Merge branch 'for-2.6.29' of git://linux-nfs.org/~bfields/linux
[linux-2.6] / fs / gfs2 / super.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2007 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/sched.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/crc32.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <linux/bio.h>
18 #include <linux/lm_interface.h>
19
20 #include "gfs2.h"
21 #include "incore.h"
22 #include "bmap.h"
23 #include "dir.h"
24 #include "glock.h"
25 #include "glops.h"
26 #include "inode.h"
27 #include "log.h"
28 #include "meta_io.h"
29 #include "quota.h"
30 #include "recovery.h"
31 #include "rgrp.h"
32 #include "super.h"
33 #include "trans.h"
34 #include "util.h"
35
36 /**
37  * gfs2_jindex_free - Clear all the journal index information
38  * @sdp: The GFS2 superblock
39  *
40  */
41
42 void gfs2_jindex_free(struct gfs2_sbd *sdp)
43 {
44         struct list_head list, *head;
45         struct gfs2_jdesc *jd;
46         struct gfs2_journal_extent *jext;
47
48         spin_lock(&sdp->sd_jindex_spin);
49         list_add(&list, &sdp->sd_jindex_list);
50         list_del_init(&sdp->sd_jindex_list);
51         sdp->sd_journals = 0;
52         spin_unlock(&sdp->sd_jindex_spin);
53
54         while (!list_empty(&list)) {
55                 jd = list_entry(list.next, struct gfs2_jdesc, jd_list);
56                 head = &jd->extent_list;
57                 while (!list_empty(head)) {
58                         jext = list_entry(head->next,
59                                           struct gfs2_journal_extent,
60                                           extent_list);
61                         list_del(&jext->extent_list);
62                         kfree(jext);
63                 }
64                 list_del(&jd->jd_list);
65                 iput(jd->jd_inode);
66                 kfree(jd);
67         }
68 }
69
70 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid)
71 {
72         struct gfs2_jdesc *jd;
73         int found = 0;
74
75         list_for_each_entry(jd, head, jd_list) {
76                 if (jd->jd_jid == jid) {
77                         found = 1;
78                         break;
79                 }
80         }
81
82         if (!found)
83                 jd = NULL;
84
85         return jd;
86 }
87
88 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid)
89 {
90         struct gfs2_jdesc *jd;
91
92         spin_lock(&sdp->sd_jindex_spin);
93         jd = jdesc_find_i(&sdp->sd_jindex_list, jid);
94         spin_unlock(&sdp->sd_jindex_spin);
95
96         return jd;
97 }
98
99 int gfs2_jdesc_check(struct gfs2_jdesc *jd)
100 {
101         struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
102         struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
103         int ar;
104         int error;
105
106         if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) ||
107             (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) {
108                 gfs2_consist_inode(ip);
109                 return -EIO;
110         }
111         jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
112
113         error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar);
114         if (!error && ar) {
115                 gfs2_consist_inode(ip);
116                 error = -EIO;
117         }
118
119         return error;
120 }
121
122 /**
123  * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one
124  * @sdp: the filesystem
125  *
126  * Returns: errno
127  */
128
129 int gfs2_make_fs_rw(struct gfs2_sbd *sdp)
130 {
131         struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode);
132         struct gfs2_glock *j_gl = ip->i_gl;
133         struct gfs2_holder t_gh;
134         struct gfs2_log_header_host head;
135         int error;
136
137         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh);
138         if (error)
139                 return error;
140
141         j_gl->gl_ops->go_inval(j_gl, DIO_METADATA);
142
143         error = gfs2_find_jhead(sdp->sd_jdesc, &head);
144         if (error)
145                 goto fail;
146
147         if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
148                 gfs2_consist(sdp);
149                 error = -EIO;
150                 goto fail;
151         }
152
153         /*  Initialize some head of the log stuff  */
154         sdp->sd_log_sequence = head.lh_sequence + 1;
155         gfs2_log_pointers_init(sdp, head.lh_blkno);
156
157         error = gfs2_quota_init(sdp);
158         if (error)
159                 goto fail;
160
161         set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags);
162
163         gfs2_glock_dq_uninit(&t_gh);
164
165         return 0;
166
167 fail:
168         t_gh.gh_flags |= GL_NOCACHE;
169         gfs2_glock_dq_uninit(&t_gh);
170
171         return error;
172 }
173
174 static void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf)
175 {
176         const struct gfs2_statfs_change *str = buf;
177
178         sc->sc_total = be64_to_cpu(str->sc_total);
179         sc->sc_free = be64_to_cpu(str->sc_free);
180         sc->sc_dinodes = be64_to_cpu(str->sc_dinodes);
181 }
182
183 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf)
184 {
185         struct gfs2_statfs_change *str = buf;
186
187         str->sc_total = cpu_to_be64(sc->sc_total);
188         str->sc_free = cpu_to_be64(sc->sc_free);
189         str->sc_dinodes = cpu_to_be64(sc->sc_dinodes);
190 }
191
192 int gfs2_statfs_init(struct gfs2_sbd *sdp)
193 {
194         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
195         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
196         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
197         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
198         struct buffer_head *m_bh, *l_bh;
199         struct gfs2_holder gh;
200         int error;
201
202         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
203                                    &gh);
204         if (error)
205                 return error;
206
207         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
208         if (error)
209                 goto out;
210
211         if (sdp->sd_args.ar_spectator) {
212                 spin_lock(&sdp->sd_statfs_spin);
213                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
214                                       sizeof(struct gfs2_dinode));
215                 spin_unlock(&sdp->sd_statfs_spin);
216         } else {
217                 error = gfs2_meta_inode_buffer(l_ip, &l_bh);
218                 if (error)
219                         goto out_m_bh;
220
221                 spin_lock(&sdp->sd_statfs_spin);
222                 gfs2_statfs_change_in(m_sc, m_bh->b_data +
223                                       sizeof(struct gfs2_dinode));
224                 gfs2_statfs_change_in(l_sc, l_bh->b_data +
225                                       sizeof(struct gfs2_dinode));
226                 spin_unlock(&sdp->sd_statfs_spin);
227
228                 brelse(l_bh);
229         }
230
231 out_m_bh:
232         brelse(m_bh);
233 out:
234         gfs2_glock_dq_uninit(&gh);
235         return 0;
236 }
237
238 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free,
239                         s64 dinodes)
240 {
241         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
242         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
243         struct buffer_head *l_bh;
244         int error;
245
246         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
247         if (error)
248                 return;
249
250         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
251
252         spin_lock(&sdp->sd_statfs_spin);
253         l_sc->sc_total += total;
254         l_sc->sc_free += free;
255         l_sc->sc_dinodes += dinodes;
256         gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode));
257         spin_unlock(&sdp->sd_statfs_spin);
258
259         brelse(l_bh);
260 }
261
262 int gfs2_statfs_sync(struct gfs2_sbd *sdp)
263 {
264         struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
265         struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode);
266         struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master;
267         struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local;
268         struct gfs2_holder gh;
269         struct buffer_head *m_bh, *l_bh;
270         int error;
271
272         error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE,
273                                    &gh);
274         if (error)
275                 return error;
276
277         error = gfs2_meta_inode_buffer(m_ip, &m_bh);
278         if (error)
279                 goto out;
280
281         spin_lock(&sdp->sd_statfs_spin);
282         gfs2_statfs_change_in(m_sc, m_bh->b_data +
283                               sizeof(struct gfs2_dinode));
284         if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) {
285                 spin_unlock(&sdp->sd_statfs_spin);
286                 goto out_bh;
287         }
288         spin_unlock(&sdp->sd_statfs_spin);
289
290         error = gfs2_meta_inode_buffer(l_ip, &l_bh);
291         if (error)
292                 goto out_bh;
293
294         error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
295         if (error)
296                 goto out_bh2;
297
298         gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1);
299
300         spin_lock(&sdp->sd_statfs_spin);
301         m_sc->sc_total += l_sc->sc_total;
302         m_sc->sc_free += l_sc->sc_free;
303         m_sc->sc_dinodes += l_sc->sc_dinodes;
304         memset(l_sc, 0, sizeof(struct gfs2_statfs_change));
305         memset(l_bh->b_data + sizeof(struct gfs2_dinode),
306                0, sizeof(struct gfs2_statfs_change));
307         spin_unlock(&sdp->sd_statfs_spin);
308
309         gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
310         gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode));
311
312         gfs2_trans_end(sdp);
313
314 out_bh2:
315         brelse(l_bh);
316 out_bh:
317         brelse(m_bh);
318 out:
319         gfs2_glock_dq_uninit(&gh);
320         return error;
321 }
322
323 struct lfcc {
324         struct list_head list;
325         struct gfs2_holder gh;
326 };
327
328 /**
329  * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all
330  *                            journals are clean
331  * @sdp: the file system
332  * @state: the state to put the transaction lock into
333  * @t_gh: the hold on the transaction lock
334  *
335  * Returns: errno
336  */
337
338 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp,
339                                     struct gfs2_holder *t_gh)
340 {
341         struct gfs2_inode *ip;
342         struct gfs2_holder ji_gh;
343         struct gfs2_jdesc *jd;
344         struct lfcc *lfcc;
345         LIST_HEAD(list);
346         struct gfs2_log_header_host lh;
347         int error;
348
349         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
350                 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL);
351                 if (!lfcc) {
352                         error = -ENOMEM;
353                         goto out;
354                 }
355                 ip = GFS2_I(jd->jd_inode);
356                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh);
357                 if (error) {
358                         kfree(lfcc);
359                         goto out;
360                 }
361                 list_add(&lfcc->list, &list);
362         }
363
364         error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED,
365                                    GL_NOCACHE, t_gh);
366
367         list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) {
368                 error = gfs2_jdesc_check(jd);
369                 if (error)
370                         break;
371                 error = gfs2_find_jhead(jd, &lh);
372                 if (error)
373                         break;
374                 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
375                         error = -EBUSY;
376                         break;
377                 }
378         }
379
380         if (error)
381                 gfs2_glock_dq_uninit(t_gh);
382
383 out:
384         while (!list_empty(&list)) {
385                 lfcc = list_entry(list.next, struct lfcc, list);
386                 list_del(&lfcc->list);
387                 gfs2_glock_dq_uninit(&lfcc->gh);
388                 kfree(lfcc);
389         }
390         gfs2_glock_dq_uninit(&ji_gh);
391         return error;
392 }
393
394 /**
395  * gfs2_freeze_fs - freezes the file system
396  * @sdp: the file system
397  *
398  * This function flushes data and meta data for all machines by
399  * aquiring the transaction log exclusively.  All journals are
400  * ensured to be in a clean state as well.
401  *
402  * Returns: errno
403  */
404
405 int gfs2_freeze_fs(struct gfs2_sbd *sdp)
406 {
407         int error = 0;
408
409         mutex_lock(&sdp->sd_freeze_lock);
410
411         if (!sdp->sd_freeze_count++) {
412                 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh);
413                 if (error)
414                         sdp->sd_freeze_count--;
415         }
416
417         mutex_unlock(&sdp->sd_freeze_lock);
418
419         return error;
420 }
421
422 /**
423  * gfs2_unfreeze_fs - unfreezes the file system
424  * @sdp: the file system
425  *
426  * This function allows the file system to proceed by unlocking
427  * the exclusively held transaction lock.  Other GFS2 nodes are
428  * now free to acquire the lock shared and go on with their lives.
429  *
430  */
431
432 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp)
433 {
434         mutex_lock(&sdp->sd_freeze_lock);
435
436         if (sdp->sd_freeze_count && !--sdp->sd_freeze_count)
437                 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh);
438
439         mutex_unlock(&sdp->sd_freeze_lock);
440 }
441