xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision 4800cd83)
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_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_dir2.h"
28 #include "xfs_mount.h"
29 #include "xfs_bmap_btree.h"
30 #include "xfs_alloc_btree.h"
31 #include "xfs_ialloc_btree.h"
32 #include "xfs_dinode.h"
33 #include "xfs_inode.h"
34 #include "xfs_btree.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_alloc.h"
37 #include "xfs_rtalloc.h"
38 #include "xfs_bmap.h"
39 #include "xfs_error.h"
40 #include "xfs_rw.h"
41 #include "xfs_quota.h"
42 #include "xfs_fsops.h"
43 #include "xfs_utils.h"
44 #include "xfs_trace.h"
45 
46 
47 STATIC void	xfs_unmountfs_wait(xfs_mount_t *);
48 
49 
50 #ifdef HAVE_PERCPU_SB
51 STATIC void	xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
52 						int);
53 STATIC void	xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
54 						int);
55 STATIC void	xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
56 #else
57 
58 #define xfs_icsb_balance_counter(mp, a, b)		do { } while (0)
59 #define xfs_icsb_balance_counter_locked(mp, a, b)	do { } while (0)
60 #endif
61 
62 static const struct {
63 	short offset;
64 	short type;	/* 0 = integer
65 			 * 1 = binary / string (no translation)
66 			 */
67 } xfs_sb_info[] = {
68     { offsetof(xfs_sb_t, sb_magicnum),   0 },
69     { offsetof(xfs_sb_t, sb_blocksize),  0 },
70     { offsetof(xfs_sb_t, sb_dblocks),    0 },
71     { offsetof(xfs_sb_t, sb_rblocks),    0 },
72     { offsetof(xfs_sb_t, sb_rextents),   0 },
73     { offsetof(xfs_sb_t, sb_uuid),       1 },
74     { offsetof(xfs_sb_t, sb_logstart),   0 },
75     { offsetof(xfs_sb_t, sb_rootino),    0 },
76     { offsetof(xfs_sb_t, sb_rbmino),     0 },
77     { offsetof(xfs_sb_t, sb_rsumino),    0 },
78     { offsetof(xfs_sb_t, sb_rextsize),   0 },
79     { offsetof(xfs_sb_t, sb_agblocks),   0 },
80     { offsetof(xfs_sb_t, sb_agcount),    0 },
81     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
82     { offsetof(xfs_sb_t, sb_logblocks),  0 },
83     { offsetof(xfs_sb_t, sb_versionnum), 0 },
84     { offsetof(xfs_sb_t, sb_sectsize),   0 },
85     { offsetof(xfs_sb_t, sb_inodesize),  0 },
86     { offsetof(xfs_sb_t, sb_inopblock),  0 },
87     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
88     { offsetof(xfs_sb_t, sb_blocklog),   0 },
89     { offsetof(xfs_sb_t, sb_sectlog),    0 },
90     { offsetof(xfs_sb_t, sb_inodelog),   0 },
91     { offsetof(xfs_sb_t, sb_inopblog),   0 },
92     { offsetof(xfs_sb_t, sb_agblklog),   0 },
93     { offsetof(xfs_sb_t, sb_rextslog),   0 },
94     { offsetof(xfs_sb_t, sb_inprogress), 0 },
95     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
96     { offsetof(xfs_sb_t, sb_icount),     0 },
97     { offsetof(xfs_sb_t, sb_ifree),      0 },
98     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
99     { offsetof(xfs_sb_t, sb_frextents),  0 },
100     { offsetof(xfs_sb_t, sb_uquotino),   0 },
101     { offsetof(xfs_sb_t, sb_gquotino),   0 },
102     { offsetof(xfs_sb_t, sb_qflags),     0 },
103     { offsetof(xfs_sb_t, sb_flags),      0 },
104     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
105     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
106     { offsetof(xfs_sb_t, sb_unit),	 0 },
107     { offsetof(xfs_sb_t, sb_width),	 0 },
108     { offsetof(xfs_sb_t, sb_dirblklog),	 0 },
109     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
110     { offsetof(xfs_sb_t, sb_logsectsize),0 },
111     { offsetof(xfs_sb_t, sb_logsunit),	 0 },
112     { offsetof(xfs_sb_t, sb_features2),	 0 },
113     { offsetof(xfs_sb_t, sb_bad_features2), 0 },
114     { sizeof(xfs_sb_t),			 0 }
115 };
116 
117 static DEFINE_MUTEX(xfs_uuid_table_mutex);
118 static int xfs_uuid_table_size;
119 static uuid_t *xfs_uuid_table;
120 
121 /*
122  * See if the UUID is unique among mounted XFS filesystems.
123  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
124  */
125 STATIC int
126 xfs_uuid_mount(
127 	struct xfs_mount	*mp)
128 {
129 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
130 	int			hole, i;
131 
132 	if (mp->m_flags & XFS_MOUNT_NOUUID)
133 		return 0;
134 
135 	if (uuid_is_nil(uuid)) {
136 		cmn_err(CE_WARN,
137 			"XFS: Filesystem %s has nil UUID - can't mount",
138 			mp->m_fsname);
139 		return XFS_ERROR(EINVAL);
140 	}
141 
142 	mutex_lock(&xfs_uuid_table_mutex);
143 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
144 		if (uuid_is_nil(&xfs_uuid_table[i])) {
145 			hole = i;
146 			continue;
147 		}
148 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
149 			goto out_duplicate;
150 	}
151 
152 	if (hole < 0) {
153 		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
154 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
155 			xfs_uuid_table_size  * sizeof(*xfs_uuid_table),
156 			KM_SLEEP);
157 		hole = xfs_uuid_table_size++;
158 	}
159 	xfs_uuid_table[hole] = *uuid;
160 	mutex_unlock(&xfs_uuid_table_mutex);
161 
162 	return 0;
163 
164  out_duplicate:
165 	mutex_unlock(&xfs_uuid_table_mutex);
166 	cmn_err(CE_WARN, "XFS: Filesystem %s has duplicate UUID - can't mount",
167 			 mp->m_fsname);
168 	return XFS_ERROR(EINVAL);
169 }
170 
171 STATIC void
172 xfs_uuid_unmount(
173 	struct xfs_mount	*mp)
174 {
175 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
176 	int			i;
177 
178 	if (mp->m_flags & XFS_MOUNT_NOUUID)
179 		return;
180 
181 	mutex_lock(&xfs_uuid_table_mutex);
182 	for (i = 0; i < xfs_uuid_table_size; i++) {
183 		if (uuid_is_nil(&xfs_uuid_table[i]))
184 			continue;
185 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
186 			continue;
187 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
188 		break;
189 	}
190 	ASSERT(i < xfs_uuid_table_size);
191 	mutex_unlock(&xfs_uuid_table_mutex);
192 }
193 
194 
195 /*
196  * Reference counting access wrappers to the perag structures.
197  * Because we never free per-ag structures, the only thing we
198  * have to protect against changes is the tree structure itself.
199  */
200 struct xfs_perag *
201 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
202 {
203 	struct xfs_perag	*pag;
204 	int			ref = 0;
205 
206 	rcu_read_lock();
207 	pag = radix_tree_lookup(&mp->m_perag_tree, agno);
208 	if (pag) {
209 		ASSERT(atomic_read(&pag->pag_ref) >= 0);
210 		ref = atomic_inc_return(&pag->pag_ref);
211 	}
212 	rcu_read_unlock();
213 	trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
214 	return pag;
215 }
216 
217 /*
218  * search from @first to find the next perag with the given tag set.
219  */
220 struct xfs_perag *
221 xfs_perag_get_tag(
222 	struct xfs_mount	*mp,
223 	xfs_agnumber_t		first,
224 	int			tag)
225 {
226 	struct xfs_perag	*pag;
227 	int			found;
228 	int			ref;
229 
230 	rcu_read_lock();
231 	found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
232 					(void **)&pag, first, 1, tag);
233 	if (found <= 0) {
234 		rcu_read_unlock();
235 		return NULL;
236 	}
237 	ref = atomic_inc_return(&pag->pag_ref);
238 	rcu_read_unlock();
239 	trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
240 	return pag;
241 }
242 
243 void
244 xfs_perag_put(struct xfs_perag *pag)
245 {
246 	int	ref;
247 
248 	ASSERT(atomic_read(&pag->pag_ref) > 0);
249 	ref = atomic_dec_return(&pag->pag_ref);
250 	trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
251 }
252 
253 STATIC void
254 __xfs_free_perag(
255 	struct rcu_head	*head)
256 {
257 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
258 
259 	ASSERT(atomic_read(&pag->pag_ref) == 0);
260 	kmem_free(pag);
261 }
262 
263 /*
264  * Free up the per-ag resources associated with the mount structure.
265  */
266 STATIC void
267 xfs_free_perag(
268 	xfs_mount_t	*mp)
269 {
270 	xfs_agnumber_t	agno;
271 	struct xfs_perag *pag;
272 
273 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
274 		spin_lock(&mp->m_perag_lock);
275 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
276 		spin_unlock(&mp->m_perag_lock);
277 		ASSERT(pag);
278 		ASSERT(atomic_read(&pag->pag_ref) == 0);
279 		call_rcu(&pag->rcu_head, __xfs_free_perag);
280 	}
281 }
282 
283 /*
284  * Check size of device based on the (data/realtime) block count.
285  * Note: this check is used by the growfs code as well as mount.
286  */
287 int
288 xfs_sb_validate_fsb_count(
289 	xfs_sb_t	*sbp,
290 	__uint64_t	nblocks)
291 {
292 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
293 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
294 
295 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
296 	if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
297 		return EFBIG;
298 #else                  /* Limited by UINT_MAX of sectors */
299 	if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
300 		return EFBIG;
301 #endif
302 	return 0;
303 }
304 
305 /*
306  * Check the validity of the SB found.
307  */
308 STATIC int
309 xfs_mount_validate_sb(
310 	xfs_mount_t	*mp,
311 	xfs_sb_t	*sbp,
312 	int		flags)
313 {
314 	/*
315 	 * If the log device and data device have the
316 	 * same device number, the log is internal.
317 	 * Consequently, the sb_logstart should be non-zero.  If
318 	 * we have a zero sb_logstart in this case, we may be trying to mount
319 	 * a volume filesystem in a non-volume manner.
320 	 */
321 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
322 		xfs_fs_mount_cmn_err(flags, "bad magic number");
323 		return XFS_ERROR(EWRONGFS);
324 	}
325 
326 	if (!xfs_sb_good_version(sbp)) {
327 		xfs_fs_mount_cmn_err(flags, "bad version");
328 		return XFS_ERROR(EWRONGFS);
329 	}
330 
331 	if (unlikely(
332 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
333 		xfs_fs_mount_cmn_err(flags,
334 			"filesystem is marked as having an external log; "
335 			"specify logdev on the\nmount command line.");
336 		return XFS_ERROR(EINVAL);
337 	}
338 
339 	if (unlikely(
340 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
341 		xfs_fs_mount_cmn_err(flags,
342 			"filesystem is marked as having an internal log; "
343 			"do not specify logdev on\nthe mount command line.");
344 		return XFS_ERROR(EINVAL);
345 	}
346 
347 	/*
348 	 * More sanity checking. These were stolen directly from
349 	 * xfs_repair.
350 	 */
351 	if (unlikely(
352 	    sbp->sb_agcount <= 0					||
353 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
354 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
355 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
356 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
357 	    sbp->sb_sectsize != (1 << sbp->sb_sectlog)			||
358 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
359 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
360 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
361 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
362 	    sbp->sb_blocksize != (1 << sbp->sb_blocklog)		||
363 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
364 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
365 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
366 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
367 	    sbp->sb_inodesize != (1 << sbp->sb_inodelog)		||
368 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
369 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
370 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
371 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
372 		xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
373 		return XFS_ERROR(EFSCORRUPTED);
374 	}
375 
376 	/*
377 	 * Sanity check AG count, size fields against data size field
378 	 */
379 	if (unlikely(
380 	    sbp->sb_dblocks == 0 ||
381 	    sbp->sb_dblocks >
382 	     (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
383 	    sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
384 			      sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
385 		xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
386 		return XFS_ERROR(EFSCORRUPTED);
387 	}
388 
389 	/*
390 	 * Until this is fixed only page-sized or smaller data blocks work.
391 	 */
392 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
393 		xfs_fs_mount_cmn_err(flags,
394 			"file system with blocksize %d bytes",
395 			sbp->sb_blocksize);
396 		xfs_fs_mount_cmn_err(flags,
397 			"only pagesize (%ld) or less will currently work.",
398 			PAGE_SIZE);
399 		return XFS_ERROR(ENOSYS);
400 	}
401 
402 	/*
403 	 * Currently only very few inode sizes are supported.
404 	 */
405 	switch (sbp->sb_inodesize) {
406 	case 256:
407 	case 512:
408 	case 1024:
409 	case 2048:
410 		break;
411 	default:
412 		xfs_fs_mount_cmn_err(flags,
413 			"inode size of %d bytes not supported",
414 			sbp->sb_inodesize);
415 		return XFS_ERROR(ENOSYS);
416 	}
417 
418 	if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
419 	    xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
420 		xfs_fs_mount_cmn_err(flags,
421 			"file system too large to be mounted on this system.");
422 		return XFS_ERROR(EFBIG);
423 	}
424 
425 	if (unlikely(sbp->sb_inprogress)) {
426 		xfs_fs_mount_cmn_err(flags, "file system busy");
427 		return XFS_ERROR(EFSCORRUPTED);
428 	}
429 
430 	/*
431 	 * Version 1 directory format has never worked on Linux.
432 	 */
433 	if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
434 		xfs_fs_mount_cmn_err(flags,
435 			"file system using version 1 directory format");
436 		return XFS_ERROR(ENOSYS);
437 	}
438 
439 	return 0;
440 }
441 
442 int
443 xfs_initialize_perag(
444 	xfs_mount_t	*mp,
445 	xfs_agnumber_t	agcount,
446 	xfs_agnumber_t	*maxagi)
447 {
448 	xfs_agnumber_t	index, max_metadata;
449 	xfs_agnumber_t	first_initialised = 0;
450 	xfs_perag_t	*pag;
451 	xfs_agino_t	agino;
452 	xfs_ino_t	ino;
453 	xfs_sb_t	*sbp = &mp->m_sb;
454 	int		error = -ENOMEM;
455 
456 	/*
457 	 * Walk the current per-ag tree so we don't try to initialise AGs
458 	 * that already exist (growfs case). Allocate and insert all the
459 	 * AGs we don't find ready for initialisation.
460 	 */
461 	for (index = 0; index < agcount; index++) {
462 		pag = xfs_perag_get(mp, index);
463 		if (pag) {
464 			xfs_perag_put(pag);
465 			continue;
466 		}
467 		if (!first_initialised)
468 			first_initialised = index;
469 
470 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
471 		if (!pag)
472 			goto out_unwind;
473 		pag->pag_agno = index;
474 		pag->pag_mount = mp;
475 		spin_lock_init(&pag->pag_ici_lock);
476 		mutex_init(&pag->pag_ici_reclaim_lock);
477 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
478 		spin_lock_init(&pag->pag_buf_lock);
479 		pag->pag_buf_tree = RB_ROOT;
480 
481 		if (radix_tree_preload(GFP_NOFS))
482 			goto out_unwind;
483 
484 		spin_lock(&mp->m_perag_lock);
485 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
486 			BUG();
487 			spin_unlock(&mp->m_perag_lock);
488 			radix_tree_preload_end();
489 			error = -EEXIST;
490 			goto out_unwind;
491 		}
492 		spin_unlock(&mp->m_perag_lock);
493 		radix_tree_preload_end();
494 	}
495 
496 	/*
497 	 * If we mount with the inode64 option, or no inode overflows
498 	 * the legacy 32-bit address space clear the inode32 option.
499 	 */
500 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
501 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
502 
503 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
504 		mp->m_flags |= XFS_MOUNT_32BITINODES;
505 	else
506 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
507 
508 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
509 		/*
510 		 * Calculate how much should be reserved for inodes to meet
511 		 * the max inode percentage.
512 		 */
513 		if (mp->m_maxicount) {
514 			__uint64_t	icount;
515 
516 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
517 			do_div(icount, 100);
518 			icount += sbp->sb_agblocks - 1;
519 			do_div(icount, sbp->sb_agblocks);
520 			max_metadata = icount;
521 		} else {
522 			max_metadata = agcount;
523 		}
524 
525 		for (index = 0; index < agcount; index++) {
526 			ino = XFS_AGINO_TO_INO(mp, index, agino);
527 			if (ino > XFS_MAXINUMBER_32) {
528 				index++;
529 				break;
530 			}
531 
532 			pag = xfs_perag_get(mp, index);
533 			pag->pagi_inodeok = 1;
534 			if (index < max_metadata)
535 				pag->pagf_metadata = 1;
536 			xfs_perag_put(pag);
537 		}
538 	} else {
539 		for (index = 0; index < agcount; index++) {
540 			pag = xfs_perag_get(mp, index);
541 			pag->pagi_inodeok = 1;
542 			xfs_perag_put(pag);
543 		}
544 	}
545 
546 	if (maxagi)
547 		*maxagi = index;
548 	return 0;
549 
550 out_unwind:
551 	kmem_free(pag);
552 	for (; index > first_initialised; index--) {
553 		pag = radix_tree_delete(&mp->m_perag_tree, index);
554 		kmem_free(pag);
555 	}
556 	return error;
557 }
558 
559 void
560 xfs_sb_from_disk(
561 	xfs_sb_t	*to,
562 	xfs_dsb_t	*from)
563 {
564 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
565 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
566 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
567 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
568 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
569 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
570 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
571 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
572 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
573 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
574 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
575 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
576 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
577 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
578 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
579 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
580 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
581 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
582 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
583 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
584 	to->sb_blocklog = from->sb_blocklog;
585 	to->sb_sectlog = from->sb_sectlog;
586 	to->sb_inodelog = from->sb_inodelog;
587 	to->sb_inopblog = from->sb_inopblog;
588 	to->sb_agblklog = from->sb_agblklog;
589 	to->sb_rextslog = from->sb_rextslog;
590 	to->sb_inprogress = from->sb_inprogress;
591 	to->sb_imax_pct = from->sb_imax_pct;
592 	to->sb_icount = be64_to_cpu(from->sb_icount);
593 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
594 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
595 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
596 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
597 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
598 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
599 	to->sb_flags = from->sb_flags;
600 	to->sb_shared_vn = from->sb_shared_vn;
601 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
602 	to->sb_unit = be32_to_cpu(from->sb_unit);
603 	to->sb_width = be32_to_cpu(from->sb_width);
604 	to->sb_dirblklog = from->sb_dirblklog;
605 	to->sb_logsectlog = from->sb_logsectlog;
606 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
607 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
608 	to->sb_features2 = be32_to_cpu(from->sb_features2);
609 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
610 }
611 
612 /*
613  * Copy in core superblock to ondisk one.
614  *
615  * The fields argument is mask of superblock fields to copy.
616  */
617 void
618 xfs_sb_to_disk(
619 	xfs_dsb_t	*to,
620 	xfs_sb_t	*from,
621 	__int64_t	fields)
622 {
623 	xfs_caddr_t	to_ptr = (xfs_caddr_t)to;
624 	xfs_caddr_t	from_ptr = (xfs_caddr_t)from;
625 	xfs_sb_field_t	f;
626 	int		first;
627 	int		size;
628 
629 	ASSERT(fields);
630 	if (!fields)
631 		return;
632 
633 	while (fields) {
634 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
635 		first = xfs_sb_info[f].offset;
636 		size = xfs_sb_info[f + 1].offset - first;
637 
638 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
639 
640 		if (size == 1 || xfs_sb_info[f].type == 1) {
641 			memcpy(to_ptr + first, from_ptr + first, size);
642 		} else {
643 			switch (size) {
644 			case 2:
645 				*(__be16 *)(to_ptr + first) =
646 					cpu_to_be16(*(__u16 *)(from_ptr + first));
647 				break;
648 			case 4:
649 				*(__be32 *)(to_ptr + first) =
650 					cpu_to_be32(*(__u32 *)(from_ptr + first));
651 				break;
652 			case 8:
653 				*(__be64 *)(to_ptr + first) =
654 					cpu_to_be64(*(__u64 *)(from_ptr + first));
655 				break;
656 			default:
657 				ASSERT(0);
658 			}
659 		}
660 
661 		fields &= ~(1LL << f);
662 	}
663 }
664 
665 /*
666  * xfs_readsb
667  *
668  * Does the initial read of the superblock.
669  */
670 int
671 xfs_readsb(xfs_mount_t *mp, int flags)
672 {
673 	unsigned int	sector_size;
674 	xfs_buf_t	*bp;
675 	int		error;
676 
677 	ASSERT(mp->m_sb_bp == NULL);
678 	ASSERT(mp->m_ddev_targp != NULL);
679 
680 	/*
681 	 * Allocate a (locked) buffer to hold the superblock.
682 	 * This will be kept around at all times to optimize
683 	 * access to the superblock.
684 	 */
685 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
686 
687 reread:
688 	bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
689 					XFS_SB_DADDR, sector_size, 0);
690 	if (!bp) {
691 		xfs_fs_mount_cmn_err(flags, "SB buffer read failed");
692 		return EIO;
693 	}
694 
695 	/*
696 	 * Initialize the mount structure from the superblock.
697 	 * But first do some basic consistency checking.
698 	 */
699 	xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
700 	error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
701 	if (error) {
702 		xfs_fs_mount_cmn_err(flags, "SB validate failed");
703 		goto release_buf;
704 	}
705 
706 	/*
707 	 * We must be able to do sector-sized and sector-aligned IO.
708 	 */
709 	if (sector_size > mp->m_sb.sb_sectsize) {
710 		xfs_fs_mount_cmn_err(flags,
711 			"device supports only %u byte sectors (not %u)",
712 			sector_size, mp->m_sb.sb_sectsize);
713 		error = ENOSYS;
714 		goto release_buf;
715 	}
716 
717 	/*
718 	 * If device sector size is smaller than the superblock size,
719 	 * re-read the superblock so the buffer is correctly sized.
720 	 */
721 	if (sector_size < mp->m_sb.sb_sectsize) {
722 		xfs_buf_relse(bp);
723 		sector_size = mp->m_sb.sb_sectsize;
724 		goto reread;
725 	}
726 
727 	/* Initialize per-cpu counters */
728 	xfs_icsb_reinit_counters(mp);
729 
730 	mp->m_sb_bp = bp;
731 	xfs_buf_unlock(bp);
732 	return 0;
733 
734 release_buf:
735 	xfs_buf_relse(bp);
736 	return error;
737 }
738 
739 
740 /*
741  * xfs_mount_common
742  *
743  * Mount initialization code establishing various mount
744  * fields from the superblock associated with the given
745  * mount structure
746  */
747 STATIC void
748 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
749 {
750 	mp->m_agfrotor = mp->m_agirotor = 0;
751 	spin_lock_init(&mp->m_agirotor_lock);
752 	mp->m_maxagi = mp->m_sb.sb_agcount;
753 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
754 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
755 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
756 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
757 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
758 	mp->m_blockmask = sbp->sb_blocksize - 1;
759 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
760 	mp->m_blockwmask = mp->m_blockwsize - 1;
761 
762 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
763 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
764 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
765 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
766 
767 	mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
768 	mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
769 	mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
770 	mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
771 
772 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
773 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
774 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
775 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
776 
777 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
778 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
779 					sbp->sb_inopblock);
780 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
781 }
782 
783 /*
784  * xfs_initialize_perag_data
785  *
786  * Read in each per-ag structure so we can count up the number of
787  * allocated inodes, free inodes and used filesystem blocks as this
788  * information is no longer persistent in the superblock. Once we have
789  * this information, write it into the in-core superblock structure.
790  */
791 STATIC int
792 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
793 {
794 	xfs_agnumber_t	index;
795 	xfs_perag_t	*pag;
796 	xfs_sb_t	*sbp = &mp->m_sb;
797 	uint64_t	ifree = 0;
798 	uint64_t	ialloc = 0;
799 	uint64_t	bfree = 0;
800 	uint64_t	bfreelst = 0;
801 	uint64_t	btree = 0;
802 	int		error;
803 
804 	for (index = 0; index < agcount; index++) {
805 		/*
806 		 * read the agf, then the agi. This gets us
807 		 * all the information we need and populates the
808 		 * per-ag structures for us.
809 		 */
810 		error = xfs_alloc_pagf_init(mp, NULL, index, 0);
811 		if (error)
812 			return error;
813 
814 		error = xfs_ialloc_pagi_init(mp, NULL, index);
815 		if (error)
816 			return error;
817 		pag = xfs_perag_get(mp, index);
818 		ifree += pag->pagi_freecount;
819 		ialloc += pag->pagi_count;
820 		bfree += pag->pagf_freeblks;
821 		bfreelst += pag->pagf_flcount;
822 		btree += pag->pagf_btreeblks;
823 		xfs_perag_put(pag);
824 	}
825 	/*
826 	 * Overwrite incore superblock counters with just-read data
827 	 */
828 	spin_lock(&mp->m_sb_lock);
829 	sbp->sb_ifree = ifree;
830 	sbp->sb_icount = ialloc;
831 	sbp->sb_fdblocks = bfree + bfreelst + btree;
832 	spin_unlock(&mp->m_sb_lock);
833 
834 	/* Fixup the per-cpu counters as well. */
835 	xfs_icsb_reinit_counters(mp);
836 
837 	return 0;
838 }
839 
840 /*
841  * Update alignment values based on mount options and sb values
842  */
843 STATIC int
844 xfs_update_alignment(xfs_mount_t *mp)
845 {
846 	xfs_sb_t	*sbp = &(mp->m_sb);
847 
848 	if (mp->m_dalign) {
849 		/*
850 		 * If stripe unit and stripe width are not multiples
851 		 * of the fs blocksize turn off alignment.
852 		 */
853 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
854 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
855 			if (mp->m_flags & XFS_MOUNT_RETERR) {
856 				cmn_err(CE_WARN,
857 					"XFS: alignment check 1 failed");
858 				return XFS_ERROR(EINVAL);
859 			}
860 			mp->m_dalign = mp->m_swidth = 0;
861 		} else {
862 			/*
863 			 * Convert the stripe unit and width to FSBs.
864 			 */
865 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
866 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
867 				if (mp->m_flags & XFS_MOUNT_RETERR) {
868 					return XFS_ERROR(EINVAL);
869 				}
870 				xfs_fs_cmn_err(CE_WARN, mp,
871 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
872 					mp->m_dalign, mp->m_swidth,
873 					sbp->sb_agblocks);
874 
875 				mp->m_dalign = 0;
876 				mp->m_swidth = 0;
877 			} else if (mp->m_dalign) {
878 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
879 			} else {
880 				if (mp->m_flags & XFS_MOUNT_RETERR) {
881 					xfs_fs_cmn_err(CE_WARN, mp,
882 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
883                                         	mp->m_dalign,
884 						mp->m_blockmask +1);
885 					return XFS_ERROR(EINVAL);
886 				}
887 				mp->m_swidth = 0;
888 			}
889 		}
890 
891 		/*
892 		 * Update superblock with new values
893 		 * and log changes
894 		 */
895 		if (xfs_sb_version_hasdalign(sbp)) {
896 			if (sbp->sb_unit != mp->m_dalign) {
897 				sbp->sb_unit = mp->m_dalign;
898 				mp->m_update_flags |= XFS_SB_UNIT;
899 			}
900 			if (sbp->sb_width != mp->m_swidth) {
901 				sbp->sb_width = mp->m_swidth;
902 				mp->m_update_flags |= XFS_SB_WIDTH;
903 			}
904 		}
905 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
906 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
907 			mp->m_dalign = sbp->sb_unit;
908 			mp->m_swidth = sbp->sb_width;
909 	}
910 
911 	return 0;
912 }
913 
914 /*
915  * Set the maximum inode count for this filesystem
916  */
917 STATIC void
918 xfs_set_maxicount(xfs_mount_t *mp)
919 {
920 	xfs_sb_t	*sbp = &(mp->m_sb);
921 	__uint64_t	icount;
922 
923 	if (sbp->sb_imax_pct) {
924 		/*
925 		 * Make sure the maximum inode count is a multiple
926 		 * of the units we allocate inodes in.
927 		 */
928 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
929 		do_div(icount, 100);
930 		do_div(icount, mp->m_ialloc_blks);
931 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
932 				   sbp->sb_inopblog;
933 	} else {
934 		mp->m_maxicount = 0;
935 	}
936 }
937 
938 /*
939  * Set the default minimum read and write sizes unless
940  * already specified in a mount option.
941  * We use smaller I/O sizes when the file system
942  * is being used for NFS service (wsync mount option).
943  */
944 STATIC void
945 xfs_set_rw_sizes(xfs_mount_t *mp)
946 {
947 	xfs_sb_t	*sbp = &(mp->m_sb);
948 	int		readio_log, writeio_log;
949 
950 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
951 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
952 			readio_log = XFS_WSYNC_READIO_LOG;
953 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
954 		} else {
955 			readio_log = XFS_READIO_LOG_LARGE;
956 			writeio_log = XFS_WRITEIO_LOG_LARGE;
957 		}
958 	} else {
959 		readio_log = mp->m_readio_log;
960 		writeio_log = mp->m_writeio_log;
961 	}
962 
963 	if (sbp->sb_blocklog > readio_log) {
964 		mp->m_readio_log = sbp->sb_blocklog;
965 	} else {
966 		mp->m_readio_log = readio_log;
967 	}
968 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
969 	if (sbp->sb_blocklog > writeio_log) {
970 		mp->m_writeio_log = sbp->sb_blocklog;
971 	} else {
972 		mp->m_writeio_log = writeio_log;
973 	}
974 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
975 }
976 
977 /*
978  * precalculate the low space thresholds for dynamic speculative preallocation.
979  */
980 void
981 xfs_set_low_space_thresholds(
982 	struct xfs_mount	*mp)
983 {
984 	int i;
985 
986 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
987 		__uint64_t space = mp->m_sb.sb_dblocks;
988 
989 		do_div(space, 100);
990 		mp->m_low_space[i] = space * (i + 1);
991 	}
992 }
993 
994 
995 /*
996  * Set whether we're using inode alignment.
997  */
998 STATIC void
999 xfs_set_inoalignment(xfs_mount_t *mp)
1000 {
1001 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
1002 	    mp->m_sb.sb_inoalignmt >=
1003 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1004 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1005 	else
1006 		mp->m_inoalign_mask = 0;
1007 	/*
1008 	 * If we are using stripe alignment, check whether
1009 	 * the stripe unit is a multiple of the inode alignment
1010 	 */
1011 	if (mp->m_dalign && mp->m_inoalign_mask &&
1012 	    !(mp->m_dalign & mp->m_inoalign_mask))
1013 		mp->m_sinoalign = mp->m_dalign;
1014 	else
1015 		mp->m_sinoalign = 0;
1016 }
1017 
1018 /*
1019  * Check that the data (and log if separate) are an ok size.
1020  */
1021 STATIC int
1022 xfs_check_sizes(xfs_mount_t *mp)
1023 {
1024 	xfs_buf_t	*bp;
1025 	xfs_daddr_t	d;
1026 
1027 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1028 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1029 		cmn_err(CE_WARN, "XFS: filesystem size mismatch detected");
1030 		return XFS_ERROR(EFBIG);
1031 	}
1032 	bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1033 					d - XFS_FSS_TO_BB(mp, 1),
1034 					BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1035 	if (!bp) {
1036 		cmn_err(CE_WARN, "XFS: last sector read failed");
1037 		return EIO;
1038 	}
1039 	xfs_buf_relse(bp);
1040 
1041 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
1042 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1043 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1044 			cmn_err(CE_WARN, "XFS: log size mismatch detected");
1045 			return XFS_ERROR(EFBIG);
1046 		}
1047 		bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1048 					d - XFS_FSB_TO_BB(mp, 1),
1049 					XFS_FSB_TO_B(mp, 1), 0);
1050 		if (!bp) {
1051 			cmn_err(CE_WARN, "XFS: log device read failed");
1052 			return EIO;
1053 		}
1054 		xfs_buf_relse(bp);
1055 	}
1056 	return 0;
1057 }
1058 
1059 /*
1060  * Clear the quotaflags in memory and in the superblock.
1061  */
1062 int
1063 xfs_mount_reset_sbqflags(
1064 	struct xfs_mount	*mp)
1065 {
1066 	int			error;
1067 	struct xfs_trans	*tp;
1068 
1069 	mp->m_qflags = 0;
1070 
1071 	/*
1072 	 * It is OK to look at sb_qflags here in mount path,
1073 	 * without m_sb_lock.
1074 	 */
1075 	if (mp->m_sb.sb_qflags == 0)
1076 		return 0;
1077 	spin_lock(&mp->m_sb_lock);
1078 	mp->m_sb.sb_qflags = 0;
1079 	spin_unlock(&mp->m_sb_lock);
1080 
1081 	/*
1082 	 * If the fs is readonly, let the incore superblock run
1083 	 * with quotas off but don't flush the update out to disk
1084 	 */
1085 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1086 		return 0;
1087 
1088 #ifdef QUOTADEBUG
1089 	xfs_fs_cmn_err(CE_NOTE, mp, "Writing superblock quota changes");
1090 #endif
1091 
1092 	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1093 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1094 				      XFS_DEFAULT_LOG_COUNT);
1095 	if (error) {
1096 		xfs_trans_cancel(tp, 0);
1097 		xfs_fs_cmn_err(CE_ALERT, mp,
1098 			"xfs_mount_reset_sbqflags: Superblock update failed!");
1099 		return error;
1100 	}
1101 
1102 	xfs_mod_sb(tp, XFS_SB_QFLAGS);
1103 	return xfs_trans_commit(tp, 0);
1104 }
1105 
1106 __uint64_t
1107 xfs_default_resblks(xfs_mount_t *mp)
1108 {
1109 	__uint64_t resblks;
1110 
1111 	/*
1112 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
1113 	 * smaller.  This is intended to cover concurrent allocation
1114 	 * transactions when we initially hit enospc. These each require a 4
1115 	 * block reservation. Hence by default we cover roughly 2000 concurrent
1116 	 * allocation reservations.
1117 	 */
1118 	resblks = mp->m_sb.sb_dblocks;
1119 	do_div(resblks, 20);
1120 	resblks = min_t(__uint64_t, resblks, 8192);
1121 	return resblks;
1122 }
1123 
1124 /*
1125  * This function does the following on an initial mount of a file system:
1126  *	- reads the superblock from disk and init the mount struct
1127  *	- if we're a 32-bit kernel, do a size check on the superblock
1128  *		so we don't mount terabyte filesystems
1129  *	- init mount struct realtime fields
1130  *	- allocate inode hash table for fs
1131  *	- init directory manager
1132  *	- perform recovery and init the log manager
1133  */
1134 int
1135 xfs_mountfs(
1136 	xfs_mount_t	*mp)
1137 {
1138 	xfs_sb_t	*sbp = &(mp->m_sb);
1139 	xfs_inode_t	*rip;
1140 	__uint64_t	resblks;
1141 	uint		quotamount = 0;
1142 	uint		quotaflags = 0;
1143 	int		error = 0;
1144 
1145 	xfs_mount_common(mp, sbp);
1146 
1147 	/*
1148 	 * Check for a mismatched features2 values.  Older kernels
1149 	 * read & wrote into the wrong sb offset for sb_features2
1150 	 * on some platforms due to xfs_sb_t not being 64bit size aligned
1151 	 * when sb_features2 was added, which made older superblock
1152 	 * reading/writing routines swap it as a 64-bit value.
1153 	 *
1154 	 * For backwards compatibility, we make both slots equal.
1155 	 *
1156 	 * If we detect a mismatched field, we OR the set bits into the
1157 	 * existing features2 field in case it has already been modified; we
1158 	 * don't want to lose any features.  We then update the bad location
1159 	 * with the ORed value so that older kernels will see any features2
1160 	 * flags, and mark the two fields as needing updates once the
1161 	 * transaction subsystem is online.
1162 	 */
1163 	if (xfs_sb_has_mismatched_features2(sbp)) {
1164 		cmn_err(CE_WARN,
1165 			"XFS: correcting sb_features alignment problem");
1166 		sbp->sb_features2 |= sbp->sb_bad_features2;
1167 		sbp->sb_bad_features2 = sbp->sb_features2;
1168 		mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1169 
1170 		/*
1171 		 * Re-check for ATTR2 in case it was found in bad_features2
1172 		 * slot.
1173 		 */
1174 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1175 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
1176 			mp->m_flags |= XFS_MOUNT_ATTR2;
1177 	}
1178 
1179 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1180 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1181 		xfs_sb_version_removeattr2(&mp->m_sb);
1182 		mp->m_update_flags |= XFS_SB_FEATURES2;
1183 
1184 		/* update sb_versionnum for the clearing of the morebits */
1185 		if (!sbp->sb_features2)
1186 			mp->m_update_flags |= XFS_SB_VERSIONNUM;
1187 	}
1188 
1189 	/*
1190 	 * Check if sb_agblocks is aligned at stripe boundary
1191 	 * If sb_agblocks is NOT aligned turn off m_dalign since
1192 	 * allocator alignment is within an ag, therefore ag has
1193 	 * to be aligned at stripe boundary.
1194 	 */
1195 	error = xfs_update_alignment(mp);
1196 	if (error)
1197 		goto out;
1198 
1199 	xfs_alloc_compute_maxlevels(mp);
1200 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1201 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1202 	xfs_ialloc_compute_maxlevels(mp);
1203 
1204 	xfs_set_maxicount(mp);
1205 
1206 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1207 
1208 	error = xfs_uuid_mount(mp);
1209 	if (error)
1210 		goto out;
1211 
1212 	/*
1213 	 * Set the minimum read and write sizes
1214 	 */
1215 	xfs_set_rw_sizes(mp);
1216 
1217 	/* set the low space thresholds for dynamic preallocation */
1218 	xfs_set_low_space_thresholds(mp);
1219 
1220 	/*
1221 	 * Set the inode cluster size.
1222 	 * This may still be overridden by the file system
1223 	 * block size if it is larger than the chosen cluster size.
1224 	 */
1225 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1226 
1227 	/*
1228 	 * Set inode alignment fields
1229 	 */
1230 	xfs_set_inoalignment(mp);
1231 
1232 	/*
1233 	 * Check that the data (and log if separate) are an ok size.
1234 	 */
1235 	error = xfs_check_sizes(mp);
1236 	if (error)
1237 		goto out_remove_uuid;
1238 
1239 	/*
1240 	 * Initialize realtime fields in the mount structure
1241 	 */
1242 	error = xfs_rtmount_init(mp);
1243 	if (error) {
1244 		cmn_err(CE_WARN, "XFS: RT mount failed");
1245 		goto out_remove_uuid;
1246 	}
1247 
1248 	/*
1249 	 *  Copies the low order bits of the timestamp and the randomly
1250 	 *  set "sequence" number out of a UUID.
1251 	 */
1252 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1253 
1254 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
1255 
1256 	xfs_dir_mount(mp);
1257 
1258 	/*
1259 	 * Initialize the attribute manager's entries.
1260 	 */
1261 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1262 
1263 	/*
1264 	 * Initialize the precomputed transaction reservations values.
1265 	 */
1266 	xfs_trans_init(mp);
1267 
1268 	/*
1269 	 * Allocate and initialize the per-ag data.
1270 	 */
1271 	spin_lock_init(&mp->m_perag_lock);
1272 	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1273 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1274 	if (error) {
1275 		cmn_err(CE_WARN, "XFS: Failed per-ag init: %d", error);
1276 		goto out_remove_uuid;
1277 	}
1278 
1279 	if (!sbp->sb_logblocks) {
1280 		cmn_err(CE_WARN, "XFS: no log defined");
1281 		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1282 		error = XFS_ERROR(EFSCORRUPTED);
1283 		goto out_free_perag;
1284 	}
1285 
1286 	/*
1287 	 * log's mount-time initialization. Perform 1st part recovery if needed
1288 	 */
1289 	error = xfs_log_mount(mp, mp->m_logdev_targp,
1290 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1291 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1292 	if (error) {
1293 		cmn_err(CE_WARN, "XFS: log mount failed");
1294 		goto out_free_perag;
1295 	}
1296 
1297 	/*
1298 	 * Now the log is mounted, we know if it was an unclean shutdown or
1299 	 * not. If it was, with the first phase of recovery has completed, we
1300 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1301 	 * but they are recovered transactionally in the second recovery phase
1302 	 * later.
1303 	 *
1304 	 * Hence we can safely re-initialise incore superblock counters from
1305 	 * the per-ag data. These may not be correct if the filesystem was not
1306 	 * cleanly unmounted, so we need to wait for recovery to finish before
1307 	 * doing this.
1308 	 *
1309 	 * If the filesystem was cleanly unmounted, then we can trust the
1310 	 * values in the superblock to be correct and we don't need to do
1311 	 * anything here.
1312 	 *
1313 	 * If we are currently making the filesystem, the initialisation will
1314 	 * fail as the perag data is in an undefined state.
1315 	 */
1316 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1317 	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1318 	     !mp->m_sb.sb_inprogress) {
1319 		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1320 		if (error)
1321 			goto out_free_perag;
1322 	}
1323 
1324 	/*
1325 	 * Get and sanity-check the root inode.
1326 	 * Save the pointer to it in the mount structure.
1327 	 */
1328 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1329 	if (error) {
1330 		cmn_err(CE_WARN, "XFS: failed to read root inode");
1331 		goto out_log_dealloc;
1332 	}
1333 
1334 	ASSERT(rip != NULL);
1335 
1336 	if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1337 		cmn_err(CE_WARN, "XFS: corrupted root inode");
1338 		cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1339 			XFS_BUFTARG_NAME(mp->m_ddev_targp),
1340 			(unsigned long long)rip->i_ino);
1341 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
1342 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1343 				 mp);
1344 		error = XFS_ERROR(EFSCORRUPTED);
1345 		goto out_rele_rip;
1346 	}
1347 	mp->m_rootip = rip;	/* save it */
1348 
1349 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1350 
1351 	/*
1352 	 * Initialize realtime inode pointers in the mount structure
1353 	 */
1354 	error = xfs_rtmount_inodes(mp);
1355 	if (error) {
1356 		/*
1357 		 * Free up the root inode.
1358 		 */
1359 		cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1360 		goto out_rele_rip;
1361 	}
1362 
1363 	/*
1364 	 * If this is a read-only mount defer the superblock updates until
1365 	 * the next remount into writeable mode.  Otherwise we would never
1366 	 * perform the update e.g. for the root filesystem.
1367 	 */
1368 	if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1369 		error = xfs_mount_log_sb(mp, mp->m_update_flags);
1370 		if (error) {
1371 			cmn_err(CE_WARN, "XFS: failed to write sb changes");
1372 			goto out_rtunmount;
1373 		}
1374 	}
1375 
1376 	/*
1377 	 * Initialise the XFS quota management subsystem for this mount
1378 	 */
1379 	if (XFS_IS_QUOTA_RUNNING(mp)) {
1380 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1381 		if (error)
1382 			goto out_rtunmount;
1383 	} else {
1384 		ASSERT(!XFS_IS_QUOTA_ON(mp));
1385 
1386 		/*
1387 		 * If a file system had quotas running earlier, but decided to
1388 		 * mount without -o uquota/pquota/gquota options, revoke the
1389 		 * quotachecked license.
1390 		 */
1391 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1392 			cmn_err(CE_NOTE,
1393 				"XFS: resetting qflags for filesystem %s",
1394 				mp->m_fsname);
1395 
1396 			error = xfs_mount_reset_sbqflags(mp);
1397 			if (error)
1398 				return error;
1399 		}
1400 	}
1401 
1402 	/*
1403 	 * Finish recovering the file system.  This part needed to be
1404 	 * delayed until after the root and real-time bitmap inodes
1405 	 * were consistently read in.
1406 	 */
1407 	error = xfs_log_mount_finish(mp);
1408 	if (error) {
1409 		cmn_err(CE_WARN, "XFS: log mount finish failed");
1410 		goto out_rtunmount;
1411 	}
1412 
1413 	/*
1414 	 * Complete the quota initialisation, post-log-replay component.
1415 	 */
1416 	if (quotamount) {
1417 		ASSERT(mp->m_qflags == 0);
1418 		mp->m_qflags = quotaflags;
1419 
1420 		xfs_qm_mount_quotas(mp);
1421 	}
1422 
1423 	/*
1424 	 * Now we are mounted, reserve a small amount of unused space for
1425 	 * privileged transactions. This is needed so that transaction
1426 	 * space required for critical operations can dip into this pool
1427 	 * when at ENOSPC. This is needed for operations like create with
1428 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1429 	 * are not allowed to use this reserved space.
1430 	 *
1431 	 * This may drive us straight to ENOSPC on mount, but that implies
1432 	 * we were already there on the last unmount. Warn if this occurs.
1433 	 */
1434 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1435 		resblks = xfs_default_resblks(mp);
1436 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1437 		if (error)
1438 			cmn_err(CE_WARN, "XFS: Unable to allocate reserve "
1439 				"blocks. Continuing without a reserve pool.");
1440 	}
1441 
1442 	return 0;
1443 
1444  out_rtunmount:
1445 	xfs_rtunmount_inodes(mp);
1446  out_rele_rip:
1447 	IRELE(rip);
1448  out_log_dealloc:
1449 	xfs_log_unmount(mp);
1450  out_free_perag:
1451 	xfs_free_perag(mp);
1452  out_remove_uuid:
1453 	xfs_uuid_unmount(mp);
1454  out:
1455 	return error;
1456 }
1457 
1458 /*
1459  * This flushes out the inodes,dquots and the superblock, unmounts the
1460  * log and makes sure that incore structures are freed.
1461  */
1462 void
1463 xfs_unmountfs(
1464 	struct xfs_mount	*mp)
1465 {
1466 	__uint64_t		resblks;
1467 	int			error;
1468 
1469 	xfs_qm_unmount_quotas(mp);
1470 	xfs_rtunmount_inodes(mp);
1471 	IRELE(mp->m_rootip);
1472 
1473 	/*
1474 	 * We can potentially deadlock here if we have an inode cluster
1475 	 * that has been freed has its buffer still pinned in memory because
1476 	 * the transaction is still sitting in a iclog. The stale inodes
1477 	 * on that buffer will have their flush locks held until the
1478 	 * transaction hits the disk and the callbacks run. the inode
1479 	 * flush takes the flush lock unconditionally and with nothing to
1480 	 * push out the iclog we will never get that unlocked. hence we
1481 	 * need to force the log first.
1482 	 */
1483 	xfs_log_force(mp, XFS_LOG_SYNC);
1484 
1485 	/*
1486 	 * Do a delwri reclaim pass first so that as many dirty inodes are
1487 	 * queued up for IO as possible. Then flush the buffers before making
1488 	 * a synchronous path to catch all the remaining inodes are reclaimed.
1489 	 * This makes the reclaim process as quick as possible by avoiding
1490 	 * synchronous writeout and blocking on inodes already in the delwri
1491 	 * state as much as possible.
1492 	 */
1493 	xfs_reclaim_inodes(mp, 0);
1494 	XFS_bflush(mp->m_ddev_targp);
1495 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1496 
1497 	xfs_qm_unmount(mp);
1498 
1499 	/*
1500 	 * Flush out the log synchronously so that we know for sure
1501 	 * that nothing is pinned.  This is important because bflush()
1502 	 * will skip pinned buffers.
1503 	 */
1504 	xfs_log_force(mp, XFS_LOG_SYNC);
1505 
1506 	xfs_binval(mp->m_ddev_targp);
1507 	if (mp->m_rtdev_targp) {
1508 		xfs_binval(mp->m_rtdev_targp);
1509 	}
1510 
1511 	/*
1512 	 * Unreserve any blocks we have so that when we unmount we don't account
1513 	 * the reserved free space as used. This is really only necessary for
1514 	 * lazy superblock counting because it trusts the incore superblock
1515 	 * counters to be absolutely correct on clean unmount.
1516 	 *
1517 	 * We don't bother correcting this elsewhere for lazy superblock
1518 	 * counting because on mount of an unclean filesystem we reconstruct the
1519 	 * correct counter value and this is irrelevant.
1520 	 *
1521 	 * For non-lazy counter filesystems, this doesn't matter at all because
1522 	 * we only every apply deltas to the superblock and hence the incore
1523 	 * value does not matter....
1524 	 */
1525 	resblks = 0;
1526 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1527 	if (error)
1528 		cmn_err(CE_WARN, "XFS: Unable to free reserved block pool. "
1529 				"Freespace may not be correct on next mount.");
1530 
1531 	error = xfs_log_sbcount(mp, 1);
1532 	if (error)
1533 		cmn_err(CE_WARN, "XFS: Unable to update superblock counters. "
1534 				"Freespace may not be correct on next mount.");
1535 	xfs_unmountfs_writesb(mp);
1536 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
1537 	xfs_log_unmount_write(mp);
1538 	xfs_log_unmount(mp);
1539 	xfs_uuid_unmount(mp);
1540 
1541 #if defined(DEBUG)
1542 	xfs_errortag_clearall(mp, 0);
1543 #endif
1544 	xfs_free_perag(mp);
1545 }
1546 
1547 STATIC void
1548 xfs_unmountfs_wait(xfs_mount_t *mp)
1549 {
1550 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1551 		xfs_wait_buftarg(mp->m_logdev_targp);
1552 	if (mp->m_rtdev_targp)
1553 		xfs_wait_buftarg(mp->m_rtdev_targp);
1554 	xfs_wait_buftarg(mp->m_ddev_targp);
1555 }
1556 
1557 int
1558 xfs_fs_writable(xfs_mount_t *mp)
1559 {
1560 	return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1561 		(mp->m_flags & XFS_MOUNT_RDONLY));
1562 }
1563 
1564 /*
1565  * xfs_log_sbcount
1566  *
1567  * Called either periodically to keep the on disk superblock values
1568  * roughly up to date or from unmount to make sure the values are
1569  * correct on a clean unmount.
1570  *
1571  * Note this code can be called during the process of freezing, so
1572  * we may need to use the transaction allocator which does not not
1573  * block when the transaction subsystem is in its frozen state.
1574  */
1575 int
1576 xfs_log_sbcount(
1577 	xfs_mount_t	*mp,
1578 	uint		sync)
1579 {
1580 	xfs_trans_t	*tp;
1581 	int		error;
1582 
1583 	if (!xfs_fs_writable(mp))
1584 		return 0;
1585 
1586 	xfs_icsb_sync_counters(mp, 0);
1587 
1588 	/*
1589 	 * we don't need to do this if we are updating the superblock
1590 	 * counters on every modification.
1591 	 */
1592 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1593 		return 0;
1594 
1595 	tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1596 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1597 					XFS_DEFAULT_LOG_COUNT);
1598 	if (error) {
1599 		xfs_trans_cancel(tp, 0);
1600 		return error;
1601 	}
1602 
1603 	xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1604 	if (sync)
1605 		xfs_trans_set_sync(tp);
1606 	error = xfs_trans_commit(tp, 0);
1607 	return error;
1608 }
1609 
1610 int
1611 xfs_unmountfs_writesb(xfs_mount_t *mp)
1612 {
1613 	xfs_buf_t	*sbp;
1614 	int		error = 0;
1615 
1616 	/*
1617 	 * skip superblock write if fs is read-only, or
1618 	 * if we are doing a forced umount.
1619 	 */
1620 	if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1621 		XFS_FORCED_SHUTDOWN(mp))) {
1622 
1623 		sbp = xfs_getsb(mp, 0);
1624 
1625 		XFS_BUF_UNDONE(sbp);
1626 		XFS_BUF_UNREAD(sbp);
1627 		XFS_BUF_UNDELAYWRITE(sbp);
1628 		XFS_BUF_WRITE(sbp);
1629 		XFS_BUF_UNASYNC(sbp);
1630 		ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1631 		xfsbdstrat(mp, sbp);
1632 		error = xfs_buf_iowait(sbp);
1633 		if (error)
1634 			xfs_ioerror_alert("xfs_unmountfs_writesb",
1635 					  mp, sbp, XFS_BUF_ADDR(sbp));
1636 		xfs_buf_relse(sbp);
1637 	}
1638 	return error;
1639 }
1640 
1641 /*
1642  * xfs_mod_sb() can be used to copy arbitrary changes to the
1643  * in-core superblock into the superblock buffer to be logged.
1644  * It does not provide the higher level of locking that is
1645  * needed to protect the in-core superblock from concurrent
1646  * access.
1647  */
1648 void
1649 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1650 {
1651 	xfs_buf_t	*bp;
1652 	int		first;
1653 	int		last;
1654 	xfs_mount_t	*mp;
1655 	xfs_sb_field_t	f;
1656 
1657 	ASSERT(fields);
1658 	if (!fields)
1659 		return;
1660 	mp = tp->t_mountp;
1661 	bp = xfs_trans_getsb(tp, mp, 0);
1662 	first = sizeof(xfs_sb_t);
1663 	last = 0;
1664 
1665 	/* translate/copy */
1666 
1667 	xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1668 
1669 	/* find modified range */
1670 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1671 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1672 	last = xfs_sb_info[f + 1].offset - 1;
1673 
1674 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1675 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1676 	first = xfs_sb_info[f].offset;
1677 
1678 	xfs_trans_log_buf(tp, bp, first, last);
1679 }
1680 
1681 
1682 /*
1683  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1684  * a delta to a specified field in the in-core superblock.  Simply
1685  * switch on the field indicated and apply the delta to that field.
1686  * Fields are not allowed to dip below zero, so if the delta would
1687  * do this do not apply it and return EINVAL.
1688  *
1689  * The m_sb_lock must be held when this routine is called.
1690  */
1691 STATIC int
1692 xfs_mod_incore_sb_unlocked(
1693 	xfs_mount_t	*mp,
1694 	xfs_sb_field_t	field,
1695 	int64_t		delta,
1696 	int		rsvd)
1697 {
1698 	int		scounter;	/* short counter for 32 bit fields */
1699 	long long	lcounter;	/* long counter for 64 bit fields */
1700 	long long	res_used, rem;
1701 
1702 	/*
1703 	 * With the in-core superblock spin lock held, switch
1704 	 * on the indicated field.  Apply the delta to the
1705 	 * proper field.  If the fields value would dip below
1706 	 * 0, then do not apply the delta and return EINVAL.
1707 	 */
1708 	switch (field) {
1709 	case XFS_SBS_ICOUNT:
1710 		lcounter = (long long)mp->m_sb.sb_icount;
1711 		lcounter += delta;
1712 		if (lcounter < 0) {
1713 			ASSERT(0);
1714 			return XFS_ERROR(EINVAL);
1715 		}
1716 		mp->m_sb.sb_icount = lcounter;
1717 		return 0;
1718 	case XFS_SBS_IFREE:
1719 		lcounter = (long long)mp->m_sb.sb_ifree;
1720 		lcounter += delta;
1721 		if (lcounter < 0) {
1722 			ASSERT(0);
1723 			return XFS_ERROR(EINVAL);
1724 		}
1725 		mp->m_sb.sb_ifree = lcounter;
1726 		return 0;
1727 	case XFS_SBS_FDBLOCKS:
1728 		lcounter = (long long)
1729 			mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1730 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1731 
1732 		if (delta > 0) {		/* Putting blocks back */
1733 			if (res_used > delta) {
1734 				mp->m_resblks_avail += delta;
1735 			} else {
1736 				rem = delta - res_used;
1737 				mp->m_resblks_avail = mp->m_resblks;
1738 				lcounter += rem;
1739 			}
1740 		} else {				/* Taking blocks away */
1741 			lcounter += delta;
1742 			if (lcounter >= 0) {
1743 				mp->m_sb.sb_fdblocks = lcounter +
1744 							XFS_ALLOC_SET_ASIDE(mp);
1745 				return 0;
1746 			}
1747 
1748 			/*
1749 			 * We are out of blocks, use any available reserved
1750 			 * blocks if were allowed to.
1751 			 */
1752 			if (!rsvd)
1753 				return XFS_ERROR(ENOSPC);
1754 
1755 			lcounter = (long long)mp->m_resblks_avail + delta;
1756 			if (lcounter >= 0) {
1757 				mp->m_resblks_avail = lcounter;
1758 				return 0;
1759 			}
1760 			printk_once(KERN_WARNING
1761 				"Filesystem \"%s\": reserve blocks depleted! "
1762 				"Consider increasing reserve pool size.",
1763 				mp->m_fsname);
1764 			return XFS_ERROR(ENOSPC);
1765 		}
1766 
1767 		mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1768 		return 0;
1769 	case XFS_SBS_FREXTENTS:
1770 		lcounter = (long long)mp->m_sb.sb_frextents;
1771 		lcounter += delta;
1772 		if (lcounter < 0) {
1773 			return XFS_ERROR(ENOSPC);
1774 		}
1775 		mp->m_sb.sb_frextents = lcounter;
1776 		return 0;
1777 	case XFS_SBS_DBLOCKS:
1778 		lcounter = (long long)mp->m_sb.sb_dblocks;
1779 		lcounter += delta;
1780 		if (lcounter < 0) {
1781 			ASSERT(0);
1782 			return XFS_ERROR(EINVAL);
1783 		}
1784 		mp->m_sb.sb_dblocks = lcounter;
1785 		return 0;
1786 	case XFS_SBS_AGCOUNT:
1787 		scounter = mp->m_sb.sb_agcount;
1788 		scounter += delta;
1789 		if (scounter < 0) {
1790 			ASSERT(0);
1791 			return XFS_ERROR(EINVAL);
1792 		}
1793 		mp->m_sb.sb_agcount = scounter;
1794 		return 0;
1795 	case XFS_SBS_IMAX_PCT:
1796 		scounter = mp->m_sb.sb_imax_pct;
1797 		scounter += delta;
1798 		if (scounter < 0) {
1799 			ASSERT(0);
1800 			return XFS_ERROR(EINVAL);
1801 		}
1802 		mp->m_sb.sb_imax_pct = scounter;
1803 		return 0;
1804 	case XFS_SBS_REXTSIZE:
1805 		scounter = mp->m_sb.sb_rextsize;
1806 		scounter += delta;
1807 		if (scounter < 0) {
1808 			ASSERT(0);
1809 			return XFS_ERROR(EINVAL);
1810 		}
1811 		mp->m_sb.sb_rextsize = scounter;
1812 		return 0;
1813 	case XFS_SBS_RBMBLOCKS:
1814 		scounter = mp->m_sb.sb_rbmblocks;
1815 		scounter += delta;
1816 		if (scounter < 0) {
1817 			ASSERT(0);
1818 			return XFS_ERROR(EINVAL);
1819 		}
1820 		mp->m_sb.sb_rbmblocks = scounter;
1821 		return 0;
1822 	case XFS_SBS_RBLOCKS:
1823 		lcounter = (long long)mp->m_sb.sb_rblocks;
1824 		lcounter += delta;
1825 		if (lcounter < 0) {
1826 			ASSERT(0);
1827 			return XFS_ERROR(EINVAL);
1828 		}
1829 		mp->m_sb.sb_rblocks = lcounter;
1830 		return 0;
1831 	case XFS_SBS_REXTENTS:
1832 		lcounter = (long long)mp->m_sb.sb_rextents;
1833 		lcounter += delta;
1834 		if (lcounter < 0) {
1835 			ASSERT(0);
1836 			return XFS_ERROR(EINVAL);
1837 		}
1838 		mp->m_sb.sb_rextents = lcounter;
1839 		return 0;
1840 	case XFS_SBS_REXTSLOG:
1841 		scounter = mp->m_sb.sb_rextslog;
1842 		scounter += delta;
1843 		if (scounter < 0) {
1844 			ASSERT(0);
1845 			return XFS_ERROR(EINVAL);
1846 		}
1847 		mp->m_sb.sb_rextslog = scounter;
1848 		return 0;
1849 	default:
1850 		ASSERT(0);
1851 		return XFS_ERROR(EINVAL);
1852 	}
1853 }
1854 
1855 /*
1856  * xfs_mod_incore_sb() is used to change a field in the in-core
1857  * superblock structure by the specified delta.  This modification
1858  * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1859  * routine to do the work.
1860  */
1861 int
1862 xfs_mod_incore_sb(
1863 	struct xfs_mount	*mp,
1864 	xfs_sb_field_t		field,
1865 	int64_t			delta,
1866 	int			rsvd)
1867 {
1868 	int			status;
1869 
1870 #ifdef HAVE_PERCPU_SB
1871 	ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1872 #endif
1873 	spin_lock(&mp->m_sb_lock);
1874 	status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1875 	spin_unlock(&mp->m_sb_lock);
1876 
1877 	return status;
1878 }
1879 
1880 /*
1881  * Change more than one field in the in-core superblock structure at a time.
1882  *
1883  * The fields and changes to those fields are specified in the array of
1884  * xfs_mod_sb structures passed in.  Either all of the specified deltas
1885  * will be applied or none of them will.  If any modified field dips below 0,
1886  * then all modifications will be backed out and EINVAL will be returned.
1887  *
1888  * Note that this function may not be used for the superblock values that
1889  * are tracked with the in-memory per-cpu counters - a direct call to
1890  * xfs_icsb_modify_counters is required for these.
1891  */
1892 int
1893 xfs_mod_incore_sb_batch(
1894 	struct xfs_mount	*mp,
1895 	xfs_mod_sb_t		*msb,
1896 	uint			nmsb,
1897 	int			rsvd)
1898 {
1899 	xfs_mod_sb_t		*msbp = &msb[0];
1900 	int			error = 0;
1901 
1902 	/*
1903 	 * Loop through the array of mod structures and apply each individually.
1904 	 * If any fail, then back out all those which have already been applied.
1905 	 * Do all of this within the scope of the m_sb_lock so that all of the
1906 	 * changes will be atomic.
1907 	 */
1908 	spin_lock(&mp->m_sb_lock);
1909 	for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1910 		ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1911 		       msbp->msb_field > XFS_SBS_FDBLOCKS);
1912 
1913 		error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1914 						   msbp->msb_delta, rsvd);
1915 		if (error)
1916 			goto unwind;
1917 	}
1918 	spin_unlock(&mp->m_sb_lock);
1919 	return 0;
1920 
1921 unwind:
1922 	while (--msbp >= msb) {
1923 		error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1924 						   -msbp->msb_delta, rsvd);
1925 		ASSERT(error == 0);
1926 	}
1927 	spin_unlock(&mp->m_sb_lock);
1928 	return error;
1929 }
1930 
1931 /*
1932  * xfs_getsb() is called to obtain the buffer for the superblock.
1933  * The buffer is returned locked and read in from disk.
1934  * The buffer should be released with a call to xfs_brelse().
1935  *
1936  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1937  * the superblock buffer if it can be locked without sleeping.
1938  * If it can't then we'll return NULL.
1939  */
1940 xfs_buf_t *
1941 xfs_getsb(
1942 	xfs_mount_t	*mp,
1943 	int		flags)
1944 {
1945 	xfs_buf_t	*bp;
1946 
1947 	ASSERT(mp->m_sb_bp != NULL);
1948 	bp = mp->m_sb_bp;
1949 	if (flags & XBF_TRYLOCK) {
1950 		if (!XFS_BUF_CPSEMA(bp)) {
1951 			return NULL;
1952 		}
1953 	} else {
1954 		XFS_BUF_PSEMA(bp, PRIBIO);
1955 	}
1956 	XFS_BUF_HOLD(bp);
1957 	ASSERT(XFS_BUF_ISDONE(bp));
1958 	return bp;
1959 }
1960 
1961 /*
1962  * Used to free the superblock along various error paths.
1963  */
1964 void
1965 xfs_freesb(
1966 	struct xfs_mount	*mp)
1967 {
1968 	struct xfs_buf		*bp = mp->m_sb_bp;
1969 
1970 	xfs_buf_lock(bp);
1971 	mp->m_sb_bp = NULL;
1972 	xfs_buf_relse(bp);
1973 }
1974 
1975 /*
1976  * Used to log changes to the superblock unit and width fields which could
1977  * be altered by the mount options, as well as any potential sb_features2
1978  * fixup. Only the first superblock is updated.
1979  */
1980 int
1981 xfs_mount_log_sb(
1982 	xfs_mount_t	*mp,
1983 	__int64_t	fields)
1984 {
1985 	xfs_trans_t	*tp;
1986 	int		error;
1987 
1988 	ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1989 			 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1990 			 XFS_SB_VERSIONNUM));
1991 
1992 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1993 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1994 				XFS_DEFAULT_LOG_COUNT);
1995 	if (error) {
1996 		xfs_trans_cancel(tp, 0);
1997 		return error;
1998 	}
1999 	xfs_mod_sb(tp, fields);
2000 	error = xfs_trans_commit(tp, 0);
2001 	return error;
2002 }
2003 
2004 /*
2005  * If the underlying (data/log/rt) device is readonly, there are some
2006  * operations that cannot proceed.
2007  */
2008 int
2009 xfs_dev_is_read_only(
2010 	struct xfs_mount	*mp,
2011 	char			*message)
2012 {
2013 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
2014 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
2015 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
2016 		cmn_err(CE_NOTE,
2017 			"XFS: %s required on read-only device.", message);
2018 		cmn_err(CE_NOTE,
2019 			"XFS: write access unavailable, cannot proceed.");
2020 		return EROFS;
2021 	}
2022 	return 0;
2023 }
2024 
2025 #ifdef HAVE_PERCPU_SB
2026 /*
2027  * Per-cpu incore superblock counters
2028  *
2029  * Simple concept, difficult implementation
2030  *
2031  * Basically, replace the incore superblock counters with a distributed per cpu
2032  * counter for contended fields (e.g.  free block count).
2033  *
2034  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2035  * hence needs to be accurately read when we are running low on space. Hence
2036  * there is a method to enable and disable the per-cpu counters based on how
2037  * much "stuff" is available in them.
2038  *
2039  * Basically, a counter is enabled if there is enough free resource to justify
2040  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2041  * ENOSPC), then we disable the counters to synchronise all callers and
2042  * re-distribute the available resources.
2043  *
2044  * If, once we redistributed the available resources, we still get a failure,
2045  * we disable the per-cpu counter and go through the slow path.
2046  *
2047  * The slow path is the current xfs_mod_incore_sb() function.  This means that
2048  * when we disable a per-cpu counter, we need to drain its resources back to
2049  * the global superblock. We do this after disabling the counter to prevent
2050  * more threads from queueing up on the counter.
2051  *
2052  * Essentially, this means that we still need a lock in the fast path to enable
2053  * synchronisation between the global counters and the per-cpu counters. This
2054  * is not a problem because the lock will be local to a CPU almost all the time
2055  * and have little contention except when we get to ENOSPC conditions.
2056  *
2057  * Basically, this lock becomes a barrier that enables us to lock out the fast
2058  * path while we do things like enabling and disabling counters and
2059  * synchronising the counters.
2060  *
2061  * Locking rules:
2062  *
2063  * 	1. m_sb_lock before picking up per-cpu locks
2064  * 	2. per-cpu locks always picked up via for_each_online_cpu() order
2065  * 	3. accurate counter sync requires m_sb_lock + per cpu locks
2066  * 	4. modifying per-cpu counters requires holding per-cpu lock
2067  * 	5. modifying global counters requires holding m_sb_lock
2068  *	6. enabling or disabling a counter requires holding the m_sb_lock
2069  *	   and _none_ of the per-cpu locks.
2070  *
2071  * Disabled counters are only ever re-enabled by a balance operation
2072  * that results in more free resources per CPU than a given threshold.
2073  * To ensure counters don't remain disabled, they are rebalanced when
2074  * the global resource goes above a higher threshold (i.e. some hysteresis
2075  * is present to prevent thrashing).
2076  */
2077 
2078 #ifdef CONFIG_HOTPLUG_CPU
2079 /*
2080  * hot-plug CPU notifier support.
2081  *
2082  * We need a notifier per filesystem as we need to be able to identify
2083  * the filesystem to balance the counters out. This is achieved by
2084  * having a notifier block embedded in the xfs_mount_t and doing pointer
2085  * magic to get the mount pointer from the notifier block address.
2086  */
2087 STATIC int
2088 xfs_icsb_cpu_notify(
2089 	struct notifier_block *nfb,
2090 	unsigned long action,
2091 	void *hcpu)
2092 {
2093 	xfs_icsb_cnts_t *cntp;
2094 	xfs_mount_t	*mp;
2095 
2096 	mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2097 	cntp = (xfs_icsb_cnts_t *)
2098 			per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2099 	switch (action) {
2100 	case CPU_UP_PREPARE:
2101 	case CPU_UP_PREPARE_FROZEN:
2102 		/* Easy Case - initialize the area and locks, and
2103 		 * then rebalance when online does everything else for us. */
2104 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2105 		break;
2106 	case CPU_ONLINE:
2107 	case CPU_ONLINE_FROZEN:
2108 		xfs_icsb_lock(mp);
2109 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2110 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2111 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2112 		xfs_icsb_unlock(mp);
2113 		break;
2114 	case CPU_DEAD:
2115 	case CPU_DEAD_FROZEN:
2116 		/* Disable all the counters, then fold the dead cpu's
2117 		 * count into the total on the global superblock and
2118 		 * re-enable the counters. */
2119 		xfs_icsb_lock(mp);
2120 		spin_lock(&mp->m_sb_lock);
2121 		xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2122 		xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2123 		xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2124 
2125 		mp->m_sb.sb_icount += cntp->icsb_icount;
2126 		mp->m_sb.sb_ifree += cntp->icsb_ifree;
2127 		mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2128 
2129 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2130 
2131 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2132 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2133 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2134 		spin_unlock(&mp->m_sb_lock);
2135 		xfs_icsb_unlock(mp);
2136 		break;
2137 	}
2138 
2139 	return NOTIFY_OK;
2140 }
2141 #endif /* CONFIG_HOTPLUG_CPU */
2142 
2143 int
2144 xfs_icsb_init_counters(
2145 	xfs_mount_t	*mp)
2146 {
2147 	xfs_icsb_cnts_t *cntp;
2148 	int		i;
2149 
2150 	mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2151 	if (mp->m_sb_cnts == NULL)
2152 		return -ENOMEM;
2153 
2154 #ifdef CONFIG_HOTPLUG_CPU
2155 	mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2156 	mp->m_icsb_notifier.priority = 0;
2157 	register_hotcpu_notifier(&mp->m_icsb_notifier);
2158 #endif /* CONFIG_HOTPLUG_CPU */
2159 
2160 	for_each_online_cpu(i) {
2161 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2162 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2163 	}
2164 
2165 	mutex_init(&mp->m_icsb_mutex);
2166 
2167 	/*
2168 	 * start with all counters disabled so that the
2169 	 * initial balance kicks us off correctly
2170 	 */
2171 	mp->m_icsb_counters = -1;
2172 	return 0;
2173 }
2174 
2175 void
2176 xfs_icsb_reinit_counters(
2177 	xfs_mount_t	*mp)
2178 {
2179 	xfs_icsb_lock(mp);
2180 	/*
2181 	 * start with all counters disabled so that the
2182 	 * initial balance kicks us off correctly
2183 	 */
2184 	mp->m_icsb_counters = -1;
2185 	xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2186 	xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2187 	xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2188 	xfs_icsb_unlock(mp);
2189 }
2190 
2191 void
2192 xfs_icsb_destroy_counters(
2193 	xfs_mount_t	*mp)
2194 {
2195 	if (mp->m_sb_cnts) {
2196 		unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2197 		free_percpu(mp->m_sb_cnts);
2198 	}
2199 	mutex_destroy(&mp->m_icsb_mutex);
2200 }
2201 
2202 STATIC void
2203 xfs_icsb_lock_cntr(
2204 	xfs_icsb_cnts_t	*icsbp)
2205 {
2206 	while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2207 		ndelay(1000);
2208 	}
2209 }
2210 
2211 STATIC void
2212 xfs_icsb_unlock_cntr(
2213 	xfs_icsb_cnts_t	*icsbp)
2214 {
2215 	clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2216 }
2217 
2218 
2219 STATIC void
2220 xfs_icsb_lock_all_counters(
2221 	xfs_mount_t	*mp)
2222 {
2223 	xfs_icsb_cnts_t *cntp;
2224 	int		i;
2225 
2226 	for_each_online_cpu(i) {
2227 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2228 		xfs_icsb_lock_cntr(cntp);
2229 	}
2230 }
2231 
2232 STATIC void
2233 xfs_icsb_unlock_all_counters(
2234 	xfs_mount_t	*mp)
2235 {
2236 	xfs_icsb_cnts_t *cntp;
2237 	int		i;
2238 
2239 	for_each_online_cpu(i) {
2240 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2241 		xfs_icsb_unlock_cntr(cntp);
2242 	}
2243 }
2244 
2245 STATIC void
2246 xfs_icsb_count(
2247 	xfs_mount_t	*mp,
2248 	xfs_icsb_cnts_t	*cnt,
2249 	int		flags)
2250 {
2251 	xfs_icsb_cnts_t *cntp;
2252 	int		i;
2253 
2254 	memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2255 
2256 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2257 		xfs_icsb_lock_all_counters(mp);
2258 
2259 	for_each_online_cpu(i) {
2260 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2261 		cnt->icsb_icount += cntp->icsb_icount;
2262 		cnt->icsb_ifree += cntp->icsb_ifree;
2263 		cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2264 	}
2265 
2266 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2267 		xfs_icsb_unlock_all_counters(mp);
2268 }
2269 
2270 STATIC int
2271 xfs_icsb_counter_disabled(
2272 	xfs_mount_t	*mp,
2273 	xfs_sb_field_t	field)
2274 {
2275 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2276 	return test_bit(field, &mp->m_icsb_counters);
2277 }
2278 
2279 STATIC void
2280 xfs_icsb_disable_counter(
2281 	xfs_mount_t	*mp,
2282 	xfs_sb_field_t	field)
2283 {
2284 	xfs_icsb_cnts_t	cnt;
2285 
2286 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2287 
2288 	/*
2289 	 * If we are already disabled, then there is nothing to do
2290 	 * here. We check before locking all the counters to avoid
2291 	 * the expensive lock operation when being called in the
2292 	 * slow path and the counter is already disabled. This is
2293 	 * safe because the only time we set or clear this state is under
2294 	 * the m_icsb_mutex.
2295 	 */
2296 	if (xfs_icsb_counter_disabled(mp, field))
2297 		return;
2298 
2299 	xfs_icsb_lock_all_counters(mp);
2300 	if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2301 		/* drain back to superblock */
2302 
2303 		xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2304 		switch(field) {
2305 		case XFS_SBS_ICOUNT:
2306 			mp->m_sb.sb_icount = cnt.icsb_icount;
2307 			break;
2308 		case XFS_SBS_IFREE:
2309 			mp->m_sb.sb_ifree = cnt.icsb_ifree;
2310 			break;
2311 		case XFS_SBS_FDBLOCKS:
2312 			mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2313 			break;
2314 		default:
2315 			BUG();
2316 		}
2317 	}
2318 
2319 	xfs_icsb_unlock_all_counters(mp);
2320 }
2321 
2322 STATIC void
2323 xfs_icsb_enable_counter(
2324 	xfs_mount_t	*mp,
2325 	xfs_sb_field_t	field,
2326 	uint64_t	count,
2327 	uint64_t	resid)
2328 {
2329 	xfs_icsb_cnts_t	*cntp;
2330 	int		i;
2331 
2332 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2333 
2334 	xfs_icsb_lock_all_counters(mp);
2335 	for_each_online_cpu(i) {
2336 		cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2337 		switch (field) {
2338 		case XFS_SBS_ICOUNT:
2339 			cntp->icsb_icount = count + resid;
2340 			break;
2341 		case XFS_SBS_IFREE:
2342 			cntp->icsb_ifree = count + resid;
2343 			break;
2344 		case XFS_SBS_FDBLOCKS:
2345 			cntp->icsb_fdblocks = count + resid;
2346 			break;
2347 		default:
2348 			BUG();
2349 			break;
2350 		}
2351 		resid = 0;
2352 	}
2353 	clear_bit(field, &mp->m_icsb_counters);
2354 	xfs_icsb_unlock_all_counters(mp);
2355 }
2356 
2357 void
2358 xfs_icsb_sync_counters_locked(
2359 	xfs_mount_t	*mp,
2360 	int		flags)
2361 {
2362 	xfs_icsb_cnts_t	cnt;
2363 
2364 	xfs_icsb_count(mp, &cnt, flags);
2365 
2366 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2367 		mp->m_sb.sb_icount = cnt.icsb_icount;
2368 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2369 		mp->m_sb.sb_ifree = cnt.icsb_ifree;
2370 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2371 		mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2372 }
2373 
2374 /*
2375  * Accurate update of per-cpu counters to incore superblock
2376  */
2377 void
2378 xfs_icsb_sync_counters(
2379 	xfs_mount_t	*mp,
2380 	int		flags)
2381 {
2382 	spin_lock(&mp->m_sb_lock);
2383 	xfs_icsb_sync_counters_locked(mp, flags);
2384 	spin_unlock(&mp->m_sb_lock);
2385 }
2386 
2387 /*
2388  * Balance and enable/disable counters as necessary.
2389  *
2390  * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2391  * chosen to be the same number as single on disk allocation chunk per CPU, and
2392  * free blocks is something far enough zero that we aren't going thrash when we
2393  * get near ENOSPC. We also need to supply a minimum we require per cpu to
2394  * prevent looping endlessly when xfs_alloc_space asks for more than will
2395  * be distributed to a single CPU but each CPU has enough blocks to be
2396  * reenabled.
2397  *
2398  * Note that we can be called when counters are already disabled.
2399  * xfs_icsb_disable_counter() optimises the counter locking in this case to
2400  * prevent locking every per-cpu counter needlessly.
2401  */
2402 
2403 #define XFS_ICSB_INO_CNTR_REENABLE	(uint64_t)64
2404 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2405 		(uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2406 STATIC void
2407 xfs_icsb_balance_counter_locked(
2408 	xfs_mount_t	*mp,
2409 	xfs_sb_field_t  field,
2410 	int		min_per_cpu)
2411 {
2412 	uint64_t	count, resid;
2413 	int		weight = num_online_cpus();
2414 	uint64_t	min = (uint64_t)min_per_cpu;
2415 
2416 	/* disable counter and sync counter */
2417 	xfs_icsb_disable_counter(mp, field);
2418 
2419 	/* update counters  - first CPU gets residual*/
2420 	switch (field) {
2421 	case XFS_SBS_ICOUNT:
2422 		count = mp->m_sb.sb_icount;
2423 		resid = do_div(count, weight);
2424 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2425 			return;
2426 		break;
2427 	case XFS_SBS_IFREE:
2428 		count = mp->m_sb.sb_ifree;
2429 		resid = do_div(count, weight);
2430 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2431 			return;
2432 		break;
2433 	case XFS_SBS_FDBLOCKS:
2434 		count = mp->m_sb.sb_fdblocks;
2435 		resid = do_div(count, weight);
2436 		if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2437 			return;
2438 		break;
2439 	default:
2440 		BUG();
2441 		count = resid = 0;	/* quiet, gcc */
2442 		break;
2443 	}
2444 
2445 	xfs_icsb_enable_counter(mp, field, count, resid);
2446 }
2447 
2448 STATIC void
2449 xfs_icsb_balance_counter(
2450 	xfs_mount_t	*mp,
2451 	xfs_sb_field_t  fields,
2452 	int		min_per_cpu)
2453 {
2454 	spin_lock(&mp->m_sb_lock);
2455 	xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2456 	spin_unlock(&mp->m_sb_lock);
2457 }
2458 
2459 int
2460 xfs_icsb_modify_counters(
2461 	xfs_mount_t	*mp,
2462 	xfs_sb_field_t	field,
2463 	int64_t		delta,
2464 	int		rsvd)
2465 {
2466 	xfs_icsb_cnts_t	*icsbp;
2467 	long long	lcounter;	/* long counter for 64 bit fields */
2468 	int		ret = 0;
2469 
2470 	might_sleep();
2471 again:
2472 	preempt_disable();
2473 	icsbp = this_cpu_ptr(mp->m_sb_cnts);
2474 
2475 	/*
2476 	 * if the counter is disabled, go to slow path
2477 	 */
2478 	if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2479 		goto slow_path;
2480 	xfs_icsb_lock_cntr(icsbp);
2481 	if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2482 		xfs_icsb_unlock_cntr(icsbp);
2483 		goto slow_path;
2484 	}
2485 
2486 	switch (field) {
2487 	case XFS_SBS_ICOUNT:
2488 		lcounter = icsbp->icsb_icount;
2489 		lcounter += delta;
2490 		if (unlikely(lcounter < 0))
2491 			goto balance_counter;
2492 		icsbp->icsb_icount = lcounter;
2493 		break;
2494 
2495 	case XFS_SBS_IFREE:
2496 		lcounter = icsbp->icsb_ifree;
2497 		lcounter += delta;
2498 		if (unlikely(lcounter < 0))
2499 			goto balance_counter;
2500 		icsbp->icsb_ifree = lcounter;
2501 		break;
2502 
2503 	case XFS_SBS_FDBLOCKS:
2504 		BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2505 
2506 		lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2507 		lcounter += delta;
2508 		if (unlikely(lcounter < 0))
2509 			goto balance_counter;
2510 		icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2511 		break;
2512 	default:
2513 		BUG();
2514 		break;
2515 	}
2516 	xfs_icsb_unlock_cntr(icsbp);
2517 	preempt_enable();
2518 	return 0;
2519 
2520 slow_path:
2521 	preempt_enable();
2522 
2523 	/*
2524 	 * serialise with a mutex so we don't burn lots of cpu on
2525 	 * the superblock lock. We still need to hold the superblock
2526 	 * lock, however, when we modify the global structures.
2527 	 */
2528 	xfs_icsb_lock(mp);
2529 
2530 	/*
2531 	 * Now running atomically.
2532 	 *
2533 	 * If the counter is enabled, someone has beaten us to rebalancing.
2534 	 * Drop the lock and try again in the fast path....
2535 	 */
2536 	if (!(xfs_icsb_counter_disabled(mp, field))) {
2537 		xfs_icsb_unlock(mp);
2538 		goto again;
2539 	}
2540 
2541 	/*
2542 	 * The counter is currently disabled. Because we are
2543 	 * running atomically here, we know a rebalance cannot
2544 	 * be in progress. Hence we can go straight to operating
2545 	 * on the global superblock. We do not call xfs_mod_incore_sb()
2546 	 * here even though we need to get the m_sb_lock. Doing so
2547 	 * will cause us to re-enter this function and deadlock.
2548 	 * Hence we get the m_sb_lock ourselves and then call
2549 	 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2550 	 * directly on the global counters.
2551 	 */
2552 	spin_lock(&mp->m_sb_lock);
2553 	ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2554 	spin_unlock(&mp->m_sb_lock);
2555 
2556 	/*
2557 	 * Now that we've modified the global superblock, we
2558 	 * may be able to re-enable the distributed counters
2559 	 * (e.g. lots of space just got freed). After that
2560 	 * we are done.
2561 	 */
2562 	if (ret != ENOSPC)
2563 		xfs_icsb_balance_counter(mp, field, 0);
2564 	xfs_icsb_unlock(mp);
2565 	return ret;
2566 
2567 balance_counter:
2568 	xfs_icsb_unlock_cntr(icsbp);
2569 	preempt_enable();
2570 
2571 	/*
2572 	 * We may have multiple threads here if multiple per-cpu
2573 	 * counters run dry at the same time. This will mean we can
2574 	 * do more balances than strictly necessary but it is not
2575 	 * the common slowpath case.
2576 	 */
2577 	xfs_icsb_lock(mp);
2578 
2579 	/*
2580 	 * running atomically.
2581 	 *
2582 	 * This will leave the counter in the correct state for future
2583 	 * accesses. After the rebalance, we simply try again and our retry
2584 	 * will either succeed through the fast path or slow path without
2585 	 * another balance operation being required.
2586 	 */
2587 	xfs_icsb_balance_counter(mp, field, delta);
2588 	xfs_icsb_unlock(mp);
2589 	goto again;
2590 }
2591 
2592 #endif
2593