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");
175 if (sb->sb_bsize_shift != ffs(sb->sb_bsize) - 1) {
176 pr_warn("Invalid block size shift\n");
182 static void end_bio_io_page(struct bio *bio)
184 struct page *page = bio->bi_private;
187 SetPageUptodate(page);
189 pr_warn("error %d reading superblock\n", bio->bi_status);
193 static void gfs2_sb_in(struct gfs2_sbd *sdp, const void *buf)
195 struct gfs2_sb_host *sb = &sdp->sd_sb;
196 struct super_block *s = sdp->sd_vfs;
197 const struct gfs2_sb *str = buf;
199 sb->sb_magic = be32_to_cpu(str->sb_header.mh_magic);
200 sb->sb_type = be32_to_cpu(str->sb_header.mh_type);
201 sb->sb_format = be32_to_cpu(str->sb_header.mh_format);
202 sb->sb_fs_format = be32_to_cpu(str->sb_fs_format);
203 sb->sb_multihost_format = be32_to_cpu(str->sb_multihost_format);
204 sb->sb_bsize = be32_to_cpu(str->sb_bsize);
205 sb->sb_bsize_shift = be32_to_cpu(str->sb_bsize_shift);
206 sb->sb_master_dir.no_addr = be64_to_cpu(str->sb_master_dir.no_addr);
207 sb->sb_master_dir.no_formal_ino = be64_to_cpu(str->sb_master_dir.no_formal_ino);
208 sb->sb_root_dir.no_addr = be64_to_cpu(str->sb_root_dir.no_addr);
209 sb->sb_root_dir.no_formal_ino = be64_to_cpu(str->sb_root_dir.no_formal_ino);
211 memcpy(sb->sb_lockproto, str->sb_lockproto, GFS2_LOCKNAME_LEN);
212 memcpy(sb->sb_locktable, str->sb_locktable, GFS2_LOCKNAME_LEN);
213 memcpy(&s->s_uuid, str->sb_uuid, 16);
217 * gfs2_read_super - Read the gfs2 super block from disk
218 * @sdp: The GFS2 super block
219 * @sector: The location of the super block
220 * @error: The error code to return
222 * This uses the bio functions to read the super block from disk
223 * because we want to be 100% sure that we never read cached data.
224 * A super block is read twice only during each GFS2 mount and is
225 * never written to by the filesystem. The first time its read no
226 * locks are held, and the only details which are looked at are those
227 * relating to the locking protocol. Once locking is up and working,
228 * the sb is read again under the lock to establish the location of
229 * the master directory (contains pointers to journals etc) and the
232 * Returns: 0 on success or error
235 static int gfs2_read_super(struct gfs2_sbd *sdp, sector_t sector, int silent)
237 struct super_block *sb = sdp->sd_vfs;
242 page = alloc_page(GFP_NOFS);
246 ClearPageUptodate(page);
247 ClearPageDirty(page);
250 bio = bio_alloc(GFP_NOFS, 1);
251 bio->bi_iter.bi_sector = sector * (sb->s_blocksize >> 9);
252 bio_set_dev(bio, sb->s_bdev);
253 bio_add_page(bio, page, PAGE_SIZE, 0);
255 bio->bi_end_io = end_bio_io_page;
256 bio->bi_private = page;
257 bio_set_op_attrs(bio, REQ_OP_READ, REQ_META);
259 wait_on_page_locked(page);
261 if (!PageUptodate(page)) {
269 return gfs2_check_sb(sdp, silent);
273 * gfs2_read_sb - Read super block
274 * @sdp: The GFS2 superblock
275 * @silent: Don't print message if mount fails
279 static int gfs2_read_sb(struct gfs2_sbd *sdp, int silent)
281 u32 hash_blocks, ind_blocks, leaf_blocks;
286 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
289 fs_err(sdp, "can't read superblock\n");
293 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
294 GFS2_BASIC_BLOCK_SHIFT;
295 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
296 sdp->sd_diptrs = (sdp->sd_sb.sb_bsize -
297 sizeof(struct gfs2_dinode)) / sizeof(u64);
298 sdp->sd_inptrs = (sdp->sd_sb.sb_bsize -
299 sizeof(struct gfs2_meta_header)) / sizeof(u64);
300 sdp->sd_jbsize = sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header);
301 sdp->sd_hash_bsize = sdp->sd_sb.sb_bsize / 2;
302 sdp->sd_hash_bsize_shift = sdp->sd_sb.sb_bsize_shift - 1;
303 sdp->sd_hash_ptrs = sdp->sd_hash_bsize / sizeof(u64);
304 sdp->sd_qc_per_block = (sdp->sd_sb.sb_bsize -
305 sizeof(struct gfs2_meta_header)) /
306 sizeof(struct gfs2_quota_change);
307 sdp->sd_blocks_per_bitmap = (sdp->sd_sb.sb_bsize -
308 sizeof(struct gfs2_meta_header))
309 * GFS2_NBBY; /* not the rgrp bitmap, subsequent bitmaps only */
311 /* Compute maximum reservation required to add a entry to a directory */
313 hash_blocks = DIV_ROUND_UP(sizeof(u64) * BIT(GFS2_DIR_MAX_DEPTH),
317 for (tmp_blocks = hash_blocks; tmp_blocks > sdp->sd_diptrs;) {
318 tmp_blocks = DIV_ROUND_UP(tmp_blocks, sdp->sd_inptrs);
319 ind_blocks += tmp_blocks;
322 leaf_blocks = 2 + GFS2_DIR_MAX_DEPTH;
324 sdp->sd_max_dirres = hash_blocks + ind_blocks + leaf_blocks;
326 sdp->sd_heightsize[0] = sdp->sd_sb.sb_bsize -
327 sizeof(struct gfs2_dinode);
328 sdp->sd_heightsize[1] = sdp->sd_sb.sb_bsize * sdp->sd_diptrs;
333 space = sdp->sd_heightsize[x - 1] * sdp->sd_inptrs;
335 m = do_div(d, sdp->sd_inptrs);
337 if (d != sdp->sd_heightsize[x - 1] || m)
339 sdp->sd_heightsize[x] = space;
341 sdp->sd_max_height = x;
342 sdp->sd_heightsize[x] = ~0;
343 gfs2_assert(sdp, sdp->sd_max_height <= GFS2_MAX_META_HEIGHT);
345 sdp->sd_jheightsize[0] = sdp->sd_sb.sb_bsize -
346 sizeof(struct gfs2_dinode);
347 sdp->sd_jheightsize[1] = sdp->sd_jbsize * sdp->sd_diptrs;
352 space = sdp->sd_jheightsize[x - 1] * sdp->sd_inptrs;
354 m = do_div(d, sdp->sd_inptrs);
356 if (d != sdp->sd_jheightsize[x - 1] || m)
358 sdp->sd_jheightsize[x] = space;
360 sdp->sd_max_jheight = x;
361 sdp->sd_jheightsize[x] = ~0;
362 gfs2_assert(sdp, sdp->sd_max_jheight <= GFS2_MAX_META_HEIGHT);
364 sdp->sd_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
365 sizeof(struct gfs2_leaf)) /
366 GFS2_MIN_DIRENT_SIZE;
370 static int init_names(struct gfs2_sbd *sdp, int silent)
375 proto = sdp->sd_args.ar_lockproto;
376 table = sdp->sd_args.ar_locktable;
378 /* Try to autodetect */
380 if (!proto[0] || !table[0]) {
381 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
386 proto = sdp->sd_sb.sb_lockproto;
388 table = sdp->sd_sb.sb_locktable;
392 table = sdp->sd_vfs->s_id;
394 BUILD_BUG_ON(GFS2_LOCKNAME_LEN > GFS2_FSNAME_LEN);
396 strscpy(sdp->sd_proto_name, proto, GFS2_LOCKNAME_LEN);
397 strscpy(sdp->sd_table_name, table, GFS2_LOCKNAME_LEN);
399 table = sdp->sd_table_name;
400 while ((table = strchr(table, '/')))
406 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
414 error = gfs2_glock_nq_num(sdp,
415 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
416 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
419 fs_err(sdp, "can't acquire mount glock: %d\n", error);
423 error = gfs2_glock_nq_num(sdp,
424 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
426 LM_FLAG_NOEXP | GL_EXACT,
429 fs_err(sdp, "can't acquire live glock: %d\n", error);
433 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
434 CREATE, &sdp->sd_rename_gl);
436 fs_err(sdp, "can't create rename glock: %d\n", error);
440 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
441 CREATE, &sdp->sd_freeze_gl);
443 fs_err(sdp, "can't create transaction glock: %d\n", error);
450 gfs2_glock_put(sdp->sd_freeze_gl);
452 gfs2_glock_put(sdp->sd_rename_gl);
454 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
456 gfs2_glock_dq_uninit(mount_gh);
461 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
462 u64 no_addr, const char *name)
464 struct gfs2_sbd *sdp = sb->s_fs_info;
465 struct dentry *dentry;
468 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
469 GFS2_BLKST_FREE /* ignore */);
471 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
472 return PTR_ERR(inode);
474 dentry = d_make_root(inode);
476 fs_err(sdp, "can't alloc %s dentry\n", name);
483 static int init_sb(struct gfs2_sbd *sdp, int silent)
485 struct super_block *sb = sdp->sd_vfs;
486 struct gfs2_holder sb_gh;
490 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
491 LM_ST_SHARED, 0, &sb_gh);
493 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
497 ret = gfs2_read_sb(sdp, silent);
499 fs_err(sdp, "can't read superblock: %d\n", ret);
503 /* Set up the buffer cache and SB for real */
504 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
506 fs_err(sdp, "FS block size (%u) is too small for device "
508 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
511 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
513 fs_err(sdp, "FS block size (%u) is too big for machine "
515 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
518 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
520 /* Get the root inode */
521 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
522 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
526 /* Get the master inode */
527 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
528 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
530 dput(sdp->sd_root_dir);
533 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
535 gfs2_glock_dq_uninit(&sb_gh);
539 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
541 char *message = "FIRSTMOUNT=Done";
542 char *envp[] = { message, NULL };
544 fs_info(sdp, "first mount done, others may mount\n");
546 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
547 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
549 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
553 * gfs2_jindex_hold - Grab a lock on the jindex
554 * @sdp: The GFS2 superblock
555 * @ji_gh: the holder for the jindex glock
560 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
562 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
565 struct gfs2_jdesc *jd;
570 mutex_lock(&sdp->sd_jindex_mutex);
573 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
577 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
578 name.hash = gfs2_disk_hash(name.name, name.len);
580 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
581 if (error == -ENOENT) {
586 gfs2_glock_dq_uninit(ji_gh);
592 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
596 INIT_LIST_HEAD(&jd->extent_list);
597 INIT_LIST_HEAD(&jd->jd_revoke_list);
599 INIT_WORK(&jd->jd_work, gfs2_recover_func);
600 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
601 if (!jd->jd_inode || IS_ERR(jd->jd_inode)) {
605 error = PTR_ERR(jd->jd_inode);
610 spin_lock(&sdp->sd_jindex_spin);
611 jd->jd_jid = sdp->sd_journals++;
612 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
613 spin_unlock(&sdp->sd_jindex_spin);
616 mutex_unlock(&sdp->sd_jindex_mutex);
622 * check_journal_clean - Make sure a journal is clean for a spectator mount
623 * @sdp: The GFS2 superblock
624 * @jd: The journal descriptor
626 * Returns: 0 if the journal is clean or locked, else an error
628 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
631 struct gfs2_holder j_gh;
632 struct gfs2_log_header_host head;
633 struct gfs2_inode *ip;
635 ip = GFS2_I(jd->jd_inode);
636 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
637 GL_EXACT | GL_NOCACHE, &j_gh);
639 fs_err(sdp, "Error locking journal for spectator mount.\n");
642 error = gfs2_jdesc_check(jd);
644 fs_err(sdp, "Error checking journal for spectator mount.\n");
647 error = gfs2_find_jhead(jd, &head);
649 fs_err(sdp, "Error parsing journal for spectator mount.\n");
652 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
654 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
655 "must not be a spectator.\n", jd->jd_jid);
659 gfs2_glock_dq_uninit(&j_gh);
663 static int init_journal(struct gfs2_sbd *sdp, int undo)
665 struct inode *master = d_inode(sdp->sd_master_dir);
666 struct gfs2_holder ji_gh;
667 struct gfs2_inode *ip;
676 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
677 if (IS_ERR(sdp->sd_jindex)) {
678 fs_err(sdp, "can't lookup journal index: %d\n", error);
679 return PTR_ERR(sdp->sd_jindex);
682 /* Load in the journal index special file */
684 error = gfs2_jindex_hold(sdp, &ji_gh);
686 fs_err(sdp, "can't read journal index: %d\n", error);
691 if (!gfs2_jindex_size(sdp)) {
692 fs_err(sdp, "no journals!\n");
696 atomic_set(&sdp->sd_log_blks_needed, 0);
697 if (sdp->sd_args.ar_spectator) {
698 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
699 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
700 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
701 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
703 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
704 fs_err(sdp, "can't mount journal #%u\n",
705 sdp->sd_lockstruct.ls_jid);
706 fs_err(sdp, "there are only %u journals (0 - %u)\n",
707 gfs2_jindex_size(sdp),
708 gfs2_jindex_size(sdp) - 1);
711 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
713 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
715 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
716 &sdp->sd_journal_gh);
718 fs_err(sdp, "can't acquire journal glock: %d\n", error);
722 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
723 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
724 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
727 fs_err(sdp, "can't acquire journal inode glock: %d\n",
729 goto fail_journal_gh;
732 error = gfs2_jdesc_check(sdp->sd_jdesc);
734 fs_err(sdp, "my journal (%u) is bad: %d\n",
735 sdp->sd_jdesc->jd_jid, error);
738 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
739 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
740 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
742 /* Map the extents for this journal's blocks */
743 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
745 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
747 if (sdp->sd_lockstruct.ls_first) {
749 for (x = 0; x < sdp->sd_journals; x++) {
750 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
752 if (sdp->sd_args.ar_spectator) {
753 error = check_journal_clean(sdp, jd);
758 error = gfs2_recover_journal(jd, true);
760 fs_err(sdp, "error recovering journal %u: %d\n",
766 gfs2_others_may_mount(sdp);
767 } else if (!sdp->sd_args.ar_spectator) {
768 error = gfs2_recover_journal(sdp->sd_jdesc, true);
770 fs_err(sdp, "error recovering my journal: %d\n", error);
775 sdp->sd_log_idle = 1;
776 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
777 gfs2_glock_dq_uninit(&ji_gh);
779 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
783 if (!sdp->sd_args.ar_spectator)
784 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
786 if (!sdp->sd_args.ar_spectator)
787 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
789 gfs2_jindex_free(sdp);
791 gfs2_glock_dq_uninit(&ji_gh);
793 iput(sdp->sd_jindex);
797 static struct lock_class_key gfs2_quota_imutex_key;
799 static int init_inodes(struct gfs2_sbd *sdp, int undo)
802 struct inode *master = d_inode(sdp->sd_master_dir);
807 error = init_journal(sdp, undo);
808 complete_all(&sdp->sd_journal_ready);
812 /* Read in the master statfs inode */
813 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
814 if (IS_ERR(sdp->sd_statfs_inode)) {
815 error = PTR_ERR(sdp->sd_statfs_inode);
816 fs_err(sdp, "can't read in statfs inode: %d\n", error);
820 /* Read in the resource index inode */
821 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
822 if (IS_ERR(sdp->sd_rindex)) {
823 error = PTR_ERR(sdp->sd_rindex);
824 fs_err(sdp, "can't get resource index inode: %d\n", error);
827 sdp->sd_rindex_uptodate = 0;
829 /* Read in the quota inode */
830 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
831 if (IS_ERR(sdp->sd_quota_inode)) {
832 error = PTR_ERR(sdp->sd_quota_inode);
833 fs_err(sdp, "can't get quota file inode: %d\n", error);
837 * i_mutex on quota files is special. Since this inode is hidden system
838 * file, we are safe to define locking ourselves.
840 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
841 &gfs2_quota_imutex_key);
843 error = gfs2_rindex_update(sdp);
850 iput(sdp->sd_quota_inode);
852 gfs2_clear_rgrpd(sdp);
853 iput(sdp->sd_rindex);
855 iput(sdp->sd_statfs_inode);
857 init_journal(sdp, UNDO);
862 static int init_per_node(struct gfs2_sbd *sdp, int undo)
864 struct inode *pn = NULL;
867 struct gfs2_inode *ip;
868 struct inode *master = d_inode(sdp->sd_master_dir);
870 if (sdp->sd_args.ar_spectator)
876 pn = gfs2_lookup_simple(master, "per_node");
879 fs_err(sdp, "can't find per_node directory: %d\n", error);
883 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
884 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
885 if (IS_ERR(sdp->sd_sc_inode)) {
886 error = PTR_ERR(sdp->sd_sc_inode);
887 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
891 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
892 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
893 if (IS_ERR(sdp->sd_qc_inode)) {
894 error = PTR_ERR(sdp->sd_qc_inode);
895 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
902 ip = GFS2_I(sdp->sd_sc_inode);
903 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
906 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
910 ip = GFS2_I(sdp->sd_qc_inode);
911 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
914 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
921 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
923 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
925 iput(sdp->sd_qc_inode);
927 iput(sdp->sd_sc_inode);
933 static const match_table_t nolock_tokens = {
934 { Opt_jid, "jid=%d", },
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 pr_info("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))
998 /* Obsolete, but left for backward compat purposes */
1001 ret = match_int(&tmp[0], &option);
1002 if (ret || (option != 0 && option != 1))
1003 goto hostdata_error;
1004 ls->ls_first = option;
1009 fs_info(sdp, "unknown hostdata (%s)\n", o);
1014 if (lm->lm_mount == NULL) {
1015 fs_info(sdp, "Now mounting FS...\n");
1016 complete_all(&sdp->sd_locking_init);
1019 ret = lm->lm_mount(sdp, table);
1021 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1022 complete_all(&sdp->sd_locking_init);
1026 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1028 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1029 if (likely(!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) &&
1031 lm->lm_unmount(sdp);
1034 static int wait_on_journal(struct gfs2_sbd *sdp)
1036 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1039 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1043 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1045 struct super_block *sb = sdp->sd_vfs;
1048 char *envp[] = { ro, spectator, NULL };
1049 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1050 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1051 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1055 * fill_super - Read in superblock
1056 * @sb: The VFS superblock
1057 * @data: Mount options
1058 * @silent: Don't complain if it's not a GFS2 filesystem
1063 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1065 struct gfs2_sbd *sdp;
1066 struct gfs2_holder mount_gh;
1071 pr_warn("can't alloc struct gfs2_sbd\n");
1074 sdp->sd_args = *args;
1076 if (sdp->sd_args.ar_spectator) {
1077 sb->s_flags |= MS_RDONLY;
1078 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1080 if (sdp->sd_args.ar_posix_acl)
1081 sb->s_flags |= MS_POSIXACL;
1082 if (sdp->sd_args.ar_nobarrier)
1083 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1085 sb->s_flags |= MS_NOSEC;
1086 sb->s_magic = GFS2_MAGIC;
1087 sb->s_op = &gfs2_super_ops;
1088 sb->s_d_op = &gfs2_dops;
1089 sb->s_export_op = &gfs2_export_ops;
1090 sb->s_xattr = gfs2_xattr_handlers;
1091 sb->s_qcop = &gfs2_quotactl_ops;
1092 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1093 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1094 sb->s_time_gran = 1;
1095 sb->s_maxbytes = MAX_LFS_FILESIZE;
1097 /* Set up the buffer cache and fill in some fake block size values
1098 to allow us to read-in the on-disk superblock. */
1099 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1100 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1101 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1102 GFS2_BASIC_BLOCK_SHIFT;
1103 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1105 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1106 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1107 if (sdp->sd_args.ar_statfs_quantum) {
1108 sdp->sd_tune.gt_statfs_slow = 0;
1109 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1111 sdp->sd_tune.gt_statfs_slow = 1;
1112 sdp->sd_tune.gt_statfs_quantum = 30;
1115 error = init_names(sdp, silent);
1117 /* In this case, we haven't initialized sysfs, so we have to
1118 manually free the sdp. */
1119 free_percpu(sdp->sd_lkstats);
1121 sb->s_fs_info = NULL;
1125 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1127 error = gfs2_sys_fs_add(sdp);
1129 * If we hit an error here, gfs2_sys_fs_add will have called function
1130 * kobject_put which causes the sysfs usage count to go to zero, which
1131 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1132 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1133 * kobject_put to free sdp.
1138 gfs2_create_debugfs_file(sdp);
1140 error = gfs2_lm_mount(sdp, silent);
1144 error = init_locking(sdp, &mount_gh, DO);
1148 error = init_sb(sdp, silent);
1152 error = wait_on_journal(sdp);
1157 * If user space has failed to join the cluster or some similar
1158 * failure has occurred, then the journal id will contain a
1159 * negative (error) number. This will then be returned to the
1160 * caller (of the mount syscall). We do this even for spectator
1161 * mounts (which just write a jid of 0 to indicate "ok" even though
1162 * the jid is unused in the spectator case)
1164 if (sdp->sd_lockstruct.ls_jid < 0) {
1165 error = sdp->sd_lockstruct.ls_jid;
1166 sdp->sd_lockstruct.ls_jid = 0;
1170 if (sdp->sd_args.ar_spectator)
1171 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1172 sdp->sd_table_name);
1174 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1175 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1177 error = init_inodes(sdp, DO);
1181 error = init_per_node(sdp, DO);
1185 error = gfs2_statfs_init(sdp);
1187 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1191 if (sb_rdonly(sb)) {
1192 struct gfs2_holder freeze_gh;
1194 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
1195 GL_EXACT, &freeze_gh);
1197 fs_err(sdp, "can't make FS RO: %d\n", error);
1200 gfs2_glock_dq_uninit(&freeze_gh);
1202 error = gfs2_make_fs_rw(sdp);
1204 fs_err(sdp, "can't make FS RW: %d\n", error);
1209 gfs2_glock_dq_uninit(&mount_gh);
1210 gfs2_online_uevent(sdp);
1214 init_per_node(sdp, UNDO);
1216 init_inodes(sdp, UNDO);
1218 if (sdp->sd_root_dir)
1219 dput(sdp->sd_root_dir);
1220 if (sdp->sd_master_dir)
1221 dput(sdp->sd_master_dir);
1226 init_locking(sdp, &mount_gh, UNDO);
1228 complete_all(&sdp->sd_journal_ready);
1229 gfs2_gl_hash_clear(sdp);
1230 gfs2_lm_unmount(sdp);
1232 gfs2_delete_debugfs_file(sdp);
1233 free_percpu(sdp->sd_lkstats);
1234 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1235 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1236 gfs2_sys_fs_del(sdp);
1237 sb->s_fs_info = NULL;
1241 static int set_gfs2_super(struct super_block *s, void *data)
1244 s->s_dev = s->s_bdev->bd_dev;
1245 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1249 static int test_gfs2_super(struct super_block *s, void *ptr)
1251 struct block_device *bdev = ptr;
1252 return (bdev == s->s_bdev);
1256 * gfs2_mount - Get the GFS2 superblock
1257 * @fs_type: The GFS2 filesystem type
1258 * @flags: Mount flags
1259 * @dev_name: The name of the device
1260 * @data: The mount arguments
1262 * Q. Why not use get_sb_bdev() ?
1263 * A. We need to select one of two root directories to mount, independent
1264 * of whether this is the initial, or subsequent, mount of this sb
1266 * Returns: 0 or -ve on error
1269 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1270 const char *dev_name, void *data)
1272 struct block_device *bdev;
1273 struct super_block *s;
1274 fmode_t mode = FMODE_READ | FMODE_EXCL;
1276 struct gfs2_args args;
1277 struct gfs2_sbd *sdp;
1279 if (!(flags & MS_RDONLY))
1280 mode |= FMODE_WRITE;
1282 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1284 return ERR_CAST(bdev);
1287 * once the super is inserted into the list by sget, s_umount
1288 * will protect the lockfs code from trying to start a snapshot
1289 * while we are mounting
1291 mutex_lock(&bdev->bd_fsfreeze_mutex);
1292 if (bdev->bd_fsfreeze_count > 0) {
1293 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1297 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1298 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1305 * s_umount nests inside bd_mutex during
1306 * __invalidate_device(). blkdev_put() acquires
1307 * bd_mutex and can't be called under s_umount. Drop
1308 * s_umount temporarily. This is safe as we're
1309 * holding an active reference.
1311 up_write(&s->s_umount);
1312 blkdev_put(bdev, mode);
1313 down_write(&s->s_umount);
1315 /* s_mode must be set before deactivate_locked_super calls */
1319 memset(&args, 0, sizeof(args));
1320 args.ar_quota = GFS2_QUOTA_DEFAULT;
1321 args.ar_data = GFS2_DATA_DEFAULT;
1322 args.ar_commit = 30;
1323 args.ar_statfs_quantum = 30;
1324 args.ar_quota_quantum = 60;
1325 args.ar_errors = GFS2_ERRORS_DEFAULT;
1327 error = gfs2_mount_args(&args, data);
1329 pr_warn("can't parse mount arguments\n");
1335 if ((flags ^ s->s_flags) & MS_RDONLY)
1338 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1339 sb_set_blocksize(s, block_size(bdev));
1340 error = fill_super(s, &args, flags & MS_SILENT ? 1 : 0);
1343 s->s_flags |= MS_ACTIVE;
1349 return dget(sdp->sd_master_dir);
1351 return dget(sdp->sd_root_dir);
1354 deactivate_locked_super(s);
1355 return ERR_PTR(error);
1357 blkdev_put(bdev, mode);
1358 return ERR_PTR(error);
1361 static int set_meta_super(struct super_block *s, void *ptr)
1366 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1367 int flags, const char *dev_name, void *data)
1369 struct super_block *s;
1370 struct gfs2_sbd *sdp;
1374 if (!dev_name || !*dev_name)
1375 return ERR_PTR(-EINVAL);
1377 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1379 pr_warn("path_lookup on %s returned error %d\n",
1381 return ERR_PTR(error);
1383 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1384 path.dentry->d_sb->s_bdev);
1387 pr_warn("gfs2 mount does not exist\n");
1390 if ((flags ^ s->s_flags) & MS_RDONLY) {
1391 deactivate_locked_super(s);
1392 return ERR_PTR(-EBUSY);
1395 return dget(sdp->sd_master_dir);
1398 static void gfs2_kill_sb(struct super_block *sb)
1400 struct gfs2_sbd *sdp = sb->s_fs_info;
1403 kill_block_super(sb);
1407 gfs2_log_flush(sdp, NULL, SYNC_FLUSH);
1408 dput(sdp->sd_root_dir);
1409 dput(sdp->sd_master_dir);
1410 sdp->sd_root_dir = NULL;
1411 sdp->sd_master_dir = NULL;
1412 shrink_dcache_sb(sb);
1413 free_percpu(sdp->sd_lkstats);
1414 kill_block_super(sb);
1417 struct file_system_type gfs2_fs_type = {
1419 .fs_flags = FS_REQUIRES_DEV,
1420 .mount = gfs2_mount,
1421 .kill_sb = gfs2_kill_sb,
1422 .owner = THIS_MODULE,
1424 MODULE_ALIAS_FS("gfs2");
1426 struct file_system_type gfs2meta_fs_type = {
1428 .fs_flags = FS_REQUIRES_DEV,
1429 .mount = gfs2_mount_meta,
1430 .owner = THIS_MODULE,
1432 MODULE_ALIAS_FS("gfs2meta");