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_net.h>
51 #include <lustre/lustre_idl.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
241 * cannot be turned on/off on-fly;
242 * temporary used by SOM.
246 * original unit size of file quota and
247 * upper limitation for adjust file qunit
249 unsigned long lqc_iunit_sz;
251 * Trigger dqacq when available file
252 * quota less than this value, trigger
253 * dqrel when available file quota
254 * more than this value + 1 iunit
256 unsigned long lqc_itune_sz;
258 * original unit size of block quota and
259 * upper limitation for adjust block qunit
261 unsigned long lqc_bunit_sz;
262 /** See comment of lqc_itune_sz */
263 unsigned long lqc_btune_sz;
264 /** all lustre_qunit_size structures */
265 struct lustre_hash *lqc_lqs_hash;
269 * the values below are relative to how master change its qunit sizes
272 * this affects the boundary of
273 * shrinking and enlarging qunit size. default=4
275 unsigned long lqc_cqs_boundary_factor;
276 /** the least value of block qunit */
277 unsigned long lqc_cqs_least_bunit;
278 /** the least value of inode qunit */
279 unsigned long lqc_cqs_least_iunit;
281 * when enlarging, qunit size will
282 * mutilple it; when shrinking,
283 * qunit size will divide it
285 unsigned long lqc_cqs_qs_factor;
287 * avoid ping-pong effect of
288 * adjusting qunit size. How many
289 * seconds must be waited between
290 * enlarging and shinking qunit
293 int lqc_switch_seconds;
295 * when blk qunit reaches this value,
296 * later write reqs from client should be sync b=16642
299 /** guard lqc_imp_valid now */
302 * when mds isn't connected, threads
303 * on osts who send the quota reqs
304 * with wait==1 will be put here b=14840
306 cfs_waitq_t lqc_wait_for_qmaster;
307 struct proc_dir_entry *lqc_proc_dir;
308 /** lquota statistics */
309 struct lprocfs_stats *lqc_stats;
312 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
313 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
315 struct lustre_qunit_size {
316 struct hlist_node lqs_hash; /** the hash entry */
317 unsigned int lqs_id; /** id of user/group */
318 unsigned long lqs_flags; /** is user/group; FULLBUF or LESSBUF */
319 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
321 * Trigger dqacq when available file quota
322 * less than this value, trigger dqrel
323 * when more than this value + 1 iunit
325 unsigned long lqs_itune_sz;
326 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
327 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
328 /** the blocks reached ost and don't finish */
329 unsigned long lqs_bwrite_pending;
330 /** the inodes reached mds and don't finish */
331 unsigned long lqs_iwrite_pending;
332 /** when inodes are allocated/released, this value will record it */
333 long long lqs_ino_rec;
334 /** when blocks are allocated/released, this value will record it */
335 long long lqs_blk_rec;
336 atomic_t lqs_refcount;
337 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
338 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
340 unsigned long long lqs_key; /** hash key */
341 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
344 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
345 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
346 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
348 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
349 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
350 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
352 /* In the hash for lustre_qunit_size, the key is decided by
353 * grp_or_usr and uid/gid, in here, I combine these two values,
354 * which will make comparing easier and more efficient */
355 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
356 #define LQS_KEY_ID(key) (key & 0xffffffff)
357 #define LQS_KEY_GRP(key) (key >> 32)
359 static inline void lqs_getref(struct lustre_qunit_size *lqs)
361 atomic_inc(&lqs->lqs_refcount);
362 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n",
363 lqs, atomic_read(&lqs->lqs_refcount));
366 static inline void lqs_putref(struct lustre_qunit_size *lqs)
368 LASSERT(atomic_read(&lqs->lqs_refcount) > 0);
370 /* killing last ref, let's let hash table kill it */
371 if (atomic_read(&lqs->lqs_refcount) == 1) {
372 lustre_hash_del(lqs->lqs_ctxt->lqc_lqs_hash,
373 &lqs->lqs_key, &lqs->lqs_hash);
376 atomic_dec(&lqs->lqs_refcount);
377 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n",
378 lqs, atomic_read(&lqs->lqs_refcount));
383 static inline void lqs_initref(struct lustre_qunit_size *lqs)
385 atomic_set(&lqs->lqs_refcount, 0);
390 struct lustre_quota_info {
393 struct lustre_quota_ctxt {
396 #define QUOTA_MASTER_READY(qctxt)
397 #define QUOTA_MASTER_UNREADY(qctxt)
399 #endif /* !__KERNEL__ */
403 #define LL_DQUOT_OFF(sb) do {} while(0)
405 struct lustre_quota_info {
408 struct lustre_quota_ctxt {
411 #endif /* !HAVE_QUOTA_SUPPORT */
413 /* If the (quota limit < qunit * slave count), the slave which can't
414 * acquire qunit should set it's local limit as MIN_QLIMIT */
417 struct quotacheck_thread_args {
418 struct obd_export *qta_exp; /** obd export */
419 struct obd_device *qta_obd; /** obd device */
420 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
421 struct super_block *qta_sb; /** obd super block */
422 atomic_t *qta_sem; /** obt_quotachecking */
425 struct obd_trans_info;
426 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
427 struct obd_trans_info *oti, int isblk);
430 int (*quota_init) (void);
431 int (*quota_exit) (void);
432 int (*quota_setup) (struct obd_device *);
433 int (*quota_cleanup) (struct obd_device *);
435 * For quota master, close admin quota files
437 int (*quota_fs_cleanup) (struct obd_device *);
438 int (*quota_ctl) (struct obd_device *, struct obd_export *,
439 struct obd_quotactl *);
440 int (*quota_check) (struct obd_device *, struct obd_export *,
441 struct obd_quotactl *);
442 int (*quota_recovery) (struct obd_device *);
445 * For quota master/slave, adjust quota limit after fs operation
447 int (*quota_adjust) (struct obd_device *, const unsigned int[],
448 const unsigned int[], int, int);
451 * For quota slave, set import, trigger quota recovery,
452 * For quota master, set lqc_setup
454 int (*quota_setinfo) (struct obd_device *, void *);
457 * For quota slave, clear import when relative import is invalid
459 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
462 * For quota slave, set proper thread resoure capability
464 int (*quota_enforce) (struct obd_device *, unsigned int);
467 * For quota slave, check whether specified uid/gid is over quota
469 int (*quota_getflag) (struct obd_device *, struct obdo *);
473 * For quota slave, acquire/release quota from master if needed
475 int (*quota_acquire) (struct obd_device *, const unsigned int [],
476 struct obd_trans_info *, int);
479 * For quota slave, check whether specified uid/gid's remaining quota
480 * can finish a block_write or inode_create rpc. It updates the pending
481 * record of block and inode, acquires quota if necessary
483 int (*quota_chkquota) (struct obd_device *, const unsigned int [],
484 int [], int, quota_acquire,
485 struct obd_trans_info *, int, struct inode *,
489 * For quota client, the actions after the pending write is committed
491 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
496 * For quota client, poll if the quota check done
498 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
501 * For quota client, check whether specified uid/gid is over quota
503 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
506 * For quota client, set over quota flag for specifed uid/gid
508 int (*quota_setdq) (struct client_obd *, const unsigned int [],
512 * For adjusting qunit size b=10600
514 int (*quota_adjust_qunit) (struct obd_export *exp,
515 struct quota_adjust_qunit *oqaq,
516 struct lustre_quota_ctxt *qctxt);
520 #define Q_COPY(out, in, member) (out)->member = (in)->member
522 #define QUOTA_OP(interface, op) interface->quota_ ## op
524 #define QUOTA_CHECK_OP(interface, op) \
528 if (!QUOTA_OP(interface, op)) { \
529 CERROR("no quota operation: " #op "\n"); \
530 RETURN(-EOPNOTSUPP); \
534 static inline int lquota_init(quota_interface_t *interface)
539 QUOTA_CHECK_OP(interface, init);
540 rc = QUOTA_OP(interface, init)();
544 static inline int lquota_exit(quota_interface_t *interface)
549 QUOTA_CHECK_OP(interface, exit);
550 rc = QUOTA_OP(interface, exit)();
554 static inline int lquota_setup(quota_interface_t *interface,
555 struct obd_device *obd)
560 QUOTA_CHECK_OP(interface, setup);
561 rc = QUOTA_OP(interface, setup)(obd);
565 static inline int lquota_cleanup(quota_interface_t *interface,
566 struct obd_device *obd)
571 QUOTA_CHECK_OP(interface, cleanup);
572 rc = QUOTA_OP(interface, cleanup)(obd);
576 static inline int lquota_fs_cleanup(quota_interface_t *interface,
577 struct obd_device *obd)
582 QUOTA_CHECK_OP(interface, fs_cleanup);
583 rc = QUOTA_OP(interface, fs_cleanup)(obd);
587 static inline int lquota_recovery(quota_interface_t *interface,
588 struct obd_device *obd)
593 QUOTA_CHECK_OP(interface, recovery);
594 rc = QUOTA_OP(interface, recovery)(obd);
598 static inline int lquota_check(quota_interface_t *interface,
599 struct obd_device *obd,
600 struct obd_export *exp,
601 struct obd_quotactl *oqctl)
606 QUOTA_CHECK_OP(interface, check);
607 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
611 static inline int lquota_ctl(quota_interface_t *interface,
612 struct obd_device *obd,
613 struct obd_quotactl *oqctl)
618 QUOTA_CHECK_OP(interface, ctl);
619 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
623 static inline int lquota_adjust(quota_interface_t *interface,
624 struct obd_device *obd,
625 const unsigned int qcids[],
626 const unsigned int qpids[],
632 QUOTA_CHECK_OP(interface, adjust);
633 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
637 static inline int lquota_chkdq(quota_interface_t *interface,
638 struct client_obd *cli, const unsigned int qid[])
643 QUOTA_CHECK_OP(interface, chkdq);
644 rc = QUOTA_OP(interface, chkdq)(cli, qid);
648 static inline int lquota_setdq(quota_interface_t *interface,
649 struct client_obd *cli, const unsigned int qid[],
650 obd_flag valid, obd_flag flags)
655 QUOTA_CHECK_OP(interface, setdq);
656 rc = QUOTA_OP(interface, setdq)(cli, qid, valid, flags);
660 static inline int lquota_poll_check(quota_interface_t *interface,
661 struct obd_export *exp,
662 struct if_quotacheck *qchk)
667 QUOTA_CHECK_OP(interface, poll_check);
668 rc = QUOTA_OP(interface, poll_check)(exp, qchk);
672 static inline int lquota_setinfo(quota_interface_t *interface,
673 struct obd_device *obd,
679 QUOTA_CHECK_OP(interface, setinfo);
680 rc = QUOTA_OP(interface, setinfo)(obd, data);
684 static inline int lquota_clearinfo(quota_interface_t *interface,
685 struct obd_export *exp,
686 struct obd_device *obd)
691 QUOTA_CHECK_OP(interface, clearinfo);
692 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
696 static inline int lquota_enforce(quota_interface_t *interface,
697 struct obd_device *obd,
703 QUOTA_CHECK_OP(interface, enforce);
704 rc = QUOTA_OP(interface, enforce)(obd, ignore);
708 static inline int lquota_getflag(quota_interface_t *interface,
709 struct obd_device *obd, struct obdo *oa)
714 QUOTA_CHECK_OP(interface, getflag);
715 rc = QUOTA_OP(interface, getflag)(obd, oa);
720 static inline int lquota_chkquota(quota_interface_t *interface,
721 struct obd_device *obd,
722 const unsigned int id[], int pending[],
723 int count, struct obd_trans_info *oti,
724 int isblk, void *data, int frags)
729 QUOTA_CHECK_OP(interface, chkquota);
730 QUOTA_CHECK_OP(interface, acquire);
731 rc = QUOTA_OP(interface, chkquota)(obd, id, pending, count,
732 QUOTA_OP(interface, acquire), oti,
733 isblk, (struct inode *)data, frags);
737 static inline int lquota_pending_commit(quota_interface_t *interface,
738 struct obd_device *obd,
739 const unsigned int id[],
740 int pending[], int isblk)
745 QUOTA_CHECK_OP(interface, pending_commit);
746 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
752 extern quota_interface_t osc_quota_interface;
753 extern quota_interface_t lov_quota_interface;
754 extern quota_interface_t mdc_quota_interface;
755 extern quota_interface_t lmv_quota_interface;
771 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
772 "admin_quotafile_v2.usr", /** user admin quotafile */\
773 "admin_quotafile_v2.grp" /** group admin quotafile */\
776 #endif /* _LUSTRE_QUOTA_H */