2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/blkdev.h>
16 #include <linux/kthread.h>
17 #include <linux/export.h>
18 #include <linux/namei.h>
19 #include <linux/mount.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/quotaops.h>
37 #include "trace_gfs2.h"
43 * gfs2_tune_init - Fill a gfs2_tune structure with default values
48 static void gfs2_tune_init(struct gfs2_tune *gt)
50 spin_lock_init(>->gt_spin);
52 gt->gt_quota_simul_sync = 64;
53 gt->gt_quota_warn_period = 10;
54 gt->gt_quota_scale_num = 1;
55 gt->gt_quota_scale_den = 1;
56 gt->gt_new_files_jdata = 0;
57 gt->gt_max_readahead = 1 << 18;
58 gt->gt_complain_secs = 10;
61 static struct gfs2_sbd *init_sbd(struct super_block *sb)
65 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
71 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
72 if (!sdp->sd_lkstats) {
77 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
78 gfs2_tune_init(&sdp->sd_tune);
80 init_waitqueue_head(&sdp->sd_glock_wait);
81 atomic_set(&sdp->sd_glock_disposal, 0);
82 init_completion(&sdp->sd_locking_init);
83 spin_lock_init(&sdp->sd_statfs_spin);
85 spin_lock_init(&sdp->sd_rindex_spin);
86 sdp->sd_rindex_tree.rb_node = NULL;
88 INIT_LIST_HEAD(&sdp->sd_jindex_list);
89 spin_lock_init(&sdp->sd_jindex_spin);
90 mutex_init(&sdp->sd_jindex_mutex);
92 INIT_LIST_HEAD(&sdp->sd_quota_list);
93 mutex_init(&sdp->sd_quota_mutex);
94 init_waitqueue_head(&sdp->sd_quota_wait);
95 INIT_LIST_HEAD(&sdp->sd_trunc_list);
96 spin_lock_init(&sdp->sd_trunc_lock);
98 spin_lock_init(&sdp->sd_log_lock);
99 atomic_set(&sdp->sd_log_pinned, 0);
100 INIT_LIST_HEAD(&sdp->sd_log_le_buf);
101 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
102 INIT_LIST_HEAD(&sdp->sd_log_le_rg);
103 INIT_LIST_HEAD(&sdp->sd_log_le_databuf);
104 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
106 init_waitqueue_head(&sdp->sd_log_waitq);
107 init_waitqueue_head(&sdp->sd_logd_waitq);
108 spin_lock_init(&sdp->sd_ail_lock);
109 INIT_LIST_HEAD(&sdp->sd_ail1_list);
110 INIT_LIST_HEAD(&sdp->sd_ail2_list);
112 init_rwsem(&sdp->sd_log_flush_lock);
113 atomic_set(&sdp->sd_log_in_flight, 0);
114 init_waitqueue_head(&sdp->sd_log_flush_wait);
116 INIT_LIST_HEAD(&sdp->sd_revoke_list);
118 mutex_init(&sdp->sd_freeze_lock);
125 * gfs2_check_sb - Check superblock
126 * @sdp: the filesystem
127 * @sb: The superblock
128 * @silent: Don't print a message if the check fails
130 * Checks the version code of the FS is one that we understand how to
131 * read and that the sizes of the various on-disk structures have not
135 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
137 struct gfs2_sb_host *sb = &sdp->sd_sb;
139 if (sb->sb_magic != GFS2_MAGIC ||
140 sb->sb_type != GFS2_METATYPE_SB) {
142 printk(KERN_WARNING "GFS2: not a GFS2 filesystem\n");
146 /* If format numbers match exactly, we're done. */
148 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
149 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
152 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
157 static void end_bio_io_page(struct bio *bio, int error)
159 struct page *page = bio->bi_private;
162 SetPageUptodate(page);
164 printk(KERN_WARNING "gfs2: error %d reading superblock\n", error);
168 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
170 struct gfs2_sb_host *sb = &sdp->sd_sb;
171 struct super_block *s = sdp->sd_vfs;
172 const struct gfs2_sb *str = buf;
174 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
175 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
176 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
177 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
178 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
179 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
180 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
181 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
182 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
183 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
184 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
186 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
187 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
188 memcpy(s->s_uuid, str->sb_uuid, 16);
192 * gfs2_read_super - Read the gfs2 super block from disk
193 * @sdp: The GFS2 super block
194 * @sector: The location of the super block
195 * @error: The error code to return
197 * This uses the bio functions to read the super block from disk
198 * because we want to be 100% sure that we never read cached data.
199 * A super block is read twice only during each GFS2 mount and is
200 * never written to by the filesystem. The first time its read no
201 * locks are held, and the only details which are looked at are those
202 * relating to the locking protocol. Once locking is up and working,
203 * the sb is read again under the lock to establish the location of
204 * the master directory (contains pointers to journals etc) and the
207 * Returns: 0 on success or error
210 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
212 struct super_block *sb = sdp->sd_vfs;
217 page = alloc_page(GFP_NOFS);
221 ClearPageUptodate(page);
222 ClearPageDirty(page);
225 bio = bio_alloc(GFP_NOFS, 1);
226 bio->bi_sector = sector * (sb->s_blocksize >> 9);
227 bio->bi_bdev = sb->s_bdev;
228 bio_add_page(bio, page, PAGE_SIZE, 0);
230 bio->bi_end_io = end_bio_io_page;
231 bio->bi_private = page;
232 submit_bio(READ_SYNC | REQ_META, bio);
233 wait_on_page_locked(page);
235 if (!PageUptodate(page)) {
243 return gfs2_check_sb(sdp, silent);
247 * gfs2_read_sb - Read super block
248 * @sdp: The GFS2 superblock
249 * @silent: Don't print message if mount fails
253 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
255 u32 hash_blocks, ind_blocks, leaf_blocks;
260 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
263 fs_err(sdp, "can't read superblock\n");
267 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
268 GFS2_BASIC_BLOCK_SHIFT;
269 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
270 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
271 sizeof(struct gfs2_dinode)) / sizeof(u64);
272 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
273 sizeof(struct gfs2_meta_header)) / sizeof(u64);
274 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
275 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
276 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
277 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
278 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
279 sizeof(struct gfs2_meta_header)) /
280 sizeof(struct gfs2_quota_change);
282 /* Compute maximum reservation required to add a entry to a directory */
284 hash_blocks = DIV_ROUND_UP(sizeof(u64) * (1 << GFS2_DIR_MAX_DEPTH),
288 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
289 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
290 ind_blocks += tmp_blocks;
293 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
295 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
297 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
298 sizeof(struct gfs2_dinode);
299 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
304 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
306 m = do_div(d, sdp->sd_inptrs);
308 if (d != sdp->sd_heightsize[x - 1] || m)
310 sdp->sd_heightsize[x] = space;
312 sdp->sd_max_height = x;
313 sdp->sd_heightsize[x] = ~0;
314 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
316 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
317 sizeof(struct gfs2_dinode);
318 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
323 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
325 m = do_div(d, sdp->sd_inptrs);
327 if (d != sdp->sd_jheightsize[x - 1] || m)
329 sdp->sd_jheightsize[x] = space;
331 sdp->sd_max_jheight = x;
332 sdp->sd_jheightsize[x] = ~0;
333 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
338 static int init_names(struct gfs2_sbd *sdp, int silent)
343 proto = sdp->sd_args.ar_lockproto;
344 table = sdp->sd_args.ar_locktable;
346 /* Try to autodetect */
348 if (!proto[0] || !table[0]) {
349 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
354 proto = sdp->sd_sb.sb_lockproto;
356 table = sdp->sd_sb.sb_locktable;
360 table = sdp->sd_vfs->s_id;
362 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
363 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
365 table = sdp->sd_table_name;
366 while ((table = strchr(table, '/')))
372 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
380 error = gfs2_glock_nq_num(sdp,
381 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
382 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
385 fs_err(sdp, "can't acquire mount glock: %d\n", error);
389 error = gfs2_glock_nq_num(sdp,
390 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
392 LM_FLAG_NOEXP | GL_EXACT,
395 fs_err(sdp, "can't acquire live glock: %d\n", error);
399 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
400 CREATE, &sdp->sd_rename_gl);
402 fs_err(sdp, "can't create rename glock: %d\n", error);
406 error = gfs2_glock_get(sdp, GFS2_TRANS_LOCK, &gfs2_trans_glops,
407 CREATE, &sdp->sd_trans_gl);
409 fs_err(sdp, "can't create transaction glock: %d\n", error);
416 gfs2_glock_put(sdp->sd_trans_gl);
418 gfs2_glock_put(sdp->sd_rename_gl);
420 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
422 gfs2_glock_dq_uninit(mount_gh);
427 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
428 u64 no_addr, const char *name)
430 struct gfs2_sbd *sdp = sb->s_fs_info;
431 struct dentry *dentry;
434 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0, 0);
436 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
437 return PTR_ERR(inode);
439 dentry = d_make_root(inode);
441 fs_err(sdp, "can't alloc %s dentry\n", name);
448 static int init_sb(struct gfs2_sbd *sdp, int silent)
450 struct super_block *sb = sdp->sd_vfs;
451 struct gfs2_holder sb_gh;
455 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
456 LM_ST_SHARED, 0, &sb_gh);
458 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
462 ret = gfs2_read_sb(sdp, silent);
464 fs_err(sdp, "can't read superblock: %d\n", ret);
468 /* Set up the buffer cache and SB for real */
469 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
471 fs_err(sdp, "FS block size (%u) is too small for device "
473 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
476 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
478 fs_err(sdp, "FS block size (%u) is too big for machine "
480 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
483 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
485 /* Get the root inode */
486 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
487 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
491 /* Get the master inode */
492 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
493 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
495 dput(sdp->sd_root_dir);
498 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
500 gfs2_glock_dq_uninit(&sb_gh);
505 * map_journal_extents - create a reusable "extent" mapping from all logical
506 * blocks to all physical blocks for the given journal. This will save
507 * us time when writing journal blocks. Most journals will have only one
508 * extent that maps all their logical blocks. That's because gfs2.mkfs
509 * arranges the journal blocks sequentially to maximize performance.
510 * So the extent would map the first block for the entire file length.
511 * However, gfs2_jadd can happen while file activity is happening, so
512 * those journals may not be sequential. Less likely is the case where
513 * the users created their own journals by mounting the metafs and
514 * laying it out. But it's still possible. These journals might have
517 * TODO: This should be done in bigger chunks rather than one block at a time,
518 * but since it's only done at mount time, I'm not worried about the
521 static int map_journal_extents(struct gfs2_sbd *sdp)
523 struct gfs2_jdesc *jd = sdp->sd_jdesc;
525 u64 db, prev_db; /* logical block, disk block, prev disk block */
526 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
527 struct gfs2_journal_extent *jext = NULL;
528 struct buffer_head bh;
533 for (lb = 0; lb < i_size_read(jd->jd_inode) >> sdp->sd_sb.sb_bsize_shift; lb++) {
536 bh.b_size = 1 << ip->i_inode.i_blkbits;
537 rc = gfs2_block_map(jd->jd_inode, lb, &bh, 0);
540 printk(KERN_INFO "GFS2 journal mapping error %d: lb="
541 "%u db=%llu\n", rc, lb, (unsigned long long)db);
544 if (!prev_db || db != prev_db + 1) {
545 jext = kzalloc(sizeof(struct gfs2_journal_extent),
548 printk(KERN_INFO "GFS2 error: out of memory "
549 "mapping journal extents.\n");
556 list_add_tail(&jext->extent_list, &jd->extent_list);
565 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
567 char *message = "FIRSTMOUNT=Done";
568 char *envp[] = { message, NULL };
570 fs_info(sdp, "first mount done, others may mount\n");
572 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
573 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
575 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
579 * gfs2_jindex_hold - Grab a lock on the jindex
580 * @sdp: The GFS2 superblock
581 * @ji_gh: the holder for the jindex glock
586 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
588 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
591 struct gfs2_jdesc *jd;
596 mutex_lock(&sdp->sd_jindex_mutex);
599 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
603 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
604 name.hash = gfs2_disk_hash(name.name, name.len);
606 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
607 if (error == -ENOENT) {
612 gfs2_glock_dq_uninit(ji_gh);
618 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
622 INIT_LIST_HEAD(&jd->extent_list);
623 INIT_WORK(&jd->jd_work, gfs2_recover_func);
624 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
625 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
629 error = PTR_ERR(jd->jd_inode);
634 spin_lock(&sdp->sd_jindex_spin);
635 jd->jd_jid = sdp->sd_journals++;
636 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
637 spin_unlock(&sdp->sd_jindex_spin);
640 mutex_unlock(&sdp->sd_jindex_mutex);
645 static int init_journal(struct gfs2_sbd *sdp, int undo)
647 struct inode *master = sdp->sd_master_dir->d_inode;
648 struct gfs2_holder ji_gh;
649 struct gfs2_inode *ip;
658 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
659 if (IS_ERR(sdp->sd_jindex)) {
660 fs_err(sdp, "can't lookup journal index: %d\n", error);
661 return PTR_ERR(sdp->sd_jindex);
664 /* Load in the journal index special file */
666 error = gfs2_jindex_hold(sdp, &ji_gh);
668 fs_err(sdp, "can't read journal index: %d\n", error);
673 if (!gfs2_jindex_size(sdp)) {
674 fs_err(sdp, "no journals!\n");
678 if (sdp->sd_args.ar_spectator) {
679 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
680 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
681 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
682 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
684 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
685 fs_err(sdp, "can't mount journal #%u\n",
686 sdp->sd_lockstruct.ls_jid);
687 fs_err(sdp, "there are only %u journals (0 - %u)\n",
688 gfs2_jindex_size(sdp),
689 gfs2_jindex_size(sdp) - 1);
692 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
694 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
696 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
697 &sdp->sd_journal_gh);
699 fs_err(sdp, "can't acquire journal glock: %d\n", error);
703 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
704 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
705 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
708 fs_err(sdp, "can't acquire journal inode glock: %d\n",
710 goto fail_journal_gh;
713 error = gfs2_jdesc_check(sdp->sd_jdesc);
715 fs_err(sdp, "my journal (%u) is bad: %d\n",
716 sdp->sd_jdesc->jd_jid, error);
719 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
720 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
721 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
723 /* Map the extents for this journal's blocks */
724 map_journal_extents(sdp);
726 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
728 if (sdp->sd_lockstruct.ls_first) {
730 for (x = 0; x < sdp->sd_journals; x++) {
731 error = gfs2_recover_journal(gfs2_jdesc_find(sdp, x),
734 fs_err(sdp, "error recovering journal %u: %d\n",
740 gfs2_others_may_mount(sdp);
741 } else if (!sdp->sd_args.ar_spectator) {
742 error = gfs2_recover_journal(sdp->sd_jdesc, true);
744 fs_err(sdp, "error recovering my journal: %d\n", error);
749 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
750 gfs2_glock_dq_uninit(&ji_gh);
756 if (!sdp->sd_args.ar_spectator)
757 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
759 if (!sdp->sd_args.ar_spectator)
760 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
762 gfs2_jindex_free(sdp);
764 gfs2_glock_dq_uninit(&ji_gh);
766 iput(sdp->sd_jindex);
771 static int init_inodes(struct gfs2_sbd *sdp, int undo)
774 struct inode *master = sdp->sd_master_dir->d_inode;
779 error = init_journal(sdp, undo);
783 /* Read in the master statfs inode */
784 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
785 if (IS_ERR(sdp->sd_statfs_inode)) {
786 error = PTR_ERR(sdp->sd_statfs_inode);
787 fs_err(sdp, "can't read in statfs inode: %d\n", error);
791 /* Read in the resource index inode */
792 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
793 if (IS_ERR(sdp->sd_rindex)) {
794 error = PTR_ERR(sdp->sd_rindex);
795 fs_err(sdp, "can't get resource index inode: %d\n", error);
798 sdp->sd_rindex_uptodate = 0;
800 /* Read in the quota inode */
801 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
802 if (IS_ERR(sdp->sd_quota_inode)) {
803 error = PTR_ERR(sdp->sd_quota_inode);
804 fs_err(sdp, "can't get quota file inode: %d\n", error);
808 error = gfs2_rindex_update(sdp);
815 iput(sdp->sd_quota_inode);
817 gfs2_clear_rgrpd(sdp);
818 iput(sdp->sd_rindex);
820 iput(sdp->sd_statfs_inode);
822 init_journal(sdp, UNDO);
827 static int init_per_node(struct gfs2_sbd *sdp, int undo)
829 struct inode *pn = NULL;
832 struct gfs2_inode *ip;
833 struct inode *master = sdp->sd_master_dir->d_inode;
835 if (sdp->sd_args.ar_spectator)
841 pn = gfs2_lookup_simple(master, "per_node");
844 fs_err(sdp, "can't find per_node directory: %d\n", error);
848 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
849 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
850 if (IS_ERR(sdp->sd_sc_inode)) {
851 error = PTR_ERR(sdp->sd_sc_inode);
852 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
856 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
857 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
858 if (IS_ERR(sdp->sd_qc_inode)) {
859 error = PTR_ERR(sdp->sd_qc_inode);
860 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
867 ip = GFS2_I(sdp->sd_sc_inode);
868 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
871 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
875 ip = GFS2_I(sdp->sd_qc_inode);
876 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
879 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
886 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
888 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
890 iput(sdp->sd_qc_inode);
892 iput(sdp->sd_sc_inode);
899 static int init_threads(struct gfs2_sbd *sdp, int undo)
901 struct task_struct *p;
907 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
910 fs_err(sdp, "can't start logd thread: %d\n", error);
913 sdp->sd_logd_process = p;
915 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
918 fs_err(sdp, "can't start quotad thread: %d\n", error);
921 sdp->sd_quotad_process = p;
927 kthread_stop(sdp->sd_quotad_process);
929 kthread_stop(sdp->sd_logd_process);
933 static const match_table_t nolock_tokens = {
934 { Opt_jid, "jid=%d\n", },
938 static const struct lm_lockops nolock_ops = {
939 .lm_proto_name = "lock_nolock",
940 .lm_put_lock = gfs2_glock_free,
941 .lm_tokens = &nolock_tokens,
945 * gfs2_lm_mount - mount a locking protocol
946 * @sdp: the filesystem
947 * @args: mount arguments
948 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
953 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
955 const struct lm_lockops *lm;
956 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
957 struct gfs2_args *args = &sdp->sd_args;
958 const char *proto = sdp->sd_proto_name;
959 const char *table = sdp->sd_table_name;
963 if (!strcmp("lock_nolock", proto)) {
965 sdp->sd_args.ar_localflocks = 1;
966 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
967 } else if (!strcmp("lock_dlm", proto)) {
971 printk(KERN_INFO "GFS2: can't find protocol %s\n", proto);
975 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
980 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
981 substring_t tmp[MAX_OPT_ARGS];
987 token = match_token(o, *lm->lm_tokens, tmp);
990 ret = match_int(&tmp[0], &option);
991 if (ret || option < 0)
993 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
997 /* Obsolete, but left for backward compat purposes */
1000 ret = match_int(&tmp[0], &option);
1001 if (ret || (option != 0 && option != 1))
1002 goto hostdata_error;
1003 ls->ls_first = option;
1006 ret = match_int(&tmp[0], &option);
1007 if (ret || (option != 0 && option != 1))
1008 goto hostdata_error;
1009 ls->ls_nodir = option;
1014 fs_info(sdp, "unknown hostdata (%s)\n", o);
1019 if (lm->lm_mount == NULL) {
1020 fs_info(sdp, "Now mounting FS...\n");
1021 complete_all(&sdp->sd_locking_init);
1024 ret = lm->lm_mount(sdp, table);
1026 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1027 complete_all(&sdp->sd_locking_init);
1031 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1033 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1034 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1036 lm->lm_unmount(sdp);
1039 static int gfs2_journalid_wait(void *word)
1041 if (signal_pending(current))
1047 static int wait_on_journal(struct gfs2_sbd *sdp)
1049 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1052 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, gfs2_journalid_wait, TASK_INTERRUPTIBLE);
1055 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1057 struct super_block *sb = sdp->sd_vfs;
1060 char *envp[] = { ro, spectator, NULL };
1061 sprintf(ro, "RDONLY=%d", (sb->s_flags & MS_RDONLY) ? 1 : 0);
1062 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1063 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1067 * fill_super - Read in superblock
1068 * @sb: The VFS superblock
1069 * @data: Mount options
1070 * @silent: Don't complain if it's not a GFS2 filesystem
1075 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1077 struct gfs2_sbd *sdp;
1078 struct gfs2_holder mount_gh;
1083 printk(KERN_WARNING "GFS2: can't alloc struct gfs2_sbd\n");
1086 sdp->sd_args = *args;
1088 if (sdp->sd_args.ar_spectator) {
1089 sb->s_flags |= MS_RDONLY;
1090 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1092 if (sdp->sd_args.ar_posix_acl)
1093 sb->s_flags |= MS_POSIXACL;
1094 if (sdp->sd_args.ar_nobarrier)
1095 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1097 sb->s_flags |= MS_NOSEC;
1098 sb->s_magic = GFS2_MAGIC;
1099 sb->s_op = &gfs2_super_ops;
1100 sb->s_d_op = &gfs2_dops;
1101 sb->s_export_op = &gfs2_export_ops;
1102 sb->s_xattr = gfs2_xattr_handlers;
1103 sb->s_qcop = &gfs2_quotactl_ops;
1104 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1105 sb->s_time_gran = 1;
1106 sb->s_maxbytes = MAX_LFS_FILESIZE;
1108 /* Set up the buffer cache and fill in some fake block size values
1109 to allow us to read-in the on-disk superblock. */
1110 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1111 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1112 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1113 GFS2_BASIC_BLOCK_SHIFT;
1114 sdp->sd_fsb2bb = 1 << sdp->sd_fsb2bb_shift;
1116 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1117 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1118 if (sdp->sd_args.ar_statfs_quantum) {
1119 sdp->sd_tune.gt_statfs_slow = 0;
1120 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1122 sdp->sd_tune.gt_statfs_slow = 1;
1123 sdp->sd_tune.gt_statfs_quantum = 30;
1126 error = init_names(sdp, silent);
1130 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s", sdp->sd_table_name);
1132 gfs2_create_debugfs_file(sdp);
1134 error = gfs2_sys_fs_add(sdp);
1138 error = gfs2_lm_mount(sdp, silent);
1142 error = init_locking(sdp, &mount_gh, DO);
1146 error = init_sb(sdp, silent);
1150 error = wait_on_journal(sdp);
1155 * If user space has failed to join the cluster or some similar
1156 * failure has occurred, then the journal id will contain a
1157 * negative (error) number. This will then be returned to the
1158 * caller (of the mount syscall). We do this even for spectator
1159 * mounts (which just write a jid of 0 to indicate "ok" even though
1160 * the jid is unused in the spectator case)
1162 if (sdp->sd_lockstruct.ls_jid < 0) {
1163 error = sdp->sd_lockstruct.ls_jid;
1164 sdp->sd_lockstruct.ls_jid = 0;
1168 if (sdp->sd_args.ar_spectator)
1169 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.s",
1170 sdp->sd_table_name);
1172 snprintf(sdp->sd_fsname, GFS2_FSNAME_LEN, "%s.%u",
1173 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
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);
1228 free_percpu(sdp->sd_lkstats);
1230 sb->s_fs_info = NULL;
1234 static int set_gfs2_super(struct super_block *s, void *data)
1237 s->s_dev = s->s_bdev->bd_dev;
1240 * We set the bdi here to the queue backing, file systems can
1241 * overwrite this in ->fill_super()
1243 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
1247 static int test_gfs2_super(struct super_block *s, void *ptr)
1249 struct block_device *bdev = ptr;
1250 return (bdev == s->s_bdev);
1254 * gfs2_mount - Get the GFS2 superblock
1255 * @fs_type: The GFS2 filesystem type
1256 * @flags: Mount flags
1257 * @dev_name: The name of the device
1258 * @data: The mount arguments
1260 * Q. Why not use get_sb_bdev() ?
1261 * A. We need to select one of two root directories to mount, independent
1262 * of whether this is the initial, or subsequent, mount of this sb
1264 * Returns: 0 or -ve on error
1267 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1268 const char *dev_name, void *data)
1270 struct block_device *bdev;
1271 struct super_block *s;
1272 fmode_t mode = FMODE_READ | FMODE_EXCL;
1274 struct gfs2_args args;
1275 struct gfs2_sbd *sdp;
1277 if (!(flags & MS_RDONLY))
1278 mode |= FMODE_WRITE;
1280 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1282 return ERR_CAST(bdev);
1285 * once the super is inserted into the list by sget, s_umount
1286 * will protect the lockfs code from trying to start a snapshot
1287 * while we are mounting
1289 mutex_lock(&bdev->bd_fsfreeze_mutex);
1290 if (bdev->bd_fsfreeze_count > 0) {
1291 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1295 s = sget(fs_type, test_gfs2_super, set_gfs2_super, bdev);
1296 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1302 blkdev_put(bdev, mode);
1304 memset(&args, 0, sizeof(args));
1305 args.ar_quota = GFS2_QUOTA_DEFAULT;
1306 args.ar_data = GFS2_DATA_DEFAULT;
1307 args.ar_commit = 30;
1308 args.ar_statfs_quantum = 30;
1309 args.ar_quota_quantum = 60;
1310 args.ar_errors = GFS2_ERRORS_DEFAULT;
1312 error = gfs2_mount_args(&args, data);
1314 printk(KERN_WARNING "GFS2: can't parse mount arguments\n");
1320 if ((flags ^ s->s_flags) & MS_RDONLY)
1323 char b[BDEVNAME_SIZE];
1327 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1328 sb_set_blocksize(s, block_size(bdev));
1329 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1332 s->s_flags |= MS_ACTIVE;
1338 return dget(sdp->sd_master_dir);
1340 return dget(sdp->sd_root_dir);
1343 deactivate_locked_super(s);
1344 return ERR_PTR(error);
1346 blkdev_put(bdev, mode);
1347 return ERR_PTR(error);
1350 static int set_meta_super(struct super_block *s, void *ptr)
1355 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1356 int flags, const char *dev_name, void *data)
1358 struct super_block *s;
1359 struct gfs2_sbd *sdp;
1363 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1365 printk(KERN_WARNING "GFS2: path_lookup on %s returned error %d\n",
1367 return ERR_PTR(error);
1369 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super,
1370 path.dentry->d_inode->i_sb->s_bdev);
1373 printk(KERN_WARNING "GFS2: gfs2 mount does not exist\n");
1376 if ((flags ^ s->s_flags) & MS_RDONLY) {
1377 deactivate_locked_super(s);
1378 return ERR_PTR(-EBUSY);
1381 return dget(sdp->sd_master_dir);
1384 static void gfs2_kill_sb(struct super_block *sb)
1386 struct gfs2_sbd *sdp = sb->s_fs_info;
1389 kill_block_super(sb);
1393 gfs2_meta_syncfs(sdp);
1394 dput(sdp->sd_root_dir);
1395 dput(sdp->sd_master_dir);
1396 sdp->sd_root_dir = NULL;
1397 sdp->sd_master_dir = NULL;
1398 shrink_dcache_sb(sb);
1399 kill_block_super(sb);
1400 gfs2_delete_debugfs_file(sdp);
1401 free_percpu(sdp->sd_lkstats);
1405 struct file_system_type gfs2_fs_type = {
1407 .fs_flags = FS_REQUIRES_DEV,
1408 .mount = gfs2_mount,
1409 .kill_sb = gfs2_kill_sb,
1410 .owner = THIS_MODULE,
1413 struct file_system_type gfs2meta_fs_type = {
1415 .fs_flags = FS_REQUIRES_DEV,
1416 .mount = gfs2_mount_meta,
1417 .owner = THIS_MODULE,