]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - fs/gfs2/glops.c
pagemap: do not leak physical addresses to non-privileged userspace
[mirror_ubuntu-bionic-kernel.git] / fs / gfs2 / glops.c
1 /*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
8 */
9
10 #include <linux/spinlock.h>
11 #include <linux/completion.h>
12 #include <linux/buffer_head.h>
13 #include <linux/gfs2_ondisk.h>
14 #include <linux/bio.h>
15 #include <linux/posix_acl.h>
16
17 #include "gfs2.h"
18 #include "incore.h"
19 #include "bmap.h"
20 #include "glock.h"
21 #include "glops.h"
22 #include "inode.h"
23 #include "log.h"
24 #include "meta_io.h"
25 #include "recovery.h"
26 #include "rgrp.h"
27 #include "util.h"
28 #include "trans.h"
29 #include "dir.h"
30
31 struct workqueue_struct *gfs2_freeze_wq;
32
33 static void gfs2_ail_error(struct gfs2_glock *gl, const struct buffer_head *bh)
34 {
35 fs_err(gl->gl_sbd, "AIL buffer %p: blocknr %llu state 0x%08lx mapping %p page state 0x%lx\n",
36 bh, (unsigned long long)bh->b_blocknr, bh->b_state,
37 bh->b_page->mapping, bh->b_page->flags);
38 fs_err(gl->gl_sbd, "AIL glock %u:%llu mapping %p\n",
39 gl->gl_name.ln_type, gl->gl_name.ln_number,
40 gfs2_glock2aspace(gl));
41 gfs2_lm_withdraw(gl->gl_sbd, "AIL error\n");
42 }
43
44 /**
45 * __gfs2_ail_flush - remove all buffers for a given lock from the AIL
46 * @gl: the glock
47 * @fsync: set when called from fsync (not all buffers will be clean)
48 *
49 * None of the buffers should be dirty, locked, or pinned.
50 */
51
52 static void __gfs2_ail_flush(struct gfs2_glock *gl, bool fsync,
53 unsigned int nr_revokes)
54 {
55 struct gfs2_sbd *sdp = gl->gl_sbd;
56 struct list_head *head = &gl->gl_ail_list;
57 struct gfs2_bufdata *bd, *tmp;
58 struct buffer_head *bh;
59 const unsigned long b_state = (1UL << BH_Dirty)|(1UL << BH_Pinned)|(1UL << BH_Lock);
60
61 gfs2_log_lock(sdp);
62 spin_lock(&sdp->sd_ail_lock);
63 list_for_each_entry_safe_reverse(bd, tmp, head, bd_ail_gl_list) {
64 if (nr_revokes == 0)
65 break;
66 bh = bd->bd_bh;
67 if (bh->b_state & b_state) {
68 if (fsync)
69 continue;
70 gfs2_ail_error(gl, bh);
71 }
72 gfs2_trans_add_revoke(sdp, bd);
73 nr_revokes--;
74 }
75 GLOCK_BUG_ON(gl, !fsync && atomic_read(&gl->gl_ail_count));
76 spin_unlock(&sdp->sd_ail_lock);
77 gfs2_log_unlock(sdp);
78 }
79
80
81 static void gfs2_ail_empty_gl(struct gfs2_glock *gl)
82 {
83 struct gfs2_sbd *sdp = gl->gl_sbd;
84 struct gfs2_trans tr;
85
86 memset(&tr, 0, sizeof(tr));
87 INIT_LIST_HEAD(&tr.tr_buf);
88 INIT_LIST_HEAD(&tr.tr_databuf);
89 tr.tr_revokes = atomic_read(&gl->gl_ail_count);
90
91 if (!tr.tr_revokes)
92 return;
93
94 /* A shortened, inline version of gfs2_trans_begin()
95 * tr->alloced is not set since the transaction structure is
96 * on the stack */
97 tr.tr_reserved = 1 + gfs2_struct2blk(sdp, tr.tr_revokes, sizeof(u64));
98 tr.tr_ip = _RET_IP_;
99 if (gfs2_log_reserve(sdp, tr.tr_reserved) < 0)
100 return;
101 WARN_ON_ONCE(current->journal_info);
102 current->journal_info = &tr;
103
104 __gfs2_ail_flush(gl, 0, tr.tr_revokes);
105
106 gfs2_trans_end(sdp);
107 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
108 }
109
110 void gfs2_ail_flush(struct gfs2_glock *gl, bool fsync)
111 {
112 struct gfs2_sbd *sdp = gl->gl_sbd;
113 unsigned int revokes = atomic_read(&gl->gl_ail_count);
114 unsigned int max_revokes = (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_log_descriptor)) / sizeof(u64);
115 int ret;
116
117 if (!revokes)
118 return;
119
120 while (revokes > max_revokes)
121 max_revokes += (sdp->sd_sb.sb_bsize - sizeof(struct gfs2_meta_header)) / sizeof(u64);
122
123 ret = gfs2_trans_begin(sdp, 0, max_revokes);
124 if (ret)
125 return;
126 __gfs2_ail_flush(gl, fsync, max_revokes);
127 gfs2_trans_end(sdp);
128 gfs2_log_flush(sdp, NULL, NORMAL_FLUSH);
129 }
130
131 /**
132 * rgrp_go_sync - sync out the metadata for this glock
133 * @gl: the glock
134 *
135 * Called when demoting or unlocking an EX glock. We must flush
136 * to disk all dirty buffers/pages relating to this glock, and must not
137 * not return to caller to demote/unlock the glock until I/O is complete.
138 */
139
140 static void rgrp_go_sync(struct gfs2_glock *gl)
141 {
142 struct gfs2_sbd *sdp = gl->gl_sbd;
143 struct address_space *mapping = &sdp->sd_aspace;
144 struct gfs2_rgrpd *rgd;
145 int error;
146
147 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
148 return;
149 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
150
151 gfs2_log_flush(sdp, gl, NORMAL_FLUSH);
152 filemap_fdatawrite_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
153 error = filemap_fdatawait_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
154 mapping_set_error(mapping, error);
155 gfs2_ail_empty_gl(gl);
156
157 spin_lock(&gl->gl_spin);
158 rgd = gl->gl_object;
159 if (rgd)
160 gfs2_free_clones(rgd);
161 spin_unlock(&gl->gl_spin);
162 }
163
164 /**
165 * rgrp_go_inval - invalidate the metadata for this glock
166 * @gl: the glock
167 * @flags:
168 *
169 * We never used LM_ST_DEFERRED with resource groups, so that we
170 * should always see the metadata flag set here.
171 *
172 */
173
174 static void rgrp_go_inval(struct gfs2_glock *gl, int flags)
175 {
176 struct gfs2_sbd *sdp = gl->gl_sbd;
177 struct address_space *mapping = &sdp->sd_aspace;
178
179 WARN_ON_ONCE(!(flags & DIO_METADATA));
180 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
181 truncate_inode_pages_range(mapping, gl->gl_vm.start, gl->gl_vm.end);
182
183 if (gl->gl_object) {
184 struct gfs2_rgrpd *rgd = (struct gfs2_rgrpd *)gl->gl_object;
185 rgd->rd_flags &= ~GFS2_RDF_UPTODATE;
186 }
187 }
188
189 /**
190 * inode_go_sync - Sync the dirty data and/or metadata for an inode glock
191 * @gl: the glock protecting the inode
192 *
193 */
194
195 static void inode_go_sync(struct gfs2_glock *gl)
196 {
197 struct gfs2_inode *ip = gl->gl_object;
198 struct address_space *metamapping = gfs2_glock2aspace(gl);
199 int error;
200
201 if (ip && !S_ISREG(ip->i_inode.i_mode))
202 ip = NULL;
203 if (ip) {
204 if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
205 unmap_shared_mapping_range(ip->i_inode.i_mapping, 0, 0);
206 inode_dio_wait(&ip->i_inode);
207 }
208 if (!test_and_clear_bit(GLF_DIRTY, &gl->gl_flags))
209 return;
210
211 GLOCK_BUG_ON(gl, gl->gl_state != LM_ST_EXCLUSIVE);
212
213 gfs2_log_flush(gl->gl_sbd, gl, NORMAL_FLUSH);
214 filemap_fdatawrite(metamapping);
215 if (ip) {
216 struct address_space *mapping = ip->i_inode.i_mapping;
217 filemap_fdatawrite(mapping);
218 error = filemap_fdatawait(mapping);
219 mapping_set_error(mapping, error);
220 }
221 error = filemap_fdatawait(metamapping);
222 mapping_set_error(metamapping, error);
223 gfs2_ail_empty_gl(gl);
224 /*
225 * Writeback of the data mapping may cause the dirty flag to be set
226 * so we have to clear it again here.
227 */
228 smp_mb__before_atomic();
229 clear_bit(GLF_DIRTY, &gl->gl_flags);
230 }
231
232 /**
233 * inode_go_inval - prepare a inode glock to be released
234 * @gl: the glock
235 * @flags:
236 *
237 * Normally we invalidate everything, but if we are moving into
238 * LM_ST_DEFERRED from LM_ST_SHARED or LM_ST_EXCLUSIVE then we
239 * can keep hold of the metadata, since it won't have changed.
240 *
241 */
242
243 static void inode_go_inval(struct gfs2_glock *gl, int flags)
244 {
245 struct gfs2_inode *ip = gl->gl_object;
246
247 gfs2_assert_withdraw(gl->gl_sbd, !atomic_read(&gl->gl_ail_count));
248
249 if (flags & DIO_METADATA) {
250 struct address_space *mapping = gfs2_glock2aspace(gl);
251 truncate_inode_pages(mapping, 0);
252 if (ip) {
253 set_bit(GIF_INVALID, &ip->i_flags);
254 forget_all_cached_acls(&ip->i_inode);
255 gfs2_dir_hash_inval(ip);
256 }
257 }
258
259 if (ip == GFS2_I(gl->gl_sbd->sd_rindex)) {
260 gfs2_log_flush(gl->gl_sbd, NULL, NORMAL_FLUSH);
261 gl->gl_sbd->sd_rindex_uptodate = 0;
262 }
263 if (ip && S_ISREG(ip->i_inode.i_mode))
264 truncate_inode_pages(ip->i_inode.i_mapping, 0);
265 }
266
267 /**
268 * inode_go_demote_ok - Check to see if it's ok to unlock an inode glock
269 * @gl: the glock
270 *
271 * Returns: 1 if it's ok
272 */
273
274 static int inode_go_demote_ok(const struct gfs2_glock *gl)
275 {
276 struct gfs2_sbd *sdp = gl->gl_sbd;
277 struct gfs2_holder *gh;
278
279 if (sdp->sd_jindex == gl->gl_object || sdp->sd_rindex == gl->gl_object)
280 return 0;
281
282 if (!list_empty(&gl->gl_holders)) {
283 gh = list_entry(gl->gl_holders.next, struct gfs2_holder, gh_list);
284 if (gh->gh_list.next != &gl->gl_holders)
285 return 0;
286 }
287
288 return 1;
289 }
290
291 /**
292 * gfs2_set_nlink - Set the inode's link count based on on-disk info
293 * @inode: The inode in question
294 * @nlink: The link count
295 *
296 * If the link count has hit zero, it must never be raised, whatever the
297 * on-disk inode might say. When new struct inodes are created the link
298 * count is set to 1, so that we can safely use this test even when reading
299 * in on disk information for the first time.
300 */
301
302 static void gfs2_set_nlink(struct inode *inode, u32 nlink)
303 {
304 /*
305 * We will need to review setting the nlink count here in the
306 * light of the forthcoming ro bind mount work. This is a reminder
307 * to do that.
308 */
309 if ((inode->i_nlink != nlink) && (inode->i_nlink != 0)) {
310 if (nlink == 0)
311 clear_nlink(inode);
312 else
313 set_nlink(inode, nlink);
314 }
315 }
316
317 static int gfs2_dinode_in(struct gfs2_inode *ip, const void *buf)
318 {
319 const struct gfs2_dinode *str = buf;
320 struct timespec atime;
321 u16 height, depth;
322
323 if (unlikely(ip->i_no_addr != be64_to_cpu(str->di_num.no_addr)))
324 goto corrupt;
325 ip->i_no_formal_ino = be64_to_cpu(str->di_num.no_formal_ino);
326 ip->i_inode.i_mode = be32_to_cpu(str->di_mode);
327 ip->i_inode.i_rdev = 0;
328 switch (ip->i_inode.i_mode & S_IFMT) {
329 case S_IFBLK:
330 case S_IFCHR:
331 ip->i_inode.i_rdev = MKDEV(be32_to_cpu(str->di_major),
332 be32_to_cpu(str->di_minor));
333 break;
334 };
335
336 i_uid_write(&ip->i_inode, be32_to_cpu(str->di_uid));
337 i_gid_write(&ip->i_inode, be32_to_cpu(str->di_gid));
338 gfs2_set_nlink(&ip->i_inode, be32_to_cpu(str->di_nlink));
339 i_size_write(&ip->i_inode, be64_to_cpu(str->di_size));
340 gfs2_set_inode_blocks(&ip->i_inode, be64_to_cpu(str->di_blocks));
341 atime.tv_sec = be64_to_cpu(str->di_atime);
342 atime.tv_nsec = be32_to_cpu(str->di_atime_nsec);
343 if (timespec_compare(&ip->i_inode.i_atime, &atime) < 0)
344 ip->i_inode.i_atime = atime;
345 ip->i_inode.i_mtime.tv_sec = be64_to_cpu(str->di_mtime);
346 ip->i_inode.i_mtime.tv_nsec = be32_to_cpu(str->di_mtime_nsec);
347 ip->i_inode.i_ctime.tv_sec = be64_to_cpu(str->di_ctime);
348 ip->i_inode.i_ctime.tv_nsec = be32_to_cpu(str->di_ctime_nsec);
349
350 ip->i_goal = be64_to_cpu(str->di_goal_meta);
351 ip->i_generation = be64_to_cpu(str->di_generation);
352
353 ip->i_diskflags = be32_to_cpu(str->di_flags);
354 ip->i_eattr = be64_to_cpu(str->di_eattr);
355 /* i_diskflags and i_eattr must be set before gfs2_set_inode_flags() */
356 gfs2_set_inode_flags(&ip->i_inode);
357 height = be16_to_cpu(str->di_height);
358 if (unlikely(height > GFS2_MAX_META_HEIGHT))
359 goto corrupt;
360 ip->i_height = (u8)height;
361
362 depth = be16_to_cpu(str->di_depth);
363 if (unlikely(depth > GFS2_DIR_MAX_DEPTH))
364 goto corrupt;
365 ip->i_depth = (u8)depth;
366 ip->i_entries = be32_to_cpu(str->di_entries);
367
368 if (S_ISREG(ip->i_inode.i_mode))
369 gfs2_set_aops(&ip->i_inode);
370
371 return 0;
372 corrupt:
373 gfs2_consist_inode(ip);
374 return -EIO;
375 }
376
377 /**
378 * gfs2_inode_refresh - Refresh the incore copy of the dinode
379 * @ip: The GFS2 inode
380 *
381 * Returns: errno
382 */
383
384 int gfs2_inode_refresh(struct gfs2_inode *ip)
385 {
386 struct buffer_head *dibh;
387 int error;
388
389 error = gfs2_meta_inode_buffer(ip, &dibh);
390 if (error)
391 return error;
392
393 error = gfs2_dinode_in(ip, dibh->b_data);
394 brelse(dibh);
395 clear_bit(GIF_INVALID, &ip->i_flags);
396
397 return error;
398 }
399
400 /**
401 * inode_go_lock - operation done after an inode lock is locked by a process
402 * @gl: the glock
403 * @flags:
404 *
405 * Returns: errno
406 */
407
408 static int inode_go_lock(struct gfs2_holder *gh)
409 {
410 struct gfs2_glock *gl = gh->gh_gl;
411 struct gfs2_sbd *sdp = gl->gl_sbd;
412 struct gfs2_inode *ip = gl->gl_object;
413 int error = 0;
414
415 if (!ip || (gh->gh_flags & GL_SKIP))
416 return 0;
417
418 if (test_bit(GIF_INVALID, &ip->i_flags)) {
419 error = gfs2_inode_refresh(ip);
420 if (error)
421 return error;
422 }
423
424 if (gh->gh_state != LM_ST_DEFERRED)
425 inode_dio_wait(&ip->i_inode);
426
427 if ((ip->i_diskflags & GFS2_DIF_TRUNC_IN_PROG) &&
428 (gl->gl_state == LM_ST_EXCLUSIVE) &&
429 (gh->gh_state == LM_ST_EXCLUSIVE)) {
430 spin_lock(&sdp->sd_trunc_lock);
431 if (list_empty(&ip->i_trunc_list))
432 list_add(&sdp->sd_trunc_list, &ip->i_trunc_list);
433 spin_unlock(&sdp->sd_trunc_lock);
434 wake_up(&sdp->sd_quota_wait);
435 return 1;
436 }
437
438 return error;
439 }
440
441 /**
442 * inode_go_dump - print information about an inode
443 * @seq: The iterator
444 * @ip: the inode
445 *
446 */
447
448 static void inode_go_dump(struct seq_file *seq, const struct gfs2_glock *gl)
449 {
450 const struct gfs2_inode *ip = gl->gl_object;
451 if (ip == NULL)
452 return;
453 gfs2_print_dbg(seq, " I: n:%llu/%llu t:%u f:0x%02lx d:0x%08x s:%llu\n",
454 (unsigned long long)ip->i_no_formal_ino,
455 (unsigned long long)ip->i_no_addr,
456 IF2DT(ip->i_inode.i_mode), ip->i_flags,
457 (unsigned int)ip->i_diskflags,
458 (unsigned long long)i_size_read(&ip->i_inode));
459 }
460
461 /**
462 * freeze_go_sync - promote/demote the freeze glock
463 * @gl: the glock
464 * @state: the requested state
465 * @flags:
466 *
467 */
468
469 static void freeze_go_sync(struct gfs2_glock *gl)
470 {
471 int error = 0;
472 struct gfs2_sbd *sdp = gl->gl_sbd;
473
474 if (gl->gl_state == LM_ST_SHARED &&
475 test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
476 atomic_set(&sdp->sd_freeze_state, SFS_STARTING_FREEZE);
477 error = freeze_super(sdp->sd_vfs);
478 if (error) {
479 printk(KERN_INFO "GFS2: couldn't freeze filesystem: %d\n", error);
480 gfs2_assert_withdraw(sdp, 0);
481 }
482 queue_work(gfs2_freeze_wq, &sdp->sd_freeze_work);
483 gfs2_log_flush(sdp, NULL, FREEZE_FLUSH);
484 }
485 }
486
487 /**
488 * freeze_go_xmote_bh - After promoting/demoting the freeze glock
489 * @gl: the glock
490 *
491 */
492
493 static int freeze_go_xmote_bh(struct gfs2_glock *gl, struct gfs2_holder *gh)
494 {
495 struct gfs2_sbd *sdp = gl->gl_sbd;
496 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
497 struct gfs2_glock *j_gl = ip->i_gl;
498 struct gfs2_log_header_host head;
499 int error;
500
501 if (test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags)) {
502 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
503
504 error = gfs2_find_jhead(sdp->sd_jdesc, &head);
505 if (error)
506 gfs2_consist(sdp);
507 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT))
508 gfs2_consist(sdp);
509
510 /* Initialize some head of the log stuff */
511 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) {
512 sdp->sd_log_sequence = head.lh_sequence + 1;
513 gfs2_log_pointers_init(sdp, head.lh_blkno);
514 }
515 }
516 return 0;
517 }
518
519 /**
520 * trans_go_demote_ok
521 * @gl: the glock
522 *
523 * Always returns 0
524 */
525
526 static int freeze_go_demote_ok(const struct gfs2_glock *gl)
527 {
528 return 0;
529 }
530
531 /**
532 * iopen_go_callback - schedule the dcache entry for the inode to be deleted
533 * @gl: the glock
534 *
535 * gl_spin lock is held while calling this
536 */
537 static void iopen_go_callback(struct gfs2_glock *gl, bool remote)
538 {
539 struct gfs2_inode *ip = (struct gfs2_inode *)gl->gl_object;
540 struct gfs2_sbd *sdp = gl->gl_sbd;
541
542 if (!remote || (sdp->sd_vfs->s_flags & MS_RDONLY))
543 return;
544
545 if (gl->gl_demote_state == LM_ST_UNLOCKED &&
546 gl->gl_state == LM_ST_SHARED && ip) {
547 gl->gl_lockref.count++;
548 if (queue_work(gfs2_delete_workqueue, &gl->gl_delete) == 0)
549 gl->gl_lockref.count--;
550 }
551 }
552
553 const struct gfs2_glock_operations gfs2_meta_glops = {
554 .go_type = LM_TYPE_META,
555 };
556
557 const struct gfs2_glock_operations gfs2_inode_glops = {
558 .go_sync = inode_go_sync,
559 .go_inval = inode_go_inval,
560 .go_demote_ok = inode_go_demote_ok,
561 .go_lock = inode_go_lock,
562 .go_dump = inode_go_dump,
563 .go_type = LM_TYPE_INODE,
564 .go_flags = GLOF_ASPACE,
565 };
566
567 const struct gfs2_glock_operations gfs2_rgrp_glops = {
568 .go_sync = rgrp_go_sync,
569 .go_inval = rgrp_go_inval,
570 .go_lock = gfs2_rgrp_go_lock,
571 .go_unlock = gfs2_rgrp_go_unlock,
572 .go_dump = gfs2_rgrp_dump,
573 .go_type = LM_TYPE_RGRP,
574 .go_flags = GLOF_LVB,
575 };
576
577 const struct gfs2_glock_operations gfs2_freeze_glops = {
578 .go_sync = freeze_go_sync,
579 .go_xmote_bh = freeze_go_xmote_bh,
580 .go_demote_ok = freeze_go_demote_ok,
581 .go_type = LM_TYPE_NONDISK,
582 };
583
584 const struct gfs2_glock_operations gfs2_iopen_glops = {
585 .go_type = LM_TYPE_IOPEN,
586 .go_callback = iopen_go_callback,
587 };
588
589 const struct gfs2_glock_operations gfs2_flock_glops = {
590 .go_type = LM_TYPE_FLOCK,
591 };
592
593 const struct gfs2_glock_operations gfs2_nondisk_glops = {
594 .go_type = LM_TYPE_NONDISK,
595 };
596
597 const struct gfs2_glock_operations gfs2_quota_glops = {
598 .go_type = LM_TYPE_QUOTA,
599 .go_flags = GLOF_LVB,
600 };
601
602 const struct gfs2_glock_operations gfs2_journal_glops = {
603 .go_type = LM_TYPE_JOURNAL,
604 };
605
606 const struct gfs2_glock_operations *gfs2_glops_list[] = {
607 [LM_TYPE_META] = &gfs2_meta_glops,
608 [LM_TYPE_INODE] = &gfs2_inode_glops,
609 [LM_TYPE_RGRP] = &gfs2_rgrp_glops,
610 [LM_TYPE_IOPEN] = &gfs2_iopen_glops,
611 [LM_TYPE_FLOCK] = &gfs2_flock_glops,
612 [LM_TYPE_NONDISK] = &gfs2_nondisk_glops,
613 [LM_TYPE_QUOTA] = &gfs2_quota_glops,
614 [LM_TYPE_JOURNAL] = &gfs2_journal_glops,
615 };
616