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 
68 void xfs_xlatesb(void *, xfs_sb_t *, int, xfs_arch_t, __int64_t);
69 
70 static struct {
71     short offset;
72     short type;     /* 0 = integer
73 		* 1 = binary / string (no translation)
74 		*/
75 } xfs_sb_info[] = {
76     { offsetof(xfs_sb_t, sb_magicnum),   0 },
77     { offsetof(xfs_sb_t, sb_blocksize),  0 },
78     { offsetof(xfs_sb_t, sb_dblocks),    0 },
79     { offsetof(xfs_sb_t, sb_rblocks),    0 },
80     { offsetof(xfs_sb_t, sb_rextents),   0 },
81     { offsetof(xfs_sb_t, sb_uuid),       1 },
82     { offsetof(xfs_sb_t, sb_logstart),   0 },
83     { offsetof(xfs_sb_t, sb_rootino),    0 },
84     { offsetof(xfs_sb_t, sb_rbmino),     0 },
85     { offsetof(xfs_sb_t, sb_rsumino),    0 },
86     { offsetof(xfs_sb_t, sb_rextsize),   0 },
87     { offsetof(xfs_sb_t, sb_agblocks),   0 },
88     { offsetof(xfs_sb_t, sb_agcount),    0 },
89     { offsetof(xfs_sb_t, sb_rbmblocks),  0 },
90     { offsetof(xfs_sb_t, sb_logblocks),  0 },
91     { offsetof(xfs_sb_t, sb_versionnum), 0 },
92     { offsetof(xfs_sb_t, sb_sectsize),   0 },
93     { offsetof(xfs_sb_t, sb_inodesize),  0 },
94     { offsetof(xfs_sb_t, sb_inopblock),  0 },
95     { offsetof(xfs_sb_t, sb_fname[0]),   1 },
96     { offsetof(xfs_sb_t, sb_blocklog),   0 },
97     { offsetof(xfs_sb_t, sb_sectlog),    0 },
98     { offsetof(xfs_sb_t, sb_inodelog),   0 },
99     { offsetof(xfs_sb_t, sb_inopblog),   0 },
100     { offsetof(xfs_sb_t, sb_agblklog),   0 },
101     { offsetof(xfs_sb_t, sb_rextslog),   0 },
102     { offsetof(xfs_sb_t, sb_inprogress), 0 },
103     { offsetof(xfs_sb_t, sb_imax_pct),   0 },
104     { offsetof(xfs_sb_t, sb_icount),     0 },
105     { offsetof(xfs_sb_t, sb_ifree),      0 },
106     { offsetof(xfs_sb_t, sb_fdblocks),   0 },
107     { offsetof(xfs_sb_t, sb_frextents),  0 },
108     { offsetof(xfs_sb_t, sb_uquotino),   0 },
109     { offsetof(xfs_sb_t, sb_gquotino),   0 },
110     { offsetof(xfs_sb_t, sb_qflags),     0 },
111     { offsetof(xfs_sb_t, sb_flags),      0 },
112     { offsetof(xfs_sb_t, sb_shared_vn),  0 },
113     { offsetof(xfs_sb_t, sb_inoalignmt), 0 },
114     { offsetof(xfs_sb_t, sb_unit),	 0 },
115     { offsetof(xfs_sb_t, sb_width),	 0 },
116     { offsetof(xfs_sb_t, sb_dirblklog),	 0 },
117     { offsetof(xfs_sb_t, sb_logsectlog), 0 },
118     { offsetof(xfs_sb_t, sb_logsectsize),0 },
119     { offsetof(xfs_sb_t, sb_logsunit),	 0 },
120     { offsetof(xfs_sb_t, sb_features2),	 0 },
121     { sizeof(xfs_sb_t),			 0 }
122 };
123 
124 /*
125  * Return a pointer to an initialized xfs_mount structure.
126  */
127 xfs_mount_t *
xfs_mount_init(void)128 xfs_mount_init(void)
129 {
130 	xfs_mount_t *mp;
131 
132 	mp = kmem_zalloc(sizeof(*mp), KM_SLEEP);
133 
134 	AIL_LOCKINIT(&mp->m_ail_lock, "xfs_ail");
135 	spinlock_init(&mp->m_sb_lock, "xfs_sb");
136 	mutex_init(&mp->m_ilock, MUTEX_DEFAULT, "xfs_ilock");
137 	initnsema(&mp->m_growlock, 1, "xfs_grow");
138 	/*
139 	 * Initialize the AIL.
140 	 */
141 	xfs_trans_ail_init(mp);
142 
143 	atomic_set(&mp->m_active_trans, 0);
144 
145 	return mp;
146 }
147 
148 /*
149  * Free up the resources associated with a mount structure.  Assume that
150  * the structure was initially zeroed, so we can tell which fields got
151  * initialized.
152  */
153 void
xfs_mount_free(xfs_mount_t * mp,int remove_bhv)154 xfs_mount_free(
155 	xfs_mount_t *mp,
156 	int	    remove_bhv)
157 {
158 	if (mp->m_ihash)
159 		xfs_ihash_free(mp);
160 	if (mp->m_chash)
161 		xfs_chash_free(mp);
162 
163 	if (mp->m_perag) {
164 		int	agno;
165 
166 		for (agno = 0; agno < mp->m_maxagi; agno++)
167 			if (mp->m_perag[agno].pagb_list)
168 				kmem_free(mp->m_perag[agno].pagb_list,
169 						sizeof(xfs_perag_busy_t) *
170 							XFS_PAGB_NUM_SLOTS);
171 		kmem_free(mp->m_perag,
172 			  sizeof(xfs_perag_t) * mp->m_sb.sb_agcount);
173 	}
174 
175 	AIL_LOCK_DESTROY(&mp->m_ail_lock);
176 	spinlock_destroy(&mp->m_sb_lock);
177 	mutex_destroy(&mp->m_ilock);
178 	freesema(&mp->m_growlock);
179 	if (mp->m_quotainfo)
180 		XFS_QM_DONE(mp);
181 
182 	if (mp->m_fsname != NULL)
183 		kmem_free(mp->m_fsname, mp->m_fsname_len);
184 
185 	if (remove_bhv) {
186 		struct vfs	*vfsp = XFS_MTOVFS(mp);
187 
188 		bhv_remove_all_vfsops(vfsp, 0);
189 		VFS_REMOVEBHV(vfsp, &mp->m_bhv);
190 	}
191 
192 	kmem_free(mp, sizeof(xfs_mount_t));
193 }
194 
195 
196 /*
197  * Check the validity of the SB found.
198  */
199 STATIC int
xfs_mount_validate_sb(xfs_mount_t * mp,xfs_sb_t * sbp)200 xfs_mount_validate_sb(
201 	xfs_mount_t	*mp,
202 	xfs_sb_t	*sbp)
203 {
204 	/*
205 	 * If the log device and data device have the
206 	 * same device number, the log is internal.
207 	 * Consequently, the sb_logstart should be non-zero.  If
208 	 * we have a zero sb_logstart in this case, we may be trying to mount
209 	 * a volume filesystem in a non-volume manner.
210 	 */
211 	if (sbp->sb_magicnum != XFS_SB_MAGIC) {
212 		cmn_err(CE_WARN, "XFS: bad magic number");
213 		return XFS_ERROR(EWRONGFS);
214 	}
215 
216 	if (!XFS_SB_GOOD_VERSION(sbp)) {
217 		cmn_err(CE_WARN, "XFS: bad version");
218 		return XFS_ERROR(EWRONGFS);
219 	}
220 
221 	if (unlikely(
222 	    sbp->sb_logstart == 0 && mp->m_logdev_targp == mp->m_ddev_targp)) {
223 		cmn_err(CE_WARN,
224 	"XFS: filesystem is marked as having an external log; "
225 	"specify logdev on the\nmount command line.");
226 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(1)",
227 				     XFS_ERRLEVEL_HIGH, mp, sbp);
228 		return XFS_ERROR(EFSCORRUPTED);
229 	}
230 
231 	if (unlikely(
232 	    sbp->sb_logstart != 0 && mp->m_logdev_targp != mp->m_ddev_targp)) {
233 		cmn_err(CE_WARN,
234 	"XFS: filesystem is marked as having an internal log; "
235 	"don't specify logdev on\nthe mount command line.");
236 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(2)",
237 				     XFS_ERRLEVEL_HIGH, mp, sbp);
238 		return XFS_ERROR(EFSCORRUPTED);
239 	}
240 
241 	/*
242 	 * More sanity checking. These were stolen directly from
243 	 * xfs_repair.
244 	 */
245 	if (unlikely(
246 	    sbp->sb_agcount <= 0					||
247 	    sbp->sb_sectsize < XFS_MIN_SECTORSIZE			||
248 	    sbp->sb_sectsize > XFS_MAX_SECTORSIZE			||
249 	    sbp->sb_sectlog < XFS_MIN_SECTORSIZE_LOG			||
250 	    sbp->sb_sectlog > XFS_MAX_SECTORSIZE_LOG			||
251 	    sbp->sb_blocksize < XFS_MIN_BLOCKSIZE			||
252 	    sbp->sb_blocksize > XFS_MAX_BLOCKSIZE			||
253 	    sbp->sb_blocklog < XFS_MIN_BLOCKSIZE_LOG			||
254 	    sbp->sb_blocklog > XFS_MAX_BLOCKSIZE_LOG			||
255 	    sbp->sb_inodesize < XFS_DINODE_MIN_SIZE			||
256 	    sbp->sb_inodesize > XFS_DINODE_MAX_SIZE			||
257 	    (sbp->sb_rextsize * sbp->sb_blocksize > XFS_MAX_RTEXTSIZE)	||
258 	    (sbp->sb_rextsize * sbp->sb_blocksize < XFS_MIN_RTEXTSIZE)	||
259 	    sbp->sb_imax_pct > 100)) {
260 		cmn_err(CE_WARN, "XFS: SB sanity check 1 failed");
261 		XFS_CORRUPTION_ERROR("xfs_mount_validate_sb(3)",
262 				     XFS_ERRLEVEL_LOW, mp, sbp);
263 		return XFS_ERROR(EFSCORRUPTED);
264 	}
265 
266 	/*
267 	 * Sanity check AG count, size fields against data size field
268 	 */
269 	if (unlikely(
270 	    sbp->sb_dblocks == 0 ||
271 	    sbp->sb_dblocks >
272 	     (xfs_drfsbno_t)sbp->sb_agcount * sbp->sb_agblocks ||
273 	    sbp->sb_dblocks < (xfs_drfsbno_t)(sbp->sb_agcount - 1) *
274 			      sbp->sb_agblocks + XFS_MIN_AG_BLOCKS)) {
275 		cmn_err(CE_WARN, "XFS: SB sanity check 2 failed");
276 		XFS_ERROR_REPORT("xfs_mount_validate_sb(4)",
277 				 XFS_ERRLEVEL_LOW, mp);
278 		return XFS_ERROR(EFSCORRUPTED);
279 	}
280 
281 	ASSERT(PAGE_SHIFT >= sbp->sb_blocklog);
282 	ASSERT(sbp->sb_blocklog >= BBSHIFT);
283 
284 #if XFS_BIG_BLKNOS     /* Limited by ULONG_MAX of page cache index */
285 	if (unlikely(
286 	    (sbp->sb_dblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX ||
287 	    (sbp->sb_rblocks >> (PAGE_SHIFT - sbp->sb_blocklog)) > ULONG_MAX)) {
288 #else                  /* Limited by UINT_MAX of sectors */
289 	if (unlikely(
290 	    (sbp->sb_dblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX ||
291 	    (sbp->sb_rblocks << (sbp->sb_blocklog - BBSHIFT)) > UINT_MAX)) {
292 #endif
293 		cmn_err(CE_WARN,
294 	"XFS: File system is too large to be mounted on this system.");
295 		return XFS_ERROR(E2BIG);
296 	}
297 
298 	if (unlikely(sbp->sb_inprogress)) {
299 		cmn_err(CE_WARN, "XFS: file system busy");
300 		XFS_ERROR_REPORT("xfs_mount_validate_sb(5)",
301 				 XFS_ERRLEVEL_LOW, mp);
302 		return XFS_ERROR(EFSCORRUPTED);
303 	}
304 
305 	/*
306 	 * Until this is fixed only page-sized or smaller data blocks work.
307 	 */
308 	if (unlikely(sbp->sb_blocksize > PAGE_SIZE)) {
309 		cmn_err(CE_WARN,
310 		"XFS: Attempted to mount file system with blocksize %d bytes",
311 			sbp->sb_blocksize);
312 		cmn_err(CE_WARN,
313 		"XFS: Only page-sized (%d) or less blocksizes currently work.",
314 			PAGE_SIZE);
315 		return XFS_ERROR(ENOSYS);
316 	}
317 
318 	return 0;
319 }
320 
321 xfs_agnumber_t
322 xfs_initialize_perag(xfs_mount_t *mp, xfs_agnumber_t agcount)
323 {
324 	xfs_agnumber_t	index, max_metadata;
325 	xfs_perag_t	*pag;
326 	xfs_agino_t	agino;
327 	xfs_ino_t	ino;
328 	xfs_sb_t	*sbp = &mp->m_sb;
329 	xfs_ino_t	max_inum = XFS_MAXINUMBER_32;
330 
331 	/* Check to see if the filesystem can overflow 32 bit inodes */
332 	agino = XFS_OFFBNO_TO_AGINO(mp, sbp->sb_agblocks - 1, 0);
333 	ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
334 
335 	/* Clear the mount flag if no inode can overflow 32 bits
336 	 * on this filesystem, or if specifically requested..
337 	 */
338 	if ((mp->m_flags & XFS_MOUNT_32BITINOOPT) && ino > max_inum) {
339 		mp->m_flags |= XFS_MOUNT_32BITINODES;
340 	} else {
341 		mp->m_flags &= ~XFS_MOUNT_32BITINODES;
342 	}
343 
344 	/* If we can overflow then setup the ag headers accordingly */
345 	if (mp->m_flags & XFS_MOUNT_32BITINODES) {
346 		/* Calculate how much should be reserved for inodes to
347 		 * meet the max inode percentage.
348 		 */
349 		if (mp->m_maxicount) {
350 			__uint64_t	icount;
351 
352 			icount = sbp->sb_dblocks * sbp->sb_imax_pct;
353 			do_div(icount, 100);
354 			icount += sbp->sb_agblocks - 1;
355 			do_div(icount, mp->m_ialloc_blks);
356 			max_metadata = icount;
357 		} else {
358 			max_metadata = agcount;
359 		}
360 		for (index = 0; index < agcount; index++) {
361 			ino = XFS_AGINO_TO_INO(mp, index, agino);
362 			if (ino > max_inum) {
363 				index++;
364 				break;
365 			}
366 
367 			/* This ag is prefered for inodes */
368 			pag = &mp->m_perag[index];
369 			pag->pagi_inodeok = 1;
370 			if (index < max_metadata)
371 				pag->pagf_metadata = 1;
372 		}
373 	} else {
374 		/* Setup default behavior for smaller filesystems */
375 		for (index = 0; index < agcount; index++) {
376 			pag = &mp->m_perag[index];
377 			pag->pagi_inodeok = 1;
378 		}
379 	}
380 	return index;
381 }
382 
383 /*
384  * xfs_xlatesb
385  *
386  *     data       - on disk version of sb
387  *     sb         - a superblock
388  *     dir        - conversion direction: <0 - convert sb to buf
389  *                                        >0 - convert buf to sb
390  *     arch       - architecture to read/write from/to buf
391  *     fields     - which fields to copy (bitmask)
392  */
393 void
394 xfs_xlatesb(
395 	void		*data,
396 	xfs_sb_t	*sb,
397 	int		dir,
398 	xfs_arch_t	arch,
399 	__int64_t	fields)
400 {
401 	xfs_caddr_t	buf_ptr;
402 	xfs_caddr_t	mem_ptr;
403 	xfs_sb_field_t	f;
404 	int		first;
405 	int		size;
406 
407 	ASSERT(dir);
408 	ASSERT(fields);
409 
410 	if (!fields)
411 		return;
412 
413 	buf_ptr = (xfs_caddr_t)data;
414 	mem_ptr = (xfs_caddr_t)sb;
415 
416 	while (fields) {
417 		f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
418 		first = xfs_sb_info[f].offset;
419 		size = xfs_sb_info[f + 1].offset - first;
420 
421 		ASSERT(xfs_sb_info[f].type == 0 || xfs_sb_info[f].type == 1);
422 
423 		if (arch == ARCH_NOCONVERT ||
424 		    size == 1 ||
425 		    xfs_sb_info[f].type == 1) {
426 			if (dir > 0) {
427 				memcpy(mem_ptr + first, buf_ptr + first, size);
428 			} else {
429 				memcpy(buf_ptr + first, mem_ptr + first, size);
430 			}
431 		} else {
432 			switch (size) {
433 			case 2:
434 				INT_XLATE(*(__uint16_t*)(buf_ptr+first),
435 					  *(__uint16_t*)(mem_ptr+first),
436 					  dir, arch);
437 				break;
438 			case 4:
439 				INT_XLATE(*(__uint32_t*)(buf_ptr+first),
440 					  *(__uint32_t*)(mem_ptr+first),
441 					  dir, arch);
442 				break;
443 			case 8:
444 				INT_XLATE(*(__uint64_t*)(buf_ptr+first),
445 					  *(__uint64_t*)(mem_ptr+first), dir, arch);
446 				break;
447 			default:
448 				ASSERT(0);
449 			}
450 		}
451 
452 		fields &= ~(1LL << f);
453 	}
454 }
455 
456 /*
457  * xfs_readsb
458  *
459  * Does the initial read of the superblock.
460  */
461 int
462 xfs_readsb(xfs_mount_t *mp)
463 {
464 	unsigned int	sector_size;
465 	unsigned int	extra_flags;
466 	xfs_buf_t	*bp;
467 	xfs_sb_t	*sbp;
468 	int		error;
469 
470 	ASSERT(mp->m_sb_bp == NULL);
471 	ASSERT(mp->m_ddev_targp != NULL);
472 
473 	/*
474 	 * Allocate a (locked) buffer to hold the superblock.
475 	 * This will be kept around at all times to optimize
476 	 * access to the superblock.
477 	 */
478 	sector_size = xfs_getsize_buftarg(mp->m_ddev_targp);
479 	extra_flags = XFS_BUF_LOCK | XFS_BUF_MANAGE | XFS_BUF_MAPPED;
480 
481 	bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
482 				BTOBB(sector_size), extra_flags);
483 	if (!bp || XFS_BUF_ISERROR(bp)) {
484 		cmn_err(CE_WARN, "XFS: SB read failed");
485 		error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
486 		goto fail;
487 	}
488 	ASSERT(XFS_BUF_ISBUSY(bp));
489 	ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
490 
491 	/*
492 	 * Initialize the mount structure from the superblock.
493 	 * But first do some basic consistency checking.
494 	 */
495 	sbp = XFS_BUF_TO_SBP(bp);
496 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), 1,
497 				ARCH_CONVERT, XFS_SB_ALL_BITS);
498 
499 	error = xfs_mount_validate_sb(mp, &(mp->m_sb));
500 	if (error) {
501 		cmn_err(CE_WARN, "XFS: SB validate failed");
502 		goto fail;
503 	}
504 
505 	/*
506 	 * We must be able to do sector-sized and sector-aligned IO.
507 	 */
508 	if (sector_size > mp->m_sb.sb_sectsize) {
509 		cmn_err(CE_WARN,
510 			"XFS: device supports only %u byte sectors (not %u)",
511 			sector_size, mp->m_sb.sb_sectsize);
512 		error = ENOSYS;
513 		goto fail;
514 	}
515 
516 	/*
517 	 * If device sector size is smaller than the superblock size,
518 	 * re-read the superblock so the buffer is correctly sized.
519 	 */
520 	if (sector_size < mp->m_sb.sb_sectsize) {
521 		XFS_BUF_UNMANAGE(bp);
522 		xfs_buf_relse(bp);
523 		sector_size = mp->m_sb.sb_sectsize;
524 		bp = xfs_buf_read_flags(mp->m_ddev_targp, XFS_SB_DADDR,
525 					BTOBB(sector_size), extra_flags);
526 		if (!bp || XFS_BUF_ISERROR(bp)) {
527 			cmn_err(CE_WARN, "XFS: SB re-read failed");
528 			error = bp ? XFS_BUF_GETERROR(bp) : ENOMEM;
529 			goto fail;
530 		}
531 		ASSERT(XFS_BUF_ISBUSY(bp));
532 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
533 	}
534 
535 	mp->m_sb_bp = bp;
536 	xfs_buf_relse(bp);
537 	ASSERT(XFS_BUF_VALUSEMA(bp) > 0);
538 	return 0;
539 
540  fail:
541 	if (bp) {
542 		XFS_BUF_UNMANAGE(bp);
543 		xfs_buf_relse(bp);
544 	}
545 	return error;
546 }
547 
548 
549 /*
550  * xfs_mount_common
551  *
552  * Mount initialization code establishing various mount
553  * fields from the superblock associated with the given
554  * mount structure
555  */
556 void
557 xfs_mount_common(xfs_mount_t *mp, xfs_sb_t *sbp)
558 {
559 	int	i;
560 
561 	mp->m_agfrotor = mp->m_agirotor = 0;
562 	spinlock_init(&mp->m_agirotor_lock, "m_agirotor_lock");
563 	mp->m_maxagi = mp->m_sb.sb_agcount;
564 	mp->m_blkbit_log = sbp->sb_blocklog + XFS_NBBYLOG;
565 	mp->m_blkbb_log = sbp->sb_blocklog - BBSHIFT;
566 	mp->m_sectbb_log = sbp->sb_sectlog - BBSHIFT;
567 	mp->m_agno_log = xfs_highbit32(sbp->sb_agcount - 1) + 1;
568 	mp->m_agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
569 	mp->m_litino = sbp->sb_inodesize -
570 		((uint)sizeof(xfs_dinode_core_t) + (uint)sizeof(xfs_agino_t));
571 	mp->m_blockmask = sbp->sb_blocksize - 1;
572 	mp->m_blockwsize = sbp->sb_blocksize >> XFS_WORDLOG;
573 	mp->m_blockwmask = mp->m_blockwsize - 1;
574 	INIT_LIST_HEAD(&mp->m_del_inodes);
575 
576 	/*
577 	 * Setup for attributes, in case they get created.
578 	 * This value is for inodes getting attributes for the first time,
579 	 * the per-inode value is for old attribute values.
580 	 */
581 	ASSERT(sbp->sb_inodesize >= 256 && sbp->sb_inodesize <= 2048);
582 	switch (sbp->sb_inodesize) {
583 	case 256:
584 		mp->m_attroffset = XFS_LITINO(mp) - XFS_BMDR_SPACE_CALC(2);
585 		break;
586 	case 512:
587 	case 1024:
588 	case 2048:
589 		mp->m_attroffset = XFS_BMDR_SPACE_CALC(12);
590 		break;
591 	default:
592 		ASSERT(0);
593 	}
594 	ASSERT(mp->m_attroffset < XFS_LITINO(mp));
595 
596 	for (i = 0; i < 2; i++) {
597 		mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
598 			xfs_alloc, i == 0);
599 		mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
600 			xfs_alloc, i == 0);
601 	}
602 	for (i = 0; i < 2; i++) {
603 		mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
604 			xfs_bmbt, i == 0);
605 		mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
606 			xfs_bmbt, i == 0);
607 	}
608 	for (i = 0; i < 2; i++) {
609 		mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
610 			xfs_inobt, i == 0);
611 		mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
612 			xfs_inobt, i == 0);
613 	}
614 
615 	mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
616 	mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
617 					sbp->sb_inopblock);
618 	mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
619 }
620 /*
621  * xfs_mountfs
622  *
623  * This function does the following on an initial mount of a file system:
624  *	- reads the superblock from disk and init the mount struct
625  *	- if we're a 32-bit kernel, do a size check on the superblock
626  *		so we don't mount terabyte filesystems
627  *	- init mount struct realtime fields
628  *	- allocate inode hash table for fs
629  *	- init directory manager
630  *	- perform recovery and init the log manager
631  */
632 int
633 xfs_mountfs(
634 	vfs_t		*vfsp,
635 	xfs_mount_t	*mp,
636 	int		mfsi_flags)
637 {
638 	xfs_buf_t	*bp;
639 	xfs_sb_t	*sbp = &(mp->m_sb);
640 	xfs_inode_t	*rip;
641 	vnode_t		*rvp = NULL;
642 	int		readio_log, writeio_log;
643 	xfs_daddr_t	d;
644 	__uint64_t	ret64;
645 	__int64_t	update_flags;
646 	uint		quotamount, quotaflags;
647 	int		agno;
648 	int		uuid_mounted = 0;
649 	int		error = 0;
650 
651 	if (mp->m_sb_bp == NULL) {
652 		if ((error = xfs_readsb(mp))) {
653 			return (error);
654 		}
655 	}
656 	xfs_mount_common(mp, sbp);
657 
658 	/*
659 	 * Check if sb_agblocks is aligned at stripe boundary
660 	 * If sb_agblocks is NOT aligned turn off m_dalign since
661 	 * allocator alignment is within an ag, therefore ag has
662 	 * to be aligned at stripe boundary.
663 	 */
664 	update_flags = 0LL;
665 	if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
666 		/*
667 		 * If stripe unit and stripe width are not multiples
668 		 * of the fs blocksize turn off alignment.
669 		 */
670 		if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
671 		    (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
672 			if (mp->m_flags & XFS_MOUNT_RETERR) {
673 				cmn_err(CE_WARN,
674 					"XFS: alignment check 1 failed");
675 				error = XFS_ERROR(EINVAL);
676 				goto error1;
677 			}
678 			mp->m_dalign = mp->m_swidth = 0;
679 		} else {
680 			/*
681 			 * Convert the stripe unit and width to FSBs.
682 			 */
683 			mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
684 			if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
685 				if (mp->m_flags & XFS_MOUNT_RETERR) {
686 					error = XFS_ERROR(EINVAL);
687 					goto error1;
688 				}
689 				xfs_fs_cmn_err(CE_WARN, mp,
690 "stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
691 					mp->m_dalign, mp->m_swidth,
692 					sbp->sb_agblocks);
693 
694 				mp->m_dalign = 0;
695 				mp->m_swidth = 0;
696 			} else if (mp->m_dalign) {
697 				mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
698 			} else {
699 				if (mp->m_flags & XFS_MOUNT_RETERR) {
700 					xfs_fs_cmn_err(CE_WARN, mp,
701 "stripe alignment turned off: sunit(%d) less than bsize(%d)",
702                                         	mp->m_dalign,
703 						mp->m_blockmask +1);
704 					error = XFS_ERROR(EINVAL);
705 					goto error1;
706 				}
707 				mp->m_swidth = 0;
708 			}
709 		}
710 
711 		/*
712 		 * Update superblock with new values
713 		 * and log changes
714 		 */
715 		if (XFS_SB_VERSION_HASDALIGN(sbp)) {
716 			if (sbp->sb_unit != mp->m_dalign) {
717 				sbp->sb_unit = mp->m_dalign;
718 				update_flags |= XFS_SB_UNIT;
719 			}
720 			if (sbp->sb_width != mp->m_swidth) {
721 				sbp->sb_width = mp->m_swidth;
722 				update_flags |= XFS_SB_WIDTH;
723 			}
724 		}
725 	} else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
726 		    XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
727 			mp->m_dalign = sbp->sb_unit;
728 			mp->m_swidth = sbp->sb_width;
729 	}
730 
731 	xfs_alloc_compute_maxlevels(mp);
732 	xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
733 	xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
734 	xfs_ialloc_compute_maxlevels(mp);
735 
736 	if (sbp->sb_imax_pct) {
737 		__uint64_t	icount;
738 
739 		/* Make sure the maximum inode count is a multiple of the
740 		 * units we allocate inodes in.
741 		 */
742 
743 		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
744 		do_div(icount, 100);
745 		do_div(icount, mp->m_ialloc_blks);
746 		mp->m_maxicount = (icount * mp->m_ialloc_blks)  <<
747 				   sbp->sb_inopblog;
748 	} else
749 		mp->m_maxicount = 0;
750 
751 	mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
752 
753 	/*
754 	 * XFS uses the uuid from the superblock as the unique
755 	 * identifier for fsid.  We can not use the uuid from the volume
756 	 * since a single partition filesystem is identical to a single
757 	 * partition volume/filesystem.
758 	 */
759 	if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
760 	    (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
761 		if (xfs_uuid_mount(mp)) {
762 			error = XFS_ERROR(EINVAL);
763 			goto error1;
764 		}
765 		uuid_mounted=1;
766 		ret64 = uuid_hash64(&sbp->sb_uuid);
767 		memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
768 	}
769 
770 	/*
771 	 * Set the default minimum read and write sizes unless
772 	 * already specified in a mount option.
773 	 * We use smaller I/O sizes when the file system
774 	 * is being used for NFS service (wsync mount option).
775 	 */
776 	if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
777 		if (mp->m_flags & XFS_MOUNT_WSYNC) {
778 			readio_log = XFS_WSYNC_READIO_LOG;
779 			writeio_log = XFS_WSYNC_WRITEIO_LOG;
780 		} else {
781 			readio_log = XFS_READIO_LOG_LARGE;
782 			writeio_log = XFS_WRITEIO_LOG_LARGE;
783 		}
784 	} else {
785 		readio_log = mp->m_readio_log;
786 		writeio_log = mp->m_writeio_log;
787 	}
788 
789 	/*
790 	 * Set the number of readahead buffers to use based on
791 	 * physical memory size.
792 	 */
793 	if (xfs_physmem <= 4096)		/* <= 16MB */
794 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
795 	else if (xfs_physmem <= 8192)	/* <= 32MB */
796 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
797 	else
798 		mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
799 	if (sbp->sb_blocklog > readio_log) {
800 		mp->m_readio_log = sbp->sb_blocklog;
801 	} else {
802 		mp->m_readio_log = readio_log;
803 	}
804 	mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
805 	if (sbp->sb_blocklog > writeio_log) {
806 		mp->m_writeio_log = sbp->sb_blocklog;
807 	} else {
808 		mp->m_writeio_log = writeio_log;
809 	}
810 	mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
811 
812 	/*
813 	 * Set the inode cluster size based on the physical memory
814 	 * size.  This may still be overridden by the file system
815 	 * block size if it is larger than the chosen cluster size.
816 	 */
817 	if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
818 		mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
819 	} else {
820 		mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
821 	}
822 	/*
823 	 * Set whether we're using inode alignment.
824 	 */
825 	if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
826 	    mp->m_sb.sb_inoalignmt >=
827 	    XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
828 		mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
829 	else
830 		mp->m_inoalign_mask = 0;
831 	/*
832 	 * If we are using stripe alignment, check whether
833 	 * the stripe unit is a multiple of the inode alignment
834 	 */
835 	if (mp->m_dalign && mp->m_inoalign_mask &&
836 	    !(mp->m_dalign & mp->m_inoalign_mask))
837 		mp->m_sinoalign = mp->m_dalign;
838 	else
839 		mp->m_sinoalign = 0;
840 	/*
841 	 * Check that the data (and log if separate) are an ok size.
842 	 */
843 	d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
844 	if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
845 		cmn_err(CE_WARN, "XFS: size check 1 failed");
846 		error = XFS_ERROR(E2BIG);
847 		goto error1;
848 	}
849 	error = xfs_read_buf(mp, mp->m_ddev_targp,
850 			     d - XFS_FSS_TO_BB(mp, 1),
851 			     XFS_FSS_TO_BB(mp, 1), 0, &bp);
852 	if (!error) {
853 		xfs_buf_relse(bp);
854 	} else {
855 		cmn_err(CE_WARN, "XFS: size check 2 failed");
856 		if (error == ENOSPC) {
857 			error = XFS_ERROR(E2BIG);
858 		}
859 		goto error1;
860 	}
861 
862 	if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
863 	    mp->m_logdev_targp != mp->m_ddev_targp) {
864 		d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
865 		if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
866 			cmn_err(CE_WARN, "XFS: size check 3 failed");
867 			error = XFS_ERROR(E2BIG);
868 			goto error1;
869 		}
870 		error = xfs_read_buf(mp, mp->m_logdev_targp,
871 				     d - XFS_FSB_TO_BB(mp, 1),
872 				     XFS_FSB_TO_BB(mp, 1), 0, &bp);
873 		if (!error) {
874 			xfs_buf_relse(bp);
875 		} else {
876 			cmn_err(CE_WARN, "XFS: size check 3 failed");
877 			if (error == ENOSPC) {
878 				error = XFS_ERROR(E2BIG);
879 			}
880 			goto error1;
881 		}
882 	}
883 
884 	/*
885 	 * Initialize realtime fields in the mount structure
886 	 */
887 	if ((error = xfs_rtmount_init(mp))) {
888 		cmn_err(CE_WARN, "XFS: RT mount failed");
889 		goto error1;
890 	}
891 
892 	/*
893 	 * For client case we are done now
894 	 */
895 	if (mfsi_flags & XFS_MFSI_CLIENT) {
896 		return(0);
897 	}
898 
899 	/*
900 	 *  Copies the low order bits of the timestamp and the randomly
901 	 *  set "sequence" number out of a UUID.
902 	 */
903 	uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
904 
905 	/*
906 	 *  The vfs structure needs to have a file system independent
907 	 *  way of checking for the invariant file system ID.  Since it
908 	 *  can't look at mount structures it has a pointer to the data
909 	 *  in the mount structure.
910 	 *
911 	 *  File systems that don't support user level file handles (i.e.
912 	 *  all of them except for XFS) will leave vfs_altfsid as NULL.
913 	 */
914 	vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
915 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
916 
917 	/*
918 	 * Select the right directory manager.
919 	 */
920 	mp->m_dirops =
921 		XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
922 			xfsv2_dirops :
923 			xfsv1_dirops;
924 
925 	/*
926 	 * Initialize directory manager's entries.
927 	 */
928 	XFS_DIR_MOUNT(mp);
929 
930 	/*
931 	 * Initialize the attribute manager's entries.
932 	 */
933 	mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
934 
935 	/*
936 	 * Initialize the precomputed transaction reservations values.
937 	 */
938 	xfs_trans_init(mp);
939 
940 	/*
941 	 * Allocate and initialize the inode hash table for this
942 	 * file system.
943 	 */
944 	xfs_ihash_init(mp);
945 	xfs_chash_init(mp);
946 
947 	/*
948 	 * Allocate and initialize the per-ag data.
949 	 */
950 	init_rwsem(&mp->m_peraglock);
951 	mp->m_perag =
952 		kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
953 
954 	mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
955 
956 	/*
957 	 * log's mount-time initialization. Perform 1st part recovery if needed
958 	 */
959 	if (likely(sbp->sb_logblocks > 0)) {	/* check for volume case */
960 		error = xfs_log_mount(mp, mp->m_logdev_targp,
961 				      XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
962 				      XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
963 		if (error) {
964 			cmn_err(CE_WARN, "XFS: log mount failed");
965 			goto error2;
966 		}
967 	} else {	/* No log has been defined */
968 		cmn_err(CE_WARN, "XFS: no log defined");
969 		XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
970 		error = XFS_ERROR(EFSCORRUPTED);
971 		goto error2;
972 	}
973 
974 	/*
975 	 * Get and sanity-check the root inode.
976 	 * Save the pointer to it in the mount structure.
977 	 */
978 	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
979 	if (error) {
980 		cmn_err(CE_WARN, "XFS: failed to read root inode");
981 		goto error3;
982 	}
983 
984 	ASSERT(rip != NULL);
985 	rvp = XFS_ITOV(rip);
986 
987 	if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
988 		cmn_err(CE_WARN, "XFS: corrupted root inode");
989 		prdev("Root inode %llu is not a directory",
990 		      mp->m_ddev_targp, (unsigned long long)rip->i_ino);
991 		xfs_iunlock(rip, XFS_ILOCK_EXCL);
992 		XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
993 				 mp);
994 		error = XFS_ERROR(EFSCORRUPTED);
995 		goto error4;
996 	}
997 	mp->m_rootip = rip;	/* save it */
998 
999 	xfs_iunlock(rip, XFS_ILOCK_EXCL);
1000 
1001 	/*
1002 	 * Initialize realtime inode pointers in the mount structure
1003 	 */
1004 	if ((error = xfs_rtmount_inodes(mp))) {
1005 		/*
1006 		 * Free up the root inode.
1007 		 */
1008 		cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1009 		goto error4;
1010 	}
1011 
1012 	/*
1013 	 * If fs is not mounted readonly, then update the superblock
1014 	 * unit and width changes.
1015 	 */
1016 	if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1017 		xfs_mount_log_sbunit(mp, update_flags);
1018 
1019 	/*
1020 	 * Initialise the XFS quota management subsystem for this mount
1021 	 */
1022 	if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1023 		goto error4;
1024 
1025 	/*
1026 	 * Finish recovering the file system.  This part needed to be
1027 	 * delayed until after the root and real-time bitmap inodes
1028 	 * were consistently read in.
1029 	 */
1030 	error = xfs_log_mount_finish(mp, mfsi_flags);
1031 	if (error) {
1032 		cmn_err(CE_WARN, "XFS: log mount finish failed");
1033 		goto error4;
1034 	}
1035 
1036 	/*
1037 	 * Complete the quota initialisation, post-log-replay component.
1038 	 */
1039 	if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1040 		goto error4;
1041 
1042 	return 0;
1043 
1044  error4:
1045 	/*
1046 	 * Free up the root inode.
1047 	 */
1048 	VN_RELE(rvp);
1049  error3:
1050 	xfs_log_unmount_dealloc(mp);
1051  error2:
1052 	xfs_ihash_free(mp);
1053 	xfs_chash_free(mp);
1054 	for (agno = 0; agno < sbp->sb_agcount; agno++)
1055 		if (mp->m_perag[agno].pagb_list)
1056 			kmem_free(mp->m_perag[agno].pagb_list,
1057 			  sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1058 	kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1059 	mp->m_perag = NULL;
1060 	/* FALLTHROUGH */
1061  error1:
1062 	if (uuid_mounted)
1063 		xfs_uuid_unmount(mp);
1064 	xfs_freesb(mp);
1065 	return error;
1066 }
1067 
1068 /*
1069  * xfs_unmountfs
1070  *
1071  * This flushes out the inodes,dquots and the superblock, unmounts the
1072  * log and makes sure that incore structures are freed.
1073  */
1074 int
1075 xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1076 {
1077 	struct vfs	*vfsp = XFS_MTOVFS(mp);
1078 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1079 	int64_t		fsid;
1080 #endif
1081 
1082 	xfs_iflush_all(mp, XFS_FLUSH_ALL);
1083 
1084 	XFS_QM_DQPURGEALL(mp,
1085 		XFS_QMOPT_UQUOTA | XFS_QMOPT_GQUOTA | XFS_QMOPT_UMOUNTING);
1086 
1087 	/*
1088 	 * Flush out the log synchronously so that we know for sure
1089 	 * that nothing is pinned.  This is important because bflush()
1090 	 * will skip pinned buffers.
1091 	 */
1092 	xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1093 
1094 	xfs_binval(mp->m_ddev_targp);
1095 	if (mp->m_rtdev_targp) {
1096 		xfs_binval(mp->m_rtdev_targp);
1097 	}
1098 
1099 	xfs_unmountfs_writesb(mp);
1100 
1101 	xfs_unmountfs_wait(mp); 		/* wait for async bufs */
1102 
1103 	xfs_log_unmount(mp);			/* Done! No more fs ops. */
1104 
1105 	xfs_freesb(mp);
1106 
1107 	/*
1108 	 * All inodes from this mount point should be freed.
1109 	 */
1110 	ASSERT(mp->m_inodes == NULL);
1111 
1112 	/*
1113 	 * We may have bufs that are in the process of getting written still.
1114 	 * We must wait for the I/O completion of those. The sync flag here
1115 	 * does a two pass iteration thru the bufcache.
1116 	 */
1117 	if (XFS_FORCED_SHUTDOWN(mp)) {
1118 		xfs_incore_relse(mp->m_ddev_targp, 0, 1); /* synchronous */
1119 	}
1120 
1121 	xfs_unmountfs_close(mp, cr);
1122 	if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1123 		xfs_uuid_unmount(mp);
1124 
1125 #if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1126 	/*
1127 	 * clear all error tags on this filesystem
1128 	 */
1129 	memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1130 	xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1131 #endif
1132 	XFS_IODONE(vfsp);
1133 	xfs_mount_free(mp, 1);
1134 	return 0;
1135 }
1136 
1137 void
1138 xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1139 {
1140 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1141 		xfs_free_buftarg(mp->m_logdev_targp, 1);
1142 	if (mp->m_rtdev_targp)
1143 		xfs_free_buftarg(mp->m_rtdev_targp, 1);
1144 	xfs_free_buftarg(mp->m_ddev_targp, 0);
1145 }
1146 
1147 void
1148 xfs_unmountfs_wait(xfs_mount_t *mp)
1149 {
1150 	if (mp->m_logdev_targp != mp->m_ddev_targp)
1151 		xfs_wait_buftarg(mp->m_logdev_targp);
1152 	if (mp->m_rtdev_targp)
1153 		xfs_wait_buftarg(mp->m_rtdev_targp);
1154 	xfs_wait_buftarg(mp->m_ddev_targp);
1155 }
1156 
1157 int
1158 xfs_unmountfs_writesb(xfs_mount_t *mp)
1159 {
1160 	xfs_buf_t	*sbp;
1161 	xfs_sb_t	*sb;
1162 	int		error = 0;
1163 
1164 	/*
1165 	 * skip superblock write if fs is read-only, or
1166 	 * if we are doing a forced umount.
1167 	 */
1168 	sbp = xfs_getsb(mp, 0);
1169 	if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1170 		XFS_FORCED_SHUTDOWN(mp))) {
1171 		/*
1172 		 * mark shared-readonly if desired
1173 		 */
1174 		sb = XFS_BUF_TO_SBP(sbp);
1175 		if (mp->m_mk_sharedro) {
1176 			if (!(sb->sb_flags & XFS_SBF_READONLY))
1177 				sb->sb_flags |= XFS_SBF_READONLY;
1178 			if (!XFS_SB_VERSION_HASSHARED(sb))
1179 				XFS_SB_VERSION_ADDSHARED(sb);
1180 			xfs_fs_cmn_err(CE_NOTE, mp,
1181 				"Unmounting, marking shared read-only");
1182 		}
1183 		XFS_BUF_UNDONE(sbp);
1184 		XFS_BUF_UNREAD(sbp);
1185 		XFS_BUF_UNDELAYWRITE(sbp);
1186 		XFS_BUF_WRITE(sbp);
1187 		XFS_BUF_UNASYNC(sbp);
1188 		ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1189 		xfsbdstrat(mp, sbp);
1190 		/* Nevermind errors we might get here. */
1191 		error = xfs_iowait(sbp);
1192 		if (error)
1193 			xfs_ioerror_alert("xfs_unmountfs_writesb",
1194 					  mp, sbp, XFS_BUF_ADDR(sbp));
1195 		if (error && mp->m_mk_sharedro)
1196 			xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting.  Filesystem may not be marked shared readonly");
1197 	}
1198 	xfs_buf_relse(sbp);
1199 	return (error);
1200 }
1201 
1202 /*
1203  * xfs_mod_sb() can be used to copy arbitrary changes to the
1204  * in-core superblock into the superblock buffer to be logged.
1205  * It does not provide the higher level of locking that is
1206  * needed to protect the in-core superblock from concurrent
1207  * access.
1208  */
1209 void
1210 xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1211 {
1212 	xfs_buf_t	*bp;
1213 	int		first;
1214 	int		last;
1215 	xfs_mount_t	*mp;
1216 	xfs_sb_t	*sbp;
1217 	xfs_sb_field_t	f;
1218 
1219 	ASSERT(fields);
1220 	if (!fields)
1221 		return;
1222 	mp = tp->t_mountp;
1223 	bp = xfs_trans_getsb(tp, mp, 0);
1224 	sbp = XFS_BUF_TO_SBP(bp);
1225 	first = sizeof(xfs_sb_t);
1226 	last = 0;
1227 
1228 	/* translate/copy */
1229 
1230 	xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, ARCH_CONVERT, fields);
1231 
1232 	/* find modified range */
1233 
1234 	f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1235 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1236 	first = xfs_sb_info[f].offset;
1237 
1238 	f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1239 	ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1240 	last = xfs_sb_info[f + 1].offset - 1;
1241 
1242 	xfs_trans_log_buf(tp, bp, first, last);
1243 }
1244 
1245 /*
1246  * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1247  * a delta to a specified field in the in-core superblock.  Simply
1248  * switch on the field indicated and apply the delta to that field.
1249  * Fields are not allowed to dip below zero, so if the delta would
1250  * do this do not apply it and return EINVAL.
1251  *
1252  * The SB_LOCK must be held when this routine is called.
1253  */
1254 STATIC int
1255 xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1256 			int delta, int rsvd)
1257 {
1258 	int		scounter;	/* short counter for 32 bit fields */
1259 	long long	lcounter;	/* long counter for 64 bit fields */
1260 	long long	res_used, rem;
1261 
1262 	/*
1263 	 * With the in-core superblock spin lock held, switch
1264 	 * on the indicated field.  Apply the delta to the
1265 	 * proper field.  If the fields value would dip below
1266 	 * 0, then do not apply the delta and return EINVAL.
1267 	 */
1268 	switch (field) {
1269 	case XFS_SBS_ICOUNT:
1270 		lcounter = (long long)mp->m_sb.sb_icount;
1271 		lcounter += delta;
1272 		if (lcounter < 0) {
1273 			ASSERT(0);
1274 			return (XFS_ERROR(EINVAL));
1275 		}
1276 		mp->m_sb.sb_icount = lcounter;
1277 		return (0);
1278 	case XFS_SBS_IFREE:
1279 		lcounter = (long long)mp->m_sb.sb_ifree;
1280 		lcounter += delta;
1281 		if (lcounter < 0) {
1282 			ASSERT(0);
1283 			return (XFS_ERROR(EINVAL));
1284 		}
1285 		mp->m_sb.sb_ifree = lcounter;
1286 		return (0);
1287 	case XFS_SBS_FDBLOCKS:
1288 
1289 		lcounter = (long long)mp->m_sb.sb_fdblocks;
1290 		res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1291 
1292 		if (delta > 0) {		/* Putting blocks back */
1293 			if (res_used > delta) {
1294 				mp->m_resblks_avail += delta;
1295 			} else {
1296 				rem = delta - res_used;
1297 				mp->m_resblks_avail = mp->m_resblks;
1298 				lcounter += rem;
1299 			}
1300 		} else {				/* Taking blocks away */
1301 
1302 			lcounter += delta;
1303 
1304 		/*
1305 		 * If were out of blocks, use any available reserved blocks if
1306 		 * were allowed to.
1307 		 */
1308 
1309 			if (lcounter < 0) {
1310 				if (rsvd) {
1311 					lcounter = (long long)mp->m_resblks_avail + delta;
1312 					if (lcounter < 0) {
1313 						return (XFS_ERROR(ENOSPC));
1314 					}
1315 					mp->m_resblks_avail = lcounter;
1316 					return (0);
1317 				} else {	/* not reserved */
1318 					return (XFS_ERROR(ENOSPC));
1319 				}
1320 			}
1321 		}
1322 
1323 		mp->m_sb.sb_fdblocks = lcounter;
1324 		return (0);
1325 	case XFS_SBS_FREXTENTS:
1326 		lcounter = (long long)mp->m_sb.sb_frextents;
1327 		lcounter += delta;
1328 		if (lcounter < 0) {
1329 			return (XFS_ERROR(ENOSPC));
1330 		}
1331 		mp->m_sb.sb_frextents = lcounter;
1332 		return (0);
1333 	case XFS_SBS_DBLOCKS:
1334 		lcounter = (long long)mp->m_sb.sb_dblocks;
1335 		lcounter += delta;
1336 		if (lcounter < 0) {
1337 			ASSERT(0);
1338 			return (XFS_ERROR(EINVAL));
1339 		}
1340 		mp->m_sb.sb_dblocks = lcounter;
1341 		return (0);
1342 	case XFS_SBS_AGCOUNT:
1343 		scounter = mp->m_sb.sb_agcount;
1344 		scounter += delta;
1345 		if (scounter < 0) {
1346 			ASSERT(0);
1347 			return (XFS_ERROR(EINVAL));
1348 		}
1349 		mp->m_sb.sb_agcount = scounter;
1350 		return (0);
1351 	case XFS_SBS_IMAX_PCT:
1352 		scounter = mp->m_sb.sb_imax_pct;
1353 		scounter += delta;
1354 		if (scounter < 0) {
1355 			ASSERT(0);
1356 			return (XFS_ERROR(EINVAL));
1357 		}
1358 		mp->m_sb.sb_imax_pct = scounter;
1359 		return (0);
1360 	case XFS_SBS_REXTSIZE:
1361 		scounter = mp->m_sb.sb_rextsize;
1362 		scounter += delta;
1363 		if (scounter < 0) {
1364 			ASSERT(0);
1365 			return (XFS_ERROR(EINVAL));
1366 		}
1367 		mp->m_sb.sb_rextsize = scounter;
1368 		return (0);
1369 	case XFS_SBS_RBMBLOCKS:
1370 		scounter = mp->m_sb.sb_rbmblocks;
1371 		scounter += delta;
1372 		if (scounter < 0) {
1373 			ASSERT(0);
1374 			return (XFS_ERROR(EINVAL));
1375 		}
1376 		mp->m_sb.sb_rbmblocks = scounter;
1377 		return (0);
1378 	case XFS_SBS_RBLOCKS:
1379 		lcounter = (long long)mp->m_sb.sb_rblocks;
1380 		lcounter += delta;
1381 		if (lcounter < 0) {
1382 			ASSERT(0);
1383 			return (XFS_ERROR(EINVAL));
1384 		}
1385 		mp->m_sb.sb_rblocks = lcounter;
1386 		return (0);
1387 	case XFS_SBS_REXTENTS:
1388 		lcounter = (long long)mp->m_sb.sb_rextents;
1389 		lcounter += delta;
1390 		if (lcounter < 0) {
1391 			ASSERT(0);
1392 			return (XFS_ERROR(EINVAL));
1393 		}
1394 		mp->m_sb.sb_rextents = lcounter;
1395 		return (0);
1396 	case XFS_SBS_REXTSLOG:
1397 		scounter = mp->m_sb.sb_rextslog;
1398 		scounter += delta;
1399 		if (scounter < 0) {
1400 			ASSERT(0);
1401 			return (XFS_ERROR(EINVAL));
1402 		}
1403 		mp->m_sb.sb_rextslog = scounter;
1404 		return (0);
1405 	default:
1406 		ASSERT(0);
1407 		return (XFS_ERROR(EINVAL));
1408 	}
1409 }
1410 
1411 /*
1412  * xfs_mod_incore_sb() is used to change a field in the in-core
1413  * superblock structure by the specified delta.  This modification
1414  * is protected by the SB_LOCK.  Just use the xfs_mod_incore_sb_unlocked()
1415  * routine to do the work.
1416  */
1417 int
1418 xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1419 {
1420 	unsigned long	s;
1421 	int	status;
1422 
1423 	s = XFS_SB_LOCK(mp);
1424 	status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1425 	XFS_SB_UNLOCK(mp, s);
1426 	return (status);
1427 }
1428 
1429 /*
1430  * xfs_mod_incore_sb_batch() is used to change more than one field
1431  * in the in-core superblock structure at a time.  This modification
1432  * is protected by a lock internal to this module.  The fields and
1433  * changes to those fields are specified in the array of xfs_mod_sb
1434  * structures passed in.
1435  *
1436  * Either all of the specified deltas will be applied or none of
1437  * them will.  If any modified field dips below 0, then all modifications
1438  * will be backed out and EINVAL will be returned.
1439  */
1440 int
1441 xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1442 {
1443 	unsigned long	s;
1444 	int		status=0;
1445 	xfs_mod_sb_t	*msbp;
1446 
1447 	/*
1448 	 * Loop through the array of mod structures and apply each
1449 	 * individually.  If any fail, then back out all those
1450 	 * which have already been applied.  Do all of this within
1451 	 * the scope of the SB_LOCK so that all of the changes will
1452 	 * be atomic.
1453 	 */
1454 	s = XFS_SB_LOCK(mp);
1455 	msbp = &msb[0];
1456 	for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1457 		/*
1458 		 * Apply the delta at index n.  If it fails, break
1459 		 * from the loop so we'll fall into the undo loop
1460 		 * below.
1461 		 */
1462 		status = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1463 						    msbp->msb_delta, rsvd);
1464 		if (status != 0) {
1465 			break;
1466 		}
1467 	}
1468 
1469 	/*
1470 	 * If we didn't complete the loop above, then back out
1471 	 * any changes made to the superblock.  If you add code
1472 	 * between the loop above and here, make sure that you
1473 	 * preserve the value of status. Loop back until
1474 	 * we step below the beginning of the array.  Make sure
1475 	 * we don't touch anything back there.
1476 	 */
1477 	if (status != 0) {
1478 		msbp--;
1479 		while (msbp >= msb) {
1480 			status = xfs_mod_incore_sb_unlocked(mp,
1481 				    msbp->msb_field, -(msbp->msb_delta), rsvd);
1482 			ASSERT(status == 0);
1483 			msbp--;
1484 		}
1485 	}
1486 	XFS_SB_UNLOCK(mp, s);
1487 	return (status);
1488 }
1489 
1490 /*
1491  * xfs_getsb() is called to obtain the buffer for the superblock.
1492  * The buffer is returned locked and read in from disk.
1493  * The buffer should be released with a call to xfs_brelse().
1494  *
1495  * If the flags parameter is BUF_TRYLOCK, then we'll only return
1496  * the superblock buffer if it can be locked without sleeping.
1497  * If it can't then we'll return NULL.
1498  */
1499 xfs_buf_t *
1500 xfs_getsb(
1501 	xfs_mount_t	*mp,
1502 	int		flags)
1503 {
1504 	xfs_buf_t	*bp;
1505 
1506 	ASSERT(mp->m_sb_bp != NULL);
1507 	bp = mp->m_sb_bp;
1508 	if (flags & XFS_BUF_TRYLOCK) {
1509 		if (!XFS_BUF_CPSEMA(bp)) {
1510 			return NULL;
1511 		}
1512 	} else {
1513 		XFS_BUF_PSEMA(bp, PRIBIO);
1514 	}
1515 	XFS_BUF_HOLD(bp);
1516 	ASSERT(XFS_BUF_ISDONE(bp));
1517 	return (bp);
1518 }
1519 
1520 /*
1521  * Used to free the superblock along various error paths.
1522  */
1523 void
1524 xfs_freesb(
1525 	xfs_mount_t	*mp)
1526 {
1527 	xfs_buf_t	*bp;
1528 
1529 	/*
1530 	 * Use xfs_getsb() so that the buffer will be locked
1531 	 * when we call xfs_buf_relse().
1532 	 */
1533 	bp = xfs_getsb(mp, 0);
1534 	XFS_BUF_UNMANAGE(bp);
1535 	xfs_buf_relse(bp);
1536 	mp->m_sb_bp = NULL;
1537 }
1538 
1539 /*
1540  * See if the UUID is unique among mounted XFS filesystems.
1541  * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1542  */
1543 STATIC int
1544 xfs_uuid_mount(
1545 	xfs_mount_t	*mp)
1546 {
1547 	if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1548 		cmn_err(CE_WARN,
1549 			"XFS: Filesystem %s has nil UUID - can't mount",
1550 			mp->m_fsname);
1551 		return -1;
1552 	}
1553 	if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1554 		cmn_err(CE_WARN,
1555 			"XFS: Filesystem %s has duplicate UUID - can't mount",
1556 			mp->m_fsname);
1557 		return -1;
1558 	}
1559 	return 0;
1560 }
1561 
1562 /*
1563  * Remove filesystem from the UUID table.
1564  */
1565 STATIC void
1566 xfs_uuid_unmount(
1567 	xfs_mount_t	*mp)
1568 {
1569 	uuid_table_remove(&mp->m_sb.sb_uuid);
1570 }
1571 
1572 /*
1573  * Used to log changes to the superblock unit and width fields which could
1574  * be altered by the mount options. Only the first superblock is updated.
1575  */
1576 STATIC void
1577 xfs_mount_log_sbunit(
1578 	xfs_mount_t	*mp,
1579 	__int64_t	fields)
1580 {
1581 	xfs_trans_t	*tp;
1582 
1583 	ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1584 
1585 	tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1586 	if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1587 				XFS_DEFAULT_LOG_COUNT)) {
1588 		xfs_trans_cancel(tp, 0);
1589 		return;
1590 	}
1591 	xfs_mod_sb(tp, fields);
1592 	xfs_trans_commit(tp, 0, NULL);
1593 }
1594