xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision a1e58bbd)
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_dmapi.h"
29 #include "xfs_mount.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_alloc_btree.h"
32 #include "xfs_ialloc_btree.h"
33 #include "xfs_dir2_sf.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_ialloc.h"
39 #include "xfs_alloc.h"
40 #include "xfs_rtalloc.h"
41 #include "xfs_bmap.h"
42 #include "xfs_error.h"
43 #include "xfs_rw.h"
44 #include "xfs_quota.h"
45 #include "xfs_fsops.h"
46 
47 STATIC void	xfs_mount_log_sb(xfs_mount_t *, __int64_t);
48 STATIC int	xfs_uuid_mount(xfs_mount_t *);
49 STATIC void	xfs_uuid_unmount(xfs_mount_t *mp);
50 STATIC void	xfs_unmountfs_wait(xfs_mount_t *);
51 
52 
53 #ifdef HAVE_PERCPU_SB
54 STATIC void	xfs_icsb_destroy_counters(xfs_mount_t *);
55 STATIC void	xfs_icsb_balance_counter(xfs_mount_t *, xfs_sb_field_t,
56 						int, int);
57 STATIC void	xfs_icsb_sync_counters(xfs_mount_t *);
58 STATIC int	xfs_icsb_modify_counters(xfs_mount_t *, xfs_sb_field_t,
59 						int64_t, int);
60 STATIC int	xfs_icsb_disable_counter(xfs_mount_t *, xfs_sb_field_t);
61 
62 #else
63 
64 #define xfs_icsb_destroy_counters(mp)			do { } while (0)
65 #define xfs_icsb_balance_counter(mp, a, b, c)		do { } while (0)
66 #define xfs_icsb_sync_counters(mp)			do { } while (0)
67 #define xfs_icsb_modify_counters(mp, a, b, c)		do { } while (0)
68 
69 #endif
70 
71 static const struct {
72 	short offset;
73 	short type;	/* 0 = integer
74 			 * 1 = binary / string (no translation)
75 			 */
76 } xfs_sb_info[] = {
77     { offsetof(xfs_sb_t, sb_magicnum),   0 },
78     { offsetof(xfs_sb_t, sb_blocksize),  0 },
79     { offsetof(xfs_sb_t, sb_dblocks),    0 },
80     { offsetof(xfs_sb_t, sb_rblocks),    0 },
81     { offsetof(xfs_sb_t, sb_rextents),   0 },
82     { offsetof(xfs_sb_t, sb_uuid),       1 },
83     { offsetof(xfs_sb_t, sb_logstart),   0 },
84     { offsetof(xfs_sb_t, sb_rootino),    0 },
85     { offsetof(xfs_sb_t, sb_rbmino),     0 },
86     { offsetof(xfs_sb_t, sb_rsumino),    0 },
87     { offsetof(xfs_sb_t, sb_rextsize),   0 },
88     { offsetof(xfs_sb_t, sb_agblocks),   0 },
89     { offsetof(xfs_sb_t, sb_agcount),    0 },
90     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
91     { offsetof(xfs_sb_t, sb_logblocks),  0 },
92     { offsetof(xfs_sb_t, sb_versionnum), 0 },
93     { offsetof(xfs_sb_t, sb_sectsize),   0 },
94     { offsetof(xfs_sb_t, sb_inodesize),  0 },
95     { offsetof(xfs_sb_t, sb_inopblock),  0 },
96     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
97     { offsetof(xfs_sb_t, sb_blocklog),   0 },
98     { offsetof(xfs_sb_t, sb_sectlog),    0 },
99     { offsetof(xfs_sb_t, sb_inodelog),   0 },
100     { offsetof(xfs_sb_t, sb_inopblog),   0 },
101     { offsetof(xfs_sb_t, sb_agblklog),   0 },
102     { offsetof(xfs_sb_t, sb_rextslog),   0 },
103     { offsetof(xfs_sb_t, sb_inprogress), 0 },
104     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
105     { offsetof(xfs_sb_t, sb_icount),     0 },
106     { offsetof(xfs_sb_t, sb_ifree),      0 },
107     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
108     { offsetof(xfs_sb_t, sb_frextents),  0 },
109     { offsetof(xfs_sb_t, sb_uquotino),   0 },
110     { offsetof(xfs_sb_t, sb_gquotino),   0 },
111     { offsetof(xfs_sb_t, sb_qflags),     0 },
112     { offsetof(xfs_sb_t, sb_flags),      0 },
113     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
114     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
115     { offsetof(xfs_sb_t, sb_unit),	 0 },
116     { offsetof(xfs_sb_t, sb_width),	 0 },
117     { offsetof(xfs_sb_t, sb_dirblklog),	 0 },
118     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
119     { offsetof(xfs_sb_t, sb_logsectsize),0 },
120     { offsetof(xfs_sb_t, sb_logsunit),	 0 },
121     { offsetof(xfs_sb_t, sb_features2),	 0 },
122     { offsetof(xfs_sb_t, sb_bad_features2), 0 },
123     { sizeof(xfs_sb_t),			 0 }
124 };
125 
126 /*
127  * Return a pointer to an initialized xfs_mount structure.
128  */
129 xfs_mount_t *
130 xfs_mount_init(void)
131 {
132 	xfs_mount_t *mp;
133 
134 	mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
135 
136 	if (xfs_icsb_init_counters(mp)) {
137 		mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
138 	}
139 
140 	spin_lock_init(&mp->m_sb_lock);
141 	mutex_init(&mp->m_ilock);
142 	mutex_init(&mp->m_growlock);
143 	atomic_set(&mp->m_active_trans, 0);
144 
145 	return mp;
146 }
147 
148 /*
149  * Free up the resources associated with a mount structure.  Assume that
150  * the structure was initially zeroed, so we can tell which fields got
151  * initialized.
152  */
153 void
154 xfs_mount_free(
155 	xfs_mount_t	*mp)
156 {
157 	if (mp->m_perag) {
158 		int	agno;
159 
160 		for (agno = 0; agno < mp->m_maxagi; agno++)
161 			if (mp->m_perag[agno].pagb_list)
162 				kmem_free(mp->m_perag[agno].pagb_list,
163 						sizeof(xfs_perag_busy_t) *
164 							XFS_PAGB_NUM_SLOTS);
165 		kmem_free(mp->m_perag,
166 			  sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
167 	}
168 
169 	spinlock_destroy(&mp->m_ail_lock);
170 	spinlock_destroy(&mp->m_sb_lock);
171 	mutex_destroy(&mp->m_ilock);
172 	mutex_destroy(&mp->m_growlock);
173 	if (mp->m_quotainfo)
174 		XFS_QM_DONE(mp);
175 
176 	if (mp->m_fsname != NULL)
177 		kmem_free(mp->m_fsname, mp->m_fsname_len);
178 	if (mp->m_rtname != NULL)
179 		kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
180 	if (mp->m_logname != NULL)
181 		kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
182 
183 	xfs_icsb_destroy_counters(mp);
184 }
185 
186 /*
187  * Check size of device based on the (data/realtime) block count.
188  * Note: this check is used by the growfs code as well as mount.
189  */
190 int
191 xfs_sb_validate_fsb_count(
192 	xfs_sb_t	*sbp,
193 	__uint64_t	nblocks)
194 {
195 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
196 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
197 
198 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
199 	if (nblocks >> (PAGE_CACHE_SHIFT - sbp->sb_blocklog) > ULONG_MAX)
200 		return E2BIG;
201 #else                  /* Limited by UINT_MAX of sectors */
202 	if (nblocks << (sbp->sb_blocklog - BBSHIFT) > UINT_MAX)
203 		return E2BIG;
204 #endif
205 	return 0;
206 }
207 
208 /*
209  * Check the validity of the SB found.
210  */
211 STATIC int
212 xfs_mount_validate_sb(
213 	xfs_mount_t	*mp,
214 	xfs_sb_t	*sbp,
215 	int		flags)
216 {
217 	/*
218 	 * If the log device and data device have the
219 	 * same device number, the log is internal.
220 	 * Consequently, the sb_logstart should be non-zero.  If
221 	 * we have a zero sb_logstart in this case, we may be trying to mount
222 	 * a volume filesystem in a non-volume manner.
223 	 */
224 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
225 		xfs_fs_mount_cmn_err(flags, "bad magic number");
226 		return XFS_ERROR(EWRONGFS);
227 	}
228 
229 	if (!xfs_sb_good_version(sbp)) {
230 		xfs_fs_mount_cmn_err(flags, "bad version");
231 		return XFS_ERROR(EWRONGFS);
232 	}
233 
234 	if (unlikely(
235 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
236 		xfs_fs_mount_cmn_err(flags,
237 			"filesystem is marked as having an external log; "
238 			"specify logdev on the\nmount command line.");
239 		return XFS_ERROR(EINVAL);
240 	}
241 
242 	if (unlikely(
243 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
244 		xfs_fs_mount_cmn_err(flags,
245 			"filesystem is marked as having an internal log; "
246 			"do not specify logdev on\nthe mount command line.");
247 		return XFS_ERROR(EINVAL);
248 	}
249 
250 	/*
251 	 * More sanity checking. These were stolen directly from
252 	 * xfs_repair.
253 	 */
254 	if (unlikely(
255 	    sbp->sb_agcount <= 0					||
256 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
257 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
258 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
259 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
260 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
261 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
262 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
263 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
264 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
265 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
266 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
267 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
268 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
269 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
270 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
271 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
272 		xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
273 		return XFS_ERROR(EFSCORRUPTED);
274 	}
275 
276 	/*
277 	 * Sanity check AG count, size fields against data size field
278 	 */
279 	if (unlikely(
280 	    sbp->sb_dblocks == 0 ||
281 	    sbp->sb_dblocks >
282 	     (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
283 	    sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
284 			      sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
285 		xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
286 		return XFS_ERROR(EFSCORRUPTED);
287 	}
288 
289 	if (xfs_sb_validate_fsb_count(sbp, sbp->sb_dblocks) ||
290 	    xfs_sb_validate_fsb_count(sbp, sbp->sb_rblocks)) {
291 		xfs_fs_mount_cmn_err(flags,
292 			"file system too large to be mounted on this system.");
293 		return XFS_ERROR(E2BIG);
294 	}
295 
296 	if (unlikely(sbp->sb_inprogress)) {
297 		xfs_fs_mount_cmn_err(flags, "file system busy");
298 		return XFS_ERROR(EFSCORRUPTED);
299 	}
300 
301 	/*
302 	 * Version 1 directory format has never worked on Linux.
303 	 */
304 	if (unlikely(!xfs_sb_version_hasdirv2(sbp))) {
305 		xfs_fs_mount_cmn_err(flags,
306 			"file system using version 1 directory format");
307 		return XFS_ERROR(ENOSYS);
308 	}
309 
310 	/*
311 	 * Until this is fixed only page-sized or smaller data blocks work.
312 	 */
313 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
314 		xfs_fs_mount_cmn_err(flags,
315 			"file system with blocksize %d bytes",
316 			sbp->sb_blocksize);
317 		xfs_fs_mount_cmn_err(flags,
318 			"only pagesize (%ld) or less will currently work.",
319 			PAGE_SIZE);
320 		return XFS_ERROR(ENOSYS);
321 	}
322 
323 	return 0;
324 }
325 
326 STATIC void
327 xfs_initialize_perag_icache(
328 	xfs_perag_t	*pag)
329 {
330 	if (!pag->pag_ici_init) {
331 		rwlock_init(&pag->pag_ici_lock);
332 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
333 		pag->pag_ici_init = 1;
334 	}
335 }
336 
337 xfs_agnumber_t
338 xfs_initialize_perag(
339 	xfs_mount_t	*mp,
340 	xfs_agnumber_t	agcount)
341 {
342 	xfs_agnumber_t	index, max_metadata;
343 	xfs_perag_t	*pag;
344 	xfs_agino_t	agino;
345 	xfs_ino_t	ino;
346 	xfs_sb_t	*sbp = &mp->m_sb;
347 	xfs_ino_t	max_inum = XFS_MAXINUMBER_32;
348 
349 	/* Check to see if the filesystem can overflow 32 bit inodes */
350 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
351 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
352 
353 	/* Clear the mount flag if no inode can overflow 32 bits
354 	 * on this filesystem, or if specifically requested..
355 	 */
356 	if ((mp->m_flags & XFS_MOUNT_SMALL_INUMS) && ino > max_inum) {
357 		mp->m_flags |= XFS_MOUNT_32BITINODES;
358 	} else {
359 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
360 	}
361 
362 	/* If we can overflow then setup the ag headers accordingly */
363 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
364 		/* Calculate how much should be reserved for inodes to
365 		 * meet the max inode percentage.
366 		 */
367 		if (mp->m_maxicount) {
368 			__uint64_t	icount;
369 
370 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
371 			do_div(icount, 100);
372 			icount += sbp->sb_agblocks - 1;
373 			do_div(icount, sbp->sb_agblocks);
374 			max_metadata = icount;
375 		} else {
376 			max_metadata = agcount;
377 		}
378 		for (index = 0; index < agcount; index++) {
379 			ino = XFS_AGINO_TO_INO(mp, index, agino);
380 			if (ino > max_inum) {
381 				index++;
382 				break;
383 			}
384 
385 			/* This ag is preferred for inodes */
386 			pag = &mp->m_perag[index];
387 			pag->pagi_inodeok = 1;
388 			if (index < max_metadata)
389 				pag->pagf_metadata = 1;
390 			xfs_initialize_perag_icache(pag);
391 		}
392 	} else {
393 		/* Setup default behavior for smaller filesystems */
394 		for (index = 0; index < agcount; index++) {
395 			pag = &mp->m_perag[index];
396 			pag->pagi_inodeok = 1;
397 			xfs_initialize_perag_icache(pag);
398 		}
399 	}
400 	return index;
401 }
402 
403 void
404 xfs_sb_from_disk(
405 	xfs_sb_t	*to,
406 	xfs_dsb_t	*from)
407 {
408 	to->sb_magicnum = be32_to_cpu(from->sb_magicnum);
409 	to->sb_blocksize = be32_to_cpu(from->sb_blocksize);
410 	to->sb_dblocks = be64_to_cpu(from->sb_dblocks);
411 	to->sb_rblocks = be64_to_cpu(from->sb_rblocks);
412 	to->sb_rextents = be64_to_cpu(from->sb_rextents);
413 	memcpy(&to->sb_uuid, &from->sb_uuid, sizeof(to->sb_uuid));
414 	to->sb_logstart = be64_to_cpu(from->sb_logstart);
415 	to->sb_rootino = be64_to_cpu(from->sb_rootino);
416 	to->sb_rbmino = be64_to_cpu(from->sb_rbmino);
417 	to->sb_rsumino = be64_to_cpu(from->sb_rsumino);
418 	to->sb_rextsize = be32_to_cpu(from->sb_rextsize);
419 	to->sb_agblocks = be32_to_cpu(from->sb_agblocks);
420 	to->sb_agcount = be32_to_cpu(from->sb_agcount);
421 	to->sb_rbmblocks = be32_to_cpu(from->sb_rbmblocks);
422 	to->sb_logblocks = be32_to_cpu(from->sb_logblocks);
423 	to->sb_versionnum = be16_to_cpu(from->sb_versionnum);
424 	to->sb_sectsize = be16_to_cpu(from->sb_sectsize);
425 	to->sb_inodesize = be16_to_cpu(from->sb_inodesize);
426 	to->sb_inopblock = be16_to_cpu(from->sb_inopblock);
427 	memcpy(&to->sb_fname, &from->sb_fname, sizeof(to->sb_fname));
428 	to->sb_blocklog = from->sb_blocklog;
429 	to->sb_sectlog = from->sb_sectlog;
430 	to->sb_inodelog = from->sb_inodelog;
431 	to->sb_inopblog = from->sb_inopblog;
432 	to->sb_agblklog = from->sb_agblklog;
433 	to->sb_rextslog = from->sb_rextslog;
434 	to->sb_inprogress = from->sb_inprogress;
435 	to->sb_imax_pct = from->sb_imax_pct;
436 	to->sb_icount = be64_to_cpu(from->sb_icount);
437 	to->sb_ifree = be64_to_cpu(from->sb_ifree);
438 	to->sb_fdblocks = be64_to_cpu(from->sb_fdblocks);
439 	to->sb_frextents = be64_to_cpu(from->sb_frextents);
440 	to->sb_uquotino = be64_to_cpu(from->sb_uquotino);
441 	to->sb_gquotino = be64_to_cpu(from->sb_gquotino);
442 	to->sb_qflags = be16_to_cpu(from->sb_qflags);
443 	to->sb_flags = from->sb_flags;
444 	to->sb_shared_vn = from->sb_shared_vn;
445 	to->sb_inoalignmt = be32_to_cpu(from->sb_inoalignmt);
446 	to->sb_unit = be32_to_cpu(from->sb_unit);
447 	to->sb_width = be32_to_cpu(from->sb_width);
448 	to->sb_dirblklog = from->sb_dirblklog;
449 	to->sb_logsectlog = from->sb_logsectlog;
450 	to->sb_logsectsize = be16_to_cpu(from->sb_logsectsize);
451 	to->sb_logsunit = be32_to_cpu(from->sb_logsunit);
452 	to->sb_features2 = be32_to_cpu(from->sb_features2);
453 	to->sb_bad_features2 = be32_to_cpu(from->sb_bad_features2);
454 }
455 
456 /*
457  * Copy in core superblock to ondisk one.
458  *
459  * The fields argument is mask of superblock fields to copy.
460  */
461 void
462 xfs_sb_to_disk(
463 	xfs_dsb_t	*to,
464 	xfs_sb_t	*from,
465 	__int64_t	fields)
466 {
467 	xfs_caddr_t	to_ptr = (xfs_caddr_t)to;
468 	xfs_caddr_t	from_ptr = (xfs_caddr_t)from;
469 	xfs_sb_field_t	f;
470 	int		first;
471 	int		size;
472 
473 	ASSERT(fields);
474 	if (!fields)
475 		return;
476 
477 	while (fields) {
478 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
479 		first = xfs_sb_info[f].offset;
480 		size = xfs_sb_info[f + 1].offset - first;
481 
482 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
483 
484 		if (size == 1 || xfs_sb_info[f].type == 1) {
485 			memcpy(to_ptr + first, from_ptr + first, size);
486 		} else {
487 			switch (size) {
488 			case 2:
489 				*(__be16 *)(to_ptr + first) =
490 					cpu_to_be16(*(__u16 *)(from_ptr + first));
491 				break;
492 			case 4:
493 				*(__be32 *)(to_ptr + first) =
494 					cpu_to_be32(*(__u32 *)(from_ptr + first));
495 				break;
496 			case 8:
497 				*(__be64 *)(to_ptr + first) =
498 					cpu_to_be64(*(__u64 *)(from_ptr + first));
499 				break;
500 			default:
501 				ASSERT(0);
502 			}
503 		}
504 
505 		fields &= ~(1LL << f);
506 	}
507 }
508 
509 /*
510  * xfs_readsb
511  *
512  * Does the initial read of the superblock.
513  */
514 int
515 xfs_readsb(xfs_mount_t *mp, int flags)
516 {
517 	unsigned int	sector_size;
518 	unsigned int	extra_flags;
519 	xfs_buf_t	*bp;
520 	int		error;
521 
522 	ASSERT(mp->m_sb_bp == NULL);
523 	ASSERT(mp->m_ddev_targp != NULL);
524 
525 	/*
526 	 * Allocate a (locked) buffer to hold the superblock.
527 	 * This will be kept around at all times to optimize
528 	 * access to the superblock.
529 	 */
530 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
531 	extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
532 
533 	bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
534 				BTOBB(sector_size), extra_flags);
535 	if (!bp || XFS_BUF_ISERROR(bp)) {
536 		xfs_fs_mount_cmn_err(flags, "SB read failed");
537 		error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
538 		goto fail;
539 	}
540 	ASSERT(XFS_BUF_ISBUSY(bp));
541 	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
542 
543 	/*
544 	 * Initialize the mount structure from the superblock.
545 	 * But first do some basic consistency checking.
546 	 */
547 	xfs_sb_from_disk(&mp->m_sb, XFS_BUF_TO_SBP(bp));
548 
549 	error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
550 	if (error) {
551 		xfs_fs_mount_cmn_err(flags, "SB validate failed");
552 		goto fail;
553 	}
554 
555 	/*
556 	 * We must be able to do sector-sized and sector-aligned IO.
557 	 */
558 	if (sector_size > mp->m_sb.sb_sectsize) {
559 		xfs_fs_mount_cmn_err(flags,
560 			"device supports only %u byte sectors (not %u)",
561 			sector_size, mp->m_sb.sb_sectsize);
562 		error = ENOSYS;
563 		goto fail;
564 	}
565 
566 	/*
567 	 * If device sector size is smaller than the superblock size,
568 	 * re-read the superblock so the buffer is correctly sized.
569 	 */
570 	if (sector_size < mp->m_sb.sb_sectsize) {
571 		XFS_BUF_UNMANAGE(bp);
572 		xfs_buf_relse(bp);
573 		sector_size = mp->m_sb.sb_sectsize;
574 		bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
575 					BTOBB(sector_size), extra_flags);
576 		if (!bp || XFS_BUF_ISERROR(bp)) {
577 			xfs_fs_mount_cmn_err(flags, "SB re-read failed");
578 			error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
579 			goto fail;
580 		}
581 		ASSERT(XFS_BUF_ISBUSY(bp));
582 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
583 	}
584 
585 	/* Initialize per-cpu counters */
586 	xfs_icsb_reinit_counters(mp);
587 
588 	mp->m_sb_bp = bp;
589 	xfs_buf_relse(bp);
590 	ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
591 	return 0;
592 
593  fail:
594 	if (bp) {
595 		XFS_BUF_UNMANAGE(bp);
596 		xfs_buf_relse(bp);
597 	}
598 	return error;
599 }
600 
601 
602 /*
603  * xfs_mount_common
604  *
605  * Mount initialization code establishing various mount
606  * fields from the superblock associated with the given
607  * mount structure
608  */
609 STATIC void
610 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
611 {
612 	int	i;
613 
614 	mp->m_agfrotor = mp->m_agirotor = 0;
615 	spin_lock_init(&mp->m_agirotor_lock);
616 	mp->m_maxagi = mp->m_sb.sb_agcount;
617 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
618 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
619 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
620 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
621 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
622 	mp->m_litino = sbp->sb_inodesize -
623 		((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
624 	mp->m_blockmask = sbp->sb_blocksize - 1;
625 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
626 	mp->m_blockwmask = mp->m_blockwsize - 1;
627 	INIT_LIST_HEAD(&mp->m_del_inodes);
628 
629 	/*
630 	 * Setup for attributes, in case they get created.
631 	 * This value is for inodes getting attributes for the first time,
632 	 * the per-inode value is for old attribute values.
633 	 */
634 	ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
635 	switch (sbp->sb_inodesize) {
636 	case 256:
637 		mp->m_attroffset = XFS_LITINO(mp) -
638 				   XFS_BMDR_SPACE_CALC(MINABTPTRS);
639 		break;
640 	case 512:
641 	case 1024:
642 	case 2048:
643 		mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
644 		break;
645 	default:
646 		ASSERT(0);
647 	}
648 	ASSERT(mp->m_attroffset < XFS_LITINO(mp));
649 
650 	for (i = 0; i < 2; i++) {
651 		mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
652 			xfs_alloc, i == 0);
653 		mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
654 			xfs_alloc, i == 0);
655 	}
656 	for (i = 0; i < 2; i++) {
657 		mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
658 			xfs_bmbt, i == 0);
659 		mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
660 			xfs_bmbt, i == 0);
661 	}
662 	for (i = 0; i < 2; i++) {
663 		mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
664 			xfs_inobt, i == 0);
665 		mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
666 			xfs_inobt, i == 0);
667 	}
668 
669 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
670 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
671 					sbp->sb_inopblock);
672 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
673 }
674 
675 /*
676  * xfs_initialize_perag_data
677  *
678  * Read in each per-ag structure so we can count up the number of
679  * allocated inodes, free inodes and used filesystem blocks as this
680  * information is no longer persistent in the superblock. Once we have
681  * this information, write it into the in-core superblock structure.
682  */
683 STATIC int
684 xfs_initialize_perag_data(xfs_mount_t *mp, xfs_agnumber_t agcount)
685 {
686 	xfs_agnumber_t	index;
687 	xfs_perag_t	*pag;
688 	xfs_sb_t	*sbp = &mp->m_sb;
689 	uint64_t	ifree = 0;
690 	uint64_t	ialloc = 0;
691 	uint64_t	bfree = 0;
692 	uint64_t	bfreelst = 0;
693 	uint64_t	btree = 0;
694 	int		error;
695 
696 	for (index = 0; index < agcount; index++) {
697 		/*
698 		 * read the agf, then the agi. This gets us
699 		 * all the inforamtion we need and populates the
700 		 * per-ag structures for us.
701 		 */
702 		error = xfs_alloc_pagf_init(mp, NULL, index, 0);
703 		if (error)
704 			return error;
705 
706 		error = xfs_ialloc_pagi_init(mp, NULL, index);
707 		if (error)
708 			return error;
709 		pag = &mp->m_perag[index];
710 		ifree += pag->pagi_freecount;
711 		ialloc += pag->pagi_count;
712 		bfree += pag->pagf_freeblks;
713 		bfreelst += pag->pagf_flcount;
714 		btree += pag->pagf_btreeblks;
715 	}
716 	/*
717 	 * Overwrite incore superblock counters with just-read data
718 	 */
719 	spin_lock(&mp->m_sb_lock);
720 	sbp->sb_ifree = ifree;
721 	sbp->sb_icount = ialloc;
722 	sbp->sb_fdblocks = bfree + bfreelst + btree;
723 	spin_unlock(&mp->m_sb_lock);
724 
725 	/* Fixup the per-cpu counters as well. */
726 	xfs_icsb_reinit_counters(mp);
727 
728 	return 0;
729 }
730 
731 /*
732  * Update alignment values based on mount options and sb values
733  */
734 STATIC int
735 xfs_update_alignment(xfs_mount_t *mp, int mfsi_flags, __uint64_t *update_flags)
736 {
737 	xfs_sb_t	*sbp = &(mp->m_sb);
738 
739 	if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
740 		/*
741 		 * If stripe unit and stripe width are not multiples
742 		 * of the fs blocksize turn off alignment.
743 		 */
744 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
745 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
746 			if (mp->m_flags & XFS_MOUNT_RETERR) {
747 				cmn_err(CE_WARN,
748 					"XFS: alignment check 1 failed");
749 				return XFS_ERROR(EINVAL);
750 			}
751 			mp->m_dalign = mp->m_swidth = 0;
752 		} else {
753 			/*
754 			 * Convert the stripe unit and width to FSBs.
755 			 */
756 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
757 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
758 				if (mp->m_flags & XFS_MOUNT_RETERR) {
759 					return XFS_ERROR(EINVAL);
760 				}
761 				xfs_fs_cmn_err(CE_WARN, mp,
762 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
763 					mp->m_dalign, mp->m_swidth,
764 					sbp->sb_agblocks);
765 
766 				mp->m_dalign = 0;
767 				mp->m_swidth = 0;
768 			} else if (mp->m_dalign) {
769 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
770 			} else {
771 				if (mp->m_flags & XFS_MOUNT_RETERR) {
772 					xfs_fs_cmn_err(CE_WARN, mp,
773 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
774                                         	mp->m_dalign,
775 						mp->m_blockmask +1);
776 					return XFS_ERROR(EINVAL);
777 				}
778 				mp->m_swidth = 0;
779 			}
780 		}
781 
782 		/*
783 		 * Update superblock with new values
784 		 * and log changes
785 		 */
786 		if (xfs_sb_version_hasdalign(sbp)) {
787 			if (sbp->sb_unit != mp->m_dalign) {
788 				sbp->sb_unit = mp->m_dalign;
789 				*update_flags |= XFS_SB_UNIT;
790 			}
791 			if (sbp->sb_width != mp->m_swidth) {
792 				sbp->sb_width = mp->m_swidth;
793 				*update_flags |= XFS_SB_WIDTH;
794 			}
795 		}
796 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
797 		    xfs_sb_version_hasdalign(&mp->m_sb)) {
798 			mp->m_dalign = sbp->sb_unit;
799 			mp->m_swidth = sbp->sb_width;
800 	}
801 
802 	return 0;
803 }
804 
805 /*
806  * Set the maximum inode count for this filesystem
807  */
808 STATIC void
809 xfs_set_maxicount(xfs_mount_t *mp)
810 {
811 	xfs_sb_t	*sbp = &(mp->m_sb);
812 	__uint64_t	icount;
813 
814 	if (sbp->sb_imax_pct) {
815 		/*
816 		 * Make sure the maximum inode count is a multiple
817 		 * of the units we allocate inodes in.
818 		 */
819 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
820 		do_div(icount, 100);
821 		do_div(icount, mp->m_ialloc_blks);
822 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
823 				   sbp->sb_inopblog;
824 	} else {
825 		mp->m_maxicount = 0;
826 	}
827 }
828 
829 /*
830  * Set the default minimum read and write sizes unless
831  * already specified in a mount option.
832  * We use smaller I/O sizes when the file system
833  * is being used for NFS service (wsync mount option).
834  */
835 STATIC void
836 xfs_set_rw_sizes(xfs_mount_t *mp)
837 {
838 	xfs_sb_t	*sbp = &(mp->m_sb);
839 	int		readio_log, writeio_log;
840 
841 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
842 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
843 			readio_log = XFS_WSYNC_READIO_LOG;
844 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
845 		} else {
846 			readio_log = XFS_READIO_LOG_LARGE;
847 			writeio_log = XFS_WRITEIO_LOG_LARGE;
848 		}
849 	} else {
850 		readio_log = mp->m_readio_log;
851 		writeio_log = mp->m_writeio_log;
852 	}
853 
854 	if (sbp->sb_blocklog > readio_log) {
855 		mp->m_readio_log = sbp->sb_blocklog;
856 	} else {
857 		mp->m_readio_log = readio_log;
858 	}
859 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
860 	if (sbp->sb_blocklog > writeio_log) {
861 		mp->m_writeio_log = sbp->sb_blocklog;
862 	} else {
863 		mp->m_writeio_log = writeio_log;
864 	}
865 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
866 }
867 
868 /*
869  * Set whether we're using inode alignment.
870  */
871 STATIC void
872 xfs_set_inoalignment(xfs_mount_t *mp)
873 {
874 	if (xfs_sb_version_hasalign(&mp->m_sb) &&
875 	    mp->m_sb.sb_inoalignmt >=
876 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
877 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
878 	else
879 		mp->m_inoalign_mask = 0;
880 	/*
881 	 * If we are using stripe alignment, check whether
882 	 * the stripe unit is a multiple of the inode alignment
883 	 */
884 	if (mp->m_dalign && mp->m_inoalign_mask &&
885 	    !(mp->m_dalign & mp->m_inoalign_mask))
886 		mp->m_sinoalign = mp->m_dalign;
887 	else
888 		mp->m_sinoalign = 0;
889 }
890 
891 /*
892  * Check that the data (and log if separate) are an ok size.
893  */
894 STATIC int
895 xfs_check_sizes(xfs_mount_t *mp, int mfsi_flags)
896 {
897 	xfs_buf_t	*bp;
898 	xfs_daddr_t	d;
899 	int		error;
900 
901 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
902 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
903 		cmn_err(CE_WARN, "XFS: size check 1 failed");
904 		return XFS_ERROR(E2BIG);
905 	}
906 	error = xfs_read_buf(mp, mp->m_ddev_targp,
907 			     d - XFS_FSS_TO_BB(mp, 1),
908 			     XFS_FSS_TO_BB(mp, 1), 0, &bp);
909 	if (!error) {
910 		xfs_buf_relse(bp);
911 	} else {
912 		cmn_err(CE_WARN, "XFS: size check 2 failed");
913 		if (error == ENOSPC)
914 			error = XFS_ERROR(E2BIG);
915 		return error;
916 	}
917 
918 	if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
919 	    mp->m_logdev_targp != mp->m_ddev_targp) {
920 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
921 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
922 			cmn_err(CE_WARN, "XFS: size check 3 failed");
923 			return XFS_ERROR(E2BIG);
924 		}
925 		error = xfs_read_buf(mp, mp->m_logdev_targp,
926 				     d - XFS_FSB_TO_BB(mp, 1),
927 				     XFS_FSB_TO_BB(mp, 1), 0, &bp);
928 		if (!error) {
929 			xfs_buf_relse(bp);
930 		} else {
931 			cmn_err(CE_WARN, "XFS: size check 3 failed");
932 			if (error == ENOSPC)
933 				error = XFS_ERROR(E2BIG);
934 			return error;
935 		}
936 	}
937 	return 0;
938 }
939 
940 /*
941  * xfs_mountfs
942  *
943  * This function does the following on an initial mount of a file system:
944  *	- reads the superblock from disk and init the mount struct
945  *	- if we're a 32-bit kernel, do a size check on the superblock
946  *		so we don't mount terabyte filesystems
947  *	- init mount struct realtime fields
948  *	- allocate inode hash table for fs
949  *	- init directory manager
950  *	- perform recovery and init the log manager
951  */
952 int
953 xfs_mountfs(
954 	xfs_mount_t	*mp,
955 	int		mfsi_flags)
956 {
957 	xfs_sb_t	*sbp = &(mp->m_sb);
958 	xfs_inode_t	*rip;
959 	bhv_vnode_t	*rvp = NULL;
960 	__uint64_t	resblks;
961 	__int64_t	update_flags = 0LL;
962 	uint		quotamount, quotaflags;
963 	int		agno;
964 	int		uuid_mounted = 0;
965 	int		error = 0;
966 
967 	if (mp->m_sb_bp == NULL) {
968 		error = xfs_readsb(mp, mfsi_flags);
969 		if (error)
970 			return error;
971 	}
972 	xfs_mount_common(mp, sbp);
973 
974 	/*
975 	 * Check for a mismatched features2 values.  Older kernels
976 	 * read & wrote into the wrong sb offset for sb_features2
977 	 * on some platforms due to xfs_sb_t not being 64bit size aligned
978 	 * when sb_features2 was added, which made older superblock
979 	 * reading/writing routines swap it as a 64-bit value.
980 	 *
981 	 * For backwards compatibility, we make both slots equal.
982 	 *
983 	 * If we detect a mismatched field, we OR the set bits into the
984 	 * existing features2 field in case it has already been modified; we
985 	 * don't want to lose any features.  We then update the bad location
986 	 * with the ORed value so that older kernels will see any features2
987 	 * flags, and mark the two fields as needing updates once the
988 	 * transaction subsystem is online.
989 	 */
990 	if (xfs_sb_has_mismatched_features2(sbp)) {
991 		cmn_err(CE_WARN,
992 			"XFS: correcting sb_features alignment problem");
993 		sbp->sb_features2 |= sbp->sb_bad_features2;
994 		sbp->sb_bad_features2 = sbp->sb_features2;
995 		update_flags |= XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2;
996 
997 		/*
998 		 * Re-check for ATTR2 in case it was found in bad_features2
999 		 * slot.
1000 		 */
1001 		if (xfs_sb_version_hasattr2(&mp->m_sb))
1002 			mp->m_flags |= XFS_MOUNT_ATTR2;
1003 
1004 	}
1005 
1006 	/*
1007 	 * Check if sb_agblocks is aligned at stripe boundary
1008 	 * If sb_agblocks is NOT aligned turn off m_dalign since
1009 	 * allocator alignment is within an ag, therefore ag has
1010 	 * to be aligned at stripe boundary.
1011 	 */
1012 	error = xfs_update_alignment(mp, mfsi_flags, &update_flags);
1013 	if (error)
1014 		goto error1;
1015 
1016 	xfs_alloc_compute_maxlevels(mp);
1017 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
1018 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
1019 	xfs_ialloc_compute_maxlevels(mp);
1020 
1021 	xfs_set_maxicount(mp);
1022 
1023 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
1024 
1025 	/*
1026 	 * XFS uses the uuid from the superblock as the unique
1027 	 * identifier for fsid.  We can not use the uuid from the volume
1028 	 * since a single partition filesystem is identical to a single
1029 	 * partition volume/filesystem.
1030 	 */
1031 	if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
1032 	    (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
1033 		if (xfs_uuid_mount(mp)) {
1034 			error = XFS_ERROR(EINVAL);
1035 			goto error1;
1036 		}
1037 		uuid_mounted=1;
1038 	}
1039 
1040 	/*
1041 	 * Set the minimum read and write sizes
1042 	 */
1043 	xfs_set_rw_sizes(mp);
1044 
1045 	/*
1046 	 * Set the inode cluster size.
1047 	 * This may still be overridden by the file system
1048 	 * block size if it is larger than the chosen cluster size.
1049 	 */
1050 	mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
1051 
1052 	/*
1053 	 * Set inode alignment fields
1054 	 */
1055 	xfs_set_inoalignment(mp);
1056 
1057 	/*
1058 	 * Check that the data (and log if separate) are an ok size.
1059 	 */
1060 	error = xfs_check_sizes(mp, mfsi_flags);
1061 	if (error)
1062 		goto error1;
1063 
1064 	/*
1065 	 * Initialize realtime fields in the mount structure
1066 	 */
1067 	error = xfs_rtmount_init(mp);
1068 	if (error) {
1069 		cmn_err(CE_WARN, "XFS: RT mount failed");
1070 		goto error1;
1071 	}
1072 
1073 	/*
1074 	 * For client case we are done now
1075 	 */
1076 	if (mfsi_flags & XFS_MFSI_CLIENT) {
1077 		return 0;
1078 	}
1079 
1080 	/*
1081 	 *  Copies the low order bits of the timestamp and the randomly
1082 	 *  set "sequence" number out of a UUID.
1083 	 */
1084 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
1085 
1086 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
1087 
1088 	xfs_dir_mount(mp);
1089 
1090 	/*
1091 	 * Initialize the attribute manager's entries.
1092 	 */
1093 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
1094 
1095 	/*
1096 	 * Initialize the precomputed transaction reservations values.
1097 	 */
1098 	xfs_trans_init(mp);
1099 
1100 	/*
1101 	 * Allocate and initialize the per-ag data.
1102 	 */
1103 	init_rwsem(&mp->m_peraglock);
1104 	mp->m_perag =
1105 		kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
1106 
1107 	mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
1108 
1109 	/*
1110 	 * log's mount-time initialization. Perform 1st part recovery if needed
1111 	 */
1112 	if (likely(sbp->sb_logblocks > 0)) {	/* check for volume case */
1113 		error = xfs_log_mount(mp, mp->m_logdev_targp,
1114 				      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
1115 				      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
1116 		if (error) {
1117 			cmn_err(CE_WARN, "XFS: log mount failed");
1118 			goto error2;
1119 		}
1120 	} else {	/* No log has been defined */
1121 		cmn_err(CE_WARN, "XFS: no log defined");
1122 		XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
1123 		error = XFS_ERROR(EFSCORRUPTED);
1124 		goto error2;
1125 	}
1126 
1127 	/*
1128 	 * Now the log is mounted, we know if it was an unclean shutdown or
1129 	 * not. If it was, with the first phase of recovery has completed, we
1130 	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
1131 	 * but they are recovered transactionally in the second recovery phase
1132 	 * later.
1133 	 *
1134 	 * Hence we can safely re-initialise incore superblock counters from
1135 	 * the per-ag data. These may not be correct if the filesystem was not
1136 	 * cleanly unmounted, so we need to wait for recovery to finish before
1137 	 * doing this.
1138 	 *
1139 	 * If the filesystem was cleanly unmounted, then we can trust the
1140 	 * values in the superblock to be correct and we don't need to do
1141 	 * anything here.
1142 	 *
1143 	 * If we are currently making the filesystem, the initialisation will
1144 	 * fail as the perag data is in an undefined state.
1145 	 */
1146 
1147 	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
1148 	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
1149 	     !mp->m_sb.sb_inprogress) {
1150 		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
1151 		if (error) {
1152 			goto error2;
1153 		}
1154 	}
1155 	/*
1156 	 * Get and sanity-check the root inode.
1157 	 * Save the pointer to it in the mount structure.
1158 	 */
1159 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
1160 	if (error) {
1161 		cmn_err(CE_WARN, "XFS: failed to read root inode");
1162 		goto error3;
1163 	}
1164 
1165 	ASSERT(rip != NULL);
1166 	rvp = XFS_ITOV(rip);
1167 
1168 	if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
1169 		cmn_err(CE_WARN, "XFS: corrupted root inode");
1170 		cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
1171 			XFS_BUFTARG_NAME(mp->m_ddev_targp),
1172 			(unsigned long long)rip->i_ino);
1173 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
1174 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
1175 				 mp);
1176 		error = XFS_ERROR(EFSCORRUPTED);
1177 		goto error4;
1178 	}
1179 	mp->m_rootip = rip;	/* save it */
1180 
1181 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1182 
1183 	/*
1184 	 * Initialize realtime inode pointers in the mount structure
1185 	 */
1186 	error = xfs_rtmount_inodes(mp);
1187 	if (error) {
1188 		/*
1189 		 * Free up the root inode.
1190 		 */
1191 		cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1192 		goto error4;
1193 	}
1194 
1195 	/*
1196 	 * If fs is not mounted readonly, then update the superblock changes.
1197 	 */
1198 	if (update_flags && !(mp->m_flags & XFS_MOUNT_RDONLY))
1199 		xfs_mount_log_sb(mp, update_flags);
1200 
1201 	/*
1202 	 * Initialise the XFS quota management subsystem for this mount
1203 	 */
1204 	error = XFS_QM_INIT(mp, &quotamount, &quotaflags);
1205 	if (error)
1206 		goto error4;
1207 
1208 	/*
1209 	 * Finish recovering the file system.  This part needed to be
1210 	 * delayed until after the root and real-time bitmap inodes
1211 	 * were consistently read in.
1212 	 */
1213 	error = xfs_log_mount_finish(mp, mfsi_flags);
1214 	if (error) {
1215 		cmn_err(CE_WARN, "XFS: log mount finish failed");
1216 		goto error4;
1217 	}
1218 
1219 	/*
1220 	 * Complete the quota initialisation, post-log-replay component.
1221 	 */
1222 	error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags);
1223 	if (error)
1224 		goto error4;
1225 
1226 	/*
1227 	 * Now we are mounted, reserve a small amount of unused space for
1228 	 * privileged transactions. This is needed so that transaction
1229 	 * space required for critical operations can dip into this pool
1230 	 * when at ENOSPC. This is needed for operations like create with
1231 	 * attr, unwritten extent conversion at ENOSPC, etc. Data allocations
1232 	 * are not allowed to use this reserved space.
1233 	 *
1234 	 * We default to 5% or 1024 fsbs of space reserved, whichever is smaller.
1235 	 * This may drive us straight to ENOSPC on mount, but that implies
1236 	 * we were already there on the last unmount.
1237 	 */
1238 	resblks = mp->m_sb.sb_dblocks;
1239 	do_div(resblks, 20);
1240 	resblks = min_t(__uint64_t, resblks, 1024);
1241 	xfs_reserve_blocks(mp, &resblks, NULL);
1242 
1243 	return 0;
1244 
1245  error4:
1246 	/*
1247 	 * Free up the root inode.
1248 	 */
1249 	VN_RELE(rvp);
1250  error3:
1251 	xfs_log_unmount_dealloc(mp);
1252  error2:
1253 	for (agno = 0; agno < sbp->sb_agcount; agno++)
1254 		if (mp->m_perag[agno].pagb_list)
1255 			kmem_free(mp->m_perag[agno].pagb_list,
1256 			  sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1257 	kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1258 	mp->m_perag = NULL;
1259 	/* FALLTHROUGH */
1260  error1:
1261 	if (uuid_mounted)
1262 		xfs_uuid_unmount(mp);
1263 	xfs_freesb(mp);
1264 	return error;
1265 }
1266 
1267 /*
1268  * xfs_unmountfs
1269  *
1270  * This flushes out the inodes,dquots and the superblock, unmounts the
1271  * log and makes sure that incore structures are freed.
1272  */
1273 int
1274 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1275 {
1276 	__uint64_t	resblks;
1277 
1278 	/*
1279 	 * We can potentially deadlock here if we have an inode cluster
1280 	 * that has been freed has it's buffer still pinned in memory because
1281 	 * the transaction is still sitting in a iclog. The stale inodes
1282 	 * on that buffer will have their flush locks held until the
1283 	 * transaction hits the disk and the callbacks run. the inode
1284 	 * flush takes the flush lock unconditionally and with nothing to
1285 	 * push out the iclog we will never get that unlocked. hence we
1286 	 * need to force the log first.
1287 	 */
1288 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1289 	xfs_iflush_all(mp);
1290 
1291 	XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1292 
1293 	/*
1294 	 * Flush out the log synchronously so that we know for sure
1295 	 * that nothing is pinned.  This is important because bflush()
1296 	 * will skip pinned buffers.
1297 	 */
1298 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1299 
1300 	xfs_binval(mp->m_ddev_targp);
1301 	if (mp->m_rtdev_targp) {
1302 		xfs_binval(mp->m_rtdev_targp);
1303 	}
1304 
1305 	/*
1306 	 * Unreserve any blocks we have so that when we unmount we don't account
1307 	 * the reserved free space as used. This is really only necessary for
1308 	 * lazy superblock counting because it trusts the incore superblock
1309 	 * counters to be aboslutely correct on clean unmount.
1310 	 *
1311 	 * We don't bother correcting this elsewhere for lazy superblock
1312 	 * counting because on mount of an unclean filesystem we reconstruct the
1313 	 * correct counter value and this is irrelevant.
1314 	 *
1315 	 * For non-lazy counter filesystems, this doesn't matter at all because
1316 	 * we only every apply deltas to the superblock and hence the incore
1317 	 * value does not matter....
1318 	 */
1319 	resblks = 0;
1320 	xfs_reserve_blocks(mp, &resblks, NULL);
1321 
1322 	xfs_log_sbcount(mp, 1);
1323 	xfs_unmountfs_writesb(mp);
1324 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
1325 	xfs_log_unmount(mp);			/* Done! No more fs ops. */
1326 
1327 	xfs_freesb(mp);
1328 
1329 	/*
1330 	 * All inodes from this mount point should be freed.
1331 	 */
1332 	ASSERT(mp->m_inodes == NULL);
1333 
1334 	xfs_unmountfs_close(mp, cr);
1335 	if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1336 		xfs_uuid_unmount(mp);
1337 
1338 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1339 	xfs_errortag_clearall(mp, 0);
1340 #endif
1341 	xfs_mount_free(mp);
1342 	return 0;
1343 }
1344 
1345 void
1346 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1347 {
1348 	if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
1349 		xfs_free_buftarg(mp->m_logdev_targp, 1);
1350 	if (mp->m_rtdev_targp)
1351 		xfs_free_buftarg(mp->m_rtdev_targp, 1);
1352 	xfs_free_buftarg(mp->m_ddev_targp, 0);
1353 }
1354 
1355 STATIC void
1356 xfs_unmountfs_wait(xfs_mount_t *mp)
1357 {
1358 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1359 		xfs_wait_buftarg(mp->m_logdev_targp);
1360 	if (mp->m_rtdev_targp)
1361 		xfs_wait_buftarg(mp->m_rtdev_targp);
1362 	xfs_wait_buftarg(mp->m_ddev_targp);
1363 }
1364 
1365 int
1366 xfs_fs_writable(xfs_mount_t *mp)
1367 {
1368 	return !(xfs_test_for_freeze(mp) || XFS_FORCED_SHUTDOWN(mp) ||
1369 		(mp->m_flags & XFS_MOUNT_RDONLY));
1370 }
1371 
1372 /*
1373  * xfs_log_sbcount
1374  *
1375  * Called either periodically to keep the on disk superblock values
1376  * roughly up to date or from unmount to make sure the values are
1377  * correct on a clean unmount.
1378  *
1379  * Note this code can be called during the process of freezing, so
1380  * we may need to use the transaction allocator which does not not
1381  * block when the transaction subsystem is in its frozen state.
1382  */
1383 int
1384 xfs_log_sbcount(
1385 	xfs_mount_t	*mp,
1386 	uint		sync)
1387 {
1388 	xfs_trans_t	*tp;
1389 	int		error;
1390 
1391 	if (!xfs_fs_writable(mp))
1392 		return 0;
1393 
1394 	xfs_icsb_sync_counters(mp);
1395 
1396 	/*
1397 	 * we don't need to do this if we are updating the superblock
1398 	 * counters on every modification.
1399 	 */
1400 	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
1401 		return 0;
1402 
1403 	tp = _xfs_trans_alloc(mp, XFS_TRANS_SB_COUNT);
1404 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1405 					XFS_DEFAULT_LOG_COUNT);
1406 	if (error) {
1407 		xfs_trans_cancel(tp, 0);
1408 		return error;
1409 	}
1410 
1411 	xfs_mod_sb(tp, XFS_SB_IFREE | XFS_SB_ICOUNT | XFS_SB_FDBLOCKS);
1412 	if (sync)
1413 		xfs_trans_set_sync(tp);
1414 	xfs_trans_commit(tp, 0);
1415 
1416 	return 0;
1417 }
1418 
1419 STATIC void
1420 xfs_mark_shared_ro(
1421 	xfs_mount_t	*mp,
1422 	xfs_buf_t	*bp)
1423 {
1424 	xfs_dsb_t	*sb = XFS_BUF_TO_SBP(bp);
1425 	__uint16_t	version;
1426 
1427 	if (!(sb->sb_flags & XFS_SBF_READONLY))
1428 		sb->sb_flags |= XFS_SBF_READONLY;
1429 
1430 	version = be16_to_cpu(sb->sb_versionnum);
1431 	if ((version & XFS_SB_VERSION_NUMBITS) != XFS_SB_VERSION_4 ||
1432 	    !(version & XFS_SB_VERSION_SHAREDBIT))
1433 		version |= XFS_SB_VERSION_SHAREDBIT;
1434 	sb->sb_versionnum = cpu_to_be16(version);
1435 }
1436 
1437 int
1438 xfs_unmountfs_writesb(xfs_mount_t *mp)
1439 {
1440 	xfs_buf_t	*sbp;
1441 	int		error = 0;
1442 
1443 	/*
1444 	 * skip superblock write if fs is read-only, or
1445 	 * if we are doing a forced umount.
1446 	 */
1447 	if (!((mp->m_flags & XFS_MOUNT_RDONLY) ||
1448 		XFS_FORCED_SHUTDOWN(mp))) {
1449 
1450 		sbp = xfs_getsb(mp, 0);
1451 
1452 		/*
1453 		 * mark shared-readonly if desired
1454 		 */
1455 		if (mp->m_mk_sharedro)
1456 			xfs_mark_shared_ro(mp, sbp);
1457 
1458 		XFS_BUF_UNDONE(sbp);
1459 		XFS_BUF_UNREAD(sbp);
1460 		XFS_BUF_UNDELAYWRITE(sbp);
1461 		XFS_BUF_WRITE(sbp);
1462 		XFS_BUF_UNASYNC(sbp);
1463 		ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1464 		xfsbdstrat(mp, sbp);
1465 		/* Nevermind errors we might get here. */
1466 		error = xfs_iowait(sbp);
1467 		if (error)
1468 			xfs_ioerror_alert("xfs_unmountfs_writesb",
1469 					  mp, sbp, XFS_BUF_ADDR(sbp));
1470 		if (error && mp->m_mk_sharedro)
1471 			xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
1472 		xfs_buf_relse(sbp);
1473 	}
1474 	return error;
1475 }
1476 
1477 /*
1478  * xfs_mod_sb() can be used to copy arbitrary changes to the
1479  * in-core superblock into the superblock buffer to be logged.
1480  * It does not provide the higher level of locking that is
1481  * needed to protect the in-core superblock from concurrent
1482  * access.
1483  */
1484 void
1485 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1486 {
1487 	xfs_buf_t	*bp;
1488 	int		first;
1489 	int		last;
1490 	xfs_mount_t	*mp;
1491 	xfs_sb_field_t	f;
1492 
1493 	ASSERT(fields);
1494 	if (!fields)
1495 		return;
1496 	mp = tp->t_mountp;
1497 	bp = xfs_trans_getsb(tp, mp, 0);
1498 	first = sizeof(xfs_sb_t);
1499 	last = 0;
1500 
1501 	/* translate/copy */
1502 
1503 	xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, fields);
1504 
1505 	/* find modified range */
1506 
1507 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1508 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1509 	first = xfs_sb_info[f].offset;
1510 
1511 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1512 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1513 	last = xfs_sb_info[f + 1].offset - 1;
1514 
1515 	xfs_trans_log_buf(tp, bp, first, last);
1516 }
1517 
1518 
1519 /*
1520  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1521  * a delta to a specified field in the in-core superblock.  Simply
1522  * switch on the field indicated and apply the delta to that field.
1523  * Fields are not allowed to dip below zero, so if the delta would
1524  * do this do not apply it and return EINVAL.
1525  *
1526  * The m_sb_lock must be held when this routine is called.
1527  */
1528 int
1529 xfs_mod_incore_sb_unlocked(
1530 	xfs_mount_t	*mp,
1531 	xfs_sb_field_t	field,
1532 	int64_t		delta,
1533 	int		rsvd)
1534 {
1535 	int		scounter;	/* short counter for 32 bit fields */
1536 	long long	lcounter;	/* long counter for 64 bit fields */
1537 	long long	res_used, rem;
1538 
1539 	/*
1540 	 * With the in-core superblock spin lock held, switch
1541 	 * on the indicated field.  Apply the delta to the
1542 	 * proper field.  If the fields value would dip below
1543 	 * 0, then do not apply the delta and return EINVAL.
1544 	 */
1545 	switch (field) {
1546 	case XFS_SBS_ICOUNT:
1547 		lcounter = (long long)mp->m_sb.sb_icount;
1548 		lcounter += delta;
1549 		if (lcounter < 0) {
1550 			ASSERT(0);
1551 			return XFS_ERROR(EINVAL);
1552 		}
1553 		mp->m_sb.sb_icount = lcounter;
1554 		return 0;
1555 	case XFS_SBS_IFREE:
1556 		lcounter = (long long)mp->m_sb.sb_ifree;
1557 		lcounter += delta;
1558 		if (lcounter < 0) {
1559 			ASSERT(0);
1560 			return XFS_ERROR(EINVAL);
1561 		}
1562 		mp->m_sb.sb_ifree = lcounter;
1563 		return 0;
1564 	case XFS_SBS_FDBLOCKS:
1565 		lcounter = (long long)
1566 			mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1567 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1568 
1569 		if (delta > 0) {		/* Putting blocks back */
1570 			if (res_used > delta) {
1571 				mp->m_resblks_avail += delta;
1572 			} else {
1573 				rem = delta - res_used;
1574 				mp->m_resblks_avail = mp->m_resblks;
1575 				lcounter += rem;
1576 			}
1577 		} else {				/* Taking blocks away */
1578 
1579 			lcounter += delta;
1580 
1581 		/*
1582 		 * If were out of blocks, use any available reserved blocks if
1583 		 * were allowed to.
1584 		 */
1585 
1586 			if (lcounter < 0) {
1587 				if (rsvd) {
1588 					lcounter = (long long)mp->m_resblks_avail + delta;
1589 					if (lcounter < 0) {
1590 						return XFS_ERROR(ENOSPC);
1591 					}
1592 					mp->m_resblks_avail = lcounter;
1593 					return 0;
1594 				} else {	/* not reserved */
1595 					return XFS_ERROR(ENOSPC);
1596 				}
1597 			}
1598 		}
1599 
1600 		mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1601 		return 0;
1602 	case XFS_SBS_FREXTENTS:
1603 		lcounter = (long long)mp->m_sb.sb_frextents;
1604 		lcounter += delta;
1605 		if (lcounter < 0) {
1606 			return XFS_ERROR(ENOSPC);
1607 		}
1608 		mp->m_sb.sb_frextents = lcounter;
1609 		return 0;
1610 	case XFS_SBS_DBLOCKS:
1611 		lcounter = (long long)mp->m_sb.sb_dblocks;
1612 		lcounter += delta;
1613 		if (lcounter < 0) {
1614 			ASSERT(0);
1615 			return XFS_ERROR(EINVAL);
1616 		}
1617 		mp->m_sb.sb_dblocks = lcounter;
1618 		return 0;
1619 	case XFS_SBS_AGCOUNT:
1620 		scounter = mp->m_sb.sb_agcount;
1621 		scounter += delta;
1622 		if (scounter < 0) {
1623 			ASSERT(0);
1624 			return XFS_ERROR(EINVAL);
1625 		}
1626 		mp->m_sb.sb_agcount = scounter;
1627 		return 0;
1628 	case XFS_SBS_IMAX_PCT:
1629 		scounter = mp->m_sb.sb_imax_pct;
1630 		scounter += delta;
1631 		if (scounter < 0) {
1632 			ASSERT(0);
1633 			return XFS_ERROR(EINVAL);
1634 		}
1635 		mp->m_sb.sb_imax_pct = scounter;
1636 		return 0;
1637 	case XFS_SBS_REXTSIZE:
1638 		scounter = mp->m_sb.sb_rextsize;
1639 		scounter += delta;
1640 		if (scounter < 0) {
1641 			ASSERT(0);
1642 			return XFS_ERROR(EINVAL);
1643 		}
1644 		mp->m_sb.sb_rextsize = scounter;
1645 		return 0;
1646 	case XFS_SBS_RBMBLOCKS:
1647 		scounter = mp->m_sb.sb_rbmblocks;
1648 		scounter += delta;
1649 		if (scounter < 0) {
1650 			ASSERT(0);
1651 			return XFS_ERROR(EINVAL);
1652 		}
1653 		mp->m_sb.sb_rbmblocks = scounter;
1654 		return 0;
1655 	case XFS_SBS_RBLOCKS:
1656 		lcounter = (long long)mp->m_sb.sb_rblocks;
1657 		lcounter += delta;
1658 		if (lcounter < 0) {
1659 			ASSERT(0);
1660 			return XFS_ERROR(EINVAL);
1661 		}
1662 		mp->m_sb.sb_rblocks = lcounter;
1663 		return 0;
1664 	case XFS_SBS_REXTENTS:
1665 		lcounter = (long long)mp->m_sb.sb_rextents;
1666 		lcounter += delta;
1667 		if (lcounter < 0) {
1668 			ASSERT(0);
1669 			return XFS_ERROR(EINVAL);
1670 		}
1671 		mp->m_sb.sb_rextents = lcounter;
1672 		return 0;
1673 	case XFS_SBS_REXTSLOG:
1674 		scounter = mp->m_sb.sb_rextslog;
1675 		scounter += delta;
1676 		if (scounter < 0) {
1677 			ASSERT(0);
1678 			return XFS_ERROR(EINVAL);
1679 		}
1680 		mp->m_sb.sb_rextslog = scounter;
1681 		return 0;
1682 	default:
1683 		ASSERT(0);
1684 		return XFS_ERROR(EINVAL);
1685 	}
1686 }
1687 
1688 /*
1689  * xfs_mod_incore_sb() is used to change a field in the in-core
1690  * superblock structure by the specified delta.  This modification
1691  * is protected by the m_sb_lock.  Just use the xfs_mod_incore_sb_unlocked()
1692  * routine to do the work.
1693  */
1694 int
1695 xfs_mod_incore_sb(
1696 	xfs_mount_t	*mp,
1697 	xfs_sb_field_t	field,
1698 	int64_t		delta,
1699 	int		rsvd)
1700 {
1701 	int	status;
1702 
1703 	/* check for per-cpu counters */
1704 	switch (field) {
1705 #ifdef HAVE_PERCPU_SB
1706 	case XFS_SBS_ICOUNT:
1707 	case XFS_SBS_IFREE:
1708 	case XFS_SBS_FDBLOCKS:
1709 		if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1710 			status = xfs_icsb_modify_counters(mp, field,
1711 							delta, rsvd);
1712 			break;
1713 		}
1714 		/* FALLTHROUGH */
1715 #endif
1716 	default:
1717 		spin_lock(&mp->m_sb_lock);
1718 		status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1719 		spin_unlock(&mp->m_sb_lock);
1720 		break;
1721 	}
1722 
1723 	return status;
1724 }
1725 
1726 /*
1727  * xfs_mod_incore_sb_batch() is used to change more than one field
1728  * in the in-core superblock structure at a time.  This modification
1729  * is protected by a lock internal to this module.  The fields and
1730  * changes to those fields are specified in the array of xfs_mod_sb
1731  * structures passed in.
1732  *
1733  * Either all of the specified deltas will be applied or none of
1734  * them will.  If any modified field dips below 0, then all modifications
1735  * will be backed out and EINVAL will be returned.
1736  */
1737 int
1738 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1739 {
1740 	int		status=0;
1741 	xfs_mod_sb_t	*msbp;
1742 
1743 	/*
1744 	 * Loop through the array of mod structures and apply each
1745 	 * individually.  If any fail, then back out all those
1746 	 * which have already been applied.  Do all of this within
1747 	 * the scope of the m_sb_lock so that all of the changes will
1748 	 * be atomic.
1749 	 */
1750 	spin_lock(&mp->m_sb_lock);
1751 	msbp = &msb[0];
1752 	for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1753 		/*
1754 		 * Apply the delta at index n.  If it fails, break
1755 		 * from the loop so we'll fall into the undo loop
1756 		 * below.
1757 		 */
1758 		switch (msbp->msb_field) {
1759 #ifdef HAVE_PERCPU_SB
1760 		case XFS_SBS_ICOUNT:
1761 		case XFS_SBS_IFREE:
1762 		case XFS_SBS_FDBLOCKS:
1763 			if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1764 				spin_unlock(&mp->m_sb_lock);
1765 				status = xfs_icsb_modify_counters(mp,
1766 							msbp->msb_field,
1767 							msbp->msb_delta, rsvd);
1768 				spin_lock(&mp->m_sb_lock);
1769 				break;
1770 			}
1771 			/* FALLTHROUGH */
1772 #endif
1773 		default:
1774 			status = xfs_mod_incore_sb_unlocked(mp,
1775 						msbp->msb_field,
1776 						msbp->msb_delta, rsvd);
1777 			break;
1778 		}
1779 
1780 		if (status != 0) {
1781 			break;
1782 		}
1783 	}
1784 
1785 	/*
1786 	 * If we didn't complete the loop above, then back out
1787 	 * any changes made to the superblock.  If you add code
1788 	 * between the loop above and here, make sure that you
1789 	 * preserve the value of status. Loop back until
1790 	 * we step below the beginning of the array.  Make sure
1791 	 * we don't touch anything back there.
1792 	 */
1793 	if (status != 0) {
1794 		msbp--;
1795 		while (msbp >= msb) {
1796 			switch (msbp->msb_field) {
1797 #ifdef HAVE_PERCPU_SB
1798 			case XFS_SBS_ICOUNT:
1799 			case XFS_SBS_IFREE:
1800 			case XFS_SBS_FDBLOCKS:
1801 				if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1802 					spin_unlock(&mp->m_sb_lock);
1803 					status = xfs_icsb_modify_counters(mp,
1804 							msbp->msb_field,
1805 							-(msbp->msb_delta),
1806 							rsvd);
1807 					spin_lock(&mp->m_sb_lock);
1808 					break;
1809 				}
1810 				/* FALLTHROUGH */
1811 #endif
1812 			default:
1813 				status = xfs_mod_incore_sb_unlocked(mp,
1814 							msbp->msb_field,
1815 							-(msbp->msb_delta),
1816 							rsvd);
1817 				break;
1818 			}
1819 			ASSERT(status == 0);
1820 			msbp--;
1821 		}
1822 	}
1823 	spin_unlock(&mp->m_sb_lock);
1824 	return status;
1825 }
1826 
1827 /*
1828  * xfs_getsb() is called to obtain the buffer for the superblock.
1829  * The buffer is returned locked and read in from disk.
1830  * The buffer should be released with a call to xfs_brelse().
1831  *
1832  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1833  * the superblock buffer if it can be locked without sleeping.
1834  * If it can't then we'll return NULL.
1835  */
1836 xfs_buf_t *
1837 xfs_getsb(
1838 	xfs_mount_t	*mp,
1839 	int		flags)
1840 {
1841 	xfs_buf_t	*bp;
1842 
1843 	ASSERT(mp->m_sb_bp != NULL);
1844 	bp = mp->m_sb_bp;
1845 	if (flags & XFS_BUF_TRYLOCK) {
1846 		if (!XFS_BUF_CPSEMA(bp)) {
1847 			return NULL;
1848 		}
1849 	} else {
1850 		XFS_BUF_PSEMA(bp, PRIBIO);
1851 	}
1852 	XFS_BUF_HOLD(bp);
1853 	ASSERT(XFS_BUF_ISDONE(bp));
1854 	return bp;
1855 }
1856 
1857 /*
1858  * Used to free the superblock along various error paths.
1859  */
1860 void
1861 xfs_freesb(
1862 	xfs_mount_t	*mp)
1863 {
1864 	xfs_buf_t	*bp;
1865 
1866 	/*
1867 	 * Use xfs_getsb() so that the buffer will be locked
1868 	 * when we call xfs_buf_relse().
1869 	 */
1870 	bp = xfs_getsb(mp, 0);
1871 	XFS_BUF_UNMANAGE(bp);
1872 	xfs_buf_relse(bp);
1873 	mp->m_sb_bp = NULL;
1874 }
1875 
1876 /*
1877  * See if the UUID is unique among mounted XFS filesystems.
1878  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1879  */
1880 STATIC int
1881 xfs_uuid_mount(
1882 	xfs_mount_t	*mp)
1883 {
1884 	if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1885 		cmn_err(CE_WARN,
1886 			"XFS: Filesystem %s has nil UUID - can't mount",
1887 			mp->m_fsname);
1888 		return -1;
1889 	}
1890 	if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1891 		cmn_err(CE_WARN,
1892 			"XFS: Filesystem %s has duplicate UUID - can't mount",
1893 			mp->m_fsname);
1894 		return -1;
1895 	}
1896 	return 0;
1897 }
1898 
1899 /*
1900  * Remove filesystem from the UUID table.
1901  */
1902 STATIC void
1903 xfs_uuid_unmount(
1904 	xfs_mount_t	*mp)
1905 {
1906 	uuid_table_remove(&mp->m_sb.sb_uuid);
1907 }
1908 
1909 /*
1910  * Used to log changes to the superblock unit and width fields which could
1911  * be altered by the mount options, as well as any potential sb_features2
1912  * fixup. Only the first superblock is updated.
1913  */
1914 STATIC void
1915 xfs_mount_log_sb(
1916 	xfs_mount_t	*mp,
1917 	__int64_t	fields)
1918 {
1919 	xfs_trans_t	*tp;
1920 
1921 	ASSERT(fields & (XFS_SB_UNIT | XFS_SB_WIDTH | XFS_SB_UUID |
1922 			 XFS_SB_FEATURES2 | XFS_SB_BAD_FEATURES2));
1923 
1924 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1925 	if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1926 				XFS_DEFAULT_LOG_COUNT)) {
1927 		xfs_trans_cancel(tp, 0);
1928 		return;
1929 	}
1930 	xfs_mod_sb(tp, fields);
1931 	xfs_trans_commit(tp, 0);
1932 }
1933 
1934 
1935 #ifdef HAVE_PERCPU_SB
1936 /*
1937  * Per-cpu incore superblock counters
1938  *
1939  * Simple concept, difficult implementation
1940  *
1941  * Basically, replace the incore superblock counters with a distributed per cpu
1942  * counter for contended fields (e.g.  free block count).
1943  *
1944  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1945  * hence needs to be accurately read when we are running low on space. Hence
1946  * there is a method to enable and disable the per-cpu counters based on how
1947  * much "stuff" is available in them.
1948  *
1949  * Basically, a counter is enabled if there is enough free resource to justify
1950  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1951  * ENOSPC), then we disable the counters to synchronise all callers and
1952  * re-distribute the available resources.
1953  *
1954  * If, once we redistributed the available resources, we still get a failure,
1955  * we disable the per-cpu counter and go through the slow path.
1956  *
1957  * The slow path is the current xfs_mod_incore_sb() function.  This means that
1958  * when we disable a per-cpu counter, we need to drain it's resources back to
1959  * the global superblock. We do this after disabling the counter to prevent
1960  * more threads from queueing up on the counter.
1961  *
1962  * Essentially, this means that we still need a lock in the fast path to enable
1963  * synchronisation between the global counters and the per-cpu counters. This
1964  * is not a problem because the lock will be local to a CPU almost all the time
1965  * and have little contention except when we get to ENOSPC conditions.
1966  *
1967  * Basically, this lock becomes a barrier that enables us to lock out the fast
1968  * path while we do things like enabling and disabling counters and
1969  * synchronising the counters.
1970  *
1971  * Locking rules:
1972  *
1973  * 	1. m_sb_lock before picking up per-cpu locks
1974  * 	2. per-cpu locks always picked up via for_each_online_cpu() order
1975  * 	3. accurate counter sync requires m_sb_lock + per cpu locks
1976  * 	4. modifying per-cpu counters requires holding per-cpu lock
1977  * 	5. modifying global counters requires holding m_sb_lock
1978  *	6. enabling or disabling a counter requires holding the m_sb_lock
1979  *	   and _none_ of the per-cpu locks.
1980  *
1981  * Disabled counters are only ever re-enabled by a balance operation
1982  * that results in more free resources per CPU than a given threshold.
1983  * To ensure counters don't remain disabled, they are rebalanced when
1984  * the global resource goes above a higher threshold (i.e. some hysteresis
1985  * is present to prevent thrashing).
1986  */
1987 
1988 #ifdef CONFIG_HOTPLUG_CPU
1989 /*
1990  * hot-plug CPU notifier support.
1991  *
1992  * We need a notifier per filesystem as we need to be able to identify
1993  * the filesystem to balance the counters out. This is achieved by
1994  * having a notifier block embedded in the xfs_mount_t and doing pointer
1995  * magic to get the mount pointer from the notifier block address.
1996  */
1997 STATIC int
1998 xfs_icsb_cpu_notify(
1999 	struct notifier_block *nfb,
2000 	unsigned long action,
2001 	void *hcpu)
2002 {
2003 	xfs_icsb_cnts_t *cntp;
2004 	xfs_mount_t	*mp;
2005 
2006 	mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
2007 	cntp = (xfs_icsb_cnts_t *)
2008 			per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
2009 	switch (action) {
2010 	case CPU_UP_PREPARE:
2011 	case CPU_UP_PREPARE_FROZEN:
2012 		/* Easy Case - initialize the area and locks, and
2013 		 * then rebalance when online does everything else for us. */
2014 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2015 		break;
2016 	case CPU_ONLINE:
2017 	case CPU_ONLINE_FROZEN:
2018 		xfs_icsb_lock(mp);
2019 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
2020 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
2021 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
2022 		xfs_icsb_unlock(mp);
2023 		break;
2024 	case CPU_DEAD:
2025 	case CPU_DEAD_FROZEN:
2026 		/* Disable all the counters, then fold the dead cpu's
2027 		 * count into the total on the global superblock and
2028 		 * re-enable the counters. */
2029 		xfs_icsb_lock(mp);
2030 		spin_lock(&mp->m_sb_lock);
2031 		xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
2032 		xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
2033 		xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
2034 
2035 		mp->m_sb.sb_icount += cntp->icsb_icount;
2036 		mp->m_sb.sb_ifree += cntp->icsb_ifree;
2037 		mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
2038 
2039 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2040 
2041 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
2042 					 XFS_ICSB_SB_LOCKED, 0);
2043 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
2044 					 XFS_ICSB_SB_LOCKED, 0);
2045 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
2046 					 XFS_ICSB_SB_LOCKED, 0);
2047 		spin_unlock(&mp->m_sb_lock);
2048 		xfs_icsb_unlock(mp);
2049 		break;
2050 	}
2051 
2052 	return NOTIFY_OK;
2053 }
2054 #endif /* CONFIG_HOTPLUG_CPU */
2055 
2056 int
2057 xfs_icsb_init_counters(
2058 	xfs_mount_t	*mp)
2059 {
2060 	xfs_icsb_cnts_t *cntp;
2061 	int		i;
2062 
2063 	mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
2064 	if (mp->m_sb_cnts == NULL)
2065 		return -ENOMEM;
2066 
2067 #ifdef CONFIG_HOTPLUG_CPU
2068 	mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
2069 	mp->m_icsb_notifier.priority = 0;
2070 	register_hotcpu_notifier(&mp->m_icsb_notifier);
2071 #endif /* CONFIG_HOTPLUG_CPU */
2072 
2073 	for_each_online_cpu(i) {
2074 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2075 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
2076 	}
2077 
2078 	mutex_init(&mp->m_icsb_mutex);
2079 
2080 	/*
2081 	 * start with all counters disabled so that the
2082 	 * initial balance kicks us off correctly
2083 	 */
2084 	mp->m_icsb_counters = -1;
2085 	return 0;
2086 }
2087 
2088 void
2089 xfs_icsb_reinit_counters(
2090 	xfs_mount_t	*mp)
2091 {
2092 	xfs_icsb_lock(mp);
2093 	/*
2094 	 * start with all counters disabled so that the
2095 	 * initial balance kicks us off correctly
2096 	 */
2097 	mp->m_icsb_counters = -1;
2098 	xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
2099 	xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
2100 	xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
2101 	xfs_icsb_unlock(mp);
2102 }
2103 
2104 STATIC void
2105 xfs_icsb_destroy_counters(
2106 	xfs_mount_t	*mp)
2107 {
2108 	if (mp->m_sb_cnts) {
2109 		unregister_hotcpu_notifier(&mp->m_icsb_notifier);
2110 		free_percpu(mp->m_sb_cnts);
2111 	}
2112 	mutex_destroy(&mp->m_icsb_mutex);
2113 }
2114 
2115 STATIC_INLINE void
2116 xfs_icsb_lock_cntr(
2117 	xfs_icsb_cnts_t	*icsbp)
2118 {
2119 	while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
2120 		ndelay(1000);
2121 	}
2122 }
2123 
2124 STATIC_INLINE void
2125 xfs_icsb_unlock_cntr(
2126 	xfs_icsb_cnts_t	*icsbp)
2127 {
2128 	clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
2129 }
2130 
2131 
2132 STATIC_INLINE void
2133 xfs_icsb_lock_all_counters(
2134 	xfs_mount_t	*mp)
2135 {
2136 	xfs_icsb_cnts_t *cntp;
2137 	int		i;
2138 
2139 	for_each_online_cpu(i) {
2140 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2141 		xfs_icsb_lock_cntr(cntp);
2142 	}
2143 }
2144 
2145 STATIC_INLINE void
2146 xfs_icsb_unlock_all_counters(
2147 	xfs_mount_t	*mp)
2148 {
2149 	xfs_icsb_cnts_t *cntp;
2150 	int		i;
2151 
2152 	for_each_online_cpu(i) {
2153 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2154 		xfs_icsb_unlock_cntr(cntp);
2155 	}
2156 }
2157 
2158 STATIC void
2159 xfs_icsb_count(
2160 	xfs_mount_t	*mp,
2161 	xfs_icsb_cnts_t	*cnt,
2162 	int		flags)
2163 {
2164 	xfs_icsb_cnts_t *cntp;
2165 	int		i;
2166 
2167 	memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
2168 
2169 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2170 		xfs_icsb_lock_all_counters(mp);
2171 
2172 	for_each_online_cpu(i) {
2173 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
2174 		cnt->icsb_icount += cntp->icsb_icount;
2175 		cnt->icsb_ifree += cntp->icsb_ifree;
2176 		cnt->icsb_fdblocks += cntp->icsb_fdblocks;
2177 	}
2178 
2179 	if (!(flags & XFS_ICSB_LAZY_COUNT))
2180 		xfs_icsb_unlock_all_counters(mp);
2181 }
2182 
2183 STATIC int
2184 xfs_icsb_counter_disabled(
2185 	xfs_mount_t	*mp,
2186 	xfs_sb_field_t	field)
2187 {
2188 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2189 	return test_bit(field, &mp->m_icsb_counters);
2190 }
2191 
2192 STATIC int
2193 xfs_icsb_disable_counter(
2194 	xfs_mount_t	*mp,
2195 	xfs_sb_field_t	field)
2196 {
2197 	xfs_icsb_cnts_t	cnt;
2198 
2199 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2200 
2201 	/*
2202 	 * If we are already disabled, then there is nothing to do
2203 	 * here. We check before locking all the counters to avoid
2204 	 * the expensive lock operation when being called in the
2205 	 * slow path and the counter is already disabled. This is
2206 	 * safe because the only time we set or clear this state is under
2207 	 * the m_icsb_mutex.
2208 	 */
2209 	if (xfs_icsb_counter_disabled(mp, field))
2210 		return 0;
2211 
2212 	xfs_icsb_lock_all_counters(mp);
2213 	if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
2214 		/* drain back to superblock */
2215 
2216 		xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
2217 		switch(field) {
2218 		case XFS_SBS_ICOUNT:
2219 			mp->m_sb.sb_icount = cnt.icsb_icount;
2220 			break;
2221 		case XFS_SBS_IFREE:
2222 			mp->m_sb.sb_ifree = cnt.icsb_ifree;
2223 			break;
2224 		case XFS_SBS_FDBLOCKS:
2225 			mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2226 			break;
2227 		default:
2228 			BUG();
2229 		}
2230 	}
2231 
2232 	xfs_icsb_unlock_all_counters(mp);
2233 
2234 	return 0;
2235 }
2236 
2237 STATIC void
2238 xfs_icsb_enable_counter(
2239 	xfs_mount_t	*mp,
2240 	xfs_sb_field_t	field,
2241 	uint64_t	count,
2242 	uint64_t	resid)
2243 {
2244 	xfs_icsb_cnts_t	*cntp;
2245 	int		i;
2246 
2247 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
2248 
2249 	xfs_icsb_lock_all_counters(mp);
2250 	for_each_online_cpu(i) {
2251 		cntp = per_cpu_ptr(mp->m_sb_cnts, i);
2252 		switch (field) {
2253 		case XFS_SBS_ICOUNT:
2254 			cntp->icsb_icount = count + resid;
2255 			break;
2256 		case XFS_SBS_IFREE:
2257 			cntp->icsb_ifree = count + resid;
2258 			break;
2259 		case XFS_SBS_FDBLOCKS:
2260 			cntp->icsb_fdblocks = count + resid;
2261 			break;
2262 		default:
2263 			BUG();
2264 			break;
2265 		}
2266 		resid = 0;
2267 	}
2268 	clear_bit(field, &mp->m_icsb_counters);
2269 	xfs_icsb_unlock_all_counters(mp);
2270 }
2271 
2272 void
2273 xfs_icsb_sync_counters_flags(
2274 	xfs_mount_t	*mp,
2275 	int		flags)
2276 {
2277 	xfs_icsb_cnts_t	cnt;
2278 
2279 	/* Pass 1: lock all counters */
2280 	if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2281 		spin_lock(&mp->m_sb_lock);
2282 
2283 	xfs_icsb_count(mp, &cnt, flags);
2284 
2285 	/* Step 3: update mp->m_sb fields */
2286 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2287 		mp->m_sb.sb_icount = cnt.icsb_icount;
2288 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2289 		mp->m_sb.sb_ifree = cnt.icsb_ifree;
2290 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2291 		mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2292 
2293 	if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2294 		spin_unlock(&mp->m_sb_lock);
2295 }
2296 
2297 /*
2298  * Accurate update of per-cpu counters to incore superblock
2299  */
2300 STATIC void
2301 xfs_icsb_sync_counters(
2302 	xfs_mount_t	*mp)
2303 {
2304 	xfs_icsb_sync_counters_flags(mp, 0);
2305 }
2306 
2307 /*
2308  * Balance and enable/disable counters as necessary.
2309  *
2310  * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2311  * chosen to be the same number as single on disk allocation chunk per CPU, and
2312  * free blocks is something far enough zero that we aren't going thrash when we
2313  * get near ENOSPC. We also need to supply a minimum we require per cpu to
2314  * prevent looping endlessly when xfs_alloc_space asks for more than will
2315  * be distributed to a single CPU but each CPU has enough blocks to be
2316  * reenabled.
2317  *
2318  * Note that we can be called when counters are already disabled.
2319  * xfs_icsb_disable_counter() optimises the counter locking in this case to
2320  * prevent locking every per-cpu counter needlessly.
2321  */
2322 
2323 #define XFS_ICSB_INO_CNTR_REENABLE	(uint64_t)64
2324 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2325 		(uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2326 STATIC void
2327 xfs_icsb_balance_counter(
2328 	xfs_mount_t	*mp,
2329 	xfs_sb_field_t  field,
2330 	int		flags,
2331 	int		min_per_cpu)
2332 {
2333 	uint64_t	count, resid;
2334 	int		weight = num_online_cpus();
2335 	uint64_t	min = (uint64_t)min_per_cpu;
2336 
2337 	if (!(flags & XFS_ICSB_SB_LOCKED))
2338 		spin_lock(&mp->m_sb_lock);
2339 
2340 	/* disable counter and sync counter */
2341 	xfs_icsb_disable_counter(mp, field);
2342 
2343 	/* update counters  - first CPU gets residual*/
2344 	switch (field) {
2345 	case XFS_SBS_ICOUNT:
2346 		count = mp->m_sb.sb_icount;
2347 		resid = do_div(count, weight);
2348 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2349 			goto out;
2350 		break;
2351 	case XFS_SBS_IFREE:
2352 		count = mp->m_sb.sb_ifree;
2353 		resid = do_div(count, weight);
2354 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2355 			goto out;
2356 		break;
2357 	case XFS_SBS_FDBLOCKS:
2358 		count = mp->m_sb.sb_fdblocks;
2359 		resid = do_div(count, weight);
2360 		if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2361 			goto out;
2362 		break;
2363 	default:
2364 		BUG();
2365 		count = resid = 0;	/* quiet, gcc */
2366 		break;
2367 	}
2368 
2369 	xfs_icsb_enable_counter(mp, field, count, resid);
2370 out:
2371 	if (!(flags & XFS_ICSB_SB_LOCKED))
2372 		spin_unlock(&mp->m_sb_lock);
2373 }
2374 
2375 STATIC int
2376 xfs_icsb_modify_counters(
2377 	xfs_mount_t	*mp,
2378 	xfs_sb_field_t	field,
2379 	int64_t		delta,
2380 	int		rsvd)
2381 {
2382 	xfs_icsb_cnts_t	*icsbp;
2383 	long long	lcounter;	/* long counter for 64 bit fields */
2384 	int		cpu, ret = 0;
2385 
2386 	might_sleep();
2387 again:
2388 	cpu = get_cpu();
2389 	icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2390 
2391 	/*
2392 	 * if the counter is disabled, go to slow path
2393 	 */
2394 	if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2395 		goto slow_path;
2396 	xfs_icsb_lock_cntr(icsbp);
2397 	if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2398 		xfs_icsb_unlock_cntr(icsbp);
2399 		goto slow_path;
2400 	}
2401 
2402 	switch (field) {
2403 	case XFS_SBS_ICOUNT:
2404 		lcounter = icsbp->icsb_icount;
2405 		lcounter += delta;
2406 		if (unlikely(lcounter < 0))
2407 			goto balance_counter;
2408 		icsbp->icsb_icount = lcounter;
2409 		break;
2410 
2411 	case XFS_SBS_IFREE:
2412 		lcounter = icsbp->icsb_ifree;
2413 		lcounter += delta;
2414 		if (unlikely(lcounter < 0))
2415 			goto balance_counter;
2416 		icsbp->icsb_ifree = lcounter;
2417 		break;
2418 
2419 	case XFS_SBS_FDBLOCKS:
2420 		BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2421 
2422 		lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2423 		lcounter += delta;
2424 		if (unlikely(lcounter < 0))
2425 			goto balance_counter;
2426 		icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2427 		break;
2428 	default:
2429 		BUG();
2430 		break;
2431 	}
2432 	xfs_icsb_unlock_cntr(icsbp);
2433 	put_cpu();
2434 	return 0;
2435 
2436 slow_path:
2437 	put_cpu();
2438 
2439 	/*
2440 	 * serialise with a mutex so we don't burn lots of cpu on
2441 	 * the superblock lock. We still need to hold the superblock
2442 	 * lock, however, when we modify the global structures.
2443 	 */
2444 	xfs_icsb_lock(mp);
2445 
2446 	/*
2447 	 * Now running atomically.
2448 	 *
2449 	 * If the counter is enabled, someone has beaten us to rebalancing.
2450 	 * Drop the lock and try again in the fast path....
2451 	 */
2452 	if (!(xfs_icsb_counter_disabled(mp, field))) {
2453 		xfs_icsb_unlock(mp);
2454 		goto again;
2455 	}
2456 
2457 	/*
2458 	 * The counter is currently disabled. Because we are
2459 	 * running atomically here, we know a rebalance cannot
2460 	 * be in progress. Hence we can go straight to operating
2461 	 * on the global superblock. We do not call xfs_mod_incore_sb()
2462 	 * here even though we need to get the m_sb_lock. Doing so
2463 	 * will cause us to re-enter this function and deadlock.
2464 	 * Hence we get the m_sb_lock ourselves and then call
2465 	 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2466 	 * directly on the global counters.
2467 	 */
2468 	spin_lock(&mp->m_sb_lock);
2469 	ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2470 	spin_unlock(&mp->m_sb_lock);
2471 
2472 	/*
2473 	 * Now that we've modified the global superblock, we
2474 	 * may be able to re-enable the distributed counters
2475 	 * (e.g. lots of space just got freed). After that
2476 	 * we are done.
2477 	 */
2478 	if (ret != ENOSPC)
2479 		xfs_icsb_balance_counter(mp, field, 0, 0);
2480 	xfs_icsb_unlock(mp);
2481 	return ret;
2482 
2483 balance_counter:
2484 	xfs_icsb_unlock_cntr(icsbp);
2485 	put_cpu();
2486 
2487 	/*
2488 	 * We may have multiple threads here if multiple per-cpu
2489 	 * counters run dry at the same time. This will mean we can
2490 	 * do more balances than strictly necessary but it is not
2491 	 * the common slowpath case.
2492 	 */
2493 	xfs_icsb_lock(mp);
2494 
2495 	/*
2496 	 * running atomically.
2497 	 *
2498 	 * This will leave the counter in the correct state for future
2499 	 * accesses. After the rebalance, we simply try again and our retry
2500 	 * will either succeed through the fast path or slow path without
2501 	 * another balance operation being required.
2502 	 */
2503 	xfs_icsb_balance_counter(mp, field, 0, delta);
2504 	xfs_icsb_unlock(mp);
2505 	goto again;
2506 }
2507 
2508 #endif
2509