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 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
97 /* A shortened, inline version of gfs2_trans_begin()
98 * tr->alloced is not set since the transaction structure is
100 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
102 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
104 WARN_ON_ONCE(current->journal_info);
105 current->journal_info = &tr;
107 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
110 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
111 GFS2_LFC_AIL_EMPTY_GL);
114 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
116 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
117 unsigned int revokes = atomic_read(&gl->gl_ail_count);
118 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
124 while (revokes > max_revokes)
125 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
127 ret = gfs2_trans_begin(sdp, 0, max_revokes);
130 __gfs2_ail_flush(gl, fsync, max_revokes);
132 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
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, GFS2_LOG_HEAD_FLUSH_NORMAL |
163 GFS2_LFC_RGRP_GO_SYNC);
164 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
165 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
166 mapping_set_error(mapping, error);
167 gfs2_ail_empty_gl(gl);
169 spin_lock(&gl->gl_lockref.lock);
172 gfs2_free_clones(rgd);
173 spin_unlock(&gl->gl_lockref.lock);
177 * rgrp_go_inval - invalidate the metadata for this glock
181 * We never used LM_ST_DEFERRED with resource groups, so that we
182 * should always see the metadata flag set here.
186 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
188 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
189 struct address_space *mapping = &sdp->sd_aspace;
190 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
193 gfs2_rgrp_brelse(rgd);
195 WARN_ON_ONCE(!(flags & DIO_METADATA));
196 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
197 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
200 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
203 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
205 struct gfs2_inode *ip;
207 spin_lock(&gl->gl_lockref.lock);
210 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
211 spin_unlock(&gl->gl_lockref.lock);
215 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
217 struct gfs2_rgrpd *rgd;
219 spin_lock(&gl->gl_lockref.lock);
221 spin_unlock(&gl->gl_lockref.lock);
226 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
231 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
232 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
236 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
237 * @gl: the glock protecting the inode
241 static void inode_go_sync(struct gfs2_glock *gl)
243 struct gfs2_inode *ip = gfs2_glock2inode(gl);
244 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
245 struct address_space *metamapping = gfs2_glock2aspace(gl);
249 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
250 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
251 inode_dio_wait(&ip->i_inode);
253 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
256 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
258 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
259 GFS2_LFC_INODE_GO_SYNC);
260 filemap_fdatawrite(metamapping);
262 struct address_space *mapping = ip->i_inode.i_mapping;
263 filemap_fdatawrite(mapping);
264 error = filemap_fdatawait(mapping);
265 mapping_set_error(mapping, error);
267 error = filemap_fdatawait(metamapping);
268 mapping_set_error(metamapping, error);
269 gfs2_ail_empty_gl(gl);
271 * Writeback of the data mapping may cause the dirty flag to be set
272 * so we have to clear it again here.
274 smp_mb__before_atomic();
275 clear_bit(GLF_DIRTY, &gl->gl_flags);
278 gfs2_clear_glop_pending(ip);
282 * inode_go_inval - prepare a inode glock to be released
286 * Normally we invalidate everything, but if we are moving into
287 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
288 * can keep hold of the metadata, since it won't have changed.
292 static void inode_go_inval(struct gfs2_glock *gl, int flags)
294 struct gfs2_inode *ip = gfs2_glock2inode(gl);
296 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
298 if (flags & DIO_METADATA) {
299 struct address_space *mapping = gfs2_glock2aspace(gl);
300 truncate_inode_pages(mapping, 0);
302 set_bit(GIF_INVALID, &ip->i_flags);
303 forget_all_cached_acls(&ip->i_inode);
304 security_inode_invalidate_secctx(&ip->i_inode);
305 gfs2_dir_hash_inval(ip);
309 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
310 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
311 GFS2_LOG_HEAD_FLUSH_NORMAL |
312 GFS2_LFC_INODE_GO_INVAL);
313 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
315 if (ip && S_ISREG(ip->i_inode.i_mode))
316 truncate_inode_pages(ip->i_inode.i_mapping, 0);
318 gfs2_clear_glop_pending(ip);
322 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
325 * Returns: 1 if it's ok
328 static int inode_go_demote_ok(const struct gfs2_glock *gl)
330 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
332 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
338 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
340 const struct gfs2_dinode *str = buf;
341 struct timespec64 atime;
344 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
346 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
347 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
348 ip->i_inode.i_rdev = 0;
349 switch (ip->i_inode.i_mode & S_IFMT) {
352 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
353 be32_to_cpu(str->di_minor));
357 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
358 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
359 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
360 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
361 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
362 atime.tv_sec = be64_to_cpu(str->di_atime);
363 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
364 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
365 ip->i_inode.i_atime = atime;
366 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
367 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
368 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
369 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
371 ip->i_goal = be64_to_cpu(str->di_goal_meta);
372 ip->i_generation = be64_to_cpu(str->di_generation);
374 ip->i_diskflags = be32_to_cpu(str->di_flags);
375 ip->i_eattr = be64_to_cpu(str->di_eattr);
376 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
377 gfs2_set_inode_flags(&ip->i_inode);
378 height = be16_to_cpu(str->di_height);
379 if (unlikely(height > GFS2_MAX_META_HEIGHT))
381 ip->i_height = (u8)height;
383 depth = be16_to_cpu(str->di_depth);
384 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
386 ip->i_depth = (u8)depth;
387 ip->i_entries = be32_to_cpu(str->di_entries);
389 if (S_ISREG(ip->i_inode.i_mode))
390 gfs2_set_aops(&ip->i_inode);
394 gfs2_consist_inode(ip);
399 * gfs2_inode_refresh - Refresh the incore copy of the dinode
400 * @ip: The GFS2 inode
405 int gfs2_inode_refresh(struct gfs2_inode *ip)
407 struct buffer_head *dibh;
410 error = gfs2_meta_inode_buffer(ip, &dibh);
414 error = gfs2_dinode_in(ip, dibh->b_data);
416 clear_bit(GIF_INVALID, &ip->i_flags);
422 * inode_go_lock - operation done after an inode lock is locked by a process
429 static int inode_go_lock(struct gfs2_holder *gh)
431 struct gfs2_glock *gl = gh->gh_gl;
432 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
433 struct gfs2_inode *ip = gl->gl_object;
436 if (!ip || (gh->gh_flags & GL_SKIP))
439 if (test_bit(GIF_INVALID, &ip->i_flags)) {
440 error = gfs2_inode_refresh(ip);
445 if (gh->gh_state != LM_ST_DEFERRED)
446 inode_dio_wait(&ip->i_inode);
448 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
449 (gl->gl_state == LM_ST_EXCLUSIVE) &&
450 (gh->gh_state == LM_ST_EXCLUSIVE)) {
451 spin_lock(&sdp->sd_trunc_lock);
452 if (list_empty(&ip->i_trunc_list))
453 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
454 spin_unlock(&sdp->sd_trunc_lock);
455 wake_up(&sdp->sd_quota_wait);
463 * inode_go_dump - print information about an inode
469 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl)
471 struct gfs2_inode *ip = gl->gl_object;
472 struct inode *inode = &ip->i_inode;
473 unsigned long nrpages;
478 xa_lock_irq(&inode->i_data.i_pages);
479 nrpages = inode->i_data.nrpages;
480 xa_unlock_irq(&inode->i_data.i_pages);
482 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu p:%lu\n",
483 (unsigned long long)ip->i_no_formal_ino,
484 (unsigned long long)ip->i_no_addr,
485 IF2DT(ip->i_inode.i_mode), ip->i_flags,
486 (unsigned int)ip->i_diskflags,
487 (unsigned long long)i_size_read(inode), nrpages);
491 * freeze_go_sync - promote/demote the freeze glock
493 * @state: the requested state
498 static void freeze_go_sync(struct gfs2_glock *gl)
501 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
503 if (gl->gl_state == LM_ST_SHARED &&
504 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
505 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
506 error = freeze_super(sdp->sd_vfs);
508 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
509 gfs2_assert_withdraw(sdp, 0);
511 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
512 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
513 GFS2_LFC_FREEZE_GO_SYNC);
518 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
523 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
525 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
526 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
527 struct gfs2_glock *j_gl = ip->i_gl;
528 struct gfs2_log_header_host head;
531 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
532 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
534 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
537 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
540 /* Initialize some head of the log stuff */
541 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
542 sdp->sd_log_sequence = head.lh_sequence + 1;
543 gfs2_log_pointers_init(sdp, head.lh_blkno);
556 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
562 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
565 * gl_lockref.lock lock is held while calling this
567 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
569 struct gfs2_inode *ip = gl->gl_object;
570 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
572 if (!remote || sb_rdonly(sdp->sd_vfs))
575 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
576 gl->gl_state == LM_ST_SHARED && ip) {
577 gl->gl_lockref.count++;
578 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
579 gl->gl_lockref.count--;
583 const struct gfs2_glock_operations gfs2_meta_glops = {
584 .go_type = LM_TYPE_META,
587 const struct gfs2_glock_operations gfs2_inode_glops = {
588 .go_sync = inode_go_sync,
589 .go_inval = inode_go_inval,
590 .go_demote_ok = inode_go_demote_ok,
591 .go_lock = inode_go_lock,
592 .go_dump = inode_go_dump,
593 .go_type = LM_TYPE_INODE,
594 .go_flags = GLOF_ASPACE | GLOF_LRU,
597 const struct gfs2_glock_operations gfs2_rgrp_glops = {
598 .go_sync = rgrp_go_sync,
599 .go_inval = rgrp_go_inval,
600 .go_lock = gfs2_rgrp_go_lock,
601 .go_unlock = gfs2_rgrp_go_unlock,
602 .go_dump = gfs2_rgrp_dump,
603 .go_type = LM_TYPE_RGRP,
604 .go_flags = GLOF_LVB,
607 const struct gfs2_glock_operations gfs2_freeze_glops = {
608 .go_sync = freeze_go_sync,
609 .go_xmote_bh = freeze_go_xmote_bh,
610 .go_demote_ok = freeze_go_demote_ok,
611 .go_type = LM_TYPE_NONDISK,
614 const struct gfs2_glock_operations gfs2_iopen_glops = {
615 .go_type = LM_TYPE_IOPEN,
616 .go_callback = iopen_go_callback,
617 .go_flags = GLOF_LRU,
620 const struct gfs2_glock_operations gfs2_flock_glops = {
621 .go_type = LM_TYPE_FLOCK,
622 .go_flags = GLOF_LRU,
625 const struct gfs2_glock_operations gfs2_nondisk_glops = {
626 .go_type = LM_TYPE_NONDISK,
629 const struct gfs2_glock_operations gfs2_quota_glops = {
630 .go_type = LM_TYPE_QUOTA,
631 .go_flags = GLOF_LVB | GLOF_LRU,
634 const struct gfs2_glock_operations gfs2_journal_glops = {
635 .go_type = LM_TYPE_JOURNAL,
638 const struct gfs2_glock_operations *gfs2_glops_list[] = {
639 [LM_TYPE_META] = &gfs2_meta_glops,
640 [LM_TYPE_INODE] = &gfs2_inode_glops,
641 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
642 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
643 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
644 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
645 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
646 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,