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 #include <linux/spinlock.h>
8 #include <linux/completion.h>
9 #include <linux/buffer_head.h>
10 #include <linux/gfs2_ondisk.h>
11 #include <linux/bio.h>
12 #include <linux/posix_acl.h>
13 #include <linux/security.h>
30 struct workqueue_struct *gfs2_freeze_wq;
32 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
34 fs_err(gl->gl_name.ln_sbd,
35 "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page "
37 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
38 bh->b_page->mapping, bh->b_page->flags);
39 fs_err(gl->gl_name.ln_sbd, "AIL glock %u:%llu mapping %p\n",
40 gl->gl_name.ln_type, gl->gl_name.ln_number,
41 gfs2_glock2aspace(gl));
42 gfs2_lm_withdraw(gl->gl_name.ln_sbd, "AIL error\n");
46 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
48 * @fsync: set when called from fsync (not all buffers will be clean)
50 * None of the buffers should be dirty, locked, or pinned.
53 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
54 unsigned int nr_revokes)
56 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
57 struct list_head *head = &gl->gl_ail_list;
58 struct gfs2_bufdata *bd, *tmp;
59 struct buffer_head *bh;
60 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
63 spin_lock(&sdp->sd_ail_lock);
64 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
68 if (bh->b_state & b_state) {
71 gfs2_ail_error(gl, bh);
73 gfs2_trans_add_revoke(sdp, bd);
76 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
77 spin_unlock(&sdp->sd_ail_lock);
82 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
84 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
87 memset(&tr, 0, sizeof(tr));
88 INIT_LIST_HEAD(&tr.tr_buf);
89 INIT_LIST_HEAD(&tr.tr_databuf);
90 INIT_LIST_HEAD(&tr.tr_ail1_list);
91 INIT_LIST_HEAD(&tr.tr_ail2_list);
92 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
99 * We have nothing on the ail, but there could be revokes on
100 * the sdp revoke queue, in which case, we still want to flush
101 * the log and wait for it to finish.
103 * If the sdp revoke list is empty too, we might still have an
104 * io outstanding for writing revokes, so we should wait for
105 * it before returning.
107 * If none of these conditions are true, our revokes are all
108 * flushed and we can return.
111 have_revokes = !list_empty(&sdp->sd_log_revokes);
112 log_in_flight = atomic_read(&sdp->sd_log_in_flight);
113 gfs2_log_unlock(sdp);
121 /* A shortened, inline version of gfs2_trans_begin()
122 * tr->alloced is not set since the transaction structure is
124 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
126 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
128 WARN_ON_ONCE(current->journal_info);
129 current->journal_info = &tr;
131 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
135 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
136 GFS2_LFC_AIL_EMPTY_GL);
139 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
141 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
142 unsigned int revokes = atomic_read(&gl->gl_ail_count);
143 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
149 while (revokes > max_revokes)
150 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
152 ret = gfs2_trans_begin(sdp, 0, max_revokes);
155 __gfs2_ail_flush(gl, fsync, max_revokes);
157 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
162 * rgrp_go_sync - sync out the metadata for this glock
165 * Called when demoting or unlocking an EX glock. We must flush
166 * to disk all dirty buffers/pages relating to this glock, and must not
167 * return to caller to demote/unlock the glock until I/O is complete.
170 static void rgrp_go_sync(struct gfs2_glock *gl)
172 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
173 struct address_space *mapping = &sdp->sd_aspace;
174 struct gfs2_rgrpd *rgd;
177 spin_lock(&gl->gl_lockref.lock);
180 gfs2_rgrp_brelse(rgd);
181 spin_unlock(&gl->gl_lockref.lock);
183 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
185 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
187 gfs2_log_flush(sdp, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
188 GFS2_LFC_RGRP_GO_SYNC);
189 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
190 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
191 mapping_set_error(mapping, error);
192 gfs2_ail_empty_gl(gl);
194 spin_lock(&gl->gl_lockref.lock);
197 gfs2_free_clones(rgd);
198 spin_unlock(&gl->gl_lockref.lock);
202 * rgrp_go_inval - invalidate the metadata for this glock
206 * We never used LM_ST_DEFERRED with resource groups, so that we
207 * should always see the metadata flag set here.
211 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
213 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
214 struct address_space *mapping = &sdp->sd_aspace;
215 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
218 gfs2_rgrp_brelse(rgd);
220 WARN_ON_ONCE(!(flags & DIO_METADATA));
221 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
222 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
225 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
228 static struct gfs2_inode *gfs2_glock2inode(struct gfs2_glock *gl)
230 struct gfs2_inode *ip;
232 spin_lock(&gl->gl_lockref.lock);
235 set_bit(GIF_GLOP_PENDING, &ip->i_flags);
236 spin_unlock(&gl->gl_lockref.lock);
240 struct gfs2_rgrpd *gfs2_glock2rgrp(struct gfs2_glock *gl)
242 struct gfs2_rgrpd *rgd;
244 spin_lock(&gl->gl_lockref.lock);
246 spin_unlock(&gl->gl_lockref.lock);
251 static void gfs2_clear_glop_pending(struct gfs2_inode *ip)
256 clear_bit_unlock(GIF_GLOP_PENDING, &ip->i_flags);
257 wake_up_bit(&ip->i_flags, GIF_GLOP_PENDING);
261 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
262 * @gl: the glock protecting the inode
266 static void inode_go_sync(struct gfs2_glock *gl)
268 struct gfs2_inode *ip = gfs2_glock2inode(gl);
269 int isreg = ip && S_ISREG(ip->i_inode.i_mode);
270 struct address_space *metamapping = gfs2_glock2aspace(gl);
274 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
275 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
276 inode_dio_wait(&ip->i_inode);
278 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
281 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
283 gfs2_log_flush(gl->gl_name.ln_sbd, gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
284 GFS2_LFC_INODE_GO_SYNC);
285 filemap_fdatawrite(metamapping);
287 struct address_space *mapping = ip->i_inode.i_mapping;
288 filemap_fdatawrite(mapping);
289 error = filemap_fdatawait(mapping);
290 mapping_set_error(mapping, error);
292 error = filemap_fdatawait(metamapping);
293 mapping_set_error(metamapping, error);
294 gfs2_ail_empty_gl(gl);
296 * Writeback of the data mapping may cause the dirty flag to be set
297 * so we have to clear it again here.
299 smp_mb__before_atomic();
300 clear_bit(GLF_DIRTY, &gl->gl_flags);
303 gfs2_clear_glop_pending(ip);
307 * inode_go_inval - prepare a inode glock to be released
311 * Normally we invalidate everything, but if we are moving into
312 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
313 * can keep hold of the metadata, since it won't have changed.
317 static void inode_go_inval(struct gfs2_glock *gl, int flags)
319 struct gfs2_inode *ip = gfs2_glock2inode(gl);
321 gfs2_assert_withdraw(gl->gl_name.ln_sbd, !atomic_read(&gl->gl_ail_count));
323 if (flags & DIO_METADATA) {
324 struct address_space *mapping = gfs2_glock2aspace(gl);
325 truncate_inode_pages(mapping, 0);
327 set_bit(GIF_INVALID, &ip->i_flags);
328 forget_all_cached_acls(&ip->i_inode);
329 security_inode_invalidate_secctx(&ip->i_inode);
330 gfs2_dir_hash_inval(ip);
334 if (ip == GFS2_I(gl->gl_name.ln_sbd->sd_rindex)) {
335 gfs2_log_flush(gl->gl_name.ln_sbd, NULL,
336 GFS2_LOG_HEAD_FLUSH_NORMAL |
337 GFS2_LFC_INODE_GO_INVAL);
338 gl->gl_name.ln_sbd->sd_rindex_uptodate = 0;
340 if (ip && S_ISREG(ip->i_inode.i_mode))
341 truncate_inode_pages(ip->i_inode.i_mapping, 0);
343 gfs2_clear_glop_pending(ip);
347 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
350 * Returns: 1 if it's ok
353 static int inode_go_demote_ok(const struct gfs2_glock *gl)
355 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
357 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
363 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
365 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
366 const struct gfs2_dinode *str = buf;
367 struct timespec64 atime;
370 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
372 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
373 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
374 ip->i_inode.i_rdev = 0;
375 switch (ip->i_inode.i_mode & S_IFMT) {
378 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
379 be32_to_cpu(str->di_minor));
383 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
384 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
385 set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
386 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
387 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
388 atime.tv_sec = be64_to_cpu(str->di_atime);
389 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
390 if (timespec64_compare(&ip->i_inode.i_atime, &atime) < 0)
391 ip->i_inode.i_atime = atime;
392 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
393 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
394 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
395 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
397 ip->i_goal = be64_to_cpu(str->di_goal_meta);
398 ip->i_generation = be64_to_cpu(str->di_generation);
400 ip->i_diskflags = be32_to_cpu(str->di_flags);
401 ip->i_eattr = be64_to_cpu(str->di_eattr);
402 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
403 gfs2_set_inode_flags(&ip->i_inode);
404 height = be16_to_cpu(str->di_height);
405 if (unlikely(height > sdp->sd_max_height))
407 ip->i_height = (u8)height;
409 depth = be16_to_cpu(str->di_depth);
410 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
412 ip->i_depth = (u8)depth;
413 ip->i_entries = be32_to_cpu(str->di_entries);
415 if (gfs2_is_stuffed(ip) && ip->i_inode.i_size > gfs2_max_stuffed_size(ip))
418 if (S_ISREG(ip->i_inode.i_mode))
419 gfs2_set_aops(&ip->i_inode);
423 gfs2_consist_inode(ip);
428 * gfs2_inode_refresh - Refresh the incore copy of the dinode
429 * @ip: The GFS2 inode
434 int gfs2_inode_refresh(struct gfs2_inode *ip)
436 struct buffer_head *dibh;
439 error = gfs2_meta_inode_buffer(ip, &dibh);
443 error = gfs2_dinode_in(ip, dibh->b_data);
445 clear_bit(GIF_INVALID, &ip->i_flags);
451 * inode_go_lock - operation done after an inode lock is locked by a process
458 static int inode_go_lock(struct gfs2_holder *gh)
460 struct gfs2_glock *gl = gh->gh_gl;
461 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
462 struct gfs2_inode *ip = gl->gl_object;
465 if (!ip || (gh->gh_flags & GL_SKIP))
468 if (test_bit(GIF_INVALID, &ip->i_flags)) {
469 error = gfs2_inode_refresh(ip);
474 if (gh->gh_state != LM_ST_DEFERRED)
475 inode_dio_wait(&ip->i_inode);
477 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
478 (gl->gl_state == LM_ST_EXCLUSIVE) &&
479 (gh->gh_state == LM_ST_EXCLUSIVE)) {
480 spin_lock(&sdp->sd_trunc_lock);
481 if (list_empty(&ip->i_trunc_list))
482 list_add(&ip->i_trunc_list, &sdp->sd_trunc_list);
483 spin_unlock(&sdp->sd_trunc_lock);
484 wake_up(&sdp->sd_quota_wait);
492 * inode_go_dump - print information about an inode
495 * @fs_id_buf: file system id (may be empty)
499 static void inode_go_dump(struct seq_file *seq, struct gfs2_glock *gl,
500 const char *fs_id_buf)
502 struct gfs2_inode *ip = gl->gl_object;
503 struct inode *inode = &ip->i_inode;
504 unsigned long nrpages;
509 xa_lock_irq(&inode->i_data.i_pages);
510 nrpages = inode->i_data.nrpages;
511 xa_unlock_irq(&inode->i_data.i_pages);
513 gfs2_print_dbg(seq, "%s I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu "
514 "p:%lu\n", fs_id_buf,
515 (unsigned long long)ip->i_no_formal_ino,
516 (unsigned long long)ip->i_no_addr,
517 IF2DT(ip->i_inode.i_mode), ip->i_flags,
518 (unsigned int)ip->i_diskflags,
519 (unsigned long long)i_size_read(inode), nrpages);
523 * freeze_go_sync - promote/demote the freeze glock
525 * @state: the requested state
530 static void freeze_go_sync(struct gfs2_glock *gl)
533 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
535 if (gl->gl_state == LM_ST_SHARED &&
536 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
537 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
538 error = freeze_super(sdp->sd_vfs);
540 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
542 gfs2_assert_withdraw(sdp, 0);
544 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
545 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
546 GFS2_LFC_FREEZE_GO_SYNC);
551 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
556 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
558 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
559 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
560 struct gfs2_glock *j_gl = ip->i_gl;
561 struct gfs2_log_header_host head;
564 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
565 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
567 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
570 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
573 /* Initialize some head of the log stuff */
574 if (!test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
575 sdp->sd_log_sequence = head.lh_sequence + 1;
576 gfs2_log_pointers_init(sdp, head.lh_blkno);
589 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
595 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
598 * gl_lockref.lock lock is held while calling this
600 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
602 struct gfs2_inode *ip = gl->gl_object;
603 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
605 if (!remote || sb_rdonly(sdp->sd_vfs))
608 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
609 gl->gl_state == LM_ST_SHARED && ip) {
610 gl->gl_lockref.count++;
611 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
612 gl->gl_lockref.count--;
616 const struct gfs2_glock_operations gfs2_meta_glops = {
617 .go_type = LM_TYPE_META,
620 const struct gfs2_glock_operations gfs2_inode_glops = {
621 .go_sync = inode_go_sync,
622 .go_inval = inode_go_inval,
623 .go_demote_ok = inode_go_demote_ok,
624 .go_lock = inode_go_lock,
625 .go_dump = inode_go_dump,
626 .go_type = LM_TYPE_INODE,
627 .go_flags = GLOF_ASPACE | GLOF_LRU,
630 const struct gfs2_glock_operations gfs2_rgrp_glops = {
631 .go_sync = rgrp_go_sync,
632 .go_inval = rgrp_go_inval,
633 .go_lock = gfs2_rgrp_go_lock,
634 .go_unlock = gfs2_rgrp_go_unlock,
635 .go_dump = gfs2_rgrp_dump,
636 .go_type = LM_TYPE_RGRP,
637 .go_flags = GLOF_LVB,
640 const struct gfs2_glock_operations gfs2_freeze_glops = {
641 .go_sync = freeze_go_sync,
642 .go_xmote_bh = freeze_go_xmote_bh,
643 .go_demote_ok = freeze_go_demote_ok,
644 .go_type = LM_TYPE_NONDISK,
647 const struct gfs2_glock_operations gfs2_iopen_glops = {
648 .go_type = LM_TYPE_IOPEN,
649 .go_callback = iopen_go_callback,
650 .go_flags = GLOF_LRU,
653 const struct gfs2_glock_operations gfs2_flock_glops = {
654 .go_type = LM_TYPE_FLOCK,
655 .go_flags = GLOF_LRU,
658 const struct gfs2_glock_operations gfs2_nondisk_glops = {
659 .go_type = LM_TYPE_NONDISK,
662 const struct gfs2_glock_operations gfs2_quota_glops = {
663 .go_type = LM_TYPE_QUOTA,
664 .go_flags = GLOF_LVB | GLOF_LRU,
667 const struct gfs2_glock_operations gfs2_journal_glops = {
668 .go_type = LM_TYPE_JOURNAL,
671 const struct gfs2_glock_operations *gfs2_glops_list[] = {
672 [LM_TYPE_META] = &gfs2_meta_glops,
673 [LM_TYPE_INODE] = &gfs2_inode_glops,
674 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
675 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
676 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
677 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
678 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
679 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,