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>
41 #include <linux/spinlock.h>
42 #include <linux/completion.h>
43 #include <linux/buffer_head.h>
44 #include <linux/sort.h>
46 #include <linux/bio.h>
47 #include <linux/gfs2_ondisk.h>
48 #include <linux/kthread.h>
49 #include <linux/freezer.h>
50 #include <linux/dqblk_xfs.h>
69 struct gfs2_quota_change_host {
71 u32 qc_flags; /* GFS2_QCF_... */
75 static LIST_HEAD(qd_lru_list);
76 static atomic_t qd_lru_count = ATOMIC_INIT(0);
77 static DEFINE_SPINLOCK(qd_lru_lock);
79 int gfs2_shrink_qd_memory(int nr, gfp_t gfp_mask)
81 struct gfs2_quota_data *qd;
87 if (!(gfp_mask & __GFP_FS))
90 spin_lock(&qd_lru_lock);
91 while (nr && !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 qd2offset(struct gfs2_quota_data *qd)
124 offset = 2 * (u64)qd->qd_id + !test_bit(QDF_USER, &qd->qd_flags);
125 offset *= sizeof(struct gfs2_quota);
130 static int qd_alloc(struct gfs2_sbd *sdp, int user, u32 id,
131 struct gfs2_quota_data **qdp)
133 struct gfs2_quota_data *qd;
136 qd = kmem_cache_zalloc(gfs2_quotad_cachep, GFP_NOFS);
140 atomic_set(&qd->qd_count, 1);
143 set_bit(QDF_USER, &qd->qd_flags);
145 INIT_LIST_HEAD(&qd->qd_reclaim);
147 error = gfs2_glock_get(sdp, 2 * (u64)id + !user,
148 &gfs2_quota_glops, CREATE, &qd->qd_gl);
157 kmem_cache_free(gfs2_quotad_cachep, qd);
161 static int qd_get(struct gfs2_sbd *sdp, int user, u32 id,
162 struct gfs2_quota_data **qdp)
164 struct gfs2_quota_data *qd = NULL, *new_qd = NULL;
171 spin_lock(&qd_lru_lock);
172 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
173 if (qd->qd_id == id &&
174 !test_bit(QDF_USER, &qd->qd_flags) == !user) {
175 if (!atomic_read(&qd->qd_count) &&
176 !list_empty(&qd->qd_reclaim)) {
177 /* Remove it from reclaim list */
178 list_del_init(&qd->qd_reclaim);
179 atomic_dec(&qd_lru_count);
181 atomic_inc(&qd->qd_count);
192 list_add(&qd->qd_list, &sdp->sd_quota_list);
193 atomic_inc(&sdp->sd_quota_count);
197 spin_unlock(&qd_lru_lock);
201 gfs2_glock_put(new_qd->qd_gl);
202 kmem_cache_free(gfs2_quotad_cachep, new_qd);
208 error = qd_alloc(sdp, user, id, &new_qd);
214 static void qd_hold(struct gfs2_quota_data *qd)
216 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
217 gfs2_assert(sdp, atomic_read(&qd->qd_count));
218 atomic_inc(&qd->qd_count);
221 static void qd_put(struct gfs2_quota_data *qd)
223 if (atomic_dec_and_lock(&qd->qd_count, &qd_lru_lock)) {
224 /* Add to the reclaim list */
225 list_add_tail(&qd->qd_reclaim, &qd_lru_list);
226 atomic_inc(&qd_lru_count);
227 spin_unlock(&qd_lru_lock);
231 static int slot_get(struct gfs2_quota_data *qd)
233 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
234 unsigned int c, o = 0, b;
235 unsigned char byte = 0;
237 spin_lock(&qd_lru_lock);
239 if (qd->qd_slot_count++) {
240 spin_unlock(&qd_lru_lock);
244 for (c = 0; c < sdp->sd_quota_chunks; c++)
245 for (o = 0; o < PAGE_SIZE; o++) {
246 byte = sdp->sd_quota_bitmap[c][o];
254 for (b = 0; b < 8; b++)
255 if (!(byte & (1 << b)))
257 qd->qd_slot = c * (8 * PAGE_SIZE) + o * 8 + b;
259 if (qd->qd_slot >= sdp->sd_quota_slots)
262 sdp->sd_quota_bitmap[c][o] |= 1 << b;
264 spin_unlock(&qd_lru_lock);
270 spin_unlock(&qd_lru_lock);
274 static void slot_hold(struct gfs2_quota_data *qd)
276 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
278 spin_lock(&qd_lru_lock);
279 gfs2_assert(sdp, qd->qd_slot_count);
281 spin_unlock(&qd_lru_lock);
284 static void slot_put(struct gfs2_quota_data *qd)
286 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
288 spin_lock(&qd_lru_lock);
289 gfs2_assert(sdp, qd->qd_slot_count);
290 if (!--qd->qd_slot_count) {
291 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, qd->qd_slot, 0);
294 spin_unlock(&qd_lru_lock);
297 static int bh_get(struct gfs2_quota_data *qd)
299 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
300 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
301 unsigned int block, offset;
302 struct buffer_head *bh;
304 struct buffer_head bh_map = { .b_state = 0, .b_blocknr = 0 };
306 mutex_lock(&sdp->sd_quota_mutex);
308 if (qd->qd_bh_count++) {
309 mutex_unlock(&sdp->sd_quota_mutex);
313 block = qd->qd_slot / sdp->sd_qc_per_block;
314 offset = qd->qd_slot % sdp->sd_qc_per_block;
316 bh_map.b_size = 1 << ip->i_inode.i_blkbits;
317 error = gfs2_block_map(&ip->i_inode, block, &bh_map, 0);
320 error = gfs2_meta_read(ip->i_gl, bh_map.b_blocknr, DIO_WAIT, &bh);
324 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC))
328 qd->qd_bh_qc = (struct gfs2_quota_change *)
329 (bh->b_data + sizeof(struct gfs2_meta_header) +
330 offset * sizeof(struct gfs2_quota_change));
332 mutex_unlock(&sdp->sd_quota_mutex);
340 mutex_unlock(&sdp->sd_quota_mutex);
344 static void bh_put(struct gfs2_quota_data *qd)
346 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
348 mutex_lock(&sdp->sd_quota_mutex);
349 gfs2_assert(sdp, qd->qd_bh_count);
350 if (!--qd->qd_bh_count) {
355 mutex_unlock(&sdp->sd_quota_mutex);
358 static int qd_fish(struct gfs2_sbd *sdp, struct gfs2_quota_data **qdp)
360 struct gfs2_quota_data *qd = NULL;
366 if (sdp->sd_vfs->s_flags & MS_RDONLY)
369 spin_lock(&qd_lru_lock);
371 list_for_each_entry(qd, &sdp->sd_quota_list, qd_list) {
372 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
373 !test_bit(QDF_CHANGE, &qd->qd_flags) ||
374 qd->qd_sync_gen >= sdp->sd_quota_sync_gen)
377 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
379 set_bit(QDF_LOCKED, &qd->qd_flags);
380 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
381 atomic_inc(&qd->qd_count);
382 qd->qd_change_sync = qd->qd_change;
383 gfs2_assert_warn(sdp, qd->qd_slot_count);
393 spin_unlock(&qd_lru_lock);
396 gfs2_assert_warn(sdp, qd->qd_change_sync);
399 clear_bit(QDF_LOCKED, &qd->qd_flags);
411 static int qd_trylock(struct gfs2_quota_data *qd)
413 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
415 if (sdp->sd_vfs->s_flags & MS_RDONLY)
418 spin_lock(&qd_lru_lock);
420 if (test_bit(QDF_LOCKED, &qd->qd_flags) ||
421 !test_bit(QDF_CHANGE, &qd->qd_flags)) {
422 spin_unlock(&qd_lru_lock);
426 list_move_tail(&qd->qd_list, &sdp->sd_quota_list);
428 set_bit(QDF_LOCKED, &qd->qd_flags);
429 gfs2_assert_warn(sdp, atomic_read(&qd->qd_count));
430 atomic_inc(&qd->qd_count);
431 qd->qd_change_sync = qd->qd_change;
432 gfs2_assert_warn(sdp, qd->qd_slot_count);
435 spin_unlock(&qd_lru_lock);
437 gfs2_assert_warn(sdp, qd->qd_change_sync);
439 clear_bit(QDF_LOCKED, &qd->qd_flags);
448 static void qd_unlock(struct gfs2_quota_data *qd)
450 gfs2_assert_warn(qd->qd_gl->gl_sbd,
451 test_bit(QDF_LOCKED, &qd->qd_flags));
452 clear_bit(QDF_LOCKED, &qd->qd_flags);
458 static int qdsb_get(struct gfs2_sbd *sdp, int user, u32 id,
459 struct gfs2_quota_data **qdp)
463 error = qd_get(sdp, user, id, qdp);
467 error = slot_get(*qdp);
471 error = bh_get(*qdp);
484 static void qdsb_put(struct gfs2_quota_data *qd)
491 int gfs2_quota_hold(struct gfs2_inode *ip, u32 uid, u32 gid)
493 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
494 struct gfs2_alloc *al = ip->i_alloc;
495 struct gfs2_quota_data **qd = al->al_qd;
498 if (gfs2_assert_warn(sdp, !al->al_qd_num) ||
499 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags)))
502 if (sdp->sd_args.ar_quota == GFS2_QUOTA_OFF)
505 error = qdsb_get(sdp, QUOTA_USER, ip->i_inode.i_uid, qd);
511 error = qdsb_get(sdp, QUOTA_GROUP, ip->i_inode.i_gid, qd);
517 if (uid != NO_QUOTA_CHANGE && uid != ip->i_inode.i_uid) {
518 error = qdsb_get(sdp, QUOTA_USER, uid, qd);
525 if (gid != NO_QUOTA_CHANGE && gid != ip->i_inode.i_gid) {
526 error = qdsb_get(sdp, QUOTA_GROUP, gid, qd);
535 gfs2_quota_unhold(ip);
539 void gfs2_quota_unhold(struct gfs2_inode *ip)
541 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
542 struct gfs2_alloc *al = ip->i_alloc;
545 gfs2_assert_warn(sdp, !test_bit(GIF_QD_LOCKED, &ip->i_flags));
547 for (x = 0; x < al->al_qd_num; x++) {
548 qdsb_put(al->al_qd[x]);
554 static int sort_qd(const void *a, const void *b)
556 const struct gfs2_quota_data *qd_a = *(const struct gfs2_quota_data **)a;
557 const struct gfs2_quota_data *qd_b = *(const struct gfs2_quota_data **)b;
559 if (!test_bit(QDF_USER, &qd_a->qd_flags) !=
560 !test_bit(QDF_USER, &qd_b->qd_flags)) {
561 if (test_bit(QDF_USER, &qd_a->qd_flags))
566 if (qd_a->qd_id < qd_b->qd_id)
568 if (qd_a->qd_id > qd_b->qd_id)
574 static void do_qc(struct gfs2_quota_data *qd, s64 change)
576 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
577 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
578 struct gfs2_quota_change *qc = qd->qd_bh_qc;
581 mutex_lock(&sdp->sd_quota_mutex);
582 gfs2_trans_add_bh(ip->i_gl, qd->qd_bh, 1);
584 if (!test_bit(QDF_CHANGE, &qd->qd_flags)) {
587 if (test_bit(QDF_USER, &qd->qd_flags))
588 qc->qc_flags = cpu_to_be32(GFS2_QCF_USER);
589 qc->qc_id = cpu_to_be32(qd->qd_id);
592 x = be64_to_cpu(qc->qc_change) + change;
593 qc->qc_change = cpu_to_be64(x);
595 spin_lock(&qd_lru_lock);
597 spin_unlock(&qd_lru_lock);
600 gfs2_assert_warn(sdp, test_bit(QDF_CHANGE, &qd->qd_flags));
601 clear_bit(QDF_CHANGE, &qd->qd_flags);
606 } else if (!test_and_set_bit(QDF_CHANGE, &qd->qd_flags)) {
611 mutex_unlock(&sdp->sd_quota_mutex);
615 * gfs2_adjust_quota - adjust record of current block usage
616 * @ip: The quota inode
617 * @loc: Offset of the entry in the quota file
618 * @change: The amount of change to record
619 * @qd: The quota data
621 * This function was mostly borrowed from gfs2_block_truncate_page which was
622 * in turn mostly borrowed from ext3
624 * Returns: 0 or -ve on error
627 static int gfs2_adjust_quota(struct gfs2_inode *ip, loff_t loc,
628 s64 change, struct gfs2_quota_data *qd)
630 struct inode *inode = &ip->i_inode;
631 struct address_space *mapping = inode->i_mapping;
632 unsigned long index = loc >> PAGE_CACHE_SHIFT;
633 unsigned offset = loc & (PAGE_CACHE_SIZE - 1);
634 unsigned blocksize, iblock, pos;
635 struct buffer_head *bh;
638 struct gfs2_quota *qp;
642 if (gfs2_is_stuffed(ip))
643 gfs2_unstuff_dinode(ip, NULL);
645 page = grab_cache_page(mapping, index);
649 blocksize = inode->i_sb->s_blocksize;
650 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
652 if (!page_has_buffers(page))
653 create_empty_buffers(page, blocksize, 0);
655 bh = page_buffers(page);
657 while (offset >= pos) {
658 bh = bh->b_this_page;
663 if (!buffer_mapped(bh)) {
664 gfs2_block_map(inode, iblock, bh, 1);
665 if (!buffer_mapped(bh))
669 if (PageUptodate(page))
670 set_buffer_uptodate(bh);
672 if (!buffer_uptodate(bh)) {
673 ll_rw_block(READ_META, 1, &bh);
675 if (!buffer_uptodate(bh))
679 gfs2_trans_add_bh(ip->i_gl, bh, 0);
681 kaddr = kmap_atomic(page, KM_USER0);
683 value = (s64)be64_to_cpu(qp->qu_value) + change;
684 qp->qu_value = cpu_to_be64(value);
685 qd->qd_qb.qb_value = qp->qu_value;
686 flush_dcache_page(page);
687 kunmap_atomic(kaddr, KM_USER0);
691 page_cache_release(page);
695 static int do_sync(unsigned int num_qd, struct gfs2_quota_data **qda)
697 struct gfs2_sbd *sdp = (*qda)->qd_gl->gl_sbd;
698 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
699 unsigned int data_blocks, ind_blocks;
700 struct gfs2_holder *ghs, i_gh;
702 struct gfs2_quota_data *qd;
704 unsigned int nalloc = 0, blocks;
705 struct gfs2_alloc *al = NULL;
708 gfs2_write_calc_reserv(ip, sizeof(struct gfs2_quota),
709 &data_blocks, &ind_blocks);
711 ghs = kcalloc(num_qd, sizeof(struct gfs2_holder), GFP_NOFS);
715 sort(qda, num_qd, sizeof(struct gfs2_quota_data *), sort_qd, NULL);
716 for (qx = 0; qx < num_qd; qx++) {
717 error = gfs2_glock_nq_init(qda[qx]->qd_gl, LM_ST_EXCLUSIVE,
718 GL_NOCACHE, &ghs[qx]);
723 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &i_gh);
727 for (x = 0; x < num_qd; x++) {
730 offset = qd2offset(qda[x]);
731 error = gfs2_write_alloc_required(ip, offset,
732 sizeof(struct gfs2_quota),
740 al = gfs2_alloc_get(ip);
746 * 1 blk for unstuffing inode if stuffed. We add this extra
747 * block to the reservation unconditionally. If the inode
748 * doesn't need unstuffing, the block will be released to the
749 * rgrp since it won't be allocated during the transaction
751 al->al_requested = 1;
752 /* +1 in the end for block requested above for unstuffing */
753 blocks = num_qd * data_blocks + RES_DINODE + num_qd + 1;
756 al->al_requested += nalloc * (data_blocks + ind_blocks);
757 error = gfs2_inplace_reserve(ip);
762 blocks += al->al_rgd->rd_length + nalloc * ind_blocks + RES_STATFS;
764 error = gfs2_trans_begin(sdp, blocks, 0);
768 for (x = 0; x < num_qd; x++) {
770 offset = qd2offset(qd);
771 error = gfs2_adjust_quota(ip, offset, qd->qd_change_sync,
772 (struct gfs2_quota_data *)qd);
776 do_qc(qd, -qd->qd_change_sync);
784 gfs2_inplace_release(ip);
788 gfs2_glock_dq_uninit(&i_gh);
791 gfs2_glock_dq_uninit(&ghs[qx]);
793 gfs2_log_flush(ip->i_gl->gl_sbd, ip->i_gl);
797 static int do_glock(struct gfs2_quota_data *qd, int force_refresh,
798 struct gfs2_holder *q_gh)
800 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
801 struct gfs2_inode *ip = GFS2_I(sdp->sd_quota_inode);
802 struct gfs2_holder i_gh;
805 struct gfs2_quota_lvb *qlvb;
808 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_SHARED, 0, q_gh);
812 qd->qd_qb = *(struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
814 if (force_refresh || qd->qd_qb.qb_magic != cpu_to_be32(GFS2_MAGIC)) {
816 gfs2_glock_dq_uninit(q_gh);
817 error = gfs2_glock_nq_init(qd->qd_gl, LM_ST_EXCLUSIVE,
822 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &i_gh);
826 memset(&q, 0, sizeof(struct gfs2_quota));
828 error = gfs2_internal_read(ip, NULL, (char *)&q, &pos, sizeof(q));
831 if ((error < sizeof(q)) && force_refresh) {
835 gfs2_glock_dq_uninit(&i_gh);
837 qlvb = (struct gfs2_quota_lvb *)qd->qd_gl->gl_lvb;
838 qlvb->qb_magic = cpu_to_be32(GFS2_MAGIC);
840 qlvb->qb_limit = q.qu_limit;
841 qlvb->qb_warn = q.qu_warn;
842 qlvb->qb_value = q.qu_value;
845 gfs2_glock_dq_uninit(q_gh);
853 gfs2_glock_dq_uninit(&i_gh);
855 gfs2_glock_dq_uninit(q_gh);
859 int gfs2_quota_lock(struct gfs2_inode *ip, u32 uid, u32 gid)
861 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
862 struct gfs2_alloc *al = ip->i_alloc;
866 gfs2_quota_hold(ip, uid, gid);
868 if (capable(CAP_SYS_RESOURCE) ||
869 sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
872 sort(al->al_qd, al->al_qd_num, sizeof(struct gfs2_quota_data *),
875 for (x = 0; x < al->al_qd_num; x++) {
876 error = do_glock(al->al_qd[x], NO_FORCE, &al->al_qd_ghs[x]);
882 set_bit(GIF_QD_LOCKED, &ip->i_flags);
885 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
886 gfs2_quota_unhold(ip);
892 static int need_sync(struct gfs2_quota_data *qd)
894 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
895 struct gfs2_tune *gt = &sdp->sd_tune;
897 unsigned int num, den;
900 if (!qd->qd_qb.qb_limit)
903 spin_lock(&qd_lru_lock);
904 value = qd->qd_change;
905 spin_unlock(&qd_lru_lock);
907 spin_lock(>->gt_spin);
908 num = gt->gt_quota_scale_num;
909 den = gt->gt_quota_scale_den;
910 spin_unlock(>->gt_spin);
914 else if ((s64)be64_to_cpu(qd->qd_qb.qb_value) >=
915 (s64)be64_to_cpu(qd->qd_qb.qb_limit))
918 value *= gfs2_jindex_size(sdp) * num;
919 value = div_s64(value, den);
920 value += (s64)be64_to_cpu(qd->qd_qb.qb_value);
921 if (value < (s64)be64_to_cpu(qd->qd_qb.qb_limit))
928 void gfs2_quota_unlock(struct gfs2_inode *ip)
930 struct gfs2_alloc *al = ip->i_alloc;
931 struct gfs2_quota_data *qda[4];
932 unsigned int count = 0;
935 if (!test_and_clear_bit(GIF_QD_LOCKED, &ip->i_flags))
938 for (x = 0; x < al->al_qd_num; x++) {
939 struct gfs2_quota_data *qd;
943 sync = need_sync(qd);
945 gfs2_glock_dq_uninit(&al->al_qd_ghs[x]);
947 if (sync && qd_trylock(qd))
953 for (x = 0; x < count; x++)
958 gfs2_quota_unhold(ip);
963 static int print_message(struct gfs2_quota_data *qd, char *type)
965 struct gfs2_sbd *sdp = qd->qd_gl->gl_sbd;
967 printk(KERN_INFO "GFS2: fsid=%s: quota %s for %s %u\r\n",
968 sdp->sd_fsname, type,
969 (test_bit(QDF_USER, &qd->qd_flags)) ? "user" : "group",
975 int gfs2_quota_check(struct gfs2_inode *ip, u32 uid, u32 gid)
977 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
978 struct gfs2_alloc *al = ip->i_alloc;
979 struct gfs2_quota_data *qd;
984 if (!test_bit(GIF_QD_LOCKED, &ip->i_flags))
987 if (sdp->sd_args.ar_quota != GFS2_QUOTA_ON)
990 for (x = 0; x < al->al_qd_num; x++) {
993 if (!((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
994 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))))
997 value = (s64)be64_to_cpu(qd->qd_qb.qb_value);
998 spin_lock(&qd_lru_lock);
999 value += qd->qd_change;
1000 spin_unlock(&qd_lru_lock);
1002 if (be64_to_cpu(qd->qd_qb.qb_limit) && (s64)be64_to_cpu(qd->qd_qb.qb_limit) < value) {
1003 print_message(qd, "exceeded");
1006 } else if (be64_to_cpu(qd->qd_qb.qb_warn) &&
1007 (s64)be64_to_cpu(qd->qd_qb.qb_warn) < value &&
1008 time_after_eq(jiffies, qd->qd_last_warn +
1010 gt_quota_warn_period) * HZ)) {
1011 error = print_message(qd, "warning");
1012 qd->qd_last_warn = jiffies;
1019 void gfs2_quota_change(struct gfs2_inode *ip, s64 change,
1022 struct gfs2_alloc *al = ip->i_alloc;
1023 struct gfs2_quota_data *qd;
1026 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), change))
1028 if (ip->i_diskflags & GFS2_DIF_SYSTEM)
1031 for (x = 0; x < al->al_qd_num; x++) {
1034 if ((qd->qd_id == uid && test_bit(QDF_USER, &qd->qd_flags)) ||
1035 (qd->qd_id == gid && !test_bit(QDF_USER, &qd->qd_flags))) {
1041 int gfs2_quota_sync(struct super_block *sb, int type)
1043 struct gfs2_sbd *sdp = sb->s_fs_info;
1044 struct gfs2_quota_data **qda;
1045 unsigned int max_qd = gfs2_tune_get(sdp, gt_quota_simul_sync);
1046 unsigned int num_qd;
1050 sdp->sd_quota_sync_gen++;
1052 qda = kcalloc(max_qd, sizeof(struct gfs2_quota_data *), GFP_KERNEL);
1060 error = qd_fish(sdp, qda + num_qd);
1061 if (error || !qda[num_qd])
1063 if (++num_qd == max_qd)
1069 error = do_sync(num_qd, qda);
1071 for (x = 0; x < num_qd; x++)
1072 qda[x]->qd_sync_gen =
1073 sdp->sd_quota_sync_gen;
1075 for (x = 0; x < num_qd; x++)
1078 } while (!error && num_qd == max_qd);
1085 int gfs2_quota_refresh(struct gfs2_sbd *sdp, int user, u32 id)
1087 struct gfs2_quota_data *qd;
1088 struct gfs2_holder q_gh;
1091 error = qd_get(sdp, user, id, &qd);
1095 error = do_glock(qd, FORCE, &q_gh);
1097 gfs2_glock_dq_uninit(&q_gh);
1103 static void gfs2_quota_change_in(struct gfs2_quota_change_host *qc, const void *buf)
1105 const struct gfs2_quota_change *str = buf;
1107 qc->qc_change = be64_to_cpu(str->qc_change);
1108 qc->qc_flags = be32_to_cpu(str->qc_flags);
1109 qc->qc_id = be32_to_cpu(str->qc_id);
1112 int gfs2_quota_init(struct gfs2_sbd *sdp)
1114 struct gfs2_inode *ip = GFS2_I(sdp->sd_qc_inode);
1115 unsigned int blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift;
1116 unsigned int x, slot = 0;
1117 unsigned int found = 0;
1122 if (!ip->i_disksize || ip->i_disksize > (64 << 20) ||
1123 ip->i_disksize & (sdp->sd_sb.sb_bsize - 1)) {
1124 gfs2_consist_inode(ip);
1127 sdp->sd_quota_slots = blocks * sdp->sd_qc_per_block;
1128 sdp->sd_quota_chunks = DIV_ROUND_UP(sdp->sd_quota_slots, 8 * PAGE_SIZE);
1132 sdp->sd_quota_bitmap = kcalloc(sdp->sd_quota_chunks,
1133 sizeof(unsigned char *), GFP_NOFS);
1134 if (!sdp->sd_quota_bitmap)
1137 for (x = 0; x < sdp->sd_quota_chunks; x++) {
1138 sdp->sd_quota_bitmap[x] = kzalloc(PAGE_SIZE, GFP_NOFS);
1139 if (!sdp->sd_quota_bitmap[x])
1143 for (x = 0; x < blocks; x++) {
1144 struct buffer_head *bh;
1149 error = gfs2_extent_map(&ip->i_inode, x, &new, &dblock, &extlen);
1154 bh = gfs2_meta_ra(ip->i_gl, dblock, extlen);
1157 if (gfs2_metatype_check(sdp, bh, GFS2_METATYPE_QC)) {
1162 for (y = 0; y < sdp->sd_qc_per_block && slot < sdp->sd_quota_slots;
1164 struct gfs2_quota_change_host qc;
1165 struct gfs2_quota_data *qd;
1167 gfs2_quota_change_in(&qc, bh->b_data +
1168 sizeof(struct gfs2_meta_header) +
1169 y * sizeof(struct gfs2_quota_change));
1173 error = qd_alloc(sdp, (qc.qc_flags & GFS2_QCF_USER),
1180 set_bit(QDF_CHANGE, &qd->qd_flags);
1181 qd->qd_change = qc.qc_change;
1183 qd->qd_slot_count = 1;
1185 spin_lock(&qd_lru_lock);
1186 gfs2_icbit_munge(sdp, sdp->sd_quota_bitmap, slot, 1);
1187 list_add(&qd->qd_list, &sdp->sd_quota_list);
1188 atomic_inc(&sdp->sd_quota_count);
1189 spin_unlock(&qd_lru_lock);
1200 fs_info(sdp, "found %u quota changes\n", found);
1205 gfs2_quota_cleanup(sdp);
1209 void gfs2_quota_cleanup(struct gfs2_sbd *sdp)
1211 struct list_head *head = &sdp->sd_quota_list;
1212 struct gfs2_quota_data *qd;
1215 spin_lock(&qd_lru_lock);
1216 while (!list_empty(head)) {
1217 qd = list_entry(head->prev, struct gfs2_quota_data, qd_list);
1219 if (atomic_read(&qd->qd_count) > 1 ||
1220 (atomic_read(&qd->qd_count) &&
1221 !test_bit(QDF_CHANGE, &qd->qd_flags))) {
1222 list_move(&qd->qd_list, head);
1223 spin_unlock(&qd_lru_lock);
1225 spin_lock(&qd_lru_lock);
1229 list_del(&qd->qd_list);
1230 /* Also remove if this qd exists in the reclaim list */
1231 if (!list_empty(&qd->qd_reclaim)) {
1232 list_del_init(&qd->qd_reclaim);
1233 atomic_dec(&qd_lru_count);
1235 atomic_dec(&sdp->sd_quota_count);
1236 spin_unlock(&qd_lru_lock);
1238 if (!atomic_read(&qd->qd_count)) {
1239 gfs2_assert_warn(sdp, !qd->qd_change);
1240 gfs2_assert_warn(sdp, !qd->qd_slot_count);
1242 gfs2_assert_warn(sdp, qd->qd_slot_count == 1);
1243 gfs2_assert_warn(sdp, !qd->qd_bh_count);
1245 gfs2_glock_put(qd->qd_gl);
1246 kmem_cache_free(gfs2_quotad_cachep, qd);
1248 spin_lock(&qd_lru_lock);
1250 spin_unlock(&qd_lru_lock);
1252 gfs2_assert_warn(sdp, !atomic_read(&sdp->sd_quota_count));
1254 if (sdp->sd_quota_bitmap) {
1255 for (x = 0; x < sdp->sd_quota_chunks; x++)
1256 kfree(sdp->sd_quota_bitmap[x]);
1257 kfree(sdp->sd_quota_bitmap);
1261 static void quotad_error(struct gfs2_sbd *sdp, const char *msg, int error)
1263 if (error == 0 || error == -EROFS)
1265 if (!test_bit(SDF_SHUTDOWN, &sdp->sd_flags))
1266 fs_err(sdp, "gfs2_quotad: %s error %d\n", msg, error);
1269 static void quotad_check_timeo(struct gfs2_sbd *sdp, const char *msg,
1270 int (*fxn)(struct super_block *sb, int type),
1271 unsigned long t, unsigned long *timeo,
1272 unsigned int *new_timeo)
1275 int error = fxn(sdp->sd_vfs, 0);
1276 quotad_error(sdp, msg, error);
1277 *timeo = gfs2_tune_get_i(&sdp->sd_tune, new_timeo) * HZ;
1283 static void quotad_check_trunc_list(struct gfs2_sbd *sdp)
1285 struct gfs2_inode *ip;
1289 spin_lock(&sdp->sd_trunc_lock);
1290 if (!list_empty(&sdp->sd_trunc_list)) {
1291 ip = list_entry(sdp->sd_trunc_list.next,
1292 struct gfs2_inode, i_trunc_list);
1293 list_del_init(&ip->i_trunc_list);
1295 spin_unlock(&sdp->sd_trunc_lock);
1298 gfs2_glock_finish_truncate(ip);
1303 * gfs2_quotad - Write cached quota changes into the quota file
1304 * @sdp: Pointer to GFS2 superblock
1308 int gfs2_quotad(void *data)
1310 struct gfs2_sbd *sdp = data;
1311 struct gfs2_tune *tune = &sdp->sd_tune;
1312 unsigned long statfs_timeo = 0;
1313 unsigned long quotad_timeo = 0;
1314 unsigned long t = 0;
1318 while (!kthread_should_stop()) {
1320 /* Update the master statfs file */
1321 quotad_check_timeo(sdp, "statfs", gfs2_statfs_sync, t,
1322 &statfs_timeo, &tune->gt_statfs_quantum);
1324 /* Update quota file */
1325 quotad_check_timeo(sdp, "sync", gfs2_quota_sync, t,
1326 "ad_timeo, &tune->gt_quota_quantum);
1328 /* Check for & recover partially truncated inodes */
1329 quotad_check_trunc_list(sdp);
1331 if (freezing(current))
1333 t = min(quotad_timeo, statfs_timeo);
1335 prepare_to_wait(&sdp->sd_quota_wait, &wait, TASK_INTERRUPTIBLE);
1336 spin_lock(&sdp->sd_trunc_lock);
1337 empty = list_empty(&sdp->sd_trunc_list);
1338 spin_unlock(&sdp->sd_trunc_lock);
1340 t -= schedule_timeout(t);
1343 finish_wait(&sdp->sd_quota_wait, &wait);
1349 static int gfs2_quota_get_xstate(struct super_block *sb,
1350 struct fs_quota_stat *fqs)
1352 struct gfs2_sbd *sdp = sb->s_fs_info;
1354 memset(fqs, 0, sizeof(struct fs_quota_stat));
1355 fqs->qs_version = FS_QSTAT_VERSION;
1356 if (sdp->sd_args.ar_quota == GFS2_QUOTA_ON)
1357 fqs->qs_flags = (XFS_QUOTA_UDQ_ENFD | XFS_QUOTA_GDQ_ENFD);
1358 else if (sdp->sd_args.ar_quota == GFS2_QUOTA_ACCOUNT)
1359 fqs->qs_flags = (XFS_QUOTA_UDQ_ACCT | XFS_QUOTA_GDQ_ACCT);
1360 if (sdp->sd_quota_inode) {
1361 fqs->qs_uquota.qfs_ino = GFS2_I(sdp->sd_quota_inode)->i_no_addr;
1362 fqs->qs_uquota.qfs_nblks = sdp->sd_quota_inode->i_blocks;
1364 fqs->qs_uquota.qfs_nextents = 1; /* unsupported */
1365 fqs->qs_gquota = fqs->qs_uquota; /* its the same inode in both cases */
1366 fqs->qs_incoredqs = atomic_read(&qd_lru_count);
1370 const struct quotactl_ops gfs2_quotactl_ops = {
1371 .quota_sync = gfs2_quota_sync,
1372 .get_xstate = gfs2_quota_get_xstate,