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, 2014, 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 the of the user or group for which we would
94 * like to access disk usage.
95 * \param capa - is the capability, not used.
97 * \retval +ve - success : exact match
98 * \retval -ve - failure
100 static int osd_acct_index_lookup(const struct lu_env *env,
101 struct dt_object *dtobj,
102 struct dt_rec *dtrec,
103 const struct dt_key *dtkey,
104 struct lustre_capa *capa)
106 struct osd_thread_info *info = osd_oti_get(env);
107 #ifdef 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((void *)dqblk, 0, sizeof(struct obd_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 #ifdef HAVE_DQUOT_FS_DISK_QUOTA
132 rec->bspace = dqblk->d_bcount;
133 rec->ispace = dqblk->d_icount;
135 rec->bspace = dqblk->dqb_curspace;
136 rec->ispace = dqblk->dqb_curinodes;
141 #define QUOTA_IT_READ_ERROR(it, rc) \
142 CERROR("%s: Error while trying to read quota information, " \
143 "failed with %d\n", \
144 osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
147 * Initialize osd Iterator for given osd index object.
149 * \param dt - osd index object
150 * \param attr - not used
151 * \param capa - BYPASS_CAPA
153 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
154 struct dt_object *dt,
155 __u32 attr, struct lustre_capa *capa)
157 struct osd_thread_info *info = osd_oti_get(env);
158 struct osd_it_quota *it;
159 struct lu_object *lo = &dt->do_lu;
160 struct osd_object *obj = osd_dt_obj(dt);
164 LASSERT(lu_object_exists(lo));
167 RETURN(ERR_PTR(-ENOMEM));
169 if (info->oti_it_inline) {
172 RETURN(ERR_PTR(-ENOMEM));
174 it = &info->oti_it_quota;
175 info->oti_it_inline = 1;
178 memset(it, 0, sizeof(*it));
181 INIT_LIST_HEAD(&it->oiq_list);
183 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
184 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
186 /* NB: we don't need to store the tree depth since it is always
187 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
189 RETURN((struct dt_it *)it);
193 * Free given iterator.
195 * \param di - osd iterator
197 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
199 struct osd_thread_info *info = osd_oti_get(env);
200 struct osd_it_quota *it = (struct osd_it_quota *)di;
201 struct osd_quota_leaf *leaf, *tmp;
204 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
206 list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
207 list_del_init(&leaf->oql_link);
211 if (it != &info->oti_it_quota)
214 info->oti_it_inline = 0;
220 * Move Iterator to record specified by \a key, if the \a key isn't found,
221 * move to the first valid record.
223 * \param di - osd iterator
224 * \param key - uid or gid
226 * \retval +ve - di points to the first valid record
227 * \retval +1 - di points to exact matched key
228 * \retval -ve - failure
230 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
231 const struct dt_key *key)
233 struct osd_it_quota *it = (struct osd_it_quota *)di;
234 const struct lu_fid *fid =
235 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
236 int type = fid2type(fid);
237 qid_t dqid = *(qid_t *)key;
243 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
244 LUSTRE_DQTREEOFF, 0, it);
245 if (offset > 0) { /* Found */
247 } else if (offset < 0) { /* Error */
248 QUOTA_IT_READ_ERROR(it, (int)offset);
252 /* The @key is not found, move to the first valid entry */
253 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
266 * \param di - osd iterator
268 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
273 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
275 struct osd_quota_leaf *leaf;
280 INIT_LIST_HEAD(&leaf->oql_link);
281 leaf->oql_blk = it->oiq_blk[depth];
282 list_add_tail(&leaf->oql_link, &it->oiq_list);
287 * Move on to the next valid entry.
289 * \param di - osd iterator
291 * \retval +ve - iterator reached the end
292 * \retval 0 - iterator has not reached the end yet
293 * \retval -ve - unexpected failure
295 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
297 struct osd_it_quota *it = (struct osd_it_quota *)di;
298 const struct lu_fid *fid =
299 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
300 int type = fid2type(fid);
306 /* Let's first check if there are any remaining valid entry in the
307 * current leaf block. Start with the next entry after the current one.
309 depth = LUSTRE_DQTREEDEPTH;
310 index = it->oiq_index[depth];
311 if (++index < LUSTRE_DQSTRINBLK) {
312 /* Search for the next valid entry from current index */
313 rc = walk_block_dqentry(env, it->oiq_obj, type,
314 it->oiq_blk[depth], index, it);
316 QUOTA_IT_READ_ERROR(it, rc);
318 } else if (rc == 0) {
319 /* Found on entry, @it is already updated to the
320 * new position in walk_block_dqentry(). */
323 rc = osd_it_add_processed(it, depth);
328 rc = osd_it_add_processed(it, depth);
334 /* We have consumed all the entries of the current leaf block, move on
335 * to the next one. */
338 /* We keep searching as long as walk_tree_dqentry() returns +1
339 * (= no valid entry found). */
340 for (; depth >= 0 && rc > 0; depth--) {
341 index = it->oiq_index[depth];
344 rc = walk_tree_dqentry(env, it->oiq_obj, type,
345 it->oiq_blk[depth], depth, index, it);
349 QUOTA_IT_READ_ERROR(it, rc);
354 * Return pointer to the key under iterator.
356 * \param di - osd iterator
358 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
359 const struct dt_it *di)
361 struct osd_it_quota *it = (struct osd_it_quota *)di;
364 RETURN((struct dt_key *)&it->oiq_id);
368 * Return size of key under iterator (in bytes)
370 * \param di - osd iterator
372 static int osd_it_acct_key_size(const struct lu_env *env,
373 const struct dt_it *di)
375 struct osd_it_quota *it = (struct osd_it_quota *)di;
378 RETURN((int)sizeof(it->oiq_id));
382 * Return pointer to the record under iterator.
384 * \param di - osd iterator
385 * \param attr - not used
387 static int osd_it_acct_rec(const struct lu_env *env,
388 const struct dt_it *di,
389 struct dt_rec *dtrec, __u32 attr)
391 struct osd_it_quota *it = (struct osd_it_quota *)di;
392 const struct dt_key *key = osd_it_acct_key(env, di);
397 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key,
399 RETURN(rc > 0 ? 0 : rc);
403 * Returns cookie for current Iterator position.
405 * \param di - osd iterator
407 static __u64 osd_it_acct_store(const struct lu_env *env,
408 const struct dt_it *di)
410 struct osd_it_quota *it = (struct osd_it_quota *)di;
417 * Restore iterator from cookie. if the \a hash isn't found,
418 * restore the first valid record.
420 * \param di - osd iterator
421 * \param hash - iterator location cookie
423 * \retval +ve - di points to the first valid record
424 * \retval +1 - di points to exact matched hash
425 * \retval -ve - failure
427 static int osd_it_acct_load(const struct lu_env *env,
428 const struct dt_it *di, __u64 hash)
431 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
432 (const struct dt_key *)&hash));
436 * Index and Iterator operations for accounting objects
438 const struct dt_index_operations osd_acct_index_ops = {
439 .dio_lookup = osd_acct_index_lookup,
441 .init = osd_it_acct_init,
442 .fini = osd_it_acct_fini,
443 .get = osd_it_acct_get,
444 .put = osd_it_acct_put,
445 .next = osd_it_acct_next,
446 .key = osd_it_acct_key,
447 .key_size = osd_it_acct_key_size,
448 .rec = osd_it_acct_rec,
449 .store = osd_it_acct_store,
450 .load = osd_it_acct_load
454 static inline void osd_quota_swab(char *ptr, size_t size)
458 LASSERT((size & (sizeof(__u64) - 1)) == 0);
460 for (offset = 0; offset < size; offset += sizeof(__u64))
461 __swab64s((__u64 *)(ptr + offset));
464 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
465 const struct dt_rec *rec,
466 union lquota_rec *quota_rec)
469 struct iam_descr *descr;
471 LASSERT(obj->oo_dir != NULL);
472 descr = obj->oo_dir->od_container.ic_descr;
474 memcpy(quota_rec, rec, descr->id_rec_size);
476 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
477 return (const struct dt_rec *)quota_rec;
483 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
486 struct iam_descr *descr;
488 LASSERT(obj->oo_dir != NULL);
489 descr = obj->oo_dir->od_container.ic_descr;
491 osd_quota_swab((char *)rec, descr->id_rec_size);
497 static inline int osd_qid_type(struct osd_thandle *oh, int i)
499 return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
502 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
504 oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
508 * Reserve journal credits for quota files update first, then call
509 * ->op_begin() to perform quota enforcement.
511 * \param env - the environment passed by the caller
512 * \param oh - osd transaction handle
513 * \param qi - quota id & space required for this operation
514 * \param obj - osd object, could be NULL when it's under create
515 * \param enforce - whether to perform quota enforcement
516 * \param flags - if the operation is write, return no user quota, no
517 * group quota, or sync commit flags to the caller
519 * \retval 0 - success
520 * \retval -ve - failure
522 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
523 struct lquota_id_info *qi, struct osd_object *obj,
524 bool enforce, int *flags)
526 struct osd_thread_info *info = osd_oti_get(env);
527 struct osd_device *dev = info->oti_dev;
528 struct qsd_instance *qsd = dev->od_quota_slave;
529 struct inode *inode = NULL;
535 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
538 for (i = 0; i < oh->ot_id_cnt; i++) {
539 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
540 osd_qid_type(oh, i) == qi->lqi_type) {
547 /* we need to account for credits for this new ID */
548 if (i >= OSD_MAX_UGID_CNT) {
549 CERROR("Too many(%d) trans qids!\n", i + 1);
554 inode = obj->oo_inode;
556 /* root ID entry should be always present in the quota file */
557 if (qi->lqi_id.qid_uid == 0) {
560 /* used space for this ID could be dropped to zero,
561 * reserve extra credits for removing ID entry from
563 if (qi->lqi_space < 0)
564 crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
565 /* reserve credits for adding ID entry to the quota
566 * file if the i_dquot isn't initialized yet. */
567 else if (inode == NULL ||
568 inode->i_dquot[qi->lqi_type] == NULL)
569 crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
574 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
576 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
577 osd_qid_set_type(oh, i, qi->lqi_type);
581 if (unlikely(qsd == NULL))
582 /* quota slave instance hasn't been allocated yet */
587 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
592 * Wrapper for osd_declare_qid()
594 * \param env - the environment passed by the caller
595 * \param uid - user id of the inode
596 * \param gid - group id of the inode
597 * \param space - how many blocks/inodes will be consumed/released
598 * \param oh - osd transaction handle
599 * \param obj - osd object, could be NULL when it's under create
600 * \param is_blk - block quota or inode quota?
601 * \param flags - if the operation is write, return no user quota, no
602 * group quota, or sync commit flags to the caller
603 * \param force - set to 1 when changes are performed by root user and thus
604 * can't failed with EDQUOT
606 * \retval 0 - success
607 * \retval -ve - failure
609 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
610 long long space, struct osd_thandle *oh,
611 struct osd_object *obj, bool is_blk, int *flags,
614 struct osd_thread_info *info = osd_oti_get(env);
615 struct lquota_id_info *qi = &info->oti_qi;
616 int rcu, rcg; /* user & group rc */
619 /* let's start with user quota */
620 qi->lqi_id.qid_uid = uid;
621 qi->lqi_type = USRQUOTA;
622 qi->lqi_space = space;
623 qi->lqi_is_blk = is_blk;
624 rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
626 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
627 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
630 /* For non-fatal error, we want to continue to get the noquota flags
631 * for group id. This is only for commit write, which has @flags passed
632 * in. See osd_declare_write_commit().
633 * When force is set to true, we also want to proceed with the gid */
634 if (rcu && (rcu != -EDQUOT || flags == NULL))
637 /* and now group quota */
638 qi->lqi_id.qid_gid = gid;
639 qi->lqi_type = GRPQUOTA;
640 rcg = osd_declare_qid(env, oh, qi, obj, true, flags);
642 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
643 /* as before, ignore EDQUOT & EINPROGRESS for root */
646 RETURN(rcu ? rcu : rcg);
649 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0)
650 /* Following code is used to migrate old admin quota files (in Linux quota
651 * file v2 format) into the new quota global indexes (in IAM format). */
653 /* copied from osd_it_acct_get(), only changed the 'type' to -1 */
654 static int osd_it_admin_get(const struct lu_env *env, struct dt_it *di,
655 const struct dt_key *key)
657 struct osd_it_quota *it = (struct osd_it_quota *)di;
659 qid_t dqid = *(qid_t *)key;
664 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
665 LUSTRE_DQTREEOFF, 0, it);
666 if (offset > 0) { /* Found */
668 } else if (offset < 0) { /* Error */
669 QUOTA_IT_READ_ERROR(it, (int)offset);
673 /* The @key is not found, move to the first valid entry */
674 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
677 /* no valid entry found */
682 static int osd_it_admin_load(const struct lu_env *env,
683 const struct dt_it *di, __u64 hash)
688 rc = osd_it_admin_get(env, (struct dt_it *)di,
689 (const struct dt_key *)&hash);
693 static int osd_it_admin_rec(const struct lu_env *env,
694 const struct dt_it *di,
695 struct dt_rec *dtrec, __u32 attr)
697 struct osd_it_quota *it = (struct osd_it_quota *)di;
701 struct lustre_disk_dqblk_v2 *dqblk =
702 (struct lustre_disk_dqblk_v2 *)dtrec;
706 buf.lb_len = sizeof(*dqblk);
708 pos = it->oiq_offset;
709 rc = dt_record_read(env, &it->oiq_obj->oo_dt, &buf, &pos);
713 /* copied from osd_it_acct_next(), only changed the 'type' to -1 */
714 static int osd_it_admin_next(const struct lu_env *env, struct dt_it *di)
716 struct osd_it_quota *it = (struct osd_it_quota *)di;
722 /* Let's first check if there are any remaining valid entry in the
723 * current leaf block. Start with the next entry after the current one.
725 depth = LUSTRE_DQTREEDEPTH;
726 index = it->oiq_index[depth];
727 if (++index < LUSTRE_DQSTRINBLK) {
728 /* Search for the next valid entry from current index */
729 rc = walk_block_dqentry(env, it->oiq_obj, type,
730 it->oiq_blk[depth], index, it);
732 QUOTA_IT_READ_ERROR(it, rc);
734 } else if (rc == 0) {
735 /* Found on entry, @it is already updated to the
736 * new position in walk_block_dqentry(). */
739 rc = osd_it_add_processed(it, depth);
744 rc = osd_it_add_processed(it, depth);
750 /* We have consumed all the entries of the current leaf block, move on
751 * to the next one. */
754 /* We keep searching as long as walk_tree_dqentry() returns +1
755 * (= no valid entry found). */
756 for (; depth >= 0 && rc > 0; depth--) {
757 index = it->oiq_index[depth];
760 rc = walk_tree_dqentry(env, it->oiq_obj, type,
761 it->oiq_blk[depth], depth, index, it);
765 QUOTA_IT_READ_ERROR(it, rc);
769 static const struct dt_index_operations osd_admin_index_ops = {
770 .dio_lookup = osd_acct_index_lookup,
772 .init = osd_it_acct_init,
773 .fini = osd_it_acct_fini,
774 .get = osd_it_admin_get,
775 .put = osd_it_acct_put,
776 .next = osd_it_admin_next,
777 .key = osd_it_acct_key,
778 .key_size = osd_it_acct_key_size,
779 .rec = osd_it_admin_rec,
780 .store = osd_it_acct_store,
781 .load = osd_it_admin_load
785 static int convert_quota_file(const struct lu_env *env,
786 struct dt_object *old, struct dt_object *new,
789 const struct dt_it_ops *iops = &old->do_index_ops->dio_it;
790 struct osd_object *obj;
795 struct lquota_glb_rec *glb_rec = NULL;
798 struct lustre_disk_dqblk_v2 *dqblk = NULL;
799 struct lustre_disk_dqinfo *dqinfo = NULL;
802 obj = osd_dt_obj(old);
803 LASSERT(obj->oo_inode);
805 if (i_size_read(obj->oo_inode) == 0)
808 /* allocate buffers */
809 OBD_ALLOC_PTR(dqinfo);
813 OBD_ALLOC_PTR(glb_rec);
815 GOTO(out, rc = -ENOMEM);
817 OBD_ALLOC_PTR(dqblk);
819 GOTO(out, rc = -ENOMEM);
821 /* convert the old igrace/bgrace */
823 buf.lb_len = sizeof(*dqinfo);
824 pos = LUSTRE_DQINFOOFF;
826 rc = dt_record_read(env, old, &buf, &pos);
830 /* keep it in little endian */
831 grace = isblk ? dqinfo->dqi_bgrace : dqinfo->dqi_igrace;
833 glb_rec->qbr_time = grace;
834 rc = lquota_disk_write_glb(env, new, 0, glb_rec);
837 glb_rec->qbr_time = 0;
840 /* iterate the old admin file, insert each record into the
842 it = iops->init(env, old, 0, BYPASS_CAPA);
844 GOTO(out, rc = PTR_ERR(it));
846 rc = iops->load(env, it, 0);
848 GOTO(out_it, rc = 0);
853 key = iops->key(env, it);
855 GOTO(out_it, rc = PTR_ERR(key));
857 /* skip the root user/group */
858 if (*((__u64 *)key) == 0)
861 rc = iops->rec(env, it, (struct dt_rec *)dqblk, 0);
865 /* keep the value in little endian */
866 glb_rec->qbr_hardlimit = isblk ? dqblk->dqb_bhardlimit :
867 dqblk->dqb_ihardlimit;
868 glb_rec->qbr_softlimit = isblk ? dqblk->dqb_bsoftlimit :
869 dqblk->dqb_isoftlimit;
871 rc = lquota_disk_write_glb(env, new, *((__u64 *)key), glb_rec);
875 rc = iops->next(env, it);
889 OBD_FREE_PTR(glb_rec);
891 OBD_FREE_PTR(dqinfo);
895 /* Nobdy else can access the global index now, it's safe to truncate and
897 static int truncate_quota_index(const struct lu_env *env, struct dt_object *dt,
898 const struct dt_index_features *feat)
900 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
902 struct lu_attr *attr;
903 struct osd_thandle *oth;
906 struct iam_container *bag = &(osd_dt_obj(dt))->oo_dir->od_container;
907 struct lu_buf *lb = &osd_oti_get(env)->oti_buf;
910 LASSERT(bag->ic_root_bh != NULL);
911 iam_container_fini(bag);
913 LASSERT(fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB);
920 attr->la_valid = LA_SIZE;
922 th = dt_trans_create(env, &osd->od_dt_dev);
928 rc = dt_declare_punch(env, dt, 0, OBD_OBJECT_EOF, th);
932 rc = dt_declare_attr_set(env, dt, attr, th);
936 inode = osd_dt_obj(dt)->oo_inode;
939 /* iam_lfix_create() writes two blocks at the beginning */
940 lb->lb_len = osd_sb(osd)->s_blocksize * 2;
941 rc = dt_declare_record_write(env, dt, lb, 0, th);
945 rc = dt_trans_start_local(env, &osd->od_dt_dev, th);
949 dt_write_lock(env, dt, 0);
950 rc = dt_punch(env, dt, 0, OBD_OBJECT_EOF, th, BYPASS_CAPA);
954 rc = dt_attr_set(env, dt, attr, th, BYPASS_CAPA);
958 oth = container_of(th, struct osd_thandle, ot_super);
960 if (feat->dif_flags & DT_IND_VARKEY)
961 rc = iam_lvar_create(osd_dt_obj(dt)->oo_inode,
962 feat->dif_keysize_max,
964 feat->dif_recsize_max, oth->ot_handle);
966 rc = iam_lfix_create(osd_dt_obj(dt)->oo_inode,
967 feat->dif_keysize_max,
969 feat->dif_recsize_max, oth->ot_handle);
971 dt_write_unlock(env, dt);
973 dt_trans_stop(env, &osd->od_dt_dev, th);
977 rc = iam_container_setup(bag);
979 iam_container_fini(bag);
984 static int set_quota_index_version(const struct lu_env *env,
985 struct dt_object *dt,
986 dt_obj_version_t version)
988 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
993 th = dt_trans_create(env, &osd->od_dt_dev);
997 rc = dt_declare_version_set(env, dt, th);
1001 rc = dt_trans_start_local(env, &osd->od_dt_dev, th);
1006 dt_version_set(env, dt, version, th);
1008 dt_trans_stop(env, &osd->od_dt_dev, th);
1012 int osd_quota_migration(const struct lu_env *env, struct dt_object *dt,
1013 const struct dt_index_features *feat)
1015 struct osd_thread_info *oti = osd_oti_get(env);
1016 struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
1017 struct dt_object *root, *parent = NULL, *admin = NULL;
1018 dt_obj_version_t version;
1020 bool isblk = false, converted = false;
1024 /* not newly created global index */
1025 version = dt_version_get(env, dt);
1030 rc = dt_root_get(env, &osd->od_dt_dev, &oti->oti_fid);
1032 CERROR("%s: Can't get root FID, rc:%d\n", osd->od_svname, rc);
1036 root = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
1038 CERROR("%s: Failed to locate root "DFID", rc:%ld\n",
1039 osd->od_svname, PFID(&oti->oti_fid), PTR_ERR(root));
1040 RETURN(PTR_ERR(root));
1043 /* locate /OBJECTS */
1044 rc = dt_lookup_dir(env, root, OBJECTS, &oti->oti_fid);
1045 if (rc == -ENOENT) {
1048 CERROR("%s: Failed to lookup %s, rc:%d\n",
1049 osd->od_svname, OBJECTS, rc);
1053 parent = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
1054 if (IS_ERR(parent)) {
1055 CERROR("%s: Failed to locate %s "DFID", rc:%ld\n",
1056 osd->od_svname, OBJECTS, PFID(&oti->oti_fid),
1058 GOTO(out, rc = PTR_ERR(parent));
1061 /* locate quota admin file */
1062 if (feat == &dt_quota_iusr_features) {
1065 } else if (feat == &dt_quota_busr_features) {
1068 } else if (feat == &dt_quota_igrp_features) {
1076 rc = dt_lookup_dir(env, parent, fname, &oti->oti_fid);
1077 if (rc == -ENOENT) {
1080 CERROR("%s: Failed to lookup %s, rc:%d\n",
1081 osd->od_svname, fname, rc);
1085 admin = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
1086 if (IS_ERR(admin)) {
1087 CERROR("%s: Failed to locate %s "DFID", rc:%d\n",
1088 osd->od_svname, fname, PFID(&oti->oti_fid), rc);
1089 GOTO(out, rc = PTR_ERR(admin));
1092 if (!dt_object_exists(admin)) {
1093 CERROR("%s: Old admin file %s doesn't exist, but is still "
1094 " referenced in parent directory.\n",
1095 osd->od_svname, fname);
1096 GOTO(out, rc = -ENOENT);
1099 /* truncate the new quota index file in case of any leftovers
1100 * from last failed migration */
1101 rc = truncate_quota_index(env, dt, feat);
1103 CERROR("%s: Failed to truncate the quota index "DFID", rc:%d\n",
1104 osd->od_svname, PFID(lu_object_fid(&dt->do_lu)), rc);
1108 /* set up indexing operations for the admin file */
1109 admin->do_index_ops = &osd_admin_index_ops;
1111 LCONSOLE_INFO("%s: Migrate %s quota from old admin quota file(%s) to "
1112 "new IAM quota index("DFID").\n", osd->od_svname,
1113 isblk ? "block" : "inode", fname,
1114 PFID(lu_object_fid(&dt->do_lu)));
1116 /* iterate the admin quota file, and insert each record into
1117 * the new index file */
1118 rc = convert_quota_file(env, admin, dt, isblk);
1120 CERROR("%s: Migrate old admin quota file(%s) failed, rc:%d\n",
1121 osd->od_svname, fname, rc);
1124 /* if no migration happen, we need to set the default grace time. */
1125 if (!converted && rc == 0) {
1126 struct lquota_glb_rec *rec = &oti->oti_quota_rec.lqr_glb_rec;
1128 rec->qbr_hardlimit = 0;
1129 rec->qbr_softlimit = 0;
1130 rec->qbr_granted = 0;
1131 rec->qbr_time = isblk ? MAX_DQ_TIME : MAX_IQ_TIME;
1133 rc = lquota_disk_write_glb(env, dt, 0, rec);
1135 CERROR("%s: Failed to set default grace time for "
1136 "index("DFID"), rc:%d\n", osd->od_svname,
1137 PFID(lu_object_fid(&dt->do_lu)), rc);
1140 /* bump index version to 1 (or 2 if migration happened), so the
1141 * migration will be skipped next time. */
1143 rc = set_quota_index_version(env , dt, converted ? 2 : 1);
1145 CERROR("%s: Failed to set quota index("DFID") "
1146 "version, rc:%d\n", osd->od_svname,
1147 PFID(lu_object_fid(&dt->do_lu)), rc);
1150 if (admin && !IS_ERR(admin))
1151 lu_object_put(env, &admin->do_lu);
1152 if (parent && !IS_ERR(parent))
1153 lu_object_put(env, &parent->do_lu);
1154 lu_object_put(env, &root->do_lu);
1158 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 53, 0) */