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 <dt_object.h>
32 #include <lustre_quota.h>
33 #include "osd_internal.h"
36 * Helpers function to find out the quota type (USRQUOTA/GRPQUOTA) of a
39 static inline int fid2type(const struct lu_fid *fid)
41 LASSERT(fid_is_acct(fid));
42 switch (fid_oid(fid)) {
47 case ACCT_PROJECT_OID:
51 LASSERTF(0, "invalid fid for quota type: %u\n", fid_oid(fid));
56 * Space Accounting Management
60 * Look up an accounting object based on its fid.
62 * \param info - is the osd thread info passed by the caller
63 * \param osd - is the osd device
64 * \param fid - is the fid of the accounting object we want to look up
65 * \param id - is the osd_inode_id struct to fill with the inode number of
66 * the quota file if the lookup is successful
68 int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
69 const struct lu_fid *fid, struct osd_inode_id *id)
71 struct super_block *sb = osd_sb(osd);
74 LASSERT(fid_is_acct(fid));
76 if (!ldiskfs_has_feature_quota(sb))
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_feature_project(sb)) {
94 prj_quota = LDISKFS_SB(sb)->s_es->s_prj_quota_inum;
95 id->oii_ino = le32_to_cpu(prj_quota);
102 if (!ldiskfs_valid_inum(sb, id->oii_ino) &&
103 id->oii_ino != LDISKFS_USR_QUOTA_INO &&
104 id->oii_ino != LDISKFS_GRP_QUOTA_INO)
111 * Return space usage (#blocks & #inodes) consumed by a given uid or gid.
113 * \param env - is the environment passed by the caller
114 * \param dtobj - is the accounting object
115 * \param dtrec - is the record to fill with space usage information
116 * \param dtkey - is the id of the user or group for which we would
117 * like to access disk usage.
119 * \retval +ve - success : exact match
120 * \retval -ve - failure
122 static int osd_acct_index_lookup(const struct lu_env *env,
123 struct dt_object *dtobj,
124 struct dt_rec *dtrec,
125 const struct dt_key *dtkey)
127 struct osd_thread_info *info = osd_oti_get(env);
128 #if defined(HAVE_DQUOT_QC_DQBLK)
129 struct qc_dqblk *dqblk = &info->oti_qdq;
131 struct fs_disk_quota *dqblk = &info->oti_fdq;
133 struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
134 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
135 __u64 id = *((__u64 *)dtkey);
142 type = fid2type(lu_object_fid(&dtobj->do_lu));
143 memset(dqblk, 0, sizeof(*dqblk));
144 qid = make_kqid(&init_user_ns, type, id);
145 rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
148 #if defined(HAVE_DQUOT_QC_DQBLK)
149 rec->bspace = dqblk->d_space;
150 rec->ispace = dqblk->d_ino_count;
152 rec->bspace = dqblk->d_bcount;
153 rec->ispace = dqblk->d_icount;
158 #define QUOTA_IT_READ_ERROR(it, rc) \
159 CERROR("%s: Error while trying to read quota information, " \
160 "failed with %d\n", \
161 osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc) \
164 * Initialize osd Iterator for given osd index object.
166 * \param dt - osd index object
167 * \param attr - not used
169 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
170 struct dt_object *dt,
173 struct osd_it_quota *it;
174 struct lu_object *lo = &dt->do_lu;
175 struct osd_object *obj = osd_dt_obj(dt);
179 LASSERT(lu_object_exists(lo));
183 RETURN(ERR_PTR(-ENOMEM));
187 INIT_LIST_HEAD(&it->oiq_list);
189 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
190 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
193 * NB: we don't need to store the tree depth since it is always
194 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
197 RETURN((struct dt_it *)it);
201 * Free given iterator.
203 * \param di - osd iterator
205 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
207 struct osd_it_quota *it = (struct osd_it_quota *)di;
208 struct osd_quota_leaf *leaf, *tmp;
212 osd_object_put(env, it->oiq_obj);
214 list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
215 list_del_init(&leaf->oql_link);
225 * Move Iterator to record specified by \a key, if the \a key isn't found,
226 * move to the first valid record.
228 * \param di - osd iterator
229 * \param key - uid or gid
231 * \retval +ve - di points to the first valid record
232 * \retval +1 - di points to exact matched key
233 * \retval -ve - failure
235 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
236 const struct dt_key *key)
238 struct osd_it_quota *it = (struct osd_it_quota *)di;
239 const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
241 qid_t dqid = *(qid_t *)key;
246 type = fid2type(fid);
248 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
249 LUSTRE_DQTREEOFF, 0, it);
250 if (offset > 0) { /* Found */
252 } else if (offset < 0) { /* Error */
253 QUOTA_IT_READ_ERROR(it, (int)offset);
257 /* The @key is not found, move to the first valid entry */
258 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
271 * \param di - osd iterator
273 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
278 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
280 struct osd_quota_leaf *leaf;
285 INIT_LIST_HEAD(&leaf->oql_link);
286 leaf->oql_blk = it->oiq_blk[depth];
287 list_add_tail(&leaf->oql_link, &it->oiq_list);
292 * Move on to the next valid entry.
294 * \param di - osd iterator
296 * \retval +ve - iterator reached the end
297 * \retval 0 - iterator has not reached the end yet
298 * \retval -ve - unexpected failure
300 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
302 struct osd_it_quota *it = (struct osd_it_quota *)di;
303 const struct lu_fid *fid = lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
310 type = fid2type(fid);
313 * Let's first check if there are any remaining valid entry in the
314 * current leaf block. Start with the next entry after the current one.
316 depth = LUSTRE_DQTREEDEPTH;
317 index = it->oiq_index[depth];
318 if (++index < LUSTRE_DQSTRINBLK) {
319 /* Search for the next valid entry from current index */
320 rc = walk_block_dqentry(env, it->oiq_obj, type,
321 it->oiq_blk[depth], index, it);
323 QUOTA_IT_READ_ERROR(it, rc);
325 } else if (rc == 0) {
327 * Found on entry, @it is already updated to the
328 * new position in walk_block_dqentry().
332 rc = osd_it_add_processed(it, depth);
337 rc = osd_it_add_processed(it, depth);
344 * We have consumed all the entries of the current leaf block, move on
350 * We keep searching as long as walk_tree_dqentry() returns +1
351 * (= no valid entry found).
353 for (; depth >= 0 && rc > 0; depth--) {
354 index = it->oiq_index[depth];
357 rc = walk_tree_dqentry(env, it->oiq_obj, type,
358 it->oiq_blk[depth], depth, index, it);
362 QUOTA_IT_READ_ERROR(it, rc);
367 * Return pointer to the key under iterator.
369 * \param di - osd iterator
371 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
372 const struct dt_it *di)
374 struct osd_it_quota *it = (struct osd_it_quota *)di;
377 RETURN((struct dt_key *)&it->oiq_id);
381 * Return size of key under iterator (in bytes)
383 * \param di - osd iterator
385 static int osd_it_acct_key_size(const struct lu_env *env,
386 const struct dt_it *di)
388 struct osd_it_quota *it = (struct osd_it_quota *)di;
391 RETURN((int)sizeof(it->oiq_id));
395 * Return pointer to the record under iterator.
397 * \param di - osd iterator
398 * \param attr - not used
400 static int osd_it_acct_rec(const struct lu_env *env,
401 const struct dt_it *di,
402 struct dt_rec *dtrec, __u32 attr)
404 struct osd_it_quota *it = (struct osd_it_quota *)di;
405 const struct dt_key *key = osd_it_acct_key(env, di);
410 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
411 RETURN(rc > 0 ? 0 : rc);
415 * Returns cookie for current Iterator position.
417 * \param di - osd iterator
419 static __u64 osd_it_acct_store(const struct lu_env *env,
420 const struct dt_it *di)
422 struct osd_it_quota *it = (struct osd_it_quota *)di;
429 * Restore iterator from cookie. if the \a hash isn't found,
430 * restore the first valid record.
432 * \param di - osd iterator
433 * \param hash - iterator location cookie
435 * \retval +ve - di points to the first valid record
436 * \retval +1 - di points to exact matched hash
437 * \retval -ve - failure
439 static int osd_it_acct_load(const struct lu_env *env,
440 const struct dt_it *di, __u64 hash)
442 struct osd_it_quota *it = (struct osd_it_quota *)di;
447 * LU-8999 - If it is called to resume the iteration, calling
448 * osd_it_acct_get could change the block orders in the lower level
449 * of the quota tree, which are saved in osd_it_quota->oiq_blk.
451 if (it->oiq_id != 0 && it->oiq_id == hash)
454 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
455 (const struct dt_key *)&hash));
459 * Index and Iterator operations for accounting objects
461 const struct dt_index_operations osd_acct_index_ops = {
462 .dio_lookup = osd_acct_index_lookup,
464 .init = osd_it_acct_init,
465 .fini = osd_it_acct_fini,
466 .get = osd_it_acct_get,
467 .put = osd_it_acct_put,
468 .next = osd_it_acct_next,
469 .key = osd_it_acct_key,
470 .key_size = osd_it_acct_key_size,
471 .rec = osd_it_acct_rec,
472 .store = osd_it_acct_store,
473 .load = osd_it_acct_load
477 static inline void osd_quota_swab(char *ptr, size_t size)
481 LASSERT((size & (sizeof(__u64) - 1)) == 0);
483 for (offset = 0; offset < size; offset += sizeof(__u64))
484 __swab64s((__u64 *)(ptr + offset));
487 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
488 const struct dt_rec *rec,
489 union lquota_rec *quota_rec)
492 struct iam_descr *descr;
494 LASSERT(obj->oo_dir != NULL);
495 descr = obj->oo_dir->od_container.ic_descr;
497 memcpy(quota_rec, rec, descr->id_rec_size);
499 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
500 return (const struct dt_rec *)quota_rec;
506 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
509 struct iam_descr *descr;
511 LASSERT(obj->oo_dir != NULL);
512 descr = obj->oo_dir->od_container.ic_descr;
514 osd_quota_swab((char *)rec, descr->id_rec_size);
518 static inline int osd_qid_type(struct osd_thandle *oh, int i)
520 return oh->ot_id_types[i];
524 * Reserve journal credits for quota files update first, then call
525 * ->op_begin() to perform quota enforcement.
527 * \param env - the environment passed by the caller
528 * \param oh - osd transaction handle
529 * \param qi - quota id & space required for this operation
530 * \param obj - osd object, could be NULL when it's under create
531 * \param enforce - whether to perform quota enforcement
532 * \param flags - if the operation is write, return no user quota, no
533 * group quota, or sync commit flags to the caller
535 * \retval 0 - success
536 * \retval -ve - failure
538 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
539 struct lquota_id_info *qi, struct osd_object *obj,
540 bool enforce, enum osd_quota_local_flags *local_flags)
542 struct osd_device *dev;
543 struct qsd_instance *qsd;
544 struct inode *inode = NULL;
546 __u8 res = qi->lqi_is_blk ? LQUOTA_RES_DT : LQUOTA_RES_MD;
552 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
555 dev = osd_dt_dev(oh->ot_super.th_dev);
556 LASSERT(dev != NULL);
558 if (res == LQUOTA_RES_DT)
559 qsd = dev->od_quota_slave_dt;
561 qsd = dev->od_quota_slave_md;
563 for (i = 0; i < oh->ot_id_cnt; i++) {
564 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
565 oh->ot_id_res[i] == res &&
566 oh->ot_id_types[i] == qi->lqi_type) {
573 /* we need to account for credits for this new ID */
574 if (i >= OSD_MAX_UGID_CNT) {
575 CERROR("Too many(%d) trans qids!\n", i + 1);
580 inode = obj->oo_inode;
582 if (qi->lqi_id.qid_uid == 0) {
583 /* root ID should be always present in the quota file */
586 /* can't rely on the current state as it can change
588 * if used space for this ID could be dropped to zero,
589 * reserve extra credits for removing ID entry from
592 if (qi->lqi_space < 0)
593 crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
595 crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
598 osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
600 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
601 oh->ot_id_types[i] = qi->lqi_type;
602 oh->ot_id_res[i] = res;
606 if (unlikely(qsd == NULL))
607 /* quota slave instance hasn't been allocated yet */
612 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi,
618 * Wrapper for osd_declare_qid()
620 * \param env - the environment passed by the caller
621 * \param uid - user id of the inode
622 * \param gid - group id of the inode
623 * \param space - how many blocks/inodes will be consumed/released
624 * \param oh - osd transaction handle
625 * \param obj - osd object, could be NULL when it's under create
626 * \param flags - if the operation is write, return no user quota, no
627 * group quota, or sync commit flags to the caller
628 * \param osd_qid_flags - indicate this is a inode/block accounting
629 * and whether changes are performed by root user
631 * \retval 0 - success
632 * \retval -ve - failure
634 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
635 __u32 projid, long long space, struct osd_thandle *oh,
636 struct osd_object *obj,
637 enum osd_quota_local_flags *local_flags,
638 enum osd_qid_declare_flags osd_qid_declare_flags)
640 struct osd_thread_info *info = osd_oti_get(env);
641 struct lquota_id_info *qi = &info->oti_qi;
642 int rcu, rcg, rcp = 0; /* user & group & project rc */
643 struct thandle *th = &oh->ot_super;
644 bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE) ||
648 /* let's start with user quota */
649 qi->lqi_id.qid_uid = uid;
650 qi->lqi_type = USRQUOTA;
651 qi->lqi_space = space;
652 qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK);
653 rcu = osd_declare_qid(env, oh, qi, obj, true, local_flags);
655 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
656 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
660 * For non-fatal error, we want to continue to get the noquota flags
661 * for group id. This is only for commit write, which has @flags passed
662 * in. See osd_declare_write_commit().
663 * When force is set to true, we also want to proceed with the gid
665 if (rcu && (rcu != -EDQUOT || local_flags == NULL))
668 /* and now group quota */
669 qi->lqi_id.qid_gid = gid;
670 qi->lqi_type = GRPQUOTA;
671 rcg = osd_declare_qid(env, oh, qi, obj, true, local_flags);
673 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
674 /* as before, ignore EDQUOT & EINPROGRESS for root */
677 #ifdef HAVE_PROJECT_QUOTA
678 if (rcg && (rcg != -EDQUOT || local_flags == NULL))
681 /* and now project quota */
682 qi->lqi_id.qid_projid = projid;
683 qi->lqi_type = PRJQUOTA;
684 rcp = osd_declare_qid(env, oh, qi, obj, true, local_flags);
686 if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS)) {
687 CDEBUG(D_QUOTA, "forced to ignore quota flags = %#x\n",
688 local_flags ? *local_flags : -1);
689 /* as before, ignore EDQUOT & EINPROGRESS for root */
694 RETURN(rcu ? rcu : (rcg ? rcg : rcp));