2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/gfs2_ondisk.h>
14 #include <linux/bio.h>
15 #include <linux/posix_acl.h>
16 #include <linux/security.h>
32 struct workqueue_struct *gfs2_freeze_wq;
34 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
36 fs_err(gl->gl_name.ln_sbd,
37 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
39 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
40 bh->b_page->mapping, bh->b_page->flags);
41 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
42 gl->gl_name.ln_type, gl->gl_name.ln_number,
43 gfs2_glock2aspace(gl));
44 gfs2_lm_withdraw(gl->gl_name.ln_sbd, "AIL error\n");
48 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
50 * @fsync: set when called from fsync (not all buffers will be clean)
52 * None of the buffers should be dirty, locked, or pinned.
55 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
56 unsigned int nr_revokes)
58 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
59 struct list_head *head = &gl->gl_ail_list;
60 struct gfs2_bufdata *bd, *tmp;
61 struct buffer_head *bh;
62 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
65 spin_lock(&sdp->sd_ail_lock);
66 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
70 if (bh->b_state & b_state) {
73 gfs2_ail_error(gl, bh);
75 gfs2_trans_add_revoke(sdp, bd);
78 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
79 spin_unlock(&sdp->sd_ail_lock);
84 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
86 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
89 memset(&tr, 0, sizeof(tr));
90 INIT_LIST_HEAD(&tr.tr_buf);
91 INIT_LIST_HEAD(&tr.tr_databuf);
92 INIT_LIST_HEAD(&tr.tr_ail1_list);
93 INIT_LIST_HEAD(&tr.tr_ail2_list);
94 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
99 /* A shortened, inline version of gfs2_trans_begin()
100 * tr->alloced is not set since the transaction structure is
102 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
104 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
106 WARN_ON_ONCE(current->journal_info);
107 current->journal_info = &tr;
109 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
112 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
115 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
117 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
118 unsigned int revokes = atomic_read(&gl->gl_ail_count);
119 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
125 while (revokes > max_revokes)
126 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
128 ret = gfs2_trans_begin(sdp, 0, max_revokes);
131 __gfs2_ail_flush(gl, fsync, max_revokes);
133 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
137 * rgrp_go_sync - sync out the metadata for this glock
140 * Called when demoting or unlocking an EX glock. We must flush
141 * to disk all dirty buffers/pages relating to this glock, and must not
142 * return to caller to demote/unlock the glock until I/O is complete.
145 static void rgrp_go_sync(struct gfs2_glock *gl)
147 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
148 struct address_space *mapping = &sdp->sd_aspace;
149 struct gfs2_rgrpd *rgd;
152 spin_lock(&gl->gl_lockref.lock);
155 gfs2_rgrp_brelse(rgd);
156 spin_unlock(&gl->gl_lockref.lock);
158 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
160 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
162 gfs2_log_flush(sdp, gl, NORMAL_FLUSH);
163 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
164 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
165 mapping_set_error(mapping, error);
166 gfs2_ail_empty_gl(gl);
168 spin_lock(&gl->gl_lockref.lock);
171 gfs2_free_clones(rgd);
172 spin_unlock(&gl->gl_lockref.lock);
176 * rgrp_go_inval - invalidate the metadata for this glock
180 * We never used LM_ST_DEFERRED with resource groups, so that we
181 * should always see the metadata flag set here.
185 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
187 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
188 struct address_space *mapping = &sdp->sd_aspace;
189 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
192 gfs2_rgrp_brelse(rgd);
194 WARN_ON_ONCE(!(flags & DIO_METADATA));
195 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
196 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
199 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
202 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
204 struct gfs2_inode *ip;
206 spin_lock(&gl->gl_lockref.lock);
209 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
210 spin_unlock(&gl->gl_lockref.lock);
214 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
216 struct gfs2_rgrpd *rgd;
218 spin_lock(&gl->gl_lockref.lock);
220 spin_unlock(&gl->gl_lockref.lock);
225 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
230 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
231 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
235 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
236 * @gl: the glock protecting the inode
240 static void inode_go_sync(struct gfs2_glock *gl)
242 struct gfs2_inode *ip = gfs2_glock2inode(gl);
243 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
244 struct address_space *metamapping = gfs2_glock2aspace(gl);
248 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
249 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
250 inode_dio_wait(&ip->i_inode);
252 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
255 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
257 gfs2_log_flush(gl->gl_name.ln_sbd, gl, NORMAL_FLUSH);
258 filemap_fdatawrite(metamapping);
260 struct address_space *mapping = ip->i_inode.i_mapping;
261 filemap_fdatawrite(mapping);
262 error = filemap_fdatawait(mapping);
263 mapping_set_error(mapping, error);
265 error = filemap_fdatawait(metamapping);
266 mapping_set_error(metamapping, error);
267 gfs2_ail_empty_gl(gl);
269 * Writeback of the data mapping may cause the dirty flag to be set
270 * so we have to clear it again here.
272 smp_mb__before_atomic();
273 clear_bit(GLF_DIRTY, &gl->gl_flags);
276 gfs2_clear_glop_pending(ip);
280 * inode_go_inval - prepare a inode glock to be released
284 * Normally we invalidate everything, but if we are moving into
285 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
286 * can keep hold of the metadata, since it won't have changed.
290 static void inode_go_inval(struct gfs2_glock *gl, int flags)
292 struct gfs2_inode *ip = gfs2_glock2inode(gl);
294 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
296 if (flags & DIO_METADATA) {
297 struct address_space *mapping = gfs2_glock2aspace(gl);
298 truncate_inode_pages(mapping, 0);
300 set_bit(GIF_INVALID, &ip->i_flags);
301 forget_all_cached_acls(&ip->i_inode);
302 security_inode_invalidate_secctx(&ip->i_inode);
303 gfs2_dir_hash_inval(ip);
307 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
308 gfs2_log_flush(gl->gl_name.ln_sbd, NULL, NORMAL_FLUSH);
309 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
311 if (ip && S_ISREG(ip->i_inode.i_mode))
312 truncate_inode_pages(ip->i_inode.i_mapping, 0);
314 gfs2_clear_glop_pending(ip);
318 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
321 * Returns: 1 if it's ok
324 static int inode_go_demote_ok(const struct gfs2_glock *gl)
326 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
328 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
334 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
336 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
337 const struct gfs2_dinode *str = buf;
338 struct timespec atime;
341 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
343 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
344 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
345 ip->i_inode.i_rdev = 0;
346 switch (ip->i_inode.i_mode & S_IFMT) {
349 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
350 be32_to_cpu(str->di_minor));
354 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
355 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
356 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
357 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
358 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
359 atime.tv_sec = be64_to_cpu(str->di_atime);
360 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
361 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
362 ip->i_inode.i_atime = atime;
363 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
364 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
365 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
366 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
368 ip->i_goal = be64_to_cpu(str->di_goal_meta);
369 ip->i_generation = be64_to_cpu(str->di_generation);
371 ip->i_diskflags = be32_to_cpu(str->di_flags);
372 ip->i_eattr = be64_to_cpu(str->di_eattr);
373 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
374 gfs2_set_inode_flags(&ip->i_inode);
375 height = be16_to_cpu(str->di_height);
376 if (unlikely(height > sdp->sd_max_height))
378 ip->i_height = (u8)height;
380 depth = be16_to_cpu(str->di_depth);
381 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
383 ip->i_depth = (u8)depth;
384 ip->i_entries = be32_to_cpu(str->di_entries);
386 if (S_ISREG(ip->i_inode.i_mode))
387 gfs2_set_aops(&ip->i_inode);
391 gfs2_consist_inode(ip);
396 * gfs2_inode_refresh - Refresh the incore copy of the dinode
397 * @ip: The GFS2 inode
402 int gfs2_inode_refresh(struct gfs2_inode *ip)
404 struct buffer_head *dibh;
407 error = gfs2_meta_inode_buffer(ip, &dibh);
411 error = gfs2_dinode_in(ip, dibh->b_data);
413 clear_bit(GIF_INVALID, &ip->i_flags);
419 * inode_go_lock - operation done after an inode lock is locked by a process
426 static int inode_go_lock(struct gfs2_holder *gh)
428 struct gfs2_glock *gl = gh->gh_gl;
429 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
430 struct gfs2_inode *ip = gl->gl_object;
433 if (!ip || (gh->gh_flags & GL_SKIP))
436 if (test_bit(GIF_INVALID, &ip->i_flags)) {
437 error = gfs2_inode_refresh(ip);
442 if (gh->gh_state != LM_ST_DEFERRED)
443 inode_dio_wait(&ip->i_inode);
445 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
446 (gl->gl_state == LM_ST_EXCLUSIVE) &&
447 (gh->gh_state == LM_ST_EXCLUSIVE)) {
448 spin_lock(&sdp->sd_trunc_lock);
449 if (list_empty(&ip->i_trunc_list))
450 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
451 spin_unlock(&sdp->sd_trunc_lock);
452 wake_up(&sdp->sd_quota_wait);
460 * inode_go_dump - print information about an inode
466 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
468 const struct gfs2_inode *ip = gl->gl_object;
471 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
472 (unsigned long long)ip->i_no_formal_ino,
473 (unsigned long long)ip->i_no_addr,
474 IF2DT(ip->i_inode.i_mode), ip->i_flags,
475 (unsigned int)ip->i_diskflags,
476 (unsigned long long)i_size_read(&ip->i_inode));
480 * freeze_go_sync - promote/demote the freeze glock
482 * @state: the requested state
487 static void freeze_go_sync(struct gfs2_glock *gl)
490 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
492 if (gl->gl_state == LM_ST_SHARED &&
493 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
494 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
495 error = freeze_super(sdp->sd_vfs);
497 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
498 gfs2_assert_withdraw(sdp, 0);
500 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
501 gfs2_log_flush(sdp, NULL, FREEZE_FLUSH);
506 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
511 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
513 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
514 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
515 struct gfs2_glock *j_gl = ip->i_gl;
516 struct gfs2_log_header_host head;
519 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
520 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
522 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
525 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
528 /* Initialize some head of the log stuff */
529 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
530 sdp->sd_log_sequence = head.lh_sequence + 1;
531 gfs2_log_pointers_init(sdp, head.lh_blkno);
544 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
550 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
553 * gl_lockref.lock lock is held while calling this
555 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
557 struct gfs2_inode *ip = gl->gl_object;
558 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
560 if (!remote || sb_rdonly(sdp->sd_vfs))
563 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
564 gl->gl_state == LM_ST_SHARED && ip) {
565 gl->gl_lockref.count++;
566 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
567 gl->gl_lockref.count--;
571 const struct gfs2_glock_operations gfs2_meta_glops = {
572 .go_type = LM_TYPE_META,
575 const struct gfs2_glock_operations gfs2_inode_glops = {
576 .go_sync = inode_go_sync,
577 .go_inval = inode_go_inval,
578 .go_demote_ok = inode_go_demote_ok,
579 .go_lock = inode_go_lock,
580 .go_dump = inode_go_dump,
581 .go_type = LM_TYPE_INODE,
582 .go_flags = GLOF_ASPACE | GLOF_LRU,
585 const struct gfs2_glock_operations gfs2_rgrp_glops = {
586 .go_sync = rgrp_go_sync,
587 .go_inval = rgrp_go_inval,
588 .go_lock = gfs2_rgrp_go_lock,
589 .go_unlock = gfs2_rgrp_go_unlock,
590 .go_dump = gfs2_rgrp_dump,
591 .go_type = LM_TYPE_RGRP,
592 .go_flags = GLOF_LVB,
595 const struct gfs2_glock_operations gfs2_freeze_glops = {
596 .go_sync = freeze_go_sync,
597 .go_xmote_bh = freeze_go_xmote_bh,
598 .go_demote_ok = freeze_go_demote_ok,
599 .go_type = LM_TYPE_NONDISK,
602 const struct gfs2_glock_operations gfs2_iopen_glops = {
603 .go_type = LM_TYPE_IOPEN,
604 .go_callback = iopen_go_callback,
605 .go_flags = GLOF_LRU,
608 const struct gfs2_glock_operations gfs2_flock_glops = {
609 .go_type = LM_TYPE_FLOCK,
610 .go_flags = GLOF_LRU,
613 const struct gfs2_glock_operations gfs2_nondisk_glops = {
614 .go_type = LM_TYPE_NONDISK,
617 const struct gfs2_glock_operations gfs2_quota_glops = {
618 .go_type = LM_TYPE_QUOTA,
619 .go_flags = GLOF_LVB | GLOF_LRU,
622 const struct gfs2_glock_operations gfs2_journal_glops = {
623 .go_type = LM_TYPE_JOURNAL,
626 const struct gfs2_glock_operations *gfs2_glops_list[] = {
627 [LM_TYPE_META] = &gfs2_meta_glops,
628 [LM_TYPE_INODE] = &gfs2_inode_glops,
629 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
630 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
631 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
632 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
633 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
634 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,