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, 2017, 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\n", 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_feature_quota(sb))
78 id->oii_gen = OSD_OII_NOGEN;
79 switch (fid2type(fid)) {
82 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum);
86 le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum);
89 #ifdef HAVE_PROJECT_QUOTA
90 if (ldiskfs_has_feature_project(sb)) {
93 prj_quota = LDISKFS_SB(sb)->s_es->s_prj_quota_inum;
94 id->oii_ino = le32_to_cpu(prj_quota);
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;
130 struct fs_disk_quota *dqblk = &info->oti_fdq;
132 struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
133 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
134 __u64 id = *((__u64 *)dtkey);
141 type = fid2type(lu_object_fid(&dtobj->do_lu));
142 memset(dqblk, 0, sizeof(*dqblk));
143 qid = make_kqid(&init_user_ns, type, id);
144 rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
147 #if defined(HAVE_DQUOT_QC_DQBLK)
148 rec->bspace = dqblk->d_space;
149 rec->ispace = dqblk->d_ino_count;
151 rec->bspace = dqblk->d_bcount;
152 rec->ispace = dqblk->d_icount;
157 #define QUOTA_IT_READ_ERROR(it, rc) \
158 CERROR("%s: Error while trying to read quota information, " \
159 "failed with %d\n", \
160 osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc) \
163 * Initialize osd Iterator for given osd index object.
165 * \param dt - osd index object
166 * \param attr - not used
168 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
169 struct dt_object *dt,
172 struct osd_it_quota *it;
173 struct lu_object *lo = &dt->do_lu;
174 struct osd_object *obj = osd_dt_obj(dt);
178 LASSERT(lu_object_exists(lo));
182 RETURN(ERR_PTR(-ENOMEM));
186 INIT_LIST_HEAD(&it->oiq_list);
188 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
189 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
192 * NB: we don't need to store the tree depth since it is always
193 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
196 RETURN((struct dt_it *)it);
200 * Free given iterator.
202 * \param di - osd iterator
204 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
206 struct osd_it_quota *it = (struct osd_it_quota *)di;
207 struct osd_quota_leaf *leaf, *tmp;
211 osd_object_put(env, it->oiq_obj);
213 list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
214 list_del_init(&leaf->oql_link);
224 * Move Iterator to record specified by \a key, if the \a key isn't found,
225 * move to the first valid record.
227 * \param di - osd iterator
228 * \param key - uid or gid
230 * \retval +ve - di points to the first valid record
231 * \retval +1 - di points to exact matched key
232 * \retval -ve - failure
234 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
235 const struct dt_key *key)
237 struct osd_it_quota *it = (struct osd_it_quota *)di;
238 const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
240 qid_t dqid = *(qid_t *)key;
245 type = fid2type(fid);
247 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
248 LUSTRE_DQTREEOFF, 0, it);
249 if (offset > 0) { /* Found */
251 } else if (offset < 0) { /* Error */
252 QUOTA_IT_READ_ERROR(it, (int)offset);
256 /* The @key is not found, move to the first valid entry */
257 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
270 * \param di - osd iterator
272 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
277 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
279 struct osd_quota_leaf *leaf;
284 INIT_LIST_HEAD(&leaf->oql_link);
285 leaf->oql_blk = it->oiq_blk[depth];
286 list_add_tail(&leaf->oql_link, &it->oiq_list);
291 * Move on to the next valid entry.
293 * \param di - osd iterator
295 * \retval +ve - iterator reached the end
296 * \retval 0 - iterator has not reached the end yet
297 * \retval -ve - unexpected failure
299 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
301 struct osd_it_quota *it = (struct osd_it_quota *)di;
302 const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
309 type = fid2type(fid);
312 * Let's first check if there are any remaining valid entry in the
313 * current leaf block. Start with the next entry after the current one.
315 depth = LUSTRE_DQTREEDEPTH;
316 index = it->oiq_index[depth];
317 if (++index < LUSTRE_DQSTRINBLK) {
318 /* Search for the next valid entry from current index */
319 rc = walk_block_dqentry(env, it->oiq_obj, type,
320 it->oiq_blk[depth], index, it);
322 QUOTA_IT_READ_ERROR(it, rc);
324 } else if (rc == 0) {
326 * Found on entry, @it is already updated to the
327 * new position in walk_block_dqentry().
331 rc = osd_it_add_processed(it, depth);
336 rc = osd_it_add_processed(it, depth);
343 * We have consumed all the entries of the current leaf block, move on
349 * We keep searching as long as walk_tree_dqentry() returns +1
350 * (= no valid entry found).
352 for (; depth >= 0 && rc > 0; depth--) {
353 index = it->oiq_index[depth];
356 rc = walk_tree_dqentry(env, it->oiq_obj, type,
357 it->oiq_blk[depth], depth, index, it);
361 QUOTA_IT_READ_ERROR(it, rc);
366 * Return pointer to the key under iterator.
368 * \param di - osd iterator
370 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
371 const struct dt_it *di)
373 struct osd_it_quota *it = (struct osd_it_quota *)di;
376 RETURN((struct dt_key *)&it->oiq_id);
380 * Return size of key under iterator (in bytes)
382 * \param di - osd iterator
384 static int osd_it_acct_key_size(const struct lu_env *env,
385 const struct dt_it *di)
387 struct osd_it_quota *it = (struct osd_it_quota *)di;
390 RETURN((int)sizeof(it->oiq_id));
394 * Return pointer to the record under iterator.
396 * \param di - osd iterator
397 * \param attr - not used
399 static int osd_it_acct_rec(const struct lu_env *env,
400 const struct dt_it *di,
401 struct dt_rec *dtrec, __u32 attr)
403 struct osd_it_quota *it = (struct osd_it_quota *)di;
404 const struct dt_key *key = osd_it_acct_key(env, di);
409 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
410 RETURN(rc > 0 ? 0 : rc);
414 * Returns cookie for current Iterator position.
416 * \param di - osd iterator
418 static __u64 osd_it_acct_store(const struct lu_env *env,
419 const struct dt_it *di)
421 struct osd_it_quota *it = (struct osd_it_quota *)di;
428 * Restore iterator from cookie. if the \a hash isn't found,
429 * restore the first valid record.
431 * \param di - osd iterator
432 * \param hash - iterator location cookie
434 * \retval +ve - di points to the first valid record
435 * \retval +1 - di points to exact matched hash
436 * \retval -ve - failure
438 static int osd_it_acct_load(const struct lu_env *env,
439 const struct dt_it *di, __u64 hash)
441 struct osd_it_quota *it = (struct osd_it_quota *)di;
446 * LU-8999 - If it is called to resume the iteration, calling
447 * osd_it_acct_get could change the block orders in the lower level
448 * of the quota tree, which are saved in osd_it_quota->oiq_blk.
450 if (it->oiq_id != 0 && it->oiq_id == hash)
453 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
454 (const struct dt_key *)&hash));
458 * Index and Iterator operations for accounting objects
460 const struct dt_index_operations osd_acct_index_ops = {
461 .dio_lookup = osd_acct_index_lookup,
463 .init = osd_it_acct_init,
464 .fini = osd_it_acct_fini,
465 .get = osd_it_acct_get,
466 .put = osd_it_acct_put,
467 .next = osd_it_acct_next,
468 .key = osd_it_acct_key,
469 .key_size = osd_it_acct_key_size,
470 .rec = osd_it_acct_rec,
471 .store = osd_it_acct_store,
472 .load = osd_it_acct_load
476 static inline void osd_quota_swab(char *ptr, size_t size)
480 LASSERT((size & (sizeof(__u64) - 1)) == 0);
482 for (offset = 0; offset < size; offset += sizeof(__u64))
483 __swab64s((__u64 *)(ptr + offset));
486 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
487 const struct dt_rec *rec,
488 union lquota_rec *quota_rec)
491 struct iam_descr *descr;
493 LASSERT(obj->oo_dir != NULL);
494 descr = obj->oo_dir->od_container.ic_descr;
496 memcpy(quota_rec, rec, descr->id_rec_size);
498 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
499 return (const struct dt_rec *)quota_rec;
505 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
508 struct iam_descr *descr;
510 LASSERT(obj->oo_dir != NULL);
511 descr = obj->oo_dir->od_container.ic_descr;
513 osd_quota_swab((char *)rec, descr->id_rec_size);
517 static inline int osd_qid_type(struct osd_thandle *oh, int i)
519 return oh->ot_id_types[i];
523 * Reserve journal credits for quota files update first, then call
524 * ->op_begin() to perform quota enforcement.
526 * \param env - the environment passed by the caller
527 * \param oh - osd transaction handle
528 * \param qi - quota id & space required for this operation
529 * \param obj - osd object, could be NULL when it's under create
530 * \param enforce - whether to perform quota enforcement
531 * \param flags - if the operation is write, return no user quota, no
532 * group quota, or sync commit flags to the caller
534 * \retval 0 - success
535 * \retval -ve - failure
537 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
538 struct lquota_id_info *qi, struct osd_object *obj,
539 bool enforce, enum osd_quota_local_flags *local_flags)
541 struct osd_device *dev;
542 struct qsd_instance *qsd;
543 struct inode *inode = NULL;
545 __u8 res = qi->lqi_is_blk ? LQUOTA_RES_DT : LQUOTA_RES_MD;
551 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
554 dev = osd_dt_dev(oh->ot_super.th_dev);
555 LASSERT(dev != NULL);
557 if (res == LQUOTA_RES_DT)
558 qsd = dev->od_quota_slave_dt;
560 qsd = dev->od_quota_slave_md;
562 for (i = 0; i < oh->ot_id_cnt; i++) {
563 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
564 oh->ot_id_res[i] == res &&
565 oh->ot_id_types[i] == qi->lqi_type) {
572 /* we need to account for credits for this new ID */
573 if (i >= OSD_MAX_UGID_CNT) {
574 CERROR("Too many(%d) trans qids!\n", i + 1);
579 inode = obj->oo_inode;
581 if (qi->lqi_id.qid_uid == 0) {
582 /* root ID should be always present in the quota file */
585 /* can't rely on the current state as it can change
587 * if used space for this ID could be dropped to zero,
588 * reserve extra credits for removing ID entry from
591 if (qi->lqi_space < 0)
592 crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
594 crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
597 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
599 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
600 oh->ot_id_types[i] = qi->lqi_type;
601 oh->ot_id_res[i] = res;
605 if (unlikely(qsd == NULL))
606 /* quota slave instance hasn't been allocated yet */
611 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi,
617 * Wrapper for osd_declare_qid()
619 * \param env - the environment passed by the caller
620 * \param uid - user id of the inode
621 * \param gid - group id of the inode
622 * \param space - how many blocks/inodes will be consumed/released
623 * \param oh - osd transaction handle
624 * \param obj - osd object, could be NULL when it's under create
625 * \param flags - if the operation is write, return no user quota, no
626 * group quota, or sync commit flags to the caller
627 * \param osd_qid_flags - indicate this is a inode/block accounting
628 * and whether changes are performed by root user
630 * \retval 0 - success
631 * \retval -ve - failure
633 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
634 __u32 projid, long long space, struct osd_thandle *oh,
635 struct osd_object *obj,
636 enum osd_quota_local_flags *local_flags,
637 enum osd_qid_declare_flags osd_qid_declare_flags)
639 struct osd_thread_info *info = osd_oti_get(env);
640 struct lquota_id_info *qi = &info->oti_qi;
641 int rcu, rcg, rcp = 0; /* user & group & project rc */
642 struct thandle *th = &oh->ot_super;
643 bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE) ||
647 /* let's start with user quota */
648 qi->lqi_id.qid_uid = uid;
649 qi->lqi_type = USRQUOTA;
650 qi->lqi_space = space;
651 qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK);
652 rcu = osd_declare_qid(env, oh, qi, obj, true, local_flags);
654 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
655 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
659 * For non-fatal error, we want to continue to get the noquota flags
660 * for group id. This is only for commit write, which has @flags passed
661 * in. See osd_declare_write_commit().
662 * When force is set to true, we also want to proceed with the gid
664 if (rcu && (rcu != -EDQUOT || local_flags == NULL))
667 /* and now group quota */
668 qi->lqi_id.qid_gid = gid;
669 qi->lqi_type = GRPQUOTA;
670 rcg = osd_declare_qid(env, oh, qi, obj, true, local_flags);
672 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
673 /* as before, ignore EDQUOT & EINPROGRESS for root */
676 #ifdef HAVE_PROJECT_QUOTA
677 if (rcg && (rcg != -EDQUOT || local_flags == NULL))
680 /* and now project quota */
681 qi->lqi_id.qid_projid = projid;
682 qi->lqi_type = PRJQUOTA;
683 rcp = osd_declare_qid(env, oh, qi, obj, true, local_flags);
685 if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS)) {
686 CDEBUG(D_QUOTA, "forced to ignore quota flags = %#x\n",
687 local_flags ? *local_flags : -1);
688 /* as before, ignore EDQUOT & EINPROGRESS for root */
693 RETURN(rcu ? rcu : (rcg ? rcg : rcp));