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)
80 #define MAX_IQ_TIME 604800 /* (7*24*60*60) 1 week */
84 #define MAX_DQ_TIME 604800 /* (7*24*60*60) 1 week */
91 LQUOTA_FIRST_STAT = 0,
94 * these four are for measuring quota requests, for both of
95 * quota master and quota slaves
97 LQUOTA_SYNC_ACQ = LQUOTA_FIRST_STAT,
104 * these four measure how much time I/O threads spend on dealing
105 * with quota before and after writing data or creating files,
106 * only for quota slaves(lquota_chkquota and lquota_pending_commit)
108 LQUOTA_WAIT_FOR_CHK_BLK,
109 LQUOTA_WAIT_FOR_CHK_INO,
110 LQUOTA_WAIT_FOR_COMMIT_BLK,
111 LQUOTA_WAIT_FOR_COMMIT_INO,
115 * these two are for measuring time waiting return of quota reqs
116 * (qctxt_wait_pending_dqacq), only for quota salves
118 LQUOTA_WAIT_PENDING_BLK_QUOTA,
119 LQUOTA_WAIT_PENDING_INO_QUOTA,
123 * these two are for those when they are calling
124 * qctxt_wait_pending_dqacq, the quota req has returned already,
125 * only for quota salves
127 LQUOTA_NOWAIT_PENDING_BLK_QUOTA,
128 LQUOTA_NOWAIT_PENDING_INO_QUOTA,
132 * these are for quota ctl
138 * these are for adjust quota qunit, for both of
139 * quota master and quota slaves
147 /* structures to access admin quotafile */
148 struct lustre_mem_dqinfo {
149 unsigned int dqi_bgrace;
150 unsigned int dqi_igrace;
151 unsigned long dqi_flags;
152 unsigned int dqi_blocks;
153 unsigned int dqi_free_blk;
154 unsigned int dqi_free_entry;
157 struct lustre_quota_info {
158 struct file *qi_files[MAXQUOTAS];
159 struct lustre_mem_dqinfo qi_info[MAXQUOTAS];
160 lustre_quota_version_t qi_version;
163 struct lustre_mem_dqblk {
164 __u64 dqb_bhardlimit; /**< absolute limit on disk blks alloc */
165 __u64 dqb_bsoftlimit; /**< preferred limit on disk blks */
166 __u64 dqb_curspace; /**< current used space */
167 __u64 dqb_ihardlimit; /**< absolute limit on allocated inodes */
168 __u64 dqb_isoftlimit; /**< preferred inode limit */
169 __u64 dqb_curinodes; /**< current # allocated inodes */
170 time_t dqb_btime; /**< time limit for excessive disk use */
171 time_t dqb_itime; /**< time limit for excessive inode use */
174 struct lustre_dquot {
175 /** Hash list in memory, protect by dquot_hash_lock */
177 /** Protect the data in lustre_dquot */
178 cfs_mutex_t dq_mutex;
180 cfs_atomic_t dq_refcnt;
181 /** Pointer of quota info it belongs to */
182 struct lustre_quota_info *dq_info;
183 /** Offset of dquot on disk */
185 /** ID this applies to (uid, gid) */
187 /** Type fo quota (USRQUOTA, GRPQUOUTA) */
189 /** See DQ_ in quota.h */
190 unsigned long dq_flags;
191 /** Diskquota usage */
192 struct lustre_mem_dqblk dq_dqb;
200 /* set inode quota limitation on a quota uid/gid */
201 #define QI_SET (1 << 30)
202 /* set block quota limitation on a quota uid/gid */
203 #define QB_SET (1 << 31)
206 #define QFILE_RD_INFO 2
207 #define QFILE_WR_INFO 3
208 #define QFILE_INIT_INFO 4
209 #define QFILE_RD_DQUOT 5
210 #define QFILE_WR_DQUOT 6
211 #define QFILE_CONVERT 7
213 /* admin quotafile operations */
214 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
215 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
216 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
217 int lustre_read_dquot(struct lustre_dquot *dquot);
218 int lustre_commit_dquot(struct lustre_dquot *dquot);
219 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
220 int lustre_get_qids(struct file *file, struct inode *inode, int type,
222 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
224 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
227 /* user quota is turned on on filter */
228 #define LQC_USRQUOTA_FLAG (1 << 0)
229 /* group quota is turned on on filter */
230 #define LQC_GRPQUOTA_FLAG (1 << 1)
232 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
233 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
235 struct lustre_quota_ctxt {
236 /** superblock this applies to */
237 struct super_block *lqc_sb;
238 /** obd_device_target for obt_rwsem */
239 struct obd_device_target *lqc_obt;
240 /** import used to send dqacq/dqrel RPC */
241 struct obd_import *lqc_import;
242 /** dqacq/dqrel RPC handler, only for quota master */
243 dqacq_handler_t lqc_handler;
245 unsigned long lqc_flags;
247 unsigned long lqc_recovery:1, /** Doing recovery */
249 * the function of change qunit size
252 lqc_valid:1, /** this qctxt is valid or not */
254 * tell whether of not quota_type has
255 * been processed, so that the master
256 * knows when it can start processing
257 * incoming acq/rel quota requests
261 * original unit size of file quota and
262 * upper limitation for adjust file qunit
264 unsigned long lqc_iunit_sz;
266 * Trigger dqacq when available file
267 * quota less than this value, trigger
268 * dqrel when available file quota
269 * more than this value + 1 iunit
271 unsigned long lqc_itune_sz;
273 * original unit size of block quota and
274 * upper limitation for adjust block qunit
276 unsigned long lqc_bunit_sz;
277 /** See comment of lqc_itune_sz */
278 unsigned long lqc_btune_sz;
279 /** all lustre_qunit_size structures */
280 cfs_hash_t *lqc_lqs_hash;
284 * the values below are relative to how master change its qunit sizes
287 * this affects the boundary of
288 * shrinking and enlarging qunit size. default=4
290 unsigned long lqc_cqs_boundary_factor;
291 /** the least value of block qunit */
292 unsigned long lqc_cqs_least_bunit;
293 /** the least value of inode qunit */
294 unsigned long lqc_cqs_least_iunit;
296 * when enlarging, qunit size will
297 * mutilple it; when shrinking,
298 * qunit size will divide it
300 unsigned long lqc_cqs_qs_factor;
302 * avoid ping-pong effect of
303 * adjusting qunit size. How many
304 * seconds must be waited between
305 * enlarging and shinking qunit
308 int lqc_switch_seconds;
310 * when blk qunit reaches this value,
311 * later write reqs from client should be sync b=16642
314 /** guard lqc_imp_valid now */
315 cfs_spinlock_t lqc_lock;
317 * when mds isn't connected, threads
318 * on osts who send the quota reqs
319 * with wait==1 will be put here b=14840
321 cfs_waitq_t lqc_wait_for_qmaster;
322 struct proc_dir_entry *lqc_proc_dir;
323 /** lquota statistics */
324 struct lprocfs_stats *lqc_stats;
325 /** the number of used hashed lqs */
326 cfs_atomic_t lqc_lqs;
327 /** no lqs are in use */
328 cfs_waitq_t lqc_lqs_waitq;
331 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
332 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
334 struct lustre_qunit_size {
335 cfs_hlist_node_t lqs_hash; /** the hash entry */
336 unsigned int lqs_id; /** id of user/group */
337 unsigned long lqs_flags; /** 31st bit is QB_SET, 30th bit is QI_SET
338 * other bits are same as LQUOTA_FLAGS_*
340 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
342 * Trigger dqacq when available file quota
343 * less than this value, trigger dqrel
344 * when more than this value + 1 iunit
346 unsigned long lqs_itune_sz;
347 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
348 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
349 /** the blocks reached ost and don't finish */
350 unsigned long lqs_bwrite_pending;
351 /** the inodes reached mds and don't finish */
352 unsigned long lqs_iwrite_pending;
353 /** when inodes are allocated/released, this value will record it */
354 long long lqs_ino_rec;
355 /** when blocks are allocated/released, this value will record it */
356 long long lqs_blk_rec;
357 cfs_atomic_t lqs_refcount;
358 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
359 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
360 cfs_spinlock_t lqs_lock;
361 unsigned long long lqs_key; /** hash key */
362 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
365 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
366 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
367 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
368 #define LQS_IS_RECOVERY(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_RECOVERY)
369 #define LQS_IS_SETQUOTA(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_SETQUOTA)
371 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
372 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
373 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
374 #define LQS_SET_RECOVERY(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_RECOVERY)
375 #define LQS_SET_SETQUOTA(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_SETQUOTA)
377 #define LQS_CLEAR_RECOVERY(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_RECOVERY)
378 #define LQS_CLEAR_SETQUOTA(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_SETQUOTA)
380 /* In the hash for lustre_qunit_size, the key is decided by
381 * grp_or_usr and uid/gid, in here, I combine these two values,
382 * which will make comparing easier and more efficient */
383 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
384 #define LQS_KEY_ID(key) (key & 0xffffffff)
385 #define LQS_KEY_GRP(key) (key >> 32)
387 static inline void lqs_getref(struct lustre_qunit_size *lqs)
389 int count = cfs_atomic_inc_return(&lqs->lqs_refcount);
391 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count);
394 static inline void lqs_putref(struct lustre_qunit_size *lqs)
396 int count = cfs_atomic_read(&lqs->lqs_refcount);
399 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count - 1);
401 if (cfs_atomic_dec_and_test(&lqs->lqs_refcount)) {
402 if (cfs_atomic_dec_and_test(&lqs->lqs_ctxt->lqc_lqs))
403 cfs_waitq_signal(&lqs->lqs_ctxt->lqc_lqs_waitq);
410 struct lustre_quota_info {
413 struct lustre_quota_ctxt {
416 #define QUOTA_MASTER_READY(qctxt)
417 #define QUOTA_MASTER_UNREADY(qctxt)
419 #endif /* !__KERNEL__ */
421 /* If the (quota limit < qunit * slave count), the slave which can't
422 * acquire qunit should set it's local limit as MIN_QLIMIT */
425 struct quotacheck_thread_args {
426 struct obd_export *qta_exp; /** obd export */
427 struct obd_device *qta_obd; /** obd device */
428 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
429 struct super_block *qta_sb; /** obd super block */
430 cfs_semaphore_t *qta_sem; /** obt_quotachecking */
433 struct obd_trans_info;
434 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
435 struct obd_trans_info *oti, int isblk);
438 int (*quota_init) (void);
439 int (*quota_exit) (void);
440 int (*quota_setup) (struct obd_device *);
441 int (*quota_cleanup) (struct obd_device *);
443 * For quota master, close admin quota files
445 int (*quota_fs_cleanup) (struct obd_device *);
446 int (*quota_ctl) (struct obd_device *, struct obd_export *,
447 struct obd_quotactl *);
448 int (*quota_check) (struct obd_device *, struct obd_export *,
449 struct obd_quotactl *);
450 int (*quota_recovery) (struct obd_device *);
453 * For quota master/slave, adjust quota limit after fs operation
455 int (*quota_adjust) (struct obd_device *, const unsigned int[],
456 const unsigned int[], int, int);
459 * For quota slave, set import, trigger quota recovery,
460 * For quota master, set lqc_setup
462 int (*quota_setinfo) (struct obd_device *, void *);
465 * For quota slave, clear import when relative import is invalid
467 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
470 * For quota slave, set proper thread resoure capability
472 int (*quota_enforce) (struct obd_device *, unsigned int);
475 * For quota slave, check whether specified uid/gid is over quota
477 int (*quota_getflag) (struct obd_device *, struct obdo *);
481 * For quota slave, acquire/release quota from master if needed
483 int (*quota_acquire) (struct obd_device *, const unsigned int [],
484 struct obd_trans_info *, int);
487 * For quota slave, check whether specified uid/gid's remaining quota
488 * can finish a block_write or inode_create rpc. It updates the pending
489 * record of block and inode, acquires quota if necessary
491 int (*quota_chkquota) (struct obd_device *, struct obd_export *,
492 const unsigned int [], int [],
493 int, quota_acquire, struct obd_trans_info *,
494 int, struct inode *, int);
497 * For quota client, the actions after the pending write is committed
499 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
504 * For quota client, poll if the quota check done
506 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
509 * For quota client, check whether specified uid/gid is over quota
511 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
514 * For quota client, set over quota flag for specifed uid/gid
516 int (*quota_setdq) (struct client_obd *, const unsigned int [],
520 * For adjusting qunit size b=10600
522 int (*quota_adjust_qunit) (struct obd_export *exp,
523 struct quota_adjust_qunit *oqaq,
524 struct lustre_quota_ctxt *qctxt,
525 struct ptlrpc_request_set *rqset);
529 #define Q_COPY(out, in, member) (out)->member = (in)->member
531 #define QUOTA_OP(interface, op) interface->quota_ ## op
533 #define QUOTA_CHECK_OP(interface, op) \
537 if (!QUOTA_OP(interface, op)) { \
538 CERROR("no quota operation: " #op "\n"); \
539 RETURN(-EOPNOTSUPP); \
543 static inline int lquota_init(quota_interface_t *interface)
548 QUOTA_CHECK_OP(interface, init);
549 rc = QUOTA_OP(interface, init)();
553 static inline int lquota_exit(quota_interface_t *interface)
558 QUOTA_CHECK_OP(interface, exit);
559 rc = QUOTA_OP(interface, exit)();
563 static inline int lquota_setup(quota_interface_t *interface,
564 struct obd_device *obd)
569 QUOTA_CHECK_OP(interface, setup);
570 rc = QUOTA_OP(interface, setup)(obd);
574 static inline int lquota_cleanup(quota_interface_t *interface,
575 struct obd_device *obd)
580 QUOTA_CHECK_OP(interface, cleanup);
581 rc = QUOTA_OP(interface, cleanup)(obd);
585 static inline int lquota_fs_cleanup(quota_interface_t *interface,
586 struct obd_device *obd)
591 QUOTA_CHECK_OP(interface, fs_cleanup);
592 rc = QUOTA_OP(interface, fs_cleanup)(obd);
596 static inline int lquota_recovery(quota_interface_t *interface,
597 struct obd_device *obd)
602 QUOTA_CHECK_OP(interface, recovery);
603 rc = QUOTA_OP(interface, recovery)(obd);
607 static inline int lquota_check(quota_interface_t *interface,
608 struct obd_device *obd,
609 struct obd_export *exp,
610 struct obd_quotactl *oqctl)
615 QUOTA_CHECK_OP(interface, check);
616 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
620 static inline int lquota_ctl(quota_interface_t *interface,
621 struct obd_device *obd,
622 struct obd_quotactl *oqctl)
627 QUOTA_CHECK_OP(interface, ctl);
628 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
632 static inline int lquota_adjust(quota_interface_t *interface,
633 struct obd_device *obd,
634 const unsigned int qcids[],
635 const unsigned int qpids[],
641 QUOTA_CHECK_OP(interface, adjust);
642 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
646 static inline int lquota_setinfo(quota_interface_t *interface,
647 struct obd_device *obd,
653 QUOTA_CHECK_OP(interface, setinfo);
654 rc = QUOTA_OP(interface, setinfo)(obd, data);
658 static inline int lquota_clearinfo(quota_interface_t *interface,
659 struct obd_export *exp,
660 struct obd_device *obd)
665 QUOTA_CHECK_OP(interface, clearinfo);
666 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
670 static inline int lquota_enforce(quota_interface_t *interface,
671 struct obd_device *obd,
677 QUOTA_CHECK_OP(interface, enforce);
678 rc = QUOTA_OP(interface, enforce)(obd, ignore);
682 static inline int lquota_getflag(quota_interface_t *interface,
683 struct obd_device *obd, struct obdo *oa)
688 QUOTA_CHECK_OP(interface, getflag);
689 rc = QUOTA_OP(interface, getflag)(obd, oa);
694 static inline int lquota_chkquota(quota_interface_t *interface,
695 struct obd_device *obd,
696 struct obd_export *exp,
697 const unsigned int id[], int pending[],
698 int count, struct obd_trans_info *oti,
699 int isblk, void *data, int frags)
704 QUOTA_CHECK_OP(interface, chkquota);
705 QUOTA_CHECK_OP(interface, acquire);
706 rc = QUOTA_OP(interface, chkquota)(obd, exp, id, pending, count,
707 QUOTA_OP(interface, acquire), oti,
708 isblk, (struct inode *)data, frags);
712 static inline int lquota_pending_commit(quota_interface_t *interface,
713 struct obd_device *obd,
714 const unsigned int id[],
715 int pending[], int isblk)
720 QUOTA_CHECK_OP(interface, pending_commit);
721 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
741 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
742 "admin_quotafile_v2.usr", /** user admin quotafile */\
743 "admin_quotafile_v2.grp" /** group admin quotafile */\
747 * Definitions of structures for vfsv0 quota format
748 * Source linux/fs/quota/quotaio_v2.h
750 * The following definitions are normally found in private kernel headers.
751 * However, some sites build Lustre against kernel development headers rather
752 * than than full kernel source, so we provide them here for compatibility.
755 # if !defined(HAVE_QUOTAIO_H) && !defined(HAVE_FS_QUOTA_QUOTAIO_H) && \
756 !defined(HAVE_FS_QUOTAIO_H)
758 #include <linux/types.h>
759 #include <linux/quota.h>
761 #define V2_INITQMAGICS {\
762 0xd9c01f11, /* USRQUOTA */\
763 0xd9c01927 /* GRPQUOTA */\
766 /* Header with type and version specific information */
767 struct v2_disk_dqinfo {
768 __le32 dqi_bgrace; /* Time before block soft limit becomes hard limit */
769 __le32 dqi_igrace; /* Time before inode soft limit becomes hard limit */
770 __le32 dqi_flags; /* Flags for quotafile (DQF_*) */
771 __le32 dqi_blocks; /* Number of blocks in file */
772 __le32 dqi_free_blk; /* Number of first free block in the list */
773 __le32 dqi_free_entry; /* Number of block with at least one free entry */
776 /* First generic header */
777 struct v2_disk_dqheader {
778 __le32 dqh_magic; /* Magic number identifying file */
779 __le32 dqh_version; /* File version */
781 #define V2_DQINFOOFF sizeof(struct v2_disk_dqheader) /* Offset of info header in file */
782 #define QT_TREEOFF 1 /* Offset of tree in file in blocks */
783 #define V2_DQTREEOFF QT_TREEOFF
785 # endif /* !defined(HAVE_QUOTAIO_V1_H) ... */
786 #endif /* __KERNEL__ */
790 #endif /* _LUSTRE_QUOTA_H */