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/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
36 #include "trace_gfs2.h"
42 * gfs2_tune_init - Fill a gfs2_tune structure with default values
47 static void gfs2_tune_init(struct gfs2_tune *gt)
49 spin_lock_init(>->gt_spin);
51 gt->gt_quota_simul_sync = 64;
52 gt->gt_quota_warn_period = 10;
53 gt->gt_quota_scale_num = 1;
54 gt->gt_quota_scale_den = 1;
55 gt->gt_new_files_jdata = 0;
56 gt->gt_max_readahead = 1 << 18;
57 gt->gt_complain_secs = 10;
60 static struct gfs2_sbd *init_sbd(struct super_block *sb)
64 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
70 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
71 gfs2_tune_init(&sdp->sd_tune);
73 init_waitqueue_head(&sdp->sd_glock_wait);
74 atomic_set(&sdp->sd_glock_disposal, 0);
75 spin_lock_init(&sdp->sd_statfs_spin);
77 spin_lock_init(&sdp->sd_rindex_spin);
78 mutex_init(&sdp->sd_rindex_mutex);
79 INIT_LIST_HEAD(&sdp->sd_rindex_list);
80 INIT_LIST_HEAD(&sdp->sd_rindex_mru_list);
82 INIT_LIST_HEAD(&sdp->sd_jindex_list);
83 spin_lock_init(&sdp->sd_jindex_spin);
84 mutex_init(&sdp->sd_jindex_mutex);
86 INIT_LIST_HEAD(&sdp->sd_quota_list);
87 mutex_init(&sdp->sd_quota_mutex);
88 init_waitqueue_head(&sdp->sd_quota_wait);
89 INIT_LIST_HEAD(&sdp->sd_trunc_list);
90 spin_lock_init(&sdp->sd_trunc_lock);
92 spin_lock_init(&sdp->sd_log_lock);
93 atomic_set(&sdp->sd_log_pinned, 0);
94 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
95 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
96 INIT_LIST_HEAD(&sdp->sd_log_le_rg);
97 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
98 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
100 init_waitqueue_head(&sdp->sd_log_waitq);
101 init_waitqueue_head(&sdp->sd_logd_waitq);
102 INIT_LIST_HEAD(&sdp->sd_ail1_list);
103 INIT_LIST_HEAD(&sdp->sd_ail2_list);
105 init_rwsem(&sdp->sd_log_flush_lock);
106 atomic_set(&sdp->sd_log_in_flight, 0);
107 init_waitqueue_head(&sdp->sd_log_flush_wait);
109 INIT_LIST_HEAD(&sdp->sd_revoke_list);
111 mutex_init(&sdp->sd_freeze_lock);
118 * gfs2_check_sb - Check superblock
119 * @sdp: the filesystem
120 * @sb: The superblock
121 * @silent: Don't print a message if the check fails
123 * Checks the version code of the FS is one that we understand how to
124 * read and that the sizes of the various on-disk structures have not
128 static int gfs2_check_sb(struct gfs2_sbd *sdp, struct gfs2_sb_host *sb, int silent)
130 if (sb->sb_magic != GFS2_MAGIC ||
131 sb->sb_type != GFS2_METATYPE_SB) {
133 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
137 /* If format numbers match exactly, we're done. */
139 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
140 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
143 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
148 static void end_bio_io_page(struct bio *bio, int error)
150 struct page *page = bio->bi_private;
153 SetPageUptodate(page);
155 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
159 static void gfs2_sb_in(struct gfs2_sb_host *sb, const void *buf)
161 const struct gfs2_sb *str = buf;
163 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
164 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
165 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
166 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
167 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
168 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
169 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
170 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
171 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
172 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
173 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
175 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
176 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
177 memcpy(sb->sb_uuid, str->sb_uuid, 16);
181 * gfs2_read_super - Read the gfs2 super block from disk
182 * @sdp: The GFS2 super block
183 * @sector: The location of the super block
184 * @error: The error code to return
186 * This uses the bio functions to read the super block from disk
187 * because we want to be 100% sure that we never read cached data.
188 * A super block is read twice only during each GFS2 mount and is
189 * never written to by the filesystem. The first time its read no
190 * locks are held, and the only details which are looked at are those
191 * relating to the locking protocol. Once locking is up and working,
192 * the sb is read again under the lock to establish the location of
193 * the master directory (contains pointers to journals etc) and the
196 * Returns: 0 on success or error
199 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector)
201 struct super_block *sb = sdp->sd_vfs;
206 page = alloc_page(GFP_NOFS);
210 ClearPageUptodate(page);
211 ClearPageDirty(page);
214 bio = bio_alloc(GFP_NOFS, 1);
215 bio->bi_sector = sector * (sb->s_blocksize >> 9);
216 bio->bi_bdev = sb->s_bdev;
217 bio_add_page(bio, page, PAGE_SIZE, 0);
219 bio->bi_end_io = end_bio_io_page;
220 bio->bi_private = page;
221 submit_bio(READ_SYNC | REQ_META, bio);
222 wait_on_page_locked(page);
224 if (!PageUptodate(page)) {
229 gfs2_sb_in(&sdp->sd_sb, p);
236 * gfs2_read_sb - Read super block
237 * @sdp: The GFS2 superblock
238 * @silent: Don't print message if mount fails
242 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
244 u32 hash_blocks, ind_blocks, leaf_blocks;
249 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
252 fs_err(sdp, "can't read superblock\n");
256 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
260 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
261 GFS2_BASIC_BLOCK_SHIFT;
262 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
263 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
264 sizeof(struct gfs2_dinode)) / sizeof(u64);
265 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
266 sizeof(struct gfs2_meta_header)) / sizeof(u64);
267 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
268 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
269 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
270 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
271 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
272 sizeof(struct gfs2_meta_header)) /
273 sizeof(struct gfs2_quota_change);
275 /* Compute maximum reservation required to add a entry to a directory */
277 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
281 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
282 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
283 ind_blocks += tmp_blocks;
286 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
288 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
290 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
291 sizeof(struct gfs2_dinode);
292 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
297 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
299 m = do_div(d, sdp->sd_inptrs);
301 if (d != sdp->sd_heightsize[x - 1] || m)
303 sdp->sd_heightsize[x] = space;
305 sdp->sd_max_height = x;
306 sdp->sd_heightsize[x] = ~0;
307 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
309 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
310 sizeof(struct gfs2_dinode);
311 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
316 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
318 m = do_div(d, sdp->sd_inptrs);
320 if (d != sdp->sd_jheightsize[x - 1] || m)
322 sdp->sd_jheightsize[x] = space;
324 sdp->sd_max_jheight = x;
325 sdp->sd_jheightsize[x] = ~0;
326 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
331 static int init_names(struct gfs2_sbd *sdp, int silent)
336 proto = sdp->sd_args.ar_lockproto;
337 table = sdp->sd_args.ar_locktable;
339 /* Try to autodetect */
341 if (!proto[0] || !table[0]) {
342 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift);
346 error = gfs2_check_sb(sdp, &sdp->sd_sb, silent);
351 proto = sdp->sd_sb.sb_lockproto;
353 table = sdp->sd_sb.sb_locktable;
357 table = sdp->sd_vfs->s_id;
359 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
360 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
362 table = sdp->sd_table_name;
363 while ((table = strchr(table, '/')))
370 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
378 error = gfs2_glock_nq_num(sdp,
379 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
380 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
383 fs_err(sdp, "can't acquire mount glock: %d\n", error);
387 error = gfs2_glock_nq_num(sdp,
388 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
390 LM_FLAG_NOEXP | GL_EXACT,
393 fs_err(sdp, "can't acquire live glock: %d\n", error);
397 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
398 CREATE, &sdp->sd_rename_gl);
400 fs_err(sdp, "can't create rename glock: %d\n", error);
404 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
405 CREATE, &sdp->sd_trans_gl);
407 fs_err(sdp, "can't create transaction glock: %d\n", error);
414 gfs2_glock_put(sdp->sd_trans_gl);
416 gfs2_glock_put(sdp->sd_rename_gl);
418 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
420 gfs2_glock_dq_uninit(mount_gh);
425 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
426 u64 no_addr, const char *name)
428 struct gfs2_sbd *sdp = sb->s_fs_info;
429 struct dentry *dentry;
432 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0);
434 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
435 return PTR_ERR(inode);
437 dentry = d_alloc_root(inode);
439 fs_err(sdp, "can't alloc %s dentry\n", name);
443 dentry->d_op = &gfs2_dops;
448 static int init_sb(struct gfs2_sbd *sdp, int silent)
450 struct super_block *sb = sdp->sd_vfs;
451 struct gfs2_holder sb_gh;
455 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
456 LM_ST_SHARED, 0, &sb_gh);
458 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
462 ret = gfs2_read_sb(sdp, silent);
464 fs_err(sdp, "can't read superblock: %d\n", ret);
468 /* Set up the buffer cache and SB for real */
469 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
471 fs_err(sdp, "FS block size (%u) is too small for device "
473 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
476 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
478 fs_err(sdp, "FS block size (%u) is too big for machine "
480 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
483 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
485 /* Get the root inode */
486 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
487 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
491 /* Get the master inode */
492 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
493 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
495 dput(sdp->sd_root_dir);
498 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
500 gfs2_glock_dq_uninit(&sb_gh);
505 * map_journal_extents - create a reusable "extent" mapping from all logical
506 * blocks to all physical blocks for the given journal. This will save
507 * us time when writing journal blocks. Most journals will have only one
508 * extent that maps all their logical blocks. That's because gfs2.mkfs
509 * arranges the journal blocks sequentially to maximize performance.
510 * So the extent would map the first block for the entire file length.
511 * However, gfs2_jadd can happen while file activity is happening, so
512 * those journals may not be sequential. Less likely is the case where
513 * the users created their own journals by mounting the metafs and
514 * laying it out. But it's still possible. These journals might have
517 * TODO: This should be done in bigger chunks rather than one block at a time,
518 * but since it's only done at mount time, I'm not worried about the
521 static int map_journal_extents(struct gfs2_sbd *sdp)
523 struct gfs2_jdesc *jd = sdp->sd_jdesc;
525 u64 db, prev_db; /* logical block, disk block, prev disk block */
526 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
527 struct gfs2_journal_extent *jext = NULL;
528 struct buffer_head bh;
533 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
536 bh.b_size = 1 << ip->i_inode.i_blkbits;
537 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
540 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
541 "%u db=%llu\n", rc, lb, (unsigned long long)db);
544 if (!prev_db || db != prev_db + 1) {
545 jext = kzalloc(sizeof(struct gfs2_journal_extent),
548 printk(KERN_INFO "GFS2 error: out of memory "
549 "mapping journal extents.\n");
556 list_add_tail(&jext->extent_list, &jd->extent_list);
565 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
567 char *message = "FIRSTMOUNT=Done";
568 char *envp[] = { message, NULL };
569 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
570 ls->ls_first_done = 1;
571 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
575 * gfs2_jindex_hold - Grab a lock on the jindex
576 * @sdp: The GFS2 superblock
577 * @ji_gh: the holder for the jindex glock
582 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
584 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
587 struct gfs2_jdesc *jd;
592 mutex_lock(&sdp->sd_jindex_mutex);
595 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
599 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
600 name.hash = gfs2_disk_hash(name.name, name.len);
602 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
603 if (error == -ENOENT) {
608 gfs2_glock_dq_uninit(ji_gh);
614 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
618 INIT_LIST_HEAD(&jd->extent_list);
619 INIT_WORK(&jd->jd_work, gfs2_recover_func);
620 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
621 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
625 error = PTR_ERR(jd->jd_inode);
630 spin_lock(&sdp->sd_jindex_spin);
631 jd->jd_jid = sdp->sd_journals++;
632 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
633 spin_unlock(&sdp->sd_jindex_spin);
636 mutex_unlock(&sdp->sd_jindex_mutex);
641 static int init_journal(struct gfs2_sbd *sdp, int undo)
643 struct inode *master = sdp->sd_master_dir->d_inode;
644 struct gfs2_holder ji_gh;
645 struct gfs2_inode *ip;
654 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
655 if (IS_ERR(sdp->sd_jindex)) {
656 fs_err(sdp, "can't lookup journal index: %d\n", error);
657 return PTR_ERR(sdp->sd_jindex);
659 ip = GFS2_I(sdp->sd_jindex);
661 /* Load in the journal index special file */
663 error = gfs2_jindex_hold(sdp, &ji_gh);
665 fs_err(sdp, "can't read journal index: %d\n", error);
670 if (!gfs2_jindex_size(sdp)) {
671 fs_err(sdp, "no journals!\n");
675 if (sdp->sd_args.ar_spectator) {
676 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
677 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
678 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
679 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
681 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
682 fs_err(sdp, "can't mount journal #%u\n",
683 sdp->sd_lockstruct.ls_jid);
684 fs_err(sdp, "there are only %u journals (0 - %u)\n",
685 gfs2_jindex_size(sdp),
686 gfs2_jindex_size(sdp) - 1);
689 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
691 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
693 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
694 &sdp->sd_journal_gh);
696 fs_err(sdp, "can't acquire journal glock: %d\n", error);
700 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
701 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
702 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
705 fs_err(sdp, "can't acquire journal inode glock: %d\n",
707 goto fail_journal_gh;
710 error = gfs2_jdesc_check(sdp->sd_jdesc);
712 fs_err(sdp, "my journal (%u) is bad: %d\n",
713 sdp->sd_jdesc->jd_jid, error);
716 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
717 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
718 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
720 /* Map the extents for this journal's blocks */
721 map_journal_extents(sdp);
723 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
725 if (sdp->sd_lockstruct.ls_first) {
727 for (x = 0; x < sdp->sd_journals; x++) {
728 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
731 fs_err(sdp, "error recovering journal %u: %d\n",
737 gfs2_others_may_mount(sdp);
738 } else if (!sdp->sd_args.ar_spectator) {
739 error = gfs2_recover_journal(sdp->sd_jdesc, true);
741 fs_err(sdp, "error recovering my journal: %d\n", error);
746 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
747 gfs2_glock_dq_uninit(&ji_gh);
753 if (!sdp->sd_args.ar_spectator)
754 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
756 if (!sdp->sd_args.ar_spectator)
757 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
759 gfs2_jindex_free(sdp);
761 gfs2_glock_dq_uninit(&ji_gh);
763 iput(sdp->sd_jindex);
768 static int init_inodes(struct gfs2_sbd *sdp, int undo)
771 struct gfs2_inode *ip;
772 struct inode *master = sdp->sd_master_dir->d_inode;
777 error = init_journal(sdp, undo);
781 /* Read in the master statfs inode */
782 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
783 if (IS_ERR(sdp->sd_statfs_inode)) {
784 error = PTR_ERR(sdp->sd_statfs_inode);
785 fs_err(sdp, "can't read in statfs inode: %d\n", error);
789 /* Read in the resource index inode */
790 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
791 if (IS_ERR(sdp->sd_rindex)) {
792 error = PTR_ERR(sdp->sd_rindex);
793 fs_err(sdp, "can't get resource index inode: %d\n", error);
796 ip = GFS2_I(sdp->sd_rindex);
797 sdp->sd_rindex_uptodate = 0;
799 /* Read in the quota inode */
800 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
801 if (IS_ERR(sdp->sd_quota_inode)) {
802 error = PTR_ERR(sdp->sd_quota_inode);
803 fs_err(sdp, "can't get quota file inode: %d\n", error);
809 iput(sdp->sd_quota_inode);
811 gfs2_clear_rgrpd(sdp);
812 iput(sdp->sd_rindex);
814 iput(sdp->sd_statfs_inode);
816 init_journal(sdp, UNDO);
821 static int init_per_node(struct gfs2_sbd *sdp, int undo)
823 struct inode *pn = NULL;
826 struct gfs2_inode *ip;
827 struct inode *master = sdp->sd_master_dir->d_inode;
829 if (sdp->sd_args.ar_spectator)
835 pn = gfs2_lookup_simple(master, "per_node");
838 fs_err(sdp, "can't find per_node directory: %d\n", error);
842 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
843 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
844 if (IS_ERR(sdp->sd_sc_inode)) {
845 error = PTR_ERR(sdp->sd_sc_inode);
846 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
850 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
851 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
852 if (IS_ERR(sdp->sd_qc_inode)) {
853 error = PTR_ERR(sdp->sd_qc_inode);
854 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
861 ip = GFS2_I(sdp->sd_sc_inode);
862 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
865 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
869 ip = GFS2_I(sdp->sd_qc_inode);
870 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
873 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
880 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
882 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
884 iput(sdp->sd_qc_inode);
886 iput(sdp->sd_sc_inode);
893 static int init_threads(struct gfs2_sbd *sdp, int undo)
895 struct task_struct *p;
901 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
904 fs_err(sdp, "can't start logd thread: %d\n", error);
907 sdp->sd_logd_process = p;
909 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
912 fs_err(sdp, "can't start quotad thread: %d\n", error);
915 sdp->sd_quotad_process = p;
921 kthread_stop(sdp->sd_quotad_process);
923 kthread_stop(sdp->sd_logd_process);
927 static const match_table_t nolock_tokens = {
928 { Opt_jid, "jid=%d\n", },
932 static void nolock_put_lock(struct kmem_cache *cachep, struct gfs2_glock *gl)
934 struct gfs2_sbd *sdp = gl->gl_sbd;
935 kmem_cache_free(cachep, gl);
936 if (atomic_dec_and_test(&sdp->sd_glock_disposal))
937 wake_up(&sdp->sd_glock_wait);
940 static const struct lm_lockops nolock_ops = {
941 .lm_proto_name = "lock_nolock",
942 .lm_put_lock = nolock_put_lock,
943 .lm_tokens = &nolock_tokens,
947 * gfs2_lm_mount - mount a locking protocol
948 * @sdp: the filesystem
949 * @args: mount arguments
950 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
955 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
957 const struct lm_lockops *lm;
958 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
959 struct gfs2_args *args = &sdp->sd_args;
960 const char *proto = sdp->sd_proto_name;
961 const char *table = sdp->sd_table_name;
966 if (!strcmp("lock_nolock", proto)) {
968 sdp->sd_args.ar_localflocks = 1;
969 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
970 } else if (!strcmp("lock_dlm", proto)) {
974 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
978 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
983 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
984 substring_t tmp[MAX_OPT_ARGS];
990 token = match_token(o, *lm->lm_tokens, tmp);
993 ret = match_int(&tmp[0], &option);
994 if (ret || option < 0)
996 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
1000 /* Obsolete, but left for backward compat purposes */
1003 ret = match_int(&tmp[0], &option);
1004 if (ret || (option != 0 && option != 1))
1005 goto hostdata_error;
1006 ls->ls_first = option;
1009 ret = match_int(&tmp[0], &option);
1010 if (ret || (option != 0 && option != 1))
1011 goto hostdata_error;
1012 ls->ls_nodir = option;
1017 fs_info(sdp, "unknown hostdata (%s)\n", o);
1022 if (sdp->sd_args.ar_spectator)
1023 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s", table);
1025 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u", table,
1026 sdp->sd_lockstruct.ls_jid);
1028 fsname = strchr(table, ':');
1031 if (lm->lm_mount == NULL) {
1032 fs_info(sdp, "Now mounting FS...\n");
1035 ret = lm->lm_mount(sdp, fsname);
1037 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1041 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1043 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1044 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1046 lm->lm_unmount(sdp);
1049 static int gfs2_journalid_wait(void *word)
1051 if (signal_pending(current))
1057 static int wait_on_journal(struct gfs2_sbd *sdp)
1059 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1062 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1065 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1067 struct super_block *sb = sdp->sd_vfs;
1070 char *envp[] = { ro, spectator, NULL };
1071 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1072 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1073 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1077 * fill_super - Read in superblock
1078 * @sb: The VFS superblock
1079 * @data: Mount options
1080 * @silent: Don't complain if it's not a GFS2 filesystem
1085 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1087 struct gfs2_sbd *sdp;
1088 struct gfs2_holder mount_gh;
1093 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1096 sdp->sd_args = *args;
1098 if (sdp->sd_args.ar_spectator) {
1099 sb->s_flags |= MS_RDONLY;
1100 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
1102 if (sdp->sd_args.ar_posix_acl)
1103 sb->s_flags |= MS_POSIXACL;
1104 if (sdp->sd_args.ar_nobarrier)
1105 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1107 sb->s_magic = GFS2_MAGIC;
1108 sb->s_op = &gfs2_super_ops;
1109 sb->s_export_op = &gfs2_export_ops;
1110 sb->s_xattr = gfs2_xattr_handlers;
1111 sb->s_qcop = &gfs2_quotactl_ops;
1112 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1113 sb->s_time_gran = 1;
1114 sb->s_maxbytes = MAX_LFS_FILESIZE;
1116 /* Set up the buffer cache and fill in some fake block size values
1117 to allow us to read-in the on-disk superblock. */
1118 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1119 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1120 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1121 GFS2_BASIC_BLOCK_SHIFT;
1122 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1124 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1125 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1126 if (sdp->sd_args.ar_statfs_quantum) {
1127 sdp->sd_tune.gt_statfs_slow = 0;
1128 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1131 sdp->sd_tune.gt_statfs_slow = 1;
1132 sdp->sd_tune.gt_statfs_quantum = 30;
1135 error = init_names(sdp, silent);
1139 gfs2_create_debugfs_file(sdp);
1141 error = gfs2_sys_fs_add(sdp);
1145 error = gfs2_lm_mount(sdp, silent);
1149 error = init_locking(sdp, &mount_gh, DO);
1153 error = init_sb(sdp, silent);
1157 error = wait_on_journal(sdp);
1162 * If user space has failed to join the cluster or some similar
1163 * failure has occurred, then the journal id will contain a
1164 * negative (error) number. This will then be returned to the
1165 * caller (of the mount syscall). We do this even for spectator
1166 * mounts (which just write a jid of 0 to indicate "ok" even though
1167 * the jid is unused in the spectator case)
1169 if (sdp->sd_lockstruct.ls_jid < 0) {
1170 error = sdp->sd_lockstruct.ls_jid;
1171 sdp->sd_lockstruct.ls_jid = 0;
1175 error = init_inodes(sdp, DO);
1179 error = init_per_node(sdp, DO);
1183 error = gfs2_statfs_init(sdp);
1185 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1189 error = init_threads(sdp, DO);
1193 if (!(sb->s_flags & MS_RDONLY)) {
1194 error = gfs2_make_fs_rw(sdp);
1196 fs_err(sdp, "can't make FS RW: %d\n", error);
1201 gfs2_glock_dq_uninit(&mount_gh);
1202 gfs2_online_uevent(sdp);
1206 init_threads(sdp, UNDO);
1208 init_per_node(sdp, UNDO);
1210 init_inodes(sdp, UNDO);
1212 if (sdp->sd_root_dir)
1213 dput(sdp->sd_root_dir);
1214 if (sdp->sd_master_dir)
1215 dput(sdp->sd_master_dir);
1220 init_locking(sdp, &mount_gh, UNDO);
1222 gfs2_gl_hash_clear(sdp);
1223 gfs2_lm_unmount(sdp);
1225 gfs2_sys_fs_del(sdp);
1227 gfs2_delete_debugfs_file(sdp);
1229 sb->s_fs_info = NULL;
1233 static int set_gfs2_super(struct super_block *s, void *data)
1236 s->s_dev = s->s_bdev->bd_dev;
1239 * We set the bdi here to the queue backing, file systems can
1240 * overwrite this in ->fill_super()
1242 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1246 static int test_gfs2_super(struct super_block *s, void *ptr)
1248 struct block_device *bdev = ptr;
1249 return (bdev == s->s_bdev);
1253 * gfs2_mount - Get the GFS2 superblock
1254 * @fs_type: The GFS2 filesystem type
1255 * @flags: Mount flags
1256 * @dev_name: The name of the device
1257 * @data: The mount arguments
1259 * Q. Why not use get_sb_bdev() ?
1260 * A. We need to select one of two root directories to mount, independent
1261 * of whether this is the initial, or subsequent, mount of this sb
1263 * Returns: 0 or -ve on error
1266 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1267 const char *dev_name, void *data)
1269 struct block_device *bdev;
1270 struct super_block *s;
1271 fmode_t mode = FMODE_READ;
1273 struct gfs2_args args;
1274 struct gfs2_sbd *sdp;
1276 if (!(flags & MS_RDONLY))
1277 mode |= FMODE_WRITE;
1279 bdev = open_bdev_exclusive(dev_name, mode, fs_type);
1281 return ERR_CAST(bdev);
1284 * once the super is inserted into the list by sget, s_umount
1285 * will protect the lockfs code from trying to start a snapshot
1286 * while we are mounting
1288 mutex_lock(&bdev->bd_fsfreeze_mutex);
1289 if (bdev->bd_fsfreeze_count > 0) {
1290 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1294 s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1295 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1301 close_bdev_exclusive(bdev, mode);
1303 memset(&args, 0, sizeof(args));
1304 args.ar_quota = GFS2_QUOTA_DEFAULT;
1305 args.ar_data = GFS2_DATA_DEFAULT;
1306 args.ar_commit = 30;
1307 args.ar_statfs_quantum = 30;
1308 args.ar_quota_quantum = 60;
1309 args.ar_errors = GFS2_ERRORS_DEFAULT;
1311 error = gfs2_mount_args(&args, data);
1313 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1319 if ((flags ^ s->s_flags) & MS_RDONLY)
1322 char b[BDEVNAME_SIZE];
1326 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1327 sb_set_blocksize(s, block_size(bdev));
1328 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1331 s->s_flags |= MS_ACTIVE;
1337 return dget(sdp->sd_master_dir);
1339 return dget(sdp->sd_root_dir);
1342 deactivate_locked_super(s);
1343 return ERR_PTR(error);
1345 close_bdev_exclusive(bdev, mode);
1346 return ERR_PTR(error);
1349 static int set_meta_super(struct super_block *s, void *ptr)
1354 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1355 int flags, const char *dev_name, void *data)
1357 struct super_block *s;
1358 struct gfs2_sbd *sdp;
1362 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1364 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1366 return ERR_PTR(error);
1368 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1369 path.dentry->d_inode->i_sb->s_bdev);
1372 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1375 if ((flags ^ s->s_flags) & MS_RDONLY) {
1376 deactivate_locked_super(s);
1377 return ERR_PTR(-EBUSY);
1380 return dget(sdp->sd_master_dir);
1383 static void gfs2_kill_sb(struct super_block *sb)
1385 struct gfs2_sbd *sdp = sb->s_fs_info;
1388 kill_block_super(sb);
1392 gfs2_meta_syncfs(sdp);
1393 dput(sdp->sd_root_dir);
1394 dput(sdp->sd_master_dir);
1395 sdp->sd_root_dir = NULL;
1396 sdp->sd_master_dir = NULL;
1397 shrink_dcache_sb(sb);
1398 kill_block_super(sb);
1399 gfs2_delete_debugfs_file(sdp);
1403 struct file_system_type gfs2_fs_type = {
1405 .fs_flags = FS_REQUIRES_DEV,
1406 .mount = gfs2_mount,
1407 .kill_sb = gfs2_kill_sb,
1408 .owner = THIS_MODULE,
1411 struct file_system_type gfs2meta_fs_type = {
1413 .fs_flags = FS_REQUIRES_DEV,
1414 .mount = gfs2_mount_meta,
1415 .owner = THIS_MODULE,