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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/sched.h>
13 #include <linux/slab.h>
14 #include <linux/spinlock.h>
15 #include <linux/completion.h>
16 #include <linux/buffer_head.h>
17 #include <linux/blkdev.h>
18 #include <linux/kthread.h>
19 #include <linux/export.h>
20 #include <linux/namei.h>
21 #include <linux/mount.h>
22 #include <linux/gfs2_ondisk.h>
23 #include <linux/quotaops.h>
24 #include <linux/lockdep.h>
25 #include <linux/module.h>
42 #include "trace_gfs2.h"
48 * gfs2_tune_init - Fill a gfs2_tune structure with default values
53 static void gfs2_tune_init(struct gfs2_tune
*gt
)
55 spin_lock_init(>
->gt_spin
);
57 gt
->gt_quota_warn_period
= 10;
58 gt
->gt_quota_scale_num
= 1;
59 gt
->gt_quota_scale_den
= 1;
60 gt
->gt_new_files_jdata
= 0;
61 gt
->gt_max_readahead
= 1 << 18;
62 gt
->gt_complain_secs
= 10;
65 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
68 struct address_space
*mapping
;
70 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
76 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
77 if (!sdp
->sd_lkstats
) {
82 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
83 gfs2_tune_init(&sdp
->sd_tune
);
85 init_waitqueue_head(&sdp
->sd_glock_wait
);
86 atomic_set(&sdp
->sd_glock_disposal
, 0);
87 init_completion(&sdp
->sd_locking_init
);
88 init_completion(&sdp
->sd_wdack
);
89 spin_lock_init(&sdp
->sd_statfs_spin
);
91 spin_lock_init(&sdp
->sd_rindex_spin
);
92 sdp
->sd_rindex_tree
.rb_node
= NULL
;
94 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
95 spin_lock_init(&sdp
->sd_jindex_spin
);
96 mutex_init(&sdp
->sd_jindex_mutex
);
97 init_completion(&sdp
->sd_journal_ready
);
99 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
100 mutex_init(&sdp
->sd_quota_mutex
);
101 mutex_init(&sdp
->sd_quota_sync_mutex
);
102 init_waitqueue_head(&sdp
->sd_quota_wait
);
103 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
104 spin_lock_init(&sdp
->sd_trunc_lock
);
105 spin_lock_init(&sdp
->sd_bitmap_lock
);
107 mapping
= &sdp
->sd_aspace
;
109 address_space_init_once(mapping
);
110 mapping
->a_ops
= &gfs2_rgrp_aops
;
111 mapping
->host
= sb
->s_bdev
->bd_inode
;
113 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
114 mapping
->private_data
= NULL
;
115 mapping
->writeback_index
= 0;
117 spin_lock_init(&sdp
->sd_log_lock
);
118 atomic_set(&sdp
->sd_log_pinned
, 0);
119 INIT_LIST_HEAD(&sdp
->sd_log_le_revoke
);
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 atomic_set(&sdp
->sd_reserving_log
, 0);
132 init_waitqueue_head(&sdp
->sd_reserving_log_wait
);
133 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
134 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
135 mutex_init(&sdp
->sd_freeze_mutex
);
142 * gfs2_check_sb - Check superblock
143 * @sdp: the filesystem
144 * @sb: The superblock
145 * @silent: Don't print a message if the check fails
147 * Checks the version code of the FS is one that we understand how to
148 * read and that the sizes of the various on-disk structures have not
152 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
154 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
156 if (sb
->sb_magic
!= GFS2_MAGIC
||
157 sb
->sb_type
!= GFS2_METATYPE_SB
) {
159 pr_warn("not a GFS2 filesystem\n");
163 /* If format numbers match exactly, we're done. */
165 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
166 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
169 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
174 static void end_bio_io_page(struct bio
*bio
)
176 struct page
*page
= bio
->bi_private
;
179 SetPageUptodate(page
);
181 pr_warn("error %d reading superblock\n", bio
->bi_error
);
185 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
187 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
188 struct super_block
*s
= sdp
->sd_vfs
;
189 const struct gfs2_sb
*str
= buf
;
191 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
192 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
193 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
194 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
195 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
196 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
197 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
198 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
199 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
200 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
201 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
203 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
204 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
205 memcpy(s
->s_uuid
, str
->sb_uuid
, 16);
209 * gfs2_read_super - Read the gfs2 super block from disk
210 * @sdp: The GFS2 super block
211 * @sector: The location of the super block
212 * @error: The error code to return
214 * This uses the bio functions to read the super block from disk
215 * because we want to be 100% sure that we never read cached data.
216 * A super block is read twice only during each GFS2 mount and is
217 * never written to by the filesystem. The first time its read no
218 * locks are held, and the only details which are looked at are those
219 * relating to the locking protocol. Once locking is up and working,
220 * the sb is read again under the lock to establish the location of
221 * the master directory (contains pointers to journals etc) and the
224 * Returns: 0 on success or error
227 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
229 struct super_block
*sb
= sdp
->sd_vfs
;
234 page
= alloc_page(GFP_NOFS
);
238 ClearPageUptodate(page
);
239 ClearPageDirty(page
);
242 bio
= bio_alloc(GFP_NOFS
, 1);
243 bio
->bi_iter
.bi_sector
= sector
* (sb
->s_blocksize
>> 9);
244 bio
->bi_bdev
= sb
->s_bdev
;
245 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
247 bio
->bi_end_io
= end_bio_io_page
;
248 bio
->bi_private
= page
;
249 bio
->bi_rw
= READ_SYNC
| REQ_META
;
251 wait_on_page_locked(page
);
253 if (!PageUptodate(page
)) {
261 return gfs2_check_sb(sdp
, silent
);
265 * gfs2_read_sb - Read super block
266 * @sdp: The GFS2 superblock
267 * @silent: Don't print message if mount fails
271 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
273 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
278 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
281 fs_err(sdp
, "can't read superblock\n");
285 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
286 GFS2_BASIC_BLOCK_SHIFT
;
287 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
288 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
289 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
290 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
291 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
292 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
293 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
294 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
295 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
296 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
297 sizeof(struct gfs2_meta_header
)) /
298 sizeof(struct gfs2_quota_change
);
299 sdp
->sd_blocks_per_bitmap
= (sdp
->sd_sb
.sb_bsize
-
300 sizeof(struct gfs2_meta_header
))
301 * GFS2_NBBY
; /* not the rgrp bitmap, subsequent bitmaps only */
303 /* Compute maximum reservation required to add a entry to a directory */
305 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * (1 << GFS2_DIR_MAX_DEPTH
),
309 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
310 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
311 ind_blocks
+= tmp_blocks
;
314 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
316 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
318 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
319 sizeof(struct gfs2_dinode
);
320 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
325 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
327 m
= do_div(d
, sdp
->sd_inptrs
);
329 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
331 sdp
->sd_heightsize
[x
] = space
;
333 sdp
->sd_max_height
= x
;
334 sdp
->sd_heightsize
[x
] = ~0;
335 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
337 sdp
->sd_jheightsize
[0] = sdp
->sd_sb
.sb_bsize
-
338 sizeof(struct gfs2_dinode
);
339 sdp
->sd_jheightsize
[1] = sdp
->sd_jbsize
* sdp
->sd_diptrs
;
344 space
= sdp
->sd_jheightsize
[x
- 1] * sdp
->sd_inptrs
;
346 m
= do_div(d
, sdp
->sd_inptrs
);
348 if (d
!= sdp
->sd_jheightsize
[x
- 1] || m
)
350 sdp
->sd_jheightsize
[x
] = space
;
352 sdp
->sd_max_jheight
= x
;
353 sdp
->sd_jheightsize
[x
] = ~0;
354 gfs2_assert(sdp
, sdp
->sd_max_jheight
<= GFS2_MAX_META_HEIGHT
);
356 sdp
->sd_max_dents_per_leaf
= (sdp
->sd_sb
.sb_bsize
-
357 sizeof(struct gfs2_leaf
)) /
358 GFS2_MIN_DIRENT_SIZE
;
362 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
367 proto
= sdp
->sd_args
.ar_lockproto
;
368 table
= sdp
->sd_args
.ar_locktable
;
370 /* Try to autodetect */
372 if (!proto
[0] || !table
[0]) {
373 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
378 proto
= sdp
->sd_sb
.sb_lockproto
;
380 table
= sdp
->sd_sb
.sb_locktable
;
384 table
= sdp
->sd_vfs
->s_id
;
386 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
387 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
389 table
= sdp
->sd_table_name
;
390 while ((table
= strchr(table
, '/')))
396 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
404 error
= gfs2_glock_nq_num(sdp
,
405 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
406 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
409 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
413 error
= gfs2_glock_nq_num(sdp
,
414 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
416 LM_FLAG_NOEXP
| GL_EXACT
,
419 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
423 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
424 CREATE
, &sdp
->sd_rename_gl
);
426 fs_err(sdp
, "can't create rename glock: %d\n", error
);
430 error
= gfs2_glock_get(sdp
, GFS2_FREEZE_LOCK
, &gfs2_freeze_glops
,
431 CREATE
, &sdp
->sd_freeze_gl
);
433 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
440 gfs2_glock_put(sdp
->sd_freeze_gl
);
442 gfs2_glock_put(sdp
->sd_rename_gl
);
444 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
446 gfs2_glock_dq_uninit(mount_gh
);
451 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
452 u64 no_addr
, const char *name
)
454 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
455 struct dentry
*dentry
;
458 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0);
460 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
461 return PTR_ERR(inode
);
463 dentry
= d_make_root(inode
);
465 fs_err(sdp
, "can't alloc %s dentry\n", name
);
472 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
474 struct super_block
*sb
= sdp
->sd_vfs
;
475 struct gfs2_holder sb_gh
;
479 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
480 LM_ST_SHARED
, 0, &sb_gh
);
482 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
486 ret
= gfs2_read_sb(sdp
, silent
);
488 fs_err(sdp
, "can't read superblock: %d\n", ret
);
492 /* Set up the buffer cache and SB for real */
493 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
495 fs_err(sdp
, "FS block size (%u) is too small for device "
497 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
500 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
502 fs_err(sdp
, "FS block size (%u) is too big for machine "
504 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
507 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
509 /* Get the root inode */
510 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
511 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
515 /* Get the master inode */
516 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
517 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
519 dput(sdp
->sd_root_dir
);
522 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
524 gfs2_glock_dq_uninit(&sb_gh
);
528 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
530 char *message
= "FIRSTMOUNT=Done";
531 char *envp
[] = { message
, NULL
};
533 fs_info(sdp
, "first mount done, others may mount\n");
535 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
536 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
538 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
542 * gfs2_jindex_hold - Grab a lock on the jindex
543 * @sdp: The GFS2 superblock
544 * @ji_gh: the holder for the jindex glock
549 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
551 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
554 struct gfs2_jdesc
*jd
;
559 mutex_lock(&sdp
->sd_jindex_mutex
);
562 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
566 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
567 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
569 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
570 if (error
== -ENOENT
) {
575 gfs2_glock_dq_uninit(ji_gh
);
581 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
585 INIT_LIST_HEAD(&jd
->extent_list
);
586 INIT_LIST_HEAD(&jd
->jd_revoke_list
);
588 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
589 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
590 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
594 error
= PTR_ERR(jd
->jd_inode
);
599 spin_lock(&sdp
->sd_jindex_spin
);
600 jd
->jd_jid
= sdp
->sd_journals
++;
601 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
602 spin_unlock(&sdp
->sd_jindex_spin
);
605 mutex_unlock(&sdp
->sd_jindex_mutex
);
611 * check_journal_clean - Make sure a journal is clean for a spectator mount
612 * @sdp: The GFS2 superblock
613 * @jd: The journal descriptor
615 * Returns: 0 if the journal is clean or locked, else an error
617 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
620 struct gfs2_holder j_gh
;
621 struct gfs2_log_header_host head
;
622 struct gfs2_inode
*ip
;
624 ip
= GFS2_I(jd
->jd_inode
);
625 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
626 GL_EXACT
| GL_NOCACHE
, &j_gh
);
628 fs_err(sdp
, "Error locking journal for spectator mount.\n");
631 error
= gfs2_jdesc_check(jd
);
633 fs_err(sdp
, "Error checking journal for spectator mount.\n");
636 error
= gfs2_find_jhead(jd
, &head
);
638 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
641 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
643 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
644 "must not be a spectator.\n", jd
->jd_jid
);
648 gfs2_glock_dq_uninit(&j_gh
);
652 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
654 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
655 struct gfs2_holder ji_gh
;
656 struct gfs2_inode
*ip
;
665 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
666 if (IS_ERR(sdp
->sd_jindex
)) {
667 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
668 return PTR_ERR(sdp
->sd_jindex
);
671 /* Load in the journal index special file */
673 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
675 fs_err(sdp
, "can't read journal index: %d\n", error
);
680 if (!gfs2_jindex_size(sdp
)) {
681 fs_err(sdp
, "no journals!\n");
685 if (sdp
->sd_args
.ar_spectator
) {
686 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
687 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
688 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
689 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
691 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
692 fs_err(sdp
, "can't mount journal #%u\n",
693 sdp
->sd_lockstruct
.ls_jid
);
694 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
695 gfs2_jindex_size(sdp
),
696 gfs2_jindex_size(sdp
) - 1);
699 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
701 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
703 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
704 &sdp
->sd_journal_gh
);
706 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
710 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
711 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
712 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
715 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
717 goto fail_journal_gh
;
720 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
722 fs_err(sdp
, "my journal (%u) is bad: %d\n",
723 sdp
->sd_jdesc
->jd_jid
, error
);
726 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
727 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
728 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
730 /* Map the extents for this journal's blocks */
731 gfs2_map_journal_extents(sdp
, sdp
->sd_jdesc
);
733 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
735 if (sdp
->sd_lockstruct
.ls_first
) {
737 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
738 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
740 if (sdp
->sd_args
.ar_spectator
) {
741 error
= check_journal_clean(sdp
, jd
);
746 error
= gfs2_recover_journal(jd
, true);
748 fs_err(sdp
, "error recovering journal %u: %d\n",
754 gfs2_others_may_mount(sdp
);
755 } else if (!sdp
->sd_args
.ar_spectator
) {
756 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
758 fs_err(sdp
, "error recovering my journal: %d\n", error
);
763 sdp
->sd_log_idle
= 1;
764 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
765 gfs2_glock_dq_uninit(&ji_gh
);
767 INIT_WORK(&sdp
->sd_freeze_work
, gfs2_freeze_func
);
771 if (!sdp
->sd_args
.ar_spectator
)
772 gfs2_glock_dq_uninit(&sdp
->sd_jinode_gh
);
774 if (!sdp
->sd_args
.ar_spectator
)
775 gfs2_glock_dq_uninit(&sdp
->sd_journal_gh
);
777 gfs2_jindex_free(sdp
);
779 gfs2_glock_dq_uninit(&ji_gh
);
781 iput(sdp
->sd_jindex
);
785 static struct lock_class_key gfs2_quota_imutex_key
;
787 static int init_inodes(struct gfs2_sbd
*sdp
, int undo
)
790 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
795 error
= init_journal(sdp
, undo
);
796 complete_all(&sdp
->sd_journal_ready
);
800 /* Read in the master statfs inode */
801 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
802 if (IS_ERR(sdp
->sd_statfs_inode
)) {
803 error
= PTR_ERR(sdp
->sd_statfs_inode
);
804 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
808 /* Read in the resource index inode */
809 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
810 if (IS_ERR(sdp
->sd_rindex
)) {
811 error
= PTR_ERR(sdp
->sd_rindex
);
812 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
815 sdp
->sd_rindex_uptodate
= 0;
817 /* Read in the quota inode */
818 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
819 if (IS_ERR(sdp
->sd_quota_inode
)) {
820 error
= PTR_ERR(sdp
->sd_quota_inode
);
821 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
825 * i_mutex on quota files is special. Since this inode is hidden system
826 * file, we are safe to define locking ourselves.
828 lockdep_set_class(&sdp
->sd_quota_inode
->i_rwsem
,
829 &gfs2_quota_imutex_key
);
831 error
= gfs2_rindex_update(sdp
);
838 iput(sdp
->sd_quota_inode
);
840 gfs2_clear_rgrpd(sdp
);
841 iput(sdp
->sd_rindex
);
843 iput(sdp
->sd_statfs_inode
);
845 init_journal(sdp
, UNDO
);
850 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
852 struct inode
*pn
= NULL
;
855 struct gfs2_inode
*ip
;
856 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
858 if (sdp
->sd_args
.ar_spectator
)
864 pn
= gfs2_lookup_simple(master
, "per_node");
867 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
871 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
872 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
873 if (IS_ERR(sdp
->sd_sc_inode
)) {
874 error
= PTR_ERR(sdp
->sd_sc_inode
);
875 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
879 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
880 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
881 if (IS_ERR(sdp
->sd_qc_inode
)) {
882 error
= PTR_ERR(sdp
->sd_qc_inode
);
883 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
890 ip
= GFS2_I(sdp
->sd_sc_inode
);
891 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
894 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
898 ip
= GFS2_I(sdp
->sd_qc_inode
);
899 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
902 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
909 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
911 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
913 iput(sdp
->sd_qc_inode
);
915 iput(sdp
->sd_sc_inode
);
921 static const match_table_t nolock_tokens
= {
922 { Opt_jid
, "jid=%d\n", },
926 static const struct lm_lockops nolock_ops
= {
927 .lm_proto_name
= "lock_nolock",
928 .lm_put_lock
= gfs2_glock_free
,
929 .lm_tokens
= &nolock_tokens
,
933 * gfs2_lm_mount - mount a locking protocol
934 * @sdp: the filesystem
935 * @args: mount arguments
936 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
941 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
943 const struct lm_lockops
*lm
;
944 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
945 struct gfs2_args
*args
= &sdp
->sd_args
;
946 const char *proto
= sdp
->sd_proto_name
;
947 const char *table
= sdp
->sd_table_name
;
951 if (!strcmp("lock_nolock", proto
)) {
953 sdp
->sd_args
.ar_localflocks
= 1;
954 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
955 } else if (!strcmp("lock_dlm", proto
)) {
959 pr_info("can't find protocol %s\n", proto
);
963 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
968 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
969 substring_t tmp
[MAX_OPT_ARGS
];
975 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
978 ret
= match_int(&tmp
[0], &option
);
979 if (ret
|| option
< 0)
981 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
986 /* Obsolete, but left for backward compat purposes */
989 ret
= match_int(&tmp
[0], &option
);
990 if (ret
|| (option
!= 0 && option
!= 1))
992 ls
->ls_first
= option
;
997 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
1002 if (lm
->lm_mount
== NULL
) {
1003 fs_info(sdp
, "Now mounting FS...\n");
1004 complete_all(&sdp
->sd_locking_init
);
1007 ret
= lm
->lm_mount(sdp
, table
);
1009 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
1010 complete_all(&sdp
->sd_locking_init
);
1014 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1016 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1017 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1019 lm
->lm_unmount(sdp
);
1022 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1024 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1027 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, TASK_INTERRUPTIBLE
)
1031 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1033 struct super_block
*sb
= sdp
->sd_vfs
;
1036 char *envp
[] = { ro
, spectator
, NULL
};
1037 sprintf(ro
, "RDONLY=%d", (sb
->s_flags
& MS_RDONLY
) ? 1 : 0);
1038 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1039 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1043 * fill_super - Read in superblock
1044 * @sb: The VFS superblock
1045 * @data: Mount options
1046 * @silent: Don't complain if it's not a GFS2 filesystem
1051 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1053 struct gfs2_sbd
*sdp
;
1054 struct gfs2_holder mount_gh
;
1059 pr_warn("can't alloc struct gfs2_sbd\n");
1062 sdp
->sd_args
= *args
;
1064 if (sdp
->sd_args
.ar_spectator
) {
1065 sb
->s_flags
|= MS_RDONLY
;
1066 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1068 if (sdp
->sd_args
.ar_posix_acl
)
1069 sb
->s_flags
|= MS_POSIXACL
;
1070 if (sdp
->sd_args
.ar_nobarrier
)
1071 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1073 sb
->s_flags
|= MS_NOSEC
;
1074 sb
->s_magic
= GFS2_MAGIC
;
1075 sb
->s_op
= &gfs2_super_ops
;
1076 sb
->s_d_op
= &gfs2_dops
;
1077 sb
->s_export_op
= &gfs2_export_ops
;
1078 sb
->s_xattr
= gfs2_xattr_handlers
;
1079 sb
->s_qcop
= &gfs2_quotactl_ops
;
1080 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
1081 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1082 sb
->s_time_gran
= 1;
1083 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1085 /* Set up the buffer cache and fill in some fake block size values
1086 to allow us to read-in the on-disk superblock. */
1087 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1088 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1089 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1090 GFS2_BASIC_BLOCK_SHIFT
;
1091 sdp
->sd_fsb2bb
= 1 << sdp
->sd_fsb2bb_shift
;
1093 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1094 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1095 if (sdp
->sd_args
.ar_statfs_quantum
) {
1096 sdp
->sd_tune
.gt_statfs_slow
= 0;
1097 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1099 sdp
->sd_tune
.gt_statfs_slow
= 1;
1100 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1103 error
= init_names(sdp
, silent
);
1105 /* In this case, we haven't initialized sysfs, so we have to
1106 manually free the sdp. */
1107 free_percpu(sdp
->sd_lkstats
);
1109 sb
->s_fs_info
= NULL
;
1113 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s", sdp
->sd_table_name
);
1115 error
= gfs2_sys_fs_add(sdp
);
1117 * If we hit an error here, gfs2_sys_fs_add will have called function
1118 * kobject_put which causes the sysfs usage count to go to zero, which
1119 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1120 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1121 * kobject_put to free sdp.
1126 gfs2_create_debugfs_file(sdp
);
1128 error
= gfs2_lm_mount(sdp
, silent
);
1132 error
= init_locking(sdp
, &mount_gh
, DO
);
1136 error
= init_sb(sdp
, silent
);
1140 error
= wait_on_journal(sdp
);
1145 * If user space has failed to join the cluster or some similar
1146 * failure has occurred, then the journal id will contain a
1147 * negative (error) number. This will then be returned to the
1148 * caller (of the mount syscall). We do this even for spectator
1149 * mounts (which just write a jid of 0 to indicate "ok" even though
1150 * the jid is unused in the spectator case)
1152 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1153 error
= sdp
->sd_lockstruct
.ls_jid
;
1154 sdp
->sd_lockstruct
.ls_jid
= 0;
1158 if (sdp
->sd_args
.ar_spectator
)
1159 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.s",
1160 sdp
->sd_table_name
);
1162 snprintf(sdp
->sd_fsname
, GFS2_FSNAME_LEN
, "%s.%u",
1163 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1165 error
= init_inodes(sdp
, DO
);
1169 error
= init_per_node(sdp
, DO
);
1173 error
= gfs2_statfs_init(sdp
);
1175 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1179 if (!(sb
->s_flags
& MS_RDONLY
)) {
1180 error
= gfs2_make_fs_rw(sdp
);
1182 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1187 gfs2_glock_dq_uninit(&mount_gh
);
1188 gfs2_online_uevent(sdp
);
1192 init_per_node(sdp
, UNDO
);
1194 init_inodes(sdp
, UNDO
);
1196 if (sdp
->sd_root_dir
)
1197 dput(sdp
->sd_root_dir
);
1198 if (sdp
->sd_master_dir
)
1199 dput(sdp
->sd_master_dir
);
1204 init_locking(sdp
, &mount_gh
, UNDO
);
1206 complete_all(&sdp
->sd_journal_ready
);
1207 gfs2_gl_hash_clear(sdp
);
1208 gfs2_lm_unmount(sdp
);
1210 gfs2_delete_debugfs_file(sdp
);
1211 free_percpu(sdp
->sd_lkstats
);
1212 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1213 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1214 gfs2_sys_fs_del(sdp
);
1215 sb
->s_fs_info
= NULL
;
1219 static int set_gfs2_super(struct super_block
*s
, void *data
)
1222 s
->s_dev
= s
->s_bdev
->bd_dev
;
1225 * We set the bdi here to the queue backing, file systems can
1226 * overwrite this in ->fill_super()
1228 s
->s_bdi
= &bdev_get_queue(s
->s_bdev
)->backing_dev_info
;
1232 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1234 struct block_device
*bdev
= ptr
;
1235 return (bdev
== s
->s_bdev
);
1239 * gfs2_mount - Get the GFS2 superblock
1240 * @fs_type: The GFS2 filesystem type
1241 * @flags: Mount flags
1242 * @dev_name: The name of the device
1243 * @data: The mount arguments
1245 * Q. Why not use get_sb_bdev() ?
1246 * A. We need to select one of two root directories to mount, independent
1247 * of whether this is the initial, or subsequent, mount of this sb
1249 * Returns: 0 or -ve on error
1252 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1253 const char *dev_name
, void *data
)
1255 struct block_device
*bdev
;
1256 struct super_block
*s
;
1257 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1259 struct gfs2_args args
;
1260 struct gfs2_sbd
*sdp
;
1262 if (!(flags
& MS_RDONLY
))
1263 mode
|= FMODE_WRITE
;
1265 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1267 return ERR_CAST(bdev
);
1270 * once the super is inserted into the list by sget, s_umount
1271 * will protect the lockfs code from trying to start a snapshot
1272 * while we are mounting
1274 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1275 if (bdev
->bd_fsfreeze_count
> 0) {
1276 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1280 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1281 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1288 * s_umount nests inside bd_mutex during
1289 * __invalidate_device(). blkdev_put() acquires
1290 * bd_mutex and can't be called under s_umount. Drop
1291 * s_umount temporarily. This is safe as we're
1292 * holding an active reference.
1294 up_write(&s
->s_umount
);
1295 blkdev_put(bdev
, mode
);
1296 down_write(&s
->s_umount
);
1298 /* s_mode must be set before deactivate_locked_super calls */
1302 memset(&args
, 0, sizeof(args
));
1303 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1304 args
.ar_data
= GFS2_DATA_DEFAULT
;
1305 args
.ar_commit
= 30;
1306 args
.ar_statfs_quantum
= 30;
1307 args
.ar_quota_quantum
= 60;
1308 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1310 error
= gfs2_mount_args(&args
, data
);
1312 pr_warn("can't parse mount arguments\n");
1318 if ((flags
^ s
->s_flags
) & MS_RDONLY
)
1321 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1322 sb_set_blocksize(s
, block_size(bdev
));
1323 error
= fill_super(s
, &args
, flags
& MS_SILENT
? 1 : 0);
1326 s
->s_flags
|= MS_ACTIVE
;
1332 return dget(sdp
->sd_master_dir
);
1334 return dget(sdp
->sd_root_dir
);
1337 deactivate_locked_super(s
);
1338 return ERR_PTR(error
);
1340 blkdev_put(bdev
, mode
);
1341 return ERR_PTR(error
);
1344 static int set_meta_super(struct super_block
*s
, void *ptr
)
1349 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1350 int flags
, const char *dev_name
, void *data
)
1352 struct super_block
*s
;
1353 struct gfs2_sbd
*sdp
;
1357 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1359 pr_warn("path_lookup on %s returned error %d\n",
1361 return ERR_PTR(error
);
1363 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1364 path
.dentry
->d_sb
->s_bdev
);
1367 pr_warn("gfs2 mount does not exist\n");
1370 if ((flags
^ s
->s_flags
) & MS_RDONLY
) {
1371 deactivate_locked_super(s
);
1372 return ERR_PTR(-EBUSY
);
1375 return dget(sdp
->sd_master_dir
);
1378 static void gfs2_kill_sb(struct super_block
*sb
)
1380 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1383 kill_block_super(sb
);
1387 gfs2_log_flush(sdp
, NULL
, SYNC_FLUSH
);
1388 dput(sdp
->sd_root_dir
);
1389 dput(sdp
->sd_master_dir
);
1390 sdp
->sd_root_dir
= NULL
;
1391 sdp
->sd_master_dir
= NULL
;
1392 shrink_dcache_sb(sb
);
1393 gfs2_delete_debugfs_file(sdp
);
1394 free_percpu(sdp
->sd_lkstats
);
1395 kill_block_super(sb
);
1398 struct file_system_type gfs2_fs_type
= {
1400 .fs_flags
= FS_REQUIRES_DEV
,
1401 .mount
= gfs2_mount
,
1402 .kill_sb
= gfs2_kill_sb
,
1403 .owner
= THIS_MODULE
,
1405 MODULE_ALIAS_FS("gfs2");
1407 struct file_system_type gfs2meta_fs_type
= {
1409 .fs_flags
= FS_REQUIRES_DEV
,
1410 .mount
= gfs2_mount_meta
,
1411 .owner
= THIS_MODULE
,
1413 MODULE_ALIAS_FS("gfs2meta");