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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
39 #ifndef _LUSTRE_QUOTA_H
40 #define _LUSTRE_QUOTA_H
42 /** \defgroup quota quota
47 #if defined(__linux__)
48 #include <linux/lustre_quota.h>
49 #elif defined(__APPLE__)
50 #include <darwin/lustre_quota.h>
51 #elif defined(__WINNT__)
52 #include <winnt/lustre_quota.h>
54 #error Unsupported operating system.
57 #include <lustre_net.h>
58 #include <lustre/lustre_idl.h>
60 #include <obd_support.h>
69 #ifndef QUOTABLOCK_BITS
70 #define QUOTABLOCK_BITS 10
73 #ifndef QUOTABLOCK_SIZE
74 #define QUOTABLOCK_SIZE (1 << QUOTABLOCK_BITS)
78 #define toqb(x) (((x) + QUOTABLOCK_SIZE - 1) >> QUOTABLOCK_BITS)
81 #ifdef HAVE_QUOTA_SUPPORT
84 #define MAX_IQ_TIME 604800 /* (7*24*60*60) 1 week */
88 #define MAX_DQ_TIME 604800 /* (7*24*60*60) 1 week */
95 LQUOTA_FIRST_STAT = 0,
98 * these four are for measuring quota requests, for both of
99 * quota master and quota slaves
101 LQUOTA_SYNC_ACQ = LQUOTA_FIRST_STAT,
108 * these four measure how much time I/O threads spend on dealing
109 * with quota before and after writing data or creating files,
110 * only for quota slaves(lquota_chkquota and lquota_pending_commit)
112 LQUOTA_WAIT_FOR_CHK_BLK,
113 LQUOTA_WAIT_FOR_CHK_INO,
114 LQUOTA_WAIT_FOR_COMMIT_BLK,
115 LQUOTA_WAIT_FOR_COMMIT_INO,
119 * these two are for measuring time waiting return of quota reqs
120 * (qctxt_wait_pending_dqacq), only for quota salves
122 LQUOTA_WAIT_PENDING_BLK_QUOTA,
123 LQUOTA_WAIT_PENDING_INO_QUOTA,
127 * these two are for those when they are calling
128 * qctxt_wait_pending_dqacq, the quota req has returned already,
129 * only for quota salves
131 LQUOTA_NOWAIT_PENDING_BLK_QUOTA,
132 LQUOTA_NOWAIT_PENDING_INO_QUOTA,
136 * these are for quota ctl
142 * these are for adjust quota qunit, for both of
143 * quota master and quota slaves
151 /* structures to access admin quotafile */
152 struct lustre_mem_dqinfo {
153 unsigned int dqi_bgrace;
154 unsigned int dqi_igrace;
155 unsigned long dqi_flags;
156 unsigned int dqi_blocks;
157 unsigned int dqi_free_blk;
158 unsigned int dqi_free_entry;
161 struct lustre_quota_info {
162 struct file *qi_files[MAXQUOTAS];
163 struct lustre_mem_dqinfo qi_info[MAXQUOTAS];
164 lustre_quota_version_t qi_version;
167 struct lustre_mem_dqblk {
168 __u64 dqb_bhardlimit; /**< absolute limit on disk blks alloc */
169 __u64 dqb_bsoftlimit; /**< preferred limit on disk blks */
170 __u64 dqb_curspace; /**< current used space */
171 __u64 dqb_ihardlimit; /**< absolute limit on allocated inodes */
172 __u64 dqb_isoftlimit; /**< preferred inode limit */
173 __u64 dqb_curinodes; /**< current # allocated inodes */
174 time_t dqb_btime; /**< time limit for excessive disk use */
175 time_t dqb_itime; /**< time limit for excessive inode use */
178 struct lustre_dquot {
179 /** Hash list in memory, protect by dquot_hash_lock */
181 /** Protect the data in lustre_dquot */
182 cfs_mutex_t dq_mutex;
184 cfs_atomic_t dq_refcnt;
185 /** Pointer of quota info it belongs to */
186 struct lustre_quota_info *dq_info;
187 /** Offset of dquot on disk */
189 /** ID this applies to (uid, gid) */
191 /** Type fo quota (USRQUOTA, GRPQUOUTA) */
193 /** See DQ_ in quota.h */
194 unsigned long dq_flags;
195 /** Diskquota usage */
196 struct lustre_mem_dqblk dq_dqb;
204 /* set inode quota limitation on a quota uid/gid */
205 #define QI_SET (1 << 30)
206 /* set block quota limitation on a quota uid/gid */
207 #define QB_SET (1 << 31)
210 #define QFILE_RD_INFO 2
211 #define QFILE_WR_INFO 3
212 #define QFILE_INIT_INFO 4
213 #define QFILE_RD_DQUOT 5
214 #define QFILE_WR_DQUOT 6
215 #define QFILE_CONVERT 7
217 /* admin quotafile operations */
218 int lustre_check_quota_file(struct lustre_quota_info *lqi, int type);
219 int lustre_read_quota_info(struct lustre_quota_info *lqi, int type);
220 int lustre_write_quota_info(struct lustre_quota_info *lqi, int type);
221 int lustre_read_dquot(struct lustre_dquot *dquot);
222 int lustre_commit_dquot(struct lustre_dquot *dquot);
223 int lustre_init_quota_info(struct lustre_quota_info *lqi, int type);
224 int lustre_get_qids(struct file *file, struct inode *inode, int type,
226 int lustre_quota_convert(struct lustre_quota_info *lqi, int type);
228 typedef int (*dqacq_handler_t) (struct obd_device * obd, struct qunit_data * qd,
231 /* user quota is turned on on filter */
232 #define LQC_USRQUOTA_FLAG (1 << 0)
233 /* group quota is turned on on filter */
234 #define LQC_GRPQUOTA_FLAG (1 << 1)
236 #define UGQUOTA2LQC(id) ((Q_TYPEMATCH(id, USRQUOTA) ? LQC_USRQUOTA_FLAG : 0) | \
237 (Q_TYPEMATCH(id, GRPQUOTA) ? LQC_GRPQUOTA_FLAG : 0))
239 struct lustre_quota_ctxt {
240 /** superblock this applies to */
241 struct super_block *lqc_sb;
242 /** obd_device_target for obt_rwsem */
243 struct obd_device_target *lqc_obt;
244 /** import used to send dqacq/dqrel RPC */
245 struct obd_import *lqc_import;
246 /** dqacq/dqrel RPC handler, only for quota master */
247 dqacq_handler_t lqc_handler;
249 unsigned long lqc_flags;
251 unsigned long lqc_recovery:1, /** Doing recovery */
253 * the function of change qunit size
256 lqc_valid:1, /** this qctxt is valid or not */
258 * tell whether of not quota_type has
259 * been processed, so that the master
260 * knows when it can start processing
261 * incoming acq/rel quota requests
265 * original unit size of file quota and
266 * upper limitation for adjust file qunit
268 unsigned long lqc_iunit_sz;
270 * Trigger dqacq when available file
271 * quota less than this value, trigger
272 * dqrel when available file quota
273 * more than this value + 1 iunit
275 unsigned long lqc_itune_sz;
277 * original unit size of block quota and
278 * upper limitation for adjust block qunit
280 unsigned long lqc_bunit_sz;
281 /** See comment of lqc_itune_sz */
282 unsigned long lqc_btune_sz;
283 /** all lustre_qunit_size structures */
284 cfs_hash_t *lqc_lqs_hash;
288 * the values below are relative to how master change its qunit sizes
291 * this affects the boundary of
292 * shrinking and enlarging qunit size. default=4
294 unsigned long lqc_cqs_boundary_factor;
295 /** the least value of block qunit */
296 unsigned long lqc_cqs_least_bunit;
297 /** the least value of inode qunit */
298 unsigned long lqc_cqs_least_iunit;
300 * when enlarging, qunit size will
301 * mutilple it; when shrinking,
302 * qunit size will divide it
304 unsigned long lqc_cqs_qs_factor;
306 * avoid ping-pong effect of
307 * adjusting qunit size. How many
308 * seconds must be waited between
309 * enlarging and shinking qunit
312 int lqc_switch_seconds;
314 * when blk qunit reaches this value,
315 * later write reqs from client should be sync b=16642
318 /** guard lqc_imp_valid now */
319 cfs_spinlock_t lqc_lock;
321 * when mds isn't connected, threads
322 * on osts who send the quota reqs
323 * with wait==1 will be put here b=14840
325 cfs_waitq_t lqc_wait_for_qmaster;
326 struct proc_dir_entry *lqc_proc_dir;
327 /** lquota statistics */
328 struct lprocfs_stats *lqc_stats;
329 /** the number of used hashed lqs */
330 cfs_atomic_t lqc_lqs;
331 /** no lqs are in use */
332 cfs_waitq_t lqc_lqs_waitq;
335 #define QUOTA_MASTER_READY(qctxt) (qctxt)->lqc_setup = 1
336 #define QUOTA_MASTER_UNREADY(qctxt) (qctxt)->lqc_setup = 0
338 struct lustre_qunit_size {
339 cfs_hlist_node_t lqs_hash; /** the hash entry */
340 unsigned int lqs_id; /** id of user/group */
341 unsigned long lqs_flags; /** 31st bit is QB_SET, 30th bit is QI_SET
342 * other bits are same as LQUOTA_FLAGS_*
344 unsigned long lqs_iunit_sz; /** Unit size of file quota currently */
346 * Trigger dqacq when available file quota
347 * less than this value, trigger dqrel
348 * when more than this value + 1 iunit
350 unsigned long lqs_itune_sz;
351 unsigned long lqs_bunit_sz; /** Unit size of block quota currently */
352 unsigned long lqs_btune_sz; /** See comment of lqs itune sz */
353 /** the blocks reached ost and don't finish */
354 unsigned long lqs_bwrite_pending;
355 /** the inodes reached mds and don't finish */
356 unsigned long lqs_iwrite_pending;
357 /** when inodes are allocated/released, this value will record it */
358 long long lqs_ino_rec;
359 /** when blocks are allocated/released, this value will record it */
360 long long lqs_blk_rec;
361 cfs_atomic_t lqs_refcount;
362 cfs_time_t lqs_last_bshrink; /** time of last block shrink */
363 cfs_time_t lqs_last_ishrink; /** time of last inode shrink */
364 cfs_spinlock_t lqs_lock;
365 unsigned long long lqs_key; /** hash key */
366 struct lustre_quota_ctxt *lqs_ctxt; /** quota ctxt */
369 #define LQS_IS_GRP(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_GRP)
370 #define LQS_IS_ADJBLK(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJBLK)
371 #define LQS_IS_ADJINO(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_ADJINO)
372 #define LQS_IS_RECOVERY(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_RECOVERY)
373 #define LQS_IS_SETQUOTA(lqs) ((lqs)->lqs_flags & LQUOTA_FLAGS_SETQUOTA)
375 #define LQS_SET_GRP(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_GRP)
376 #define LQS_SET_ADJBLK(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJBLK)
377 #define LQS_SET_ADJINO(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_ADJINO)
378 #define LQS_SET_RECOVERY(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_RECOVERY)
379 #define LQS_SET_SETQUOTA(lqs) ((lqs)->lqs_flags |= LQUOTA_FLAGS_SETQUOTA)
381 #define LQS_CLEAR_RECOVERY(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_RECOVERY)
382 #define LQS_CLEAR_SETQUOTA(lqs) ((lqs)->lqs_flags &= ~LQUOTA_FLAGS_SETQUOTA)
384 /* In the hash for lustre_qunit_size, the key is decided by
385 * grp_or_usr and uid/gid, in here, I combine these two values,
386 * which will make comparing easier and more efficient */
387 #define LQS_KEY(is_grp, id) ((is_grp ? 1ULL << 32: 0) + id)
388 #define LQS_KEY_ID(key) (key & 0xffffffff)
389 #define LQS_KEY_GRP(key) (key >> 32)
391 static inline void lqs_getref(struct lustre_qunit_size *lqs)
393 int count = cfs_atomic_inc_return(&lqs->lqs_refcount);
395 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count);
398 static inline void lqs_putref(struct lustre_qunit_size *lqs)
400 int count = cfs_atomic_read(&lqs->lqs_refcount);
403 CDEBUG(D_INFO, "lqs=%p refcount %d\n", lqs, count - 1);
405 if (cfs_atomic_dec_and_test(&lqs->lqs_refcount)) {
406 if (cfs_atomic_dec_and_test(&lqs->lqs_ctxt->lqc_lqs))
407 cfs_waitq_signal(&lqs->lqs_ctxt->lqc_lqs_waitq);
414 struct lustre_quota_info {
417 struct lustre_quota_ctxt {
420 #define QUOTA_MASTER_READY(qctxt)
421 #define QUOTA_MASTER_UNREADY(qctxt)
423 #endif /* !__KERNEL__ */
427 #define LL_DQUOT_OFF(sb) do {} while(0)
429 struct lustre_quota_info {
432 struct lustre_quota_ctxt {
435 #endif /* !HAVE_QUOTA_SUPPORT */
437 /* If the (quota limit < qunit * slave count), the slave which can't
438 * acquire qunit should set it's local limit as MIN_QLIMIT */
441 struct quotacheck_thread_args {
442 struct obd_export *qta_exp; /** obd export */
443 struct obd_device *qta_obd; /** obd device */
444 struct obd_quotactl qta_oqctl; /** obd_quotactl args */
445 struct super_block *qta_sb; /** obd super block */
446 cfs_semaphore_t *qta_sem; /** obt_quotachecking */
449 struct obd_trans_info;
450 typedef int (*quota_acquire)(struct obd_device *obd, const unsigned int id[],
451 struct obd_trans_info *oti, int isblk);
454 int (*quota_init) (void);
455 int (*quota_exit) (void);
456 int (*quota_setup) (struct obd_device *);
457 int (*quota_cleanup) (struct obd_device *);
459 * For quota master, close admin quota files
461 int (*quota_fs_cleanup) (struct obd_device *);
462 int (*quota_ctl) (struct obd_device *, struct obd_export *,
463 struct obd_quotactl *);
464 int (*quota_check) (struct obd_device *, struct obd_export *,
465 struct obd_quotactl *);
466 int (*quota_recovery) (struct obd_device *);
469 * For quota master/slave, adjust quota limit after fs operation
471 int (*quota_adjust) (struct obd_device *, const unsigned int[],
472 const unsigned int[], int, int);
475 * For quota slave, set import, trigger quota recovery,
476 * For quota master, set lqc_setup
478 int (*quota_setinfo) (struct obd_device *, void *);
481 * For quota slave, clear import when relative import is invalid
483 int (*quota_clearinfo) (struct obd_export *, struct obd_device *);
486 * For quota slave, set proper thread resoure capability
488 int (*quota_enforce) (struct obd_device *, unsigned int);
491 * For quota slave, check whether specified uid/gid is over quota
493 int (*quota_getflag) (struct obd_device *, struct obdo *);
497 * For quota slave, acquire/release quota from master if needed
499 int (*quota_acquire) (struct obd_device *, const unsigned int [],
500 struct obd_trans_info *, int);
503 * For quota slave, check whether specified uid/gid's remaining quota
504 * can finish a block_write or inode_create rpc. It updates the pending
505 * record of block and inode, acquires quota if necessary
507 int (*quota_chkquota) (struct obd_device *, struct obd_export *,
508 const unsigned int [], int [],
509 int, quota_acquire, struct obd_trans_info *,
510 int, struct inode *, int);
513 * For quota client, the actions after the pending write is committed
515 int (*quota_pending_commit) (struct obd_device *, const unsigned int [],
520 * For quota client, poll if the quota check done
522 int (*quota_poll_check) (struct obd_export *, struct if_quotacheck *);
525 * For quota client, check whether specified uid/gid is over quota
527 int (*quota_chkdq) (struct client_obd *, const unsigned int []);
530 * For quota client, set over quota flag for specifed uid/gid
532 int (*quota_setdq) (struct client_obd *, const unsigned int [],
536 * For adjusting qunit size b=10600
538 int (*quota_adjust_qunit) (struct obd_export *exp,
539 struct quota_adjust_qunit *oqaq,
540 struct lustre_quota_ctxt *qctxt,
541 struct ptlrpc_request_set *rqset);
545 #define Q_COPY(out, in, member) (out)->member = (in)->member
547 #define QUOTA_OP(interface, op) interface->quota_ ## op
549 #define QUOTA_CHECK_OP(interface, op) \
553 if (!QUOTA_OP(interface, op)) { \
554 CERROR("no quota operation: " #op "\n"); \
555 RETURN(-EOPNOTSUPP); \
559 static inline int lquota_init(quota_interface_t *interface)
564 QUOTA_CHECK_OP(interface, init);
565 rc = QUOTA_OP(interface, init)();
569 static inline int lquota_exit(quota_interface_t *interface)
574 QUOTA_CHECK_OP(interface, exit);
575 rc = QUOTA_OP(interface, exit)();
579 static inline int lquota_setup(quota_interface_t *interface,
580 struct obd_device *obd)
585 QUOTA_CHECK_OP(interface, setup);
586 rc = QUOTA_OP(interface, setup)(obd);
590 static inline int lquota_cleanup(quota_interface_t *interface,
591 struct obd_device *obd)
596 QUOTA_CHECK_OP(interface, cleanup);
597 rc = QUOTA_OP(interface, cleanup)(obd);
601 static inline int lquota_fs_cleanup(quota_interface_t *interface,
602 struct obd_device *obd)
607 QUOTA_CHECK_OP(interface, fs_cleanup);
608 rc = QUOTA_OP(interface, fs_cleanup)(obd);
612 static inline int lquota_recovery(quota_interface_t *interface,
613 struct obd_device *obd)
618 QUOTA_CHECK_OP(interface, recovery);
619 rc = QUOTA_OP(interface, recovery)(obd);
623 static inline int lquota_check(quota_interface_t *interface,
624 struct obd_device *obd,
625 struct obd_export *exp,
626 struct obd_quotactl *oqctl)
631 QUOTA_CHECK_OP(interface, check);
632 rc = QUOTA_OP(interface, check)(obd, exp, oqctl);
636 static inline int lquota_ctl(quota_interface_t *interface,
637 struct obd_device *obd,
638 struct obd_quotactl *oqctl)
643 QUOTA_CHECK_OP(interface, ctl);
644 rc = QUOTA_OP(interface, ctl)(obd, NULL, oqctl);
648 static inline int lquota_adjust(quota_interface_t *interface,
649 struct obd_device *obd,
650 const unsigned int qcids[],
651 const unsigned int qpids[],
657 QUOTA_CHECK_OP(interface, adjust);
658 ret = QUOTA_OP(interface, adjust)(obd, qcids, qpids, rc, opc);
662 static inline int lquota_setinfo(quota_interface_t *interface,
663 struct obd_device *obd,
669 QUOTA_CHECK_OP(interface, setinfo);
670 rc = QUOTA_OP(interface, setinfo)(obd, data);
674 static inline int lquota_clearinfo(quota_interface_t *interface,
675 struct obd_export *exp,
676 struct obd_device *obd)
681 QUOTA_CHECK_OP(interface, clearinfo);
682 rc = QUOTA_OP(interface, clearinfo)(exp, obd);
686 static inline int lquota_enforce(quota_interface_t *interface,
687 struct obd_device *obd,
693 QUOTA_CHECK_OP(interface, enforce);
694 rc = QUOTA_OP(interface, enforce)(obd, ignore);
698 static inline int lquota_getflag(quota_interface_t *interface,
699 struct obd_device *obd, struct obdo *oa)
704 QUOTA_CHECK_OP(interface, getflag);
705 rc = QUOTA_OP(interface, getflag)(obd, oa);
710 static inline int lquota_chkquota(quota_interface_t *interface,
711 struct obd_device *obd,
712 struct obd_export *exp,
713 const unsigned int id[], int pending[],
714 int count, struct obd_trans_info *oti,
715 int isblk, void *data, int frags)
720 QUOTA_CHECK_OP(interface, chkquota);
721 QUOTA_CHECK_OP(interface, acquire);
722 rc = QUOTA_OP(interface, chkquota)(obd, exp, id, pending, count,
723 QUOTA_OP(interface, acquire), oti,
724 isblk, (struct inode *)data, frags);
728 static inline int lquota_pending_commit(quota_interface_t *interface,
729 struct obd_device *obd,
730 const unsigned int id[],
731 int pending[], int isblk)
736 QUOTA_CHECK_OP(interface, pending_commit);
737 rc = QUOTA_OP(interface, pending_commit)(obd, id, pending, isblk);
757 #define LUSTRE_ADMIN_QUOTAFILES_V2 {\
758 "admin_quotafile_v2.usr", /** user admin quotafile */\
759 "admin_quotafile_v2.grp" /** group admin quotafile */\
763 * Definitions of structures for vfsv0 quota format
764 * Source linux/fs/quota/quotaio_v2.h
766 * The following definitions are normally found in private kernel headers.
767 * However, some sites build Lustre against kernel development headers rather
768 * than than full kernel source, so we provide them here for compatibility.
771 # if !defined(HAVE_QUOTAIO_H) && !defined(HAVE_FS_QUOTA_QUOTAIO_H) && \
772 !defined(HAVE_FS_QUOTAIO_H)
774 #include <linux/types.h>
775 #include <linux/quota.h>
777 #define V2_INITQMAGICS {\
778 0xd9c01f11, /* USRQUOTA */\
779 0xd9c01927 /* GRPQUOTA */\
782 /* Header with type and version specific information */
783 struct v2_disk_dqinfo {
784 __le32 dqi_bgrace; /* Time before block soft limit becomes hard limit */
785 __le32 dqi_igrace; /* Time before inode soft limit becomes hard limit */
786 __le32 dqi_flags; /* Flags for quotafile (DQF_*) */
787 __le32 dqi_blocks; /* Number of blocks in file */
788 __le32 dqi_free_blk; /* Number of first free block in the list */
789 __le32 dqi_free_entry; /* Number of block with at least one free entry */
792 /* First generic header */
793 struct v2_disk_dqheader {
794 __le32 dqh_magic; /* Magic number identifying file */
795 __le32 dqh_version; /* File version */
797 #define V2_DQINFOOFF sizeof(struct v2_disk_dqheader) /* Offset of info header in file */
798 #define QT_TREEOFF 1 /* Offset of tree in file in blocks */
799 #define V2_DQTREEOFF QT_TREEOFF
801 # endif /* !defined(HAVE_QUOTAIO_V1_H) ... */
802 #endif /* __KERNEL__ */
806 #endif /* _LUSTRE_QUOTA_H */