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) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _OFD_INTERNAL_H
38 #define _OFD_INTERNAL_H
41 #include <obd_class.h>
42 #include <dt_object.h>
43 #include <md_object.h>
44 #include <lustre_fid.h>
46 #include <lustre_capa.h>
48 #define OFD_INIT_OBJID 0
49 #define OFD_ROCOMPAT_SUPP (0)
50 #define OFD_INCOMPAT_SUPP (OBD_INCOMPAT_GROUPS | OBD_INCOMPAT_OST | \
51 OBD_INCOMPAT_COMMON_LR)
52 #define OFD_PRECREATE_BATCH_DEFAULT (FILTER_SUBDIR_COUNT * 4)
54 /* on small filesystems we should not precreate too many objects in
55 * a single transaction, otherwise we can overflow transactions */
56 #define OFD_PRECREATE_SMALL_FS (1024ULL * 1024 * 1024)
57 #define OFD_PRECREATE_BATCH_SMALL 8
59 /* Limit the returned fields marked valid to those that we actually might set */
60 #define OFD_VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
61 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
63 /* per-client-per-object persistent state (LRU) */
65 cfs_list_t fmd_list; /* linked to fed_mod_list */
66 struct lu_fid fmd_fid; /* FID being written to */
67 __u64 fmd_mactime_xid; /* xid highest {m,a,c}time setattr */
68 cfs_time_t fmd_expire; /* time when the fmd should expire */
69 int fmd_refcount; /* reference counter - list holds 1 */
72 #define OFD_FMD_MAX_NUM_DEFAULT 128
73 #define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * HZ)
77 LPROC_OFD_STATS_READ = 0,
78 LPROC_OFD_STATS_WRITE,
79 LPROC_OFD_STATS_GETATTR,
80 LPROC_OFD_STATS_SETATTR,
81 LPROC_OFD_STATS_PUNCH,
83 LPROC_OFD_STATS_DESTROY,
84 LPROC_OFD_STATS_CREATE,
85 LPROC_OFD_STATS_STATFS,
86 LPROC_OFD_STATS_GET_INFO,
87 LPROC_OFD_STATS_SET_INFO,
88 LPROC_OFD_STATS_QUOTACTL,
92 static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
93 char *jobid, long amount)
95 if (exp->exp_obd && exp->exp_obd->obd_stats)
96 lprocfs_counter_add(exp->exp_obd->obd_stats, opcode, amount);
98 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
99 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
100 lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
102 if (exp->exp_nid_stats != NULL &&
103 exp->exp_nid_stats->nid_stats != NULL) {
104 lprocfs_counter_add(exp->exp_nid_stats->nid_stats, opcode,
112 spinlock_t os_last_oid_lock;
113 struct mutex os_create_lock;
114 cfs_atomic_t os_refc;
115 struct dt_object *os_lastid_obj;
116 unsigned long os_destroys_in_progress:1;
120 struct dt_device ofd_dt_dev;
121 struct dt_device *ofd_osd;
122 struct obd_export *ofd_osd_exp;
123 struct dt_device_param ofd_dt_conf;
124 /* DLM name-space for meta-data locks maintained by this server */
125 struct ldlm_namespace *ofd_namespace;
127 /* transaction callbacks */
128 struct dt_txn_callback ofd_txn_cb;
131 struct lu_target ofd_lut;
132 struct dt_object *ofd_health_check_file;
134 int ofd_subdir_count;
136 cfs_list_t ofd_seq_list;
137 rwlock_t ofd_seq_list_lock;
139 int ofd_precreate_batch;
140 spinlock_t ofd_batch_lock;
142 /* protect all statfs-related counters */
143 spinlock_t ofd_osfs_lock;
144 /* statfs optimization: we cache a bit */
145 struct obd_statfs ofd_osfs;
148 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
149 obd_size ofd_osfs_unstable;
151 /* counters used during statfs update, protected by ofd_osfs_lock.
152 * record when some statfs refresh are in progress */
153 int ofd_statfs_inflight;
154 /* track writes completed while statfs refresh is underway.
155 * tracking is only effective when ofd_statfs_inflight > 1 */
156 obd_size ofd_osfs_inflight;
158 /* grants: all values in bytes */
159 /* grant lock to protect all grant counters */
160 spinlock_t ofd_grant_lock;
161 /* total amount of dirty data reported by clients in incoming obdo */
162 obd_size ofd_tot_dirty;
163 /* sum of filesystem space granted to clients for async writes */
164 obd_size ofd_tot_granted;
165 /* grant used by I/Os in progress (between prepare and commit) */
166 obd_size ofd_tot_pending;
167 /* free space threshold over which we stop granting space to clients
168 * ofd_grant_ratio is stored as a fixed-point fraction using
169 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
172 /* number of clients using grants */
173 int ofd_tot_granted_clients;
175 /* ofd mod data: ofd_device wide values */
176 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
177 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
179 spinlock_t ofd_flags_lock;
180 unsigned long ofd_raid_degraded:1,
181 /* sync journal on writes */
183 /* shall we grant space to clients not
184 * supporting OBD_CONNECT_GRANT_PARAM? */
185 ofd_grant_compat_disable:1;
186 struct seq_server_site ofd_seq_site;
189 static inline struct ofd_device *ofd_dev(struct lu_device *d)
191 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
194 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
196 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
199 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
201 return ofd_dev(exp->exp_obd->obd_lu_dev);
204 static inline char *ofd_name(struct ofd_device *ofd)
206 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
210 struct lu_object_header ofo_header;
211 struct dt_object ofo_obj;
215 static inline struct ofd_object *ofd_obj(struct lu_object *o)
217 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
220 static inline int ofd_object_exists(struct ofd_object *obj)
222 LASSERT(obj != NULL);
223 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
225 return lu_object_exists(&obj->ofo_obj.do_lu);
228 static inline struct dt_object *fo2dt(struct ofd_object *obj)
230 return &obj->ofo_obj;
233 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
235 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
237 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
240 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
242 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
245 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
246 const struct ofd_object *obj)
248 /* TODO: see mdd_object_capa() */
252 static inline void ofd_read_lock(const struct lu_env *env,
253 struct ofd_object *fo)
255 struct dt_object *next = ofd_object_child(fo);
257 next->do_ops->do_read_lock(env, next, 0);
260 static inline void ofd_read_unlock(const struct lu_env *env,
261 struct ofd_object *fo)
263 struct dt_object *next = ofd_object_child(fo);
265 next->do_ops->do_read_unlock(env, next);
268 static inline void ofd_write_lock(const struct lu_env *env,
269 struct ofd_object *fo)
271 struct dt_object *next = ofd_object_child(fo);
273 next->do_ops->do_write_lock(env, next, 0);
276 static inline void ofd_write_unlock(const struct lu_env *env,
277 struct ofd_object *fo)
279 struct dt_object *next = ofd_object_child(fo);
281 next->do_ops->do_write_unlock(env, next);
285 * Common data shared by obdofd-level handlers. This is allocated per-thread
286 * to reduce stack consumption.
288 struct ofd_thread_info {
289 const struct lu_env *fti_env;
291 struct obd_export *fti_exp;
294 __u64 fti_pre_version;
295 __u32 fti_has_trans:1, /* has txn already */
298 struct lu_fid fti_fid;
299 struct lu_attr fti_attr;
300 struct lu_attr fti_attr2;
301 struct ldlm_res_id fti_resid;
302 struct filter_fid fti_mds_fid;
303 struct ost_id fti_ostid;
304 struct ofd_object *fti_obj;
306 char name[64]; /* for ofd_init0() */
307 struct obd_statfs osfs; /* for obdofd_statfs() */
310 /* Ops object filename */
311 struct lu_name fti_name;
312 struct dt_object_format fti_dof;
313 struct lu_buf fti_buf;
316 /* Space used by the I/O, used by grant code */
317 unsigned long fti_used;
318 struct ost_lvb fti_lvb;
321 extern void target_recovery_fini(struct obd_device *obd);
322 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
325 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
326 int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
327 struct lustre_capa *capa, __u64 opc);
328 void ofd_free_capa_keys(struct ofd_device *ofd);
331 extern struct lu_context_key ofd_thread_key;
332 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd);
335 extern struct obd_ops ofd_obd_ops;
336 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
337 struct obd_statfs *osfs, __u64 max_age,
339 int ofd_orphans_destroy(const struct lu_env *env, struct obd_export *exp,
340 struct ofd_device *ofd, struct obdo *oa);
341 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
342 const struct lu_fid *fid, int orphan);
343 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
344 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
347 obd_id ofd_seq_last_oid(struct ofd_seq *oseq);
348 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id);
349 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
350 struct ofd_seq *oseq);
351 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
352 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq);
353 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
355 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
356 struct obd_device *obd);
357 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
358 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
359 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
361 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
364 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
365 struct obdo *oa, int objcount, struct obd_ioobj *obj,
366 struct niobuf_remote *rnb, int *nr_local,
367 struct niobuf_local *lnb, struct obd_trans_info *oti,
368 struct lustre_capa *capa);
369 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
370 struct obdo *oa, int objcount, struct obd_ioobj *obj,
371 struct niobuf_remote *rnb, int npages,
372 struct niobuf_local *lnb, struct obd_trans_info *oti,
376 struct thandle *ofd_trans_create(const struct lu_env *env,
377 struct ofd_device *ofd);
378 int ofd_trans_start(const struct lu_env *env,
379 struct ofd_device *ofd, struct ofd_object *fo,
381 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
382 struct thandle *th, int rc);
383 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
388 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
389 void ofd_stats_counter_init(struct lprocfs_stats *stats);
391 static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
393 memset(lvars, 0, sizeof(*lvars));
395 static inline void ofd_stats_counter_init(struct lprocfs_stats *stats) {}
399 struct ofd_object *ofd_object_find(const struct lu_env *env,
400 struct ofd_device *ofd,
401 const struct lu_fid *fid);
402 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
403 struct ofd_device *ofd,
404 const struct lu_fid *fid,
405 struct lu_attr *attr);
406 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
407 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
408 obd_id id, struct ofd_seq *oseq, int nr, int sync);
410 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
411 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
412 struct lu_attr *la, struct filter_fid *ff);
413 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
414 __u64 start, __u64 end, struct lu_attr *la,
415 struct filter_fid *ff);
416 int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
417 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
419 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
420 struct lu_attr *la, int is_setattr);
423 struct ofd_object *ofd_object_find_exists(const struct lu_env *env,
424 struct ofd_device *ofd,
427 struct ofd_object *fo;
429 fo = ofd_object_find(env, ofd, fid);
430 if (!IS_ERR(fo) && !ofd_object_exists(fo)) {
431 ofd_object_put(env, fo);
432 fo = ERR_PTR(-ENOENT);
438 #define OFD_GRANT_RATIO_SHIFT 8
439 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
441 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
444 static inline int ofd_grant_ratio_conv(int percentage)
446 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
449 static inline int ofd_grant_param_supp(struct obd_export *exp)
451 return !!(exp_connect_flags(exp) & OBD_CONNECT_GRANT_PARAM);
454 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
455 * That's 4KB=2^12 which is the biggest block size known to work whatever
456 * the client's page size is. */
457 #define COMPAT_BSIZE_SHIFT 12
458 static inline int ofd_grant_compat(struct obd_export *exp,
459 struct ofd_device *ofd)
461 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
462 * a block size > page size and consume PAGE_CACHE_SIZE of grant when
463 * dirtying a page regardless of the block size */
464 return !!(ofd_obd(ofd)->obd_self_export != exp &&
465 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
466 !ofd_grant_param_supp(exp));
469 static inline int ofd_grant_prohibit(struct obd_export *exp,
470 struct ofd_device *ofd)
472 /* When ofd_grant_compat_disable is set, we don't grant any space to
473 * clients not supporting OBD_CONNECT_GRANT_PARAM.
474 * Otherwise, space granted to such a client is inflated since it
475 * consumes PAGE_CACHE_SIZE of grant space per block */
476 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
479 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
480 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
481 obd_size want, bool conservative);
482 void ofd_grant_discard(struct obd_export *exp);
483 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
485 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
486 struct obdo *oa, struct niobuf_remote *rnb,
488 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
489 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
492 int ofd_fmd_init(void);
493 void ofd_fmd_exit(void);
494 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
496 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
498 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
499 void ofd_fmd_expire(struct obd_export *exp);
500 void ofd_fmd_cleanup(struct obd_export *exp);
502 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
504 #define ofd_fmd_drop(exp, fid) do {} while (0)
508 int ofd_fid_set_index(const struct lu_env *env, struct ofd_device *ofd,
510 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd);
511 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd);
514 extern struct ldlm_valblock_ops ofd_lvbo;
517 int ofd_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp,
518 void *req_cookie, ldlm_mode_t mode, __u64 flags,
521 static inline struct ofd_thread_info *ofd_info(const struct lu_env *env)
523 struct ofd_thread_info *info;
525 lu_env_refill((void *)env);
526 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
531 static inline struct ofd_thread_info *ofd_info_init(const struct lu_env *env,
532 struct obd_export *exp)
534 struct ofd_thread_info *info;
536 info = ofd_info(env);
537 LASSERT(info->fti_exp == NULL);
538 LASSERT(info->fti_env == NULL);
539 LASSERT(info->fti_attr.la_valid == 0);
543 info->fti_pre_version = 0;
544 info->fti_transno = 0;
545 info->fti_has_trans = 0;
549 static inline struct ofd_thread_info *tsi2ofd_info(struct tgt_session_info *tsi)
551 struct ptlrpc_request *req = tgt_ses_req(tsi);
552 struct ofd_thread_info *info;
554 info = ofd_info(tsi->tsi_env);
555 LASSERT(info->fti_exp == NULL);
556 LASSERT(info->fti_env == NULL);
557 LASSERT(info->fti_attr.la_valid == 0);
559 info->fti_env = tsi->tsi_env;
560 info->fti_exp = tsi->tsi_exp;
561 info->fti_has_trans = 0;
563 info->fti_xid = req->rq_xid;
564 /** VBR: take versions from request */
565 if (req->rq_reqmsg != NULL &&
566 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
567 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
569 info->fti_pre_version = pre_version ? pre_version[0] : 0;
570 info->fti_transno = lustre_msg_get_transno(req->rq_reqmsg);
575 static inline void ofd_oti2info(struct ofd_thread_info *info,
576 struct obd_trans_info *oti)
578 info->fti_xid = oti->oti_xid;
579 info->fti_transno = oti->oti_transno;
580 info->fti_pre_version = oti->oti_pre_version;
583 static inline void ofd_info2oti(struct ofd_thread_info *info,
584 struct obd_trans_info *oti)
586 oti->oti_xid = info->fti_xid;
587 LASSERTF(ergo(oti->oti_transno > 0,
588 oti->oti_transno == info->fti_transno),
589 "Overwrite replay transno "LPX64" by "LPX64"\n",
590 oti->oti_transno, info->fti_transno);
591 oti->oti_transno = info->fti_transno;
592 oti->oti_pre_version = info->fti_pre_version;
595 /* sync on lock cancel is useless when we force a journal flush,
596 * and if we enable async journal commit, we should also turn on
597 * sync on lock cancel if it is not enabled already. */
598 static inline void ofd_slc_set(struct ofd_device *ofd)
600 if (ofd->ofd_syncjournal == 1)
601 ofd->ofd_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
602 else if (ofd->ofd_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
603 ofd->ofd_lut.lut_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
606 static inline void ofd_prepare_fidea(struct filter_fid *ff,
607 const struct obdo *oa)
609 /* packing fid and converting it to LE for storing into EA.
610 * Here ->o_stripe_idx should be filled by LOV and rest of
611 * fields - by client. */
612 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
613 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
614 /* XXX: we are ignoring o_parent_ver here, since this should
615 * be the same for all objects in this fileset. */
616 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
619 /* niobuf_remote has no rnb_ prefix in master */
620 #define rnb_offset offset
621 #define rnb_flags flags
623 /* the same for niobuf_local */
624 #define lnb_flags flags
627 #endif /* _OFD_INTERNAL_H */