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, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, Whamcloud, Inc.
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 /** \defgroup quota quota
45 #if defined(__linux__)
46 #include <linux/lustre_quota.h>
47 #elif defined(__APPLE__)
48 #include <darwin/lustre_quota.h>
49 #elif defined(__WINNT__)
50 #include <winnt/lustre_quota.h>
52 #error Unsupported operating system.
55 #include <lustre_net.h>
56 #include <lustre/lustre_idl.h>
58 #include <obd_support.h>
67 #ifndef QUOTABLOCK_BITS
68 #define QUOTABLOCK_BITS 10
71 #ifndef QUOTABLOCK_SIZE
72 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
76 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
79 #ifdef HAVE_QUOTA_SUPPORT
82 #define MAX_IQ_TIME 604800 /* (7*24*60*60) 1 week */
86 #define MAX_DQ_TIME 604800 /* (7*24*60*60) 1 week */
93 LQUOTA_FIRST_STAT = 0,
96 * these four are for measuring quota requests, for both of
97 * quota master and quota slaves
99 LQUOTA_SYNC_ACQ = LQUOTA_FIRST_STAT,
106 * these four measure how much time I/O threads spend on dealing
107 * with quota before and after writing data or creating files,
108 * only for quota slaves(lquota_chkquota and lquota_pending_commit)
110 LQUOTA_WAIT_FOR_CHK_BLK,
111 LQUOTA_WAIT_FOR_CHK_INO,
112 LQUOTA_WAIT_FOR_COMMIT_BLK,
113 LQUOTA_WAIT_FOR_COMMIT_INO,
117 * these two are for measuring time waiting return of quota reqs
118 * (qctxt_wait_pending_dqacq), only for quota salves
120 LQUOTA_WAIT_PENDING_BLK_QUOTA,
121 LQUOTA_WAIT_PENDING_INO_QUOTA,
125 * these two are for those when they are calling
126 * qctxt_wait_pending_dqacq, the quota req has returned already,
127 * only for quota salves
129 LQUOTA_NOWAIT_PENDING_BLK_QUOTA,
130 LQUOTA_NOWAIT_PENDING_INO_QUOTA,
134 * these are for quota ctl
140 * these are for adjust quota qunit, for both of
141 * quota master and quota slaves
149 /* structures to access admin quotafile */
150 struct lustre_mem_dqinfo {
151 unsigned int dqi_bgrace;
152 unsigned int dqi_igrace;
153 unsigned long dqi_flags;
154 unsigned int dqi_blocks;
155 unsigned int dqi_free_blk;
156 unsigned int dqi_free_entry;
159 struct lustre_quota_info {
160 struct file *qi_files[MAXQUOTAS];
161 struct lustre_mem_dqinfo qi_info[MAXQUOTAS];
162 lustre_quota_version_t qi_version;
165 struct lustre_mem_dqblk {
166 __u64 dqb_bhardlimit; /**< absolute limit on disk blks alloc */
167 __u64 dqb_bsoftlimit; /**< preferred limit on disk blks */
168 __u64 dqb_curspace; /**< current used space */
169 __u64 dqb_ihardlimit; /**< absolute limit on allocated inodes */
170 __u64 dqb_isoftlimit; /**< preferred inode limit */
171 __u64 dqb_curinodes; /**< current # allocated inodes */
172 time_t dqb_btime; /**< time limit for excessive disk use */
173 time_t dqb_itime; /**< time limit for excessive inode use */
176 struct lustre_dquot {
177 /** Hash list in memory, protect by dquot_hash_lock */
179 /** Protect the data in lustre_dquot */
180 cfs_mutex_t dq_mutex;
182 cfs_atomic_t dq_refcnt;
183 /** Pointer of quota info it belongs to */
184 struct lustre_quota_info *dq_info;
185 /** Offset of dquot on disk */
187 /** ID this applies to (uid, gid) */
189 /** Type fo quota (USRQUOTA, GRPQUOUTA) */
191 /** See DQ_ in quota.h */
192 unsigned long dq_flags;
193 /** Diskquota usage */
194 struct lustre_mem_dqblk dq_dqb;
202 /* set inode quota limitation on a quota uid/gid */
203 #define QI_SET (1 << 30)
204 /* set block quota limitation on a quota uid/gid */
205 #define QB_SET (1 << 31)
208 #define QFILE_RD_INFO 2
209 #define QFILE_WR_INFO 3
210 #define QFILE_INIT_INFO 4
211 #define QFILE_RD_DQUOT 5
212 #define QFILE_WR_DQUOT 6
213 #define QFILE_CONVERT 7
215 /* admin quotafile operations */
216 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
217 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
218 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
219 int lustre_read_dquot(struct lustre_dquot *dquot);
220 int lustre_commit_dquot(struct lustre_dquot *dquot);
221 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
222 int lustre_get_qids(struct file *file, struct inode *inode, int type,
224 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
226 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
229 /* user quota is turned on on filter */
230 #define LQC_USRQUOTA_FLAG (1 << 0)
231 /* group quota is turned on on filter */
232 #define LQC_GRPQUOTA_FLAG (1 << 1)
234 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
235 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
237 struct lustre_quota_ctxt {
238 /** superblock this applies to */
239 struct super_block *lqc_sb;
240 /** obd_device_target for obt_rwsem */
241 struct obd_device_target *lqc_obt;
242 /** import used to send dqacq/dqrel RPC */
243 struct obd_import *lqc_import;
244 /** dqacq/dqrel RPC handler, only for quota master */
245 dqacq_handler_t lqc_handler;
247 unsigned long lqc_flags;
249 unsigned long lqc_recovery:1, /** Doing recovery */
251 * the function of change qunit size
254 lqc_valid:1, /** this qctxt is valid or not */
256 * tell whether of not quota_type has
257 * been processed, so that the master
258 * knows when it can start processing
259 * incoming acq/rel quota requests
263 * original unit size of file quota and
264 * upper limitation for adjust file qunit
266 unsigned long lqc_iunit_sz;
268 * Trigger dqacq when available file
269 * quota less than this value, trigger
270 * dqrel when available file quota
271 * more than this value + 1 iunit
273 unsigned long lqc_itune_sz;
275 * original unit size of block quota and
276 * upper limitation for adjust block qunit
278 unsigned long lqc_bunit_sz;
279 /** See comment of lqc_itune_sz */
280 unsigned long lqc_btune_sz;
281 /** all lustre_qunit_size structures */
282 cfs_hash_t *lqc_lqs_hash;
286 * the values below are relative to how master change its qunit sizes
289 * this affects the boundary of
290 * shrinking and enlarging qunit size. default=4
292 unsigned long lqc_cqs_boundary_factor;
293 /** the least value of block qunit */
294 unsigned long lqc_cqs_least_bunit;
295 /** the least value of inode qunit */
296 unsigned long lqc_cqs_least_iunit;
298 * when enlarging, qunit size will
299 * mutilple it; when shrinking,
300 * qunit size will divide it
302 unsigned long lqc_cqs_qs_factor;
304 * avoid ping-pong effect of
305 * adjusting qunit size. How many
306 * seconds must be waited between
307 * enlarging and shinking qunit
310 int lqc_switch_seconds;
312 * when blk qunit reaches this value,
313 * later write reqs from client should be sync b=16642
316 /** guard lqc_imp_valid now */
317 cfs_spinlock_t lqc_lock;
319 * when mds isn't connected, threads
320 * on osts who send the quota reqs
321 * with wait==1 will be put here b=14840
323 cfs_waitq_t lqc_wait_for_qmaster;
324 struct proc_dir_entry *lqc_proc_dir;
325 /** lquota statistics */
326 struct lprocfs_stats *lqc_stats;
327 /** the number of used hashed lqs */
328 cfs_atomic_t lqc_lqs;
329 /** no lqs are in use */
330 cfs_waitq_t lqc_lqs_waitq;
333 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
334 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
336 struct lustre_qunit_size {
337 cfs_hlist_node_t lqs_hash; /** the hash entry */
338 unsigned int lqs_id; /** id of user/group */
339 unsigned long lqs_flags; /** 31st bit is QB_SET, 30th bit is QI_SET
340 * other bits are same as LQUOTA_FLAGS_*
342 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
344 * Trigger dqacq when available file quota
345 * less than this value, trigger dqrel
346 * when more than this value + 1 iunit
348 unsigned long lqs_itune_sz;
349 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
350 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
351 /** the blocks reached ost and don't finish */
352 unsigned long lqs_bwrite_pending;
353 /** the inodes reached mds and don't finish */
354 unsigned long lqs_iwrite_pending;
355 /** when inodes are allocated/released, this value will record it */
356 long long lqs_ino_rec;
357 /** when blocks are allocated/released, this value will record it */
358 long long lqs_blk_rec;
359 cfs_atomic_t lqs_refcount;
360 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
361 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
362 cfs_spinlock_t lqs_lock;
363 unsigned long long lqs_key; /** hash key */
364 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
367 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
368 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
369 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
370 #define LQS_IS_RECOVERY(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_RECOVERY)
371 #define LQS_IS_SETQUOTA(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_SETQUOTA)
373 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
374 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
375 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
376 #define LQS_SET_RECOVERY(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_RECOVERY)
377 #define LQS_SET_SETQUOTA(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_SETQUOTA)
379 #define LQS_CLEAR_RECOVERY(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_RECOVERY)
380 #define LQS_CLEAR_SETQUOTA(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_SETQUOTA)
382 /* In the hash for lustre_qunit_size, the key is decided by
383 * grp_or_usr and uid/gid, in here, I combine these two values,
384 * which will make comparing easier and more efficient */
385 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
386 #define LQS_KEY_ID(key) (key & 0xffffffff)
387 #define LQS_KEY_GRP(key) (key >> 32)
389 static inline void lqs_getref(struct lustre_qunit_size *lqs)
391 int count = cfs_atomic_inc_return(&lqs->lqs_refcount);
393 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count);
396 static inline void lqs_putref(struct lustre_qunit_size *lqs)
398 int count = cfs_atomic_read(&lqs->lqs_refcount);
401 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count - 1);
403 if (cfs_atomic_dec_and_test(&lqs->lqs_refcount)) {
404 if (cfs_atomic_dec_and_test(&lqs->lqs_ctxt->lqc_lqs))
405 cfs_waitq_signal(&lqs->lqs_ctxt->lqc_lqs_waitq);
412 struct lustre_quota_info {
415 struct lustre_quota_ctxt {
418 #define QUOTA_MASTER_READY(qctxt)
419 #define QUOTA_MASTER_UNREADY(qctxt)
421 #endif /* !__KERNEL__ */
425 #define LL_DQUOT_OFF(sb) do {} while(0)
427 struct lustre_quota_info {
430 struct lustre_quota_ctxt {
433 #endif /* !HAVE_QUOTA_SUPPORT */
435 /* If the (quota limit < qunit * slave count), the slave which can't
436 * acquire qunit should set it's local limit as MIN_QLIMIT */
439 struct quotacheck_thread_args {
440 struct obd_export *qta_exp; /** obd export */
441 struct obd_device *qta_obd; /** obd device */
442 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
443 struct super_block *qta_sb; /** obd super block */
444 cfs_semaphore_t *qta_sem; /** obt_quotachecking */
447 struct obd_trans_info;
448 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
449 struct obd_trans_info *oti, int isblk);
452 int (*quota_init) (void);
453 int (*quota_exit) (void);
454 int (*quota_setup) (struct obd_device *);
455 int (*quota_cleanup) (struct obd_device *);
457 * For quota master, close admin quota files
459 int (*quota_fs_cleanup) (struct obd_device *);
460 int (*quota_ctl) (struct obd_device *, struct obd_export *,
461 struct obd_quotactl *);
462 int (*quota_check) (struct obd_device *, struct obd_export *,
463 struct obd_quotactl *);
464 int (*quota_recovery) (struct obd_device *);
467 * For quota master/slave, adjust quota limit after fs operation
469 int (*quota_adjust) (struct obd_device *, const unsigned int[],
470 const unsigned int[], int, int);
473 * For quota slave, set import, trigger quota recovery,
474 * For quota master, set lqc_setup
476 int (*quota_setinfo) (struct obd_device *, void *);
479 * For quota slave, clear import when relative import is invalid
481 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
484 * For quota slave, set proper thread resoure capability
486 int (*quota_enforce) (struct obd_device *, unsigned int);
489 * For quota slave, check whether specified uid/gid is over quota
491 int (*quota_getflag) (struct obd_device *, struct obdo *);
495 * For quota slave, acquire/release quota from master if needed
497 int (*quota_acquire) (struct obd_device *, const unsigned int [],
498 struct obd_trans_info *, int);
501 * For quota slave, check whether specified uid/gid's remaining quota
502 * can finish a block_write or inode_create rpc. It updates the pending
503 * record of block and inode, acquires quota if necessary
505 int (*quota_chkquota) (struct obd_device *, struct obd_export *,
506 const unsigned int [], int [],
507 int, quota_acquire, struct obd_trans_info *,
508 int, struct inode *, int);
511 * For quota client, the actions after the pending write is committed
513 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
518 * For quota client, poll if the quota check done
520 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
523 * For quota client, check whether specified uid/gid is over quota
525 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
528 * For quota client, set over quota flag for specifed uid/gid
530 int (*quota_setdq) (struct client_obd *, const unsigned int [],
534 * For adjusting qunit size b=10600
536 int (*quota_adjust_qunit) (struct obd_export *exp,
537 struct quota_adjust_qunit *oqaq,
538 struct lustre_quota_ctxt *qctxt,
539 struct ptlrpc_request_set *rqset);
543 #define Q_COPY(out, in, member) (out)->member = (in)->member
545 #define QUOTA_OP(interface, op) interface->quota_ ## op
547 #define QUOTA_CHECK_OP(interface, op) \
551 if (!QUOTA_OP(interface, op)) { \
552 CERROR("no quota operation: " #op "\n"); \
553 RETURN(-EOPNOTSUPP); \
557 static inline int lquota_init(quota_interface_t *interface)
562 QUOTA_CHECK_OP(interface, init);
563 rc = QUOTA_OP(interface, init)();
567 static inline int lquota_exit(quota_interface_t *interface)
572 QUOTA_CHECK_OP(interface, exit);
573 rc = QUOTA_OP(interface, exit)();
577 static inline int lquota_setup(quota_interface_t *interface,
578 struct obd_device *obd)
583 QUOTA_CHECK_OP(interface, setup);
584 rc = QUOTA_OP(interface, setup)(obd);
588 static inline int lquota_cleanup(quota_interface_t *interface,
589 struct obd_device *obd)
594 QUOTA_CHECK_OP(interface, cleanup);
595 rc = QUOTA_OP(interface, cleanup)(obd);
599 static inline int lquota_fs_cleanup(quota_interface_t *interface,
600 struct obd_device *obd)
605 QUOTA_CHECK_OP(interface, fs_cleanup);
606 rc = QUOTA_OP(interface, fs_cleanup)(obd);
610 static inline int lquota_recovery(quota_interface_t *interface,
611 struct obd_device *obd)
616 QUOTA_CHECK_OP(interface, recovery);
617 rc = QUOTA_OP(interface, recovery)(obd);
621 static inline int lquota_check(quota_interface_t *interface,
622 struct obd_device *obd,
623 struct obd_export *exp,
624 struct obd_quotactl *oqctl)
629 QUOTA_CHECK_OP(interface, check);
630 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
634 static inline int lquota_ctl(quota_interface_t *interface,
635 struct obd_device *obd,
636 struct obd_quotactl *oqctl)
641 QUOTA_CHECK_OP(interface, ctl);
642 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
646 static inline int lquota_adjust(quota_interface_t *interface,
647 struct obd_device *obd,
648 const unsigned int qcids[],
649 const unsigned int qpids[],
655 QUOTA_CHECK_OP(interface, adjust);
656 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
660 static inline int lquota_setinfo(quota_interface_t *interface,
661 struct obd_device *obd,
667 QUOTA_CHECK_OP(interface, setinfo);
668 rc = QUOTA_OP(interface, setinfo)(obd, data);
672 static inline int lquota_clearinfo(quota_interface_t *interface,
673 struct obd_export *exp,
674 struct obd_device *obd)
679 QUOTA_CHECK_OP(interface, clearinfo);
680 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
684 static inline int lquota_enforce(quota_interface_t *interface,
685 struct obd_device *obd,
691 QUOTA_CHECK_OP(interface, enforce);
692 rc = QUOTA_OP(interface, enforce)(obd, ignore);
696 static inline int lquota_getflag(quota_interface_t *interface,
697 struct obd_device *obd, struct obdo *oa)
702 QUOTA_CHECK_OP(interface, getflag);
703 rc = QUOTA_OP(interface, getflag)(obd, oa);
708 static inline int lquota_chkquota(quota_interface_t *interface,
709 struct obd_device *obd,
710 struct obd_export *exp,
711 const unsigned int id[], int pending[],
712 int count, struct obd_trans_info *oti,
713 int isblk, void *data, int frags)
718 QUOTA_CHECK_OP(interface, chkquota);
719 QUOTA_CHECK_OP(interface, acquire);
720 rc = QUOTA_OP(interface, chkquota)(obd, exp, id, pending, count,
721 QUOTA_OP(interface, acquire), oti,
722 isblk, (struct inode *)data, frags);
726 static inline int lquota_pending_commit(quota_interface_t *interface,
727 struct obd_device *obd,
728 const unsigned int id[],
729 int pending[], int isblk)
734 QUOTA_CHECK_OP(interface, pending_commit);
735 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
755 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
756 "admin_quotafile_v2.usr", /** user admin quotafile */\
757 "admin_quotafile_v2.grp" /** group admin quotafile */\
761 * Definitions of structures for vfsv0 quota format
762 * Source linux/fs/quota/quotaio_v2.h
764 * The following definitions are normally found in private kernel headers.
765 * However, some sites build Lustre against kernel development headers rather
766 * than than full kernel source, so we provide them here for compatibility.
769 # if !defined(HAVE_QUOTAIO_H) && !defined(HAVE_FS_QUOTA_QUOTAIO_H) && \
770 !defined(HAVE_FS_QUOTAIO_H)
772 #include <linux/types.h>
773 #include <linux/quota.h>
775 #define V2_INITQMAGICS {\
776 0xd9c01f11, /* USRQUOTA */\
777 0xd9c01927 /* GRPQUOTA */\
780 /* Header with type and version specific information */
781 struct v2_disk_dqinfo {
782 __le32 dqi_bgrace; /* Time before block soft limit becomes hard limit */
783 __le32 dqi_igrace; /* Time before inode soft limit becomes hard limit */
784 __le32 dqi_flags; /* Flags for quotafile (DQF_*) */
785 __le32 dqi_blocks; /* Number of blocks in file */
786 __le32 dqi_free_blk; /* Number of first free block in the list */
787 __le32 dqi_free_entry; /* Number of block with at least one free entry */
790 /* First generic header */
791 struct v2_disk_dqheader {
792 __le32 dqh_magic; /* Magic number identifying file */
793 __le32 dqh_version; /* File version */
795 #define V2_DQINFOOFF sizeof(struct v2_disk_dqheader) /* Offset of info header in file */
796 #define QT_TREEOFF 1 /* Offset of tree in file in blocks */
797 #define V2_DQTREEOFF QT_TREEOFF
799 # endif /* !defined(HAVE_QUOTAIO_V1_H) ... */
800 #endif /* __KERNEL__ */
804 #endif /* _LUSTRE_QUOTA_H */