1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _LUSTRE_QUOTA_H
38 #define _LUSTRE_QUOTA_H
40 #if defined(__linux__)
41 #include <linux/lustre_quota.h>
42 #elif defined(__APPLE__)
43 #include <darwin/lustre_quota.h>
44 #elif defined(__WINNT__)
45 #include <winnt/lustre_quota.h>
47 #error Unsupported operating system.
50 #include <lustre/lustre_idl.h>
51 #include <lustre_net.h>
53 #include <obd_support.h>
54 #include <class_hash.h>
63 #ifdef HAVE_QUOTA_SUPPORT
69 LQUOTA_FIRST_STAT = 0,
72 * these four are for measuring quota requests, for both of
73 * quota master and quota slaves
75 LQUOTA_SYNC_ACQ = LQUOTA_FIRST_STAT,
82 * these four measure how much time I/O threads spend on dealing
83 * with quota before and after writing data or creating files,
84 * only for quota slaves(lquota_chkquota and lquota_pending_commit)
86 LQUOTA_WAIT_FOR_CHK_BLK,
87 LQUOTA_WAIT_FOR_CHK_INO,
88 LQUOTA_WAIT_FOR_COMMIT_BLK,
89 LQUOTA_WAIT_FOR_COMMIT_INO,
93 * these two are for measuring time waiting return of quota reqs
94 * (qctxt_wait_pending_dqacq), only for quota salves
96 LQUOTA_WAIT_PENDING_BLK_QUOTA,
97 LQUOTA_WAIT_PENDING_INO_QUOTA,
101 * these two are for those when they are calling
102 * qctxt_wait_pending_dqacq, the quota req has returned already,
103 * only for quota salves
105 LQUOTA_NOWAIT_PENDING_BLK_QUOTA,
106 LQUOTA_NOWAIT_PENDING_INO_QUOTA,
110 * these are for quota ctl
116 * these are for adjust quota qunit, for both of
117 * quota master and quota slaves
125 /* structures to access admin quotafile */
126 struct lustre_mem_dqinfo {
127 unsigned int dqi_bgrace;
128 unsigned int dqi_igrace;
129 unsigned long dqi_flags;
130 unsigned int dqi_blocks;
131 unsigned int dqi_free_blk;
132 unsigned int dqi_free_entry;
135 struct lustre_quota_info {
136 struct file *qi_files[MAXQUOTAS];
137 struct lustre_mem_dqinfo qi_info[MAXQUOTAS];
138 lustre_quota_version_t qi_version;
141 #define DQ_STATUS_AVAIL 0x0 /* Available dquot */
142 #define DQ_STATUS_SET 0x01 /* Sombody is setting dquot */
143 #define DQ_STATUS_RECOVERY 0x02 /* dquot is in recovery */
145 struct lustre_mem_dqblk {
146 __u64 dqb_bhardlimit; /**< absolute limit on disk blks alloc */
147 __u64 dqb_bsoftlimit; /**< preferred limit on disk blks */
148 __u64 dqb_curspace; /**< current used space */
149 __u64 dqb_ihardlimit; /**< absolute limit on allocated inodes */
150 __u64 dqb_isoftlimit; /**< preferred inode limit */
151 __u64 dqb_curinodes; /**< current # allocated inodes */
152 time_t dqb_btime; /**< time limit for excessive disk use */
153 time_t dqb_itime; /**< time limit for excessive inode use */
156 struct lustre_dquot {
157 /** Hash list in memory, protect by dquot_hash_lock */
158 struct list_head dq_hash;
159 /** Protect the data in lustre_dquot */
160 struct semaphore dq_sem;
163 /** Pointer of quota info it belongs to */
164 struct lustre_quota_info *dq_info;
165 /** Offset of dquot on disk */
167 /** ID this applies to (uid, gid) */
169 /** Type fo quota (USRQUOTA, GRPQUOUTA) */
171 /** See DQ_STATUS_ */
172 unsigned short dq_status;
173 /** See DQ_ in quota.h */
174 unsigned long dq_flags;
175 /** Diskquota usage */
176 struct lustre_mem_dqblk dq_dqb;
180 struct list_head di_link;
185 #define QFILE_RD_INFO 2
186 #define QFILE_WR_INFO 3
187 #define QFILE_INIT_INFO 4
188 #define QFILE_RD_DQUOT 5
189 #define QFILE_WR_DQUOT 6
190 #define QFILE_CONVERT 7
192 /* admin quotafile operations */
193 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
194 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
195 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
196 int lustre_read_dquot(struct lustre_dquot *dquot);
197 int lustre_commit_dquot(struct lustre_dquot *dquot);
198 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
199 int lustre_get_qids(struct file *file, struct inode *inode, int type,
200 struct list_head *list);
201 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
203 #define LL_DQUOT_OFF(sb) DQUOT_OFF(sb)
205 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
208 /* user quota is turned on on filter */
209 #define LQC_USRQUOTA_FLAG (1 << 0)
210 /* group quota is turned on on filter */
211 #define LQC_GRPQUOTA_FLAG (1 << 1)
213 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
214 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
216 struct lustre_quota_ctxt {
217 /** superblock this applies to */
218 struct super_block *lqc_sb;
219 /** obd_device_target for obt_rwsem */
220 struct obd_device_target *lqc_obt;
221 /** import used to send dqacq/dqrel RPC */
222 struct obd_import *lqc_import;
223 /** dqacq/dqrel RPC handler, only for quota master */
224 dqacq_handler_t lqc_handler;
226 unsigned long lqc_flags;
228 unsigned long lqc_recovery:1, /** Doing recovery */
230 * the function of change qunit size
233 lqc_valid:1, /** this qctxt is valid or not */
235 * tell whether of not quota_type has
236 * been processed, so that the master
237 * knows when it can start processing
238 * incoming acq/rel quota requests
242 * original unit size of file quota and
243 * upper limitation for adjust file qunit
245 unsigned long lqc_iunit_sz;
247 * Trigger dqacq when available file
248 * quota less than this value, trigger
249 * dqrel when available file quota
250 * more than this value + 1 iunit
252 unsigned long lqc_itune_sz;
254 * original unit size of block quota and
255 * upper limitation for adjust block qunit
257 unsigned long lqc_bunit_sz;
258 /** See comment of lqc_itune_sz */
259 unsigned long lqc_btune_sz;
260 /** all lustre_qunit_size structures */
261 struct lustre_hash *lqc_lqs_hash;
265 * the values below are relative to how master change its qunit sizes
268 * this affects the boundary of
269 * shrinking and enlarging qunit size. default=4
271 unsigned long lqc_cqs_boundary_factor;
272 /** the least value of block qunit */
273 unsigned long lqc_cqs_least_bunit;
274 /** the least value of inode qunit */
275 unsigned long lqc_cqs_least_iunit;
277 * when enlarging, qunit size will
278 * mutilple it; when shrinking,
279 * qunit size will divide it
281 unsigned long lqc_cqs_qs_factor;
283 * avoid ping-pong effect of
284 * adjusting qunit size. How many
285 * seconds must be waited between
286 * enlarging and shinking qunit
289 int lqc_switch_seconds;
291 * when blk qunit reaches this value,
292 * later write reqs from client should be sync b=16642
295 /** guard lqc_imp_valid now */
298 * when mds isn't connected, threads
299 * on osts who send the quota reqs
300 * with wait==1 will be put here b=14840
302 cfs_waitq_t lqc_wait_for_qmaster;
303 struct proc_dir_entry *lqc_proc_dir;
304 /** lquota statistics */
305 struct lprocfs_stats *lqc_stats;
308 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
309 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
311 struct lustre_qunit_size {
312 struct hlist_node lqs_hash; /** the hash entry */
313 unsigned int lqs_id; /** id of user/group */
314 unsigned long lqs_flags; /** is user/group; FULLBUF or LESSBUF */
315 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
317 * Trigger dqacq when available file quota
318 * less than this value, trigger dqrel
319 * when more than this value + 1 iunit
321 unsigned long lqs_itune_sz;
322 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
323 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
324 /** the blocks reached ost and don't finish */
325 unsigned long lqs_bwrite_pending;
326 /** the inodes reached mds and don't finish */
327 unsigned long lqs_iwrite_pending;
328 /** when inodes are allocated/released, this value will record it */
329 long long lqs_ino_rec;
330 /** when blocks are allocated/released, this value will record it */
331 long long lqs_blk_rec;
332 atomic_t lqs_refcount;
333 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
334 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
336 struct quota_adjust_qunit lqs_key; /** hash key */
337 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
340 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
341 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
342 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
344 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
345 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
346 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
348 static inline void lqs_getref(struct lustre_qunit_size *lqs)
350 atomic_inc(&lqs->lqs_refcount);
351 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n",
352 lqs, atomic_read(&lqs->lqs_refcount));
355 static inline void lqs_putref(struct lustre_qunit_size *lqs)
357 LASSERT(atomic_read(&lqs->lqs_refcount) > 0);
359 /* killing last ref, let's let hash table kill it */
360 if (atomic_read(&lqs->lqs_refcount) == 1) {
361 lustre_hash_del(lqs->lqs_ctxt->lqc_lqs_hash,
362 &lqs->lqs_key, &lqs->lqs_hash);
365 atomic_dec(&lqs->lqs_refcount);
366 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n",
367 lqs, atomic_read(&lqs->lqs_refcount));
372 static inline void lqs_initref(struct lustre_qunit_size *lqs)
374 atomic_set(&lqs->lqs_refcount, 0);
379 struct lustre_quota_info {
382 struct lustre_quota_ctxt {
385 #define QUOTA_MASTER_READY(qctxt)
386 #define QUOTA_MASTER_UNREADY(qctxt)
388 #endif /* !__KERNEL__ */
392 #define LL_DQUOT_OFF(sb) do {} while(0)
394 struct lustre_quota_info {
397 struct lustre_quota_ctxt {
400 #endif /* !HAVE_QUOTA_SUPPORT */
402 /* If the (quota limit < qunit * slave count), the slave which can't
403 * acquire qunit should set it's local limit as MIN_QLIMIT */
406 struct quotacheck_thread_args {
407 struct obd_export *qta_exp; /** obd export */
408 struct obd_device *qta_obd; /** obd device */
409 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
410 struct super_block *qta_sb; /** obd super block */
411 atomic_t *qta_sem; /** obt_quotachecking */
414 struct obd_trans_info;
415 typedef int (*quota_acquire)(struct obd_device *obd, unsigned int uid,
416 unsigned int gid, struct obd_trans_info *oti,
420 int (*quota_init) (void);
421 int (*quota_exit) (void);
422 int (*quota_setup) (struct obd_device *);
423 int (*quota_cleanup) (struct obd_device *);
425 * For quota master, close admin quota files
427 int (*quota_fs_cleanup) (struct obd_device *);
428 int (*quota_ctl) (struct obd_device *, struct obd_export *,
429 struct obd_quotactl *);
430 int (*quota_check) (struct obd_device *, struct obd_export *,
431 struct obd_quotactl *);
432 int (*quota_recovery) (struct obd_device *);
435 * For quota master/slave, adjust quota limit after fs operation
437 int (*quota_adjust) (struct obd_device *, unsigned int[],
438 unsigned int[], int, int);
441 * For quota slave, set import, trigger quota recovery,
442 * For quota master, set lqc_setup
444 int (*quota_setinfo) (struct obd_device *, void *);
447 * For quota slave, clear import when relative import is invalid
449 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
452 * For quota slave, set proper thread resoure capability
454 int (*quota_enforce) (struct obd_device *, unsigned int);
457 * For quota slave, check whether specified uid/gid is over quota
459 int (*quota_getflag) (struct obd_device *, struct obdo *);
463 * For quota slave, acquire/release quota from master if needed
465 int (*quota_acquire) (struct obd_device *, unsigned int, unsigned int,
466 struct obd_trans_info *, int);
469 * For quota slave, check whether specified uid/gid's remaining quota
470 * can finish a block_write or inode_create rpc. It updates the pending
471 * record of block and inode, acquires quota if necessary
473 int (*quota_chkquota) (struct obd_device *, unsigned int, unsigned int,
474 int, int *, quota_acquire,
475 struct obd_trans_info *, int, struct inode *,
479 * For quota client, the actions after the pending write is committed
481 int (*quota_pending_commit) (struct obd_device *, unsigned int,
482 unsigned int, int, int);
486 * For quota client, poll if the quota check done
488 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
491 * For quota client, check whether specified uid/gid is over quota
493 int (*quota_chkdq) (struct client_obd *, unsigned int, unsigned int);
496 * For quota client, set over quota flag for specifed uid/gid
498 int (*quota_setdq) (struct client_obd *, unsigned int, unsigned int,
502 * For adjusting qunit size b=10600
504 int (*quota_adjust_qunit) (struct obd_export *exp,
505 struct quota_adjust_qunit *oqaq,
506 struct lustre_quota_ctxt *qctxt);
510 #define Q_COPY(out, in, member) (out)->member = (in)->member
512 #define QUOTA_OP(interface, op) interface->quota_ ## op
514 #define QUOTA_CHECK_OP(interface, op) \
518 if (!QUOTA_OP(interface, op)) { \
519 CERROR("no quota operation: " #op "\n"); \
520 RETURN(-EOPNOTSUPP); \
524 static inline int lquota_init(quota_interface_t *interface)
529 QUOTA_CHECK_OP(interface, init);
530 rc = QUOTA_OP(interface, init)();
534 static inline int lquota_exit(quota_interface_t *interface)
539 QUOTA_CHECK_OP(interface, exit);
540 rc = QUOTA_OP(interface, exit)();
544 static inline int lquota_setup(quota_interface_t *interface,
545 struct obd_device *obd)
550 QUOTA_CHECK_OP(interface, setup);
551 rc = QUOTA_OP(interface, setup)(obd);
555 static inline int lquota_cleanup(quota_interface_t *interface,
556 struct obd_device *obd)
561 QUOTA_CHECK_OP(interface, cleanup);
562 rc = QUOTA_OP(interface, cleanup)(obd);
566 static inline int lquota_fs_cleanup(quota_interface_t *interface,
567 struct obd_device *obd)
572 QUOTA_CHECK_OP(interface, fs_cleanup);
573 rc = QUOTA_OP(interface, fs_cleanup)(obd);
577 static inline int lquota_recovery(quota_interface_t *interface,
578 struct obd_device *obd)
583 QUOTA_CHECK_OP(interface, recovery);
584 rc = QUOTA_OP(interface, recovery)(obd);
588 static inline int lquota_check(quota_interface_t *interface,
589 struct obd_device *obd,
590 struct obd_export *exp,
591 struct obd_quotactl *oqctl)
596 QUOTA_CHECK_OP(interface, check);
597 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
601 static inline int lquota_ctl(quota_interface_t *interface,
602 struct obd_device *obd,
603 struct obd_quotactl *oqctl)
608 QUOTA_CHECK_OP(interface, ctl);
609 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
613 static inline int lquota_adjust(quota_interface_t *interface,
614 struct obd_device *obd,
615 unsigned int qcids[],
616 unsigned int qpids[],
622 QUOTA_CHECK_OP(interface, adjust);
623 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
627 static inline int lquota_chkdq(quota_interface_t *interface,
628 struct client_obd *cli,
629 unsigned int uid, unsigned int gid)
634 QUOTA_CHECK_OP(interface, chkdq);
635 rc = QUOTA_OP(interface, chkdq)(cli, uid, gid);
639 static inline int lquota_setdq(quota_interface_t *interface,
640 struct client_obd *cli,
641 unsigned int uid, unsigned int gid,
642 obd_flag valid, obd_flag flags)
647 QUOTA_CHECK_OP(interface, setdq);
648 rc = QUOTA_OP(interface, setdq)(cli, uid, gid, valid, flags);
652 static inline int lquota_poll_check(quota_interface_t *interface,
653 struct obd_export *exp,
654 struct if_quotacheck *qchk)
659 QUOTA_CHECK_OP(interface, poll_check);
660 rc = QUOTA_OP(interface, poll_check)(exp, qchk);
664 static inline int lquota_setinfo(quota_interface_t *interface,
665 struct obd_device *obd,
671 QUOTA_CHECK_OP(interface, setinfo);
672 rc = QUOTA_OP(interface, setinfo)(obd, data);
676 static inline int lquota_clearinfo(quota_interface_t *interface,
677 struct obd_export *exp,
678 struct obd_device *obd)
683 QUOTA_CHECK_OP(interface, clearinfo);
684 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
688 static inline int lquota_enforce(quota_interface_t *interface,
689 struct obd_device *obd,
695 QUOTA_CHECK_OP(interface, enforce);
696 rc = QUOTA_OP(interface, enforce)(obd, ignore);
700 static inline int lquota_getflag(quota_interface_t *interface,
701 struct obd_device *obd, struct obdo *oa)
706 QUOTA_CHECK_OP(interface, getflag);
707 rc = QUOTA_OP(interface, getflag)(obd, oa);
712 static inline int lquota_chkquota(quota_interface_t *interface,
713 struct obd_device *obd,
714 unsigned int uid, unsigned int gid, int count,
715 int *flag, struct obd_trans_info *oti,
716 int isblk, void *data, int frags)
721 QUOTA_CHECK_OP(interface, chkquota);
722 QUOTA_CHECK_OP(interface, acquire);
723 rc = QUOTA_OP(interface, chkquota)(obd, uid, gid, count, flag,
724 QUOTA_OP(interface, acquire), oti,
725 isblk, (struct inode *)data, frags);
729 static inline int lquota_pending_commit(quota_interface_t *interface,
730 struct obd_device *obd,
731 unsigned int uid, unsigned int gid,
732 int pending, int isblk)
737 QUOTA_CHECK_OP(interface, pending_commit);
738 rc = QUOTA_OP(interface, pending_commit)(obd, uid, gid, pending, isblk);
744 extern quota_interface_t osc_quota_interface;
745 extern quota_interface_t lov_quota_interface;
746 extern quota_interface_t mdc_quota_interface;
747 extern quota_interface_t lmv_quota_interface;
763 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
764 "admin_quotafile_v2.usr", /** user admin quotafile */\
765 "admin_quotafile_v2.grp" /** group admin quotafile */\
768 #endif /* _LUSTRE_QUOTA_H */