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 (OBJ_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)
75 #define OFD_SOFT_SYNC_LIMIT_DEFAULT 16
79 LPROC_OFD_STATS_READ = 0,
80 LPROC_OFD_STATS_WRITE,
81 LPROC_OFD_STATS_GETATTR,
82 LPROC_OFD_STATS_SETATTR,
83 LPROC_OFD_STATS_PUNCH,
85 LPROC_OFD_STATS_DESTROY,
86 LPROC_OFD_STATS_CREATE,
87 LPROC_OFD_STATS_STATFS,
88 LPROC_OFD_STATS_GET_INFO,
89 LPROC_OFD_STATS_SET_INFO,
90 LPROC_OFD_STATS_QUOTACTL,
94 static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
95 char *jobid, long amount)
97 if (exp->exp_obd && exp->exp_obd->obd_stats)
98 lprocfs_counter_add(exp->exp_obd->obd_stats, opcode, amount);
100 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
101 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
102 lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
104 if (exp->exp_nid_stats != NULL &&
105 exp->exp_nid_stats->nid_stats != NULL) {
106 lprocfs_counter_add(exp->exp_nid_stats->nid_stats, opcode,
114 spinlock_t os_last_oid_lock;
115 struct mutex os_create_lock;
117 struct dt_object *os_lastid_obj;
118 unsigned long os_destroys_in_progress:1;
122 struct dt_device ofd_dt_dev;
123 struct dt_device *ofd_osd;
124 struct obd_export *ofd_osd_exp;
125 struct dt_device_param ofd_dt_conf;
126 /* DLM name-space for meta-data locks maintained by this server */
127 struct ldlm_namespace *ofd_namespace;
130 struct lu_target ofd_lut;
131 struct dt_object *ofd_health_check_file;
133 int ofd_subdir_count;
135 cfs_list_t ofd_seq_list;
136 rwlock_t ofd_seq_list_lock;
138 int ofd_precreate_batch;
139 spinlock_t ofd_batch_lock;
141 /* protect all statfs-related counters */
142 spinlock_t ofd_osfs_lock;
143 /* statfs optimization: we cache a bit */
144 struct obd_statfs ofd_osfs;
147 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
148 obd_size ofd_osfs_unstable;
150 /* counters used during statfs update, protected by ofd_osfs_lock.
151 * record when some statfs refresh are in progress */
152 int ofd_statfs_inflight;
153 /* track writes completed while statfs refresh is underway.
154 * tracking is only effective when ofd_statfs_inflight > 1 */
155 obd_size ofd_osfs_inflight;
157 /* grants: all values in bytes */
158 /* grant lock to protect all grant counters */
159 spinlock_t ofd_grant_lock;
160 /* total amount of dirty data reported by clients in incoming obdo */
161 obd_size ofd_tot_dirty;
162 /* sum of filesystem space granted to clients for async writes */
163 obd_size ofd_tot_granted;
164 /* grant used by I/Os in progress (between prepare and commit) */
165 obd_size ofd_tot_pending;
166 /* free space threshold over which we stop granting space to clients
167 * ofd_grant_ratio is stored as a fixed-point fraction using
168 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
171 /* number of clients using grants */
172 int ofd_tot_granted_clients;
174 /* ofd mod data: ofd_device wide values */
175 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
176 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
178 spinlock_t ofd_flags_lock;
179 unsigned long ofd_raid_degraded:1,
180 /* sync journal on writes */
182 /* shall we grant space to clients not
183 * supporting OBD_CONNECT_GRANT_PARAM? */
184 ofd_grant_compat_disable:1,
185 /* Protected by ofd_lastid_rwsem. */
186 ofd_lastid_rebuilding:1;
187 struct seq_server_site ofd_seq_site;
188 /* the limit of SOFT_SYNC RPCs that will trigger a soft sync */
189 unsigned int ofd_soft_sync_limit;
190 /* Protect ::ofd_lastid_rebuilding */
191 struct rw_semaphore ofd_lastid_rwsem;
192 __u64 ofd_lastid_gen;
195 static inline struct ofd_device *ofd_dev(struct lu_device *d)
197 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
200 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
202 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
205 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
207 return ofd_dev(exp->exp_obd->obd_lu_dev);
210 static inline char *ofd_name(struct ofd_device *ofd)
212 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
216 struct lu_object_header ofo_header;
217 struct dt_object ofo_obj;
221 static inline struct ofd_object *ofd_obj(struct lu_object *o)
223 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
226 static inline int ofd_object_exists(struct ofd_object *obj)
228 LASSERT(obj != NULL);
229 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
231 return lu_object_exists(&obj->ofo_obj.do_lu);
234 static inline struct dt_object *fo2dt(struct ofd_object *obj)
236 return &obj->ofo_obj;
239 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
241 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
243 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
246 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
248 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
251 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
252 const struct ofd_object *obj)
254 /* TODO: see mdd_object_capa() */
258 static inline void ofd_read_lock(const struct lu_env *env,
259 struct ofd_object *fo)
261 struct dt_object *next = ofd_object_child(fo);
263 next->do_ops->do_read_lock(env, next, 0);
266 static inline void ofd_read_unlock(const struct lu_env *env,
267 struct ofd_object *fo)
269 struct dt_object *next = ofd_object_child(fo);
271 next->do_ops->do_read_unlock(env, next);
274 static inline void ofd_write_lock(const struct lu_env *env,
275 struct ofd_object *fo)
277 struct dt_object *next = ofd_object_child(fo);
279 next->do_ops->do_write_lock(env, next, 0);
282 static inline void ofd_write_unlock(const struct lu_env *env,
283 struct ofd_object *fo)
285 struct dt_object *next = ofd_object_child(fo);
287 next->do_ops->do_write_unlock(env, next);
291 * Common data shared by obdofd-level handlers. This is allocated per-thread
292 * to reduce stack consumption.
294 struct ofd_thread_info {
295 const struct lu_env *fti_env;
297 struct obd_export *fti_exp;
299 __u64 fti_pre_version;
301 struct lu_fid fti_fid;
302 struct lu_attr fti_attr;
303 struct lu_attr fti_attr2;
304 struct ldlm_res_id fti_resid;
305 struct filter_fid fti_mds_fid;
306 struct ost_id fti_ostid;
307 struct ofd_object *fti_obj;
309 char name[64]; /* for ofd_init0() */
310 struct obd_statfs osfs; /* for obdofd_statfs() */
313 /* Ops object filename */
314 struct lu_name fti_name;
315 struct dt_object_format fti_dof;
316 struct lu_buf fti_buf;
319 /* Space used by the I/O, used by grant code */
320 unsigned long fti_used;
321 struct ost_lvb fti_lvb;
324 extern void target_recovery_fini(struct obd_device *obd);
325 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
328 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
329 int ofd_auth_capa(struct obd_export *exp, const struct lu_fid *fid,
330 obd_seq seq, struct lustre_capa *capa, __u64 opc);
331 void ofd_free_capa_keys(struct ofd_device *ofd);
334 extern struct lu_context_key ofd_thread_key;
335 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd);
336 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
337 struct lu_fid *fid, struct ll_user_fiemap *fiemap);
340 extern struct obd_ops ofd_obd_ops;
341 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
342 struct obd_statfs *osfs, __u64 max_age,
344 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
345 const struct lu_fid *fid, int orphan);
346 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
347 struct obd_statfs *osfs, __u64 max_age, __u32 flags);
348 int ofd_obd_disconnect(struct obd_export *exp);
351 obd_id ofd_seq_last_oid(struct ofd_seq *oseq);
352 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id);
353 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
354 struct ofd_seq *oseq);
355 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
356 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq);
357 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
359 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
360 struct obd_device *obd);
361 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
362 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
363 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
365 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
366 void ofd_seqs_free(const struct lu_env *env, struct ofd_device *ofd);
369 int ofd_preprw(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 *nr_local,
372 struct niobuf_local *lnb, struct obd_trans_info *oti,
373 struct lustre_capa *capa);
374 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
375 struct obdo *oa, int objcount, struct obd_ioobj *obj,
376 struct niobuf_remote *rnb, int npages,
377 struct niobuf_local *lnb, struct obd_trans_info *oti,
381 struct thandle *ofd_trans_create(const struct lu_env *env,
382 struct ofd_device *ofd);
383 int ofd_trans_start(const struct lu_env *env,
384 struct ofd_device *ofd, struct ofd_object *fo,
386 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
387 struct thandle *th, int rc);
388 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
393 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
394 void ofd_stats_counter_init(struct lprocfs_stats *stats);
396 static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
398 memset(lvars, 0, sizeof(*lvars));
400 static inline void ofd_stats_counter_init(struct lprocfs_stats *stats) {}
404 struct ofd_object *ofd_object_find(const struct lu_env *env,
405 struct ofd_device *ofd,
406 const struct lu_fid *fid);
407 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
408 struct ofd_device *ofd,
409 const struct lu_fid *fid,
410 struct lu_attr *attr);
411 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
412 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
413 obd_id id, struct ofd_seq *oseq, int nr, int sync);
415 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
416 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
417 struct lu_attr *la, struct filter_fid *ff);
418 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
419 __u64 start, __u64 end, struct lu_attr *la,
420 struct filter_fid *ff);
421 int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
422 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
424 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
425 struct lu_attr *la, int is_setattr);
428 struct ofd_object *ofd_object_find_exists(const struct lu_env *env,
429 struct ofd_device *ofd,
430 const struct lu_fid *fid)
432 struct ofd_object *fo;
434 fo = ofd_object_find(env, ofd, fid);
435 if (!IS_ERR(fo) && !ofd_object_exists(fo)) {
436 ofd_object_put(env, fo);
437 fo = ERR_PTR(-ENOENT);
443 #define OFD_GRANT_RATIO_SHIFT 8
444 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
446 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
449 static inline int ofd_grant_ratio_conv(int percentage)
451 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
454 static inline int ofd_grant_param_supp(struct obd_export *exp)
456 return !!(exp_connect_flags(exp) & OBD_CONNECT_GRANT_PARAM);
459 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
460 * That's 4KB=2^12 which is the biggest block size known to work whatever
461 * the client's page size is. */
462 #define COMPAT_BSIZE_SHIFT 12
463 static inline int ofd_grant_compat(struct obd_export *exp,
464 struct ofd_device *ofd)
466 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
467 * a block size > page size and consume PAGE_CACHE_SIZE of grant when
468 * dirtying a page regardless of the block size */
469 return !!(ofd_obd(ofd)->obd_self_export != exp &&
470 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
471 !ofd_grant_param_supp(exp));
474 static inline int ofd_grant_prohibit(struct obd_export *exp,
475 struct ofd_device *ofd)
477 /* When ofd_grant_compat_disable is set, we don't grant any space to
478 * clients not supporting OBD_CONNECT_GRANT_PARAM.
479 * Otherwise, space granted to such a client is inflated since it
480 * consumes PAGE_CACHE_SIZE of grant space per block */
481 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
484 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
485 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
486 obd_size want, bool conservative);
487 void ofd_grant_discard(struct obd_export *exp);
488 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
490 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
491 struct obdo *oa, struct niobuf_remote *rnb,
493 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
494 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
497 int ofd_fmd_init(void);
498 void ofd_fmd_exit(void);
499 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
500 const struct lu_fid *fid);
501 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
502 const struct lu_fid *fid);
503 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
504 void ofd_fmd_expire(struct obd_export *exp);
505 void ofd_fmd_cleanup(struct obd_export *exp);
507 void ofd_fmd_drop(struct obd_export *exp, const struct lu_fid *fid);
509 #define ofd_fmd_drop(exp, fid) do {} while (0)
513 int ofd_fid_set_index(const struct lu_env *env, struct ofd_device *ofd,
515 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd);
516 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd);
519 extern struct ldlm_valblock_ops ofd_lvbo;
522 int ofd_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp,
523 void *req_cookie, ldlm_mode_t mode, __u64 flags,
526 static inline struct ofd_thread_info *ofd_info(const struct lu_env *env)
528 struct ofd_thread_info *info;
530 lu_env_refill((void *)env);
531 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
536 static inline struct ofd_thread_info *ofd_info_init(const struct lu_env *env,
537 struct obd_export *exp)
539 struct ofd_thread_info *info;
541 info = ofd_info(env);
542 LASSERT(info->fti_exp == NULL);
543 LASSERT(info->fti_env == NULL);
544 LASSERT(info->fti_attr.la_valid == 0);
548 info->fti_pre_version = 0;
552 static inline struct ofd_thread_info *tsi2ofd_info(struct tgt_session_info *tsi)
554 struct ptlrpc_request *req = tgt_ses_req(tsi);
555 struct ofd_thread_info *info;
557 info = ofd_info(tsi->tsi_env);
558 LASSERT(info->fti_exp == NULL);
559 LASSERT(info->fti_env == NULL);
560 LASSERT(info->fti_attr.la_valid == 0);
562 info->fti_env = tsi->tsi_env;
563 info->fti_exp = tsi->tsi_exp;
565 info->fti_xid = req->rq_xid;
566 /** VBR: take versions from request */
567 if (req->rq_reqmsg != NULL &&
568 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
569 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
571 info->fti_pre_version = pre_version ? pre_version[0] : 0;
576 static inline void ofd_oti2info(struct ofd_thread_info *info,
577 struct obd_trans_info *oti)
579 info->fti_xid = oti->oti_xid;
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 oti->oti_pre_version = info->fti_pre_version;
590 /* sync on lock cancel is useless when we force a journal flush,
591 * and if we enable async journal commit, we should also turn on
592 * sync on lock cancel if it is not enabled already. */
593 static inline void ofd_slc_set(struct ofd_device *ofd)
595 if (ofd->ofd_syncjournal == 1)
596 ofd->ofd_lut.lut_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
597 else if (ofd->ofd_lut.lut_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
598 ofd->ofd_lut.lut_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
601 static inline void ofd_prepare_fidea(struct filter_fid *ff,
602 const struct obdo *oa)
604 /* packing fid and converting it to LE for storing into EA.
605 * Here ->o_stripe_idx should be filled by LOV and rest of
606 * fields - by client. */
607 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
608 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
609 /* XXX: we are ignoring o_parent_ver here, since this should
610 * be the same for all objects in this fileset. */
611 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
614 /* niobuf_remote has no rnb_ prefix in master */
615 #define rnb_offset offset
616 #define rnb_flags flags
618 /* the same for niobuf_local */
619 #define lnb_flags flags
622 #endif /* _OFD_INTERNAL_H */