4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2012, 2015, Intel Corporation.
25 * Use is subject to license terms.
27 * Author: Johann Lombardi <johann@whamcloud.com>
28 * Author: Niu Yawei <niu@whamcloud.com>
31 #include <lustre_quota.h>
32 #include "osd_internal.h"
35 * Helpers function to find out the quota type (USRQUOTA/GRPQUOTA) of a
38 static inline int fid2type(const struct lu_fid *fid)
40 LASSERT(fid_is_acct(fid));
41 switch (fid_oid(fid)) {
46 case ACCT_PROJECT_OID:
50 LASSERTF(0, "invalid fid for quota type: %u", fid_oid(fid));
55 * Space Accounting Management
59 * Look up an accounting object based on its fid.
61 * \param info - is the osd thread info passed by the caller
62 * \param osd - is the osd device
63 * \param fid - is the fid of the accounting object we want to look up
64 * \param id - is the osd_inode_id struct to fill with the inode number of
65 * the quota file if the lookup is successful
67 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
68 const struct lu_fid *fid, struct osd_inode_id *id)
70 struct super_block *sb = osd_sb(osd);
73 LASSERT(fid_is_acct(fid));
75 if (!LDISKFS_HAS_RO_COMPAT_FEATURE(sb,
76 LDISKFS_FEATURE_RO_COMPAT_QUOTA))
79 id->oii_gen = OSD_OII_NOGEN;
80 switch (fid2type(fid)) {
83 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum);
87 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum);
90 #ifdef HAVE_PROJECT_QUOTA
91 if (LDISKFS_HAS_RO_COMPAT_FEATURE(sb,
92 LDISKFS_FEATURE_RO_COMPAT_PROJECT))
94 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum);
101 if (!ldiskfs_valid_inum(sb, id->oii_ino) &&
102 id->oii_ino != LDISKFS_USR_QUOTA_INO &&
103 id->oii_ino != LDISKFS_GRP_QUOTA_INO)
110 * Return space usage (#blocks & #inodes) consumed by a given uid or gid.
112 * \param env - is the environment passed by the caller
113 * \param dtobj - is the accounting object
114 * \param dtrec - is the record to fill with space usage information
115 * \param dtkey - is the id of the user or group for which we would
116 * like to access disk usage.
118 * \retval +ve - success : exact match
119 * \retval -ve - failure
121 static int osd_acct_index_lookup(const struct lu_env *env,
122 struct dt_object *dtobj,
123 struct dt_rec *dtrec,
124 const struct dt_key *dtkey)
126 struct osd_thread_info *info = osd_oti_get(env);
127 #if defined(HAVE_DQUOT_QC_DQBLK)
128 struct qc_dqblk *dqblk = &info->oti_qdq;
129 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
130 struct fs_disk_quota *dqblk = &info->oti_fdq;
132 struct if_dqblk *dqblk = &info->oti_dqblk;
134 struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
135 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
136 __u64 id = *((__u64 *)dtkey);
138 #ifdef HAVE_DQUOT_KQID
145 type = fid2type(lu_object_fid(&dtobj->do_lu));
146 memset(dqblk, 0, sizeof(*dqblk));
147 #ifdef HAVE_DQUOT_KQID
148 qid = make_kqid(&init_user_ns, type, id);
149 rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
151 rc = sb->s_qcop->get_dqblk(sb, type, (qid_t) id, dqblk);
155 #if defined(HAVE_DQUOT_QC_DQBLK)
156 rec->bspace = dqblk->d_space;
157 rec->ispace = dqblk->d_ino_count;
158 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
159 rec->bspace = dqblk->d_bcount;
160 rec->ispace = dqblk->d_icount;
162 rec->bspace = dqblk->dqb_curspace;
163 rec->ispace = dqblk->dqb_curinodes;
168 #define QUOTA_IT_READ_ERROR(it, rc) \
169 CERROR("%s: Error while trying to read quota information, " \
170 "failed with %d\n", \
171 osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
174 * Initialize osd Iterator for given osd index object.
176 * \param dt - osd index object
177 * \param attr - not used
179 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
180 struct dt_object *dt,
183 struct osd_it_quota *it;
184 struct lu_object *lo = &dt->do_lu;
185 struct osd_object *obj = osd_dt_obj(dt);
189 LASSERT(lu_object_exists(lo));
193 RETURN(ERR_PTR(-ENOMEM));
197 INIT_LIST_HEAD(&it->oiq_list);
199 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
200 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
202 /* NB: we don't need to store the tree depth since it is always
203 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
205 RETURN((struct dt_it *)it);
209 * Free given iterator.
211 * \param di - osd iterator
213 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
215 struct osd_it_quota *it = (struct osd_it_quota *)di;
216 struct osd_quota_leaf *leaf, *tmp;
219 osd_object_put(env, it->oiq_obj);
221 list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
222 list_del_init(&leaf->oql_link);
232 * Move Iterator to record specified by \a key, if the \a key isn't found,
233 * move to the first valid record.
235 * \param di - osd iterator
236 * \param key - uid or gid
238 * \retval +ve - di points to the first valid record
239 * \retval +1 - di points to exact matched key
240 * \retval -ve - failure
242 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
243 const struct dt_key *key)
245 struct osd_it_quota *it = (struct osd_it_quota *)di;
246 const struct lu_fid *fid =
247 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
249 qid_t dqid = *(qid_t *)key;
254 type = fid2type(fid);
256 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
257 LUSTRE_DQTREEOFF, 0, it);
258 if (offset > 0) { /* Found */
260 } else if (offset < 0) { /* Error */
261 QUOTA_IT_READ_ERROR(it, (int)offset);
265 /* The @key is not found, move to the first valid entry */
266 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
279 * \param di - osd iterator
281 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
286 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
288 struct osd_quota_leaf *leaf;
293 INIT_LIST_HEAD(&leaf->oql_link);
294 leaf->oql_blk = it->oiq_blk[depth];
295 list_add_tail(&leaf->oql_link, &it->oiq_list);
300 * Move on to the next valid entry.
302 * \param di - osd iterator
304 * \retval +ve - iterator reached the end
305 * \retval 0 - iterator has not reached the end yet
306 * \retval -ve - unexpected failure
308 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
310 struct osd_it_quota *it = (struct osd_it_quota *)di;
311 const struct lu_fid *fid =
312 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
319 type = fid2type(fid);
321 /* Let's first check if there are any remaining valid entry in the
322 * current leaf block. Start with the next entry after the current one.
324 depth = LUSTRE_DQTREEDEPTH;
325 index = it->oiq_index[depth];
326 if (++index < LUSTRE_DQSTRINBLK) {
327 /* Search for the next valid entry from current index */
328 rc = walk_block_dqentry(env, it->oiq_obj, type,
329 it->oiq_blk[depth], index, it);
331 QUOTA_IT_READ_ERROR(it, rc);
333 } else if (rc == 0) {
334 /* Found on entry, @it is already updated to the
335 * new position in walk_block_dqentry(). */
338 rc = osd_it_add_processed(it, depth);
343 rc = osd_it_add_processed(it, depth);
349 /* We have consumed all the entries of the current leaf block, move on
350 * to the next one. */
353 /* We keep searching as long as walk_tree_dqentry() returns +1
354 * (= no valid entry found). */
355 for (; depth >= 0 && rc > 0; depth--) {
356 index = it->oiq_index[depth];
359 rc = walk_tree_dqentry(env, it->oiq_obj, type,
360 it->oiq_blk[depth], depth, index, it);
364 QUOTA_IT_READ_ERROR(it, rc);
369 * Return pointer to the key under iterator.
371 * \param di - osd iterator
373 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
374 const struct dt_it *di)
376 struct osd_it_quota *it = (struct osd_it_quota *)di;
379 RETURN((struct dt_key *)&it->oiq_id);
383 * Return size of key under iterator (in bytes)
385 * \param di - osd iterator
387 static int osd_it_acct_key_size(const struct lu_env *env,
388 const struct dt_it *di)
390 struct osd_it_quota *it = (struct osd_it_quota *)di;
393 RETURN((int)sizeof(it->oiq_id));
397 * Return pointer to the record under iterator.
399 * \param di - osd iterator
400 * \param attr - not used
402 static int osd_it_acct_rec(const struct lu_env *env,
403 const struct dt_it *di,
404 struct dt_rec *dtrec, __u32 attr)
406 struct osd_it_quota *it = (struct osd_it_quota *)di;
407 const struct dt_key *key = osd_it_acct_key(env, di);
412 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
413 RETURN(rc > 0 ? 0 : rc);
417 * Returns cookie for current Iterator position.
419 * \param di - osd iterator
421 static __u64 osd_it_acct_store(const struct lu_env *env,
422 const struct dt_it *di)
424 struct osd_it_quota *it = (struct osd_it_quota *)di;
431 * Restore iterator from cookie. if the \a hash isn't found,
432 * restore the first valid record.
434 * \param di - osd iterator
435 * \param hash - iterator location cookie
437 * \retval +ve - di points to the first valid record
438 * \retval +1 - di points to exact matched hash
439 * \retval -ve - failure
441 static int osd_it_acct_load(const struct lu_env *env,
442 const struct dt_it *di, __u64 hash)
445 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
446 (const struct dt_key *)&hash));
450 * Index and Iterator operations for accounting objects
452 const struct dt_index_operations osd_acct_index_ops = {
453 .dio_lookup = osd_acct_index_lookup,
455 .init = osd_it_acct_init,
456 .fini = osd_it_acct_fini,
457 .get = osd_it_acct_get,
458 .put = osd_it_acct_put,
459 .next = osd_it_acct_next,
460 .key = osd_it_acct_key,
461 .key_size = osd_it_acct_key_size,
462 .rec = osd_it_acct_rec,
463 .store = osd_it_acct_store,
464 .load = osd_it_acct_load
468 static inline void osd_quota_swab(char *ptr, size_t size)
472 LASSERT((size & (sizeof(__u64) - 1)) == 0);
474 for (offset = 0; offset < size; offset += sizeof(__u64))
475 __swab64s((__u64 *)(ptr + offset));
478 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
479 const struct dt_rec *rec,
480 union lquota_rec *quota_rec)
483 struct iam_descr *descr;
485 LASSERT(obj->oo_dir != NULL);
486 descr = obj->oo_dir->od_container.ic_descr;
488 memcpy(quota_rec, rec, descr->id_rec_size);
490 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
491 return (const struct dt_rec *)quota_rec;
497 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
500 struct iam_descr *descr;
502 LASSERT(obj->oo_dir != NULL);
503 descr = obj->oo_dir->od_container.ic_descr;
505 osd_quota_swab((char *)rec, descr->id_rec_size);
511 static inline int osd_qid_type(struct osd_thandle *oh, int i)
513 return oh->ot_id_types[i];
517 * Reserve journal credits for quota files update first, then call
518 * ->op_begin() to perform quota enforcement.
520 * \param env - the environment passed by the caller
521 * \param oh - osd transaction handle
522 * \param qi - quota id & space required for this operation
523 * \param obj - osd object, could be NULL when it's under create
524 * \param enforce - whether to perform quota enforcement
525 * \param flags - if the operation is write, return no user quota, no
526 * group quota, or sync commit flags to the caller
528 * \retval 0 - success
529 * \retval -ve - failure
531 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
532 struct lquota_id_info *qi, struct osd_object *obj,
533 bool enforce, int *flags)
535 struct osd_device *dev;
536 struct qsd_instance *qsd;
537 struct inode *inode = NULL;
543 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
546 dev = osd_dt_dev(oh->ot_super.th_dev);
547 LASSERT(dev != NULL);
549 qsd = dev->od_quota_slave;
551 for (i = 0; i < oh->ot_id_cnt; i++) {
552 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
553 oh->ot_id_types[i] == qi->lqi_type) {
560 /* we need to account for credits for this new ID */
561 if (i >= OSD_MAX_UGID_CNT) {
562 CERROR("Too many(%d) trans qids!\n", i + 1);
567 inode = obj->oo_inode;
569 /* root ID entry should be always present in the quota file */
570 if (qi->lqi_id.qid_uid == 0) {
573 /* used space for this ID could be dropped to zero,
574 * reserve extra credits for removing ID entry from
576 if (qi->lqi_space < 0)
577 crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
578 /* reserve credits for adding ID entry to the quota
579 * file if the i_dquot isn't initialized yet. */
580 else if (inode == NULL ||
581 #ifdef HAVE_EXT4_INFO_DQUOT
582 LDISKFS_I(inode)->i_dquot[qi->lqi_type] == NULL)
584 inode->i_dquot[qi->lqi_type] == NULL)
586 crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
591 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
593 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
594 oh->ot_id_types[i] = qi->lqi_type;
598 if (unlikely(qsd == NULL))
599 /* quota slave instance hasn't been allocated yet */
604 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
609 * Wrapper for osd_declare_qid()
611 * \param env - the environment passed by the caller
612 * \param uid - user id of the inode
613 * \param gid - group id of the inode
614 * \param space - how many blocks/inodes will be consumed/released
615 * \param oh - osd transaction handle
616 * \param obj - osd object, could be NULL when it's under create
617 * \param flags - if the operation is write, return no user quota, no
618 * group quota, or sync commit flags to the caller
619 * \param osd_qid_flags - indicate this is a inode/block accounting
620 * and whether changes are performed by root user
622 * \retval 0 - success
623 * \retval -ve - failure
625 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
626 __u32 projid, long long space, struct osd_thandle *oh,
627 struct osd_object *obj, int *flags,
628 enum osd_qid_declare_flags osd_qid_declare_flags)
630 struct osd_thread_info *info = osd_oti_get(env);
631 struct lquota_id_info *qi = &info->oti_qi;
632 int rcu, rcg, rcp = 0; /* user & group & project rc */
633 bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE);
636 /* let's start with user quota */
637 qi->lqi_id.qid_uid = uid;
638 qi->lqi_type = USRQUOTA;
639 qi->lqi_space = space;
640 qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK);
641 rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
643 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
644 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
647 /* For non-fatal error, we want to continue to get the noquota flags
648 * for group id. This is only for commit write, which has @flags passed
649 * in. See osd_declare_write_commit().
650 * When force is set to true, we also want to proceed with the gid */
651 if (rcu && (rcu != -EDQUOT || flags == NULL))
654 /* and now group quota */
655 qi->lqi_id.qid_gid = gid;
656 qi->lqi_type = GRPQUOTA;
657 rcg = osd_declare_qid(env, oh, qi, obj, true, flags);
659 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
660 /* as before, ignore EDQUOT & EINPROGRESS for root */
663 #ifdef HAVE_PROJECT_QUOTA
664 if (rcg && (rcg != -EDQUOT || flags == NULL))
667 /* and now project quota */
668 qi->lqi_id.qid_projid = projid;
669 qi->lqi_type = PRJQUOTA;
670 rcp = osd_declare_qid(env, oh, qi, obj, true, flags);
672 if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS))
673 /* as before, ignore EDQUOT & EINPROGRESS for root */
677 RETURN(rcu ? rcu : (rcg ? rcg : rcp));