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
gfs2_jindex_free(struct gfs2_sbd * sdp)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 sdp->sd_jdesc = NULL;
71 while (!list_empty(&list)) {
72 jd = list_first_entry(&list, struct gfs2_jdesc, jd_list);
73 gfs2_free_journal_extents(jd);
74 list_del(&jd->jd_list);
75 iput(jd->jd_inode);
76 jd->jd_inode = NULL;
77 kfree(jd);
78 }
79 }
80
jdesc_find_i(struct list_head * head,unsigned int jid)81 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
82 {
83 struct gfs2_jdesc *jd;
84 int found = 0;
85
86 list_for_each_entry(jd, head, jd_list) {
87 if (jd->jd_jid == jid) {
88 found = 1;
89 break;
90 }
91 }
92
93 if (!found)
94 jd = NULL;
95
96 return jd;
97 }
98
gfs2_jdesc_find(struct gfs2_sbd * sdp,unsigned int jid)99 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
100 {
101 struct gfs2_jdesc *jd;
102
103 spin_lock(&sdp->sd_jindex_spin);
104 jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
105 spin_unlock(&sdp->sd_jindex_spin);
106
107 return jd;
108 }
109
gfs2_jdesc_check(struct gfs2_jdesc * jd)110 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
111 {
112 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
113 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
114 u64 size = i_size_read(jd->jd_inode);
115
116 if (gfs2_check_internal_file_size(jd->jd_inode, 8 << 20, BIT(30)))
117 return -EIO;
118
119 jd->jd_blocks = size >> sdp->sd_sb.sb_bsize_shift;
120
121 if (gfs2_write_alloc_required(ip, 0, size)) {
122 gfs2_consist_inode(ip);
123 return -EIO;
124 }
125
126 return 0;
127 }
128
129 /**
130 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
131 * @sdp: the filesystem
132 *
133 * Returns: errno
134 */
135
gfs2_make_fs_rw(struct gfs2_sbd * sdp)136 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
137 {
138 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
139 struct gfs2_glock *j_gl = ip->i_gl;
140 struct gfs2_log_header_host head;
141 int error;
142
143 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
144 if (gfs2_withdrawn(sdp))
145 return -EIO;
146
147 error = gfs2_find_jhead(sdp->sd_jdesc, &head, false);
148 if (error || gfs2_withdrawn(sdp))
149 return error;
150
151 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
152 gfs2_consist(sdp);
153 return -EIO;
154 }
155
156 /* Initialize some head of the log stuff */
157 sdp->sd_log_sequence = head.lh_sequence + 1;
158 gfs2_log_pointers_init(sdp, head.lh_blkno);
159
160 error = gfs2_quota_init(sdp);
161 if (!error && !gfs2_withdrawn(sdp))
162 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
163 return error;
164 }
165
gfs2_statfs_change_in(struct gfs2_statfs_change_host * sc,const void * buf)166 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
167 {
168 const struct gfs2_statfs_change *str = buf;
169
170 sc->sc_total = be64_to_cpu(str->sc_total);
171 sc->sc_free = be64_to_cpu(str->sc_free);
172 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
173 }
174
gfs2_statfs_change_out(const struct gfs2_statfs_change_host * sc,void * buf)175 void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
176 {
177 struct gfs2_statfs_change *str = buf;
178
179 str->sc_total = cpu_to_be64(sc->sc_total);
180 str->sc_free = cpu_to_be64(sc->sc_free);
181 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
182 }
183
gfs2_statfs_init(struct gfs2_sbd * sdp)184 int gfs2_statfs_init(struct gfs2_sbd *sdp)
185 {
186 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
187 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
188 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
189 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
190 struct buffer_head *m_bh, *l_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 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
210 if (error)
211 goto out_m_bh;
212
213 spin_lock(&sdp->sd_statfs_spin);
214 gfs2_statfs_change_in(m_sc, m_bh->b_data +
215 sizeof(struct gfs2_dinode));
216 gfs2_statfs_change_in(l_sc, l_bh->b_data +
217 sizeof(struct gfs2_dinode));
218 spin_unlock(&sdp->sd_statfs_spin);
219
220 brelse(l_bh);
221 }
222
223 out_m_bh:
224 brelse(m_bh);
225 out:
226 gfs2_glock_dq_uninit(&gh);
227 return 0;
228 }
229
gfs2_statfs_change(struct gfs2_sbd * sdp,s64 total,s64 free,s64 dinodes)230 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
231 s64 dinodes)
232 {
233 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
234 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
235 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
236 struct buffer_head *l_bh;
237 s64 x, y;
238 int need_sync = 0;
239 int error;
240
241 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
242 if (error)
243 return;
244
245 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
246
247 spin_lock(&sdp->sd_statfs_spin);
248 l_sc->sc_total += total;
249 l_sc->sc_free += free;
250 l_sc->sc_dinodes += dinodes;
251 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
252 if (sdp->sd_args.ar_statfs_percent) {
253 x = 100 * l_sc->sc_free;
254 y = m_sc->sc_free * sdp->sd_args.ar_statfs_percent;
255 if (x >= y || x <= -y)
256 need_sync = 1;
257 }
258 spin_unlock(&sdp->sd_statfs_spin);
259
260 brelse(l_bh);
261 if (need_sync)
262 gfs2_wake_up_statfs(sdp);
263 }
264
update_statfs(struct gfs2_sbd * sdp,struct buffer_head * m_bh,struct buffer_head * l_bh)265 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh,
266 struct buffer_head *l_bh)
267 {
268 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
269 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
270 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
271 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
272
273 gfs2_trans_add_meta(l_ip->i_gl, l_bh);
274 gfs2_trans_add_meta(m_ip->i_gl, m_bh);
275
276 spin_lock(&sdp->sd_statfs_spin);
277 m_sc->sc_total += l_sc->sc_total;
278 m_sc->sc_free += l_sc->sc_free;
279 m_sc->sc_dinodes += l_sc->sc_dinodes;
280 memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
281 memset(l_bh->b_data + sizeof(struct gfs2_dinode),
282 0, sizeof(struct gfs2_statfs_change));
283 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
284 spin_unlock(&sdp->sd_statfs_spin);
285 }
286
gfs2_statfs_sync(struct super_block * sb,int type)287 int gfs2_statfs_sync(struct super_block *sb, int type)
288 {
289 struct gfs2_sbd *sdp = sb->s_fs_info;
290 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
291 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
292 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
293 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
294 struct gfs2_holder gh;
295 struct buffer_head *m_bh, *l_bh;
296 int error;
297
298 sb_start_write(sb);
299 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
300 &gh);
301 if (error)
302 goto out;
303
304 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
305 if (error)
306 goto out_unlock;
307
308 spin_lock(&sdp->sd_statfs_spin);
309 gfs2_statfs_change_in(m_sc, m_bh->b_data +
310 sizeof(struct gfs2_dinode));
311 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
312 spin_unlock(&sdp->sd_statfs_spin);
313 goto out_bh;
314 }
315 spin_unlock(&sdp->sd_statfs_spin);
316
317 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
318 if (error)
319 goto out_bh;
320
321 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
322 if (error)
323 goto out_bh2;
324
325 update_statfs(sdp, m_bh, l_bh);
326 sdp->sd_statfs_force_sync = 0;
327
328 gfs2_trans_end(sdp);
329
330 out_bh2:
331 brelse(l_bh);
332 out_bh:
333 brelse(m_bh);
334 out_unlock:
335 gfs2_glock_dq_uninit(&gh);
336 out:
337 sb_end_write(sb);
338 return error;
339 }
340
341 struct lfcc {
342 struct list_head list;
343 struct gfs2_holder gh;
344 };
345
346 /**
347 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
348 * journals are clean
349 * @sdp: the file system
350 * @state: the state to put the transaction lock into
351 * @t_gh: the hold on the transaction lock
352 *
353 * Returns: errno
354 */
355
gfs2_lock_fs_check_clean(struct gfs2_sbd * sdp)356 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp)
357 {
358 struct gfs2_inode *ip;
359 struct gfs2_jdesc *jd;
360 struct lfcc *lfcc;
361 LIST_HEAD(list);
362 struct gfs2_log_header_host lh;
363 int error;
364
365 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
366 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
367 if (!lfcc) {
368 error = -ENOMEM;
369 goto out;
370 }
371 ip = GFS2_I(jd->jd_inode);
372 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
373 if (error) {
374 kfree(lfcc);
375 goto out;
376 }
377 list_add(&lfcc->list, &list);
378 }
379
380 error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_EXCLUSIVE,
381 LM_FLAG_NOEXP, &sdp->sd_freeze_gh);
382 if (error)
383 goto out;
384
385 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
386 error = gfs2_jdesc_check(jd);
387 if (error)
388 break;
389 error = gfs2_find_jhead(jd, &lh, false);
390 if (error)
391 break;
392 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
393 error = -EBUSY;
394 break;
395 }
396 }
397
398 if (error)
399 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
400
401 out:
402 while (!list_empty(&list)) {
403 lfcc = list_first_entry(&list, struct lfcc, list);
404 list_del(&lfcc->list);
405 gfs2_glock_dq_uninit(&lfcc->gh);
406 kfree(lfcc);
407 }
408 return error;
409 }
410
gfs2_dinode_out(const struct gfs2_inode * ip,void * buf)411 void gfs2_dinode_out(const struct gfs2_inode *ip, void *buf)
412 {
413 struct gfs2_dinode *str = buf;
414
415 str->di_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
416 str->di_header.mh_type = cpu_to_be32(GFS2_METATYPE_DI);
417 str->di_header.mh_format = cpu_to_be32(GFS2_FORMAT_DI);
418 str->di_num.no_addr = cpu_to_be64(ip->i_no_addr);
419 str->di_num.no_formal_ino = cpu_to_be64(ip->i_no_formal_ino);
420 str->di_mode = cpu_to_be32(ip->i_inode.i_mode);
421 str->di_uid = cpu_to_be32(i_uid_read(&ip->i_inode));
422 str->di_gid = cpu_to_be32(i_gid_read(&ip->i_inode));
423 str->di_nlink = cpu_to_be32(ip->i_inode.i_nlink);
424 str->di_size = cpu_to_be64(i_size_read(&ip->i_inode));
425 str->di_blocks = cpu_to_be64(gfs2_get_inode_blocks(&ip->i_inode));
426 str->di_atime = cpu_to_be64(ip->i_inode.i_atime.tv_sec);
427 str->di_mtime = cpu_to_be64(ip->i_inode.i_mtime.tv_sec);
428 str->di_ctime = cpu_to_be64(ip->i_inode.i_ctime.tv_sec);
429
430 str->di_goal_meta = cpu_to_be64(ip->i_goal);
431 str->di_goal_data = cpu_to_be64(ip->i_goal);
432 str->di_generation = cpu_to_be64(ip->i_generation);
433
434 str->di_flags = cpu_to_be32(ip->i_diskflags);
435 str->di_height = cpu_to_be16(ip->i_height);
436 str->di_payload_format = cpu_to_be32(S_ISDIR(ip->i_inode.i_mode) &&
437 !(ip->i_diskflags & GFS2_DIF_EXHASH) ?
438 GFS2_FORMAT_DE : 0);
439 str->di_depth = cpu_to_be16(ip->i_depth);
440 str->di_entries = cpu_to_be32(ip->i_entries);
441
442 str->di_eattr = cpu_to_be64(ip->i_eattr);
443 str->di_atime_nsec = cpu_to_be32(ip->i_inode.i_atime.tv_nsec);
444 str->di_mtime_nsec = cpu_to_be32(ip->i_inode.i_mtime.tv_nsec);
445 str->di_ctime_nsec = cpu_to_be32(ip->i_inode.i_ctime.tv_nsec);
446 }
447
448 /**
449 * gfs2_write_inode - Make sure the inode is stable on the disk
450 * @inode: The inode
451 * @wbc: The writeback control structure
452 *
453 * Returns: errno
454 */
455
gfs2_write_inode(struct inode * inode,struct writeback_control * wbc)456 static int gfs2_write_inode(struct inode *inode, struct writeback_control *wbc)
457 {
458 struct gfs2_inode *ip = GFS2_I(inode);
459 struct gfs2_sbd *sdp = GFS2_SB(inode);
460 struct address_space *metamapping = gfs2_glock2aspace(ip->i_gl);
461 struct backing_dev_info *bdi = inode_to_bdi(metamapping->host);
462 int ret = 0;
463 bool flush_all = (wbc->sync_mode == WB_SYNC_ALL || gfs2_is_jdata(ip));
464
465 if (flush_all)
466 gfs2_log_flush(GFS2_SB(inode), ip->i_gl,
467 GFS2_LOG_HEAD_FLUSH_NORMAL |
468 GFS2_LFC_WRITE_INODE);
469 if (bdi->wb.dirty_exceeded)
470 gfs2_ail1_flush(sdp, wbc);
471 else
472 filemap_fdatawrite(metamapping);
473 if (flush_all)
474 ret = filemap_fdatawait(metamapping);
475 if (ret)
476 mark_inode_dirty_sync(inode);
477 else {
478 spin_lock(&inode->i_lock);
479 if (!(inode->i_flags & I_DIRTY))
480 gfs2_ordered_del_inode(ip);
481 spin_unlock(&inode->i_lock);
482 }
483 return ret;
484 }
485
486 /**
487 * gfs2_dirty_inode - check for atime updates
488 * @inode: The inode in question
489 * @flags: The type of dirty
490 *
491 * Unfortunately it can be called under any combination of inode
492 * glock and transaction lock, so we have to check carefully.
493 *
494 * At the moment this deals only with atime - it should be possible
495 * to expand that role in future, once a review of the locking has
496 * been carried out.
497 */
498
gfs2_dirty_inode(struct inode * inode,int flags)499 static void gfs2_dirty_inode(struct inode *inode, int flags)
500 {
501 struct gfs2_inode *ip = GFS2_I(inode);
502 struct gfs2_sbd *sdp = GFS2_SB(inode);
503 struct buffer_head *bh;
504 struct gfs2_holder gh;
505 int need_unlock = 0;
506 int need_endtrans = 0;
507 int ret;
508
509 if (!(flags & I_DIRTY_INODE))
510 return;
511 if (unlikely(gfs2_withdrawn(sdp)))
512 return;
513 if (!gfs2_glock_is_locked_by_me(ip->i_gl)) {
514 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
515 if (ret) {
516 fs_err(sdp, "dirty_inode: glock %d\n", ret);
517 gfs2_dump_glock(NULL, ip->i_gl, true);
518 return;
519 }
520 need_unlock = 1;
521 } else if (WARN_ON_ONCE(ip->i_gl->gl_state != LM_ST_EXCLUSIVE))
522 return;
523
524 if (current->journal_info == NULL) {
525 ret = gfs2_trans_begin(sdp, RES_DINODE, 0);
526 if (ret) {
527 fs_err(sdp, "dirty_inode: gfs2_trans_begin %d\n", ret);
528 goto out;
529 }
530 need_endtrans = 1;
531 }
532
533 ret = gfs2_meta_inode_buffer(ip, &bh);
534 if (ret == 0) {
535 gfs2_trans_add_meta(ip->i_gl, bh);
536 gfs2_dinode_out(ip, bh->b_data);
537 brelse(bh);
538 }
539
540 if (need_endtrans)
541 gfs2_trans_end(sdp);
542 out:
543 if (need_unlock)
544 gfs2_glock_dq_uninit(&gh);
545 }
546
547 /**
548 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one
549 * @sdp: the filesystem
550 *
551 * Returns: errno
552 */
553
gfs2_make_fs_ro(struct gfs2_sbd * sdp)554 int gfs2_make_fs_ro(struct gfs2_sbd *sdp)
555 {
556 int error = 0;
557 int log_write_allowed = test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
558
559 gfs2_flush_delete_work(sdp);
560 if (!log_write_allowed && current == sdp->sd_quotad_process)
561 fs_warn(sdp, "The quotad daemon is withdrawing.\n");
562 else if (sdp->sd_quotad_process)
563 kthread_stop(sdp->sd_quotad_process);
564 sdp->sd_quotad_process = NULL;
565
566 if (!log_write_allowed && current == sdp->sd_logd_process)
567 fs_warn(sdp, "The logd daemon is withdrawing.\n");
568 else if (sdp->sd_logd_process)
569 kthread_stop(sdp->sd_logd_process);
570 sdp->sd_logd_process = NULL;
571
572 if (log_write_allowed) {
573 gfs2_quota_sync(sdp->sd_vfs, 0);
574 gfs2_statfs_sync(sdp->sd_vfs, 0);
575
576 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_SHUTDOWN |
577 GFS2_LFC_MAKE_FS_RO);
578 wait_event(sdp->sd_reserving_log_wait,
579 atomic_read(&sdp->sd_reserving_log) == 0);
580 gfs2_assert_warn(sdp, atomic_read(&sdp->sd_log_blks_free) ==
581 sdp->sd_jdesc->jd_blocks);
582 } else {
583 wait_event_timeout(sdp->sd_reserving_log_wait,
584 atomic_read(&sdp->sd_reserving_log) == 0,
585 HZ * 5);
586 }
587 gfs2_quota_cleanup(sdp);
588
589 if (!log_write_allowed)
590 sdp->sd_vfs->s_flags |= SB_RDONLY;
591
592 return error;
593 }
594
595 /**
596 * gfs2_put_super - Unmount the filesystem
597 * @sb: The VFS superblock
598 *
599 */
600
gfs2_put_super(struct super_block * sb)601 static void gfs2_put_super(struct super_block *sb)
602 {
603 struct gfs2_sbd *sdp = sb->s_fs_info;
604 int error;
605 struct gfs2_jdesc *jd;
606
607 /* No more recovery requests */
608 set_bit(SDF_NORECOVERY, &sdp->sd_flags);
609 smp_mb();
610
611 /* Wait on outstanding recovery */
612 restart:
613 spin_lock(&sdp->sd_jindex_spin);
614 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
615 if (!test_bit(JDF_RECOVERY, &jd->jd_flags))
616 continue;
617 spin_unlock(&sdp->sd_jindex_spin);
618 wait_on_bit(&jd->jd_flags, JDF_RECOVERY,
619 TASK_UNINTERRUPTIBLE);
620 goto restart;
621 }
622 spin_unlock(&sdp->sd_jindex_spin);
623
624 if (!sb_rdonly(sb)) {
625 error = gfs2_make_fs_ro(sdp);
626 if (error)
627 gfs2_io_error(sdp);
628 }
629 WARN_ON(gfs2_withdrawing(sdp));
630
631 /* At this point, we're through modifying the disk */
632
633 /* Release stuff */
634
635 iput(sdp->sd_jindex);
636 iput(sdp->sd_statfs_inode);
637 iput(sdp->sd_rindex);
638 iput(sdp->sd_quota_inode);
639
640 gfs2_glock_put(sdp->sd_rename_gl);
641 gfs2_glock_put(sdp->sd_freeze_gl);
642
643 if (!sdp->sd_args.ar_spectator) {
644 if (gfs2_holder_initialized(&sdp->sd_journal_gh))
645 gfs2_glock_dq_uninit(&sdp->sd_journal_gh);
646 if (gfs2_holder_initialized(&sdp->sd_jinode_gh))
647 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh);
648 gfs2_glock_dq_uninit(&sdp->sd_sc_gh);
649 gfs2_glock_dq_uninit(&sdp->sd_qc_gh);
650 free_local_statfs_inodes(sdp);
651 iput(sdp->sd_qc_inode);
652 }
653
654 gfs2_glock_dq_uninit(&sdp->sd_live_gh);
655 gfs2_clear_rgrpd(sdp);
656 gfs2_jindex_free(sdp);
657 /* Take apart glock structures and buffer lists */
658 gfs2_gl_hash_clear(sdp);
659 truncate_inode_pages_final(&sdp->sd_aspace);
660 gfs2_delete_debugfs_file(sdp);
661 /* Unmount the locking protocol */
662 gfs2_lm_unmount(sdp);
663
664 /* At this point, we're through participating in the lockspace */
665 gfs2_sys_fs_del(sdp);
666 free_sbd(sdp);
667 }
668
669 /**
670 * gfs2_sync_fs - sync the filesystem
671 * @sb: the superblock
672 *
673 * Flushes the log to disk.
674 */
675
gfs2_sync_fs(struct super_block * sb,int wait)676 static int gfs2_sync_fs(struct super_block *sb, int wait)
677 {
678 struct gfs2_sbd *sdp = sb->s_fs_info;
679
680 gfs2_quota_sync(sb, -1);
681 if (wait)
682 gfs2_log_flush(sdp, NULL, GFS2_LOG_HEAD_FLUSH_NORMAL |
683 GFS2_LFC_SYNC_FS);
684 return sdp->sd_log_error;
685 }
686
gfs2_freeze_func(struct work_struct * work)687 void gfs2_freeze_func(struct work_struct *work)
688 {
689 int error;
690 struct gfs2_holder freeze_gh;
691 struct gfs2_sbd *sdp = container_of(work, struct gfs2_sbd, sd_freeze_work);
692 struct super_block *sb = sdp->sd_vfs;
693
694 atomic_inc(&sb->s_active);
695 error = gfs2_freeze_lock(sdp, &freeze_gh, 0);
696 if (error) {
697 gfs2_assert_withdraw(sdp, 0);
698 } else {
699 atomic_set(&sdp->sd_freeze_state, SFS_UNFROZEN);
700 error = thaw_super(sb);
701 if (error) {
702 fs_info(sdp, "GFS2: couldn't thaw filesystem: %d\n",
703 error);
704 gfs2_assert_withdraw(sdp, 0);
705 }
706 gfs2_freeze_unlock(&freeze_gh);
707 }
708 deactivate_super(sb);
709 clear_bit_unlock(SDF_FS_FROZEN, &sdp->sd_flags);
710 wake_up_bit(&sdp->sd_flags, SDF_FS_FROZEN);
711 return;
712 }
713
714 /**
715 * gfs2_freeze - prevent further writes to the filesystem
716 * @sb: the VFS structure for the filesystem
717 *
718 */
719
gfs2_freeze(struct super_block * sb)720 static int gfs2_freeze(struct super_block *sb)
721 {
722 struct gfs2_sbd *sdp = sb->s_fs_info;
723 int error;
724
725 mutex_lock(&sdp->sd_freeze_mutex);
726 if (atomic_read(&sdp->sd_freeze_state) != SFS_UNFROZEN) {
727 error = -EBUSY;
728 goto out;
729 }
730
731 for (;;) {
732 if (gfs2_withdrawn(sdp)) {
733 error = -EINVAL;
734 goto out;
735 }
736
737 error = gfs2_lock_fs_check_clean(sdp);
738 if (!error)
739 break;
740
741 if (error == -EBUSY)
742 fs_err(sdp, "waiting for recovery before freeze\n");
743 else if (error == -EIO) {
744 fs_err(sdp, "Fatal IO error: cannot freeze gfs2 due "
745 "to recovery error.\n");
746 goto out;
747 } else {
748 fs_err(sdp, "error freezing FS: %d\n", error);
749 }
750 fs_err(sdp, "retrying...\n");
751 msleep(1000);
752 }
753 set_bit(SDF_FS_FROZEN, &sdp->sd_flags);
754 out:
755 mutex_unlock(&sdp->sd_freeze_mutex);
756 return error;
757 }
758
759 /**
760 * gfs2_unfreeze - reallow writes to the filesystem
761 * @sb: the VFS structure for the filesystem
762 *
763 */
764
gfs2_unfreeze(struct super_block * sb)765 static int gfs2_unfreeze(struct super_block *sb)
766 {
767 struct gfs2_sbd *sdp = sb->s_fs_info;
768
769 mutex_lock(&sdp->sd_freeze_mutex);
770 if (atomic_read(&sdp->sd_freeze_state) != SFS_FROZEN ||
771 !gfs2_holder_initialized(&sdp->sd_freeze_gh)) {
772 mutex_unlock(&sdp->sd_freeze_mutex);
773 return -EINVAL;
774 }
775
776 gfs2_freeze_unlock(&sdp->sd_freeze_gh);
777 mutex_unlock(&sdp->sd_freeze_mutex);
778 return wait_on_bit(&sdp->sd_flags, SDF_FS_FROZEN, TASK_INTERRUPTIBLE);
779 }
780
781 /**
782 * statfs_fill - fill in the sg for a given RG
783 * @rgd: the RG
784 * @sc: the sc structure
785 *
786 * Returns: 0 on success, -ESTALE if the LVB is invalid
787 */
788
statfs_slow_fill(struct gfs2_rgrpd * rgd,struct gfs2_statfs_change_host * sc)789 static int statfs_slow_fill(struct gfs2_rgrpd *rgd,
790 struct gfs2_statfs_change_host *sc)
791 {
792 gfs2_rgrp_verify(rgd);
793 sc->sc_total += rgd->rd_data;
794 sc->sc_free += rgd->rd_free;
795 sc->sc_dinodes += rgd->rd_dinodes;
796 return 0;
797 }
798
799 /**
800 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking
801 * @sdp: the filesystem
802 * @sc: the sc info that will be returned
803 *
804 * Any error (other than a signal) will cause this routine to fall back
805 * to the synchronous version.
806 *
807 * FIXME: This really shouldn't busy wait like this.
808 *
809 * Returns: errno
810 */
811
gfs2_statfs_slow(struct gfs2_sbd * sdp,struct gfs2_statfs_change_host * sc)812 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
813 {
814 struct gfs2_rgrpd *rgd_next;
815 struct gfs2_holder *gha, *gh;
816 unsigned int slots = 64;
817 unsigned int x;
818 int done;
819 int error = 0, err;
820
821 memset(sc, 0, sizeof(struct gfs2_statfs_change_host));
822 gha = kmalloc_array(slots, sizeof(struct gfs2_holder), GFP_KERNEL);
823 if (!gha)
824 return -ENOMEM;
825 for (x = 0; x < slots; x++)
826 gfs2_holder_mark_uninitialized(gha + x);
827
828 rgd_next = gfs2_rgrpd_get_first(sdp);
829
830 for (;;) {
831 done = 1;
832
833 for (x = 0; x < slots; x++) {
834 gh = gha + x;
835
836 if (gfs2_holder_initialized(gh) && gfs2_glock_poll(gh)) {
837 err = gfs2_glock_wait(gh);
838 if (err) {
839 gfs2_holder_uninit(gh);
840 error = err;
841 } else {
842 if (!error) {
843 struct gfs2_rgrpd *rgd =
844 gfs2_glock2rgrp(gh->gh_gl);
845
846 error = statfs_slow_fill(rgd, sc);
847 }
848 gfs2_glock_dq_uninit(gh);
849 }
850 }
851
852 if (gfs2_holder_initialized(gh))
853 done = 0;
854 else if (rgd_next && !error) {
855 error = gfs2_glock_nq_init(rgd_next->rd_gl,
856 LM_ST_SHARED,
857 GL_ASYNC,
858 gh);
859 rgd_next = gfs2_rgrpd_get_next(rgd_next);
860 done = 0;
861 }
862
863 if (signal_pending(current))
864 error = -ERESTARTSYS;
865 }
866
867 if (done)
868 break;
869
870 yield();
871 }
872
873 kfree(gha);
874 return error;
875 }
876
877 /**
878 * gfs2_statfs_i - Do a statfs
879 * @sdp: the filesystem
880 * @sg: the sg structure
881 *
882 * Returns: errno
883 */
884
gfs2_statfs_i(struct gfs2_sbd * sdp,struct gfs2_statfs_change_host * sc)885 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc)
886 {
887 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
888 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
889
890 spin_lock(&sdp->sd_statfs_spin);
891
892 *sc = *m_sc;
893 sc->sc_total += l_sc->sc_total;
894 sc->sc_free += l_sc->sc_free;
895 sc->sc_dinodes += l_sc->sc_dinodes;
896
897 spin_unlock(&sdp->sd_statfs_spin);
898
899 if (sc->sc_free < 0)
900 sc->sc_free = 0;
901 if (sc->sc_free > sc->sc_total)
902 sc->sc_free = sc->sc_total;
903 if (sc->sc_dinodes < 0)
904 sc->sc_dinodes = 0;
905
906 return 0;
907 }
908
909 /**
910 * gfs2_statfs - Gather and return stats about the filesystem
911 * @sb: The superblock
912 * @statfsbuf: The buffer
913 *
914 * Returns: 0 on success or error code
915 */
916
gfs2_statfs(struct dentry * dentry,struct kstatfs * buf)917 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
918 {
919 struct super_block *sb = dentry->d_sb;
920 struct gfs2_sbd *sdp = sb->s_fs_info;
921 struct gfs2_statfs_change_host sc;
922 int error;
923
924 error = gfs2_rindex_update(sdp);
925 if (error)
926 return error;
927
928 if (gfs2_tune_get(sdp, gt_statfs_slow))
929 error = gfs2_statfs_slow(sdp, &sc);
930 else
931 error = gfs2_statfs_i(sdp, &sc);
932
933 if (error)
934 return error;
935
936 buf->f_type = GFS2_MAGIC;
937 buf->f_bsize = sdp->sd_sb.sb_bsize;
938 buf->f_blocks = sc.sc_total;
939 buf->f_bfree = sc.sc_free;
940 buf->f_bavail = sc.sc_free;
941 buf->f_files = sc.sc_dinodes + sc.sc_free;
942 buf->f_ffree = sc.sc_free;
943 buf->f_namelen = GFS2_FNAMESIZE;
944
945 return 0;
946 }
947
948 /**
949 * gfs2_drop_inode - Drop an inode (test for remote unlink)
950 * @inode: The inode to drop
951 *
952 * If we've received a callback on an iopen lock then it's because a
953 * remote node tried to deallocate the inode but failed due to this node
954 * still having the inode open. Here we mark the link count zero
955 * since we know that it must have reached zero if the GLF_DEMOTE flag
956 * is set on the iopen glock. If we didn't do a disk read since the
957 * remote node removed the final link then we might otherwise miss
958 * this event. This check ensures that this node will deallocate the
959 * inode's blocks, or alternatively pass the baton on to another
960 * node for later deallocation.
961 */
962
gfs2_drop_inode(struct inode * inode)963 static int gfs2_drop_inode(struct inode *inode)
964 {
965 struct gfs2_inode *ip = GFS2_I(inode);
966
967 if (!test_bit(GIF_FREE_VFS_INODE, &ip->i_flags) &&
968 inode->i_nlink &&
969 gfs2_holder_initialized(&ip->i_iopen_gh)) {
970 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
971 if (test_bit(GLF_DEMOTE, &gl->gl_flags))
972 clear_nlink(inode);
973 }
974
975 /*
976 * When under memory pressure when an inode's link count has dropped to
977 * zero, defer deleting the inode to the delete workqueue. This avoids
978 * calling into DLM under memory pressure, which can deadlock.
979 */
980 if (!inode->i_nlink &&
981 unlikely(current->flags & PF_MEMALLOC) &&
982 gfs2_holder_initialized(&ip->i_iopen_gh)) {
983 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
984
985 gfs2_glock_hold(gl);
986 if (!gfs2_queue_delete_work(gl, 0))
987 gfs2_glock_queue_put(gl);
988 return false;
989 }
990
991 return generic_drop_inode(inode);
992 }
993
is_ancestor(const struct dentry * d1,const struct dentry * d2)994 static int is_ancestor(const struct dentry *d1, const struct dentry *d2)
995 {
996 do {
997 if (d1 == d2)
998 return 1;
999 d1 = d1->d_parent;
1000 } while (!IS_ROOT(d1));
1001 return 0;
1002 }
1003
1004 /**
1005 * gfs2_show_options - Show mount options for /proc/mounts
1006 * @s: seq_file structure
1007 * @root: root of this (sub)tree
1008 *
1009 * Returns: 0 on success or error code
1010 */
1011
gfs2_show_options(struct seq_file * s,struct dentry * root)1012 static int gfs2_show_options(struct seq_file *s, struct dentry *root)
1013 {
1014 struct gfs2_sbd *sdp = root->d_sb->s_fs_info;
1015 struct gfs2_args *args = &sdp->sd_args;
1016 int val;
1017
1018 if (is_ancestor(root, sdp->sd_master_dir))
1019 seq_puts(s, ",meta");
1020 if (args->ar_lockproto[0])
1021 seq_show_option(s, "lockproto", args->ar_lockproto);
1022 if (args->ar_locktable[0])
1023 seq_show_option(s, "locktable", args->ar_locktable);
1024 if (args->ar_hostdata[0])
1025 seq_show_option(s, "hostdata", args->ar_hostdata);
1026 if (args->ar_spectator)
1027 seq_puts(s, ",spectator");
1028 if (args->ar_localflocks)
1029 seq_puts(s, ",localflocks");
1030 if (args->ar_debug)
1031 seq_puts(s, ",debug");
1032 if (args->ar_posix_acl)
1033 seq_puts(s, ",acl");
1034 if (args->ar_quota != GFS2_QUOTA_DEFAULT) {
1035 char *state;
1036 switch (args->ar_quota) {
1037 case GFS2_QUOTA_OFF:
1038 state = "off";
1039 break;
1040 case GFS2_QUOTA_ACCOUNT:
1041 state = "account";
1042 break;
1043 case GFS2_QUOTA_ON:
1044 state = "on";
1045 break;
1046 default:
1047 state = "unknown";
1048 break;
1049 }
1050 seq_printf(s, ",quota=%s", state);
1051 }
1052 if (args->ar_suiddir)
1053 seq_puts(s, ",suiddir");
1054 if (args->ar_data != GFS2_DATA_DEFAULT) {
1055 char *state;
1056 switch (args->ar_data) {
1057 case GFS2_DATA_WRITEBACK:
1058 state = "writeback";
1059 break;
1060 case GFS2_DATA_ORDERED:
1061 state = "ordered";
1062 break;
1063 default:
1064 state = "unknown";
1065 break;
1066 }
1067 seq_printf(s, ",data=%s", state);
1068 }
1069 if (args->ar_discard)
1070 seq_puts(s, ",discard");
1071 val = sdp->sd_tune.gt_logd_secs;
1072 if (val != 30)
1073 seq_printf(s, ",commit=%d", val);
1074 val = sdp->sd_tune.gt_statfs_quantum;
1075 if (val != 30)
1076 seq_printf(s, ",statfs_quantum=%d", val);
1077 else if (sdp->sd_tune.gt_statfs_slow)
1078 seq_puts(s, ",statfs_quantum=0");
1079 val = sdp->sd_tune.gt_quota_quantum;
1080 if (val != 60)
1081 seq_printf(s, ",quota_quantum=%d", val);
1082 if (args->ar_statfs_percent)
1083 seq_printf(s, ",statfs_percent=%d", args->ar_statfs_percent);
1084 if (args->ar_errors != GFS2_ERRORS_DEFAULT) {
1085 const char *state;
1086
1087 switch (args->ar_errors) {
1088 case GFS2_ERRORS_WITHDRAW:
1089 state = "withdraw";
1090 break;
1091 case GFS2_ERRORS_PANIC:
1092 state = "panic";
1093 break;
1094 default:
1095 state = "unknown";
1096 break;
1097 }
1098 seq_printf(s, ",errors=%s", state);
1099 }
1100 if (test_bit(SDF_NOBARRIERS, &sdp->sd_flags))
1101 seq_puts(s, ",nobarrier");
1102 if (test_bit(SDF_DEMOTE, &sdp->sd_flags))
1103 seq_puts(s, ",demote_interface_used");
1104 if (args->ar_rgrplvb)
1105 seq_puts(s, ",rgrplvb");
1106 if (args->ar_loccookie)
1107 seq_puts(s, ",loccookie");
1108 return 0;
1109 }
1110
gfs2_final_release_pages(struct gfs2_inode * ip)1111 static void gfs2_final_release_pages(struct gfs2_inode *ip)
1112 {
1113 struct inode *inode = &ip->i_inode;
1114 struct gfs2_glock *gl = ip->i_gl;
1115
1116 truncate_inode_pages(gfs2_glock2aspace(ip->i_gl), 0);
1117 truncate_inode_pages(&inode->i_data, 0);
1118
1119 if (atomic_read(&gl->gl_revokes) == 0) {
1120 clear_bit(GLF_LFLUSH, &gl->gl_flags);
1121 clear_bit(GLF_DIRTY, &gl->gl_flags);
1122 }
1123 }
1124
gfs2_dinode_dealloc(struct gfs2_inode * ip)1125 static int gfs2_dinode_dealloc(struct gfs2_inode *ip)
1126 {
1127 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1128 struct gfs2_rgrpd *rgd;
1129 struct gfs2_holder gh;
1130 int error;
1131
1132 if (gfs2_get_inode_blocks(&ip->i_inode) != 1) {
1133 gfs2_consist_inode(ip);
1134 return -EIO;
1135 }
1136
1137 error = gfs2_rindex_update(sdp);
1138 if (error)
1139 return error;
1140
1141 error = gfs2_quota_hold(ip, NO_UID_QUOTA_CHANGE, NO_GID_QUOTA_CHANGE);
1142 if (error)
1143 return error;
1144
1145 rgd = gfs2_blk2rgrpd(sdp, ip->i_no_addr, 1);
1146 if (!rgd) {
1147 gfs2_consist_inode(ip);
1148 error = -EIO;
1149 goto out_qs;
1150 }
1151
1152 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &gh);
1153 if (error)
1154 goto out_qs;
1155
1156 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_STATFS + RES_QUOTA,
1157 sdp->sd_jdesc->jd_blocks);
1158 if (error)
1159 goto out_rg_gunlock;
1160
1161 gfs2_free_di(rgd, ip);
1162
1163 gfs2_final_release_pages(ip);
1164
1165 gfs2_trans_end(sdp);
1166
1167 out_rg_gunlock:
1168 gfs2_glock_dq_uninit(&gh);
1169 out_qs:
1170 gfs2_quota_unhold(ip);
1171 return error;
1172 }
1173
1174 /**
1175 * gfs2_glock_put_eventually
1176 * @gl: The glock to put
1177 *
1178 * When under memory pressure, trigger a deferred glock put to make sure we
1179 * won't call into DLM and deadlock. Otherwise, put the glock directly.
1180 */
1181
gfs2_glock_put_eventually(struct gfs2_glock * gl)1182 static void gfs2_glock_put_eventually(struct gfs2_glock *gl)
1183 {
1184 if (current->flags & PF_MEMALLOC)
1185 gfs2_glock_queue_put(gl);
1186 else
1187 gfs2_glock_put(gl);
1188 }
1189
gfs2_upgrade_iopen_glock(struct inode * inode)1190 static bool gfs2_upgrade_iopen_glock(struct inode *inode)
1191 {
1192 struct gfs2_inode *ip = GFS2_I(inode);
1193 struct gfs2_sbd *sdp = GFS2_SB(inode);
1194 struct gfs2_holder *gh = &ip->i_iopen_gh;
1195 long timeout = 5 * HZ;
1196 int error;
1197
1198 gh->gh_flags |= GL_NOCACHE;
1199 gfs2_glock_dq_wait(gh);
1200
1201 /*
1202 * If there are no other lock holders, we'll get the lock immediately.
1203 * Otherwise, the other nodes holding the lock will be notified about
1204 * our locking request. If they don't have the inode open, they'll
1205 * evict the cached inode and release the lock. Otherwise, if they
1206 * poke the inode glock, we'll take this as an indication that they
1207 * still need the iopen glock and that they'll take care of deleting
1208 * the inode when they're done. As a last resort, if another node
1209 * keeps holding the iopen glock without showing any activity on the
1210 * inode glock, we'll eventually time out.
1211 *
1212 * Note that we're passing the LM_FLAG_TRY_1CB flag to the first
1213 * locking request as an optimization to notify lock holders as soon as
1214 * possible. Without that flag, they'd be notified implicitly by the
1215 * second locking request.
1216 */
1217
1218 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, gh);
1219 error = gfs2_glock_nq(gh);
1220 if (error != GLR_TRYFAILED)
1221 return !error;
1222
1223 gfs2_holder_reinit(LM_ST_EXCLUSIVE, GL_ASYNC | GL_NOCACHE, gh);
1224 error = gfs2_glock_nq(gh);
1225 if (error)
1226 return false;
1227
1228 timeout = wait_event_interruptible_timeout(sdp->sd_async_glock_wait,
1229 !test_bit(HIF_WAIT, &gh->gh_iflags) ||
1230 test_bit(GLF_DEMOTE, &ip->i_gl->gl_flags),
1231 timeout);
1232 if (!test_bit(HIF_HOLDER, &gh->gh_iflags)) {
1233 gfs2_glock_dq(gh);
1234 return false;
1235 }
1236 return true;
1237 }
1238
1239 /**
1240 * evict_should_delete - determine whether the inode is eligible for deletion
1241 * @inode: The inode to evict
1242 *
1243 * This function determines whether the evicted inode is eligible to be deleted
1244 * and locks the inode glock.
1245 *
1246 * Returns: the fate of the dinode
1247 */
evict_should_delete(struct inode * inode,struct gfs2_holder * gh)1248 static enum dinode_demise evict_should_delete(struct inode *inode,
1249 struct gfs2_holder *gh)
1250 {
1251 struct gfs2_inode *ip = GFS2_I(inode);
1252 struct super_block *sb = inode->i_sb;
1253 struct gfs2_sbd *sdp = sb->s_fs_info;
1254 int ret;
1255
1256 if (test_bit(GIF_ALLOC_FAILED, &ip->i_flags)) {
1257 BUG_ON(!gfs2_glock_is_locked_by_me(ip->i_gl));
1258 goto should_delete;
1259 }
1260
1261 if (test_bit(GIF_DEFERRED_DELETE, &ip->i_flags))
1262 return SHOULD_DEFER_EVICTION;
1263
1264 /* Deletes should never happen under memory pressure anymore. */
1265 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
1266 return SHOULD_DEFER_EVICTION;
1267
1268 /* Must not read inode block until block type has been verified */
1269 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_SKIP, gh);
1270 if (unlikely(ret)) {
1271 glock_clear_object(ip->i_iopen_gh.gh_gl, ip);
1272 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1273 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
1274 return SHOULD_DEFER_EVICTION;
1275 }
1276
1277 if (gfs2_inode_already_deleted(ip->i_gl, ip->i_no_formal_ino))
1278 return SHOULD_NOT_DELETE_DINODE;
1279 ret = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED);
1280 if (ret)
1281 return SHOULD_NOT_DELETE_DINODE;
1282
1283 if (test_bit(GIF_INVALID, &ip->i_flags)) {
1284 ret = gfs2_inode_refresh(ip);
1285 if (ret)
1286 return SHOULD_NOT_DELETE_DINODE;
1287 }
1288
1289 /*
1290 * The inode may have been recreated in the meantime.
1291 */
1292 if (inode->i_nlink)
1293 return SHOULD_NOT_DELETE_DINODE;
1294
1295 should_delete:
1296 if (gfs2_holder_initialized(&ip->i_iopen_gh) &&
1297 test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1298 if (!gfs2_upgrade_iopen_glock(inode)) {
1299 gfs2_holder_uninit(&ip->i_iopen_gh);
1300 return SHOULD_NOT_DELETE_DINODE;
1301 }
1302 }
1303 return SHOULD_DELETE_DINODE;
1304 }
1305
1306 /**
1307 * evict_unlinked_inode - delete the pieces of an unlinked evicted inode
1308 * @inode: The inode to evict
1309 */
evict_unlinked_inode(struct inode * inode)1310 static int evict_unlinked_inode(struct inode *inode)
1311 {
1312 struct gfs2_inode *ip = GFS2_I(inode);
1313 int ret;
1314
1315 if (S_ISDIR(inode->i_mode) &&
1316 (ip->i_diskflags & GFS2_DIF_EXHASH)) {
1317 ret = gfs2_dir_exhash_dealloc(ip);
1318 if (ret)
1319 goto out;
1320 }
1321
1322 if (ip->i_eattr) {
1323 ret = gfs2_ea_dealloc(ip);
1324 if (ret)
1325 goto out;
1326 }
1327
1328 if (!gfs2_is_stuffed(ip)) {
1329 ret = gfs2_file_dealloc(ip);
1330 if (ret)
1331 goto out;
1332 }
1333
1334 /* We're about to clear the bitmap for the dinode, but as soon as we
1335 do, gfs2_create_inode can create another inode at the same block
1336 location and try to set gl_object again. We clear gl_object here so
1337 that subsequent inode creates don't see an old gl_object. */
1338 glock_clear_object(ip->i_gl, ip);
1339 ret = gfs2_dinode_dealloc(ip);
1340 gfs2_inode_remember_delete(ip->i_gl, ip->i_no_formal_ino);
1341 out:
1342 return ret;
1343 }
1344
1345 /*
1346 * evict_linked_inode - evict an inode whose dinode has not been unlinked
1347 * @inode: The inode to evict
1348 */
evict_linked_inode(struct inode * inode)1349 static int evict_linked_inode(struct inode *inode)
1350 {
1351 struct super_block *sb = inode->i_sb;
1352 struct gfs2_sbd *sdp = sb->s_fs_info;
1353 struct gfs2_inode *ip = GFS2_I(inode);
1354 struct address_space *metamapping;
1355 int ret;
1356
1357 gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL |
1358 GFS2_LFC_EVICT_INODE);
1359 metamapping = gfs2_glock2aspace(ip->i_gl);
1360 if (test_bit(GLF_DIRTY, &ip->i_gl->gl_flags)) {
1361 filemap_fdatawrite(metamapping);
1362 filemap_fdatawait(metamapping);
1363 }
1364 write_inode_now(inode, 1);
1365 gfs2_ail_flush(ip->i_gl, 0);
1366
1367 ret = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks);
1368 if (ret)
1369 return ret;
1370
1371 /* Needs to be done before glock release & also in a transaction */
1372 truncate_inode_pages(&inode->i_data, 0);
1373 truncate_inode_pages(metamapping, 0);
1374 gfs2_trans_end(sdp);
1375 return 0;
1376 }
1377
1378 /**
1379 * gfs2_evict_inode - Remove an inode from cache
1380 * @inode: The inode to evict
1381 *
1382 * There are three cases to consider:
1383 * 1. i_nlink == 0, we are final opener (and must deallocate)
1384 * 2. i_nlink == 0, we are not the final opener (and cannot deallocate)
1385 * 3. i_nlink > 0
1386 *
1387 * If the fs is read only, then we have to treat all cases as per #3
1388 * since we are unable to do any deallocation. The inode will be
1389 * deallocated by the next read/write node to attempt an allocation
1390 * in the same resource group
1391 *
1392 * We have to (at the moment) hold the inodes main lock to cover
1393 * the gap between unlocking the shared lock on the iopen lock and
1394 * taking the exclusive lock. I'd rather do a shared -> exclusive
1395 * conversion on the iopen lock, but we can change that later. This
1396 * is safe, just less efficient.
1397 */
1398
gfs2_evict_inode(struct inode * inode)1399 static void gfs2_evict_inode(struct inode *inode)
1400 {
1401 struct super_block *sb = inode->i_sb;
1402 struct gfs2_sbd *sdp = sb->s_fs_info;
1403 struct gfs2_inode *ip = GFS2_I(inode);
1404 struct gfs2_holder gh;
1405 int ret;
1406
1407 if (test_bit(GIF_FREE_VFS_INODE, &ip->i_flags)) {
1408 clear_inode(inode);
1409 return;
1410 }
1411
1412 if (inode->i_nlink || sb_rdonly(sb))
1413 goto out;
1414
1415 /*
1416 * In case of an incomplete mount, gfs2_evict_inode() may be called for
1417 * system files without having an active journal to write to. In that
1418 * case, skip the filesystem evict.
1419 */
1420 if (!sdp->sd_jdesc)
1421 goto out;
1422
1423 gfs2_holder_mark_uninitialized(&gh);
1424 ret = evict_should_delete(inode, &gh);
1425 if (ret == SHOULD_DEFER_EVICTION)
1426 goto out;
1427 if (ret == SHOULD_DELETE_DINODE)
1428 ret = evict_unlinked_inode(inode);
1429 else
1430 ret = evict_linked_inode(inode);
1431
1432 if (gfs2_rs_active(&ip->i_res))
1433 gfs2_rs_deltree(&ip->i_res);
1434
1435 if (gfs2_holder_initialized(&gh)) {
1436 glock_clear_object(ip->i_gl, ip);
1437 gfs2_glock_dq_uninit(&gh);
1438 }
1439 if (ret && ret != GLR_TRYFAILED && ret != -EROFS)
1440 fs_warn(sdp, "gfs2_evict_inode: %d\n", ret);
1441 out:
1442 truncate_inode_pages_final(&inode->i_data);
1443 if (ip->i_qadata)
1444 gfs2_assert_warn(sdp, ip->i_qadata->qa_ref == 0);
1445 gfs2_rs_delete(ip, NULL);
1446 gfs2_ordered_del_inode(ip);
1447 clear_inode(inode);
1448 gfs2_dir_hash_inval(ip);
1449 if (gfs2_holder_initialized(&ip->i_iopen_gh)) {
1450 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl;
1451
1452 glock_clear_object(gl, ip);
1453 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) {
1454 ip->i_iopen_gh.gh_flags |= GL_NOCACHE;
1455 gfs2_glock_dq(&ip->i_iopen_gh);
1456 }
1457 gfs2_glock_hold(gl);
1458 gfs2_holder_uninit(&ip->i_iopen_gh);
1459 gfs2_glock_put_eventually(gl);
1460 }
1461 if (ip->i_gl) {
1462 glock_clear_object(ip->i_gl, ip);
1463 wait_on_bit_io(&ip->i_flags, GIF_GLOP_PENDING, TASK_UNINTERRUPTIBLE);
1464 gfs2_glock_add_to_lru(ip->i_gl);
1465 gfs2_glock_put_eventually(ip->i_gl);
1466 ip->i_gl = NULL;
1467 }
1468 }
1469
gfs2_alloc_inode(struct super_block * sb)1470 static struct inode *gfs2_alloc_inode(struct super_block *sb)
1471 {
1472 struct gfs2_inode *ip;
1473
1474 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
1475 if (!ip)
1476 return NULL;
1477 ip->i_flags = 0;
1478 ip->i_gl = NULL;
1479 gfs2_holder_mark_uninitialized(&ip->i_iopen_gh);
1480 memset(&ip->i_res, 0, sizeof(ip->i_res));
1481 RB_CLEAR_NODE(&ip->i_res.rs_node);
1482 ip->i_rahead = 0;
1483 return &ip->i_inode;
1484 }
1485
gfs2_free_inode(struct inode * inode)1486 static void gfs2_free_inode(struct inode *inode)
1487 {
1488 kmem_cache_free(gfs2_inode_cachep, GFS2_I(inode));
1489 }
1490
free_local_statfs_inodes(struct gfs2_sbd * sdp)1491 extern void free_local_statfs_inodes(struct gfs2_sbd *sdp)
1492 {
1493 struct local_statfs_inode *lsi, *safe;
1494
1495 /* Run through the statfs inodes list to iput and free memory */
1496 list_for_each_entry_safe(lsi, safe, &sdp->sd_sc_inodes_list, si_list) {
1497 if (lsi->si_jid == sdp->sd_jdesc->jd_jid)
1498 sdp->sd_sc_inode = NULL; /* belongs to this node */
1499 if (lsi->si_sc_inode)
1500 iput(lsi->si_sc_inode);
1501 list_del(&lsi->si_list);
1502 kfree(lsi);
1503 }
1504 }
1505
find_local_statfs_inode(struct gfs2_sbd * sdp,unsigned int index)1506 extern struct inode *find_local_statfs_inode(struct gfs2_sbd *sdp,
1507 unsigned int index)
1508 {
1509 struct local_statfs_inode *lsi;
1510
1511 /* Return the local (per node) statfs inode in the
1512 * sdp->sd_sc_inodes_list corresponding to the 'index'. */
1513 list_for_each_entry(lsi, &sdp->sd_sc_inodes_list, si_list) {
1514 if (lsi->si_jid == index)
1515 return lsi->si_sc_inode;
1516 }
1517 return NULL;
1518 }
1519
1520 const struct super_operations gfs2_super_ops = {
1521 .alloc_inode = gfs2_alloc_inode,
1522 .free_inode = gfs2_free_inode,
1523 .write_inode = gfs2_write_inode,
1524 .dirty_inode = gfs2_dirty_inode,
1525 .evict_inode = gfs2_evict_inode,
1526 .put_super = gfs2_put_super,
1527 .sync_fs = gfs2_sync_fs,
1528 .freeze_super = gfs2_freeze,
1529 .thaw_super = gfs2_unfreeze,
1530 .statfs = gfs2_statfs,
1531 .drop_inode = gfs2_drop_inode,
1532 .show_options = gfs2_show_options,
1533 };
1534
1535