[GFS2] Fix quota do_list operation hang
[linux-2.6.git] / fs / gfs2 / quota.c
1 /*
2  * Copyright (C) Sistina Software, Inc.  1997-2003 All rights reserved.
3  * Copyright (C) 2004-2006 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 /*
11  * Quota change tags are associated with each transaction that allocates or
12  * deallocates space.  Those changes are accumulated locally to each node (in a
13  * per-node file) and then are periodically synced to the quota file.  This
14  * avoids the bottleneck of constantly touching the quota file, but introduces
15  * fuzziness in the current usage value of IDs that are being used on different
16  * nodes in the cluster simultaneously.  So, it is possible for a user on
17  * multiple nodes to overrun their quota, but that overrun is controlable.
18  * Since quota tags are part of transactions, there is no need to a quota check
19  * program to be run on node crashes or anything like that.
20  *
21  * There are couple of knobs that let the administrator manage the quota
22  * fuzziness.  "quota_quantum" sets the maximum time a quota change can be
23  * sitting on one node before being synced to the quota file.  (The default is
24  * 60 seconds.)  Another knob, "quota_scale" controls how quickly the frequency
25  * of quota file syncs increases as the user moves closer to their limit.  The
26  * more frequent the syncs, the more accurate the quota enforcement, but that
27  * means that there is more contention between the nodes for the quota file.
28  * The default value is one.  This sets the maximum theoretical quota overrun
29  * (with infinite node with infinite bandwidth) to twice the user's limit.  (In
30  * practice, the maximum overrun you see should be much less.)  A "quota_scale"
31  * number greater than one makes quota syncs more frequent and reduces the
32  * maximum overrun.  Numbers less than one (but greater than zero) make quota
33  * syncs less frequent.
34  *
35  * GFS quotas also use per-ID Lock Value Blocks (LVBs) to cache the contents of
36  * the quota file, so it is not being constantly read.
37  */
38
39 #include <linux/sched.h>
40 #include <linux/slab.h>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
45 #include <linux/fs.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/lm_interface.h>
49
50 #include "gfs2.h"
51 #include "incore.h"
52 #include "bmap.h"
53 #include "glock.h"
54 #include "glops.h"
55 #include "log.h"
56 #include "meta_io.h"
57 #include "quota.h"
58 #include "rgrp.h"
59 #include "super.h"
60 #include "trans.h"
61 #include "inode.h"
62 #include "ops_file.h"
63 #include "ops_address.h"
64 #include "util.h"
65
66 #define QUOTA_USER 1
67 #define QUOTA_GROUP 0
68
69 struct gfs2_quota_host {
70         u64 qu_limit;
71         u64 qu_warn;
72         s64 qu_value;
73         u32 qu_ll_next;
74 };
75
76 struct gfs2_quota_change_host {
77         u64 qc_change;
78         u32 qc_flags; /* GFS2_QCF_... */
79         u32 qc_id;
80 };
81
82 static u64 qd2offset(struct gfs2_quota_data *qd)
83 {
84         u64 offset;
85
86         offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
87         offset *= sizeof(struct gfs2_quota);
88
89         return offset;
90 }
91
92 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
93                     struct gfs2_quota_data **qdp)
94 {
95         struct gfs2_quota_data *qd;
96         int error;
97
98         qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
99         if (!qd)
100                 return -ENOMEM;
101
102         qd->qd_count = 1;
103         qd->qd_id = id;
104         if (user)
105                 set_bit(QDF_USER, &qd->qd_flags);
106         qd->qd_slot = -1;
107
108         error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
109                               &gfs2_quota_glops, CREATE, &qd->qd_gl);
110         if (error)
111                 goto fail;
112
113         error = gfs2_lvb_hold(qd->qd_gl);
114         gfs2_glock_put(qd->qd_gl);
115         if (error)
116                 goto fail;
117
118         *qdp = qd;
119
120         return 0;
121
122 fail:
123         kfree(qd);
124         return error;
125 }
126
127 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
128                   struct gfs2_quota_data **qdp)
129 {
130         struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
131         int error, found;
132
133         *qdp = NULL;
134
135         for (;;) {
136                 found = 0;
137                 spin_lock(&sdp->sd_quota_spin);
138                 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
139                         if (qd->qd_id == id &&
140                             !test_bit(QDF_USER, &qd->qd_flags) == !user) {
141                                 qd->qd_count++;
142                                 found = 1;
143                                 break;
144                         }
145                 }
146
147                 if (!found)
148                         qd = NULL;
149
150                 if (!qd && new_qd) {
151                         qd = new_qd;
152                         list_add(&qd->qd_list, &sdp->sd_quota_list);
153                         atomic_inc(&sdp->sd_quota_count);
154                         new_qd = NULL;
155                 }
156
157                 spin_unlock(&sdp->sd_quota_spin);
158
159                 if (qd || !create) {
160                         if (new_qd) {
161                                 gfs2_lvb_unhold(new_qd->qd_gl);
162                                 kfree(new_qd);
163                         }
164                         *qdp = qd;
165                         return 0;
166                 }
167
168                 error = qd_alloc(sdp, user, id, &new_qd);
169                 if (error)
170                         return error;
171         }
172 }
173
174 static void qd_hold(struct gfs2_quota_data *qd)
175 {
176         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
177
178         spin_lock(&sdp->sd_quota_spin);
179         gfs2_assert(sdp, qd->qd_count);
180         qd->qd_count++;
181         spin_unlock(&sdp->sd_quota_spin);
182 }
183
184 static void qd_put(struct gfs2_quota_data *qd)
185 {
186         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
187         spin_lock(&sdp->sd_quota_spin);
188         gfs2_assert(sdp, qd->qd_count);
189         if (!--qd->qd_count)
190                 qd->qd_last_touched = jiffies;
191         spin_unlock(&sdp->sd_quota_spin);
192 }
193
194 static int slot_get(struct gfs2_quota_data *qd)
195 {
196         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
197         unsigned int c, o = 0, b;
198         unsigned char byte = 0;
199
200         spin_lock(&sdp->sd_quota_spin);
201
202         if (qd->qd_slot_count++) {
203                 spin_unlock(&sdp->sd_quota_spin);
204                 return 0;
205         }
206
207         for (c = 0; c < sdp->sd_quota_chunks; c++)
208                 for (o = 0; o < PAGE_SIZE; o++) {
209                         byte = sdp->sd_quota_bitmap[c][o];
210                         if (byte != 0xFF)
211                                 goto found;
212                 }
213
214         goto fail;
215
216 found:
217         for (b = 0; b < 8; b++)
218                 if (!(byte & (1 << b)))
219                         break;
220         qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
221
222         if (qd->qd_slot >= sdp->sd_quota_slots)
223                 goto fail;
224
225         sdp->sd_quota_bitmap[c][o] |= 1 << b;
226
227         spin_unlock(&sdp->sd_quota_spin);
228
229         return 0;
230
231 fail:
232         qd->qd_slot_count--;
233         spin_unlock(&sdp->sd_quota_spin);
234         return -ENOSPC;
235 }
236
237 static void slot_hold(struct gfs2_quota_data *qd)
238 {
239         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
240
241         spin_lock(&sdp->sd_quota_spin);
242         gfs2_assert(sdp, qd->qd_slot_count);
243         qd->qd_slot_count++;
244         spin_unlock(&sdp->sd_quota_spin);
245 }
246
247 static void slot_put(struct gfs2_quota_data *qd)
248 {
249         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
250
251         spin_lock(&sdp->sd_quota_spin);
252         gfs2_assert(sdp, qd->qd_slot_count);
253         if (!--qd->qd_slot_count) {
254                 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
255                 qd->qd_slot = -1;
256         }
257         spin_unlock(&sdp->sd_quota_spin);
258 }
259
260 static int bh_get(struct gfs2_quota_data *qd)
261 {
262         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
263         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
264         unsigned int block, offset;
265         struct buffer_head *bh;
266         int error;
267         struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
268
269         mutex_lock(&sdp->sd_quota_mutex);
270
271         if (qd->qd_bh_count++) {
272                 mutex_unlock(&sdp->sd_quota_mutex);
273                 return 0;
274         }
275
276         block = qd->qd_slot / sdp->sd_qc_per_block;
277         offset = qd->qd_slot % sdp->sd_qc_per_block;;
278
279         bh_map.b_size = 1 << ip->i_inode.i_blkbits;
280         error = gfs2_block_map(&ip->i_inode, block, 0, &bh_map);
281         if (error)
282                 goto fail;
283         error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
284         if (error)
285                 goto fail;
286         error = -EIO;
287         if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
288                 goto fail_brelse;
289
290         qd->qd_bh = bh;
291         qd->qd_bh_qc = (struct gfs2_quota_change *)
292                 (bh->b_data + sizeof(struct gfs2_meta_header) +
293                  offset * sizeof(struct gfs2_quota_change));
294
295         mutex_unlock(&sdp->sd_quota_mutex);
296
297         return 0;
298
299 fail_brelse:
300         brelse(bh);
301 fail:
302         qd->qd_bh_count--;
303         mutex_unlock(&sdp->sd_quota_mutex);
304         return error;
305 }
306
307 static void bh_put(struct gfs2_quota_data *qd)
308 {
309         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
310
311         mutex_lock(&sdp->sd_quota_mutex);
312         gfs2_assert(sdp, qd->qd_bh_count);
313         if (!--qd->qd_bh_count) {
314                 brelse(qd->qd_bh);
315                 qd->qd_bh = NULL;
316                 qd->qd_bh_qc = NULL;
317         }
318         mutex_unlock(&sdp->sd_quota_mutex);
319 }
320
321 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
322 {
323         struct gfs2_quota_data *qd = NULL;
324         int error;
325         int found = 0;
326
327         *qdp = NULL;
328
329         if (sdp->sd_vfs->s_flags & MS_RDONLY)
330                 return 0;
331
332         spin_lock(&sdp->sd_quota_spin);
333
334         list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
335                 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
336                     !test_bit(QDF_CHANGE, &qd->qd_flags) ||
337                     qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
338                         continue;
339
340                 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
341
342                 set_bit(QDF_LOCKED, &qd->qd_flags);
343                 gfs2_assert_warn(sdp, qd->qd_count);
344                 qd->qd_count++;
345                 qd->qd_change_sync = qd->qd_change;
346                 gfs2_assert_warn(sdp, qd->qd_slot_count);
347                 qd->qd_slot_count++;
348                 found = 1;
349
350                 break;
351         }
352
353         if (!found)
354                 qd = NULL;
355
356         spin_unlock(&sdp->sd_quota_spin);
357
358         if (qd) {
359                 gfs2_assert_warn(sdp, qd->qd_change_sync);
360                 error = bh_get(qd);
361                 if (error) {
362                         clear_bit(QDF_LOCKED, &qd->qd_flags);
363                         slot_put(qd);
364                         qd_put(qd);
365                         return error;
366                 }
367         }
368
369         *qdp = qd;
370
371         return 0;
372 }
373
374 static int qd_trylock(struct gfs2_quota_data *qd)
375 {
376         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
377
378         if (sdp->sd_vfs->s_flags & MS_RDONLY)
379                 return 0;
380
381         spin_lock(&sdp->sd_quota_spin);
382
383         if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
384             !test_bit(QDF_CHANGE, &qd->qd_flags)) {
385                 spin_unlock(&sdp->sd_quota_spin);
386                 return 0;
387         }
388
389         list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
390
391         set_bit(QDF_LOCKED, &qd->qd_flags);
392         gfs2_assert_warn(sdp, qd->qd_count);
393         qd->qd_count++;
394         qd->qd_change_sync = qd->qd_change;
395         gfs2_assert_warn(sdp, qd->qd_slot_count);
396         qd->qd_slot_count++;
397
398         spin_unlock(&sdp->sd_quota_spin);
399
400         gfs2_assert_warn(sdp, qd->qd_change_sync);
401         if (bh_get(qd)) {
402                 clear_bit(QDF_LOCKED, &qd->qd_flags);
403                 slot_put(qd);
404                 qd_put(qd);
405                 return 0;
406         }
407
408         return 1;
409 }
410
411 static void qd_unlock(struct gfs2_quota_data *qd)
412 {
413         gfs2_assert_warn(qd->qd_gl->gl_sbd,
414                          test_bit(QDF_LOCKED, &qd->qd_flags));
415         clear_bit(QDF_LOCKED, &qd->qd_flags);
416         bh_put(qd);
417         slot_put(qd);
418         qd_put(qd);
419 }
420
421 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
422                     struct gfs2_quota_data **qdp)
423 {
424         int error;
425
426         error = qd_get(sdp, user, id, create, qdp);
427         if (error)
428                 return error;
429
430         error = slot_get(*qdp);
431         if (error)
432                 goto fail;
433
434         error = bh_get(*qdp);
435         if (error)
436                 goto fail_slot;
437
438         return 0;
439
440 fail_slot:
441         slot_put(*qdp);
442 fail:
443         qd_put(*qdp);
444         return error;
445 }
446
447 static void qdsb_put(struct gfs2_quota_data *qd)
448 {
449         bh_put(qd);
450         slot_put(qd);
451         qd_put(qd);
452 }
453
454 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
455 {
456         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
457         struct gfs2_alloc *al = &ip->i_alloc;
458         struct gfs2_quota_data **qd = al->al_qd;
459         int error;
460
461         if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
462             gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
463                 return -EIO;
464
465         if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
466                 return 0;
467
468         error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, CREATE, qd);
469         if (error)
470                 goto out;
471         al->al_qd_num++;
472         qd++;
473
474         error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, CREATE, qd);
475         if (error)
476                 goto out;
477         al->al_qd_num++;
478         qd++;
479
480         if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
481                 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
482                 if (error)
483                         goto out;
484                 al->al_qd_num++;
485                 qd++;
486         }
487
488         if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
489                 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
490                 if (error)
491                         goto out;
492                 al->al_qd_num++;
493                 qd++;
494         }
495
496 out:
497         if (error)
498                 gfs2_quota_unhold(ip);
499         return error;
500 }
501
502 void gfs2_quota_unhold(struct gfs2_inode *ip)
503 {
504         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
505         struct gfs2_alloc *al = &ip->i_alloc;
506         unsigned int x;
507
508         gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
509
510         for (x = 0; x < al->al_qd_num; x++) {
511                 qdsb_put(al->al_qd[x]);
512                 al->al_qd[x] = NULL;
513         }
514         al->al_qd_num = 0;
515 }
516
517 static int sort_qd(const void *a, const void *b)
518 {
519         const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
520         const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
521
522         if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
523             !test_bit(QDF_USER, &qd_b->qd_flags)) {
524                 if (test_bit(QDF_USER, &qd_a->qd_flags))
525                         return -1;
526                 else
527                         return 1;
528         }
529         if (qd_a->qd_id < qd_b->qd_id)
530                 return -1;
531         if (qd_a->qd_id > qd_b->qd_id)
532                 return 1;
533
534         return 0;
535 }
536
537 static void do_qc(struct gfs2_quota_data *qd, s64 change)
538 {
539         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
540         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
541         struct gfs2_quota_change *qc = qd->qd_bh_qc;
542         s64 x;
543
544         mutex_lock(&sdp->sd_quota_mutex);
545         gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
546
547         if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
548                 qc->qc_change = 0;
549                 qc->qc_flags = 0;
550                 if (test_bit(QDF_USER, &qd->qd_flags))
551                         qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
552                 qc->qc_id = cpu_to_be32(qd->qd_id);
553         }
554
555         x = be64_to_cpu(qc->qc_change) + change;
556         qc->qc_change = cpu_to_be64(x);
557
558         spin_lock(&sdp->sd_quota_spin);
559         qd->qd_change = x;
560         spin_unlock(&sdp->sd_quota_spin);
561
562         if (!x) {
563                 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
564                 clear_bit(QDF_CHANGE, &qd->qd_flags);
565                 qc->qc_flags = 0;
566                 qc->qc_id = 0;
567                 slot_put(qd);
568                 qd_put(qd);
569         } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
570                 qd_hold(qd);
571                 slot_hold(qd);
572         }
573
574         mutex_unlock(&sdp->sd_quota_mutex);
575 }
576
577 static void gfs2_quota_in(struct gfs2_quota_host *qu, const void *buf)
578 {
579         const struct gfs2_quota *str = buf;
580
581         qu->qu_limit = be64_to_cpu(str->qu_limit);
582         qu->qu_warn = be64_to_cpu(str->qu_warn);
583         qu->qu_value = be64_to_cpu(str->qu_value);
584         qu->qu_ll_next = be32_to_cpu(str->qu_ll_next);
585 }
586
587 static void gfs2_quota_out(const struct gfs2_quota_host *qu, void *buf)
588 {
589         struct gfs2_quota *str = buf;
590
591         str->qu_limit = cpu_to_be64(qu->qu_limit);
592         str->qu_warn = cpu_to_be64(qu->qu_warn);
593         str->qu_value = cpu_to_be64(qu->qu_value);
594         str->qu_ll_next = cpu_to_be32(qu->qu_ll_next);
595         memset(&str->qu_reserved, 0, sizeof(str->qu_reserved));
596 }
597
598 /**
599  * gfs2_adjust_quota
600  *
601  * This function was mostly borrowed from gfs2_block_truncate_page which was
602  * in turn mostly borrowed from ext3
603  */
604 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
605                              s64 change, struct gfs2_quota_data *qd)
606 {
607         struct inode *inode = &ip->i_inode;
608         struct address_space *mapping = inode->i_mapping;
609         unsigned long index = loc >> PAGE_CACHE_SHIFT;
610         unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
611         unsigned blocksize, iblock, pos;
612         struct buffer_head *bh;
613         struct page *page;
614         void *kaddr;
615         char *ptr;
616         struct gfs2_quota_host qp;
617         s64 value;
618         int err = -EIO;
619
620         if (gfs2_is_stuffed(ip)) {
621                 struct gfs2_alloc *al = NULL;
622                 al = gfs2_alloc_get(ip);
623                 /* just request 1 blk */
624                 al->al_requested = 1;
625                 gfs2_inplace_reserve(ip);
626                 gfs2_unstuff_dinode(ip, NULL);
627                 gfs2_inplace_release(ip);
628                 gfs2_alloc_put(ip);
629         }
630         page = grab_cache_page(mapping, index);
631         if (!page)
632                 return -ENOMEM;
633
634         blocksize = inode->i_sb->s_blocksize;
635         iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
636
637         if (!page_has_buffers(page))
638                 create_empty_buffers(page, blocksize, 0);
639
640         bh = page_buffers(page);
641         pos = blocksize;
642         while (offset >= pos) {
643                 bh = bh->b_this_page;
644                 iblock++;
645                 pos += blocksize;
646         }
647
648         if (!buffer_mapped(bh)) {
649                 gfs2_get_block(inode, iblock, bh, 1);
650                 if (!buffer_mapped(bh))
651                         goto unlock;
652         }
653
654         if (PageUptodate(page))
655                 set_buffer_uptodate(bh);
656
657         if (!buffer_uptodate(bh)) {
658                 ll_rw_block(READ_META, 1, &bh);
659                 wait_on_buffer(bh);
660                 if (!buffer_uptodate(bh))
661                         goto unlock;
662         }
663
664         gfs2_trans_add_bh(ip->i_gl, bh, 0);
665
666         kaddr = kmap_atomic(page, KM_USER0);
667         ptr = kaddr + offset;
668         gfs2_quota_in(&qp, ptr);
669         qp.qu_value += change;
670         value = qp.qu_value;
671         gfs2_quota_out(&qp, ptr);
672         flush_dcache_page(page);
673         kunmap_atomic(kaddr, KM_USER0);
674         err = 0;
675         qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
676         qd->qd_qb.qb_value = cpu_to_be64(value);
677         ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_magic = cpu_to_be32(GFS2_MAGIC);
678         ((struct gfs2_quota_lvb*)(qd->qd_gl->gl_lvb))->qb_value = cpu_to_be64(value);
679 unlock:
680         unlock_page(page);
681         page_cache_release(page);
682         return err;
683 }
684
685 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
686 {
687         struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
688         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
689         unsigned int data_blocks, ind_blocks;
690         struct gfs2_holder *ghs, i_gh;
691         unsigned int qx, x;
692         struct gfs2_quota_data *qd;
693         loff_t offset;
694         unsigned int nalloc = 0;
695         struct gfs2_alloc *al = NULL;
696         int error;
697
698         gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
699                               &data_blocks, &ind_blocks);
700
701         ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
702         if (!ghs)
703                 return -ENOMEM;
704
705         sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
706         for (qx = 0; qx < num_qd; qx++) {
707                 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
708                                            LM_ST_EXCLUSIVE,
709                                            GL_NOCACHE, &ghs[qx]);
710                 if (error)
711                         goto out;
712         }
713
714         error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
715         if (error)
716                 goto out;
717
718         for (x = 0; x < num_qd; x++) {
719                 int alloc_required;
720
721                 offset = qd2offset(qda[x]);
722                 error = gfs2_write_alloc_required(ip, offset,
723                                                   sizeof(struct gfs2_quota),
724                                                   &alloc_required);
725                 if (error)
726                         goto out_gunlock;
727                 if (alloc_required)
728                         nalloc++;
729         }
730
731         if (nalloc) {
732                 al = gfs2_alloc_get(ip);
733
734                 al->al_requested = nalloc * (data_blocks + ind_blocks);
735
736                 error = gfs2_inplace_reserve(ip);
737                 if (error)
738                         goto out_alloc;
739
740                 error = gfs2_trans_begin(sdp,
741                                          al->al_rgd->rd_length +
742                                          num_qd * data_blocks +
743                                          nalloc * ind_blocks +
744                                          RES_DINODE + num_qd +
745                                          RES_STATFS, 0);
746                 if (error)
747                         goto out_ipres;
748         } else {
749                 error = gfs2_trans_begin(sdp,
750                                          num_qd * data_blocks +
751                                          RES_DINODE + num_qd, 0);
752                 if (error)
753                         goto out_gunlock;
754         }
755
756         for (x = 0; x < num_qd; x++) {
757                 qd = qda[x];
758                 offset = qd2offset(qd);
759                 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
760                                           (struct gfs2_quota_data *)
761                                           qd);
762                 if (error)
763                         goto out_end_trans;
764
765                 do_qc(qd, -qd->qd_change_sync);
766         }
767
768         error = 0;
769
770 out_end_trans:
771         gfs2_trans_end(sdp);
772 out_ipres:
773         if (nalloc)
774                 gfs2_inplace_release(ip);
775 out_alloc:
776         if (nalloc)
777                 gfs2_alloc_put(ip);
778 out_gunlock:
779         gfs2_glock_dq_uninit(&i_gh);
780 out:
781         while (qx--)
782                 gfs2_glock_dq_uninit(&ghs[qx]);
783         kfree(ghs);
784         gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
785         return error;
786 }
787
788 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
789                     struct gfs2_holder *q_gh)
790 {
791         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
792         struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
793         struct gfs2_holder i_gh;
794         struct gfs2_quota_host q;
795         char buf[sizeof(struct gfs2_quota)];
796         struct file_ra_state ra_state;
797         int error;
798         struct gfs2_quota_lvb *qlvb;
799
800         file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
801 restart:
802         error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
803         if (error)
804                 return error;
805
806         qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
807
808         if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
809                 loff_t pos;
810                 gfs2_glock_dq_uninit(q_gh);
811                 error = gfs2_glock_nq_init(qd->qd_gl,
812                                           LM_ST_EXCLUSIVE, GL_NOCACHE,
813                                           q_gh);
814                 if (error)
815                         return error;
816
817                 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
818                 if (error)
819                         goto fail;
820
821                 memset(buf, 0, sizeof(struct gfs2_quota));
822                 pos = qd2offset(qd);
823                 error = gfs2_internal_read(ip, &ra_state, buf,
824                                            &pos, sizeof(struct gfs2_quota));
825                 if (error < 0)
826                         goto fail_gunlock;
827
828                 gfs2_glock_dq_uninit(&i_gh);
829
830
831                 gfs2_quota_in(&q, buf);
832                 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
833                 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
834                 qlvb->__pad = 0;
835                 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
836                 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
837                 qlvb->qb_value = cpu_to_be64(q.qu_value);
838                 qd->qd_qb = *qlvb;
839
840                 if (gfs2_glock_is_blocking(qd->qd_gl)) {
841                         gfs2_glock_dq_uninit(q_gh);
842                         force_refresh = 0;
843                         goto restart;
844                 }
845         }
846
847         return 0;
848
849 fail_gunlock:
850         gfs2_glock_dq_uninit(&i_gh);
851 fail:
852         gfs2_glock_dq_uninit(q_gh);
853         return error;
854 }
855
856 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
857 {
858         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
859         struct gfs2_alloc *al = &ip->i_alloc;
860         unsigned int x;
861         int error = 0;
862
863         gfs2_quota_hold(ip, uid, gid);
864
865         if (capable(CAP_SYS_RESOURCE) ||
866             sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
867                 return 0;
868
869         sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
870              sort_qd, NULL);
871
872         for (x = 0; x < al->al_qd_num; x++) {
873                 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
874                 if (error)
875                         break;
876         }
877
878         if (!error)
879                 set_bit(GIF_QD_LOCKED, &ip->i_flags);
880         else {
881                 while (x--)
882                         gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
883                 gfs2_quota_unhold(ip);
884         }
885
886         return error;
887 }
888
889 static int need_sync(struct gfs2_quota_data *qd)
890 {
891         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
892         struct gfs2_tune *gt = &sdp->sd_tune;
893         s64 value;
894         unsigned int num, den;
895         int do_sync = 1;
896
897         if (!qd->qd_qb.qb_limit)
898                 return 0;
899
900         spin_lock(&sdp->sd_quota_spin);
901         value = qd->qd_change;
902         spin_unlock(&sdp->sd_quota_spin);
903
904         spin_lock(&gt->gt_spin);
905         num = gt->gt_quota_scale_num;
906         den = gt->gt_quota_scale_den;
907         spin_unlock(&gt->gt_spin);
908
909         if (value < 0)
910                 do_sync = 0;
911         else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
912                  (s64)be64_to_cpu(qd->qd_qb.qb_limit))
913                 do_sync = 0;
914         else {
915                 value *= gfs2_jindex_size(sdp) * num;
916                 do_div(value, den);
917                 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
918                 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
919                         do_sync = 0;
920         }
921
922         return do_sync;
923 }
924
925 void gfs2_quota_unlock(struct gfs2_inode *ip)
926 {
927         struct gfs2_alloc *al = &ip->i_alloc;
928         struct gfs2_quota_data *qda[4];
929         unsigned int count = 0;
930         unsigned int x;
931
932         if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
933                 goto out;
934
935         for (x = 0; x < al->al_qd_num; x++) {
936                 struct gfs2_quota_data *qd;
937                 int sync;
938
939                 qd = al->al_qd[x];
940                 sync = need_sync(qd);
941
942                 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
943
944                 if (sync && qd_trylock(qd))
945                         qda[count++] = qd;
946         }
947
948         if (count) {
949                 do_sync(count, qda);
950                 for (x = 0; x < count; x++)
951                         qd_unlock(qda[x]);
952         }
953
954 out:
955         gfs2_quota_unhold(ip);
956 }
957
958 #define MAX_LINE 256
959
960 static int print_message(struct gfs2_quota_data *qd, char *type)
961 {
962         struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
963
964         printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
965                sdp->sd_fsname, type,
966                (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
967                qd->qd_id);
968
969         return 0;
970 }
971
972 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
973 {
974         struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
975         struct gfs2_alloc *al = &ip->i_alloc;
976         struct gfs2_quota_data *qd;
977         s64 value;
978         unsigned int x;
979         int error = 0;
980
981         if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
982                 return 0;
983
984         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
985                 return 0;
986
987         for (x = 0; x < al->al_qd_num; x++) {
988                 qd = al->al_qd[x];
989
990                 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
991                       (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
992                         continue;
993
994                 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
995                 spin_lock(&sdp->sd_quota_spin);
996                 value += qd->qd_change;
997                 spin_unlock(&sdp->sd_quota_spin);
998
999                 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1000                         print_message(qd, "exceeded");
1001                         error = -EDQUOT;
1002                         break;
1003                 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1004                            (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1005                            time_after_eq(jiffies, qd->qd_last_warn +
1006                                          gfs2_tune_get(sdp,
1007                                                 gt_quota_warn_period) * HZ)) {
1008                         error = print_message(qd, "warning");
1009                         qd->qd_last_warn = jiffies;
1010                 }
1011         }
1012
1013         return error;
1014 }
1015
1016 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1017                        u32 uid, u32 gid)
1018 {
1019         struct gfs2_alloc *al = &ip->i_alloc;
1020         struct gfs2_quota_data *qd;
1021         unsigned int x;
1022         unsigned int found = 0;
1023
1024         if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1025                 return;
1026         if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
1027                 return;
1028
1029         for (x = 0; x < al->al_qd_num; x++) {
1030                 qd = al->al_qd[x];
1031
1032                 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1033                     (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1034                         do_qc(qd, change);
1035                         found++;
1036                 }
1037         }
1038 }
1039
1040 int gfs2_quota_sync(struct gfs2_sbd *sdp)
1041 {
1042         struct gfs2_quota_data **qda;
1043         unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1044         unsigned int num_qd;
1045         unsigned int x;
1046         int error = 0;
1047
1048         sdp->sd_quota_sync_gen++;
1049
1050         qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1051         if (!qda)
1052                 return -ENOMEM;
1053
1054         do {
1055                 num_qd = 0;
1056
1057                 for (;;) {
1058                         error = qd_fish(sdp, qda + num_qd);
1059                         if (error || !qda[num_qd])
1060                                 break;
1061                         if (++num_qd == max_qd)
1062                                 break;
1063                 }
1064
1065                 if (num_qd) {
1066                         if (!error)
1067                                 error = do_sync(num_qd, qda);
1068                         if (!error)
1069                                 for (x = 0; x < num_qd; x++)
1070                                         qda[x]->qd_sync_gen =
1071                                                 sdp->sd_quota_sync_gen;
1072
1073                         for (x = 0; x < num_qd; x++)
1074                                 qd_unlock(qda[x]);
1075                 }
1076         } while (!error && num_qd == max_qd);
1077
1078         kfree(qda);
1079
1080         return error;
1081 }
1082
1083 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1084 {
1085         struct gfs2_quota_data *qd;
1086         struct gfs2_holder q_gh;
1087         int error;
1088
1089         error = qd_get(sdp, user, id, CREATE, &qd);
1090         if (error)
1091                 return error;
1092
1093         error = do_glock(qd, FORCE, &q_gh);
1094         if (!error)
1095                 gfs2_glock_dq_uninit(&q_gh);
1096
1097         qd_put(qd);
1098
1099         return error;
1100 }
1101
1102 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1103 {
1104         const struct gfs2_quota_change *str = buf;
1105
1106         qc->qc_change = be64_to_cpu(str->qc_change);
1107         qc->qc_flags = be32_to_cpu(str->qc_flags);
1108         qc->qc_id = be32_to_cpu(str->qc_id);
1109 }
1110
1111 int gfs2_quota_init(struct gfs2_sbd *sdp)
1112 {
1113         struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1114         unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1115         unsigned int x, slot = 0;
1116         unsigned int found = 0;
1117         u64 dblock;
1118         u32 extlen = 0;
1119         int error;
1120
1121         if (!ip->i_di.di_size || ip->i_di.di_size > (64 << 20) ||
1122             ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1123                 gfs2_consist_inode(ip);
1124                 return -EIO;
1125         }
1126         sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1127         sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1128
1129         error = -ENOMEM;
1130
1131         sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1132                                        sizeof(unsigned char *), GFP_KERNEL);
1133         if (!sdp->sd_quota_bitmap)
1134                 return error;
1135
1136         for (x = 0; x < sdp->sd_quota_chunks; x++) {
1137                 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1138                 if (!sdp->sd_quota_bitmap[x])
1139                         goto fail;
1140         }
1141
1142         for (x = 0; x < blocks; x++) {
1143                 struct buffer_head *bh;
1144                 unsigned int y;
1145
1146                 if (!extlen) {
1147                         int new = 0;
1148                         error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1149                         if (error)
1150                                 goto fail;
1151                 }
1152                 error = -EIO;
1153                 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1154                 if (!bh)
1155                         goto fail;
1156                 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1157                         brelse(bh);
1158                         goto fail;
1159                 }
1160
1161                 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1162                      y++, slot++) {
1163                         struct gfs2_quota_change_host qc;
1164                         struct gfs2_quota_data *qd;
1165
1166                         gfs2_quota_change_in(&qc, bh->b_data +
1167                                           sizeof(struct gfs2_meta_header) +
1168                                           y * sizeof(struct gfs2_quota_change));
1169                         if (!qc.qc_change)
1170                                 continue;
1171
1172                         error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1173                                          qc.qc_id, &qd);
1174                         if (error) {
1175                                 brelse(bh);
1176                                 goto fail;
1177                         }
1178
1179                         set_bit(QDF_CHANGE, &qd->qd_flags);
1180                         qd->qd_change = qc.qc_change;
1181                         qd->qd_slot = slot;
1182                         qd->qd_slot_count = 1;
1183                         qd->qd_last_touched = jiffies;
1184
1185                         spin_lock(&sdp->sd_quota_spin);
1186                         gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1187                         list_add(&qd->qd_list, &sdp->sd_quota_list);
1188                         atomic_inc(&sdp->sd_quota_count);
1189                         spin_unlock(&sdp->sd_quota_spin);
1190
1191                         found++;
1192                 }
1193
1194                 brelse(bh);
1195                 dblock++;
1196                 extlen--;
1197         }
1198
1199         if (found)
1200                 fs_info(sdp, "found %u quota changes\n", found);
1201
1202         return 0;
1203
1204 fail:
1205         gfs2_quota_cleanup(sdp);
1206         return error;
1207 }
1208
1209 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1210 {
1211         struct gfs2_quota_data *qd, *safe;
1212         LIST_HEAD(dead);
1213
1214         spin_lock(&sdp->sd_quota_spin);
1215         list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1216                 if (!qd->qd_count &&
1217                     time_after_eq(jiffies, qd->qd_last_touched +
1218                                 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1219                         list_move(&qd->qd_list, &dead);
1220                         gfs2_assert_warn(sdp,
1221                                          atomic_read(&sdp->sd_quota_count) > 0);
1222                         atomic_dec(&sdp->sd_quota_count);
1223                 }
1224         }
1225         spin_unlock(&sdp->sd_quota_spin);
1226
1227         while (!list_empty(&dead)) {
1228                 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1229                 list_del(&qd->qd_list);
1230
1231                 gfs2_assert_warn(sdp, !qd->qd_change);
1232                 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1233                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1234
1235                 gfs2_lvb_unhold(qd->qd_gl);
1236                 kfree(qd);
1237         }
1238 }
1239
1240 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1241 {
1242         struct list_head *head = &sdp->sd_quota_list;
1243         struct gfs2_quota_data *qd;
1244         unsigned int x;
1245
1246         spin_lock(&sdp->sd_quota_spin);
1247         while (!list_empty(head)) {
1248                 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1249
1250                 if (qd->qd_count > 1 ||
1251                     (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1252                         list_move(&qd->qd_list, head);
1253                         spin_unlock(&sdp->sd_quota_spin);
1254                         schedule();
1255                         spin_lock(&sdp->sd_quota_spin);
1256                         continue;
1257                 }
1258
1259                 list_del(&qd->qd_list);
1260                 atomic_dec(&sdp->sd_quota_count);
1261                 spin_unlock(&sdp->sd_quota_spin);
1262
1263                 if (!qd->qd_count) {
1264                         gfs2_assert_warn(sdp, !qd->qd_change);
1265                         gfs2_assert_warn(sdp, !qd->qd_slot_count);
1266                 } else
1267                         gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1268                 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1269
1270                 gfs2_lvb_unhold(qd->qd_gl);
1271                 kfree(qd);
1272
1273                 spin_lock(&sdp->sd_quota_spin);
1274         }
1275         spin_unlock(&sdp->sd_quota_spin);
1276
1277         gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1278
1279         if (sdp->sd_quota_bitmap) {
1280                 for (x = 0; x < sdp->sd_quota_chunks; x++)
1281                         kfree(sdp->sd_quota_bitmap[x]);
1282                 kfree(sdp->sd_quota_bitmap);
1283         }
1284 }
1285