perf/x86: Add a few more comments
[linux-2.6/btrfs-unstable.git] / fs / gfs2 / glops.c
blob3bf0631b5d56d8a76a17b9f70e3fa1cc4b9ee2f0
1 /*
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.
8 */
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>
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "inode.h"
23 #include "log.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "util.h"
28 #include "trans.h"
29 #include "dir.h"
31 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
33 fs_err(gl->gl_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
34 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
35 bh->b_page->mapping, bh->b_page->flags);
36 fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
37 gl->gl_name.ln_type, gl->gl_name.ln_number,
38 gfs2_glock2aspace(gl));
39 gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
42 /**
43 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
44 * @gl: the glock
45 * @fsync: set when called from fsync (not all buffers will be clean)
47 * None of the buffers should be dirty, locked, or pinned.
50 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
51 unsigned int nr_revokes)
53 struct gfs2_sbd *sdp = gl->gl_sbd;
54 struct list_head *head = &gl->gl_ail_list;
55 struct gfs2_bufdata *bd, *tmp;
56 struct buffer_head *bh;
57 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
59 gfs2_log_lock(sdp);
60 spin_lock(&sdp->sd_ail_lock);
61 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
62 if (nr_revokes == 0)
63 break;
64 bh = bd->bd_bh;
65 if (bh->b_state & b_state) {
66 if (fsync)
67 continue;
68 gfs2_ail_error(gl, bh);
70 gfs2_trans_add_revoke(sdp, bd);
71 nr_revokes--;
73 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
74 spin_unlock(&sdp->sd_ail_lock);
75 gfs2_log_unlock(sdp);
79 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
81 struct gfs2_sbd *sdp = gl->gl_sbd;
82 struct gfs2_trans tr;
84 memset(&tr, 0, sizeof(tr));
85 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
87 if (!tr.tr_revokes)
88 return;
90 /* A shortened, inline version of gfs2_trans_begin() */
91 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
92 tr.tr_ip = (unsigned long)__builtin_return_address(0);
93 sb_start_intwrite(sdp->sd_vfs);
94 gfs2_log_reserve(sdp, tr.tr_reserved);
95 WARN_ON_ONCE(current->journal_info);
96 current->journal_info = &tr;
98 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
100 gfs2_trans_end(sdp);
101 gfs2_log_flush(sdp, NULL);
104 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
106 struct gfs2_sbd *sdp = gl->gl_sbd;
107 unsigned int revokes = atomic_read(&gl->gl_ail_count);
108 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
109 int ret;
111 if (!revokes)
112 return;
114 while (revokes > max_revokes)
115 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
117 ret = gfs2_trans_begin(sdp, 0, max_revokes);
118 if (ret)
119 return;
120 __gfs2_ail_flush(gl, fsync, max_revokes);
121 gfs2_trans_end(sdp);
122 gfs2_log_flush(sdp, NULL);
126 * rgrp_go_sync - sync out the metadata for this glock
127 * @gl: the glock
129 * Called when demoting or unlocking an EX glock. We must flush
130 * to disk all dirty buffers/pages relating to this glock, and must not
131 * not return to caller to demote/unlock the glock until I/O is complete.
134 static void rgrp_go_sync(struct gfs2_glock *gl)
136 struct gfs2_sbd *sdp = gl->gl_sbd;
137 struct address_space *mapping = &sdp->sd_aspace;
138 struct gfs2_rgrpd *rgd;
139 int error;
141 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
142 return;
143 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
145 gfs2_log_flush(sdp, gl);
146 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
147 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
148 mapping_set_error(mapping, error);
149 gfs2_ail_empty_gl(gl);
151 spin_lock(&gl->gl_spin);
152 rgd = gl->gl_object;
153 if (rgd)
154 gfs2_free_clones(rgd);
155 spin_unlock(&gl->gl_spin);
159 * rgrp_go_inval - invalidate the metadata for this glock
160 * @gl: the glock
161 * @flags:
163 * We never used LM_ST_DEFERRED with resource groups, so that we
164 * should always see the metadata flag set here.
168 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
170 struct gfs2_sbd *sdp = gl->gl_sbd;
171 struct address_space *mapping = &sdp->sd_aspace;
173 WARN_ON_ONCE(!(flags & DIO_METADATA));
174 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
175 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
177 if (gl->gl_object) {
178 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
179 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
184 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
185 * @gl: the glock protecting the inode
189 static void inode_go_sync(struct gfs2_glock *gl)
191 struct gfs2_inode *ip = gl->gl_object;
192 struct address_space *metamapping = gfs2_glock2aspace(gl);
193 int error;
195 if (ip && !S_ISREG(ip->i_inode.i_mode))
196 ip = NULL;
197 if (ip) {
198 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
199 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
200 inode_dio_wait(&ip->i_inode);
202 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
203 return;
205 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
207 gfs2_log_flush(gl->gl_sbd, gl);
208 filemap_fdatawrite(metamapping);
209 if (ip) {
210 struct address_space *mapping = ip->i_inode.i_mapping;
211 filemap_fdatawrite(mapping);
212 error = filemap_fdatawait(mapping);
213 mapping_set_error(mapping, error);
215 error = filemap_fdatawait(metamapping);
216 mapping_set_error(metamapping, error);
217 gfs2_ail_empty_gl(gl);
219 * Writeback of the data mapping may cause the dirty flag to be set
220 * so we have to clear it again here.
222 smp_mb__before_clear_bit();
223 clear_bit(GLF_DIRTY, &gl->gl_flags);
227 * inode_go_inval - prepare a inode glock to be released
228 * @gl: the glock
229 * @flags:
231 * Normally we invlidate everything, but if we are moving into
232 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
233 * can keep hold of the metadata, since it won't have changed.
237 static void inode_go_inval(struct gfs2_glock *gl, int flags)
239 struct gfs2_inode *ip = gl->gl_object;
241 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
243 if (flags & DIO_METADATA) {
244 struct address_space *mapping = gfs2_glock2aspace(gl);
245 truncate_inode_pages(mapping, 0);
246 if (ip) {
247 set_bit(GIF_INVALID, &ip->i_flags);
248 forget_all_cached_acls(&ip->i_inode);
249 gfs2_dir_hash_inval(ip);
253 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
254 gfs2_log_flush(gl->gl_sbd, NULL);
255 gl->gl_sbd->sd_rindex_uptodate = 0;
257 if (ip && S_ISREG(ip->i_inode.i_mode))
258 truncate_inode_pages(ip->i_inode.i_mapping, 0);
262 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
263 * @gl: the glock
265 * Returns: 1 if it's ok
268 static int inode_go_demote_ok(const struct gfs2_glock *gl)
270 struct gfs2_sbd *sdp = gl->gl_sbd;
271 struct gfs2_holder *gh;
273 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
274 return 0;
276 if (!list_empty(&gl->gl_holders)) {
277 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
278 if (gh->gh_list.next != &gl->gl_holders)
279 return 0;
282 return 1;
286 * gfs2_set_nlink - Set the inode's link count based on on-disk info
287 * @inode: The inode in question
288 * @nlink: The link count
290 * If the link count has hit zero, it must never be raised, whatever the
291 * on-disk inode might say. When new struct inodes are created the link
292 * count is set to 1, so that we can safely use this test even when reading
293 * in on disk information for the first time.
296 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
299 * We will need to review setting the nlink count here in the
300 * light of the forthcoming ro bind mount work. This is a reminder
301 * to do that.
303 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
304 if (nlink == 0)
305 clear_nlink(inode);
306 else
307 set_nlink(inode, nlink);
311 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
313 const struct gfs2_dinode *str = buf;
314 struct timespec atime;
315 u16 height, depth;
317 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
318 goto corrupt;
319 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
320 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
321 ip->i_inode.i_rdev = 0;
322 switch (ip->i_inode.i_mode & S_IFMT) {
323 case S_IFBLK:
324 case S_IFCHR:
325 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
326 be32_to_cpu(str->di_minor));
327 break;
330 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
331 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
332 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
333 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
334 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
335 atime.tv_sec = be64_to_cpu(str->di_atime);
336 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
337 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
338 ip->i_inode.i_atime = atime;
339 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
340 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
341 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
342 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
344 ip->i_goal = be64_to_cpu(str->di_goal_meta);
345 ip->i_generation = be64_to_cpu(str->di_generation);
347 ip->i_diskflags = be32_to_cpu(str->di_flags);
348 ip->i_eattr = be64_to_cpu(str->di_eattr);
349 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
350 gfs2_set_inode_flags(&ip->i_inode);
351 height = be16_to_cpu(str->di_height);
352 if (unlikely(height > GFS2_MAX_META_HEIGHT))
353 goto corrupt;
354 ip->i_height = (u8)height;
356 depth = be16_to_cpu(str->di_depth);
357 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
358 goto corrupt;
359 ip->i_depth = (u8)depth;
360 ip->i_entries = be32_to_cpu(str->di_entries);
362 if (S_ISREG(ip->i_inode.i_mode))
363 gfs2_set_aops(&ip->i_inode);
365 return 0;
366 corrupt:
367 gfs2_consist_inode(ip);
368 return -EIO;
372 * gfs2_inode_refresh - Refresh the incore copy of the dinode
373 * @ip: The GFS2 inode
375 * Returns: errno
378 int gfs2_inode_refresh(struct gfs2_inode *ip)
380 struct buffer_head *dibh;
381 int error;
383 error = gfs2_meta_inode_buffer(ip, &dibh);
384 if (error)
385 return error;
387 error = gfs2_dinode_in(ip, dibh->b_data);
388 brelse(dibh);
389 clear_bit(GIF_INVALID, &ip->i_flags);
391 return error;
395 * inode_go_lock - operation done after an inode lock is locked by a process
396 * @gl: the glock
397 * @flags:
399 * Returns: errno
402 static int inode_go_lock(struct gfs2_holder *gh)
404 struct gfs2_glock *gl = gh->gh_gl;
405 struct gfs2_sbd *sdp = gl->gl_sbd;
406 struct gfs2_inode *ip = gl->gl_object;
407 int error = 0;
409 if (!ip || (gh->gh_flags & GL_SKIP))
410 return 0;
412 if (test_bit(GIF_INVALID, &ip->i_flags)) {
413 error = gfs2_inode_refresh(ip);
414 if (error)
415 return error;
418 if (gh->gh_state != LM_ST_DEFERRED)
419 inode_dio_wait(&ip->i_inode);
421 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
422 (gl->gl_state == LM_ST_EXCLUSIVE) &&
423 (gh->gh_state == LM_ST_EXCLUSIVE)) {
424 spin_lock(&sdp->sd_trunc_lock);
425 if (list_empty(&ip->i_trunc_list))
426 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
427 spin_unlock(&sdp->sd_trunc_lock);
428 wake_up(&sdp->sd_quota_wait);
429 return 1;
432 return error;
436 * inode_go_dump - print information about an inode
437 * @seq: The iterator
438 * @ip: the inode
442 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
444 const struct gfs2_inode *ip = gl->gl_object;
445 if (ip == NULL)
446 return;
447 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
448 (unsigned long long)ip->i_no_formal_ino,
449 (unsigned long long)ip->i_no_addr,
450 IF2DT(ip->i_inode.i_mode), ip->i_flags,
451 (unsigned int)ip->i_diskflags,
452 (unsigned long long)i_size_read(&ip->i_inode));
456 * trans_go_sync - promote/demote the transaction glock
457 * @gl: the glock
458 * @state: the requested state
459 * @flags:
463 static void trans_go_sync(struct gfs2_glock *gl)
465 struct gfs2_sbd *sdp = gl->gl_sbd;
467 if (gl->gl_state != LM_ST_UNLOCKED &&
468 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
469 gfs2_meta_syncfs(sdp);
470 gfs2_log_shutdown(sdp);
475 * trans_go_xmote_bh - After promoting/demoting the transaction glock
476 * @gl: the glock
480 static int trans_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
482 struct gfs2_sbd *sdp = gl->gl_sbd;
483 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
484 struct gfs2_glock *j_gl = ip->i_gl;
485 struct gfs2_log_header_host head;
486 int error;
488 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
489 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
491 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
492 if (error)
493 gfs2_consist(sdp);
494 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
495 gfs2_consist(sdp);
497 /* Initialize some head of the log stuff */
498 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
499 sdp->sd_log_sequence = head.lh_sequence + 1;
500 gfs2_log_pointers_init(sdp, head.lh_blkno);
503 return 0;
507 * trans_go_demote_ok
508 * @gl: the glock
510 * Always returns 0
513 static int trans_go_demote_ok(const struct gfs2_glock *gl)
515 return 0;
519 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
520 * @gl: the glock
522 * gl_spin lock is held while calling this
524 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
526 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
527 struct gfs2_sbd *sdp = gl->gl_sbd;
529 if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
530 return;
532 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
533 gl->gl_state == LM_ST_SHARED && ip) {
534 gl->gl_lockref.count++;
535 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
536 gl->gl_lockref.count--;
540 const struct gfs2_glock_operations gfs2_meta_glops = {
541 .go_type = LM_TYPE_META,
544 const struct gfs2_glock_operations gfs2_inode_glops = {
545 .go_sync = inode_go_sync,
546 .go_inval = inode_go_inval,
547 .go_demote_ok = inode_go_demote_ok,
548 .go_lock = inode_go_lock,
549 .go_dump = inode_go_dump,
550 .go_type = LM_TYPE_INODE,
551 .go_flags = GLOF_ASPACE,
554 const struct gfs2_glock_operations gfs2_rgrp_glops = {
555 .go_sync = rgrp_go_sync,
556 .go_inval = rgrp_go_inval,
557 .go_lock = gfs2_rgrp_go_lock,
558 .go_unlock = gfs2_rgrp_go_unlock,
559 .go_dump = gfs2_rgrp_dump,
560 .go_type = LM_TYPE_RGRP,
561 .go_flags = GLOF_LVB,
564 const struct gfs2_glock_operations gfs2_trans_glops = {
565 .go_sync = trans_go_sync,
566 .go_xmote_bh = trans_go_xmote_bh,
567 .go_demote_ok = trans_go_demote_ok,
568 .go_type = LM_TYPE_NONDISK,
571 const struct gfs2_glock_operations gfs2_iopen_glops = {
572 .go_type = LM_TYPE_IOPEN,
573 .go_callback = iopen_go_callback,
576 const struct gfs2_glock_operations gfs2_flock_glops = {
577 .go_type = LM_TYPE_FLOCK,
580 const struct gfs2_glock_operations gfs2_nondisk_glops = {
581 .go_type = LM_TYPE_NONDISK,
584 const struct gfs2_glock_operations gfs2_quota_glops = {
585 .go_type = LM_TYPE_QUOTA,
586 .go_flags = GLOF_LVB,
589 const struct gfs2_glock_operations gfs2_journal_glops = {
590 .go_type = LM_TYPE_JOURNAL,
593 const struct gfs2_glock_operations *gfs2_glops_list[] = {
594 [LM_TYPE_META] = &gfs2_meta_glops,
595 [LM_TYPE_INODE] = &gfs2_inode_glops,
596 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
597 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
598 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
599 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
600 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
601 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,