xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision 171f1bc7)
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 #ifdef HAVE_PERCPU_SB
48 STATIC void	xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
49 						int);
50 STATIC void	xfs_icsb_balance_counter_locked(xfs_mount_t *, xfs_sb_field_t,
51 						int);
52 STATIC void	xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
53 #else
54 
55 #define xfs_icsb_balance_counter(mp, a, b)		do { } while (0)
56 #define xfs_icsb_balance_counter_locked(mp, a, b)	do { } while (0)
57 #endif
58 
59 static const struct {
60 	short offset;
61 	short type;	/* 0 = integer
62 			 * 1 = binary / string (no translation)
63 			 */
64 } xfs_sb_info[] = {
65     { offsetof(xfs_sb_t, sb_magicnum),   0 },
66     { offsetof(xfs_sb_t, sb_blocksize),  0 },
67     { offsetof(xfs_sb_t, sb_dblocks),    0 },
68     { offsetof(xfs_sb_t, sb_rblocks),    0 },
69     { offsetof(xfs_sb_t, sb_rextents),   0 },
70     { offsetof(xfs_sb_t, sb_uuid),       1 },
71     { offsetof(xfs_sb_t, sb_logstart),   0 },
72     { offsetof(xfs_sb_t, sb_rootino),    0 },
73     { offsetof(xfs_sb_t, sb_rbmino),     0 },
74     { offsetof(xfs_sb_t, sb_rsumino),    0 },
75     { offsetof(xfs_sb_t, sb_rextsize),   0 },
76     { offsetof(xfs_sb_t, sb_agblocks),   0 },
77     { offsetof(xfs_sb_t, sb_agcount),    0 },
78     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
79     { offsetof(xfs_sb_t, sb_logblocks),  0 },
80     { offsetof(xfs_sb_t, sb_versionnum), 0 },
81     { offsetof(xfs_sb_t, sb_sectsize),   0 },
82     { offsetof(xfs_sb_t, sb_inodesize),  0 },
83     { offsetof(xfs_sb_t, sb_inopblock),  0 },
84     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
85     { offsetof(xfs_sb_t, sb_blocklog),   0 },
86     { offsetof(xfs_sb_t, sb_sectlog),    0 },
87     { offsetof(xfs_sb_t, sb_inodelog),   0 },
88     { offsetof(xfs_sb_t, sb_inopblog),   0 },
89     { offsetof(xfs_sb_t, sb_agblklog),   0 },
90     { offsetof(xfs_sb_t, sb_rextslog),   0 },
91     { offsetof(xfs_sb_t, sb_inprogress), 0 },
92     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
93     { offsetof(xfs_sb_t, sb_icount),     0 },
94     { offsetof(xfs_sb_t, sb_ifree),      0 },
95     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
96     { offsetof(xfs_sb_t, sb_frextents),  0 },
97     { offsetof(xfs_sb_t, sb_uquotino),   0 },
98     { offsetof(xfs_sb_t, sb_gquotino),   0 },
99     { offsetof(xfs_sb_t, sb_qflags),     0 },
100     { offsetof(xfs_sb_t, sb_flags),      0 },
101     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
102     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
103     { offsetof(xfs_sb_t, sb_unit),	 0 },
104     { offsetof(xfs_sb_t, sb_width),	 0 },
105     { offsetof(xfs_sb_t, sb_dirblklog),	 0 },
106     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
107     { offsetof(xfs_sb_t, sb_logsectsize),0 },
108     { offsetof(xfs_sb_t, sb_logsunit),	 0 },
109     { offsetof(xfs_sb_t, sb_features2),	 0 },
110     { offsetof(xfs_sb_t, sb_bad_features2), 0 },
111     { sizeof(xfs_sb_t),			 0 }
112 };
113 
114 static DEFINE_MUTEX(xfs_uuid_table_mutex);
115 static int xfs_uuid_table_size;
116 static uuid_t *xfs_uuid_table;
117 
118 /*
119  * See if the UUID is unique among mounted XFS filesystems.
120  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
121  */
122 STATIC int
123 xfs_uuid_mount(
124 	struct xfs_mount	*mp)
125 {
126 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
127 	int			hole, i;
128 
129 	if (mp->m_flags & XFS_MOUNT_NOUUID)
130 		return 0;
131 
132 	if (uuid_is_nil(uuid)) {
133 		xfs_warn(mp, "Filesystem has nil UUID - can't mount");
134 		return XFS_ERROR(EINVAL);
135 	}
136 
137 	mutex_lock(&xfs_uuid_table_mutex);
138 	for (i = 0, hole = -1; i < xfs_uuid_table_size; i++) {
139 		if (uuid_is_nil(&xfs_uuid_table[i])) {
140 			hole = i;
141 			continue;
142 		}
143 		if (uuid_equal(uuid, &xfs_uuid_table[i]))
144 			goto out_duplicate;
145 	}
146 
147 	if (hole < 0) {
148 		xfs_uuid_table = kmem_realloc(xfs_uuid_table,
149 			(xfs_uuid_table_size + 1) * sizeof(*xfs_uuid_table),
150 			xfs_uuid_table_size  * sizeof(*xfs_uuid_table),
151 			KM_SLEEP);
152 		hole = xfs_uuid_table_size++;
153 	}
154 	xfs_uuid_table[hole] = *uuid;
155 	mutex_unlock(&xfs_uuid_table_mutex);
156 
157 	return 0;
158 
159  out_duplicate:
160 	mutex_unlock(&xfs_uuid_table_mutex);
161 	xfs_warn(mp, "Filesystem has duplicate UUID - can't mount");
162 	return XFS_ERROR(EINVAL);
163 }
164 
165 STATIC void
166 xfs_uuid_unmount(
167 	struct xfs_mount	*mp)
168 {
169 	uuid_t			*uuid = &mp->m_sb.sb_uuid;
170 	int			i;
171 
172 	if (mp->m_flags & XFS_MOUNT_NOUUID)
173 		return;
174 
175 	mutex_lock(&xfs_uuid_table_mutex);
176 	for (i = 0; i < xfs_uuid_table_size; i++) {
177 		if (uuid_is_nil(&xfs_uuid_table[i]))
178 			continue;
179 		if (!uuid_equal(uuid, &xfs_uuid_table[i]))
180 			continue;
181 		memset(&xfs_uuid_table[i], 0, sizeof(uuid_t));
182 		break;
183 	}
184 	ASSERT(i < xfs_uuid_table_size);
185 	mutex_unlock(&xfs_uuid_table_mutex);
186 }
187 
188 
189 /*
190  * Reference counting access wrappers to the perag structures.
191  * Because we never free per-ag structures, the only thing we
192  * have to protect against changes is the tree structure itself.
193  */
194 struct xfs_perag *
195 xfs_perag_get(struct xfs_mount *mp, xfs_agnumber_t agno)
196 {
197 	struct xfs_perag	*pag;
198 	int			ref = 0;
199 
200 	rcu_read_lock();
201 	pag = radix_tree_lookup(&mp->m_perag_tree, agno);
202 	if (pag) {
203 		ASSERT(atomic_read(&pag->pag_ref) >= 0);
204 		ref = atomic_inc_return(&pag->pag_ref);
205 	}
206 	rcu_read_unlock();
207 	trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
208 	return pag;
209 }
210 
211 /*
212  * search from @first to find the next perag with the given tag set.
213  */
214 struct xfs_perag *
215 xfs_perag_get_tag(
216 	struct xfs_mount	*mp,
217 	xfs_agnumber_t		first,
218 	int			tag)
219 {
220 	struct xfs_perag	*pag;
221 	int			found;
222 	int			ref;
223 
224 	rcu_read_lock();
225 	found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
226 					(void **)&pag, first, 1, tag);
227 	if (found <= 0) {
228 		rcu_read_unlock();
229 		return NULL;
230 	}
231 	ref = atomic_inc_return(&pag->pag_ref);
232 	rcu_read_unlock();
233 	trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
234 	return pag;
235 }
236 
237 void
238 xfs_perag_put(struct xfs_perag *pag)
239 {
240 	int	ref;
241 
242 	ASSERT(atomic_read(&pag->pag_ref) > 0);
243 	ref = atomic_dec_return(&pag->pag_ref);
244 	trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
245 }
246 
247 STATIC void
248 __xfs_free_perag(
249 	struct rcu_head	*head)
250 {
251 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
252 
253 	ASSERT(atomic_read(&pag->pag_ref) == 0);
254 	kmem_free(pag);
255 }
256 
257 /*
258  * Free up the per-ag resources associated with the mount structure.
259  */
260 STATIC void
261 xfs_free_perag(
262 	xfs_mount_t	*mp)
263 {
264 	xfs_agnumber_t	agno;
265 	struct xfs_perag *pag;
266 
267 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
268 		spin_lock(&mp->m_perag_lock);
269 		pag = radix_tree_delete(&mp->m_perag_tree, agno);
270 		spin_unlock(&mp->m_perag_lock);
271 		ASSERT(pag);
272 		ASSERT(atomic_read(&pag->pag_ref) == 0);
273 		call_rcu(&pag->rcu_head, __xfs_free_perag);
274 	}
275 }
276 
277 /*
278  * Check size of device based on the (data/realtime) block count.
279  * Note: this check is used by the growfs code as well as mount.
280  */
281 int
282 xfs_sb_validate_fsb_count(
283 	xfs_sb_t	*sbp,
284 	__uint64_t	nblocks)
285 {
286 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
287 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
288 
289 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
290 	if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
291 		return EFBIG;
292 #else                  /* Limited by UINT_MAX of sectors */
293 	if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
294 		return EFBIG;
295 #endif
296 	return 0;
297 }
298 
299 /*
300  * Check the validity of the SB found.
301  */
302 STATIC int
303 xfs_mount_validate_sb(
304 	xfs_mount_t	*mp,
305 	xfs_sb_t	*sbp,
306 	int		flags)
307 {
308 	int		loud = !(flags & XFS_MFSI_QUIET);
309 
310 	/*
311 	 * If the log device and data device have the
312 	 * same device number, the log is internal.
313 	 * Consequently, the sb_logstart should be non-zero.  If
314 	 * we have a zero sb_logstart in this case, we may be trying to mount
315 	 * a volume filesystem in a non-volume manner.
316 	 */
317 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
318 		if (loud)
319 			xfs_warn(mp, "bad magic number");
320 		return XFS_ERROR(EWRONGFS);
321 	}
322 
323 	if (!xfs_sb_good_version(sbp)) {
324 		if (loud)
325 			xfs_warn(mp, "bad version");
326 		return XFS_ERROR(EWRONGFS);
327 	}
328 
329 	if (unlikely(
330 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
331 		if (loud)
332 			xfs_warn(mp,
333 		"filesystem is marked as having an external log; "
334 		"specify logdev on the mount command line.");
335 		return XFS_ERROR(EINVAL);
336 	}
337 
338 	if (unlikely(
339 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
340 		if (loud)
341 			xfs_warn(mp,
342 		"filesystem is marked as having an internal log; "
343 		"do not specify logdev on the mount command line.");
344 		return XFS_ERROR(EINVAL);
345 	}
346 
347 	/*
348 	 * More sanity checking.  Most of 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 	    sbp->sb_dblocks == 0					||
373 	    sbp->sb_dblocks > XFS_MAX_DBLOCKS(sbp)			||
374 	    sbp->sb_dblocks < XFS_MIN_DBLOCKS(sbp))) {
375 		if (loud)
376 			XFS_CORRUPTION_ERROR("SB sanity check failed",
377 				XFS_ERRLEVEL_LOW, mp, sbp);
378 		return XFS_ERROR(EFSCORRUPTED);
379 	}
380 
381 	/*
382 	 * Until this is fixed only page-sized or smaller data blocks work.
383 	 */
384 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
385 		if (loud) {
386 			xfs_warn(mp,
387 		"File system with blocksize %d bytes. "
388 		"Only pagesize (%ld) or less will currently work.",
389 				sbp->sb_blocksize, PAGE_SIZE);
390 		}
391 		return XFS_ERROR(ENOSYS);
392 	}
393 
394 	/*
395 	 * Currently only very few inode sizes are supported.
396 	 */
397 	switch (sbp->sb_inodesize) {
398 	case 256:
399 	case 512:
400 	case 1024:
401 	case 2048:
402 		break;
403 	default:
404 		if (loud)
405 			xfs_warn(mp, "inode size of %d bytes not supported",
406 				sbp->sb_inodesize);
407 		return XFS_ERROR(ENOSYS);
408 	}
409 
410 	if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
411 	    xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
412 		if (loud)
413 			xfs_warn(mp,
414 		"file system too large to be mounted on this system.");
415 		return XFS_ERROR(EFBIG);
416 	}
417 
418 	if (unlikely(sbp->sb_inprogress)) {
419 		if (loud)
420 			xfs_warn(mp, "file system busy");
421 		return XFS_ERROR(EFSCORRUPTED);
422 	}
423 
424 	/*
425 	 * Version 1 directory format has never worked on Linux.
426 	 */
427 	if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
428 		if (loud)
429 			xfs_warn(mp,
430 				"file system using version 1 directory format");
431 		return XFS_ERROR(ENOSYS);
432 	}
433 
434 	return 0;
435 }
436 
437 int
438 xfs_initialize_perag(
439 	xfs_mount_t	*mp,
440 	xfs_agnumber_t	agcount,
441 	xfs_agnumber_t	*maxagi)
442 {
443 	xfs_agnumber_t	index, max_metadata;
444 	xfs_agnumber_t	first_initialised = 0;
445 	xfs_perag_t	*pag;
446 	xfs_agino_t	agino;
447 	xfs_ino_t	ino;
448 	xfs_sb_t	*sbp = &mp->m_sb;
449 	int		error = -ENOMEM;
450 
451 	/*
452 	 * Walk the current per-ag tree so we don't try to initialise AGs
453 	 * that already exist (growfs case). Allocate and insert all the
454 	 * AGs we don't find ready for initialisation.
455 	 */
456 	for (index = 0; index < agcount; index++) {
457 		pag = xfs_perag_get(mp, index);
458 		if (pag) {
459 			xfs_perag_put(pag);
460 			continue;
461 		}
462 		if (!first_initialised)
463 			first_initialised = index;
464 
465 		pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
466 		if (!pag)
467 			goto out_unwind;
468 		pag->pag_agno = index;
469 		pag->pag_mount = mp;
470 		spin_lock_init(&pag->pag_ici_lock);
471 		mutex_init(&pag->pag_ici_reclaim_lock);
472 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
473 		spin_lock_init(&pag->pag_buf_lock);
474 		pag->pag_buf_tree = RB_ROOT;
475 
476 		if (radix_tree_preload(GFP_NOFS))
477 			goto out_unwind;
478 
479 		spin_lock(&mp->m_perag_lock);
480 		if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
481 			BUG();
482 			spin_unlock(&mp->m_perag_lock);
483 			radix_tree_preload_end();
484 			error = -EEXIST;
485 			goto out_unwind;
486 		}
487 		spin_unlock(&mp->m_perag_lock);
488 		radix_tree_preload_end();
489 	}
490 
491 	/*
492 	 * If we mount with the inode64 option, or no inode overflows
493 	 * the legacy 32-bit address space clear the inode32 option.
494 	 */
495 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
496 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
497 
498 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > XFS_MAXINUMBER_32)
499 		mp->m_flags |= XFS_MOUNT_32BITINODES;
500 	else
501 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
502 
503 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
504 		/*
505 		 * Calculate how much should be reserved for inodes to meet
506 		 * the max inode percentage.
507 		 */
508 		if (mp->m_maxicount) {
509 			__uint64_t	icount;
510 
511 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
512 			do_div(icount, 100);
513 			icount += sbp->sb_agblocks - 1;
514 			do_div(icount, sbp->sb_agblocks);
515 			max_metadata = icount;
516 		} else {
517 			max_metadata = agcount;
518 		}
519 
520 		for (index = 0; index < agcount; index++) {
521 			ino = XFS_AGINO_TO_INO(mp, index, agino);
522 			if (ino > XFS_MAXINUMBER_32) {
523 				index++;
524 				break;
525 			}
526 
527 			pag = xfs_perag_get(mp, index);
528 			pag->pagi_inodeok = 1;
529 			if (index < max_metadata)
530 				pag->pagf_metadata = 1;
531 			xfs_perag_put(pag);
532 		}
533 	} else {
534 		for (index = 0; index < agcount; index++) {
535 			pag = xfs_perag_get(mp, index);
536 			pag->pagi_inodeok = 1;
537 			xfs_perag_put(pag);
538 		}
539 	}
540 
541 	if (maxagi)
542 		*maxagi = index;
543 	return 0;
544 
545 out_unwind:
546 	kmem_free(pag);
547 	for (; index > first_initialised; index--) {
548 		pag = radix_tree_delete(&mp->m_perag_tree, index);
549 		kmem_free(pag);
550 	}
551 	return error;
552 }
553 
554 void
555 xfs_sb_from_disk(
556 	xfs_sb_t	*to,
557 	xfs_dsb_t	*from)
558 {
559 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
560 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
561 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
562 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
563 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
564 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
565 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
566 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
567 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
568 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
569 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
570 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
571 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
572 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
573 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
574 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
575 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
576 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
577 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
578 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
579 	to->sb_blocklog = from->sb_blocklog;
580 	to->sb_sectlog = from->sb_sectlog;
581 	to->sb_inodelog = from->sb_inodelog;
582 	to->sb_inopblog = from->sb_inopblog;
583 	to->sb_agblklog = from->sb_agblklog;
584 	to->sb_rextslog = from->sb_rextslog;
585 	to->sb_inprogress = from->sb_inprogress;
586 	to->sb_imax_pct = from->sb_imax_pct;
587 	to->sb_icount = be64_to_cpu(from->sb_icount);
588 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
589 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
590 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
591 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
592 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
593 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
594 	to->sb_flags = from->sb_flags;
595 	to->sb_shared_vn = from->sb_shared_vn;
596 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
597 	to->sb_unit = be32_to_cpu(from->sb_unit);
598 	to->sb_width = be32_to_cpu(from->sb_width);
599 	to->sb_dirblklog = from->sb_dirblklog;
600 	to->sb_logsectlog = from->sb_logsectlog;
601 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
602 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
603 	to->sb_features2 = be32_to_cpu(from->sb_features2);
604 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
605 }
606 
607 /*
608  * Copy in core superblock to ondisk one.
609  *
610  * The fields argument is mask of superblock fields to copy.
611  */
612 void
613 xfs_sb_to_disk(
614 	xfs_dsb_t	*to,
615 	xfs_sb_t	*from,
616 	__int64_t	fields)
617 {
618 	xfs_caddr_t	to_ptr = (xfs_caddr_t)to;
619 	xfs_caddr_t	from_ptr = (xfs_caddr_t)from;
620 	xfs_sb_field_t	f;
621 	int		first;
622 	int		size;
623 
624 	ASSERT(fields);
625 	if (!fields)
626 		return;
627 
628 	while (fields) {
629 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
630 		first = xfs_sb_info[f].offset;
631 		size = xfs_sb_info[f + 1].offset - first;
632 
633 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
634 
635 		if (size == 1 || xfs_sb_info[f].type == 1) {
636 			memcpy(to_ptr + first, from_ptr + first, size);
637 		} else {
638 			switch (size) {
639 			case 2:
640 				*(__be16 *)(to_ptr + first) =
641 					cpu_to_be16(*(__u16 *)(from_ptr + first));
642 				break;
643 			case 4:
644 				*(__be32 *)(to_ptr + first) =
645 					cpu_to_be32(*(__u32 *)(from_ptr + first));
646 				break;
647 			case 8:
648 				*(__be64 *)(to_ptr + first) =
649 					cpu_to_be64(*(__u64 *)(from_ptr + first));
650 				break;
651 			default:
652 				ASSERT(0);
653 			}
654 		}
655 
656 		fields &= ~(1LL << f);
657 	}
658 }
659 
660 /*
661  * xfs_readsb
662  *
663  * Does the initial read of the superblock.
664  */
665 int
666 xfs_readsb(xfs_mount_t *mp, int flags)
667 {
668 	unsigned int	sector_size;
669 	xfs_buf_t	*bp;
670 	int		error;
671 	int		loud = !(flags & XFS_MFSI_QUIET);
672 
673 	ASSERT(mp->m_sb_bp == NULL);
674 	ASSERT(mp->m_ddev_targp != NULL);
675 
676 	/*
677 	 * Allocate a (locked) buffer to hold the superblock.
678 	 * This will be kept around at all times to optimize
679 	 * access to the superblock.
680 	 */
681 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
682 
683 reread:
684 	bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
685 					XFS_SB_DADDR, sector_size, 0);
686 	if (!bp) {
687 		if (loud)
688 			xfs_warn(mp, "SB buffer read failed");
689 		return EIO;
690 	}
691 
692 	/*
693 	 * Initialize the mount structure from the superblock.
694 	 * But first do some basic consistency checking.
695 	 */
696 	xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
697 	error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
698 	if (error) {
699 		if (loud)
700 			xfs_warn(mp, "SB validate failed");
701 		goto release_buf;
702 	}
703 
704 	/*
705 	 * We must be able to do sector-sized and sector-aligned IO.
706 	 */
707 	if (sector_size > mp->m_sb.sb_sectsize) {
708 		if (loud)
709 			xfs_warn(mp, "device supports %u byte sectors (not %u)",
710 				sector_size, mp->m_sb.sb_sectsize);
711 		error = ENOSYS;
712 		goto release_buf;
713 	}
714 
715 	/*
716 	 * If device sector size is smaller than the superblock size,
717 	 * re-read the superblock so the buffer is correctly sized.
718 	 */
719 	if (sector_size < mp->m_sb.sb_sectsize) {
720 		xfs_buf_relse(bp);
721 		sector_size = mp->m_sb.sb_sectsize;
722 		goto reread;
723 	}
724 
725 	/* Initialize per-cpu counters */
726 	xfs_icsb_reinit_counters(mp);
727 
728 	mp->m_sb_bp = bp;
729 	xfs_buf_unlock(bp);
730 	return 0;
731 
732 release_buf:
733 	xfs_buf_relse(bp);
734 	return error;
735 }
736 
737 
738 /*
739  * xfs_mount_common
740  *
741  * Mount initialization code establishing various mount
742  * fields from the superblock associated with the given
743  * mount structure
744  */
745 STATIC void
746 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
747 {
748 	mp->m_agfrotor = mp->m_agirotor = 0;
749 	spin_lock_init(&mp->m_agirotor_lock);
750 	mp->m_maxagi = mp->m_sb.sb_agcount;
751 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
752 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
753 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
754 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
755 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
756 	mp->m_blockmask = sbp->sb_blocksize - 1;
757 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
758 	mp->m_blockwmask = mp->m_blockwsize - 1;
759 
760 	mp->m_alloc_mxr[0] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 1);
761 	mp->m_alloc_mxr[1] = xfs_allocbt_maxrecs(mp, sbp->sb_blocksize, 0);
762 	mp->m_alloc_mnr[0] = mp->m_alloc_mxr[0] / 2;
763 	mp->m_alloc_mnr[1] = mp->m_alloc_mxr[1] / 2;
764 
765 	mp->m_inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
766 	mp->m_inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
767 	mp->m_inobt_mnr[0] = mp->m_inobt_mxr[0] / 2;
768 	mp->m_inobt_mnr[1] = mp->m_inobt_mxr[1] / 2;
769 
770 	mp->m_bmap_dmxr[0] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 1);
771 	mp->m_bmap_dmxr[1] = xfs_bmbt_maxrecs(mp, sbp->sb_blocksize, 0);
772 	mp->m_bmap_dmnr[0] = mp->m_bmap_dmxr[0] / 2;
773 	mp->m_bmap_dmnr[1] = mp->m_bmap_dmxr[1] / 2;
774 
775 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
776 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
777 					sbp->sb_inopblock);
778 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
779 }
780 
781 /*
782  * xfs_initialize_perag_data
783  *
784  * Read in each per-ag structure so we can count up the number of
785  * allocated inodes, free inodes and used filesystem blocks as this
786  * information is no longer persistent in the superblock. Once we have
787  * this information, write it into the in-core superblock structure.
788  */
789 STATIC int
790 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
791 {
792 	xfs_agnumber_t	index;
793 	xfs_perag_t	*pag;
794 	xfs_sb_t	*sbp = &mp->m_sb;
795 	uint64_t	ifree = 0;
796 	uint64_t	ialloc = 0;
797 	uint64_t	bfree = 0;
798 	uint64_t	bfreelst = 0;
799 	uint64_t	btree = 0;
800 	int		error;
801 
802 	for (index = 0; index < agcount; index++) {
803 		/*
804 		 * read the agf, then the agi. This gets us
805 		 * all the information we need and populates the
806 		 * per-ag structures for us.
807 		 */
808 		error = xfs_alloc_pagf_init(mp, NULL, index, 0);
809 		if (error)
810 			return error;
811 
812 		error = xfs_ialloc_pagi_init(mp, NULL, index);
813 		if (error)
814 			return error;
815 		pag = xfs_perag_get(mp, index);
816 		ifree += pag->pagi_freecount;
817 		ialloc += pag->pagi_count;
818 		bfree += pag->pagf_freeblks;
819 		bfreelst += pag->pagf_flcount;
820 		btree += pag->pagf_btreeblks;
821 		xfs_perag_put(pag);
822 	}
823 	/*
824 	 * Overwrite incore superblock counters with just-read data
825 	 */
826 	spin_lock(&mp->m_sb_lock);
827 	sbp->sb_ifree = ifree;
828 	sbp->sb_icount = ialloc;
829 	sbp->sb_fdblocks = bfree + bfreelst + btree;
830 	spin_unlock(&mp->m_sb_lock);
831 
832 	/* Fixup the per-cpu counters as well. */
833 	xfs_icsb_reinit_counters(mp);
834 
835 	return 0;
836 }
837 
838 /*
839  * Update alignment values based on mount options and sb values
840  */
841 STATIC int
842 xfs_update_alignment(xfs_mount_t *mp)
843 {
844 	xfs_sb_t	*sbp = &(mp->m_sb);
845 
846 	if (mp->m_dalign) {
847 		/*
848 		 * If stripe unit and stripe width are not multiples
849 		 * of the fs blocksize turn off alignment.
850 		 */
851 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
852 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
853 			if (mp->m_flags & XFS_MOUNT_RETERR) {
854 				xfs_warn(mp, "alignment check failed: "
855 					 "(sunit/swidth vs. blocksize)");
856 				return XFS_ERROR(EINVAL);
857 			}
858 			mp->m_dalign = mp->m_swidth = 0;
859 		} else {
860 			/*
861 			 * Convert the stripe unit and width to FSBs.
862 			 */
863 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
864 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
865 				if (mp->m_flags & XFS_MOUNT_RETERR) {
866 					xfs_warn(mp, "alignment check failed: "
867 						 "(sunit/swidth vs. ag size)");
868 					return XFS_ERROR(EINVAL);
869 				}
870 				xfs_warn(mp,
871 		"stripe alignment turned off: sunit(%d)/swidth(%d) "
872 		"incompatible with agsize(%d)",
873 					mp->m_dalign, mp->m_swidth,
874 					sbp->sb_agblocks);
875 
876 				mp->m_dalign = 0;
877 				mp->m_swidth = 0;
878 			} else if (mp->m_dalign) {
879 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
880 			} else {
881 				if (mp->m_flags & XFS_MOUNT_RETERR) {
882 					xfs_warn(mp, "alignment check failed: "
883 						"sunit(%d) less than bsize(%d)",
884 						mp->m_dalign,
885 						mp->m_blockmask +1);
886 					return XFS_ERROR(EINVAL);
887 				}
888 				mp->m_swidth = 0;
889 			}
890 		}
891 
892 		/*
893 		 * Update superblock with new values
894 		 * and log changes
895 		 */
896 		if (xfs_sb_version_hasdalign(sbp)) {
897 			if (sbp->sb_unit != mp->m_dalign) {
898 				sbp->sb_unit = mp->m_dalign;
899 				mp->m_update_flags |= XFS_SB_UNIT;
900 			}
901 			if (sbp->sb_width != mp->m_swidth) {
902 				sbp->sb_width = mp->m_swidth;
903 				mp->m_update_flags |= XFS_SB_WIDTH;
904 			}
905 		}
906 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
907 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
908 			mp->m_dalign = sbp->sb_unit;
909 			mp->m_swidth = sbp->sb_width;
910 	}
911 
912 	return 0;
913 }
914 
915 /*
916  * Set the maximum inode count for this filesystem
917  */
918 STATIC void
919 xfs_set_maxicount(xfs_mount_t *mp)
920 {
921 	xfs_sb_t	*sbp = &(mp->m_sb);
922 	__uint64_t	icount;
923 
924 	if (sbp->sb_imax_pct) {
925 		/*
926 		 * Make sure the maximum inode count is a multiple
927 		 * of the units we allocate inodes in.
928 		 */
929 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
930 		do_div(icount, 100);
931 		do_div(icount, mp->m_ialloc_blks);
932 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
933 				   sbp->sb_inopblog;
934 	} else {
935 		mp->m_maxicount = 0;
936 	}
937 }
938 
939 /*
940  * Set the default minimum read and write sizes unless
941  * already specified in a mount option.
942  * We use smaller I/O sizes when the file system
943  * is being used for NFS service (wsync mount option).
944  */
945 STATIC void
946 xfs_set_rw_sizes(xfs_mount_t *mp)
947 {
948 	xfs_sb_t	*sbp = &(mp->m_sb);
949 	int		readio_log, writeio_log;
950 
951 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
952 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
953 			readio_log = XFS_WSYNC_READIO_LOG;
954 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
955 		} else {
956 			readio_log = XFS_READIO_LOG_LARGE;
957 			writeio_log = XFS_WRITEIO_LOG_LARGE;
958 		}
959 	} else {
960 		readio_log = mp->m_readio_log;
961 		writeio_log = mp->m_writeio_log;
962 	}
963 
964 	if (sbp->sb_blocklog > readio_log) {
965 		mp->m_readio_log = sbp->sb_blocklog;
966 	} else {
967 		mp->m_readio_log = readio_log;
968 	}
969 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
970 	if (sbp->sb_blocklog > writeio_log) {
971 		mp->m_writeio_log = sbp->sb_blocklog;
972 	} else {
973 		mp->m_writeio_log = writeio_log;
974 	}
975 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
976 }
977 
978 /*
979  * precalculate the low space thresholds for dynamic speculative preallocation.
980  */
981 void
982 xfs_set_low_space_thresholds(
983 	struct xfs_mount	*mp)
984 {
985 	int i;
986 
987 	for (i = 0; i < XFS_LOWSP_MAX; i++) {
988 		__uint64_t space = mp->m_sb.sb_dblocks;
989 
990 		do_div(space, 100);
991 		mp->m_low_space[i] = space * (i + 1);
992 	}
993 }
994 
995 
996 /*
997  * Set whether we're using inode alignment.
998  */
999 STATIC void
1000 xfs_set_inoalignment(xfs_mount_t *mp)
1001 {
1002 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
1003 	    mp->m_sb.sb_inoalignmt >=
1004 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
1005 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
1006 	else
1007 		mp->m_inoalign_mask = 0;
1008 	/*
1009 	 * If we are using stripe alignment, check whether
1010 	 * the stripe unit is a multiple of the inode alignment
1011 	 */
1012 	if (mp->m_dalign && mp->m_inoalign_mask &&
1013 	    !(mp->m_dalign & mp->m_inoalign_mask))
1014 		mp->m_sinoalign = mp->m_dalign;
1015 	else
1016 		mp->m_sinoalign = 0;
1017 }
1018 
1019 /*
1020  * Check that the data (and log if separate) are an ok size.
1021  */
1022 STATIC int
1023 xfs_check_sizes(xfs_mount_t *mp)
1024 {
1025 	xfs_buf_t	*bp;
1026 	xfs_daddr_t	d;
1027 
1028 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
1029 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
1030 		xfs_warn(mp, "filesystem size mismatch detected");
1031 		return XFS_ERROR(EFBIG);
1032 	}
1033 	bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
1034 					d - XFS_FSS_TO_BB(mp, 1),
1035 					BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
1036 	if (!bp) {
1037 		xfs_warn(mp, "last sector read failed");
1038 		return EIO;
1039 	}
1040 	xfs_buf_relse(bp);
1041 
1042 	if (mp->m_logdev_targp != mp->m_ddev_targp) {
1043 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
1044 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
1045 			xfs_warn(mp, "log size mismatch detected");
1046 			return XFS_ERROR(EFBIG);
1047 		}
1048 		bp = xfs_buf_read_uncached(mp, mp->m_logdev_targp,
1049 					d - XFS_FSB_TO_BB(mp, 1),
1050 					XFS_FSB_TO_B(mp, 1), 0);
1051 		if (!bp) {
1052 			xfs_warn(mp, "log device read failed");
1053 			return EIO;
1054 		}
1055 		xfs_buf_relse(bp);
1056 	}
1057 	return 0;
1058 }
1059 
1060 /*
1061  * Clear the quotaflags in memory and in the superblock.
1062  */
1063 int
1064 xfs_mount_reset_sbqflags(
1065 	struct xfs_mount	*mp)
1066 {
1067 	int			error;
1068 	struct xfs_trans	*tp;
1069 
1070 	mp->m_qflags = 0;
1071 
1072 	/*
1073 	 * It is OK to look at sb_qflags here in mount path,
1074 	 * without m_sb_lock.
1075 	 */
1076 	if (mp->m_sb.sb_qflags == 0)
1077 		return 0;
1078 	spin_lock(&mp->m_sb_lock);
1079 	mp->m_sb.sb_qflags = 0;
1080 	spin_unlock(&mp->m_sb_lock);
1081 
1082 	/*
1083 	 * If the fs is readonly, let the incore superblock run
1084 	 * with quotas off but don't flush the update out to disk
1085 	 */
1086 	if (mp->m_flags & XFS_MOUNT_RDONLY)
1087 		return 0;
1088 
1089 	tp = xfs_trans_alloc(mp, XFS_TRANS_QM_SBCHANGE);
1090 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1091 				      XFS_DEFAULT_LOG_COUNT);
1092 	if (error) {
1093 		xfs_trans_cancel(tp, 0);
1094 		xfs_alert(mp, "%s: Superblock update failed!", __func__);
1095 		return error;
1096 	}
1097 
1098 	xfs_mod_sb(tp, XFS_SB_QFLAGS);
1099 	return xfs_trans_commit(tp, 0);
1100 }
1101 
1102 __uint64_t
1103 xfs_default_resblks(xfs_mount_t *mp)
1104 {
1105 	__uint64_t resblks;
1106 
1107 	/*
1108 	 * We default to 5% or 8192 fsbs of space reserved, whichever is
1109 	 * smaller.  This is intended to cover concurrent allocation
1110 	 * transactions when we initially hit enospc. These each require a 4
1111 	 * block reservation. Hence by default we cover roughly 2000 concurrent
1112 	 * allocation reservations.
1113 	 */
1114 	resblks = mp->m_sb.sb_dblocks;
1115 	do_div(resblks, 20);
1116 	resblks = min_t(__uint64_t, resblks, 8192);
1117 	return resblks;
1118 }
1119 
1120 /*
1121  * This function does the following on an initial mount of a file system:
1122  *	- reads the superblock from disk and init the mount struct
1123  *	- if we're a 32-bit kernel, do a size check on the superblock
1124  *		so we don't mount terabyte filesystems
1125  *	- init mount struct realtime fields
1126  *	- allocate inode hash table for fs
1127  *	- init directory manager
1128  *	- perform recovery and init the log manager
1129  */
1130 int
1131 xfs_mountfs(
1132 	xfs_mount_t	*mp)
1133 {
1134 	xfs_sb_t	*sbp = &(mp->m_sb);
1135 	xfs_inode_t	*rip;
1136 	__uint64_t	resblks;
1137 	uint		quotamount = 0;
1138 	uint		quotaflags = 0;
1139 	int		error = 0;
1140 
1141 	xfs_mount_common(mp, sbp);
1142 
1143 	/*
1144 	 * Check for a mismatched features2 values.  Older kernels
1145 	 * read & wrote into the wrong sb offset for sb_features2
1146 	 * on some platforms due to xfs_sb_t not being 64bit size aligned
1147 	 * when sb_features2 was added, which made older superblock
1148 	 * reading/writing routines swap it as a 64-bit value.
1149 	 *
1150 	 * For backwards compatibility, we make both slots equal.
1151 	 *
1152 	 * If we detect a mismatched field, we OR the set bits into the
1153 	 * existing features2 field in case it has already been modified; we
1154 	 * don't want to lose any features.  We then update the bad location
1155 	 * with the ORed value so that older kernels will see any features2
1156 	 * flags, and mark the two fields as needing updates once the
1157 	 * transaction subsystem is online.
1158 	 */
1159 	if (xfs_sb_has_mismatched_features2(sbp)) {
1160 		xfs_warn(mp, "correcting sb_features alignment problem");
1161 		sbp->sb_features2 |= sbp->sb_bad_features2;
1162 		sbp->sb_bad_features2 = sbp->sb_features2;
1163 		mp->m_update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
1164 
1165 		/*
1166 		 * Re-check for ATTR2 in case it was found in bad_features2
1167 		 * slot.
1168 		 */
1169 		if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1170 		   !(mp->m_flags & XFS_MOUNT_NOATTR2))
1171 			mp->m_flags |= XFS_MOUNT_ATTR2;
1172 	}
1173 
1174 	if (xfs_sb_version_hasattr2(&mp->m_sb) &&
1175 	   (mp->m_flags & XFS_MOUNT_NOATTR2)) {
1176 		xfs_sb_version_removeattr2(&mp->m_sb);
1177 		mp->m_update_flags |= XFS_SB_FEATURES2;
1178 
1179 		/* update sb_versionnum for the clearing of the morebits */
1180 		if (!sbp->sb_features2)
1181 			mp->m_update_flags |= XFS_SB_VERSIONNUM;
1182 	}
1183 
1184 	/*
1185 	 * Check if sb_agblocks is aligned at stripe boundary
1186 	 * If sb_agblocks is NOT aligned turn off m_dalign since
1187 	 * allocator alignment is within an ag, therefore ag has
1188 	 * to be aligned at stripe boundary.
1189 	 */
1190 	error = xfs_update_alignment(mp);
1191 	if (error)
1192 		goto out;
1193 
1194 	xfs_alloc_compute_maxlevels(mp);
1195 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1196 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1197 	xfs_ialloc_compute_maxlevels(mp);
1198 
1199 	xfs_set_maxicount(mp);
1200 
1201 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1202 
1203 	error = xfs_uuid_mount(mp);
1204 	if (error)
1205 		goto out;
1206 
1207 	/*
1208 	 * Set the minimum read and write sizes
1209 	 */
1210 	xfs_set_rw_sizes(mp);
1211 
1212 	/* set the low space thresholds for dynamic preallocation */
1213 	xfs_set_low_space_thresholds(mp);
1214 
1215 	/*
1216 	 * Set the inode cluster size.
1217 	 * This may still be overridden by the file system
1218 	 * block size if it is larger than the chosen cluster size.
1219 	 */
1220 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1221 
1222 	/*
1223 	 * Set inode alignment fields
1224 	 */
1225 	xfs_set_inoalignment(mp);
1226 
1227 	/*
1228 	 * Check that the data (and log if separate) are an ok size.
1229 	 */
1230 	error = xfs_check_sizes(mp);
1231 	if (error)
1232 		goto out_remove_uuid;
1233 
1234 	/*
1235 	 * Initialize realtime fields in the mount structure
1236 	 */
1237 	error = xfs_rtmount_init(mp);
1238 	if (error) {
1239 		xfs_warn(mp, "RT mount failed");
1240 		goto out_remove_uuid;
1241 	}
1242 
1243 	/*
1244 	 *  Copies the low order bits of the timestamp and the randomly
1245 	 *  set "sequence" number out of a UUID.
1246 	 */
1247 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1248 
1249 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
1250 
1251 	xfs_dir_mount(mp);
1252 
1253 	/*
1254 	 * Initialize the attribute manager's entries.
1255 	 */
1256 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1257 
1258 	/*
1259 	 * Initialize the precomputed transaction reservations values.
1260 	 */
1261 	xfs_trans_init(mp);
1262 
1263 	/*
1264 	 * Allocate and initialize the per-ag data.
1265 	 */
1266 	spin_lock_init(&mp->m_perag_lock);
1267 	INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1268 	error = xfs_initialize_perag(mp, sbp->sb_agcount, &mp->m_maxagi);
1269 	if (error) {
1270 		xfs_warn(mp, "Failed per-ag init: %d", error);
1271 		goto out_remove_uuid;
1272 	}
1273 
1274 	if (!sbp->sb_logblocks) {
1275 		xfs_warn(mp, "no log defined");
1276 		XFS_ERROR_REPORT("xfs_mountfs", XFS_ERRLEVEL_LOW, mp);
1277 		error = XFS_ERROR(EFSCORRUPTED);
1278 		goto out_free_perag;
1279 	}
1280 
1281 	/*
1282 	 * log's mount-time initialization. Perform 1st part recovery if needed
1283 	 */
1284 	error = xfs_log_mount(mp, mp->m_logdev_targp,
1285 			      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1286 			      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1287 	if (error) {
1288 		xfs_warn(mp, "log mount failed");
1289 		goto out_free_perag;
1290 	}
1291 
1292 	/*
1293 	 * Now the log is mounted, we know if it was an unclean shutdown or
1294 	 * not. If it was, with the first phase of recovery has completed, we
1295 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1296 	 * but they are recovered transactionally in the second recovery phase
1297 	 * later.
1298 	 *
1299 	 * Hence we can safely re-initialise incore superblock counters from
1300 	 * the per-ag data. These may not be correct if the filesystem was not
1301 	 * cleanly unmounted, so we need to wait for recovery to finish before
1302 	 * doing this.
1303 	 *
1304 	 * If the filesystem was cleanly unmounted, then we can trust the
1305 	 * values in the superblock to be correct and we don't need to do
1306 	 * anything here.
1307 	 *
1308 	 * If we are currently making the filesystem, the initialisation will
1309 	 * fail as the perag data is in an undefined state.
1310 	 */
1311 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1312 	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1313 	     !mp->m_sb.sb_inprogress) {
1314 		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1315 		if (error)
1316 			goto out_free_perag;
1317 	}
1318 
1319 	/*
1320 	 * Get and sanity-check the root inode.
1321 	 * Save the pointer to it in the mount structure.
1322 	 */
1323 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip);
1324 	if (error) {
1325 		xfs_warn(mp, "failed to read root inode");
1326 		goto out_log_dealloc;
1327 	}
1328 
1329 	ASSERT(rip != NULL);
1330 
1331 	if (unlikely(!S_ISDIR(rip->i_d.di_mode))) {
1332 		xfs_warn(mp, "corrupted root inode %llu: not a directory",
1333 			(unsigned long long)rip->i_ino);
1334 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
1335 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1336 				 mp);
1337 		error = XFS_ERROR(EFSCORRUPTED);
1338 		goto out_rele_rip;
1339 	}
1340 	mp->m_rootip = rip;	/* save it */
1341 
1342 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1343 
1344 	/*
1345 	 * Initialize realtime inode pointers in the mount structure
1346 	 */
1347 	error = xfs_rtmount_inodes(mp);
1348 	if (error) {
1349 		/*
1350 		 * Free up the root inode.
1351 		 */
1352 		xfs_warn(mp, "failed to read RT inodes");
1353 		goto out_rele_rip;
1354 	}
1355 
1356 	/*
1357 	 * If this is a read-only mount defer the superblock updates until
1358 	 * the next remount into writeable mode.  Otherwise we would never
1359 	 * perform the update e.g. for the root filesystem.
1360 	 */
1361 	if (mp->m_update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY)) {
1362 		error = xfs_mount_log_sb(mp, mp->m_update_flags);
1363 		if (error) {
1364 			xfs_warn(mp, "failed to write sb changes");
1365 			goto out_rtunmount;
1366 		}
1367 	}
1368 
1369 	/*
1370 	 * Initialise the XFS quota management subsystem for this mount
1371 	 */
1372 	if (XFS_IS_QUOTA_RUNNING(mp)) {
1373 		error = xfs_qm_newmount(mp, &quotamount, &quotaflags);
1374 		if (error)
1375 			goto out_rtunmount;
1376 	} else {
1377 		ASSERT(!XFS_IS_QUOTA_ON(mp));
1378 
1379 		/*
1380 		 * If a file system had quotas running earlier, but decided to
1381 		 * mount without -o uquota/pquota/gquota options, revoke the
1382 		 * quotachecked license.
1383 		 */
1384 		if (mp->m_sb.sb_qflags & XFS_ALL_QUOTA_ACCT) {
1385 			xfs_notice(mp, "resetting quota flags");
1386 			error = xfs_mount_reset_sbqflags(mp);
1387 			if (error)
1388 				return error;
1389 		}
1390 	}
1391 
1392 	/*
1393 	 * Finish recovering the file system.  This part needed to be
1394 	 * delayed until after the root and real-time bitmap inodes
1395 	 * were consistently read in.
1396 	 */
1397 	error = xfs_log_mount_finish(mp);
1398 	if (error) {
1399 		xfs_warn(mp, "log mount finish failed");
1400 		goto out_rtunmount;
1401 	}
1402 
1403 	/*
1404 	 * Complete the quota initialisation, post-log-replay component.
1405 	 */
1406 	if (quotamount) {
1407 		ASSERT(mp->m_qflags == 0);
1408 		mp->m_qflags = quotaflags;
1409 
1410 		xfs_qm_mount_quotas(mp);
1411 	}
1412 
1413 	/*
1414 	 * Now we are mounted, reserve a small amount of unused space for
1415 	 * privileged transactions. This is needed so that transaction
1416 	 * space required for critical operations can dip into this pool
1417 	 * when at ENOSPC. This is needed for operations like create with
1418 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1419 	 * are not allowed to use this reserved space.
1420 	 *
1421 	 * This may drive us straight to ENOSPC on mount, but that implies
1422 	 * we were already there on the last unmount. Warn if this occurs.
1423 	 */
1424 	if (!(mp->m_flags & XFS_MOUNT_RDONLY)) {
1425 		resblks = xfs_default_resblks(mp);
1426 		error = xfs_reserve_blocks(mp, &resblks, NULL);
1427 		if (error)
1428 			xfs_warn(mp,
1429 	"Unable to allocate reserve blocks. Continuing without reserve pool.");
1430 	}
1431 
1432 	return 0;
1433 
1434  out_rtunmount:
1435 	xfs_rtunmount_inodes(mp);
1436  out_rele_rip:
1437 	IRELE(rip);
1438  out_log_dealloc:
1439 	xfs_log_unmount(mp);
1440  out_free_perag:
1441 	xfs_free_perag(mp);
1442  out_remove_uuid:
1443 	xfs_uuid_unmount(mp);
1444  out:
1445 	return error;
1446 }
1447 
1448 /*
1449  * This flushes out the inodes,dquots and the superblock, unmounts the
1450  * log and makes sure that incore structures are freed.
1451  */
1452 void
1453 xfs_unmountfs(
1454 	struct xfs_mount	*mp)
1455 {
1456 	__uint64_t		resblks;
1457 	int			error;
1458 
1459 	xfs_qm_unmount_quotas(mp);
1460 	xfs_rtunmount_inodes(mp);
1461 	IRELE(mp->m_rootip);
1462 
1463 	/*
1464 	 * We can potentially deadlock here if we have an inode cluster
1465 	 * that has been freed has its buffer still pinned in memory because
1466 	 * the transaction is still sitting in a iclog. The stale inodes
1467 	 * on that buffer will have their flush locks held until the
1468 	 * transaction hits the disk and the callbacks run. the inode
1469 	 * flush takes the flush lock unconditionally and with nothing to
1470 	 * push out the iclog we will never get that unlocked. hence we
1471 	 * need to force the log first.
1472 	 */
1473 	xfs_log_force(mp, XFS_LOG_SYNC);
1474 
1475 	/*
1476 	 * Do a delwri reclaim pass first so that as many dirty inodes are
1477 	 * queued up for IO as possible. Then flush the buffers before making
1478 	 * a synchronous path to catch all the remaining inodes are reclaimed.
1479 	 * This makes the reclaim process as quick as possible by avoiding
1480 	 * synchronous writeout and blocking on inodes already in the delwri
1481 	 * state as much as possible.
1482 	 */
1483 	xfs_reclaim_inodes(mp, 0);
1484 	xfs_flush_buftarg(mp->m_ddev_targp, 1);
1485 	xfs_reclaim_inodes(mp, SYNC_WAIT);
1486 
1487 	xfs_qm_unmount(mp);
1488 
1489 	/*
1490 	 * Flush out the log synchronously so that we know for sure
1491 	 * that nothing is pinned.  This is important because bflush()
1492 	 * will skip pinned buffers.
1493 	 */
1494 	xfs_log_force(mp, XFS_LOG_SYNC);
1495 
1496 	/*
1497 	 * Unreserve any blocks we have so that when we unmount we don't account
1498 	 * the reserved free space as used. This is really only necessary for
1499 	 * lazy superblock counting because it trusts the incore superblock
1500 	 * counters to be absolutely correct on clean unmount.
1501 	 *
1502 	 * We don't bother correcting this elsewhere for lazy superblock
1503 	 * counting because on mount of an unclean filesystem we reconstruct the
1504 	 * correct counter value and this is irrelevant.
1505 	 *
1506 	 * For non-lazy counter filesystems, this doesn't matter at all because
1507 	 * we only every apply deltas to the superblock and hence the incore
1508 	 * value does not matter....
1509 	 */
1510 	resblks = 0;
1511 	error = xfs_reserve_blocks(mp, &resblks, NULL);
1512 	if (error)
1513 		xfs_warn(mp, "Unable to free reserved block pool. "
1514 				"Freespace may not be correct on next mount.");
1515 
1516 	error = xfs_log_sbcount(mp);
1517 	if (error)
1518 		xfs_warn(mp, "Unable to update superblock counters. "
1519 				"Freespace may not be correct on next mount.");
1520 	xfs_unmountfs_writesb(mp);
1521 
1522 	/*
1523 	 * Make sure all buffers have been flushed and completed before
1524 	 * unmounting the log.
1525 	 */
1526 	error = xfs_flush_buftarg(mp->m_ddev_targp, 1);
1527 	if (error)
1528 		xfs_warn(mp, "%d busy buffers during unmount.", error);
1529 	xfs_wait_buftarg(mp->m_ddev_targp);
1530 
1531 	xfs_log_unmount_write(mp);
1532 	xfs_log_unmount(mp);
1533 	xfs_uuid_unmount(mp);
1534 
1535 #if defined(DEBUG)
1536 	xfs_errortag_clearall(mp, 0);
1537 #endif
1538 	xfs_free_perag(mp);
1539 }
1540 
1541 int
1542 xfs_fs_writable(xfs_mount_t *mp)
1543 {
1544 	return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1545 		(mp->m_flags & XFS_MOUNT_RDONLY));
1546 }
1547 
1548 /*
1549  * xfs_log_sbcount
1550  *
1551  * Sync the superblock counters to disk.
1552  *
1553  * Note this code can be called during the process of freezing, so
1554  * we may need to use the transaction allocator which does not
1555  * block when the transaction subsystem is in its frozen state.
1556  */
1557 int
1558 xfs_log_sbcount(xfs_mount_t *mp)
1559 {
1560 	xfs_trans_t	*tp;
1561 	int		error;
1562 
1563 	if (!xfs_fs_writable(mp))
1564 		return 0;
1565 
1566 	xfs_icsb_sync_counters(mp, 0);
1567 
1568 	/*
1569 	 * we don't need to do this if we are updating the superblock
1570 	 * counters on every modification.
1571 	 */
1572 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1573 		return 0;
1574 
1575 	tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT, KM_SLEEP);
1576 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1577 					XFS_DEFAULT_LOG_COUNT);
1578 	if (error) {
1579 		xfs_trans_cancel(tp, 0);
1580 		return error;
1581 	}
1582 
1583 	xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1584 	xfs_trans_set_sync(tp);
1585 	error = xfs_trans_commit(tp, 0);
1586 	return error;
1587 }
1588 
1589 int
1590 xfs_unmountfs_writesb(xfs_mount_t *mp)
1591 {
1592 	xfs_buf_t	*sbp;
1593 	int		error = 0;
1594 
1595 	/*
1596 	 * skip superblock write if fs is read-only, or
1597 	 * if we are doing a forced umount.
1598 	 */
1599 	if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1600 		XFS_FORCED_SHUTDOWN(mp))) {
1601 
1602 		sbp = xfs_getsb(mp, 0);
1603 
1604 		XFS_BUF_UNDONE(sbp);
1605 		XFS_BUF_UNREAD(sbp);
1606 		xfs_buf_delwri_dequeue(sbp);
1607 		XFS_BUF_WRITE(sbp);
1608 		XFS_BUF_UNASYNC(sbp);
1609 		ASSERT(sbp->b_target == mp->m_ddev_targp);
1610 		xfsbdstrat(mp, sbp);
1611 		error = xfs_buf_iowait(sbp);
1612 		if (error)
1613 			xfs_buf_ioerror_alert(sbp, __func__);
1614 		xfs_buf_relse(sbp);
1615 	}
1616 	return error;
1617 }
1618 
1619 /*
1620  * xfs_mod_sb() can be used to copy arbitrary changes to the
1621  * in-core superblock into the superblock buffer to be logged.
1622  * It does not provide the higher level of locking that is
1623  * needed to protect the in-core superblock from concurrent
1624  * access.
1625  */
1626 void
1627 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1628 {
1629 	xfs_buf_t	*bp;
1630 	int		first;
1631 	int		last;
1632 	xfs_mount_t	*mp;
1633 	xfs_sb_field_t	f;
1634 
1635 	ASSERT(fields);
1636 	if (!fields)
1637 		return;
1638 	mp = tp->t_mountp;
1639 	bp = xfs_trans_getsb(tp, mp, 0);
1640 	first = sizeof(xfs_sb_t);
1641 	last = 0;
1642 
1643 	/* translate/copy */
1644 
1645 	xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1646 
1647 	/* find modified range */
1648 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1649 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1650 	last = xfs_sb_info[f + 1].offset - 1;
1651 
1652 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1653 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1654 	first = xfs_sb_info[f].offset;
1655 
1656 	xfs_trans_log_buf(tp, bp, first, last);
1657 }
1658 
1659 
1660 /*
1661  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1662  * a delta to a specified field in the in-core superblock.  Simply
1663  * switch on the field indicated and apply the delta to that field.
1664  * Fields are not allowed to dip below zero, so if the delta would
1665  * do this do not apply it and return EINVAL.
1666  *
1667  * The m_sb_lock must be held when this routine is called.
1668  */
1669 STATIC int
1670 xfs_mod_incore_sb_unlocked(
1671 	xfs_mount_t	*mp,
1672 	xfs_sb_field_t	field,
1673 	int64_t		delta,
1674 	int		rsvd)
1675 {
1676 	int		scounter;	/* short counter for 32 bit fields */
1677 	long long	lcounter;	/* long counter for 64 bit fields */
1678 	long long	res_used, rem;
1679 
1680 	/*
1681 	 * With the in-core superblock spin lock held, switch
1682 	 * on the indicated field.  Apply the delta to the
1683 	 * proper field.  If the fields value would dip below
1684 	 * 0, then do not apply the delta and return EINVAL.
1685 	 */
1686 	switch (field) {
1687 	case XFS_SBS_ICOUNT:
1688 		lcounter = (long long)mp->m_sb.sb_icount;
1689 		lcounter += delta;
1690 		if (lcounter < 0) {
1691 			ASSERT(0);
1692 			return XFS_ERROR(EINVAL);
1693 		}
1694 		mp->m_sb.sb_icount = lcounter;
1695 		return 0;
1696 	case XFS_SBS_IFREE:
1697 		lcounter = (long long)mp->m_sb.sb_ifree;
1698 		lcounter += delta;
1699 		if (lcounter < 0) {
1700 			ASSERT(0);
1701 			return XFS_ERROR(EINVAL);
1702 		}
1703 		mp->m_sb.sb_ifree = lcounter;
1704 		return 0;
1705 	case XFS_SBS_FDBLOCKS:
1706 		lcounter = (long long)
1707 			mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1708 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1709 
1710 		if (delta > 0) {		/* Putting blocks back */
1711 			if (res_used > delta) {
1712 				mp->m_resblks_avail += delta;
1713 			} else {
1714 				rem = delta - res_used;
1715 				mp->m_resblks_avail = mp->m_resblks;
1716 				lcounter += rem;
1717 			}
1718 		} else {				/* Taking blocks away */
1719 			lcounter += delta;
1720 			if (lcounter >= 0) {
1721 				mp->m_sb.sb_fdblocks = lcounter +
1722 							XFS_ALLOC_SET_ASIDE(mp);
1723 				return 0;
1724 			}
1725 
1726 			/*
1727 			 * We are out of blocks, use any available reserved
1728 			 * blocks if were allowed to.
1729 			 */
1730 			if (!rsvd)
1731 				return XFS_ERROR(ENOSPC);
1732 
1733 			lcounter = (long long)mp->m_resblks_avail + delta;
1734 			if (lcounter >= 0) {
1735 				mp->m_resblks_avail = lcounter;
1736 				return 0;
1737 			}
1738 			printk_once(KERN_WARNING
1739 				"Filesystem \"%s\": reserve blocks depleted! "
1740 				"Consider increasing reserve pool size.",
1741 				mp->m_fsname);
1742 			return XFS_ERROR(ENOSPC);
1743 		}
1744 
1745 		mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1746 		return 0;
1747 	case XFS_SBS_FREXTENTS:
1748 		lcounter = (long long)mp->m_sb.sb_frextents;
1749 		lcounter += delta;
1750 		if (lcounter < 0) {
1751 			return XFS_ERROR(ENOSPC);
1752 		}
1753 		mp->m_sb.sb_frextents = lcounter;
1754 		return 0;
1755 	case XFS_SBS_DBLOCKS:
1756 		lcounter = (long long)mp->m_sb.sb_dblocks;
1757 		lcounter += delta;
1758 		if (lcounter < 0) {
1759 			ASSERT(0);
1760 			return XFS_ERROR(EINVAL);
1761 		}
1762 		mp->m_sb.sb_dblocks = lcounter;
1763 		return 0;
1764 	case XFS_SBS_AGCOUNT:
1765 		scounter = mp->m_sb.sb_agcount;
1766 		scounter += delta;
1767 		if (scounter < 0) {
1768 			ASSERT(0);
1769 			return XFS_ERROR(EINVAL);
1770 		}
1771 		mp->m_sb.sb_agcount = scounter;
1772 		return 0;
1773 	case XFS_SBS_IMAX_PCT:
1774 		scounter = mp->m_sb.sb_imax_pct;
1775 		scounter += delta;
1776 		if (scounter < 0) {
1777 			ASSERT(0);
1778 			return XFS_ERROR(EINVAL);
1779 		}
1780 		mp->m_sb.sb_imax_pct = scounter;
1781 		return 0;
1782 	case XFS_SBS_REXTSIZE:
1783 		scounter = mp->m_sb.sb_rextsize;
1784 		scounter += delta;
1785 		if (scounter < 0) {
1786 			ASSERT(0);
1787 			return XFS_ERROR(EINVAL);
1788 		}
1789 		mp->m_sb.sb_rextsize = scounter;
1790 		return 0;
1791 	case XFS_SBS_RBMBLOCKS:
1792 		scounter = mp->m_sb.sb_rbmblocks;
1793 		scounter += delta;
1794 		if (scounter < 0) {
1795 			ASSERT(0);
1796 			return XFS_ERROR(EINVAL);
1797 		}
1798 		mp->m_sb.sb_rbmblocks = scounter;
1799 		return 0;
1800 	case XFS_SBS_RBLOCKS:
1801 		lcounter = (long long)mp->m_sb.sb_rblocks;
1802 		lcounter += delta;
1803 		if (lcounter < 0) {
1804 			ASSERT(0);
1805 			return XFS_ERROR(EINVAL);
1806 		}
1807 		mp->m_sb.sb_rblocks = lcounter;
1808 		return 0;
1809 	case XFS_SBS_REXTENTS:
1810 		lcounter = (long long)mp->m_sb.sb_rextents;
1811 		lcounter += delta;
1812 		if (lcounter < 0) {
1813 			ASSERT(0);
1814 			return XFS_ERROR(EINVAL);
1815 		}
1816 		mp->m_sb.sb_rextents = lcounter;
1817 		return 0;
1818 	case XFS_SBS_REXTSLOG:
1819 		scounter = mp->m_sb.sb_rextslog;
1820 		scounter += delta;
1821 		if (scounter < 0) {
1822 			ASSERT(0);
1823 			return XFS_ERROR(EINVAL);
1824 		}
1825 		mp->m_sb.sb_rextslog = scounter;
1826 		return 0;
1827 	default:
1828 		ASSERT(0);
1829 		return XFS_ERROR(EINVAL);
1830 	}
1831 }
1832 
1833 /*
1834  * xfs_mod_incore_sb() is used to change a field in the in-core
1835  * superblock structure by the specified delta.  This modification
1836  * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1837  * routine to do the work.
1838  */
1839 int
1840 xfs_mod_incore_sb(
1841 	struct xfs_mount	*mp,
1842 	xfs_sb_field_t		field,
1843 	int64_t			delta,
1844 	int			rsvd)
1845 {
1846 	int			status;
1847 
1848 #ifdef HAVE_PERCPU_SB
1849 	ASSERT(field < XFS_SBS_ICOUNT || field > XFS_SBS_FDBLOCKS);
1850 #endif
1851 	spin_lock(&mp->m_sb_lock);
1852 	status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1853 	spin_unlock(&mp->m_sb_lock);
1854 
1855 	return status;
1856 }
1857 
1858 /*
1859  * Change more than one field in the in-core superblock structure at a time.
1860  *
1861  * The fields and changes to those fields are specified in the array of
1862  * xfs_mod_sb structures passed in.  Either all of the specified deltas
1863  * will be applied or none of them will.  If any modified field dips below 0,
1864  * then all modifications will be backed out and EINVAL will be returned.
1865  *
1866  * Note that this function may not be used for the superblock values that
1867  * are tracked with the in-memory per-cpu counters - a direct call to
1868  * xfs_icsb_modify_counters is required for these.
1869  */
1870 int
1871 xfs_mod_incore_sb_batch(
1872 	struct xfs_mount	*mp,
1873 	xfs_mod_sb_t		*msb,
1874 	uint			nmsb,
1875 	int			rsvd)
1876 {
1877 	xfs_mod_sb_t		*msbp;
1878 	int			error = 0;
1879 
1880 	/*
1881 	 * Loop through the array of mod structures and apply each individually.
1882 	 * If any fail, then back out all those which have already been applied.
1883 	 * Do all of this within the scope of the m_sb_lock so that all of the
1884 	 * changes will be atomic.
1885 	 */
1886 	spin_lock(&mp->m_sb_lock);
1887 	for (msbp = msb; msbp < (msb + nmsb); msbp++) {
1888 		ASSERT(msbp->msb_field < XFS_SBS_ICOUNT ||
1889 		       msbp->msb_field > XFS_SBS_FDBLOCKS);
1890 
1891 		error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1892 						   msbp->msb_delta, rsvd);
1893 		if (error)
1894 			goto unwind;
1895 	}
1896 	spin_unlock(&mp->m_sb_lock);
1897 	return 0;
1898 
1899 unwind:
1900 	while (--msbp >= msb) {
1901 		error = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1902 						   -msbp->msb_delta, rsvd);
1903 		ASSERT(error == 0);
1904 	}
1905 	spin_unlock(&mp->m_sb_lock);
1906 	return error;
1907 }
1908 
1909 /*
1910  * xfs_getsb() is called to obtain the buffer for the superblock.
1911  * The buffer is returned locked and read in from disk.
1912  * The buffer should be released with a call to xfs_brelse().
1913  *
1914  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1915  * the superblock buffer if it can be locked without sleeping.
1916  * If it can't then we'll return NULL.
1917  */
1918 struct xfs_buf *
1919 xfs_getsb(
1920 	struct xfs_mount	*mp,
1921 	int			flags)
1922 {
1923 	struct xfs_buf		*bp = mp->m_sb_bp;
1924 
1925 	if (!xfs_buf_trylock(bp)) {
1926 		if (flags & XBF_TRYLOCK)
1927 			return NULL;
1928 		xfs_buf_lock(bp);
1929 	}
1930 
1931 	xfs_buf_hold(bp);
1932 	ASSERT(XFS_BUF_ISDONE(bp));
1933 	return bp;
1934 }
1935 
1936 /*
1937  * Used to free the superblock along various error paths.
1938  */
1939 void
1940 xfs_freesb(
1941 	struct xfs_mount	*mp)
1942 {
1943 	struct xfs_buf		*bp = mp->m_sb_bp;
1944 
1945 	xfs_buf_lock(bp);
1946 	mp->m_sb_bp = NULL;
1947 	xfs_buf_relse(bp);
1948 }
1949 
1950 /*
1951  * Used to log changes to the superblock unit and width fields which could
1952  * be altered by the mount options, as well as any potential sb_features2
1953  * fixup. Only the first superblock is updated.
1954  */
1955 int
1956 xfs_mount_log_sb(
1957 	xfs_mount_t	*mp,
1958 	__int64_t	fields)
1959 {
1960 	xfs_trans_t	*tp;
1961 	int		error;
1962 
1963 	ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1964 			 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2 |
1965 			 XFS_SB_VERSIONNUM));
1966 
1967 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1968 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1969 				XFS_DEFAULT_LOG_COUNT);
1970 	if (error) {
1971 		xfs_trans_cancel(tp, 0);
1972 		return error;
1973 	}
1974 	xfs_mod_sb(tp, fields);
1975 	error = xfs_trans_commit(tp, 0);
1976 	return error;
1977 }
1978 
1979 /*
1980  * If the underlying (data/log/rt) device is readonly, there are some
1981  * operations that cannot proceed.
1982  */
1983 int
1984 xfs_dev_is_read_only(
1985 	struct xfs_mount	*mp,
1986 	char			*message)
1987 {
1988 	if (xfs_readonly_buftarg(mp->m_ddev_targp) ||
1989 	    xfs_readonly_buftarg(mp->m_logdev_targp) ||
1990 	    (mp->m_rtdev_targp && xfs_readonly_buftarg(mp->m_rtdev_targp))) {
1991 		xfs_notice(mp, "%s required on read-only device.", message);
1992 		xfs_notice(mp, "write access unavailable, cannot proceed.");
1993 		return EROFS;
1994 	}
1995 	return 0;
1996 }
1997 
1998 #ifdef HAVE_PERCPU_SB
1999 /*
2000  * Per-cpu incore superblock counters
2001  *
2002  * Simple concept, difficult implementation
2003  *
2004  * Basically, replace the incore superblock counters with a distributed per cpu
2005  * counter for contended fields (e.g.  free block count).
2006  *
2007  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
2008  * hence needs to be accurately read when we are running low on space. Hence
2009  * there is a method to enable and disable the per-cpu counters based on how
2010  * much "stuff" is available in them.
2011  *
2012  * Basically, a counter is enabled if there is enough free resource to justify
2013  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
2014  * ENOSPC), then we disable the counters to synchronise all callers and
2015  * re-distribute the available resources.
2016  *
2017  * If, once we redistributed the available resources, we still get a failure,
2018  * we disable the per-cpu counter and go through the slow path.
2019  *
2020  * The slow path is the current xfs_mod_incore_sb() function.  This means that
2021  * when we disable a per-cpu counter, we need to drain its resources back to
2022  * the global superblock. We do this after disabling the counter to prevent
2023  * more threads from queueing up on the counter.
2024  *
2025  * Essentially, this means that we still need a lock in the fast path to enable
2026  * synchronisation between the global counters and the per-cpu counters. This
2027  * is not a problem because the lock will be local to a CPU almost all the time
2028  * and have little contention except when we get to ENOSPC conditions.
2029  *
2030  * Basically, this lock becomes a barrier that enables us to lock out the fast
2031  * path while we do things like enabling and disabling counters and
2032  * synchronising the counters.
2033  *
2034  * Locking rules:
2035  *
2036  * 	1. m_sb_lock before picking up per-cpu locks
2037  * 	2. per-cpu locks always picked up via for_each_online_cpu() order
2038  * 	3. accurate counter sync requires m_sb_lock + per cpu locks
2039  * 	4. modifying per-cpu counters requires holding per-cpu lock
2040  * 	5. modifying global counters requires holding m_sb_lock
2041  *	6. enabling or disabling a counter requires holding the m_sb_lock
2042  *	   and _none_ of the per-cpu locks.
2043  *
2044  * Disabled counters are only ever re-enabled by a balance operation
2045  * that results in more free resources per CPU than a given threshold.
2046  * To ensure counters don't remain disabled, they are rebalanced when
2047  * the global resource goes above a higher threshold (i.e. some hysteresis
2048  * is present to prevent thrashing).
2049  */
2050 
2051 #ifdef CONFIG_HOTPLUG_CPU
2052 /*
2053  * hot-plug CPU notifier support.
2054  *
2055  * We need a notifier per filesystem as we need to be able to identify
2056  * the filesystem to balance the counters out. This is achieved by
2057  * having a notifier block embedded in the xfs_mount_t and doing pointer
2058  * magic to get the mount pointer from the notifier block address.
2059  */
2060 STATIC int
2061 xfs_icsb_cpu_notify(
2062 	struct notifier_block *nfb,
2063 	unsigned long action,
2064 	void *hcpu)
2065 {
2066 	xfs_icsb_cnts_t *cntp;
2067 	xfs_mount_t	*mp;
2068 
2069 	mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2070 	cntp = (xfs_icsb_cnts_t *)
2071 			per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2072 	switch (action) {
2073 	case CPU_UP_PREPARE:
2074 	case CPU_UP_PREPARE_FROZEN:
2075 		/* Easy Case - initialize the area and locks, and
2076 		 * then rebalance when online does everything else for us. */
2077 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2078 		break;
2079 	case CPU_ONLINE:
2080 	case CPU_ONLINE_FROZEN:
2081 		xfs_icsb_lock(mp);
2082 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2083 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2084 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2085 		xfs_icsb_unlock(mp);
2086 		break;
2087 	case CPU_DEAD:
2088 	case CPU_DEAD_FROZEN:
2089 		/* Disable all the counters, then fold the dead cpu's
2090 		 * count into the total on the global superblock and
2091 		 * re-enable the counters. */
2092 		xfs_icsb_lock(mp);
2093 		spin_lock(&mp->m_sb_lock);
2094 		xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2095 		xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2096 		xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2097 
2098 		mp->m_sb.sb_icount += cntp->icsb_icount;
2099 		mp->m_sb.sb_ifree += cntp->icsb_ifree;
2100 		mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2101 
2102 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2103 
2104 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_ICOUNT, 0);
2105 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_IFREE, 0);
2106 		xfs_icsb_balance_counter_locked(mp, XFS_SBS_FDBLOCKS, 0);
2107 		spin_unlock(&mp->m_sb_lock);
2108 		xfs_icsb_unlock(mp);
2109 		break;
2110 	}
2111 
2112 	return NOTIFY_OK;
2113 }
2114 #endif /* CONFIG_HOTPLUG_CPU */
2115 
2116 int
2117 xfs_icsb_init_counters(
2118 	xfs_mount_t	*mp)
2119 {
2120 	xfs_icsb_cnts_t *cntp;
2121 	int		i;
2122 
2123 	mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2124 	if (mp->m_sb_cnts == NULL)
2125 		return -ENOMEM;
2126 
2127 #ifdef CONFIG_HOTPLUG_CPU
2128 	mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2129 	mp->m_icsb_notifier.priority = 0;
2130 	register_hotcpu_notifier(&mp->m_icsb_notifier);
2131 #endif /* CONFIG_HOTPLUG_CPU */
2132 
2133 	for_each_online_cpu(i) {
2134 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2135 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2136 	}
2137 
2138 	mutex_init(&mp->m_icsb_mutex);
2139 
2140 	/*
2141 	 * start with all counters disabled so that the
2142 	 * initial balance kicks us off correctly
2143 	 */
2144 	mp->m_icsb_counters = -1;
2145 	return 0;
2146 }
2147 
2148 void
2149 xfs_icsb_reinit_counters(
2150 	xfs_mount_t	*mp)
2151 {
2152 	xfs_icsb_lock(mp);
2153 	/*
2154 	 * start with all counters disabled so that the
2155 	 * initial balance kicks us off correctly
2156 	 */
2157 	mp->m_icsb_counters = -1;
2158 	xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0);
2159 	xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0);
2160 	xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0);
2161 	xfs_icsb_unlock(mp);
2162 }
2163 
2164 void
2165 xfs_icsb_destroy_counters(
2166 	xfs_mount_t	*mp)
2167 {
2168 	if (mp->m_sb_cnts) {
2169 		unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2170 		free_percpu(mp->m_sb_cnts);
2171 	}
2172 	mutex_destroy(&mp->m_icsb_mutex);
2173 }
2174 
2175 STATIC void
2176 xfs_icsb_lock_cntr(
2177 	xfs_icsb_cnts_t	*icsbp)
2178 {
2179 	while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2180 		ndelay(1000);
2181 	}
2182 }
2183 
2184 STATIC void
2185 xfs_icsb_unlock_cntr(
2186 	xfs_icsb_cnts_t	*icsbp)
2187 {
2188 	clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2189 }
2190 
2191 
2192 STATIC void
2193 xfs_icsb_lock_all_counters(
2194 	xfs_mount_t	*mp)
2195 {
2196 	xfs_icsb_cnts_t *cntp;
2197 	int		i;
2198 
2199 	for_each_online_cpu(i) {
2200 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2201 		xfs_icsb_lock_cntr(cntp);
2202 	}
2203 }
2204 
2205 STATIC void
2206 xfs_icsb_unlock_all_counters(
2207 	xfs_mount_t	*mp)
2208 {
2209 	xfs_icsb_cnts_t *cntp;
2210 	int		i;
2211 
2212 	for_each_online_cpu(i) {
2213 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2214 		xfs_icsb_unlock_cntr(cntp);
2215 	}
2216 }
2217 
2218 STATIC void
2219 xfs_icsb_count(
2220 	xfs_mount_t	*mp,
2221 	xfs_icsb_cnts_t	*cnt,
2222 	int		flags)
2223 {
2224 	xfs_icsb_cnts_t *cntp;
2225 	int		i;
2226 
2227 	memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2228 
2229 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2230 		xfs_icsb_lock_all_counters(mp);
2231 
2232 	for_each_online_cpu(i) {
2233 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2234 		cnt->icsb_icount += cntp->icsb_icount;
2235 		cnt->icsb_ifree += cntp->icsb_ifree;
2236 		cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2237 	}
2238 
2239 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2240 		xfs_icsb_unlock_all_counters(mp);
2241 }
2242 
2243 STATIC int
2244 xfs_icsb_counter_disabled(
2245 	xfs_mount_t	*mp,
2246 	xfs_sb_field_t	field)
2247 {
2248 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2249 	return test_bit(field, &mp->m_icsb_counters);
2250 }
2251 
2252 STATIC void
2253 xfs_icsb_disable_counter(
2254 	xfs_mount_t	*mp,
2255 	xfs_sb_field_t	field)
2256 {
2257 	xfs_icsb_cnts_t	cnt;
2258 
2259 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2260 
2261 	/*
2262 	 * If we are already disabled, then there is nothing to do
2263 	 * here. We check before locking all the counters to avoid
2264 	 * the expensive lock operation when being called in the
2265 	 * slow path and the counter is already disabled. This is
2266 	 * safe because the only time we set or clear this state is under
2267 	 * the m_icsb_mutex.
2268 	 */
2269 	if (xfs_icsb_counter_disabled(mp, field))
2270 		return;
2271 
2272 	xfs_icsb_lock_all_counters(mp);
2273 	if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2274 		/* drain back to superblock */
2275 
2276 		xfs_icsb_count(mp, &cnt, XFS_ICSB_LAZY_COUNT);
2277 		switch(field) {
2278 		case XFS_SBS_ICOUNT:
2279 			mp->m_sb.sb_icount = cnt.icsb_icount;
2280 			break;
2281 		case XFS_SBS_IFREE:
2282 			mp->m_sb.sb_ifree = cnt.icsb_ifree;
2283 			break;
2284 		case XFS_SBS_FDBLOCKS:
2285 			mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2286 			break;
2287 		default:
2288 			BUG();
2289 		}
2290 	}
2291 
2292 	xfs_icsb_unlock_all_counters(mp);
2293 }
2294 
2295 STATIC void
2296 xfs_icsb_enable_counter(
2297 	xfs_mount_t	*mp,
2298 	xfs_sb_field_t	field,
2299 	uint64_t	count,
2300 	uint64_t	resid)
2301 {
2302 	xfs_icsb_cnts_t	*cntp;
2303 	int		i;
2304 
2305 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2306 
2307 	xfs_icsb_lock_all_counters(mp);
2308 	for_each_online_cpu(i) {
2309 		cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2310 		switch (field) {
2311 		case XFS_SBS_ICOUNT:
2312 			cntp->icsb_icount = count + resid;
2313 			break;
2314 		case XFS_SBS_IFREE:
2315 			cntp->icsb_ifree = count + resid;
2316 			break;
2317 		case XFS_SBS_FDBLOCKS:
2318 			cntp->icsb_fdblocks = count + resid;
2319 			break;
2320 		default:
2321 			BUG();
2322 			break;
2323 		}
2324 		resid = 0;
2325 	}
2326 	clear_bit(field, &mp->m_icsb_counters);
2327 	xfs_icsb_unlock_all_counters(mp);
2328 }
2329 
2330 void
2331 xfs_icsb_sync_counters_locked(
2332 	xfs_mount_t	*mp,
2333 	int		flags)
2334 {
2335 	xfs_icsb_cnts_t	cnt;
2336 
2337 	xfs_icsb_count(mp, &cnt, flags);
2338 
2339 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2340 		mp->m_sb.sb_icount = cnt.icsb_icount;
2341 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2342 		mp->m_sb.sb_ifree = cnt.icsb_ifree;
2343 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2344 		mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2345 }
2346 
2347 /*
2348  * Accurate update of per-cpu counters to incore superblock
2349  */
2350 void
2351 xfs_icsb_sync_counters(
2352 	xfs_mount_t	*mp,
2353 	int		flags)
2354 {
2355 	spin_lock(&mp->m_sb_lock);
2356 	xfs_icsb_sync_counters_locked(mp, flags);
2357 	spin_unlock(&mp->m_sb_lock);
2358 }
2359 
2360 /*
2361  * Balance and enable/disable counters as necessary.
2362  *
2363  * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2364  * chosen to be the same number as single on disk allocation chunk per CPU, and
2365  * free blocks is something far enough zero that we aren't going thrash when we
2366  * get near ENOSPC. We also need to supply a minimum we require per cpu to
2367  * prevent looping endlessly when xfs_alloc_space asks for more than will
2368  * be distributed to a single CPU but each CPU has enough blocks to be
2369  * reenabled.
2370  *
2371  * Note that we can be called when counters are already disabled.
2372  * xfs_icsb_disable_counter() optimises the counter locking in this case to
2373  * prevent locking every per-cpu counter needlessly.
2374  */
2375 
2376 #define XFS_ICSB_INO_CNTR_REENABLE	(uint64_t)64
2377 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2378 		(uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2379 STATIC void
2380 xfs_icsb_balance_counter_locked(
2381 	xfs_mount_t	*mp,
2382 	xfs_sb_field_t  field,
2383 	int		min_per_cpu)
2384 {
2385 	uint64_t	count, resid;
2386 	int		weight = num_online_cpus();
2387 	uint64_t	min = (uint64_t)min_per_cpu;
2388 
2389 	/* disable counter and sync counter */
2390 	xfs_icsb_disable_counter(mp, field);
2391 
2392 	/* update counters  - first CPU gets residual*/
2393 	switch (field) {
2394 	case XFS_SBS_ICOUNT:
2395 		count = mp->m_sb.sb_icount;
2396 		resid = do_div(count, weight);
2397 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2398 			return;
2399 		break;
2400 	case XFS_SBS_IFREE:
2401 		count = mp->m_sb.sb_ifree;
2402 		resid = do_div(count, weight);
2403 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2404 			return;
2405 		break;
2406 	case XFS_SBS_FDBLOCKS:
2407 		count = mp->m_sb.sb_fdblocks;
2408 		resid = do_div(count, weight);
2409 		if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2410 			return;
2411 		break;
2412 	default:
2413 		BUG();
2414 		count = resid = 0;	/* quiet, gcc */
2415 		break;
2416 	}
2417 
2418 	xfs_icsb_enable_counter(mp, field, count, resid);
2419 }
2420 
2421 STATIC void
2422 xfs_icsb_balance_counter(
2423 	xfs_mount_t	*mp,
2424 	xfs_sb_field_t  fields,
2425 	int		min_per_cpu)
2426 {
2427 	spin_lock(&mp->m_sb_lock);
2428 	xfs_icsb_balance_counter_locked(mp, fields, min_per_cpu);
2429 	spin_unlock(&mp->m_sb_lock);
2430 }
2431 
2432 int
2433 xfs_icsb_modify_counters(
2434 	xfs_mount_t	*mp,
2435 	xfs_sb_field_t	field,
2436 	int64_t		delta,
2437 	int		rsvd)
2438 {
2439 	xfs_icsb_cnts_t	*icsbp;
2440 	long long	lcounter;	/* long counter for 64 bit fields */
2441 	int		ret = 0;
2442 
2443 	might_sleep();
2444 again:
2445 	preempt_disable();
2446 	icsbp = this_cpu_ptr(mp->m_sb_cnts);
2447 
2448 	/*
2449 	 * if the counter is disabled, go to slow path
2450 	 */
2451 	if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2452 		goto slow_path;
2453 	xfs_icsb_lock_cntr(icsbp);
2454 	if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2455 		xfs_icsb_unlock_cntr(icsbp);
2456 		goto slow_path;
2457 	}
2458 
2459 	switch (field) {
2460 	case XFS_SBS_ICOUNT:
2461 		lcounter = icsbp->icsb_icount;
2462 		lcounter += delta;
2463 		if (unlikely(lcounter < 0))
2464 			goto balance_counter;
2465 		icsbp->icsb_icount = lcounter;
2466 		break;
2467 
2468 	case XFS_SBS_IFREE:
2469 		lcounter = icsbp->icsb_ifree;
2470 		lcounter += delta;
2471 		if (unlikely(lcounter < 0))
2472 			goto balance_counter;
2473 		icsbp->icsb_ifree = lcounter;
2474 		break;
2475 
2476 	case XFS_SBS_FDBLOCKS:
2477 		BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2478 
2479 		lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2480 		lcounter += delta;
2481 		if (unlikely(lcounter < 0))
2482 			goto balance_counter;
2483 		icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2484 		break;
2485 	default:
2486 		BUG();
2487 		break;
2488 	}
2489 	xfs_icsb_unlock_cntr(icsbp);
2490 	preempt_enable();
2491 	return 0;
2492 
2493 slow_path:
2494 	preempt_enable();
2495 
2496 	/*
2497 	 * serialise with a mutex so we don't burn lots of cpu on
2498 	 * the superblock lock. We still need to hold the superblock
2499 	 * lock, however, when we modify the global structures.
2500 	 */
2501 	xfs_icsb_lock(mp);
2502 
2503 	/*
2504 	 * Now running atomically.
2505 	 *
2506 	 * If the counter is enabled, someone has beaten us to rebalancing.
2507 	 * Drop the lock and try again in the fast path....
2508 	 */
2509 	if (!(xfs_icsb_counter_disabled(mp, field))) {
2510 		xfs_icsb_unlock(mp);
2511 		goto again;
2512 	}
2513 
2514 	/*
2515 	 * The counter is currently disabled. Because we are
2516 	 * running atomically here, we know a rebalance cannot
2517 	 * be in progress. Hence we can go straight to operating
2518 	 * on the global superblock. We do not call xfs_mod_incore_sb()
2519 	 * here even though we need to get the m_sb_lock. Doing so
2520 	 * will cause us to re-enter this function and deadlock.
2521 	 * Hence we get the m_sb_lock ourselves and then call
2522 	 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2523 	 * directly on the global counters.
2524 	 */
2525 	spin_lock(&mp->m_sb_lock);
2526 	ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2527 	spin_unlock(&mp->m_sb_lock);
2528 
2529 	/*
2530 	 * Now that we've modified the global superblock, we
2531 	 * may be able to re-enable the distributed counters
2532 	 * (e.g. lots of space just got freed). After that
2533 	 * we are done.
2534 	 */
2535 	if (ret != ENOSPC)
2536 		xfs_icsb_balance_counter(mp, field, 0);
2537 	xfs_icsb_unlock(mp);
2538 	return ret;
2539 
2540 balance_counter:
2541 	xfs_icsb_unlock_cntr(icsbp);
2542 	preempt_enable();
2543 
2544 	/*
2545 	 * We may have multiple threads here if multiple per-cpu
2546 	 * counters run dry at the same time. This will mean we can
2547 	 * do more balances than strictly necessary but it is not
2548 	 * the common slowpath case.
2549 	 */
2550 	xfs_icsb_lock(mp);
2551 
2552 	/*
2553 	 * running atomically.
2554 	 *
2555 	 * This will leave the counter in the correct state for future
2556 	 * accesses. After the rebalance, we simply try again and our retry
2557 	 * will either succeed through the fast path or slow path without
2558 	 * another balance operation being required.
2559 	 */
2560 	xfs_icsb_balance_counter(mp, field, delta);
2561 	xfs_icsb_unlock(mp);
2562 	goto again;
2563 }
2564 
2565 #endif
2566