]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3a8a9a10 | 3 | * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
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 | |
e9fc2aa0 | 7 | * of the GNU General Public License version 2. |
b3b94faa DT |
8 | */ |
9 | ||
10 | #include <linux/sched.h> | |
11 | #include <linux/slab.h> | |
12 | #include <linux/spinlock.h> | |
13 | #include <linux/completion.h> | |
14 | #include <linux/buffer_head.h> | |
b3b94faa DT |
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> | |
5c676f6d | 20 | #include <linux/gfs2_ondisk.h> |
feaa7bba | 21 | #include <linux/crc32.h> |
7d308590 | 22 | #include <linux/lm_interface.h> |
b3b94faa DT |
23 | |
24 | #include "gfs2.h" | |
5c676f6d | 25 | #include "incore.h" |
b3b94faa DT |
26 | #include "glock.h" |
27 | #include "inode.h" | |
28 | #include "lm.h" | |
29 | #include "log.h" | |
30 | #include "mount.h" | |
31 | #include "ops_super.h" | |
b3b94faa DT |
32 | #include "quota.h" |
33 | #include "recovery.h" | |
34 | #include "rgrp.h" | |
35 | #include "super.h" | |
36 | #include "sys.h" | |
5c676f6d | 37 | #include "util.h" |
feaa7bba SW |
38 | #include "trans.h" |
39 | #include "dir.h" | |
40 | #include "eattr.h" | |
41 | #include "bmap.h" | |
b3b94faa DT |
42 | |
43 | /** | |
44 | * gfs2_write_inode - Make sure the inode is stable on the disk | |
45 | * @inode: The inode | |
46 | * @sync: synchronous write flag | |
47 | * | |
48 | * Returns: errno | |
49 | */ | |
50 | ||
51 | static int gfs2_write_inode(struct inode *inode, int sync) | |
52 | { | |
feaa7bba SW |
53 | struct gfs2_inode *ip = GFS2_I(inode); |
54 | ||
55 | /* Check this is a "normal" inode */ | |
bba9dfd8 | 56 | if (inode->i_private) { |
feaa7bba SW |
57 | if (current->flags & PF_MEMALLOC) |
58 | return 0; | |
59 | if (sync) | |
3a8476dd | 60 | gfs2_log_flush(GFS2_SB(inode), ip->i_gl); |
feaa7bba | 61 | } |
b3b94faa DT |
62 | |
63 | return 0; | |
64 | } | |
65 | ||
66 | /** | |
67 | * gfs2_put_super - Unmount the filesystem | |
68 | * @sb: The VFS superblock | |
69 | * | |
70 | */ | |
71 | ||
72 | static void gfs2_put_super(struct super_block *sb) | |
73 | { | |
5c676f6d | 74 | struct gfs2_sbd *sdp = sb->s_fs_info; |
b3b94faa DT |
75 | int error; |
76 | ||
77 | if (!sdp) | |
78 | return; | |
79 | ||
86384605 | 80 | if (!strncmp(sb->s_type->name, "gfs2meta", 8)) |
5acd3967 | 81 | return; /* Nothing to do */ |
86384605 | 82 | |
b3b94faa DT |
83 | /* Unfreeze the filesystem, if we need to */ |
84 | ||
f55ab26a | 85 | mutex_lock(&sdp->sd_freeze_lock); |
b3b94faa DT |
86 | if (sdp->sd_freeze_count) |
87 | gfs2_glock_dq_uninit(&sdp->sd_freeze_gh); | |
f55ab26a | 88 | mutex_unlock(&sdp->sd_freeze_lock); |
b3b94faa | 89 | |
b3b94faa DT |
90 | kthread_stop(sdp->sd_quotad_process); |
91 | kthread_stop(sdp->sd_logd_process); | |
92 | kthread_stop(sdp->sd_recoverd_process); | |
93 | while (sdp->sd_glockd_num--) | |
94 | kthread_stop(sdp->sd_glockd_process[sdp->sd_glockd_num]); | |
b3b94faa DT |
95 | |
96 | if (!(sb->s_flags & MS_RDONLY)) { | |
97 | error = gfs2_make_fs_ro(sdp); | |
98 | if (error) | |
99 | gfs2_io_error(sdp); | |
100 | } | |
b3b94faa DT |
101 | /* At this point, we're through modifying the disk */ |
102 | ||
103 | /* Release stuff */ | |
104 | ||
f42faf4f SW |
105 | iput(sdp->sd_master_dir); |
106 | iput(sdp->sd_jindex); | |
107 | iput(sdp->sd_inum_inode); | |
108 | iput(sdp->sd_statfs_inode); | |
109 | iput(sdp->sd_rindex); | |
110 | iput(sdp->sd_quota_inode); | |
b3b94faa DT |
111 | |
112 | gfs2_glock_put(sdp->sd_rename_gl); | |
113 | gfs2_glock_put(sdp->sd_trans_gl); | |
114 | ||
115 | if (!sdp->sd_args.ar_spectator) { | |
116 | gfs2_glock_dq_uninit(&sdp->sd_journal_gh); | |
117 | gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); | |
118 | gfs2_glock_dq_uninit(&sdp->sd_ir_gh); | |
119 | gfs2_glock_dq_uninit(&sdp->sd_sc_gh); | |
b3b94faa | 120 | gfs2_glock_dq_uninit(&sdp->sd_qc_gh); |
f42faf4f SW |
121 | iput(sdp->sd_ir_inode); |
122 | iput(sdp->sd_sc_inode); | |
f42faf4f | 123 | iput(sdp->sd_qc_inode); |
b3b94faa DT |
124 | } |
125 | ||
126 | gfs2_glock_dq_uninit(&sdp->sd_live_gh); | |
b3b94faa DT |
127 | gfs2_clear_rgrpd(sdp); |
128 | gfs2_jindex_free(sdp); | |
b3b94faa DT |
129 | /* Take apart glock structures and buffer lists */ |
130 | gfs2_gl_hash_clear(sdp, WAIT); | |
b3b94faa DT |
131 | /* Unmount the locking protocol */ |
132 | gfs2_lm_unmount(sdp); | |
133 | ||
134 | /* At this point, we're through participating in the lockspace */ | |
b3b94faa | 135 | gfs2_sys_fs_del(sdp); |
ff6af411 | 136 | kfree(sdp); |
b3b94faa DT |
137 | } |
138 | ||
139 | /** | |
4a221953 SW |
140 | * gfs2_write_super |
141 | * @sb: the superblock | |
b3b94faa | 142 | * |
b3b94faa DT |
143 | */ |
144 | ||
145 | static void gfs2_write_super(struct super_block *sb) | |
146 | { | |
4a221953 SW |
147 | sb->s_dirt = 0; |
148 | } | |
149 | ||
150 | /** | |
151 | * gfs2_sync_fs - sync the filesystem | |
152 | * @sb: the superblock | |
153 | * | |
154 | * Flushes the log to disk. | |
155 | */ | |
156 | static int gfs2_sync_fs(struct super_block *sb, int wait) | |
157 | { | |
158 | sb->s_dirt = 0; | |
b004157a SW |
159 | if (wait) |
160 | gfs2_log_flush(sb->s_fs_info, NULL); | |
4a221953 | 161 | return 0; |
b3b94faa DT |
162 | } |
163 | ||
164 | /** | |
165 | * gfs2_write_super_lockfs - prevent further writes to the filesystem | |
166 | * @sb: the VFS structure for the filesystem | |
167 | * | |
168 | */ | |
169 | ||
170 | static void gfs2_write_super_lockfs(struct super_block *sb) | |
171 | { | |
5c676f6d | 172 | struct gfs2_sbd *sdp = sb->s_fs_info; |
b3b94faa DT |
173 | int error; |
174 | ||
c3780511 DT |
175 | if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) |
176 | return; | |
177 | ||
b3b94faa DT |
178 | for (;;) { |
179 | error = gfs2_freeze_fs(sdp); | |
180 | if (!error) | |
181 | break; | |
182 | ||
183 | switch (error) { | |
184 | case -EBUSY: | |
185 | fs_err(sdp, "waiting for recovery before freeze\n"); | |
186 | break; | |
187 | ||
188 | default: | |
189 | fs_err(sdp, "error freezing FS: %d\n", error); | |
190 | break; | |
191 | } | |
192 | ||
193 | fs_err(sdp, "retrying...\n"); | |
194 | msleep(1000); | |
195 | } | |
196 | } | |
197 | ||
198 | /** | |
199 | * gfs2_unlockfs - reallow writes to the filesystem | |
200 | * @sb: the VFS structure for the filesystem | |
201 | * | |
202 | */ | |
203 | ||
204 | static void gfs2_unlockfs(struct super_block *sb) | |
205 | { | |
2bdbc5d7 | 206 | gfs2_unfreeze_fs(sb->s_fs_info); |
b3b94faa DT |
207 | } |
208 | ||
209 | /** | |
210 | * gfs2_statfs - Gather and return stats about the filesystem | |
211 | * @sb: The superblock | |
212 | * @statfsbuf: The buffer | |
213 | * | |
214 | * Returns: 0 on success or error code | |
215 | */ | |
216 | ||
0c0834a3 | 217 | static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) |
b3b94faa | 218 | { |
0c0834a3 | 219 | struct super_block *sb = dentry->d_inode->i_sb; |
5c676f6d | 220 | struct gfs2_sbd *sdp = sb->s_fs_info; |
bd209cc0 | 221 | struct gfs2_statfs_change_host sc; |
b3b94faa DT |
222 | int error; |
223 | ||
b3b94faa DT |
224 | if (gfs2_tune_get(sdp, gt_statfs_slow)) |
225 | error = gfs2_statfs_slow(sdp, &sc); | |
226 | else | |
227 | error = gfs2_statfs_i(sdp, &sc); | |
228 | ||
229 | if (error) | |
230 | return error; | |
231 | ||
b3b94faa DT |
232 | buf->f_type = GFS2_MAGIC; |
233 | buf->f_bsize = sdp->sd_sb.sb_bsize; | |
234 | buf->f_blocks = sc.sc_total; | |
235 | buf->f_bfree = sc.sc_free; | |
236 | buf->f_bavail = sc.sc_free; | |
237 | buf->f_files = sc.sc_dinodes + sc.sc_free; | |
238 | buf->f_ffree = sc.sc_free; | |
239 | buf->f_namelen = GFS2_FNAMESIZE; | |
240 | ||
241 | return 0; | |
242 | } | |
243 | ||
244 | /** | |
245 | * gfs2_remount_fs - called when the FS is remounted | |
246 | * @sb: the filesystem | |
247 | * @flags: the remount flags | |
248 | * @data: extra data passed in (not used right now) | |
249 | * | |
250 | * Returns: errno | |
251 | */ | |
252 | ||
253 | static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data) | |
254 | { | |
5c676f6d | 255 | struct gfs2_sbd *sdp = sb->s_fs_info; |
b3b94faa DT |
256 | int error; |
257 | ||
b3b94faa DT |
258 | error = gfs2_mount_args(sdp, data, 1); |
259 | if (error) | |
260 | return error; | |
261 | ||
262 | if (sdp->sd_args.ar_spectator) | |
263 | *flags |= MS_RDONLY; | |
264 | else { | |
265 | if (*flags & MS_RDONLY) { | |
266 | if (!(sb->s_flags & MS_RDONLY)) | |
267 | error = gfs2_make_fs_ro(sdp); | |
268 | } else if (!(*flags & MS_RDONLY) && | |
269 | (sb->s_flags & MS_RDONLY)) { | |
270 | error = gfs2_make_fs_rw(sdp); | |
271 | } | |
272 | } | |
273 | ||
274 | if (*flags & (MS_NOATIME | MS_NODIRATIME)) | |
275 | set_bit(SDF_NOATIME, &sdp->sd_flags); | |
276 | else | |
277 | clear_bit(SDF_NOATIME, &sdp->sd_flags); | |
278 | ||
279 | /* Don't let the VFS update atimes. GFS2 handles this itself. */ | |
280 | *flags |= MS_NOATIME | MS_NODIRATIME; | |
281 | ||
282 | return error; | |
283 | } | |
284 | ||
3b8249f6 SW |
285 | /** |
286 | * gfs2_drop_inode - Drop an inode (test for remote unlink) | |
287 | * @inode: The inode to drop | |
288 | * | |
289 | * If we've received a callback on an iopen lock then its because a | |
290 | * remote node tried to deallocate the inode but failed due to this node | |
291 | * still having the inode open. Here we mark the link count zero | |
292 | * since we know that it must have reached zero if the GLF_DEMOTE flag | |
293 | * is set on the iopen glock. If we didn't do a disk read since the | |
294 | * remote node removed the final link then we might otherwise miss | |
295 | * this event. This check ensures that this node will deallocate the | |
296 | * inode's blocks, or alternatively pass the baton on to another | |
297 | * node for later deallocation. | |
298 | */ | |
299 | static void gfs2_drop_inode(struct inode *inode) | |
300 | { | |
301 | if (inode->i_private && inode->i_nlink) { | |
302 | struct gfs2_inode *ip = GFS2_I(inode); | |
303 | struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; | |
304 | if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags)) | |
305 | clear_nlink(inode); | |
306 | } | |
307 | generic_drop_inode(inode); | |
308 | } | |
309 | ||
b3b94faa DT |
310 | /** |
311 | * gfs2_clear_inode - Deallocate an inode when VFS is done with it | |
312 | * @inode: The VFS inode | |
313 | * | |
314 | */ | |
315 | ||
316 | static void gfs2_clear_inode(struct inode *inode) | |
317 | { | |
feaa7bba SW |
318 | /* This tells us its a "real" inode and not one which only |
319 | * serves to contain an address space (see rgrp.c, meta_io.c) | |
320 | * which therefore doesn't have its own glocks. | |
321 | */ | |
bba9dfd8 | 322 | if (inode->i_private) { |
feaa7bba | 323 | struct gfs2_inode *ip = GFS2_I(inode); |
feaa7bba | 324 | ip->i_gl->gl_object = NULL; |
b3b94faa | 325 | gfs2_glock_schedule_for_reclaim(ip->i_gl); |
feaa7bba SW |
326 | gfs2_glock_put(ip->i_gl); |
327 | ip->i_gl = NULL; | |
d93cfa98 AD |
328 | if (ip->i_iopen_gh.gh_gl) { |
329 | ip->i_iopen_gh.gh_gl->gl_object = NULL; | |
feaa7bba | 330 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); |
d93cfa98 | 331 | } |
b3b94faa DT |
332 | } |
333 | } | |
334 | ||
335 | /** | |
336 | * gfs2_show_options - Show mount options for /proc/mounts | |
337 | * @s: seq_file structure | |
338 | * @mnt: vfsmount | |
339 | * | |
340 | * Returns: 0 on success or error code | |
341 | */ | |
342 | ||
343 | static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt) | |
344 | { | |
5c676f6d | 345 | struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info; |
b3b94faa DT |
346 | struct gfs2_args *args = &sdp->sd_args; |
347 | ||
b3b94faa DT |
348 | if (args->ar_lockproto[0]) |
349 | seq_printf(s, ",lockproto=%s", args->ar_lockproto); | |
350 | if (args->ar_locktable[0]) | |
351 | seq_printf(s, ",locktable=%s", args->ar_locktable); | |
352 | if (args->ar_hostdata[0]) | |
353 | seq_printf(s, ",hostdata=%s", args->ar_hostdata); | |
354 | if (args->ar_spectator) | |
355 | seq_printf(s, ",spectator"); | |
356 | if (args->ar_ignore_local_fs) | |
357 | seq_printf(s, ",ignore_local_fs"); | |
358 | if (args->ar_localflocks) | |
359 | seq_printf(s, ",localflocks"); | |
360 | if (args->ar_localcaching) | |
361 | seq_printf(s, ",localcaching"); | |
362 | if (args->ar_debug) | |
363 | seq_printf(s, ",debug"); | |
364 | if (args->ar_upgrade) | |
365 | seq_printf(s, ",upgrade"); | |
366 | if (args->ar_num_glockd != GFS2_GLOCKD_DEFAULT) | |
367 | seq_printf(s, ",num_glockd=%u", args->ar_num_glockd); | |
368 | if (args->ar_posix_acl) | |
369 | seq_printf(s, ",acl"); | |
370 | if (args->ar_quota != GFS2_QUOTA_DEFAULT) { | |
371 | char *state; | |
372 | switch (args->ar_quota) { | |
373 | case GFS2_QUOTA_OFF: | |
374 | state = "off"; | |
375 | break; | |
376 | case GFS2_QUOTA_ACCOUNT: | |
377 | state = "account"; | |
378 | break; | |
379 | case GFS2_QUOTA_ON: | |
380 | state = "on"; | |
381 | break; | |
382 | default: | |
383 | state = "unknown"; | |
384 | break; | |
385 | } | |
386 | seq_printf(s, ",quota=%s", state); | |
387 | } | |
388 | if (args->ar_suiddir) | |
389 | seq_printf(s, ",suiddir"); | |
390 | if (args->ar_data != GFS2_DATA_DEFAULT) { | |
391 | char *state; | |
392 | switch (args->ar_data) { | |
393 | case GFS2_DATA_WRITEBACK: | |
394 | state = "writeback"; | |
395 | break; | |
396 | case GFS2_DATA_ORDERED: | |
397 | state = "ordered"; | |
398 | break; | |
399 | default: | |
400 | state = "unknown"; | |
401 | break; | |
402 | } | |
403 | seq_printf(s, ",data=%s", state); | |
404 | } | |
405 | ||
406 | return 0; | |
407 | } | |
408 | ||
907b9bce | 409 | /* |
feaa7bba SW |
410 | * We have to (at the moment) hold the inodes main lock to cover |
411 | * the gap between unlocking the shared lock on the iopen lock and | |
412 | * taking the exclusive lock. I'd rather do a shared -> exclusive | |
413 | * conversion on the iopen lock, but we can change that later. This | |
414 | * is safe, just less efficient. | |
415 | */ | |
416 | static void gfs2_delete_inode(struct inode *inode) | |
417 | { | |
418 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
419 | struct gfs2_inode *ip = GFS2_I(inode); | |
420 | struct gfs2_holder gh; | |
421 | int error; | |
422 | ||
bba9dfd8 | 423 | if (!inode->i_private) |
feaa7bba SW |
424 | goto out; |
425 | ||
d93cfa98 | 426 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); |
feaa7bba SW |
427 | if (unlikely(error)) { |
428 | gfs2_glock_dq_uninit(&ip->i_iopen_gh); | |
429 | goto out; | |
430 | } | |
431 | ||
d93cfa98 | 432 | gfs2_glock_dq_wait(&ip->i_iopen_gh); |
feaa7bba SW |
433 | gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh); |
434 | error = gfs2_glock_nq(&ip->i_iopen_gh); | |
435 | if (error) | |
436 | goto out_uninit; | |
437 | ||
b60623c2 | 438 | if (S_ISDIR(inode->i_mode) && |
feaa7bba SW |
439 | (ip->i_di.di_flags & GFS2_DIF_EXHASH)) { |
440 | error = gfs2_dir_exhash_dealloc(ip); | |
441 | if (error) | |
442 | goto out_unlock; | |
443 | } | |
444 | ||
445 | if (ip->i_di.di_eattr) { | |
446 | error = gfs2_ea_dealloc(ip); | |
447 | if (error) | |
448 | goto out_unlock; | |
449 | } | |
450 | ||
451 | if (!gfs2_is_stuffed(ip)) { | |
452 | error = gfs2_file_dealloc(ip); | |
453 | if (error) | |
454 | goto out_unlock; | |
455 | } | |
456 | ||
457 | error = gfs2_dinode_dealloc(ip); | |
49686f71 SW |
458 | /* |
459 | * Must do this before unlock to avoid trying to write back | |
460 | * potentially dirty data now that inode no longer exists | |
461 | * on disk. | |
462 | */ | |
463 | truncate_inode_pages(&inode->i_data, 0); | |
feaa7bba SW |
464 | |
465 | out_unlock: | |
466 | gfs2_glock_dq(&ip->i_iopen_gh); | |
467 | out_uninit: | |
468 | gfs2_holder_uninit(&ip->i_iopen_gh); | |
469 | gfs2_glock_dq_uninit(&gh); | |
3b8249f6 | 470 | if (error && error != GLR_TRYFAILED) |
feaa7bba SW |
471 | fs_warn(sdp, "gfs2_delete_inode: %d\n", error); |
472 | out: | |
473 | truncate_inode_pages(&inode->i_data, 0); | |
474 | clear_inode(inode); | |
475 | } | |
476 | ||
477 | ||
478 | ||
320dd101 SW |
479 | static struct inode *gfs2_alloc_inode(struct super_block *sb) |
480 | { | |
320dd101 SW |
481 | struct gfs2_inode *ip; |
482 | ||
483 | ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL); | |
484 | if (ip) { | |
485 | ip->i_flags = 0; | |
486 | ip->i_gl = NULL; | |
320dd101 SW |
487 | ip->i_last_pfault = jiffies; |
488 | } | |
489 | return &ip->i_inode; | |
490 | } | |
491 | ||
492 | static void gfs2_destroy_inode(struct inode *inode) | |
493 | { | |
494 | kmem_cache_free(gfs2_inode_cachep, inode); | |
495 | } | |
496 | ||
ee9b6d61 | 497 | const struct super_operations gfs2_super_ops = { |
4a221953 SW |
498 | .alloc_inode = gfs2_alloc_inode, |
499 | .destroy_inode = gfs2_destroy_inode, | |
500 | .write_inode = gfs2_write_inode, | |
501 | .delete_inode = gfs2_delete_inode, | |
502 | .put_super = gfs2_put_super, | |
503 | .write_super = gfs2_write_super, | |
504 | .sync_fs = gfs2_sync_fs, | |
505 | .write_super_lockfs = gfs2_write_super_lockfs, | |
506 | .unlockfs = gfs2_unlockfs, | |
507 | .statfs = gfs2_statfs, | |
508 | .remount_fs = gfs2_remount_fs, | |
509 | .clear_inode = gfs2_clear_inode, | |
3b8249f6 | 510 | .drop_inode = gfs2_drop_inode, |
4a221953 | 511 | .show_options = gfs2_show_options, |
b3b94faa DT |
512 | }; |
513 |