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;
184 /* set inode quota limitation on a quota uid/gid */
185 #define QI_SET (1 << 30)
186 /* set block quota limitation on a quota uid/gid */
187 #define QB_SET (1 << 31)
190 #define QFILE_RD_INFO 2
191 #define QFILE_WR_INFO 3
192 #define QFILE_INIT_INFO 4
193 #define QFILE_RD_DQUOT 5
194 #define QFILE_WR_DQUOT 6
195 #define QFILE_CONVERT 7
197 /* admin quotafile operations */
198 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
199 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
200 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
201 int lustre_read_dquot(struct lustre_dquot *dquot);
202 int lustre_commit_dquot(struct lustre_dquot *dquot);
203 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
204 int lustre_get_qids(struct file *file, struct inode *inode, int type,
205 struct list_head *list);
206 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
208 #define LL_DQUOT_OFF(sb) DQUOT_OFF(sb)
210 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
213 /* user quota is turned on on filter */
214 #define LQC_USRQUOTA_FLAG (1 << 0)
215 /* group quota is turned on on filter */
216 #define LQC_GRPQUOTA_FLAG (1 << 1)
218 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
219 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
221 struct lustre_quota_ctxt {
222 /** superblock this applies to */
223 struct super_block *lqc_sb;
224 /** obd_device_target for obt_rwsem */
225 struct obd_device_target *lqc_obt;
226 /** import used to send dqacq/dqrel RPC */
227 struct obd_import *lqc_import;
228 /** dqacq/dqrel RPC handler, only for quota master */
229 dqacq_handler_t lqc_handler;
231 unsigned long lqc_flags;
233 unsigned long lqc_recovery:1, /** Doing recovery */
235 * the function of change qunit size
238 lqc_valid:1, /** this qctxt is valid or not */
240 * tell whether of not quota_type has
241 * been processed, so that the master
242 * knows when it can start processing
243 * incoming acq/rel quota requests
246 * cannot be turned on/off on-fly;
247 * temporary used by SOM.
251 * original unit size of file quota and
252 * upper limitation for adjust file qunit
254 unsigned long lqc_iunit_sz;
256 * Trigger dqacq when available file
257 * quota less than this value, trigger
258 * dqrel when available file quota
259 * more than this value + 1 iunit
261 unsigned long lqc_itune_sz;
263 * original unit size of block quota and
264 * upper limitation for adjust block qunit
266 unsigned long lqc_bunit_sz;
267 /** See comment of lqc_itune_sz */
268 unsigned long lqc_btune_sz;
269 /** all lustre_qunit_size structures */
270 struct lustre_hash *lqc_lqs_hash;
274 * the values below are relative to how master change its qunit sizes
277 * this affects the boundary of
278 * shrinking and enlarging qunit size. default=4
280 unsigned long lqc_cqs_boundary_factor;
281 /** the least value of block qunit */
282 unsigned long lqc_cqs_least_bunit;
283 /** the least value of inode qunit */
284 unsigned long lqc_cqs_least_iunit;
286 * when enlarging, qunit size will
287 * mutilple it; when shrinking,
288 * qunit size will divide it
290 unsigned long lqc_cqs_qs_factor;
292 * avoid ping-pong effect of
293 * adjusting qunit size. How many
294 * seconds must be waited between
295 * enlarging and shinking qunit
298 int lqc_switch_seconds;
300 * when blk qunit reaches this value,
301 * later write reqs from client should be sync b=16642
304 /** guard lqc_imp_valid now */
307 * when mds isn't connected, threads
308 * on osts who send the quota reqs
309 * with wait==1 will be put here b=14840
311 cfs_waitq_t lqc_wait_for_qmaster;
312 struct proc_dir_entry *lqc_proc_dir;
313 /** lquota statistics */
314 struct lprocfs_stats *lqc_stats;
315 /** the number of used hashed lqs */
317 /** no lqs are in use */
318 cfs_waitq_t lqc_lqs_waitq;
321 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
322 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
324 struct lustre_qunit_size {
325 struct hlist_node lqs_hash; /** the hash entry */
326 unsigned int lqs_id; /** id of user/group */
327 unsigned long lqs_flags; /** 31st bit is QB_SET, 30th bit is QI_SET
328 * other bits are same as LQUOTA_FLAGS_*
330 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
332 * Trigger dqacq when available file quota
333 * less than this value, trigger dqrel
334 * when more than this value + 1 iunit
336 unsigned long lqs_itune_sz;
337 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
338 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
339 /** the blocks reached ost and don't finish */
340 unsigned long lqs_bwrite_pending;
341 /** the inodes reached mds and don't finish */
342 unsigned long lqs_iwrite_pending;
343 /** when inodes are allocated/released, this value will record it */
344 long long lqs_ino_rec;
345 /** when blocks are allocated/released, this value will record it */
346 long long lqs_blk_rec;
347 atomic_t lqs_refcount;
348 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
349 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
351 unsigned long long lqs_key; /** hash key */
352 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
355 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
356 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
357 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
359 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
360 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
361 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
363 /* In the hash for lustre_qunit_size, the key is decided by
364 * grp_or_usr and uid/gid, in here, I combine these two values,
365 * which will make comparing easier and more efficient */
366 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
367 #define LQS_KEY_ID(key) (key & 0xffffffff)
368 #define LQS_KEY_GRP(key) (key >> 32)
370 static inline void __lqs_getref(struct lustre_qunit_size *lqs)
372 int count = atomic_inc_return(&lqs->lqs_refcount);
374 if (count == 2) /* quota_create_lqs */
375 atomic_inc(&lqs->lqs_ctxt->lqc_lqs);
376 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n", lqs, count);
379 static inline void lqs_getref(struct lustre_qunit_size *lqs)
384 static inline void __lqs_putref(struct lustre_qunit_size *lqs, int del)
386 int count = atomic_read(&lqs->lqs_refcount);
390 CDEBUG(D_QUOTA, "lqs=%p refcount to be 0\n", lqs);
392 /* killing last ref, let's let hash table kill it */
393 lustre_hash_del(lqs->lqs_ctxt->lqc_lqs_hash,
394 &lqs->lqs_key, &lqs->lqs_hash);
397 atomic_dec(&lqs->lqs_refcount);
400 count = atomic_dec_return(&lqs->lqs_refcount);
402 if (atomic_dec_and_test(&lqs->lqs_ctxt->lqc_lqs))
403 cfs_waitq_signal(&lqs->lqs_ctxt->lqc_lqs_waitq);
404 CDEBUG(D_QUOTA, "lqs=%p refcount %d\n", lqs, count);
408 static inline void lqs_putref(struct lustre_qunit_size *lqs)
410 __lqs_putref(lqs, 1);
413 static inline void lqs_initref(struct lustre_qunit_size *lqs)
415 atomic_set(&lqs->lqs_refcount, 0);
420 struct lustre_quota_info {
423 struct lustre_quota_ctxt {
426 #define QUOTA_MASTER_READY(qctxt)
427 #define QUOTA_MASTER_UNREADY(qctxt)
429 #endif /* !__KERNEL__ */
433 #define LL_DQUOT_OFF(sb) do {} while(0)
435 struct lustre_quota_info {
438 struct lustre_quota_ctxt {
441 #endif /* !HAVE_QUOTA_SUPPORT */
443 /* If the (quota limit < qunit * slave count), the slave which can't
444 * acquire qunit should set it's local limit as MIN_QLIMIT */
447 struct quotacheck_thread_args {
448 struct obd_export *qta_exp; /** obd export */
449 struct obd_device *qta_obd; /** obd device */
450 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
451 struct super_block *qta_sb; /** obd super block */
452 struct semaphore *qta_sem; /** obt_quotachecking */
455 struct obd_trans_info;
456 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
457 struct obd_trans_info *oti, int isblk);
460 int (*quota_init) (void);
461 int (*quota_exit) (void);
462 int (*quota_setup) (struct obd_device *);
463 int (*quota_cleanup) (struct obd_device *);
465 * For quota master, close admin quota files
467 int (*quota_fs_cleanup) (struct obd_device *);
468 int (*quota_ctl) (struct obd_device *, struct obd_export *,
469 struct obd_quotactl *);
470 int (*quota_check) (struct obd_device *, struct obd_export *,
471 struct obd_quotactl *);
472 int (*quota_recovery) (struct obd_device *);
475 * For quota master/slave, adjust quota limit after fs operation
477 int (*quota_adjust) (struct obd_device *, const unsigned int[],
478 const unsigned int[], int, int);
481 * For quota slave, set import, trigger quota recovery,
482 * For quota master, set lqc_setup
484 int (*quota_setinfo) (struct obd_device *, void *);
487 * For quota slave, clear import when relative import is invalid
489 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
492 * For quota slave, set proper thread resoure capability
494 int (*quota_enforce) (struct obd_device *, unsigned int);
497 * For quota slave, check whether specified uid/gid is over quota
499 int (*quota_getflag) (struct obd_device *, struct obdo *);
503 * For quota slave, acquire/release quota from master if needed
505 int (*quota_acquire) (struct obd_device *, const unsigned int [],
506 struct obd_trans_info *, int);
509 * For quota slave, check whether specified uid/gid's remaining quota
510 * can finish a block_write or inode_create rpc. It updates the pending
511 * record of block and inode, acquires quota if necessary
513 int (*quota_chkquota) (struct obd_device *, const unsigned int [],
514 int [], int, quota_acquire,
515 struct obd_trans_info *, int, struct inode *,
519 * For quota client, the actions after the pending write is committed
521 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
526 * For quota client, poll if the quota check done
528 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
531 * For quota client, check whether specified uid/gid is over quota
533 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
536 * For quota client, set over quota flag for specifed uid/gid
538 int (*quota_setdq) (struct client_obd *, const unsigned int [],
542 * For adjusting qunit size b=10600
544 int (*quota_adjust_qunit) (struct obd_export *exp,
545 struct quota_adjust_qunit *oqaq,
546 struct lustre_quota_ctxt *qctxt);
550 #define Q_COPY(out, in, member) (out)->member = (in)->member
552 #define QUOTA_OP(interface, op) interface->quota_ ## op
554 #define QUOTA_CHECK_OP(interface, op) \
558 if (!QUOTA_OP(interface, op)) { \
559 CERROR("no quota operation: " #op "\n"); \
560 RETURN(-EOPNOTSUPP); \
564 static inline int lquota_init(quota_interface_t *interface)
569 QUOTA_CHECK_OP(interface, init);
570 rc = QUOTA_OP(interface, init)();
574 static inline int lquota_exit(quota_interface_t *interface)
579 QUOTA_CHECK_OP(interface, exit);
580 rc = QUOTA_OP(interface, exit)();
584 static inline int lquota_setup(quota_interface_t *interface,
585 struct obd_device *obd)
590 QUOTA_CHECK_OP(interface, setup);
591 rc = QUOTA_OP(interface, setup)(obd);
595 static inline int lquota_cleanup(quota_interface_t *interface,
596 struct obd_device *obd)
601 QUOTA_CHECK_OP(interface, cleanup);
602 rc = QUOTA_OP(interface, cleanup)(obd);
606 static inline int lquota_fs_cleanup(quota_interface_t *interface,
607 struct obd_device *obd)
612 QUOTA_CHECK_OP(interface, fs_cleanup);
613 rc = QUOTA_OP(interface, fs_cleanup)(obd);
617 static inline int lquota_recovery(quota_interface_t *interface,
618 struct obd_device *obd)
623 QUOTA_CHECK_OP(interface, recovery);
624 rc = QUOTA_OP(interface, recovery)(obd);
628 static inline int lquota_check(quota_interface_t *interface,
629 struct obd_device *obd,
630 struct obd_export *exp,
631 struct obd_quotactl *oqctl)
636 QUOTA_CHECK_OP(interface, check);
637 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
641 static inline int lquota_ctl(quota_interface_t *interface,
642 struct obd_device *obd,
643 struct obd_quotactl *oqctl)
648 QUOTA_CHECK_OP(interface, ctl);
649 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
653 static inline int lquota_adjust(quota_interface_t *interface,
654 struct obd_device *obd,
655 const unsigned int qcids[],
656 const unsigned int qpids[],
662 QUOTA_CHECK_OP(interface, adjust);
663 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
667 static inline int lquota_chkdq(quota_interface_t *interface,
668 struct client_obd *cli, const unsigned int qid[])
673 QUOTA_CHECK_OP(interface, chkdq);
674 rc = QUOTA_OP(interface, chkdq)(cli, qid);
678 static inline int lquota_setdq(quota_interface_t *interface,
679 struct client_obd *cli, const unsigned int qid[],
680 obd_flag valid, obd_flag flags)
685 QUOTA_CHECK_OP(interface, setdq);
686 rc = QUOTA_OP(interface, setdq)(cli, qid, valid, flags);
690 static inline int lquota_poll_check(quota_interface_t *interface,
691 struct obd_export *exp,
692 struct if_quotacheck *qchk)
697 QUOTA_CHECK_OP(interface, poll_check);
698 rc = QUOTA_OP(interface, poll_check)(exp, qchk);
702 static inline int lquota_setinfo(quota_interface_t *interface,
703 struct obd_device *obd,
709 QUOTA_CHECK_OP(interface, setinfo);
710 rc = QUOTA_OP(interface, setinfo)(obd, data);
714 static inline int lquota_clearinfo(quota_interface_t *interface,
715 struct obd_export *exp,
716 struct obd_device *obd)
721 QUOTA_CHECK_OP(interface, clearinfo);
722 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
726 static inline int lquota_enforce(quota_interface_t *interface,
727 struct obd_device *obd,
733 QUOTA_CHECK_OP(interface, enforce);
734 rc = QUOTA_OP(interface, enforce)(obd, ignore);
738 static inline int lquota_getflag(quota_interface_t *interface,
739 struct obd_device *obd, struct obdo *oa)
744 QUOTA_CHECK_OP(interface, getflag);
745 rc = QUOTA_OP(interface, getflag)(obd, oa);
750 static inline int lquota_chkquota(quota_interface_t *interface,
751 struct obd_device *obd,
752 const unsigned int id[], int pending[],
753 int count, struct obd_trans_info *oti,
754 int isblk, void *data, int frags)
759 QUOTA_CHECK_OP(interface, chkquota);
760 QUOTA_CHECK_OP(interface, acquire);
761 rc = QUOTA_OP(interface, chkquota)(obd, id, pending, count,
762 QUOTA_OP(interface, acquire), oti,
763 isblk, (struct inode *)data, frags);
767 static inline int lquota_pending_commit(quota_interface_t *interface,
768 struct obd_device *obd,
769 const unsigned int id[],
770 int pending[], int isblk)
775 QUOTA_CHECK_OP(interface, pending_commit);
776 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
782 extern quota_interface_t osc_quota_interface;
783 extern quota_interface_t lov_quota_interface;
784 extern quota_interface_t mdc_quota_interface;
785 extern quota_interface_t lmv_quota_interface;
801 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
802 "admin_quotafile_v2.usr", /** user admin quotafile */\
803 "admin_quotafile_v2.grp" /** group admin quotafile */\
806 #endif /* _LUSTRE_QUOTA_H */