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>
26 #include <linux/backing-dev.h>
43 #include "trace_gfs2.h"
50 * gfs2_tune_init - Fill a gfs2_tune structure with default values
55 static void gfs2_tune_init(struct gfs2_tune
*gt
)
57 spin_lock_init(>
->gt_spin
);
59 gt
->gt_quota_warn_period
= 10;
60 gt
->gt_quota_scale_num
= 1;
61 gt
->gt_quota_scale_den
= 1;
62 gt
->gt_new_files_jdata
= 0;
63 gt
->gt_max_readahead
= BIT(18);
64 gt
->gt_complain_secs
= 10;
67 static struct gfs2_sbd
*init_sbd(struct super_block
*sb
)
70 struct address_space
*mapping
;
72 sdp
= kzalloc(sizeof(struct gfs2_sbd
), GFP_KERNEL
);
77 sdp
->sd_lkstats
= alloc_percpu(struct gfs2_pcpu_lkstats
);
78 if (!sdp
->sd_lkstats
) {
84 set_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
);
85 gfs2_tune_init(&sdp
->sd_tune
);
87 init_waitqueue_head(&sdp
->sd_glock_wait
);
88 atomic_set(&sdp
->sd_glock_disposal
, 0);
89 init_completion(&sdp
->sd_locking_init
);
90 init_completion(&sdp
->sd_wdack
);
91 spin_lock_init(&sdp
->sd_statfs_spin
);
93 spin_lock_init(&sdp
->sd_rindex_spin
);
94 sdp
->sd_rindex_tree
.rb_node
= NULL
;
96 INIT_LIST_HEAD(&sdp
->sd_jindex_list
);
97 spin_lock_init(&sdp
->sd_jindex_spin
);
98 mutex_init(&sdp
->sd_jindex_mutex
);
99 init_completion(&sdp
->sd_journal_ready
);
101 INIT_LIST_HEAD(&sdp
->sd_quota_list
);
102 mutex_init(&sdp
->sd_quota_mutex
);
103 mutex_init(&sdp
->sd_quota_sync_mutex
);
104 init_waitqueue_head(&sdp
->sd_quota_wait
);
105 INIT_LIST_HEAD(&sdp
->sd_trunc_list
);
106 spin_lock_init(&sdp
->sd_trunc_lock
);
107 spin_lock_init(&sdp
->sd_bitmap_lock
);
109 mapping
= &sdp
->sd_aspace
;
111 address_space_init_once(mapping
);
112 mapping
->a_ops
= &gfs2_rgrp_aops
;
113 mapping
->host
= sb
->s_bdev
->bd_inode
;
115 mapping_set_gfp_mask(mapping
, GFP_NOFS
);
116 mapping
->private_data
= NULL
;
117 mapping
->writeback_index
= 0;
119 spin_lock_init(&sdp
->sd_log_lock
);
120 atomic_set(&sdp
->sd_log_pinned
, 0);
121 INIT_LIST_HEAD(&sdp
->sd_log_le_revoke
);
122 INIT_LIST_HEAD(&sdp
->sd_log_le_ordered
);
123 spin_lock_init(&sdp
->sd_ordered_lock
);
125 init_waitqueue_head(&sdp
->sd_log_waitq
);
126 init_waitqueue_head(&sdp
->sd_logd_waitq
);
127 spin_lock_init(&sdp
->sd_ail_lock
);
128 INIT_LIST_HEAD(&sdp
->sd_ail1_list
);
129 INIT_LIST_HEAD(&sdp
->sd_ail2_list
);
131 init_rwsem(&sdp
->sd_log_flush_lock
);
132 atomic_set(&sdp
->sd_log_in_flight
, 0);
133 atomic_set(&sdp
->sd_reserving_log
, 0);
134 init_waitqueue_head(&sdp
->sd_reserving_log_wait
);
135 init_waitqueue_head(&sdp
->sd_log_flush_wait
);
136 atomic_set(&sdp
->sd_freeze_state
, SFS_UNFROZEN
);
137 mutex_init(&sdp
->sd_freeze_mutex
);
144 * gfs2_check_sb - Check superblock
145 * @sdp: the filesystem
146 * @sb: The superblock
147 * @silent: Don't print a message if the check fails
149 * Checks the version code of the FS is one that we understand how to
150 * read and that the sizes of the various on-disk structures have not
154 static int gfs2_check_sb(struct gfs2_sbd
*sdp
, int silent
)
156 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
158 if (sb
->sb_magic
!= GFS2_MAGIC
||
159 sb
->sb_type
!= GFS2_METATYPE_SB
) {
161 pr_warn("not a GFS2 filesystem\n");
165 /* If format numbers match exactly, we're done. */
167 if (sb
->sb_fs_format
== GFS2_FORMAT_FS
&&
168 sb
->sb_multihost_format
== GFS2_FORMAT_MULTI
)
171 fs_warn(sdp
, "Unknown on-disk format, unable to mount\n");
176 static void end_bio_io_page(struct bio
*bio
)
178 struct page
*page
= bio
->bi_private
;
181 SetPageUptodate(page
);
183 pr_warn("error %d reading superblock\n", bio
->bi_status
);
187 static void gfs2_sb_in(struct gfs2_sbd
*sdp
, const void *buf
)
189 struct gfs2_sb_host
*sb
= &sdp
->sd_sb
;
190 struct super_block
*s
= sdp
->sd_vfs
;
191 const struct gfs2_sb
*str
= buf
;
193 sb
->sb_magic
= be32_to_cpu(str
->sb_header
.mh_magic
);
194 sb
->sb_type
= be32_to_cpu(str
->sb_header
.mh_type
);
195 sb
->sb_format
= be32_to_cpu(str
->sb_header
.mh_format
);
196 sb
->sb_fs_format
= be32_to_cpu(str
->sb_fs_format
);
197 sb
->sb_multihost_format
= be32_to_cpu(str
->sb_multihost_format
);
198 sb
->sb_bsize
= be32_to_cpu(str
->sb_bsize
);
199 sb
->sb_bsize_shift
= be32_to_cpu(str
->sb_bsize_shift
);
200 sb
->sb_master_dir
.no_addr
= be64_to_cpu(str
->sb_master_dir
.no_addr
);
201 sb
->sb_master_dir
.no_formal_ino
= be64_to_cpu(str
->sb_master_dir
.no_formal_ino
);
202 sb
->sb_root_dir
.no_addr
= be64_to_cpu(str
->sb_root_dir
.no_addr
);
203 sb
->sb_root_dir
.no_formal_ino
= be64_to_cpu(str
->sb_root_dir
.no_formal_ino
);
205 memcpy(sb
->sb_lockproto
, str
->sb_lockproto
, GFS2_LOCKNAME_LEN
);
206 memcpy(sb
->sb_locktable
, str
->sb_locktable
, GFS2_LOCKNAME_LEN
);
207 memcpy(&s
->s_uuid
, str
->sb_uuid
, 16);
211 * gfs2_read_super - Read the gfs2 super block from disk
212 * @sdp: The GFS2 super block
213 * @sector: The location of the super block
214 * @error: The error code to return
216 * This uses the bio functions to read the super block from disk
217 * because we want to be 100% sure that we never read cached data.
218 * A super block is read twice only during each GFS2 mount and is
219 * never written to by the filesystem. The first time its read no
220 * locks are held, and the only details which are looked at are those
221 * relating to the locking protocol. Once locking is up and working,
222 * the sb is read again under the lock to establish the location of
223 * the master directory (contains pointers to journals etc) and the
226 * Returns: 0 on success or error
229 static int gfs2_read_super(struct gfs2_sbd
*sdp
, sector_t sector
, int silent
)
231 struct super_block
*sb
= sdp
->sd_vfs
;
236 page
= alloc_page(GFP_NOFS
);
240 ClearPageUptodate(page
);
241 ClearPageDirty(page
);
244 bio
= bio_alloc(GFP_NOFS
, 1);
245 bio
->bi_iter
.bi_sector
= sector
* (sb
->s_blocksize
>> 9);
246 bio_set_dev(bio
, sb
->s_bdev
);
247 bio_add_page(bio
, page
, PAGE_SIZE
, 0);
249 bio
->bi_end_io
= end_bio_io_page
;
250 bio
->bi_private
= page
;
251 bio_set_op_attrs(bio
, REQ_OP_READ
, REQ_META
);
253 wait_on_page_locked(page
);
255 if (!PageUptodate(page
)) {
263 return gfs2_check_sb(sdp
, silent
);
267 * gfs2_read_sb - Read super block
268 * @sdp: The GFS2 superblock
269 * @silent: Don't print message if mount fails
273 static int gfs2_read_sb(struct gfs2_sbd
*sdp
, int silent
)
275 u32 hash_blocks
, ind_blocks
, leaf_blocks
;
280 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
283 fs_err(sdp
, "can't read superblock\n");
287 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
288 GFS2_BASIC_BLOCK_SHIFT
;
289 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
290 sdp
->sd_diptrs
= (sdp
->sd_sb
.sb_bsize
-
291 sizeof(struct gfs2_dinode
)) / sizeof(u64
);
292 sdp
->sd_inptrs
= (sdp
->sd_sb
.sb_bsize
-
293 sizeof(struct gfs2_meta_header
)) / sizeof(u64
);
294 sdp
->sd_jbsize
= sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_meta_header
);
295 sdp
->sd_hash_bsize
= sdp
->sd_sb
.sb_bsize
/ 2;
296 sdp
->sd_hash_bsize_shift
= sdp
->sd_sb
.sb_bsize_shift
- 1;
297 sdp
->sd_hash_ptrs
= sdp
->sd_hash_bsize
/ sizeof(u64
);
298 sdp
->sd_qc_per_block
= (sdp
->sd_sb
.sb_bsize
-
299 sizeof(struct gfs2_meta_header
)) /
300 sizeof(struct gfs2_quota_change
);
301 sdp
->sd_blocks_per_bitmap
= (sdp
->sd_sb
.sb_bsize
-
302 sizeof(struct gfs2_meta_header
))
303 * GFS2_NBBY
; /* not the rgrp bitmap, subsequent bitmaps only */
305 /* Compute maximum reservation required to add a entry to a directory */
307 hash_blocks
= DIV_ROUND_UP(sizeof(u64
) * BIT(GFS2_DIR_MAX_DEPTH
),
311 for (tmp_blocks
= hash_blocks
; tmp_blocks
> sdp
->sd_diptrs
;) {
312 tmp_blocks
= DIV_ROUND_UP(tmp_blocks
, sdp
->sd_inptrs
);
313 ind_blocks
+= tmp_blocks
;
316 leaf_blocks
= 2 + GFS2_DIR_MAX_DEPTH
;
318 sdp
->sd_max_dirres
= hash_blocks
+ ind_blocks
+ leaf_blocks
;
320 sdp
->sd_heightsize
[0] = sdp
->sd_sb
.sb_bsize
-
321 sizeof(struct gfs2_dinode
);
322 sdp
->sd_heightsize
[1] = sdp
->sd_sb
.sb_bsize
* sdp
->sd_diptrs
;
327 space
= sdp
->sd_heightsize
[x
- 1] * sdp
->sd_inptrs
;
329 m
= do_div(d
, sdp
->sd_inptrs
);
331 if (d
!= sdp
->sd_heightsize
[x
- 1] || m
)
333 sdp
->sd_heightsize
[x
] = space
;
335 sdp
->sd_max_height
= x
;
336 sdp
->sd_heightsize
[x
] = ~0;
337 gfs2_assert(sdp
, sdp
->sd_max_height
<= GFS2_MAX_META_HEIGHT
);
339 sdp
->sd_max_dents_per_leaf
= (sdp
->sd_sb
.sb_bsize
-
340 sizeof(struct gfs2_leaf
)) /
341 GFS2_MIN_DIRENT_SIZE
;
345 static int init_names(struct gfs2_sbd
*sdp
, int silent
)
350 proto
= sdp
->sd_args
.ar_lockproto
;
351 table
= sdp
->sd_args
.ar_locktable
;
353 /* Try to autodetect */
355 if (!proto
[0] || !table
[0]) {
356 error
= gfs2_read_super(sdp
, GFS2_SB_ADDR
>> sdp
->sd_fsb2bb_shift
, silent
);
361 proto
= sdp
->sd_sb
.sb_lockproto
;
363 table
= sdp
->sd_sb
.sb_locktable
;
367 table
= sdp
->sd_vfs
->s_id
;
369 strlcpy(sdp
->sd_proto_name
, proto
, GFS2_FSNAME_LEN
);
370 strlcpy(sdp
->sd_table_name
, table
, GFS2_FSNAME_LEN
);
372 table
= sdp
->sd_table_name
;
373 while ((table
= strchr(table
, '/')))
379 static int init_locking(struct gfs2_sbd
*sdp
, struct gfs2_holder
*mount_gh
,
387 error
= gfs2_glock_nq_num(sdp
,
388 GFS2_MOUNT_LOCK
, &gfs2_nondisk_glops
,
389 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
| GL_NOCACHE
,
392 fs_err(sdp
, "can't acquire mount glock: %d\n", error
);
396 error
= gfs2_glock_nq_num(sdp
,
397 GFS2_LIVE_LOCK
, &gfs2_nondisk_glops
,
399 LM_FLAG_NOEXP
| GL_EXACT
,
402 fs_err(sdp
, "can't acquire live glock: %d\n", error
);
406 error
= gfs2_glock_get(sdp
, GFS2_RENAME_LOCK
, &gfs2_nondisk_glops
,
407 CREATE
, &sdp
->sd_rename_gl
);
409 fs_err(sdp
, "can't create rename glock: %d\n", error
);
413 error
= gfs2_glock_get(sdp
, GFS2_FREEZE_LOCK
, &gfs2_freeze_glops
,
414 CREATE
, &sdp
->sd_freeze_gl
);
416 fs_err(sdp
, "can't create transaction glock: %d\n", error
);
423 gfs2_glock_put(sdp
->sd_freeze_gl
);
425 gfs2_glock_put(sdp
->sd_rename_gl
);
427 gfs2_glock_dq_uninit(&sdp
->sd_live_gh
);
429 gfs2_glock_dq_uninit(mount_gh
);
434 static int gfs2_lookup_root(struct super_block
*sb
, struct dentry
**dptr
,
435 u64 no_addr
, const char *name
)
437 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
438 struct dentry
*dentry
;
441 inode
= gfs2_inode_lookup(sb
, DT_DIR
, no_addr
, 0,
442 GFS2_BLKST_FREE
/* ignore */);
444 fs_err(sdp
, "can't read in %s inode: %ld\n", name
, PTR_ERR(inode
));
445 return PTR_ERR(inode
);
447 dentry
= d_make_root(inode
);
449 fs_err(sdp
, "can't alloc %s dentry\n", name
);
456 static int init_sb(struct gfs2_sbd
*sdp
, int silent
)
458 struct super_block
*sb
= sdp
->sd_vfs
;
459 struct gfs2_holder sb_gh
;
463 ret
= gfs2_glock_nq_num(sdp
, GFS2_SB_LOCK
, &gfs2_meta_glops
,
464 LM_ST_SHARED
, 0, &sb_gh
);
466 fs_err(sdp
, "can't acquire superblock glock: %d\n", ret
);
470 ret
= gfs2_read_sb(sdp
, silent
);
472 fs_err(sdp
, "can't read superblock: %d\n", ret
);
476 /* Set up the buffer cache and SB for real */
477 if (sdp
->sd_sb
.sb_bsize
< bdev_logical_block_size(sb
->s_bdev
)) {
479 fs_err(sdp
, "FS block size (%u) is too small for device "
481 sdp
->sd_sb
.sb_bsize
, bdev_logical_block_size(sb
->s_bdev
));
484 if (sdp
->sd_sb
.sb_bsize
> PAGE_SIZE
) {
486 fs_err(sdp
, "FS block size (%u) is too big for machine "
488 sdp
->sd_sb
.sb_bsize
, (unsigned int)PAGE_SIZE
);
491 sb_set_blocksize(sb
, sdp
->sd_sb
.sb_bsize
);
493 /* Get the root inode */
494 no_addr
= sdp
->sd_sb
.sb_root_dir
.no_addr
;
495 ret
= gfs2_lookup_root(sb
, &sdp
->sd_root_dir
, no_addr
, "root");
499 /* Get the master inode */
500 no_addr
= sdp
->sd_sb
.sb_master_dir
.no_addr
;
501 ret
= gfs2_lookup_root(sb
, &sdp
->sd_master_dir
, no_addr
, "master");
503 dput(sdp
->sd_root_dir
);
506 sb
->s_root
= dget(sdp
->sd_args
.ar_meta
? sdp
->sd_master_dir
: sdp
->sd_root_dir
);
508 gfs2_glock_dq_uninit(&sb_gh
);
512 static void gfs2_others_may_mount(struct gfs2_sbd
*sdp
)
514 char *message
= "FIRSTMOUNT=Done";
515 char *envp
[] = { message
, NULL
};
517 fs_info(sdp
, "first mount done, others may mount\n");
519 if (sdp
->sd_lockstruct
.ls_ops
->lm_first_done
)
520 sdp
->sd_lockstruct
.ls_ops
->lm_first_done(sdp
);
522 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_CHANGE
, envp
);
526 * gfs2_jindex_hold - Grab a lock on the jindex
527 * @sdp: The GFS2 superblock
528 * @ji_gh: the holder for the jindex glock
533 static int gfs2_jindex_hold(struct gfs2_sbd
*sdp
, struct gfs2_holder
*ji_gh
)
535 struct gfs2_inode
*dip
= GFS2_I(sdp
->sd_jindex
);
538 struct gfs2_jdesc
*jd
;
543 mutex_lock(&sdp
->sd_jindex_mutex
);
546 error
= gfs2_glock_nq_init(dip
->i_gl
, LM_ST_SHARED
, 0, ji_gh
);
550 name
.len
= sprintf(buf
, "journal%u", sdp
->sd_journals
);
551 name
.hash
= gfs2_disk_hash(name
.name
, name
.len
);
553 error
= gfs2_dir_check(sdp
->sd_jindex
, &name
, NULL
);
554 if (error
== -ENOENT
) {
559 gfs2_glock_dq_uninit(ji_gh
);
565 jd
= kzalloc(sizeof(struct gfs2_jdesc
), GFP_KERNEL
);
569 INIT_LIST_HEAD(&jd
->extent_list
);
570 INIT_LIST_HEAD(&jd
->jd_revoke_list
);
572 INIT_WORK(&jd
->jd_work
, gfs2_recover_func
);
573 jd
->jd_inode
= gfs2_lookupi(sdp
->sd_jindex
, &name
, 1);
574 if (!jd
->jd_inode
|| IS_ERR(jd
->jd_inode
)) {
578 error
= PTR_ERR(jd
->jd_inode
);
583 spin_lock(&sdp
->sd_jindex_spin
);
584 jd
->jd_jid
= sdp
->sd_journals
++;
585 list_add_tail(&jd
->jd_list
, &sdp
->sd_jindex_list
);
586 spin_unlock(&sdp
->sd_jindex_spin
);
589 mutex_unlock(&sdp
->sd_jindex_mutex
);
595 * check_journal_clean - Make sure a journal is clean for a spectator mount
596 * @sdp: The GFS2 superblock
597 * @jd: The journal descriptor
599 * Returns: 0 if the journal is clean or locked, else an error
601 static int check_journal_clean(struct gfs2_sbd
*sdp
, struct gfs2_jdesc
*jd
)
604 struct gfs2_holder j_gh
;
605 struct gfs2_log_header_host head
;
606 struct gfs2_inode
*ip
;
608 ip
= GFS2_I(jd
->jd_inode
);
609 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_NOEXP
|
610 GL_EXACT
| GL_NOCACHE
, &j_gh
);
612 fs_err(sdp
, "Error locking journal for spectator mount.\n");
615 error
= gfs2_jdesc_check(jd
);
617 fs_err(sdp
, "Error checking journal for spectator mount.\n");
620 error
= gfs2_find_jhead(jd
, &head
);
622 fs_err(sdp
, "Error parsing journal for spectator mount.\n");
625 if (!(head
.lh_flags
& GFS2_LOG_HEAD_UNMOUNT
)) {
627 fs_err(sdp
, "jid=%u: Journal is dirty, so the first mounter "
628 "must not be a spectator.\n", jd
->jd_jid
);
632 gfs2_glock_dq_uninit(&j_gh
);
636 static int init_journal(struct gfs2_sbd
*sdp
, int undo
)
638 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
639 struct gfs2_holder ji_gh
;
640 struct gfs2_inode
*ip
;
649 sdp
->sd_jindex
= gfs2_lookup_simple(master
, "jindex");
650 if (IS_ERR(sdp
->sd_jindex
)) {
651 fs_err(sdp
, "can't lookup journal index: %d\n", error
);
652 return PTR_ERR(sdp
->sd_jindex
);
655 /* Load in the journal index special file */
657 error
= gfs2_jindex_hold(sdp
, &ji_gh
);
659 fs_err(sdp
, "can't read journal index: %d\n", error
);
664 if (!gfs2_jindex_size(sdp
)) {
665 fs_err(sdp
, "no journals!\n");
669 atomic_set(&sdp
->sd_log_blks_needed
, 0);
670 if (sdp
->sd_args
.ar_spectator
) {
671 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, 0);
672 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
673 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
674 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
676 if (sdp
->sd_lockstruct
.ls_jid
>= gfs2_jindex_size(sdp
)) {
677 fs_err(sdp
, "can't mount journal #%u\n",
678 sdp
->sd_lockstruct
.ls_jid
);
679 fs_err(sdp
, "there are only %u journals (0 - %u)\n",
680 gfs2_jindex_size(sdp
),
681 gfs2_jindex_size(sdp
) - 1);
684 sdp
->sd_jdesc
= gfs2_jdesc_find(sdp
, sdp
->sd_lockstruct
.ls_jid
);
686 error
= gfs2_glock_nq_num(sdp
, sdp
->sd_lockstruct
.ls_jid
,
688 LM_ST_EXCLUSIVE
, LM_FLAG_NOEXP
,
689 &sdp
->sd_journal_gh
);
691 fs_err(sdp
, "can't acquire journal glock: %d\n", error
);
695 ip
= GFS2_I(sdp
->sd_jdesc
->jd_inode
);
696 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
,
697 LM_FLAG_NOEXP
| GL_EXACT
| GL_NOCACHE
,
700 fs_err(sdp
, "can't acquire journal inode glock: %d\n",
702 goto fail_journal_gh
;
705 error
= gfs2_jdesc_check(sdp
->sd_jdesc
);
707 fs_err(sdp
, "my journal (%u) is bad: %d\n",
708 sdp
->sd_jdesc
->jd_jid
, error
);
711 atomic_set(&sdp
->sd_log_blks_free
, sdp
->sd_jdesc
->jd_blocks
);
712 atomic_set(&sdp
->sd_log_thresh1
, 2*sdp
->sd_jdesc
->jd_blocks
/5);
713 atomic_set(&sdp
->sd_log_thresh2
, 4*sdp
->sd_jdesc
->jd_blocks
/5);
715 /* Map the extents for this journal's blocks */
716 gfs2_map_journal_extents(sdp
, sdp
->sd_jdesc
);
718 trace_gfs2_log_blocks(sdp
, atomic_read(&sdp
->sd_log_blks_free
));
720 if (sdp
->sd_lockstruct
.ls_first
) {
722 for (x
= 0; x
< sdp
->sd_journals
; x
++) {
723 struct gfs2_jdesc
*jd
= gfs2_jdesc_find(sdp
, x
);
725 if (sdp
->sd_args
.ar_spectator
) {
726 error
= check_journal_clean(sdp
, jd
);
731 error
= gfs2_recover_journal(jd
, true);
733 fs_err(sdp
, "error recovering journal %u: %d\n",
739 gfs2_others_may_mount(sdp
);
740 } else if (!sdp
->sd_args
.ar_spectator
) {
741 error
= gfs2_recover_journal(sdp
->sd_jdesc
, true);
743 fs_err(sdp
, "error recovering my journal: %d\n", error
);
748 sdp
->sd_log_idle
= 1;
749 set_bit(SDF_JOURNAL_CHECKED
, &sdp
->sd_flags
);
750 gfs2_glock_dq_uninit(&ji_gh
);
752 INIT_WORK(&sdp
->sd_freeze_work
, gfs2_freeze_func
);
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
= d_inode(sdp
->sd_master_dir
);
780 error
= init_journal(sdp
, undo
);
781 complete_all(&sdp
->sd_journal_ready
);
785 /* Read in the master statfs inode */
786 sdp
->sd_statfs_inode
= gfs2_lookup_simple(master
, "statfs");
787 if (IS_ERR(sdp
->sd_statfs_inode
)) {
788 error
= PTR_ERR(sdp
->sd_statfs_inode
);
789 fs_err(sdp
, "can't read in statfs inode: %d\n", error
);
793 /* Read in the resource index inode */
794 sdp
->sd_rindex
= gfs2_lookup_simple(master
, "rindex");
795 if (IS_ERR(sdp
->sd_rindex
)) {
796 error
= PTR_ERR(sdp
->sd_rindex
);
797 fs_err(sdp
, "can't get resource index inode: %d\n", error
);
800 sdp
->sd_rindex_uptodate
= 0;
802 /* Read in the quota inode */
803 sdp
->sd_quota_inode
= gfs2_lookup_simple(master
, "quota");
804 if (IS_ERR(sdp
->sd_quota_inode
)) {
805 error
= PTR_ERR(sdp
->sd_quota_inode
);
806 fs_err(sdp
, "can't get quota file inode: %d\n", error
);
810 * i_rwsem on quota files is special. Since this inode is hidden system
811 * file, we are safe to define locking ourselves.
813 lockdep_set_class(&sdp
->sd_quota_inode
->i_rwsem
,
814 &gfs2_quota_imutex_key
);
816 error
= gfs2_rindex_update(sdp
);
823 iput(sdp
->sd_quota_inode
);
825 gfs2_clear_rgrpd(sdp
);
826 iput(sdp
->sd_rindex
);
828 iput(sdp
->sd_statfs_inode
);
830 init_journal(sdp
, UNDO
);
835 static int init_per_node(struct gfs2_sbd
*sdp
, int undo
)
837 struct inode
*pn
= NULL
;
840 struct gfs2_inode
*ip
;
841 struct inode
*master
= d_inode(sdp
->sd_master_dir
);
843 if (sdp
->sd_args
.ar_spectator
)
849 pn
= gfs2_lookup_simple(master
, "per_node");
852 fs_err(sdp
, "can't find per_node directory: %d\n", error
);
856 sprintf(buf
, "statfs_change%u", sdp
->sd_jdesc
->jd_jid
);
857 sdp
->sd_sc_inode
= gfs2_lookup_simple(pn
, buf
);
858 if (IS_ERR(sdp
->sd_sc_inode
)) {
859 error
= PTR_ERR(sdp
->sd_sc_inode
);
860 fs_err(sdp
, "can't find local \"sc\" file: %d\n", error
);
864 sprintf(buf
, "quota_change%u", sdp
->sd_jdesc
->jd_jid
);
865 sdp
->sd_qc_inode
= gfs2_lookup_simple(pn
, buf
);
866 if (IS_ERR(sdp
->sd_qc_inode
)) {
867 error
= PTR_ERR(sdp
->sd_qc_inode
);
868 fs_err(sdp
, "can't find local \"qc\" file: %d\n", error
);
875 ip
= GFS2_I(sdp
->sd_sc_inode
);
876 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
879 fs_err(sdp
, "can't lock local \"sc\" file: %d\n", error
);
883 ip
= GFS2_I(sdp
->sd_qc_inode
);
884 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, 0,
887 fs_err(sdp
, "can't lock local \"qc\" file: %d\n", error
);
894 gfs2_glock_dq_uninit(&sdp
->sd_qc_gh
);
896 gfs2_glock_dq_uninit(&sdp
->sd_sc_gh
);
898 iput(sdp
->sd_qc_inode
);
900 iput(sdp
->sd_sc_inode
);
906 static const match_table_t nolock_tokens
= {
907 { Opt_jid
, "jid=%d\n", },
911 static const struct lm_lockops nolock_ops
= {
912 .lm_proto_name
= "lock_nolock",
913 .lm_put_lock
= gfs2_glock_free
,
914 .lm_tokens
= &nolock_tokens
,
918 * gfs2_lm_mount - mount a locking protocol
919 * @sdp: the filesystem
920 * @args: mount arguments
921 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
926 static int gfs2_lm_mount(struct gfs2_sbd
*sdp
, int silent
)
928 const struct lm_lockops
*lm
;
929 struct lm_lockstruct
*ls
= &sdp
->sd_lockstruct
;
930 struct gfs2_args
*args
= &sdp
->sd_args
;
931 const char *proto
= sdp
->sd_proto_name
;
932 const char *table
= sdp
->sd_table_name
;
936 if (!strcmp("lock_nolock", proto
)) {
938 sdp
->sd_args
.ar_localflocks
= 1;
939 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
940 } else if (!strcmp("lock_dlm", proto
)) {
944 pr_info("can't find protocol %s\n", proto
);
948 fs_info(sdp
, "Trying to join cluster \"%s\", \"%s\"\n", proto
, table
);
953 for (options
= args
->ar_hostdata
; (o
= strsep(&options
, ":")); ) {
954 substring_t tmp
[MAX_OPT_ARGS
];
960 token
= match_token(o
, *lm
->lm_tokens
, tmp
);
963 ret
= match_int(&tmp
[0], &option
);
964 if (ret
|| option
< 0)
966 if (test_and_clear_bit(SDF_NOJOURNALID
, &sdp
->sd_flags
))
971 /* Obsolete, but left for backward compat purposes */
974 ret
= match_int(&tmp
[0], &option
);
975 if (ret
|| (option
!= 0 && option
!= 1))
977 ls
->ls_first
= option
;
982 fs_info(sdp
, "unknown hostdata (%s)\n", o
);
987 if (lm
->lm_mount
== NULL
) {
988 fs_info(sdp
, "Now mounting FS...\n");
989 complete_all(&sdp
->sd_locking_init
);
992 ret
= lm
->lm_mount(sdp
, table
);
994 fs_info(sdp
, "Joined cluster. Now mounting FS...\n");
995 complete_all(&sdp
->sd_locking_init
);
999 void gfs2_lm_unmount(struct gfs2_sbd
*sdp
)
1001 const struct lm_lockops
*lm
= sdp
->sd_lockstruct
.ls_ops
;
1002 if (likely(!test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)) &&
1004 lm
->lm_unmount(sdp
);
1007 static int wait_on_journal(struct gfs2_sbd
*sdp
)
1009 if (sdp
->sd_lockstruct
.ls_ops
->lm_mount
== NULL
)
1012 return wait_on_bit(&sdp
->sd_flags
, SDF_NOJOURNALID
, TASK_INTERRUPTIBLE
)
1016 void gfs2_online_uevent(struct gfs2_sbd
*sdp
)
1018 struct super_block
*sb
= sdp
->sd_vfs
;
1021 char *envp
[] = { ro
, spectator
, NULL
};
1022 sprintf(ro
, "RDONLY=%d", sb_rdonly(sb
));
1023 sprintf(spectator
, "SPECTATOR=%d", sdp
->sd_args
.ar_spectator
? 1 : 0);
1024 kobject_uevent_env(&sdp
->sd_kobj
, KOBJ_ONLINE
, envp
);
1028 * fill_super - Read in superblock
1029 * @sb: The VFS superblock
1030 * @data: Mount options
1031 * @silent: Don't complain if it's not a GFS2 filesystem
1036 static int fill_super(struct super_block
*sb
, struct gfs2_args
*args
, int silent
)
1038 struct gfs2_sbd
*sdp
;
1039 struct gfs2_holder mount_gh
;
1044 pr_warn("can't alloc struct gfs2_sbd\n");
1047 sdp
->sd_args
= *args
;
1049 if (sdp
->sd_args
.ar_spectator
) {
1050 sb
->s_flags
|= SB_RDONLY
;
1051 set_bit(SDF_RORECOVERY
, &sdp
->sd_flags
);
1053 if (sdp
->sd_args
.ar_posix_acl
)
1054 sb
->s_flags
|= SB_POSIXACL
;
1055 if (sdp
->sd_args
.ar_nobarrier
)
1056 set_bit(SDF_NOBARRIERS
, &sdp
->sd_flags
);
1058 sb
->s_flags
|= SB_NOSEC
;
1059 sb
->s_magic
= GFS2_MAGIC
;
1060 sb
->s_op
= &gfs2_super_ops
;
1061 sb
->s_d_op
= &gfs2_dops
;
1062 sb
->s_export_op
= &gfs2_export_ops
;
1063 sb
->s_xattr
= gfs2_xattr_handlers
;
1064 sb
->s_qcop
= &gfs2_quotactl_ops
;
1065 sb
->s_quota_types
= QTYPE_MASK_USR
| QTYPE_MASK_GRP
;
1066 sb_dqopt(sb
)->flags
|= DQUOT_QUOTA_SYS_FILE
;
1067 sb
->s_time_gran
= 1;
1068 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1070 /* Set up the buffer cache and fill in some fake block size values
1071 to allow us to read-in the on-disk superblock. */
1072 sdp
->sd_sb
.sb_bsize
= sb_min_blocksize(sb
, GFS2_BASIC_BLOCK
);
1073 sdp
->sd_sb
.sb_bsize_shift
= sb
->s_blocksize_bits
;
1074 sdp
->sd_fsb2bb_shift
= sdp
->sd_sb
.sb_bsize_shift
-
1075 GFS2_BASIC_BLOCK_SHIFT
;
1076 sdp
->sd_fsb2bb
= BIT(sdp
->sd_fsb2bb_shift
);
1078 sdp
->sd_tune
.gt_logd_secs
= sdp
->sd_args
.ar_commit
;
1079 sdp
->sd_tune
.gt_quota_quantum
= sdp
->sd_args
.ar_quota_quantum
;
1080 if (sdp
->sd_args
.ar_statfs_quantum
) {
1081 sdp
->sd_tune
.gt_statfs_slow
= 0;
1082 sdp
->sd_tune
.gt_statfs_quantum
= sdp
->sd_args
.ar_statfs_quantum
;
1084 sdp
->sd_tune
.gt_statfs_slow
= 1;
1085 sdp
->sd_tune
.gt_statfs_quantum
= 30;
1088 error
= init_names(sdp
, silent
);
1090 /* In this case, we haven't initialized sysfs, so we have to
1091 manually free the sdp. */
1092 free_percpu(sdp
->sd_lkstats
);
1094 sb
->s_fs_info
= NULL
;
1098 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s", sdp
->sd_table_name
);
1100 error
= gfs2_sys_fs_add(sdp
);
1102 * If we hit an error here, gfs2_sys_fs_add will have called function
1103 * kobject_put which causes the sysfs usage count to go to zero, which
1104 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1105 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1106 * kobject_put to free sdp.
1111 gfs2_create_debugfs_file(sdp
);
1113 error
= gfs2_lm_mount(sdp
, silent
);
1117 error
= init_locking(sdp
, &mount_gh
, DO
);
1121 error
= init_sb(sdp
, silent
);
1125 error
= wait_on_journal(sdp
);
1130 * If user space has failed to join the cluster or some similar
1131 * failure has occurred, then the journal id will contain a
1132 * negative (error) number. This will then be returned to the
1133 * caller (of the mount syscall). We do this even for spectator
1134 * mounts (which just write a jid of 0 to indicate "ok" even though
1135 * the jid is unused in the spectator case)
1137 if (sdp
->sd_lockstruct
.ls_jid
< 0) {
1138 error
= sdp
->sd_lockstruct
.ls_jid
;
1139 sdp
->sd_lockstruct
.ls_jid
= 0;
1143 if (sdp
->sd_args
.ar_spectator
)
1144 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.s",
1145 sdp
->sd_table_name
);
1147 snprintf(sdp
->sd_fsname
, sizeof(sdp
->sd_fsname
), "%s.%u",
1148 sdp
->sd_table_name
, sdp
->sd_lockstruct
.ls_jid
);
1150 error
= init_inodes(sdp
, DO
);
1154 error
= init_per_node(sdp
, DO
);
1158 error
= gfs2_statfs_init(sdp
);
1160 fs_err(sdp
, "can't initialize statfs subsystem: %d\n", error
);
1164 if (!sb_rdonly(sb
)) {
1165 error
= gfs2_make_fs_rw(sdp
);
1167 fs_err(sdp
, "can't make FS RW: %d\n", error
);
1172 gfs2_glock_dq_uninit(&mount_gh
);
1173 gfs2_online_uevent(sdp
);
1177 init_per_node(sdp
, UNDO
);
1179 init_inodes(sdp
, UNDO
);
1181 if (sdp
->sd_root_dir
)
1182 dput(sdp
->sd_root_dir
);
1183 if (sdp
->sd_master_dir
)
1184 dput(sdp
->sd_master_dir
);
1189 init_locking(sdp
, &mount_gh
, UNDO
);
1191 complete_all(&sdp
->sd_journal_ready
);
1192 gfs2_gl_hash_clear(sdp
);
1193 gfs2_lm_unmount(sdp
);
1195 gfs2_delete_debugfs_file(sdp
);
1196 free_percpu(sdp
->sd_lkstats
);
1197 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1198 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1199 gfs2_sys_fs_del(sdp
);
1200 sb
->s_fs_info
= NULL
;
1204 static int set_gfs2_super(struct super_block
*s
, void *data
)
1207 s
->s_dev
= s
->s_bdev
->bd_dev
;
1208 s
->s_bdi
= bdi_get(s
->s_bdev
->bd_bdi
);
1212 static int test_gfs2_super(struct super_block
*s
, void *ptr
)
1214 struct block_device
*bdev
= ptr
;
1215 return (bdev
== s
->s_bdev
);
1219 * gfs2_mount - Get the GFS2 superblock
1220 * @fs_type: The GFS2 filesystem type
1221 * @flags: Mount flags
1222 * @dev_name: The name of the device
1223 * @data: The mount arguments
1225 * Q. Why not use get_sb_bdev() ?
1226 * A. We need to select one of two root directories to mount, independent
1227 * of whether this is the initial, or subsequent, mount of this sb
1229 * Returns: 0 or -ve on error
1232 static struct dentry
*gfs2_mount(struct file_system_type
*fs_type
, int flags
,
1233 const char *dev_name
, void *data
)
1235 struct block_device
*bdev
;
1236 struct super_block
*s
;
1237 fmode_t mode
= FMODE_READ
| FMODE_EXCL
;
1239 struct gfs2_args args
;
1240 struct gfs2_sbd
*sdp
;
1242 if (!(flags
& SB_RDONLY
))
1243 mode
|= FMODE_WRITE
;
1245 bdev
= blkdev_get_by_path(dev_name
, mode
, fs_type
);
1247 return ERR_CAST(bdev
);
1250 * once the super is inserted into the list by sget, s_umount
1251 * will protect the lockfs code from trying to start a snapshot
1252 * while we are mounting
1254 mutex_lock(&bdev
->bd_fsfreeze_mutex
);
1255 if (bdev
->bd_fsfreeze_count
> 0) {
1256 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1260 s
= sget(fs_type
, test_gfs2_super
, set_gfs2_super
, flags
, bdev
);
1261 mutex_unlock(&bdev
->bd_fsfreeze_mutex
);
1268 * s_umount nests inside bd_mutex during
1269 * __invalidate_device(). blkdev_put() acquires
1270 * bd_mutex and can't be called under s_umount. Drop
1271 * s_umount temporarily. This is safe as we're
1272 * holding an active reference.
1274 up_write(&s
->s_umount
);
1275 blkdev_put(bdev
, mode
);
1276 down_write(&s
->s_umount
);
1278 /* s_mode must be set before deactivate_locked_super calls */
1282 memset(&args
, 0, sizeof(args
));
1283 args
.ar_quota
= GFS2_QUOTA_DEFAULT
;
1284 args
.ar_data
= GFS2_DATA_DEFAULT
;
1285 args
.ar_commit
= 30;
1286 args
.ar_statfs_quantum
= 30;
1287 args
.ar_quota_quantum
= 60;
1288 args
.ar_errors
= GFS2_ERRORS_DEFAULT
;
1290 error
= gfs2_mount_args(&args
, data
);
1292 pr_warn("can't parse mount arguments\n");
1298 if ((flags
^ s
->s_flags
) & SB_RDONLY
)
1301 snprintf(s
->s_id
, sizeof(s
->s_id
), "%pg", bdev
);
1302 sb_set_blocksize(s
, block_size(bdev
));
1303 error
= fill_super(s
, &args
, flags
& SB_SILENT
? 1 : 0);
1306 s
->s_flags
|= SB_ACTIVE
;
1312 return dget(sdp
->sd_master_dir
);
1314 return dget(sdp
->sd_root_dir
);
1317 deactivate_locked_super(s
);
1318 return ERR_PTR(error
);
1320 blkdev_put(bdev
, mode
);
1321 return ERR_PTR(error
);
1324 static int set_meta_super(struct super_block
*s
, void *ptr
)
1329 static struct dentry
*gfs2_mount_meta(struct file_system_type
*fs_type
,
1330 int flags
, const char *dev_name
, void *data
)
1332 struct super_block
*s
;
1333 struct gfs2_sbd
*sdp
;
1337 if (!dev_name
|| !*dev_name
)
1338 return ERR_PTR(-EINVAL
);
1340 error
= kern_path(dev_name
, LOOKUP_FOLLOW
, &path
);
1342 pr_warn("path_lookup on %s returned error %d\n",
1344 return ERR_PTR(error
);
1346 s
= sget(&gfs2_fs_type
, test_gfs2_super
, set_meta_super
, flags
,
1347 path
.dentry
->d_sb
->s_bdev
);
1350 pr_warn("gfs2 mount does not exist\n");
1353 if ((flags
^ s
->s_flags
) & SB_RDONLY
) {
1354 deactivate_locked_super(s
);
1355 return ERR_PTR(-EBUSY
);
1358 return dget(sdp
->sd_master_dir
);
1361 static void gfs2_kill_sb(struct super_block
*sb
)
1363 struct gfs2_sbd
*sdp
= sb
->s_fs_info
;
1366 kill_block_super(sb
);
1370 gfs2_log_flush(sdp
, NULL
, GFS2_LOG_HEAD_FLUSH_SYNC
| GFS2_LFC_KILL_SB
);
1371 dput(sdp
->sd_root_dir
);
1372 dput(sdp
->sd_master_dir
);
1373 sdp
->sd_root_dir
= NULL
;
1374 sdp
->sd_master_dir
= NULL
;
1375 shrink_dcache_sb(sb
);
1376 free_percpu(sdp
->sd_lkstats
);
1377 kill_block_super(sb
);
1380 struct file_system_type gfs2_fs_type
= {
1382 .fs_flags
= FS_REQUIRES_DEV
,
1383 .mount
= gfs2_mount
,
1384 .kill_sb
= gfs2_kill_sb
,
1385 .owner
= THIS_MODULE
,
1387 MODULE_ALIAS_FS("gfs2");
1389 struct file_system_type gfs2meta_fs_type
= {
1391 .fs_flags
= FS_REQUIRES_DEV
,
1392 .mount
= gfs2_mount_meta
,
1393 .owner
= THIS_MODULE
,
1395 MODULE_ALIAS_FS("gfs2meta");