blob: 9c593fd50c6a50ec6d2c6f25ff2c150d1c250574 [file] [log] [blame]
David Brazdil0f672f62019-12-10 10:32:29 +00001// SPDX-License-Identifier: GPL-2.0-only
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00002/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00005 */
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"
David Brazdil0f672f62019-12-10 10:32:29 +000045#include "lops.h"
Andrew Scullb4b6d4a2019-01-02 15:54:55 +000046
47/**
48 * gfs2_jindex_free - Clear all the journal index information
49 * @sdp: The GFS2 superblock
50 *
51 */
52
53void gfs2_jindex_free(struct gfs2_sbd *sdp)
54{
55 struct list_head list;
56 struct gfs2_jdesc *jd;
57
58 spin_lock(&sdp->sd_jindex_spin);
59 list_add(&list, &sdp->sd_jindex_list);
60 list_del_init(&sdp->sd_jindex_list);
61 sdp->sd_journals = 0;
62 spin_unlock(&sdp->sd_jindex_spin);
63
64 while (!list_empty(&list)) {
65 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
66 gfs2_free_journal_extents(jd);
67 list_del(&jd->jd_list);
68 iput(jd->jd_inode);
69 kfree(jd);
70 }
71}
72
73static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
74{
75 struct gfs2_jdesc *jd;
76 int found = 0;
77
78 list_for_each_entry(jd, head, jd_list) {
79 if (jd->jd_jid == jid) {
80 found = 1;
81 break;
82 }
83 }
84
85 if (!found)
86 jd = NULL;
87
88 return jd;
89}
90
91struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
92{
93 struct gfs2_jdesc *jd;
94
95 spin_lock(&sdp->sd_jindex_spin);
96 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
97 spin_unlock(&sdp->sd_jindex_spin);
98
99 return jd;
100}
101
102int gfs2_jdesc_check(struct gfs2_jdesc *jd)
103{
104 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
105 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
106 u64 size = i_size_read(jd->jd_inode);
107
108 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
109 return -EIO;
110
111 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
112
113 if (gfs2_write_alloc_required(ip, 0, size)) {
114 gfs2_consist_inode(ip);
115 return -EIO;
116 }
117
118 return 0;
119}
120
121static int init_threads(struct gfs2_sbd *sdp)
122{
123 struct task_struct *p;
124 int error = 0;
125
126 p = kthread_run(gfs2_logd, sdp, "gfs2_logd");
127 if (IS_ERR(p)) {
128 error = PTR_ERR(p);
129 fs_err(sdp, "can't start logd thread: %d\n", error);
130 return error;
131 }
132 sdp->sd_logd_process = p;
133
134 p = kthread_run(gfs2_quotad, sdp, "gfs2_quotad");
135 if (IS_ERR(p)) {
136 error = PTR_ERR(p);
137 fs_err(sdp, "can't start quotad thread: %d\n", error);
138 goto fail;
139 }
140 sdp->sd_quotad_process = p;
141 return 0;
142
143fail:
144 kthread_stop(sdp->sd_logd_process);
David Brazdil0f672f62019-12-10 10:32:29 +0000145 sdp->sd_logd_process = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000146 return error;
147}
148
149/**
150 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
151 * @sdp: the filesystem
152 *
153 * Returns: errno
154 */
155
156int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
157{
158 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
159 struct gfs2_glock *j_gl = ip->i_gl;
160 struct gfs2_holder freeze_gh;
161 struct gfs2_log_header_host head;
162 int error;
163
164 error = init_threads(sdp);
165 if (error)
166 return error;
167
168 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
169 &freeze_gh);
170 if (error)
171 goto fail_threads;
172
173 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
174
David Brazdil0f672f62019-12-10 10:32:29 +0000175 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000176 if (error)
177 goto fail;
178
179 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
180 gfs2_consist(sdp);
181 error = -EIO;
182 goto fail;
183 }
184
185 /* Initialize some head of the log stuff */
186 sdp->sd_log_sequence = head.lh_sequence + 1;
187 gfs2_log_pointers_init(sdp, head.lh_blkno);
188
189 error = gfs2_quota_init(sdp);
190 if (error)
191 goto fail;
192
193 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
194
195 gfs2_glock_dq_uninit(&freeze_gh);
196
197 return 0;
198
199fail:
200 freeze_gh.gh_flags |= GL_NOCACHE;
201 gfs2_glock_dq_uninit(&freeze_gh);
202fail_threads:
David Brazdil0f672f62019-12-10 10:32:29 +0000203 if (sdp->sd_quotad_process)
204 kthread_stop(sdp->sd_quotad_process);
205 sdp->sd_quotad_process = NULL;
206 if (sdp->sd_logd_process)
207 kthread_stop(sdp->sd_logd_process);
208 sdp->sd_logd_process = NULL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000209 return error;
210}
211
212void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
213{
214 const struct gfs2_statfs_change *str = buf;
215
216 sc->sc_total = be64_to_cpu(str->sc_total);
217 sc->sc_free = be64_to_cpu(str->sc_free);
218 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
219}
220
221static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
222{
223 struct gfs2_statfs_change *str = buf;
224
225 str->sc_total = cpu_to_be64(sc->sc_total);
226 str->sc_free = cpu_to_be64(sc->sc_free);
227 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
228}
229
230int gfs2_statfs_init(struct gfs2_sbd *sdp)
231{
232 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
233 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
234 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
235 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
236 struct buffer_head *m_bh, *l_bh;
237 struct gfs2_holder gh;
238 int error;
239
240 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
241 &gh);
242 if (error)
243 return error;
244
245 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
246 if (error)
247 goto out;
248
249 if (sdp->sd_args.ar_spectator) {
250 spin_lock(&sdp->sd_statfs_spin);
251 gfs2_statfs_change_in(m_sc, m_bh->b_data +
252 sizeof(struct gfs2_dinode));
253 spin_unlock(&sdp->sd_statfs_spin);
254 } else {
255 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
256 if (error)
257 goto out_m_bh;
258
259 spin_lock(&sdp->sd_statfs_spin);
260 gfs2_statfs_change_in(m_sc, m_bh->b_data +
261 sizeof(struct gfs2_dinode));
262 gfs2_statfs_change_in(l_sc, l_bh->b_data +
263 sizeof(struct gfs2_dinode));
264 spin_unlock(&sdp->sd_statfs_spin);
265
266 brelse(l_bh);
267 }
268
269out_m_bh:
270 brelse(m_bh);
271out:
272 gfs2_glock_dq_uninit(&gh);
273 return 0;
274}
275
276void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
277 s64 dinodes)
278{
279 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
280 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
281 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
282 struct buffer_head *l_bh;
283 s64 x, y;
284 int need_sync = 0;
285 int error;
286
287 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
288 if (error)
289 return;
290
291 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
292
293 spin_lock(&sdp->sd_statfs_spin);
294 l_sc->sc_total += total;
295 l_sc->sc_free += free;
296 l_sc->sc_dinodes += dinodes;
297 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
298 if (sdp->sd_args.ar_statfs_percent) {
299 x = 100 * l_sc->sc_free;
300 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
301 if (x >= y || x <= -y)
302 need_sync = 1;
303 }
304 spin_unlock(&sdp->sd_statfs_spin);
305
306 brelse(l_bh);
307 if (need_sync)
308 gfs2_wake_up_statfs(sdp);
309}
310
311void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
312 struct buffer_head *l_bh)
313{
314 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
315 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
316 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
317 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
318
319 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
320 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
321
322 spin_lock(&sdp->sd_statfs_spin);
323 m_sc->sc_total += l_sc->sc_total;
324 m_sc->sc_free += l_sc->sc_free;
325 m_sc->sc_dinodes += l_sc->sc_dinodes;
326 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
327 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
328 0, sizeof(struct gfs2_statfs_change));
329 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
330 spin_unlock(&sdp->sd_statfs_spin);
331}
332
333int gfs2_statfs_sync(struct super_block *sb, int type)
334{
335 struct gfs2_sbd *sdp = sb->s_fs_info;
336 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
337 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
338 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
339 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
340 struct gfs2_holder gh;
341 struct buffer_head *m_bh, *l_bh;
342 int error;
343
344 sb_start_write(sb);
345 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
346 &gh);
347 if (error)
348 goto out;
349
350 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
351 if (error)
352 goto out_unlock;
353
354 spin_lock(&sdp->sd_statfs_spin);
355 gfs2_statfs_change_in(m_sc, m_bh->b_data +
356 sizeof(struct gfs2_dinode));
357 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
358 spin_unlock(&sdp->sd_statfs_spin);
359 goto out_bh;
360 }
361 spin_unlock(&sdp->sd_statfs_spin);
362
363 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
364 if (error)
365 goto out_bh;
366
367 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
368 if (error)
369 goto out_bh2;
370
371 update_statfs(sdp, m_bh, l_bh);
372 sdp->sd_statfs_force_sync = 0;
373
374 gfs2_trans_end(sdp);
375
376out_bh2:
377 brelse(l_bh);
378out_bh:
379 brelse(m_bh);
380out_unlock:
381 gfs2_glock_dq_uninit(&gh);
382out:
383 sb_end_write(sb);
384 return error;
385}
386
387struct lfcc {
388 struct list_head list;
389 struct gfs2_holder gh;
390};
391
392/**
393 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
394 * journals are clean
395 * @sdp: the file system
396 * @state: the state to put the transaction lock into
397 * @t_gh: the hold on the transaction lock
398 *
399 * Returns: errno
400 */
401
402static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
403 struct gfs2_holder *freeze_gh)
404{
405 struct gfs2_inode *ip;
406 struct gfs2_jdesc *jd;
407 struct lfcc *lfcc;
408 LIST_HEAD(list);
409 struct gfs2_log_header_host lh;
410 int error;
411
412 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
413 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
414 if (!lfcc) {
415 error = -ENOMEM;
416 goto out;
417 }
418 ip = GFS2_I(jd->jd_inode);
419 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
420 if (error) {
421 kfree(lfcc);
422 goto out;
423 }
424 list_add(&lfcc->list, &list);
425 }
426
427 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
428 GL_NOCACHE, freeze_gh);
429
430 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
431 error = gfs2_jdesc_check(jd);
432 if (error)
433 break;
David Brazdil0f672f62019-12-10 10:32:29 +0000434 error = gfs2_find_jhead(jd, &lh, false);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000435 if (error)
436 break;
437 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
438 error = -EBUSY;
439 break;
440 }
441 }
442
443 if (error)
444 gfs2_glock_dq_uninit(freeze_gh);
445
446out:
447 while (!list_empty(&list)) {
448 lfcc = list_entry(list.next, struct lfcc, list);
449 list_del(&lfcc->list);
450 gfs2_glock_dq_uninit(&lfcc->gh);
451 kfree(lfcc);
452 }
453 return error;
454}
455
456void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
457{
458 struct gfs2_dinode *str = buf;
459
460 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
461 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
462 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
463 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
464 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
465 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
466 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
467 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
468 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
469 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
470 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
471 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
472 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
473 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
474
475 str->di_goal_meta = cpu_to_be64(ip->i_goal);
476 str->di_goal_data = cpu_to_be64(ip->i_goal);
477 str->di_generation = cpu_to_be64(ip->i_generation);
478
479 str->di_flags = cpu_to_be32(ip->i_diskflags);
480 str->di_height = cpu_to_be16(ip->i_height);
481 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
482 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
483 GFS2_FORMAT_DE : 0);
484 str->di_depth = cpu_to_be16(ip->i_depth);
485 str->di_entries = cpu_to_be32(ip->i_entries);
486
487 str->di_eattr = cpu_to_be64(ip->i_eattr);
488 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
489 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
490 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
491}
492
493/**
494 * gfs2_write_inode - Make sure the inode is stable on the disk
495 * @inode: The inode
496 * @wbc: The writeback control structure
497 *
498 * Returns: errno
499 */
500
501static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
502{
503 struct gfs2_inode *ip = GFS2_I(inode);
504 struct gfs2_sbd *sdp = GFS2_SB(inode);
505 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
506 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
507 int ret = 0;
508 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
509
510 if (flush_all)
511 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
512 GFS2_LOG_HEAD_FLUSH_NORMAL |
513 GFS2_LFC_WRITE_INODE);
514 if (bdi->wb.dirty_exceeded)
515 gfs2_ail1_flush(sdp, wbc);
516 else
517 filemap_fdatawrite(metamapping);
518 if (flush_all)
519 ret = filemap_fdatawait(metamapping);
520 if (ret)
521 mark_inode_dirty_sync(inode);
522 else {
523 spin_lock(&inode->i_lock);
524 if (!(inode->i_flags & I_DIRTY))
525 gfs2_ordered_del_inode(ip);
526 spin_unlock(&inode->i_lock);
527 }
528 return ret;
529}
530
531/**
532 * gfs2_dirty_inode - check for atime updates
533 * @inode: The inode in question
534 * @flags: The type of dirty
535 *
536 * Unfortunately it can be called under any combination of inode
537 * glock and transaction lock, so we have to check carefully.
538 *
539 * At the moment this deals only with atime - it should be possible
540 * to expand that role in future, once a review of the locking has
541 * been carried out.
542 */
543
544static void gfs2_dirty_inode(struct inode *inode, int flags)
545{
546 struct gfs2_inode *ip = GFS2_I(inode);
547 struct gfs2_sbd *sdp = GFS2_SB(inode);
548 struct buffer_head *bh;
549 struct gfs2_holder gh;
550 int need_unlock = 0;
551 int need_endtrans = 0;
552 int ret;
553
554 if (!(flags & I_DIRTY_INODE))
555 return;
David Brazdil0f672f62019-12-10 10:32:29 +0000556 if (unlikely(test_bit(SDF_WITHDRAWN, &sdp->sd_flags)))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000557 return;
558 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
559 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
560 if (ret) {
561 fs_err(sdp, "dirty_inode: glock %d\n", ret);
562 return;
563 }
564 need_unlock = 1;
565 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
566 return;
567
568 if (current->journal_info == NULL) {
569 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
570 if (ret) {
571 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
572 goto out;
573 }
574 need_endtrans = 1;
575 }
576
577 ret = gfs2_meta_inode_buffer(ip, &bh);
578 if (ret == 0) {
579 gfs2_trans_add_meta(ip->i_gl, bh);
580 gfs2_dinode_out(ip, bh->b_data);
581 brelse(bh);
582 }
583
584 if (need_endtrans)
585 gfs2_trans_end(sdp);
586out:
587 if (need_unlock)
588 gfs2_glock_dq_uninit(&gh);
589}
590
591/**
592 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
593 * @sdp: the filesystem
594 *
595 * Returns: errno
596 */
597
David Brazdil0f672f62019-12-10 10:32:29 +0000598int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000599{
600 struct gfs2_holder freeze_gh;
601 int error;
602
603 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, GL_NOCACHE,
604 &freeze_gh);
David Brazdil0f672f62019-12-10 10:32:29 +0000605 if (error && !test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000606 return error;
607
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000608 flush_workqueue(gfs2_delete_workqueue);
David Brazdil0f672f62019-12-10 10:32:29 +0000609 if (sdp->sd_quotad_process)
610 kthread_stop(sdp->sd_quotad_process);
611 sdp->sd_quotad_process = NULL;
612 if (sdp->sd_logd_process)
613 kthread_stop(sdp->sd_logd_process);
614 sdp->sd_logd_process = NULL;
615
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000616 gfs2_quota_sync(sdp->sd_vfs, 0);
617 gfs2_statfs_sync(sdp->sd_vfs, 0);
618
619 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
620 GFS2_LFC_MAKE_FS_RO);
621 wait_event(sdp->sd_reserving_log_wait, atomic_read(&sdp->sd_reserving_log) == 0);
622 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) == sdp->sd_jdesc->jd_blocks);
623
624 if (gfs2_holder_initialized(&freeze_gh))
625 gfs2_glock_dq_uninit(&freeze_gh);
626
627 gfs2_quota_cleanup(sdp);
628
629 return error;
630}
631
632/**
633 * gfs2_put_super - Unmount the filesystem
634 * @sb: The VFS superblock
635 *
636 */
637
638static void gfs2_put_super(struct super_block *sb)
639{
640 struct gfs2_sbd *sdp = sb->s_fs_info;
641 int error;
642 struct gfs2_jdesc *jd;
643
644 /* No more recovery requests */
645 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
646 smp_mb();
647
648 /* Wait on outstanding recovery */
649restart:
650 spin_lock(&sdp->sd_jindex_spin);
651 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
652 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
653 continue;
654 spin_unlock(&sdp->sd_jindex_spin);
655 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
656 TASK_UNINTERRUPTIBLE);
657 goto restart;
658 }
659 spin_unlock(&sdp->sd_jindex_spin);
660
661 if (!sb_rdonly(sb)) {
662 error = gfs2_make_fs_ro(sdp);
663 if (error)
664 gfs2_io_error(sdp);
665 }
666 /* At this point, we're through modifying the disk */
667
668 /* Release stuff */
669
670 iput(sdp->sd_jindex);
671 iput(sdp->sd_statfs_inode);
672 iput(sdp->sd_rindex);
673 iput(sdp->sd_quota_inode);
674
675 gfs2_glock_put(sdp->sd_rename_gl);
676 gfs2_glock_put(sdp->sd_freeze_gl);
677
678 if (!sdp->sd_args.ar_spectator) {
679 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
680 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
681 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
682 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
683 iput(sdp->sd_sc_inode);
684 iput(sdp->sd_qc_inode);
685 }
686
687 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
688 gfs2_clear_rgrpd(sdp);
689 gfs2_jindex_free(sdp);
690 /* Take apart glock structures and buffer lists */
691 gfs2_gl_hash_clear(sdp);
Olivier Deprez0e641232021-09-23 10:07:05 +0200692 truncate_inode_pages_final(&sdp->sd_aspace);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000693 gfs2_delete_debugfs_file(sdp);
694 /* Unmount the locking protocol */
695 gfs2_lm_unmount(sdp);
696
697 /* At this point, we're through participating in the lockspace */
698 gfs2_sys_fs_del(sdp);
Olivier Deprez0e641232021-09-23 10:07:05 +0200699 free_sbd(sdp);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000700}
701
702/**
703 * gfs2_sync_fs - sync the filesystem
704 * @sb: the superblock
705 *
706 * Flushes the log to disk.
707 */
708
709static int gfs2_sync_fs(struct super_block *sb, int wait)
710{
711 struct gfs2_sbd *sdp = sb->s_fs_info;
712
713 gfs2_quota_sync(sb, -1);
714 if (wait)
715 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
716 GFS2_LFC_SYNC_FS);
717 return sdp->sd_log_error;
718}
719
720void gfs2_freeze_func(struct work_struct *work)
721{
722 int error;
723 struct gfs2_holder freeze_gh;
724 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
725 struct super_block *sb = sdp->sd_vfs;
726
727 atomic_inc(&sb->s_active);
728 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED, 0,
729 &freeze_gh);
730 if (error) {
David Brazdil0f672f62019-12-10 10:32:29 +0000731 fs_info(sdp, "GFS2: couldn't get freeze lock : %d\n", error);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000732 gfs2_assert_withdraw(sdp, 0);
David Brazdil0f672f62019-12-10 10:32:29 +0000733 } else {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000734 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
735 error = thaw_super(sb);
736 if (error) {
David Brazdil0f672f62019-12-10 10:32:29 +0000737 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
738 error);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000739 gfs2_assert_withdraw(sdp, 0);
740 }
741 if (!test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags))
742 freeze_gh.gh_flags |= GL_NOCACHE;
743 gfs2_glock_dq_uninit(&freeze_gh);
744 }
745 deactivate_super(sb);
David Brazdil0f672f62019-12-10 10:32:29 +0000746 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
747 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000748 return;
749}
750
751/**
752 * gfs2_freeze - prevent further writes to the filesystem
753 * @sb: the VFS structure for the filesystem
754 *
755 */
756
757static int gfs2_freeze(struct super_block *sb)
758{
759 struct gfs2_sbd *sdp = sb->s_fs_info;
Olivier Deprez0e641232021-09-23 10:07:05 +0200760 int error;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000761
762 mutex_lock(&sdp->sd_freeze_mutex);
Olivier Deprez0e641232021-09-23 10:07:05 +0200763 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
764 error = -EBUSY;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000765 goto out;
Olivier Deprez0e641232021-09-23 10:07:05 +0200766 }
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000767
David Brazdil0f672f62019-12-10 10:32:29 +0000768 if (test_bit(SDF_WITHDRAWN, &sdp->sd_flags)) {
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000769 error = -EINVAL;
770 goto out;
771 }
772
773 for (;;) {
774 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
775 if (!error)
776 break;
777
David Brazdil0f672f62019-12-10 10:32:29 +0000778 if (error == -EBUSY)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000779 fs_err(sdp, "waiting for recovery before freeze\n");
David Brazdil0f672f62019-12-10 10:32:29 +0000780 else
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000781 fs_err(sdp, "error freezing FS: %d\n", error);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000782
783 fs_err(sdp, "retrying...\n");
784 msleep(1000);
785 }
David Brazdil0f672f62019-12-10 10:32:29 +0000786 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000787out:
788 mutex_unlock(&sdp->sd_freeze_mutex);
789 return error;
790}
791
792/**
793 * gfs2_unfreeze - reallow writes to the filesystem
794 * @sb: the VFS structure for the filesystem
795 *
796 */
797
798static int gfs2_unfreeze(struct super_block *sb)
799{
800 struct gfs2_sbd *sdp = sb->s_fs_info;
801
802 mutex_lock(&sdp->sd_freeze_mutex);
Olivier Deprez0e641232021-09-23 10:07:05 +0200803 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000804 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
805 mutex_unlock(&sdp->sd_freeze_mutex);
Olivier Deprez0e641232021-09-23 10:07:05 +0200806 return -EINVAL;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000807 }
808
809 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
810 mutex_unlock(&sdp->sd_freeze_mutex);
David Brazdil0f672f62019-12-10 10:32:29 +0000811 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000812}
813
814/**
815 * statfs_fill - fill in the sg for a given RG
816 * @rgd: the RG
817 * @sc: the sc structure
818 *
819 * Returns: 0 on success, -ESTALE if the LVB is invalid
820 */
821
822static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
823 struct gfs2_statfs_change_host *sc)
824{
825 gfs2_rgrp_verify(rgd);
826 sc->sc_total += rgd->rd_data;
827 sc->sc_free += rgd->rd_free;
828 sc->sc_dinodes += rgd->rd_dinodes;
829 return 0;
830}
831
832/**
833 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
834 * @sdp: the filesystem
835 * @sc: the sc info that will be returned
836 *
837 * Any error (other than a signal) will cause this routine to fall back
838 * to the synchronous version.
839 *
840 * FIXME: This really shouldn't busy wait like this.
841 *
842 * Returns: errno
843 */
844
845static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
846{
847 struct gfs2_rgrpd *rgd_next;
848 struct gfs2_holder *gha, *gh;
849 unsigned int slots = 64;
850 unsigned int x;
851 int done;
852 int error = 0, err;
853
854 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
855 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
856 if (!gha)
857 return -ENOMEM;
858 for (x = 0; x < slots; x++)
859 gfs2_holder_mark_uninitialized(gha + x);
860
861 rgd_next = gfs2_rgrpd_get_first(sdp);
862
863 for (;;) {
864 done = 1;
865
866 for (x = 0; x < slots; x++) {
867 gh = gha + x;
868
869 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
870 err = gfs2_glock_wait(gh);
871 if (err) {
872 gfs2_holder_uninit(gh);
873 error = err;
874 } else {
875 if (!error) {
876 struct gfs2_rgrpd *rgd =
877 gfs2_glock2rgrp(gh->gh_gl);
878
879 error = statfs_slow_fill(rgd, sc);
880 }
881 gfs2_glock_dq_uninit(gh);
882 }
883 }
884
885 if (gfs2_holder_initialized(gh))
886 done = 0;
887 else if (rgd_next && !error) {
888 error = gfs2_glock_nq_init(rgd_next->rd_gl,
889 LM_ST_SHARED,
890 GL_ASYNC,
891 gh);
892 rgd_next = gfs2_rgrpd_get_next(rgd_next);
893 done = 0;
894 }
895
896 if (signal_pending(current))
897 error = -ERESTARTSYS;
898 }
899
900 if (done)
901 break;
902
903 yield();
904 }
905
906 kfree(gha);
907 return error;
908}
909
910/**
911 * gfs2_statfs_i - Do a statfs
912 * @sdp: the filesystem
913 * @sg: the sg structure
914 *
915 * Returns: errno
916 */
917
918static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
919{
920 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
921 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
922
923 spin_lock(&sdp->sd_statfs_spin);
924
925 *sc = *m_sc;
926 sc->sc_total += l_sc->sc_total;
927 sc->sc_free += l_sc->sc_free;
928 sc->sc_dinodes += l_sc->sc_dinodes;
929
930 spin_unlock(&sdp->sd_statfs_spin);
931
932 if (sc->sc_free < 0)
933 sc->sc_free = 0;
934 if (sc->sc_free > sc->sc_total)
935 sc->sc_free = sc->sc_total;
936 if (sc->sc_dinodes < 0)
937 sc->sc_dinodes = 0;
938
939 return 0;
940}
941
942/**
943 * gfs2_statfs - Gather and return stats about the filesystem
944 * @sb: The superblock
945 * @statfsbuf: The buffer
946 *
947 * Returns: 0 on success or error code
948 */
949
950static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
951{
952 struct super_block *sb = dentry->d_sb;
953 struct gfs2_sbd *sdp = sb->s_fs_info;
954 struct gfs2_statfs_change_host sc;
955 int error;
956
957 error = gfs2_rindex_update(sdp);
958 if (error)
959 return error;
960
961 if (gfs2_tune_get(sdp, gt_statfs_slow))
962 error = gfs2_statfs_slow(sdp, &sc);
963 else
964 error = gfs2_statfs_i(sdp, &sc);
965
966 if (error)
967 return error;
968
969 buf->f_type = GFS2_MAGIC;
970 buf->f_bsize = sdp->sd_sb.sb_bsize;
971 buf->f_blocks = sc.sc_total;
972 buf->f_bfree = sc.sc_free;
973 buf->f_bavail = sc.sc_free;
974 buf->f_files = sc.sc_dinodes + sc.sc_free;
975 buf->f_ffree = sc.sc_free;
976 buf->f_namelen = GFS2_FNAMESIZE;
977
978 return 0;
979}
980
981/**
Andrew Scullb4b6d4a2019-01-02 15:54:55 +0000982 * gfs2_drop_inode - Drop an inode (test for remote unlink)
983 * @inode: The inode to drop
984 *
985 * If we've received a callback on an iopen lock then it's because a
986 * remote node tried to deallocate the inode but failed due to this node
987 * still having the inode open. Here we mark the link count zero
988 * since we know that it must have reached zero if the GLF_DEMOTE flag
989 * is set on the iopen glock. If we didn't do a disk read since the
990 * remote node removed the final link then we might otherwise miss
991 * this event. This check ensures that this node will deallocate the
992 * inode's blocks, or alternatively pass the baton on to another
993 * node for later deallocation.
994 */
995
996static int gfs2_drop_inode(struct inode *inode)
997{
998 struct gfs2_inode *ip = GFS2_I(inode);
999
1000 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
1001 inode->i_nlink &&
1002 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1003 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1004 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
1005 clear_nlink(inode);
1006 }
1007
1008 /*
1009 * When under memory pressure when an inode's link count has dropped to
1010 * zero, defer deleting the inode to the delete workqueue. This avoids
1011 * calling into DLM under memory pressure, which can deadlock.
1012 */
1013 if (!inode->i_nlink &&
1014 unlikely(current->flags & PF_MEMALLOC) &&
1015 gfs2_holder_initialized(&ip->i_iopen_gh)) {
1016 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1017
1018 gfs2_glock_hold(gl);
1019 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
1020 gfs2_glock_queue_put(gl);
1021 return false;
1022 }
1023
1024 return generic_drop_inode(inode);
1025}
1026
1027static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
1028{
1029 do {
1030 if (d1 == d2)
1031 return 1;
1032 d1 = d1->d_parent;
1033 } while (!IS_ROOT(d1));
1034 return 0;
1035}
1036
1037/**
1038 * gfs2_show_options - Show mount options for /proc/mounts
1039 * @s: seq_file structure
1040 * @root: root of this (sub)tree
1041 *
1042 * Returns: 0 on success or error code
1043 */
1044
1045static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1046{
1047 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1048 struct gfs2_args *args = &sdp->sd_args;
1049 int val;
1050
1051 if (is_ancestor(root, sdp->sd_master_dir))
1052 seq_puts(s, ",meta");
1053 if (args->ar_lockproto[0])
1054 seq_show_option(s, "lockproto", args->ar_lockproto);
1055 if (args->ar_locktable[0])
1056 seq_show_option(s, "locktable", args->ar_locktable);
1057 if (args->ar_hostdata[0])
1058 seq_show_option(s, "hostdata", args->ar_hostdata);
1059 if (args->ar_spectator)
1060 seq_puts(s, ",spectator");
1061 if (args->ar_localflocks)
1062 seq_puts(s, ",localflocks");
1063 if (args->ar_debug)
1064 seq_puts(s, ",debug");
1065 if (args->ar_posix_acl)
1066 seq_puts(s, ",acl");
1067 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1068 char *state;
1069 switch (args->ar_quota) {
1070 case GFS2_QUOTA_OFF:
1071 state = "off";
1072 break;
1073 case GFS2_QUOTA_ACCOUNT:
1074 state = "account";
1075 break;
1076 case GFS2_QUOTA_ON:
1077 state = "on";
1078 break;
1079 default:
1080 state = "unknown";
1081 break;
1082 }
1083 seq_printf(s, ",quota=%s", state);
1084 }
1085 if (args->ar_suiddir)
1086 seq_puts(s, ",suiddir");
1087 if (args->ar_data != GFS2_DATA_DEFAULT) {
1088 char *state;
1089 switch (args->ar_data) {
1090 case GFS2_DATA_WRITEBACK:
1091 state = "writeback";
1092 break;
1093 case GFS2_DATA_ORDERED:
1094 state = "ordered";
1095 break;
1096 default:
1097 state = "unknown";
1098 break;
1099 }
1100 seq_printf(s, ",data=%s", state);
1101 }
1102 if (args->ar_discard)
1103 seq_puts(s, ",discard");
1104 val = sdp->sd_tune.gt_logd_secs;
1105 if (val != 30)
1106 seq_printf(s, ",commit=%d", val);
1107 val = sdp->sd_tune.gt_statfs_quantum;
1108 if (val != 30)
1109 seq_printf(s, ",statfs_quantum=%d", val);
1110 else if (sdp->sd_tune.gt_statfs_slow)
1111 seq_puts(s, ",statfs_quantum=0");
1112 val = sdp->sd_tune.gt_quota_quantum;
1113 if (val != 60)
1114 seq_printf(s, ",quota_quantum=%d", val);
1115 if (args->ar_statfs_percent)
1116 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1117 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1118 const char *state;
1119
1120 switch (args->ar_errors) {
1121 case GFS2_ERRORS_WITHDRAW:
1122 state = "withdraw";
1123 break;
1124 case GFS2_ERRORS_PANIC:
1125 state = "panic";
1126 break;
1127 default:
1128 state = "unknown";
1129 break;
1130 }
1131 seq_printf(s, ",errors=%s", state);
1132 }
1133 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1134 seq_puts(s, ",nobarrier");
1135 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1136 seq_puts(s, ",demote_interface_used");
1137 if (args->ar_rgrplvb)
1138 seq_puts(s, ",rgrplvb");
1139 if (args->ar_loccookie)
1140 seq_puts(s, ",loccookie");
1141 return 0;
1142}
1143
1144static void gfs2_final_release_pages(struct gfs2_inode *ip)
1145{
1146 struct inode *inode = &ip->i_inode;
1147 struct gfs2_glock *gl = ip->i_gl;
1148
1149 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1150 truncate_inode_pages(&inode->i_data, 0);
1151
1152 if (atomic_read(&gl->gl_revokes) == 0) {
1153 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1154 clear_bit(GLF_DIRTY, &gl->gl_flags);
1155 }
1156}
1157
1158static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1159{
1160 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1161 struct gfs2_rgrpd *rgd;
1162 struct gfs2_holder gh;
1163 int error;
1164
1165 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1166 gfs2_consist_inode(ip);
1167 return -EIO;
1168 }
1169
1170 error = gfs2_rindex_update(sdp);
1171 if (error)
1172 return error;
1173
1174 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1175 if (error)
1176 return error;
1177
1178 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1179 if (!rgd) {
1180 gfs2_consist_inode(ip);
1181 error = -EIO;
1182 goto out_qs;
1183 }
1184
1185 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1186 if (error)
1187 goto out_qs;
1188
1189 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1190 sdp->sd_jdesc->jd_blocks);
1191 if (error)
1192 goto out_rg_gunlock;
1193
1194 gfs2_free_di(rgd, ip);
1195
1196 gfs2_final_release_pages(ip);
1197
1198 gfs2_trans_end(sdp);
1199
1200out_rg_gunlock:
1201 gfs2_glock_dq_uninit(&gh);
1202out_qs:
1203 gfs2_quota_unhold(ip);
1204 return error;
1205}
1206
1207/**
1208 * gfs2_glock_put_eventually
1209 * @gl: The glock to put
1210 *
1211 * When under memory pressure, trigger a deferred glock put to make sure we
1212 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1213 */
1214
1215static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1216{
1217 if (current->flags & PF_MEMALLOC)
1218 gfs2_glock_queue_put(gl);
1219 else
1220 gfs2_glock_put(gl);
1221}
1222
1223/**
1224 * gfs2_evict_inode - Remove an inode from cache
1225 * @inode: The inode to evict
1226 *
1227 * There are three cases to consider:
1228 * 1. i_nlink == 0, we are final opener (and must deallocate)
1229 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1230 * 3. i_nlink > 0
1231 *
1232 * If the fs is read only, then we have to treat all cases as per #3
1233 * since we are unable to do any deallocation. The inode will be
1234 * deallocated by the next read/write node to attempt an allocation
1235 * in the same resource group
1236 *
1237 * We have to (at the moment) hold the inodes main lock to cover
1238 * the gap between unlocking the shared lock on the iopen lock and
1239 * taking the exclusive lock. I'd rather do a shared -> exclusive
1240 * conversion on the iopen lock, but we can change that later. This
1241 * is safe, just less efficient.
1242 */
1243
1244static void gfs2_evict_inode(struct inode *inode)
1245{
1246 struct super_block *sb = inode->i_sb;
1247 struct gfs2_sbd *sdp = sb->s_fs_info;
1248 struct gfs2_inode *ip = GFS2_I(inode);
1249 struct gfs2_holder gh;
1250 struct address_space *metamapping;
1251 int error;
1252
1253 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1254 clear_inode(inode);
1255 return;
1256 }
1257
1258 if (inode->i_nlink || sb_rdonly(sb))
1259 goto out;
1260
1261 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1262 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1263 gfs2_holder_mark_uninitialized(&gh);
1264 goto alloc_failed;
1265 }
1266
1267 /* Deletes should never happen under memory pressure anymore. */
1268 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1269 goto out;
1270
1271 /* Must not read inode block until block type has been verified */
1272 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, &gh);
1273 if (unlikely(error)) {
1274 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1275 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1276 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1277 goto out;
1278 }
1279
1280 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1281 if (error)
1282 goto out_truncate;
1283
1284 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1285 error = gfs2_inode_refresh(ip);
1286 if (error)
1287 goto out_truncate;
1288 }
1289
1290 /*
1291 * The inode may have been recreated in the meantime.
1292 */
1293 if (inode->i_nlink)
1294 goto out_truncate;
1295
1296alloc_failed:
1297 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1298 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1299 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1300 gfs2_glock_dq_wait(&ip->i_iopen_gh);
1301 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE,
1302 &ip->i_iopen_gh);
1303 error = gfs2_glock_nq(&ip->i_iopen_gh);
1304 if (error)
1305 goto out_truncate;
1306 }
1307
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001308 if (S_ISDIR(inode->i_mode) &&
1309 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1310 error = gfs2_dir_exhash_dealloc(ip);
1311 if (error)
1312 goto out_unlock;
1313 }
1314
1315 if (ip->i_eattr) {
1316 error = gfs2_ea_dealloc(ip);
1317 if (error)
1318 goto out_unlock;
1319 }
1320
1321 if (!gfs2_is_stuffed(ip)) {
1322 error = gfs2_file_dealloc(ip);
1323 if (error)
1324 goto out_unlock;
1325 }
1326
1327 /* We're about to clear the bitmap for the dinode, but as soon as we
1328 do, gfs2_create_inode can create another inode at the same block
1329 location and try to set gl_object again. We clear gl_object here so
1330 that subsequent inode creates don't see an old gl_object. */
1331 glock_clear_object(ip->i_gl, ip);
1332 error = gfs2_dinode_dealloc(ip);
1333 goto out_unlock;
1334
1335out_truncate:
1336 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1337 GFS2_LFC_EVICT_INODE);
1338 metamapping = gfs2_glock2aspace(ip->i_gl);
1339 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1340 filemap_fdatawrite(metamapping);
1341 filemap_fdatawait(metamapping);
1342 }
1343 write_inode_now(inode, 1);
1344 gfs2_ail_flush(ip->i_gl, 0);
1345
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001346 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1347 if (error)
1348 goto out_unlock;
1349 /* Needs to be done before glock release & also in a transaction */
1350 truncate_inode_pages(&inode->i_data, 0);
1351 truncate_inode_pages(metamapping, 0);
1352 gfs2_trans_end(sdp);
1353
1354out_unlock:
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001355 if (gfs2_rs_active(&ip->i_res))
1356 gfs2_rs_deltree(&ip->i_res);
1357
1358 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1359 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1360 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1361 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1362 gfs2_glock_dq(&ip->i_iopen_gh);
1363 }
1364 gfs2_holder_uninit(&ip->i_iopen_gh);
1365 }
1366 if (gfs2_holder_initialized(&gh)) {
1367 glock_clear_object(ip->i_gl, ip);
1368 gfs2_glock_dq_uninit(&gh);
1369 }
1370 if (error && error != GLR_TRYFAILED && error != -EROFS)
1371 fs_warn(sdp, "gfs2_evict_inode: %d\n", error);
1372out:
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001373 truncate_inode_pages_final(&inode->i_data);
1374 gfs2_rsqa_delete(ip, NULL);
1375 gfs2_ordered_del_inode(ip);
1376 clear_inode(inode);
1377 gfs2_dir_hash_inval(ip);
1378 glock_clear_object(ip->i_gl, ip);
1379 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1380 gfs2_glock_add_to_lru(ip->i_gl);
1381 gfs2_glock_put_eventually(ip->i_gl);
1382 ip->i_gl = NULL;
1383 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1384 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1385
1386 glock_clear_object(gl, ip);
1387 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1388 gfs2_glock_hold(gl);
1389 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1390 gfs2_glock_put_eventually(gl);
1391 }
1392}
1393
1394static struct inode *gfs2_alloc_inode(struct super_block *sb)
1395{
1396 struct gfs2_inode *ip;
1397
1398 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
David Brazdil0f672f62019-12-10 10:32:29 +00001399 if (!ip)
1400 return NULL;
1401 ip->i_flags = 0;
1402 ip->i_gl = NULL;
1403 memset(&ip->i_res, 0, sizeof(ip->i_res));
1404 RB_CLEAR_NODE(&ip->i_res.rs_node);
1405 ip->i_rahead = 0;
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001406 return &ip->i_inode;
1407}
1408
David Brazdil0f672f62019-12-10 10:32:29 +00001409static void gfs2_free_inode(struct inode *inode)
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001410{
David Brazdil0f672f62019-12-10 10:32:29 +00001411 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001412}
1413
1414const struct super_operations gfs2_super_ops = {
1415 .alloc_inode = gfs2_alloc_inode,
David Brazdil0f672f62019-12-10 10:32:29 +00001416 .free_inode = gfs2_free_inode,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001417 .write_inode = gfs2_write_inode,
1418 .dirty_inode = gfs2_dirty_inode,
1419 .evict_inode = gfs2_evict_inode,
1420 .put_super = gfs2_put_super,
1421 .sync_fs = gfs2_sync_fs,
1422 .freeze_super = gfs2_freeze,
1423 .thaw_super = gfs2_unfreeze,
1424 .statfs = gfs2_statfs,
Andrew Scullb4b6d4a2019-01-02 15:54:55 +00001425 .drop_inode = gfs2_drop_inode,
1426 .show_options = gfs2_show_options,
1427};
1428