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>
53 #include <linux/lockref.h>
69 struct gfs2_quota_change_host {
71 u32 qc_flags; /* GFS2_QCF_... */
75 static LIST_HEAD(qd_lru_list);
76 static atomic_t qd_lru_count = ATOMIC_INIT(0);
77 static DEFINE_SPINLOCK(qd_lru_lock);
79 unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
80 struct shrink_control *sc)
82 struct gfs2_quota_data *qd;
84 int nr_to_scan = sc->nr_to_scan;
87 if (!(sc->gfp_mask & __GFP_FS))
90 spin_lock(&qd_lru_lock);
91 while (nr_to_scan && !list_empty(&qd_lru_list)) {
92 qd = list_entry(qd_lru_list.next,
93 struct gfs2_quota_data, qd_reclaim);
94 sdp = qd->qd_gl->gl_sbd;
96 /* Free from the filesystem-specific list */
97 list_del(&qd->qd_list);
99 gfs2_assert_warn(sdp, !qd->qd_change);
100 gfs2_assert_warn(sdp, !qd->qd_slot_count);
101 gfs2_assert_warn(sdp, !qd->qd_bh_count);
103 gfs2_glock_put(qd->qd_gl);
104 atomic_dec(&sdp->sd_quota_count);
106 /* Delete it from the common reclaim list */
107 list_del_init(&qd->qd_reclaim);
108 atomic_dec(&qd_lru_count);
109 spin_unlock(&qd_lru_lock);
110 kmem_cache_free(gfs2_quotad_cachep, qd);
111 spin_lock(&qd_lru_lock);
115 spin_unlock(&qd_lru_lock);
119 unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
120 struct shrink_control *sc)
122 return vfs_pressure_ratio(atomic_read(&qd_lru_count));
125 static u64 qd2index(struct gfs2_quota_data *qd)
127 struct kqid qid = qd->qd_id;
128 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
129 ((qid.type == USRQUOTA) ? 0 : 1);
132 static u64 qd2offset(struct gfs2_quota_data *qd)
136 offset = qd2index(qd);
137 offset *= sizeof(struct gfs2_quota);
142 static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
143 struct gfs2_quota_data **qdp)
145 struct gfs2_quota_data *qd;
148 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
152 qd->qd_lockref.count = 1;
153 spin_lock_init(&qd->qd_lockref.lock);
156 INIT_LIST_HEAD(&qd->qd_reclaim);
158 error = gfs2_glock_get(sdp, qd2index(qd),
159 &gfs2_quota_glops, CREATE, &qd->qd_gl);
168 kmem_cache_free(gfs2_quotad_cachep, qd);
172 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
173 struct gfs2_quota_data **qdp)
175 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
182 spin_lock(&qd_lru_lock);
183 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
184 if (qid_eq(qd->qd_id, qid)) {
185 lockref_get(&qd->qd_lockref);
186 if (!list_empty(&qd->qd_reclaim)) {
187 /* Remove it from reclaim list */
188 list_del_init(&qd->qd_reclaim);
189 atomic_dec(&qd_lru_count);
201 list_add(&qd->qd_list, &sdp->sd_quota_list);
202 atomic_inc(&sdp->sd_quota_count);
206 spin_unlock(&qd_lru_lock);
210 gfs2_glock_put(new_qd->qd_gl);
211 kmem_cache_free(gfs2_quotad_cachep, new_qd);
217 error = qd_alloc(sdp, qid, &new_qd);
223 static void qd_hold(struct gfs2_quota_data *qd)
225 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
226 gfs2_assert(sdp, !__lockref_is_dead(&qd->qd_lockref));
227 lockref_get(&qd->qd_lockref);
230 static void qd_put(struct gfs2_quota_data *qd)
232 spin_lock(&qd_lru_lock);
234 if (!lockref_put_or_lock(&qd->qd_lockref)) {
236 /* Add to the reclaim list */
237 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
238 atomic_inc(&qd_lru_count);
240 spin_unlock(&qd->qd_lockref.lock);
243 spin_unlock(&qd_lru_lock);
246 static int slot_get(struct gfs2_quota_data *qd)
248 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
249 unsigned int c, o = 0, b;
250 unsigned char byte = 0;
252 spin_lock(&qd_lru_lock);
254 if (qd->qd_slot_count++) {
255 spin_unlock(&qd_lru_lock);
259 for (c = 0; c < sdp->sd_quota_chunks; c++)
260 for (o = 0; o < PAGE_SIZE; o++) {
261 byte = sdp->sd_quota_bitmap[c][o];
269 for (b = 0; b < 8; b++)
270 if (!(byte & (1 << b)))
272 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
274 if (qd->qd_slot >= sdp->sd_quota_slots)
277 sdp->sd_quota_bitmap[c][o] |= 1 << b;
279 spin_unlock(&qd_lru_lock);
285 spin_unlock(&qd_lru_lock);
289 static void slot_hold(struct gfs2_quota_data *qd)
291 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
293 spin_lock(&qd_lru_lock);
294 gfs2_assert(sdp, qd->qd_slot_count);
296 spin_unlock(&qd_lru_lock);
299 static void gfs2_icbit_munge(struct gfs2_sbd *sdp, unsigned char **bitmap,
300 unsigned int bit, int new_value)
302 unsigned int c, o, b = bit;
305 c = b / (8 * PAGE_SIZE);
310 old_value = (bitmap[c][o] & (1 << b));
311 gfs2_assert_withdraw(sdp, !old_value != !new_value);
314 bitmap[c][o] |= 1 << b;
316 bitmap[c][o] &= ~(1 << b);
319 static void slot_put(struct gfs2_quota_data *qd)
321 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
323 spin_lock(&qd_lru_lock);
324 gfs2_assert(sdp, qd->qd_slot_count);
325 if (!--qd->qd_slot_count) {
326 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
329 spin_unlock(&qd_lru_lock);
332 static int bh_get(struct gfs2_quota_data *qd)
334 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
335 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
336 unsigned int block, offset;
337 struct buffer_head *bh;
339 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
341 mutex_lock(&sdp->sd_quota_mutex);
343 if (qd->qd_bh_count++) {
344 mutex_unlock(&sdp->sd_quota_mutex);
348 block = qd->qd_slot / sdp->sd_qc_per_block;
349 offset = qd->qd_slot % sdp->sd_qc_per_block;
351 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
352 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
355 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
359 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
363 qd->qd_bh_qc = (struct gfs2_quota_change *)
364 (bh->b_data + sizeof(struct gfs2_meta_header) +
365 offset * sizeof(struct gfs2_quota_change));
367 mutex_unlock(&sdp->sd_quota_mutex);
375 mutex_unlock(&sdp->sd_quota_mutex);
379 static void bh_put(struct gfs2_quota_data *qd)
381 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
383 mutex_lock(&sdp->sd_quota_mutex);
384 gfs2_assert(sdp, qd->qd_bh_count);
385 if (!--qd->qd_bh_count) {
390 mutex_unlock(&sdp->sd_quota_mutex);
393 static int qd_check_sync(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd,
396 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
397 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
398 (sync_gen && (qd->qd_sync_gen >= *sync_gen)))
401 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
403 set_bit(QDF_LOCKED, &qd->qd_flags);
404 gfs2_assert_warn(sdp, !__lockref_is_dead(&qd->qd_lockref));
405 lockref_get(&qd->qd_lockref);
406 qd->qd_change_sync = qd->qd_change;
407 gfs2_assert_warn(sdp, qd->qd_slot_count);
412 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
414 struct gfs2_quota_data *qd = NULL;
420 if (sdp->sd_vfs->s_flags & MS_RDONLY)
423 spin_lock(&qd_lru_lock);
425 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
426 found = qd_check_sync(sdp, qd, &sdp->sd_quota_sync_gen);
434 spin_unlock(&qd_lru_lock);
437 gfs2_assert_warn(sdp, qd->qd_change_sync);
440 clear_bit(QDF_LOCKED, &qd->qd_flags);
452 static void qd_unlock(struct gfs2_quota_data *qd)
454 gfs2_assert_warn(qd->qd_gl->gl_sbd,
455 test_bit(QDF_LOCKED, &qd->qd_flags));
456 clear_bit(QDF_LOCKED, &qd->qd_flags);
462 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
463 struct gfs2_quota_data **qdp)
467 error = qd_get(sdp, qid, qdp);
471 error = slot_get(*qdp);
475 error = bh_get(*qdp);
488 static void qdsb_put(struct gfs2_quota_data *qd)
495 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
497 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
498 struct gfs2_quota_data **qd;
501 if (ip->i_res == NULL) {
502 error = gfs2_rs_alloc(ip);
507 qd = ip->i_res->rs_qa_qd;
509 if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
510 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
513 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
516 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
519 ip->i_res->rs_qa_qd_num++;
522 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
525 ip->i_res->rs_qa_qd_num++;
528 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
529 !uid_eq(uid, ip->i_inode.i_uid)) {
530 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
533 ip->i_res->rs_qa_qd_num++;
537 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
538 !gid_eq(gid, ip->i_inode.i_gid)) {
539 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
542 ip->i_res->rs_qa_qd_num++;
548 gfs2_quota_unhold(ip);
552 void gfs2_quota_unhold(struct gfs2_inode *ip)
554 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
557 if (ip->i_res == NULL)
559 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
561 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
562 qdsb_put(ip->i_res->rs_qa_qd[x]);
563 ip->i_res->rs_qa_qd[x] = NULL;
565 ip->i_res->rs_qa_qd_num = 0;
568 static int sort_qd(const void *a, const void *b)
570 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
571 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
573 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
575 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
580 static void do_qc(struct gfs2_quota_data *qd, s64 change)
582 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
583 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
584 struct gfs2_quota_change *qc = qd->qd_bh_qc;
587 mutex_lock(&sdp->sd_quota_mutex);
588 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
590 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
593 if (qd->qd_id.type == USRQUOTA)
594 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
595 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
598 x = be64_to_cpu(qc->qc_change) + change;
599 qc->qc_change = cpu_to_be64(x);
601 spin_lock(&qd_lru_lock);
603 spin_unlock(&qd_lru_lock);
606 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
607 clear_bit(QDF_CHANGE, &qd->qd_flags);
612 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
617 mutex_unlock(&sdp->sd_quota_mutex);
621 * gfs2_adjust_quota - adjust record of current block usage
622 * @ip: The quota inode
623 * @loc: Offset of the entry in the quota file
624 * @change: The amount of usage change to record
625 * @qd: The quota data
626 * @fdq: The updated limits to record
628 * This function was mostly borrowed from gfs2_block_truncate_page which was
629 * in turn mostly borrowed from ext3
631 * Returns: 0 or -ve on error
634 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
635 s64 change, struct gfs2_quota_data *qd,
636 struct fs_disk_quota *fdq)
638 struct inode *inode = &ip->i_inode;
639 struct gfs2_sbd *sdp = GFS2_SB(inode);
640 struct address_space *mapping = inode->i_mapping;
641 unsigned long index = loc >> PAGE_CACHE_SHIFT;
642 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
643 unsigned blocksize, iblock, pos;
644 struct buffer_head *bh;
647 struct gfs2_quota q, *qp;
651 if (gfs2_is_stuffed(ip)) {
652 err = gfs2_unstuff_dinode(ip, NULL);
657 memset(&q, 0, sizeof(struct gfs2_quota));
658 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
664 qp->qu_value = be64_to_cpu(qp->qu_value);
665 qp->qu_value += change;
666 qp->qu_value = cpu_to_be64(qp->qu_value);
667 qd->qd_qb.qb_value = qp->qu_value;
669 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
670 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
671 qd->qd_qb.qb_warn = qp->qu_warn;
673 if (fdq->d_fieldmask & FS_DQ_BHARD) {
674 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
675 qd->qd_qb.qb_limit = qp->qu_limit;
677 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
678 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
679 qd->qd_qb.qb_value = qp->qu_value;
683 /* Write the quota into the quota file on disk */
685 nbytes = sizeof(struct gfs2_quota);
687 page = find_or_create_page(mapping, index, GFP_NOFS);
691 blocksize = inode->i_sb->s_blocksize;
692 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
694 if (!page_has_buffers(page))
695 create_empty_buffers(page, blocksize, 0);
697 bh = page_buffers(page);
699 while (offset >= pos) {
700 bh = bh->b_this_page;
705 if (!buffer_mapped(bh)) {
706 gfs2_block_map(inode, iblock, bh, 1);
707 if (!buffer_mapped(bh))
709 /* If it's a newly allocated disk block for quota, zero it */
711 zero_user(page, pos - blocksize, bh->b_size);
714 if (PageUptodate(page))
715 set_buffer_uptodate(bh);
717 if (!buffer_uptodate(bh)) {
718 ll_rw_block(READ | REQ_META, 1, &bh);
720 if (!buffer_uptodate(bh))
724 gfs2_trans_add_data(ip->i_gl, bh);
726 kaddr = kmap_atomic(page);
727 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
728 nbytes = PAGE_CACHE_SIZE - offset;
729 memcpy(kaddr + offset, ptr, nbytes);
730 flush_dcache_page(page);
731 kunmap_atomic(kaddr);
733 page_cache_release(page);
735 /* If quota straddles page boundary, we need to update the rest of the
736 * quota at the beginning of the next page */
737 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
739 nbytes = sizeof(struct gfs2_quota) - nbytes;
745 size = loc + sizeof(struct gfs2_quota);
746 if (size > inode->i_size)
747 i_size_write(inode, size);
748 inode->i_mtime = inode->i_atime = CURRENT_TIME;
749 mark_inode_dirty(inode);
754 page_cache_release(page);
758 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
760 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
761 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
762 struct gfs2_alloc_parms ap = { .aflags = 0, };
763 unsigned int data_blocks, ind_blocks;
764 struct gfs2_holder *ghs, i_gh;
766 struct gfs2_quota_data *qd;
769 unsigned int nalloc = 0, blocks;
772 error = gfs2_rs_alloc(ip);
776 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
777 &data_blocks, &ind_blocks);
779 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
783 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
784 mutex_lock(&ip->i_inode.i_mutex);
785 for (qx = 0; qx < num_qd; qx++) {
786 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
787 GL_NOCACHE, &ghs[qx]);
792 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
796 for (x = 0; x < num_qd; x++) {
797 offset = qd2offset(qda[x]);
798 if (gfs2_write_alloc_required(ip, offset,
799 sizeof(struct gfs2_quota)))
804 * 1 blk for unstuffing inode if stuffed. We add this extra
805 * block to the reservation unconditionally. If the inode
806 * doesn't need unstuffing, the block will be released to the
807 * rgrp since it won't be allocated during the transaction
809 /* +3 in the end for unstuffing block, inode size update block
810 * and another block in case quota straddles page boundary and
811 * two blocks need to be updated instead of 1 */
812 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
814 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
815 ap.target = reserved;
816 error = gfs2_inplace_reserve(ip, &ap);
821 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
823 error = gfs2_trans_begin(sdp, blocks, 0);
827 for (x = 0; x < num_qd; x++) {
829 offset = qd2offset(qd);
830 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
834 do_qc(qd, -qd->qd_change_sync);
835 set_bit(QDF_REFRESH, &qd->qd_flags);
843 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, (char *)&q, &pos, sizeof(q));
869 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
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_lksb.sb_lvbptr;
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, kuid_t uid, kgid_t gid)
927 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
928 struct gfs2_quota_data *qd;
932 error = gfs2_quota_hold(ip, uid, gid);
936 if (capable(CAP_SYS_RESOURCE) ||
937 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
940 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
941 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
943 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
944 int force = NO_FORCE;
945 qd = ip->i_res->rs_qa_qd[x];
946 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
948 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
954 set_bit(GIF_QD_LOCKED, &ip->i_flags);
957 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
958 gfs2_quota_unhold(ip);
964 static int need_sync(struct gfs2_quota_data *qd)
966 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
967 struct gfs2_tune *gt = &sdp->sd_tune;
969 unsigned int num, den;
972 if (!qd->qd_qb.qb_limit)
975 spin_lock(&qd_lru_lock);
976 value = qd->qd_change;
977 spin_unlock(&qd_lru_lock);
979 spin_lock(>->gt_spin);
980 num = gt->gt_quota_scale_num;
981 den = gt->gt_quota_scale_den;
982 spin_unlock(>->gt_spin);
986 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
987 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
990 value *= gfs2_jindex_size(sdp) * num;
991 value = div_s64(value, den);
992 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
993 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1000 void gfs2_quota_unlock(struct gfs2_inode *ip)
1002 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1003 struct gfs2_quota_data *qda[4];
1004 unsigned int count = 0;
1008 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1011 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1012 struct gfs2_quota_data *qd;
1015 qd = ip->i_res->rs_qa_qd[x];
1016 sync = need_sync(qd);
1018 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1022 spin_lock(&qd_lru_lock);
1023 found = qd_check_sync(sdp, qd, NULL);
1024 spin_unlock(&qd_lru_lock);
1029 gfs2_assert_warn(sdp, qd->qd_change_sync);
1031 clear_bit(QDF_LOCKED, &qd->qd_flags);
1041 do_sync(count, qda);
1042 for (x = 0; x < count; x++)
1047 gfs2_quota_unhold(ip);
1050 #define MAX_LINE 256
1052 static int print_message(struct gfs2_quota_data *qd, char *type)
1054 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1056 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1057 sdp->sd_fsname, type,
1058 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1059 from_kqid(&init_user_ns, qd->qd_id));
1064 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1066 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1067 struct gfs2_quota_data *qd;
1072 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1075 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1078 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1079 qd = ip->i_res->rs_qa_qd[x];
1081 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1082 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1085 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1086 spin_lock(&qd_lru_lock);
1087 value += qd->qd_change;
1088 spin_unlock(&qd_lru_lock);
1090 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1091 print_message(qd, "exceeded");
1092 quota_send_warning(qd->qd_id,
1093 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1097 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1098 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1099 time_after_eq(jiffies, qd->qd_last_warn +
1101 gt_quota_warn_period) * HZ)) {
1102 quota_send_warning(qd->qd_id,
1103 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1104 error = print_message(qd, "warning");
1105 qd->qd_last_warn = jiffies;
1112 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1113 kuid_t uid, kgid_t gid)
1115 struct gfs2_quota_data *qd;
1118 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1120 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1123 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1124 qd = ip->i_res->rs_qa_qd[x];
1126 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1127 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1133 int gfs2_quota_sync(struct super_block *sb, int type)
1135 struct gfs2_sbd *sdp = sb->s_fs_info;
1136 struct gfs2_quota_data **qda;
1137 unsigned int max_qd = PAGE_SIZE/sizeof(struct gfs2_holder);
1138 unsigned int num_qd;
1142 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1146 mutex_lock(&sdp->sd_quota_sync_mutex);
1147 sdp->sd_quota_sync_gen++;
1153 error = qd_fish(sdp, qda + num_qd);
1154 if (error || !qda[num_qd])
1156 if (++num_qd == max_qd)
1162 error = do_sync(num_qd, qda);
1164 for (x = 0; x < num_qd; x++)
1165 qda[x]->qd_sync_gen =
1166 sdp->sd_quota_sync_gen;
1168 for (x = 0; x < num_qd; x++)
1171 } while (!error && num_qd == max_qd);
1173 mutex_unlock(&sdp->sd_quota_sync_mutex);
1179 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1181 struct gfs2_quota_data *qd;
1182 struct gfs2_holder q_gh;
1185 error = qd_get(sdp, qid, &qd);
1189 error = do_glock(qd, FORCE, &q_gh);
1191 gfs2_glock_dq_uninit(&q_gh);
1197 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1199 const struct gfs2_quota_change *str = buf;
1201 qc->qc_change = be64_to_cpu(str->qc_change);
1202 qc->qc_flags = be32_to_cpu(str->qc_flags);
1203 qc->qc_id = make_kqid(&init_user_ns,
1204 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1205 be32_to_cpu(str->qc_id));
1208 int gfs2_quota_init(struct gfs2_sbd *sdp)
1210 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1211 u64 size = i_size_read(sdp->sd_qc_inode);
1212 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1213 unsigned int x, slot = 0;
1214 unsigned int found = 0;
1219 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1222 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1223 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1227 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1228 sizeof(unsigned char *), GFP_NOFS);
1229 if (!sdp->sd_quota_bitmap)
1232 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1233 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1234 if (!sdp->sd_quota_bitmap[x])
1238 for (x = 0; x < blocks; x++) {
1239 struct buffer_head *bh;
1244 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1249 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1252 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1257 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1259 struct gfs2_quota_change_host qc;
1260 struct gfs2_quota_data *qd;
1262 gfs2_quota_change_in(&qc, bh->b_data +
1263 sizeof(struct gfs2_meta_header) +
1264 y * sizeof(struct gfs2_quota_change));
1268 error = qd_alloc(sdp, qc.qc_id, &qd);
1274 set_bit(QDF_CHANGE, &qd->qd_flags);
1275 qd->qd_change = qc.qc_change;
1277 qd->qd_slot_count = 1;
1279 spin_lock(&qd_lru_lock);
1280 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1281 list_add(&qd->qd_list, &sdp->sd_quota_list);
1282 atomic_inc(&sdp->sd_quota_count);
1283 spin_unlock(&qd_lru_lock);
1294 fs_info(sdp, "found %u quota changes\n", found);
1299 gfs2_quota_cleanup(sdp);
1303 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1305 struct list_head *head = &sdp->sd_quota_list;
1306 struct gfs2_quota_data *qd;
1309 spin_lock(&qd_lru_lock);
1310 while (!list_empty(head)) {
1311 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1314 * To be removed in due course... we should be able to
1315 * ensure that all refs to the qd have done by this point
1316 * so that this rather odd test is not required
1318 spin_lock(&qd->qd_lockref.lock);
1319 if (qd->qd_lockref.count > 1 ||
1320 (qd->qd_lockref.count && !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1321 spin_unlock(&qd->qd_lockref.lock);
1322 list_move(&qd->qd_list, head);
1323 spin_unlock(&qd_lru_lock);
1325 spin_lock(&qd_lru_lock);
1328 spin_unlock(&qd->qd_lockref.lock);
1330 list_del(&qd->qd_list);
1331 /* Also remove if this qd exists in the reclaim list */
1332 if (!list_empty(&qd->qd_reclaim)) {
1333 list_del_init(&qd->qd_reclaim);
1334 atomic_dec(&qd_lru_count);
1336 atomic_dec(&sdp->sd_quota_count);
1337 spin_unlock(&qd_lru_lock);
1339 if (!qd->qd_lockref.count) {
1340 gfs2_assert_warn(sdp, !qd->qd_change);
1341 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1343 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1344 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1346 gfs2_glock_put(qd->qd_gl);
1347 kmem_cache_free(gfs2_quotad_cachep, qd);
1349 spin_lock(&qd_lru_lock);
1351 spin_unlock(&qd_lru_lock);
1353 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1355 if (sdp->sd_quota_bitmap) {
1356 for (x = 0; x < sdp->sd_quota_chunks; x++)
1357 kfree(sdp->sd_quota_bitmap[x]);
1358 kfree(sdp->sd_quota_bitmap);
1362 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1364 if (error == 0 || error == -EROFS)
1366 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1367 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1370 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1371 int (*fxn)(struct super_block *sb, int type),
1372 unsigned long t, unsigned long *timeo,
1373 unsigned int *new_timeo)
1376 int error = fxn(sdp->sd_vfs, 0);
1377 quotad_error(sdp, msg, error);
1378 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1384 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1386 struct gfs2_inode *ip;
1390 spin_lock(&sdp->sd_trunc_lock);
1391 if (!list_empty(&sdp->sd_trunc_list)) {
1392 ip = list_entry(sdp->sd_trunc_list.next,
1393 struct gfs2_inode, i_trunc_list);
1394 list_del_init(&ip->i_trunc_list);
1396 spin_unlock(&sdp->sd_trunc_lock);
1399 gfs2_glock_finish_truncate(ip);
1403 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1404 if (!sdp->sd_statfs_force_sync) {
1405 sdp->sd_statfs_force_sync = 1;
1406 wake_up(&sdp->sd_quota_wait);
1412 * gfs2_quotad - Write cached quota changes into the quota file
1413 * @sdp: Pointer to GFS2 superblock
1417 int gfs2_quotad(void *data)
1419 struct gfs2_sbd *sdp = data;
1420 struct gfs2_tune *tune = &sdp->sd_tune;
1421 unsigned long statfs_timeo = 0;
1422 unsigned long quotad_timeo = 0;
1423 unsigned long t = 0;
1427 while (!kthread_should_stop()) {
1429 /* Update the master statfs file */
1430 if (sdp->sd_statfs_force_sync) {
1431 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1432 quotad_error(sdp, "statfs", error);
1433 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1436 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1438 &tune->gt_statfs_quantum);
1440 /* Update quota file */
1441 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1442 "ad_timeo, &tune->gt_quota_quantum);
1444 /* Check for & recover partially truncated inodes */
1445 quotad_check_trunc_list(sdp);
1449 t = min(quotad_timeo, statfs_timeo);
1451 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1452 spin_lock(&sdp->sd_trunc_lock);
1453 empty = list_empty(&sdp->sd_trunc_list);
1454 spin_unlock(&sdp->sd_trunc_lock);
1455 if (empty && !sdp->sd_statfs_force_sync)
1456 t -= schedule_timeout(t);
1459 finish_wait(&sdp->sd_quota_wait, &wait);
1465 static int gfs2_quota_get_xstate(struct super_block *sb,
1466 struct fs_quota_stat *fqs)
1468 struct gfs2_sbd *sdp = sb->s_fs_info;
1470 memset(fqs, 0, sizeof(struct fs_quota_stat));
1471 fqs->qs_version = FS_QSTAT_VERSION;
1473 switch (sdp->sd_args.ar_quota) {
1475 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1477 case GFS2_QUOTA_ACCOUNT:
1478 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1480 case GFS2_QUOTA_OFF:
1484 if (sdp->sd_quota_inode) {
1485 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1486 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1488 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1489 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1490 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1494 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1495 struct fs_disk_quota *fdq)
1497 struct gfs2_sbd *sdp = sb->s_fs_info;
1498 struct gfs2_quota_lvb *qlvb;
1499 struct gfs2_quota_data *qd;
1500 struct gfs2_holder q_gh;
1503 memset(fdq, 0, sizeof(struct fs_disk_quota));
1505 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1506 return -ESRCH; /* Crazy XFS error code */
1508 if ((qid.type != USRQUOTA) &&
1509 (qid.type != GRPQUOTA))
1512 error = qd_get(sdp, qid, &qd);
1515 error = do_glock(qd, FORCE, &q_gh);
1519 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1520 fdq->d_version = FS_DQUOT_VERSION;
1521 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1522 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1523 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1524 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1525 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1527 gfs2_glock_dq_uninit(&q_gh);
1533 /* GFS2 only supports a subset of the XFS fields */
1534 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1536 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1537 struct fs_disk_quota *fdq)
1539 struct gfs2_sbd *sdp = sb->s_fs_info;
1540 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1541 struct gfs2_quota_data *qd;
1542 struct gfs2_holder q_gh, i_gh;
1543 unsigned int data_blocks, ind_blocks;
1544 unsigned int blocks = 0;
1549 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1550 return -ESRCH; /* Crazy XFS error code */
1552 if ((qid.type != USRQUOTA) &&
1553 (qid.type != GRPQUOTA))
1556 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1559 error = qd_get(sdp, qid, &qd);
1563 error = gfs2_rs_alloc(ip);
1567 mutex_lock(&ip->i_inode.i_mutex);
1568 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1571 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1575 /* Check for existing entry, if none then alloc new blocks */
1576 error = update_qd(sdp, qd);
1580 /* If nothing has changed, this is a no-op */
1581 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1582 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1583 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1585 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1586 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1587 fdq->d_fieldmask ^= FS_DQ_BHARD;
1589 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1590 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1591 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1593 if (fdq->d_fieldmask == 0)
1596 offset = qd2offset(qd);
1597 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1598 if (gfs2_is_stuffed(ip))
1600 if (alloc_required) {
1601 struct gfs2_alloc_parms ap = { .aflags = 0, };
1602 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1603 &data_blocks, &ind_blocks);
1604 blocks = 1 + data_blocks + ind_blocks;
1606 error = gfs2_inplace_reserve(ip, &ap);
1609 blocks += gfs2_rg_blocks(ip, blocks);
1612 /* Some quotas span block boundaries and can update two blocks,
1613 adding an extra block to the transaction to handle such quotas */
1614 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1619 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1621 gfs2_trans_end(sdp);
1624 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);
1636 const struct quotactl_ops gfs2_quotactl_ops = {
1637 .quota_sync = gfs2_quota_sync,
1638 .get_xstate = gfs2_quota_get_xstate,
1639 .get_dqblk = gfs2_get_dqblk,
1640 .set_dqblk = gfs2_set_dqblk,