xref: /linux/fs/xfs/xfs_mount.c (revision d67b569f5f620c0fb95d5212642746b7ba9d29e4)
1 /*
2  * Copyright (c) 2000-2004 Silicon Graphics, Inc.  All Rights Reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it would be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11  *
12  * Further, this software is distributed without any warranty that it is
13  * free of the rightful claim of any third person regarding infringement
14  * or the like.  Any license provided herein, whether implied or
15  * otherwise, applies only to this software file.  Patent licenses, if
16  * any, provided herein do not apply to combinations of this program with
17  * other software, or any other product whatsoever.
18  *
19  * You should have received a copy of the GNU General Public License along
20  * with this program; if not, write the Free Software Foundation, Inc., 59
21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24  * Mountain View, CA  94043, or:
25  *
26  * http://www.sgi.com
27  *
28  * For further information regarding this notice, see:
29  *
30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31  */
32 
33 #include "xfs.h"
34 #include "xfs_macros.h"
35 #include "xfs_types.h"
36 #include "xfs_inum.h"
37 #include "xfs_log.h"
38 #include "xfs_trans.h"
39 #include "xfs_sb.h"
40 #include "xfs_ag.h"
41 #include "xfs_dir.h"
42 #include "xfs_dir2.h"
43 #include "xfs_dmapi.h"
44 #include "xfs_mount.h"
45 #include "xfs_alloc_btree.h"
46 #include "xfs_bmap_btree.h"
47 #include "xfs_ialloc_btree.h"
48 #include "xfs_btree.h"
49 #include "xfs_ialloc.h"
50 #include "xfs_attr_sf.h"
51 #include "xfs_dir_sf.h"
52 #include "xfs_dir2_sf.h"
53 #include "xfs_dinode.h"
54 #include "xfs_inode.h"
55 #include "xfs_alloc.h"
56 #include "xfs_rtalloc.h"
57 #include "xfs_bmap.h"
58 #include "xfs_error.h"
59 #include "xfs_bit.h"
60 #include "xfs_rw.h"
61 #include "xfs_quota.h"
62 #include "xfs_fsops.h"
63 
64 STATIC void	xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
65 STATIC int	xfs_uuid_mount(xfs_mount_t *);
66 STATIC void	xfs_uuid_unmount(xfs_mount_t *mp);
67 STATIC void	xfs_unmountfs_wait(xfs_mount_t *);
68 
69 static struct {
70     short offset;
71     short type;     /* 0 = integer
72 		* 1 = binary / string (no translation)
73 		*/
74 } xfs_sb_info[] = {
75     { offsetof(xfs_sb_t, sb_magicnum),   0 },
76     { offsetof(xfs_sb_t, sb_blocksize),  0 },
77     { offsetof(xfs_sb_t, sb_dblocks),    0 },
78     { offsetof(xfs_sb_t, sb_rblocks),    0 },
79     { offsetof(xfs_sb_t, sb_rextents),   0 },
80     { offsetof(xfs_sb_t, sb_uuid),       1 },
81     { offsetof(xfs_sb_t, sb_logstart),   0 },
82     { offsetof(xfs_sb_t, sb_rootino),    0 },
83     { offsetof(xfs_sb_t, sb_rbmino),     0 },
84     { offsetof(xfs_sb_t, sb_rsumino),    0 },
85     { offsetof(xfs_sb_t, sb_rextsize),   0 },
86     { offsetof(xfs_sb_t, sb_agblocks),   0 },
87     { offsetof(xfs_sb_t, sb_agcount),    0 },
88     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
89     { offsetof(xfs_sb_t, sb_logblocks),  0 },
90     { offsetof(xfs_sb_t, sb_versionnum), 0 },
91     { offsetof(xfs_sb_t, sb_sectsize),   0 },
92     { offsetof(xfs_sb_t, sb_inodesize),  0 },
93     { offsetof(xfs_sb_t, sb_inopblock),  0 },
94     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
95     { offsetof(xfs_sb_t, sb_blocklog),   0 },
96     { offsetof(xfs_sb_t, sb_sectlog),    0 },
97     { offsetof(xfs_sb_t, sb_inodelog),   0 },
98     { offsetof(xfs_sb_t, sb_inopblog),   0 },
99     { offsetof(xfs_sb_t, sb_agblklog),   0 },
100     { offsetof(xfs_sb_t, sb_rextslog),   0 },
101     { offsetof(xfs_sb_t, sb_inprogress), 0 },
102     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
103     { offsetof(xfs_sb_t, sb_icount),     0 },
104     { offsetof(xfs_sb_t, sb_ifree),      0 },
105     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
106     { offsetof(xfs_sb_t, sb_frextents),  0 },
107     { offsetof(xfs_sb_t, sb_uquotino),   0 },
108     { offsetof(xfs_sb_t, sb_gquotino),   0 },
109     { offsetof(xfs_sb_t, sb_qflags),     0 },
110     { offsetof(xfs_sb_t, sb_flags),      0 },
111     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
112     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
113     { offsetof(xfs_sb_t, sb_unit),	 0 },
114     { offsetof(xfs_sb_t, sb_width),	 0 },
115     { offsetof(xfs_sb_t, sb_dirblklog),	 0 },
116     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
117     { offsetof(xfs_sb_t, sb_logsectsize),0 },
118     { offsetof(xfs_sb_t, sb_logsunit),	 0 },
119     { offsetof(xfs_sb_t, sb_features2),	 0 },
120     { sizeof(xfs_sb_t),			 0 }
121 };
122 
123 /*
124  * Return a pointer to an initialized xfs_mount structure.
125  */
126 xfs_mount_t *
127 xfs_mount_init(void)
128 {
129 	xfs_mount_t *mp;
130 
131 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
132 
133 	AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
134 	spinlock_init(&mp->m_sb_lock, "xfs_sb");
135 	mutex_init(&mp->m_ilock, MUTEX_DEFAULT, "xfs_ilock");
136 	initnsema(&mp->m_growlock, 1, "xfs_grow");
137 	/*
138 	 * Initialize the AIL.
139 	 */
140 	xfs_trans_ail_init(mp);
141 
142 	atomic_set(&mp->m_active_trans, 0);
143 
144 	return mp;
145 }
146 
147 /*
148  * Free up the resources associated with a mount structure.  Assume that
149  * the structure was initially zeroed, so we can tell which fields got
150  * initialized.
151  */
152 void
153 xfs_mount_free(
154 	xfs_mount_t *mp,
155 	int	    remove_bhv)
156 {
157 	if (mp->m_ihash)
158 		xfs_ihash_free(mp);
159 	if (mp->m_chash)
160 		xfs_chash_free(mp);
161 
162 	if (mp->m_perag) {
163 		int	agno;
164 
165 		for (agno = 0; agno < mp->m_maxagi; agno++)
166 			if (mp->m_perag[agno].pagb_list)
167 				kmem_free(mp->m_perag[agno].pagb_list,
168 						sizeof(xfs_perag_busy_t) *
169 							XFS_PAGB_NUM_SLOTS);
170 		kmem_free(mp->m_perag,
171 			  sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
172 	}
173 
174 	AIL_LOCK_DESTROY(&mp->m_ail_lock);
175 	spinlock_destroy(&mp->m_sb_lock);
176 	mutex_destroy(&mp->m_ilock);
177 	freesema(&mp->m_growlock);
178 	if (mp->m_quotainfo)
179 		XFS_QM_DONE(mp);
180 
181 	if (mp->m_fsname != NULL)
182 		kmem_free(mp->m_fsname, mp->m_fsname_len);
183 
184 	if (remove_bhv) {
185 		struct vfs	*vfsp = XFS_MTOVFS(mp);
186 
187 		bhv_remove_all_vfsops(vfsp, 0);
188 		VFS_REMOVEBHV(vfsp, &mp->m_bhv);
189 	}
190 
191 	kmem_free(mp, sizeof(xfs_mount_t));
192 }
193 
194 
195 /*
196  * Check the validity of the SB found.
197  */
198 STATIC int
199 xfs_mount_validate_sb(
200 	xfs_mount_t	*mp,
201 	xfs_sb_t	*sbp)
202 {
203 	/*
204 	 * If the log device and data device have the
205 	 * same device number, the log is internal.
206 	 * Consequently, the sb_logstart should be non-zero.  If
207 	 * we have a zero sb_logstart in this case, we may be trying to mount
208 	 * a volume filesystem in a non-volume manner.
209 	 */
210 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
211 		cmn_err(CE_WARN, "XFS: bad magic number");
212 		return XFS_ERROR(EWRONGFS);
213 	}
214 
215 	if (!XFS_SB_GOOD_VERSION(sbp)) {
216 		cmn_err(CE_WARN, "XFS: bad version");
217 		return XFS_ERROR(EWRONGFS);
218 	}
219 
220 	if (unlikely(
221 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
222 		cmn_err(CE_WARN,
223 	"XFS: filesystem is marked as having an external log; "
224 	"specify logdev on the\nmount command line.");
225 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
226 				     XFS_ERRLEVEL_HIGH, mp, sbp);
227 		return XFS_ERROR(EFSCORRUPTED);
228 	}
229 
230 	if (unlikely(
231 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
232 		cmn_err(CE_WARN,
233 	"XFS: filesystem is marked as having an internal log; "
234 	"don't specify logdev on\nthe mount command line.");
235 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
236 				     XFS_ERRLEVEL_HIGH, mp, sbp);
237 		return XFS_ERROR(EFSCORRUPTED);
238 	}
239 
240 	/*
241 	 * More sanity checking. These were stolen directly from
242 	 * xfs_repair.
243 	 */
244 	if (unlikely(
245 	    sbp->sb_agcount <= 0					||
246 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
247 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
248 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
249 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
250 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
251 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
252 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
253 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
254 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
255 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
256 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
257 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
258 	    sbp->sb_imax_pct > 100)) {
259 		cmn_err(CE_WARN, "XFS: SB sanity check 1 failed");
260 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
261 				     XFS_ERRLEVEL_LOW, mp, sbp);
262 		return XFS_ERROR(EFSCORRUPTED);
263 	}
264 
265 	/*
266 	 * Sanity check AG count, size fields against data size field
267 	 */
268 	if (unlikely(
269 	    sbp->sb_dblocks == 0 ||
270 	    sbp->sb_dblocks >
271 	     (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
272 	    sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
273 			      sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
274 		cmn_err(CE_WARN, "XFS: SB sanity check 2 failed");
275 		XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
276 				 XFS_ERRLEVEL_LOW, mp);
277 		return XFS_ERROR(EFSCORRUPTED);
278 	}
279 
280 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
281 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
282 
283 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
284 	if (unlikely(
285 	    (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
286 	    (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
287 #else                  /* Limited by UINT_MAX of sectors */
288 	if (unlikely(
289 	    (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
290 	    (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
291 #endif
292 		cmn_err(CE_WARN,
293 	"XFS: File system is too large to be mounted on this system.");
294 		return XFS_ERROR(E2BIG);
295 	}
296 
297 	if (unlikely(sbp->sb_inprogress)) {
298 		cmn_err(CE_WARN, "XFS: file system busy");
299 		XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
300 				 XFS_ERRLEVEL_LOW, mp);
301 		return XFS_ERROR(EFSCORRUPTED);
302 	}
303 
304 	/*
305 	 * Version 1 directory format has never worked on Linux.
306 	 */
307 	if (unlikely(!XFS_SB_VERSION_HASDIRV2(sbp))) {
308 		cmn_err(CE_WARN,
309 	"XFS: Attempted to mount file system using version 1 directory format");
310 		return XFS_ERROR(ENOSYS);
311 	}
312 
313 	/*
314 	 * Until this is fixed only page-sized or smaller data blocks work.
315 	 */
316 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
317 		cmn_err(CE_WARN,
318 		"XFS: Attempted to mount file system with blocksize %d bytes",
319 			sbp->sb_blocksize);
320 		cmn_err(CE_WARN,
321 		"XFS: Only page-sized (%d) or less blocksizes currently work.",
322 			PAGE_SIZE);
323 		return XFS_ERROR(ENOSYS);
324 	}
325 
326 	return 0;
327 }
328 
329 xfs_agnumber_t
330 xfs_initialize_perag(xfs_mount_t *mp, xfs_agnumber_t agcount)
331 {
332 	xfs_agnumber_t	index, max_metadata;
333 	xfs_perag_t	*pag;
334 	xfs_agino_t	agino;
335 	xfs_ino_t	ino;
336 	xfs_sb_t	*sbp = &mp->m_sb;
337 	xfs_ino_t	max_inum = XFS_MAXINUMBER_32;
338 
339 	/* Check to see if the filesystem can overflow 32 bit inodes */
340 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
341 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
342 
343 	/* Clear the mount flag if no inode can overflow 32 bits
344 	 * on this filesystem, or if specifically requested..
345 	 */
346 	if ((mp->m_flags & XFS_MOUNT_32BITINOOPT) && ino > max_inum) {
347 		mp->m_flags |= XFS_MOUNT_32BITINODES;
348 	} else {
349 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
350 	}
351 
352 	/* If we can overflow then setup the ag headers accordingly */
353 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
354 		/* Calculate how much should be reserved for inodes to
355 		 * meet the max inode percentage.
356 		 */
357 		if (mp->m_maxicount) {
358 			__uint64_t	icount;
359 
360 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
361 			do_div(icount, 100);
362 			icount += sbp->sb_agblocks - 1;
363 			do_div(icount, mp->m_ialloc_blks);
364 			max_metadata = icount;
365 		} else {
366 			max_metadata = agcount;
367 		}
368 		for (index = 0; index < agcount; index++) {
369 			ino = XFS_AGINO_TO_INO(mp, index, agino);
370 			if (ino > max_inum) {
371 				index++;
372 				break;
373 			}
374 
375 			/* This ag is prefered for inodes */
376 			pag = &mp->m_perag[index];
377 			pag->pagi_inodeok = 1;
378 			if (index < max_metadata)
379 				pag->pagf_metadata = 1;
380 		}
381 	} else {
382 		/* Setup default behavior for smaller filesystems */
383 		for (index = 0; index < agcount; index++) {
384 			pag = &mp->m_perag[index];
385 			pag->pagi_inodeok = 1;
386 		}
387 	}
388 	return index;
389 }
390 
391 /*
392  * xfs_xlatesb
393  *
394  *     data       - on disk version of sb
395  *     sb         - a superblock
396  *     dir        - conversion direction: <0 - convert sb to buf
397  *                                        >0 - convert buf to sb
398  *     fields     - which fields to copy (bitmask)
399  */
400 void
401 xfs_xlatesb(
402 	void		*data,
403 	xfs_sb_t	*sb,
404 	int		dir,
405 	__int64_t	fields)
406 {
407 	xfs_caddr_t	buf_ptr;
408 	xfs_caddr_t	mem_ptr;
409 	xfs_sb_field_t	f;
410 	int		first;
411 	int		size;
412 
413 	ASSERT(dir);
414 	ASSERT(fields);
415 
416 	if (!fields)
417 		return;
418 
419 	buf_ptr = (xfs_caddr_t)data;
420 	mem_ptr = (xfs_caddr_t)sb;
421 
422 	while (fields) {
423 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
424 		first = xfs_sb_info[f].offset;
425 		size = xfs_sb_info[f + 1].offset - first;
426 
427 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
428 
429 		if (size == 1 || xfs_sb_info[f].type == 1) {
430 			if (dir > 0) {
431 				memcpy(mem_ptr + first, buf_ptr + first, size);
432 			} else {
433 				memcpy(buf_ptr + first, mem_ptr + first, size);
434 			}
435 		} else {
436 			switch (size) {
437 			case 2:
438 				INT_XLATE(*(__uint16_t*)(buf_ptr+first),
439 					  *(__uint16_t*)(mem_ptr+first),
440 					  dir, ARCH_CONVERT);
441 				break;
442 			case 4:
443 				INT_XLATE(*(__uint32_t*)(buf_ptr+first),
444 					  *(__uint32_t*)(mem_ptr+first),
445 					  dir, ARCH_CONVERT);
446 				break;
447 			case 8:
448 				INT_XLATE(*(__uint64_t*)(buf_ptr+first),
449 					  *(__uint64_t*)(mem_ptr+first), dir, ARCH_CONVERT);
450 				break;
451 			default:
452 				ASSERT(0);
453 			}
454 		}
455 
456 		fields &= ~(1LL << f);
457 	}
458 }
459 
460 /*
461  * xfs_readsb
462  *
463  * Does the initial read of the superblock.
464  */
465 int
466 xfs_readsb(xfs_mount_t *mp)
467 {
468 	unsigned int	sector_size;
469 	unsigned int	extra_flags;
470 	xfs_buf_t	*bp;
471 	xfs_sb_t	*sbp;
472 	int		error;
473 
474 	ASSERT(mp->m_sb_bp == NULL);
475 	ASSERT(mp->m_ddev_targp != NULL);
476 
477 	/*
478 	 * Allocate a (locked) buffer to hold the superblock.
479 	 * This will be kept around at all times to optimize
480 	 * access to the superblock.
481 	 */
482 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
483 	extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
484 
485 	bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
486 				BTOBB(sector_size), extra_flags);
487 	if (!bp || XFS_BUF_ISERROR(bp)) {
488 		cmn_err(CE_WARN, "XFS: SB read failed");
489 		error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
490 		goto fail;
491 	}
492 	ASSERT(XFS_BUF_ISBUSY(bp));
493 	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
494 
495 	/*
496 	 * Initialize the mount structure from the superblock.
497 	 * But first do some basic consistency checking.
498 	 */
499 	sbp = XFS_BUF_TO_SBP(bp);
500 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1, XFS_SB_ALL_BITS);
501 
502 	error = xfs_mount_validate_sb(mp, &(mp->m_sb));
503 	if (error) {
504 		cmn_err(CE_WARN, "XFS: SB validate failed");
505 		goto fail;
506 	}
507 
508 	/*
509 	 * We must be able to do sector-sized and sector-aligned IO.
510 	 */
511 	if (sector_size > mp->m_sb.sb_sectsize) {
512 		cmn_err(CE_WARN,
513 			"XFS: device supports only %u byte sectors (not %u)",
514 			sector_size, mp->m_sb.sb_sectsize);
515 		error = ENOSYS;
516 		goto fail;
517 	}
518 
519 	/*
520 	 * If device sector size is smaller than the superblock size,
521 	 * re-read the superblock so the buffer is correctly sized.
522 	 */
523 	if (sector_size < mp->m_sb.sb_sectsize) {
524 		XFS_BUF_UNMANAGE(bp);
525 		xfs_buf_relse(bp);
526 		sector_size = mp->m_sb.sb_sectsize;
527 		bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
528 					BTOBB(sector_size), extra_flags);
529 		if (!bp || XFS_BUF_ISERROR(bp)) {
530 			cmn_err(CE_WARN, "XFS: SB re-read failed");
531 			error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
532 			goto fail;
533 		}
534 		ASSERT(XFS_BUF_ISBUSY(bp));
535 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
536 	}
537 
538 	mp->m_sb_bp = bp;
539 	xfs_buf_relse(bp);
540 	ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
541 	return 0;
542 
543  fail:
544 	if (bp) {
545 		XFS_BUF_UNMANAGE(bp);
546 		xfs_buf_relse(bp);
547 	}
548 	return error;
549 }
550 
551 
552 /*
553  * xfs_mount_common
554  *
555  * Mount initialization code establishing various mount
556  * fields from the superblock associated with the given
557  * mount structure
558  */
559 STATIC void
560 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
561 {
562 	int	i;
563 
564 	mp->m_agfrotor = mp->m_agirotor = 0;
565 	spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
566 	mp->m_maxagi = mp->m_sb.sb_agcount;
567 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
568 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
569 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
570 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
571 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
572 	mp->m_litino = sbp->sb_inodesize -
573 		((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
574 	mp->m_blockmask = sbp->sb_blocksize - 1;
575 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
576 	mp->m_blockwmask = mp->m_blockwsize - 1;
577 	INIT_LIST_HEAD(&mp->m_del_inodes);
578 
579 	/*
580 	 * Setup for attributes, in case they get created.
581 	 * This value is for inodes getting attributes for the first time,
582 	 * the per-inode value is for old attribute values.
583 	 */
584 	ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
585 	switch (sbp->sb_inodesize) {
586 	case 256:
587 		mp->m_attroffset = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(2);
588 		break;
589 	case 512:
590 	case 1024:
591 	case 2048:
592 		mp->m_attroffset = XFS_BMDR_SPACE_CALC(12);
593 		break;
594 	default:
595 		ASSERT(0);
596 	}
597 	ASSERT(mp->m_attroffset < XFS_LITINO(mp));
598 
599 	for (i = 0; i < 2; i++) {
600 		mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
601 			xfs_alloc, i == 0);
602 		mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
603 			xfs_alloc, i == 0);
604 	}
605 	for (i = 0; i < 2; i++) {
606 		mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
607 			xfs_bmbt, i == 0);
608 		mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
609 			xfs_bmbt, i == 0);
610 	}
611 	for (i = 0; i < 2; i++) {
612 		mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
613 			xfs_inobt, i == 0);
614 		mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
615 			xfs_inobt, i == 0);
616 	}
617 
618 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
619 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
620 					sbp->sb_inopblock);
621 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
622 }
623 /*
624  * xfs_mountfs
625  *
626  * This function does the following on an initial mount of a file system:
627  *	- reads the superblock from disk and init the mount struct
628  *	- if we're a 32-bit kernel, do a size check on the superblock
629  *		so we don't mount terabyte filesystems
630  *	- init mount struct realtime fields
631  *	- allocate inode hash table for fs
632  *	- init directory manager
633  *	- perform recovery and init the log manager
634  */
635 int
636 xfs_mountfs(
637 	vfs_t		*vfsp,
638 	xfs_mount_t	*mp,
639 	int		mfsi_flags)
640 {
641 	xfs_buf_t	*bp;
642 	xfs_sb_t	*sbp = &(mp->m_sb);
643 	xfs_inode_t	*rip;
644 	vnode_t		*rvp = NULL;
645 	int		readio_log, writeio_log;
646 	xfs_daddr_t	d;
647 	__uint64_t	ret64;
648 	__int64_t	update_flags;
649 	uint		quotamount, quotaflags;
650 	int		agno;
651 	int		uuid_mounted = 0;
652 	int		error = 0;
653 
654 	if (mp->m_sb_bp == NULL) {
655 		if ((error = xfs_readsb(mp))) {
656 			return (error);
657 		}
658 	}
659 	xfs_mount_common(mp, sbp);
660 
661 	/*
662 	 * Check if sb_agblocks is aligned at stripe boundary
663 	 * If sb_agblocks is NOT aligned turn off m_dalign since
664 	 * allocator alignment is within an ag, therefore ag has
665 	 * to be aligned at stripe boundary.
666 	 */
667 	update_flags = 0LL;
668 	if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
669 		/*
670 		 * If stripe unit and stripe width are not multiples
671 		 * of the fs blocksize turn off alignment.
672 		 */
673 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
674 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
675 			if (mp->m_flags & XFS_MOUNT_RETERR) {
676 				cmn_err(CE_WARN,
677 					"XFS: alignment check 1 failed");
678 				error = XFS_ERROR(EINVAL);
679 				goto error1;
680 			}
681 			mp->m_dalign = mp->m_swidth = 0;
682 		} else {
683 			/*
684 			 * Convert the stripe unit and width to FSBs.
685 			 */
686 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
687 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
688 				if (mp->m_flags & XFS_MOUNT_RETERR) {
689 					error = XFS_ERROR(EINVAL);
690 					goto error1;
691 				}
692 				xfs_fs_cmn_err(CE_WARN, mp,
693 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
694 					mp->m_dalign, mp->m_swidth,
695 					sbp->sb_agblocks);
696 
697 				mp->m_dalign = 0;
698 				mp->m_swidth = 0;
699 			} else if (mp->m_dalign) {
700 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
701 			} else {
702 				if (mp->m_flags & XFS_MOUNT_RETERR) {
703 					xfs_fs_cmn_err(CE_WARN, mp,
704 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
705                                         	mp->m_dalign,
706 						mp->m_blockmask +1);
707 					error = XFS_ERROR(EINVAL);
708 					goto error1;
709 				}
710 				mp->m_swidth = 0;
711 			}
712 		}
713 
714 		/*
715 		 * Update superblock with new values
716 		 * and log changes
717 		 */
718 		if (XFS_SB_VERSION_HASDALIGN(sbp)) {
719 			if (sbp->sb_unit != mp->m_dalign) {
720 				sbp->sb_unit = mp->m_dalign;
721 				update_flags |= XFS_SB_UNIT;
722 			}
723 			if (sbp->sb_width != mp->m_swidth) {
724 				sbp->sb_width = mp->m_swidth;
725 				update_flags |= XFS_SB_WIDTH;
726 			}
727 		}
728 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
729 		    XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
730 			mp->m_dalign = sbp->sb_unit;
731 			mp->m_swidth = sbp->sb_width;
732 	}
733 
734 	xfs_alloc_compute_maxlevels(mp);
735 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
736 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
737 	xfs_ialloc_compute_maxlevels(mp);
738 
739 	if (sbp->sb_imax_pct) {
740 		__uint64_t	icount;
741 
742 		/* Make sure the maximum inode count is a multiple of the
743 		 * units we allocate inodes in.
744 		 */
745 
746 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
747 		do_div(icount, 100);
748 		do_div(icount, mp->m_ialloc_blks);
749 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
750 				   sbp->sb_inopblog;
751 	} else
752 		mp->m_maxicount = 0;
753 
754 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
755 
756 	/*
757 	 * XFS uses the uuid from the superblock as the unique
758 	 * identifier for fsid.  We can not use the uuid from the volume
759 	 * since a single partition filesystem is identical to a single
760 	 * partition volume/filesystem.
761 	 */
762 	if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
763 	    (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
764 		if (xfs_uuid_mount(mp)) {
765 			error = XFS_ERROR(EINVAL);
766 			goto error1;
767 		}
768 		uuid_mounted=1;
769 		ret64 = uuid_hash64(&sbp->sb_uuid);
770 		memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
771 	}
772 
773 	/*
774 	 * Set the default minimum read and write sizes unless
775 	 * already specified in a mount option.
776 	 * We use smaller I/O sizes when the file system
777 	 * is being used for NFS service (wsync mount option).
778 	 */
779 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
780 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
781 			readio_log = XFS_WSYNC_READIO_LOG;
782 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
783 		} else {
784 			readio_log = XFS_READIO_LOG_LARGE;
785 			writeio_log = XFS_WRITEIO_LOG_LARGE;
786 		}
787 	} else {
788 		readio_log = mp->m_readio_log;
789 		writeio_log = mp->m_writeio_log;
790 	}
791 
792 	/*
793 	 * Set the number of readahead buffers to use based on
794 	 * physical memory size.
795 	 */
796 	if (xfs_physmem <= 4096)		/* <= 16MB */
797 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
798 	else if (xfs_physmem <= 8192)	/* <= 32MB */
799 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
800 	else
801 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
802 	if (sbp->sb_blocklog > readio_log) {
803 		mp->m_readio_log = sbp->sb_blocklog;
804 	} else {
805 		mp->m_readio_log = readio_log;
806 	}
807 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
808 	if (sbp->sb_blocklog > writeio_log) {
809 		mp->m_writeio_log = sbp->sb_blocklog;
810 	} else {
811 		mp->m_writeio_log = writeio_log;
812 	}
813 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
814 
815 	/*
816 	 * Set the inode cluster size based on the physical memory
817 	 * size.  This may still be overridden by the file system
818 	 * block size if it is larger than the chosen cluster size.
819 	 */
820 	if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
821 		mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
822 	} else {
823 		mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
824 	}
825 	/*
826 	 * Set whether we're using inode alignment.
827 	 */
828 	if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
829 	    mp->m_sb.sb_inoalignmt >=
830 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
831 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
832 	else
833 		mp->m_inoalign_mask = 0;
834 	/*
835 	 * If we are using stripe alignment, check whether
836 	 * the stripe unit is a multiple of the inode alignment
837 	 */
838 	if (mp->m_dalign && mp->m_inoalign_mask &&
839 	    !(mp->m_dalign & mp->m_inoalign_mask))
840 		mp->m_sinoalign = mp->m_dalign;
841 	else
842 		mp->m_sinoalign = 0;
843 	/*
844 	 * Check that the data (and log if separate) are an ok size.
845 	 */
846 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
847 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
848 		cmn_err(CE_WARN, "XFS: size check 1 failed");
849 		error = XFS_ERROR(E2BIG);
850 		goto error1;
851 	}
852 	error = xfs_read_buf(mp, mp->m_ddev_targp,
853 			     d - XFS_FSS_TO_BB(mp, 1),
854 			     XFS_FSS_TO_BB(mp, 1), 0, &bp);
855 	if (!error) {
856 		xfs_buf_relse(bp);
857 	} else {
858 		cmn_err(CE_WARN, "XFS: size check 2 failed");
859 		if (error == ENOSPC) {
860 			error = XFS_ERROR(E2BIG);
861 		}
862 		goto error1;
863 	}
864 
865 	if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
866 	    mp->m_logdev_targp != mp->m_ddev_targp) {
867 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
868 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
869 			cmn_err(CE_WARN, "XFS: size check 3 failed");
870 			error = XFS_ERROR(E2BIG);
871 			goto error1;
872 		}
873 		error = xfs_read_buf(mp, mp->m_logdev_targp,
874 				     d - XFS_FSB_TO_BB(mp, 1),
875 				     XFS_FSB_TO_BB(mp, 1), 0, &bp);
876 		if (!error) {
877 			xfs_buf_relse(bp);
878 		} else {
879 			cmn_err(CE_WARN, "XFS: size check 3 failed");
880 			if (error == ENOSPC) {
881 				error = XFS_ERROR(E2BIG);
882 			}
883 			goto error1;
884 		}
885 	}
886 
887 	/*
888 	 * Initialize realtime fields in the mount structure
889 	 */
890 	if ((error = xfs_rtmount_init(mp))) {
891 		cmn_err(CE_WARN, "XFS: RT mount failed");
892 		goto error1;
893 	}
894 
895 	/*
896 	 * For client case we are done now
897 	 */
898 	if (mfsi_flags & XFS_MFSI_CLIENT) {
899 		return(0);
900 	}
901 
902 	/*
903 	 *  Copies the low order bits of the timestamp and the randomly
904 	 *  set "sequence" number out of a UUID.
905 	 */
906 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
907 
908 	/*
909 	 *  The vfs structure needs to have a file system independent
910 	 *  way of checking for the invariant file system ID.  Since it
911 	 *  can't look at mount structures it has a pointer to the data
912 	 *  in the mount structure.
913 	 *
914 	 *  File systems that don't support user level file handles (i.e.
915 	 *  all of them except for XFS) will leave vfs_altfsid as NULL.
916 	 */
917 	vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
918 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
919 
920 	/*
921 	 * Select the right directory manager.
922 	 */
923 	mp->m_dirops =
924 		XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
925 			xfsv2_dirops :
926 			xfsv1_dirops;
927 
928 	/*
929 	 * Initialize directory manager's entries.
930 	 */
931 	XFS_DIR_MOUNT(mp);
932 
933 	/*
934 	 * Initialize the attribute manager's entries.
935 	 */
936 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
937 
938 	/*
939 	 * Initialize the precomputed transaction reservations values.
940 	 */
941 	xfs_trans_init(mp);
942 
943 	/*
944 	 * Allocate and initialize the inode hash table for this
945 	 * file system.
946 	 */
947 	xfs_ihash_init(mp);
948 	xfs_chash_init(mp);
949 
950 	/*
951 	 * Allocate and initialize the per-ag data.
952 	 */
953 	init_rwsem(&mp->m_peraglock);
954 	mp->m_perag =
955 		kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
956 
957 	mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
958 
959 	/*
960 	 * log's mount-time initialization. Perform 1st part recovery if needed
961 	 */
962 	if (likely(sbp->sb_logblocks > 0)) {	/* check for volume case */
963 		error = xfs_log_mount(mp, mp->m_logdev_targp,
964 				      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
965 				      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
966 		if (error) {
967 			cmn_err(CE_WARN, "XFS: log mount failed");
968 			goto error2;
969 		}
970 	} else {	/* No log has been defined */
971 		cmn_err(CE_WARN, "XFS: no log defined");
972 		XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
973 		error = XFS_ERROR(EFSCORRUPTED);
974 		goto error2;
975 	}
976 
977 	/*
978 	 * Get and sanity-check the root inode.
979 	 * Save the pointer to it in the mount structure.
980 	 */
981 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
982 	if (error) {
983 		cmn_err(CE_WARN, "XFS: failed to read root inode");
984 		goto error3;
985 	}
986 
987 	ASSERT(rip != NULL);
988 	rvp = XFS_ITOV(rip);
989 
990 	if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
991 		cmn_err(CE_WARN, "XFS: corrupted root inode");
992 		prdev("Root inode %llu is not a directory",
993 		      mp->m_ddev_targp, (unsigned long long)rip->i_ino);
994 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
995 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
996 				 mp);
997 		error = XFS_ERROR(EFSCORRUPTED);
998 		goto error4;
999 	}
1000 	mp->m_rootip = rip;	/* save it */
1001 
1002 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1003 
1004 	/*
1005 	 * Initialize realtime inode pointers in the mount structure
1006 	 */
1007 	if ((error = xfs_rtmount_inodes(mp))) {
1008 		/*
1009 		 * Free up the root inode.
1010 		 */
1011 		cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1012 		goto error4;
1013 	}
1014 
1015 	/*
1016 	 * If fs is not mounted readonly, then update the superblock
1017 	 * unit and width changes.
1018 	 */
1019 	if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1020 		xfs_mount_log_sbunit(mp, update_flags);
1021 
1022 	/*
1023 	 * Initialise the XFS quota management subsystem for this mount
1024 	 */
1025 	if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1026 		goto error4;
1027 
1028 	/*
1029 	 * Finish recovering the file system.  This part needed to be
1030 	 * delayed until after the root and real-time bitmap inodes
1031 	 * were consistently read in.
1032 	 */
1033 	error = xfs_log_mount_finish(mp, mfsi_flags);
1034 	if (error) {
1035 		cmn_err(CE_WARN, "XFS: log mount finish failed");
1036 		goto error4;
1037 	}
1038 
1039 	/*
1040 	 * Complete the quota initialisation, post-log-replay component.
1041 	 */
1042 	if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1043 		goto error4;
1044 
1045 	return 0;
1046 
1047  error4:
1048 	/*
1049 	 * Free up the root inode.
1050 	 */
1051 	VN_RELE(rvp);
1052  error3:
1053 	xfs_log_unmount_dealloc(mp);
1054  error2:
1055 	xfs_ihash_free(mp);
1056 	xfs_chash_free(mp);
1057 	for (agno = 0; agno < sbp->sb_agcount; agno++)
1058 		if (mp->m_perag[agno].pagb_list)
1059 			kmem_free(mp->m_perag[agno].pagb_list,
1060 			  sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1061 	kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1062 	mp->m_perag = NULL;
1063 	/* FALLTHROUGH */
1064  error1:
1065 	if (uuid_mounted)
1066 		xfs_uuid_unmount(mp);
1067 	xfs_freesb(mp);
1068 	return error;
1069 }
1070 
1071 /*
1072  * xfs_unmountfs
1073  *
1074  * This flushes out the inodes,dquots and the superblock, unmounts the
1075  * log and makes sure that incore structures are freed.
1076  */
1077 int
1078 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1079 {
1080 	struct vfs	*vfsp = XFS_MTOVFS(mp);
1081 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1082 	int64_t		fsid;
1083 #endif
1084 
1085 	xfs_iflush_all(mp);
1086 
1087 	XFS_QM_DQPURGEALL(mp,
1088 		XFS_QMOPT_UQUOTA | XFS_QMOPT_GQUOTA | XFS_QMOPT_UMOUNTING);
1089 
1090 	/*
1091 	 * Flush out the log synchronously so that we know for sure
1092 	 * that nothing is pinned.  This is important because bflush()
1093 	 * will skip pinned buffers.
1094 	 */
1095 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1096 
1097 	xfs_binval(mp->m_ddev_targp);
1098 	if (mp->m_rtdev_targp) {
1099 		xfs_binval(mp->m_rtdev_targp);
1100 	}
1101 
1102 	xfs_unmountfs_writesb(mp);
1103 
1104 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
1105 
1106 	xfs_log_unmount(mp);			/* Done! No more fs ops. */
1107 
1108 	xfs_freesb(mp);
1109 
1110 	/*
1111 	 * All inodes from this mount point should be freed.
1112 	 */
1113 	ASSERT(mp->m_inodes == NULL);
1114 
1115 	xfs_unmountfs_close(mp, cr);
1116 	if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1117 		xfs_uuid_unmount(mp);
1118 
1119 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1120 	/*
1121 	 * clear all error tags on this filesystem
1122 	 */
1123 	memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1124 	xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1125 #endif
1126 	XFS_IODONE(vfsp);
1127 	xfs_mount_free(mp, 1);
1128 	return 0;
1129 }
1130 
1131 void
1132 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1133 {
1134 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1135 		xfs_free_buftarg(mp->m_logdev_targp, 1);
1136 	if (mp->m_rtdev_targp)
1137 		xfs_free_buftarg(mp->m_rtdev_targp, 1);
1138 	xfs_free_buftarg(mp->m_ddev_targp, 0);
1139 }
1140 
1141 STATIC void
1142 xfs_unmountfs_wait(xfs_mount_t *mp)
1143 {
1144 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1145 		xfs_wait_buftarg(mp->m_logdev_targp);
1146 	if (mp->m_rtdev_targp)
1147 		xfs_wait_buftarg(mp->m_rtdev_targp);
1148 	xfs_wait_buftarg(mp->m_ddev_targp);
1149 }
1150 
1151 int
1152 xfs_unmountfs_writesb(xfs_mount_t *mp)
1153 {
1154 	xfs_buf_t	*sbp;
1155 	xfs_sb_t	*sb;
1156 	int		error = 0;
1157 
1158 	/*
1159 	 * skip superblock write if fs is read-only, or
1160 	 * if we are doing a forced umount.
1161 	 */
1162 	sbp = xfs_getsb(mp, 0);
1163 	if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1164 		XFS_FORCED_SHUTDOWN(mp))) {
1165 		/*
1166 		 * mark shared-readonly if desired
1167 		 */
1168 		sb = XFS_BUF_TO_SBP(sbp);
1169 		if (mp->m_mk_sharedro) {
1170 			if (!(sb->sb_flags & XFS_SBF_READONLY))
1171 				sb->sb_flags |= XFS_SBF_READONLY;
1172 			if (!XFS_SB_VERSION_HASSHARED(sb))
1173 				XFS_SB_VERSION_ADDSHARED(sb);
1174 			xfs_fs_cmn_err(CE_NOTE, mp,
1175 				"Unmounting, marking shared read-only");
1176 		}
1177 		XFS_BUF_UNDONE(sbp);
1178 		XFS_BUF_UNREAD(sbp);
1179 		XFS_BUF_UNDELAYWRITE(sbp);
1180 		XFS_BUF_WRITE(sbp);
1181 		XFS_BUF_UNASYNC(sbp);
1182 		ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1183 		xfsbdstrat(mp, sbp);
1184 		/* Nevermind errors we might get here. */
1185 		error = xfs_iowait(sbp);
1186 		if (error)
1187 			xfs_ioerror_alert("xfs_unmountfs_writesb",
1188 					  mp, sbp, XFS_BUF_ADDR(sbp));
1189 		if (error && mp->m_mk_sharedro)
1190 			xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
1191 	}
1192 	xfs_buf_relse(sbp);
1193 	return (error);
1194 }
1195 
1196 /*
1197  * xfs_mod_sb() can be used to copy arbitrary changes to the
1198  * in-core superblock into the superblock buffer to be logged.
1199  * It does not provide the higher level of locking that is
1200  * needed to protect the in-core superblock from concurrent
1201  * access.
1202  */
1203 void
1204 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1205 {
1206 	xfs_buf_t	*bp;
1207 	int		first;
1208 	int		last;
1209 	xfs_mount_t	*mp;
1210 	xfs_sb_t	*sbp;
1211 	xfs_sb_field_t	f;
1212 
1213 	ASSERT(fields);
1214 	if (!fields)
1215 		return;
1216 	mp = tp->t_mountp;
1217 	bp = xfs_trans_getsb(tp, mp, 0);
1218 	sbp = XFS_BUF_TO_SBP(bp);
1219 	first = sizeof(xfs_sb_t);
1220 	last = 0;
1221 
1222 	/* translate/copy */
1223 
1224 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1225 
1226 	/* find modified range */
1227 
1228 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1229 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1230 	first = xfs_sb_info[f].offset;
1231 
1232 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1233 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1234 	last = xfs_sb_info[f + 1].offset - 1;
1235 
1236 	xfs_trans_log_buf(tp, bp, first, last);
1237 }
1238 
1239 /*
1240  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1241  * a delta to a specified field in the in-core superblock.  Simply
1242  * switch on the field indicated and apply the delta to that field.
1243  * Fields are not allowed to dip below zero, so if the delta would
1244  * do this do not apply it and return EINVAL.
1245  *
1246  * The SB_LOCK must be held when this routine is called.
1247  */
1248 STATIC int
1249 xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1250 			int delta, int rsvd)
1251 {
1252 	int		scounter;	/* short counter for 32 bit fields */
1253 	long long	lcounter;	/* long counter for 64 bit fields */
1254 	long long	res_used, rem;
1255 
1256 	/*
1257 	 * With the in-core superblock spin lock held, switch
1258 	 * on the indicated field.  Apply the delta to the
1259 	 * proper field.  If the fields value would dip below
1260 	 * 0, then do not apply the delta and return EINVAL.
1261 	 */
1262 	switch (field) {
1263 	case XFS_SBS_ICOUNT:
1264 		lcounter = (long long)mp->m_sb.sb_icount;
1265 		lcounter += delta;
1266 		if (lcounter < 0) {
1267 			ASSERT(0);
1268 			return (XFS_ERROR(EINVAL));
1269 		}
1270 		mp->m_sb.sb_icount = lcounter;
1271 		return (0);
1272 	case XFS_SBS_IFREE:
1273 		lcounter = (long long)mp->m_sb.sb_ifree;
1274 		lcounter += delta;
1275 		if (lcounter < 0) {
1276 			ASSERT(0);
1277 			return (XFS_ERROR(EINVAL));
1278 		}
1279 		mp->m_sb.sb_ifree = lcounter;
1280 		return (0);
1281 	case XFS_SBS_FDBLOCKS:
1282 
1283 		lcounter = (long long)mp->m_sb.sb_fdblocks;
1284 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1285 
1286 		if (delta > 0) {		/* Putting blocks back */
1287 			if (res_used > delta) {
1288 				mp->m_resblks_avail += delta;
1289 			} else {
1290 				rem = delta - res_used;
1291 				mp->m_resblks_avail = mp->m_resblks;
1292 				lcounter += rem;
1293 			}
1294 		} else {				/* Taking blocks away */
1295 
1296 			lcounter += delta;
1297 
1298 		/*
1299 		 * If were out of blocks, use any available reserved blocks if
1300 		 * were allowed to.
1301 		 */
1302 
1303 			if (lcounter < 0) {
1304 				if (rsvd) {
1305 					lcounter = (long long)mp->m_resblks_avail + delta;
1306 					if (lcounter < 0) {
1307 						return (XFS_ERROR(ENOSPC));
1308 					}
1309 					mp->m_resblks_avail = lcounter;
1310 					return (0);
1311 				} else {	/* not reserved */
1312 					return (XFS_ERROR(ENOSPC));
1313 				}
1314 			}
1315 		}
1316 
1317 		mp->m_sb.sb_fdblocks = lcounter;
1318 		return (0);
1319 	case XFS_SBS_FREXTENTS:
1320 		lcounter = (long long)mp->m_sb.sb_frextents;
1321 		lcounter += delta;
1322 		if (lcounter < 0) {
1323 			return (XFS_ERROR(ENOSPC));
1324 		}
1325 		mp->m_sb.sb_frextents = lcounter;
1326 		return (0);
1327 	case XFS_SBS_DBLOCKS:
1328 		lcounter = (long long)mp->m_sb.sb_dblocks;
1329 		lcounter += delta;
1330 		if (lcounter < 0) {
1331 			ASSERT(0);
1332 			return (XFS_ERROR(EINVAL));
1333 		}
1334 		mp->m_sb.sb_dblocks = lcounter;
1335 		return (0);
1336 	case XFS_SBS_AGCOUNT:
1337 		scounter = mp->m_sb.sb_agcount;
1338 		scounter += delta;
1339 		if (scounter < 0) {
1340 			ASSERT(0);
1341 			return (XFS_ERROR(EINVAL));
1342 		}
1343 		mp->m_sb.sb_agcount = scounter;
1344 		return (0);
1345 	case XFS_SBS_IMAX_PCT:
1346 		scounter = mp->m_sb.sb_imax_pct;
1347 		scounter += delta;
1348 		if (scounter < 0) {
1349 			ASSERT(0);
1350 			return (XFS_ERROR(EINVAL));
1351 		}
1352 		mp->m_sb.sb_imax_pct = scounter;
1353 		return (0);
1354 	case XFS_SBS_REXTSIZE:
1355 		scounter = mp->m_sb.sb_rextsize;
1356 		scounter += delta;
1357 		if (scounter < 0) {
1358 			ASSERT(0);
1359 			return (XFS_ERROR(EINVAL));
1360 		}
1361 		mp->m_sb.sb_rextsize = scounter;
1362 		return (0);
1363 	case XFS_SBS_RBMBLOCKS:
1364 		scounter = mp->m_sb.sb_rbmblocks;
1365 		scounter += delta;
1366 		if (scounter < 0) {
1367 			ASSERT(0);
1368 			return (XFS_ERROR(EINVAL));
1369 		}
1370 		mp->m_sb.sb_rbmblocks = scounter;
1371 		return (0);
1372 	case XFS_SBS_RBLOCKS:
1373 		lcounter = (long long)mp->m_sb.sb_rblocks;
1374 		lcounter += delta;
1375 		if (lcounter < 0) {
1376 			ASSERT(0);
1377 			return (XFS_ERROR(EINVAL));
1378 		}
1379 		mp->m_sb.sb_rblocks = lcounter;
1380 		return (0);
1381 	case XFS_SBS_REXTENTS:
1382 		lcounter = (long long)mp->m_sb.sb_rextents;
1383 		lcounter += delta;
1384 		if (lcounter < 0) {
1385 			ASSERT(0);
1386 			return (XFS_ERROR(EINVAL));
1387 		}
1388 		mp->m_sb.sb_rextents = lcounter;
1389 		return (0);
1390 	case XFS_SBS_REXTSLOG:
1391 		scounter = mp->m_sb.sb_rextslog;
1392 		scounter += delta;
1393 		if (scounter < 0) {
1394 			ASSERT(0);
1395 			return (XFS_ERROR(EINVAL));
1396 		}
1397 		mp->m_sb.sb_rextslog = scounter;
1398 		return (0);
1399 	default:
1400 		ASSERT(0);
1401 		return (XFS_ERROR(EINVAL));
1402 	}
1403 }
1404 
1405 /*
1406  * xfs_mod_incore_sb() is used to change a field in the in-core
1407  * superblock structure by the specified delta.  This modification
1408  * is protected by the SB_LOCK.  Just use the xfs_mod_incore_sb_unlocked()
1409  * routine to do the work.
1410  */
1411 int
1412 xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1413 {
1414 	unsigned long	s;
1415 	int	status;
1416 
1417 	s = XFS_SB_LOCK(mp);
1418 	status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1419 	XFS_SB_UNLOCK(mp, s);
1420 	return (status);
1421 }
1422 
1423 /*
1424  * xfs_mod_incore_sb_batch() is used to change more than one field
1425  * in the in-core superblock structure at a time.  This modification
1426  * is protected by a lock internal to this module.  The fields and
1427  * changes to those fields are specified in the array of xfs_mod_sb
1428  * structures passed in.
1429  *
1430  * Either all of the specified deltas will be applied or none of
1431  * them will.  If any modified field dips below 0, then all modifications
1432  * will be backed out and EINVAL will be returned.
1433  */
1434 int
1435 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1436 {
1437 	unsigned long	s;
1438 	int		status=0;
1439 	xfs_mod_sb_t	*msbp;
1440 
1441 	/*
1442 	 * Loop through the array of mod structures and apply each
1443 	 * individually.  If any fail, then back out all those
1444 	 * which have already been applied.  Do all of this within
1445 	 * the scope of the SB_LOCK so that all of the changes will
1446 	 * be atomic.
1447 	 */
1448 	s = XFS_SB_LOCK(mp);
1449 	msbp = &msb[0];
1450 	for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1451 		/*
1452 		 * Apply the delta at index n.  If it fails, break
1453 		 * from the loop so we'll fall into the undo loop
1454 		 * below.
1455 		 */
1456 		status = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1457 						    msbp->msb_delta, rsvd);
1458 		if (status != 0) {
1459 			break;
1460 		}
1461 	}
1462 
1463 	/*
1464 	 * If we didn't complete the loop above, then back out
1465 	 * any changes made to the superblock.  If you add code
1466 	 * between the loop above and here, make sure that you
1467 	 * preserve the value of status. Loop back until
1468 	 * we step below the beginning of the array.  Make sure
1469 	 * we don't touch anything back there.
1470 	 */
1471 	if (status != 0) {
1472 		msbp--;
1473 		while (msbp >= msb) {
1474 			status = xfs_mod_incore_sb_unlocked(mp,
1475 				    msbp->msb_field, -(msbp->msb_delta), rsvd);
1476 			ASSERT(status == 0);
1477 			msbp--;
1478 		}
1479 	}
1480 	XFS_SB_UNLOCK(mp, s);
1481 	return (status);
1482 }
1483 
1484 /*
1485  * xfs_getsb() is called to obtain the buffer for the superblock.
1486  * The buffer is returned locked and read in from disk.
1487  * The buffer should be released with a call to xfs_brelse().
1488  *
1489  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1490  * the superblock buffer if it can be locked without sleeping.
1491  * If it can't then we'll return NULL.
1492  */
1493 xfs_buf_t *
1494 xfs_getsb(
1495 	xfs_mount_t	*mp,
1496 	int		flags)
1497 {
1498 	xfs_buf_t	*bp;
1499 
1500 	ASSERT(mp->m_sb_bp != NULL);
1501 	bp = mp->m_sb_bp;
1502 	if (flags & XFS_BUF_TRYLOCK) {
1503 		if (!XFS_BUF_CPSEMA(bp)) {
1504 			return NULL;
1505 		}
1506 	} else {
1507 		XFS_BUF_PSEMA(bp, PRIBIO);
1508 	}
1509 	XFS_BUF_HOLD(bp);
1510 	ASSERT(XFS_BUF_ISDONE(bp));
1511 	return (bp);
1512 }
1513 
1514 /*
1515  * Used to free the superblock along various error paths.
1516  */
1517 void
1518 xfs_freesb(
1519 	xfs_mount_t	*mp)
1520 {
1521 	xfs_buf_t	*bp;
1522 
1523 	/*
1524 	 * Use xfs_getsb() so that the buffer will be locked
1525 	 * when we call xfs_buf_relse().
1526 	 */
1527 	bp = xfs_getsb(mp, 0);
1528 	XFS_BUF_UNMANAGE(bp);
1529 	xfs_buf_relse(bp);
1530 	mp->m_sb_bp = NULL;
1531 }
1532 
1533 /*
1534  * See if the UUID is unique among mounted XFS filesystems.
1535  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1536  */
1537 STATIC int
1538 xfs_uuid_mount(
1539 	xfs_mount_t	*mp)
1540 {
1541 	if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1542 		cmn_err(CE_WARN,
1543 			"XFS: Filesystem %s has nil UUID - can't mount",
1544 			mp->m_fsname);
1545 		return -1;
1546 	}
1547 	if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1548 		cmn_err(CE_WARN,
1549 			"XFS: Filesystem %s has duplicate UUID - can't mount",
1550 			mp->m_fsname);
1551 		return -1;
1552 	}
1553 	return 0;
1554 }
1555 
1556 /*
1557  * Remove filesystem from the UUID table.
1558  */
1559 STATIC void
1560 xfs_uuid_unmount(
1561 	xfs_mount_t	*mp)
1562 {
1563 	uuid_table_remove(&mp->m_sb.sb_uuid);
1564 }
1565 
1566 /*
1567  * Used to log changes to the superblock unit and width fields which could
1568  * be altered by the mount options. Only the first superblock is updated.
1569  */
1570 STATIC void
1571 xfs_mount_log_sbunit(
1572 	xfs_mount_t	*mp,
1573 	__int64_t	fields)
1574 {
1575 	xfs_trans_t	*tp;
1576 
1577 	ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1578 
1579 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1580 	if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1581 				XFS_DEFAULT_LOG_COUNT)) {
1582 		xfs_trans_cancel(tp, 0);
1583 		return;
1584 	}
1585 	xfs_mod_sb(tp, fields);
1586 	xfs_trans_commit(tp, 0, NULL);
1587 }
1588