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>
62 #ifndef QUOTABLOCK_BITS
63 #define QUOTABLOCK_BITS 10
66 #ifndef QUOTABLOCK_SIZE
67 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
71 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
74 #ifdef HAVE_QUOTA_SUPPORT
77 #define MAX_IQ_TIME 604800 /* (7*24*60*60) 1 week */
81 #define MAX_DQ_TIME 604800 /* (7*24*60*60) 1 week */
88 LQUOTA_FIRST_STAT = 0,
91 * these four are for measuring quota requests, for both of
92 * quota master and quota slaves
94 LQUOTA_SYNC_ACQ = LQUOTA_FIRST_STAT,
101 * these four measure how much time I/O threads spend on dealing
102 * with quota before and after writing data or creating files,
103 * only for quota slaves(lquota_chkquota and lquota_pending_commit)
105 LQUOTA_WAIT_FOR_CHK_BLK,
106 LQUOTA_WAIT_FOR_CHK_INO,
107 LQUOTA_WAIT_FOR_COMMIT_BLK,
108 LQUOTA_WAIT_FOR_COMMIT_INO,
112 * these two are for measuring time waiting return of quota reqs
113 * (qctxt_wait_pending_dqacq), only for quota salves
115 LQUOTA_WAIT_PENDING_BLK_QUOTA,
116 LQUOTA_WAIT_PENDING_INO_QUOTA,
120 * these two are for those when they are calling
121 * qctxt_wait_pending_dqacq, the quota req has returned already,
122 * only for quota salves
124 LQUOTA_NOWAIT_PENDING_BLK_QUOTA,
125 LQUOTA_NOWAIT_PENDING_INO_QUOTA,
129 * these are for quota ctl
135 * these are for adjust quota qunit, for both of
136 * quota master and quota slaves
144 /* structures to access admin quotafile */
145 struct lustre_mem_dqinfo {
146 unsigned int dqi_bgrace;
147 unsigned int dqi_igrace;
148 unsigned long dqi_flags;
149 unsigned int dqi_blocks;
150 unsigned int dqi_free_blk;
151 unsigned int dqi_free_entry;
154 struct lustre_quota_info {
155 struct file *qi_files[MAXQUOTAS];
156 struct lustre_mem_dqinfo qi_info[MAXQUOTAS];
157 lustre_quota_version_t qi_version;
160 #define DQ_STATUS_AVAIL 0x0 /* Available dquot */
161 #define DQ_STATUS_SET 0x01 /* Sombody is setting dquot */
162 #define DQ_STATUS_RECOVERY 0x02 /* dquot is in recovery */
164 struct lustre_mem_dqblk {
165 __u64 dqb_bhardlimit; /**< absolute limit on disk blks alloc */
166 __u64 dqb_bsoftlimit; /**< preferred limit on disk blks */
167 __u64 dqb_curspace; /**< current used space */
168 __u64 dqb_ihardlimit; /**< absolute limit on allocated inodes */
169 __u64 dqb_isoftlimit; /**< preferred inode limit */
170 __u64 dqb_curinodes; /**< current # allocated inodes */
171 time_t dqb_btime; /**< time limit for excessive disk use */
172 time_t dqb_itime; /**< time limit for excessive inode use */
175 struct lustre_dquot {
176 /** Hash list in memory, protect by dquot_hash_lock */
178 /** Protect the data in lustre_dquot */
179 cfs_semaphore_t dq_sem;
182 /** Pointer of quota info it belongs to */
183 struct lustre_quota_info *dq_info;
184 /** Offset of dquot on disk */
186 /** ID this applies to (uid, gid) */
188 /** Type fo quota (USRQUOTA, GRPQUOUTA) */
190 /** See DQ_STATUS_ */
191 unsigned short dq_status;
192 /** See DQ_ in quota.h */
193 unsigned long dq_flags;
194 /** Diskquota usage */
195 struct lustre_mem_dqblk dq_dqb;
203 /* set inode quota limitation on a quota uid/gid */
204 #define QI_SET (1 << 30)
205 /* set block quota limitation on a quota uid/gid */
206 #define QB_SET (1 << 31)
209 #define QFILE_RD_INFO 2
210 #define QFILE_WR_INFO 3
211 #define QFILE_INIT_INFO 4
212 #define QFILE_RD_DQUOT 5
213 #define QFILE_WR_DQUOT 6
214 #define QFILE_CONVERT 7
216 /* admin quotafile operations */
217 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
218 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
219 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
220 int lustre_read_dquot(struct lustre_dquot *dquot);
221 int lustre_commit_dquot(struct lustre_dquot *dquot);
222 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
223 int lustre_get_qids(struct file *file, struct inode *inode, int type,
225 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
227 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
231 #ifdef HAVE_VFS_DQ_OFF
232 #define LL_DQUOT_OFF(sb, remount) vfs_dq_off(sb, remount)
234 #define LL_DQUOT_OFF(sb, remount) DQUOT_OFF(sb)
238 #define LL_DQUOT_OFF(sb) DQUOT_OFF(sb)
241 /* user quota is turned on on filter */
242 #define LQC_USRQUOTA_FLAG (1 << 0)
243 /* group quota is turned on on filter */
244 #define LQC_GRPQUOTA_FLAG (1 << 1)
246 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
247 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
249 struct lustre_quota_ctxt {
250 /** superblock this applies to */
251 struct super_block *lqc_sb;
252 /** obd_device_target for obt_rwsem */
253 struct obd_device_target *lqc_obt;
254 /** import used to send dqacq/dqrel RPC */
255 struct obd_import *lqc_import;
256 /** dqacq/dqrel RPC handler, only for quota master */
257 dqacq_handler_t lqc_handler;
259 unsigned long lqc_flags;
261 unsigned long lqc_recovery:1, /** Doing recovery */
263 * the function of change qunit size
266 lqc_valid:1, /** this qctxt is valid or not */
268 * tell whether of not quota_type has
269 * been processed, so that the master
270 * knows when it can start processing
271 * incoming acq/rel quota requests
275 * original unit size of file quota and
276 * upper limitation for adjust file qunit
278 unsigned long lqc_iunit_sz;
280 * Trigger dqacq when available file
281 * quota less than this value, trigger
282 * dqrel when available file quota
283 * more than this value + 1 iunit
285 unsigned long lqc_itune_sz;
287 * original unit size of block quota and
288 * upper limitation for adjust block qunit
290 unsigned long lqc_bunit_sz;
291 /** See comment of lqc_itune_sz */
292 unsigned long lqc_btune_sz;
293 /** all lustre_qunit_size structures */
294 cfs_hash_t *lqc_lqs_hash;
298 * the values below are relative to how master change its qunit sizes
301 * this affects the boundary of
302 * shrinking and enlarging qunit size. default=4
304 unsigned long lqc_cqs_boundary_factor;
305 /** the least value of block qunit */
306 unsigned long lqc_cqs_least_bunit;
307 /** the least value of inode qunit */
308 unsigned long lqc_cqs_least_iunit;
310 * when enlarging, qunit size will
311 * mutilple it; when shrinking,
312 * qunit size will divide it
314 unsigned long lqc_cqs_qs_factor;
316 * avoid ping-pong effect of
317 * adjusting qunit size. How many
318 * seconds must be waited between
319 * enlarging and shinking qunit
322 int lqc_switch_seconds;
324 * when blk qunit reaches this value,
325 * later write reqs from client should be sync b=16642
328 /** guard lqc_imp_valid now */
329 cfs_spinlock_t lqc_lock;
331 * when mds isn't connected, threads
332 * on osts who send the quota reqs
333 * with wait==1 will be put here b=14840
335 cfs_waitq_t lqc_wait_for_qmaster;
336 struct proc_dir_entry *lqc_proc_dir;
337 /** lquota statistics */
338 struct lprocfs_stats *lqc_stats;
339 /** the number of used hashed lqs */
340 cfs_atomic_t lqc_lqs;
341 /** no lqs are in use */
342 cfs_waitq_t lqc_lqs_waitq;
345 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
346 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
348 struct lustre_qunit_size {
349 cfs_hlist_node_t lqs_hash; /** the hash entry */
350 unsigned int lqs_id; /** id of user/group */
351 unsigned long lqs_flags; /** 31st bit is QB_SET, 30th bit is QI_SET
352 * other bits are same as LQUOTA_FLAGS_*
354 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
356 * Trigger dqacq when available file quota
357 * less than this value, trigger dqrel
358 * when more than this value + 1 iunit
360 unsigned long lqs_itune_sz;
361 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
362 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
363 /** the blocks reached ost and don't finish */
364 unsigned long lqs_bwrite_pending;
365 /** the inodes reached mds and don't finish */
366 unsigned long lqs_iwrite_pending;
367 /** when inodes are allocated/released, this value will record it */
368 long long lqs_ino_rec;
369 /** when blocks are allocated/released, this value will record it */
370 long long lqs_blk_rec;
371 cfs_atomic_t lqs_refcount;
372 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
373 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
374 cfs_spinlock_t lqs_lock;
375 unsigned long long lqs_key; /** hash key */
376 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
379 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
380 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
381 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
383 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
384 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
385 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
387 /* In the hash for lustre_qunit_size, the key is decided by
388 * grp_or_usr and uid/gid, in here, I combine these two values,
389 * which will make comparing easier and more efficient */
390 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
391 #define LQS_KEY_ID(key) (key & 0xffffffff)
392 #define LQS_KEY_GRP(key) (key >> 32)
394 static inline void __lqs_getref(struct lustre_qunit_size *lqs)
396 int count = cfs_atomic_inc_return(&lqs->lqs_refcount);
398 if (count == 2) /* quota_create_lqs */
399 cfs_atomic_inc(&lqs->lqs_ctxt->lqc_lqs);
400 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count);
403 static inline void lqs_getref(struct lustre_qunit_size *lqs)
408 static inline void __lqs_putref(struct lustre_qunit_size *lqs)
410 LASSERT(cfs_atomic_read(&lqs->lqs_refcount) > 0);
412 if (cfs_atomic_dec_return(&lqs->lqs_refcount) == 1)
413 if (cfs_atomic_dec_and_test(&lqs->lqs_ctxt->lqc_lqs))
414 cfs_waitq_signal(&lqs->lqs_ctxt->lqc_lqs_waitq);
415 CDEBUG(D_INFO, "lqs=%p refcount %d\n",
416 lqs, cfs_atomic_read(&lqs->lqs_refcount));
419 static inline void lqs_putref(struct lustre_qunit_size *lqs)
424 static inline void lqs_initref(struct lustre_qunit_size *lqs)
426 cfs_atomic_set(&lqs->lqs_refcount, 0);
431 struct lustre_quota_info {
434 struct lustre_quota_ctxt {
437 #define QUOTA_MASTER_READY(qctxt)
438 #define QUOTA_MASTER_UNREADY(qctxt)
440 #endif /* !__KERNEL__ */
444 #define LL_DQUOT_OFF(sb, remount) do {} while(0)
446 struct lustre_quota_info {
449 struct lustre_quota_ctxt {
452 #endif /* !HAVE_QUOTA_SUPPORT */
454 /* If the (quota limit < qunit * slave count), the slave which can't
455 * acquire qunit should set it's local limit as MIN_QLIMIT */
458 struct quotacheck_thread_args {
459 struct obd_export *qta_exp; /** obd export */
460 struct obd_device *qta_obd; /** obd device */
461 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
462 struct super_block *qta_sb; /** obd super block */
463 cfs_semaphore_t *qta_sem; /** obt_quotachecking */
466 struct obd_trans_info;
467 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
468 struct obd_trans_info *oti, int isblk);
471 int (*quota_init) (void);
472 int (*quota_exit) (void);
473 int (*quota_setup) (struct obd_device *);
474 int (*quota_cleanup) (struct obd_device *);
476 * For quota master, close admin quota files
478 int (*quota_fs_cleanup) (struct obd_device *);
479 int (*quota_ctl) (struct obd_device *, struct obd_export *,
480 struct obd_quotactl *);
481 int (*quota_check) (struct obd_device *, struct obd_export *,
482 struct obd_quotactl *);
483 int (*quota_recovery) (struct obd_device *);
486 * For quota master/slave, adjust quota limit after fs operation
488 int (*quota_adjust) (struct obd_device *, const unsigned int[],
489 const unsigned int[], int, int);
492 * For quota slave, set import, trigger quota recovery,
493 * For quota master, set lqc_setup
495 int (*quota_setinfo) (struct obd_device *, void *);
498 * For quota slave, clear import when relative import is invalid
500 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
503 * For quota slave, set proper thread resoure capability
505 int (*quota_enforce) (struct obd_device *, unsigned int);
508 * For quota slave, check whether specified uid/gid is over quota
510 int (*quota_getflag) (struct obd_device *, struct obdo *);
514 * For quota slave, acquire/release quota from master if needed
516 int (*quota_acquire) (struct obd_device *, const unsigned int [],
517 struct obd_trans_info *, int);
520 * For quota slave, check whether specified uid/gid's remaining quota
521 * can finish a block_write or inode_create rpc. It updates the pending
522 * record of block and inode, acquires quota if necessary
524 int (*quota_chkquota) (struct obd_device *, struct obd_export *,
525 const unsigned int [], int [],
526 int, quota_acquire, struct obd_trans_info *,
527 int, struct inode *, int);
530 * For quota client, the actions after the pending write is committed
532 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
537 * For quota client, poll if the quota check done
539 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
542 * For quota client, check whether specified uid/gid is over quota
544 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
547 * For quota client, set over quota flag for specifed uid/gid
549 int (*quota_setdq) (struct client_obd *, const unsigned int [],
553 * For adjusting qunit size b=10600
555 int (*quota_adjust_qunit) (struct obd_export *exp,
556 struct quota_adjust_qunit *oqaq,
557 struct lustre_quota_ctxt *qctxt);
561 #define Q_COPY(out, in, member) (out)->member = (in)->member
563 #define QUOTA_OP(interface, op) interface->quota_ ## op
565 #define QUOTA_CHECK_OP(interface, op) \
569 if (!QUOTA_OP(interface, op)) { \
570 CERROR("no quota operation: " #op "\n"); \
571 RETURN(-EOPNOTSUPP); \
575 static inline int lquota_init(quota_interface_t *interface)
580 QUOTA_CHECK_OP(interface, init);
581 rc = QUOTA_OP(interface, init)();
585 static inline int lquota_exit(quota_interface_t *interface)
590 QUOTA_CHECK_OP(interface, exit);
591 rc = QUOTA_OP(interface, exit)();
595 static inline int lquota_setup(quota_interface_t *interface,
596 struct obd_device *obd)
601 QUOTA_CHECK_OP(interface, setup);
602 rc = QUOTA_OP(interface, setup)(obd);
606 static inline int lquota_cleanup(quota_interface_t *interface,
607 struct obd_device *obd)
612 QUOTA_CHECK_OP(interface, cleanup);
613 rc = QUOTA_OP(interface, cleanup)(obd);
617 static inline int lquota_fs_cleanup(quota_interface_t *interface,
618 struct obd_device *obd)
623 QUOTA_CHECK_OP(interface, fs_cleanup);
624 rc = QUOTA_OP(interface, fs_cleanup)(obd);
628 static inline int lquota_recovery(quota_interface_t *interface,
629 struct obd_device *obd)
634 QUOTA_CHECK_OP(interface, recovery);
635 rc = QUOTA_OP(interface, recovery)(obd);
639 static inline int lquota_check(quota_interface_t *interface,
640 struct obd_device *obd,
641 struct obd_export *exp,
642 struct obd_quotactl *oqctl)
647 QUOTA_CHECK_OP(interface, check);
648 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
652 static inline int lquota_ctl(quota_interface_t *interface,
653 struct obd_device *obd,
654 struct obd_quotactl *oqctl)
659 QUOTA_CHECK_OP(interface, ctl);
660 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
664 static inline int lquota_adjust(quota_interface_t *interface,
665 struct obd_device *obd,
666 const unsigned int qcids[],
667 const unsigned int qpids[],
673 QUOTA_CHECK_OP(interface, adjust);
674 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
678 static inline int lquota_chkdq(quota_interface_t *interface,
679 struct client_obd *cli, const unsigned int qid[])
684 QUOTA_CHECK_OP(interface, chkdq);
685 rc = QUOTA_OP(interface, chkdq)(cli, qid);
689 static inline int lquota_setdq(quota_interface_t *interface,
690 struct client_obd *cli, const unsigned int qid[],
691 obd_flag valid, obd_flag flags)
696 QUOTA_CHECK_OP(interface, setdq);
697 rc = QUOTA_OP(interface, setdq)(cli, qid, valid, flags);
701 static inline int lquota_poll_check(quota_interface_t *interface,
702 struct obd_export *exp,
703 struct if_quotacheck *qchk)
708 QUOTA_CHECK_OP(interface, poll_check);
709 rc = QUOTA_OP(interface, poll_check)(exp, qchk);
713 static inline int lquota_setinfo(quota_interface_t *interface,
714 struct obd_device *obd,
720 QUOTA_CHECK_OP(interface, setinfo);
721 rc = QUOTA_OP(interface, setinfo)(obd, data);
725 static inline int lquota_clearinfo(quota_interface_t *interface,
726 struct obd_export *exp,
727 struct obd_device *obd)
732 QUOTA_CHECK_OP(interface, clearinfo);
733 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
737 static inline int lquota_enforce(quota_interface_t *interface,
738 struct obd_device *obd,
744 QUOTA_CHECK_OP(interface, enforce);
745 rc = QUOTA_OP(interface, enforce)(obd, ignore);
749 static inline int lquota_getflag(quota_interface_t *interface,
750 struct obd_device *obd, struct obdo *oa)
755 QUOTA_CHECK_OP(interface, getflag);
756 rc = QUOTA_OP(interface, getflag)(obd, oa);
761 static inline int lquota_chkquota(quota_interface_t *interface,
762 struct obd_device *obd,
763 struct obd_export *exp,
764 const unsigned int id[], int pending[],
765 int count, struct obd_trans_info *oti,
766 int isblk, void *data, int frags)
771 QUOTA_CHECK_OP(interface, chkquota);
772 QUOTA_CHECK_OP(interface, acquire);
773 rc = QUOTA_OP(interface, chkquota)(obd, exp, id, pending, count,
774 QUOTA_OP(interface, acquire), oti,
775 isblk, (struct inode *)data, frags);
779 static inline int lquota_pending_commit(quota_interface_t *interface,
780 struct obd_device *obd,
781 const unsigned int id[],
782 int pending[], int isblk)
787 QUOTA_CHECK_OP(interface, pending_commit);
788 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
794 extern quota_interface_t osc_quota_interface;
795 extern quota_interface_t lov_quota_interface;
796 extern quota_interface_t mdc_quota_interface;
797 extern quota_interface_t lmv_quota_interface;
813 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
814 "admin_quotafile_v2.usr", /** user admin quotafile */\
815 "admin_quotafile_v2.grp" /** group admin quotafile */\
818 #endif /* _LUSTRE_QUOTA_H */