2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/blkdev.h>
16 #include <linux/kthread.h>
17 #include <linux/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
22 #include <linux/lockdep.h>
38 #include "trace_gfs2.h"
44 * gfs2_tune_init - Fill a gfs2_tune structure with default values
49 static void gfs2_tune_init(struct gfs2_tune
*gt
)
51 spin_lock_init(>
->gt_spin
);
53 gt
->gt_quota_simul_sync
= 64;
54 gt
->gt_quota_warn_period
= 10;
55 gt
->gt_quota_scale_num
= 1;
56 gt
->gt_quota_scale_den
= 1;
57 gt
->gt_new_files_jdata
= 0;
58 gt
->gt_max_readahead
= 1 << 18;
59 gt
->gt_complain_secs
= 10;
62 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
66 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
72 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
73 if (!sdp
->sd_lkstats
) {
78 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
79 gfs2_tune_init(&sdp
->sd_tune
);
81 init_waitqueue_head(&sdp
->sd_glock_wait
);
82 atomic_set(&sdp
->sd_glock_disposal
, 0);
83 init_completion(&sdp
->sd_locking_init
);
84 spin_lock_init(&sdp
->sd_statfs_spin
);
86 spin_lock_init(&sdp
->sd_rindex_spin
);
87 sdp
->sd_rindex_tree
.rb_node
= NULL
;
89 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
90 spin_lock_init(&sdp
->sd_jindex_spin
);
91 mutex_init(&sdp
->sd_jindex_mutex
);
93 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
94 mutex_init(&sdp
->sd_quota_mutex
);
95 init_waitqueue_head(&sdp
->sd_quota_wait
);
96 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
97 spin_lock_init(&sdp
->sd_trunc_lock
);
99 spin_lock_init(&sdp
->sd_log_lock
);
100 atomic_set(&sdp
->sd_log_pinned
, 0);
101 INIT_LIST_HEAD(&sdp
->sd_log_le_buf
);
102 INIT_LIST_HEAD(&sdp
->sd_log_le_revoke
);
103 INIT_LIST_HEAD(&sdp
->sd_log_le_databuf
);
104 INIT_LIST_HEAD(&sdp
->sd_log_le_ordered
);
106 init_waitqueue_head(&sdp
->sd_log_waitq
);
107 init_waitqueue_head(&sdp
->sd_logd_waitq
);
108 spin_lock_init(&sdp
->sd_ail_lock
);
109 INIT_LIST_HEAD(&sdp
->sd_ail1_list
);
110 INIT_LIST_HEAD(&sdp
->sd_ail2_list
);
112 init_rwsem(&sdp
->sd_log_flush_lock
);
113 atomic_set(&sdp
->sd_log_in_flight
, 0);
114 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
116 INIT_LIST_HEAD(&sdp
->sd_revoke_list
);
118 mutex_init(&sdp
->sd_freeze_lock
);
125 * gfs2_check_sb - Check superblock
126 * @sdp: the filesystem
127 * @sb: The superblock
128 * @silent: Don't print a message if the check fails
130 * Checks the version code of the FS is one that we understand how to
131 * read and that the sizes of the various on-disk structures have not
135 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
137 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
139 if (sb
->sb_magic
!= GFS2_MAGIC
||
140 sb
->sb_type
!= GFS2_METATYPE_SB
) {
142 printk(KERN_WARNING
"GFS2: not a GFS2 filesystem\n");
146 /* If format numbers match exactly, we're done. */
148 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
149 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
152 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
157 static void end_bio_io_page(struct bio
*bio
, int error
)
159 struct page
*page
= bio
->bi_private
;
162 SetPageUptodate(page
);
164 printk(KERN_WARNING
"gfs2: error %d reading superblock\n", error
);
168 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
170 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
171 struct super_block
*s
= sdp
->sd_vfs
;
172 const struct gfs2_sb
*str
= buf
;
174 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
175 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
176 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
177 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
178 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
179 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
180 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
181 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
182 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
183 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
184 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
186 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
187 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
188 memcpy(s
->s_uuid
, str
->sb_uuid
, 16);
192 * gfs2_read_super - Read the gfs2 super block from disk
193 * @sdp: The GFS2 super block
194 * @sector: The location of the super block
195 * @error: The error code to return
197 * This uses the bio functions to read the super block from disk
198 * because we want to be 100% sure that we never read cached data.
199 * A super block is read twice only during each GFS2 mount and is
200 * never written to by the filesystem. The first time its read no
201 * locks are held, and the only details which are looked at are those
202 * relating to the locking protocol. Once locking is up and working,
203 * the sb is read again under the lock to establish the location of
204 * the master directory (contains pointers to journals etc) and the
207 * Returns: 0 on success or error
210 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
212 struct super_block
*sb
= sdp
->sd_vfs
;
217 page
= alloc_page(GFP_NOFS
);
221 ClearPageUptodate(page
);
222 ClearPageDirty(page
);
225 bio
= bio_alloc(GFP_NOFS
, 1);
226 bio
->bi_sector
= sector
* (sb
->s_blocksize
>> 9);
227 bio
->bi_bdev
= sb
->s_bdev
;
228 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
230 bio
->bi_end_io
= end_bio_io_page
;
231 bio
->bi_private
= page
;
232 submit_bio(READ_SYNC
| REQ_META
, bio
);
233 wait_on_page_locked(page
);
235 if (!PageUptodate(page
)) {
243 return gfs2_check_sb(sdp
, silent
);
247 * gfs2_read_sb - Read super block
248 * @sdp: The GFS2 superblock
249 * @silent: Don't print message if mount fails
253 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
255 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
260 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
263 fs_err(sdp
, "can't read superblock\n");
267 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
268 GFS2_BASIC_BLOCK_SHIFT
;
269 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
270 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
271 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
272 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
273 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
274 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
275 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
276 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
277 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
278 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
279 sizeof(struct gfs2_meta_header
)) /
280 sizeof(struct gfs2_quota_change
);
282 /* Compute maximum reservation required to add a entry to a directory */
284 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * (1 << GFS2_DIR_MAX_DEPTH
),
288 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
289 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
290 ind_blocks
+= tmp_blocks
;
293 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
295 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
297 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
298 sizeof(struct gfs2_dinode
);
299 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
304 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
306 m
= do_div(d
, sdp
->sd_inptrs
);
308 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
310 sdp
->sd_heightsize
[x
] = space
;
312 sdp
->sd_max_height
= x
;
313 sdp
->sd_heightsize
[x
] = ~0;
314 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
316 sdp
->sd_jheightsize
[0] = sdp
->sd_sb
.sb_bsize
-
317 sizeof(struct gfs2_dinode
);
318 sdp
->sd_jheightsize
[1] = sdp
->sd_jbsize
* sdp
->sd_diptrs
;
323 space
= sdp
->sd_jheightsize
[x
- 1] * sdp
->sd_inptrs
;
325 m
= do_div(d
, sdp
->sd_inptrs
);
327 if (d
!= sdp
->sd_jheightsize
[x
- 1] || m
)
329 sdp
->sd_jheightsize
[x
] = space
;
331 sdp
->sd_max_jheight
= x
;
332 sdp
->sd_jheightsize
[x
] = ~0;
333 gfs2_assert(sdp
, sdp
->sd_max_jheight
<= GFS2_MAX_META_HEIGHT
);
338 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
343 proto
= sdp
->sd_args
.ar_lockproto
;
344 table
= sdp
->sd_args
.ar_locktable
;
346 /* Try to autodetect */
348 if (!proto
[0] || !table
[0]) {
349 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
354 proto
= sdp
->sd_sb
.sb_lockproto
;
356 table
= sdp
->sd_sb
.sb_locktable
;
360 table
= sdp
->sd_vfs
->s_id
;
362 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
363 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
365 table
= sdp
->sd_table_name
;
366 while ((table
= strchr(table
, '/')))
372 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
380 error
= gfs2_glock_nq_num(sdp
,
381 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
382 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
385 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
389 error
= gfs2_glock_nq_num(sdp
,
390 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
392 LM_FLAG_NOEXP
| GL_EXACT
,
395 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
399 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
400 CREATE
, &sdp
->sd_rename_gl
);
402 fs_err(sdp
, "can't create rename glock: %d\n", error
);
406 error
= gfs2_glock_get(sdp
, GFS2_TRANS_LOCK
, &gfs2_trans_glops
,
407 CREATE
, &sdp
->sd_trans_gl
);
409 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
416 gfs2_glock_put(sdp
->sd_trans_gl
);
418 gfs2_glock_put(sdp
->sd_rename_gl
);
420 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
422 gfs2_glock_dq_uninit(mount_gh
);
427 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
428 u64 no_addr
, const char *name
)
430 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
431 struct dentry
*dentry
;
434 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0, 0);
436 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
437 return PTR_ERR(inode
);
439 dentry
= d_make_root(inode
);
441 fs_err(sdp
, "can't alloc %s dentry\n", name
);
448 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
450 struct super_block
*sb
= sdp
->sd_vfs
;
451 struct gfs2_holder sb_gh
;
455 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
456 LM_ST_SHARED
, 0, &sb_gh
);
458 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
462 ret
= gfs2_read_sb(sdp
, silent
);
464 fs_err(sdp
, "can't read superblock: %d\n", ret
);
468 /* Set up the buffer cache and SB for real */
469 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
471 fs_err(sdp
, "FS block size (%u) is too small for device "
473 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
476 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
478 fs_err(sdp
, "FS block size (%u) is too big for machine "
480 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
483 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
485 /* Get the root inode */
486 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
487 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
491 /* Get the master inode */
492 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
493 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
495 dput(sdp
->sd_root_dir
);
498 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
500 gfs2_glock_dq_uninit(&sb_gh
);
505 * map_journal_extents - create a reusable "extent" mapping from all logical
506 * blocks to all physical blocks for the given journal. This will save
507 * us time when writing journal blocks. Most journals will have only one
508 * extent that maps all their logical blocks. That's because gfs2.mkfs
509 * arranges the journal blocks sequentially to maximize performance.
510 * So the extent would map the first block for the entire file length.
511 * However, gfs2_jadd can happen while file activity is happening, so
512 * those journals may not be sequential. Less likely is the case where
513 * the users created their own journals by mounting the metafs and
514 * laying it out. But it's still possible. These journals might have
517 * TODO: This should be done in bigger chunks rather than one block at a time,
518 * but since it's only done at mount time, I'm not worried about the
521 static int map_journal_extents(struct gfs2_sbd
*sdp
)
523 struct gfs2_jdesc
*jd
= sdp
->sd_jdesc
;
525 u64 db
, prev_db
; /* logical block, disk block, prev disk block */
526 struct gfs2_inode
*ip
= GFS2_I(jd
->jd_inode
);
527 struct gfs2_journal_extent
*jext
= NULL
;
528 struct buffer_head bh
;
533 for (lb
= 0; lb
< i_size_read(jd
->jd_inode
) >> sdp
->sd_sb
.sb_bsize_shift
; lb
++) {
536 bh
.b_size
= 1 << ip
->i_inode
.i_blkbits
;
537 rc
= gfs2_block_map(jd
->jd_inode
, lb
, &bh
, 0);
540 printk(KERN_INFO
"GFS2 journal mapping error %d: lb="
541 "%u db=%llu\n", rc
, lb
, (unsigned long long)db
);
544 if (!prev_db
|| db
!= prev_db
+ 1) {
545 jext
= kzalloc(sizeof(struct gfs2_journal_extent
),
548 printk(KERN_INFO
"GFS2 error: out of memory "
549 "mapping journal extents.\n");
556 list_add_tail(&jext
->extent_list
, &jd
->extent_list
);
565 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
567 char *message
= "FIRSTMOUNT=Done";
568 char *envp
[] = { message
, NULL
};
570 fs_info(sdp
, "first mount done, others may mount\n");
572 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
573 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
575 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
579 * gfs2_jindex_hold - Grab a lock on the jindex
580 * @sdp: The GFS2 superblock
581 * @ji_gh: the holder for the jindex glock
586 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
588 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
591 struct gfs2_jdesc
*jd
;
596 mutex_lock(&sdp
->sd_jindex_mutex
);
599 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
603 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
604 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
606 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
607 if (error
== -ENOENT
) {
612 gfs2_glock_dq_uninit(ji_gh
);
618 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
622 INIT_LIST_HEAD(&jd
->extent_list
);
623 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
624 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
625 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
629 error
= PTR_ERR(jd
->jd_inode
);
634 spin_lock(&sdp
->sd_jindex_spin
);
635 jd
->jd_jid
= sdp
->sd_journals
++;
636 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
637 spin_unlock(&sdp
->sd_jindex_spin
);
640 mutex_unlock(&sdp
->sd_jindex_mutex
);
645 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
647 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
648 struct gfs2_holder ji_gh
;
649 struct gfs2_inode
*ip
;
658 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
659 if (IS_ERR(sdp
->sd_jindex
)) {
660 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
661 return PTR_ERR(sdp
->sd_jindex
);
664 /* Load in the journal index special file */
666 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
668 fs_err(sdp
, "can't read journal index: %d\n", error
);
673 if (!gfs2_jindex_size(sdp
)) {
674 fs_err(sdp
, "no journals!\n");
678 if (sdp
->sd_args
.ar_spectator
) {
679 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
680 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
681 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
682 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
684 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
685 fs_err(sdp
, "can't mount journal #%u\n",
686 sdp
->sd_lockstruct
.ls_jid
);
687 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
688 gfs2_jindex_size(sdp
),
689 gfs2_jindex_size(sdp
) - 1);
692 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
694 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
696 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
697 &sdp
->sd_journal_gh
);
699 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
703 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
704 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
705 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
708 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
710 goto fail_journal_gh
;
713 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
715 fs_err(sdp
, "my journal (%u) is bad: %d\n",
716 sdp
->sd_jdesc
->jd_jid
, error
);
719 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
720 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
721 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
723 /* Map the extents for this journal's blocks */
724 map_journal_extents(sdp
);
726 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
728 if (sdp
->sd_lockstruct
.ls_first
) {
730 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
731 error
= gfs2_recover_journal(gfs2_jdesc_find(sdp
, x
),
734 fs_err(sdp
, "error recovering journal %u: %d\n",
740 gfs2_others_may_mount(sdp
);
741 } else if (!sdp
->sd_args
.ar_spectator
) {
742 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
744 fs_err(sdp
, "error recovering my journal: %d\n", error
);
749 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
750 gfs2_glock_dq_uninit(&ji_gh
);
756 if (!sdp
->sd_args
.ar_spectator
)
757 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
759 if (!sdp
->sd_args
.ar_spectator
)
760 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
762 gfs2_jindex_free(sdp
);
764 gfs2_glock_dq_uninit(&ji_gh
);
766 iput(sdp
->sd_jindex
);
770 static struct lock_class_key gfs2_quota_imutex_key
;
772 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
775 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
780 error
= init_journal(sdp
, undo
);
784 /* Read in the master statfs inode */
785 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
786 if (IS_ERR(sdp
->sd_statfs_inode
)) {
787 error
= PTR_ERR(sdp
->sd_statfs_inode
);
788 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
792 /* Read in the resource index inode */
793 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
794 if (IS_ERR(sdp
->sd_rindex
)) {
795 error
= PTR_ERR(sdp
->sd_rindex
);
796 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
799 sdp
->sd_rindex_uptodate
= 0;
801 /* Read in the quota inode */
802 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
803 if (IS_ERR(sdp
->sd_quota_inode
)) {
804 error
= PTR_ERR(sdp
->sd_quota_inode
);
805 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
809 * i_mutex on quota files is special. Since this inode is hidden system
810 * file, we are safe to define locking ourselves.
812 lockdep_set_class(&sdp
->sd_quota_inode
->i_mutex
,
813 &gfs2_quota_imutex_key
);
815 error
= gfs2_rindex_update(sdp
);
822 iput(sdp
->sd_quota_inode
);
824 gfs2_clear_rgrpd(sdp
);
825 iput(sdp
->sd_rindex
);
827 iput(sdp
->sd_statfs_inode
);
829 init_journal(sdp
, UNDO
);
834 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
836 struct inode
*pn
= NULL
;
839 struct gfs2_inode
*ip
;
840 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
842 if (sdp
->sd_args
.ar_spectator
)
848 pn
= gfs2_lookup_simple(master
, "per_node");
851 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
855 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
856 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
857 if (IS_ERR(sdp
->sd_sc_inode
)) {
858 error
= PTR_ERR(sdp
->sd_sc_inode
);
859 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
863 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
864 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
865 if (IS_ERR(sdp
->sd_qc_inode
)) {
866 error
= PTR_ERR(sdp
->sd_qc_inode
);
867 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
874 ip
= GFS2_I(sdp
->sd_sc_inode
);
875 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
878 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
882 ip
= GFS2_I(sdp
->sd_qc_inode
);
883 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
886 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
893 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
895 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
897 iput(sdp
->sd_qc_inode
);
899 iput(sdp
->sd_sc_inode
);
906 static int init_threads(struct gfs2_sbd
*sdp
, int undo
)
908 struct task_struct
*p
;
914 p
= kthread_run(gfs2_logd
, sdp
, "gfs2_logd");
917 fs_err(sdp
, "can't start logd thread: %d\n", error
);
920 sdp
->sd_logd_process
= p
;
922 p
= kthread_run(gfs2_quotad
, sdp
, "gfs2_quotad");
925 fs_err(sdp
, "can't start quotad thread: %d\n", error
);
928 sdp
->sd_quotad_process
= p
;
934 kthread_stop(sdp
->sd_quotad_process
);
936 kthread_stop(sdp
->sd_logd_process
);
940 static const match_table_t nolock_tokens
= {
941 { Opt_jid
, "jid=%d\n", },
945 static const struct lm_lockops nolock_ops
= {
946 .lm_proto_name
= "lock_nolock",
947 .lm_put_lock
= gfs2_glock_free
,
948 .lm_tokens
= &nolock_tokens
,
952 * gfs2_lm_mount - mount a locking protocol
953 * @sdp: the filesystem
954 * @args: mount arguments
955 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
960 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
962 const struct lm_lockops
*lm
;
963 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
964 struct gfs2_args
*args
= &sdp
->sd_args
;
965 const char *proto
= sdp
->sd_proto_name
;
966 const char *table
= sdp
->sd_table_name
;
970 if (!strcmp("lock_nolock", proto
)) {
972 sdp
->sd_args
.ar_localflocks
= 1;
973 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
974 } else if (!strcmp("lock_dlm", proto
)) {
978 printk(KERN_INFO
"GFS2: can't find protocol %s\n", proto
);
982 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
987 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
988 substring_t tmp
[MAX_OPT_ARGS
];
994 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
997 ret
= match_int(&tmp
[0], &option
);
998 if (ret
|| option
< 0)
1000 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
1001 ls
->ls_jid
= option
;
1005 /* Obsolete, but left for backward compat purposes */
1008 ret
= match_int(&tmp
[0], &option
);
1009 if (ret
|| (option
!= 0 && option
!= 1))
1010 goto hostdata_error
;
1011 ls
->ls_first
= option
;
1016 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
1021 if (lm
->lm_mount
== NULL
) {
1022 fs_info(sdp
, "Now mounting FS...\n");
1023 complete_all(&sdp
->sd_locking_init
);
1026 ret
= lm
->lm_mount(sdp
, table
);
1028 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
1029 complete_all(&sdp
->sd_locking_init
);
1033 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1035 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1036 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1038 lm
->lm_unmount(sdp
);
1041 static int gfs2_journalid_wait(void *word
)
1043 if (signal_pending(current
))
1049 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1051 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1054 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, gfs2_journalid_wait
, TASK_INTERRUPTIBLE
);
1057 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1059 struct super_block
*sb
= sdp
->sd_vfs
;
1062 char *envp
[] = { ro
, spectator
, NULL
};
1063 sprintf(ro
, "RDONLY=%d", (sb
->s_flags
& MS_RDONLY
) ? 1 : 0);
1064 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1065 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1069 * fill_super - Read in superblock
1070 * @sb: The VFS superblock
1071 * @data: Mount options
1072 * @silent: Don't complain if it's not a GFS2 filesystem
1077 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1079 struct gfs2_sbd
*sdp
;
1080 struct gfs2_holder mount_gh
;
1085 printk(KERN_WARNING
"GFS2: can't alloc struct gfs2_sbd\n");
1088 sdp
->sd_args
= *args
;
1090 if (sdp
->sd_args
.ar_spectator
) {
1091 sb
->s_flags
|= MS_RDONLY
;
1092 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1094 if (sdp
->sd_args
.ar_posix_acl
)
1095 sb
->s_flags
|= MS_POSIXACL
;
1096 if (sdp
->sd_args
.ar_nobarrier
)
1097 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1099 sb
->s_flags
|= MS_NOSEC
;
1100 sb
->s_magic
= GFS2_MAGIC
;
1101 sb
->s_op
= &gfs2_super_ops
;
1102 sb
->s_d_op
= &gfs2_dops
;
1103 sb
->s_export_op
= &gfs2_export_ops
;
1104 sb
->s_xattr
= gfs2_xattr_handlers
;
1105 sb
->s_qcop
= &gfs2_quotactl_ops
;
1106 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1107 sb
->s_time_gran
= 1;
1108 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1110 /* Set up the buffer cache and fill in some fake block size values
1111 to allow us to read-in the on-disk superblock. */
1112 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1113 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1114 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1115 GFS2_BASIC_BLOCK_SHIFT
;
1116 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
1118 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1119 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1120 if (sdp
->sd_args
.ar_statfs_quantum
) {
1121 sdp
->sd_tune
.gt_statfs_slow
= 0;
1122 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1124 sdp
->sd_tune
.gt_statfs_slow
= 1;
1125 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1128 error
= init_names(sdp
, silent
);
1130 /* In this case, we haven't initialized sysfs, so we have to
1131 manually free the sdp. */
1132 free_percpu(sdp
->sd_lkstats
);
1134 sb
->s_fs_info
= NULL
;
1138 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s", sdp
->sd_table_name
);
1140 error
= gfs2_sys_fs_add(sdp
);
1142 * If we hit an error here, gfs2_sys_fs_add will have called function
1143 * kobject_put which causes the sysfs usage count to go to zero, which
1144 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1145 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1146 * kobject_put to free sdp.
1151 gfs2_create_debugfs_file(sdp
);
1153 error
= gfs2_lm_mount(sdp
, silent
);
1157 error
= init_locking(sdp
, &mount_gh
, DO
);
1161 error
= init_sb(sdp
, silent
);
1165 error
= wait_on_journal(sdp
);
1170 * If user space has failed to join the cluster or some similar
1171 * failure has occurred, then the journal id will contain a
1172 * negative (error) number. This will then be returned to the
1173 * caller (of the mount syscall). We do this even for spectator
1174 * mounts (which just write a jid of 0 to indicate "ok" even though
1175 * the jid is unused in the spectator case)
1177 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1178 error
= sdp
->sd_lockstruct
.ls_jid
;
1179 sdp
->sd_lockstruct
.ls_jid
= 0;
1183 if (sdp
->sd_args
.ar_spectator
)
1184 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.s",
1185 sdp
->sd_table_name
);
1187 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.%u",
1188 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1190 error
= init_inodes(sdp
, DO
);
1194 error
= init_per_node(sdp
, DO
);
1198 error
= gfs2_statfs_init(sdp
);
1200 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1204 error
= init_threads(sdp
, DO
);
1208 if (!(sb
->s_flags
& MS_RDONLY
)) {
1209 error
= gfs2_make_fs_rw(sdp
);
1211 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1216 gfs2_glock_dq_uninit(&mount_gh
);
1217 gfs2_online_uevent(sdp
);
1221 init_threads(sdp
, UNDO
);
1223 init_per_node(sdp
, UNDO
);
1225 init_inodes(sdp
, UNDO
);
1227 if (sdp
->sd_root_dir
)
1228 dput(sdp
->sd_root_dir
);
1229 if (sdp
->sd_master_dir
)
1230 dput(sdp
->sd_master_dir
);
1235 init_locking(sdp
, &mount_gh
, UNDO
);
1237 gfs2_gl_hash_clear(sdp
);
1238 gfs2_lm_unmount(sdp
);
1240 gfs2_delete_debugfs_file(sdp
);
1241 free_percpu(sdp
->sd_lkstats
);
1242 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1243 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1244 gfs2_sys_fs_del(sdp
);
1245 sb
->s_fs_info
= NULL
;
1249 static int set_gfs2_super(struct super_block
*s
, void *data
)
1252 s
->s_dev
= s
->s_bdev
->bd_dev
;
1255 * We set the bdi here to the queue backing, file systems can
1256 * overwrite this in ->fill_super()
1258 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
1262 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1264 struct block_device
*bdev
= ptr
;
1265 return (bdev
== s
->s_bdev
);
1269 * gfs2_mount - Get the GFS2 superblock
1270 * @fs_type: The GFS2 filesystem type
1271 * @flags: Mount flags
1272 * @dev_name: The name of the device
1273 * @data: The mount arguments
1275 * Q. Why not use get_sb_bdev() ?
1276 * A. We need to select one of two root directories to mount, independent
1277 * of whether this is the initial, or subsequent, mount of this sb
1279 * Returns: 0 or -ve on error
1282 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1283 const char *dev_name
, void *data
)
1285 struct block_device
*bdev
;
1286 struct super_block
*s
;
1287 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1289 struct gfs2_args args
;
1290 struct gfs2_sbd
*sdp
;
1292 if (!(flags
& MS_RDONLY
))
1293 mode
|= FMODE_WRITE
;
1295 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1297 return ERR_CAST(bdev
);
1300 * once the super is inserted into the list by sget, s_umount
1301 * will protect the lockfs code from trying to start a snapshot
1302 * while we are mounting
1304 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1305 if (bdev
->bd_fsfreeze_count
> 0) {
1306 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1310 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1311 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1317 blkdev_put(bdev
, mode
);
1319 memset(&args
, 0, sizeof(args
));
1320 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1321 args
.ar_data
= GFS2_DATA_DEFAULT
;
1322 args
.ar_commit
= 30;
1323 args
.ar_statfs_quantum
= 30;
1324 args
.ar_quota_quantum
= 60;
1325 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1327 error
= gfs2_mount_args(&args
, data
);
1329 printk(KERN_WARNING
"GFS2: can't parse mount arguments\n");
1335 if ((flags
^ s
->s_flags
) & MS_RDONLY
)
1338 char b
[BDEVNAME_SIZE
];
1341 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
1342 sb_set_blocksize(s
, block_size(bdev
));
1343 error
= fill_super(s
, &args
, flags
& MS_SILENT
? 1 : 0);
1346 s
->s_flags
|= MS_ACTIVE
;
1352 return dget(sdp
->sd_master_dir
);
1354 return dget(sdp
->sd_root_dir
);
1357 deactivate_locked_super(s
);
1358 return ERR_PTR(error
);
1360 blkdev_put(bdev
, mode
);
1361 return ERR_PTR(error
);
1364 static int set_meta_super(struct super_block
*s
, void *ptr
)
1369 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1370 int flags
, const char *dev_name
, void *data
)
1372 struct super_block
*s
;
1373 struct gfs2_sbd
*sdp
;
1377 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1379 printk(KERN_WARNING
"GFS2: path_lookup on %s returned error %d\n",
1381 return ERR_PTR(error
);
1383 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1384 path
.dentry
->d_inode
->i_sb
->s_bdev
);
1387 printk(KERN_WARNING
"GFS2: gfs2 mount does not exist\n");
1390 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1391 deactivate_locked_super(s
);
1392 return ERR_PTR(-EBUSY
);
1395 return dget(sdp
->sd_master_dir
);
1398 static void gfs2_kill_sb(struct super_block
*sb
)
1400 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1403 kill_block_super(sb
);
1407 gfs2_meta_syncfs(sdp
);
1408 dput(sdp
->sd_root_dir
);
1409 dput(sdp
->sd_master_dir
);
1410 sdp
->sd_root_dir
= NULL
;
1411 sdp
->sd_master_dir
= NULL
;
1412 shrink_dcache_sb(sb
);
1413 gfs2_delete_debugfs_file(sdp
);
1414 free_percpu(sdp
->sd_lkstats
);
1415 kill_block_super(sb
);
1418 struct file_system_type gfs2_fs_type
= {
1420 .fs_flags
= FS_REQUIRES_DEV
,
1421 .mount
= gfs2_mount
,
1422 .kill_sb
= gfs2_kill_sb
,
1423 .owner
= THIS_MODULE
,
1426 struct file_system_type gfs2meta_fs_type
= {
1428 .fs_flags
= FS_REQUIRES_DEV
,
1429 .mount
= gfs2_mount_meta
,
1430 .owner
= THIS_MODULE
,