blob: a93ef802db6b099f958becad8514f9ca65b054b8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scottd8cc8902005-11-02 10:34:53 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc. All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
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
64STATIC void xfs_mount_log_sbunit(xfs_mount_t *, __int64_t);
65STATIC int xfs_uuid_mount(xfs_mount_t *);
66STATIC void xfs_uuid_unmount(xfs_mount_t *mp);
Christoph Hellwigba0f32d2005-06-21 15:36:52 +100067STATIC void xfs_unmountfs_wait(xfs_mount_t *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69static 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 */
126xfs_mount_t *
127xfs_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 */
152void
153xfs_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 */
198STATIC int
199xfs_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 /*
Nathan Scottde206142005-05-05 13:24:13 -0700305 * 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 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 * 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,
Christoph Hellwigda1650a2005-11-02 10:21:35 +1100321 "XFS: Only page-sized (%ld) or less blocksizes currently work.",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 PAGE_SIZE);
323 return XFS_ERROR(ENOSYS);
324 }
325
326 return 0;
327}
328
329xfs_agnumber_t
330xfs_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 */
400void
401xfs_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 */
465int
466xfs_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 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000559STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560xfs_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:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100587 mp->m_attroffset = XFS_LITINO(mp) -
588 XFS_BMDR_SPACE_CALC(MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 break;
590 case 512:
591 case 1024:
592 case 2048:
Nathan Scottd8cc8902005-11-02 10:34:53 +1100593 mp->m_attroffset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 break;
595 default:
596 ASSERT(0);
597 }
598 ASSERT(mp->m_attroffset < XFS_LITINO(mp));
599
600 for (i = 0; i < 2; i++) {
601 mp->m_alloc_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
602 xfs_alloc, i == 0);
603 mp->m_alloc_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
604 xfs_alloc, i == 0);
605 }
606 for (i = 0; i < 2; i++) {
607 mp->m_bmap_dmxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
608 xfs_bmbt, i == 0);
609 mp->m_bmap_dmnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
610 xfs_bmbt, i == 0);
611 }
612 for (i = 0; i < 2; i++) {
613 mp->m_inobt_mxr[i] = XFS_BTREE_BLOCK_MAXRECS(sbp->sb_blocksize,
614 xfs_inobt, i == 0);
615 mp->m_inobt_mnr[i] = XFS_BTREE_BLOCK_MINRECS(sbp->sb_blocksize,
616 xfs_inobt, i == 0);
617 }
618
619 mp->m_bsize = XFS_FSB_TO_BB(mp, 1);
620 mp->m_ialloc_inos = (int)MAX((__uint16_t)XFS_INODES_PER_CHUNK,
621 sbp->sb_inopblock);
622 mp->m_ialloc_blks = mp->m_ialloc_inos >> sbp->sb_inopblog;
623}
624/*
625 * xfs_mountfs
626 *
627 * This function does the following on an initial mount of a file system:
628 * - reads the superblock from disk and init the mount struct
629 * - if we're a 32-bit kernel, do a size check on the superblock
630 * so we don't mount terabyte filesystems
631 * - init mount struct realtime fields
632 * - allocate inode hash table for fs
633 * - init directory manager
634 * - perform recovery and init the log manager
635 */
636int
637xfs_mountfs(
638 vfs_t *vfsp,
639 xfs_mount_t *mp,
640 int mfsi_flags)
641{
642 xfs_buf_t *bp;
643 xfs_sb_t *sbp = &(mp->m_sb);
644 xfs_inode_t *rip;
645 vnode_t *rvp = NULL;
646 int readio_log, writeio_log;
647 xfs_daddr_t d;
648 __uint64_t ret64;
649 __int64_t update_flags;
650 uint quotamount, quotaflags;
651 int agno;
652 int uuid_mounted = 0;
653 int error = 0;
654
655 if (mp->m_sb_bp == NULL) {
656 if ((error = xfs_readsb(mp))) {
657 return (error);
658 }
659 }
660 xfs_mount_common(mp, sbp);
661
662 /*
663 * Check if sb_agblocks is aligned at stripe boundary
664 * If sb_agblocks is NOT aligned turn off m_dalign since
665 * allocator alignment is within an ag, therefore ag has
666 * to be aligned at stripe boundary.
667 */
668 update_flags = 0LL;
669 if (mp->m_dalign && !(mfsi_flags & XFS_MFSI_SECOND)) {
670 /*
671 * If stripe unit and stripe width are not multiples
672 * of the fs blocksize turn off alignment.
673 */
674 if ((BBTOB(mp->m_dalign) & mp->m_blockmask) ||
675 (BBTOB(mp->m_swidth) & mp->m_blockmask)) {
676 if (mp->m_flags & XFS_MOUNT_RETERR) {
677 cmn_err(CE_WARN,
678 "XFS: alignment check 1 failed");
679 error = XFS_ERROR(EINVAL);
680 goto error1;
681 }
682 mp->m_dalign = mp->m_swidth = 0;
683 } else {
684 /*
685 * Convert the stripe unit and width to FSBs.
686 */
687 mp->m_dalign = XFS_BB_TO_FSBT(mp, mp->m_dalign);
688 if (mp->m_dalign && (sbp->sb_agblocks % mp->m_dalign)) {
689 if (mp->m_flags & XFS_MOUNT_RETERR) {
690 error = XFS_ERROR(EINVAL);
691 goto error1;
692 }
693 xfs_fs_cmn_err(CE_WARN, mp,
694"stripe alignment turned off: sunit(%d)/swidth(%d) incompatible with agsize(%d)",
695 mp->m_dalign, mp->m_swidth,
696 sbp->sb_agblocks);
697
698 mp->m_dalign = 0;
699 mp->m_swidth = 0;
700 } else if (mp->m_dalign) {
701 mp->m_swidth = XFS_BB_TO_FSBT(mp, mp->m_swidth);
702 } else {
703 if (mp->m_flags & XFS_MOUNT_RETERR) {
704 xfs_fs_cmn_err(CE_WARN, mp,
705"stripe alignment turned off: sunit(%d) less than bsize(%d)",
706 mp->m_dalign,
707 mp->m_blockmask +1);
708 error = XFS_ERROR(EINVAL);
709 goto error1;
710 }
711 mp->m_swidth = 0;
712 }
713 }
714
715 /*
716 * Update superblock with new values
717 * and log changes
718 */
719 if (XFS_SB_VERSION_HASDALIGN(sbp)) {
720 if (sbp->sb_unit != mp->m_dalign) {
721 sbp->sb_unit = mp->m_dalign;
722 update_flags |= XFS_SB_UNIT;
723 }
724 if (sbp->sb_width != mp->m_swidth) {
725 sbp->sb_width = mp->m_swidth;
726 update_flags |= XFS_SB_WIDTH;
727 }
728 }
729 } else if ((mp->m_flags & XFS_MOUNT_NOALIGN) != XFS_MOUNT_NOALIGN &&
730 XFS_SB_VERSION_HASDALIGN(&mp->m_sb)) {
731 mp->m_dalign = sbp->sb_unit;
732 mp->m_swidth = sbp->sb_width;
733 }
734
735 xfs_alloc_compute_maxlevels(mp);
736 xfs_bmap_compute_maxlevels(mp, XFS_DATA_FORK);
737 xfs_bmap_compute_maxlevels(mp, XFS_ATTR_FORK);
738 xfs_ialloc_compute_maxlevels(mp);
739
740 if (sbp->sb_imax_pct) {
741 __uint64_t icount;
742
743 /* Make sure the maximum inode count is a multiple of the
744 * units we allocate inodes in.
745 */
746
747 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
748 do_div(icount, 100);
749 do_div(icount, mp->m_ialloc_blks);
750 mp->m_maxicount = (icount * mp->m_ialloc_blks) <<
751 sbp->sb_inopblog;
752 } else
753 mp->m_maxicount = 0;
754
755 mp->m_maxioffset = xfs_max_file_offset(sbp->sb_blocklog);
756
757 /*
758 * XFS uses the uuid from the superblock as the unique
759 * identifier for fsid. We can not use the uuid from the volume
760 * since a single partition filesystem is identical to a single
761 * partition volume/filesystem.
762 */
763 if ((mfsi_flags & XFS_MFSI_SECOND) == 0 &&
764 (mp->m_flags & XFS_MOUNT_NOUUID) == 0) {
765 if (xfs_uuid_mount(mp)) {
766 error = XFS_ERROR(EINVAL);
767 goto error1;
768 }
769 uuid_mounted=1;
770 ret64 = uuid_hash64(&sbp->sb_uuid);
771 memcpy(&vfsp->vfs_fsid, &ret64, sizeof(ret64));
772 }
773
774 /*
775 * Set the default minimum read and write sizes unless
776 * already specified in a mount option.
777 * We use smaller I/O sizes when the file system
778 * is being used for NFS service (wsync mount option).
779 */
780 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
781 if (mp->m_flags & XFS_MOUNT_WSYNC) {
782 readio_log = XFS_WSYNC_READIO_LOG;
783 writeio_log = XFS_WSYNC_WRITEIO_LOG;
784 } else {
785 readio_log = XFS_READIO_LOG_LARGE;
786 writeio_log = XFS_WRITEIO_LOG_LARGE;
787 }
788 } else {
789 readio_log = mp->m_readio_log;
790 writeio_log = mp->m_writeio_log;
791 }
792
793 /*
794 * Set the number of readahead buffers to use based on
795 * physical memory size.
796 */
797 if (xfs_physmem <= 4096) /* <= 16MB */
798 mp->m_nreadaheads = XFS_RW_NREADAHEAD_16MB;
799 else if (xfs_physmem <= 8192) /* <= 32MB */
800 mp->m_nreadaheads = XFS_RW_NREADAHEAD_32MB;
801 else
802 mp->m_nreadaheads = XFS_RW_NREADAHEAD_K32;
803 if (sbp->sb_blocklog > readio_log) {
804 mp->m_readio_log = sbp->sb_blocklog;
805 } else {
806 mp->m_readio_log = readio_log;
807 }
808 mp->m_readio_blocks = 1 << (mp->m_readio_log - sbp->sb_blocklog);
809 if (sbp->sb_blocklog > writeio_log) {
810 mp->m_writeio_log = sbp->sb_blocklog;
811 } else {
812 mp->m_writeio_log = writeio_log;
813 }
814 mp->m_writeio_blocks = 1 << (mp->m_writeio_log - sbp->sb_blocklog);
815
816 /*
817 * Set the inode cluster size based on the physical memory
818 * size. This may still be overridden by the file system
819 * block size if it is larger than the chosen cluster size.
820 */
821 if (xfs_physmem <= btoc(32 * 1024 * 1024)) { /* <= 32 MB */
822 mp->m_inode_cluster_size = XFS_INODE_SMALL_CLUSTER_SIZE;
823 } else {
824 mp->m_inode_cluster_size = XFS_INODE_BIG_CLUSTER_SIZE;
825 }
826 /*
827 * Set whether we're using inode alignment.
828 */
829 if (XFS_SB_VERSION_HASALIGN(&mp->m_sb) &&
830 mp->m_sb.sb_inoalignmt >=
831 XFS_B_TO_FSBT(mp, mp->m_inode_cluster_size))
832 mp->m_inoalign_mask = mp->m_sb.sb_inoalignmt - 1;
833 else
834 mp->m_inoalign_mask = 0;
835 /*
836 * If we are using stripe alignment, check whether
837 * the stripe unit is a multiple of the inode alignment
838 */
839 if (mp->m_dalign && mp->m_inoalign_mask &&
840 !(mp->m_dalign & mp->m_inoalign_mask))
841 mp->m_sinoalign = mp->m_dalign;
842 else
843 mp->m_sinoalign = 0;
844 /*
845 * Check that the data (and log if separate) are an ok size.
846 */
847 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks);
848 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_dblocks) {
849 cmn_err(CE_WARN, "XFS: size check 1 failed");
850 error = XFS_ERROR(E2BIG);
851 goto error1;
852 }
853 error = xfs_read_buf(mp, mp->m_ddev_targp,
854 d - XFS_FSS_TO_BB(mp, 1),
855 XFS_FSS_TO_BB(mp, 1), 0, &bp);
856 if (!error) {
857 xfs_buf_relse(bp);
858 } else {
859 cmn_err(CE_WARN, "XFS: size check 2 failed");
860 if (error == ENOSPC) {
861 error = XFS_ERROR(E2BIG);
862 }
863 goto error1;
864 }
865
866 if (((mfsi_flags & XFS_MFSI_CLIENT) == 0) &&
867 mp->m_logdev_targp != mp->m_ddev_targp) {
868 d = (xfs_daddr_t)XFS_FSB_TO_BB(mp, mp->m_sb.sb_logblocks);
869 if (XFS_BB_TO_FSB(mp, d) != mp->m_sb.sb_logblocks) {
870 cmn_err(CE_WARN, "XFS: size check 3 failed");
871 error = XFS_ERROR(E2BIG);
872 goto error1;
873 }
874 error = xfs_read_buf(mp, mp->m_logdev_targp,
875 d - XFS_FSB_TO_BB(mp, 1),
876 XFS_FSB_TO_BB(mp, 1), 0, &bp);
877 if (!error) {
878 xfs_buf_relse(bp);
879 } else {
880 cmn_err(CE_WARN, "XFS: size check 3 failed");
881 if (error == ENOSPC) {
882 error = XFS_ERROR(E2BIG);
883 }
884 goto error1;
885 }
886 }
887
888 /*
889 * Initialize realtime fields in the mount structure
890 */
891 if ((error = xfs_rtmount_init(mp))) {
892 cmn_err(CE_WARN, "XFS: RT mount failed");
893 goto error1;
894 }
895
896 /*
897 * For client case we are done now
898 */
899 if (mfsi_flags & XFS_MFSI_CLIENT) {
900 return(0);
901 }
902
903 /*
904 * Copies the low order bits of the timestamp and the randomly
905 * set "sequence" number out of a UUID.
906 */
907 uuid_getnodeuniq(&sbp->sb_uuid, mp->m_fixedfsid);
908
909 /*
910 * The vfs structure needs to have a file system independent
911 * way of checking for the invariant file system ID. Since it
912 * can't look at mount structures it has a pointer to the data
913 * in the mount structure.
914 *
915 * File systems that don't support user level file handles (i.e.
916 * all of them except for XFS) will leave vfs_altfsid as NULL.
917 */
918 vfsp->vfs_altfsid = (xfs_fsid_t *)mp->m_fixedfsid;
919 mp->m_dmevmask = 0; /* not persistent; set after each mount */
920
921 /*
922 * Select the right directory manager.
923 */
924 mp->m_dirops =
925 XFS_SB_VERSION_HASDIRV2(&mp->m_sb) ?
926 xfsv2_dirops :
927 xfsv1_dirops;
928
929 /*
930 * Initialize directory manager's entries.
931 */
932 XFS_DIR_MOUNT(mp);
933
934 /*
935 * Initialize the attribute manager's entries.
936 */
937 mp->m_attr_magicpct = (mp->m_sb.sb_blocksize * 37) / 100;
938
939 /*
940 * Initialize the precomputed transaction reservations values.
941 */
942 xfs_trans_init(mp);
943
944 /*
945 * Allocate and initialize the inode hash table for this
946 * file system.
947 */
948 xfs_ihash_init(mp);
949 xfs_chash_init(mp);
950
951 /*
952 * Allocate and initialize the per-ag data.
953 */
954 init_rwsem(&mp->m_peraglock);
955 mp->m_perag =
956 kmem_zalloc(sbp->sb_agcount * sizeof(xfs_perag_t), KM_SLEEP);
957
958 mp->m_maxagi = xfs_initialize_perag(mp, sbp->sb_agcount);
959
960 /*
961 * log's mount-time initialization. Perform 1st part recovery if needed
962 */
963 if (likely(sbp->sb_logblocks > 0)) { /* check for volume case */
964 error = xfs_log_mount(mp, mp->m_logdev_targp,
965 XFS_FSB_TO_DADDR(mp, sbp->sb_logstart),
966 XFS_FSB_TO_BB(mp, sbp->sb_logblocks));
967 if (error) {
968 cmn_err(CE_WARN, "XFS: log mount failed");
969 goto error2;
970 }
971 } else { /* No log has been defined */
972 cmn_err(CE_WARN, "XFS: no log defined");
973 XFS_ERROR_REPORT("xfs_mountfs_int(1)", XFS_ERRLEVEL_LOW, mp);
974 error = XFS_ERROR(EFSCORRUPTED);
975 goto error2;
976 }
977
978 /*
979 * Get and sanity-check the root inode.
980 * Save the pointer to it in the mount structure.
981 */
982 error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
983 if (error) {
984 cmn_err(CE_WARN, "XFS: failed to read root inode");
985 goto error3;
986 }
987
988 ASSERT(rip != NULL);
989 rvp = XFS_ITOV(rip);
990
991 if (unlikely((rip->i_d.di_mode & S_IFMT) != S_IFDIR)) {
992 cmn_err(CE_WARN, "XFS: corrupted root inode");
993 prdev("Root inode %llu is not a directory",
994 mp->m_ddev_targp, (unsigned long long)rip->i_ino);
995 xfs_iunlock(rip, XFS_ILOCK_EXCL);
996 XFS_ERROR_REPORT("xfs_mountfs_int(2)", XFS_ERRLEVEL_LOW,
997 mp);
998 error = XFS_ERROR(EFSCORRUPTED);
999 goto error4;
1000 }
1001 mp->m_rootip = rip; /* save it */
1002
1003 xfs_iunlock(rip, XFS_ILOCK_EXCL);
1004
1005 /*
1006 * Initialize realtime inode pointers in the mount structure
1007 */
1008 if ((error = xfs_rtmount_inodes(mp))) {
1009 /*
1010 * Free up the root inode.
1011 */
1012 cmn_err(CE_WARN, "XFS: failed to read RT inodes");
1013 goto error4;
1014 }
1015
1016 /*
1017 * If fs is not mounted readonly, then update the superblock
1018 * unit and width changes.
1019 */
1020 if (update_flags && !(vfsp->vfs_flag & VFS_RDONLY))
1021 xfs_mount_log_sbunit(mp, update_flags);
1022
1023 /*
1024 * Initialise the XFS quota management subsystem for this mount
1025 */
1026 if ((error = XFS_QM_INIT(mp, &quotamount, &quotaflags)))
1027 goto error4;
1028
1029 /*
1030 * Finish recovering the file system. This part needed to be
1031 * delayed until after the root and real-time bitmap inodes
1032 * were consistently read in.
1033 */
1034 error = xfs_log_mount_finish(mp, mfsi_flags);
1035 if (error) {
1036 cmn_err(CE_WARN, "XFS: log mount finish failed");
1037 goto error4;
1038 }
1039
1040 /*
1041 * Complete the quota initialisation, post-log-replay component.
1042 */
1043 if ((error = XFS_QM_MOUNT(mp, quotamount, quotaflags, mfsi_flags)))
1044 goto error4;
1045
1046 return 0;
1047
1048 error4:
1049 /*
1050 * Free up the root inode.
1051 */
1052 VN_RELE(rvp);
1053 error3:
1054 xfs_log_unmount_dealloc(mp);
1055 error2:
1056 xfs_ihash_free(mp);
1057 xfs_chash_free(mp);
1058 for (agno = 0; agno < sbp->sb_agcount; agno++)
1059 if (mp->m_perag[agno].pagb_list)
1060 kmem_free(mp->m_perag[agno].pagb_list,
1061 sizeof(xfs_perag_busy_t) * XFS_PAGB_NUM_SLOTS);
1062 kmem_free(mp->m_perag, sbp->sb_agcount * sizeof(xfs_perag_t));
1063 mp->m_perag = NULL;
1064 /* FALLTHROUGH */
1065 error1:
1066 if (uuid_mounted)
1067 xfs_uuid_unmount(mp);
1068 xfs_freesb(mp);
1069 return error;
1070}
1071
1072/*
1073 * xfs_unmountfs
1074 *
1075 * This flushes out the inodes,dquots and the superblock, unmounts the
1076 * log and makes sure that incore structures are freed.
1077 */
1078int
1079xfs_unmountfs(xfs_mount_t *mp, struct cred *cr)
1080{
1081 struct vfs *vfsp = XFS_MTOVFS(mp);
1082#if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1083 int64_t fsid;
1084#endif
1085
Christoph Hellwigefa80272005-06-21 15:37:17 +10001086 xfs_iflush_all(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 XFS_QM_DQPURGEALL(mp,
1089 XFS_QMOPT_UQUOTA | XFS_QMOPT_GQUOTA | XFS_QMOPT_UMOUNTING);
1090
1091 /*
1092 * Flush out the log synchronously so that we know for sure
1093 * that nothing is pinned. This is important because bflush()
1094 * will skip pinned buffers.
1095 */
1096 xfs_log_force(mp, (xfs_lsn_t)0, XFS_LOG_FORCE | XFS_LOG_SYNC);
1097
1098 xfs_binval(mp->m_ddev_targp);
1099 if (mp->m_rtdev_targp) {
1100 xfs_binval(mp->m_rtdev_targp);
1101 }
1102
1103 xfs_unmountfs_writesb(mp);
1104
1105 xfs_unmountfs_wait(mp); /* wait for async bufs */
1106
1107 xfs_log_unmount(mp); /* Done! No more fs ops. */
1108
1109 xfs_freesb(mp);
1110
1111 /*
1112 * All inodes from this mount point should be freed.
1113 */
1114 ASSERT(mp->m_inodes == NULL);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 xfs_unmountfs_close(mp, cr);
1117 if ((mp->m_flags & XFS_MOUNT_NOUUID) == 0)
1118 xfs_uuid_unmount(mp);
1119
1120#if defined(DEBUG) || defined(INDUCE_IO_ERROR)
1121 /*
1122 * clear all error tags on this filesystem
1123 */
1124 memcpy(&fsid, &vfsp->vfs_fsid, sizeof(int64_t));
1125 xfs_errortag_clearall_umount(fsid, mp->m_fsname, 0);
1126#endif
1127 XFS_IODONE(vfsp);
1128 xfs_mount_free(mp, 1);
1129 return 0;
1130}
1131
1132void
1133xfs_unmountfs_close(xfs_mount_t *mp, struct cred *cr)
1134{
1135 if (mp->m_logdev_targp != mp->m_ddev_targp)
1136 xfs_free_buftarg(mp->m_logdev_targp, 1);
1137 if (mp->m_rtdev_targp)
1138 xfs_free_buftarg(mp->m_rtdev_targp, 1);
1139 xfs_free_buftarg(mp->m_ddev_targp, 0);
1140}
1141
Christoph Hellwigba0f32d2005-06-21 15:36:52 +10001142STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143xfs_unmountfs_wait(xfs_mount_t *mp)
1144{
1145 if (mp->m_logdev_targp != mp->m_ddev_targp)
1146 xfs_wait_buftarg(mp->m_logdev_targp);
1147 if (mp->m_rtdev_targp)
1148 xfs_wait_buftarg(mp->m_rtdev_targp);
1149 xfs_wait_buftarg(mp->m_ddev_targp);
1150}
1151
1152int
1153xfs_unmountfs_writesb(xfs_mount_t *mp)
1154{
1155 xfs_buf_t *sbp;
1156 xfs_sb_t *sb;
1157 int error = 0;
1158
1159 /*
1160 * skip superblock write if fs is read-only, or
1161 * if we are doing a forced umount.
1162 */
1163 sbp = xfs_getsb(mp, 0);
1164 if (!(XFS_MTOVFS(mp)->vfs_flag & VFS_RDONLY ||
1165 XFS_FORCED_SHUTDOWN(mp))) {
1166 /*
1167 * mark shared-readonly if desired
1168 */
1169 sb = XFS_BUF_TO_SBP(sbp);
1170 if (mp->m_mk_sharedro) {
1171 if (!(sb->sb_flags & XFS_SBF_READONLY))
1172 sb->sb_flags |= XFS_SBF_READONLY;
1173 if (!XFS_SB_VERSION_HASSHARED(sb))
1174 XFS_SB_VERSION_ADDSHARED(sb);
1175 xfs_fs_cmn_err(CE_NOTE, mp,
1176 "Unmounting, marking shared read-only");
1177 }
1178 XFS_BUF_UNDONE(sbp);
1179 XFS_BUF_UNREAD(sbp);
1180 XFS_BUF_UNDELAYWRITE(sbp);
1181 XFS_BUF_WRITE(sbp);
1182 XFS_BUF_UNASYNC(sbp);
1183 ASSERT(XFS_BUF_TARGET(sbp) == mp->m_ddev_targp);
1184 xfsbdstrat(mp, sbp);
1185 /* Nevermind errors we might get here. */
1186 error = xfs_iowait(sbp);
1187 if (error)
1188 xfs_ioerror_alert("xfs_unmountfs_writesb",
1189 mp, sbp, XFS_BUF_ADDR(sbp));
1190 if (error && mp->m_mk_sharedro)
1191 xfs_fs_cmn_err(CE_ALERT, mp, "Superblock write error detected while unmounting. Filesystem may not be marked shared readonly");
1192 }
1193 xfs_buf_relse(sbp);
1194 return (error);
1195}
1196
1197/*
1198 * xfs_mod_sb() can be used to copy arbitrary changes to the
1199 * in-core superblock into the superblock buffer to be logged.
1200 * It does not provide the higher level of locking that is
1201 * needed to protect the in-core superblock from concurrent
1202 * access.
1203 */
1204void
1205xfs_mod_sb(xfs_trans_t *tp, __int64_t fields)
1206{
1207 xfs_buf_t *bp;
1208 int first;
1209 int last;
1210 xfs_mount_t *mp;
1211 xfs_sb_t *sbp;
1212 xfs_sb_field_t f;
1213
1214 ASSERT(fields);
1215 if (!fields)
1216 return;
1217 mp = tp->t_mountp;
1218 bp = xfs_trans_getsb(tp, mp, 0);
1219 sbp = XFS_BUF_TO_SBP(bp);
1220 first = sizeof(xfs_sb_t);
1221 last = 0;
1222
1223 /* translate/copy */
1224
1225 xfs_xlatesb(XFS_BUF_PTR(bp), &(mp->m_sb), -1, fields);
1226
1227 /* find modified range */
1228
1229 f = (xfs_sb_field_t)xfs_lowbit64((__uint64_t)fields);
1230 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1231 first = xfs_sb_info[f].offset;
1232
1233 f = (xfs_sb_field_t)xfs_highbit64((__uint64_t)fields);
1234 ASSERT((1LL << f) & XFS_SB_MOD_BITS);
1235 last = xfs_sb_info[f + 1].offset - 1;
1236
1237 xfs_trans_log_buf(tp, bp, first, last);
1238}
1239
1240/*
1241 * xfs_mod_incore_sb_unlocked() is a utility routine common used to apply
1242 * a delta to a specified field in the in-core superblock. Simply
1243 * switch on the field indicated and apply the delta to that field.
1244 * Fields are not allowed to dip below zero, so if the delta would
1245 * do this do not apply it and return EINVAL.
1246 *
1247 * The SB_LOCK must be held when this routine is called.
1248 */
1249STATIC int
1250xfs_mod_incore_sb_unlocked(xfs_mount_t *mp, xfs_sb_field_t field,
1251 int delta, int rsvd)
1252{
1253 int scounter; /* short counter for 32 bit fields */
1254 long long lcounter; /* long counter for 64 bit fields */
1255 long long res_used, rem;
1256
1257 /*
1258 * With the in-core superblock spin lock held, switch
1259 * on the indicated field. Apply the delta to the
1260 * proper field. If the fields value would dip below
1261 * 0, then do not apply the delta and return EINVAL.
1262 */
1263 switch (field) {
1264 case XFS_SBS_ICOUNT:
1265 lcounter = (long long)mp->m_sb.sb_icount;
1266 lcounter += delta;
1267 if (lcounter < 0) {
1268 ASSERT(0);
1269 return (XFS_ERROR(EINVAL));
1270 }
1271 mp->m_sb.sb_icount = lcounter;
1272 return (0);
1273 case XFS_SBS_IFREE:
1274 lcounter = (long long)mp->m_sb.sb_ifree;
1275 lcounter += delta;
1276 if (lcounter < 0) {
1277 ASSERT(0);
1278 return (XFS_ERROR(EINVAL));
1279 }
1280 mp->m_sb.sb_ifree = lcounter;
1281 return (0);
1282 case XFS_SBS_FDBLOCKS:
1283
1284 lcounter = (long long)mp->m_sb.sb_fdblocks;
1285 res_used = (long long)(mp->m_resblks - mp->m_resblks_avail);
1286
1287 if (delta > 0) { /* Putting blocks back */
1288 if (res_used > delta) {
1289 mp->m_resblks_avail += delta;
1290 } else {
1291 rem = delta - res_used;
1292 mp->m_resblks_avail = mp->m_resblks;
1293 lcounter += rem;
1294 }
1295 } else { /* Taking blocks away */
1296
1297 lcounter += delta;
1298
1299 /*
1300 * If were out of blocks, use any available reserved blocks if
1301 * were allowed to.
1302 */
1303
1304 if (lcounter < 0) {
1305 if (rsvd) {
1306 lcounter = (long long)mp->m_resblks_avail + delta;
1307 if (lcounter < 0) {
1308 return (XFS_ERROR(ENOSPC));
1309 }
1310 mp->m_resblks_avail = lcounter;
1311 return (0);
1312 } else { /* not reserved */
1313 return (XFS_ERROR(ENOSPC));
1314 }
1315 }
1316 }
1317
1318 mp->m_sb.sb_fdblocks = lcounter;
1319 return (0);
1320 case XFS_SBS_FREXTENTS:
1321 lcounter = (long long)mp->m_sb.sb_frextents;
1322 lcounter += delta;
1323 if (lcounter < 0) {
1324 return (XFS_ERROR(ENOSPC));
1325 }
1326 mp->m_sb.sb_frextents = lcounter;
1327 return (0);
1328 case XFS_SBS_DBLOCKS:
1329 lcounter = (long long)mp->m_sb.sb_dblocks;
1330 lcounter += delta;
1331 if (lcounter < 0) {
1332 ASSERT(0);
1333 return (XFS_ERROR(EINVAL));
1334 }
1335 mp->m_sb.sb_dblocks = lcounter;
1336 return (0);
1337 case XFS_SBS_AGCOUNT:
1338 scounter = mp->m_sb.sb_agcount;
1339 scounter += delta;
1340 if (scounter < 0) {
1341 ASSERT(0);
1342 return (XFS_ERROR(EINVAL));
1343 }
1344 mp->m_sb.sb_agcount = scounter;
1345 return (0);
1346 case XFS_SBS_IMAX_PCT:
1347 scounter = mp->m_sb.sb_imax_pct;
1348 scounter += delta;
1349 if (scounter < 0) {
1350 ASSERT(0);
1351 return (XFS_ERROR(EINVAL));
1352 }
1353 mp->m_sb.sb_imax_pct = scounter;
1354 return (0);
1355 case XFS_SBS_REXTSIZE:
1356 scounter = mp->m_sb.sb_rextsize;
1357 scounter += delta;
1358 if (scounter < 0) {
1359 ASSERT(0);
1360 return (XFS_ERROR(EINVAL));
1361 }
1362 mp->m_sb.sb_rextsize = scounter;
1363 return (0);
1364 case XFS_SBS_RBMBLOCKS:
1365 scounter = mp->m_sb.sb_rbmblocks;
1366 scounter += delta;
1367 if (scounter < 0) {
1368 ASSERT(0);
1369 return (XFS_ERROR(EINVAL));
1370 }
1371 mp->m_sb.sb_rbmblocks = scounter;
1372 return (0);
1373 case XFS_SBS_RBLOCKS:
1374 lcounter = (long long)mp->m_sb.sb_rblocks;
1375 lcounter += delta;
1376 if (lcounter < 0) {
1377 ASSERT(0);
1378 return (XFS_ERROR(EINVAL));
1379 }
1380 mp->m_sb.sb_rblocks = lcounter;
1381 return (0);
1382 case XFS_SBS_REXTENTS:
1383 lcounter = (long long)mp->m_sb.sb_rextents;
1384 lcounter += delta;
1385 if (lcounter < 0) {
1386 ASSERT(0);
1387 return (XFS_ERROR(EINVAL));
1388 }
1389 mp->m_sb.sb_rextents = lcounter;
1390 return (0);
1391 case XFS_SBS_REXTSLOG:
1392 scounter = mp->m_sb.sb_rextslog;
1393 scounter += delta;
1394 if (scounter < 0) {
1395 ASSERT(0);
1396 return (XFS_ERROR(EINVAL));
1397 }
1398 mp->m_sb.sb_rextslog = scounter;
1399 return (0);
1400 default:
1401 ASSERT(0);
1402 return (XFS_ERROR(EINVAL));
1403 }
1404}
1405
1406/*
1407 * xfs_mod_incore_sb() is used to change a field in the in-core
1408 * superblock structure by the specified delta. This modification
1409 * is protected by the SB_LOCK. Just use the xfs_mod_incore_sb_unlocked()
1410 * routine to do the work.
1411 */
1412int
1413xfs_mod_incore_sb(xfs_mount_t *mp, xfs_sb_field_t field, int delta, int rsvd)
1414{
1415 unsigned long s;
1416 int status;
1417
1418 s = XFS_SB_LOCK(mp);
1419 status = xfs_mod_incore_sb_unlocked(mp, field, delta, rsvd);
1420 XFS_SB_UNLOCK(mp, s);
1421 return (status);
1422}
1423
1424/*
1425 * xfs_mod_incore_sb_batch() is used to change more than one field
1426 * in the in-core superblock structure at a time. This modification
1427 * is protected by a lock internal to this module. The fields and
1428 * changes to those fields are specified in the array of xfs_mod_sb
1429 * structures passed in.
1430 *
1431 * Either all of the specified deltas will be applied or none of
1432 * them will. If any modified field dips below 0, then all modifications
1433 * will be backed out and EINVAL will be returned.
1434 */
1435int
1436xfs_mod_incore_sb_batch(xfs_mount_t *mp, xfs_mod_sb_t *msb, uint nmsb, int rsvd)
1437{
1438 unsigned long s;
1439 int status=0;
1440 xfs_mod_sb_t *msbp;
1441
1442 /*
1443 * Loop through the array of mod structures and apply each
1444 * individually. If any fail, then back out all those
1445 * which have already been applied. Do all of this within
1446 * the scope of the SB_LOCK so that all of the changes will
1447 * be atomic.
1448 */
1449 s = XFS_SB_LOCK(mp);
1450 msbp = &msb[0];
1451 for (msbp = &msbp[0]; msbp < (msb + nmsb); msbp++) {
1452 /*
1453 * Apply the delta at index n. If it fails, break
1454 * from the loop so we'll fall into the undo loop
1455 * below.
1456 */
1457 status = xfs_mod_incore_sb_unlocked(mp, msbp->msb_field,
1458 msbp->msb_delta, rsvd);
1459 if (status != 0) {
1460 break;
1461 }
1462 }
1463
1464 /*
1465 * If we didn't complete the loop above, then back out
1466 * any changes made to the superblock. If you add code
1467 * between the loop above and here, make sure that you
1468 * preserve the value of status. Loop back until
1469 * we step below the beginning of the array. Make sure
1470 * we don't touch anything back there.
1471 */
1472 if (status != 0) {
1473 msbp--;
1474 while (msbp >= msb) {
1475 status = xfs_mod_incore_sb_unlocked(mp,
1476 msbp->msb_field, -(msbp->msb_delta), rsvd);
1477 ASSERT(status == 0);
1478 msbp--;
1479 }
1480 }
1481 XFS_SB_UNLOCK(mp, s);
1482 return (status);
1483}
1484
1485/*
1486 * xfs_getsb() is called to obtain the buffer for the superblock.
1487 * The buffer is returned locked and read in from disk.
1488 * The buffer should be released with a call to xfs_brelse().
1489 *
1490 * If the flags parameter is BUF_TRYLOCK, then we'll only return
1491 * the superblock buffer if it can be locked without sleeping.
1492 * If it can't then we'll return NULL.
1493 */
1494xfs_buf_t *
1495xfs_getsb(
1496 xfs_mount_t *mp,
1497 int flags)
1498{
1499 xfs_buf_t *bp;
1500
1501 ASSERT(mp->m_sb_bp != NULL);
1502 bp = mp->m_sb_bp;
1503 if (flags & XFS_BUF_TRYLOCK) {
1504 if (!XFS_BUF_CPSEMA(bp)) {
1505 return NULL;
1506 }
1507 } else {
1508 XFS_BUF_PSEMA(bp, PRIBIO);
1509 }
1510 XFS_BUF_HOLD(bp);
1511 ASSERT(XFS_BUF_ISDONE(bp));
1512 return (bp);
1513}
1514
1515/*
1516 * Used to free the superblock along various error paths.
1517 */
1518void
1519xfs_freesb(
1520 xfs_mount_t *mp)
1521{
1522 xfs_buf_t *bp;
1523
1524 /*
1525 * Use xfs_getsb() so that the buffer will be locked
1526 * when we call xfs_buf_relse().
1527 */
1528 bp = xfs_getsb(mp, 0);
1529 XFS_BUF_UNMANAGE(bp);
1530 xfs_buf_relse(bp);
1531 mp->m_sb_bp = NULL;
1532}
1533
1534/*
1535 * See if the UUID is unique among mounted XFS filesystems.
1536 * Mount fails if UUID is nil or a FS with the same UUID is already mounted.
1537 */
1538STATIC int
1539xfs_uuid_mount(
1540 xfs_mount_t *mp)
1541{
1542 if (uuid_is_nil(&mp->m_sb.sb_uuid)) {
1543 cmn_err(CE_WARN,
1544 "XFS: Filesystem %s has nil UUID - can't mount",
1545 mp->m_fsname);
1546 return -1;
1547 }
1548 if (!uuid_table_insert(&mp->m_sb.sb_uuid)) {
1549 cmn_err(CE_WARN,
1550 "XFS: Filesystem %s has duplicate UUID - can't mount",
1551 mp->m_fsname);
1552 return -1;
1553 }
1554 return 0;
1555}
1556
1557/*
1558 * Remove filesystem from the UUID table.
1559 */
1560STATIC void
1561xfs_uuid_unmount(
1562 xfs_mount_t *mp)
1563{
1564 uuid_table_remove(&mp->m_sb.sb_uuid);
1565}
1566
1567/*
1568 * Used to log changes to the superblock unit and width fields which could
1569 * be altered by the mount options. Only the first superblock is updated.
1570 */
1571STATIC void
1572xfs_mount_log_sbunit(
1573 xfs_mount_t *mp,
1574 __int64_t fields)
1575{
1576 xfs_trans_t *tp;
1577
1578 ASSERT(fields & (XFS_SB_UNIT|XFS_SB_WIDTH|XFS_SB_UUID));
1579
1580 tp = xfs_trans_alloc(mp, XFS_TRANS_SB_UNIT);
1581 if (xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
1582 XFS_DEFAULT_LOG_COUNT)) {
1583 xfs_trans_cancel(tp, 0);
1584 return;
1585 }
1586 xfs_mod_sb(tp, fields);
1587 xfs_trans_commit(tp, 0, NULL);
1588}