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>
23 #include <linux/module.h>
40 #include "trace_gfs2.h"
46 * gfs2_tune_init - Fill a gfs2_tune structure with default values
51 static void gfs2_tune_init(struct gfs2_tune
*gt
)
53 spin_lock_init(>
->gt_spin
);
55 gt
->gt_quota_warn_period
= 10;
56 gt
->gt_quota_scale_num
= 1;
57 gt
->gt_quota_scale_den
= 1;
58 gt
->gt_new_files_jdata
= 0;
59 gt
->gt_max_readahead
= 1 << 18;
60 gt
->gt_complain_secs
= 10;
63 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
66 struct address_space
*mapping
;
68 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
74 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
75 if (!sdp
->sd_lkstats
) {
80 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
81 gfs2_tune_init(&sdp
->sd_tune
);
83 init_waitqueue_head(&sdp
->sd_glock_wait
);
84 atomic_set(&sdp
->sd_glock_disposal
, 0);
85 init_completion(&sdp
->sd_locking_init
);
86 init_completion(&sdp
->sd_wdack
);
87 spin_lock_init(&sdp
->sd_statfs_spin
);
89 spin_lock_init(&sdp
->sd_rindex_spin
);
90 sdp
->sd_rindex_tree
.rb_node
= NULL
;
92 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
93 spin_lock_init(&sdp
->sd_jindex_spin
);
94 mutex_init(&sdp
->sd_jindex_mutex
);
96 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
97 mutex_init(&sdp
->sd_quota_mutex
);
98 mutex_init(&sdp
->sd_quota_sync_mutex
);
99 init_waitqueue_head(&sdp
->sd_quota_wait
);
100 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
101 spin_lock_init(&sdp
->sd_trunc_lock
);
102 spin_lock_init(&sdp
->sd_bitmap_lock
);
104 mapping
= &sdp
->sd_aspace
;
106 address_space_init_once(mapping
);
107 mapping
->a_ops
= &gfs2_meta_aops
;
108 mapping
->host
= sb
->s_bdev
->bd_inode
;
110 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
111 mapping
->private_data
= NULL
;
112 mapping
->backing_dev_info
= sb
->s_bdi
;
113 mapping
->writeback_index
= 0;
115 spin_lock_init(&sdp
->sd_log_lock
);
116 atomic_set(&sdp
->sd_log_pinned
, 0);
117 INIT_LIST_HEAD(&sdp
->sd_log_le_buf
);
118 INIT_LIST_HEAD(&sdp
->sd_log_le_revoke
);
119 INIT_LIST_HEAD(&sdp
->sd_log_le_databuf
);
120 INIT_LIST_HEAD(&sdp
->sd_log_le_ordered
);
121 spin_lock_init(&sdp
->sd_ordered_lock
);
123 init_waitqueue_head(&sdp
->sd_log_waitq
);
124 init_waitqueue_head(&sdp
->sd_logd_waitq
);
125 spin_lock_init(&sdp
->sd_ail_lock
);
126 INIT_LIST_HEAD(&sdp
->sd_ail1_list
);
127 INIT_LIST_HEAD(&sdp
->sd_ail2_list
);
129 init_rwsem(&sdp
->sd_log_flush_lock
);
130 atomic_set(&sdp
->sd_log_in_flight
, 0);
131 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
133 INIT_LIST_HEAD(&sdp
->sd_revoke_list
);
140 * gfs2_check_sb - Check superblock
141 * @sdp: the filesystem
142 * @sb: The superblock
143 * @silent: Don't print a message if the check fails
145 * Checks the version code of the FS is one that we understand how to
146 * read and that the sizes of the various on-disk structures have not
150 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
152 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
154 if (sb
->sb_magic
!= GFS2_MAGIC
||
155 sb
->sb_type
!= GFS2_METATYPE_SB
) {
157 printk(KERN_WARNING
"GFS2: not a GFS2 filesystem\n");
161 /* If format numbers match exactly, we're done. */
163 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
164 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
167 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
172 static void end_bio_io_page(struct bio
*bio
, int error
)
174 struct page
*page
= bio
->bi_private
;
177 SetPageUptodate(page
);
179 printk(KERN_WARNING
"gfs2: error %d reading superblock\n", error
);
183 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
185 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
186 struct super_block
*s
= sdp
->sd_vfs
;
187 const struct gfs2_sb
*str
= buf
;
189 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
190 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
191 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
192 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
193 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
194 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
195 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
196 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
197 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
198 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
199 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
201 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
202 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
203 memcpy(s
->s_uuid
, str
->sb_uuid
, 16);
207 * gfs2_read_super - Read the gfs2 super block from disk
208 * @sdp: The GFS2 super block
209 * @sector: The location of the super block
210 * @error: The error code to return
212 * This uses the bio functions to read the super block from disk
213 * because we want to be 100% sure that we never read cached data.
214 * A super block is read twice only during each GFS2 mount and is
215 * never written to by the filesystem. The first time its read no
216 * locks are held, and the only details which are looked at are those
217 * relating to the locking protocol. Once locking is up and working,
218 * the sb is read again under the lock to establish the location of
219 * the master directory (contains pointers to journals etc) and the
222 * Returns: 0 on success or error
225 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
227 struct super_block
*sb
= sdp
->sd_vfs
;
232 page
= alloc_page(GFP_NOFS
);
236 ClearPageUptodate(page
);
237 ClearPageDirty(page
);
240 bio
= bio_alloc(GFP_NOFS
, 1);
241 bio
->bi_iter
.bi_sector
= sector
* (sb
->s_blocksize
>> 9);
242 bio
->bi_bdev
= sb
->s_bdev
;
243 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
245 bio
->bi_end_io
= end_bio_io_page
;
246 bio
->bi_private
= page
;
247 submit_bio(READ_SYNC
| REQ_META
, bio
);
248 wait_on_page_locked(page
);
250 if (!PageUptodate(page
)) {
258 return gfs2_check_sb(sdp
, silent
);
262 * gfs2_read_sb - Read super block
263 * @sdp: The GFS2 superblock
264 * @silent: Don't print message if mount fails
268 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
270 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
275 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
278 fs_err(sdp
, "can't read superblock\n");
282 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
283 GFS2_BASIC_BLOCK_SHIFT
;
284 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
285 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
286 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
287 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
288 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
289 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
290 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
291 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
292 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
293 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
294 sizeof(struct gfs2_meta_header
)) /
295 sizeof(struct gfs2_quota_change
);
296 sdp
->sd_blocks_per_bitmap
= (sdp
->sd_sb
.sb_bsize
-
297 sizeof(struct gfs2_meta_header
))
298 * GFS2_NBBY
; /* not the rgrp bitmap, subsequent bitmaps only */
300 /* Compute maximum reservation required to add a entry to a directory */
302 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * (1 << GFS2_DIR_MAX_DEPTH
),
306 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
307 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
308 ind_blocks
+= tmp_blocks
;
311 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
313 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
315 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
316 sizeof(struct gfs2_dinode
);
317 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
322 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
324 m
= do_div(d
, sdp
->sd_inptrs
);
326 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
328 sdp
->sd_heightsize
[x
] = space
;
330 sdp
->sd_max_height
= x
;
331 sdp
->sd_heightsize
[x
] = ~0;
332 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
334 sdp
->sd_jheightsize
[0] = sdp
->sd_sb
.sb_bsize
-
335 sizeof(struct gfs2_dinode
);
336 sdp
->sd_jheightsize
[1] = sdp
->sd_jbsize
* sdp
->sd_diptrs
;
341 space
= sdp
->sd_jheightsize
[x
- 1] * sdp
->sd_inptrs
;
343 m
= do_div(d
, sdp
->sd_inptrs
);
345 if (d
!= sdp
->sd_jheightsize
[x
- 1] || m
)
347 sdp
->sd_jheightsize
[x
] = space
;
349 sdp
->sd_max_jheight
= x
;
350 sdp
->sd_jheightsize
[x
] = ~0;
351 gfs2_assert(sdp
, sdp
->sd_max_jheight
<= GFS2_MAX_META_HEIGHT
);
356 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
361 proto
= sdp
->sd_args
.ar_lockproto
;
362 table
= sdp
->sd_args
.ar_locktable
;
364 /* Try to autodetect */
366 if (!proto
[0] || !table
[0]) {
367 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
372 proto
= sdp
->sd_sb
.sb_lockproto
;
374 table
= sdp
->sd_sb
.sb_locktable
;
378 table
= sdp
->sd_vfs
->s_id
;
380 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
381 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
383 table
= sdp
->sd_table_name
;
384 while ((table
= strchr(table
, '/')))
390 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
398 error
= gfs2_glock_nq_num(sdp
,
399 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
400 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
403 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
407 error
= gfs2_glock_nq_num(sdp
,
408 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
410 LM_FLAG_NOEXP
| GL_EXACT
,
413 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
417 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
418 CREATE
, &sdp
->sd_rename_gl
);
420 fs_err(sdp
, "can't create rename glock: %d\n", error
);
424 error
= gfs2_glock_get(sdp
, GFS2_TRANS_LOCK
, &gfs2_trans_glops
,
425 CREATE
, &sdp
->sd_trans_gl
);
427 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
434 gfs2_glock_put(sdp
->sd_trans_gl
);
436 gfs2_glock_put(sdp
->sd_rename_gl
);
438 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
440 gfs2_glock_dq_uninit(mount_gh
);
445 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
446 u64 no_addr
, const char *name
)
448 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
449 struct dentry
*dentry
;
452 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0, 0);
454 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
455 return PTR_ERR(inode
);
457 dentry
= d_make_root(inode
);
459 fs_err(sdp
, "can't alloc %s dentry\n", name
);
466 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
468 struct super_block
*sb
= sdp
->sd_vfs
;
469 struct gfs2_holder sb_gh
;
473 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
474 LM_ST_SHARED
, 0, &sb_gh
);
476 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
480 ret
= gfs2_read_sb(sdp
, silent
);
482 fs_err(sdp
, "can't read superblock: %d\n", ret
);
486 /* Set up the buffer cache and SB for real */
487 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
489 fs_err(sdp
, "FS block size (%u) is too small for device "
491 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
494 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
496 fs_err(sdp
, "FS block size (%u) is too big for machine "
498 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
501 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
503 /* Get the root inode */
504 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
505 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
509 /* Get the master inode */
510 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
511 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
513 dput(sdp
->sd_root_dir
);
516 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
518 gfs2_glock_dq_uninit(&sb_gh
);
523 * map_journal_extents - create a reusable "extent" mapping from all logical
524 * blocks to all physical blocks for the given journal. This will save
525 * us time when writing journal blocks. Most journals will have only one
526 * extent that maps all their logical blocks. That's because gfs2.mkfs
527 * arranges the journal blocks sequentially to maximize performance.
528 * So the extent would map the first block for the entire file length.
529 * However, gfs2_jadd can happen while file activity is happening, so
530 * those journals may not be sequential. Less likely is the case where
531 * the users created their own journals by mounting the metafs and
532 * laying it out. But it's still possible. These journals might have
535 * TODO: This should be done in bigger chunks rather than one block at a time,
536 * but since it's only done at mount time, I'm not worried about the
539 static int map_journal_extents(struct gfs2_sbd
*sdp
)
541 struct gfs2_jdesc
*jd
= sdp
->sd_jdesc
;
543 u64 db
, prev_db
; /* logical block, disk block, prev disk block */
544 struct gfs2_inode
*ip
= GFS2_I(jd
->jd_inode
);
545 struct gfs2_journal_extent
*jext
= NULL
;
546 struct buffer_head bh
;
551 for (lb
= 0; lb
< i_size_read(jd
->jd_inode
) >> sdp
->sd_sb
.sb_bsize_shift
; lb
++) {
554 bh
.b_size
= 1 << ip
->i_inode
.i_blkbits
;
555 rc
= gfs2_block_map(jd
->jd_inode
, lb
, &bh
, 0);
558 printk(KERN_INFO
"GFS2 journal mapping error %d: lb="
559 "%u db=%llu\n", rc
, lb
, (unsigned long long)db
);
562 if (!prev_db
|| db
!= prev_db
+ 1) {
563 jext
= kzalloc(sizeof(struct gfs2_journal_extent
),
566 printk(KERN_INFO
"GFS2 error: out of memory "
567 "mapping journal extents.\n");
574 list_add_tail(&jext
->extent_list
, &jd
->extent_list
);
583 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
585 char *message
= "FIRSTMOUNT=Done";
586 char *envp
[] = { message
, NULL
};
588 fs_info(sdp
, "first mount done, others may mount\n");
590 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
591 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
593 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
597 * gfs2_jindex_hold - Grab a lock on the jindex
598 * @sdp: The GFS2 superblock
599 * @ji_gh: the holder for the jindex glock
604 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
606 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
609 struct gfs2_jdesc
*jd
;
614 mutex_lock(&sdp
->sd_jindex_mutex
);
617 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
621 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
622 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
624 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
625 if (error
== -ENOENT
) {
630 gfs2_glock_dq_uninit(ji_gh
);
636 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
640 INIT_LIST_HEAD(&jd
->extent_list
);
641 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
642 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
643 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
647 error
= PTR_ERR(jd
->jd_inode
);
652 spin_lock(&sdp
->sd_jindex_spin
);
653 jd
->jd_jid
= sdp
->sd_journals
++;
654 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
655 spin_unlock(&sdp
->sd_jindex_spin
);
658 mutex_unlock(&sdp
->sd_jindex_mutex
);
664 * check_journal_clean - Make sure a journal is clean for a spectator mount
665 * @sdp: The GFS2 superblock
666 * @jd: The journal descriptor
668 * Returns: 0 if the journal is clean or locked, else an error
670 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
673 struct gfs2_holder j_gh
;
674 struct gfs2_log_header_host head
;
675 struct gfs2_inode
*ip
;
677 ip
= GFS2_I(jd
->jd_inode
);
678 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
679 GL_EXACT
| GL_NOCACHE
, &j_gh
);
681 fs_err(sdp
, "Error locking journal for spectator mount.\n");
684 error
= gfs2_jdesc_check(jd
);
686 fs_err(sdp
, "Error checking journal for spectator mount.\n");
689 error
= gfs2_find_jhead(jd
, &head
);
691 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
694 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
696 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
697 "must not be a spectator.\n", jd
->jd_jid
);
701 gfs2_glock_dq_uninit(&j_gh
);
705 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
707 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
708 struct gfs2_holder ji_gh
;
709 struct gfs2_inode
*ip
;
718 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
719 if (IS_ERR(sdp
->sd_jindex
)) {
720 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
721 return PTR_ERR(sdp
->sd_jindex
);
724 /* Load in the journal index special file */
726 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
728 fs_err(sdp
, "can't read journal index: %d\n", error
);
733 if (!gfs2_jindex_size(sdp
)) {
734 fs_err(sdp
, "no journals!\n");
738 if (sdp
->sd_args
.ar_spectator
) {
739 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
740 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
741 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
742 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
744 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
745 fs_err(sdp
, "can't mount journal #%u\n",
746 sdp
->sd_lockstruct
.ls_jid
);
747 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
748 gfs2_jindex_size(sdp
),
749 gfs2_jindex_size(sdp
) - 1);
752 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
754 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
756 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
757 &sdp
->sd_journal_gh
);
759 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
763 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
764 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
765 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
768 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
770 goto fail_journal_gh
;
773 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
775 fs_err(sdp
, "my journal (%u) is bad: %d\n",
776 sdp
->sd_jdesc
->jd_jid
, error
);
779 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
780 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
781 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
783 /* Map the extents for this journal's blocks */
784 map_journal_extents(sdp
);
786 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
788 if (sdp
->sd_lockstruct
.ls_first
) {
790 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
791 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
793 if (sdp
->sd_args
.ar_spectator
) {
794 error
= check_journal_clean(sdp
, jd
);
799 error
= gfs2_recover_journal(jd
, true);
801 fs_err(sdp
, "error recovering journal %u: %d\n",
807 gfs2_others_may_mount(sdp
);
808 } else if (!sdp
->sd_args
.ar_spectator
) {
809 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
811 fs_err(sdp
, "error recovering my journal: %d\n", error
);
816 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
817 gfs2_glock_dq_uninit(&ji_gh
);
823 if (!sdp
->sd_args
.ar_spectator
)
824 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
826 if (!sdp
->sd_args
.ar_spectator
)
827 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
829 gfs2_jindex_free(sdp
);
831 gfs2_glock_dq_uninit(&ji_gh
);
833 iput(sdp
->sd_jindex
);
837 static struct lock_class_key gfs2_quota_imutex_key
;
839 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
842 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
847 error
= init_journal(sdp
, undo
);
851 /* Read in the master statfs inode */
852 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
853 if (IS_ERR(sdp
->sd_statfs_inode
)) {
854 error
= PTR_ERR(sdp
->sd_statfs_inode
);
855 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
859 /* Read in the resource index inode */
860 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
861 if (IS_ERR(sdp
->sd_rindex
)) {
862 error
= PTR_ERR(sdp
->sd_rindex
);
863 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
866 sdp
->sd_rindex_uptodate
= 0;
868 /* Read in the quota inode */
869 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
870 if (IS_ERR(sdp
->sd_quota_inode
)) {
871 error
= PTR_ERR(sdp
->sd_quota_inode
);
872 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
876 * i_mutex on quota files is special. Since this inode is hidden system
877 * file, we are safe to define locking ourselves.
879 lockdep_set_class(&sdp
->sd_quota_inode
->i_mutex
,
880 &gfs2_quota_imutex_key
);
882 error
= gfs2_rindex_update(sdp
);
889 iput(sdp
->sd_quota_inode
);
891 gfs2_clear_rgrpd(sdp
);
892 iput(sdp
->sd_rindex
);
894 iput(sdp
->sd_statfs_inode
);
896 init_journal(sdp
, UNDO
);
901 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
903 struct inode
*pn
= NULL
;
906 struct gfs2_inode
*ip
;
907 struct inode
*master
= sdp
->sd_master_dir
->d_inode
;
909 if (sdp
->sd_args
.ar_spectator
)
915 pn
= gfs2_lookup_simple(master
, "per_node");
918 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
922 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
923 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
924 if (IS_ERR(sdp
->sd_sc_inode
)) {
925 error
= PTR_ERR(sdp
->sd_sc_inode
);
926 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
930 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
931 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
932 if (IS_ERR(sdp
->sd_qc_inode
)) {
933 error
= PTR_ERR(sdp
->sd_qc_inode
);
934 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
941 ip
= GFS2_I(sdp
->sd_sc_inode
);
942 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
945 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
949 ip
= GFS2_I(sdp
->sd_qc_inode
);
950 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
953 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
960 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
962 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
964 iput(sdp
->sd_qc_inode
);
966 iput(sdp
->sd_sc_inode
);
973 static const match_table_t nolock_tokens
= {
974 { Opt_jid
, "jid=%d\n", },
978 static const struct lm_lockops nolock_ops
= {
979 .lm_proto_name
= "lock_nolock",
980 .lm_put_lock
= gfs2_glock_free
,
981 .lm_tokens
= &nolock_tokens
,
985 * gfs2_lm_mount - mount a locking protocol
986 * @sdp: the filesystem
987 * @args: mount arguments
988 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
993 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
995 const struct lm_lockops
*lm
;
996 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
997 struct gfs2_args
*args
= &sdp
->sd_args
;
998 const char *proto
= sdp
->sd_proto_name
;
999 const char *table
= sdp
->sd_table_name
;
1003 if (!strcmp("lock_nolock", proto
)) {
1005 sdp
->sd_args
.ar_localflocks
= 1;
1006 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
1007 } else if (!strcmp("lock_dlm", proto
)) {
1011 printk(KERN_INFO
"GFS2: can't find protocol %s\n", proto
);
1015 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
1020 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
1021 substring_t tmp
[MAX_OPT_ARGS
];
1027 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
1030 ret
= match_int(&tmp
[0], &option
);
1031 if (ret
|| option
< 0)
1032 goto hostdata_error
;
1033 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
1034 ls
->ls_jid
= option
;
1038 /* Obsolete, but left for backward compat purposes */
1041 ret
= match_int(&tmp
[0], &option
);
1042 if (ret
|| (option
!= 0 && option
!= 1))
1043 goto hostdata_error
;
1044 ls
->ls_first
= option
;
1049 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
1054 if (lm
->lm_mount
== NULL
) {
1055 fs_info(sdp
, "Now mounting FS...\n");
1056 complete_all(&sdp
->sd_locking_init
);
1059 ret
= lm
->lm_mount(sdp
, table
);
1061 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
1062 complete_all(&sdp
->sd_locking_init
);
1066 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1068 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1069 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1071 lm
->lm_unmount(sdp
);
1074 static int gfs2_journalid_wait(void *word
)
1076 if (signal_pending(current
))
1082 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1084 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1087 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, gfs2_journalid_wait
, TASK_INTERRUPTIBLE
);
1090 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1092 struct super_block
*sb
= sdp
->sd_vfs
;
1095 char *envp
[] = { ro
, spectator
, NULL
};
1096 sprintf(ro
, "RDONLY=%d", (sb
->s_flags
& MS_RDONLY
) ? 1 : 0);
1097 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1098 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1102 * fill_super - Read in superblock
1103 * @sb: The VFS superblock
1104 * @data: Mount options
1105 * @silent: Don't complain if it's not a GFS2 filesystem
1110 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1112 struct gfs2_sbd
*sdp
;
1113 struct gfs2_holder mount_gh
;
1118 printk(KERN_WARNING
"GFS2: can't alloc struct gfs2_sbd\n");
1121 sdp
->sd_args
= *args
;
1123 if (sdp
->sd_args
.ar_spectator
) {
1124 sb
->s_flags
|= MS_RDONLY
;
1125 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1127 if (sdp
->sd_args
.ar_posix_acl
)
1128 sb
->s_flags
|= MS_POSIXACL
;
1129 if (sdp
->sd_args
.ar_nobarrier
)
1130 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1132 sb
->s_flags
|= MS_NOSEC
;
1133 sb
->s_magic
= GFS2_MAGIC
;
1134 sb
->s_op
= &gfs2_super_ops
;
1135 sb
->s_d_op
= &gfs2_dops
;
1136 sb
->s_export_op
= &gfs2_export_ops
;
1137 sb
->s_xattr
= gfs2_xattr_handlers
;
1138 sb
->s_qcop
= &gfs2_quotactl_ops
;
1139 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1140 sb
->s_time_gran
= 1;
1141 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1143 /* Set up the buffer cache and fill in some fake block size values
1144 to allow us to read-in the on-disk superblock. */
1145 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1146 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1147 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1148 GFS2_BASIC_BLOCK_SHIFT
;
1149 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
1151 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1152 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1153 if (sdp
->sd_args
.ar_statfs_quantum
) {
1154 sdp
->sd_tune
.gt_statfs_slow
= 0;
1155 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1157 sdp
->sd_tune
.gt_statfs_slow
= 1;
1158 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1161 error
= init_names(sdp
, silent
);
1163 /* In this case, we haven't initialized sysfs, so we have to
1164 manually free the sdp. */
1165 free_percpu(sdp
->sd_lkstats
);
1167 sb
->s_fs_info
= NULL
;
1171 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s", sdp
->sd_table_name
);
1173 error
= gfs2_sys_fs_add(sdp
);
1175 * If we hit an error here, gfs2_sys_fs_add will have called function
1176 * kobject_put which causes the sysfs usage count to go to zero, which
1177 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1178 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1179 * kobject_put to free sdp.
1184 gfs2_create_debugfs_file(sdp
);
1186 error
= gfs2_lm_mount(sdp
, silent
);
1190 error
= init_locking(sdp
, &mount_gh
, DO
);
1194 error
= init_sb(sdp
, silent
);
1198 error
= wait_on_journal(sdp
);
1203 * If user space has failed to join the cluster or some similar
1204 * failure has occurred, then the journal id will contain a
1205 * negative (error) number. This will then be returned to the
1206 * caller (of the mount syscall). We do this even for spectator
1207 * mounts (which just write a jid of 0 to indicate "ok" even though
1208 * the jid is unused in the spectator case)
1210 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1211 error
= sdp
->sd_lockstruct
.ls_jid
;
1212 sdp
->sd_lockstruct
.ls_jid
= 0;
1216 if (sdp
->sd_args
.ar_spectator
)
1217 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.s",
1218 sdp
->sd_table_name
);
1220 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.%u",
1221 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1223 error
= init_inodes(sdp
, DO
);
1227 error
= init_per_node(sdp
, DO
);
1231 error
= gfs2_statfs_init(sdp
);
1233 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1237 if (!(sb
->s_flags
& MS_RDONLY
)) {
1238 error
= gfs2_make_fs_rw(sdp
);
1240 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1245 gfs2_glock_dq_uninit(&mount_gh
);
1246 gfs2_online_uevent(sdp
);
1250 init_per_node(sdp
, UNDO
);
1252 init_inodes(sdp
, UNDO
);
1254 if (sdp
->sd_root_dir
)
1255 dput(sdp
->sd_root_dir
);
1256 if (sdp
->sd_master_dir
)
1257 dput(sdp
->sd_master_dir
);
1262 init_locking(sdp
, &mount_gh
, UNDO
);
1264 gfs2_gl_hash_clear(sdp
);
1265 gfs2_lm_unmount(sdp
);
1267 gfs2_delete_debugfs_file(sdp
);
1268 free_percpu(sdp
->sd_lkstats
);
1269 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1270 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1271 gfs2_sys_fs_del(sdp
);
1272 sb
->s_fs_info
= NULL
;
1276 static int set_gfs2_super(struct super_block
*s
, void *data
)
1279 s
->s_dev
= s
->s_bdev
->bd_dev
;
1282 * We set the bdi here to the queue backing, file systems can
1283 * overwrite this in ->fill_super()
1285 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
1289 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1291 struct block_device
*bdev
= ptr
;
1292 return (bdev
== s
->s_bdev
);
1296 * gfs2_mount - Get the GFS2 superblock
1297 * @fs_type: The GFS2 filesystem type
1298 * @flags: Mount flags
1299 * @dev_name: The name of the device
1300 * @data: The mount arguments
1302 * Q. Why not use get_sb_bdev() ?
1303 * A. We need to select one of two root directories to mount, independent
1304 * of whether this is the initial, or subsequent, mount of this sb
1306 * Returns: 0 or -ve on error
1309 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1310 const char *dev_name
, void *data
)
1312 struct block_device
*bdev
;
1313 struct super_block
*s
;
1314 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1316 struct gfs2_args args
;
1317 struct gfs2_sbd
*sdp
;
1319 if (!(flags
& MS_RDONLY
))
1320 mode
|= FMODE_WRITE
;
1322 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1324 return ERR_CAST(bdev
);
1327 * once the super is inserted into the list by sget, s_umount
1328 * will protect the lockfs code from trying to start a snapshot
1329 * while we are mounting
1331 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1332 if (bdev
->bd_fsfreeze_count
> 0) {
1333 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1337 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1338 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1345 * s_umount nests inside bd_mutex during
1346 * __invalidate_device(). blkdev_put() acquires
1347 * bd_mutex and can't be called under s_umount. Drop
1348 * s_umount temporarily. This is safe as we're
1349 * holding an active reference.
1351 up_write(&s
->s_umount
);
1352 blkdev_put(bdev
, mode
);
1353 down_write(&s
->s_umount
);
1356 memset(&args
, 0, sizeof(args
));
1357 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1358 args
.ar_data
= GFS2_DATA_DEFAULT
;
1359 args
.ar_commit
= 30;
1360 args
.ar_statfs_quantum
= 30;
1361 args
.ar_quota_quantum
= 60;
1362 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1364 error
= gfs2_mount_args(&args
, data
);
1366 printk(KERN_WARNING
"GFS2: can't parse mount arguments\n");
1372 if ((flags
^ s
->s_flags
) & MS_RDONLY
)
1375 char b
[BDEVNAME_SIZE
];
1378 strlcpy(s
->s_id
, bdevname(bdev
, b
), sizeof(s
->s_id
));
1379 sb_set_blocksize(s
, block_size(bdev
));
1380 error
= fill_super(s
, &args
, flags
& MS_SILENT
? 1 : 0);
1383 s
->s_flags
|= MS_ACTIVE
;
1389 return dget(sdp
->sd_master_dir
);
1391 return dget(sdp
->sd_root_dir
);
1394 deactivate_locked_super(s
);
1395 return ERR_PTR(error
);
1397 blkdev_put(bdev
, mode
);
1398 return ERR_PTR(error
);
1401 static int set_meta_super(struct super_block
*s
, void *ptr
)
1406 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1407 int flags
, const char *dev_name
, void *data
)
1409 struct super_block
*s
;
1410 struct gfs2_sbd
*sdp
;
1414 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1416 printk(KERN_WARNING
"GFS2: path_lookup on %s returned error %d\n",
1418 return ERR_PTR(error
);
1420 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1421 path
.dentry
->d_inode
->i_sb
->s_bdev
);
1424 printk(KERN_WARNING
"GFS2: gfs2 mount does not exist\n");
1427 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1428 deactivate_locked_super(s
);
1429 return ERR_PTR(-EBUSY
);
1432 return dget(sdp
->sd_master_dir
);
1435 static void gfs2_kill_sb(struct super_block
*sb
)
1437 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1440 kill_block_super(sb
);
1444 gfs2_meta_syncfs(sdp
);
1445 dput(sdp
->sd_root_dir
);
1446 dput(sdp
->sd_master_dir
);
1447 sdp
->sd_root_dir
= NULL
;
1448 sdp
->sd_master_dir
= NULL
;
1449 shrink_dcache_sb(sb
);
1450 gfs2_delete_debugfs_file(sdp
);
1451 free_percpu(sdp
->sd_lkstats
);
1452 kill_block_super(sb
);
1455 struct file_system_type gfs2_fs_type
= {
1457 .fs_flags
= FS_REQUIRES_DEV
,
1458 .mount
= gfs2_mount
,
1459 .kill_sb
= gfs2_kill_sb
,
1460 .owner
= THIS_MODULE
,
1462 MODULE_ALIAS_FS("gfs2");
1464 struct file_system_type gfs2meta_fs_type
= {
1466 .fs_flags
= FS_REQUIRES_DEV
,
1467 .mount
= gfs2_mount_meta
,
1468 .owner
= THIS_MODULE
,
1470 MODULE_ALIAS_FS("gfs2meta");