
303 			xfs_warn(mp, "SB validate failed with error %d.", error);
304 		/* bad CRC means corrupted metadata */
305 		if (error == -EFSBADCRC)
306 			error = -EFSCORRUPTED;
307 		return error;
308 	}
309 
310 	/*
311 	 * Initialize the mount structure from the superblock.
312 	 */
313 	xfs_sb_from_disk(sbp, bp->b_addr);
314 
315 	/*
316 	 * If we haven't validated the superblock, do so now before we try
317 	 * to check the sector size and reread the superblock appropriately.
318 	 */
319 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
320 		if (loud)
321 			xfs_warn(mp, "Invalid superblock magic number");
322 		error = -EINVAL;
323 		goto release_buf;
324 	}
325 
326 	/*
327 	 * We must be able to do sector-sized and sector-aligned IO.
328 	 */
329 	if (sector_size > sbp->sb_sectsize) {
330 		if (loud)
331 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
332 				sector_size, sbp->sb_sectsize);
333 		error = -ENOSYS;
334 		goto release_buf;
335 	}
336 
337 	if (buf_ops == NULL) {
338 		/*
339 		 * Re-read the superblock so the buffer is correctly sized,
340 		 * and properly verified.
341 		 */
342 		xfs_buf_relse(bp);
343 		sector_size = sbp->sb_sectsize;
344 		buf_ops = loud ? &xfs_sb_buf_ops : &xfs_sb_quiet_buf_ops;
345 		goto reread;
346 	}
347 
348 	xfs_reinit_percpu_counters(mp);
349 
350 	/* no need to be quiet anymore, so reset the buf ops */
351 	bp->b_ops = &xfs_sb_buf_ops;
352 
353 	mp->m_sb_bp = bp;
354 	xfs_buf_unlock(bp);
355 	return 0;
356 
357 release_buf:
358 	xfs_buf_relse(bp);
359 	return error;
360 }
361 
362 /*
363  * If the sunit/swidth change would move the precomputed root inode value, we
364  * must reject the ondisk change because repair will stumble over that.
365  * However, we allow the mount to proceed because we never rejected this
366  * combination before.  Returns true to update the sb, false otherwise.
367  */
368 static inline int
xfs_check_new_dalign(struct xfs_mount * mp,int new_dalign,bool * update_sb)369 xfs_check_new_dalign(
370 	struct xfs_mount	*mp,
371 	int			new_dalign,
372 	bool			*update_sb)
373 {
374 	struct xfs_sb		*sbp = &mp->m_sb;
375 	xfs_ino_t		calc_ino;
376 
377 	calc_ino = xfs_ialloc_calc_rootino(mp, new_dalign);
378 	trace_xfs_check_new_dalign(mp, new_dalign, calc_ino);
379 
380 	if (sbp->sb_rootino == calc_ino) {
381 		*update_sb = true;
382 		return 0;
383 	}
384 
385 	xfs_warn(mp,
386 "Cannot change stripe alignment; would require moving root inode.");
387 
388 	/*
389 	 * XXX: Next time we add a new incompat feature, this should start
390 	 * returning -EINVAL to fail the mount.  Until then, spit out a warning
391 	 * that we're ignoring the administrator's instructions.
392 	 */
393 	xfs_warn(mp, "Skipping superblock stripe alignment update.");
394 	*update_sb = false;
395 	return 0;
396 }
397 
398 /*
399  * If we were provided with new sunit/swidth values as mount options, make sure
400  * that they pass basic alignment and superblock feature checks, and convert
401  * them into the same units (FSB) that everything else expects.  This step
402  * /must/ be done before computing the inode geometry.
403  */
404 STATIC int
xfs_validate_new_dalign(struct xfs_mount * mp)405 xfs_validate_new_dalign(
406 	struct xfs_mount	*mp)
407 {
408 	if (mp->m_dalign == 0)
409 		return 0;
410 
411 	/*
412 	 * If stripe unit and stripe width are not multiples
413 	 * of the fs blocksize turn off alignment.
414 	 */
415 	if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
416 	    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
417 		xfs_warn(mp,
418 	"alignment check failed: sunit/swidth vs. blocksize(%d)",
419 			mp->m_sb.sb_blocksize);
420 		return -EINVAL;
421 	} else {
422 		/*
423 		 * Convert the stripe unit and width to FSBs.
424 		 */
425 		mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
426 		if (mp->m_dalign && (mp->m_sb.sb_agblocks % mp->m_dalign)) {
427 			xfs_warn(mp,
428 		"alignment check failed: sunit/swidth vs. agsize(%d)",
429 				 mp->m_sb.sb_agblocks);
430 			return -EINVAL;
431 		} else if (mp->m_dalign) {
432 			mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
433 		} else {
434 			xfs_warn(mp,
435 		"alignment check failed: sunit(%d) less than bsize(%d)",
436 				 mp->m_dalign, mp->m_sb.sb_blocksize);
437 			return -EINVAL;
438 		}
439 	}
440 
441 	if (!xfs_sb_version_hasdalign(&mp->m_sb)) {
442 		xfs_warn(mp,
443 "cannot change alignment: superblock does not support data alignment");
444 		return -EINVAL;
445 	}
446 
447 	return 0;
448 }
449 
450 /* Update alignment values based on mount options and sb values. */
451 STATIC int
xfs_update_alignment(struct xfs_mount * mp)452 xfs_update_alignment(
453 	struct xfs_mount	*mp)
454 {
455 	struct xfs_sb		*sbp = &mp->m_sb;
456 
457 	if (mp->m_dalign) {
458 		bool		update_sb;
459 		int		error;
460 
461 		if (sbp->sb_unit == mp->m_dalign &&
462 		    sbp->sb_width == mp->m_swidth)
463 			return 0;
464 
465 		error = xfs_check_new_dalign(mp, mp->m_dalign, &update_sb);
466 		if (error || !update_sb)
467 			return error;
468 
469 		sbp->sb_unit = mp->m_dalign;
470 		sbp->sb_width = mp->m_swidth;
471 		mp->m_update_sb = true;
472 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
473 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
474 		mp->m_dalign = sbp->sb_unit;
475 		mp->m_swidth = sbp->sb_width;
476 	}
477 
478 	return 0;
479 }
480 
481 /*
482  * precalculate the low space thresholds for dynamic speculative preallocation.
483  */
484 void
xfs_set_low_space_thresholds(struct xfs_mount * mp)485 xfs_set_low_space_thresholds(
486 	struct xfs_mount	*mp)
487 {
488 	int i;
489 
490 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
491 		uint64_t space = mp->m_sb.sb_dblocks;
492 
493 		do_div(space, 100);
494 		mp->m_low_space[i] = space * (i + 1);
495 	}
496 }
497 
498 /*
499  * Check that the data (and log if separate) is an ok size.
500  */
501 STATIC int
xfs_check_sizes(struct xfs_mount * mp)502 xfs_check_sizes(
503 	struct xfs_mount *mp)
504 {
505 	struct xfs_buf	*bp;
506 	xfs_daddr_t	d;
507 	int		error;
508 
509 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
510 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
511 		xfs_warn(mp, "filesystem size mismatch detected");
512 		return -EFBIG;
513 	}
514 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
515 					d - XFS_FSS_TO_BB(mp, 1),
516 					XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
517 	if (error) {
518 		xfs_warn(mp, "last sector read failed");
519 		return error;
520 	}
521 	xfs_buf_relse(bp);
522 
523 	if (mp->m_logdev_targp == mp->m_ddev_targp)
524 		return 0;
525 
526 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
527 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
528 		xfs_warn(mp, "log size mismatch detected");
529 		return -EFBIG;
530 	}
531 	error = xfs_buf_read_uncached(mp->m_logdev_targp,
532 					d - XFS_FSB_TO_BB(mp, 1),
533 					XFS_FSB_TO_BB(mp, 1), 0, &bp, NULL);
534 	if (error) {
535 		xfs_warn(mp, "log device read failed");
536 		return error;
537 	}
538 	xfs_buf_relse(bp);
539 	return 0;
540 }
541 
542 /*
543  * Clear the quotaflags in memory and in the superblock.
544  */
545 int
xfs_mount_reset_sbqflags(struct xfs_mount * mp)546 xfs_mount_reset_sbqflags(
547 	struct xfs_mount	*mp)
548 {
549 	mp->m_qflags = 0;
550 
551 	/* It is OK to look at sb_qflags in the mount path without m_sb_lock. */
552 	if (mp->m_sb.sb_qflags == 0)
553 		return 0;
554 	spin_lock(&mp->m_sb_lock);
555 	mp->m_sb.sb_qflags = 0;
556 	spin_unlock(&mp->m_sb_lock);
557 
558 	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
559 		return 0;
560 
561 	return xfs_sync_sb(mp, false);
562 }
563 
564 uint64_t
xfs_default_resblks(xfs_mount_t * mp)565 xfs_default_resblks(xfs_mount_t *mp)
566 {
567 	uint64_t resblks;
568 
569 	/*
570 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
571 	 * smaller.  This is intended to cover concurrent allocation
572 	 * transactions when we initially hit enospc. These each require a 4
573 	 * block reservation. Hence by default we cover roughly 2000 concurrent
574 	 * allocation reservations.
575 	 */
576 	resblks = mp->m_sb.sb_dblocks;
577 	do_div(resblks, 20);
578 	resblks = min_t(uint64_t, resblks, 8192);
579 	return resblks;
580 }
581 
582 /* Ensure the summary counts are correct. */
583 STATIC int
xfs_check_summary_counts(struct xfs_mount * mp)584 xfs_check_summary_counts(
585 	struct xfs_mount	*mp)
586 {
587 	/*
588 	 * The AG0 superblock verifier rejects in-progress filesystems,
589 	 * so we should never see the flag set this far into mounting.
590 	 */
591 	if (mp->m_sb.sb_inprogress) {
592 		xfs_err(mp, "sb_inprogress set after log recovery??");
593 		WARN_ON(1);
594 		return -EFSCORRUPTED;
595 	}
596 
597 	/*
598 	 * Now the log is mounted, we know if it was an unclean shutdown or
599 	 * not. If it was, with the first phase of recovery has completed, we
600 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
601 	 * but they are recovered transactionally in the second recovery phase
602 	 * later.
603 	 *
604 	 * If the log was clean when we mounted, we can check the summary
605 	 * counters.  If any of them are obviously incorrect, we can recompute
606 	 * them from the AGF headers in the next step.
607 	 */
608 	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
609 	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
610 	     !xfs_verify_icount(mp, mp->m_sb.sb_icount) ||
611 	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
612 		xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
613 
614 	/*
615 	 * We can safely re-initialise incore superblock counters from the
616 	 * per-ag data. These may not be correct if the filesystem was not
617 	 * cleanly unmounted, so we waited for recovery to finish before doing
618 	 * this.
619 	 *
620 	 * If the filesystem was cleanly unmounted or the previous check did
621 	 * not flag anything weird, then we can trust the values in the
622 	 * superblock to be correct and we don't need to do anything here.
623 	 * Otherwise, recalculate the summary counters.
624 	 */
625 	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
626 	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
627 	    !xfs_fs_has_sickness(mp, XFS_SICK_FS_COUNTERS))
628 		return 0;
629 
630 	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
631 }
632 
633 /*
634  * Flush and reclaim dirty inodes in preparation for unmount. Inodes and
635  * internal inode structures can be sitting in the CIL and AIL at this point,
636  * so we need to unpin them, write them back and/or reclaim them before unmount
637  * can proceed.
638  *
639  * An inode cluster that has been freed can have its buffer still pinned in
640  * memory because the transaction is still sitting in a iclog. The stale inodes
641  * on that buffer will be pinned to the buffer until the transaction hits the
642  * disk and the callbacks run. Pushing the AIL will skip the stale inodes and
643  * may never see the pinned buffer, so nothing will push out the iclog and
644  * unpin the buffer.
645  *
646  * Hence we need to force the log to unpin everything first. However, log
647  * forces don't wait for the discards they issue to complete, so we have to
648  * explicitly wait for them to complete here as well.
649  *
650  * Then we can tell the world we are unmounting so that error handling knows
651  * that the filesystem is going away and we should error out anything that we
652  * have been retrying in the background.  This will prevent never-ending
653  * retries in AIL pushing from hanging the unmount.
654  *
655  * Finally, we can push the AIL to clean all the remaining dirty objects, then
656  * reclaim the remaining inodes that are still in memory at this point in time.
657  */
658 static void
xfs_unmount_flush_inodes(struct xfs_mount * mp)659 xfs_unmount_flush_inodes(
660 	struct xfs_mount	*mp)
661 {
662 	xfs_log_force(mp, XFS_LOG_SYNC);
663 	xfs_extent_busy_wait_all(mp);
664 	flush_workqueue(xfs_discard_wq);
665 
666 	mp->m_flags |= XFS_MOUNT_UNMOUNTING;
667 
668 	xfs_ail_push_all_sync(mp->m_ail);
669 	cancel_delayed_work_sync(&mp->m_reclaim_work);
670 	xfs_reclaim_inodes(mp);
671 	xfs_health_unmount(mp);
672 }
673 
674 /*
675  * This function does the following on an initial mount of a file system:
676  *	- reads the superblock from disk and init the mount struct
677  *	- if we're a 32-bit kernel, do a size check on the superblock
678  *		so we don't mount terabyte filesystems
679  *	- init mount struct realtime fields
680  *	- allocate inode hash table for fs
681  *	- init directory manager
682  *	- perform recovery and init the log manager
683  */
684 int
xfs_mountfs(struct xfs_mount * mp)685 xfs_mountfs(
686 	struct xfs_mount	*mp)
687 {
688 	struct xfs_sb		*sbp = &(mp->m_sb);
689 	struct xfs_inode	*rip;
690 	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
691 	uint64_t		resblks;
692 	uint			quotamount = 0;
693 	uint			quotaflags = 0;
694 	int			error = 0;
695 
696 	xfs_sb_mount_common(mp, sbp);
697 
698 	/*
699 	 * Check for a mismatched features2 values.  Older kernels read & wrote
700 	 * into the wrong sb offset for sb_features2 on some platforms due to
701 	 * xfs_sb_t not being 64bit size aligned when sb_features2 was added,
702 	 * which made older superblock reading/writing routines swap it as a
703 	 * 64-bit value.
704 	 *
705 	 * For backwards compatibility, we make both slots equal.
706 	 *
707 	 * If we detect a mismatched field, we OR the set bits into the existing
708 	 * features2 field in case it has already been modified; we don't want
709 	 * to lose any features.  We then update the bad location with the ORed
710 	 * value so that older kernels will see any features2 flags. The
711 	 * superblock writeback code ensures the new sb_features2 is copied to
712 	 * sb_bad_features2 before it is logged or written to disk.
713 	 */
714 	if (xfs_sb_has_mismatched_features2(sbp)) {
715 		xfs_warn(mp, "correcting sb_features alignment problem");
716 		sbp->sb_features2 |= sbp->sb_bad_features2;
717 		mp->m_update_sb = true;
718 
719 		/*
720 		 * Re-check for ATTR2 in case it was found in bad_features2
721 		 * slot.
722 		 */
723 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
724 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
725 			mp->m_flags |= XFS_MOUNT_ATTR2;
726 	}
727 
728 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
729 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
730 		xfs_sb_version_removeattr2(&mp->m_sb);
731 		mp->m_update_sb = true;
732 
733 		/* update sb_versionnum for the clearing of the morebits */
734 		if (!sbp->sb_features2)
735 			mp->m_update_sb = true;
736 	}
737 
738 	/* always use v2 inodes by default now */
739 	if (!(mp->m_sb.sb_versionnum & XFS_SB_VERSION_NLINKBIT)) {
740 		mp->m_sb.sb_versionnum |= XFS_SB_VERSION_NLINKBIT;
741 		mp->m_update_sb = true;
742 	}
743 
744 	/*
745 	 * If we were given new sunit/swidth options, do some basic validation
746 	 * checks and convert the incore dalign and swidth values to the
747 	 * same units (FSB) that everything else uses.  This /must/ happen
748 	 * before computing the inode geometry.
749 	 */
750 	error = xfs_validate_new_dalign(mp);
751 	if (error)
752 		goto out;
753 
754 	xfs_alloc_compute_maxlevels(mp);
755 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
756 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
757 	xfs_ialloc_setup_geometry(mp);
758 	xfs_rmapbt_compute_maxlevels(mp);
759 	xfs_refcountbt_compute_maxlevels(mp);
760 
761 	/*
762 	 * Check if sb_agblocks is aligned at stripe boundary.  If sb_agblocks
763 	 * is NOT aligned turn off m_dalign since allocator alignment is within
764 	 * an ag, therefore ag has to be aligned at stripe boundary.  Note that
765 	 * we must compute the free space and rmap btree geometry before doing
766 	 * this.
767 	 */
768 	error = xfs_update_alignment(mp);
769 	if (error)
770 		goto out;
771 
772 	/* enable fail_at_unmount as default */
773 	mp->m_fail_unmount = true;
774 
775 	error = xfs_sysfs_init(&mp->m_kobj, &xfs_mp_ktype,
776 			       NULL, mp->m_super->s_id);
777 	if (error)
778 		goto out;
779 
780 	error = xfs_sysfs_init(&mp->m_stats.xs_kobj, &xfs_stats_ktype,
781 			       &mp->m_kobj, "stats");
782 	if (error)
783 		goto out_remove_sysfs;
784 
785 	error = xfs_error_sysfs_init(mp);
786 	if (error)
787 		goto out_del_stats;
788 
789 	error = xfs_errortag_init(mp);
790 	if (error)
791 		goto out_remove_error_sysfs;
792 
793 	error = xfs_uuid_mount(mp);
794 	if (error)
795 		goto out_remove_errortag;
796 
797 	/*
798 	 * Update the preferred write size based on the information from the
799 	 * on-disk superblock.
800 	 */
801 	mp->m_allocsize_log =
802 		max_t(uint32_t, sbp->sb_blocklog, mp->m_allocsize_log);
803 	mp->m_allocsize_blocks = 1U << (mp->m_allocsize_log - sbp->sb_blocklog);
804 
805 	/* set the low space thresholds for dynamic preallocation */
806 	xfs_set_low_space_thresholds(mp);
807 
808 	/*
809 	 * If enabled, sparse inode chunk alignment is expected to match the
810 	 * cluster size. Full inode chunk alignment must match the chunk size,
811 	 * but that is checked on sb read verification...
812 	 */
813 	if (xfs_sb_version_hassparseinodes(&mp->m_sb) &&
814 	    mp->m_sb.sb_spino_align !=
815 			XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw)) {
816 		xfs_warn(mp,
817 	"Sparse inode block alignment (%u) must match cluster size (%llu).",
818 			 mp->m_sb.sb_spino_align,
819 			 XFS_B_TO_FSBT(mp, igeo->inode_cluster_size_raw));
820 		error = -EINVAL;
821 		goto out_remove_uuid;
822 	}
823 
824 	/*
825 	 * Check that the data (and log if separate) is an ok size.
826 	 */
827 	error = xfs_check_sizes(mp);
828 	if (error)
829 		goto out_remove_uuid;
830 
831 	/*
832 	 * Initialize realtime fields in the mount structure
833 	 */
834 	error = xfs_rtmount_init(mp);
835 	if (error) {
836 		xfs_warn(mp, "RT mount failed");
837 		goto out_remove_uuid;
838 	}
839 
840 	/*
841 	 *  Copies the low order bits of the timestamp and the randomly
842 	 *  set "sequence" number out of a UUID.
843 	 */
844 	mp->m_fixedfsid[0] =
845 		(get_unaligned_be16(&sbp->sb_uuid.b[8]) << 16) |
846 		 get_unaligned_be16(&sbp->sb_uuid.b[4]);
847 	mp->m_fixedfsid[1] = get_unaligned_be32(&sbp->sb_uuid.b[0]);
848 
849 	error = xfs_da_mount(mp);
850 	if (error) {
851 		xfs_warn(mp, "Failed dir/attr init: %d", error);
852 		goto out_remove_uuid;
853 	}
854 
855 	/*
856 	 * Initialize the precomputed transaction reservations values.
857 	 */
858 	xfs_trans_init(mp);
859 
860 	/*
861 	 * Allocate and initialize the per-ag data.
862 	 */
863 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
864 	if (error) {
865 		xfs_warn(mp, "Failed per-ag init: %d", error);
866 		goto out_free_dir;
867 	}
868 
869 	if (XFS_IS_CORRUPT(mp, !sbp->sb_logblocks)) {
870 		xfs_warn(mp, "no log defined");
871 		error = -EFSCORRUPTED;
872 		goto out_free_perag;
873 	}
874 
875 	/*
876 	 * Log's mount-time initialization. The first part of recovery can place
877 	 * some items on the AIL, to be handled when recovery is finished or
878 	 * cancelled.
879 	 */
880 	error = xfs_log_mount(mp, mp->m_logdev_targp,
881 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
882 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
883 	if (error) {
884 		xfs_warn(mp, "log mount failed");
885 		goto out_fail_wait;
886 	}
887 
888 	/* Make sure the summary counts are ok. */
889 	error = xfs_check_summary_counts(mp);
890 	if (error)
891 		goto out_log_dealloc;
892 
893 	/*
894 	 * Get and sanity-check the root inode.
895 	 * Save the pointer to it in the mount structure.
896 	 */
897 	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_IGET_UNTRUSTED,
898 			 XFS_ILOCK_EXCL, &rip);
899 	if (error) {
900 		xfs_warn(mp,
901 			"Failed to read root inode 0x%llx, error %d",
902 			sbp->sb_rootino, -error);
903 		goto out_log_dealloc;
904 	}
905 
906 	ASSERT(rip != NULL);
907 
908 	if (XFS_IS_CORRUPT(mp, !S_ISDIR(VFS_I(rip)->i_mode))) {
909 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
910 			(unsigned long long)rip->i_ino);
911 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
912 		error = -EFSCORRUPTED;
913 		goto out_rele_rip;
914 	}
915 	mp->m_rootip = rip;	/* save it */
916 
917 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
918 
919 	/*
920 	 * Initialize realtime inode pointers in the mount structure
921 	 */
922 	error = xfs_rtmount_inodes(mp);
923 	if (error) {
924 		/*
925 		 * Free up the root inode.
926 		 */
927 		xfs_warn(mp, "failed to read RT inodes");
928 		goto out_rele_rip;
929 	}
930 
931 	/*
932 	 * If this is a read-only mount defer the superblock updates until
933 	 * the next remount into writeable mode.  Otherwise we would never
934 	 * perform the update e.g. for the root filesystem.
935 	 */
936 	if (mp->m_update_sb && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
937 		error = xfs_sync_sb(mp, false);
938 		if (error) {
939 			xfs_warn(mp, "failed to write sb changes");
940 			goto out_rtunmount;
941 		}
942 	}
943 
944 	/*
945 	 * Initialise the XFS quota management subsystem for this mount
946 	 */
947 	if (XFS_IS_QUOTA_RUNNING(mp)) {
948 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
949 		if (error)
950 			goto out_rtunmount;
951 	} else {
952 		ASSERT(!XFS_IS_QUOTA_ON(mp));
953 
954 		/*
955 		 * If a file system had quotas running earlier, but decided to
956 		 * mount without -o uquota/pquota/gquota options, revoke the
957 		 * quotachecked license.
958 		 */
959 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
960 			xfs_notice(mp, "resetting quota flags");
961 			error = xfs_mount_reset_sbqflags(mp);
962 			if (error)
963 				goto out_rtunmount;
964 		}
965 	}
966 
967 	/*
968 	 * Finish recovering the file system.  This part needed to be delayed
969 	 * until after the root and real-time bitmap inodes were consistently
970 	 * read in.  Temporarily create per-AG space reservations for metadata
971 	 * btree shape changes because space freeing transactions (for inode
972 	 * inactivation) require the per-AG reservation in lieu of reserving
973 	 * blocks.
974 	 */
975 	error = xfs_fs_reserve_ag_blocks(mp);
976 	if (error && error == -ENOSPC)
977 		xfs_warn(mp,
978 	"ENOSPC reserving per-AG metadata pool, log recovery may fail.");
979 	error = xfs_log_mount_finish(mp);
980 	xfs_fs_unreserve_ag_blocks(mp);
981 	if (error) {
982 		xfs_warn(mp, "log mount finish failed");
983 		goto out_rtunmount;
984 	}
985 
986 	/*
987 	 * Now the log is fully replayed, we can transition to full read-only
988 	 * mode for read-only mounts. This will sync all the metadata and clean
989 	 * the log so that the recovery we just performed does not have to be
990 	 * replayed again on the next mount.
991 	 *
992 	 * We use the same quiesce mechanism as the rw->ro remount, as they are
993 	 * semantically identical operations.
994 	 */
995 	if ((mp->m_flags & (XFS_MOUNT_RDONLY|XFS_MOUNT_NORECOVERY)) ==
996 							XFS_MOUNT_RDONLY) {
997 		xfs_quiesce_attr(mp);
998 	}
999 
1000 	/*
1001 	 * Complete the quota initialisation, post-log-replay component.
1002 	 */
1003 	if (quotamount) {
1004 		ASSERT(mp->m_qflags == 0);
1005 		mp->m_qflags = quotaflags;
1006 
1007 		xfs_qm_mount_quotas(mp);
1008 	}
1009 
1010 	/*
1011 	 * Now we are mounted, reserve a small amount of unused space for
1012 	 * privileged transactions. This is needed so that transaction
1013 	 * space required for critical operations can dip into this pool
1014 	 * when at ENOSPC. This is needed for operations like create with
1015 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1016 	 * are not allowed to use this reserved space.
1017 	 *
1018 	 * This may drive us straight to ENOSPC on mount, but that implies
1019 	 * we were already there on the last unmount. Warn if this occurs.
1020 	 */
1021 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1022 		resblks = xfs_default_resblks(mp);
1023 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1024 		if (error)
1025 			xfs_warn(mp,
1026 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
1027 
1028 		/* Recover any CoW blocks that never got remapped. */
1029 		error = xfs_reflink_recover_cow(mp);
1030 		if (error) {
1031 			xfs_err(mp,
1032 	"Error %d recovering leftover CoW allocations.", error);
1033 			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1034 			goto out_quota;
1035 		}
1036 
1037 		/* Reserve AG blocks for future btree expansion. */
1038 		error = xfs_fs_reserve_ag_blocks(mp);
1039 		if (error && error != -ENOSPC)
1040 			goto out_agresv;
1041 	}
1042 
1043 	return 0;
1044 
1045  out_agresv:
1046 	xfs_fs_unreserve_ag_blocks(mp);
1047  out_quota:
1048 	xfs_qm_unmount_quotas(mp);
1049  out_rtunmount:
1050 	xfs_rtunmount_inodes(mp);
1051  out_rele_rip:
1052 	xfs_irele(rip);
1053 	/* Clean out dquots that might be in memory after quotacheck. */
1054 	xfs_qm_unmount(mp);
1055 	/*
1056 	 * Flush all inode reclamation work and flush the log.
1057 	 * We have to do this /after/ rtunmount and qm_unmount because those
1058 	 * two will have scheduled delayed reclaim for the rt/quota inodes.
1059 	 *
1060 	 * This is slightly different from the unmountfs call sequence
1061 	 * because we could be tearing down a partially set up mount.  In
1062 	 * particular, if log_mount_finish fails we bail out without calling
1063 	 * qm_unmount_quotas and therefore rely on qm_unmount to release the
1064 	 * quota inodes.
1065 	 */
1066 	xfs_unmount_flush_inodes(mp);
1067  out_log_dealloc:
1068 	xfs_log_mount_cancel(mp);
1069  out_fail_wait:
1070 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1071 		xfs_wait_buftarg(mp->m_logdev_targp);
1072 	xfs_wait_buftarg(mp->m_ddev_targp);
1073  out_free_perag:
1074 	xfs_free_perag(mp);
1075  out_free_dir:
1076 	xfs_da_unmount(mp);
1077  out_remove_uuid:
1078 	xfs_uuid_unmount(mp);
1079  out_remove_errortag:
1080 	xfs_errortag_del(mp);
1081  out_remove_error_sysfs:
1082 	xfs_error_sysfs_del(mp);
1083  out_del_stats:
1084 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1085  out_remove_sysfs:
1086 	xfs_sysfs_del(&mp->m_kobj);
1087  out:
1088 	return error;
1089 }
1090 
1091 /*
1092  * This flushes out the inodes,dquots and the superblock, unmounts the
1093  * log and makes sure that incore structures are freed.
1094  */
1095 void
xfs_unmountfs(struct xfs_mount * mp)1096 xfs_unmountfs(
1097 	struct xfs_mount	*mp)
1098 {
1099 	uint64_t		resblks;
1100 	int			error;
1101 
1102 	xfs_stop_block_reaping(mp);
1103 	xfs_fs_unreserve_ag_blocks(mp);
1104 	xfs_qm_unmount_quotas(mp);
1105 	xfs_rtunmount_inodes(mp);
1106 	xfs_irele(mp->m_rootip);
1107 
1108 	xfs_unmount_flush_inodes(mp);
1109 
1110 	xfs_qm_unmount(mp);
1111 
1112 	/*
1113 	 * Unreserve any blocks we have so that when we unmount we don't account
1114 	 * the reserved free space as used. This is really only necessary for
1115 	 * lazy superblock counting because it trusts the incore superblock
1116 	 * counters to be absolutely correct on clean unmount.
1117 	 *
1118 	 * We don't bother correcting this elsewhere for lazy superblock
1119 	 * counting because on mount of an unclean filesystem we reconstruct the
1120 	 * correct counter value and this is irrelevant.
1121 	 *
1122 	 * For non-lazy counter filesystems, this doesn't matter at all because
1123 	 * we only every apply deltas to the superblock and hence the incore
1124 	 * value does not matter....
1125 	 */
1126 	resblks = 0;
1127 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1128 	if (error)
1129 		xfs_warn(mp, "Unable to free reserved block pool. "
1130 				"Freespace may not be correct on next mount.");
1131 
1132 	error = xfs_log_sbcount(mp);
1133 	if (error)
1134 		xfs_warn(mp, "Unable to update superblock counters. "
1135 				"Freespace may not be correct on next mount.");
1136 
1137 
1138 	xfs_log_unmount(mp);
1139 	xfs_da_unmount(mp);
1140 	xfs_uuid_unmount(mp);
1141 
1142 #if defined(DEBUG)
1143 	xfs_errortag_clearall(mp);
1144 #endif
1145 	xfs_free_perag(mp);
1146 
1147 	xfs_errortag_del(mp);
1148 	xfs_error_sysfs_del(mp);
1149 	xfs_sysfs_del(&mp->m_stats.xs_kobj);
1150 	xfs_sysfs_del(&mp->m_kobj);
1151 }
1152 
1153 /*
1154  * Determine whether modifications can proceed. The caller specifies the minimum
1155  * freeze level for which modifications should not be allowed. This allows
1156  * certain operations to proceed while the freeze sequence is in progress, if
1157  * necessary.
1158  */
1159 bool
xfs_fs_writable(struct xfs_mount * mp,int level)1160 xfs_fs_writable(
1161 	struct xfs_mount	*mp,
1162 	int			level)
1163 {
1164 	ASSERT(level > SB_UNFROZEN);
1165 	if ((mp->m_super->s_writers.frozen >= level) ||
1166 	    XFS_FORCED_SHUTDOWN(mp) || (mp->m_flags & XFS_MOUNT_RDONLY))
1167 		return false;
1168 
1169 	return true;
1170 }
1171 
1172 /*
1173  * xfs_log_sbcount
1174  *
1175  * Sync the superblock counters to disk.
1176  *
1177  * Note this code can be called during the process of freezing, so we use the
1178  * transaction allocator that does not block when the transaction subsystem is
1179  * in its frozen state.
1180  */
1181 int
xfs_log_sbcount(xfs_mount_t * mp)1182 xfs_log_sbcount(xfs_mount_t *mp)
1183 {
1184 	if (!xfs_log_writable(mp))
1185 		return 0;
1186 
1187 	/*
1188 	 * we don't need to do this if we are updating the superblock
1189 	 * counters on every modification.
1190 	 */
1191 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1192 		return 0;
1193 
1194 	return xfs_sync_sb(mp, true);
1195 }
1196 
1197 /*
1198  * Deltas for the block count can vary from 1 to very large, but lock contention
1199  * only occurs on frequent small block count updates such as in the delayed
1200  * allocation path for buffered writes (page a time updates). Hence we set
1201  * a large batch count (1024) to minimise global counter updates except when
1202  * we get near to ENOSPC and we have to be very accurate with our updates.
1203  */
1204 #define XFS_FDBLOCKS_BATCH	1024
1205 int
xfs_mod_fdblocks(struct xfs_mount * mp,int64_t delta,bool rsvd)1206 xfs_mod_fdblocks(
1207 	struct xfs_mount	*mp,
1208 	int64_t			delta,
1209 	bool			rsvd)
1210 {
1211 	int64_t			lcounter;
1212 	long long		res_used;
1213 	s32			batch;
1214 
1215 	if (delta > 0) {
1216 		/*
1217 		 * If the reserve pool is depleted, put blocks back into it
1218 		 * first. Most of the time the pool is full.
1219 		 */
1220 		if (likely(mp->m_resblks == mp->m_resblks_avail)) {
1221 			percpu_counter_add(&mp->m_fdblocks, delta);
1222 			return 0;
1223 		}
1224 
1225 		spin_lock(&mp->m_sb_lock);
1226 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1227 
1228 		if (res_used > delta) {
1229 			mp->m_resblks_avail += delta;
1230 		} else {
1231 			delta -= res_used;
1232 			mp->m_resblks_avail = mp->m_resblks;
1233 			percpu_counter_add(&mp->m_fdblocks, delta);
1234 		}
1235 		spin_unlock(&mp->m_sb_lock);
1236 		return 0;
1237 	}
1238 
1239 	/*
1240 	 * Taking blocks away, need to be more accurate the closer we
1241 	 * are to zero.
1242 	 *
1243 	 * If the counter has a value of less than 2 * max batch size,
1244 	 * then make everything serialise as we are real close to
1245 	 * ENOSPC.
1246 	 */
1247 	if (__percpu_counter_compare(&mp->m_fdblocks, 2 * XFS_FDBLOCKS_BATCH,
1248 				     XFS_FDBLOCKS_BATCH) < 0)
1249 		batch = 1;
1250 	else
1251 		batch = XFS_FDBLOCKS_BATCH;
1252 
1253 	percpu_counter_add_batch(&mp->m_fdblocks, delta, batch);
1254 	if (__percpu_counter_compare(&mp->m_fdblocks, mp->m_alloc_set_aside,
1255 				     XFS_FDBLOCKS_BATCH) >= 0) {
1256 		/* we had space! */
1257 		return 0;
1258 	}
1259 
1260 	/*
1261 	 * lock up the sb for dipping into reserves before releasing the space
1262 	 * that took us to ENOSPC.
1263 	 */
1264 	spin_lock(&mp->m_sb_lock);
1265 	percpu_counter_add(&mp->m_fdblocks, -delta);
1266 	if (!rsvd)
1267 		goto fdblocks_enospc;
1268 
1269 	lcounter = (long long)mp->m_resblks_avail + delta;
1270 	if (lcounter >= 0) {
1271 		mp->m_resblks_avail = lcounter;
1272 		spin_unlock(&mp->m_sb_lock);
1273 		return 0;
1274 	}
1275 	xfs_warn_once(mp,
1276 "Reserve blocks depleted! Consider increasing reserve pool size.");
1277 
1278 fdblocks_enospc:
1279 	spin_unlock(&mp->m_sb_lock);
1280 	return -ENOSPC;
1281 }
1282 
1283 int
xfs_mod_frextents(struct xfs_mount * mp,int64_t delta)1284 xfs_mod_frextents(
1285 	struct xfs_mount	*mp,
1286 	int64_t			delta)
1287 {
1288 	int64_t			lcounter;
1289 	int			ret = 0;
1290 
1291 	spin_lock(&mp->m_sb_lock);
1292 	lcounter = mp->m_sb.sb_frextents + delta;
1293 	if (lcounter < 0)
1294 		ret = -ENOSPC;
1295 	else
1296 		mp->m_sb.sb_frextents = lcounter;
1297 	spin_unlock(&mp->m_sb_lock);
1298 	return ret;
1299 }
1300 
1301 /*
1302  * Used to free the superblock along various error paths.
1303  */
1304 void
xfs_freesb(struct xfs_mount * mp)1305 xfs_freesb(
1306 	struct xfs_mount	*mp)
1307 {
1308 	struct xfs_buf		*bp = mp->m_sb_bp;
1309 
1310 	xfs_buf_lock(bp);
1311 	mp->m_sb_bp = NULL;
1312 	xfs_buf_relse(bp);
1313 }
1314 
1315 /*
1316  * If the underlying (data/log/rt) device is readonly, there are some
1317  * operations that cannot proceed.
1318  */
1319 int
xfs_dev_is_read_only(struct xfs_mount * mp,char * message)1320 xfs_dev_is_read_only(
1321 	struct xfs_mount	*mp,
1322 	char			*message)
1323 {
1324 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1325 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1326 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1327 		xfs_notice(mp, "%s required on read-only device.", message);
1328 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1329 		return -EROFS;
1330 	}
1331 	return 0;
1332 }
1333 
1334 /* Force the summary counters to be recalculated at next mount. */
1335 void
xfs_force_summary_recalc(struct xfs_mount * mp)1336 xfs_force_summary_recalc(
1337 	struct xfs_mount	*mp)
1338 {
1339 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1340 		return;
1341 
1342 	xfs_fs_mark_sick(mp, XFS_SICK_FS_COUNTERS);
1343 }
1344 
1345 /*
1346  * Update the in-core delayed block counter.
1347  *
1348  * We prefer to update the counter without having to take a spinlock for every
1349  * counter update (i.e. batching).  Each change to delayed allocation
1350  * reservations can change can easily exceed the default percpu counter
1351  * batching, so we use a larger batch factor here.
1352  *
1353  * Note that we don't currently have any callers requiring fast summation
1354  * (e.g. percpu_counter_read) so we can use a big batch value here.
1355  */
1356 #define XFS_DELALLOC_BATCH	(4096)
1357 void
xfs_mod_delalloc(struct xfs_mount * mp,int64_t delta)1358 xfs_mod_delalloc(
1359 	struct xfs_mount	*mp,
1360 	int64_t			delta)
1361 {
1362 	percpu_counter_add_batch(&mp->m_delalloc_blks, delta,
1363 			XFS_DELALLOC_BATCH);
1364 }
1365