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 if (fid_oid(fid) == ACCT_GROUP_OID)
46 static inline int obj2type(struct dt_object *obj)
48 return fid2type(lu_object_fid(&obj->do_lu));
52 * Space Accounting Management
56 * Look up an accounting object based on its fid.
58 * \param info - is the osd thread info passed by the caller
59 * \param osd - is the osd device
60 * \param fid - is the fid of the accounting object we want to look up
61 * \param id - is the osd_inode_id struct to fill with the inode number of
62 * the quota file if the lookup is successful
64 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
65 const struct lu_fid *fid, struct osd_inode_id *id)
67 struct super_block *sb = osd_sb(osd);
68 unsigned long qf_inums[2] = {
69 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum),
70 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum)
74 LASSERT(fid_is_acct(fid));
76 if (!LDISKFS_HAS_RO_COMPAT_FEATURE(sb,
77 LDISKFS_FEATURE_RO_COMPAT_QUOTA))
80 id->oii_gen = OSD_OII_NOGEN;
81 id->oii_ino = qf_inums[fid2type(fid)];
82 if (!ldiskfs_valid_inum(sb, id->oii_ino))
88 * Return space usage (#blocks & #inodes) consumed by a given uid or gid.
90 * \param env - is the environment passed by the caller
91 * \param dtobj - is the accounting object
92 * \param dtrec - is the record to fill with space usage information
93 * \param dtkey - is the id of the user or group for which we would
94 * like to access disk usage.
96 * \retval +ve - success : exact match
97 * \retval -ve - failure
99 static int osd_acct_index_lookup(const struct lu_env *env,
100 struct dt_object *dtobj,
101 struct dt_rec *dtrec,
102 const struct dt_key *dtkey)
104 struct osd_thread_info *info = osd_oti_get(env);
105 #if defined(HAVE_DQUOT_QC_DQBLK)
106 struct qc_dqblk *dqblk = &info->oti_qdq;
107 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
108 struct fs_disk_quota *dqblk = &info->oti_fdq;
110 struct if_dqblk *dqblk = &info->oti_dqblk;
112 struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
113 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
114 __u64 id = *((__u64 *)dtkey);
116 #ifdef HAVE_DQUOT_KQID
122 memset(dqblk, 0, sizeof(*dqblk));
123 #ifdef HAVE_DQUOT_KQID
124 qid = make_kqid(&init_user_ns, obj2type(dtobj), id);
125 rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
127 rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t)id, dqblk);
131 #if defined(HAVE_DQUOT_QC_DQBLK)
132 rec->bspace = dqblk->d_space;
133 rec->ispace = dqblk->d_ino_count;
134 #elif defined(HAVE_DQUOT_FS_DISK_QUOTA)
135 rec->bspace = dqblk->d_bcount;
136 rec->ispace = dqblk->d_icount;
138 rec->bspace = dqblk->dqb_curspace;
139 rec->ispace = dqblk->dqb_curinodes;
144 #define QUOTA_IT_READ_ERROR(it, rc) \
145 CERROR("%s: Error while trying to read quota information, " \
146 "failed with %d\n", \
147 osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
150 * Initialize osd Iterator for given osd index object.
152 * \param dt - osd index object
153 * \param attr - not used
155 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
156 struct dt_object *dt,
159 struct osd_it_quota *it;
160 struct lu_object *lo = &dt->do_lu;
161 struct osd_object *obj = osd_dt_obj(dt);
165 LASSERT(lu_object_exists(lo));
169 RETURN(ERR_PTR(-ENOMEM));
173 INIT_LIST_HEAD(&it->oiq_list);
175 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
176 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
178 /* NB: we don't need to store the tree depth since it is always
179 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
181 RETURN((struct dt_it *)it);
185 * Free given iterator.
187 * \param di - osd iterator
189 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
191 struct osd_it_quota *it = (struct osd_it_quota *)di;
192 struct osd_quota_leaf *leaf, *tmp;
195 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
197 list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
198 list_del_init(&leaf->oql_link);
208 * Move Iterator to record specified by \a key, if the \a key isn't found,
209 * move to the first valid record.
211 * \param di - osd iterator
212 * \param key - uid or gid
214 * \retval +ve - di points to the first valid record
215 * \retval +1 - di points to exact matched key
216 * \retval -ve - failure
218 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
219 const struct dt_key *key)
221 struct osd_it_quota *it = (struct osd_it_quota *)di;
222 const struct lu_fid *fid =
223 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
224 int type = fid2type(fid);
225 qid_t dqid = *(qid_t *)key;
231 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
232 LUSTRE_DQTREEOFF, 0, it);
233 if (offset > 0) { /* Found */
235 } else if (offset < 0) { /* Error */
236 QUOTA_IT_READ_ERROR(it, (int)offset);
240 /* The @key is not found, move to the first valid entry */
241 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
254 * \param di - osd iterator
256 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
261 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
263 struct osd_quota_leaf *leaf;
268 INIT_LIST_HEAD(&leaf->oql_link);
269 leaf->oql_blk = it->oiq_blk[depth];
270 list_add_tail(&leaf->oql_link, &it->oiq_list);
275 * Move on to the next valid entry.
277 * \param di - osd iterator
279 * \retval +ve - iterator reached the end
280 * \retval 0 - iterator has not reached the end yet
281 * \retval -ve - unexpected failure
283 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
285 struct osd_it_quota *it = (struct osd_it_quota *)di;
286 const struct lu_fid *fid =
287 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
288 int type = fid2type(fid);
294 /* Let's first check if there are any remaining valid entry in the
295 * current leaf block. Start with the next entry after the current one.
297 depth = LUSTRE_DQTREEDEPTH;
298 index = it->oiq_index[depth];
299 if (++index < LUSTRE_DQSTRINBLK) {
300 /* Search for the next valid entry from current index */
301 rc = walk_block_dqentry(env, it->oiq_obj, type,
302 it->oiq_blk[depth], index, it);
304 QUOTA_IT_READ_ERROR(it, rc);
306 } else if (rc == 0) {
307 /* Found on entry, @it is already updated to the
308 * new position in walk_block_dqentry(). */
311 rc = osd_it_add_processed(it, depth);
316 rc = osd_it_add_processed(it, depth);
322 /* We have consumed all the entries of the current leaf block, move on
323 * to the next one. */
326 /* We keep searching as long as walk_tree_dqentry() returns +1
327 * (= no valid entry found). */
328 for (; depth >= 0 && rc > 0; depth--) {
329 index = it->oiq_index[depth];
332 rc = walk_tree_dqentry(env, it->oiq_obj, type,
333 it->oiq_blk[depth], depth, index, it);
337 QUOTA_IT_READ_ERROR(it, rc);
342 * Return pointer to the key under iterator.
344 * \param di - osd iterator
346 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
347 const struct dt_it *di)
349 struct osd_it_quota *it = (struct osd_it_quota *)di;
352 RETURN((struct dt_key *)&it->oiq_id);
356 * Return size of key under iterator (in bytes)
358 * \param di - osd iterator
360 static int osd_it_acct_key_size(const struct lu_env *env,
361 const struct dt_it *di)
363 struct osd_it_quota *it = (struct osd_it_quota *)di;
366 RETURN((int)sizeof(it->oiq_id));
370 * Return pointer to the record under iterator.
372 * \param di - osd iterator
373 * \param attr - not used
375 static int osd_it_acct_rec(const struct lu_env *env,
376 const struct dt_it *di,
377 struct dt_rec *dtrec, __u32 attr)
379 struct osd_it_quota *it = (struct osd_it_quota *)di;
380 const struct dt_key *key = osd_it_acct_key(env, di);
385 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
386 RETURN(rc > 0 ? 0 : rc);
390 * Returns cookie for current Iterator position.
392 * \param di - osd iterator
394 static __u64 osd_it_acct_store(const struct lu_env *env,
395 const struct dt_it *di)
397 struct osd_it_quota *it = (struct osd_it_quota *)di;
404 * Restore iterator from cookie. if the \a hash isn't found,
405 * restore the first valid record.
407 * \param di - osd iterator
408 * \param hash - iterator location cookie
410 * \retval +ve - di points to the first valid record
411 * \retval +1 - di points to exact matched hash
412 * \retval -ve - failure
414 static int osd_it_acct_load(const struct lu_env *env,
415 const struct dt_it *di, __u64 hash)
418 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
419 (const struct dt_key *)&hash));
423 * Index and Iterator operations for accounting objects
425 const struct dt_index_operations osd_acct_index_ops = {
426 .dio_lookup = osd_acct_index_lookup,
428 .init = osd_it_acct_init,
429 .fini = osd_it_acct_fini,
430 .get = osd_it_acct_get,
431 .put = osd_it_acct_put,
432 .next = osd_it_acct_next,
433 .key = osd_it_acct_key,
434 .key_size = osd_it_acct_key_size,
435 .rec = osd_it_acct_rec,
436 .store = osd_it_acct_store,
437 .load = osd_it_acct_load
441 static inline void osd_quota_swab(char *ptr, size_t size)
445 LASSERT((size & (sizeof(__u64) - 1)) == 0);
447 for (offset = 0; offset < size; offset += sizeof(__u64))
448 __swab64s((__u64 *)(ptr + offset));
451 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
452 const struct dt_rec *rec,
453 union lquota_rec *quota_rec)
456 struct iam_descr *descr;
458 LASSERT(obj->oo_dir != NULL);
459 descr = obj->oo_dir->od_container.ic_descr;
461 memcpy(quota_rec, rec, descr->id_rec_size);
463 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
464 return (const struct dt_rec *)quota_rec;
470 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
473 struct iam_descr *descr;
475 LASSERT(obj->oo_dir != NULL);
476 descr = obj->oo_dir->od_container.ic_descr;
478 osd_quota_swab((char *)rec, descr->id_rec_size);
484 static inline int osd_qid_type(struct osd_thandle *oh, int i)
486 return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
489 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
491 oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
495 * Reserve journal credits for quota files update first, then call
496 * ->op_begin() to perform quota enforcement.
498 * \param env - the environment passed by the caller
499 * \param oh - osd transaction handle
500 * \param qi - quota id & space required for this operation
501 * \param obj - osd object, could be NULL when it's under create
502 * \param enforce - whether to perform quota enforcement
503 * \param flags - if the operation is write, return no user quota, no
504 * group quota, or sync commit flags to the caller
506 * \retval 0 - success
507 * \retval -ve - failure
509 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
510 struct lquota_id_info *qi, struct osd_object *obj,
511 bool enforce, int *flags)
513 struct osd_device *dev;
514 struct qsd_instance *qsd;
515 struct inode *inode = NULL;
521 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
524 dev = osd_dt_dev(oh->ot_super.th_dev);
525 LASSERT(dev != NULL);
527 qsd = dev->od_quota_slave;
529 for (i = 0; i < oh->ot_id_cnt; i++) {
530 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
531 osd_qid_type(oh, i) == qi->lqi_type) {
538 /* we need to account for credits for this new ID */
539 if (i >= OSD_MAX_UGID_CNT) {
540 CERROR("Too many(%d) trans qids!\n", i + 1);
545 inode = obj->oo_inode;
547 /* root ID entry should be always present in the quota file */
548 if (qi->lqi_id.qid_uid == 0) {
551 /* used space for this ID could be dropped to zero,
552 * reserve extra credits for removing ID entry from
554 if (qi->lqi_space < 0)
555 crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
556 /* reserve credits for adding ID entry to the quota
557 * file if the i_dquot isn't initialized yet. */
558 else if (inode == NULL ||
559 inode->i_dquot[qi->lqi_type] == NULL)
560 crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
565 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
567 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
568 osd_qid_set_type(oh, i, qi->lqi_type);
572 if (unlikely(qsd == NULL))
573 /* quota slave instance hasn't been allocated yet */
578 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
583 * Wrapper for osd_declare_qid()
585 * \param env - the environment passed by the caller
586 * \param uid - user id of the inode
587 * \param gid - group id of the inode
588 * \param space - how many blocks/inodes will be consumed/released
589 * \param oh - osd transaction handle
590 * \param obj - osd object, could be NULL when it's under create
591 * \param is_blk - block quota or inode quota?
592 * \param flags - if the operation is write, return no user quota, no
593 * group quota, or sync commit flags to the caller
594 * \param force - set to 1 when changes are performed by root user and thus
595 * can't failed with EDQUOT
597 * \retval 0 - success
598 * \retval -ve - failure
600 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
601 long long space, struct osd_thandle *oh,
602 struct osd_object *obj, bool is_blk, int *flags,
605 struct osd_thread_info *info = osd_oti_get(env);
606 struct lquota_id_info *qi = &info->oti_qi;
607 int rcu, rcg; /* user & group rc */
610 /* let's start with user quota */
611 qi->lqi_id.qid_uid = uid;
612 qi->lqi_type = USRQUOTA;
613 qi->lqi_space = space;
614 qi->lqi_is_blk = is_blk;
615 rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
617 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
618 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
621 /* For non-fatal error, we want to continue to get the noquota flags
622 * for group id. This is only for commit write, which has @flags passed
623 * in. See osd_declare_write_commit().
624 * When force is set to true, we also want to proceed with the gid */
625 if (rcu && (rcu != -EDQUOT || flags == NULL))
628 /* and now group quota */
629 qi->lqi_id.qid_gid = gid;
630 qi->lqi_type = GRPQUOTA;
631 rcg = osd_declare_qid(env, oh, qi, obj, true, flags);
633 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
634 /* as before, ignore EDQUOT & EINPROGRESS for root */
637 RETURN(rcu ? rcu : rcg);
640 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt)
642 struct osd_thread_info *oti = osd_oti_get(env);
643 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
644 struct dt_object *root, *parent = NULL, *admin = NULL;
645 dt_obj_version_t version;
646 char *fname, *fnames[] = {ADMIN_USR, ADMIN_GRP};
650 /* not newly created global index */
651 version = dt_version_get(env, dt);
656 rc = dt_root_get(env, &osd->od_dt_dev, &oti->oti_fid);
658 CERROR("%s: Can't get root FID, rc:%d\n", osd->od_svname, rc);
662 root = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
664 CERROR("%s: Failed to locate root "DFID", rc:%ld\n",
665 osd->od_svname, PFID(&oti->oti_fid), PTR_ERR(root));
666 RETURN(PTR_ERR(root));
669 /* locate /OBJECTS */
670 rc = dt_lookup_dir(env, root, OBJECTS, &oti->oti_fid);
674 CERROR("%s: Failed to lookup %s, rc:%d\n",
675 osd->od_svname, OBJECTS, rc);
679 parent = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
680 if (IS_ERR(parent)) {
681 CERROR("%s: Failed to locate %s "DFID", rc:%ld\n",
682 osd->od_svname, OBJECTS, PFID(&oti->oti_fid),
684 GOTO(out, rc = PTR_ERR(parent));
687 /* locate quota admin files */
688 for (i = 0; i < 2; i++) {
690 rc = dt_lookup_dir(env, parent, fname, &oti->oti_fid);
695 CERROR("%s: Failed to lookup %s, rc:%d\n",
696 osd->od_svname, fname, rc);
700 admin = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
702 CERROR("%s: Failed to locate %s "DFID", rc:%d\n",
703 osd->od_svname, fname, PFID(&oti->oti_fid), rc);
704 GOTO(out, rc = PTR_ERR(admin));
707 if (!dt_object_exists(admin)) {
708 CERROR("%s: Old admin file %s doesn't exist, but is "
709 "still referenced in parent directory.\n",
710 osd->od_svname, fname);
711 lu_object_put(env, &admin->do_lu);
712 GOTO(out, rc = -ENOENT);
715 LCONSOLE_WARN("%s: Detected old quota admin file(%s)! If you "
716 "want to keep the old quota limits settings, "
717 "please upgrade to lower version(2.5) first to "
718 "convert them into new format.\n",
719 osd->od_svname, fname);
721 lu_object_put(env, &admin->do_lu);
722 GOTO(out, rc = -EINVAL);
725 if (parent && !IS_ERR(parent))
726 lu_object_put(env, &parent->do_lu);
727 lu_object_put(env, &root->do_lu);