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"
49 * gfs2_tune_init - Fill a gfs2_tune structure with default values
54 static void gfs2_tune_init(struct gfs2_tune *gt)
56 spin_lock_init(>->gt_spin);
58 gt->gt_quota_warn_period = 10;
59 gt->gt_quota_scale_num = 1;
60 gt->gt_quota_scale_den = 1;
61 gt->gt_new_files_jdata = 0;
62 gt->gt_max_readahead = BIT(18);
63 gt->gt_complain_secs = 10;
66 static struct gfs2_sbd *init_sbd(struct super_block *sb)
69 struct address_space *mapping;
71 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
76 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
77 if (!sdp->sd_lkstats) {
83 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
84 gfs2_tune_init(&sdp->sd_tune);
86 init_waitqueue_head(&sdp->sd_glock_wait);
87 atomic_set(&sdp->sd_glock_disposal, 0);
88 init_completion(&sdp->sd_locking_init);
89 init_completion(&sdp->sd_wdack);
90 spin_lock_init(&sdp->sd_statfs_spin);
92 spin_lock_init(&sdp->sd_rindex_spin);
93 sdp->sd_rindex_tree.rb_node = NULL;
95 INIT_LIST_HEAD(&sdp->sd_jindex_list);
96 spin_lock_init(&sdp->sd_jindex_spin);
97 mutex_init(&sdp->sd_jindex_mutex);
98 init_completion(&sdp->sd_journal_ready);
100 INIT_LIST_HEAD(&sdp->sd_quota_list);
101 mutex_init(&sdp->sd_quota_mutex);
102 mutex_init(&sdp->sd_quota_sync_mutex);
103 init_waitqueue_head(&sdp->sd_quota_wait);
104 INIT_LIST_HEAD(&sdp->sd_trunc_list);
105 spin_lock_init(&sdp->sd_trunc_lock);
106 spin_lock_init(&sdp->sd_bitmap_lock);
108 mapping = &sdp->sd_aspace;
110 address_space_init_once(mapping);
111 mapping->a_ops = &gfs2_rgrp_aops;
112 mapping->host = sb->s_bdev->bd_inode;
114 mapping_set_gfp_mask(mapping, GFP_NOFS);
115 mapping->private_data = NULL;
116 mapping->writeback_index = 0;
118 spin_lock_init(&sdp->sd_log_lock);
119 atomic_set(&sdp->sd_log_pinned, 0);
120 INIT_LIST_HEAD(&sdp->sd_log_le_revoke);
121 INIT_LIST_HEAD(&sdp->sd_log_le_ordered);
122 spin_lock_init(&sdp->sd_ordered_lock);
124 init_waitqueue_head(&sdp->sd_log_waitq);
125 init_waitqueue_head(&sdp->sd_logd_waitq);
126 spin_lock_init(&sdp->sd_ail_lock);
127 INIT_LIST_HEAD(&sdp->sd_ail1_list);
128 INIT_LIST_HEAD(&sdp->sd_ail2_list);
130 init_rwsem(&sdp->sd_log_flush_lock);
131 atomic_set(&sdp->sd_log_in_flight, 0);
132 atomic_set(&sdp->sd_reserving_log, 0);
133 init_waitqueue_head(&sdp->sd_reserving_log_wait);
134 init_waitqueue_head(&sdp->sd_log_flush_wait);
135 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
136 mutex_init(&sdp->sd_freeze_mutex);
143 * gfs2_check_sb - Check superblock
144 * @sdp: the filesystem
145 * @sb: The superblock
146 * @silent: Don't print a message if the check fails
148 * Checks the version code of the FS is one that we understand how to
149 * read and that the sizes of the various on-disk structures have not
153 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
155 struct gfs2_sb_host *sb = &sdp->sd_sb;
157 if (sb->sb_magic != GFS2_MAGIC ||
158 sb->sb_type != GFS2_METATYPE_SB) {
160 pr_warn("not a GFS2 filesystem\n");
164 if (sb->sb_fs_format != GFS2_FORMAT_FS ||
165 sb->sb_multihost_format != GFS2_FORMAT_MULTI) {
166 fs_warn(sdp, "Unknown on-disk format, unable to mount\n");
170 if (sb->sb_bsize < 512 || sb->sb_bsize > PAGE_SIZE ||
171 (sb->sb_bsize & (sb->sb_bsize - 1))) {
172 pr_warn("Invalid superblock size\n");
179 static void end_bio_io_page(struct bio *bio)
181 struct page *page = bio->bi_private;
184 SetPageUptodate(page);
186 pr_warn("error %d reading superblock\n", bio->bi_status);
190 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
192 struct gfs2_sb_host *sb = &sdp->sd_sb;
193 struct super_block *s = sdp->sd_vfs;
194 const struct gfs2_sb *str = buf;
196 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
197 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
198 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
199 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
200 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
201 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
202 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
203 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
204 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
205 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
206 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
208 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
209 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
210 memcpy(&s->s_uuid, str->sb_uuid, 16);
214 * gfs2_read_super - Read the gfs2 super block from disk
215 * @sdp: The GFS2 super block
216 * @sector: The location of the super block
217 * @error: The error code to return
219 * This uses the bio functions to read the super block from disk
220 * because we want to be 100% sure that we never read cached data.
221 * A super block is read twice only during each GFS2 mount and is
222 * never written to by the filesystem. The first time its read no
223 * locks are held, and the only details which are looked at are those
224 * relating to the locking protocol. Once locking is up and working,
225 * the sb is read again under the lock to establish the location of
226 * the master directory (contains pointers to journals etc) and the
229 * Returns: 0 on success or error
232 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
234 struct super_block *sb = sdp->sd_vfs;
239 page = alloc_page(GFP_NOFS);
243 ClearPageUptodate(page);
244 ClearPageDirty(page);
247 bio = bio_alloc(GFP_NOFS, 1);
248 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
249 bio_set_dev(bio, sb->s_bdev);
250 bio_add_page(bio, page, PAGE_SIZE, 0);
252 bio->bi_end_io = end_bio_io_page;
253 bio->bi_private = page;
254 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
256 wait_on_page_locked(page);
258 if (!PageUptodate(page)) {
266 return gfs2_check_sb(sdp, silent);
270 * gfs2_read_sb - Read super block
271 * @sdp: The GFS2 superblock
272 * @silent: Don't print message if mount fails
276 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
278 u32 hash_blocks, ind_blocks, leaf_blocks;
283 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
286 fs_err(sdp, "can't read superblock\n");
290 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
291 GFS2_BASIC_BLOCK_SHIFT;
292 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
293 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
294 sizeof(struct gfs2_dinode)) / sizeof(u64);
295 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
296 sizeof(struct gfs2_meta_header)) / sizeof(u64);
297 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
298 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
299 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
300 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
301 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
302 sizeof(struct gfs2_meta_header)) /
303 sizeof(struct gfs2_quota_change);
304 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
305 sizeof(struct gfs2_meta_header))
306 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
308 /* Compute maximum reservation required to add a entry to a directory */
310 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
314 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
315 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
316 ind_blocks += tmp_blocks;
319 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
321 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
323 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
324 sizeof(struct gfs2_dinode);
325 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
330 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
332 m = do_div(d, sdp->sd_inptrs);
334 if (d != sdp->sd_heightsize[x - 1] || m)
336 sdp->sd_heightsize[x] = space;
338 sdp->sd_max_height = x;
339 sdp->sd_heightsize[x] = ~0;
340 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
342 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
343 sizeof(struct gfs2_dinode);
344 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
349 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
351 m = do_div(d, sdp->sd_inptrs);
353 if (d != sdp->sd_jheightsize[x - 1] || m)
355 sdp->sd_jheightsize[x] = space;
357 sdp->sd_max_jheight = x;
358 sdp->sd_jheightsize[x] = ~0;
359 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
361 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
362 sizeof(struct gfs2_leaf)) /
363 GFS2_MIN_DIRENT_SIZE;
367 static int init_names(struct gfs2_sbd *sdp, int silent)
372 proto = sdp->sd_args.ar_lockproto;
373 table = sdp->sd_args.ar_locktable;
375 /* Try to autodetect */
377 if (!proto[0] || !table[0]) {
378 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
383 proto = sdp->sd_sb.sb_lockproto;
385 table = sdp->sd_sb.sb_locktable;
389 table = sdp->sd_vfs->s_id;
391 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
392 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
394 table = sdp->sd_table_name;
395 while ((table = strchr(table, '/')))
401 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
409 error = gfs2_glock_nq_num(sdp,
410 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
411 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
414 fs_err(sdp, "can't acquire mount glock: %d\n", error);
418 error = gfs2_glock_nq_num(sdp,
419 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
421 LM_FLAG_NOEXP | GL_EXACT,
424 fs_err(sdp, "can't acquire live glock: %d\n", error);
428 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
429 CREATE, &sdp->sd_rename_gl);
431 fs_err(sdp, "can't create rename glock: %d\n", error);
435 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
436 CREATE, &sdp->sd_freeze_gl);
438 fs_err(sdp, "can't create transaction glock: %d\n", error);
445 gfs2_glock_put(sdp->sd_freeze_gl);
447 gfs2_glock_put(sdp->sd_rename_gl);
449 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
451 gfs2_glock_dq_uninit(mount_gh);
456 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
457 u64 no_addr, const char *name)
459 struct gfs2_sbd *sdp = sb->s_fs_info;
460 struct dentry *dentry;
463 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
464 GFS2_BLKST_FREE /* ignore */);
466 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
467 return PTR_ERR(inode);
469 dentry = d_make_root(inode);
471 fs_err(sdp, "can't alloc %s dentry\n", name);
478 static int init_sb(struct gfs2_sbd *sdp, int silent)
480 struct super_block *sb = sdp->sd_vfs;
481 struct gfs2_holder sb_gh;
485 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
486 LM_ST_SHARED, 0, &sb_gh);
488 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
492 ret = gfs2_read_sb(sdp, silent);
494 fs_err(sdp, "can't read superblock: %d\n", ret);
498 /* Set up the buffer cache and SB for real */
499 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
501 fs_err(sdp, "FS block size (%u) is too small for device "
503 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
506 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
508 fs_err(sdp, "FS block size (%u) is too big for machine "
510 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
513 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
515 /* Get the root inode */
516 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
517 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
521 /* Get the master inode */
522 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
523 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
525 dput(sdp->sd_root_dir);
528 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
530 gfs2_glock_dq_uninit(&sb_gh);
534 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
536 char *message = "FIRSTMOUNT=Done";
537 char *envp[] = { message, NULL };
539 fs_info(sdp, "first mount done, others may mount\n");
541 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
542 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
544 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
548 * gfs2_jindex_hold - Grab a lock on the jindex
549 * @sdp: The GFS2 superblock
550 * @ji_gh: the holder for the jindex glock
555 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
557 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
560 struct gfs2_jdesc *jd;
565 mutex_lock(&sdp->sd_jindex_mutex);
568 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
572 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
573 name.hash = gfs2_disk_hash(name.name, name.len);
575 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
576 if (error == -ENOENT) {
581 gfs2_glock_dq_uninit(ji_gh);
587 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
591 INIT_LIST_HEAD(&jd->extent_list);
592 INIT_LIST_HEAD(&jd->jd_revoke_list);
594 INIT_WORK(&jd->jd_work, gfs2_recover_func);
595 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
596 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
600 error = PTR_ERR(jd->jd_inode);
605 spin_lock(&sdp->sd_jindex_spin);
606 jd->jd_jid = sdp->sd_journals++;
607 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
608 spin_unlock(&sdp->sd_jindex_spin);
611 mutex_unlock(&sdp->sd_jindex_mutex);
617 * check_journal_clean - Make sure a journal is clean for a spectator mount
618 * @sdp: The GFS2 superblock
619 * @jd: The journal descriptor
621 * Returns: 0 if the journal is clean or locked, else an error
623 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
626 struct gfs2_holder j_gh;
627 struct gfs2_log_header_host head;
628 struct gfs2_inode *ip;
630 ip = GFS2_I(jd->jd_inode);
631 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
632 GL_EXACT | GL_NOCACHE, &j_gh);
634 fs_err(sdp, "Error locking journal for spectator mount.\n");
637 error = gfs2_jdesc_check(jd);
639 fs_err(sdp, "Error checking journal for spectator mount.\n");
642 error = gfs2_find_jhead(jd, &head);
644 fs_err(sdp, "Error parsing journal for spectator mount.\n");
647 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
649 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
650 "must not be a spectator.\n", jd->jd_jid);
654 gfs2_glock_dq_uninit(&j_gh);
658 static int init_journal(struct gfs2_sbd *sdp, int undo)
660 struct inode *master = d_inode(sdp->sd_master_dir);
661 struct gfs2_holder ji_gh;
662 struct gfs2_inode *ip;
671 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
672 if (IS_ERR(sdp->sd_jindex)) {
673 fs_err(sdp, "can't lookup journal index: %d\n", error);
674 return PTR_ERR(sdp->sd_jindex);
677 /* Load in the journal index special file */
679 error = gfs2_jindex_hold(sdp, &ji_gh);
681 fs_err(sdp, "can't read journal index: %d\n", error);
686 if (!gfs2_jindex_size(sdp)) {
687 fs_err(sdp, "no journals!\n");
691 atomic_set(&sdp->sd_log_blks_needed, 0);
692 if (sdp->sd_args.ar_spectator) {
693 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
694 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
695 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
696 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
698 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
699 fs_err(sdp, "can't mount journal #%u\n",
700 sdp->sd_lockstruct.ls_jid);
701 fs_err(sdp, "there are only %u journals (0 - %u)\n",
702 gfs2_jindex_size(sdp),
703 gfs2_jindex_size(sdp) - 1);
706 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
708 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
710 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
711 &sdp->sd_journal_gh);
713 fs_err(sdp, "can't acquire journal glock: %d\n", error);
717 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
718 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
719 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
722 fs_err(sdp, "can't acquire journal inode glock: %d\n",
724 goto fail_journal_gh;
727 error = gfs2_jdesc_check(sdp->sd_jdesc);
729 fs_err(sdp, "my journal (%u) is bad: %d\n",
730 sdp->sd_jdesc->jd_jid, error);
733 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
734 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
735 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
737 /* Map the extents for this journal's blocks */
738 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
740 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
742 if (sdp->sd_lockstruct.ls_first) {
744 for (x = 0; x < sdp->sd_journals; x++) {
745 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
747 if (sdp->sd_args.ar_spectator) {
748 error = check_journal_clean(sdp, jd);
753 error = gfs2_recover_journal(jd, true);
755 fs_err(sdp, "error recovering journal %u: %d\n",
761 gfs2_others_may_mount(sdp);
762 } else if (!sdp->sd_args.ar_spectator) {
763 error = gfs2_recover_journal(sdp->sd_jdesc, true);
765 fs_err(sdp, "error recovering my journal: %d\n", error);
770 sdp->sd_log_idle = 1;
771 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
772 gfs2_glock_dq_uninit(&ji_gh);
774 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
778 if (!sdp->sd_args.ar_spectator)
779 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
781 if (!sdp->sd_args.ar_spectator)
782 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
784 gfs2_jindex_free(sdp);
786 gfs2_glock_dq_uninit(&ji_gh);
788 iput(sdp->sd_jindex);
792 static struct lock_class_key gfs2_quota_imutex_key;
794 static int init_inodes(struct gfs2_sbd *sdp, int undo)
797 struct inode *master = d_inode(sdp->sd_master_dir);
802 error = init_journal(sdp, undo);
803 complete_all(&sdp->sd_journal_ready);
807 /* Read in the master statfs inode */
808 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
809 if (IS_ERR(sdp->sd_statfs_inode)) {
810 error = PTR_ERR(sdp->sd_statfs_inode);
811 fs_err(sdp, "can't read in statfs inode: %d\n", error);
815 /* Read in the resource index inode */
816 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
817 if (IS_ERR(sdp->sd_rindex)) {
818 error = PTR_ERR(sdp->sd_rindex);
819 fs_err(sdp, "can't get resource index inode: %d\n", error);
822 sdp->sd_rindex_uptodate = 0;
824 /* Read in the quota inode */
825 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
826 if (IS_ERR(sdp->sd_quota_inode)) {
827 error = PTR_ERR(sdp->sd_quota_inode);
828 fs_err(sdp, "can't get quota file inode: %d\n", error);
832 * i_mutex on quota files is special. Since this inode is hidden system
833 * file, we are safe to define locking ourselves.
835 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
836 &gfs2_quota_imutex_key);
838 error = gfs2_rindex_update(sdp);
845 iput(sdp->sd_quota_inode);
847 gfs2_clear_rgrpd(sdp);
848 iput(sdp->sd_rindex);
850 iput(sdp->sd_statfs_inode);
852 init_journal(sdp, UNDO);
857 static int init_per_node(struct gfs2_sbd *sdp, int undo)
859 struct inode *pn = NULL;
862 struct gfs2_inode *ip;
863 struct inode *master = d_inode(sdp->sd_master_dir);
865 if (sdp->sd_args.ar_spectator)
871 pn = gfs2_lookup_simple(master, "per_node");
874 fs_err(sdp, "can't find per_node directory: %d\n", error);
878 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
879 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
880 if (IS_ERR(sdp->sd_sc_inode)) {
881 error = PTR_ERR(sdp->sd_sc_inode);
882 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
886 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
887 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
888 if (IS_ERR(sdp->sd_qc_inode)) {
889 error = PTR_ERR(sdp->sd_qc_inode);
890 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
897 ip = GFS2_I(sdp->sd_sc_inode);
898 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
901 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
905 ip = GFS2_I(sdp->sd_qc_inode);
906 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
909 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
916 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
918 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
920 iput(sdp->sd_qc_inode);
922 iput(sdp->sd_sc_inode);
928 static const match_table_t nolock_tokens = {
929 { Opt_jid, "jid=%d", },
933 static const struct lm_lockops nolock_ops = {
934 .lm_proto_name = "lock_nolock",
935 .lm_put_lock = gfs2_glock_free,
936 .lm_tokens = &nolock_tokens,
940 * gfs2_lm_mount - mount a locking protocol
941 * @sdp: the filesystem
942 * @args: mount arguments
943 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
948 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
950 const struct lm_lockops *lm;
951 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
952 struct gfs2_args *args = &sdp->sd_args;
953 const char *proto = sdp->sd_proto_name;
954 const char *table = sdp->sd_table_name;
958 if (!strcmp("lock_nolock", proto)) {
960 sdp->sd_args.ar_localflocks = 1;
961 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
962 } else if (!strcmp("lock_dlm", proto)) {
966 pr_info("can't find protocol %s\n", proto);
970 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
975 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
976 substring_t tmp[MAX_OPT_ARGS];
982 token = match_token(o, *lm->lm_tokens, tmp);
985 ret = match_int(&tmp[0], &option);
986 if (ret || option < 0)
988 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
993 /* Obsolete, but left for backward compat purposes */
996 ret = match_int(&tmp[0], &option);
997 if (ret || (option != 0 && option != 1))
999 ls->ls_first = option;
1004 fs_info(sdp, "unknown hostdata (%s)\n", o);
1009 if (lm->lm_mount == NULL) {
1010 fs_info(sdp, "Now mounting FS...\n");
1011 complete_all(&sdp->sd_locking_init);
1014 ret = lm->lm_mount(sdp, table);
1016 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1017 complete_all(&sdp->sd_locking_init);
1021 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1023 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1024 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1026 lm->lm_unmount(sdp);
1029 static int wait_on_journal(struct gfs2_sbd *sdp)
1031 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1034 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1038 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1040 struct super_block *sb = sdp->sd_vfs;
1043 char *envp[] = { ro, spectator, NULL };
1044 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1045 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1046 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1050 * fill_super - Read in superblock
1051 * @sb: The VFS superblock
1052 * @data: Mount options
1053 * @silent: Don't complain if it's not a GFS2 filesystem
1058 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1060 struct gfs2_sbd *sdp;
1061 struct gfs2_holder mount_gh;
1066 pr_warn("can't alloc struct gfs2_sbd\n");
1069 sdp->sd_args = *args;
1071 if (sdp->sd_args.ar_spectator) {
1072 sb->s_flags |= MS_RDONLY;
1073 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1075 if (sdp->sd_args.ar_posix_acl)
1076 sb->s_flags |= MS_POSIXACL;
1077 if (sdp->sd_args.ar_nobarrier)
1078 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1080 sb->s_flags |= MS_NOSEC;
1081 sb->s_magic = GFS2_MAGIC;
1082 sb->s_op = &gfs2_super_ops;
1083 sb->s_d_op = &gfs2_dops;
1084 sb->s_export_op = &gfs2_export_ops;
1085 sb->s_xattr = gfs2_xattr_handlers;
1086 sb->s_qcop = &gfs2_quotactl_ops;
1087 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1088 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1089 sb->s_time_gran = 1;
1090 sb->s_maxbytes = MAX_LFS_FILESIZE;
1092 /* Set up the buffer cache and fill in some fake block size values
1093 to allow us to read-in the on-disk superblock. */
1094 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1095 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1096 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1097 GFS2_BASIC_BLOCK_SHIFT;
1098 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1100 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1101 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1102 if (sdp->sd_args.ar_statfs_quantum) {
1103 sdp->sd_tune.gt_statfs_slow = 0;
1104 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1106 sdp->sd_tune.gt_statfs_slow = 1;
1107 sdp->sd_tune.gt_statfs_quantum = 30;
1110 error = init_names(sdp, silent);
1112 /* In this case, we haven't initialized sysfs, so we have to
1113 manually free the sdp. */
1114 free_percpu(sdp->sd_lkstats);
1116 sb->s_fs_info = NULL;
1120 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1122 error = gfs2_sys_fs_add(sdp);
1124 * If we hit an error here, gfs2_sys_fs_add will have called function
1125 * kobject_put which causes the sysfs usage count to go to zero, which
1126 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1127 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1128 * kobject_put to free sdp.
1133 gfs2_create_debugfs_file(sdp);
1135 error = gfs2_lm_mount(sdp, silent);
1139 error = init_locking(sdp, &mount_gh, DO);
1143 error = init_sb(sdp, silent);
1147 error = wait_on_journal(sdp);
1152 * If user space has failed to join the cluster or some similar
1153 * failure has occurred, then the journal id will contain a
1154 * negative (error) number. This will then be returned to the
1155 * caller (of the mount syscall). We do this even for spectator
1156 * mounts (which just write a jid of 0 to indicate "ok" even though
1157 * the jid is unused in the spectator case)
1159 if (sdp->sd_lockstruct.ls_jid < 0) {
1160 error = sdp->sd_lockstruct.ls_jid;
1161 sdp->sd_lockstruct.ls_jid = 0;
1165 if (sdp->sd_args.ar_spectator)
1166 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1167 sdp->sd_table_name);
1169 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1170 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1172 error = init_inodes(sdp, DO);
1176 error = init_per_node(sdp, DO);
1180 error = gfs2_statfs_init(sdp);
1182 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1186 if (sb_rdonly(sb)) {
1187 struct gfs2_holder freeze_gh;
1189 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1190 GL_EXACT, &freeze_gh);
1192 fs_err(sdp, "can't make FS RO: %d\n", error);
1195 gfs2_glock_dq_uninit(&freeze_gh);
1197 error = gfs2_make_fs_rw(sdp);
1199 fs_err(sdp, "can't make FS RW: %d\n", error);
1204 gfs2_glock_dq_uninit(&mount_gh);
1205 gfs2_online_uevent(sdp);
1209 init_per_node(sdp, UNDO);
1211 init_inodes(sdp, UNDO);
1213 if (sdp->sd_root_dir)
1214 dput(sdp->sd_root_dir);
1215 if (sdp->sd_master_dir)
1216 dput(sdp->sd_master_dir);
1221 init_locking(sdp, &mount_gh, UNDO);
1223 complete_all(&sdp->sd_journal_ready);
1224 gfs2_gl_hash_clear(sdp);
1225 gfs2_lm_unmount(sdp);
1227 gfs2_delete_debugfs_file(sdp);
1228 free_percpu(sdp->sd_lkstats);
1229 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1230 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1231 gfs2_sys_fs_del(sdp);
1232 sb->s_fs_info = NULL;
1236 static int set_gfs2_super(struct super_block *s, void *data)
1239 s->s_dev = s->s_bdev->bd_dev;
1240 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1244 static int test_gfs2_super(struct super_block *s, void *ptr)
1246 struct block_device *bdev = ptr;
1247 return (bdev == s->s_bdev);
1251 * gfs2_mount - Get the GFS2 superblock
1252 * @fs_type: The GFS2 filesystem type
1253 * @flags: Mount flags
1254 * @dev_name: The name of the device
1255 * @data: The mount arguments
1257 * Q. Why not use get_sb_bdev() ?
1258 * A. We need to select one of two root directories to mount, independent
1259 * of whether this is the initial, or subsequent, mount of this sb
1261 * Returns: 0 or -ve on error
1264 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1265 const char *dev_name, void *data)
1267 struct block_device *bdev;
1268 struct super_block *s;
1269 fmode_t mode = FMODE_READ | FMODE_EXCL;
1271 struct gfs2_args args;
1272 struct gfs2_sbd *sdp;
1274 if (!(flags & MS_RDONLY))
1275 mode |= FMODE_WRITE;
1277 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1279 return ERR_CAST(bdev);
1282 * once the super is inserted into the list by sget, s_umount
1283 * will protect the lockfs code from trying to start a snapshot
1284 * while we are mounting
1286 mutex_lock(&bdev->bd_fsfreeze_mutex);
1287 if (bdev->bd_fsfreeze_count > 0) {
1288 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1292 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1300 * s_umount nests inside bd_mutex during
1301 * __invalidate_device(). blkdev_put() acquires
1302 * bd_mutex and can't be called under s_umount. Drop
1303 * s_umount temporarily. This is safe as we're
1304 * holding an active reference.
1306 up_write(&s->s_umount);
1307 blkdev_put(bdev, mode);
1308 down_write(&s->s_umount);
1310 /* s_mode must be set before deactivate_locked_super calls */
1314 memset(&args, 0, sizeof(args));
1315 args.ar_quota = GFS2_QUOTA_DEFAULT;
1316 args.ar_data = GFS2_DATA_DEFAULT;
1317 args.ar_commit = 30;
1318 args.ar_statfs_quantum = 30;
1319 args.ar_quota_quantum = 60;
1320 args.ar_errors = GFS2_ERRORS_DEFAULT;
1322 error = gfs2_mount_args(&args, data);
1324 pr_warn("can't parse mount arguments\n");
1330 if ((flags ^ s->s_flags) & MS_RDONLY)
1333 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1334 sb_set_blocksize(s, block_size(bdev));
1335 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1338 s->s_flags |= MS_ACTIVE;
1344 return dget(sdp->sd_master_dir);
1346 return dget(sdp->sd_root_dir);
1349 deactivate_locked_super(s);
1350 return ERR_PTR(error);
1352 blkdev_put(bdev, mode);
1353 return ERR_PTR(error);
1356 static int set_meta_super(struct super_block *s, void *ptr)
1361 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1362 int flags, const char *dev_name, void *data)
1364 struct super_block *s;
1365 struct gfs2_sbd *sdp;
1369 if (!dev_name || !*dev_name)
1370 return ERR_PTR(-EINVAL);
1372 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1374 pr_warn("path_lookup on %s returned error %d\n",
1376 return ERR_PTR(error);
1378 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1379 path.dentry->d_sb->s_bdev);
1382 pr_warn("gfs2 mount does not exist\n");
1385 if ((flags ^ s->s_flags) & MS_RDONLY) {
1386 deactivate_locked_super(s);
1387 return ERR_PTR(-EBUSY);
1390 return dget(sdp->sd_master_dir);
1393 static void gfs2_kill_sb(struct super_block *sb)
1395 struct gfs2_sbd *sdp = sb->s_fs_info;
1398 kill_block_super(sb);
1402 gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1403 dput(sdp->sd_root_dir);
1404 dput(sdp->sd_master_dir);
1405 sdp->sd_root_dir = NULL;
1406 sdp->sd_master_dir = NULL;
1407 shrink_dcache_sb(sb);
1408 free_percpu(sdp->sd_lkstats);
1409 kill_block_super(sb);
1412 struct file_system_type gfs2_fs_type = {
1414 .fs_flags = FS_REQUIRES_DEV,
1415 .mount = gfs2_mount,
1416 .kill_sb = gfs2_kill_sb,
1417 .owner = THIS_MODULE,
1419 MODULE_ALIAS_FS("gfs2");
1421 struct file_system_type gfs2meta_fs_type = {
1423 .fs_flags = FS_REQUIRES_DEV,
1424 .mount = gfs2_mount_meta,
1425 .owner = THIS_MODULE,
1427 MODULE_ALIAS_FS("gfs2meta");