xref: /openbmc/linux/fs/gfs2/quota.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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_unlock(&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_inode.i_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_inode.i_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_inode.i_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_inode.i_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 = be64_to_cpu(qc->qc_change) + change;
543 	qc->qc_change = cpu_to_be64(x);
544 
545 	spin_lock(&sdp->sd_quota_spin);
546 	qd->qd_change = x;
547 	spin_unlock(&sdp->sd_quota_spin);
548 
549 	if (!x) {
550 		gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
551 		clear_bit(QDF_CHANGE, &qd->qd_flags);
552 		qc->qc_flags = 0;
553 		qc->qc_id = 0;
554 		slot_put(qd);
555 		qd_put(qd);
556 	} else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
557 		qd_hold(qd);
558 		slot_hold(qd);
559 	}
560 
561 	mutex_unlock(&sdp->sd_quota_mutex);
562 }
563 
564 /**
565  * gfs2_adjust_quota
566  *
567  * This function was mostly borrowed from gfs2_block_truncate_page which was
568  * in turn mostly borrowed from ext3
569  */
570 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
571 			     s64 change, struct gfs2_quota_data *qd)
572 {
573 	struct inode *inode = &ip->i_inode;
574 	struct address_space *mapping = inode->i_mapping;
575 	unsigned long index = loc >> PAGE_CACHE_SHIFT;
576 	unsigned offset = loc & (PAGE_CACHE_SHIFT - 1);
577 	unsigned blocksize, iblock, pos;
578 	struct buffer_head *bh;
579 	struct page *page;
580 	void *kaddr;
581 	__be64 *ptr;
582 	s64 value;
583 	int err = -EIO;
584 
585 	page = grab_cache_page(mapping, index);
586 	if (!page)
587 		return -ENOMEM;
588 
589 	blocksize = inode->i_sb->s_blocksize;
590 	iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
591 
592 	if (!page_has_buffers(page))
593 		create_empty_buffers(page, blocksize, 0);
594 
595 	bh = page_buffers(page);
596 	pos = blocksize;
597 	while (offset >= pos) {
598 		bh = bh->b_this_page;
599 		iblock++;
600 		pos += blocksize;
601 	}
602 
603 	if (!buffer_mapped(bh)) {
604 		gfs2_get_block(inode, iblock, bh, 1);
605 		if (!buffer_mapped(bh))
606 			goto unlock;
607 	}
608 
609 	if (PageUptodate(page))
610 		set_buffer_uptodate(bh);
611 
612 	if (!buffer_uptodate(bh)) {
613 		ll_rw_block(READ_META, 1, &bh);
614 		wait_on_buffer(bh);
615 		if (!buffer_uptodate(bh))
616 			goto unlock;
617 	}
618 
619 	gfs2_trans_add_bh(ip->i_gl, bh, 0);
620 
621 	kaddr = kmap_atomic(page, KM_USER0);
622 	ptr = kaddr + offset;
623 	value = (s64)be64_to_cpu(*ptr) + change;
624 	*ptr = cpu_to_be64(value);
625 	flush_dcache_page(page);
626 	kunmap_atomic(kaddr, KM_USER0);
627 	err = 0;
628 	qd->qd_qb.qb_magic = cpu_to_be32(GFS2_MAGIC);
629 	qd->qd_qb.qb_value = cpu_to_be64(value);
630 unlock:
631 	unlock_page(page);
632 	page_cache_release(page);
633 	return err;
634 }
635 
636 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
637 {
638 	struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
639 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
640 	unsigned int data_blocks, ind_blocks;
641 	struct gfs2_holder *ghs, i_gh;
642 	unsigned int qx, x;
643 	struct gfs2_quota_data *qd;
644 	loff_t offset;
645 	unsigned int nalloc = 0;
646 	struct gfs2_alloc *al = NULL;
647 	int error;
648 
649 	gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
650 			      &data_blocks, &ind_blocks);
651 
652 	ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_KERNEL);
653 	if (!ghs)
654 		return -ENOMEM;
655 
656 	sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
657 	for (qx = 0; qx < num_qd; qx++) {
658 		error = gfs2_glock_nq_init(qda[qx]->qd_gl,
659 					   LM_ST_EXCLUSIVE,
660 					   GL_NOCACHE, &ghs[qx]);
661 		if (error)
662 			goto out;
663 	}
664 
665 	error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
666 	if (error)
667 		goto out;
668 
669 	for (x = 0; x < num_qd; x++) {
670 		int alloc_required;
671 
672 		offset = qd2offset(qda[x]);
673 		error = gfs2_write_alloc_required(ip, offset,
674 						  sizeof(struct gfs2_quota),
675 						  &alloc_required);
676 		if (error)
677 			goto out_gunlock;
678 		if (alloc_required)
679 			nalloc++;
680 	}
681 
682 	if (nalloc) {
683 		al = gfs2_alloc_get(ip);
684 
685 		al->al_requested = nalloc * (data_blocks + ind_blocks);
686 
687 		error = gfs2_inplace_reserve(ip);
688 		if (error)
689 			goto out_alloc;
690 
691 		error = gfs2_trans_begin(sdp,
692 					 al->al_rgd->rd_ri.ri_length +
693 					 num_qd * data_blocks +
694 					 nalloc * ind_blocks +
695 					 RES_DINODE + num_qd +
696 					 RES_STATFS, 0);
697 		if (error)
698 			goto out_ipres;
699 	} else {
700 		error = gfs2_trans_begin(sdp,
701 					 num_qd * data_blocks +
702 					 RES_DINODE + num_qd, 0);
703 		if (error)
704 			goto out_gunlock;
705 	}
706 
707 	for (x = 0; x < num_qd; x++) {
708 		qd = qda[x];
709 		offset = qd2offset(qd);
710 		error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
711 					  (struct gfs2_quota_data *)
712 					  qd->qd_gl->gl_lvb);
713 		if (error)
714 			goto out_end_trans;
715 
716 		do_qc(qd, -qd->qd_change_sync);
717 	}
718 
719 	error = 0;
720 
721 out_end_trans:
722 	gfs2_trans_end(sdp);
723 out_ipres:
724 	if (nalloc)
725 		gfs2_inplace_release(ip);
726 out_alloc:
727 	if (nalloc)
728 		gfs2_alloc_put(ip);
729 out_gunlock:
730 	gfs2_glock_dq_uninit(&i_gh);
731 out:
732 	while (qx--)
733 		gfs2_glock_dq_uninit(&ghs[qx]);
734 	kfree(ghs);
735 	gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
736 	return error;
737 }
738 
739 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
740 		    struct gfs2_holder *q_gh)
741 {
742 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
743 	struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
744 	struct gfs2_holder i_gh;
745 	struct gfs2_quota_host q;
746 	char buf[sizeof(struct gfs2_quota)];
747 	struct file_ra_state ra_state;
748 	int error;
749 	struct gfs2_quota_lvb *qlvb;
750 
751 	file_ra_state_init(&ra_state, sdp->sd_quota_inode->i_mapping);
752 restart:
753 	error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
754 	if (error)
755 		return error;
756 
757 	qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
758 
759 	if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
760 		loff_t pos;
761 		gfs2_glock_dq_uninit(q_gh);
762 		error = gfs2_glock_nq_init(qd->qd_gl,
763 					  LM_ST_EXCLUSIVE, GL_NOCACHE,
764 					  q_gh);
765 		if (error)
766 			return error;
767 
768 		error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
769 		if (error)
770 			goto fail;
771 
772 		memset(buf, 0, sizeof(struct gfs2_quota));
773 		pos = qd2offset(qd);
774 		error = gfs2_internal_read(ip, &ra_state, buf,
775 					   &pos, sizeof(struct gfs2_quota));
776 		if (error < 0)
777 			goto fail_gunlock;
778 
779 		gfs2_glock_dq_uninit(&i_gh);
780 
781 
782 		gfs2_quota_in(&q, buf);
783 		qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
784 		qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
785 		qlvb->__pad = 0;
786 		qlvb->qb_limit = cpu_to_be64(q.qu_limit);
787 		qlvb->qb_warn = cpu_to_be64(q.qu_warn);
788 		qlvb->qb_value = cpu_to_be64(q.qu_value);
789 		qd->qd_qb = *qlvb;
790 
791 		if (gfs2_glock_is_blocking(qd->qd_gl)) {
792 			gfs2_glock_dq_uninit(q_gh);
793 			force_refresh = 0;
794 			goto restart;
795 		}
796 	}
797 
798 	return 0;
799 
800 fail_gunlock:
801 	gfs2_glock_dq_uninit(&i_gh);
802 fail:
803 	gfs2_glock_dq_uninit(q_gh);
804 	return error;
805 }
806 
807 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
808 {
809 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
810 	struct gfs2_alloc *al = &ip->i_alloc;
811 	unsigned int x;
812 	int error = 0;
813 
814 	gfs2_quota_hold(ip, uid, gid);
815 
816 	if (capable(CAP_SYS_RESOURCE) ||
817 	    sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
818 		return 0;
819 
820 	sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
821 	     sort_qd, NULL);
822 
823 	for (x = 0; x < al->al_qd_num; x++) {
824 		error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
825 		if (error)
826 			break;
827 	}
828 
829 	if (!error)
830 		set_bit(GIF_QD_LOCKED, &ip->i_flags);
831 	else {
832 		while (x--)
833 			gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
834 		gfs2_quota_unhold(ip);
835 	}
836 
837 	return error;
838 }
839 
840 static int need_sync(struct gfs2_quota_data *qd)
841 {
842 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
843 	struct gfs2_tune *gt = &sdp->sd_tune;
844 	s64 value;
845 	unsigned int num, den;
846 	int do_sync = 1;
847 
848 	if (!qd->qd_qb.qb_limit)
849 		return 0;
850 
851 	spin_lock(&sdp->sd_quota_spin);
852 	value = qd->qd_change;
853 	spin_unlock(&sdp->sd_quota_spin);
854 
855 	spin_lock(&gt->gt_spin);
856 	num = gt->gt_quota_scale_num;
857 	den = gt->gt_quota_scale_den;
858 	spin_unlock(&gt->gt_spin);
859 
860 	if (value < 0)
861 		do_sync = 0;
862 	else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
863 		 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
864 		do_sync = 0;
865 	else {
866 		value *= gfs2_jindex_size(sdp) * num;
867 		do_div(value, den);
868 		value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
869 		if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
870 			do_sync = 0;
871 	}
872 
873 	return do_sync;
874 }
875 
876 void gfs2_quota_unlock(struct gfs2_inode *ip)
877 {
878 	struct gfs2_alloc *al = &ip->i_alloc;
879 	struct gfs2_quota_data *qda[4];
880 	unsigned int count = 0;
881 	unsigned int x;
882 
883 	if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
884 		goto out;
885 
886 	for (x = 0; x < al->al_qd_num; x++) {
887 		struct gfs2_quota_data *qd;
888 		int sync;
889 
890 		qd = al->al_qd[x];
891 		sync = need_sync(qd);
892 
893 		gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
894 
895 		if (sync && qd_trylock(qd))
896 			qda[count++] = qd;
897 	}
898 
899 	if (count) {
900 		do_sync(count, qda);
901 		for (x = 0; x < count; x++)
902 			qd_unlock(qda[x]);
903 	}
904 
905 out:
906 	gfs2_quota_unhold(ip);
907 }
908 
909 #define MAX_LINE 256
910 
911 static int print_message(struct gfs2_quota_data *qd, char *type)
912 {
913 	struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
914 
915 	printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
916 	       sdp->sd_fsname, type,
917 	       (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
918 	       qd->qd_id);
919 
920 	return 0;
921 }
922 
923 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
924 {
925 	struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
926 	struct gfs2_alloc *al = &ip->i_alloc;
927 	struct gfs2_quota_data *qd;
928 	s64 value;
929 	unsigned int x;
930 	int error = 0;
931 
932 	if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
933 		return 0;
934 
935         if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
936                 return 0;
937 
938 	for (x = 0; x < al->al_qd_num; x++) {
939 		qd = al->al_qd[x];
940 
941 		if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
942 		      (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
943 			continue;
944 
945 		value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
946 		spin_lock(&sdp->sd_quota_spin);
947 		value += qd->qd_change;
948 		spin_unlock(&sdp->sd_quota_spin);
949 
950 		if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
951 			print_message(qd, "exceeded");
952 			error = -EDQUOT;
953 			break;
954 		} else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
955 			   (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
956 			   time_after_eq(jiffies, qd->qd_last_warn +
957 					 gfs2_tune_get(sdp,
958 						gt_quota_warn_period) * HZ)) {
959 			error = print_message(qd, "warning");
960 			qd->qd_last_warn = jiffies;
961 		}
962 	}
963 
964 	return error;
965 }
966 
967 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
968 		       u32 uid, u32 gid)
969 {
970 	struct gfs2_alloc *al = &ip->i_alloc;
971 	struct gfs2_quota_data *qd;
972 	unsigned int x;
973 	unsigned int found = 0;
974 
975 	if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
976 		return;
977 	if (ip->i_di.di_flags & GFS2_DIF_SYSTEM)
978 		return;
979 
980 	for (x = 0; x < al->al_qd_num; x++) {
981 		qd = al->al_qd[x];
982 
983 		if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
984 		    (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
985 			do_qc(qd, change);
986 			found++;
987 		}
988 	}
989 }
990 
991 int gfs2_quota_sync(struct gfs2_sbd *sdp)
992 {
993 	struct gfs2_quota_data **qda;
994 	unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
995 	unsigned int num_qd;
996 	unsigned int x;
997 	int error = 0;
998 
999 	sdp->sd_quota_sync_gen++;
1000 
1001 	qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1002 	if (!qda)
1003 		return -ENOMEM;
1004 
1005 	do {
1006 		num_qd = 0;
1007 
1008 		for (;;) {
1009 			error = qd_fish(sdp, qda + num_qd);
1010 			if (error || !qda[num_qd])
1011 				break;
1012 			if (++num_qd == max_qd)
1013 				break;
1014 		}
1015 
1016 		if (num_qd) {
1017 			if (!error)
1018 				error = do_sync(num_qd, qda);
1019 			if (!error)
1020 				for (x = 0; x < num_qd; x++)
1021 					qda[x]->qd_sync_gen =
1022 						sdp->sd_quota_sync_gen;
1023 
1024 			for (x = 0; x < num_qd; x++)
1025 				qd_unlock(qda[x]);
1026 		}
1027 	} while (!error && num_qd == max_qd);
1028 
1029 	kfree(qda);
1030 
1031 	return error;
1032 }
1033 
1034 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1035 {
1036 	struct gfs2_quota_data *qd;
1037 	struct gfs2_holder q_gh;
1038 	int error;
1039 
1040 	error = qd_get(sdp, user, id, CREATE, &qd);
1041 	if (error)
1042 		return error;
1043 
1044 	error = do_glock(qd, FORCE, &q_gh);
1045 	if (!error)
1046 		gfs2_glock_dq_uninit(&q_gh);
1047 
1048 	qd_put(qd);
1049 
1050 	return error;
1051 }
1052 
1053 int gfs2_quota_init(struct gfs2_sbd *sdp)
1054 {
1055 	struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1056 	unsigned int blocks = ip->i_di.di_size >> sdp->sd_sb.sb_bsize_shift;
1057 	unsigned int x, slot = 0;
1058 	unsigned int found = 0;
1059 	u64 dblock;
1060 	u32 extlen = 0;
1061 	int error;
1062 
1063 	if (!ip->i_di.di_size || ip->i_di.di_size > (64 << 20) ||
1064 	    ip->i_di.di_size & (sdp->sd_sb.sb_bsize - 1)) {
1065 		gfs2_consist_inode(ip);
1066 		return -EIO;
1067 	}
1068 	sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1069 	sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1070 
1071 	error = -ENOMEM;
1072 
1073 	sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1074 				       sizeof(unsigned char *), GFP_KERNEL);
1075 	if (!sdp->sd_quota_bitmap)
1076 		return error;
1077 
1078 	for (x = 0; x < sdp->sd_quota_chunks; x++) {
1079 		sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_KERNEL);
1080 		if (!sdp->sd_quota_bitmap[x])
1081 			goto fail;
1082 	}
1083 
1084 	for (x = 0; x < blocks; x++) {
1085 		struct buffer_head *bh;
1086 		unsigned int y;
1087 
1088 		if (!extlen) {
1089 			int new = 0;
1090 			error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1091 			if (error)
1092 				goto fail;
1093 		}
1094 		error = -EIO;
1095 		bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1096 		if (!bh)
1097 			goto fail;
1098 		if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1099 			brelse(bh);
1100 			goto fail;
1101 		}
1102 
1103 		for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1104 		     y++, slot++) {
1105 			struct gfs2_quota_change_host qc;
1106 			struct gfs2_quota_data *qd;
1107 
1108 			gfs2_quota_change_in(&qc, bh->b_data +
1109 					  sizeof(struct gfs2_meta_header) +
1110 					  y * sizeof(struct gfs2_quota_change));
1111 			if (!qc.qc_change)
1112 				continue;
1113 
1114 			error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1115 					 qc.qc_id, &qd);
1116 			if (error) {
1117 				brelse(bh);
1118 				goto fail;
1119 			}
1120 
1121 			set_bit(QDF_CHANGE, &qd->qd_flags);
1122 			qd->qd_change = qc.qc_change;
1123 			qd->qd_slot = slot;
1124 			qd->qd_slot_count = 1;
1125 			qd->qd_last_touched = jiffies;
1126 
1127 			spin_lock(&sdp->sd_quota_spin);
1128 			gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1129 			list_add(&qd->qd_list, &sdp->sd_quota_list);
1130 			atomic_inc(&sdp->sd_quota_count);
1131 			spin_unlock(&sdp->sd_quota_spin);
1132 
1133 			found++;
1134 		}
1135 
1136 		brelse(bh);
1137 		dblock++;
1138 		extlen--;
1139 	}
1140 
1141 	if (found)
1142 		fs_info(sdp, "found %u quota changes\n", found);
1143 
1144 	return 0;
1145 
1146 fail:
1147 	gfs2_quota_cleanup(sdp);
1148 	return error;
1149 }
1150 
1151 void gfs2_quota_scan(struct gfs2_sbd *sdp)
1152 {
1153 	struct gfs2_quota_data *qd, *safe;
1154 	LIST_HEAD(dead);
1155 
1156 	spin_lock(&sdp->sd_quota_spin);
1157 	list_for_each_entry_safe(qd, safe, &sdp->sd_quota_list, qd_list) {
1158 		if (!qd->qd_count &&
1159 		    time_after_eq(jiffies, qd->qd_last_touched +
1160 			        gfs2_tune_get(sdp, gt_quota_cache_secs) * HZ)) {
1161 			list_move(&qd->qd_list, &dead);
1162 			gfs2_assert_warn(sdp,
1163 					 atomic_read(&sdp->sd_quota_count) > 0);
1164 			atomic_dec(&sdp->sd_quota_count);
1165 		}
1166 	}
1167 	spin_unlock(&sdp->sd_quota_spin);
1168 
1169 	while (!list_empty(&dead)) {
1170 		qd = list_entry(dead.next, struct gfs2_quota_data, qd_list);
1171 		list_del(&qd->qd_list);
1172 
1173 		gfs2_assert_warn(sdp, !qd->qd_change);
1174 		gfs2_assert_warn(sdp, !qd->qd_slot_count);
1175 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1176 
1177 		gfs2_lvb_unhold(qd->qd_gl);
1178 		kfree(qd);
1179 	}
1180 }
1181 
1182 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1183 {
1184 	struct list_head *head = &sdp->sd_quota_list;
1185 	struct gfs2_quota_data *qd;
1186 	unsigned int x;
1187 
1188 	spin_lock(&sdp->sd_quota_spin);
1189 	while (!list_empty(head)) {
1190 		qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1191 
1192 		if (qd->qd_count > 1 ||
1193 		    (qd->qd_count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1194 			list_move(&qd->qd_list, head);
1195 			spin_unlock(&sdp->sd_quota_spin);
1196 			schedule();
1197 			spin_lock(&sdp->sd_quota_spin);
1198 			continue;
1199 		}
1200 
1201 		list_del(&qd->qd_list);
1202 		atomic_dec(&sdp->sd_quota_count);
1203 		spin_unlock(&sdp->sd_quota_spin);
1204 
1205 		if (!qd->qd_count) {
1206 			gfs2_assert_warn(sdp, !qd->qd_change);
1207 			gfs2_assert_warn(sdp, !qd->qd_slot_count);
1208 		} else
1209 			gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1210 		gfs2_assert_warn(sdp, !qd->qd_bh_count);
1211 
1212 		gfs2_lvb_unhold(qd->qd_gl);
1213 		kfree(qd);
1214 
1215 		spin_lock(&sdp->sd_quota_spin);
1216 	}
1217 	spin_unlock(&sdp->sd_quota_spin);
1218 
1219 	gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1220 
1221 	if (sdp->sd_quota_bitmap) {
1222 		for (x = 0; x < sdp->sd_quota_chunks; x++)
1223 			kfree(sdp->sd_quota_bitmap[x]);
1224 		kfree(sdp->sd_quota_bitmap);
1225 	}
1226 }
1227 
1228