]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/gfs2/quota.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[GFS2] Finish off ioctl support
[net-next-2.6.git] / fs / gfs2 / quota.c
CommitLineData
b3b94faa
DT		 1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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 v.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/tty.h>
45#include <linux/sort.h>
18ec7d5c 46#include <linux/fs.h>
5c676f6d 47#include <linux/gfs2_ondisk.h>
b3b94faa
DT		 48#include <asm/semaphore.h>
49
50#include "gfs2.h"
5c676f6d
SW		 51#include "lm_interface.h"
52#include "incore.h"
b3b94faa
DT		 53#include "bmap.h"
54#include "glock.h"
55#include "glops.h"
b3b94faa 56#include "log.h"
5c676f6d 57#include "lvb.h"
b3b94faa
DT		 58#include "meta_io.h"
59#include "quota.h"
60#include "rgrp.h"
61#include "super.h"
62#include "trans.h"
18ec7d5c 63#include "inode.h"
f42faf4f 64#include "ops_file.h"
18ec7d5c 65#include "ops_address.h"
5c676f6d 66#include "util.h"
b3b94faa
DT		 67
68#define QUOTA_USER 1
69#define QUOTA_GROUP 0
70
71static uint64_t qd2offset(struct gfs2_quota_data *qd)
72{
73 uint64_t offset;
74
75 offset = 2 * (uint64_t)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
76 offset *= sizeof(struct gfs2_quota);
77
78 return offset;
79}
80
81static int qd_alloc(struct gfs2_sbd *sdp, int user, uint32_t id,
82 struct gfs2_quota_data **qdp)
83{
84 struct gfs2_quota_data *qd;
85 int error;
86
87 qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
88 if (!qd)
89 return -ENOMEM;
90
91 qd->qd_count = 1;
92 qd->qd_id = id;
93 if (user)
94 set_bit(QDF_USER, &qd->qd_flags);
95 qd->qd_slot = -1;
96
97 error = gfs2_glock_get(sdp, 2 * (uint64_t)id + !user,
98 &gfs2_quota_glops, CREATE, &qd->qd_gl);
99 if (error)
100 goto fail;
101
102 error = gfs2_lvb_hold(qd->qd_gl);
103 gfs2_glock_put(qd->qd_gl);
104 if (error)
105 goto fail;
106
107 *qdp = qd;
108
109 return 0;
110
111 fail:
112 kfree(qd);
113 return error;
114}
115
116static int qd_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
117 struct gfs2_quota_data **qdp)
118{
119 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
120 int error, found;
121
122 *qdp = NULL;
123
124 for (;;) {
125 found = 0;
126 spin_lock(&sdp->sd_quota_spin);
127 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
128 if (qd->qd_id == id &&
129 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
130 qd->qd_count++;
131 found = 1;
132 break;
133 }
134 }
135
136 if (!found)
137 qd = NULL;
138
139 if (!qd && new_qd) {
140 qd = new_qd;
141 list_add(&qd->qd_list, &sdp->sd_quota_list);
142 atomic_inc(&sdp->sd_quota_count);
143 new_qd = NULL;
144 }
145
146 spin_unlock(&sdp->sd_quota_spin);
147
148 if (qd || !create) {
149 if (new_qd) {
150 gfs2_lvb_unhold(new_qd->qd_gl);
151 kfree(new_qd);
152 }
153 *qdp = qd;
154 return 0;
155 }
156
157 error = qd_alloc(sdp, user, id, &new_qd);
158 if (error)
159 return error;
160 }
161}
162
163static void qd_hold(struct gfs2_quota_data *qd)
164{
165 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
166
167 spin_lock(&sdp->sd_quota_spin);
168 gfs2_assert(sdp, qd->qd_count);
169 qd->qd_count++;
170 spin_unlock(&sdp->sd_quota_spin);
171}
172
173static void qd_put(struct gfs2_quota_data *qd)
174{
175 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
176 spin_lock(&sdp->sd_quota_spin);
177 gfs2_assert(sdp, qd->qd_count);
178 if (!--qd->qd_count)
179 qd->qd_last_touched = jiffies;
180 spin_unlock(&sdp->sd_quota_spin);
181}
182
183static int slot_get(struct gfs2_quota_data *qd)
184{
185 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
186 unsigned int c, o = 0, b;
187 unsigned char byte = 0;
188
189 spin_lock(&sdp->sd_quota_spin);
190
191 if (qd->qd_slot_count++) {
192 spin_unlock(&sdp->sd_quota_spin);
193 return 0;
194 }
195
196 for (c = 0; c < sdp->sd_quota_chunks; c++)
197 for (o = 0; o < PAGE_SIZE; o++) {
198 byte = sdp->sd_quota_bitmap[c][o];
199 if (byte != 0xFF)
200 goto found;
201 }
202
203 goto fail;
204
205 found:
206 for (b = 0; b < 8; b++)
207 if (!(byte & (1 << b)))
208 break;
209 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
210
211 if (qd->qd_slot >= sdp->sd_quota_slots)
212 goto fail;
213
214 sdp->sd_quota_bitmap[c][o] |= 1 << b;
215
216 spin_unlock(&sdp->sd_quota_spin);
217
218 return 0;
219
220 fail:
221 qd->qd_slot_count--;
222 spin_unlock(&sdp->sd_quota_spin);
223 return -ENOSPC;
224}
225
226static void slot_hold(struct gfs2_quota_data *qd)
227{
228 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
229
230 spin_lock(&sdp->sd_quota_spin);
231 gfs2_assert(sdp, qd->qd_slot_count);
232 qd->qd_slot_count++;
233 spin_unlock(&sdp->sd_quota_spin);
234}
235
236static void slot_put(struct gfs2_quota_data *qd)
237{
238 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
239
240 spin_lock(&sdp->sd_quota_spin);
241 gfs2_assert(sdp, qd->qd_slot_count);
242 if (!--qd->qd_slot_count) {
243 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
244 qd->qd_slot = -1;
245 }
246 spin_unlock(&sdp->sd_quota_spin);
247}
248
249static int bh_get(struct gfs2_quota_data *qd)
250{
251 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
5c676f6d 252 struct gfs2_inode *ip = sdp->sd_qc_inode->u.generic_ip;
b3b94faa
DT		 253 unsigned int block, offset;
254 uint64_t dblock;
255 int new = 0;
256 struct buffer_head *bh;
257 int error;
258
f55ab26a 259 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 260
261 if (qd->qd_bh_count++) {
f55ab26a 262 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 263 return 0;
264 }
265
266 block = qd->qd_slot / sdp->sd_qc_per_block;
267 offset = qd->qd_slot % sdp->sd_qc_per_block;;
268
269 error = gfs2_block_map(ip, block, &new, &dblock, NULL);
270 if (error)
271 goto fail;
272 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT, &bh);
273 if (error)
274 goto fail;
275 error = -EIO;
276 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
277 goto fail_brelse;
278
279 qd->qd_bh = bh;
280 qd->qd_bh_qc = (struct gfs2_quota_change *)
281 (bh->b_data + sizeof(struct gfs2_meta_header) +
282 offset * sizeof(struct gfs2_quota_change));
283
f55ab26a 284 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 285
286 return 0;
287
288 fail_brelse:
289 brelse(bh);
290
291 fail:
292 qd->qd_bh_count--;
f55ab26a 293 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 294 return error;
295}
296
297static void bh_put(struct gfs2_quota_data *qd)
298{
299 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
300
f55ab26a 301 mutex_lock(&sdp->sd_quota_mutex);
b3b94faa
DT		 302 gfs2_assert(sdp, qd->qd_bh_count);
303 if (!--qd->qd_bh_count) {
304 brelse(qd->qd_bh);
305 qd->qd_bh = NULL;
306 qd->qd_bh_qc = NULL;
307 }
f55ab26a 308 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 309}
310
311static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
312{
313 struct gfs2_quota_data *qd = NULL;
314 int error;
315 int found = 0;
316
317 *qdp = NULL;
318
319 if (sdp->sd_vfs->s_flags & MS_RDONLY)
320 return 0;
321
322 spin_lock(&sdp->sd_quota_spin);
323
324 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
325 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
326 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
327 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
328 continue;
329
330 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
331
332 set_bit(QDF_LOCKED, &qd->qd_flags);
333 gfs2_assert_warn(sdp, qd->qd_count);
334 qd->qd_count++;
335 qd->qd_change_sync = qd->qd_change;
336 gfs2_assert_warn(sdp, qd->qd_slot_count);
337 qd->qd_slot_count++;
338 found = 1;
339
340 break;
341 }
342
343 if (!found)
344 qd = NULL;
345
346 spin_unlock(&sdp->sd_quota_spin);
347
348 if (qd) {
349 gfs2_assert_warn(sdp, qd->qd_change_sync);
350 error = bh_get(qd);
351 if (error) {
352 clear_bit(QDF_LOCKED, &qd->qd_flags);
353 slot_put(qd);
354 qd_put(qd);
355 return error;
356 }
357 }
358
359 *qdp = qd;
360
361 return 0;
362}
363
364static int qd_trylock(struct gfs2_quota_data *qd)
365{
366 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
367
368 if (sdp->sd_vfs->s_flags & MS_RDONLY)
369 return 0;
370
371 spin_lock(&sdp->sd_quota_spin);
372
373 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
374 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
375 spin_unlock(&sdp->sd_quota_spin);
376 return 0;
377 }
378
379 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
380
381 set_bit(QDF_LOCKED, &qd->qd_flags);
382 gfs2_assert_warn(sdp, qd->qd_count);
383 qd->qd_count++;
384 qd->qd_change_sync = qd->qd_change;
385 gfs2_assert_warn(sdp, qd->qd_slot_count);
386 qd->qd_slot_count++;
387
388 spin_unlock(&sdp->sd_quota_spin);
389
390 gfs2_assert_warn(sdp, qd->qd_change_sync);
391 if (bh_get(qd)) {
392 clear_bit(QDF_LOCKED, &qd->qd_flags);
393 slot_put(qd);
394 qd_put(qd);
395 return 0;
396 }
397
398 return 1;
399}
400
401static void qd_unlock(struct gfs2_quota_data *qd)
402{
568f4c96
SW		 403 gfs2_assert_warn(qd->qd_gl->gl_sbd,
404 test_bit(QDF_LOCKED, &qd->qd_flags));
b3b94faa
DT		 405 clear_bit(QDF_LOCKED, &qd->qd_flags);
406 bh_put(qd);
407 slot_put(qd);
408 qd_put(qd);
409}
410
411static int qdsb_get(struct gfs2_sbd *sdp, int user, uint32_t id, int create,
412 struct gfs2_quota_data **qdp)
413{
414 int error;
415
416 error = qd_get(sdp, user, id, create, qdp);
417 if (error)
418 return error;
419
420 error = slot_get(*qdp);
421 if (error)
422 goto fail;
423
424 error = bh_get(*qdp);
425 if (error)
426 goto fail_slot;
427
428 return 0;
429
430 fail_slot:
431 slot_put(*qdp);
432
433 fail:
434 qd_put(*qdp);
435 return error;
436}
437
438static void qdsb_put(struct gfs2_quota_data *qd)
439{
440 bh_put(qd);
441 slot_put(qd);
442 qd_put(qd);
443}
444
445int gfs2_quota_hold(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
446{
447 struct gfs2_sbd *sdp = ip->i_sbd;
448 struct gfs2_alloc *al = &ip->i_alloc;
449 struct gfs2_quota_data **qd = al->al_qd;
450 int error;
451
452 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
453 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
454 return -EIO;
455
456 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
457 return 0;
458
459 error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
460 if (error)
461 goto out;
462 al->al_qd_num++;
463 qd++;
464
465 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
466 if (error)
467 goto out;
468 al->al_qd_num++;
469 qd++;
470
471 if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
472 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
473 if (error)
474 goto out;
475 al->al_qd_num++;
476 qd++;
477 }
478
479 if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
480 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
481 if (error)
482 goto out;
483 al->al_qd_num++;
484 qd++;
485 }
486
487 out:
488 if (error)
489 gfs2_quota_unhold(ip);
490
491 return error;
492}
493
494void gfs2_quota_unhold(struct gfs2_inode *ip)
495{
496 struct gfs2_sbd *sdp = ip->i_sbd;
497 struct gfs2_alloc *al = &ip->i_alloc;
498 unsigned int x;
499
500 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
501
502 for (x = 0; x < al->al_qd_num; x++) {
503 qdsb_put(al->al_qd[x]);
504 al->al_qd[x] = NULL;
505 }
506 al->al_qd_num = 0;
507}
508
509static int sort_qd(const void *a, const void *b)
510{
511 struct gfs2_quota_data *qd_a = *(struct gfs2_quota_data **)a;
512 struct gfs2_quota_data *qd_b = *(struct gfs2_quota_data **)b;
513 int ret = 0;
514
515 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
516 !test_bit(QDF_USER, &qd_b->qd_flags)) {
517 if (test_bit(QDF_USER, &qd_a->qd_flags))
518 ret = -1;
519 else
520 ret = 1;
521 } else {
522 if (qd_a->qd_id < qd_b->qd_id)
523 ret = -1;
524 else if (qd_a->qd_id > qd_b->qd_id)
525 ret = 1;
526 }
527
528 return ret;
529}
530
531static void do_qc(struct gfs2_quota_data *qd, int64_t change)
532{
533 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
5c676f6d 534 struct gfs2_inode *ip = sdp->sd_qc_inode->u.generic_ip;
b3b94faa
DT		 535 struct gfs2_quota_change *qc = qd->qd_bh_qc;
536 int64_t x;
537
f55ab26a 538 mutex_lock(&sdp->sd_quota_mutex);
d4e9c4c3 539 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
b3b94faa
DT		 540
541 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
542 qc->qc_change = 0;
543 qc->qc_flags = 0;
544 if (test_bit(QDF_USER, &qd->qd_flags))
545 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
546 qc->qc_id = cpu_to_be32(qd->qd_id);
547 }
548
549 x = qc->qc_change;
550 x = be64_to_cpu(x) + change;
551 qc->qc_change = cpu_to_be64(x);
552
553 spin_lock(&sdp->sd_quota_spin);
554 qd->qd_change = x;
555 spin_unlock(&sdp->sd_quota_spin);
556
557 if (!x) {
558 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
559 clear_bit(QDF_CHANGE, &qd->qd_flags);
560 qc->qc_flags = 0;
561 qc->qc_id = 0;
562 slot_put(qd);
563 qd_put(qd);
564 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
565 qd_hold(qd);
566 slot_hold(qd);
567 }
568
f55ab26a 569 mutex_unlock(&sdp->sd_quota_mutex);
b3b94faa
DT		 570}
571
18ec7d5c
SW		 572/**
573 * gfs2_adjust_quota
574 *
575 * This function was mostly borrowed from gfs2_block_truncate_page which was
576 * in turn mostly borrowed from ext3
577 */
578static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
579 int64_t change, struct gfs2_quota_data *qd)
580{
7359a19c 581 struct inode *inode = ip->i_vnode;
18ec7d5c
SW		 582 struct address_space *mapping = inode->i_mapping;
583 unsigned long index = loc >> PAGE_CACHE_SHIFT;
584 unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
585 unsigned blocksize, iblock, pos;
586 struct buffer_head *bh;
587 struct page *page;
588 void *kaddr;
589 __be64 *ptr;
590 u64 value;
591 int err = -EIO;
592
593 page = grab_cache_page(mapping, index);
594 if (!page)
595 return -ENOMEM;
596
597 blocksize = inode->i_sb->s_blocksize;
598 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
599
600 if (!page_has_buffers(page))
601 create_empty_buffers(page, blocksize, 0);
602
603 bh = page_buffers(page);
604 pos = blocksize;
605 while (offset >= pos) {
606 bh = bh->b_this_page;
607 iblock++;
608 pos += blocksize;
609 }
610
611 if (!buffer_mapped(bh)) {
612 gfs2_get_block(inode, iblock, bh, 1);
613 if (!buffer_mapped(bh))
614 goto unlock;
615 }
616
617 if (PageUptodate(page))
618 set_buffer_uptodate(bh);
619
620 if (!buffer_uptodate(bh)) {
621 ll_rw_block(READ, 1, &bh);
622 wait_on_buffer(bh);
623 if (!buffer_uptodate(bh))
624 goto unlock;
625 }
626
627 gfs2_trans_add_bh(ip->i_gl, bh, 0);
628
629 kaddr = kmap_atomic(page, KM_USER0);
630 ptr = (__be64 *)(kaddr + offset);
631 value = *ptr = cpu_to_be64(be64_to_cpu(*ptr) + change);
632 flush_dcache_page(page);
633 kunmap_atomic(kaddr, KM_USER0);
634 err = 0;
635 qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
636#if 0
637 qd->qd_qb.qb_limit = cpu_to_be64(q.qu_limit);
638 qd->qd_qb.qb_warn = cpu_to_be64(q.qu_warn);
639#endif
640 qd->qd_qb.qb_value = cpu_to_be64(value);
641unlock:
642 unlock_page(page);
643 page_cache_release(page);
644 return err;
645}
646
b3b94faa
DT		 647static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
648{
649 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
5c676f6d 650 struct gfs2_inode *ip = sdp->sd_quota_inode->u.generic_ip;
b3b94faa 651 unsigned int data_blocks, ind_blocks;
f42faf4f 652 struct file_ra_state ra_state;
b3b94faa
DT		 653 struct gfs2_holder *ghs, i_gh;
654 unsigned int qx, x;
655 struct gfs2_quota_data *qd;
f42faf4f 656 loff_t offset;
b3b94faa
DT		 657 unsigned int nalloc = 0;
658 struct gfs2_alloc *al = NULL;
659 int error;
660
661 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
662 &data_blocks, &ind_blocks);
663
664 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
665 if (!ghs)
666 return -ENOMEM;
667
668 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
669 for (qx = 0; qx < num_qd; qx++) {
670 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
671 LM_ST_EXCLUSIVE,
672 GL_NOCACHE, &ghs[qx]);
673 if (error)
674 goto out;
675 }
676
677 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
678 if (error)
679 goto out;
680
681 for (x = 0; x < num_qd; x++) {
682 int alloc_required;
683
684 offset = qd2offset(qda[x]);
685 error = gfs2_write_alloc_required(ip, offset,
686 sizeof(struct gfs2_quota),
687 &alloc_required);
688 if (error)
689 goto out_gunlock;
690 if (alloc_required)
691 nalloc++;
692 }
693
694 if (nalloc) {
695 al = gfs2_alloc_get(ip);
696
697 al->al_requested = nalloc * (data_blocks + ind_blocks);
698
699 error = gfs2_inplace_reserve(ip);
700 if (error)
701 goto out_alloc;
702
703 error = gfs2_trans_begin(sdp,
704 al->al_rgd->rd_ri.ri_length +
705 num_qd * data_blocks +
706 nalloc * ind_blocks +
707 RES_DINODE + num_qd +
708 RES_STATFS, 0);
709 if (error)
710 goto out_ipres;
711 } else {
712 error = gfs2_trans_begin(sdp,
713 num_qd * data_blocks +
714 RES_DINODE + num_qd, 0);
715 if (error)
716 goto out_gunlock;
717 }
718
f42faf4f 719 file_ra_state_init(&ra_state, ip->i_vnode->i_mapping);
b3b94faa 720 for (x = 0; x < num_qd; x++) {
b3b94faa
DT		 721 qd = qda[x];
722 offset = qd2offset(qd);
18ec7d5c 723 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
568f4c96
SW		 724 (struct gfs2_quota_data *)
725 qd->qd_gl->gl_lvb);
18ec7d5c 726 if (error)
b3b94faa 727 goto out_end_trans;
b3b94faa
DT		 728
729 do_qc(qd, -qd->qd_change_sync);
b3b94faa
DT		 730 }
731
732 error = 0;
733
734 out_end_trans:
735 gfs2_trans_end(sdp);
736
737 out_ipres:
738 if (nalloc)
739 gfs2_inplace_release(ip);
740
741 out_alloc:
742 if (nalloc)
743 gfs2_alloc_put(ip);
744
745 out_gunlock:
746 gfs2_glock_dq_uninit(&i_gh);
747
748 out:
749 while (qx--)
750 gfs2_glock_dq_uninit(&ghs[qx]);
751 kfree(ghs);
752 gfs2_log_flush_glock(ip->i_gl);
753
754 return error;
755}
756
757static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
758 struct gfs2_holder *q_gh)
759{
760 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
5c676f6d 761 struct gfs2_inode *ip = sdp->sd_quota_inode->u.generic_ip;
b3b94faa
DT		 762 struct gfs2_holder i_gh;
763 struct gfs2_quota q;
764 char buf[sizeof(struct gfs2_quota)];
f42faf4f 765 struct file_ra_state ra_state;
b3b94faa
DT		 766 int error;
767
f42faf4f 768 file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
b3b94faa
DT		 769 restart:
770 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
771 if (error)
772 return error;
773
774 gfs2_quota_lvb_in(&qd->qd_qb, qd->qd_gl->gl_lvb);
775
776 if (force_refresh || qd->qd_qb.qb_magic != GFS2_MAGIC) {
f42faf4f 777 loff_t pos;
b3b94faa
DT		 778 gfs2_glock_dq_uninit(q_gh);
779 error = gfs2_glock_nq_init(qd->qd_gl,
780 LM_ST_EXCLUSIVE, GL_NOCACHE,
781 q_gh);
782 if (error)
783 return error;
784
5c676f6d 785 error = gfs2_glock_nq_init(ip->i_gl,
b3b94faa
DT		 786 LM_ST_SHARED, 0,
787 &i_gh);
788 if (error)
789 goto fail;
790
791 memset(buf, 0, sizeof(struct gfs2_quota));
f42faf4f 792 pos = qd2offset(qd);
5c676f6d 793 error = gfs2_internal_read(ip,
f42faf4f
SW		 794 &ra_state, buf,
795 &pos,
b3b94faa
DT		 796 sizeof(struct gfs2_quota));
797 if (error < 0)
798 goto fail_gunlock;
799
800 gfs2_glock_dq_uninit(&i_gh);
801
802 gfs2_quota_in(&q, buf);
803
804 memset(&qd->qd_qb, 0, sizeof(struct gfs2_quota_lvb));
805 qd->qd_qb.qb_magic = GFS2_MAGIC;
806 qd->qd_qb.qb_limit = q.qu_limit;
807 qd->qd_qb.qb_warn = q.qu_warn;
808 qd->qd_qb.qb_value = q.qu_value;
809
810 gfs2_quota_lvb_out(&qd->qd_qb, qd->qd_gl->gl_lvb);
811
812 if (gfs2_glock_is_blocking(qd->qd_gl)) {
813 gfs2_glock_dq_uninit(q_gh);
814 force_refresh = 0;
815 goto restart;
816 }
817 }
818
819 return 0;
820
821 fail_gunlock:
822 gfs2_glock_dq_uninit(&i_gh);
823
824 fail:
825 gfs2_glock_dq_uninit(q_gh);
826
827 return error;
828}
829
830int gfs2_quota_lock(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
831{
832 struct gfs2_sbd *sdp = ip->i_sbd;
833 struct gfs2_alloc *al = &ip->i_alloc;
834 unsigned int x;
835 int error = 0;
836
837 gfs2_quota_hold(ip, uid, gid);
838
839 if (capable(CAP_SYS_RESOURCE) ||
840 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
841 return 0;
842
843 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
844 sort_qd, NULL);
845
846 for (x = 0; x < al->al_qd_num; x++) {
847 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
848 if (error)
849 break;
850 }
851
852 if (!error)
853 set_bit(GIF_QD_LOCKED, &ip->i_flags);
854 else {
855 while (x--)
856 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
857 gfs2_quota_unhold(ip);
858 }
859
860 return error;
861}
862
863static int need_sync(struct gfs2_quota_data *qd)
864{
865 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
866 struct gfs2_tune *gt = &sdp->sd_tune;
867 int64_t value;
868 unsigned int num, den;
869 int do_sync = 1;
870
871 if (!qd->qd_qb.qb_limit)
872 return 0;
873
874 spin_lock(&sdp->sd_quota_spin);
875 value = qd->qd_change;
876 spin_unlock(&sdp->sd_quota_spin);
877
878 spin_lock(&gt->gt_spin);
879 num = gt->gt_quota_scale_num;
880 den = gt->gt_quota_scale_den;
881 spin_unlock(&gt->gt_spin);
882
883 if (value < 0)
884 do_sync = 0;
885 else if (qd->qd_qb.qb_value >= (int64_t)qd->qd_qb.qb_limit)
886 do_sync = 0;
887 else {
888 value *= gfs2_jindex_size(sdp) * num;
889 do_div(value, den);
890 value += qd->qd_qb.qb_value;
891 if (value < (int64_t)qd->qd_qb.qb_limit)
892 do_sync = 0;
893 }
894
895 return do_sync;
896}
897
898void gfs2_quota_unlock(struct gfs2_inode *ip)
899{
900 struct gfs2_alloc *al = &ip->i_alloc;
901 struct gfs2_quota_data *qda[4];
902 unsigned int count = 0;
903 unsigned int x;
904
905 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
906 goto out;
907
908 for (x = 0; x < al->al_qd_num; x++) {
909 struct gfs2_quota_data *qd;
910 int sync;
911
912 qd = al->al_qd[x];
913 sync = need_sync(qd);
914
915 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
916
917 if (sync && qd_trylock(qd))
918 qda[count++] = qd;
919 }
920
921 if (count) {
922 do_sync(count, qda);
923 for (x = 0; x < count; x++)
924 qd_unlock(qda[x]);
925 }
926
927 out:
928 gfs2_quota_unhold(ip);
929}
930
931#define MAX_LINE 256
932
933static int print_message(struct gfs2_quota_data *qd, char *type)
934{
935 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
936 char *line;
937 int len;
938
939 line = kmalloc(MAX_LINE, GFP_KERNEL);
940 if (!line)
941 return -ENOMEM;
942
568f4c96
SW		 943 len = snprintf(line, MAX_LINE-1,
944 "GFS2: fsid=%s: quota %s for %s %u\r\n",
b3b94faa
DT		 945 sdp->sd_fsname, type,
946 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
947 qd->qd_id);
948 line[MAX_LINE-1] = 0;
949
950 if (current->signal) { /* Is this test still required? */
951 tty_write_message(current->signal->tty, line);
952 }
953
954 kfree(line);
955
956 return 0;
957}
958
959int gfs2_quota_check(struct gfs2_inode *ip, uint32_t uid, uint32_t gid)
960{
961 struct gfs2_sbd *sdp = ip->i_sbd;
962 struct gfs2_alloc *al = &ip->i_alloc;
963 struct gfs2_quota_data *qd;
964 int64_t value;
965 unsigned int x;
966 int error = 0;
967
968 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
969 return 0;
970
971 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
972 return 0;
973
974 for (x = 0; x < al->al_qd_num; x++) {
975 qd = al->al_qd[x];
976
977 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
978 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
979 continue;
980
981 value = qd->qd_qb.qb_value;
982 spin_lock(&sdp->sd_quota_spin);
983 value += qd->qd_change;
984 spin_unlock(&sdp->sd_quota_spin);
985
986 if (qd->qd_qb.qb_limit && (int64_t)qd->qd_qb.qb_limit < value) {
987 print_message(qd, "exceeded");
988 error = -EDQUOT;
989 break;
990 } else if (qd->qd_qb.qb_warn &&
991 (int64_t)qd->qd_qb.qb_warn < value &&
992 time_after_eq(jiffies, qd->qd_last_warn +
568f4c96
SW		 993 gfs2_tune_get(sdp,
994 gt_quota_warn_period) * HZ)) {
b3b94faa
DT		 995 error = print_message(qd, "warning");
996 qd->qd_last_warn = jiffies;
997 }
998 }
999
1000 return error;
1001}
1002
1003void gfs2_quota_change(struct gfs2_inode *ip, int64_t change,
1004 uint32_t uid, uint32_t gid)
1005{
1006 struct gfs2_alloc *al = &ip->i_alloc;
1007 struct gfs2_quota_data *qd;
1008 unsigned int x;
1009 unsigned int found = 0;
1010
1011 if (gfs2_assert_warn(ip->i_sbd, change))
1012 return;
1013 if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
1014 return;
1015
1016 for (x = 0; x < al->al_qd_num; x++) {
1017 qd = al->al_qd[x];
1018
1019 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1020 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1021 do_qc(qd, change);
1022 found++;
1023 }
1024 }
1025}
1026
1027int gfs2_quota_sync(struct gfs2_sbd *sdp)
1028{
1029 struct gfs2_quota_data **qda;
1030 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1031 unsigned int num_qd;
1032 unsigned int x;
1033 int error = 0;
1034
1035 sdp->sd_quota_sync_gen++;
1036
1037 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1038 if (!qda)
1039 return -ENOMEM;
1040
1041 do {
1042 num_qd = 0;
1043
1044 for (;;) {
1045 error = qd_fish(sdp, qda + num_qd);
1046 if (error || !qda[num_qd])
1047 break;
1048 if (++num_qd == max_qd)
1049 break;
1050 }
1051
1052 if (num_qd) {
1053 if (!error)
1054 error = do_sync(num_qd, qda);
1055 if (!error)
1056 for (x = 0; x < num_qd; x++)
1057 qda[x]->qd_sync_gen =
1058 sdp->sd_quota_sync_gen;
1059
1060 for (x = 0; x < num_qd; x++)
1061 qd_unlock(qda[x]);
1062 }
1063 } while (!error && num_qd == max_qd);
1064
1065 kfree(qda);
1066
1067 return error;
1068}
1069
1070int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, uint32_t id)
1071{
1072 struct gfs2_quota_data *qd;
1073 struct gfs2_holder q_gh;
1074 int error;
1075
1076 error = qd_get(sdp, user, id, CREATE, &qd);
1077 if (error)
1078 return error;
1079
1080 error = do_glock(qd, FORCE, &q_gh);
1081 if (!error)
1082 gfs2_glock_dq_uninit(&q_gh);
1083
1084 qd_put(qd);
1085
1086 return error;
1087}
1088
1089int gfs2_quota_read(struct gfs2_sbd *sdp, int user, uint32_t id,
1090 struct gfs2_quota *q)
1091{
1092 struct gfs2_quota_data *qd;
1093 struct gfs2_holder q_gh;
1094 int error;
1095
1096 if (((user) ? (id != current->fsuid) : (!in_group_p(id))) &&
1097 !capable(CAP_SYS_ADMIN))
1098 return -EACCES;
1099
1100 error = qd_get(sdp, user, id, CREATE, &qd);
1101 if (error)
1102 return error;
1103
1104 error = do_glock(qd, NO_FORCE, &q_gh);
1105 if (error)
1106 goto out;
1107
1108 memset(q, 0, sizeof(struct gfs2_quota));
1109 q->qu_limit = qd->qd_qb.qb_limit;
1110 q->qu_warn = qd->qd_qb.qb_warn;
1111 q->qu_value = qd->qd_qb.qb_value;
1112
1113 spin_lock(&sdp->sd_quota_spin);
1114 q->qu_value += qd->qd_change;
1115 spin_unlock(&sdp->sd_quota_spin);
1116
1117 gfs2_glock_dq_uninit(&q_gh);
1118
1119 out:
1120 qd_put(qd);
1121
1122 return error;
1123}
1124
1125int gfs2_quota_init(struct gfs2_sbd *sdp)
1126{
5c676f6d 1127 struct gfs2_inode *ip = sdp->sd_qc_inode->u.generic_ip;
b3b94faa
DT		 1128 unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1129 unsigned int x, slot = 0;
1130 unsigned int found = 0;
1131 uint64_t dblock;
1132 uint32_t extlen = 0;
1133 int error;
1134
1135 if (!ip->i_di.di_size ||
1136 ip->i_di.di_size > (64 << 20) ||
1137 ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1138 gfs2_consist_inode(ip);
1139 return -EIO;
1140 }
1141 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
5c676f6d 1142 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
b3b94faa
DT		 1143
1144 error = -ENOMEM;
1145
1146 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1147 sizeof(unsigned char *), GFP_KERNEL);
1148 if (!sdp->sd_quota_bitmap)
1149 return error;
1150
1151 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1152 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1153 if (!sdp->sd_quota_bitmap[x])
1154 goto fail;
1155 }
1156
1157 for (x = 0; x < blocks; x++) {
1158 struct buffer_head *bh;
1159 unsigned int y;
1160
1161 if (!extlen) {
1162 int new = 0;
1163 error = gfs2_block_map(ip, x, &new, &dblock, &extlen);
1164 if (error)
1165 goto fail;
1166 }
1167 gfs2_meta_ra(ip->i_gl, dblock, extlen);
1168 error = gfs2_meta_read(ip->i_gl, dblock, DIO_START | DIO_WAIT,
1169 &bh);
1170 if (error)
1171 goto fail;
1172 error = -EIO;
1173 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1174 brelse(bh);
1175 goto fail;
1176 }
1177
1178 for (y = 0;
1179 y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1180 y++, slot++) {
1181 struct gfs2_quota_change qc;
1182 struct gfs2_quota_data *qd;
1183
1184 gfs2_quota_change_in(&qc, bh->b_data +
1185 sizeof(struct gfs2_meta_header) +
1186 y * sizeof(struct gfs2_quota_change));
1187 if (!qc.qc_change)
1188 continue;
1189
1190 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1191 qc.qc_id, &qd);
1192 if (error) {
1193 brelse(bh);
1194 goto fail;
1195 }
1196
1197 set_bit(QDF_CHANGE, &qd->qd_flags);
1198 qd->qd_change = qc.qc_change;
1199 qd->qd_slot = slot;
1200 qd->qd_slot_count = 1;
1201 qd->qd_last_touched = jiffies;
1202
1203 spin_lock(&sdp->sd_quota_spin);
1204 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1205 list_add(&qd->qd_list, &sdp->sd_quota_list);
1206 atomic_inc(&sdp->sd_quota_count);
1207 spin_unlock(&sdp->sd_quota_spin);
1208
1209 found++;
1210 }
1211
1212 brelse(bh);
1213 dblock++;
1214 extlen--;
1215 }
1216
1217 if (found)
1218 fs_info(sdp, "found %u quota changes\n", found);
1219
1220 return 0;
1221
1222 fail:
1223 gfs2_quota_cleanup(sdp);
1224 return error;
1225}
1226
1227void gfs2_quota_scan(struct gfs2_sbd *sdp)
1228{
1229 struct gfs2_quota_data *qd, *safe;
1230 LIST_HEAD(dead);
1231
1232 spin_lock(&sdp->sd_quota_spin);
1233 list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1234 if (!qd->qd_count &&
1235 time_after_eq(jiffies, qd->qd_last_touched +
1236 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1237 list_move(&qd->qd_list, &dead);
1238 gfs2_assert_warn(sdp,
1239 atomic_read(&sdp->sd_quota_count) > 0);
1240 atomic_dec(&sdp->sd_quota_count);
1241 }
1242 }
1243 spin_unlock(&sdp->sd_quota_spin);
1244
1245 while (!list_empty(&dead)) {
1246 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1247 list_del(&qd->qd_list);
1248
1249 gfs2_assert_warn(sdp, !qd->qd_change);
1250 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1251 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1252
1253 gfs2_lvb_unhold(qd->qd_gl);
1254 kfree(qd);
1255 }
1256}
1257
1258void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1259{
1260 struct list_head *head = &sdp->sd_quota_list;
1261 struct gfs2_quota_data *qd;
1262 unsigned int x;
1263
1264 spin_lock(&sdp->sd_quota_spin);
1265 while (!list_empty(head)) {
1266 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1267
1268 if (qd->qd_count > 1 ||
1269 (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1270 list_move(&qd->qd_list, head);
1271 spin_unlock(&sdp->sd_quota_spin);
1272 schedule();
1273 spin_lock(&sdp->sd_quota_spin);
1274 continue;
1275 }
1276
1277 list_del(&qd->qd_list);
1278 atomic_dec(&sdp->sd_quota_count);
1279 spin_unlock(&sdp->sd_quota_spin);
1280
1281 if (!qd->qd_count) {
1282 gfs2_assert_warn(sdp, !qd->qd_change);
1283 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1284 } else
1285 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1286 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1287
1288 gfs2_lvb_unhold(qd->qd_gl);
1289 kfree(qd);
1290
1291 spin_lock(&sdp->sd_quota_spin);
1292 }
1293 spin_unlock(&sdp->sd_quota_spin);
1294
1295 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1296
1297 if (sdp->sd_quota_bitmap) {
1298 for (x = 0; x < sdp->sd_quota_chunks; x++)
1299 kfree(sdp->sd_quota_bitmap[x]);
1300 kfree(sdp->sd_quota_bitmap);
1301 }
1302}
1303
Clone of davem's net-next-2.6 tree