?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NET_SCHED]: Fix endless loops caused by inaccurate qlen counters (part 1)
[linux-2.6.22.y-op.git] / fs / gfs2 / quota.c
bloba3deae7416c92d9ed8d78beae04dbea92165a73e
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 static u64 qd2offset(struct gfs2_quota_data *qd)
70 {
71 u64 offset;
72
73 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
74 offset *= sizeof(struct gfs2_quota);
75
76 return offset;
77 }
78
79 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
80 struct gfs2_quota_data **qdp)
81 {
82 struct gfs2_quota_data *qd;
83 int error;
84
85 qd = kzalloc(sizeof(struct gfs2_quota_data), GFP_KERNEL);
86 if (!qd)
87 return -ENOMEM;
88
89 qd->qd_count = 1;
90 qd->qd_id = id;
91 if (user)
92 set_bit(QDF_USER, &qd->qd_flags);
93 qd->qd_slot = -1;
94
95 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
96 &gfs2_quota_glops, CREATE, &qd->qd_gl);
97 if (error)
98 goto fail;
99
100 error = gfs2_lvb_hold(qd->qd_gl);
101 gfs2_glock_put(qd->qd_gl);
102 if (error)
103 goto fail;
104
105 *qdp = qd;
106
107 return 0;
108
109 fail:
110 kfree(qd);
111 return error;
112 }
113
114 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
115 struct gfs2_quota_data **qdp)
116 {
117 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
118 int error, found;
119
120 *qdp = NULL;
121
122 for (;;) {
123 found = 0;
124 spin_lock(&sdp->sd_quota_spin);
125 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
126 if (qd->qd_id == id &&
127 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
128 qd->qd_count++;
129 found = 1;
130 break;
131 }
132 }
133
134 if (!found)
135 qd = NULL;
136
137 if (!qd && new_qd) {
138 qd = new_qd;
139 list_add(&qd->qd_list, &sdp->sd_quota_list);
140 atomic_inc(&sdp->sd_quota_count);
141 new_qd = NULL;
142 }
143
144 spin_unlock(&sdp->sd_quota_spin);
145
146 if (qd || !create) {
147 if (new_qd) {
148 gfs2_lvb_unhold(new_qd->qd_gl);
149 kfree(new_qd);
150 }
151 *qdp = qd;
152 return 0;
153 }
154
155 error = qd_alloc(sdp, user, id, &new_qd);
156 if (error)
157 return error;
158 }
159 }
160
161 static void qd_hold(struct gfs2_quota_data *qd)
162 {
163 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
164
165 spin_lock(&sdp->sd_quota_spin);
166 gfs2_assert(sdp, qd->qd_count);
167 qd->qd_count++;
168 spin_unlock(&sdp->sd_quota_spin);
169 }
170
171 static void qd_put(struct gfs2_quota_data *qd)
172 {
173 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
174 spin_lock(&sdp->sd_quota_spin);
175 gfs2_assert(sdp, qd->qd_count);
176 if (!--qd->qd_count)
177 qd->qd_last_touched = jiffies;
178 spin_unlock(&sdp->sd_quota_spin);
179 }
180
181 static int slot_get(struct gfs2_quota_data *qd)
182 {
183 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
184 unsigned int c, o = 0, b;
185 unsigned char byte = 0;
186
187 spin_lock(&sdp->sd_quota_spin);
188
189 if (qd->qd_slot_count++) {
190 spin_unlock(&sdp->sd_quota_spin);
191 return 0;
192 }
193
194 for (c = 0; c < sdp->sd_quota_chunks; c++)
195 for (o = 0; o < PAGE_SIZE; o++) {
196 byte = sdp->sd_quota_bitmap[c][o];
197 if (byte != 0xFF)
198 goto found;
199 }
200
201 goto fail;
202
203 found:
204 for (b = 0; b < 8; b++)
205 if (!(byte & (1 << b)))
206 break;
207 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
208
209 if (qd->qd_slot >= sdp->sd_quota_slots)
210 goto fail;
211
212 sdp->sd_quota_bitmap[c][o] |= 1 << b;
213
214 spin_unlock(&sdp->sd_quota_spin);
215
216 return 0;
217
218 fail:
219 qd->qd_slot_count--;
220 spin_unlock(&sdp->sd_quota_spin);
221 return -ENOSPC;
222 }
223
224 static void slot_hold(struct gfs2_quota_data *qd)
225 {
226 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
227
228 spin_lock(&sdp->sd_quota_spin);
229 gfs2_assert(sdp, qd->qd_slot_count);
230 qd->qd_slot_count++;
231 spin_unlock(&sdp->sd_quota_spin);
232 }
233
234 static void slot_put(struct gfs2_quota_data *qd)
235 {
236 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
237
238 spin_lock(&sdp->sd_quota_spin);
239 gfs2_assert(sdp, qd->qd_slot_count);
240 if (!--qd->qd_slot_count) {
241 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
242 qd->qd_slot = -1;
243 }
244 spin_unlock(&sdp->sd_quota_spin);
245 }
246
247 static int bh_get(struct gfs2_quota_data *qd)
248 {
249 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
250 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
251 unsigned int block, offset;
252 struct buffer_head *bh;
253 int error;
254 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
255
256 mutex_lock(&sdp->sd_quota_mutex);
257
258 if (qd->qd_bh_count++) {
259 mutex_unlock(&sdp->sd_quota_mutex);
260 return 0;
261 }
262
263 block = qd->qd_slot / sdp->sd_qc_per_block;
264 offset = qd->qd_slot % sdp->sd_qc_per_block;;
265
266 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
267 error = gfs2_block_map(&ip->i_inode, block, 0, &bh_map);
268 if (error)
269 goto fail;
270 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
271 if (error)
272 goto fail;
273 error = -EIO;
274 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
275 goto fail_brelse;
276
277 qd->qd_bh = bh;
278 qd->qd_bh_qc = (struct gfs2_quota_change *)
279 (bh->b_data + sizeof(struct gfs2_meta_header) +
280 offset * sizeof(struct gfs2_quota_change));
281
282 mutex_lock(&sdp->sd_quota_mutex);
283
284 return 0;
285
286 fail_brelse:
287 brelse(bh);
288 fail:
289 qd->qd_bh_count--;
290 mutex_unlock(&sdp->sd_quota_mutex);
291 return error;
292 }
293
294 static void bh_put(struct gfs2_quota_data *qd)
295 {
296 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
297
298 mutex_lock(&sdp->sd_quota_mutex);
299 gfs2_assert(sdp, qd->qd_bh_count);
300 if (!--qd->qd_bh_count) {
301 brelse(qd->qd_bh);
302 qd->qd_bh = NULL;
303 qd->qd_bh_qc = NULL;
304 }
305 mutex_unlock(&sdp->sd_quota_mutex);
306 }
307
308 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
309 {
310 struct gfs2_quota_data *qd = NULL;
311 int error;
312 int found = 0;
313
314 *qdp = NULL;
315
316 if (sdp->sd_vfs->s_flags & MS_RDONLY)
317 return 0;
318
319 spin_lock(&sdp->sd_quota_spin);
320
321 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
322 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
323 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
324 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
325 continue;
326
327 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
328
329 set_bit(QDF_LOCKED, &qd->qd_flags);
330 gfs2_assert_warn(sdp, qd->qd_count);
331 qd->qd_count++;
332 qd->qd_change_sync = qd->qd_change;
333 gfs2_assert_warn(sdp, qd->qd_slot_count);
334 qd->qd_slot_count++;
335 found = 1;
336
337 break;
338 }
339
340 if (!found)
341 qd = NULL;
342
343 spin_unlock(&sdp->sd_quota_spin);
344
345 if (qd) {
346 gfs2_assert_warn(sdp, qd->qd_change_sync);
347 error = bh_get(qd);
348 if (error) {
349 clear_bit(QDF_LOCKED, &qd->qd_flags);
350 slot_put(qd);
351 qd_put(qd);
352 return error;
353 }
354 }
355
356 *qdp = qd;
357
358 return 0;
359 }
360
361 static int qd_trylock(struct gfs2_quota_data *qd)
362 {
363 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
364
365 if (sdp->sd_vfs->s_flags & MS_RDONLY)
366 return 0;
367
368 spin_lock(&sdp->sd_quota_spin);
369
370 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
371 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
372 spin_unlock(&sdp->sd_quota_spin);
373 return 0;
374 }
375
376 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
377
378 set_bit(QDF_LOCKED, &qd->qd_flags);
379 gfs2_assert_warn(sdp, qd->qd_count);
380 qd->qd_count++;
381 qd->qd_change_sync = qd->qd_change;
382 gfs2_assert_warn(sdp, qd->qd_slot_count);
383 qd->qd_slot_count++;
384
385 spin_unlock(&sdp->sd_quota_spin);
386
387 gfs2_assert_warn(sdp, qd->qd_change_sync);
388 if (bh_get(qd)) {
389 clear_bit(QDF_LOCKED, &qd->qd_flags);
390 slot_put(qd);
391 qd_put(qd);
392 return 0;
393 }
394
395 return 1;
396 }
397
398 static void qd_unlock(struct gfs2_quota_data *qd)
399 {
400 gfs2_assert_warn(qd->qd_gl->gl_sbd,
401 test_bit(QDF_LOCKED, &qd->qd_flags));
402 clear_bit(QDF_LOCKED, &qd->qd_flags);
403 bh_put(qd);
404 slot_put(qd);
405 qd_put(qd);
406 }
407
408 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id, int create,
409 struct gfs2_quota_data **qdp)
410 {
411 int error;
412
413 error = qd_get(sdp, user, id, create, qdp);
414 if (error)
415 return error;
416
417 error = slot_get(*qdp);
418 if (error)
419 goto fail;
420
421 error = bh_get(*qdp);
422 if (error)
423 goto fail_slot;
424
425 return 0;
426
427 fail_slot:
428 slot_put(*qdp);
429 fail:
430 qd_put(*qdp);
431 return error;
432 }
433
434 static void qdsb_put(struct gfs2_quota_data *qd)
435 {
436 bh_put(qd);
437 slot_put(qd);
438 qd_put(qd);
439 }
440
441 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
442 {
443 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
444 struct gfs2_alloc *al = &ip->i_alloc;
445 struct gfs2_quota_data **qd = al->al_qd;
446 int error;
447
448 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
449 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
450 return -EIO;
451
452 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
453 return 0;
454
455 error = qdsb_get(sdp, QUOTA_USER, ip->i_di.di_uid, CREATE, qd);
456 if (error)
457 goto out;
458 al->al_qd_num++;
459 qd++;
460
461 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_di.di_gid, CREATE, qd);
462 if (error)
463 goto out;
464 al->al_qd_num++;
465 qd++;
466
467 if (uid != NO_QUOTA_CHANGE && uid != ip->i_di.di_uid) {
468 error = qdsb_get(sdp, QUOTA_USER, uid, CREATE, qd);
469 if (error)
470 goto out;
471 al->al_qd_num++;
472 qd++;
473 }
474
475 if (gid != NO_QUOTA_CHANGE && gid != ip->i_di.di_gid) {
476 error = qdsb_get(sdp, QUOTA_GROUP, gid, CREATE, qd);
477 if (error)
478 goto out;
479 al->al_qd_num++;
480 qd++;
481 }
482
483 out:
484 if (error)
485 gfs2_quota_unhold(ip);
486 return error;
487 }
488
489 void gfs2_quota_unhold(struct gfs2_inode *ip)
490 {
491 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
492 struct gfs2_alloc *al = &ip->i_alloc;
493 unsigned int x;
494
495 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
496
497 for (x = 0; x < al->al_qd_num; x++) {
498 qdsb_put(al->al_qd[x]);
499 al->al_qd[x] = NULL;
500 }
501 al->al_qd_num = 0;
502 }
503
504 static int sort_qd(const void *a, const void *b)
505 {
506 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
507 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
508
509 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
510 !test_bit(QDF_USER, &qd_b->qd_flags)) {
511 if (test_bit(QDF_USER, &qd_a->qd_flags))
512 return -1;
513 else
514 return 1;
515 }
516 if (qd_a->qd_id < qd_b->qd_id)
517 return -1;
518 if (qd_a->qd_id > qd_b->qd_id)
519 return 1;
520
521 return 0;
522 }
523
524 static void do_qc(struct gfs2_quota_data *qd, s64 change)
525 {
526 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
527 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
528 struct gfs2_quota_change *qc = qd->qd_bh_qc;
529 s64 x;
530
531 mutex_lock(&sdp->sd_quota_mutex);
532 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
533
534 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
535 qc->qc_change = 0;
536 qc->qc_flags = 0;
537 if (test_bit(QDF_USER, &qd->qd_flags))
538 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
539 qc->qc_id = cpu_to_be32(qd->qd_id);
540 }
541
542 x = qc->qc_change;
543 x = be64_to_cpu(x) + change;
544 qc->qc_change = cpu_to_be64(x);
545
546 spin_lock(&sdp->sd_quota_spin);
547 qd->qd_change = x;
548 spin_unlock(&sdp->sd_quota_spin);
549
550 if (!x) {
551 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
552 clear_bit(QDF_CHANGE, &qd->qd_flags);
553 qc->qc_flags = 0;
554 qc->qc_id = 0;
555 slot_put(qd);
556 qd_put(qd);
557 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
558 qd_hold(qd);
559 slot_hold(qd);
560 }
561
562 mutex_unlock(&sdp->sd_quota_mutex);
563 }
564
565 /**
566 * gfs2_adjust_quota
567 *
568 * This function was mostly borrowed from gfs2_block_truncate_page which was
569 * in turn mostly borrowed from ext3
570 */
571 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
572 s64 change, struct gfs2_quota_data *qd)
573 {
574 struct inode *inode = &ip->i_inode;
575 struct address_space *mapping = inode->i_mapping;
576 unsigned long index = loc >> PAGE_CACHE_SHIFT;
577 unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
578 unsigned blocksize, iblock, pos;
579 struct buffer_head *bh;
580 struct page *page;
581 void *kaddr;
582 __be64 *ptr;
583 s64 value;
584 int err = -EIO;
585
586 page = grab_cache_page(mapping, index);
587 if (!page)
588 return -ENOMEM;
589
590 blocksize = inode->i_sb->s_blocksize;
591 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
592
593 if (!page_has_buffers(page))
594 create_empty_buffers(page, blocksize, 0);
595
596 bh = page_buffers(page);
597 pos = blocksize;
598 while (offset >= pos) {
599 bh = bh->b_this_page;
600 iblock++;
601 pos += blocksize;
602 }
603
604 if (!buffer_mapped(bh)) {
605 gfs2_get_block(inode, iblock, bh, 1);
606 if (!buffer_mapped(bh))
607 goto unlock;
608 }
609
610 if (PageUptodate(page))
611 set_buffer_uptodate(bh);
612
613 if (!buffer_uptodate(bh)) {
614 ll_rw_block(READ_META, 1, &bh);
615 wait_on_buffer(bh);
616 if (!buffer_uptodate(bh))
617 goto unlock;
618 }
619
620 gfs2_trans_add_bh(ip->i_gl, bh, 0);
621
622 kaddr = kmap_atomic(page, KM_USER0);
623 ptr = kaddr + offset;
624 value = (s64)be64_to_cpu(*ptr) + change;
625 *ptr = cpu_to_be64(value);
626 flush_dcache_page(page);
627 kunmap_atomic(kaddr, KM_USER0);
628 err = 0;
629 qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
630 qd->qd_qb.qb_value = cpu_to_be64(value);
631 unlock:
632 unlock_page(page);
633 page_cache_release(page);
634 return err;
635 }
636
637 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
638 {
639 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
640 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
641 unsigned int data_blocks, ind_blocks;
642 struct gfs2_holder *ghs, i_gh;
643 unsigned int qx, x;
644 struct gfs2_quota_data *qd;
645 loff_t offset;
646 unsigned int nalloc = 0;
647 struct gfs2_alloc *al = NULL;
648 int error;
649
650 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
651 &data_blocks, &ind_blocks);
652
653 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
654 if (!ghs)
655 return -ENOMEM;
656
657 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
658 for (qx = 0; qx < num_qd; qx++) {
659 error = gfs2_glock_nq_init(qda[qx]->qd_gl,
660 LM_ST_EXCLUSIVE,
661 GL_NOCACHE, &ghs[qx]);
662 if (error)
663 goto out;
664 }
665
666 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
667 if (error)
668 goto out;
669
670 for (x = 0; x < num_qd; x++) {
671 int alloc_required;
672
673 offset = qd2offset(qda[x]);
674 error = gfs2_write_alloc_required(ip, offset,
675 sizeof(struct gfs2_quota),
676 &alloc_required);
677 if (error)
678 goto out_gunlock;
679 if (alloc_required)
680 nalloc++;
681 }
682
683 if (nalloc) {
684 al = gfs2_alloc_get(ip);
685
686 al->al_requested = nalloc * (data_blocks + ind_blocks);
687
688 error = gfs2_inplace_reserve(ip);
689 if (error)
690 goto out_alloc;
691
692 error = gfs2_trans_begin(sdp,
693 al->al_rgd->rd_ri.ri_length +
694 num_qd * data_blocks +
695 nalloc * ind_blocks +
696 RES_DINODE + num_qd +
697 RES_STATFS, 0);
698 if (error)
699 goto out_ipres;
700 } else {
701 error = gfs2_trans_begin(sdp,
702 num_qd * data_blocks +
703 RES_DINODE + num_qd, 0);
704 if (error)
705 goto out_gunlock;
706 }
707
708 for (x = 0; x < num_qd; x++) {
709 qd = qda[x];
710 offset = qd2offset(qd);
711 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
712 (struct gfs2_quota_data *)
713 qd->qd_gl->gl_lvb);
714 if (error)
715 goto out_end_trans;
716
717 do_qc(qd, -qd->qd_change_sync);
718 }
719
720 error = 0;
721
722 out_end_trans:
723 gfs2_trans_end(sdp);
724 out_ipres:
725 if (nalloc)
726 gfs2_inplace_release(ip);
727 out_alloc:
728 if (nalloc)
729 gfs2_alloc_put(ip);
730 out_gunlock:
731 gfs2_glock_dq_uninit(&i_gh);
732 out:
733 while (qx--)
734 gfs2_glock_dq_uninit(&ghs[qx]);
735 kfree(ghs);
736 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
737 return error;
738 }
739
740 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
741 struct gfs2_holder *q_gh)
742 {
743 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
744 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
745 struct gfs2_holder i_gh;
746 struct gfs2_quota q;
747 char buf[sizeof(struct gfs2_quota)];
748 struct file_ra_state ra_state;
749 int error;
750 struct gfs2_quota_lvb *qlvb;
751
752 file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
753 restart:
754 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
755 if (error)
756 return error;
757
758 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
759
760 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
761 loff_t pos;
762 gfs2_glock_dq_uninit(q_gh);
763 error = gfs2_glock_nq_init(qd->qd_gl,
764 LM_ST_EXCLUSIVE, GL_NOCACHE,
765 q_gh);
766 if (error)
767 return error;
768
769 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
770 if (error)
771 goto fail;
772
773 memset(buf, 0, sizeof(struct gfs2_quota));
774 pos = qd2offset(qd);
775 error = gfs2_internal_read(ip, &ra_state, buf,
776 &pos, sizeof(struct gfs2_quota));
777 if (error < 0)
778 goto fail_gunlock;
779
780 gfs2_glock_dq_uninit(&i_gh);
781
782
783 gfs2_quota_in(&q, buf);
784 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
785 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
786 qlvb->__pad = 0;
787 qlvb->qb_limit = cpu_to_be64(q.qu_limit);
788 qlvb->qb_warn = cpu_to_be64(q.qu_warn);
789 qlvb->qb_value = cpu_to_be64(q.qu_value);
790 qd->qd_qb = *qlvb;
791
792 if (gfs2_glock_is_blocking(qd->qd_gl)) {
793 gfs2_glock_dq_uninit(q_gh);
794 force_refresh = 0;
795 goto restart;
796 }
797 }
798
799 return 0;
800
801 fail_gunlock:
802 gfs2_glock_dq_uninit(&i_gh);
803 fail:
804 gfs2_glock_dq_uninit(q_gh);
805 return error;
806 }
807
808 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
809 {
810 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
811 struct gfs2_alloc *al = &ip->i_alloc;
812 unsigned int x;
813 int error = 0;
814
815 gfs2_quota_hold(ip, uid, gid);
816
817 if (capable(CAP_SYS_RESOURCE) ||
818 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
819 return 0;
820
821 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
822 sort_qd, NULL);
823
824 for (x = 0; x < al->al_qd_num; x++) {
825 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
826 if (error)
827 break;
828 }
829
830 if (!error)
831 set_bit(GIF_QD_LOCKED, &ip->i_flags);
832 else {
833 while (x--)
834 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
835 gfs2_quota_unhold(ip);
836 }
837
838 return error;
839 }
840
841 static int need_sync(struct gfs2_quota_data *qd)
842 {
843 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
844 struct gfs2_tune *gt = &sdp->sd_tune;
845 s64 value;
846 unsigned int num, den;
847 int do_sync = 1;
848
849 if (!qd->qd_qb.qb_limit)
850 return 0;
851
852 spin_lock(&sdp->sd_quota_spin);
853 value = qd->qd_change;
854 spin_unlock(&sdp->sd_quota_spin);
855
856 spin_lock(&gt->gt_spin);
857 num = gt->gt_quota_scale_num;
858 den = gt->gt_quota_scale_den;
859 spin_unlock(&gt->gt_spin);
860
861 if (value < 0)
862 do_sync = 0;
863 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
864 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
865 do_sync = 0;
866 else {
867 value *= gfs2_jindex_size(sdp) * num;
868 do_div(value, den);
869 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
870 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
871 do_sync = 0;
872 }
873
874 return do_sync;
875 }
876
877 void gfs2_quota_unlock(struct gfs2_inode *ip)
878 {
879 struct gfs2_alloc *al = &ip->i_alloc;
880 struct gfs2_quota_data *qda[4];
881 unsigned int count = 0;
882 unsigned int x;
883
884 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
885 goto out;
886
887 for (x = 0; x < al->al_qd_num; x++) {
888 struct gfs2_quota_data *qd;
889 int sync;
890
891 qd = al->al_qd[x];
892 sync = need_sync(qd);
893
894 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
895
896 if (sync && qd_trylock(qd))
897 qda[count++] = qd;
898 }
899
900 if (count) {
901 do_sync(count, qda);
902 for (x = 0; x < count; x++)
903 qd_unlock(qda[x]);
904 }
905
906 out:
907 gfs2_quota_unhold(ip);
908 }
909
910 #define MAX_LINE 256
911
912 static int print_message(struct gfs2_quota_data *qd, char *type)
913 {
914 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
915
916 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
917 sdp->sd_fsname, type,
918 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
919 qd->qd_id);
920
921 return 0;
922 }
923
924 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
925 {
926 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
927 struct gfs2_alloc *al = &ip->i_alloc;
928 struct gfs2_quota_data *qd;
929 s64 value;
930 unsigned int x;
931 int error = 0;
932
933 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
934 return 0;
935
936 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
937 return 0;
938
939 for (x = 0; x < al->al_qd_num; x++) {
940 qd = al->al_qd[x];
941
942 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
943 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
944 continue;
945
946 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
947 spin_lock(&sdp->sd_quota_spin);
948 value += qd->qd_change;
949 spin_unlock(&sdp->sd_quota_spin);
950
951 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
952 print_message(qd, "exceeded");
953 error = -EDQUOT;
954 break;
955 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
956 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
957 time_after_eq(jiffies, qd->qd_last_warn +
958 gfs2_tune_get(sdp,
959 gt_quota_warn_period) * HZ)) {
960 error = print_message(qd, "warning");
961 qd->qd_last_warn = jiffies;
962 }
963 }
964
965 return error;
966 }
967
968 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
969 u32 uid, u32 gid)
970 {
971 struct gfs2_alloc *al = &ip->i_alloc;
972 struct gfs2_quota_data *qd;
973 unsigned int x;
974 unsigned int found = 0;
975
976 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
977 return;
978 if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
979 return;
980
981 for (x = 0; x < al->al_qd_num; x++) {
982 qd = al->al_qd[x];
983
984 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
985 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
986 do_qc(qd, change);
987 found++;
988 }
989 }
990 }
991
992 int gfs2_quota_sync(struct gfs2_sbd *sdp)
993 {
994 struct gfs2_quota_data **qda;
995 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
996 unsigned int num_qd;
997 unsigned int x;
998 int error = 0;
999
1000 sdp->sd_quota_sync_gen++;
1001
1002 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1003 if (!qda)
1004 return -ENOMEM;
1005
1006 do {
1007 num_qd = 0;
1008
1009 for (;;) {
1010 error = qd_fish(sdp, qda + num_qd);
1011 if (error || !qda[num_qd])
1012 break;
1013 if (++num_qd == max_qd)
1014 break;
1015 }
1016
1017 if (num_qd) {
1018 if (!error)
1019 error = do_sync(num_qd, qda);
1020 if (!error)
1021 for (x = 0; x < num_qd; x++)
1022 qda[x]->qd_sync_gen =
1023 sdp->sd_quota_sync_gen;
1024
1025 for (x = 0; x < num_qd; x++)
1026 qd_unlock(qda[x]);
1027 }
1028 } while (!error && num_qd == max_qd);
1029
1030 kfree(qda);
1031
1032 return error;
1033 }
1034
1035 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1036 {
1037 struct gfs2_quota_data *qd;
1038 struct gfs2_holder q_gh;
1039 int error;
1040
1041 error = qd_get(sdp, user, id, CREATE, &qd);
1042 if (error)
1043 return error;
1044
1045 error = do_glock(qd, FORCE, &q_gh);
1046 if (!error)
1047 gfs2_glock_dq_uninit(&q_gh);
1048
1049 qd_put(qd);
1050
1051 return error;
1052 }
1053
1054 int gfs2_quota_init(struct gfs2_sbd *sdp)
1055 {
1056 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1057 unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1058 unsigned int x, slot = 0;
1059 unsigned int found = 0;
1060 u64 dblock;
1061 u32 extlen = 0;
1062 int error;
1063
1064 if (!ip->i_di.di_size || ip->i_di.di_size > (64 << 20) ||
1065 ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1066 gfs2_consist_inode(ip);
1067 return -EIO;
1068 }
1069 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1070 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1071
1072 error = -ENOMEM;
1073
1074 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1075 sizeof(unsigned char *), GFP_KERNEL);
1076 if (!sdp->sd_quota_bitmap)
1077 return error;
1078
1079 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1080 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1081 if (!sdp->sd_quota_bitmap[x])
1082 goto fail;
1083 }
1084
1085 for (x = 0; x < blocks; x++) {
1086 struct buffer_head *bh;
1087 unsigned int y;
1088
1089 if (!extlen) {
1090 int new = 0;
1091 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1092 if (error)
1093 goto fail;
1094 }
1095 error = -EIO;
1096 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1097 if (!bh)
1098 goto fail;
1099 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1100 brelse(bh);
1101 goto fail;
1102 }
1103
1104 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1105 y++, slot++) {
1106 struct gfs2_quota_change qc;
1107 struct gfs2_quota_data *qd;
1108
1109 gfs2_quota_change_in(&qc, bh->b_data +
1110 sizeof(struct gfs2_meta_header) +
1111 y * sizeof(struct gfs2_quota_change));
1112 if (!qc.qc_change)
1113 continue;
1114
1115 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1116 qc.qc_id, &qd);
1117 if (error) {
1118 brelse(bh);
1119 goto fail;
1120 }
1121
1122 set_bit(QDF_CHANGE, &qd->qd_flags);
1123 qd->qd_change = qc.qc_change;
1124 qd->qd_slot = slot;
1125 qd->qd_slot_count = 1;
1126 qd->qd_last_touched = jiffies;
1127
1128 spin_lock(&sdp->sd_quota_spin);
1129 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1130 list_add(&qd->qd_list, &sdp->sd_quota_list);
1131 atomic_inc(&sdp->sd_quota_count);
1132 spin_unlock(&sdp->sd_quota_spin);
1133
1134 found++;
1135 }
1136
1137 brelse(bh);
1138 dblock++;
1139 extlen--;
1140 }
1141
1142 if (found)
1143 fs_info(sdp, "found %u quota changes\n", found);
1144
1145 return 0;
1146
1147 fail:
1148 gfs2_quota_cleanup(sdp);
1149 return error;
1150 }
1151
1152 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1153 {
1154 struct gfs2_quota_data *qd, *safe;
1155 LIST_HEAD(dead);
1156
1157 spin_lock(&sdp->sd_quota_spin);
1158 list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1159 if (!qd->qd_count &&
1160 time_after_eq(jiffies, qd->qd_last_touched +
1161 gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1162 list_move(&qd->qd_list, &dead);
1163 gfs2_assert_warn(sdp,
1164 atomic_read(&sdp->sd_quota_count) > 0);
1165 atomic_dec(&sdp->sd_quota_count);
1166 }
1167 }
1168 spin_unlock(&sdp->sd_quota_spin);
1169
1170 while (!list_empty(&dead)) {
1171 qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1172 list_del(&qd->qd_list);
1173
1174 gfs2_assert_warn(sdp, !qd->qd_change);
1175 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1176 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1177
1178 gfs2_lvb_unhold(qd->qd_gl);
1179 kfree(qd);
1180 }
1181 }
1182
1183 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1184 {
1185 struct list_head *head = &sdp->sd_quota_list;
1186 struct gfs2_quota_data *qd;
1187 unsigned int x;
1188
1189 spin_lock(&sdp->sd_quota_spin);
1190 while (!list_empty(head)) {
1191 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1192
1193 if (qd->qd_count > 1 ||
1194 (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1195 list_move(&qd->qd_list, head);
1196 spin_unlock(&sdp->sd_quota_spin);
1197 schedule();
1198 spin_lock(&sdp->sd_quota_spin);
1199 continue;
1200 }
1201
1202 list_del(&qd->qd_list);
1203 atomic_dec(&sdp->sd_quota_count);
1204 spin_unlock(&sdp->sd_quota_spin);
1205
1206 if (!qd->qd_count) {
1207 gfs2_assert_warn(sdp, !qd->qd_change);
1208 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1209 } else
1210 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1211 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1212
1213 gfs2_lvb_unhold(qd->qd_gl);
1214 kfree(qd);
1215
1216 spin_lock(&sdp->sd_quota_spin);
1217 }
1218 spin_unlock(&sdp->sd_quota_spin);
1219
1220 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1221
1222 if (sdp->sd_quota_bitmap) {
1223 for (x = 0; x < sdp->sd_quota_chunks; x++)
1224 kfree(sdp->sd_quota_bitmap[x]);
1225 kfree(sdp->sd_quota_bitmap);
1226 }
1227 }
1228
linux v2.6.22.y-op