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;
52 uint64_t uidobjs, bshift;
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 bshift = fls64(osd->od_max_blksz) - 1;
60 uidobjs = osd_objs_count_estimate(refdbytes, usedobjs,
61 uidbytes >> bshift, bshift);
63 /* if we have at least 1 byte, we have at least one dnode ... */
64 uidobjs = max_t(uint64_t, uidobjs, 1);
70 * Space Accounting Management
74 * Return space usage consumed by a given uid or gid.
75 * Block usage is accurrate since it is maintained by DMU itself.
76 * However, DMU does not provide inode accounting, so the #inodes in use
77 * is estimated from the block usage and statfs information.
79 * \param env - is the environment passed by the caller
80 * \param dtobj - is the accounting object
81 * \param dtrec - is the record to fill with space usage information
82 * \param dtkey - is the id the of the user or group for which we would
83 * like to access disk usage.
85 * \retval +ve - success : exact match
86 * \retval -ve - failure
88 static int osd_acct_index_lookup(const struct lu_env *env,
89 struct dt_object *dtobj,
91 const struct dt_key *dtkey)
93 struct osd_thread_info *info = osd_oti_get(env);
94 char *buf = info->oti_buf;
95 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
96 struct osd_object *obj = osd_dt_obj(dtobj);
97 struct osd_device *osd = osd_obj2dev(obj);
102 rec->bspace = rec->ispace = 0;
104 /* convert the 64-bit uid/gid into a string */
105 sprintf(buf, "%llx", *((__u64 *)dtkey));
106 /* fetch DMU object ID (DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT) to be
108 oid = osd_quota_fid2dmu(lu_object_fid(&dtobj->do_lu));
110 /* disk usage (in bytes) is maintained by DMU.
111 * DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT are special objects which
112 * not associated with any dmu_but_t (see dnode_special_open()).
113 * As a consequence, we cannot use udmu_zap_lookup() here since it
114 * requires a valid oo_db. */
115 rc = -zap_lookup(osd->od_os, oid, buf, sizeof(uint64_t), 1,
118 /* user/group has not created anything yet */
119 CDEBUG(D_QUOTA, "%s: id %s not found in DMU accounting ZAP\n",
120 osd->od_svname, buf);
124 if (osd->od_quota_iused_est) {
125 if (rec->bspace != 0)
126 /* estimate #inodes in use */
127 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
131 /* as for inode accounting, it is not maintained by DMU, so we just
132 * use our own ZAP to track inode usage */
133 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
134 buf, sizeof(uint64_t), 1, &rec->ispace);
136 /* user/group has not created any file yet */
137 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
138 osd->od_svname, buf);
146 * Initialize osd Iterator for given osd index object.
148 * \param dt - osd index object
149 * \param attr - not used
151 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
152 struct dt_object *dt,
155 struct osd_thread_info *info = osd_oti_get(env);
156 struct osd_it_quota *it;
157 struct lu_object *lo = &dt->do_lu;
158 struct osd_device *osd = osd_dev(lo->lo_dev);
162 LASSERT(lu_object_exists(lo));
165 RETURN(ERR_PTR(-ENOMEM));
169 RETURN(ERR_PTR(-ENOMEM));
171 memset(it, 0, sizeof(*it));
172 it->oiq_oid = osd_quota_fid2dmu(lu_object_fid(lo));
174 /* initialize zap cursor */
175 rc = osd_zap_cursor_init(&it->oiq_zc, osd->od_os, it->oiq_oid, 0);
181 /* take object reference */
183 it->oiq_obj = osd_dt_obj(dt);
186 RETURN((struct dt_it *)it);
190 * Free given iterator.
192 * \param di - osd iterator
194 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
196 struct osd_it_quota *it = (struct osd_it_quota *)di;
199 osd_zap_cursor_fini(it->oiq_zc);
200 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
207 * Move on to the next valid entry.
209 * \param di - osd iterator
211 * \retval +ve - iterator reached the end
212 * \retval 0 - iterator has not reached the end yet
213 * \retval -ve - unexpected failure
215 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
217 struct osd_it_quota *it = (struct osd_it_quota *)di;
218 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
222 if (it->oiq_reset == 0)
223 zap_cursor_advance(it->oiq_zc);
225 rc = -zap_cursor_retrieve(it->oiq_zc, za);
226 if (rc == -ENOENT) /* reached the end */
232 * Return pointer to the key under iterator.
234 * \param di - osd iterator
236 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
237 const struct dt_it *di)
239 struct osd_it_quota *it = (struct osd_it_quota *)di;
240 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
245 rc = -zap_cursor_retrieve(it->oiq_zc, za);
248 rc = kstrtoull(za->za_name, 16, &it->oiq_id);
250 RETURN((struct dt_key *) &it->oiq_id);
254 * Return size of key under iterator (in bytes)
256 * \param di - osd iterator
258 static int osd_it_acct_key_size(const struct lu_env *env,
259 const struct dt_it *di)
262 RETURN((int)sizeof(uint64_t));
266 * zap_cursor_retrieve read from current record.
267 * to read bytes we need to call zap_lookup explicitly.
269 static int osd_zap_cursor_retrieve_value(const struct lu_env *env,
270 zap_cursor_t *zc, char *buf,
271 int buf_size, int *bytes_read)
273 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
276 rc = -zap_cursor_retrieve(zc, za);
277 if (unlikely(rc != 0))
280 if (unlikely(za->za_integer_length <= 0))
283 actual_size = za->za_integer_length * za->za_num_integers;
285 if (actual_size > buf_size) {
286 actual_size = buf_size;
287 buf_size = actual_size / za->za_integer_length;
289 buf_size = za->za_num_integers;
292 rc = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
293 za->za_name, za->za_integer_length,
297 *bytes_read = actual_size;
303 * Return pointer to the record under iterator.
305 * \param di - osd iterator
306 * \param attr - not used
308 static int osd_it_acct_rec(const struct lu_env *env,
309 const struct dt_it *di,
310 struct dt_rec *dtrec, __u32 attr)
312 struct osd_thread_info *info = osd_oti_get(env);
313 zap_attribute_t *za = &info->oti_za;
314 struct osd_it_quota *it = (struct osd_it_quota *)di;
315 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
316 struct osd_object *obj = it->oiq_obj;
317 struct osd_device *osd = osd_obj2dev(obj);
323 rec->ispace = rec->bspace = 0;
325 /* retrieve block usage from the DMU accounting object */
326 rc = osd_zap_cursor_retrieve_value(env, it->oiq_zc,
327 (char *)&rec->bspace,
328 sizeof(uint64_t), &bytes_read);
332 if (osd->od_quota_iused_est) {
333 if (rec->bspace != 0)
334 /* estimate #inodes in use */
335 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
339 /* retrieve key associated with the current cursor */
340 rc = -zap_cursor_retrieve(it->oiq_zc, za);
341 if (unlikely(rc != 0))
344 /* inode accounting is not maintained by DMU, so we use our own ZAP to
345 * track inode usage */
346 rc = -zap_lookup(osd->od_os, it->oiq_obj->oo_db->db_object,
347 za->za_name, sizeof(uint64_t), 1, &rec->ispace);
349 /* user/group has not created any file yet */
350 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
351 osd->od_svname, za->za_name);
359 * Returns cookie for current Iterator position.
361 * \param di - osd iterator
363 static __u64 osd_it_acct_store(const struct lu_env *env,
364 const struct dt_it *di)
366 struct osd_it_quota *it = (struct osd_it_quota *)di;
369 RETURN(osd_zap_cursor_serialize(it->oiq_zc));
373 * Restore iterator from cookie. if the \a hash isn't found,
374 * restore the first valid record.
376 * \param di - osd iterator
377 * \param hash - iterator location cookie
379 * \retval +ve - di points to exact matched key
380 * \retval 0 - di points to the first valid record
381 * \retval -ve - failure
383 static int osd_it_acct_load(const struct lu_env *env,
384 const struct dt_it *di, __u64 hash)
386 struct osd_it_quota *it = (struct osd_it_quota *)di;
387 struct osd_device *osd = osd_obj2dev(it->oiq_obj);
388 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
393 /* create new cursor pointing to the new hash */
394 rc = osd_zap_cursor_init(&zc, osd->od_os, it->oiq_oid, hash);
397 osd_zap_cursor_fini(it->oiq_zc);
401 rc = -zap_cursor_retrieve(it->oiq_zc, za);
404 else if (rc == -ENOENT)
411 * Move Iterator to record specified by \a key, if the \a key isn't found,
412 * move to the first valid record.
414 * \param di - osd iterator
415 * \param key - uid or gid
417 * \retval +ve - di points to exact matched key
418 * \retval 0 - di points to the first valid record
419 * \retval -ve - failure
421 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
422 const struct dt_key *key)
426 /* XXX: like osd_zap_it_get(), API is currently broken */
427 LASSERT(*((__u64 *)key) == 0);
429 RETURN(osd_it_acct_load(env, di, 0));
435 * \param di - osd iterator
437 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
442 * Index and Iterator operations for accounting objects
444 const struct dt_index_operations osd_acct_index_ops = {
445 .dio_lookup = osd_acct_index_lookup,
447 .init = osd_it_acct_init,
448 .fini = osd_it_acct_fini,
449 .get = osd_it_acct_get,
450 .put = osd_it_acct_put,
451 .next = osd_it_acct_next,
452 .key = osd_it_acct_key,
453 .key_size = osd_it_acct_key_size,
454 .rec = osd_it_acct_rec,
455 .store = osd_it_acct_store,
456 .load = osd_it_acct_load
461 * Quota Enforcement Management
465 * Wrapper for qsd_op_begin().
467 * \param env - the environment passed by the caller
468 * \param osd - is the osd_device
469 * \param uid - user id of the inode
470 * \param gid - group id of the inode
471 * \param space - how many blocks/inodes will be consumed/released
472 * \param oh - osd transaction handle
473 * \param is_blk - block quota or inode quota?
474 * \param flags - if the operation is write, return no user quota, no
475 * group quota, or sync commit flags to the caller
476 * \param force - set to 1 when changes are performed by root user and thus
477 * can't failed with EDQUOT
479 * \retval 0 - success
480 * \retval -ve - failure
482 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
483 qid_t uid, qid_t gid, long long space,
484 struct osd_thandle *oh, bool is_blk, int *flags,
487 struct osd_thread_info *info = osd_oti_get(env);
488 struct lquota_id_info *qi = &info->oti_qi;
489 struct qsd_instance *qsd = osd->od_quota_slave;
490 int rcu, rcg; /* user & group rc */
493 if (unlikely(qsd == NULL))
494 /* quota slave instance hasn't been allocated yet */
497 /* let's start with user quota */
498 qi->lqi_id.qid_uid = uid;
499 qi->lqi_type = USRQUOTA;
500 qi->lqi_space = space;
501 qi->lqi_is_blk = is_blk;
502 rcu = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
504 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
505 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
508 /* For non-fatal error, we want to continue to get the noquota flags
509 * for group id. This is only for commit write, which has @flags passed
510 * in. See osd_declare_write_commit().
511 * When force is set to true, we also want to proceed with the gid */
512 if (rcu && (rcu != -EDQUOT || flags == NULL))
515 /* and now group quota */
516 qi->lqi_id.qid_gid = gid;
517 qi->lqi_type = GRPQUOTA;
518 rcg = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
520 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
521 /* as before, ignore EDQUOT & EINPROGRESS for root */
524 RETURN(rcu ? rcu : rcg);