2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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.
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 for a quota check
19 * program to be run on node crashes or anything like that.
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.
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.
39 #include <linux/sched.h>
40 #include <linux/slab.h>
42 #include <linux/spinlock.h>
43 #include <linux/completion.h>
44 #include <linux/buffer_head.h>
45 #include <linux/sort.h>
47 #include <linux/bio.h>
48 #include <linux/gfs2_ondisk.h>
49 #include <linux/kthread.h>
50 #include <linux/freezer.h>
51 #include <linux/quota.h>
52 #include <linux/dqblk_xfs.h>
71 struct gfs2_quota_change_host {
73 u32 qc_flags; /* GFS2_QCF_... */
77 static LIST_HEAD(qd_lru_list);
78 static atomic_t qd_lru_count = ATOMIC_INIT(0);
79 static DEFINE_SPINLOCK(qd_lru_lock);
81 int gfs2_shrink_qd_memory(struct shrinker *shrink, struct shrink_control *sc)
83 struct gfs2_quota_data *qd;
85 int nr_to_scan = sc->nr_to_scan;
90 if (!(sc->gfp_mask & __GFP_FS))
93 spin_lock(&qd_lru_lock);
94 while (nr_to_scan && !list_empty(&qd_lru_list)) {
95 qd = list_entry(qd_lru_list.next,
96 struct gfs2_quota_data, qd_reclaim);
97 sdp = qd->qd_gl->gl_sbd;
99 /* Free from the filesystem-specific list */
100 list_del(&qd->qd_list);
102 gfs2_assert_warn(sdp, !qd->qd_change);
103 gfs2_assert_warn(sdp, !qd->qd_slot_count);
104 gfs2_assert_warn(sdp, !qd->qd_bh_count);
106 gfs2_glock_put(qd->qd_gl);
107 atomic_dec(&sdp->sd_quota_count);
109 /* Delete it from the common reclaim list */
110 list_del_init(&qd->qd_reclaim);
111 atomic_dec(&qd_lru_count);
112 spin_unlock(&qd_lru_lock);
113 kmem_cache_free(gfs2_quotad_cachep, qd);
114 spin_lock(&qd_lru_lock);
117 spin_unlock(&qd_lru_lock);
120 return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
123 static u64 qd2offset(struct gfs2_quota_data *qd)
127 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
128 offset *= sizeof(struct gfs2_quota);
133 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
134 struct gfs2_quota_data **qdp)
136 struct gfs2_quota_data *qd;
139 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
143 atomic_set(&qd->qd_count, 1);
146 set_bit(QDF_USER, &qd->qd_flags);
148 INIT_LIST_HEAD(&qd->qd_reclaim);
150 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
151 &gfs2_quota_glops, CREATE, &qd->qd_gl);
160 kmem_cache_free(gfs2_quotad_cachep, qd);
164 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id,
165 struct gfs2_quota_data **qdp)
167 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
174 spin_lock(&qd_lru_lock);
175 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
176 if (qd->qd_id == id &&
177 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
178 if (!atomic_read(&qd->qd_count) &&
179 !list_empty(&qd->qd_reclaim)) {
180 /* Remove it from reclaim list */
181 list_del_init(&qd->qd_reclaim);
182 atomic_dec(&qd_lru_count);
184 atomic_inc(&qd->qd_count);
195 list_add(&qd->qd_list, &sdp->sd_quota_list);
196 atomic_inc(&sdp->sd_quota_count);
200 spin_unlock(&qd_lru_lock);
204 gfs2_glock_put(new_qd->qd_gl);
205 kmem_cache_free(gfs2_quotad_cachep, new_qd);
211 error = qd_alloc(sdp, user, id, &new_qd);
217 static void qd_hold(struct gfs2_quota_data *qd)
219 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
220 gfs2_assert(sdp, atomic_read(&qd->qd_count));
221 atomic_inc(&qd->qd_count);
224 static void qd_put(struct gfs2_quota_data *qd)
226 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
227 /* Add to the reclaim list */
228 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
229 atomic_inc(&qd_lru_count);
230 spin_unlock(&qd_lru_lock);
234 static int slot_get(struct gfs2_quota_data *qd)
236 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
237 unsigned int c, o = 0, b;
238 unsigned char byte = 0;
240 spin_lock(&qd_lru_lock);
242 if (qd->qd_slot_count++) {
243 spin_unlock(&qd_lru_lock);
247 for (c = 0; c < sdp->sd_quota_chunks; c++)
248 for (o = 0; o < PAGE_SIZE; o++) {
249 byte = sdp->sd_quota_bitmap[c][o];
257 for (b = 0; b < 8; b++)
258 if (!(byte & (1 << b)))
260 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
262 if (qd->qd_slot >= sdp->sd_quota_slots)
265 sdp->sd_quota_bitmap[c][o] |= 1 << b;
267 spin_unlock(&qd_lru_lock);
273 spin_unlock(&qd_lru_lock);
277 static void slot_hold(struct gfs2_quota_data *qd)
279 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
281 spin_lock(&qd_lru_lock);
282 gfs2_assert(sdp, qd->qd_slot_count);
284 spin_unlock(&qd_lru_lock);
287 static void slot_put(struct gfs2_quota_data *qd)
289 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
291 spin_lock(&qd_lru_lock);
292 gfs2_assert(sdp, qd->qd_slot_count);
293 if (!--qd->qd_slot_count) {
294 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
297 spin_unlock(&qd_lru_lock);
300 static int bh_get(struct gfs2_quota_data *qd)
302 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
303 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
304 unsigned int block, offset;
305 struct buffer_head *bh;
307 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
309 mutex_lock(&sdp->sd_quota_mutex);
311 if (qd->qd_bh_count++) {
312 mutex_unlock(&sdp->sd_quota_mutex);
316 block = qd->qd_slot / sdp->sd_qc_per_block;
317 offset = qd->qd_slot % sdp->sd_qc_per_block;
319 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
320 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
323 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
327 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
331 qd->qd_bh_qc = (struct gfs2_quota_change *)
332 (bh->b_data + sizeof(struct gfs2_meta_header) +
333 offset * sizeof(struct gfs2_quota_change));
335 mutex_unlock(&sdp->sd_quota_mutex);
343 mutex_unlock(&sdp->sd_quota_mutex);
347 static void bh_put(struct gfs2_quota_data *qd)
349 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
351 mutex_lock(&sdp->sd_quota_mutex);
352 gfs2_assert(sdp, qd->qd_bh_count);
353 if (!--qd->qd_bh_count) {
358 mutex_unlock(&sdp->sd_quota_mutex);
361 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
363 struct gfs2_quota_data *qd = NULL;
369 if (sdp->sd_vfs->s_flags & MS_RDONLY)
372 spin_lock(&qd_lru_lock);
374 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
375 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
376 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
377 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
380 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
382 set_bit(QDF_LOCKED, &qd->qd_flags);
383 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
384 atomic_inc(&qd->qd_count);
385 qd->qd_change_sync = qd->qd_change;
386 gfs2_assert_warn(sdp, qd->qd_slot_count);
396 spin_unlock(&qd_lru_lock);
399 gfs2_assert_warn(sdp, qd->qd_change_sync);
402 clear_bit(QDF_LOCKED, &qd->qd_flags);
414 static int qd_trylock(struct gfs2_quota_data *qd)
416 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
418 if (sdp->sd_vfs->s_flags & MS_RDONLY)
421 spin_lock(&qd_lru_lock);
423 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
424 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
425 spin_unlock(&qd_lru_lock);
429 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
431 set_bit(QDF_LOCKED, &qd->qd_flags);
432 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
433 atomic_inc(&qd->qd_count);
434 qd->qd_change_sync = qd->qd_change;
435 gfs2_assert_warn(sdp, qd->qd_slot_count);
438 spin_unlock(&qd_lru_lock);
440 gfs2_assert_warn(sdp, qd->qd_change_sync);
442 clear_bit(QDF_LOCKED, &qd->qd_flags);
451 static void qd_unlock(struct gfs2_quota_data *qd)
453 gfs2_assert_warn(qd->qd_gl->gl_sbd,
454 test_bit(QDF_LOCKED, &qd->qd_flags));
455 clear_bit(QDF_LOCKED, &qd->qd_flags);
461 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id,
462 struct gfs2_quota_data **qdp)
466 error = qd_get(sdp, user, id, qdp);
470 error = slot_get(*qdp);
474 error = bh_get(*qdp);
487 static void qdsb_put(struct gfs2_quota_data *qd)
494 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
496 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
497 struct gfs2_alloc *al = ip->i_alloc;
498 struct gfs2_quota_data **qd = al->al_qd;
501 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
502 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
505 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
508 error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, qd);
514 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, qd);
520 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
521 error = qdsb_get(sdp, QUOTA_USER, uid, qd);
528 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
529 error = qdsb_get(sdp, QUOTA_GROUP, gid, qd);
538 gfs2_quota_unhold(ip);
542 void gfs2_quota_unhold(struct gfs2_inode *ip)
544 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
545 struct gfs2_alloc *al = ip->i_alloc;
548 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
550 for (x = 0; x < al->al_qd_num; x++) {
551 qdsb_put(al->al_qd[x]);
557 static int sort_qd(const void *a, const void *b)
559 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
560 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
562 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
563 !test_bit(QDF_USER, &qd_b->qd_flags)) {
564 if (test_bit(QDF_USER, &qd_a->qd_flags))
569 if (qd_a->qd_id < qd_b->qd_id)
571 if (qd_a->qd_id > qd_b->qd_id)
577 static void do_qc(struct gfs2_quota_data *qd, s64 change)
579 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
580 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
581 struct gfs2_quota_change *qc = qd->qd_bh_qc;
584 mutex_lock(&sdp->sd_quota_mutex);
585 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
587 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
590 if (test_bit(QDF_USER, &qd->qd_flags))
591 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
592 qc->qc_id = cpu_to_be32(qd->qd_id);
595 x = be64_to_cpu(qc->qc_change) + change;
596 qc->qc_change = cpu_to_be64(x);
598 spin_lock(&qd_lru_lock);
600 spin_unlock(&qd_lru_lock);
603 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
604 clear_bit(QDF_CHANGE, &qd->qd_flags);
609 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
614 mutex_unlock(&sdp->sd_quota_mutex);
618 * gfs2_adjust_quota - adjust record of current block usage
619 * @ip: The quota inode
620 * @loc: Offset of the entry in the quota file
621 * @change: The amount of usage change to record
622 * @qd: The quota data
623 * @fdq: The updated limits to record
625 * This function was mostly borrowed from gfs2_block_truncate_page which was
626 * in turn mostly borrowed from ext3
628 * Returns: 0 or -ve on error
631 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
632 s64 change, struct gfs2_quota_data *qd,
633 struct fs_disk_quota *fdq)
635 struct inode *inode = &ip->i_inode;
636 struct gfs2_sbd *sdp = GFS2_SB(inode);
637 struct address_space *mapping = inode->i_mapping;
638 unsigned long index = loc >> PAGE_CACHE_SHIFT;
639 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
640 unsigned blocksize, iblock, pos;
641 struct buffer_head *bh;
644 struct gfs2_quota q, *qp;
648 if (gfs2_is_stuffed(ip)) {
649 err = gfs2_unstuff_dinode(ip, NULL);
654 memset(&q, 0, sizeof(struct gfs2_quota));
655 err = gfs2_internal_read(ip, NULL, (char *)&q, &loc, sizeof(q));
661 qp->qu_value = be64_to_cpu(qp->qu_value);
662 qp->qu_value += change;
663 qp->qu_value = cpu_to_be64(qp->qu_value);
664 qd->qd_qb.qb_value = qp->qu_value;
666 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
667 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
668 qd->qd_qb.qb_warn = qp->qu_warn;
670 if (fdq->d_fieldmask & FS_DQ_BHARD) {
671 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
672 qd->qd_qb.qb_limit = qp->qu_limit;
674 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
675 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
676 qd->qd_qb.qb_value = qp->qu_value;
680 /* Write the quota into the quota file on disk */
682 nbytes = sizeof(struct gfs2_quota);
684 page = grab_cache_page(mapping, index);
688 blocksize = inode->i_sb->s_blocksize;
689 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
691 if (!page_has_buffers(page))
692 create_empty_buffers(page, blocksize, 0);
694 bh = page_buffers(page);
696 while (offset >= pos) {
697 bh = bh->b_this_page;
702 if (!buffer_mapped(bh)) {
703 gfs2_block_map(inode, iblock, bh, 1);
704 if (!buffer_mapped(bh))
706 /* If it's a newly allocated disk block for quota, zero it */
708 zero_user(page, pos - blocksize, bh->b_size);
711 if (PageUptodate(page))
712 set_buffer_uptodate(bh);
714 if (!buffer_uptodate(bh)) {
715 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
717 if (!buffer_uptodate(bh))
721 gfs2_trans_add_bh(ip->i_gl, bh, 0);
723 kaddr = kmap_atomic(page, KM_USER0);
724 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
725 nbytes = PAGE_CACHE_SIZE - offset;
726 memcpy(kaddr + offset, ptr, nbytes);
727 flush_dcache_page(page);
728 kunmap_atomic(kaddr, KM_USER0);
730 page_cache_release(page);
732 /* If quota straddles page boundary, we need to update the rest of the
733 * quota at the beginning of the next page */
734 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
736 nbytes = sizeof(struct gfs2_quota) - nbytes;
742 size = loc + sizeof(struct gfs2_quota);
743 if (size > inode->i_size)
744 i_size_write(inode, size);
745 inode->i_mtime = inode->i_atime = CURRENT_TIME;
746 mark_inode_dirty(inode);
751 page_cache_release(page);
755 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
757 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
758 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
759 unsigned int data_blocks, ind_blocks;
760 struct gfs2_holder *ghs, i_gh;
762 struct gfs2_quota_data *qd;
764 unsigned int nalloc = 0, blocks;
765 struct gfs2_alloc *al = NULL;
768 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
769 &data_blocks, &ind_blocks);
771 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
775 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
776 mutex_lock_nested(&ip->i_inode.i_mutex, I_MUTEX_QUOTA);
777 for (qx = 0; qx < num_qd; qx++) {
778 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
779 GL_NOCACHE, &ghs[qx]);
784 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
788 for (x = 0; x < num_qd; x++) {
789 offset = qd2offset(qda[x]);
790 if (gfs2_write_alloc_required(ip, offset,
791 sizeof(struct gfs2_quota)))
795 al = gfs2_alloc_get(ip);
801 * 1 blk for unstuffing inode if stuffed. We add this extra
802 * block to the reservation unconditionally. If the inode
803 * doesn't need unstuffing, the block will be released to the
804 * rgrp since it won't be allocated during the transaction
806 al->al_requested = 1;
807 /* +3 in the end for unstuffing block, inode size update block
808 * and another block in case quota straddles page boundary and
809 * two blocks need to be updated instead of 1 */
810 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
813 al->al_requested += nalloc * (data_blocks + ind_blocks);
814 error = gfs2_inplace_reserve(ip);
819 blocks += gfs2_rg_blocks(ip) + nalloc * ind_blocks + RES_STATFS;
821 error = gfs2_trans_begin(sdp, blocks, 0);
825 for (x = 0; x < num_qd; x++) {
827 offset = qd2offset(qd);
828 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
832 do_qc(qd, -qd->qd_change_sync);
833 set_bit(QDF_REFRESH, &qd->qd_flags);
841 gfs2_inplace_release(ip);
845 gfs2_glock_dq_uninit(&i_gh);
848 gfs2_glock_dq_uninit(&ghs[qx]);
849 mutex_unlock(&ip->i_inode.i_mutex);
851 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
855 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
857 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
859 struct gfs2_quota_lvb *qlvb;
863 memset(&q, 0, sizeof(struct gfs2_quota));
865 error = gfs2_internal_read(ip, NULL, (char *)&q, &pos, sizeof(q));
869 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
870 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
872 qlvb->qb_limit = q.qu_limit;
873 qlvb->qb_warn = q.qu_warn;
874 qlvb->qb_value = q.qu_value;
880 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
881 struct gfs2_holder *q_gh)
883 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
884 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
885 struct gfs2_holder i_gh;
889 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
893 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
895 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
896 gfs2_glock_dq_uninit(q_gh);
897 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
902 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
906 error = update_qd(sdp, qd);
910 gfs2_glock_dq_uninit(&i_gh);
911 gfs2_glock_dq_uninit(q_gh);
919 gfs2_glock_dq_uninit(&i_gh);
921 gfs2_glock_dq_uninit(q_gh);
925 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
927 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
928 struct gfs2_alloc *al = ip->i_alloc;
929 struct gfs2_quota_data *qd;
933 error = gfs2_quota_hold(ip, uid, gid);
937 if (capable(CAP_SYS_RESOURCE) ||
938 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
941 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
944 for (x = 0; x < al->al_qd_num; x++) {
945 int force = NO_FORCE;
947 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
949 error = do_glock(qd, force, &al->al_qd_ghs[x]);
955 set_bit(GIF_QD_LOCKED, &ip->i_flags);
958 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
959 gfs2_quota_unhold(ip);
965 static int need_sync(struct gfs2_quota_data *qd)
967 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
968 struct gfs2_tune *gt = &sdp->sd_tune;
970 unsigned int num, den;
973 if (!qd->qd_qb.qb_limit)
976 spin_lock(&qd_lru_lock);
977 value = qd->qd_change;
978 spin_unlock(&qd_lru_lock);
980 spin_lock(>->gt_spin);
981 num = gt->gt_quota_scale_num;
982 den = gt->gt_quota_scale_den;
983 spin_unlock(>->gt_spin);
987 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
988 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
991 value *= gfs2_jindex_size(sdp) * num;
992 value = div_s64(value, den);
993 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
994 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1001 void gfs2_quota_unlock(struct gfs2_inode *ip)
1003 struct gfs2_alloc *al = ip->i_alloc;
1004 struct gfs2_quota_data *qda[4];
1005 unsigned int count = 0;
1008 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1011 for (x = 0; x < al->al_qd_num; x++) {
1012 struct gfs2_quota_data *qd;
1016 sync = need_sync(qd);
1018 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
1020 if (sync && qd_trylock(qd))
1025 do_sync(count, qda);
1026 for (x = 0; x < count; x++)
1031 gfs2_quota_unhold(ip);
1034 #define MAX_LINE 256
1036 static int print_message(struct gfs2_quota_data *qd, char *type)
1038 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1040 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1041 sdp->sd_fsname, type,
1042 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
1048 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
1050 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1051 struct gfs2_alloc *al = ip->i_alloc;
1052 struct gfs2_quota_data *qd;
1057 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1060 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1063 for (x = 0; x < al->al_qd_num; x++) {
1066 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1067 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
1070 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1071 spin_lock(&qd_lru_lock);
1072 value += qd->qd_change;
1073 spin_unlock(&qd_lru_lock);
1075 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1076 print_message(qd, "exceeded");
1077 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1078 USRQUOTA : GRPQUOTA, qd->qd_id,
1079 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1083 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1084 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1085 time_after_eq(jiffies, qd->qd_last_warn +
1087 gt_quota_warn_period) * HZ)) {
1088 quota_send_warning(test_bit(QDF_USER, &qd->qd_flags) ?
1089 USRQUOTA : GRPQUOTA, qd->qd_id,
1090 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1091 error = print_message(qd, "warning");
1092 qd->qd_last_warn = jiffies;
1099 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1102 struct gfs2_alloc *al = ip->i_alloc;
1103 struct gfs2_quota_data *qd;
1106 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1108 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1111 for (x = 0; x < al->al_qd_num; x++) {
1114 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1115 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1121 int gfs2_quota_sync(struct super_block *sb, int type, int wait)
1123 struct gfs2_sbd *sdp = sb->s_fs_info;
1124 struct gfs2_quota_data **qda;
1125 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1126 unsigned int num_qd;
1130 sdp->sd_quota_sync_gen++;
1132 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1140 error = qd_fish(sdp, qda + num_qd);
1141 if (error || !qda[num_qd])
1143 if (++num_qd == max_qd)
1149 error = do_sync(num_qd, qda);
1151 for (x = 0; x < num_qd; x++)
1152 qda[x]->qd_sync_gen =
1153 sdp->sd_quota_sync_gen;
1155 for (x = 0; x < num_qd; x++)
1158 } while (!error && num_qd == max_qd);
1165 static int gfs2_quota_sync_timeo(struct super_block *sb, int type)
1167 return gfs2_quota_sync(sb, type, 0);
1170 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1172 struct gfs2_quota_data *qd;
1173 struct gfs2_holder q_gh;
1176 error = qd_get(sdp, user, id, &qd);
1180 error = do_glock(qd, FORCE, &q_gh);
1182 gfs2_glock_dq_uninit(&q_gh);
1188 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1190 const struct gfs2_quota_change *str = buf;
1192 qc->qc_change = be64_to_cpu(str->qc_change);
1193 qc->qc_flags = be32_to_cpu(str->qc_flags);
1194 qc->qc_id = be32_to_cpu(str->qc_id);
1197 int gfs2_quota_init(struct gfs2_sbd *sdp)
1199 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1200 u64 size = i_size_read(sdp->sd_qc_inode);
1201 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1202 unsigned int x, slot = 0;
1203 unsigned int found = 0;
1208 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1211 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1212 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1216 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1217 sizeof(unsigned char *), GFP_NOFS);
1218 if (!sdp->sd_quota_bitmap)
1221 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1222 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1223 if (!sdp->sd_quota_bitmap[x])
1227 for (x = 0; x < blocks; x++) {
1228 struct buffer_head *bh;
1233 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1238 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1241 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1246 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1248 struct gfs2_quota_change_host qc;
1249 struct gfs2_quota_data *qd;
1251 gfs2_quota_change_in(&qc, bh->b_data +
1252 sizeof(struct gfs2_meta_header) +
1253 y * sizeof(struct gfs2_quota_change));
1257 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1264 set_bit(QDF_CHANGE, &qd->qd_flags);
1265 qd->qd_change = qc.qc_change;
1267 qd->qd_slot_count = 1;
1269 spin_lock(&qd_lru_lock);
1270 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1271 list_add(&qd->qd_list, &sdp->sd_quota_list);
1272 atomic_inc(&sdp->sd_quota_count);
1273 spin_unlock(&qd_lru_lock);
1284 fs_info(sdp, "found %u quota changes\n", found);
1289 gfs2_quota_cleanup(sdp);
1293 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1295 struct list_head *head = &sdp->sd_quota_list;
1296 struct gfs2_quota_data *qd;
1299 spin_lock(&qd_lru_lock);
1300 while (!list_empty(head)) {
1301 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1303 if (atomic_read(&qd->qd_count) > 1 ||
1304 (atomic_read(&qd->qd_count) &&
1305 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1306 list_move(&qd->qd_list, head);
1307 spin_unlock(&qd_lru_lock);
1309 spin_lock(&qd_lru_lock);
1313 list_del(&qd->qd_list);
1314 /* Also remove if this qd exists in the reclaim list */
1315 if (!list_empty(&qd->qd_reclaim)) {
1316 list_del_init(&qd->qd_reclaim);
1317 atomic_dec(&qd_lru_count);
1319 atomic_dec(&sdp->sd_quota_count);
1320 spin_unlock(&qd_lru_lock);
1322 if (!atomic_read(&qd->qd_count)) {
1323 gfs2_assert_warn(sdp, !qd->qd_change);
1324 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1326 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1327 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1329 gfs2_glock_put(qd->qd_gl);
1330 kmem_cache_free(gfs2_quotad_cachep, qd);
1332 spin_lock(&qd_lru_lock);
1334 spin_unlock(&qd_lru_lock);
1336 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1338 if (sdp->sd_quota_bitmap) {
1339 for (x = 0; x < sdp->sd_quota_chunks; x++)
1340 kfree(sdp->sd_quota_bitmap[x]);
1341 kfree(sdp->sd_quota_bitmap);
1345 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1347 if (error == 0 || error == -EROFS)
1349 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1350 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1353 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1354 int (*fxn)(struct super_block *sb, int type),
1355 unsigned long t, unsigned long *timeo,
1356 unsigned int *new_timeo)
1359 int error = fxn(sdp->sd_vfs, 0);
1360 quotad_error(sdp, msg, error);
1361 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1367 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1369 struct gfs2_inode *ip;
1373 spin_lock(&sdp->sd_trunc_lock);
1374 if (!list_empty(&sdp->sd_trunc_list)) {
1375 ip = list_entry(sdp->sd_trunc_list.next,
1376 struct gfs2_inode, i_trunc_list);
1377 list_del_init(&ip->i_trunc_list);
1379 spin_unlock(&sdp->sd_trunc_lock);
1382 gfs2_glock_finish_truncate(ip);
1386 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1387 if (!sdp->sd_statfs_force_sync) {
1388 sdp->sd_statfs_force_sync = 1;
1389 wake_up(&sdp->sd_quota_wait);
1395 * gfs2_quotad - Write cached quota changes into the quota file
1396 * @sdp: Pointer to GFS2 superblock
1400 int gfs2_quotad(void *data)
1402 struct gfs2_sbd *sdp = data;
1403 struct gfs2_tune *tune = &sdp->sd_tune;
1404 unsigned long statfs_timeo = 0;
1405 unsigned long quotad_timeo = 0;
1406 unsigned long t = 0;
1410 while (!kthread_should_stop()) {
1412 /* Update the master statfs file */
1413 if (sdp->sd_statfs_force_sync) {
1414 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1415 quotad_error(sdp, "statfs", error);
1416 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1419 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1421 &tune->gt_statfs_quantum);
1423 /* Update quota file */
1424 quotad_check_timeo(sdp, "sync", gfs2_quota_sync_timeo, t,
1425 "ad_timeo, &tune->gt_quota_quantum);
1427 /* Check for & recover partially truncated inodes */
1428 quotad_check_trunc_list(sdp);
1430 if (freezing(current))
1432 t = min(quotad_timeo, statfs_timeo);
1434 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1435 spin_lock(&sdp->sd_trunc_lock);
1436 empty = list_empty(&sdp->sd_trunc_list);
1437 spin_unlock(&sdp->sd_trunc_lock);
1438 if (empty && !sdp->sd_statfs_force_sync)
1439 t -= schedule_timeout(t);
1442 finish_wait(&sdp->sd_quota_wait, &wait);
1448 static int gfs2_quota_get_xstate(struct super_block *sb,
1449 struct fs_quota_stat *fqs)
1451 struct gfs2_sbd *sdp = sb->s_fs_info;
1453 memset(fqs, 0, sizeof(struct fs_quota_stat));
1454 fqs->qs_version = FS_QSTAT_VERSION;
1456 switch (sdp->sd_args.ar_quota) {
1458 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1460 case GFS2_QUOTA_ACCOUNT:
1461 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1463 case GFS2_QUOTA_OFF:
1467 if (sdp->sd_quota_inode) {
1468 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1469 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1471 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1472 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1473 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1477 static int gfs2_get_dqblk(struct super_block *sb, int type, qid_t id,
1478 struct fs_disk_quota *fdq)
1480 struct gfs2_sbd *sdp = sb->s_fs_info;
1481 struct gfs2_quota_lvb *qlvb;
1482 struct gfs2_quota_data *qd;
1483 struct gfs2_holder q_gh;
1486 memset(fdq, 0, sizeof(struct fs_disk_quota));
1488 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1489 return -ESRCH; /* Crazy XFS error code */
1491 if (type == USRQUOTA)
1493 else if (type == GRPQUOTA)
1498 error = qd_get(sdp, type, id, &qd);
1501 error = do_glock(qd, FORCE, &q_gh);
1505 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
1506 fdq->d_version = FS_DQUOT_VERSION;
1507 fdq->d_flags = (type == QUOTA_USER) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1509 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1510 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1511 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1513 gfs2_glock_dq_uninit(&q_gh);
1519 /* GFS2 only supports a subset of the XFS fields */
1520 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1522 static int gfs2_set_dqblk(struct super_block *sb, int type, qid_t id,
1523 struct fs_disk_quota *fdq)
1525 struct gfs2_sbd *sdp = sb->s_fs_info;
1526 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1527 struct gfs2_quota_data *qd;
1528 struct gfs2_holder q_gh, i_gh;
1529 unsigned int data_blocks, ind_blocks;
1530 unsigned int blocks = 0;
1532 struct gfs2_alloc *al;
1536 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1537 return -ESRCH; /* Crazy XFS error code */
1542 if (fdq->d_flags != FS_USER_QUOTA)
1547 if (fdq->d_flags != FS_GROUP_QUOTA)
1554 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1556 if (fdq->d_id != id)
1559 error = qd_get(sdp, type, id, &qd);
1563 mutex_lock(&ip->i_inode.i_mutex);
1564 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1567 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1571 /* Check for existing entry, if none then alloc new blocks */
1572 error = update_qd(sdp, qd);
1576 /* If nothing has changed, this is a no-op */
1577 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1578 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1579 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1581 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1582 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1583 fdq->d_fieldmask ^= FS_DQ_BHARD;
1585 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1586 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1587 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1589 if (fdq->d_fieldmask == 0)
1592 offset = qd2offset(qd);
1593 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1594 if (gfs2_is_stuffed(ip))
1596 if (alloc_required) {
1597 al = gfs2_alloc_get(ip);
1600 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1601 &data_blocks, &ind_blocks);
1602 blocks = al->al_requested = 1 + data_blocks + ind_blocks;
1603 error = gfs2_inplace_reserve(ip);
1606 blocks += gfs2_rg_blocks(ip);
1609 /* Some quotas span block boundaries and can update two blocks,
1610 adding an extra block to the transaction to handle such quotas */
1611 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1616 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1618 gfs2_trans_end(sdp);
1620 if (alloc_required) {
1621 gfs2_inplace_release(ip);
1626 gfs2_glock_dq_uninit(&i_gh);
1628 gfs2_glock_dq_uninit(&q_gh);
1630 mutex_unlock(&ip->i_inode.i_mutex);
1635 const struct quotactl_ops gfs2_quotactl_ops = {
1636 .quota_sync = gfs2_quota_sync,
1637 .get_xstate = gfs2_quota_get_xstate,
1638 .get_dqblk = gfs2_get_dqblk,
1639 .set_dqblk = gfs2_set_dqblk,