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>
30 #include <lustre_quota.h>
32 #include "osd_internal.h"
35 * Helper function to retrieve DMU object id from fid for accounting object
37 uint64_t osd_quota_fid2dmu(const struct lu_fid *fid)
39 LASSERT(fid_is_acct(fid));
40 if (fid_oid(fid) == ACCT_GROUP_OID)
41 return DMU_GROUPUSED_OBJECT;
42 return DMU_USERUSED_OBJECT;
46 * Helper function to estimate the number of inodes in use for a give uid/gid
47 * from the block usage
49 static uint64_t osd_objset_user_iused(struct osd_device *osd, uint64_t uidbytes)
51 uint64_t refdbytes, availbytes, usedobjs, availobjs;
54 /* get fresh statfs info */
55 dmu_objset_space(osd->od_os, &refdbytes, &availbytes,
56 &usedobjs, &availobjs);
58 /* estimate the number of objects based on the disk usage */
59 uidobjs = osd_objs_count_estimate(refdbytes, usedobjs,
60 uidbytes >> SPA_MAXBLOCKSHIFT);
62 /* if we have at least 1 byte, we have at least one dnode ... */
63 uidobjs = max_t(uint64_t, uidobjs, 1);
69 * Space Accounting Management
73 * Return space usage consumed by a given uid or gid.
74 * Block usage is accurrate since it is maintained by DMU itself.
75 * However, DMU does not provide inode accounting, so the #inodes in use
76 * is estimated from the block usage and statfs information.
78 * \param env - is the environment passed by the caller
79 * \param dtobj - is the accounting object
80 * \param dtrec - is the record to fill with space usage information
81 * \param dtkey - is the id the of the user or group for which we would
82 * like to access disk usage.
84 * \retval +ve - success : exact match
85 * \retval -ve - failure
87 static int osd_acct_index_lookup(const struct lu_env *env,
88 struct dt_object *dtobj,
90 const struct dt_key *dtkey)
92 struct osd_thread_info *info = osd_oti_get(env);
93 char *buf = info->oti_buf;
94 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
95 struct osd_object *obj = osd_dt_obj(dtobj);
96 struct osd_device *osd = osd_obj2dev(obj);
101 rec->bspace = rec->ispace = 0;
103 /* convert the 64-bit uid/gid into a string */
104 sprintf(buf, "%llx", *((__u64 *)dtkey));
105 /* fetch DMU object ID (DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT) to be
107 oid = osd_quota_fid2dmu(lu_object_fid(&dtobj->do_lu));
109 /* disk usage (in bytes) is maintained by DMU.
110 * DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT are special objects which
111 * not associated with any dmu_but_t (see dnode_special_open()).
112 * As a consequence, we cannot use udmu_zap_lookup() here since it
113 * requires a valid oo_db. */
114 rc = -zap_lookup(osd->od_os, oid, buf, sizeof(uint64_t), 1,
117 /* user/group has not created anything yet */
118 CDEBUG(D_QUOTA, "%s: id %s not found in DMU accounting ZAP\n",
119 osd->od_svname, buf);
123 if (osd->od_quota_iused_est) {
124 if (rec->bspace != 0)
125 /* estimate #inodes in use */
126 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
130 /* as for inode accounting, it is not maintained by DMU, so we just
131 * use our own ZAP to track inode usage */
132 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
133 buf, sizeof(uint64_t), 1, &rec->ispace);
135 /* user/group has not created any file yet */
136 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
137 osd->od_svname, buf);
145 * Initialize osd Iterator for given osd index object.
147 * \param dt - osd index object
148 * \param attr - not used
150 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
151 struct dt_object *dt,
154 struct osd_thread_info *info = osd_oti_get(env);
155 struct osd_it_quota *it;
156 struct lu_object *lo = &dt->do_lu;
157 struct osd_device *osd = osd_dev(lo->lo_dev);
161 LASSERT(lu_object_exists(lo));
164 RETURN(ERR_PTR(-ENOMEM));
168 RETURN(ERR_PTR(-ENOMEM));
170 memset(it, 0, sizeof(*it));
171 it->oiq_oid = osd_quota_fid2dmu(lu_object_fid(lo));
173 /* initialize zap cursor */
174 rc = osd_zap_cursor_init(&it->oiq_zc, osd->od_os, it->oiq_oid, 0);
180 /* take object reference */
182 it->oiq_obj = osd_dt_obj(dt);
185 RETURN((struct dt_it *)it);
189 * Free given iterator.
191 * \param di - osd iterator
193 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
195 struct osd_it_quota *it = (struct osd_it_quota *)di;
198 osd_zap_cursor_fini(it->oiq_zc);
199 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
206 * Move on to the next valid entry.
208 * \param di - osd iterator
210 * \retval +ve - iterator reached the end
211 * \retval 0 - iterator has not reached the end yet
212 * \retval -ve - unexpected failure
214 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
216 struct osd_it_quota *it = (struct osd_it_quota *)di;
217 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
221 if (it->oiq_reset == 0)
222 zap_cursor_advance(it->oiq_zc);
224 rc = -zap_cursor_retrieve(it->oiq_zc, za);
225 if (rc == -ENOENT) /* reached the end */
231 * Return pointer to the key under iterator.
233 * \param di - osd iterator
235 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
236 const struct dt_it *di)
238 struct osd_it_quota *it = (struct osd_it_quota *)di;
239 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
244 rc = -zap_cursor_retrieve(it->oiq_zc, za);
247 rc = kstrtoull(za->za_name, 16, &it->oiq_id);
249 RETURN((struct dt_key *) &it->oiq_id);
253 * Return size of key under iterator (in bytes)
255 * \param di - osd iterator
257 static int osd_it_acct_key_size(const struct lu_env *env,
258 const struct dt_it *di)
261 RETURN((int)sizeof(uint64_t));
265 * zap_cursor_retrieve read from current record.
266 * to read bytes we need to call zap_lookup explicitly.
268 static int osd_zap_cursor_retrieve_value(const struct lu_env *env,
269 zap_cursor_t *zc, char *buf,
270 int buf_size, int *bytes_read)
272 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
275 rc = -zap_cursor_retrieve(zc, za);
276 if (unlikely(rc != 0))
279 if (unlikely(za->za_integer_length <= 0))
282 actual_size = za->za_integer_length * za->za_num_integers;
284 if (actual_size > buf_size) {
285 actual_size = buf_size;
286 buf_size = actual_size / za->za_integer_length;
288 buf_size = za->za_num_integers;
291 rc = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
292 za->za_name, za->za_integer_length,
296 *bytes_read = actual_size;
302 * Return pointer to the record under iterator.
304 * \param di - osd iterator
305 * \param attr - not used
307 static int osd_it_acct_rec(const struct lu_env *env,
308 const struct dt_it *di,
309 struct dt_rec *dtrec, __u32 attr)
311 struct osd_thread_info *info = osd_oti_get(env);
312 zap_attribute_t *za = &info->oti_za;
313 struct osd_it_quota *it = (struct osd_it_quota *)di;
314 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
315 struct osd_object *obj = it->oiq_obj;
316 struct osd_device *osd = osd_obj2dev(obj);
322 rec->ispace = rec->bspace = 0;
324 /* retrieve block usage from the DMU accounting object */
325 rc = osd_zap_cursor_retrieve_value(env, it->oiq_zc,
326 (char *)&rec->bspace,
327 sizeof(uint64_t), &bytes_read);
331 if (osd->od_quota_iused_est) {
332 if (rec->bspace != 0)
333 /* estimate #inodes in use */
334 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
338 /* retrieve key associated with the current cursor */
339 rc = -zap_cursor_retrieve(it->oiq_zc, za);
340 if (unlikely(rc != 0))
343 /* inode accounting is not maintained by DMU, so we use our own ZAP to
344 * track inode usage */
345 rc = -zap_lookup(osd->od_os, it->oiq_obj->oo_db->db_object,
346 za->za_name, sizeof(uint64_t), 1, &rec->ispace);
348 /* user/group has not created any file yet */
349 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
350 osd->od_svname, za->za_name);
358 * Returns cookie for current Iterator position.
360 * \param di - osd iterator
362 static __u64 osd_it_acct_store(const struct lu_env *env,
363 const struct dt_it *di)
365 struct osd_it_quota *it = (struct osd_it_quota *)di;
368 RETURN(osd_zap_cursor_serialize(it->oiq_zc));
372 * Restore iterator from cookie. if the \a hash isn't found,
373 * restore the first valid record.
375 * \param di - osd iterator
376 * \param hash - iterator location cookie
378 * \retval +ve - di points to exact matched key
379 * \retval 0 - di points to the first valid record
380 * \retval -ve - failure
382 static int osd_it_acct_load(const struct lu_env *env,
383 const struct dt_it *di, __u64 hash)
385 struct osd_it_quota *it = (struct osd_it_quota *)di;
386 struct osd_device *osd = osd_obj2dev(it->oiq_obj);
387 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
392 /* create new cursor pointing to the new hash */
393 rc = osd_zap_cursor_init(&zc, osd->od_os, it->oiq_oid, hash);
396 osd_zap_cursor_fini(it->oiq_zc);
400 rc = -zap_cursor_retrieve(it->oiq_zc, za);
403 else if (rc == -ENOENT)
410 * Move Iterator to record specified by \a key, if the \a key isn't found,
411 * move to the first valid record.
413 * \param di - osd iterator
414 * \param key - uid or gid
416 * \retval +ve - di points to exact matched key
417 * \retval 0 - di points to the first valid record
418 * \retval -ve - failure
420 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
421 const struct dt_key *key)
425 /* XXX: like osd_zap_it_get(), API is currently broken */
426 LASSERT(*((__u64 *)key) == 0);
428 RETURN(osd_it_acct_load(env, di, 0));
434 * \param di - osd iterator
436 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
441 * Index and Iterator operations for accounting objects
443 const struct dt_index_operations osd_acct_index_ops = {
444 .dio_lookup = osd_acct_index_lookup,
446 .init = osd_it_acct_init,
447 .fini = osd_it_acct_fini,
448 .get = osd_it_acct_get,
449 .put = osd_it_acct_put,
450 .next = osd_it_acct_next,
451 .key = osd_it_acct_key,
452 .key_size = osd_it_acct_key_size,
453 .rec = osd_it_acct_rec,
454 .store = osd_it_acct_store,
455 .load = osd_it_acct_load
460 * Quota Enforcement Management
464 * Wrapper for qsd_op_begin().
466 * \param env - the environment passed by the caller
467 * \param osd - is the osd_device
468 * \param uid - user id of the inode
469 * \param gid - group id of the inode
470 * \param space - how many blocks/inodes will be consumed/released
471 * \param oh - osd transaction handle
472 * \param is_blk - block quota or inode quota?
473 * \param flags - if the operation is write, return no user quota, no
474 * group quota, or sync commit flags to the caller
475 * \param force - set to 1 when changes are performed by root user and thus
476 * can't failed with EDQUOT
478 * \retval 0 - success
479 * \retval -ve - failure
481 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
482 qid_t uid, qid_t gid, long long space,
483 struct osd_thandle *oh, bool is_blk, int *flags,
486 struct osd_thread_info *info = osd_oti_get(env);
487 struct lquota_id_info *qi = &info->oti_qi;
488 struct qsd_instance *qsd = osd->od_quota_slave;
489 int rcu, rcg; /* user & group rc */
492 if (unlikely(qsd == NULL))
493 /* quota slave instance hasn't been allocated yet */
496 /* let's start with user quota */
497 qi->lqi_id.qid_uid = uid;
498 qi->lqi_type = USRQUOTA;
499 qi->lqi_space = space;
500 qi->lqi_is_blk = is_blk;
501 rcu = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
503 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
504 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
507 /* For non-fatal error, we want to continue to get the noquota flags
508 * for group id. This is only for commit write, which has @flags passed
509 * in. See osd_declare_write_commit().
510 * When force is set to true, we also want to proceed with the gid */
511 if (rcu && (rcu != -EDQUOT || flags == NULL))
514 /* and now group quota */
515 qi->lqi_id.qid_gid = gid;
516 qi->lqi_type = GRPQUOTA;
517 rcg = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
519 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
520 /* as before, ignore EDQUOT & EINPROGRESS for root */
523 RETURN(rcu ? rcu : rcg);