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.
83 * \param capa - is the capability, not used.
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,
92 struct lustre_capa *capa)
94 struct osd_thread_info *info = osd_oti_get(env);
95 char *buf = info->oti_buf;
96 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
97 struct osd_object *obj = osd_dt_obj(dtobj);
98 struct osd_device *osd = osd_obj2dev(obj);
103 rec->bspace = rec->ispace = 0;
105 /* convert the 64-bit uid/gid into a string */
106 sprintf(buf, "%llx", *((__u64 *)dtkey));
107 /* fetch DMU object ID (DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT) to be
109 oid = osd_quota_fid2dmu(lu_object_fid(&dtobj->do_lu));
111 /* disk usage (in bytes) is maintained by DMU.
112 * DMU_USERUSED_OBJECT/DMU_GROUPUSED_OBJECT are special objects which
113 * not associated with any dmu_but_t (see dnode_special_open()).
114 * As a consequence, we cannot use udmu_zap_lookup() here since it
115 * requires a valid oo_db. */
116 rc = -zap_lookup(osd->od_os, oid, buf, sizeof(uint64_t), 1,
119 /* user/group has not created anything yet */
120 CDEBUG(D_QUOTA, "%s: id %s not found in DMU accounting ZAP\n",
121 osd->od_svname, buf);
125 if (osd->od_quota_iused_est) {
126 if (rec->bspace != 0)
127 /* estimate #inodes in use */
128 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
132 /* as for inode accounting, it is not maintained by DMU, so we just
133 * use our own ZAP to track inode usage */
134 rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
135 buf, sizeof(uint64_t), 1, &rec->ispace);
137 /* user/group has not created any file yet */
138 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
139 osd->od_svname, buf);
147 * Initialize osd Iterator for given osd index object.
149 * \param dt - osd index object
150 * \param attr - not used
151 * \param capa - BYPASS_CAPA
153 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
154 struct dt_object *dt,
156 struct lustre_capa *capa)
158 struct osd_thread_info *info = osd_oti_get(env);
159 struct osd_it_quota *it;
160 struct lu_object *lo = &dt->do_lu;
161 struct osd_device *osd = osd_dev(lo->lo_dev);
165 LASSERT(lu_object_exists(lo));
168 RETURN(ERR_PTR(-ENOMEM));
172 RETURN(ERR_PTR(-ENOMEM));
174 memset(it, 0, sizeof(*it));
175 it->oiq_oid = osd_quota_fid2dmu(lu_object_fid(lo));
177 /* initialize zap cursor */
178 rc = osd_zap_cursor_init(&it->oiq_zc, osd->od_os, it->oiq_oid, 0);
184 /* take object reference */
186 it->oiq_obj = osd_dt_obj(dt);
189 RETURN((struct dt_it *)it);
193 * Free given iterator.
195 * \param di - osd iterator
197 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
199 struct osd_it_quota *it = (struct osd_it_quota *)di;
202 osd_zap_cursor_fini(it->oiq_zc);
203 lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
210 * Move on to the next valid entry.
212 * \param di - osd iterator
214 * \retval +ve - iterator reached the end
215 * \retval 0 - iterator has not reached the end yet
216 * \retval -ve - unexpected failure
218 static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
220 struct osd_it_quota *it = (struct osd_it_quota *)di;
221 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
225 if (it->oiq_reset == 0)
226 zap_cursor_advance(it->oiq_zc);
228 rc = -zap_cursor_retrieve(it->oiq_zc, za);
229 if (rc == -ENOENT) /* reached the end */
235 * Return pointer to the key under iterator.
237 * \param di - osd iterator
239 static struct dt_key *osd_it_acct_key(const struct lu_env *env,
240 const struct dt_it *di)
242 struct osd_it_quota *it = (struct osd_it_quota *)di;
243 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
248 rc = -zap_cursor_retrieve(it->oiq_zc, za);
251 rc = kstrtoull(za->za_name, 16, &it->oiq_id);
253 RETURN((struct dt_key *) &it->oiq_id);
257 * Return size of key under iterator (in bytes)
259 * \param di - osd iterator
261 static int osd_it_acct_key_size(const struct lu_env *env,
262 const struct dt_it *di)
265 RETURN((int)sizeof(uint64_t));
269 * zap_cursor_retrieve read from current record.
270 * to read bytes we need to call zap_lookup explicitly.
272 static int osd_zap_cursor_retrieve_value(const struct lu_env *env,
273 zap_cursor_t *zc, char *buf,
274 int buf_size, int *bytes_read)
276 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
279 rc = -zap_cursor_retrieve(zc, za);
280 if (unlikely(rc != 0))
283 if (unlikely(za->za_integer_length <= 0))
286 actual_size = za->za_integer_length * za->za_num_integers;
288 if (actual_size > buf_size) {
289 actual_size = buf_size;
290 buf_size = actual_size / za->za_integer_length;
292 buf_size = za->za_num_integers;
295 rc = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
296 za->za_name, za->za_integer_length,
300 *bytes_read = actual_size;
306 * Return pointer to the record under iterator.
308 * \param di - osd iterator
309 * \param attr - not used
311 static int osd_it_acct_rec(const struct lu_env *env,
312 const struct dt_it *di,
313 struct dt_rec *dtrec, __u32 attr)
315 struct osd_thread_info *info = osd_oti_get(env);
316 zap_attribute_t *za = &info->oti_za;
317 struct osd_it_quota *it = (struct osd_it_quota *)di;
318 struct lquota_acct_rec *rec = (struct lquota_acct_rec *)dtrec;
319 struct osd_object *obj = it->oiq_obj;
320 struct osd_device *osd = osd_obj2dev(obj);
326 rec->ispace = rec->bspace = 0;
328 /* retrieve block usage from the DMU accounting object */
329 rc = osd_zap_cursor_retrieve_value(env, it->oiq_zc,
330 (char *)&rec->bspace,
331 sizeof(uint64_t), &bytes_read);
335 if (osd->od_quota_iused_est) {
336 if (rec->bspace != 0)
337 /* estimate #inodes in use */
338 rec->ispace = osd_objset_user_iused(osd, rec->bspace);
342 /* retrieve key associated with the current cursor */
343 rc = -zap_cursor_retrieve(it->oiq_zc, za);
344 if (unlikely(rc != 0))
347 /* inode accounting is not maintained by DMU, so we use our own ZAP to
348 * track inode usage */
349 rc = -zap_lookup(osd->od_os, it->oiq_obj->oo_db->db_object,
350 za->za_name, sizeof(uint64_t), 1, &rec->ispace);
352 /* user/group has not created any file yet */
353 CDEBUG(D_QUOTA, "%s: id %s not found in accounting ZAP\n",
354 osd->od_svname, za->za_name);
362 * Returns cookie for current Iterator position.
364 * \param di - osd iterator
366 static __u64 osd_it_acct_store(const struct lu_env *env,
367 const struct dt_it *di)
369 struct osd_it_quota *it = (struct osd_it_quota *)di;
372 RETURN(osd_zap_cursor_serialize(it->oiq_zc));
376 * Restore iterator from cookie. if the \a hash isn't found,
377 * restore the first valid record.
379 * \param di - osd iterator
380 * \param hash - iterator location cookie
382 * \retval +ve - di points to exact matched key
383 * \retval 0 - di points to the first valid record
384 * \retval -ve - failure
386 static int osd_it_acct_load(const struct lu_env *env,
387 const struct dt_it *di, __u64 hash)
389 struct osd_it_quota *it = (struct osd_it_quota *)di;
390 struct osd_device *osd = osd_obj2dev(it->oiq_obj);
391 zap_attribute_t *za = &osd_oti_get(env)->oti_za;
396 /* create new cursor pointing to the new hash */
397 rc = osd_zap_cursor_init(&zc, osd->od_os, it->oiq_oid, hash);
400 osd_zap_cursor_fini(it->oiq_zc);
404 rc = -zap_cursor_retrieve(it->oiq_zc, za);
407 else if (rc == -ENOENT)
414 * Move Iterator to record specified by \a key, if the \a key isn't found,
415 * move to the first valid record.
417 * \param di - osd iterator
418 * \param key - uid or gid
420 * \retval +ve - di points to exact matched key
421 * \retval 0 - di points to the first valid record
422 * \retval -ve - failure
424 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
425 const struct dt_key *key)
429 /* XXX: like osd_zap_it_get(), API is currently broken */
430 LASSERT(*((__u64 *)key) == 0);
432 RETURN(osd_it_acct_load(env, di, 0));
438 * \param di - osd iterator
440 static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
445 * Index and Iterator operations for accounting objects
447 const struct dt_index_operations osd_acct_index_ops = {
448 .dio_lookup = osd_acct_index_lookup,
450 .init = osd_it_acct_init,
451 .fini = osd_it_acct_fini,
452 .get = osd_it_acct_get,
453 .put = osd_it_acct_put,
454 .next = osd_it_acct_next,
455 .key = osd_it_acct_key,
456 .key_size = osd_it_acct_key_size,
457 .rec = osd_it_acct_rec,
458 .store = osd_it_acct_store,
459 .load = osd_it_acct_load
464 * Quota Enforcement Management
468 * Wrapper for qsd_op_begin().
470 * \param env - the environment passed by the caller
471 * \param osd - is the osd_device
472 * \param uid - user id of the inode
473 * \param gid - group id of the inode
474 * \param space - how many blocks/inodes will be consumed/released
475 * \param oh - osd transaction handle
476 * \param is_blk - block quota or inode quota?
477 * \param flags - if the operation is write, return no user quota, no
478 * group quota, or sync commit flags to the caller
479 * \param force - set to 1 when changes are performed by root user and thus
480 * can't failed with EDQUOT
482 * \retval 0 - success
483 * \retval -ve - failure
485 int osd_declare_quota(const struct lu_env *env, struct osd_device *osd,
486 qid_t uid, qid_t gid, long long space,
487 struct osd_thandle *oh, bool is_blk, int *flags,
490 struct osd_thread_info *info = osd_oti_get(env);
491 struct lquota_id_info *qi = &info->oti_qi;
492 struct qsd_instance *qsd = osd->od_quota_slave;
493 int rcu, rcg; /* user & group rc */
496 if (unlikely(qsd == NULL))
497 /* quota slave instance hasn't been allocated yet */
500 /* let's start with user quota */
501 qi->lqi_id.qid_uid = uid;
502 qi->lqi_type = USRQUOTA;
503 qi->lqi_space = space;
504 qi->lqi_is_blk = is_blk;
505 rcu = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
507 if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
508 /* ignore EDQUOT & EINPROGRESS when changes are done by root */
511 /* For non-fatal error, we want to continue to get the noquota flags
512 * for group id. This is only for commit write, which has @flags passed
513 * in. See osd_declare_write_commit().
514 * When force is set to true, we also want to proceed with the gid */
515 if (rcu && (rcu != -EDQUOT || flags == NULL))
518 /* and now group quota */
519 qi->lqi_id.qid_gid = gid;
520 qi->lqi_type = GRPQUOTA;
521 rcg = qsd_op_begin(env, qsd, &oh->ot_quota_trans, qi, flags);
523 if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
524 /* as before, ignore EDQUOT & EINPROGRESS for root */
527 RETURN(rcu ? rcu : rcg);