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 vfs_pressure_ratio(atomic_read(&qd_lru_count));
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_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 unsigned int data_blocks, ind_blocks;
763 struct gfs2_holder *ghs, i_gh;
765 struct gfs2_quota_data *qd;
768 unsigned int nalloc = 0, blocks;
771 error = gfs2_rs_alloc(ip);
775 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
776 &data_blocks, &ind_blocks);
778 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
782 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
783 mutex_lock(&ip->i_inode.i_mutex);
784 for (qx = 0; qx < num_qd; qx++) {
785 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
786 GL_NOCACHE, &ghs[qx]);
791 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
795 for (x = 0; x < num_qd; x++) {
796 offset = qd2offset(qda[x]);
797 if (gfs2_write_alloc_required(ip, offset,
798 sizeof(struct gfs2_quota)))
803 * 1 blk for unstuffing inode if stuffed. We add this extra
804 * block to the reservation unconditionally. If the inode
805 * doesn't need unstuffing, the block will be released to the
806 * rgrp since it won't be allocated during the transaction
808 /* +3 in the end for unstuffing block, inode size update block
809 * and another block in case quota straddles page boundary and
810 * two blocks need to be updated instead of 1 */
811 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 3;
813 reserved = 1 + (nalloc * (data_blocks + ind_blocks));
814 error = gfs2_inplace_reserve(ip, reserved, 0);
819 blocks += gfs2_rg_blocks(ip, reserved) + nalloc * ind_blocks + RES_STATFS;
821 error = gfs2_trans_begin(sdp, blocks, 0);
825 for (x = 0; x < num_qd; x++) {
827 offset = qd2offset(qd);
828 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync, qd, NULL);
832 do_qc(qd, -qd->qd_change_sync);
833 set_bit(QDF_REFRESH, &qd->qd_flags);
841 gfs2_inplace_release(ip);
843 gfs2_glock_dq_uninit(&i_gh);
846 gfs2_glock_dq_uninit(&ghs[qx]);
847 mutex_unlock(&ip->i_inode.i_mutex);
849 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
853 static int update_qd(struct gfs2_sbd *sdp, struct gfs2_quota_data *qd)
855 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
857 struct gfs2_quota_lvb *qlvb;
861 memset(&q, 0, sizeof(struct gfs2_quota));
863 error = gfs2_internal_read(ip, (char *)&q, &pos, sizeof(q));
867 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
868 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
870 qlvb->qb_limit = q.qu_limit;
871 qlvb->qb_warn = q.qu_warn;
872 qlvb->qb_value = q.qu_value;
878 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
879 struct gfs2_holder *q_gh)
881 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
882 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
883 struct gfs2_holder i_gh;
887 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
891 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
893 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
894 gfs2_glock_dq_uninit(q_gh);
895 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
900 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
904 error = update_qd(sdp, qd);
908 gfs2_glock_dq_uninit(&i_gh);
909 gfs2_glock_dq_uninit(q_gh);
917 gfs2_glock_dq_uninit(&i_gh);
919 gfs2_glock_dq_uninit(q_gh);
923 int gfs2_quota_lock(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
925 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
926 struct gfs2_quota_data *qd;
930 error = gfs2_quota_hold(ip, uid, gid);
934 if (capable(CAP_SYS_RESOURCE) ||
935 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
938 sort(ip->i_res->rs_qa_qd, ip->i_res->rs_qa_qd_num,
939 sizeof(struct gfs2_quota_data *), sort_qd, NULL);
941 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
942 int force = NO_FORCE;
943 qd = ip->i_res->rs_qa_qd[x];
944 if (test_and_clear_bit(QDF_REFRESH, &qd->qd_flags))
946 error = do_glock(qd, force, &ip->i_res->rs_qa_qd_ghs[x]);
952 set_bit(GIF_QD_LOCKED, &ip->i_flags);
955 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
956 gfs2_quota_unhold(ip);
962 static int need_sync(struct gfs2_quota_data *qd)
964 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
965 struct gfs2_tune *gt = &sdp->sd_tune;
967 unsigned int num, den;
970 if (!qd->qd_qb.qb_limit)
973 spin_lock(&qd_lru_lock);
974 value = qd->qd_change;
975 spin_unlock(&qd_lru_lock);
977 spin_lock(>->gt_spin);
978 num = gt->gt_quota_scale_num;
979 den = gt->gt_quota_scale_den;
980 spin_unlock(>->gt_spin);
984 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
985 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
988 value *= gfs2_jindex_size(sdp) * num;
989 value = div_s64(value, den);
990 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
991 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
998 void gfs2_quota_unlock(struct gfs2_inode *ip)
1000 struct gfs2_quota_data *qda[4];
1001 unsigned int count = 0;
1004 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
1007 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1008 struct gfs2_quota_data *qd;
1011 qd = ip->i_res->rs_qa_qd[x];
1012 sync = need_sync(qd);
1014 gfs2_glock_dq_uninit(&ip->i_res->rs_qa_qd_ghs[x]);
1016 if (sync && qd_trylock(qd))
1021 do_sync(count, qda);
1022 for (x = 0; x < count; x++)
1027 gfs2_quota_unhold(ip);
1030 #define MAX_LINE 256
1032 static int print_message(struct gfs2_quota_data *qd, char *type)
1034 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
1036 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\n",
1037 sdp->sd_fsname, type,
1038 (qd->qd_id.type == USRQUOTA) ? "user" : "group",
1039 from_kqid(&init_user_ns, qd->qd_id));
1044 int gfs2_quota_check(struct gfs2_inode *ip, kuid_t uid, kgid_t gid)
1046 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
1047 struct gfs2_quota_data *qd;
1052 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
1055 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
1058 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1059 qd = ip->i_res->rs_qa_qd[x];
1061 if (!(qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1062 qid_eq(qd->qd_id, make_kqid_gid(gid))))
1065 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
1066 spin_lock(&qd_lru_lock);
1067 value += qd->qd_change;
1068 spin_unlock(&qd_lru_lock);
1070 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1071 print_message(qd, "exceeded");
1072 quota_send_warning(qd->qd_id,
1073 sdp->sd_vfs->s_dev, QUOTA_NL_BHARDWARN);
1077 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1078 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1079 time_after_eq(jiffies, qd->qd_last_warn +
1081 gt_quota_warn_period) * HZ)) {
1082 quota_send_warning(qd->qd_id,
1083 sdp->sd_vfs->s_dev, QUOTA_NL_BSOFTWARN);
1084 error = print_message(qd, "warning");
1085 qd->qd_last_warn = jiffies;
1092 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1093 kuid_t uid, kgid_t gid)
1095 struct gfs2_quota_data *qd;
1098 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1100 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1103 for (x = 0; x < ip->i_res->rs_qa_qd_num; x++) {
1104 qd = ip->i_res->rs_qa_qd[x];
1106 if (qid_eq(qd->qd_id, make_kqid_uid(uid)) ||
1107 qid_eq(qd->qd_id, make_kqid_gid(gid))) {
1113 int gfs2_quota_sync(struct super_block *sb, int type)
1115 struct gfs2_sbd *sdp = sb->s_fs_info;
1116 struct gfs2_quota_data **qda;
1117 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1118 unsigned int num_qd;
1122 sdp->sd_quota_sync_gen++;
1124 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1132 error = qd_fish(sdp, qda + num_qd);
1133 if (error || !qda[num_qd])
1135 if (++num_qd == max_qd)
1141 error = do_sync(num_qd, qda);
1143 for (x = 0; x < num_qd; x++)
1144 qda[x]->qd_sync_gen =
1145 sdp->sd_quota_sync_gen;
1147 for (x = 0; x < num_qd; x++)
1150 } while (!error && num_qd == max_qd);
1157 int gfs2_quota_refresh(struct gfs2_sbd *sdp, struct kqid qid)
1159 struct gfs2_quota_data *qd;
1160 struct gfs2_holder q_gh;
1163 error = qd_get(sdp, qid, &qd);
1167 error = do_glock(qd, FORCE, &q_gh);
1169 gfs2_glock_dq_uninit(&q_gh);
1175 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1177 const struct gfs2_quota_change *str = buf;
1179 qc->qc_change = be64_to_cpu(str->qc_change);
1180 qc->qc_flags = be32_to_cpu(str->qc_flags);
1181 qc->qc_id = make_kqid(&init_user_ns,
1182 (qc->qc_flags & GFS2_QCF_USER)?USRQUOTA:GRPQUOTA,
1183 be32_to_cpu(str->qc_id));
1186 int gfs2_quota_init(struct gfs2_sbd *sdp)
1188 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1189 u64 size = i_size_read(sdp->sd_qc_inode);
1190 unsigned int blocks = size >> sdp->sd_sb.sb_bsize_shift;
1191 unsigned int x, slot = 0;
1192 unsigned int found = 0;
1197 if (gfs2_check_internal_file_size(sdp->sd_qc_inode, 1, 64 << 20))
1200 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1201 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1205 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1206 sizeof(unsigned char *), GFP_NOFS);
1207 if (!sdp->sd_quota_bitmap)
1210 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1211 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1212 if (!sdp->sd_quota_bitmap[x])
1216 for (x = 0; x < blocks; x++) {
1217 struct buffer_head *bh;
1222 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1227 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1230 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1235 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1237 struct gfs2_quota_change_host qc;
1238 struct gfs2_quota_data *qd;
1240 gfs2_quota_change_in(&qc, bh->b_data +
1241 sizeof(struct gfs2_meta_header) +
1242 y * sizeof(struct gfs2_quota_change));
1246 error = qd_alloc(sdp, qc.qc_id, &qd);
1252 set_bit(QDF_CHANGE, &qd->qd_flags);
1253 qd->qd_change = qc.qc_change;
1255 qd->qd_slot_count = 1;
1257 spin_lock(&qd_lru_lock);
1258 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1259 list_add(&qd->qd_list, &sdp->sd_quota_list);
1260 atomic_inc(&sdp->sd_quota_count);
1261 spin_unlock(&qd_lru_lock);
1272 fs_info(sdp, "found %u quota changes\n", found);
1277 gfs2_quota_cleanup(sdp);
1281 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1283 struct list_head *head = &sdp->sd_quota_list;
1284 struct gfs2_quota_data *qd;
1287 spin_lock(&qd_lru_lock);
1288 while (!list_empty(head)) {
1289 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1291 if (atomic_read(&qd->qd_count) > 1 ||
1292 (atomic_read(&qd->qd_count) &&
1293 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1294 list_move(&qd->qd_list, head);
1295 spin_unlock(&qd_lru_lock);
1297 spin_lock(&qd_lru_lock);
1301 list_del(&qd->qd_list);
1302 /* Also remove if this qd exists in the reclaim list */
1303 if (!list_empty(&qd->qd_reclaim)) {
1304 list_del_init(&qd->qd_reclaim);
1305 atomic_dec(&qd_lru_count);
1307 atomic_dec(&sdp->sd_quota_count);
1308 spin_unlock(&qd_lru_lock);
1310 if (!atomic_read(&qd->qd_count)) {
1311 gfs2_assert_warn(sdp, !qd->qd_change);
1312 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1314 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1315 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1317 gfs2_glock_put(qd->qd_gl);
1318 kmem_cache_free(gfs2_quotad_cachep, qd);
1320 spin_lock(&qd_lru_lock);
1322 spin_unlock(&qd_lru_lock);
1324 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1326 if (sdp->sd_quota_bitmap) {
1327 for (x = 0; x < sdp->sd_quota_chunks; x++)
1328 kfree(sdp->sd_quota_bitmap[x]);
1329 kfree(sdp->sd_quota_bitmap);
1333 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1335 if (error == 0 || error == -EROFS)
1337 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1338 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1341 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1342 int (*fxn)(struct super_block *sb, int type),
1343 unsigned long t, unsigned long *timeo,
1344 unsigned int *new_timeo)
1347 int error = fxn(sdp->sd_vfs, 0);
1348 quotad_error(sdp, msg, error);
1349 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1355 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1357 struct gfs2_inode *ip;
1361 spin_lock(&sdp->sd_trunc_lock);
1362 if (!list_empty(&sdp->sd_trunc_list)) {
1363 ip = list_entry(sdp->sd_trunc_list.next,
1364 struct gfs2_inode, i_trunc_list);
1365 list_del_init(&ip->i_trunc_list);
1367 spin_unlock(&sdp->sd_trunc_lock);
1370 gfs2_glock_finish_truncate(ip);
1374 void gfs2_wake_up_statfs(struct gfs2_sbd *sdp) {
1375 if (!sdp->sd_statfs_force_sync) {
1376 sdp->sd_statfs_force_sync = 1;
1377 wake_up(&sdp->sd_quota_wait);
1383 * gfs2_quotad - Write cached quota changes into the quota file
1384 * @sdp: Pointer to GFS2 superblock
1388 int gfs2_quotad(void *data)
1390 struct gfs2_sbd *sdp = data;
1391 struct gfs2_tune *tune = &sdp->sd_tune;
1392 unsigned long statfs_timeo = 0;
1393 unsigned long quotad_timeo = 0;
1394 unsigned long t = 0;
1398 while (!kthread_should_stop()) {
1400 /* Update the master statfs file */
1401 if (sdp->sd_statfs_force_sync) {
1402 int error = gfs2_statfs_sync(sdp->sd_vfs, 0);
1403 quotad_error(sdp, "statfs", error);
1404 statfs_timeo = gfs2_tune_get(sdp, gt_statfs_quantum) * HZ;
1407 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1409 &tune->gt_statfs_quantum);
1411 /* Update quota file */
1412 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1413 "ad_timeo, &tune->gt_quota_quantum);
1415 /* Check for & recover partially truncated inodes */
1416 quotad_check_trunc_list(sdp);
1420 t = min(quotad_timeo, statfs_timeo);
1422 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1423 spin_lock(&sdp->sd_trunc_lock);
1424 empty = list_empty(&sdp->sd_trunc_list);
1425 spin_unlock(&sdp->sd_trunc_lock);
1426 if (empty && !sdp->sd_statfs_force_sync)
1427 t -= schedule_timeout(t);
1430 finish_wait(&sdp->sd_quota_wait, &wait);
1436 static int gfs2_quota_get_xstate(struct super_block *sb,
1437 struct fs_quota_stat *fqs)
1439 struct gfs2_sbd *sdp = sb->s_fs_info;
1441 memset(fqs, 0, sizeof(struct fs_quota_stat));
1442 fqs->qs_version = FS_QSTAT_VERSION;
1444 switch (sdp->sd_args.ar_quota) {
1446 fqs->qs_flags |= (FS_QUOTA_UDQ_ENFD | FS_QUOTA_GDQ_ENFD);
1448 case GFS2_QUOTA_ACCOUNT:
1449 fqs->qs_flags |= (FS_QUOTA_UDQ_ACCT | FS_QUOTA_GDQ_ACCT);
1451 case GFS2_QUOTA_OFF:
1455 if (sdp->sd_quota_inode) {
1456 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1457 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1459 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1460 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1461 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1465 static int gfs2_get_dqblk(struct super_block *sb, struct kqid qid,
1466 struct fs_disk_quota *fdq)
1468 struct gfs2_sbd *sdp = sb->s_fs_info;
1469 struct gfs2_quota_lvb *qlvb;
1470 struct gfs2_quota_data *qd;
1471 struct gfs2_holder q_gh;
1474 memset(fdq, 0, sizeof(struct fs_disk_quota));
1476 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1477 return -ESRCH; /* Crazy XFS error code */
1479 if ((qid.type != USRQUOTA) &&
1480 (qid.type != GRPQUOTA))
1483 error = qd_get(sdp, qid, &qd);
1486 error = do_glock(qd, FORCE, &q_gh);
1490 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lksb.sb_lvbptr;
1491 fdq->d_version = FS_DQUOT_VERSION;
1492 fdq->d_flags = (qid.type == USRQUOTA) ? FS_USER_QUOTA : FS_GROUP_QUOTA;
1493 fdq->d_id = from_kqid_munged(current_user_ns(), qid);
1494 fdq->d_blk_hardlimit = be64_to_cpu(qlvb->qb_limit) << sdp->sd_fsb2bb_shift;
1495 fdq->d_blk_softlimit = be64_to_cpu(qlvb->qb_warn) << sdp->sd_fsb2bb_shift;
1496 fdq->d_bcount = be64_to_cpu(qlvb->qb_value) << sdp->sd_fsb2bb_shift;
1498 gfs2_glock_dq_uninit(&q_gh);
1504 /* GFS2 only supports a subset of the XFS fields */
1505 #define GFS2_FIELDMASK (FS_DQ_BSOFT|FS_DQ_BHARD|FS_DQ_BCOUNT)
1507 static int gfs2_set_dqblk(struct super_block *sb, struct kqid qid,
1508 struct fs_disk_quota *fdq)
1510 struct gfs2_sbd *sdp = sb->s_fs_info;
1511 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
1512 struct gfs2_quota_data *qd;
1513 struct gfs2_holder q_gh, i_gh;
1514 unsigned int data_blocks, ind_blocks;
1515 unsigned int blocks = 0;
1520 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
1521 return -ESRCH; /* Crazy XFS error code */
1523 if ((qid.type != USRQUOTA) &&
1524 (qid.type != GRPQUOTA))
1527 if (fdq->d_fieldmask & ~GFS2_FIELDMASK)
1530 error = qd_get(sdp, qid, &qd);
1534 error = gfs2_rs_alloc(ip);
1538 mutex_lock(&ip->i_inode.i_mutex);
1539 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE, 0, &q_gh);
1542 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
1546 /* Check for existing entry, if none then alloc new blocks */
1547 error = update_qd(sdp, qd);
1551 /* If nothing has changed, this is a no-op */
1552 if ((fdq->d_fieldmask & FS_DQ_BSOFT) &&
1553 ((fdq->d_blk_softlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_warn)))
1554 fdq->d_fieldmask ^= FS_DQ_BSOFT;
1556 if ((fdq->d_fieldmask & FS_DQ_BHARD) &&
1557 ((fdq->d_blk_hardlimit >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_limit)))
1558 fdq->d_fieldmask ^= FS_DQ_BHARD;
1560 if ((fdq->d_fieldmask & FS_DQ_BCOUNT) &&
1561 ((fdq->d_bcount >> sdp->sd_fsb2bb_shift) == be64_to_cpu(qd->qd_qb.qb_value)))
1562 fdq->d_fieldmask ^= FS_DQ_BCOUNT;
1564 if (fdq->d_fieldmask == 0)
1567 offset = qd2offset(qd);
1568 alloc_required = gfs2_write_alloc_required(ip, offset, sizeof(struct gfs2_quota));
1569 if (gfs2_is_stuffed(ip))
1571 if (alloc_required) {
1572 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
1573 &data_blocks, &ind_blocks);
1574 blocks = 1 + data_blocks + ind_blocks;
1575 error = gfs2_inplace_reserve(ip, blocks, 0);
1578 blocks += gfs2_rg_blocks(ip, blocks);
1581 /* Some quotas span block boundaries and can update two blocks,
1582 adding an extra block to the transaction to handle such quotas */
1583 error = gfs2_trans_begin(sdp, blocks + RES_DINODE + 2, 0);
1588 error = gfs2_adjust_quota(ip, offset, 0, qd, fdq);
1590 gfs2_trans_end(sdp);
1593 gfs2_inplace_release(ip);
1595 gfs2_glock_dq_uninit(&i_gh);
1597 gfs2_glock_dq_uninit(&q_gh);
1599 mutex_unlock(&ip->i_inode.i_mutex);
1605 const struct quotactl_ops gfs2_quotactl_ops = {
1606 .quota_sync = gfs2_quota_sync,
1607 .get_xstate = gfs2_quota_get_xstate,
1608 .get_dqblk = gfs2_get_dqblk,
1609 .set_dqblk = gfs2_set_dqblk,