]> Pileus Git - ~andy/linux/blob - fs/ocfs2/quota_global.c
ocfs2: Remove ENTRY from masklog.
[~andy/linux] / fs / ocfs2 / quota_global.c
1 /*
2  *  Implementation of operations over global quota file
3  */
4 #include <linux/spinlock.h>
5 #include <linux/fs.h>
6 #include <linux/slab.h>
7 #include <linux/quota.h>
8 #include <linux/quotaops.h>
9 #include <linux/dqblk_qtree.h>
10 #include <linux/jiffies.h>
11 #include <linux/writeback.h>
12 #include <linux/workqueue.h>
13
14 #define MLOG_MASK_PREFIX ML_QUOTA
15 #include <cluster/masklog.h>
16
17 #include "ocfs2_fs.h"
18 #include "ocfs2.h"
19 #include "alloc.h"
20 #include "blockcheck.h"
21 #include "inode.h"
22 #include "journal.h"
23 #include "file.h"
24 #include "sysfile.h"
25 #include "dlmglue.h"
26 #include "uptodate.h"
27 #include "super.h"
28 #include "buffer_head_io.h"
29 #include "quota.h"
30
31 /*
32  * Locking of quotas with OCFS2 is rather complex. Here are rules that
33  * should be obeyed by all the functions:
34  * - any write of quota structure (either to local or global file) is protected
35  *   by dqio_mutex or dquot->dq_lock.
36  * - any modification of global quota file holds inode cluster lock, i_mutex,
37  *   and ip_alloc_sem of the global quota file (achieved by
38  *   ocfs2_lock_global_qf). It also has to hold qinfo_lock.
39  * - an allocation of new blocks for local quota file is protected by
40  *   its ip_alloc_sem
41  *
42  * A rough sketch of locking dependencies (lf = local file, gf = global file):
43  * Normal filesystem operation:
44  *   start_trans -> dqio_mutex -> write to lf
45  * Syncing of local and global file:
46  *   ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
47  *     write to gf
48  *                                                     -> write to lf
49  * Acquire dquot for the first time:
50  *   dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
51  *                                   -> alloc space for gf
52  *                                   -> start_trans -> qinfo_lock -> write to gf
53  *           -> ip_alloc_sem of lf -> alloc space for lf
54  *           -> write to lf
55  * Release last reference to dquot:
56  *   dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
57  *           -> write to lf
58  * Note that all the above operations also hold the inode cluster lock of lf.
59  * Recovery:
60  *   inode cluster lock of recovered lf
61  *     -> read bitmaps -> ip_alloc_sem of lf
62  *     -> ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
63  *        write to gf
64  */
65
66 static struct workqueue_struct *ocfs2_quota_wq = NULL;
67
68 static void qsync_work_fn(struct work_struct *work);
69
70 static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
71 {
72         struct ocfs2_global_disk_dqblk *d = dp;
73         struct mem_dqblk *m = &dquot->dq_dqb;
74
75         /* Update from disk only entries not set by the admin */
76         if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
77                 m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
78                 m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
79         }
80         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
81                 m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
82         if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
83                 m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
84                 m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
85         }
86         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
87                 m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
88         if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
89                 m->dqb_btime = le64_to_cpu(d->dqb_btime);
90         if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
91                 m->dqb_itime = le64_to_cpu(d->dqb_itime);
92         OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
93 }
94
95 static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
96 {
97         struct ocfs2_global_disk_dqblk *d = dp;
98         struct mem_dqblk *m = &dquot->dq_dqb;
99
100         d->dqb_id = cpu_to_le32(dquot->dq_id);
101         d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
102         d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
103         d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
104         d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
105         d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
106         d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
107         d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
108         d->dqb_btime = cpu_to_le64(m->dqb_btime);
109         d->dqb_itime = cpu_to_le64(m->dqb_itime);
110         d->dqb_pad1 = d->dqb_pad2 = 0;
111 }
112
113 static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
114 {
115         struct ocfs2_global_disk_dqblk *d = dp;
116         struct ocfs2_mem_dqinfo *oinfo =
117                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
118
119         if (qtree_entry_unused(&oinfo->dqi_gi, dp))
120                 return 0;
121         return le32_to_cpu(d->dqb_id) == dquot->dq_id;
122 }
123
124 struct qtree_fmt_operations ocfs2_global_ops = {
125         .mem2disk_dqblk = ocfs2_global_mem2diskdqb,
126         .disk2mem_dqblk = ocfs2_global_disk2memdqb,
127         .is_id = ocfs2_global_is_id,
128 };
129
130 int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
131 {
132         struct ocfs2_disk_dqtrailer *dqt =
133                 ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
134
135         mlog(0, "Validating quota block %llu\n",
136              (unsigned long long)bh->b_blocknr);
137
138         BUG_ON(!buffer_uptodate(bh));
139
140         /*
141          * If the ecc fails, we return the error but otherwise
142          * leave the filesystem running.  We know any error is
143          * local to this block.
144          */
145         return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
146 }
147
148 int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
149                                 struct buffer_head **bhp)
150 {
151         int rc;
152
153         *bhp = NULL;
154         rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
155                                ocfs2_validate_quota_block);
156         if (rc)
157                 mlog_errno(rc);
158         return rc;
159 }
160
161 /* Read data from global quotafile - avoid pagecache and such because we cannot
162  * afford acquiring the locks... We use quota cluster lock to serialize
163  * operations. Caller is responsible for acquiring it. */
164 ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
165                          size_t len, loff_t off)
166 {
167         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
168         struct inode *gqinode = oinfo->dqi_gqinode;
169         loff_t i_size = i_size_read(gqinode);
170         int offset = off & (sb->s_blocksize - 1);
171         sector_t blk = off >> sb->s_blocksize_bits;
172         int err = 0;
173         struct buffer_head *bh;
174         size_t toread, tocopy;
175         u64 pblock = 0, pcount = 0;
176
177         if (off > i_size)
178                 return 0;
179         if (off + len > i_size)
180                 len = i_size - off;
181         toread = len;
182         while (toread > 0) {
183                 tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
184                 if (!pcount) {
185                         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
186                                                           &pcount, NULL);
187                         if (err) {
188                                 mlog_errno(err);
189                                 return err;
190                         }
191                 } else {
192                         pcount--;
193                         pblock++;
194                 }
195                 bh = NULL;
196                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
197                 if (err) {
198                         mlog_errno(err);
199                         return err;
200                 }
201                 memcpy(data, bh->b_data + offset, tocopy);
202                 brelse(bh);
203                 offset = 0;
204                 toread -= tocopy;
205                 data += tocopy;
206                 blk++;
207         }
208         return len;
209 }
210
211 /* Write to quotafile (we know the transaction is already started and has
212  * enough credits) */
213 ssize_t ocfs2_quota_write(struct super_block *sb, int type,
214                           const char *data, size_t len, loff_t off)
215 {
216         struct mem_dqinfo *info = sb_dqinfo(sb, type);
217         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
218         struct inode *gqinode = oinfo->dqi_gqinode;
219         int offset = off & (sb->s_blocksize - 1);
220         sector_t blk = off >> sb->s_blocksize_bits;
221         int err = 0, new = 0, ja_type;
222         struct buffer_head *bh = NULL;
223         handle_t *handle = journal_current_handle();
224         u64 pblock, pcount;
225
226         if (!handle) {
227                 mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
228                      "because transaction was not started.\n",
229                      (unsigned long long)off, (unsigned long long)len);
230                 return -EIO;
231         }
232         if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
233                 WARN_ON(1);
234                 len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
235         }
236
237         if (gqinode->i_size < off + len) {
238                 loff_t rounded_end =
239                                 ocfs2_align_bytes_to_blocks(sb, off + len);
240
241                 /* Space is already allocated in ocfs2_acquire_dquot() */
242                 err = ocfs2_simple_size_update(gqinode,
243                                                oinfo->dqi_gqi_bh,
244                                                rounded_end);
245                 if (err < 0)
246                         goto out;
247                 new = 1;
248         }
249         err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
250         if (err) {
251                 mlog_errno(err);
252                 goto out;
253         }
254         /* Not rewriting whole block? */
255         if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
256             !new) {
257                 err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
258                 ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
259         } else {
260                 bh = sb_getblk(sb, pblock);
261                 if (!bh)
262                         err = -ENOMEM;
263                 ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
264         }
265         if (err) {
266                 mlog_errno(err);
267                 goto out;
268         }
269         lock_buffer(bh);
270         if (new)
271                 memset(bh->b_data, 0, sb->s_blocksize);
272         memcpy(bh->b_data + offset, data, len);
273         flush_dcache_page(bh->b_page);
274         set_buffer_uptodate(bh);
275         unlock_buffer(bh);
276         ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
277         err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
278                                       ja_type);
279         if (err < 0) {
280                 brelse(bh);
281                 goto out;
282         }
283         ocfs2_journal_dirty(handle, bh);
284         brelse(bh);
285 out:
286         if (err) {
287                 mlog_errno(err);
288                 return err;
289         }
290         gqinode->i_version++;
291         ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
292         return len;
293 }
294
295 int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
296 {
297         int status;
298         struct buffer_head *bh = NULL;
299
300         status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
301         if (status < 0)
302                 return status;
303         spin_lock(&dq_data_lock);
304         if (!oinfo->dqi_gqi_count++)
305                 oinfo->dqi_gqi_bh = bh;
306         else
307                 WARN_ON(bh != oinfo->dqi_gqi_bh);
308         spin_unlock(&dq_data_lock);
309         if (ex) {
310                 mutex_lock(&oinfo->dqi_gqinode->i_mutex);
311                 down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
312         } else {
313                 down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
314         }
315         return 0;
316 }
317
318 void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
319 {
320         if (ex) {
321                 up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
322                 mutex_unlock(&oinfo->dqi_gqinode->i_mutex);
323         } else {
324                 up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
325         }
326         ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
327         brelse(oinfo->dqi_gqi_bh);
328         spin_lock(&dq_data_lock);
329         if (!--oinfo->dqi_gqi_count)
330                 oinfo->dqi_gqi_bh = NULL;
331         spin_unlock(&dq_data_lock);
332 }
333
334 /* Read information header from global quota file */
335 int ocfs2_global_read_info(struct super_block *sb, int type)
336 {
337         struct inode *gqinode = NULL;
338         unsigned int ino[MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
339                                         GROUP_QUOTA_SYSTEM_INODE };
340         struct ocfs2_global_disk_dqinfo dinfo;
341         struct mem_dqinfo *info = sb_dqinfo(sb, type);
342         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
343         u64 pcount;
344         int status;
345
346         /* Read global header */
347         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
348                         OCFS2_INVALID_SLOT);
349         if (!gqinode) {
350                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
351                         type);
352                 status = -EINVAL;
353                 goto out_err;
354         }
355         oinfo->dqi_gi.dqi_sb = sb;
356         oinfo->dqi_gi.dqi_type = type;
357         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
358         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
359         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
360         oinfo->dqi_gqi_bh = NULL;
361         oinfo->dqi_gqi_count = 0;
362         oinfo->dqi_gqinode = gqinode;
363         status = ocfs2_lock_global_qf(oinfo, 0);
364         if (status < 0) {
365                 mlog_errno(status);
366                 goto out_err;
367         }
368
369         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
370                                              &pcount, NULL);
371         if (status < 0)
372                 goto out_unlock;
373
374         status = ocfs2_qinfo_lock(oinfo, 0);
375         if (status < 0)
376                 goto out_unlock;
377         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
378                                       sizeof(struct ocfs2_global_disk_dqinfo),
379                                       OCFS2_GLOBAL_INFO_OFF);
380         ocfs2_qinfo_unlock(oinfo, 0);
381         ocfs2_unlock_global_qf(oinfo, 0);
382         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
383                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
384                      status);
385                 if (status >= 0)
386                         status = -EIO;
387                 mlog_errno(status);
388                 goto out_err;
389         }
390         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
391         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
392         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
393         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
394         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
395         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
396         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
397         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
398                                                 OCFS2_QBLK_RESERVED_SPACE;
399         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
400         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
401         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
402                            msecs_to_jiffies(oinfo->dqi_syncms));
403
404 out_err:
405         mlog_exit(status);
406         return status;
407 out_unlock:
408         ocfs2_unlock_global_qf(oinfo, 0);
409         mlog_errno(status);
410         goto out_err;
411 }
412
413 /* Write information to global quota file. Expects exlusive lock on quota
414  * file inode and quota info */
415 static int __ocfs2_global_write_info(struct super_block *sb, int type)
416 {
417         struct mem_dqinfo *info = sb_dqinfo(sb, type);
418         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
419         struct ocfs2_global_disk_dqinfo dinfo;
420         ssize_t size;
421
422         spin_lock(&dq_data_lock);
423         info->dqi_flags &= ~DQF_INFO_DIRTY;
424         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
425         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
426         spin_unlock(&dq_data_lock);
427         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
428         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
429         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
430         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
431         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
432                                      sizeof(struct ocfs2_global_disk_dqinfo),
433                                      OCFS2_GLOBAL_INFO_OFF);
434         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
435                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
436                 if (size >= 0)
437                         size = -EIO;
438                 return size;
439         }
440         return 0;
441 }
442
443 int ocfs2_global_write_info(struct super_block *sb, int type)
444 {
445         int err;
446         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
447
448         err = ocfs2_qinfo_lock(info, 1);
449         if (err < 0)
450                 return err;
451         err = __ocfs2_global_write_info(sb, type);
452         ocfs2_qinfo_unlock(info, 1);
453         return err;
454 }
455
456 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
457 {
458         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
459
460         /*
461          * We may need to allocate tree blocks and a leaf block but not the
462          * root block
463          */
464         return oinfo->dqi_gi.dqi_qtree_depth;
465 }
466
467 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
468 {
469         /* We modify all the allocated blocks, tree root, info block and
470          * the inode */
471         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
472                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
473 }
474
475 /* Sync local information about quota modifications with global quota file.
476  * Caller must have started the transaction and obtained exclusive lock for
477  * global quota file inode */
478 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
479 {
480         int err, err2;
481         struct super_block *sb = dquot->dq_sb;
482         int type = dquot->dq_type;
483         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
484         struct ocfs2_global_disk_dqblk dqblk;
485         s64 spacechange, inodechange;
486         time_t olditime, oldbtime;
487
488         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
489                                    sizeof(struct ocfs2_global_disk_dqblk),
490                                    dquot->dq_off);
491         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
492                 if (err >= 0) {
493                         mlog(ML_ERROR, "Short read from global quota file "
494                                        "(%u read)\n", err);
495                         err = -EIO;
496                 }
497                 goto out;
498         }
499
500         /* Update space and inode usage. Get also other information from
501          * global quota file so that we don't overwrite any changes there.
502          * We are */
503         spin_lock(&dq_data_lock);
504         spacechange = dquot->dq_dqb.dqb_curspace -
505                                         OCFS2_DQUOT(dquot)->dq_origspace;
506         inodechange = dquot->dq_dqb.dqb_curinodes -
507                                         OCFS2_DQUOT(dquot)->dq_originodes;
508         olditime = dquot->dq_dqb.dqb_itime;
509         oldbtime = dquot->dq_dqb.dqb_btime;
510         ocfs2_global_disk2memdqb(dquot, &dqblk);
511         mlog(0, "Syncing global dquot %u space %lld+%lld, inodes %lld+%lld\n",
512              dquot->dq_id, dquot->dq_dqb.dqb_curspace, (long long)spacechange,
513              dquot->dq_dqb.dqb_curinodes, (long long)inodechange);
514         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
515                 dquot->dq_dqb.dqb_curspace += spacechange;
516         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
517                 dquot->dq_dqb.dqb_curinodes += inodechange;
518         /* Set properly space grace time... */
519         if (dquot->dq_dqb.dqb_bsoftlimit &&
520             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
521                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
522                     oldbtime > 0) {
523                         if (dquot->dq_dqb.dqb_btime > 0)
524                                 dquot->dq_dqb.dqb_btime =
525                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
526                         else
527                                 dquot->dq_dqb.dqb_btime = oldbtime;
528                 }
529         } else {
530                 dquot->dq_dqb.dqb_btime = 0;
531                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
532         }
533         /* Set properly inode grace time... */
534         if (dquot->dq_dqb.dqb_isoftlimit &&
535             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
536                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
537                     olditime > 0) {
538                         if (dquot->dq_dqb.dqb_itime > 0)
539                                 dquot->dq_dqb.dqb_itime =
540                                         min(dquot->dq_dqb.dqb_itime, olditime);
541                         else
542                                 dquot->dq_dqb.dqb_itime = olditime;
543                 }
544         } else {
545                 dquot->dq_dqb.dqb_itime = 0;
546                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
547         }
548         /* All information is properly updated, clear the flags */
549         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
550         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
551         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
552         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
553         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
554         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
555         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
556         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
557         spin_unlock(&dq_data_lock);
558         err = ocfs2_qinfo_lock(info, freeing);
559         if (err < 0) {
560                 mlog(ML_ERROR, "Failed to lock quota info, loosing quota write"
561                                " (type=%d, id=%u)\n", dquot->dq_type,
562                                (unsigned)dquot->dq_id);
563                 goto out;
564         }
565         if (freeing)
566                 OCFS2_DQUOT(dquot)->dq_use_count--;
567         err = qtree_write_dquot(&info->dqi_gi, dquot);
568         if (err < 0)
569                 goto out_qlock;
570         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
571                 err = qtree_release_dquot(&info->dqi_gi, dquot);
572                 if (info_dirty(sb_dqinfo(sb, type))) {
573                         err2 = __ocfs2_global_write_info(sb, type);
574                         if (!err)
575                                 err = err2;
576                 }
577         }
578 out_qlock:
579         ocfs2_qinfo_unlock(info, freeing);
580 out:
581         if (err < 0)
582                 mlog_errno(err);
583         return err;
584 }
585
586 /*
587  *  Functions for periodic syncing of dquots with global file
588  */
589 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
590 {
591         handle_t *handle;
592         struct super_block *sb = dquot->dq_sb;
593         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
594         struct ocfs2_super *osb = OCFS2_SB(sb);
595         int status = 0;
596
597         mlog(0, "id=%u qtype=%u type=%lu device=%s\n", dquot->dq_id,
598                    dquot->dq_type, type, sb->s_id);
599         if (type != dquot->dq_type)
600                 goto out;
601         status = ocfs2_lock_global_qf(oinfo, 1);
602         if (status < 0)
603                 goto out;
604
605         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
606         if (IS_ERR(handle)) {
607                 status = PTR_ERR(handle);
608                 mlog_errno(status);
609                 goto out_ilock;
610         }
611         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
612         status = ocfs2_sync_dquot(dquot);
613         if (status < 0)
614                 mlog_errno(status);
615         /* We have to write local structure as well... */
616         status = ocfs2_local_write_dquot(dquot);
617         if (status < 0)
618                 mlog_errno(status);
619         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
620         ocfs2_commit_trans(osb, handle);
621 out_ilock:
622         ocfs2_unlock_global_qf(oinfo, 1);
623 out:
624         mlog_exit(status);
625         return status;
626 }
627
628 static void qsync_work_fn(struct work_struct *work)
629 {
630         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
631                                                       struct ocfs2_mem_dqinfo,
632                                                       dqi_sync_work.work);
633         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
634
635         dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
636         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
637                            msecs_to_jiffies(oinfo->dqi_syncms));
638 }
639
640 /*
641  *  Wrappers for generic quota functions
642  */
643
644 static int ocfs2_write_dquot(struct dquot *dquot)
645 {
646         handle_t *handle;
647         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
648         int status = 0;
649
650         mlog(0, "id=%u, type=%d", dquot->dq_id, dquot->dq_type);
651
652         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
653         if (IS_ERR(handle)) {
654                 status = PTR_ERR(handle);
655                 mlog_errno(status);
656                 goto out;
657         }
658         mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
659         status = ocfs2_local_write_dquot(dquot);
660         mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
661         ocfs2_commit_trans(osb, handle);
662 out:
663         mlog_exit(status);
664         return status;
665 }
666
667 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
668 {
669         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
670         /*
671          * We modify tree, leaf block, global info, local chunk header,
672          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
673          * accounts for inode update
674          */
675         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
676                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
677                OCFS2_QINFO_WRITE_CREDITS +
678                OCFS2_INODE_UPDATE_CREDITS;
679 }
680
681 static int ocfs2_release_dquot(struct dquot *dquot)
682 {
683         handle_t *handle;
684         struct ocfs2_mem_dqinfo *oinfo =
685                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
686         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
687         int status = 0;
688
689         mlog(0, "id=%u, type=%d", dquot->dq_id, dquot->dq_type);
690
691         mutex_lock(&dquot->dq_lock);
692         /* Check whether we are not racing with some other dqget() */
693         if (atomic_read(&dquot->dq_count) > 1)
694                 goto out;
695         status = ocfs2_lock_global_qf(oinfo, 1);
696         if (status < 0)
697                 goto out;
698         handle = ocfs2_start_trans(osb,
699                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
700         if (IS_ERR(handle)) {
701                 status = PTR_ERR(handle);
702                 mlog_errno(status);
703                 goto out_ilock;
704         }
705
706         status = ocfs2_global_release_dquot(dquot);
707         if (status < 0) {
708                 mlog_errno(status);
709                 goto out_trans;
710         }
711         status = ocfs2_local_release_dquot(handle, dquot);
712         /*
713          * If we fail here, we cannot do much as global structure is
714          * already released. So just complain...
715          */
716         if (status < 0)
717                 mlog_errno(status);
718         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
719 out_trans:
720         ocfs2_commit_trans(osb, handle);
721 out_ilock:
722         ocfs2_unlock_global_qf(oinfo, 1);
723 out:
724         mutex_unlock(&dquot->dq_lock);
725         mlog_exit(status);
726         return status;
727 }
728
729 /*
730  * Read global dquot structure from disk or create it if it does
731  * not exist. Also update use count of the global structure and
732  * create structure in node-local quota file.
733  */
734 static int ocfs2_acquire_dquot(struct dquot *dquot)
735 {
736         int status = 0, err;
737         int ex = 0;
738         struct super_block *sb = dquot->dq_sb;
739         struct ocfs2_super *osb = OCFS2_SB(sb);
740         int type = dquot->dq_type;
741         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
742         struct inode *gqinode = info->dqi_gqinode;
743         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
744         handle_t *handle;
745
746         mlog(0, "id=%u, type=%d", dquot->dq_id, type);
747         mutex_lock(&dquot->dq_lock);
748         /*
749          * We need an exclusive lock, because we're going to update use count
750          * and instantiate possibly new dquot structure
751          */
752         status = ocfs2_lock_global_qf(info, 1);
753         if (status < 0)
754                 goto out;
755         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
756                 status = ocfs2_qinfo_lock(info, 0);
757                 if (status < 0)
758                         goto out_dq;
759                 status = qtree_read_dquot(&info->dqi_gi, dquot);
760                 ocfs2_qinfo_unlock(info, 0);
761                 if (status < 0)
762                         goto out_dq;
763         }
764         set_bit(DQ_READ_B, &dquot->dq_flags);
765
766         OCFS2_DQUOT(dquot)->dq_use_count++;
767         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
768         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
769         if (!dquot->dq_off) {   /* No real quota entry? */
770                 ex = 1;
771                 /*
772                  * Add blocks to quota file before we start a transaction since
773                  * locking allocators ranks above a transaction start
774                  */
775                 WARN_ON(journal_current_handle());
776                 status = ocfs2_extend_no_holes(gqinode, NULL,
777                         gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
778                         gqinode->i_size);
779                 if (status < 0)
780                         goto out_dq;
781         }
782
783         handle = ocfs2_start_trans(osb,
784                                    ocfs2_calc_global_qinit_credits(sb, type));
785         if (IS_ERR(handle)) {
786                 status = PTR_ERR(handle);
787                 goto out_dq;
788         }
789         status = ocfs2_qinfo_lock(info, ex);
790         if (status < 0)
791                 goto out_trans;
792         status = qtree_write_dquot(&info->dqi_gi, dquot);
793         if (ex && info_dirty(sb_dqinfo(sb, type))) {
794                 err = __ocfs2_global_write_info(sb, type);
795                 if (!status)
796                         status = err;
797         }
798         ocfs2_qinfo_unlock(info, ex);
799 out_trans:
800         ocfs2_commit_trans(osb, handle);
801 out_dq:
802         ocfs2_unlock_global_qf(info, 1);
803         if (status < 0)
804                 goto out;
805
806         status = ocfs2_create_local_dquot(dquot);
807         if (status < 0)
808                 goto out;
809         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
810 out:
811         mutex_unlock(&dquot->dq_lock);
812         mlog_exit(status);
813         return status;
814 }
815
816 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
817 {
818         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
819                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
820                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
821                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
822                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
823                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
824         int sync = 0;
825         int status;
826         struct super_block *sb = dquot->dq_sb;
827         int type = dquot->dq_type;
828         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
829         handle_t *handle;
830         struct ocfs2_super *osb = OCFS2_SB(sb);
831
832         mlog(0, "id=%u, type=%d", dquot->dq_id, type);
833
834         /* In case user set some limits, sync dquot immediately to global
835          * quota file so that information propagates quicker */
836         spin_lock(&dq_data_lock);
837         if (dquot->dq_flags & mask)
838                 sync = 1;
839         spin_unlock(&dq_data_lock);
840         /* This is a slight hack but we can't afford getting global quota
841          * lock if we already have a transaction started. */
842         if (!sync || journal_current_handle()) {
843                 status = ocfs2_write_dquot(dquot);
844                 goto out;
845         }
846         status = ocfs2_lock_global_qf(oinfo, 1);
847         if (status < 0)
848                 goto out;
849         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
850         if (IS_ERR(handle)) {
851                 status = PTR_ERR(handle);
852                 mlog_errno(status);
853                 goto out_ilock;
854         }
855         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
856         status = ocfs2_sync_dquot(dquot);
857         if (status < 0) {
858                 mlog_errno(status);
859                 goto out_dlock;
860         }
861         /* Now write updated local dquot structure */
862         status = ocfs2_local_write_dquot(dquot);
863 out_dlock:
864         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
865         ocfs2_commit_trans(osb, handle);
866 out_ilock:
867         ocfs2_unlock_global_qf(oinfo, 1);
868 out:
869         mlog_exit(status);
870         return status;
871 }
872
873 /* This should happen only after set_dqinfo(). */
874 static int ocfs2_write_info(struct super_block *sb, int type)
875 {
876         handle_t *handle;
877         int status = 0;
878         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
879
880         status = ocfs2_lock_global_qf(oinfo, 1);
881         if (status < 0)
882                 goto out;
883         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
884         if (IS_ERR(handle)) {
885                 status = PTR_ERR(handle);
886                 mlog_errno(status);
887                 goto out_ilock;
888         }
889         status = dquot_commit_info(sb, type);
890         ocfs2_commit_trans(OCFS2_SB(sb), handle);
891 out_ilock:
892         ocfs2_unlock_global_qf(oinfo, 1);
893 out:
894         mlog_exit(status);
895         return status;
896 }
897
898 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
899 {
900         struct ocfs2_dquot *dquot =
901                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
902
903         if (!dquot)
904                 return NULL;
905         return &dquot->dq_dquot;
906 }
907
908 static void ocfs2_destroy_dquot(struct dquot *dquot)
909 {
910         kmem_cache_free(ocfs2_dquot_cachep, dquot);
911 }
912
913 const struct dquot_operations ocfs2_quota_operations = {
914         /* We never make dquot dirty so .write_dquot is never called */
915         .acquire_dquot  = ocfs2_acquire_dquot,
916         .release_dquot  = ocfs2_release_dquot,
917         .mark_dirty     = ocfs2_mark_dquot_dirty,
918         .write_info     = ocfs2_write_info,
919         .alloc_dquot    = ocfs2_alloc_dquot,
920         .destroy_dquot  = ocfs2_destroy_dquot,
921 };
922
923 int ocfs2_quota_setup(void)
924 {
925         ocfs2_quota_wq = create_workqueue("o2quot");
926         if (!ocfs2_quota_wq)
927                 return -ENOMEM;
928         return 0;
929 }
930
931 void ocfs2_quota_shutdown(void)
932 {
933         if (ocfs2_quota_wq) {
934                 flush_workqueue(ocfs2_quota_wq);
935                 destroy_workqueue(ocfs2_quota_wq);
936                 ocfs2_quota_wq = NULL;
937         }
938 }