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>
68 struct gfs2_quota_change_host {
70 u32 qc_flags; /* GFS2_QCF_... */
74 static LIST_HEAD(qd_lru_list);
75 static atomic_t qd_lru_count = ATOMIC_INIT(0);
76 static DEFINE_SPINLOCK(qd_lru_lock);
78 int gfs2_shrink_qd_memory(struct shrinker *shrink, struct shrink_control *sc)
80 struct gfs2_quota_data *qd;
82 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);
114 spin_unlock(&qd_lru_lock);
117 return (atomic_read(&qd_lru_count) * sysctl_vfs_cache_pressure) / 100;
120 static u64 qd2index(struct gfs2_quota_data *qd)
122 struct kqid qid = qd->qd_id;
123 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
124 (qid.type == USRQUOTA) ? 0 : 1;
127 static u64 qd2offset(struct gfs2_quota_data *qd)
131 offset = qd2index(qd);
132 offset *= sizeof(struct gfs2_quota);
137 static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
138 struct gfs2_quota_data **qdp)
140 struct gfs2_quota_data *qd;
143 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
147 atomic_set(&qd->qd_count, 1);
150 INIT_LIST_HEAD(&qd->qd_reclaim);
152 error = gfs2_glock_get(sdp, qd2index(qd),
153 &gfs2_quota_glops, CREATE, &qd->qd_gl);
162 kmem_cache_free(gfs2_quotad_cachep, qd);
166 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
167 struct gfs2_quota_data **qdp)
169 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
176 spin_lock(&qd_lru_lock);
177 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
178 if (qid_eq(qd->qd_id, qid)) {
179 if (!atomic_read(&qd->qd_count) &&
180 !list_empty(&qd->qd_reclaim)) {
181 /* Remove it from reclaim list */
182 list_del_init(&qd->qd_reclaim);
183 atomic_dec(&qd_lru_count);
185 atomic_inc(&qd->qd_count);
196 list_add(&qd->qd_list, &sdp->sd_quota_list);
197 atomic_inc(&sdp->sd_quota_count);
201 spin_unlock(&qd_lru_lock);
205 gfs2_glock_put(new_qd->qd_gl);
206 kmem_cache_free(gfs2_quotad_cachep, new_qd);
212 error = qd_alloc(sdp, qid, &new_qd);
218 static void qd_hold(struct gfs2_quota_data *qd)
220 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
221 gfs2_assert(sdp, atomic_read(&qd->qd_count));
222 atomic_inc(&qd->qd_count);
225 static void qd_put(struct gfs2_quota_data *qd)
227 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
228 /* Add to the reclaim list */
229 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
230 atomic_inc(&qd_lru_count);
231 spin_unlock(&qd_lru_lock);
235 static int slot_get(struct gfs2_quota_data *qd)
237 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
238 unsigned int c, o = 0, b;
239 unsigned char byte = 0;
241 spin_lock(&qd_lru_lock);
243 if (qd->qd_slot_count++) {
244 spin_unlock(&qd_lru_lock);
248 for (c = 0; c < sdp->sd_quota_chunks; c++)
249 for (o = 0; o < PAGE_SIZE; o++) {
250 byte = sdp->sd_quota_bitmap[c][o];
258 for (b = 0; b < 8; b++)
259 if (!(byte & (1 << b)))
261 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
263 if (qd->qd_slot >= sdp->sd_quota_slots)
266 sdp->sd_quota_bitmap[c][o] |= 1 << b;
268 spin_unlock(&qd_lru_lock);
274 spin_unlock(&qd_lru_lock);
278 static void slot_hold(struct gfs2_quota_data *qd)
280 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
282 spin_lock(&qd_lru_lock);
283 gfs2_assert(sdp, qd->qd_slot_count);
285 spin_unlock(&qd_lru_lock);
288 static void slot_put(struct gfs2_quota_data *qd)
290 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
292 spin_lock(&qd_lru_lock);
293 gfs2_assert(sdp, qd->qd_slot_count);
294 if (!--qd->qd_slot_count) {
295 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
298 spin_unlock(&qd_lru_lock);
301 static int bh_get(struct gfs2_quota_data *qd)
303 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
304 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
305 unsigned int block, offset;
306 struct buffer_head *bh;
308 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
310 mutex_lock(&sdp->sd_quota_mutex);
312 if (qd->qd_bh_count++) {
313 mutex_unlock(&sdp->sd_quota_mutex);
317 block = qd->qd_slot / sdp->sd_qc_per_block;
318 offset = qd->qd_slot % sdp->sd_qc_per_block;
320 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
321 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
324 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
328 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
332 qd->qd_bh_qc = (struct gfs2_quota_change *)
333 (bh->b_data + sizeof(struct gfs2_meta_header) +
334 offset * sizeof(struct gfs2_quota_change));
336 mutex_unlock(&sdp->sd_quota_mutex);
344 mutex_unlock(&sdp->sd_quota_mutex);
348 static void bh_put(struct gfs2_quota_data *qd)
350 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
352 mutex_lock(&sdp->sd_quota_mutex);
353 gfs2_assert(sdp, qd->qd_bh_count);
354 if (!--qd->qd_bh_count) {
359 mutex_unlock(&sdp->sd_quota_mutex);
362 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
364 struct gfs2_quota_data *qd = NULL;
370 if (sdp->sd_vfs->s_flags & MS_RDONLY)
373 spin_lock(&qd_lru_lock);
375 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
376 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
377 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
378 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
381 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
383 set_bit(QDF_LOCKED, &qd->qd_flags);
384 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
385 atomic_inc(&qd->qd_count);
386 qd->qd_change_sync = qd->qd_change;
387 gfs2_assert_warn(sdp, qd->qd_slot_count);
397 spin_unlock(&qd_lru_lock);
400 gfs2_assert_warn(sdp, qd->qd_change_sync);
403 clear_bit(QDF_LOCKED, &qd->qd_flags);
415 static int qd_trylock(struct gfs2_quota_data *qd)
417 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
419 if (sdp->sd_vfs->s_flags & MS_RDONLY)
422 spin_lock(&qd_lru_lock);
424 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
425 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
426 spin_unlock(&qd_lru_lock);
430 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
432 set_bit(QDF_LOCKED, &qd->qd_flags);
433 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
434 atomic_inc(&qd->qd_count);
435 qd->qd_change_sync = qd->qd_change;
436 gfs2_assert_warn(sdp, qd->qd_slot_count);
439 spin_unlock(&qd_lru_lock);
441 gfs2_assert_warn(sdp, qd->qd_change_sync);
443 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 != NO_UID_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
529 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
532 ip->i_res->rs_qa_qd_num++;
536 if (gid != NO_GID_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
537 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
540 ip->i_res->rs_qa_qd_num++;
546 gfs2_quota_unhold(ip);
550 void gfs2_quota_unhold(struct gfs2_inode *ip)
552 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
555 if (ip->i_res == NULL)
557 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
559 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
560 qdsb_put(ip->i_res->rs_qa_qd[x]);
561 ip->i_res->rs_qa_qd[x] = NULL;
563 ip->i_res->rs_qa_qd_num = 0;
566 static int sort_qd(const void *a, const void *b)
568 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
569 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
571 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
573 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
578 static void do_qc(struct gfs2_quota_data *qd, s64 change)
580 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
581 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
582 struct gfs2_quota_change *qc = qd->qd_bh_qc;
585 mutex_lock(&sdp->sd_quota_mutex);
586 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
588 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
591 if (qd->qd_id.type == USRQUOTA)
592 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
593 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
596 x = be64_to_cpu(qc->qc_change) + change;
597 qc->qc_change = cpu_to_be64(x);
599 spin_lock(&qd_lru_lock);
601 spin_unlock(&qd_lru_lock);
604 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
605 clear_bit(QDF_CHANGE, &qd->qd_flags);
610 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
615 mutex_unlock(&sdp->sd_quota_mutex);
619 * gfs2_adjust_quota - adjust record of current block usage
620 * @ip: The quota inode
621 * @loc: Offset of the entry in the quota file
622 * @change: The amount of usage change to record
623 * @qd: The quota data
624 * @fdq: The updated limits to record
626 * This function was mostly borrowed from gfs2_block_truncate_page which was
627 * in turn mostly borrowed from ext3
629 * Returns: 0 or -ve on error
632 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
633 s64 change, struct gfs2_quota_data *qd,
634 struct fs_disk_quota *fdq)
636 struct inode *inode = &ip->i_inode;
637 struct gfs2_sbd *sdp = GFS2_SB(inode);
638 struct address_space *mapping = inode->i_mapping;
639 unsigned long index = loc >> PAGE_CACHE_SHIFT;
640 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
641 unsigned blocksize, iblock, pos;
642 struct buffer_head *bh;
645 struct gfs2_quota q, *qp;
649 if (gfs2_is_stuffed(ip)) {
650 err = gfs2_unstuff_dinode(ip, NULL);
655 memset(&q, 0, sizeof(struct gfs2_quota));
656 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
662 qp->qu_value = be64_to_cpu(qp->qu_value);
663 qp->qu_value += change;
664 qp->qu_value = cpu_to_be64(qp->qu_value);
665 qd->qd_qb.qb_value = qp->qu_value;
667 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
668 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
669 qd->qd_qb.qb_warn = qp->qu_warn;
671 if (fdq->d_fieldmask & FS_DQ_BHARD) {
672 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
673 qd->qd_qb.qb_limit = qp->qu_limit;
675 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
676 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
677 qd->qd_qb.qb_value = qp->qu_value;
681 /* Write the quota into the quota file on disk */
683 nbytes = sizeof(struct gfs2_quota);
685 page = find_or_create_page(mapping, index, GFP_NOFS);
689 blocksize = inode->i_sb->s_blocksize;
690 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
692 if (!page_has_buffers(page))
693 create_empty_buffers(page, blocksize, 0);
695 bh = page_buffers(page);
697 while (offset >= pos) {
698 bh = bh->b_this_page;
703 if (!buffer_mapped(bh)) {
704 gfs2_block_map(inode, iblock, bh, 1);
705 if (!buffer_mapped(bh))
707 /* If it's a newly allocated disk block for quota, zero it */
709 zero_user(page, pos - blocksize, bh->b_size);
712 if (PageUptodate(page))
713 set_buffer_uptodate(bh);
715 if (!buffer_uptodate(bh)) {
716 ll_rw_block(READ | REQ_META, 1, &bh);
718 if (!buffer_uptodate(bh))
722 gfs2_trans_add_bh(ip->i_gl, bh, 0);
724 kaddr = kmap_atomic(page);
725 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
726 nbytes = PAGE_CACHE_SIZE - offset;
727 memcpy(kaddr + offset, ptr, nbytes);
728 flush_dcache_page(page);
729 kunmap_atomic(kaddr);
731 page_cache_release(page);
733 /* If quota straddles page boundary, we need to update the rest of the
734 * quota at the beginning of the next page */
735 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
737 nbytes = sizeof(struct gfs2_quota) - nbytes;
743 size = loc + sizeof(struct gfs2_quota);
744 if (size > inode->i_size)
745 i_size_write(inode, size);
746 inode->i_mtime = inode->i_atime = CURRENT_TIME;
747 mark_inode_dirty(inode);
752 page_cache_release(page);
756 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
758 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
759 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
760 unsigned int data_blocks, ind_blocks;
761 struct gfs2_holder *ghs, i_gh;
763 struct gfs2_quota_data *qd;
766 unsigned int nalloc = 0, blocks;
769 error = gfs2_rs_alloc(ip);
773 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
774 &data_blocks, &ind_blocks);
776 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
780 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
781 mutex_lock(&ip->i_inode.i_mutex);
782 for (qx = 0; qx < num_qd; qx++) {
783 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
784 GL_NOCACHE, &ghs[qx]);
789 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
793 for (x = 0; x < num_qd; x++) {
794 offset = qd2offset(qda[x]);
795 if (gfs2_write_alloc_required(ip, offset,
796 sizeof(struct gfs2_quota)))
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 /* +3 in the end for unstuffing block, inode size update block
807 * and another block in case quota straddles page boundary and
808 * two blocks need to be updated instead of 1 */
809 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
811 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
812 error = gfs2_inplace_reserve(ip, reserved, 0);
817 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
819 error = gfs2_trans_begin(sdp, blocks, 0);
823 for (x = 0; x < num_qd; x++) {
825 offset = qd2offset(qd);
826 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
830 do_qc(qd, -qd->qd_change_sync);
831 set_bit(QDF_REFRESH, &qd->qd_flags);
839 gfs2_inplace_release(ip);
841 gfs2_glock_dq_uninit(&i_gh);
844 gfs2_glock_dq_uninit(&ghs[qx]);
845 mutex_unlock(&ip->i_inode.i_mutex);
847 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
851 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
853 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
855 struct gfs2_quota_lvb *qlvb;
859 memset(&q, 0, sizeof(struct gfs2_quota));
861 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
865 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
866 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
868 qlvb->qb_limit = q.qu_limit;
869 qlvb->qb_warn = q.qu_warn;
870 qlvb->qb_value = q.qu_value;
876 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
877 struct gfs2_holder *q_gh)
879 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
880 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
881 struct gfs2_holder i_gh;
885 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
889 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
891 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
892 gfs2_glock_dq_uninit(q_gh);
893 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
898 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
902 error = update_qd(sdp, qd);
906 gfs2_glock_dq_uninit(&i_gh);
907 gfs2_glock_dq_uninit(q_gh);
915 gfs2_glock_dq_uninit(&i_gh);
917 gfs2_glock_dq_uninit(q_gh);
921 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
923 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
924 struct gfs2_quota_data *qd;
928 error = gfs2_quota_hold(ip, uid, gid);
932 if (capable(CAP_SYS_RESOURCE) ||
933 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
936 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
937 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
939 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
940 int force = NO_FORCE;
941 qd = ip->i_res->rs_qa_qd[x];
942 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
944 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
950 set_bit(GIF_QD_LOCKED, &ip->i_flags);
953 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
954 gfs2_quota_unhold(ip);
960 static int need_sync(struct gfs2_quota_data *qd)
962 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
963 struct gfs2_tune *gt = &sdp->sd_tune;
965 unsigned int num, den;
968 if (!qd->qd_qb.qb_limit)
971 spin_lock(&qd_lru_lock);
972 value = qd->qd_change;
973 spin_unlock(&qd_lru_lock);
975 spin_lock(>->gt_spin);
976 num = gt->gt_quota_scale_num;
977 den = gt->gt_quota_scale_den;
978 spin_unlock(>->gt_spin);
982 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
983 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
986 value *= gfs2_jindex_size(sdp) * num;
987 value = div_s64(value, den);
988 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
989 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
996 void gfs2_quota_unlock(struct gfs2_inode *ip)
998 struct gfs2_quota_data *qda[4];
999 unsigned int count = 0;
1002 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1005 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1006 struct gfs2_quota_data *qd;
1009 qd = ip->i_res->rs_qa_qd[x];
1010 sync = need_sync(qd);
1012 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1014 if (sync && qd_trylock(qd))
1019 do_sync(count, qda);
1020 for (x = 0; x < count; x++)
1025 gfs2_quota_unhold(ip);
1028 #define MAX_LINE 256
1030 static int print_message(struct gfs2_quota_data *qd, char *type)
1032 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1034 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1035 sdp->sd_fsname, type,
1036 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1037 from_kqid(&init_user_ns, qd->qd_id));
1042 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1044 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1045 struct gfs2_quota_data *qd;
1050 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1053 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1056 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1057 qd = ip->i_res->rs_qa_qd[x];
1059 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1060 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1063 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1064 spin_lock(&qd_lru_lock);
1065 value += qd->qd_change;
1066 spin_unlock(&qd_lru_lock);
1068 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1069 print_message(qd, "exceeded");
1070 quota_send_warning(qd->qd_id,
1071 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1075 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1076 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1077 time_after_eq(jiffies, qd->qd_last_warn +
1079 gt_quota_warn_period) * HZ)) {
1080 quota_send_warning(qd->qd_id,
1081 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1082 error = print_message(qd, "warning");
1083 qd->qd_last_warn = jiffies;
1090 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1091 kuid_t uid, kgid_t gid)
1093 struct gfs2_quota_data *qd;
1096 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1098 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1101 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1102 qd = ip->i_res->rs_qa_qd[x];
1104 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1105 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1111 int gfs2_quota_sync(struct super_block *sb, int type)
1113 struct gfs2_sbd *sdp = sb->s_fs_info;
1114 struct gfs2_quota_data **qda;
1115 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1116 unsigned int num_qd;
1120 sdp->sd_quota_sync_gen++;
1122 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1130 error = qd_fish(sdp, qda + num_qd);
1131 if (error || !qda[num_qd])
1133 if (++num_qd == max_qd)
1139 error = do_sync(num_qd, qda);
1141 for (x = 0; x < num_qd; x++)
1142 qda[x]->qd_sync_gen =
1143 sdp->sd_quota_sync_gen;
1145 for (x = 0; x < num_qd; x++)
1148 } while (!error && num_qd == max_qd);
1155 static int gfs2_quota_sync_timeo(struct super_block *sb, int type)
1157 return gfs2_quota_sync(sb, type);
1160 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1162 struct gfs2_quota_data *qd;
1163 struct gfs2_holder q_gh;
1166 error = qd_get(sdp, qid, &qd);
1170 error = do_glock(qd, FORCE, &q_gh);
1172 gfs2_glock_dq_uninit(&q_gh);
1178 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1180 const struct gfs2_quota_change *str = buf;
1182 qc->qc_change = be64_to_cpu(str->qc_change);
1183 qc->qc_flags = be32_to_cpu(str->qc_flags);
1184 qc->qc_id = make_kqid(&init_user_ns,
1185 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1186 be32_to_cpu(str->qc_id));
1189 int gfs2_quota_init(struct gfs2_sbd *sdp)
1191 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1192 u64 size = i_size_read(sdp->sd_qc_inode);
1193 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1194 unsigned int x, slot = 0;
1195 unsigned int found = 0;
1200 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1203 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1204 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1208 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1209 sizeof(unsigned char *), GFP_NOFS);
1210 if (!sdp->sd_quota_bitmap)
1213 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1214 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1215 if (!sdp->sd_quota_bitmap[x])
1219 for (x = 0; x < blocks; x++) {
1220 struct buffer_head *bh;
1225 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1230 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1233 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1238 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1240 struct gfs2_quota_change_host qc;
1241 struct gfs2_quota_data *qd;
1243 gfs2_quota_change_in(&qc, bh->b_data +
1244 sizeof(struct gfs2_meta_header) +
1245 y * sizeof(struct gfs2_quota_change));
1249 error = qd_alloc(sdp, qc.qc_id, &qd);
1255 set_bit(QDF_CHANGE, &qd->qd_flags);
1256 qd->qd_change = qc.qc_change;
1258 qd->qd_slot_count = 1;
1260 spin_lock(&qd_lru_lock);
1261 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1262 list_add(&qd->qd_list, &sdp->sd_quota_list);
1263 atomic_inc(&sdp->sd_quota_count);
1264 spin_unlock(&qd_lru_lock);
1275 fs_info(sdp, "found %u quota changes\n", found);
1280 gfs2_quota_cleanup(sdp);
1284 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1286 struct list_head *head = &sdp->sd_quota_list;
1287 struct gfs2_quota_data *qd;
1290 spin_lock(&qd_lru_lock);
1291 while (!list_empty(head)) {
1292 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1294 if (atomic_read(&qd->qd_count) > 1 ||
1295 (atomic_read(&qd->qd_count) &&
1296 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1297 list_move(&qd->qd_list, head);
1298 spin_unlock(&qd_lru_lock);
1300 spin_lock(&qd_lru_lock);
1304 list_del(&qd->qd_list);
1305 /* Also remove if this qd exists in the reclaim list */
1306 if (!list_empty(&qd->qd_reclaim)) {
1307 list_del_init(&qd->qd_reclaim);
1308 atomic_dec(&qd_lru_count);
1310 atomic_dec(&sdp->sd_quota_count);
1311 spin_unlock(&qd_lru_lock);
1313 if (!atomic_read(&qd->qd_count)) {
1314 gfs2_assert_warn(sdp, !qd->qd_change);
1315 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1317 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1318 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1320 gfs2_glock_put(qd->qd_gl);
1321 kmem_cache_free(gfs2_quotad_cachep, qd);
1323 spin_lock(&qd_lru_lock);
1325 spin_unlock(&qd_lru_lock);
1327 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1329 if (sdp->sd_quota_bitmap) {
1330 for (x = 0; x < sdp->sd_quota_chunks; x++)
1331 kfree(sdp->sd_quota_bitmap[x]);
1332 kfree(sdp->sd_quota_bitmap);
1336 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1338 if (error == 0 || error == -EROFS)
1340 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1341 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1344 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1345 int (*fxn)(struct super_block *sb, int type),
1346 unsigned long t, unsigned long *timeo,
1347 unsigned int *new_timeo)
1350 int error = fxn(sdp->sd_vfs, 0);
1351 quotad_error(sdp, msg, error);
1352 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1358 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1360 struct gfs2_inode *ip;
1364 spin_lock(&sdp->sd_trunc_lock);
1365 if (!list_empty(&sdp->sd_trunc_list)) {
1366 ip = list_entry(sdp->sd_trunc_list.next,
1367 struct gfs2_inode, i_trunc_list);
1368 list_del_init(&ip->i_trunc_list);
1370 spin_unlock(&sdp->sd_trunc_lock);
1373 gfs2_glock_finish_truncate(ip);
1377 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1378 if (!sdp->sd_statfs_force_sync) {
1379 sdp->sd_statfs_force_sync = 1;
1380 wake_up(&sdp->sd_quota_wait);
1386 * gfs2_quotad - Write cached quota changes into the quota file
1387 * @sdp: Pointer to GFS2 superblock
1391 int gfs2_quotad(void *data)
1393 struct gfs2_sbd *sdp = data;
1394 struct gfs2_tune *tune = &sdp->sd_tune;
1395 unsigned long statfs_timeo = 0;
1396 unsigned long quotad_timeo = 0;
1397 unsigned long t = 0;
1401 while (!kthread_should_stop()) {
1403 /* Update the master statfs file */
1404 if (sdp->sd_statfs_force_sync) {
1405 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1406 quotad_error(sdp, "statfs", error);
1407 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1410 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1412 &tune->gt_statfs_quantum);
1414 /* Update quota file */
1415 quotad_check_timeo(sdp, "sync", gfs2_quota_sync_timeo, t,
1416 "ad_timeo, &tune->gt_quota_quantum);
1418 /* Check for & recover partially truncated inodes */
1419 quotad_check_trunc_list(sdp);
1423 t = min(quotad_timeo, statfs_timeo);
1425 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1426 spin_lock(&sdp->sd_trunc_lock);
1427 empty = list_empty(&sdp->sd_trunc_list);
1428 spin_unlock(&sdp->sd_trunc_lock);
1429 if (empty && !sdp->sd_statfs_force_sync)
1430 t -= schedule_timeout(t);
1433 finish_wait(&sdp->sd_quota_wait, &wait);
1439 static int gfs2_quota_get_xstate(struct super_block *sb,
1440 struct fs_quota_stat *fqs)
1442 struct gfs2_sbd *sdp = sb->s_fs_info;
1444 memset(fqs, 0, sizeof(struct fs_quota_stat));
1445 fqs->qs_version = FS_QSTAT_VERSION;
1447 switch (sdp->sd_args.ar_quota) {
1449 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1451 case GFS2_QUOTA_ACCOUNT:
1452 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1454 case GFS2_QUOTA_OFF:
1458 if (sdp->sd_quota_inode) {
1459 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1460 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1462 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1463 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1464 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1468 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1469 struct fs_disk_quota *fdq)
1471 struct gfs2_sbd *sdp = sb->s_fs_info;
1472 struct gfs2_quota_lvb *qlvb;
1473 struct gfs2_quota_data *qd;
1474 struct gfs2_holder q_gh;
1477 memset(fdq, 0, sizeof(struct fs_disk_quota));
1479 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1480 return -ESRCH; /* Crazy XFS error code */
1482 if ((qid.type != USRQUOTA) &&
1483 (qid.type != GRPQUOTA))
1486 error = qd_get(sdp, qid, &qd);
1489 error = do_glock(qd, FORCE, &q_gh);
1493 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1494 fdq->d_version = FS_DQUOT_VERSION;
1495 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1496 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1497 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1498 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1499 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1501 gfs2_glock_dq_uninit(&q_gh);
1507 /* GFS2 only supports a subset of the XFS fields */
1508 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1510 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1511 struct fs_disk_quota *fdq)
1513 struct gfs2_sbd *sdp = sb->s_fs_info;
1514 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1515 struct gfs2_quota_data *qd;
1516 struct gfs2_holder q_gh, i_gh;
1517 unsigned int data_blocks, ind_blocks;
1518 unsigned int blocks = 0;
1523 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1524 return -ESRCH; /* Crazy XFS error code */
1526 if ((qid.type != USRQUOTA) &&
1527 (qid.type != GRPQUOTA))
1530 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1533 error = qd_get(sdp, qid, &qd);
1537 error = gfs2_rs_alloc(ip);
1541 mutex_lock(&ip->i_inode.i_mutex);
1542 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1545 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1549 /* Check for existing entry, if none then alloc new blocks */
1550 error = update_qd(sdp, qd);
1554 /* If nothing has changed, this is a no-op */
1555 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1556 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1557 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1559 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1560 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1561 fdq->d_fieldmask ^= FS_DQ_BHARD;
1563 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1564 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1565 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1567 if (fdq->d_fieldmask == 0)
1570 offset = qd2offset(qd);
1571 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1572 if (gfs2_is_stuffed(ip))
1574 if (alloc_required) {
1575 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1576 &data_blocks, &ind_blocks);
1577 blocks = 1 + data_blocks + ind_blocks;
1578 error = gfs2_inplace_reserve(ip, blocks, 0);
1581 blocks += gfs2_rg_blocks(ip, blocks);
1584 /* Some quotas span block boundaries and can update two blocks,
1585 adding an extra block to the transaction to handle such quotas */
1586 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1591 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1593 gfs2_trans_end(sdp);
1596 gfs2_inplace_release(ip);
1598 gfs2_glock_dq_uninit(&i_gh);
1600 gfs2_glock_dq_uninit(&q_gh);
1602 mutex_unlock(&ip->i_inode.i_mutex);
1608 const struct quotactl_ops gfs2_quotactl_ops = {
1609 .quota_sync = gfs2_quota_sync,
1610 .get_xstate = gfs2_quota_get_xstate,
1611 .get_dqblk = gfs2_get_dqblk,
1612 .set_dqblk = gfs2_set_dqblk,