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) 2011, 2012, Whamcloud, Inc.
25 * Use is subject to license terms.
27 * Author: Johann Lombardi <johann@whamcloud.com>
28 * Author: Niu Yawei <niu@whamcloud.com>
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);
70 LASSERT(fid_is_acct(fid));
72 if (!LDISKFS_HAS_RO_COMPAT_FEATURE(sb,
73 LDISKFS_FEATURE_RO_COMPAT_QUOTA))
76 id->oii_gen = OSD_OII_NOGEN;
77 id->oii_ino = LDISKFS_SB(sb)->s_qf_inums[fid2type(fid)];
78 if (!ldiskfs_valid_inum(sb, id->oii_ino))
84 * Return space usage (#blocks & #inodes) consumed by a given uid or gid.
86 * \param env - is the environment passed by the caller
87 * \param dtobj - is the accounting object
88 * \param dtrec - is the record to fill with space usage information
89 * \param dtkey - is the id the of the user or group for which we would
90 * like to access disk usage.
91 * \param capa - is the capability, not used.
93 * \retval +ve - success : exact match
94 * \retval -ve - failure
96 static int osd_acct_index_lookup(const struct lu_env *env,
97 struct dt_object *dtobj,
99 const struct dt_key *dtkey,
100 struct lustre_capa *capa)
102 struct osd_thread_info *info = osd_oti_get(env);
103 struct if_dqblk *dqblk = &info->oti_dqblk;
104 struct super_block *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
105 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
106 __u64 id = *((__u64 *)dtkey);
111 memset((void *)dqblk, 0, sizeof(struct obd_dqblk));
112 rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t) id, dqblk);
115 rec->bspace = dqblk->dqb_curspace;
116 rec->ispace = dqblk->dqb_curinodes;
120 #define QUOTA_IT_READ_ERROR(it, rc) \
121 CERROR("%s: Error while trying to read quota information, " \
122 "failed with %d\n", \
123 it->oiq_obj->oo_dt.do_lu.lo_dev->ld_obd->obd_name, rc); \
126 * Initialize osd Iterator for given osd index object.
128 * \param dt - osd index object
129 * \param attr - not used
130 * \param capa - BYPASS_CAPA
132 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
133 struct dt_object *dt,
134 __u32 attr, struct lustre_capa *capa)
136 struct osd_thread_info *info = osd_oti_get(env);
137 struct osd_it_quota *it;
138 struct lu_object *lo = &dt->do_lu;
139 struct osd_object *obj = osd_dt_obj(dt);
143 LASSERT(lu_object_exists(lo));
146 RETURN(ERR_PTR(-ENOMEM));
148 it = &info->oti_it_quota;
149 memset(it, 0, sizeof(*it));
152 CFS_INIT_LIST_HEAD(&it->oiq_list);
154 /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
155 it->oiq_blk[0] = LUSTRE_DQTREEOFF;
157 /* NB: we don't need to store the tree depth since it is always
158 * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
160 RETURN((struct dt_it *)it);
164 * Free given iterator.
166 * \param di - osd iterator
168 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
170 struct osd_it_quota *it = (struct osd_it_quota *)di;
171 struct osd_quota_leaf *leaf, *tmp;
174 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
176 cfs_list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
177 cfs_list_del_init(&leaf->oql_link);
184 * Move Iterator to record specified by \a key, if the \a key isn't found,
185 * move to the first valid record.
187 * \param di - osd iterator
188 * \param key - uid or gid
190 * \retval +ve - di points to the first valid record
191 * \retval +1 - di points to exact matched key
192 * \retval -ve - failure
194 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
195 const struct dt_key *key)
197 struct osd_it_quota *it = (struct osd_it_quota *)di;
198 const struct lu_fid *fid =
199 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
200 int type = fid2type(fid);
201 qid_t dqid = *(qid_t *)key;
207 offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
208 LUSTRE_DQTREEOFF, 0, it);
209 if (offset > 0) { /* Found */
211 } else if (offset < 0) { /* Error */
212 QUOTA_IT_READ_ERROR(it, (int)offset);
216 /* The @key is not found, move to the first valid entry */
217 rc = walk_tree_dqentry(env, it->oiq_obj, type, it->oiq_blk[0], 0,
230 * \param di - osd iterator
232 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
237 static int osd_it_add_processed(struct osd_it_quota *it, int depth)
239 struct osd_quota_leaf *leaf;
244 CFS_INIT_LIST_HEAD(&leaf->oql_link);
245 leaf->oql_blk = it->oiq_blk[depth];
246 cfs_list_add_tail(&leaf->oql_link, &it->oiq_list);
251 * Move on to the next valid entry.
253 * \param di - osd iterator
255 * \retval +ve - iterator reached the end
256 * \retval 0 - iterator has not reached the end yet
257 * \retval -ve - unexpected failure
259 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
261 struct osd_it_quota *it = (struct osd_it_quota *)di;
262 const struct lu_fid *fid =
263 lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
264 int type = fid2type(fid);
270 /* Let's first check if there are any remaining valid entry in the
271 * current leaf block. Start with the next entry after the current one.
273 depth = LUSTRE_DQTREEDEPTH;
274 index = it->oiq_index[depth];
275 if (++index < LUSTRE_DQSTRINBLK) {
276 /* Search for the next valid entry from current index */
277 rc = walk_block_dqentry(env, it->oiq_obj, type,
278 it->oiq_blk[depth], index, it);
280 QUOTA_IT_READ_ERROR(it, rc);
282 } else if (rc == 0) {
283 /* Found on entry, @it is already updated to the
284 * new position in walk_block_dqentry(). */
287 rc = osd_it_add_processed(it, depth);
292 rc = osd_it_add_processed(it, depth);
298 /* We have consumed all the entries of the current leaf block, move on
299 * to the next one. */
302 /* We keep searching as long as walk_tree_dqentry() returns +1
303 * (= no valid entry found). */
304 for (; depth >= 0 && rc > 0; depth--) {
305 index = it->oiq_index[depth];
308 rc = walk_tree_dqentry(env, it->oiq_obj, type,
309 it->oiq_blk[depth], depth, index, it);
313 QUOTA_IT_READ_ERROR(it, rc);
318 * Return pointer to the key under iterator.
320 * \param di - osd iterator
322 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
323 const struct dt_it *di)
325 struct osd_it_quota *it = (struct osd_it_quota *)di;
328 RETURN((struct dt_key *)&it->oiq_id);
332 * Return size of key under iterator (in bytes)
334 * \param di - osd iterator
336 static int osd_it_acct_key_size(const struct lu_env *env,
337 const struct dt_it *di)
339 struct osd_it_quota *it = (struct osd_it_quota *)di;
342 RETURN((int)sizeof(it->oiq_id));
346 * Return pointer to the record under iterator.
348 * \param di - osd iterator
349 * \param attr - not used
351 static int osd_it_acct_rec(const struct lu_env *env,
352 const struct dt_it *di,
353 struct dt_rec *dtrec, __u32 attr)
355 struct osd_it_quota *it = (struct osd_it_quota *)di;
356 const struct dt_key *key = osd_it_acct_key(env, di);
361 rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key,
363 RETURN(rc > 0 ? 0 : rc);
367 * Returns cookie for current Iterator position.
369 * \param di - osd iterator
371 static __u64 osd_it_acct_store(const struct lu_env *env,
372 const struct dt_it *di)
374 struct osd_it_quota *it = (struct osd_it_quota *)di;
381 * Restore iterator from cookie. if the \a hash isn't found,
382 * restore the first valid record.
384 * \param di - osd iterator
385 * \param hash - iterator location cookie
387 * \retval +ve - di points to the first valid record
388 * \retval +1 - di points to exact matched hash
389 * \retval -ve - failure
391 static int osd_it_acct_load(const struct lu_env *env,
392 const struct dt_it *di, __u64 hash)
395 RETURN(osd_it_acct_get(env, (struct dt_it *)di,
396 (const struct dt_key *)&hash));
400 * Index and Iterator operations for accounting objects
402 const struct dt_index_operations osd_acct_index_ops = {
403 .dio_lookup = osd_acct_index_lookup,
405 .init = osd_it_acct_init,
406 .fini = osd_it_acct_fini,
407 .get = osd_it_acct_get,
408 .put = osd_it_acct_put,
409 .next = osd_it_acct_next,
410 .key = osd_it_acct_key,
411 .key_size = osd_it_acct_key_size,
412 .rec = osd_it_acct_rec,
413 .store = osd_it_acct_store,
414 .load = osd_it_acct_load
418 static inline void osd_quota_swab(char *ptr, size_t size)
422 LASSERT((size & (sizeof(__u64) - 1)) == 0);
424 for (offset = 0; offset < size; offset += sizeof(__u64))
425 __swab64s((__u64 *)(ptr + offset));
428 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
429 const struct dt_rec *rec,
430 union lquota_rec *quota_rec)
433 struct iam_descr *descr;
435 LASSERT(obj->oo_dir != NULL);
436 descr = obj->oo_dir->od_container.ic_descr;
438 memcpy(quota_rec, rec, descr->id_rec_size);
440 osd_quota_swab((char *)quota_rec, descr->id_rec_size);
441 return (const struct dt_rec *)quota_rec;
447 void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
450 struct iam_descr *descr;
452 LASSERT(obj->oo_dir != NULL);
453 descr = obj->oo_dir->od_container.ic_descr;
455 osd_quota_swab((char *)rec, descr->id_rec_size);
461 static inline int osd_qid_type(struct osd_thandle *oh, int i)
463 return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
466 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
468 oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
472 * Reserve journal credits for quota files update first, then call
473 * ->op_begin() to perform quota enforcement.
475 * \param env - the environment passed by the caller
476 * \param oh - osd transaction handle
477 * \param qi - quota id & space required for this operation
478 * \param allocated - dquot entry in quota accounting file has been allocated
479 * \param flags - if the operation is write, return no user quota, no
480 * group quota, or sync commit flags to the caller
482 * \retval 0 - success
483 * \retval -ve - failure
485 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
486 struct lquota_id_info *qi, bool allocated, int *flags)
488 struct osd_thread_info *info = osd_oti_get(env);
489 struct osd_device *dev = info->oti_dev;
490 struct qsd_instance *qsd = dev->od_quota_slave;
496 LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
499 for (i = 0; i < oh->ot_id_cnt; i++) {
500 if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
501 osd_qid_type(oh, i) == qi->lqi_type) {
508 /* we need to account for credits for this new ID */
509 if (i >= OSD_MAX_UGID_CNT) {
510 CERROR("Too many(%d) trans qids!\n", i + 1);
514 oh->ot_credits += (allocated || qi->lqi_id.qid_uid == 0) ?
515 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
517 oh->ot_id_array[i] = qi->lqi_id.qid_uid;
518 osd_qid_set_type(oh, i, qi->lqi_type);
522 if (unlikely(qsd == NULL))
523 /* quota slave instance hasn't been allocated yet */
527 rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
532 * Wrapper for osd_declare_qid()
534 * \param env - the environment passed by the caller
535 * \param uid - user id of the inode
536 * \param gid - group id of the inode
537 * \param space - how many blocks/inodes will be consumed/released
538 * \param oh - osd transaction handle
539 * \param is_blk - block quota or inode quota?
540 * \param allocated - dquot entry in quota accounting file has been allocated
541 * \param flags - if the operation is write, return no user quota, no
542 * group quota, or sync commit flags to the caller
543 * \param force - set to 1 when changes are performed by root user and thus
544 * can't failed with EDQUOT
546 * \retval 0 - success
547 * \retval -ve - failure
549 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
550 long long space, struct osd_thandle *oh,
551 bool is_blk, bool allocated, int *flags, bool force)
553 struct osd_thread_info *info = osd_oti_get(env);
554 struct lquota_id_info *qi = &info->oti_qi;
555 int rcu, rcg; /* user & group rc */
558 /* let's start with user quota */
559 qi->lqi_id.qid_uid = uid;
560 qi->lqi_type = USRQUOTA;
561 qi->lqi_space = space;
562 qi->lqi_is_blk = is_blk;
563 rcu = osd_declare_qid(env, oh, qi, allocated, flags);
565 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
566 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
569 /* For non-fatal error, we want to continue to get the noquota flags
570 * for group id. This is only for commit write, which has @flags passed
571 * in. See osd_declare_write_commit().
572 * When force is set to true, we also want to proceed with the gid */
573 if (rcu && (rcu != -EDQUOT || flags == NULL))
576 /* and now group quota */
577 qi->lqi_id.qid_gid = gid;
578 qi->lqi_type = GRPQUOTA;
579 rcg = osd_declare_qid(env, oh, qi, allocated, flags);
581 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
582 /* as before, ignore EDQUOT & EINPROGRESS for root */
585 RETURN(rcu ? rcu : rcg);