xref: /openbmc/linux/fs/xfs/xfs_mount.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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_sbunit(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     { sizeof(xfs_sb_t),			 0 }
123 };
124 
125 /*
126  * Return a pointer to an initialized xfs_mount structure.
127  */
128 xfs_mount_t *
129 xfs_mount_init(void)
130 {
131 	xfs_mount_t *mp;
132 
133 	mp = kmem_zalloc(sizeof(xfs_mount_t), KM_SLEEP);
134 
135 	if (xfs_icsb_init_counters(mp)) {
136 		mp->m_flags |= XFS_MOUNT_NO_PERCPU_SB;
137 	}
138 
139 	AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
140 	spinlock_init(&mp->m_sb_lock, "xfs_sb");
141 	mutex_init(&mp->m_ilock);
142 	initnsema(&mp->m_growlock, 1, "xfs_grow");
143 	/*
144 	 * Initialize the AIL.
145 	 */
146 	xfs_trans_ail_init(mp);
147 
148 	atomic_set(&mp->m_active_trans, 0);
149 
150 	return mp;
151 }
152 
153 /*
154  * Free up the resources associated with a mount structure.  Assume that
155  * the structure was initially zeroed, so we can tell which fields got
156  * initialized.
157  */
158 void
159 xfs_mount_free(
160 	xfs_mount_t	*mp,
161 	int		remove_bhv)
162 {
163 	if (mp->m_ihash)
164 		xfs_ihash_free(mp);
165 	if (mp->m_chash)
166 		xfs_chash_free(mp);
167 
168 	if (mp->m_perag) {
169 		int	agno;
170 
171 		for (agno = 0; agno < mp->m_maxagi; agno++)
172 			if (mp->m_perag[agno].pagb_list)
173 				kmem_free(mp->m_perag[agno].pagb_list,
174 						sizeof(xfs_perag_busy_t) *
175 							XFS_PAGB_NUM_SLOTS);
176 		kmem_free(mp->m_perag,
177 			  sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
178 	}
179 
180 	AIL_LOCK_DESTROY(&mp->m_ail_lock);
181 	spinlock_destroy(&mp->m_sb_lock);
182 	mutex_destroy(&mp->m_ilock);
183 	freesema(&mp->m_growlock);
184 	if (mp->m_quotainfo)
185 		XFS_QM_DONE(mp);
186 
187 	if (mp->m_fsname != NULL)
188 		kmem_free(mp->m_fsname, mp->m_fsname_len);
189 	if (mp->m_rtname != NULL)
190 		kmem_free(mp->m_rtname, strlen(mp->m_rtname) + 1);
191 	if (mp->m_logname != NULL)
192 		kmem_free(mp->m_logname, strlen(mp->m_logname) + 1);
193 
194 	if (remove_bhv) {
195 		struct bhv_vfs	*vfsp = XFS_MTOVFS(mp);
196 
197 		bhv_remove_all_vfsops(vfsp, 0);
198 		VFS_REMOVEBHV(vfsp, &mp->m_bhv);
199 	}
200 
201 	xfs_icsb_destroy_counters(mp);
202 	kmem_free(mp, sizeof(xfs_mount_t));
203 }
204 
205 
206 /*
207  * Check the validity of the SB found.
208  */
209 STATIC int
210 xfs_mount_validate_sb(
211 	xfs_mount_t	*mp,
212 	xfs_sb_t	*sbp,
213 	int		flags)
214 {
215 	/*
216 	 * If the log device and data device have the
217 	 * same device number, the log is internal.
218 	 * Consequently, the sb_logstart should be non-zero.  If
219 	 * we have a zero sb_logstart in this case, we may be trying to mount
220 	 * a volume filesystem in a non-volume manner.
221 	 */
222 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
223 		xfs_fs_mount_cmn_err(flags, "bad magic number");
224 		return XFS_ERROR(EWRONGFS);
225 	}
226 
227 	if (!XFS_SB_GOOD_VERSION(sbp)) {
228 		xfs_fs_mount_cmn_err(flags, "bad version");
229 		return XFS_ERROR(EWRONGFS);
230 	}
231 
232 	if (unlikely(
233 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
234 		xfs_fs_mount_cmn_err(flags,
235 			"filesystem is marked as having an external log; "
236 			"specify logdev on the\nmount command line.");
237 		return XFS_ERROR(EINVAL);
238 	}
239 
240 	if (unlikely(
241 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
242 		xfs_fs_mount_cmn_err(flags,
243 			"filesystem is marked as having an internal log; "
244 			"do not specify logdev on\nthe mount command line.");
245 		return XFS_ERROR(EINVAL);
246 	}
247 
248 	/*
249 	 * More sanity checking. These were stolen directly from
250 	 * xfs_repair.
251 	 */
252 	if (unlikely(
253 	    sbp->sb_agcount <= 0					||
254 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
255 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
256 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
257 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
258 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
259 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
260 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
261 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
262 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
263 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
264 	    sbp->sb_inodelog < XFS_DINODE_MIN_LOG			||
265 	    sbp->sb_inodelog > XFS_DINODE_MAX_LOG			||
266 	    (sbp->sb_blocklog - sbp->sb_inodelog != sbp->sb_inopblog)	||
267 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
268 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
269 	    (sbp->sb_imax_pct > 100 /* zero sb_imax_pct is valid */))) {
270 		xfs_fs_mount_cmn_err(flags, "SB sanity check 1 failed");
271 		return XFS_ERROR(EFSCORRUPTED);
272 	}
273 
274 	/*
275 	 * Sanity check AG count, size fields against data size field
276 	 */
277 	if (unlikely(
278 	    sbp->sb_dblocks == 0 ||
279 	    sbp->sb_dblocks >
280 	     (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
281 	    sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
282 			      sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
283 		xfs_fs_mount_cmn_err(flags, "SB sanity check 2 failed");
284 		return XFS_ERROR(EFSCORRUPTED);
285 	}
286 
287 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
288 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
289 
290 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
291 	if (unlikely(
292 	    (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
293 	    (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
294 #else                  /* Limited by UINT_MAX of sectors */
295 	if (unlikely(
296 	    (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
297 	    (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
298 #endif
299 		xfs_fs_mount_cmn_err(flags,
300 			"file system too large to be mounted on this system.");
301 		return XFS_ERROR(E2BIG);
302 	}
303 
304 	if (unlikely(sbp->sb_inprogress)) {
305 		xfs_fs_mount_cmn_err(flags, "file system busy");
306 		return XFS_ERROR(EFSCORRUPTED);
307 	}
308 
309 	/*
310 	 * Version 1 directory format has never worked on Linux.
311 	 */
312 	if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
313 		xfs_fs_mount_cmn_err(flags,
314 			"file system using version 1 directory format");
315 		return XFS_ERROR(ENOSYS);
316 	}
317 
318 	/*
319 	 * Until this is fixed only page-sized or smaller data blocks work.
320 	 */
321 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
322 		xfs_fs_mount_cmn_err(flags,
323 			"file system with blocksize %d bytes",
324 			sbp->sb_blocksize);
325 		xfs_fs_mount_cmn_err(flags,
326 			"only pagesize (%ld) or less will currently work.",
327 			PAGE_SIZE);
328 		return XFS_ERROR(ENOSYS);
329 	}
330 
331 	return 0;
332 }
333 
334 xfs_agnumber_t
335 xfs_initialize_perag(
336 	bhv_vfs_t	*vfs,
337 	xfs_mount_t	*mp,
338 	xfs_agnumber_t	agcount)
339 {
340 	xfs_agnumber_t	index, max_metadata;
341 	xfs_perag_t	*pag;
342 	xfs_agino_t	agino;
343 	xfs_ino_t	ino;
344 	xfs_sb_t	*sbp = &mp->m_sb;
345 	xfs_ino_t	max_inum = XFS_MAXINUMBER_32;
346 
347 	/* Check to see if the filesystem can overflow 32 bit inodes */
348 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
349 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
350 
351 	/* Clear the mount flag if no inode can overflow 32 bits
352 	 * on this filesystem, or if specifically requested..
353 	 */
354 	if ((vfs->vfs_flag & VFS_32BITINODES) && ino > max_inum) {
355 		mp->m_flags |= XFS_MOUNT_32BITINODES;
356 	} else {
357 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
358 	}
359 
360 	/* If we can overflow then setup the ag headers accordingly */
361 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
362 		/* Calculate how much should be reserved for inodes to
363 		 * meet the max inode percentage.
364 		 */
365 		if (mp->m_maxicount) {
366 			__uint64_t	icount;
367 
368 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
369 			do_div(icount, 100);
370 			icount += sbp->sb_agblocks - 1;
371 			do_div(icount, sbp->sb_agblocks);
372 			max_metadata = icount;
373 		} else {
374 			max_metadata = agcount;
375 		}
376 		for (index = 0; index < agcount; index++) {
377 			ino = XFS_AGINO_TO_INO(mp, index, agino);
378 			if (ino > max_inum) {
379 				index++;
380 				break;
381 			}
382 
383 			/* This ag is preferred for inodes */
384 			pag = &mp->m_perag[index];
385 			pag->pagi_inodeok = 1;
386 			if (index < max_metadata)
387 				pag->pagf_metadata = 1;
388 		}
389 	} else {
390 		/* Setup default behavior for smaller filesystems */
391 		for (index = 0; index < agcount; index++) {
392 			pag = &mp->m_perag[index];
393 			pag->pagi_inodeok = 1;
394 		}
395 	}
396 	return index;
397 }
398 
399 /*
400  * xfs_xlatesb
401  *
402  *     data       - on disk version of sb
403  *     sb         - a superblock
404  *     dir        - conversion direction: <0 - convert sb to buf
405  *                                        >0 - convert buf to sb
406  *     fields     - which fields to copy (bitmask)
407  */
408 void
409 xfs_xlatesb(
410 	void		*data,
411 	xfs_sb_t	*sb,
412 	int		dir,
413 	__int64_t	fields)
414 {
415 	xfs_caddr_t	buf_ptr;
416 	xfs_caddr_t	mem_ptr;
417 	xfs_sb_field_t	f;
418 	int		first;
419 	int		size;
420 
421 	ASSERT(dir);
422 	ASSERT(fields);
423 
424 	if (!fields)
425 		return;
426 
427 	buf_ptr = (xfs_caddr_t)data;
428 	mem_ptr = (xfs_caddr_t)sb;
429 
430 	while (fields) {
431 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
432 		first = xfs_sb_info[f].offset;
433 		size = xfs_sb_info[f + 1].offset - first;
434 
435 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
436 
437 		if (size == 1 || xfs_sb_info[f].type == 1) {
438 			if (dir > 0) {
439 				memcpy(mem_ptr + first, buf_ptr + first, size);
440 			} else {
441 				memcpy(buf_ptr + first, mem_ptr + first, size);
442 			}
443 		} else {
444 			switch (size) {
445 			case 2:
446 				INT_XLATE(*(__uint16_t*)(buf_ptr+first),
447 					  *(__uint16_t*)(mem_ptr+first),
448 					  dir, ARCH_CONVERT);
449 				break;
450 			case 4:
451 				INT_XLATE(*(__uint32_t*)(buf_ptr+first),
452 					  *(__uint32_t*)(mem_ptr+first),
453 					  dir, ARCH_CONVERT);
454 				break;
455 			case 8:
456 				INT_XLATE(*(__uint64_t*)(buf_ptr+first),
457 					  *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
458 				break;
459 			default:
460 				ASSERT(0);
461 			}
462 		}
463 
464 		fields &= ~(1LL << f);
465 	}
466 }
467 
468 /*
469  * xfs_readsb
470  *
471  * Does the initial read of the superblock.
472  */
473 int
474 xfs_readsb(xfs_mount_t *mp, int flags)
475 {
476 	unsigned int	sector_size;
477 	unsigned int	extra_flags;
478 	xfs_buf_t	*bp;
479 	xfs_sb_t	*sbp;
480 	int		error;
481 
482 	ASSERT(mp->m_sb_bp == NULL);
483 	ASSERT(mp->m_ddev_targp != NULL);
484 
485 	/*
486 	 * Allocate a (locked) buffer to hold the superblock.
487 	 * This will be kept around at all times to optimize
488 	 * access to the superblock.
489 	 */
490 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
491 	extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
492 
493 	bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
494 				BTOBB(sector_size), extra_flags);
495 	if (!bp || XFS_BUF_ISERROR(bp)) {
496 		xfs_fs_mount_cmn_err(flags, "SB read failed");
497 		error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
498 		goto fail;
499 	}
500 	ASSERT(XFS_BUF_ISBUSY(bp));
501 	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
502 
503 	/*
504 	 * Initialize the mount structure from the superblock.
505 	 * But first do some basic consistency checking.
506 	 */
507 	sbp = XFS_BUF_TO_SBP(bp);
508 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
509 
510 	error = xfs_mount_validate_sb(mp, &(mp->m_sb), flags);
511 	if (error) {
512 		xfs_fs_mount_cmn_err(flags, "SB validate failed");
513 		goto fail;
514 	}
515 
516 	/*
517 	 * We must be able to do sector-sized and sector-aligned IO.
518 	 */
519 	if (sector_size > mp->m_sb.sb_sectsize) {
520 		xfs_fs_mount_cmn_err(flags,
521 			"device supports only %u byte sectors (not %u)",
522 			sector_size, mp->m_sb.sb_sectsize);
523 		error = ENOSYS;
524 		goto fail;
525 	}
526 
527 	/*
528 	 * If device sector size is smaller than the superblock size,
529 	 * re-read the superblock so the buffer is correctly sized.
530 	 */
531 	if (sector_size < mp->m_sb.sb_sectsize) {
532 		XFS_BUF_UNMANAGE(bp);
533 		xfs_buf_relse(bp);
534 		sector_size = mp->m_sb.sb_sectsize;
535 		bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
536 					BTOBB(sector_size), extra_flags);
537 		if (!bp || XFS_BUF_ISERROR(bp)) {
538 			xfs_fs_mount_cmn_err(flags, "SB re-read failed");
539 			error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
540 			goto fail;
541 		}
542 		ASSERT(XFS_BUF_ISBUSY(bp));
543 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
544 	}
545 
546 	/* Initialize per-cpu counters */
547 	xfs_icsb_reinit_counters(mp);
548 
549 	mp->m_sb_bp = bp;
550 	xfs_buf_relse(bp);
551 	ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
552 	return 0;
553 
554  fail:
555 	if (bp) {
556 		XFS_BUF_UNMANAGE(bp);
557 		xfs_buf_relse(bp);
558 	}
559 	return error;
560 }
561 
562 
563 /*
564  * xfs_mount_common
565  *
566  * Mount initialization code establishing various mount
567  * fields from the superblock associated with the given
568  * mount structure
569  */
570 STATIC void
571 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
572 {
573 	int	i;
574 
575 	mp->m_agfrotor = mp->m_agirotor = 0;
576 	spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
577 	mp->m_maxagi = mp->m_sb.sb_agcount;
578 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
579 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
580 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
581 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
582 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
583 	mp->m_litino = sbp->sb_inodesize -
584 		((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
585 	mp->m_blockmask = sbp->sb_blocksize - 1;
586 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
587 	mp->m_blockwmask = mp->m_blockwsize - 1;
588 	INIT_LIST_HEAD(&mp->m_del_inodes);
589 
590 	/*
591 	 * Setup for attributes, in case they get created.
592 	 * This value is for inodes getting attributes for the first time,
593 	 * the per-inode value is for old attribute values.
594 	 */
595 	ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
596 	switch (sbp->sb_inodesize) {
597 	case 256:
598 		mp->m_attroffset = XFS_LITINO(mp) -
599 				   XFS_BMDR_SPACE_CALC(MINABTPTRS);
600 		break;
601 	case 512:
602 	case 1024:
603 	case 2048:
604 		mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
605 		break;
606 	default:
607 		ASSERT(0);
608 	}
609 	ASSERT(mp->m_attroffset < XFS_LITINO(mp));
610 
611 	for (i = 0; i < 2; i++) {
612 		mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
613 			xfs_alloc, i == 0);
614 		mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
615 			xfs_alloc, i == 0);
616 	}
617 	for (i = 0; i < 2; i++) {
618 		mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
619 			xfs_bmbt, i == 0);
620 		mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
621 			xfs_bmbt, i == 0);
622 	}
623 	for (i = 0; i < 2; i++) {
624 		mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
625 			xfs_inobt, i == 0);
626 		mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
627 			xfs_inobt, i == 0);
628 	}
629 
630 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
631 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
632 					sbp->sb_inopblock);
633 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
634 }
635 /*
636  * xfs_mountfs
637  *
638  * This function does the following on an initial mount of a file system:
639  *	- reads the superblock from disk and init the mount struct
640  *	- if we're a 32-bit kernel, do a size check on the superblock
641  *		so we don't mount terabyte filesystems
642  *	- init mount struct realtime fields
643  *	- allocate inode hash table for fs
644  *	- init directory manager
645  *	- perform recovery and init the log manager
646  */
647 int
648 xfs_mountfs(
649 	bhv_vfs_t	*vfsp,
650 	xfs_mount_t	*mp,
651 	int		mfsi_flags)
652 {
653 	xfs_buf_t	*bp;
654 	xfs_sb_t	*sbp = &(mp->m_sb);
655 	xfs_inode_t	*rip;
656 	bhv_vnode_t	*rvp = NULL;
657 	int		readio_log, writeio_log;
658 	xfs_daddr_t	d;
659 	__uint64_t	ret64;
660 	__int64_t	update_flags;
661 	uint		quotamount, quotaflags;
662 	int		agno;
663 	int		uuid_mounted = 0;
664 	int		error = 0;
665 
666 	if (mp->m_sb_bp == NULL) {
667 		if ((error = xfs_readsb(mp, mfsi_flags))) {
668 			return error;
669 		}
670 	}
671 	xfs_mount_common(mp, sbp);
672 
673 	/*
674 	 * Check if sb_agblocks is aligned at stripe boundary
675 	 * If sb_agblocks is NOT aligned turn off m_dalign since
676 	 * allocator alignment is within an ag, therefore ag has
677 	 * to be aligned at stripe boundary.
678 	 */
679 	update_flags = 0LL;
680 	if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
681 		/*
682 		 * If stripe unit and stripe width are not multiples
683 		 * of the fs blocksize turn off alignment.
684 		 */
685 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
686 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
687 			if (mp->m_flags & XFS_MOUNT_RETERR) {
688 				cmn_err(CE_WARN,
689 					"XFS: alignment check 1 failed");
690 				error = XFS_ERROR(EINVAL);
691 				goto error1;
692 			}
693 			mp->m_dalign = mp->m_swidth = 0;
694 		} else {
695 			/*
696 			 * Convert the stripe unit and width to FSBs.
697 			 */
698 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
699 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
700 				if (mp->m_flags & XFS_MOUNT_RETERR) {
701 					error = XFS_ERROR(EINVAL);
702 					goto error1;
703 				}
704 				xfs_fs_cmn_err(CE_WARN, mp,
705 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
706 					mp->m_dalign, mp->m_swidth,
707 					sbp->sb_agblocks);
708 
709 				mp->m_dalign = 0;
710 				mp->m_swidth = 0;
711 			} else if (mp->m_dalign) {
712 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
713 			} else {
714 				if (mp->m_flags & XFS_MOUNT_RETERR) {
715 					xfs_fs_cmn_err(CE_WARN, mp,
716 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
717                                         	mp->m_dalign,
718 						mp->m_blockmask +1);
719 					error = XFS_ERROR(EINVAL);
720 					goto error1;
721 				}
722 				mp->m_swidth = 0;
723 			}
724 		}
725 
726 		/*
727 		 * Update superblock with new values
728 		 * and log changes
729 		 */
730 		if (XFS_SB_VERSION_HASDALIGN(sbp)) {
731 			if (sbp->sb_unit != mp->m_dalign) {
732 				sbp->sb_unit = mp->m_dalign;
733 				update_flags |= XFS_SB_UNIT;
734 			}
735 			if (sbp->sb_width != mp->m_swidth) {
736 				sbp->sb_width = mp->m_swidth;
737 				update_flags |= XFS_SB_WIDTH;
738 			}
739 		}
740 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
741 		    XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
742 			mp->m_dalign = sbp->sb_unit;
743 			mp->m_swidth = sbp->sb_width;
744 	}
745 
746 	xfs_alloc_compute_maxlevels(mp);
747 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
748 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
749 	xfs_ialloc_compute_maxlevels(mp);
750 
751 	if (sbp->sb_imax_pct) {
752 		__uint64_t	icount;
753 
754 		/* Make sure the maximum inode count is a multiple of the
755 		 * units we allocate inodes in.
756 		 */
757 
758 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
759 		do_div(icount, 100);
760 		do_div(icount, mp->m_ialloc_blks);
761 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
762 				   sbp->sb_inopblog;
763 	} else
764 		mp->m_maxicount = 0;
765 
766 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
767 
768 	/*
769 	 * XFS uses the uuid from the superblock as the unique
770 	 * identifier for fsid.  We can not use the uuid from the volume
771 	 * since a single partition filesystem is identical to a single
772 	 * partition volume/filesystem.
773 	 */
774 	if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
775 	    (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
776 		if (xfs_uuid_mount(mp)) {
777 			error = XFS_ERROR(EINVAL);
778 			goto error1;
779 		}
780 		uuid_mounted=1;
781 		ret64 = uuid_hash64(&sbp->sb_uuid);
782 		memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
783 	}
784 
785 	/*
786 	 * Set the default minimum read and write sizes unless
787 	 * already specified in a mount option.
788 	 * We use smaller I/O sizes when the file system
789 	 * is being used for NFS service (wsync mount option).
790 	 */
791 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
792 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
793 			readio_log = XFS_WSYNC_READIO_LOG;
794 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
795 		} else {
796 			readio_log = XFS_READIO_LOG_LARGE;
797 			writeio_log = XFS_WRITEIO_LOG_LARGE;
798 		}
799 	} else {
800 		readio_log = mp->m_readio_log;
801 		writeio_log = mp->m_writeio_log;
802 	}
803 
804 	/*
805 	 * Set the number of readahead buffers to use based on
806 	 * physical memory size.
807 	 */
808 	if (xfs_physmem <= 4096)		/* <= 16MB */
809 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
810 	else if (xfs_physmem <= 8192)	/* <= 32MB */
811 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
812 	else
813 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
814 	if (sbp->sb_blocklog > readio_log) {
815 		mp->m_readio_log = sbp->sb_blocklog;
816 	} else {
817 		mp->m_readio_log = readio_log;
818 	}
819 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
820 	if (sbp->sb_blocklog > writeio_log) {
821 		mp->m_writeio_log = sbp->sb_blocklog;
822 	} else {
823 		mp->m_writeio_log = writeio_log;
824 	}
825 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
826 
827 	/*
828 	 * Set the inode cluster size based on the physical memory
829 	 * size.  This may still be overridden by the file system
830 	 * block size if it is larger than the chosen cluster size.
831 	 */
832 	if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
833 		mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
834 	} else {
835 		mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
836 	}
837 	/*
838 	 * Set whether we're using inode alignment.
839 	 */
840 	if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
841 	    mp->m_sb.sb_inoalignmt >=
842 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
843 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
844 	else
845 		mp->m_inoalign_mask = 0;
846 	/*
847 	 * If we are using stripe alignment, check whether
848 	 * the stripe unit is a multiple of the inode alignment
849 	 */
850 	if (mp->m_dalign && mp->m_inoalign_mask &&
851 	    !(mp->m_dalign & mp->m_inoalign_mask))
852 		mp->m_sinoalign = mp->m_dalign;
853 	else
854 		mp->m_sinoalign = 0;
855 	/*
856 	 * Check that the data (and log if separate) are an ok size.
857 	 */
858 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
859 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
860 		cmn_err(CE_WARN, "XFS: size check 1 failed");
861 		error = XFS_ERROR(E2BIG);
862 		goto error1;
863 	}
864 	error = xfs_read_buf(mp, mp->m_ddev_targp,
865 			     d - XFS_FSS_TO_BB(mp, 1),
866 			     XFS_FSS_TO_BB(mp, 1), 0, &bp);
867 	if (!error) {
868 		xfs_buf_relse(bp);
869 	} else {
870 		cmn_err(CE_WARN, "XFS: size check 2 failed");
871 		if (error == ENOSPC) {
872 			error = XFS_ERROR(E2BIG);
873 		}
874 		goto error1;
875 	}
876 
877 	if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
878 	    mp->m_logdev_targp != mp->m_ddev_targp) {
879 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
880 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
881 			cmn_err(CE_WARN, "XFS: size check 3 failed");
882 			error = XFS_ERROR(E2BIG);
883 			goto error1;
884 		}
885 		error = xfs_read_buf(mp, mp->m_logdev_targp,
886 				     d - XFS_FSB_TO_BB(mp, 1),
887 				     XFS_FSB_TO_BB(mp, 1), 0, &bp);
888 		if (!error) {
889 			xfs_buf_relse(bp);
890 		} else {
891 			cmn_err(CE_WARN, "XFS: size check 3 failed");
892 			if (error == ENOSPC) {
893 				error = XFS_ERROR(E2BIG);
894 			}
895 			goto error1;
896 		}
897 	}
898 
899 	/*
900 	 * Initialize realtime fields in the mount structure
901 	 */
902 	if ((error = xfs_rtmount_init(mp))) {
903 		cmn_err(CE_WARN, "XFS: RT mount failed");
904 		goto error1;
905 	}
906 
907 	/*
908 	 * For client case we are done now
909 	 */
910 	if (mfsi_flags & XFS_MFSI_CLIENT) {
911 		return 0;
912 	}
913 
914 	/*
915 	 *  Copies the low order bits of the timestamp and the randomly
916 	 *  set "sequence" number out of a UUID.
917 	 */
918 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
919 
920 	/*
921 	 *  The vfs structure needs to have a file system independent
922 	 *  way of checking for the invariant file system ID.  Since it
923 	 *  can't look at mount structures it has a pointer to the data
924 	 *  in the mount structure.
925 	 *
926 	 *  File systems that don't support user level file handles (i.e.
927 	 *  all of them except for XFS) will leave vfs_altfsid as NULL.
928 	 */
929 	vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
930 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
931 
932 	xfs_dir_mount(mp);
933 
934 	/*
935 	 * Initialize the attribute manager's entries.
936 	 */
937 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
938 
939 	/*
940 	 * Initialize the precomputed transaction reservations values.
941 	 */
942 	xfs_trans_init(mp);
943 
944 	/*
945 	 * Allocate and initialize the inode hash table for this
946 	 * file system.
947 	 */
948 	xfs_ihash_init(mp);
949 	xfs_chash_init(mp);
950 
951 	/*
952 	 * Allocate and initialize the per-ag data.
953 	 */
954 	init_rwsem(&mp->m_peraglock);
955 	mp->m_perag =
956 		kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
957 
958 	mp->m_maxagi = xfs_initialize_perag(vfsp, mp, sbp->sb_agcount);
959 
960 	/*
961 	 * log's mount-time initialization. Perform 1st part recovery if needed
962 	 */
963 	if (likely(sbp->sb_logblocks > 0)) {	/* check for volume case */
964 		error = xfs_log_mount(mp, mp->m_logdev_targp,
965 				      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
966 				      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
967 		if (error) {
968 			cmn_err(CE_WARN, "XFS: log mount failed");
969 			goto error2;
970 		}
971 	} else {	/* No log has been defined */
972 		cmn_err(CE_WARN, "XFS: no log defined");
973 		XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
974 		error = XFS_ERROR(EFSCORRUPTED);
975 		goto error2;
976 	}
977 
978 	/*
979 	 * Get and sanity-check the root inode.
980 	 * Save the pointer to it in the mount structure.
981 	 */
982 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
983 	if (error) {
984 		cmn_err(CE_WARN, "XFS: failed to read root inode");
985 		goto error3;
986 	}
987 
988 	ASSERT(rip != NULL);
989 	rvp = XFS_ITOV(rip);
990 
991 	if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
992 		cmn_err(CE_WARN, "XFS: corrupted root inode");
993 		cmn_err(CE_WARN, "Device %s - root %llu is not a directory",
994 			XFS_BUFTARG_NAME(mp->m_ddev_targp),
995 			(unsigned long long)rip->i_ino);
996 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
997 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
998 				 mp);
999 		error = XFS_ERROR(EFSCORRUPTED);
1000 		goto error4;
1001 	}
1002 	mp->m_rootip = rip;	/* save it */
1003 
1004 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1005 
1006 	/*
1007 	 * Initialize realtime inode pointers in the mount structure
1008 	 */
1009 	if ((error = xfs_rtmount_inodes(mp))) {
1010 		/*
1011 		 * Free up the root inode.
1012 		 */
1013 		cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1014 		goto error4;
1015 	}
1016 
1017 	/*
1018 	 * If fs is not mounted readonly, then update the superblock
1019 	 * unit and width changes.
1020 	 */
1021 	if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1022 		xfs_mount_log_sbunit(mp, update_flags);
1023 
1024 	/*
1025 	 * Initialise the XFS quota management subsystem for this mount
1026 	 */
1027 	if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1028 		goto error4;
1029 
1030 	/*
1031 	 * Finish recovering the file system.  This part needed to be
1032 	 * delayed until after the root and real-time bitmap inodes
1033 	 * were consistently read in.
1034 	 */
1035 	error = xfs_log_mount_finish(mp, mfsi_flags);
1036 	if (error) {
1037 		cmn_err(CE_WARN, "XFS: log mount finish failed");
1038 		goto error4;
1039 	}
1040 
1041 	/*
1042 	 * Complete the quota initialisation, post-log-replay component.
1043 	 */
1044 	if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1045 		goto error4;
1046 
1047 	return 0;
1048 
1049  error4:
1050 	/*
1051 	 * Free up the root inode.
1052 	 */
1053 	VN_RELE(rvp);
1054  error3:
1055 	xfs_log_unmount_dealloc(mp);
1056  error2:
1057 	xfs_ihash_free(mp);
1058 	xfs_chash_free(mp);
1059 	for (agno = 0; agno < sbp->sb_agcount; agno++)
1060 		if (mp->m_perag[agno].pagb_list)
1061 			kmem_free(mp->m_perag[agno].pagb_list,
1062 			  sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1063 	kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1064 	mp->m_perag = NULL;
1065 	/* FALLTHROUGH */
1066  error1:
1067 	if (uuid_mounted)
1068 		xfs_uuid_unmount(mp);
1069 	xfs_freesb(mp);
1070 	return error;
1071 }
1072 
1073 /*
1074  * xfs_unmountfs
1075  *
1076  * This flushes out the inodes,dquots and the superblock, unmounts the
1077  * log and makes sure that incore structures are freed.
1078  */
1079 int
1080 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1081 {
1082 	struct bhv_vfs	*vfsp = XFS_MTOVFS(mp);
1083 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1084 	int64_t		fsid;
1085 #endif
1086 
1087 	xfs_iflush_all(mp);
1088 
1089 	XFS_QM_DQPURGEALL(mp, XFS_QMOPT_QUOTALL | XFS_QMOPT_UMOUNTING);
1090 
1091 	/*
1092 	 * Flush out the log synchronously so that we know for sure
1093 	 * that nothing is pinned.  This is important because bflush()
1094 	 * will skip pinned buffers.
1095 	 */
1096 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1097 
1098 	xfs_binval(mp->m_ddev_targp);
1099 	if (mp->m_rtdev_targp) {
1100 		xfs_binval(mp->m_rtdev_targp);
1101 	}
1102 
1103 	xfs_unmountfs_writesb(mp);
1104 
1105 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
1106 
1107 	xfs_log_unmount(mp);			/* Done! No more fs ops. */
1108 
1109 	xfs_freesb(mp);
1110 
1111 	/*
1112 	 * All inodes from this mount point should be freed.
1113 	 */
1114 	ASSERT(mp->m_inodes == NULL);
1115 
1116 	xfs_unmountfs_close(mp, cr);
1117 	if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1118 		xfs_uuid_unmount(mp);
1119 
1120 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1121 	/*
1122 	 * clear all error tags on this filesystem
1123 	 */
1124 	memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1125 	xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1126 #endif
1127 	XFS_IODONE(vfsp);
1128 	xfs_mount_free(mp, 1);
1129 	return 0;
1130 }
1131 
1132 void
1133 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1134 {
1135 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1136 		xfs_free_buftarg(mp->m_logdev_targp, 1);
1137 	if (mp->m_rtdev_targp)
1138 		xfs_free_buftarg(mp->m_rtdev_targp, 1);
1139 	xfs_free_buftarg(mp->m_ddev_targp, 0);
1140 }
1141 
1142 STATIC void
1143 xfs_unmountfs_wait(xfs_mount_t *mp)
1144 {
1145 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1146 		xfs_wait_buftarg(mp->m_logdev_targp);
1147 	if (mp->m_rtdev_targp)
1148 		xfs_wait_buftarg(mp->m_rtdev_targp);
1149 	xfs_wait_buftarg(mp->m_ddev_targp);
1150 }
1151 
1152 int
1153 xfs_unmountfs_writesb(xfs_mount_t *mp)
1154 {
1155 	xfs_buf_t	*sbp;
1156 	xfs_sb_t	*sb;
1157 	int		error = 0;
1158 
1159 	/*
1160 	 * skip superblock write if fs is read-only, or
1161 	 * if we are doing a forced umount.
1162 	 */
1163 	sbp = xfs_getsb(mp, 0);
1164 	if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1165 		XFS_FORCED_SHUTDOWN(mp))) {
1166 
1167 		xfs_icsb_sync_counters(mp);
1168 
1169 		/*
1170 		 * mark shared-readonly if desired
1171 		 */
1172 		sb = XFS_BUF_TO_SBP(sbp);
1173 		if (mp->m_mk_sharedro) {
1174 			if (!(sb->sb_flags & XFS_SBF_READONLY))
1175 				sb->sb_flags |= XFS_SBF_READONLY;
1176 			if (!XFS_SB_VERSION_HASSHARED(sb))
1177 				XFS_SB_VERSION_ADDSHARED(sb);
1178 			xfs_fs_cmn_err(CE_NOTE, mp,
1179 				"Unmounting, marking shared read-only");
1180 		}
1181 		XFS_BUF_UNDONE(sbp);
1182 		XFS_BUF_UNREAD(sbp);
1183 		XFS_BUF_UNDELAYWRITE(sbp);
1184 		XFS_BUF_WRITE(sbp);
1185 		XFS_BUF_UNASYNC(sbp);
1186 		ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1187 		xfsbdstrat(mp, sbp);
1188 		/* Nevermind errors we might get here. */
1189 		error = xfs_iowait(sbp);
1190 		if (error)
1191 			xfs_ioerror_alert("xfs_unmountfs_writesb",
1192 					  mp, sbp, XFS_BUF_ADDR(sbp));
1193 		if (error && mp->m_mk_sharedro)
1194 			xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
1195 	}
1196 	xfs_buf_relse(sbp);
1197 	return error;
1198 }
1199 
1200 /*
1201  * xfs_mod_sb() can be used to copy arbitrary changes to the
1202  * in-core superblock into the superblock buffer to be logged.
1203  * It does not provide the higher level of locking that is
1204  * needed to protect the in-core superblock from concurrent
1205  * access.
1206  */
1207 void
1208 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1209 {
1210 	xfs_buf_t	*bp;
1211 	int		first;
1212 	int		last;
1213 	xfs_mount_t	*mp;
1214 	xfs_sb_t	*sbp;
1215 	xfs_sb_field_t	f;
1216 
1217 	ASSERT(fields);
1218 	if (!fields)
1219 		return;
1220 	mp = tp->t_mountp;
1221 	bp = xfs_trans_getsb(tp, mp, 0);
1222 	sbp = XFS_BUF_TO_SBP(bp);
1223 	first = sizeof(xfs_sb_t);
1224 	last = 0;
1225 
1226 	/* translate/copy */
1227 
1228 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1229 
1230 	/* find modified range */
1231 
1232 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1233 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1234 	first = xfs_sb_info[f].offset;
1235 
1236 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1237 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1238 	last = xfs_sb_info[f + 1].offset - 1;
1239 
1240 	xfs_trans_log_buf(tp, bp, first, last);
1241 }
1242 
1243 
1244 /*
1245  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1246  * a delta to a specified field in the in-core superblock.  Simply
1247  * switch on the field indicated and apply the delta to that field.
1248  * Fields are not allowed to dip below zero, so if the delta would
1249  * do this do not apply it and return EINVAL.
1250  *
1251  * The SB_LOCK must be held when this routine is called.
1252  */
1253 int
1254 xfs_mod_incore_sb_unlocked(
1255 	xfs_mount_t	*mp,
1256 	xfs_sb_field_t	field,
1257 	int64_t		delta,
1258 	int		rsvd)
1259 {
1260 	int		scounter;	/* short counter for 32 bit fields */
1261 	long long	lcounter;	/* long counter for 64 bit fields */
1262 	long long	res_used, rem;
1263 
1264 	/*
1265 	 * With the in-core superblock spin lock held, switch
1266 	 * on the indicated field.  Apply the delta to the
1267 	 * proper field.  If the fields value would dip below
1268 	 * 0, then do not apply the delta and return EINVAL.
1269 	 */
1270 	switch (field) {
1271 	case XFS_SBS_ICOUNT:
1272 		lcounter = (long long)mp->m_sb.sb_icount;
1273 		lcounter += delta;
1274 		if (lcounter < 0) {
1275 			ASSERT(0);
1276 			return XFS_ERROR(EINVAL);
1277 		}
1278 		mp->m_sb.sb_icount = lcounter;
1279 		return 0;
1280 	case XFS_SBS_IFREE:
1281 		lcounter = (long long)mp->m_sb.sb_ifree;
1282 		lcounter += delta;
1283 		if (lcounter < 0) {
1284 			ASSERT(0);
1285 			return XFS_ERROR(EINVAL);
1286 		}
1287 		mp->m_sb.sb_ifree = lcounter;
1288 		return 0;
1289 	case XFS_SBS_FDBLOCKS:
1290 		lcounter = (long long)
1291 			mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
1292 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1293 
1294 		if (delta > 0) {		/* Putting blocks back */
1295 			if (res_used > delta) {
1296 				mp->m_resblks_avail += delta;
1297 			} else {
1298 				rem = delta - res_used;
1299 				mp->m_resblks_avail = mp->m_resblks;
1300 				lcounter += rem;
1301 			}
1302 		} else {				/* Taking blocks away */
1303 
1304 			lcounter += delta;
1305 
1306 		/*
1307 		 * If were out of blocks, use any available reserved blocks if
1308 		 * were allowed to.
1309 		 */
1310 
1311 			if (lcounter < 0) {
1312 				if (rsvd) {
1313 					lcounter = (long long)mp->m_resblks_avail + delta;
1314 					if (lcounter < 0) {
1315 						return XFS_ERROR(ENOSPC);
1316 					}
1317 					mp->m_resblks_avail = lcounter;
1318 					return 0;
1319 				} else {	/* not reserved */
1320 					return XFS_ERROR(ENOSPC);
1321 				}
1322 			}
1323 		}
1324 
1325 		mp->m_sb.sb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
1326 		return 0;
1327 	case XFS_SBS_FREXTENTS:
1328 		lcounter = (long long)mp->m_sb.sb_frextents;
1329 		lcounter += delta;
1330 		if (lcounter < 0) {
1331 			return XFS_ERROR(ENOSPC);
1332 		}
1333 		mp->m_sb.sb_frextents = lcounter;
1334 		return 0;
1335 	case XFS_SBS_DBLOCKS:
1336 		lcounter = (long long)mp->m_sb.sb_dblocks;
1337 		lcounter += delta;
1338 		if (lcounter < 0) {
1339 			ASSERT(0);
1340 			return XFS_ERROR(EINVAL);
1341 		}
1342 		mp->m_sb.sb_dblocks = lcounter;
1343 		return 0;
1344 	case XFS_SBS_AGCOUNT:
1345 		scounter = mp->m_sb.sb_agcount;
1346 		scounter += delta;
1347 		if (scounter < 0) {
1348 			ASSERT(0);
1349 			return XFS_ERROR(EINVAL);
1350 		}
1351 		mp->m_sb.sb_agcount = scounter;
1352 		return 0;
1353 	case XFS_SBS_IMAX_PCT:
1354 		scounter = mp->m_sb.sb_imax_pct;
1355 		scounter += delta;
1356 		if (scounter < 0) {
1357 			ASSERT(0);
1358 			return XFS_ERROR(EINVAL);
1359 		}
1360 		mp->m_sb.sb_imax_pct = scounter;
1361 		return 0;
1362 	case XFS_SBS_REXTSIZE:
1363 		scounter = mp->m_sb.sb_rextsize;
1364 		scounter += delta;
1365 		if (scounter < 0) {
1366 			ASSERT(0);
1367 			return XFS_ERROR(EINVAL);
1368 		}
1369 		mp->m_sb.sb_rextsize = scounter;
1370 		return 0;
1371 	case XFS_SBS_RBMBLOCKS:
1372 		scounter = mp->m_sb.sb_rbmblocks;
1373 		scounter += delta;
1374 		if (scounter < 0) {
1375 			ASSERT(0);
1376 			return XFS_ERROR(EINVAL);
1377 		}
1378 		mp->m_sb.sb_rbmblocks = scounter;
1379 		return 0;
1380 	case XFS_SBS_RBLOCKS:
1381 		lcounter = (long long)mp->m_sb.sb_rblocks;
1382 		lcounter += delta;
1383 		if (lcounter < 0) {
1384 			ASSERT(0);
1385 			return XFS_ERROR(EINVAL);
1386 		}
1387 		mp->m_sb.sb_rblocks = lcounter;
1388 		return 0;
1389 	case XFS_SBS_REXTENTS:
1390 		lcounter = (long long)mp->m_sb.sb_rextents;
1391 		lcounter += delta;
1392 		if (lcounter < 0) {
1393 			ASSERT(0);
1394 			return XFS_ERROR(EINVAL);
1395 		}
1396 		mp->m_sb.sb_rextents = lcounter;
1397 		return 0;
1398 	case XFS_SBS_REXTSLOG:
1399 		scounter = mp->m_sb.sb_rextslog;
1400 		scounter += delta;
1401 		if (scounter < 0) {
1402 			ASSERT(0);
1403 			return XFS_ERROR(EINVAL);
1404 		}
1405 		mp->m_sb.sb_rextslog = scounter;
1406 		return 0;
1407 	default:
1408 		ASSERT(0);
1409 		return XFS_ERROR(EINVAL);
1410 	}
1411 }
1412 
1413 /*
1414  * xfs_mod_incore_sb() is used to change a field in the in-core
1415  * superblock structure by the specified delta.  This modification
1416  * is protected by the SB_LOCK.  Just use the xfs_mod_incore_sb_unlocked()
1417  * routine to do the work.
1418  */
1419 int
1420 xfs_mod_incore_sb(
1421 	xfs_mount_t	*mp,
1422 	xfs_sb_field_t	field,
1423 	int64_t		delta,
1424 	int		rsvd)
1425 {
1426 	unsigned long	s;
1427 	int	status;
1428 
1429 	/* check for per-cpu counters */
1430 	switch (field) {
1431 #ifdef HAVE_PERCPU_SB
1432 	case XFS_SBS_ICOUNT:
1433 	case XFS_SBS_IFREE:
1434 	case XFS_SBS_FDBLOCKS:
1435 		if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1436 			status = xfs_icsb_modify_counters(mp, field,
1437 							delta, rsvd);
1438 			break;
1439 		}
1440 		/* FALLTHROUGH */
1441 #endif
1442 	default:
1443 		s = XFS_SB_LOCK(mp);
1444 		status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1445 		XFS_SB_UNLOCK(mp, s);
1446 		break;
1447 	}
1448 
1449 	return status;
1450 }
1451 
1452 /*
1453  * xfs_mod_incore_sb_batch() is used to change more than one field
1454  * in the in-core superblock structure at a time.  This modification
1455  * is protected by a lock internal to this module.  The fields and
1456  * changes to those fields are specified in the array of xfs_mod_sb
1457  * structures passed in.
1458  *
1459  * Either all of the specified deltas will be applied or none of
1460  * them will.  If any modified field dips below 0, then all modifications
1461  * will be backed out and EINVAL will be returned.
1462  */
1463 int
1464 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1465 {
1466 	unsigned long	s;
1467 	int		status=0;
1468 	xfs_mod_sb_t	*msbp;
1469 
1470 	/*
1471 	 * Loop through the array of mod structures and apply each
1472 	 * individually.  If any fail, then back out all those
1473 	 * which have already been applied.  Do all of this within
1474 	 * the scope of the SB_LOCK so that all of the changes will
1475 	 * be atomic.
1476 	 */
1477 	s = XFS_SB_LOCK(mp);
1478 	msbp = &msb[0];
1479 	for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1480 		/*
1481 		 * Apply the delta at index n.  If it fails, break
1482 		 * from the loop so we'll fall into the undo loop
1483 		 * below.
1484 		 */
1485 		switch (msbp->msb_field) {
1486 #ifdef HAVE_PERCPU_SB
1487 		case XFS_SBS_ICOUNT:
1488 		case XFS_SBS_IFREE:
1489 		case XFS_SBS_FDBLOCKS:
1490 			if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1491 				XFS_SB_UNLOCK(mp, s);
1492 				status = xfs_icsb_modify_counters(mp,
1493 							msbp->msb_field,
1494 							msbp->msb_delta, rsvd);
1495 				s = XFS_SB_LOCK(mp);
1496 				break;
1497 			}
1498 			/* FALLTHROUGH */
1499 #endif
1500 		default:
1501 			status = xfs_mod_incore_sb_unlocked(mp,
1502 						msbp->msb_field,
1503 						msbp->msb_delta, rsvd);
1504 			break;
1505 		}
1506 
1507 		if (status != 0) {
1508 			break;
1509 		}
1510 	}
1511 
1512 	/*
1513 	 * If we didn't complete the loop above, then back out
1514 	 * any changes made to the superblock.  If you add code
1515 	 * between the loop above and here, make sure that you
1516 	 * preserve the value of status. Loop back until
1517 	 * we step below the beginning of the array.  Make sure
1518 	 * we don't touch anything back there.
1519 	 */
1520 	if (status != 0) {
1521 		msbp--;
1522 		while (msbp >= msb) {
1523 			switch (msbp->msb_field) {
1524 #ifdef HAVE_PERCPU_SB
1525 			case XFS_SBS_ICOUNT:
1526 			case XFS_SBS_IFREE:
1527 			case XFS_SBS_FDBLOCKS:
1528 				if (!(mp->m_flags & XFS_MOUNT_NO_PERCPU_SB)) {
1529 					XFS_SB_UNLOCK(mp, s);
1530 					status = xfs_icsb_modify_counters(mp,
1531 							msbp->msb_field,
1532 							-(msbp->msb_delta),
1533 							rsvd);
1534 					s = XFS_SB_LOCK(mp);
1535 					break;
1536 				}
1537 				/* FALLTHROUGH */
1538 #endif
1539 			default:
1540 				status = xfs_mod_incore_sb_unlocked(mp,
1541 							msbp->msb_field,
1542 							-(msbp->msb_delta),
1543 							rsvd);
1544 				break;
1545 			}
1546 			ASSERT(status == 0);
1547 			msbp--;
1548 		}
1549 	}
1550 	XFS_SB_UNLOCK(mp, s);
1551 	return status;
1552 }
1553 
1554 /*
1555  * xfs_getsb() is called to obtain the buffer for the superblock.
1556  * The buffer is returned locked and read in from disk.
1557  * The buffer should be released with a call to xfs_brelse().
1558  *
1559  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1560  * the superblock buffer if it can be locked without sleeping.
1561  * If it can't then we'll return NULL.
1562  */
1563 xfs_buf_t *
1564 xfs_getsb(
1565 	xfs_mount_t	*mp,
1566 	int		flags)
1567 {
1568 	xfs_buf_t	*bp;
1569 
1570 	ASSERT(mp->m_sb_bp != NULL);
1571 	bp = mp->m_sb_bp;
1572 	if (flags & XFS_BUF_TRYLOCK) {
1573 		if (!XFS_BUF_CPSEMA(bp)) {
1574 			return NULL;
1575 		}
1576 	} else {
1577 		XFS_BUF_PSEMA(bp, PRIBIO);
1578 	}
1579 	XFS_BUF_HOLD(bp);
1580 	ASSERT(XFS_BUF_ISDONE(bp));
1581 	return bp;
1582 }
1583 
1584 /*
1585  * Used to free the superblock along various error paths.
1586  */
1587 void
1588 xfs_freesb(
1589 	xfs_mount_t	*mp)
1590 {
1591 	xfs_buf_t	*bp;
1592 
1593 	/*
1594 	 * Use xfs_getsb() so that the buffer will be locked
1595 	 * when we call xfs_buf_relse().
1596 	 */
1597 	bp = xfs_getsb(mp, 0);
1598 	XFS_BUF_UNMANAGE(bp);
1599 	xfs_buf_relse(bp);
1600 	mp->m_sb_bp = NULL;
1601 }
1602 
1603 /*
1604  * See if the UUID is unique among mounted XFS filesystems.
1605  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1606  */
1607 STATIC int
1608 xfs_uuid_mount(
1609 	xfs_mount_t	*mp)
1610 {
1611 	if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1612 		cmn_err(CE_WARN,
1613 			"XFS: Filesystem %s has nil UUID - can't mount",
1614 			mp->m_fsname);
1615 		return -1;
1616 	}
1617 	if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1618 		cmn_err(CE_WARN,
1619 			"XFS: Filesystem %s has duplicate UUID - can't mount",
1620 			mp->m_fsname);
1621 		return -1;
1622 	}
1623 	return 0;
1624 }
1625 
1626 /*
1627  * Remove filesystem from the UUID table.
1628  */
1629 STATIC void
1630 xfs_uuid_unmount(
1631 	xfs_mount_t	*mp)
1632 {
1633 	uuid_table_remove(&mp->m_sb.sb_uuid);
1634 }
1635 
1636 /*
1637  * Used to log changes to the superblock unit and width fields which could
1638  * be altered by the mount options. Only the first superblock is updated.
1639  */
1640 STATIC void
1641 xfs_mount_log_sbunit(
1642 	xfs_mount_t	*mp,
1643 	__int64_t	fields)
1644 {
1645 	xfs_trans_t	*tp;
1646 
1647 	ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1648 
1649 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1650 	if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1651 				XFS_DEFAULT_LOG_COUNT)) {
1652 		xfs_trans_cancel(tp, 0);
1653 		return;
1654 	}
1655 	xfs_mod_sb(tp, fields);
1656 	xfs_trans_commit(tp, 0);
1657 }
1658 
1659 
1660 #ifdef HAVE_PERCPU_SB
1661 /*
1662  * Per-cpu incore superblock counters
1663  *
1664  * Simple concept, difficult implementation
1665  *
1666  * Basically, replace the incore superblock counters with a distributed per cpu
1667  * counter for contended fields (e.g.  free block count).
1668  *
1669  * Difficulties arise in that the incore sb is used for ENOSPC checking, and
1670  * hence needs to be accurately read when we are running low on space. Hence
1671  * there is a method to enable and disable the per-cpu counters based on how
1672  * much "stuff" is available in them.
1673  *
1674  * Basically, a counter is enabled if there is enough free resource to justify
1675  * running a per-cpu fast-path. If the per-cpu counter runs out (i.e. a local
1676  * ENOSPC), then we disable the counters to synchronise all callers and
1677  * re-distribute the available resources.
1678  *
1679  * If, once we redistributed the available resources, we still get a failure,
1680  * we disable the per-cpu counter and go through the slow path.
1681  *
1682  * The slow path is the current xfs_mod_incore_sb() function.  This means that
1683  * when we disable a per-cpu counter, we need to drain it's resources back to
1684  * the global superblock. We do this after disabling the counter to prevent
1685  * more threads from queueing up on the counter.
1686  *
1687  * Essentially, this means that we still need a lock in the fast path to enable
1688  * synchronisation between the global counters and the per-cpu counters. This
1689  * is not a problem because the lock will be local to a CPU almost all the time
1690  * and have little contention except when we get to ENOSPC conditions.
1691  *
1692  * Basically, this lock becomes a barrier that enables us to lock out the fast
1693  * path while we do things like enabling and disabling counters and
1694  * synchronising the counters.
1695  *
1696  * Locking rules:
1697  *
1698  * 	1. XFS_SB_LOCK() before picking up per-cpu locks
1699  * 	2. per-cpu locks always picked up via for_each_online_cpu() order
1700  * 	3. accurate counter sync requires XFS_SB_LOCK + per cpu locks
1701  * 	4. modifying per-cpu counters requires holding per-cpu lock
1702  * 	5. modifying global counters requires holding XFS_SB_LOCK
1703  *	6. enabling or disabling a counter requires holding the XFS_SB_LOCK
1704  *	   and _none_ of the per-cpu locks.
1705  *
1706  * Disabled counters are only ever re-enabled by a balance operation
1707  * that results in more free resources per CPU than a given threshold.
1708  * To ensure counters don't remain disabled, they are rebalanced when
1709  * the global resource goes above a higher threshold (i.e. some hysteresis
1710  * is present to prevent thrashing).
1711  */
1712 
1713 #ifdef CONFIG_HOTPLUG_CPU
1714 /*
1715  * hot-plug CPU notifier support.
1716  *
1717  * We need a notifier per filesystem as we need to be able to identify
1718  * the filesystem to balance the counters out. This is achieved by
1719  * having a notifier block embedded in the xfs_mount_t and doing pointer
1720  * magic to get the mount pointer from the notifier block address.
1721  */
1722 STATIC int
1723 xfs_icsb_cpu_notify(
1724 	struct notifier_block *nfb,
1725 	unsigned long action,
1726 	void *hcpu)
1727 {
1728 	xfs_icsb_cnts_t *cntp;
1729 	xfs_mount_t	*mp;
1730 	int		s;
1731 
1732 	mp = (xfs_mount_t *)container_of(nfb, xfs_mount_t, m_icsb_notifier);
1733 	cntp = (xfs_icsb_cnts_t *)
1734 			per_cpu_ptr(mp->m_sb_cnts, (unsigned long)hcpu);
1735 	switch (action) {
1736 	case CPU_UP_PREPARE:
1737 	case CPU_UP_PREPARE_FROZEN:
1738 		/* Easy Case - initialize the area and locks, and
1739 		 * then rebalance when online does everything else for us. */
1740 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1741 		break;
1742 	case CPU_ONLINE:
1743 	case CPU_ONLINE_FROZEN:
1744 		xfs_icsb_lock(mp);
1745 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
1746 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
1747 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
1748 		xfs_icsb_unlock(mp);
1749 		break;
1750 	case CPU_DEAD:
1751 	case CPU_DEAD_FROZEN:
1752 		/* Disable all the counters, then fold the dead cpu's
1753 		 * count into the total on the global superblock and
1754 		 * re-enable the counters. */
1755 		xfs_icsb_lock(mp);
1756 		s = XFS_SB_LOCK(mp);
1757 		xfs_icsb_disable_counter(mp, XFS_SBS_ICOUNT);
1758 		xfs_icsb_disable_counter(mp, XFS_SBS_IFREE);
1759 		xfs_icsb_disable_counter(mp, XFS_SBS_FDBLOCKS);
1760 
1761 		mp->m_sb.sb_icount += cntp->icsb_icount;
1762 		mp->m_sb.sb_ifree += cntp->icsb_ifree;
1763 		mp->m_sb.sb_fdblocks += cntp->icsb_fdblocks;
1764 
1765 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1766 
1767 		xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT,
1768 					 XFS_ICSB_SB_LOCKED, 0);
1769 		xfs_icsb_balance_counter(mp, XFS_SBS_IFREE,
1770 					 XFS_ICSB_SB_LOCKED, 0);
1771 		xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS,
1772 					 XFS_ICSB_SB_LOCKED, 0);
1773 		XFS_SB_UNLOCK(mp, s);
1774 		xfs_icsb_unlock(mp);
1775 		break;
1776 	}
1777 
1778 	return NOTIFY_OK;
1779 }
1780 #endif /* CONFIG_HOTPLUG_CPU */
1781 
1782 int
1783 xfs_icsb_init_counters(
1784 	xfs_mount_t	*mp)
1785 {
1786 	xfs_icsb_cnts_t *cntp;
1787 	int		i;
1788 
1789 	mp->m_sb_cnts = alloc_percpu(xfs_icsb_cnts_t);
1790 	if (mp->m_sb_cnts == NULL)
1791 		return -ENOMEM;
1792 
1793 #ifdef CONFIG_HOTPLUG_CPU
1794 	mp->m_icsb_notifier.notifier_call = xfs_icsb_cpu_notify;
1795 	mp->m_icsb_notifier.priority = 0;
1796 	register_hotcpu_notifier(&mp->m_icsb_notifier);
1797 #endif /* CONFIG_HOTPLUG_CPU */
1798 
1799 	for_each_online_cpu(i) {
1800 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1801 		memset(cntp, 0, sizeof(xfs_icsb_cnts_t));
1802 	}
1803 
1804 	mutex_init(&mp->m_icsb_mutex);
1805 
1806 	/*
1807 	 * start with all counters disabled so that the
1808 	 * initial balance kicks us off correctly
1809 	 */
1810 	mp->m_icsb_counters = -1;
1811 	return 0;
1812 }
1813 
1814 void
1815 xfs_icsb_reinit_counters(
1816 	xfs_mount_t	*mp)
1817 {
1818 	xfs_icsb_lock(mp);
1819 	/*
1820 	 * start with all counters disabled so that the
1821 	 * initial balance kicks us off correctly
1822 	 */
1823 	mp->m_icsb_counters = -1;
1824 	xfs_icsb_balance_counter(mp, XFS_SBS_ICOUNT, 0, 0);
1825 	xfs_icsb_balance_counter(mp, XFS_SBS_IFREE, 0, 0);
1826 	xfs_icsb_balance_counter(mp, XFS_SBS_FDBLOCKS, 0, 0);
1827 	xfs_icsb_unlock(mp);
1828 }
1829 
1830 STATIC void
1831 xfs_icsb_destroy_counters(
1832 	xfs_mount_t	*mp)
1833 {
1834 	if (mp->m_sb_cnts) {
1835 		unregister_hotcpu_notifier(&mp->m_icsb_notifier);
1836 		free_percpu(mp->m_sb_cnts);
1837 	}
1838 	mutex_destroy(&mp->m_icsb_mutex);
1839 }
1840 
1841 STATIC_INLINE void
1842 xfs_icsb_lock_cntr(
1843 	xfs_icsb_cnts_t	*icsbp)
1844 {
1845 	while (test_and_set_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags)) {
1846 		ndelay(1000);
1847 	}
1848 }
1849 
1850 STATIC_INLINE void
1851 xfs_icsb_unlock_cntr(
1852 	xfs_icsb_cnts_t	*icsbp)
1853 {
1854 	clear_bit(XFS_ICSB_FLAG_LOCK, &icsbp->icsb_flags);
1855 }
1856 
1857 
1858 STATIC_INLINE void
1859 xfs_icsb_lock_all_counters(
1860 	xfs_mount_t	*mp)
1861 {
1862 	xfs_icsb_cnts_t *cntp;
1863 	int		i;
1864 
1865 	for_each_online_cpu(i) {
1866 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1867 		xfs_icsb_lock_cntr(cntp);
1868 	}
1869 }
1870 
1871 STATIC_INLINE void
1872 xfs_icsb_unlock_all_counters(
1873 	xfs_mount_t	*mp)
1874 {
1875 	xfs_icsb_cnts_t *cntp;
1876 	int		i;
1877 
1878 	for_each_online_cpu(i) {
1879 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1880 		xfs_icsb_unlock_cntr(cntp);
1881 	}
1882 }
1883 
1884 STATIC void
1885 xfs_icsb_count(
1886 	xfs_mount_t	*mp,
1887 	xfs_icsb_cnts_t	*cnt,
1888 	int		flags)
1889 {
1890 	xfs_icsb_cnts_t *cntp;
1891 	int		i;
1892 
1893 	memset(cnt, 0, sizeof(xfs_icsb_cnts_t));
1894 
1895 	if (!(flags & XFS_ICSB_LAZY_COUNT))
1896 		xfs_icsb_lock_all_counters(mp);
1897 
1898 	for_each_online_cpu(i) {
1899 		cntp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, i);
1900 		cnt->icsb_icount += cntp->icsb_icount;
1901 		cnt->icsb_ifree += cntp->icsb_ifree;
1902 		cnt->icsb_fdblocks += cntp->icsb_fdblocks;
1903 	}
1904 
1905 	if (!(flags & XFS_ICSB_LAZY_COUNT))
1906 		xfs_icsb_unlock_all_counters(mp);
1907 }
1908 
1909 STATIC int
1910 xfs_icsb_counter_disabled(
1911 	xfs_mount_t	*mp,
1912 	xfs_sb_field_t	field)
1913 {
1914 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1915 	return test_bit(field, &mp->m_icsb_counters);
1916 }
1917 
1918 STATIC int
1919 xfs_icsb_disable_counter(
1920 	xfs_mount_t	*mp,
1921 	xfs_sb_field_t	field)
1922 {
1923 	xfs_icsb_cnts_t	cnt;
1924 
1925 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1926 
1927 	/*
1928 	 * If we are already disabled, then there is nothing to do
1929 	 * here. We check before locking all the counters to avoid
1930 	 * the expensive lock operation when being called in the
1931 	 * slow path and the counter is already disabled. This is
1932 	 * safe because the only time we set or clear this state is under
1933 	 * the m_icsb_mutex.
1934 	 */
1935 	if (xfs_icsb_counter_disabled(mp, field))
1936 		return 0;
1937 
1938 	xfs_icsb_lock_all_counters(mp);
1939 	if (!test_and_set_bit(field, &mp->m_icsb_counters)) {
1940 		/* drain back to superblock */
1941 
1942 		xfs_icsb_count(mp, &cnt, XFS_ICSB_SB_LOCKED|XFS_ICSB_LAZY_COUNT);
1943 		switch(field) {
1944 		case XFS_SBS_ICOUNT:
1945 			mp->m_sb.sb_icount = cnt.icsb_icount;
1946 			break;
1947 		case XFS_SBS_IFREE:
1948 			mp->m_sb.sb_ifree = cnt.icsb_ifree;
1949 			break;
1950 		case XFS_SBS_FDBLOCKS:
1951 			mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
1952 			break;
1953 		default:
1954 			BUG();
1955 		}
1956 	}
1957 
1958 	xfs_icsb_unlock_all_counters(mp);
1959 
1960 	return 0;
1961 }
1962 
1963 STATIC void
1964 xfs_icsb_enable_counter(
1965 	xfs_mount_t	*mp,
1966 	xfs_sb_field_t	field,
1967 	uint64_t	count,
1968 	uint64_t	resid)
1969 {
1970 	xfs_icsb_cnts_t	*cntp;
1971 	int		i;
1972 
1973 	ASSERT((field >= XFS_SBS_ICOUNT) && (field <= XFS_SBS_FDBLOCKS));
1974 
1975 	xfs_icsb_lock_all_counters(mp);
1976 	for_each_online_cpu(i) {
1977 		cntp = per_cpu_ptr(mp->m_sb_cnts, i);
1978 		switch (field) {
1979 		case XFS_SBS_ICOUNT:
1980 			cntp->icsb_icount = count + resid;
1981 			break;
1982 		case XFS_SBS_IFREE:
1983 			cntp->icsb_ifree = count + resid;
1984 			break;
1985 		case XFS_SBS_FDBLOCKS:
1986 			cntp->icsb_fdblocks = count + resid;
1987 			break;
1988 		default:
1989 			BUG();
1990 			break;
1991 		}
1992 		resid = 0;
1993 	}
1994 	clear_bit(field, &mp->m_icsb_counters);
1995 	xfs_icsb_unlock_all_counters(mp);
1996 }
1997 
1998 void
1999 xfs_icsb_sync_counters_flags(
2000 	xfs_mount_t	*mp,
2001 	int		flags)
2002 {
2003 	xfs_icsb_cnts_t	cnt;
2004 	int		s;
2005 
2006 	/* Pass 1: lock all counters */
2007 	if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2008 		s = XFS_SB_LOCK(mp);
2009 
2010 	xfs_icsb_count(mp, &cnt, flags);
2011 
2012 	/* Step 3: update mp->m_sb fields */
2013 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_ICOUNT))
2014 		mp->m_sb.sb_icount = cnt.icsb_icount;
2015 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_IFREE))
2016 		mp->m_sb.sb_ifree = cnt.icsb_ifree;
2017 	if (!xfs_icsb_counter_disabled(mp, XFS_SBS_FDBLOCKS))
2018 		mp->m_sb.sb_fdblocks = cnt.icsb_fdblocks;
2019 
2020 	if ((flags & XFS_ICSB_SB_LOCKED) == 0)
2021 		XFS_SB_UNLOCK(mp, s);
2022 }
2023 
2024 /*
2025  * Accurate update of per-cpu counters to incore superblock
2026  */
2027 STATIC void
2028 xfs_icsb_sync_counters(
2029 	xfs_mount_t	*mp)
2030 {
2031 	xfs_icsb_sync_counters_flags(mp, 0);
2032 }
2033 
2034 /*
2035  * Balance and enable/disable counters as necessary.
2036  *
2037  * Thresholds for re-enabling counters are somewhat magic.  inode counts are
2038  * chosen to be the same number as single on disk allocation chunk per CPU, and
2039  * free blocks is something far enough zero that we aren't going thrash when we
2040  * get near ENOSPC. We also need to supply a minimum we require per cpu to
2041  * prevent looping endlessly when xfs_alloc_space asks for more than will
2042  * be distributed to a single CPU but each CPU has enough blocks to be
2043  * reenabled.
2044  *
2045  * Note that we can be called when counters are already disabled.
2046  * xfs_icsb_disable_counter() optimises the counter locking in this case to
2047  * prevent locking every per-cpu counter needlessly.
2048  */
2049 
2050 #define XFS_ICSB_INO_CNTR_REENABLE	(uint64_t)64
2051 #define XFS_ICSB_FDBLK_CNTR_REENABLE(mp) \
2052 		(uint64_t)(512 + XFS_ALLOC_SET_ASIDE(mp))
2053 STATIC void
2054 xfs_icsb_balance_counter(
2055 	xfs_mount_t	*mp,
2056 	xfs_sb_field_t  field,
2057 	int		flags,
2058 	int		min_per_cpu)
2059 {
2060 	uint64_t	count, resid;
2061 	int		weight = num_online_cpus();
2062 	int		s;
2063 	uint64_t	min = (uint64_t)min_per_cpu;
2064 
2065 	if (!(flags & XFS_ICSB_SB_LOCKED))
2066 		s = XFS_SB_LOCK(mp);
2067 
2068 	/* disable counter and sync counter */
2069 	xfs_icsb_disable_counter(mp, field);
2070 
2071 	/* update counters  - first CPU gets residual*/
2072 	switch (field) {
2073 	case XFS_SBS_ICOUNT:
2074 		count = mp->m_sb.sb_icount;
2075 		resid = do_div(count, weight);
2076 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2077 			goto out;
2078 		break;
2079 	case XFS_SBS_IFREE:
2080 		count = mp->m_sb.sb_ifree;
2081 		resid = do_div(count, weight);
2082 		if (count < max(min, XFS_ICSB_INO_CNTR_REENABLE))
2083 			goto out;
2084 		break;
2085 	case XFS_SBS_FDBLOCKS:
2086 		count = mp->m_sb.sb_fdblocks;
2087 		resid = do_div(count, weight);
2088 		if (count < max(min, XFS_ICSB_FDBLK_CNTR_REENABLE(mp)))
2089 			goto out;
2090 		break;
2091 	default:
2092 		BUG();
2093 		count = resid = 0;	/* quiet, gcc */
2094 		break;
2095 	}
2096 
2097 	xfs_icsb_enable_counter(mp, field, count, resid);
2098 out:
2099 	if (!(flags & XFS_ICSB_SB_LOCKED))
2100 		XFS_SB_UNLOCK(mp, s);
2101 }
2102 
2103 int
2104 xfs_icsb_modify_counters(
2105 	xfs_mount_t	*mp,
2106 	xfs_sb_field_t	field,
2107 	int64_t		delta,
2108 	int		rsvd)
2109 {
2110 	xfs_icsb_cnts_t	*icsbp;
2111 	long long	lcounter;	/* long counter for 64 bit fields */
2112 	int		cpu, ret = 0, s;
2113 
2114 	might_sleep();
2115 again:
2116 	cpu = get_cpu();
2117 	icsbp = (xfs_icsb_cnts_t *)per_cpu_ptr(mp->m_sb_cnts, cpu);
2118 
2119 	/*
2120 	 * if the counter is disabled, go to slow path
2121 	 */
2122 	if (unlikely(xfs_icsb_counter_disabled(mp, field)))
2123 		goto slow_path;
2124 	xfs_icsb_lock_cntr(icsbp);
2125 	if (unlikely(xfs_icsb_counter_disabled(mp, field))) {
2126 		xfs_icsb_unlock_cntr(icsbp);
2127 		goto slow_path;
2128 	}
2129 
2130 	switch (field) {
2131 	case XFS_SBS_ICOUNT:
2132 		lcounter = icsbp->icsb_icount;
2133 		lcounter += delta;
2134 		if (unlikely(lcounter < 0))
2135 			goto balance_counter;
2136 		icsbp->icsb_icount = lcounter;
2137 		break;
2138 
2139 	case XFS_SBS_IFREE:
2140 		lcounter = icsbp->icsb_ifree;
2141 		lcounter += delta;
2142 		if (unlikely(lcounter < 0))
2143 			goto balance_counter;
2144 		icsbp->icsb_ifree = lcounter;
2145 		break;
2146 
2147 	case XFS_SBS_FDBLOCKS:
2148 		BUG_ON((mp->m_resblks - mp->m_resblks_avail) != 0);
2149 
2150 		lcounter = icsbp->icsb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
2151 		lcounter += delta;
2152 		if (unlikely(lcounter < 0))
2153 			goto balance_counter;
2154 		icsbp->icsb_fdblocks = lcounter + XFS_ALLOC_SET_ASIDE(mp);
2155 		break;
2156 	default:
2157 		BUG();
2158 		break;
2159 	}
2160 	xfs_icsb_unlock_cntr(icsbp);
2161 	put_cpu();
2162 	return 0;
2163 
2164 slow_path:
2165 	put_cpu();
2166 
2167 	/*
2168 	 * serialise with a mutex so we don't burn lots of cpu on
2169 	 * the superblock lock. We still need to hold the superblock
2170 	 * lock, however, when we modify the global structures.
2171 	 */
2172 	xfs_icsb_lock(mp);
2173 
2174 	/*
2175 	 * Now running atomically.
2176 	 *
2177 	 * If the counter is enabled, someone has beaten us to rebalancing.
2178 	 * Drop the lock and try again in the fast path....
2179 	 */
2180 	if (!(xfs_icsb_counter_disabled(mp, field))) {
2181 		xfs_icsb_unlock(mp);
2182 		goto again;
2183 	}
2184 
2185 	/*
2186 	 * The counter is currently disabled. Because we are
2187 	 * running atomically here, we know a rebalance cannot
2188 	 * be in progress. Hence we can go straight to operating
2189 	 * on the global superblock. We do not call xfs_mod_incore_sb()
2190 	 * here even though we need to get the SB_LOCK. Doing so
2191 	 * will cause us to re-enter this function and deadlock.
2192 	 * Hence we get the SB_LOCK ourselves and then call
2193 	 * xfs_mod_incore_sb_unlocked() as the unlocked path operates
2194 	 * directly on the global counters.
2195 	 */
2196 	s = XFS_SB_LOCK(mp);
2197 	ret = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
2198 	XFS_SB_UNLOCK(mp, s);
2199 
2200 	/*
2201 	 * Now that we've modified the global superblock, we
2202 	 * may be able to re-enable the distributed counters
2203 	 * (e.g. lots of space just got freed). After that
2204 	 * we are done.
2205 	 */
2206 	if (ret != ENOSPC)
2207 		xfs_icsb_balance_counter(mp, field, 0, 0);
2208 	xfs_icsb_unlock(mp);
2209 	return ret;
2210 
2211 balance_counter:
2212 	xfs_icsb_unlock_cntr(icsbp);
2213 	put_cpu();
2214 
2215 	/*
2216 	 * We may have multiple threads here if multiple per-cpu
2217 	 * counters run dry at the same time. This will mean we can
2218 	 * do more balances than strictly necessary but it is not
2219 	 * the common slowpath case.
2220 	 */
2221 	xfs_icsb_lock(mp);
2222 
2223 	/*
2224 	 * running atomically.
2225 	 *
2226 	 * This will leave the counter in the correct state for future
2227 	 * accesses. After the rebalance, we simply try again and our retry
2228 	 * will either succeed through the fast path or slow path without
2229 	 * another balance operation being required.
2230 	 */
2231 	xfs_icsb_balance_counter(mp, field, 0, delta);
2232 	xfs_icsb_unlock(mp);
2233 	goto again;
2234 }
2235 
2236 #endif
2237