1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9 #include <linux/sched.h>
10 #include <linux/slab.h>
11 #include <linux/spinlock.h>
12 #include <linux/completion.h>
13 #include <linux/buffer_head.h>
14 #include <linux/blkdev.h>
15 #include <linux/kthread.h>
16 #include <linux/export.h>
17 #include <linux/namei.h>
18 #include <linux/mount.h>
19 #include <linux/gfs2_ondisk.h>
20 #include <linux/quotaops.h>
21 #include <linux/lockdep.h>
22 #include <linux/module.h>
23 #include <linux/backing-dev.h>
40 #include "trace_gfs2.h"
47 * gfs2_tune_init - Fill a gfs2_tune structure with default values
52 static void gfs2_tune_init(struct gfs2_tune *gt)
54 spin_lock_init(>->gt_spin);
56 gt->gt_quota_warn_period = 10;
57 gt->gt_quota_scale_num = 1;
58 gt->gt_quota_scale_den = 1;
59 gt->gt_new_files_jdata = 0;
60 gt->gt_max_readahead = BIT(18);
61 gt->gt_complain_secs = 10;
64 void free_sbd(struct gfs2_sbd *sdp)
67 free_percpu(sdp->sd_lkstats);
71 static struct gfs2_sbd *init_sbd(struct super_block *sb)
74 struct address_space *mapping;
76 sdp = kzalloc(sizeof(struct gfs2_sbd), GFP_KERNEL);
81 sdp->sd_lkstats = alloc_percpu(struct gfs2_pcpu_lkstats);
86 set_bit(SDF_NOJOURNALID, &sdp->sd_flags);
87 gfs2_tune_init(&sdp->sd_tune);
89 init_waitqueue_head(&sdp->sd_glock_wait);
90 init_waitqueue_head(&sdp->sd_async_glock_wait);
91 atomic_set(&sdp->sd_glock_disposal, 0);
92 init_completion(&sdp->sd_locking_init);
93 init_completion(&sdp->sd_wdack);
94 spin_lock_init(&sdp->sd_statfs_spin);
96 spin_lock_init(&sdp->sd_rindex_spin);
97 sdp->sd_rindex_tree.rb_node = NULL;
99 INIT_LIST_HEAD(&sdp->sd_jindex_list);
100 spin_lock_init(&sdp->sd_jindex_spin);
101 mutex_init(&sdp->sd_jindex_mutex);
102 init_completion(&sdp->sd_journal_ready);
104 INIT_LIST_HEAD(&sdp->sd_quota_list);
105 mutex_init(&sdp->sd_quota_mutex);
106 mutex_init(&sdp->sd_quota_sync_mutex);
107 init_waitqueue_head(&sdp->sd_quota_wait);
108 INIT_LIST_HEAD(&sdp->sd_trunc_list);
109 spin_lock_init(&sdp->sd_trunc_lock);
110 spin_lock_init(&sdp->sd_bitmap_lock);
112 mapping = &sdp->sd_aspace;
114 address_space_init_once(mapping);
115 mapping->a_ops = &gfs2_rgrp_aops;
116 mapping->host = sb->s_bdev->bd_inode;
118 mapping_set_gfp_mask(mapping, GFP_NOFS);
119 mapping->private_data = NULL;
120 mapping->writeback_index = 0;
122 spin_lock_init(&sdp->sd_log_lock);
123 atomic_set(&sdp->sd_log_pinned, 0);
124 INIT_LIST_HEAD(&sdp->sd_log_revokes);
125 INIT_LIST_HEAD(&sdp->sd_log_ordered);
126 spin_lock_init(&sdp->sd_ordered_lock);
128 init_waitqueue_head(&sdp->sd_log_waitq);
129 init_waitqueue_head(&sdp->sd_logd_waitq);
130 spin_lock_init(&sdp->sd_ail_lock);
131 INIT_LIST_HEAD(&sdp->sd_ail1_list);
132 INIT_LIST_HEAD(&sdp->sd_ail2_list);
134 init_rwsem(&sdp->sd_log_flush_lock);
135 atomic_set(&sdp->sd_log_in_flight, 0);
136 atomic_set(&sdp->sd_reserving_log, 0);
137 init_waitqueue_head(&sdp->sd_reserving_log_wait);
138 init_waitqueue_head(&sdp->sd_log_flush_wait);
139 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
140 mutex_init(&sdp->sd_freeze_mutex);
150 * gfs2_check_sb - Check superblock
151 * @sdp: the filesystem
152 * @sb: The superblock
153 * @silent: Don't print a message if the check fails
155 * Checks the version code of the FS is one that we understand how to
156 * read and that the sizes of the various on-disk structures have not
160 static int gfs2_check_sb(struct gfs2_sbd *sdp, int silent)
162 struct gfs2_sb_host *sb = &sdp->sd_sb;
164 if (sb->sb_magic != GFS2_MAGIC ||
165 sb->sb_type != GFS2_METATYPE_SB) {
167 pr_warn("not a GFS2 filesystem\n");
171 /* If format numbers match exactly, we're done. */
173 if (sb->sb_fs_format == GFS2_FORMAT_FS &&
174 sb->sb_multihost_format == GFS2_FORMAT_MULTI)
177 fs_warn(sdp, "Unknown on-disk format, unable to mount\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_max_dents_per_leaf = (sdp->sd_sb.sb_bsize -
346 sizeof(struct gfs2_leaf)) /
347 GFS2_MIN_DIRENT_SIZE;
351 static int init_names(struct gfs2_sbd *sdp, int silent)
356 proto = sdp->sd_args.ar_lockproto;
357 table = sdp->sd_args.ar_locktable;
359 /* Try to autodetect */
361 if (!proto[0] || !table[0]) {
362 error = gfs2_read_super(sdp, GFS2_SB_ADDR >> sdp->sd_fsb2bb_shift, silent);
367 proto = sdp->sd_sb.sb_lockproto;
369 table = sdp->sd_sb.sb_locktable;
373 table = sdp->sd_vfs->s_id;
375 strlcpy(sdp->sd_proto_name, proto, GFS2_FSNAME_LEN);
376 strlcpy(sdp->sd_table_name, table, GFS2_FSNAME_LEN);
378 table = sdp->sd_table_name;
379 while ((table = strchr(table, '/')))
385 static int init_locking(struct gfs2_sbd *sdp, struct gfs2_holder *mount_gh,
393 error = gfs2_glock_nq_num(sdp,
394 GFS2_MOUNT_LOCK, &gfs2_nondisk_glops,
395 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP | GL_NOCACHE,
398 fs_err(sdp, "can't acquire mount glock: %d\n", error);
402 error = gfs2_glock_nq_num(sdp,
403 GFS2_LIVE_LOCK, &gfs2_nondisk_glops,
405 LM_FLAG_NOEXP | GL_EXACT,
408 fs_err(sdp, "can't acquire live glock: %d\n", error);
412 error = gfs2_glock_get(sdp, GFS2_RENAME_LOCK, &gfs2_nondisk_glops,
413 CREATE, &sdp->sd_rename_gl);
415 fs_err(sdp, "can't create rename glock: %d\n", error);
419 error = gfs2_glock_get(sdp, GFS2_FREEZE_LOCK, &gfs2_freeze_glops,
420 CREATE, &sdp->sd_freeze_gl);
422 fs_err(sdp, "can't create transaction glock: %d\n", error);
429 gfs2_glock_put(sdp->sd_freeze_gl);
431 gfs2_glock_put(sdp->sd_rename_gl);
433 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
435 gfs2_glock_dq_uninit(mount_gh);
440 static int gfs2_lookup_root(struct super_block *sb, struct dentry **dptr,
441 u64 no_addr, const char *name)
443 struct gfs2_sbd *sdp = sb->s_fs_info;
444 struct dentry *dentry;
447 inode = gfs2_inode_lookup(sb, DT_DIR, no_addr, 0,
448 GFS2_BLKST_FREE /* ignore */);
450 fs_err(sdp, "can't read in %s inode: %ld\n", name, PTR_ERR(inode));
451 return PTR_ERR(inode);
453 dentry = d_make_root(inode);
455 fs_err(sdp, "can't alloc %s dentry\n", name);
462 static int init_sb(struct gfs2_sbd *sdp, int silent)
464 struct super_block *sb = sdp->sd_vfs;
465 struct gfs2_holder sb_gh;
469 ret = gfs2_glock_nq_num(sdp, GFS2_SB_LOCK, &gfs2_meta_glops,
470 LM_ST_SHARED, 0, &sb_gh);
472 fs_err(sdp, "can't acquire superblock glock: %d\n", ret);
476 ret = gfs2_read_sb(sdp, silent);
478 fs_err(sdp, "can't read superblock: %d\n", ret);
482 /* Set up the buffer cache and SB for real */
483 if (sdp->sd_sb.sb_bsize < bdev_logical_block_size(sb->s_bdev)) {
485 fs_err(sdp, "FS block size (%u) is too small for device "
487 sdp->sd_sb.sb_bsize, bdev_logical_block_size(sb->s_bdev));
490 if (sdp->sd_sb.sb_bsize > PAGE_SIZE) {
492 fs_err(sdp, "FS block size (%u) is too big for machine "
494 sdp->sd_sb.sb_bsize, (unsigned int)PAGE_SIZE);
497 sb_set_blocksize(sb, sdp->sd_sb.sb_bsize);
499 /* Get the root inode */
500 no_addr = sdp->sd_sb.sb_root_dir.no_addr;
501 ret = gfs2_lookup_root(sb, &sdp->sd_root_dir, no_addr, "root");
505 /* Get the master inode */
506 no_addr = sdp->sd_sb.sb_master_dir.no_addr;
507 ret = gfs2_lookup_root(sb, &sdp->sd_master_dir, no_addr, "master");
509 dput(sdp->sd_root_dir);
512 sb->s_root = dget(sdp->sd_args.ar_meta ? sdp->sd_master_dir : sdp->sd_root_dir);
514 gfs2_glock_dq_uninit(&sb_gh);
518 static void gfs2_others_may_mount(struct gfs2_sbd *sdp)
520 char *message = "FIRSTMOUNT=Done";
521 char *envp[] = { message, NULL };
523 fs_info(sdp, "first mount done, others may mount\n");
525 if (sdp->sd_lockstruct.ls_ops->lm_first_done)
526 sdp->sd_lockstruct.ls_ops->lm_first_done(sdp);
528 kobject_uevent_env(&sdp->sd_kobj, KOBJ_CHANGE, envp);
532 * gfs2_jindex_hold - Grab a lock on the jindex
533 * @sdp: The GFS2 superblock
534 * @ji_gh: the holder for the jindex glock
539 static int gfs2_jindex_hold(struct gfs2_sbd *sdp, struct gfs2_holder *ji_gh)
541 struct gfs2_inode *dip = GFS2_I(sdp->sd_jindex);
544 struct gfs2_jdesc *jd;
549 mutex_lock(&sdp->sd_jindex_mutex);
552 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, ji_gh);
556 name.len = sprintf(buf, "journal%u", sdp->sd_journals);
557 name.hash = gfs2_disk_hash(name.name, name.len);
559 error = gfs2_dir_check(sdp->sd_jindex, &name, NULL);
560 if (error == -ENOENT) {
565 gfs2_glock_dq_uninit(ji_gh);
571 jd = kzalloc(sizeof(struct gfs2_jdesc), GFP_KERNEL);
575 INIT_LIST_HEAD(&jd->extent_list);
576 INIT_LIST_HEAD(&jd->jd_revoke_list);
578 INIT_WORK(&jd->jd_work, gfs2_recover_func);
579 jd->jd_inode = gfs2_lookupi(sdp->sd_jindex, &name, 1);
580 if (IS_ERR_OR_NULL(jd->jd_inode)) {
584 error = PTR_ERR(jd->jd_inode);
589 spin_lock(&sdp->sd_jindex_spin);
590 jd->jd_jid = sdp->sd_journals++;
591 list_add_tail(&jd->jd_list, &sdp->sd_jindex_list);
592 spin_unlock(&sdp->sd_jindex_spin);
595 mutex_unlock(&sdp->sd_jindex_mutex);
601 * check_journal_clean - Make sure a journal is clean for a spectator mount
602 * @sdp: The GFS2 superblock
603 * @jd: The journal descriptor
605 * Returns: 0 if the journal is clean or locked, else an error
607 static int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd)
610 struct gfs2_holder j_gh;
611 struct gfs2_log_header_host head;
612 struct gfs2_inode *ip;
614 ip = GFS2_I(jd->jd_inode);
615 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
616 GL_EXACT | GL_NOCACHE, &j_gh);
618 fs_err(sdp, "Error locking journal for spectator mount.\n");
621 error = gfs2_jdesc_check(jd);
623 fs_err(sdp, "Error checking journal for spectator mount.\n");
626 error = gfs2_find_jhead(jd, &head, false);
628 fs_err(sdp, "Error parsing journal for spectator mount.\n");
631 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
633 fs_err(sdp, "jid=%u: Journal is dirty, so the first mounter "
634 "must not be a spectator.\n", jd->jd_jid);
638 gfs2_glock_dq_uninit(&j_gh);
642 static int init_journal(struct gfs2_sbd *sdp, int undo)
644 struct inode *master = d_inode(sdp->sd_master_dir);
645 struct gfs2_holder ji_gh;
646 struct gfs2_inode *ip;
655 sdp->sd_jindex = gfs2_lookup_simple(master, "jindex");
656 if (IS_ERR(sdp->sd_jindex)) {
657 fs_err(sdp, "can't lookup journal index: %d\n", error);
658 return PTR_ERR(sdp->sd_jindex);
661 /* Load in the journal index special file */
663 error = gfs2_jindex_hold(sdp, &ji_gh);
665 fs_err(sdp, "can't read journal index: %d\n", error);
670 if (!gfs2_jindex_size(sdp)) {
671 fs_err(sdp, "no journals!\n");
675 atomic_set(&sdp->sd_log_blks_needed, 0);
676 if (sdp->sd_args.ar_spectator) {
677 sdp->sd_jdesc = gfs2_jdesc_find(sdp, 0);
678 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
679 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
680 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
682 if (sdp->sd_lockstruct.ls_jid >= gfs2_jindex_size(sdp)) {
683 fs_err(sdp, "can't mount journal #%u\n",
684 sdp->sd_lockstruct.ls_jid);
685 fs_err(sdp, "there are only %u journals (0 - %u)\n",
686 gfs2_jindex_size(sdp),
687 gfs2_jindex_size(sdp) - 1);
690 sdp->sd_jdesc = gfs2_jdesc_find(sdp, sdp->sd_lockstruct.ls_jid);
692 error = gfs2_glock_nq_num(sdp, sdp->sd_lockstruct.ls_jid,
694 LM_ST_EXCLUSIVE, LM_FLAG_NOEXP,
695 &sdp->sd_journal_gh);
697 fs_err(sdp, "can't acquire journal glock: %d\n", error);
701 ip = GFS2_I(sdp->sd_jdesc->jd_inode);
702 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED,
703 LM_FLAG_NOEXP | GL_EXACT | GL_NOCACHE,
706 fs_err(sdp, "can't acquire journal inode glock: %d\n",
708 goto fail_journal_gh;
711 error = gfs2_jdesc_check(sdp->sd_jdesc);
713 fs_err(sdp, "my journal (%u) is bad: %d\n",
714 sdp->sd_jdesc->jd_jid, error);
717 atomic_set(&sdp->sd_log_blks_free, sdp->sd_jdesc->jd_blocks);
718 atomic_set(&sdp->sd_log_thresh1, 2*sdp->sd_jdesc->jd_blocks/5);
719 atomic_set(&sdp->sd_log_thresh2, 4*sdp->sd_jdesc->jd_blocks/5);
721 /* Map the extents for this journal's blocks */
722 gfs2_map_journal_extents(sdp, sdp->sd_jdesc);
724 trace_gfs2_log_blocks(sdp, atomic_read(&sdp->sd_log_blks_free));
726 if (sdp->sd_lockstruct.ls_first) {
728 for (x = 0; x < sdp->sd_journals; x++) {
729 struct gfs2_jdesc *jd = gfs2_jdesc_find(sdp, x);
731 if (sdp->sd_args.ar_spectator) {
732 error = check_journal_clean(sdp, jd);
737 error = gfs2_recover_journal(jd, true);
739 fs_err(sdp, "error recovering journal %u: %d\n",
745 gfs2_others_may_mount(sdp);
746 } else if (!sdp->sd_args.ar_spectator) {
747 error = gfs2_recover_journal(sdp->sd_jdesc, true);
749 fs_err(sdp, "error recovering my journal: %d\n", error);
754 sdp->sd_log_idle = 1;
755 set_bit(SDF_JOURNAL_CHECKED, &sdp->sd_flags);
756 gfs2_glock_dq_uninit(&ji_gh);
758 INIT_WORK(&sdp->sd_freeze_work, gfs2_freeze_func);
762 if (!sdp->sd_args.ar_spectator)
763 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
765 if (!sdp->sd_args.ar_spectator)
766 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
768 gfs2_jindex_free(sdp);
770 gfs2_glock_dq_uninit(&ji_gh);
772 iput(sdp->sd_jindex);
776 static struct lock_class_key gfs2_quota_imutex_key;
778 static int init_inodes(struct gfs2_sbd *sdp, int undo)
781 struct inode *master = d_inode(sdp->sd_master_dir);
786 error = init_journal(sdp, undo);
787 complete_all(&sdp->sd_journal_ready);
791 /* Read in the master statfs inode */
792 sdp->sd_statfs_inode = gfs2_lookup_simple(master, "statfs");
793 if (IS_ERR(sdp->sd_statfs_inode)) {
794 error = PTR_ERR(sdp->sd_statfs_inode);
795 fs_err(sdp, "can't read in statfs inode: %d\n", error);
799 /* Read in the resource index inode */
800 sdp->sd_rindex = gfs2_lookup_simple(master, "rindex");
801 if (IS_ERR(sdp->sd_rindex)) {
802 error = PTR_ERR(sdp->sd_rindex);
803 fs_err(sdp, "can't get resource index inode: %d\n", error);
806 sdp->sd_rindex_uptodate = 0;
808 /* Read in the quota inode */
809 sdp->sd_quota_inode = gfs2_lookup_simple(master, "quota");
810 if (IS_ERR(sdp->sd_quota_inode)) {
811 error = PTR_ERR(sdp->sd_quota_inode);
812 fs_err(sdp, "can't get quota file inode: %d\n", error);
816 * i_rwsem on quota files is special. Since this inode is hidden system
817 * file, we are safe to define locking ourselves.
819 lockdep_set_class(&sdp->sd_quota_inode->i_rwsem,
820 &gfs2_quota_imutex_key);
822 error = gfs2_rindex_update(sdp);
829 iput(sdp->sd_quota_inode);
831 gfs2_clear_rgrpd(sdp);
832 iput(sdp->sd_rindex);
834 iput(sdp->sd_statfs_inode);
836 init_journal(sdp, UNDO);
841 static int init_per_node(struct gfs2_sbd *sdp, int undo)
843 struct inode *pn = NULL;
846 struct gfs2_inode *ip;
847 struct inode *master = d_inode(sdp->sd_master_dir);
849 if (sdp->sd_args.ar_spectator)
855 pn = gfs2_lookup_simple(master, "per_node");
858 fs_err(sdp, "can't find per_node directory: %d\n", error);
862 sprintf(buf, "statfs_change%u", sdp->sd_jdesc->jd_jid);
863 sdp->sd_sc_inode = gfs2_lookup_simple(pn, buf);
864 if (IS_ERR(sdp->sd_sc_inode)) {
865 error = PTR_ERR(sdp->sd_sc_inode);
866 fs_err(sdp, "can't find local \"sc\" file: %d\n", error);
870 sprintf(buf, "quota_change%u", sdp->sd_jdesc->jd_jid);
871 sdp->sd_qc_inode = gfs2_lookup_simple(pn, buf);
872 if (IS_ERR(sdp->sd_qc_inode)) {
873 error = PTR_ERR(sdp->sd_qc_inode);
874 fs_err(sdp, "can't find local \"qc\" file: %d\n", error);
881 ip = GFS2_I(sdp->sd_sc_inode);
882 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
885 fs_err(sdp, "can't lock local \"sc\" file: %d\n", error);
889 ip = GFS2_I(sdp->sd_qc_inode);
890 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0,
893 fs_err(sdp, "can't lock local \"qc\" file: %d\n", error);
900 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
902 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
904 iput(sdp->sd_qc_inode);
906 iput(sdp->sd_sc_inode);
912 static const match_table_t nolock_tokens = {
913 { Opt_jid, "jid=%d\n", },
917 static const struct lm_lockops nolock_ops = {
918 .lm_proto_name = "lock_nolock",
919 .lm_put_lock = gfs2_glock_free,
920 .lm_tokens = &nolock_tokens,
924 * gfs2_lm_mount - mount a locking protocol
925 * @sdp: the filesystem
926 * @args: mount arguments
927 * @silent: if 1, don't complain if the FS isn't a GFS2 fs
932 static int gfs2_lm_mount(struct gfs2_sbd *sdp, int silent)
934 const struct lm_lockops *lm;
935 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
936 struct gfs2_args *args = &sdp->sd_args;
937 const char *proto = sdp->sd_proto_name;
938 const char *table = sdp->sd_table_name;
942 if (!strcmp("lock_nolock", proto)) {
944 sdp->sd_args.ar_localflocks = 1;
945 #ifdef CONFIG_GFS2_FS_LOCKING_DLM
946 } else if (!strcmp("lock_dlm", proto)) {
950 pr_info("can't find protocol %s\n", proto);
954 fs_info(sdp, "Trying to join cluster \"%s\", \"%s\"\n", proto, table);
959 for (options = args->ar_hostdata; (o = strsep(&options, ":")); ) {
960 substring_t tmp[MAX_OPT_ARGS];
966 token = match_token(o, *lm->lm_tokens, tmp);
969 ret = match_int(&tmp[0], &option);
970 if (ret || option < 0)
972 if (test_and_clear_bit(SDF_NOJOURNALID, &sdp->sd_flags))
977 /* Obsolete, but left for backward compat purposes */
980 ret = match_int(&tmp[0], &option);
981 if (ret || (option != 0 && option != 1))
983 ls->ls_first = option;
988 fs_info(sdp, "unknown hostdata (%s)\n", o);
993 if (lm->lm_mount == NULL) {
994 fs_info(sdp, "Now mounting FS...\n");
995 complete_all(&sdp->sd_locking_init);
998 ret = lm->lm_mount(sdp, table);
1000 fs_info(sdp, "Joined cluster. Now mounting FS...\n");
1001 complete_all(&sdp->sd_locking_init);
1005 void gfs2_lm_unmount(struct gfs2_sbd *sdp)
1007 const struct lm_lockops *lm = sdp->sd_lockstruct.ls_ops;
1008 if (likely(!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) &&
1010 lm->lm_unmount(sdp);
1013 static int wait_on_journal(struct gfs2_sbd *sdp)
1015 if (sdp->sd_lockstruct.ls_ops->lm_mount == NULL)
1018 return wait_on_bit(&sdp->sd_flags, SDF_NOJOURNALID, TASK_INTERRUPTIBLE)
1022 void gfs2_online_uevent(struct gfs2_sbd *sdp)
1024 struct super_block *sb = sdp->sd_vfs;
1027 char *envp[] = { ro, spectator, NULL };
1028 sprintf(ro, "RDONLY=%d", sb_rdonly(sb));
1029 sprintf(spectator, "SPECTATOR=%d", sdp->sd_args.ar_spectator ? 1 : 0);
1030 kobject_uevent_env(&sdp->sd_kobj, KOBJ_ONLINE, envp);
1034 * fill_super - Read in superblock
1035 * @sb: The VFS superblock
1036 * @data: Mount options
1037 * @silent: Don't complain if it's not a GFS2 filesystem
1042 static int fill_super(struct super_block *sb, struct gfs2_args *args, int silent)
1044 struct gfs2_sbd *sdp;
1045 struct gfs2_holder mount_gh;
1050 pr_warn("can't alloc struct gfs2_sbd\n");
1053 sdp->sd_args = *args;
1055 if (sdp->sd_args.ar_spectator) {
1056 sb->s_flags |= SB_RDONLY;
1057 set_bit(SDF_RORECOVERY, &sdp->sd_flags);
1059 if (sdp->sd_args.ar_posix_acl)
1060 sb->s_flags |= SB_POSIXACL;
1061 if (sdp->sd_args.ar_nobarrier)
1062 set_bit(SDF_NOBARRIERS, &sdp->sd_flags);
1064 sb->s_flags |= SB_NOSEC;
1065 sb->s_magic = GFS2_MAGIC;
1066 sb->s_op = &gfs2_super_ops;
1067 sb->s_d_op = &gfs2_dops;
1068 sb->s_export_op = &gfs2_export_ops;
1069 sb->s_xattr = gfs2_xattr_handlers;
1070 sb->s_qcop = &gfs2_quotactl_ops;
1071 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
1072 sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
1073 sb->s_time_gran = 1;
1074 sb->s_maxbytes = MAX_LFS_FILESIZE;
1076 /* Set up the buffer cache and fill in some fake block size values
1077 to allow us to read-in the on-disk superblock. */
1078 sdp->sd_sb.sb_bsize = sb_min_blocksize(sb, GFS2_BASIC_BLOCK);
1079 sdp->sd_sb.sb_bsize_shift = sb->s_blocksize_bits;
1080 sdp->sd_fsb2bb_shift = sdp->sd_sb.sb_bsize_shift -
1081 GFS2_BASIC_BLOCK_SHIFT;
1082 sdp->sd_fsb2bb = BIT(sdp->sd_fsb2bb_shift);
1084 sdp->sd_tune.gt_logd_secs = sdp->sd_args.ar_commit;
1085 sdp->sd_tune.gt_quota_quantum = sdp->sd_args.ar_quota_quantum;
1086 if (sdp->sd_args.ar_statfs_quantum) {
1087 sdp->sd_tune.gt_statfs_slow = 0;
1088 sdp->sd_tune.gt_statfs_quantum = sdp->sd_args.ar_statfs_quantum;
1090 sdp->sd_tune.gt_statfs_slow = 1;
1091 sdp->sd_tune.gt_statfs_quantum = 30;
1094 error = init_names(sdp, silent);
1096 /* In this case, we haven't initialized sysfs, so we have to
1097 manually free the sdp. */
1099 sb->s_fs_info = NULL;
1103 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s", sdp->sd_table_name);
1105 error = gfs2_sys_fs_add(sdp);
1107 * If we hit an error here, gfs2_sys_fs_add will have called function
1108 * kobject_put which causes the sysfs usage count to go to zero, which
1109 * causes sysfs to call function gfs2_sbd_release, which frees sdp.
1110 * Subsequent error paths here will call gfs2_sys_fs_del, which also
1111 * kobject_put to free sdp.
1116 gfs2_create_debugfs_file(sdp);
1118 error = gfs2_lm_mount(sdp, silent);
1122 error = init_locking(sdp, &mount_gh, DO);
1126 error = init_sb(sdp, silent);
1130 error = wait_on_journal(sdp);
1135 * If user space has failed to join the cluster or some similar
1136 * failure has occurred, then the journal id will contain a
1137 * negative (error) number. This will then be returned to the
1138 * caller (of the mount syscall). We do this even for spectator
1139 * mounts (which just write a jid of 0 to indicate "ok" even though
1140 * the jid is unused in the spectator case)
1142 if (sdp->sd_lockstruct.ls_jid < 0) {
1143 error = sdp->sd_lockstruct.ls_jid;
1144 sdp->sd_lockstruct.ls_jid = 0;
1148 if (sdp->sd_args.ar_spectator)
1149 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.s",
1150 sdp->sd_table_name);
1152 snprintf(sdp->sd_fsname, sizeof(sdp->sd_fsname), "%s.%u",
1153 sdp->sd_table_name, sdp->sd_lockstruct.ls_jid);
1155 error = init_inodes(sdp, DO);
1159 error = init_per_node(sdp, DO);
1163 error = gfs2_statfs_init(sdp);
1165 fs_err(sdp, "can't initialize statfs subsystem: %d\n", error);
1169 if (!sb_rdonly(sb)) {
1170 error = gfs2_make_fs_rw(sdp);
1172 fs_err(sdp, "can't make FS RW: %d\n", error);
1177 gfs2_glock_dq_uninit(&mount_gh);
1178 gfs2_online_uevent(sdp);
1182 init_per_node(sdp, UNDO);
1184 init_inodes(sdp, UNDO);
1186 if (sdp->sd_root_dir)
1187 dput(sdp->sd_root_dir);
1188 if (sdp->sd_master_dir)
1189 dput(sdp->sd_master_dir);
1194 init_locking(sdp, &mount_gh, UNDO);
1196 complete_all(&sdp->sd_journal_ready);
1197 gfs2_gl_hash_clear(sdp);
1198 gfs2_lm_unmount(sdp);
1200 gfs2_delete_debugfs_file(sdp);
1201 /* gfs2_sys_fs_del must be the last thing we do, since it causes
1202 * sysfs to call function gfs2_sbd_release, which frees sdp. */
1203 gfs2_sys_fs_del(sdp);
1204 sb->s_fs_info = NULL;
1208 static int set_gfs2_super(struct super_block *s, void *data)
1211 s->s_dev = s->s_bdev->bd_dev;
1212 s->s_bdi = bdi_get(s->s_bdev->bd_bdi);
1216 static int test_gfs2_super(struct super_block *s, void *ptr)
1218 struct block_device *bdev = ptr;
1219 return (bdev == s->s_bdev);
1223 * gfs2_mount - Get the GFS2 superblock
1224 * @fs_type: The GFS2 filesystem type
1225 * @flags: Mount flags
1226 * @dev_name: The name of the device
1227 * @data: The mount arguments
1229 * Q. Why not use get_sb_bdev() ?
1230 * A. We need to select one of two root directories to mount, independent
1231 * of whether this is the initial, or subsequent, mount of this sb
1233 * Returns: 0 or -ve on error
1236 static struct dentry *gfs2_mount(struct file_system_type *fs_type, int flags,
1237 const char *dev_name, void *data)
1239 struct block_device *bdev;
1240 struct super_block *s;
1241 fmode_t mode = FMODE_READ | FMODE_EXCL;
1243 struct gfs2_args args;
1244 struct gfs2_sbd *sdp;
1246 if (!(flags & SB_RDONLY))
1247 mode |= FMODE_WRITE;
1249 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
1251 return ERR_CAST(bdev);
1254 * once the super is inserted into the list by sget, s_umount
1255 * will protect the lockfs code from trying to start a snapshot
1256 * while we are mounting
1258 mutex_lock(&bdev->bd_fsfreeze_mutex);
1259 if (bdev->bd_fsfreeze_count > 0) {
1260 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1264 s = sget(fs_type, test_gfs2_super, set_gfs2_super, flags, bdev);
1265 mutex_unlock(&bdev->bd_fsfreeze_mutex);
1272 * s_umount nests inside bd_mutex during
1273 * __invalidate_device(). blkdev_put() acquires
1274 * bd_mutex and can't be called under s_umount. Drop
1275 * s_umount temporarily. This is safe as we're
1276 * holding an active reference.
1278 up_write(&s->s_umount);
1279 blkdev_put(bdev, mode);
1280 down_write(&s->s_umount);
1282 /* s_mode must be set before deactivate_locked_super calls */
1286 memset(&args, 0, sizeof(args));
1287 args.ar_quota = GFS2_QUOTA_DEFAULT;
1288 args.ar_data = GFS2_DATA_DEFAULT;
1289 args.ar_commit = 30;
1290 args.ar_statfs_quantum = 30;
1291 args.ar_quota_quantum = 60;
1292 args.ar_errors = GFS2_ERRORS_DEFAULT;
1294 error = gfs2_mount_args(&args, data);
1296 pr_warn("can't parse mount arguments\n");
1302 if ((flags ^ s->s_flags) & SB_RDONLY)
1305 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev);
1306 sb_set_blocksize(s, block_size(bdev));
1307 error = fill_super(s, &args, flags & SB_SILENT ? 1 : 0);
1310 s->s_flags |= SB_ACTIVE;
1316 return dget(sdp->sd_master_dir);
1318 return dget(sdp->sd_root_dir);
1321 deactivate_locked_super(s);
1322 return ERR_PTR(error);
1324 blkdev_put(bdev, mode);
1325 return ERR_PTR(error);
1328 static int set_meta_super(struct super_block *s, void *ptr)
1333 static struct dentry *gfs2_mount_meta(struct file_system_type *fs_type,
1334 int flags, const char *dev_name, void *data)
1336 struct super_block *s;
1337 struct gfs2_sbd *sdp;
1341 if (!dev_name || !*dev_name)
1342 return ERR_PTR(-EINVAL);
1344 error = kern_path(dev_name, LOOKUP_FOLLOW, &path);
1346 pr_warn("path_lookup on %s returned error %d\n",
1348 return ERR_PTR(error);
1350 s = sget(&gfs2_fs_type, test_gfs2_super, set_meta_super, flags,
1351 path.dentry->d_sb->s_bdev);
1354 pr_warn("gfs2 mount does not exist\n");
1357 if ((flags ^ s->s_flags) & SB_RDONLY) {
1358 deactivate_locked_super(s);
1359 return ERR_PTR(-EBUSY);
1362 return dget(sdp->sd_master_dir);
1365 static void gfs2_kill_sb(struct super_block *sb)
1367 struct gfs2_sbd *sdp = sb->s_fs_info;
1370 kill_block_super(sb);
1374 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SYNC | GFS2_LFC_KILL_SB);
1375 dput(sdp->sd_root_dir);
1376 dput(sdp->sd_master_dir);
1377 sdp->sd_root_dir = NULL;
1378 sdp->sd_master_dir = NULL;
1379 shrink_dcache_sb(sb);
1380 kill_block_super(sb);
1383 struct file_system_type gfs2_fs_type = {
1385 .fs_flags = FS_REQUIRES_DEV,
1386 .mount = gfs2_mount,
1387 .kill_sb = gfs2_kill_sb,
1388 .owner = THIS_MODULE,
1390 MODULE_ALIAS_FS("gfs2");
1392 struct file_system_type gfs2meta_fs_type = {
1394 .fs_flags = FS_REQUIRES_DEV,
1395 .mount = gfs2_mount_meta,
1396 .owner = THIS_MODULE,
1398 MODULE_ALIAS_FS("gfs2meta");