1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/bio.h>
10 #include <linux/sched/signal.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/statfs.h>
16 #include <linux/seq_file.h>
17 #include <linux/mount.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/crc32.h>
22 #include <linux/time.h>
23 #include <linux/wait.h>
24 #include <linux/writeback.h>
25 #include <linux/backing-dev.h>
26 #include <linux/kernel.h>
27
28 #include "gfs2.h"
29 #include "incore.h"
30 #include "bmap.h"
31 #include "dir.h"
32 #include "glock.h"
33 #include "glops.h"
34 #include "inode.h"
35 #include "log.h"
36 #include "meta_io.h"
37 #include "quota.h"
38 #include "recovery.h"
39 #include "rgrp.h"
40 #include "super.h"
41 #include "trans.h"
42 #include "util.h"
43 #include "sys.h"
44 #include "xattr.h"
45 #include "lops.h"
46
47 enum dinode_demise {
48 SHOULD_DELETE_DINODE,
49 SHOULD_NOT_DELETE_DINODE,
50 SHOULD_DEFER_EVICTION,
51 };
52
53 /**
54 * gfs2_jindex_free - Clear all the journal index information
55 * @sdp: The GFS2 superblock
56 *
57 */
58
59 void gfs2_jindex_free(struct gfs2_sbd *sdp)
60 {
61 struct list_head list;
62 struct gfs2_jdesc *jd;
63
64 spin_lock(&sdp->sd_jindex_spin);
65 list_add(&list, &sdp->sd_jindex_list);
66 list_del_init(&sdp->sd_jindex_list);
67 sdp->sd_journals = 0;
68 spin_unlock(&sdp->sd_jindex_spin);
69
70 down_write(&sdp->sd_log_flush_lock);
71 sdp->sd_jdesc = NULL;
72 up_write(&sdp->sd_log_flush_lock);
73
74 while (!list_empty(&list)) {
75 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
76 BUG_ON(jd->jd_log_bio);
77 gfs2_free_journal_extents(jd);
78 list_del(&jd->jd_list);
79 iput(jd->jd_inode);
80 jd->jd_inode = NULL;
81 kfree(jd);
82 }
83 }
84
85 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
86 {
87 struct gfs2_jdesc *jd;
88
89 list_for_each_entry(jd, head, jd_list) {
90 if (jd->jd_jid == jid)
91 return jd;
92 }
93 return NULL;
94 }
95
96 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
97 {
98 struct gfs2_jdesc *jd;
99
100 spin_lock(&sdp->sd_jindex_spin);
101 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
102 spin_unlock(&sdp->sd_jindex_spin);
103
104 return jd;
105 }
106
107 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
108 {
109 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
110 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
111 u64 size = i_size_read(jd->jd_inode);
112
113 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
114 return -EIO;
115
116 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
117
118 if (gfs2_write_alloc_required(ip, 0, size)) {
119 gfs2_consist_inode(ip);
120 return -EIO;
121 }
122
123 return 0;
124 }
125
126 /**
127 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
128 * @sdp: the filesystem
129 *
130 * Returns: errno
131 */
132
133 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
134 {
135 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
136 struct gfs2_glock *j_gl = ip->i_gl;
137 struct gfs2_log_header_host head;
138 int error;
139
140 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
141 if (gfs2_withdrawing_or_withdrawn(sdp))
142 return -EIO;
143
144 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
145 if (error) {
146 gfs2_consist(sdp);
147 return error;
148 }
149
150 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
151 gfs2_consist(sdp);
152 return -EIO;
153 }
154
155 /* Initialize some head of the log stuff */
156 sdp->sd_log_sequence = head.lh_sequence + 1;
157 gfs2_log_pointers_init(sdp, head.lh_blkno);
158
159 error = gfs2_quota_init(sdp);
160 if (!error && gfs2_withdrawing_or_withdrawn(sdp))
161 error = -EIO;
162 if (!error)
163 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
164 return error;
165 }
166
167 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
168 {
169 const struct gfs2_statfs_change *str = buf;
170
171 sc->sc_total = be64_to_cpu(str->sc_total);
172 sc->sc_free = be64_to_cpu(str->sc_free);
173 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
174 }
175
176 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
177 {
178 struct gfs2_statfs_change *str = buf;
179
180 str->sc_total = cpu_to_be64(sc->sc_total);
181 str->sc_free = cpu_to_be64(sc->sc_free);
182 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
183 }
184
185 int gfs2_statfs_init(struct gfs2_sbd *sdp)
186 {
187 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
188 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
189 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
190 struct buffer_head *m_bh;
191 struct gfs2_holder gh;
192 int error;
193
194 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
195 &gh);
196 if (error)
197 return error;
198
199 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
200 if (error)
201 goto out;
202
203 if (sdp->sd_args.ar_spectator) {
204 spin_lock(&sdp->sd_statfs_spin);
205 gfs2_statfs_change_in(m_sc, m_bh->b_data +
206 sizeof(struct gfs2_dinode));
207 spin_unlock(&sdp->sd_statfs_spin);
208 } else {
209 spin_lock(&sdp->sd_statfs_spin);
210 gfs2_statfs_change_in(m_sc, m_bh->b_data +
211 sizeof(struct gfs2_dinode));
212 gfs2_statfs_change_in(l_sc, sdp->sd_sc_bh->b_data +
213 sizeof(struct gfs2_dinode));
214 spin_unlock(&sdp->sd_statfs_spin);
215
216 }
217
218 brelse(m_bh);
219 out:
220 gfs2_glock_dq_uninit(&gh);
221 return 0;
222 }
223
224 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
225 s64 dinodes)
226 {
227 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
228 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
229 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
230 s64 x, y;
231 int need_sync = 0;
232
233 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
234
235 spin_lock(&sdp->sd_statfs_spin);
236 l_sc->sc_total += total;
237 l_sc->sc_free += free;
238 l_sc->sc_dinodes += dinodes;
239 gfs2_statfs_change_out(l_sc, sdp->sd_sc_bh->b_data +
240 sizeof(struct gfs2_dinode));
241 if (sdp->sd_args.ar_statfs_percent) {
242 x = 100 * l_sc->sc_free;
243 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
244 if (x >= y || x <= -y)
245 need_sync = 1;
246 }
247 spin_unlock(&sdp->sd_statfs_spin);
248
249 if (need_sync)
250 gfs2_wake_up_statfs(sdp);
251 }
252
253 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh)
254 {
255 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
256 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
257 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
258 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
259
260 gfs2_trans_add_meta(l_ip->i_gl, sdp->sd_sc_bh);
261 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
262
263 spin_lock(&sdp->sd_statfs_spin);
264 m_sc->sc_total += l_sc->sc_total;
265 m_sc->sc_free += l_sc->sc_free;
266 m_sc->sc_dinodes += l_sc->sc_dinodes;
267 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
268 memset(sdp->sd_sc_bh->b_data + sizeof(struct gfs2_dinode),
269 0, sizeof(struct gfs2_statfs_change));
270 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
271 spin_unlock(&sdp->sd_statfs_spin);
272 }
273
274 int gfs2_statfs_sync(struct super_block *sb, int type)
275 {
276 struct gfs2_sbd *sdp = sb->s_fs_info;
277 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
278 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
279 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
280 struct gfs2_holder gh;
281 struct buffer_head *m_bh;
282 int error;
283
284 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
285 &gh);
286 if (error)
287 goto out;
288
289 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
290 if (error)
291 goto out_unlock;
292
293 spin_lock(&sdp->sd_statfs_spin);
294 gfs2_statfs_change_in(m_sc, m_bh->b_data +
295 sizeof(struct gfs2_dinode));
296 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
297 spin_unlock(&sdp->sd_statfs_spin);
298 goto out_bh;
299 }
300 spin_unlock(&sdp->sd_statfs_spin);
301
302 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
303 if (error)
304 goto out_bh;
305
306 update_statfs(sdp, m_bh);
307 sdp->sd_statfs_force_sync = 0;
308
309 gfs2_trans_end(sdp);
310
311 out_bh:
312 brelse(m_bh);
313 out_unlock:
314 gfs2_glock_dq_uninit(&gh);
315 out:
316 return error;
317 }
318
319 struct lfcc {
320 struct list_head list;
321 struct gfs2_holder gh;
322 };
323
324 /**
325 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
326 * journals are clean
327 * @sdp: the file system
328 *
329 * Returns: errno
330 */
331
332 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
333 {
334 struct gfs2_inode *ip;
335 struct gfs2_jdesc *jd;
336 struct lfcc *lfcc;
337 LIST_HEAD(list);
338 struct gfs2_log_header_host lh;
339 int error, error2;
340
341 /*
342 * Grab all the journal glocks in SH mode. We are *probably* doing
343 * that to prevent recovery.
344 */
345
346 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
347 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
348 if (!lfcc) {
349 error = -ENOMEM;
350 goto out;
351 }
352 ip = GFS2_I(jd->jd_inode);
353 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
354 if (error) {
355 kfree(lfcc);
356 goto out;
357 }
358 list_add(&lfcc->list, &list);
359 }
360
361 gfs2_freeze_unlock(sdp);
362
363 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
364 LM_FLAG_NOEXP | GL_NOPID,
365 &sdp->sd_freeze_gh);
366 if (error)
367 goto relock_shared;
368
369 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
370 error = gfs2_jdesc_check(jd);
371 if (error)
372 break;
373 error = gfs2_find_jhead(jd, &lh, false);
374 if (error)
375 break;
376 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
377 error = -EBUSY;
378 break;
379 }
380 }
381
382 if (!error)
383 goto out; /* success */
384
385 gfs2_freeze_unlock(sdp);
386
387 relock_shared:
388 error2 = gfs2_freeze_lock_shared(sdp);
389 gfs2_assert_withdraw(sdp, !error2);
390
391 out:
392 while (!list_empty(&list)) {
393 lfcc = list_first_entry(&list, struct lfcc, list);
394 list_del(&lfcc->list);
395 gfs2_glock_dq_uninit(&lfcc->gh);
396 kfree(lfcc);
397 }
398 return error;
399 }
400
401 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
402 {
403 const struct inode *inode = &ip->i_inode;
404 struct gfs2_dinode *str = buf;
405
406 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
407 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
408 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
409 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
410 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
411 str->di_mode = cpu_to_be32(inode->i_mode);
412 str->di_uid = cpu_to_be32(i_uid_read(inode));
413 str->di_gid = cpu_to_be32(i_gid_read(inode));
414 str->di_nlink = cpu_to_be32(inode->i_nlink);
415 str->di_size = cpu_to_be64(i_size_read(inode));
416 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(inode));
417 str->di_atime = cpu_to_be64(inode_get_atime_sec(inode));
418 str->di_mtime = cpu_to_be64(inode_get_mtime_sec(inode));
419 str->di_ctime = cpu_to_be64(inode_get_ctime_sec(inode));
420
421 str->di_goal_meta = cpu_to_be64(ip->i_goal);
422 str->di_goal_data = cpu_to_be64(ip->i_goal);
423 str->di_generation = cpu_to_be64(ip->i_generation);
424
425 str->di_flags = cpu_to_be32(ip->i_diskflags);
426 str->di_height = cpu_to_be16(ip->i_height);
427 str->di_payload_format = cpu_to_be32(S_ISDIR(inode->i_mode) &&
428 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
429 GFS2_FORMAT_DE : 0);
430 str->di_depth = cpu_to_be16(ip->i_depth);
431 str->di_entries = cpu_to_be32(ip->i_entries);
432
433 str->di_eattr = cpu_to_be64(ip->i_eattr);
434 str->di_atime_nsec = cpu_to_be32(inode_get_atime_nsec(inode));
435 str->di_mtime_nsec = cpu_to_be32(inode_get_mtime_nsec(inode));
436 str->di_ctime_nsec = cpu_to_be32(inode_get_ctime_nsec(inode));
437 }
438
439 /**
440 * gfs2_write_inode - Make sure the inode is stable on the disk
441 * @inode: The inode
442 * @wbc: The writeback control structure
443 *
444 * Returns: errno
445 */
446
447 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
448 {
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
451 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
452 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
453 int ret = 0;
454 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
455
456 if (flush_all)
457 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
458 GFS2_LOG_HEAD_FLUSH_NORMAL |
459 GFS2_LFC_WRITE_INODE);
460 if (bdi->wb.dirty_exceeded)
461 gfs2_ail1_flush(sdp, wbc);
462 else
463 filemap_fdatawrite(metamapping);
464 if (flush_all)
465 ret = filemap_fdatawait(metamapping);
466 if (ret)
467 mark_inode_dirty_sync(inode);
468 else {
469 spin_lock(&inode->i_lock);
470 if (!(inode->i_flags & I_DIRTY))
471 gfs2_ordered_del_inode(ip);
472 spin_unlock(&inode->i_lock);
473 }
474 return ret;
475 }
476
477 /**
478 * gfs2_dirty_inode - check for atime updates
479 * @inode: The inode in question
480 * @flags: The type of dirty
481 *
482 * Unfortunately it can be called under any combination of inode
483 * glock and freeze glock, so we have to check carefully.
484 *
485 * At the moment this deals only with atime - it should be possible
486 * to expand that role in future, once a review of the locking has
487 * been carried out.
488 */
489
490 static void gfs2_dirty_inode(struct inode *inode, int flags)
491 {
492 struct gfs2_inode *ip = GFS2_I(inode);
493 struct gfs2_sbd *sdp = GFS2_SB(inode);
494 struct buffer_head *bh;
495 struct gfs2_holder gh;
496 int need_unlock = 0;
497 int need_endtrans = 0;
498 int ret;
499
500 if (unlikely(!ip->i_gl)) {
501 /* This can only happen during incomplete inode creation. */
502 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
503 return;
504 }
505
506 if (gfs2_withdrawing_or_withdrawn(sdp))
507 return;
508 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
509 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
510 if (ret) {
511 fs_err(sdp, "dirty_inode: glock %d\n", ret);
512 gfs2_dump_glock(NULL, ip->i_gl, true);
513 return;
514 }
515 need_unlock = 1;
516 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
517 return;
518
519 if (current->journal_info == NULL) {
520 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
521 if (ret) {
522 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
523 goto out;
524 }
525 need_endtrans = 1;
526 }
527
528 ret = gfs2_meta_inode_buffer(ip, &bh);
529 if (ret == 0) {
530 gfs2_trans_add_meta(ip->i_gl, bh);
531 gfs2_dinode_out(ip, bh->b_data);
532 brelse(bh);
533 }
534
535 if (need_endtrans)
536 gfs2_trans_end(sdp);
537 out:
538 if (need_unlock)
539 gfs2_glock_dq_uninit(&gh);
540 }
541
542 /**
543 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
544 * @sdp: the filesystem
545 *
546 * Returns: errno
547 */
548
549 void gfs2_make_fs_ro(struct gfs2_sbd *sdp)
550 {
551 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
552
553 if (!test_bit(SDF_KILL, &sdp->sd_flags))
554 gfs2_flush_delete_work(sdp);
555
556 gfs2_destroy_threads(sdp);
557
558 if (log_write_allowed) {
559 gfs2_quota_sync(sdp->sd_vfs, 0);
560 gfs2_statfs_sync(sdp->sd_vfs, 0);
561
562 /* We do two log flushes here. The first one commits dirty inodes
563 * and rgrps to the journal, but queues up revokes to the ail list.
564 * The second flush writes out and removes the revokes.
565 *
566 * The first must be done before the FLUSH_SHUTDOWN code
567 * clears the LIVE flag, otherwise it will not be able to start
568 * a transaction to write its revokes, and the error will cause
569 * a withdraw of the file system. */
570 gfs2_log_flush(sdp, NULL, GFS2_LFC_MAKE_FS_RO);
571 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
572 GFS2_LFC_MAKE_FS_RO);
573 wait_event_timeout(sdp->sd_log_waitq,
574 gfs2_log_is_empty(sdp),
575 HZ * 5);
576 gfs2_assert_warn(sdp, gfs2_log_is_empty(sdp));
577 }
578 gfs2_quota_cleanup(sdp);
579 }
580
581 /**
582 * gfs2_put_super - Unmount the filesystem
583 * @sb: The VFS superblock
584 *
585 */
586
587 static void gfs2_put_super(struct super_block *sb)
588 {
589 struct gfs2_sbd *sdp = sb->s_fs_info;
590 struct gfs2_jdesc *jd;
591
592 /* No more recovery requests */
593 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
594 smp_mb();
595
596 /* Wait on outstanding recovery */
597 restart:
598 spin_lock(&sdp->sd_jindex_spin);
599 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
600 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
601 continue;
602 spin_unlock(&sdp->sd_jindex_spin);
603 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
604 TASK_UNINTERRUPTIBLE);
605 goto restart;
606 }
607 spin_unlock(&sdp->sd_jindex_spin);
608
609 if (!sb_rdonly(sb))
610 gfs2_make_fs_ro(sdp);
611 else {
612 if (gfs2_withdrawing_or_withdrawn(sdp))
613 gfs2_destroy_threads(sdp);
614
615 gfs2_quota_cleanup(sdp);
616 }
617
618 WARN_ON(gfs2_withdrawing(sdp));
619
620 /* At this point, we're through modifying the disk */
621
622 /* Release stuff */
623
624 gfs2_freeze_unlock(sdp);
625
626 iput(sdp->sd_jindex);
627 iput(sdp->sd_statfs_inode);
628 iput(sdp->sd_rindex);
629 iput(sdp->sd_quota_inode);
630
631 gfs2_glock_put(sdp->sd_rename_gl);
632 gfs2_glock_put(sdp->sd_freeze_gl);
633
634 if (!sdp->sd_args.ar_spectator) {
635 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
636 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
637 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
638 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
639 brelse(sdp->sd_sc_bh);
640 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
641 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
642 free_local_statfs_inodes(sdp);
643 iput(sdp->sd_qc_inode);
644 }
645
646 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
647 gfs2_clear_rgrpd(sdp);
648 gfs2_jindex_free(sdp);
649 /* Take apart glock structures and buffer lists */
650 gfs2_gl_hash_clear(sdp);
651 truncate_inode_pages_final(&sdp->sd_aspace);
652 gfs2_delete_debugfs_file(sdp);
653
654 gfs2_sys_fs_del(sdp);
655 free_sbd(sdp);
656 }
657
658 /**
659 * gfs2_sync_fs - sync the filesystem
660 * @sb: the superblock
661 * @wait: true to wait for completion
662 *
663 * Flushes the log to disk.
664 */
665
666 static int gfs2_sync_fs(struct super_block *sb, int wait)
667 {
668 struct gfs2_sbd *sdp = sb->s_fs_info;
669
670 gfs2_quota_sync(sb, -1);
671 if (wait)
672 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
673 GFS2_LFC_SYNC_FS);
674 return sdp->sd_log_error;
675 }
676
677 static int gfs2_do_thaw(struct gfs2_sbd *sdp)
678 {
679 struct super_block *sb = sdp->sd_vfs;
680 int error;
681
682 error = gfs2_freeze_lock_shared(sdp);
683 if (error)
684 goto fail;
685 error = thaw_super(sb, FREEZE_HOLDER_USERSPACE);
686 if (!error)
687 return 0;
688
689 fail:
690 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n", error);
691 gfs2_assert_withdraw(sdp, 0);
692 return error;
693 }
694
695 void gfs2_freeze_func(struct work_struct *work)
696 {
697 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
698 struct super_block *sb = sdp->sd_vfs;
699 int error;
700
701 mutex_lock(&sdp->sd_freeze_mutex);
702 error = -EBUSY;
703 if (test_bit(SDF_FROZEN, &sdp->sd_flags))
704 goto freeze_failed;
705
706 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
707 if (error)
708 goto freeze_failed;
709
710 gfs2_freeze_unlock(sdp);
711 set_bit(SDF_FROZEN, &sdp->sd_flags);
712
713 error = gfs2_do_thaw(sdp);
714 if (error)
715 goto out;
716
717 clear_bit(SDF_FROZEN, &sdp->sd_flags);
718 goto out;
719
720 freeze_failed:
721 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n", error);
722
723 out:
724 mutex_unlock(&sdp->sd_freeze_mutex);
725 deactivate_super(sb);
726 }
727
728 /**
729 * gfs2_freeze_super - prevent further writes to the filesystem
730 * @sb: the VFS structure for the filesystem
731 *
732 */
733
734 static int gfs2_freeze_super(struct super_block *sb, enum freeze_holder who)
735 {
736 struct gfs2_sbd *sdp = sb->s_fs_info;
737 int error;
738
739 if (!mutex_trylock(&sdp->sd_freeze_mutex))
740 return -EBUSY;
741 if (test_bit(SDF_FROZEN, &sdp->sd_flags)) {
742 mutex_unlock(&sdp->sd_freeze_mutex);
743 return -EBUSY;
744 }
745
746 for (;;) {
747 error = freeze_super(sb, FREEZE_HOLDER_USERSPACE);
748 if (error) {
749 fs_info(sdp, "GFS2: couldn't freeze filesystem: %d\n",
750 error);
751 goto out;
752 }
753
754 error = gfs2_lock_fs_check_clean(sdp);
755 if (!error) {
756 set_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
757 set_bit(SDF_FROZEN, &sdp->sd_flags);
758 break;
759 }
760
761 error = gfs2_do_thaw(sdp);
762 if (error)
763 goto out;
764
765 if (error == -EBUSY)
766 fs_err(sdp, "waiting for recovery before freeze\n");
767 else if (error == -EIO) {
768 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
769 "to recovery error.\n");
770 goto out;
771 } else {
772 fs_err(sdp, "error freezing FS: %d\n", error);
773 }
774 fs_err(sdp, "retrying...\n");
775 msleep(1000);
776 }
777
778 out:
779 mutex_unlock(&sdp->sd_freeze_mutex);
780 return error;
781 }
782
783 static int gfs2_freeze_fs(struct super_block *sb)
784 {
785 struct gfs2_sbd *sdp = sb->s_fs_info;
786
787 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
788 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_FREEZE |
789 GFS2_LFC_FREEZE_GO_SYNC);
790 if (gfs2_withdrawing_or_withdrawn(sdp))
791 return -EIO;
792 }
793 return 0;
794 }
795
796 /**
797 * gfs2_thaw_super - reallow writes to the filesystem
798 * @sb: the VFS structure for the filesystem
799 *
800 */
801
802 static int gfs2_thaw_super(struct super_block *sb, enum freeze_holder who)
803 {
804 struct gfs2_sbd *sdp = sb->s_fs_info;
805 int error;
806
807 if (!mutex_trylock(&sdp->sd_freeze_mutex))
808 return -EBUSY;
809 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags)) {
810 mutex_unlock(&sdp->sd_freeze_mutex);
811 return -EINVAL;
812 }
813
814 atomic_inc(&sb->s_active);
815 gfs2_freeze_unlock(sdp);
816
817 error = gfs2_do_thaw(sdp);
818
819 if (!error) {
820 clear_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags);
821 clear_bit(SDF_FROZEN, &sdp->sd_flags);
822 }
823 mutex_unlock(&sdp->sd_freeze_mutex);
824 deactivate_super(sb);
825 return error;
826 }
827
828 void gfs2_thaw_freeze_initiator(struct super_block *sb)
829 {
830 struct gfs2_sbd *sdp = sb->s_fs_info;
831
832 mutex_lock(&sdp->sd_freeze_mutex);
833 if (!test_bit(SDF_FREEZE_INITIATOR, &sdp->sd_flags))
834 goto out;
835
836 gfs2_freeze_unlock(sdp);
837
838 out:
839 mutex_unlock(&sdp->sd_freeze_mutex);
840 }
841
842 /**
843 * statfs_slow_fill - fill in the sg for a given RG
844 * @rgd: the RG
845 * @sc: the sc structure
846 *
847 * Returns: 0 on success, -ESTALE if the LVB is invalid
848 */
849
850 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
851 struct gfs2_statfs_change_host *sc)
852 {
853 gfs2_rgrp_verify(rgd);
854 sc->sc_total += rgd->rd_data;
855 sc->sc_free += rgd->rd_free;
856 sc->sc_dinodes += rgd->rd_dinodes;
857 return 0;
858 }
859
860 /**
861 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
862 * @sdp: the filesystem
863 * @sc: the sc info that will be returned
864 *
865 * Any error (other than a signal) will cause this routine to fall back
866 * to the synchronous version.
867 *
868 * FIXME: This really shouldn't busy wait like this.
869 *
870 * Returns: errno
871 */
872
873 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
874 {
875 struct gfs2_rgrpd *rgd_next;
876 struct gfs2_holder *gha, *gh;
877 unsigned int slots = 64;
878 unsigned int x;
879 int done;
880 int error = 0, err;
881
882 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
883 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
884 if (!gha)
885 return -ENOMEM;
886 for (x = 0; x < slots; x++)
887 gfs2_holder_mark_uninitialized(gha + x);
888
889 rgd_next = gfs2_rgrpd_get_first(sdp);
890
891 for (;;) {
892 done = 1;
893
894 for (x = 0; x < slots; x++) {
895 gh = gha + x;
896
897 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
898 err = gfs2_glock_wait(gh);
899 if (err) {
900 gfs2_holder_uninit(gh);
901 error = err;
902 } else {
903 if (!error) {
904 struct gfs2_rgrpd *rgd =
905 gfs2_glock2rgrp(gh->gh_gl);
906
907 error = statfs_slow_fill(rgd, sc);
908 }
909 gfs2_glock_dq_uninit(gh);
910 }
911 }
912
913 if (gfs2_holder_initialized(gh))
914 done = 0;
915 else if (rgd_next && !error) {
916 error = gfs2_glock_nq_init(rgd_next->rd_gl,
917 LM_ST_SHARED,
918 GL_ASYNC,
919 gh);
920 rgd_next = gfs2_rgrpd_get_next(rgd_next);
921 done = 0;
922 }
923
924 if (signal_pending(current))
925 error = -ERESTARTSYS;
926 }
927
928 if (done)
929 break;
930
931 yield();
932 }
933
934 kfree(gha);
935 return error;
936 }
937
938 /**
939 * gfs2_statfs_i - Do a statfs
940 * @sdp: the filesystem
941 * @sc: the sc structure
942 *
943 * Returns: errno
944 */
945
946 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
947 {
948 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
949 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
950
951 spin_lock(&sdp->sd_statfs_spin);
952
953 *sc = *m_sc;
954 sc->sc_total += l_sc->sc_total;
955 sc->sc_free += l_sc->sc_free;
956 sc->sc_dinodes += l_sc->sc_dinodes;
957
958 spin_unlock(&sdp->sd_statfs_spin);
959
960 if (sc->sc_free < 0)
961 sc->sc_free = 0;
962 if (sc->sc_free > sc->sc_total)
963 sc->sc_free = sc->sc_total;
964 if (sc->sc_dinodes < 0)
965 sc->sc_dinodes = 0;
966
967 return 0;
968 }
969
970 /**
971 * gfs2_statfs - Gather and return stats about the filesystem
972 * @dentry: The name of the link
973 * @buf: The buffer
974 *
975 * Returns: 0 on success or error code
976 */
977
978 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
979 {
980 struct super_block *sb = dentry->d_sb;
981 struct gfs2_sbd *sdp = sb->s_fs_info;
982 struct gfs2_statfs_change_host sc;
983 int error;
984
985 error = gfs2_rindex_update(sdp);
986 if (error)
987 return error;
988
989 if (gfs2_tune_get(sdp, gt_statfs_slow))
990 error = gfs2_statfs_slow(sdp, &sc);
991 else
992 error = gfs2_statfs_i(sdp, &sc);
993
994 if (error)
995 return error;
996
997 buf->f_type = GFS2_MAGIC;
998 buf->f_bsize = sdp->sd_sb.sb_bsize;
999 buf->f_blocks = sc.sc_total;
1000 buf->f_bfree = sc.sc_free;
1001 buf->f_bavail = sc.sc_free;
1002 buf->f_files = sc.sc_dinodes + sc.sc_free;
1003 buf->f_ffree = sc.sc_free;
1004 buf->f_namelen = GFS2_FNAMESIZE;
1005 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
1006
1007 return 0;
1008 }
1009
1010 /**
1011 * gfs2_drop_inode - Drop an inode (test for remote unlink)
1012 * @inode: The inode to drop
1013 *
1014 * If we've received a callback on an iopen lock then it's because a
1015 * remote node tried to deallocate the inode but failed due to this node
1016 * still having the inode open. Here we mark the link count zero
1017 * since we know that it must have reached zero if the GLF_DEMOTE flag
1018 * is set on the iopen glock. If we didn't do a disk read since the
1019 * remote node removed the final link then we might otherwise miss
1020 * this event. This check ensures that this node will deallocate the
1021 * inode's blocks, or alternatively pass the baton on to another
1022 * node for later deallocation.
1023 */
1024
1025 static int gfs2_drop_inode(struct inode *inode)
1026 {
1027 struct gfs2_inode *ip = GFS2_I(inode);
1028 struct gfs2_sbd *sdp = GFS2_SB(inode);
1029
1030 if (inode->i_nlink &&
1031 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1032 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1033 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1034 clear_nlink(inode);
1035 }
1036
1037 /*
1038 * When under memory pressure when an inode's link count has dropped to
1039 * zero, defer deleting the inode to the delete workqueue. This avoids
1040 * calling into DLM under memory pressure, which can deadlock.
1041 */
1042 if (!inode->i_nlink &&
1043 unlikely(current->flags & PF_MEMALLOC) &&
1044 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1045 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1046
1047 gfs2_glock_hold(gl);
1048 if (!gfs2_queue_try_to_evict(gl))
1049 gfs2_glock_put_async(gl);
1050 return 0;
1051 }
1052
1053 /*
1054 * No longer cache inodes when trying to evict them all.
1055 */
1056 if (test_bit(SDF_EVICTING, &sdp->sd_flags))
1057 return 1;
1058
1059 return generic_drop_inode(inode);
1060 }
1061
1062 /**
1063 * gfs2_show_options - Show mount options for /proc/mounts
1064 * @s: seq_file structure
1065 * @root: root of this (sub)tree
1066 *
1067 * Returns: 0 on success or error code
1068 */
1069
1070 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1071 {
1072 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1073 struct gfs2_args *args = &sdp->sd_args;
1074 unsigned int logd_secs, statfs_slow, statfs_quantum, quota_quantum;
1075
1076 spin_lock(&sdp->sd_tune.gt_spin);
1077 logd_secs = sdp->sd_tune.gt_logd_secs;
1078 quota_quantum = sdp->sd_tune.gt_quota_quantum;
1079 statfs_quantum = sdp->sd_tune.gt_statfs_quantum;
1080 statfs_slow = sdp->sd_tune.gt_statfs_slow;
1081 spin_unlock(&sdp->sd_tune.gt_spin);
1082
1083 if (is_subdir(root, sdp->sd_master_dir))
1084 seq_puts(s, ",meta");
1085 if (args->ar_lockproto[0])
1086 seq_show_option(s, "lockproto", args->ar_lockproto);
1087 if (args->ar_locktable[0])
1088 seq_show_option(s, "locktable", args->ar_locktable);
1089 if (args->ar_hostdata[0])
1090 seq_show_option(s, "hostdata", args->ar_hostdata);
1091 if (args->ar_spectator)
1092 seq_puts(s, ",spectator");
1093 if (args->ar_localflocks)
1094 seq_puts(s, ",localflocks");
1095 if (args->ar_debug)
1096 seq_puts(s, ",debug");
1097 if (args->ar_posix_acl)
1098 seq_puts(s, ",acl");
1099 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1100 char *state;
1101 switch (args->ar_quota) {
1102 case GFS2_QUOTA_OFF:
1103 state = "off";
1104 break;
1105 case GFS2_QUOTA_ACCOUNT:
1106 state = "account";
1107 break;
1108 case GFS2_QUOTA_ON:
1109 state = "on";
1110 break;
1111 case GFS2_QUOTA_QUIET:
1112 state = "quiet";
1113 break;
1114 default:
1115 state = "unknown";
1116 break;
1117 }
1118 seq_printf(s, ",quota=%s", state);
1119 }
1120 if (args->ar_suiddir)
1121 seq_puts(s, ",suiddir");
1122 if (args->ar_data != GFS2_DATA_DEFAULT) {
1123 char *state;
1124 switch (args->ar_data) {
1125 case GFS2_DATA_WRITEBACK:
1126 state = "writeback";
1127 break;
1128 case GFS2_DATA_ORDERED:
1129 state = "ordered";
1130 break;
1131 default:
1132 state = "unknown";
1133 break;
1134 }
1135 seq_printf(s, ",data=%s", state);
1136 }
1137 if (args->ar_discard)
1138 seq_puts(s, ",discard");
1139 if (logd_secs != 30)
1140 seq_printf(s, ",commit=%d", logd_secs);
1141 if (statfs_quantum != 30)
1142 seq_printf(s, ",statfs_quantum=%d", statfs_quantum);
1143 else if (statfs_slow)
1144 seq_puts(s, ",statfs_quantum=0");
1145 if (quota_quantum != 60)
1146 seq_printf(s, ",quota_quantum=%d", quota_quantum);
1147 if (args->ar_statfs_percent)
1148 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1149 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1150 const char *state;
1151
1152 switch (args->ar_errors) {
1153 case GFS2_ERRORS_WITHDRAW:
1154 state = "withdraw";
1155 break;
1156 case GFS2_ERRORS_PANIC:
1157 state = "panic";
1158 break;
1159 default:
1160 state = "unknown";
1161 break;
1162 }
1163 seq_printf(s, ",errors=%s", state);
1164 }
1165 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1166 seq_puts(s, ",nobarrier");
1167 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1168 seq_puts(s, ",demote_interface_used");
1169 if (args->ar_rgrplvb)
1170 seq_puts(s, ",rgrplvb");
1171 if (args->ar_loccookie)
1172 seq_puts(s, ",loccookie");
1173 return 0;
1174 }
1175
1176 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1177 {
1178 struct inode *inode = &ip->i_inode;
1179 struct gfs2_glock *gl = ip->i_gl;
1180
1181 if (unlikely(!gl)) {
1182 /* This can only happen during incomplete inode creation. */
1183 BUG_ON(!test_bit(GIF_ALLOC_FAILED, &ip->i_flags));
1184 return;
1185 }
1186
1187 truncate_inode_pages(gfs2_glock2aspace(gl), 0);
1188 truncate_inode_pages(&inode->i_data, 0);
1189
1190 if (atomic_read(&gl->gl_revokes) == 0) {
1191 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1192 clear_bit(GLF_DIRTY, &gl->gl_flags);
1193 }
1194 }
1195
1196 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1197 {
1198 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1199 struct gfs2_rgrpd *rgd;
1200 struct gfs2_holder gh;
1201 int error;
1202
1203 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1204 gfs2_consist_inode(ip);
1205 return -EIO;
1206 }
1207
1208 gfs2_rindex_update(sdp);
1209
1210 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1211 if (error)
1212 return error;
1213
1214 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1215 if (!rgd) {
1216 gfs2_consist_inode(ip);
1217 error = -EIO;
1218 goto out_qs;
1219 }
1220
1221 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE,
1222 LM_FLAG_NODE_SCOPE, &gh);
1223 if (error)
1224 goto out_qs;
1225
1226 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1227 sdp->sd_jdesc->jd_blocks);
1228 if (error)
1229 goto out_rg_gunlock;
1230
1231 gfs2_free_di(rgd, ip);
1232
1233 gfs2_final_release_pages(ip);
1234
1235 gfs2_trans_end(sdp);
1236
1237 out_rg_gunlock:
1238 gfs2_glock_dq_uninit(&gh);
1239 out_qs:
1240 gfs2_quota_unhold(ip);
1241 return error;
1242 }
1243
1244 /**
1245 * gfs2_glock_put_eventually
1246 * @gl: The glock to put
1247 *
1248 * When under memory pressure, trigger a deferred glock put to make sure we
1249 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1250 */
1251
1252 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1253 {
1254 if (current->flags & PF_MEMALLOC)
1255 gfs2_glock_put_async(gl);
1256 else
1257 gfs2_glock_put(gl);
1258 }
1259
1260 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1261 {
1262 struct gfs2_inode *ip = GFS2_I(inode);
1263 struct gfs2_sbd *sdp = GFS2_SB(inode);
1264 struct gfs2_holder *gh = &ip->i_iopen_gh;
1265 int error;
1266
1267 gh->gh_flags |= GL_NOCACHE;
1268 gfs2_glock_dq_wait(gh);
1269
1270 /*
1271 * If there are no other lock holders, we will immediately get
1272 * exclusive access to the iopen glock here.
1273 *
1274 * Otherwise, the other nodes holding the lock will be notified about
1275 * our locking request. If they do not have the inode open, they are
1276 * expected to evict the cached inode and release the lock, allowing us
1277 * to proceed.
1278 *
1279 * Otherwise, if they cannot evict the inode, they are expected to poke
1280 * the inode glock (note: not the iopen glock). We will notice that
1281 * and stop waiting for the iopen glock immediately. The other node(s)
1282 * are then expected to take care of deleting the inode when they no
1283 * longer use it.
1284 *
1285 * As a last resort, if another node keeps holding the iopen glock
1286 * without showing any activity on the inode glock, we will eventually
1287 * time out and fail the iopen glock upgrade.
1288 */
1289
1290 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1291 error = gfs2_glock_nq(gh);
1292 if (error)
1293 return false;
1294
1295 wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1296 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1297 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1298 5 * HZ);
1299 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1300 gfs2_glock_dq(gh);
1301 return false;
1302 }
1303 return gfs2_glock_holder_ready(gh) == 0;
1304 }
1305
1306 /**
1307 * evict_should_delete - determine whether the inode is eligible for deletion
1308 * @inode: The inode to evict
1309 * @gh: The glock holder structure
1310 *
1311 * This function determines whether the evicted inode is eligible to be deleted
1312 * and locks the inode glock.
1313 *
1314 * Returns: the fate of the dinode
1315 */
1316 static enum dinode_demise evict_should_delete(struct inode *inode,
1317 struct gfs2_holder *gh)
1318 {
1319 struct gfs2_inode *ip = GFS2_I(inode);
1320 struct super_block *sb = inode->i_sb;
1321 struct gfs2_sbd *sdp = sb->s_fs_info;
1322 int ret;
1323
1324 if (unlikely(test_bit(GIF_ALLOC_FAILED, &ip->i_flags)))
1325 goto should_delete;
1326
1327 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1328 return SHOULD_DEFER_EVICTION;
1329
1330 /* Deletes should never happen under memory pressure anymore. */
1331 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1332 return SHOULD_DEFER_EVICTION;
1333
1334 /* Must not read inode block until block type has been verified */
1335 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1336 if (unlikely(ret)) {
1337 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1338 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1339 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1340 return SHOULD_DEFER_EVICTION;
1341 }
1342
1343 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1344 return SHOULD_NOT_DELETE_DINODE;
1345 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1346 if (ret)
1347 return SHOULD_NOT_DELETE_DINODE;
1348
1349 ret = gfs2_instantiate(gh);
1350 if (ret)
1351 return SHOULD_NOT_DELETE_DINODE;
1352
1353 /*
1354 * The inode may have been recreated in the meantime.
1355 */
1356 if (inode->i_nlink)
1357 return SHOULD_NOT_DELETE_DINODE;
1358
1359 should_delete:
1360 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1361 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1362 if (!gfs2_upgrade_iopen_glock(inode)) {
1363 gfs2_holder_uninit(&ip->i_iopen_gh);
1364 return SHOULD_NOT_DELETE_DINODE;
1365 }
1366 }
1367 return SHOULD_DELETE_DINODE;
1368 }
1369
1370 /**
1371 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1372 * @inode: The inode to evict
1373 */
1374 static int evict_unlinked_inode(struct inode *inode)
1375 {
1376 struct gfs2_inode *ip = GFS2_I(inode);
1377 int ret;
1378
1379 if (S_ISDIR(inode->i_mode) &&
1380 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1381 ret = gfs2_dir_exhash_dealloc(ip);
1382 if (ret)
1383 goto out;
1384 }
1385
1386 if (ip->i_eattr) {
1387 ret = gfs2_ea_dealloc(ip);
1388 if (ret)
1389 goto out;
1390 }
1391
1392 if (!gfs2_is_stuffed(ip)) {
1393 ret = gfs2_file_dealloc(ip);
1394 if (ret)
1395 goto out;
1396 }
1397
1398 /*
1399 * As soon as we clear the bitmap for the dinode, gfs2_create_inode()
1400 * can get called to recreate it, or even gfs2_inode_lookup() if the
1401 * inode was recreated on another node in the meantime.
1402 *
1403 * However, inserting the new inode into the inode hash table will not
1404 * succeed until the old inode is removed, and that only happens after
1405 * ->evict_inode() returns. The new inode is attached to its inode and
1406 * iopen glocks after inserting it into the inode hash table, so at
1407 * that point we can be sure that both glocks are unused.
1408 */
1409
1410 ret = gfs2_dinode_dealloc(ip);
1411 if (!ret && ip->i_gl)
1412 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1413
1414 out:
1415 return ret;
1416 }
1417
1418 /*
1419 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1420 * @inode: The inode to evict
1421 */
1422 static int evict_linked_inode(struct inode *inode)
1423 {
1424 struct super_block *sb = inode->i_sb;
1425 struct gfs2_sbd *sdp = sb->s_fs_info;
1426 struct gfs2_inode *ip = GFS2_I(inode);
1427 struct address_space *metamapping;
1428 int ret;
1429
1430 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1431 GFS2_LFC_EVICT_INODE);
1432 metamapping = gfs2_glock2aspace(ip->i_gl);
1433 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1434 filemap_fdatawrite(metamapping);
1435 filemap_fdatawait(metamapping);
1436 }
1437 write_inode_now(inode, 1);
1438 gfs2_ail_flush(ip->i_gl, 0);
1439
1440 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1441 if (ret)
1442 return ret;
1443
1444 /* Needs to be done before glock release & also in a transaction */
1445 truncate_inode_pages(&inode->i_data, 0);
1446 truncate_inode_pages(metamapping, 0);
1447 gfs2_trans_end(sdp);
1448 return 0;
1449 }
1450
1451 /**
1452 * gfs2_evict_inode - Remove an inode from cache
1453 * @inode: The inode to evict
1454 *
1455 * There are three cases to consider:
1456 * 1. i_nlink == 0, we are final opener (and must deallocate)
1457 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1458 * 3. i_nlink > 0
1459 *
1460 * If the fs is read only, then we have to treat all cases as per #3
1461 * since we are unable to do any deallocation. The inode will be
1462 * deallocated by the next read/write node to attempt an allocation
1463 * in the same resource group
1464 *
1465 * We have to (at the moment) hold the inodes main lock to cover
1466 * the gap between unlocking the shared lock on the iopen lock and
1467 * taking the exclusive lock. I'd rather do a shared -> exclusive
1468 * conversion on the iopen lock, but we can change that later. This
1469 * is safe, just less efficient.
1470 */
1471
1472 static void gfs2_evict_inode(struct inode *inode)
1473 {
1474 struct super_block *sb = inode->i_sb;
1475 struct gfs2_sbd *sdp = sb->s_fs_info;
1476 struct gfs2_inode *ip = GFS2_I(inode);
1477 struct gfs2_holder gh;
1478 int ret;
1479
1480 if (inode->i_nlink || sb_rdonly(sb) || !ip->i_no_addr)
1481 goto out;
1482
1483 /*
1484 * In case of an incomplete mount, gfs2_evict_inode() may be called for
1485 * system files without having an active journal to write to. In that
1486 * case, skip the filesystem evict.
1487 */
1488 if (!sdp->sd_jdesc)
1489 goto out;
1490
1491 gfs2_holder_mark_uninitialized(&gh);
1492 ret = evict_should_delete(inode, &gh);
1493 if (ret == SHOULD_DEFER_EVICTION)
1494 goto out;
1495 if (ret == SHOULD_DELETE_DINODE)
1496 ret = evict_unlinked_inode(inode);
1497 else
1498 ret = evict_linked_inode(inode);
1499
1500 if (gfs2_rs_active(&ip->i_res))
1501 gfs2_rs_deltree(&ip->i_res);
1502
1503 if (gfs2_holder_initialized(&gh))
1504 gfs2_glock_dq_uninit(&gh);
1505 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1506 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1507 out:
1508 truncate_inode_pages_final(&inode->i_data);
1509 if (ip->i_qadata)
1510 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1511 gfs2_rs_deltree(&ip->i_res);
1512 gfs2_ordered_del_inode(ip);
1513 clear_inode(inode);
1514 gfs2_dir_hash_inval(ip);
1515 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1516 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1517
1518 glock_clear_object(gl, ip);
1519 gfs2_glock_hold(gl);
1520 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1521 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1522 gfs2_glock_put_eventually(gl);
1523 }
1524 if (ip->i_gl) {
1525 glock_clear_object(ip->i_gl, ip);
1526 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1527 gfs2_glock_put_eventually(ip->i_gl);
1528 rcu_assign_pointer(ip->i_gl, NULL);
1529 }
1530 }
1531
1532 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1533 {
1534 struct gfs2_inode *ip;
1535
1536 ip = alloc_inode_sb(sb, gfs2_inode_cachep, GFP_KERNEL);
1537 if (!ip)
1538 return NULL;
1539 ip->i_no_addr = 0;
1540 ip->i_flags = 0;
1541 ip->i_gl = NULL;
1542 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1543 memset(&ip->i_res, 0, sizeof(ip->i_res));
1544 RB_CLEAR_NODE(&ip->i_res.rs_node);
1545 ip->i_rahead = 0;
1546 return &ip->i_inode;
1547 }
1548
1549 static void gfs2_free_inode(struct inode *inode)
1550 {
1551 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1552 }
1553
1554 void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1555 {
1556 struct local_statfs_inode *lsi, *safe;
1557
1558 /* Run through the statfs inodes list to iput and free memory */
1559 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1560 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1561 sdp->sd_sc_inode = NULL; /* belongs to this node */
1562 if (lsi->si_sc_inode)
1563 iput(lsi->si_sc_inode);
1564 list_del(&lsi->si_list);
1565 kfree(lsi);
1566 }
1567 }
1568
1569 struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1570 unsigned int index)
1571 {
1572 struct local_statfs_inode *lsi;
1573
1574 /* Return the local (per node) statfs inode in the
1575 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1576 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1577 if (lsi->si_jid == index)
1578 return lsi->si_sc_inode;
1579 }
1580 return NULL;
1581 }
1582
1583 const struct super_operations gfs2_super_ops = {
1584 .alloc_inode = gfs2_alloc_inode,
1585 .free_inode = gfs2_free_inode,
1586 .write_inode = gfs2_write_inode,
1587 .dirty_inode = gfs2_dirty_inode,
1588 .evict_inode = gfs2_evict_inode,
1589 .put_super = gfs2_put_super,
1590 .sync_fs = gfs2_sync_fs,
1591 .freeze_super = gfs2_freeze_super,
1592 .freeze_fs = gfs2_freeze_fs,
1593 .thaw_super = gfs2_thaw_super,
1594 .statfs = gfs2_statfs,
1595 .drop_inode = gfs2_drop_inode,
1596 .show_options = gfs2_show_options,
1597 };
1598
1599
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.