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 unsigned long gfs2_qd_shrink_scan(struct shrinker *shrink,
79 struct shrink_control *sc)
81 struct gfs2_quota_data *qd;
83 int nr_to_scan = sc->nr_to_scan;
86 if (!(sc->gfp_mask & __GFP_FS))
89 spin_lock(&qd_lru_lock);
90 while (nr_to_scan && !list_empty(&qd_lru_list)) {
91 qd = list_entry(qd_lru_list.next,
92 struct gfs2_quota_data, qd_reclaim);
93 sdp = qd->qd_gl->gl_sbd;
95 /* Free from the filesystem-specific list */
96 list_del(&qd->qd_list);
98 gfs2_assert_warn(sdp, !qd->qd_change);
99 gfs2_assert_warn(sdp, !qd->qd_slot_count);
100 gfs2_assert_warn(sdp, !qd->qd_bh_count);
102 gfs2_glock_put(qd->qd_gl);
103 atomic_dec(&sdp->sd_quota_count);
105 /* Delete it from the common reclaim list */
106 list_del_init(&qd->qd_reclaim);
107 atomic_dec(&qd_lru_count);
108 spin_unlock(&qd_lru_lock);
109 kmem_cache_free(gfs2_quotad_cachep, qd);
110 spin_lock(&qd_lru_lock);
114 spin_unlock(&qd_lru_lock);
118 unsigned long gfs2_qd_shrink_count(struct shrinker *shrink,
119 struct shrink_control *sc)
121 return vfs_pressure_ratio(atomic_read(&qd_lru_count));
124 static u64 qd2index(struct gfs2_quota_data *qd)
126 struct kqid qid = qd->qd_id;
127 return (2 * (u64)from_kqid(&init_user_ns, qid)) +
128 ((qid.type == USRQUOTA) ? 0 : 1);
131 static u64 qd2offset(struct gfs2_quota_data *qd)
135 offset = qd2index(qd);
136 offset *= sizeof(struct gfs2_quota);
141 static int qd_alloc(struct gfs2_sbd *sdp, struct kqid qid,
142 struct gfs2_quota_data **qdp)
144 struct gfs2_quota_data *qd;
147 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
151 atomic_set(&qd->qd_count, 1);
154 INIT_LIST_HEAD(&qd->qd_reclaim);
156 error = gfs2_glock_get(sdp, qd2index(qd),
157 &gfs2_quota_glops, CREATE, &qd->qd_gl);
166 kmem_cache_free(gfs2_quotad_cachep, qd);
170 static int qd_get(struct gfs2_sbd *sdp, struct kqid qid,
171 struct gfs2_quota_data **qdp)
173 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
180 spin_lock(&qd_lru_lock);
181 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
182 if (qid_eq(qd->qd_id, qid)) {
183 if (!atomic_read(&qd->qd_count) &&
184 !list_empty(&qd->qd_reclaim)) {
185 /* Remove it from reclaim list */
186 list_del_init(&qd->qd_reclaim);
187 atomic_dec(&qd_lru_count);
189 atomic_inc(&qd->qd_count);
200 list_add(&qd->qd_list, &sdp->sd_quota_list);
201 atomic_inc(&sdp->sd_quota_count);
205 spin_unlock(&qd_lru_lock);
209 gfs2_glock_put(new_qd->qd_gl);
210 kmem_cache_free(gfs2_quotad_cachep, new_qd);
216 error = qd_alloc(sdp, qid, &new_qd);
222 static void qd_hold(struct gfs2_quota_data *qd)
224 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
225 gfs2_assert(sdp, atomic_read(&qd->qd_count));
226 atomic_inc(&qd->qd_count);
229 static void qd_put(struct gfs2_quota_data *qd)
231 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
232 /* Add to the reclaim list */
233 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
234 atomic_inc(&qd_lru_count);
235 spin_unlock(&qd_lru_lock);
239 static int slot_get(struct gfs2_quota_data *qd)
241 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
242 unsigned int c, o = 0, b;
243 unsigned char byte = 0;
245 spin_lock(&qd_lru_lock);
247 if (qd->qd_slot_count++) {
248 spin_unlock(&qd_lru_lock);
252 for (c = 0; c < sdp->sd_quota_chunks; c++)
253 for (o = 0; o < PAGE_SIZE; o++) {
254 byte = sdp->sd_quota_bitmap[c][o];
262 for (b = 0; b < 8; b++)
263 if (!(byte & (1 << b)))
265 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
267 if (qd->qd_slot >= sdp->sd_quota_slots)
270 sdp->sd_quota_bitmap[c][o] |= 1 << b;
272 spin_unlock(&qd_lru_lock);
278 spin_unlock(&qd_lru_lock);
282 static void slot_hold(struct gfs2_quota_data *qd)
284 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
286 spin_lock(&qd_lru_lock);
287 gfs2_assert(sdp, qd->qd_slot_count);
289 spin_unlock(&qd_lru_lock);
292 static void slot_put(struct gfs2_quota_data *qd)
294 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
296 spin_lock(&qd_lru_lock);
297 gfs2_assert(sdp, qd->qd_slot_count);
298 if (!--qd->qd_slot_count) {
299 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
302 spin_unlock(&qd_lru_lock);
305 static int bh_get(struct gfs2_quota_data *qd)
307 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
308 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
309 unsigned int block, offset;
310 struct buffer_head *bh;
312 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
314 mutex_lock(&sdp->sd_quota_mutex);
316 if (qd->qd_bh_count++) {
317 mutex_unlock(&sdp->sd_quota_mutex);
321 block = qd->qd_slot / sdp->sd_qc_per_block;
322 offset = qd->qd_slot % sdp->sd_qc_per_block;
324 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
325 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
328 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
332 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
336 qd->qd_bh_qc = (struct gfs2_quota_change *)
337 (bh->b_data + sizeof(struct gfs2_meta_header) +
338 offset * sizeof(struct gfs2_quota_change));
340 mutex_unlock(&sdp->sd_quota_mutex);
348 mutex_unlock(&sdp->sd_quota_mutex);
352 static void bh_put(struct gfs2_quota_data *qd)
354 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
356 mutex_lock(&sdp->sd_quota_mutex);
357 gfs2_assert(sdp, qd->qd_bh_count);
358 if (!--qd->qd_bh_count) {
363 mutex_unlock(&sdp->sd_quota_mutex);
366 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
368 struct gfs2_quota_data *qd = NULL;
374 if (sdp->sd_vfs->s_flags & MS_RDONLY)
377 spin_lock(&qd_lru_lock);
379 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
380 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
381 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
382 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
385 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
387 set_bit(QDF_LOCKED, &qd->qd_flags);
388 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
389 atomic_inc(&qd->qd_count);
390 qd->qd_change_sync = qd->qd_change;
391 gfs2_assert_warn(sdp, qd->qd_slot_count);
401 spin_unlock(&qd_lru_lock);
404 gfs2_assert_warn(sdp, qd->qd_change_sync);
407 clear_bit(QDF_LOCKED, &qd->qd_flags);
419 static int qd_trylock(struct gfs2_quota_data *qd)
421 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
423 if (sdp->sd_vfs->s_flags & MS_RDONLY)
426 spin_lock(&qd_lru_lock);
428 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
429 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
430 spin_unlock(&qd_lru_lock);
434 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
436 set_bit(QDF_LOCKED, &qd->qd_flags);
437 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
438 atomic_inc(&qd->qd_count);
439 qd->qd_change_sync = qd->qd_change;
440 gfs2_assert_warn(sdp, qd->qd_slot_count);
443 spin_unlock(&qd_lru_lock);
445 gfs2_assert_warn(sdp, qd->qd_change_sync);
447 clear_bit(QDF_LOCKED, &qd->qd_flags);
456 static void qd_unlock(struct gfs2_quota_data *qd)
458 gfs2_assert_warn(qd->qd_gl->gl_sbd,
459 test_bit(QDF_LOCKED, &qd->qd_flags));
460 clear_bit(QDF_LOCKED, &qd->qd_flags);
466 static int qdsb_get(struct gfs2_sbd *sdp, struct kqid qid,
467 struct gfs2_quota_data **qdp)
471 error = qd_get(sdp, qid, qdp);
475 error = slot_get(*qdp);
479 error = bh_get(*qdp);
492 static void qdsb_put(struct gfs2_quota_data *qd)
499 int gfs2_quota_hold(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
501 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
502 struct gfs2_quota_data **qd;
505 if (ip->i_res == NULL) {
506 error = gfs2_rs_alloc(ip);
511 qd = ip->i_res->rs_qa_qd;
513 if (gfs2_assert_warn(sdp, !ip->i_res->rs_qa_qd_num) ||
514 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
517 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
520 error = qdsb_get(sdp, make_kqid_uid(ip->i_inode.i_uid), qd);
523 ip->i_res->rs_qa_qd_num++;
526 error = qdsb_get(sdp, make_kqid_gid(ip->i_inode.i_gid), qd);
529 ip->i_res->rs_qa_qd_num++;
532 if (!uid_eq(uid, NO_UID_QUOTA_CHANGE) &&
533 !uid_eq(uid, ip->i_inode.i_uid)) {
534 error = qdsb_get(sdp, make_kqid_uid(uid), qd);
537 ip->i_res->rs_qa_qd_num++;
541 if (!gid_eq(gid, NO_GID_QUOTA_CHANGE) &&
542 !gid_eq(gid, ip->i_inode.i_gid)) {
543 error = qdsb_get(sdp, make_kqid_gid(gid), qd);
546 ip->i_res->rs_qa_qd_num++;
552 gfs2_quota_unhold(ip);
556 void gfs2_quota_unhold(struct gfs2_inode *ip)
558 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
561 if (ip->i_res == NULL)
563 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
565 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
566 qdsb_put(ip->i_res->rs_qa_qd[x]);
567 ip->i_res->rs_qa_qd[x] = NULL;
569 ip->i_res->rs_qa_qd_num = 0;
572 static int sort_qd(const void *a, const void *b)
574 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
575 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
577 if (qid_lt(qd_a->qd_id, qd_b->qd_id))
579 if (qid_lt(qd_b->qd_id, qd_a->qd_id))
584 static void do_qc(struct gfs2_quota_data *qd, s64 change)
586 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
587 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
588 struct gfs2_quota_change *qc = qd->qd_bh_qc;
591 mutex_lock(&sdp->sd_quota_mutex);
592 gfs2_trans_add_meta(ip->i_gl, qd->qd_bh);
594 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
597 if (qd->qd_id.type == USRQUOTA)
598 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
599 qc->qc_id = cpu_to_be32(from_kqid(&init_user_ns, qd->qd_id));
602 x = be64_to_cpu(qc->qc_change) + change;
603 qc->qc_change = cpu_to_be64(x);
605 spin_lock(&qd_lru_lock);
607 spin_unlock(&qd_lru_lock);
610 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
611 clear_bit(QDF_CHANGE, &qd->qd_flags);
616 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
621 mutex_unlock(&sdp->sd_quota_mutex);
625 * gfs2_adjust_quota - adjust record of current block usage
626 * @ip: The quota inode
627 * @loc: Offset of the entry in the quota file
628 * @change: The amount of usage change to record
629 * @qd: The quota data
630 * @fdq: The updated limits to record
632 * This function was mostly borrowed from gfs2_block_truncate_page which was
633 * in turn mostly borrowed from ext3
635 * Returns: 0 or -ve on error
638 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
639 s64 change, struct gfs2_quota_data *qd,
640 struct fs_disk_quota *fdq)
642 struct inode *inode = &ip->i_inode;
643 struct gfs2_sbd *sdp = GFS2_SB(inode);
644 struct address_space *mapping = inode->i_mapping;
645 unsigned long index = loc >> PAGE_CACHE_SHIFT;
646 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
647 unsigned blocksize, iblock, pos;
648 struct buffer_head *bh;
651 struct gfs2_quota q, *qp;
655 if (gfs2_is_stuffed(ip)) {
656 err = gfs2_unstuff_dinode(ip, NULL);
661 memset(&q, 0, sizeof(struct gfs2_quota));
662 err = gfs2_internal_read(ip, (char *)&q, &loc, sizeof(q));
668 qp->qu_value = be64_to_cpu(qp->qu_value);
669 qp->qu_value += change;
670 qp->qu_value = cpu_to_be64(qp->qu_value);
671 qd->qd_qb.qb_value = qp->qu_value;
673 if (fdq->d_fieldmask & FS_DQ_BSOFT) {
674 qp->qu_warn = cpu_to_be64(fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift);
675 qd->qd_qb.qb_warn = qp->qu_warn;
677 if (fdq->d_fieldmask & FS_DQ_BHARD) {
678 qp->qu_limit = cpu_to_be64(fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift);
679 qd->qd_qb.qb_limit = qp->qu_limit;
681 if (fdq->d_fieldmask & FS_DQ_BCOUNT) {
682 qp->qu_value = cpu_to_be64(fdq->d_bcount >> sdp->sd_fsb2bb_shift);
683 qd->qd_qb.qb_value = qp->qu_value;
687 /* Write the quota into the quota file on disk */
689 nbytes = sizeof(struct gfs2_quota);
691 page = find_or_create_page(mapping, index, GFP_NOFS);
695 blocksize = inode->i_sb->s_blocksize;
696 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
698 if (!page_has_buffers(page))
699 create_empty_buffers(page, blocksize, 0);
701 bh = page_buffers(page);
703 while (offset >= pos) {
704 bh = bh->b_this_page;
709 if (!buffer_mapped(bh)) {
710 gfs2_block_map(inode, iblock, bh, 1);
711 if (!buffer_mapped(bh))
713 /* If it's a newly allocated disk block for quota, zero it */
715 zero_user(page, pos - blocksize, bh->b_size);
718 if (PageUptodate(page))
719 set_buffer_uptodate(bh);
721 if (!buffer_uptodate(bh)) {
722 ll_rw_block(READ | REQ_META, 1, &bh);
724 if (!buffer_uptodate(bh))
728 gfs2_trans_add_data(ip->i_gl, bh);
730 kaddr = kmap_atomic(page);
731 if (offset + sizeof(struct gfs2_quota) > PAGE_CACHE_SIZE)
732 nbytes = PAGE_CACHE_SIZE - offset;
733 memcpy(kaddr + offset, ptr, nbytes);
734 flush_dcache_page(page);
735 kunmap_atomic(kaddr);
737 page_cache_release(page);
739 /* If quota straddles page boundary, we need to update the rest of the
740 * quota at the beginning of the next page */
741 if ((offset + sizeof(struct gfs2_quota)) > PAGE_CACHE_SIZE) {
743 nbytes = sizeof(struct gfs2_quota) - nbytes;
749 size = loc + sizeof(struct gfs2_quota);
750 if (size > inode->i_size)
751 i_size_write(inode, size);
752 inode->i_mtime = inode->i_atime = CURRENT_TIME;
753 mark_inode_dirty(inode);
758 page_cache_release(page);
762 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
764 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
765 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
766 struct gfs2_alloc_parms ap = { .aflags = 0, };
767 unsigned int data_blocks, ind_blocks;
768 struct gfs2_holder *ghs, i_gh;
770 struct gfs2_quota_data *qd;
773 unsigned int nalloc = 0, blocks;
776 error = gfs2_rs_alloc(ip);
780 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
781 &data_blocks, &ind_blocks);
783 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
787 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
788 mutex_lock(&ip->i_inode.i_mutex);
789 for (qx = 0; qx < num_qd; qx++) {
790 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
791 GL_NOCACHE, &ghs[qx]);
796 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
800 for (x = 0; x < num_qd; x++) {
801 offset = qd2offset(qda[x]);
802 if (gfs2_write_alloc_required(ip, offset,
803 sizeof(struct gfs2_quota)))
808 * 1 blk for unstuffing inode if stuffed. We add this extra
809 * block to the reservation unconditionally. If the inode
810 * doesn't need unstuffing, the block will be released to the
811 * rgrp since it won't be allocated during the transaction
813 /* +3 in the end for unstuffing block, inode size update block
814 * and another block in case quota straddles page boundary and
815 * two blocks need to be updated instead of 1 */
816 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
818 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
819 ap.target = reserved;
820 error = gfs2_inplace_reserve(ip, &ap);
825 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
827 error = gfs2_trans_begin(sdp, blocks, 0);
831 for (x = 0; x < num_qd; x++) {
833 offset = qd2offset(qd);
834 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
838 do_qc(qd, -qd->qd_change_sync);
839 set_bit(QDF_REFRESH, &qd->qd_flags);
847 gfs2_inplace_release(ip);
849 gfs2_glock_dq_uninit(&i_gh);
852 gfs2_glock_dq_uninit(&ghs[qx]);
853 mutex_unlock(&ip->i_inode.i_mutex);
855 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
859 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
861 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
863 struct gfs2_quota_lvb *qlvb;
867 memset(&q, 0, sizeof(struct gfs2_quota));
869 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
873 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
874 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
876 qlvb->qb_limit = q.qu_limit;
877 qlvb->qb_warn = q.qu_warn;
878 qlvb->qb_value = q.qu_value;
884 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
885 struct gfs2_holder *q_gh)
887 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
888 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
889 struct gfs2_holder i_gh;
893 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
897 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
899 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
900 gfs2_glock_dq_uninit(q_gh);
901 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
906 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
910 error = update_qd(sdp, qd);
914 gfs2_glock_dq_uninit(&i_gh);
915 gfs2_glock_dq_uninit(q_gh);
923 gfs2_glock_dq_uninit(&i_gh);
925 gfs2_glock_dq_uninit(q_gh);
929 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
931 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
932 struct gfs2_quota_data *qd;
936 error = gfs2_quota_hold(ip, uid, gid);
940 if (capable(CAP_SYS_RESOURCE) ||
941 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
944 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
945 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
947 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
948 int force = NO_FORCE;
949 qd = ip->i_res->rs_qa_qd[x];
950 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
952 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
958 set_bit(GIF_QD_LOCKED, &ip->i_flags);
961 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
962 gfs2_quota_unhold(ip);
968 static int need_sync(struct gfs2_quota_data *qd)
970 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
971 struct gfs2_tune *gt = &sdp->sd_tune;
973 unsigned int num, den;
976 if (!qd->qd_qb.qb_limit)
979 spin_lock(&qd_lru_lock);
980 value = qd->qd_change;
981 spin_unlock(&qd_lru_lock);
983 spin_lock(>->gt_spin);
984 num = gt->gt_quota_scale_num;
985 den = gt->gt_quota_scale_den;
986 spin_unlock(>->gt_spin);
990 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
991 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
994 value *= gfs2_jindex_size(sdp) * num;
995 value = div_s64(value, den);
996 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
997 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
1004 void gfs2_quota_unlock(struct gfs2_inode *ip)
1006 struct gfs2_quota_data *qda[4];
1007 unsigned int count = 0;
1010 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1013 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1014 struct gfs2_quota_data *qd;
1017 qd = ip->i_res->rs_qa_qd[x];
1018 sync = need_sync(qd);
1020 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1022 if (sync && qd_trylock(qd))
1027 do_sync(count, qda);
1028 for (x = 0; x < count; x++)
1033 gfs2_quota_unhold(ip);
1036 #define MAX_LINE 256
1038 static int print_message(struct gfs2_quota_data *qd, char *type)
1040 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1042 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1043 sdp->sd_fsname, type,
1044 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1045 from_kqid(&init_user_ns, qd->qd_id));
1050 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1052 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1053 struct gfs2_quota_data *qd;
1058 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1061 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1064 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1065 qd = ip->i_res->rs_qa_qd[x];
1067 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1068 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1071 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1072 spin_lock(&qd_lru_lock);
1073 value += qd->qd_change;
1074 spin_unlock(&qd_lru_lock);
1076 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1077 print_message(qd, "exceeded");
1078 quota_send_warning(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(qd->qd_id,
1089 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1090 error = print_message(qd, "warning");
1091 qd->qd_last_warn = jiffies;
1098 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1099 kuid_t uid, kgid_t gid)
1101 struct gfs2_quota_data *qd;
1104 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1106 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1109 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1110 qd = ip->i_res->rs_qa_qd[x];
1112 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1113 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1119 int gfs2_quota_sync(struct super_block *sb, int type)
1121 struct gfs2_sbd *sdp = sb->s_fs_info;
1122 struct gfs2_quota_data **qda;
1123 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1124 unsigned int num_qd;
1128 sdp->sd_quota_sync_gen++;
1130 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1138 error = qd_fish(sdp, qda + num_qd);
1139 if (error || !qda[num_qd])
1141 if (++num_qd == max_qd)
1147 error = do_sync(num_qd, qda);
1149 for (x = 0; x < num_qd; x++)
1150 qda[x]->qd_sync_gen =
1151 sdp->sd_quota_sync_gen;
1153 for (x = 0; x < num_qd; x++)
1156 } while (!error && num_qd == max_qd);
1163 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1165 struct gfs2_quota_data *qd;
1166 struct gfs2_holder q_gh;
1169 error = qd_get(sdp, qid, &qd);
1173 error = do_glock(qd, FORCE, &q_gh);
1175 gfs2_glock_dq_uninit(&q_gh);
1181 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1183 const struct gfs2_quota_change *str = buf;
1185 qc->qc_change = be64_to_cpu(str->qc_change);
1186 qc->qc_flags = be32_to_cpu(str->qc_flags);
1187 qc->qc_id = make_kqid(&init_user_ns,
1188 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1189 be32_to_cpu(str->qc_id));
1192 int gfs2_quota_init(struct gfs2_sbd *sdp)
1194 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1195 u64 size = i_size_read(sdp->sd_qc_inode);
1196 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1197 unsigned int x, slot = 0;
1198 unsigned int found = 0;
1203 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1206 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1207 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1211 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1212 sizeof(unsigned char *), GFP_NOFS);
1213 if (!sdp->sd_quota_bitmap)
1216 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1217 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1218 if (!sdp->sd_quota_bitmap[x])
1222 for (x = 0; x < blocks; x++) {
1223 struct buffer_head *bh;
1228 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1233 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1236 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1241 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1243 struct gfs2_quota_change_host qc;
1244 struct gfs2_quota_data *qd;
1246 gfs2_quota_change_in(&qc, bh->b_data +
1247 sizeof(struct gfs2_meta_header) +
1248 y * sizeof(struct gfs2_quota_change));
1252 error = qd_alloc(sdp, qc.qc_id, &qd);
1258 set_bit(QDF_CHANGE, &qd->qd_flags);
1259 qd->qd_change = qc.qc_change;
1261 qd->qd_slot_count = 1;
1263 spin_lock(&qd_lru_lock);
1264 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1265 list_add(&qd->qd_list, &sdp->sd_quota_list);
1266 atomic_inc(&sdp->sd_quota_count);
1267 spin_unlock(&qd_lru_lock);
1278 fs_info(sdp, "found %u quota changes\n", found);
1283 gfs2_quota_cleanup(sdp);
1287 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1289 struct list_head *head = &sdp->sd_quota_list;
1290 struct gfs2_quota_data *qd;
1293 spin_lock(&qd_lru_lock);
1294 while (!list_empty(head)) {
1295 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1297 if (atomic_read(&qd->qd_count) > 1 ||
1298 (atomic_read(&qd->qd_count) &&
1299 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1300 list_move(&qd->qd_list, head);
1301 spin_unlock(&qd_lru_lock);
1303 spin_lock(&qd_lru_lock);
1307 list_del(&qd->qd_list);
1308 /* Also remove if this qd exists in the reclaim list */
1309 if (!list_empty(&qd->qd_reclaim)) {
1310 list_del_init(&qd->qd_reclaim);
1311 atomic_dec(&qd_lru_count);
1313 atomic_dec(&sdp->sd_quota_count);
1314 spin_unlock(&qd_lru_lock);
1316 if (!atomic_read(&qd->qd_count)) {
1317 gfs2_assert_warn(sdp, !qd->qd_change);
1318 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1320 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1321 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1323 gfs2_glock_put(qd->qd_gl);
1324 kmem_cache_free(gfs2_quotad_cachep, qd);
1326 spin_lock(&qd_lru_lock);
1328 spin_unlock(&qd_lru_lock);
1330 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1332 if (sdp->sd_quota_bitmap) {
1333 for (x = 0; x < sdp->sd_quota_chunks; x++)
1334 kfree(sdp->sd_quota_bitmap[x]);
1335 kfree(sdp->sd_quota_bitmap);
1339 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1341 if (error == 0 || error == -EROFS)
1343 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1344 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1347 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1348 int (*fxn)(struct super_block *sb, int type),
1349 unsigned long t, unsigned long *timeo,
1350 unsigned int *new_timeo)
1353 int error = fxn(sdp->sd_vfs, 0);
1354 quotad_error(sdp, msg, error);
1355 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1361 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1363 struct gfs2_inode *ip;
1367 spin_lock(&sdp->sd_trunc_lock);
1368 if (!list_empty(&sdp->sd_trunc_list)) {
1369 ip = list_entry(sdp->sd_trunc_list.next,
1370 struct gfs2_inode, i_trunc_list);
1371 list_del_init(&ip->i_trunc_list);
1373 spin_unlock(&sdp->sd_trunc_lock);
1376 gfs2_glock_finish_truncate(ip);
1380 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1381 if (!sdp->sd_statfs_force_sync) {
1382 sdp->sd_statfs_force_sync = 1;
1383 wake_up(&sdp->sd_quota_wait);
1389 * gfs2_quotad - Write cached quota changes into the quota file
1390 * @sdp: Pointer to GFS2 superblock
1394 int gfs2_quotad(void *data)
1396 struct gfs2_sbd *sdp = data;
1397 struct gfs2_tune *tune = &sdp->sd_tune;
1398 unsigned long statfs_timeo = 0;
1399 unsigned long quotad_timeo = 0;
1400 unsigned long t = 0;
1404 while (!kthread_should_stop()) {
1406 /* Update the master statfs file */
1407 if (sdp->sd_statfs_force_sync) {
1408 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1409 quotad_error(sdp, "statfs", error);
1410 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1413 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1415 &tune->gt_statfs_quantum);
1417 /* Update quota file */
1418 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1419 "ad_timeo, &tune->gt_quota_quantum);
1421 /* Check for & recover partially truncated inodes */
1422 quotad_check_trunc_list(sdp);
1426 t = min(quotad_timeo, statfs_timeo);
1428 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1429 spin_lock(&sdp->sd_trunc_lock);
1430 empty = list_empty(&sdp->sd_trunc_list);
1431 spin_unlock(&sdp->sd_trunc_lock);
1432 if (empty && !sdp->sd_statfs_force_sync)
1433 t -= schedule_timeout(t);
1436 finish_wait(&sdp->sd_quota_wait, &wait);
1442 static int gfs2_quota_get_xstate(struct super_block *sb,
1443 struct fs_quota_stat *fqs)
1445 struct gfs2_sbd *sdp = sb->s_fs_info;
1447 memset(fqs, 0, sizeof(struct fs_quota_stat));
1448 fqs->qs_version = FS_QSTAT_VERSION;
1450 switch (sdp->sd_args.ar_quota) {
1452 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1454 case GFS2_QUOTA_ACCOUNT:
1455 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1457 case GFS2_QUOTA_OFF:
1461 if (sdp->sd_quota_inode) {
1462 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1463 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1465 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1466 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1467 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1471 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1472 struct fs_disk_quota *fdq)
1474 struct gfs2_sbd *sdp = sb->s_fs_info;
1475 struct gfs2_quota_lvb *qlvb;
1476 struct gfs2_quota_data *qd;
1477 struct gfs2_holder q_gh;
1480 memset(fdq, 0, sizeof(struct fs_disk_quota));
1482 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1483 return -ESRCH; /* Crazy XFS error code */
1485 if ((qid.type != USRQUOTA) &&
1486 (qid.type != GRPQUOTA))
1489 error = qd_get(sdp, qid, &qd);
1492 error = do_glock(qd, FORCE, &q_gh);
1496 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1497 fdq->d_version = FS_DQUOT_VERSION;
1498 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1499 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1500 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1501 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1502 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1504 gfs2_glock_dq_uninit(&q_gh);
1510 /* GFS2 only supports a subset of the XFS fields */
1511 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1513 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1514 struct fs_disk_quota *fdq)
1516 struct gfs2_sbd *sdp = sb->s_fs_info;
1517 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1518 struct gfs2_quota_data *qd;
1519 struct gfs2_holder q_gh, i_gh;
1520 unsigned int data_blocks, ind_blocks;
1521 unsigned int blocks = 0;
1526 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1527 return -ESRCH; /* Crazy XFS error code */
1529 if ((qid.type != USRQUOTA) &&
1530 (qid.type != GRPQUOTA))
1533 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1536 error = qd_get(sdp, qid, &qd);
1540 error = gfs2_rs_alloc(ip);
1544 mutex_lock(&ip->i_inode.i_mutex);
1545 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1548 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1552 /* Check for existing entry, if none then alloc new blocks */
1553 error = update_qd(sdp, qd);
1557 /* If nothing has changed, this is a no-op */
1558 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1559 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1560 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1562 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1563 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1564 fdq->d_fieldmask ^= FS_DQ_BHARD;
1566 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1567 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1568 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1570 if (fdq->d_fieldmask == 0)
1573 offset = qd2offset(qd);
1574 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1575 if (gfs2_is_stuffed(ip))
1577 if (alloc_required) {
1578 struct gfs2_alloc_parms ap = { .aflags = 0, };
1579 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1580 &data_blocks, &ind_blocks);
1581 blocks = 1 + data_blocks + ind_blocks;
1583 error = gfs2_inplace_reserve(ip, &ap);
1586 blocks += gfs2_rg_blocks(ip, blocks);
1589 /* Some quotas span block boundaries and can update two blocks,
1590 adding an extra block to the transaction to handle such quotas */
1591 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1596 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1598 gfs2_trans_end(sdp);
1601 gfs2_inplace_release(ip);
1603 gfs2_glock_dq_uninit(&i_gh);
1605 gfs2_glock_dq_uninit(&q_gh);
1607 mutex_unlock(&ip->i_inode.i_mutex);
1613 const struct quotactl_ops gfs2_quotactl_ops = {
1614 .quota_sync = gfs2_quota_sync,
1615 .get_xstate = gfs2_quota_get_xstate,
1616 .get_dqblk = gfs2_get_dqblk,
1617 .set_dqblk = gfs2_set_dqblk,