fs: dcache scale dentry refcount
[linux-2.6.git] / 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         mlog_entry_void();
347
348         /* Read global header */
349         gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
350                         OCFS2_INVALID_SLOT);
351         if (!gqinode) {
352                 mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
353                         type);
354                 status = -EINVAL;
355                 goto out_err;
356         }
357         oinfo->dqi_gi.dqi_sb = sb;
358         oinfo->dqi_gi.dqi_type = type;
359         ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
360         oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
361         oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
362         oinfo->dqi_gqi_bh = NULL;
363         oinfo->dqi_gqi_count = 0;
364         oinfo->dqi_gqinode = gqinode;
365         status = ocfs2_lock_global_qf(oinfo, 0);
366         if (status < 0) {
367                 mlog_errno(status);
368                 goto out_err;
369         }
370
371         status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
372                                              &pcount, NULL);
373         if (status < 0)
374                 goto out_unlock;
375
376         status = ocfs2_qinfo_lock(oinfo, 0);
377         if (status < 0)
378                 goto out_unlock;
379         status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
380                                       sizeof(struct ocfs2_global_disk_dqinfo),
381                                       OCFS2_GLOBAL_INFO_OFF);
382         ocfs2_qinfo_unlock(oinfo, 0);
383         ocfs2_unlock_global_qf(oinfo, 0);
384         if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
385                 mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
386                      status);
387                 if (status >= 0)
388                         status = -EIO;
389                 mlog_errno(status);
390                 goto out_err;
391         }
392         info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
393         info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
394         oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
395         oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
396         oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
397         oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
398         oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
399         oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
400                                                 OCFS2_QBLK_RESERVED_SPACE;
401         oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
402         INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
403         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
404                            msecs_to_jiffies(oinfo->dqi_syncms));
405
406 out_err:
407         mlog_exit(status);
408         return status;
409 out_unlock:
410         ocfs2_unlock_global_qf(oinfo, 0);
411         mlog_errno(status);
412         goto out_err;
413 }
414
415 /* Write information to global quota file. Expects exlusive lock on quota
416  * file inode and quota info */
417 static int __ocfs2_global_write_info(struct super_block *sb, int type)
418 {
419         struct mem_dqinfo *info = sb_dqinfo(sb, type);
420         struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
421         struct ocfs2_global_disk_dqinfo dinfo;
422         ssize_t size;
423
424         spin_lock(&dq_data_lock);
425         info->dqi_flags &= ~DQF_INFO_DIRTY;
426         dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
427         dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
428         spin_unlock(&dq_data_lock);
429         dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
430         dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
431         dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
432         dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
433         size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
434                                      sizeof(struct ocfs2_global_disk_dqinfo),
435                                      OCFS2_GLOBAL_INFO_OFF);
436         if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
437                 mlog(ML_ERROR, "Cannot write global quota info structure\n");
438                 if (size >= 0)
439                         size = -EIO;
440                 return size;
441         }
442         return 0;
443 }
444
445 int ocfs2_global_write_info(struct super_block *sb, int type)
446 {
447         int err;
448         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
449
450         err = ocfs2_qinfo_lock(info, 1);
451         if (err < 0)
452                 return err;
453         err = __ocfs2_global_write_info(sb, type);
454         ocfs2_qinfo_unlock(info, 1);
455         return err;
456 }
457
458 static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
459 {
460         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
461
462         /*
463          * We may need to allocate tree blocks and a leaf block but not the
464          * root block
465          */
466         return oinfo->dqi_gi.dqi_qtree_depth;
467 }
468
469 static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
470 {
471         /* We modify all the allocated blocks, tree root, info block and
472          * the inode */
473         return (ocfs2_global_qinit_alloc(sb, type) + 2) *
474                         OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
475 }
476
477 /* Sync local information about quota modifications with global quota file.
478  * Caller must have started the transaction and obtained exclusive lock for
479  * global quota file inode */
480 int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
481 {
482         int err, err2;
483         struct super_block *sb = dquot->dq_sb;
484         int type = dquot->dq_type;
485         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
486         struct ocfs2_global_disk_dqblk dqblk;
487         s64 spacechange, inodechange;
488         time_t olditime, oldbtime;
489
490         err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
491                                    sizeof(struct ocfs2_global_disk_dqblk),
492                                    dquot->dq_off);
493         if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
494                 if (err >= 0) {
495                         mlog(ML_ERROR, "Short read from global quota file "
496                                        "(%u read)\n", err);
497                         err = -EIO;
498                 }
499                 goto out;
500         }
501
502         /* Update space and inode usage. Get also other information from
503          * global quota file so that we don't overwrite any changes there.
504          * We are */
505         spin_lock(&dq_data_lock);
506         spacechange = dquot->dq_dqb.dqb_curspace -
507                                         OCFS2_DQUOT(dquot)->dq_origspace;
508         inodechange = dquot->dq_dqb.dqb_curinodes -
509                                         OCFS2_DQUOT(dquot)->dq_originodes;
510         olditime = dquot->dq_dqb.dqb_itime;
511         oldbtime = dquot->dq_dqb.dqb_btime;
512         ocfs2_global_disk2memdqb(dquot, &dqblk);
513         mlog(0, "Syncing global dquot %u space %lld+%lld, inodes %lld+%lld\n",
514              dquot->dq_id, dquot->dq_dqb.dqb_curspace, (long long)spacechange,
515              dquot->dq_dqb.dqb_curinodes, (long long)inodechange);
516         if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
517                 dquot->dq_dqb.dqb_curspace += spacechange;
518         if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
519                 dquot->dq_dqb.dqb_curinodes += inodechange;
520         /* Set properly space grace time... */
521         if (dquot->dq_dqb.dqb_bsoftlimit &&
522             dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
523                 if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
524                     oldbtime > 0) {
525                         if (dquot->dq_dqb.dqb_btime > 0)
526                                 dquot->dq_dqb.dqb_btime =
527                                         min(dquot->dq_dqb.dqb_btime, oldbtime);
528                         else
529                                 dquot->dq_dqb.dqb_btime = oldbtime;
530                 }
531         } else {
532                 dquot->dq_dqb.dqb_btime = 0;
533                 clear_bit(DQ_BLKS_B, &dquot->dq_flags);
534         }
535         /* Set properly inode grace time... */
536         if (dquot->dq_dqb.dqb_isoftlimit &&
537             dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
538                 if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
539                     olditime > 0) {
540                         if (dquot->dq_dqb.dqb_itime > 0)
541                                 dquot->dq_dqb.dqb_itime =
542                                         min(dquot->dq_dqb.dqb_itime, olditime);
543                         else
544                                 dquot->dq_dqb.dqb_itime = olditime;
545                 }
546         } else {
547                 dquot->dq_dqb.dqb_itime = 0;
548                 clear_bit(DQ_INODES_B, &dquot->dq_flags);
549         }
550         /* All information is properly updated, clear the flags */
551         __clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
552         __clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
553         __clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
554         __clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
555         __clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
556         __clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
557         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
558         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
559         spin_unlock(&dq_data_lock);
560         err = ocfs2_qinfo_lock(info, freeing);
561         if (err < 0) {
562                 mlog(ML_ERROR, "Failed to lock quota info, loosing quota write"
563                                " (type=%d, id=%u)\n", dquot->dq_type,
564                                (unsigned)dquot->dq_id);
565                 goto out;
566         }
567         if (freeing)
568                 OCFS2_DQUOT(dquot)->dq_use_count--;
569         err = qtree_write_dquot(&info->dqi_gi, dquot);
570         if (err < 0)
571                 goto out_qlock;
572         if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
573                 err = qtree_release_dquot(&info->dqi_gi, dquot);
574                 if (info_dirty(sb_dqinfo(sb, type))) {
575                         err2 = __ocfs2_global_write_info(sb, type);
576                         if (!err)
577                                 err = err2;
578                 }
579         }
580 out_qlock:
581         ocfs2_qinfo_unlock(info, freeing);
582 out:
583         if (err < 0)
584                 mlog_errno(err);
585         return err;
586 }
587
588 /*
589  *  Functions for periodic syncing of dquots with global file
590  */
591 static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
592 {
593         handle_t *handle;
594         struct super_block *sb = dquot->dq_sb;
595         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
596         struct ocfs2_super *osb = OCFS2_SB(sb);
597         int status = 0;
598
599         mlog_entry("id=%u qtype=%u type=%lu device=%s\n", dquot->dq_id,
600                    dquot->dq_type, type, sb->s_id);
601         if (type != dquot->dq_type)
602                 goto out;
603         status = ocfs2_lock_global_qf(oinfo, 1);
604         if (status < 0)
605                 goto out;
606
607         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
608         if (IS_ERR(handle)) {
609                 status = PTR_ERR(handle);
610                 mlog_errno(status);
611                 goto out_ilock;
612         }
613         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
614         status = ocfs2_sync_dquot(dquot);
615         if (status < 0)
616                 mlog_errno(status);
617         /* We have to write local structure as well... */
618         status = ocfs2_local_write_dquot(dquot);
619         if (status < 0)
620                 mlog_errno(status);
621         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
622         ocfs2_commit_trans(osb, handle);
623 out_ilock:
624         ocfs2_unlock_global_qf(oinfo, 1);
625 out:
626         mlog_exit(status);
627         return status;
628 }
629
630 static void qsync_work_fn(struct work_struct *work)
631 {
632         struct ocfs2_mem_dqinfo *oinfo = container_of(work,
633                                                       struct ocfs2_mem_dqinfo,
634                                                       dqi_sync_work.work);
635         struct super_block *sb = oinfo->dqi_gqinode->i_sb;
636
637         dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
638         queue_delayed_work(ocfs2_quota_wq, &oinfo->dqi_sync_work,
639                            msecs_to_jiffies(oinfo->dqi_syncms));
640 }
641
642 /*
643  *  Wrappers for generic quota functions
644  */
645
646 static int ocfs2_write_dquot(struct dquot *dquot)
647 {
648         handle_t *handle;
649         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
650         int status = 0;
651
652         mlog_entry("id=%u, type=%d", dquot->dq_id, dquot->dq_type);
653
654         handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
655         if (IS_ERR(handle)) {
656                 status = PTR_ERR(handle);
657                 mlog_errno(status);
658                 goto out;
659         }
660         mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
661         status = ocfs2_local_write_dquot(dquot);
662         mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
663         ocfs2_commit_trans(osb, handle);
664 out:
665         mlog_exit(status);
666         return status;
667 }
668
669 static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
670 {
671         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
672         /*
673          * We modify tree, leaf block, global info, local chunk header,
674          * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
675          * accounts for inode update
676          */
677         return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
678                OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
679                OCFS2_QINFO_WRITE_CREDITS +
680                OCFS2_INODE_UPDATE_CREDITS;
681 }
682
683 static int ocfs2_release_dquot(struct dquot *dquot)
684 {
685         handle_t *handle;
686         struct ocfs2_mem_dqinfo *oinfo =
687                         sb_dqinfo(dquot->dq_sb, dquot->dq_type)->dqi_priv;
688         struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
689         int status = 0;
690
691         mlog_entry("id=%u, type=%d", dquot->dq_id, dquot->dq_type);
692
693         mutex_lock(&dquot->dq_lock);
694         /* Check whether we are not racing with some other dqget() */
695         if (atomic_read(&dquot->dq_count) > 1)
696                 goto out;
697         status = ocfs2_lock_global_qf(oinfo, 1);
698         if (status < 0)
699                 goto out;
700         handle = ocfs2_start_trans(osb,
701                 ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_type));
702         if (IS_ERR(handle)) {
703                 status = PTR_ERR(handle);
704                 mlog_errno(status);
705                 goto out_ilock;
706         }
707
708         status = ocfs2_global_release_dquot(dquot);
709         if (status < 0) {
710                 mlog_errno(status);
711                 goto out_trans;
712         }
713         status = ocfs2_local_release_dquot(handle, dquot);
714         /*
715          * If we fail here, we cannot do much as global structure is
716          * already released. So just complain...
717          */
718         if (status < 0)
719                 mlog_errno(status);
720         clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
721 out_trans:
722         ocfs2_commit_trans(osb, handle);
723 out_ilock:
724         ocfs2_unlock_global_qf(oinfo, 1);
725 out:
726         mutex_unlock(&dquot->dq_lock);
727         mlog_exit(status);
728         return status;
729 }
730
731 /*
732  * Read global dquot structure from disk or create it if it does
733  * not exist. Also update use count of the global structure and
734  * create structure in node-local quota file.
735  */
736 static int ocfs2_acquire_dquot(struct dquot *dquot)
737 {
738         int status = 0, err;
739         int ex = 0;
740         struct super_block *sb = dquot->dq_sb;
741         struct ocfs2_super *osb = OCFS2_SB(sb);
742         int type = dquot->dq_type;
743         struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
744         struct inode *gqinode = info->dqi_gqinode;
745         int need_alloc = ocfs2_global_qinit_alloc(sb, type);
746         handle_t *handle;
747
748         mlog_entry("id=%u, type=%d", dquot->dq_id, type);
749         mutex_lock(&dquot->dq_lock);
750         /*
751          * We need an exclusive lock, because we're going to update use count
752          * and instantiate possibly new dquot structure
753          */
754         status = ocfs2_lock_global_qf(info, 1);
755         if (status < 0)
756                 goto out;
757         if (!test_bit(DQ_READ_B, &dquot->dq_flags)) {
758                 status = ocfs2_qinfo_lock(info, 0);
759                 if (status < 0)
760                         goto out_dq;
761                 status = qtree_read_dquot(&info->dqi_gi, dquot);
762                 ocfs2_qinfo_unlock(info, 0);
763                 if (status < 0)
764                         goto out_dq;
765         }
766         set_bit(DQ_READ_B, &dquot->dq_flags);
767
768         OCFS2_DQUOT(dquot)->dq_use_count++;
769         OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
770         OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
771         if (!dquot->dq_off) {   /* No real quota entry? */
772                 ex = 1;
773                 /*
774                  * Add blocks to quota file before we start a transaction since
775                  * locking allocators ranks above a transaction start
776                  */
777                 WARN_ON(journal_current_handle());
778                 status = ocfs2_extend_no_holes(gqinode, NULL,
779                         gqinode->i_size + (need_alloc << sb->s_blocksize_bits),
780                         gqinode->i_size);
781                 if (status < 0)
782                         goto out_dq;
783         }
784
785         handle = ocfs2_start_trans(osb,
786                                    ocfs2_calc_global_qinit_credits(sb, type));
787         if (IS_ERR(handle)) {
788                 status = PTR_ERR(handle);
789                 goto out_dq;
790         }
791         status = ocfs2_qinfo_lock(info, ex);
792         if (status < 0)
793                 goto out_trans;
794         status = qtree_write_dquot(&info->dqi_gi, dquot);
795         if (ex && info_dirty(sb_dqinfo(sb, type))) {
796                 err = __ocfs2_global_write_info(sb, type);
797                 if (!status)
798                         status = err;
799         }
800         ocfs2_qinfo_unlock(info, ex);
801 out_trans:
802         ocfs2_commit_trans(osb, handle);
803 out_dq:
804         ocfs2_unlock_global_qf(info, 1);
805         if (status < 0)
806                 goto out;
807
808         status = ocfs2_create_local_dquot(dquot);
809         if (status < 0)
810                 goto out;
811         set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
812 out:
813         mutex_unlock(&dquot->dq_lock);
814         mlog_exit(status);
815         return status;
816 }
817
818 static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
819 {
820         unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
821                              (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
822                              (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
823                              (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
824                              (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
825                              (1 << (DQ_LASTSET_B + QIF_ITIME_B));
826         int sync = 0;
827         int status;
828         struct super_block *sb = dquot->dq_sb;
829         int type = dquot->dq_type;
830         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
831         handle_t *handle;
832         struct ocfs2_super *osb = OCFS2_SB(sb);
833
834         mlog_entry("id=%u, type=%d", dquot->dq_id, type);
835
836         /* In case user set some limits, sync dquot immediately to global
837          * quota file so that information propagates quicker */
838         spin_lock(&dq_data_lock);
839         if (dquot->dq_flags & mask)
840                 sync = 1;
841         spin_unlock(&dq_data_lock);
842         /* This is a slight hack but we can't afford getting global quota
843          * lock if we already have a transaction started. */
844         if (!sync || journal_current_handle()) {
845                 status = ocfs2_write_dquot(dquot);
846                 goto out;
847         }
848         status = ocfs2_lock_global_qf(oinfo, 1);
849         if (status < 0)
850                 goto out;
851         handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
852         if (IS_ERR(handle)) {
853                 status = PTR_ERR(handle);
854                 mlog_errno(status);
855                 goto out_ilock;
856         }
857         mutex_lock(&sb_dqopt(sb)->dqio_mutex);
858         status = ocfs2_sync_dquot(dquot);
859         if (status < 0) {
860                 mlog_errno(status);
861                 goto out_dlock;
862         }
863         /* Now write updated local dquot structure */
864         status = ocfs2_local_write_dquot(dquot);
865 out_dlock:
866         mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
867         ocfs2_commit_trans(osb, handle);
868 out_ilock:
869         ocfs2_unlock_global_qf(oinfo, 1);
870 out:
871         mlog_exit(status);
872         return status;
873 }
874
875 /* This should happen only after set_dqinfo(). */
876 static int ocfs2_write_info(struct super_block *sb, int type)
877 {
878         handle_t *handle;
879         int status = 0;
880         struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
881
882         mlog_entry_void();
883
884         status = ocfs2_lock_global_qf(oinfo, 1);
885         if (status < 0)
886                 goto out;
887         handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
888         if (IS_ERR(handle)) {
889                 status = PTR_ERR(handle);
890                 mlog_errno(status);
891                 goto out_ilock;
892         }
893         status = dquot_commit_info(sb, type);
894         ocfs2_commit_trans(OCFS2_SB(sb), handle);
895 out_ilock:
896         ocfs2_unlock_global_qf(oinfo, 1);
897 out:
898         mlog_exit(status);
899         return status;
900 }
901
902 static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
903 {
904         struct ocfs2_dquot *dquot =
905                                 kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
906
907         if (!dquot)
908                 return NULL;
909         return &dquot->dq_dquot;
910 }
911
912 static void ocfs2_destroy_dquot(struct dquot *dquot)
913 {
914         kmem_cache_free(ocfs2_dquot_cachep, dquot);
915 }
916
917 const struct dquot_operations ocfs2_quota_operations = {
918         /* We never make dquot dirty so .write_dquot is never called */
919         .acquire_dquot  = ocfs2_acquire_dquot,
920         .release_dquot  = ocfs2_release_dquot,
921         .mark_dirty     = ocfs2_mark_dquot_dirty,
922         .write_info     = ocfs2_write_info,
923         .alloc_dquot    = ocfs2_alloc_dquot,
924         .destroy_dquot  = ocfs2_destroy_dquot,
925 };
926
927 int ocfs2_quota_setup(void)
928 {
929         ocfs2_quota_wq = create_workqueue("o2quot");
930         if (!ocfs2_quota_wq)
931                 return -ENOMEM;
932         return 0;
933 }
934
935 void ocfs2_quota_shutdown(void)
936 {
937         if (ocfs2_quota_wq) {
938                 flush_workqueue(ocfs2_quota_wq);
939                 destroy_workqueue(ocfs2_quota_wq);
940                 ocfs2_quota_wq = NULL;
941         }
942 }