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) 2011, 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 _OFD_INTERNAL_H
38 #define _OFD_INTERNAL_H
41 #include <obd_class.h>
42 #include <dt_object.h>
43 #include <lustre_fid.h>
45 #include <lustre_capa.h>
47 #define OFD_INIT_OBJID 0
48 #define OFD_ROCOMPAT_SUPP (0)
49 #define OFD_INCOMPAT_SUPP (OBD_INCOMPAT_GROUPS | OBD_INCOMPAT_OST | \
50 OBD_INCOMPAT_COMMON_LR)
51 #define OFD_MAX_GROUPS 256
53 /* Limit the returned fields marked valid to those that we actually might set */
54 #define OFD_VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
55 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
57 /* per-client-per-object persistent state (LRU) */
59 cfs_list_t fmd_list; /* linked to fed_mod_list */
60 struct lu_fid fmd_fid; /* FID being written to */
61 __u64 fmd_mactime_xid; /* xid highest {m,a,c}time setattr */
62 cfs_time_t fmd_expire; /* time when the fmd should expire */
63 int fmd_refcount; /* reference counter - list holds 1 */
66 #define OFD_FMD_MAX_NUM_DEFAULT 128
67 #define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * CFS_HZ)
70 LPROC_OFD_READ_BYTES = 0,
71 LPROC_OFD_WRITE_BYTES = 1,
77 LPROC_OFD_STATS_READ = 0,
78 LPROC_OFD_STATS_WRITE = 1,
79 LPROC_OFD_STATS_SETATTR = 2,
80 LPROC_OFD_STATS_PUNCH = 3,
81 LPROC_OFD_STATS_SYNC = 4,
85 static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
86 char *jobid, long amount)
88 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
89 (exp->exp_connect_flags & OBD_CONNECT_JOBSTATS))
90 lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
94 struct dt_device ofd_dt_dev;
95 struct dt_device *ofd_osd;
96 struct obd_export *ofd_osd_exp;
97 struct dt_device_param ofd_dt_conf;
98 /* DLM name-space for meta-data locks maintained by this server */
99 struct ldlm_namespace *ofd_namespace;
101 /* transaction callbacks */
102 struct dt_txn_callback ofd_txn_cb;
105 struct lu_target ofd_lut;
106 struct dt_object *ofd_last_group_file;
107 struct dt_object *ofd_health_check_file;
109 int ofd_subdir_count;
112 obd_id ofd_last_objids[OFD_MAX_GROUPS];
113 cfs_mutex_t ofd_create_locks[OFD_MAX_GROUPS];
114 struct dt_object *ofd_lastid_obj[OFD_MAX_GROUPS];
115 cfs_spinlock_t ofd_objid_lock;
116 unsigned long ofd_destroys_in_progress;
118 /* protect all statfs-related counters */
119 cfs_spinlock_t ofd_osfs_lock;
120 /* statfs optimization: we cache a bit */
121 struct obd_statfs ofd_osfs;
124 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
125 obd_size ofd_osfs_unstable;
127 /* counters used during statfs update, protected by ofd_osfs_lock.
128 * record when some statfs refresh are in progress */
129 int ofd_statfs_inflight;
130 /* track writes completed while statfs refresh is underway.
131 * tracking is only effective when ofd_statfs_inflight > 1 */
132 obd_size ofd_osfs_inflight;
134 /* grants: all values in bytes */
135 /* grant lock to protect all grant counters */
136 cfs_spinlock_t ofd_grant_lock;
137 /* total amount of dirty data reported by clients in incoming obdo */
138 obd_size ofd_tot_dirty;
139 /* sum of filesystem space granted to clients for async writes */
140 obd_size ofd_tot_granted;
141 /* grant used by I/Os in progress (between prepare and commit) */
142 obd_size ofd_tot_pending;
143 /* free space threshold over which we stop granting space to clients
144 * ofd_grant_ratio is stored as a fixed-point fraction using
145 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
148 /* number of clients using grants */
149 int ofd_tot_granted_clients;
151 /* ofd mod data: ofd_device wide values */
152 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
153 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
155 cfs_spinlock_t ofd_flags_lock;
156 unsigned long ofd_raid_degraded:1,
157 /* sync journal on writes */
159 /* sync on lock cancel */
160 ofd_sync_lock_cancel:2,
161 /* shall we grant space to clients not
162 * supporting OBD_CONNECT_GRANT_PARAM? */
163 ofd_grant_compat_disable:1;
166 static inline struct ofd_device *ofd_dev(struct lu_device *d)
168 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
171 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
173 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
176 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
178 return ofd_dev(exp->exp_obd->obd_lu_dev);
181 static inline char *ofd_name(struct ofd_device *ofd)
183 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
187 struct lu_object_header ofo_header;
188 struct dt_object ofo_obj;
192 static inline struct ofd_object *ofd_obj(struct lu_object *o)
194 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
197 static inline int ofd_object_exists(struct ofd_object *obj)
199 LASSERT(obj != NULL);
200 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
202 return lu_object_exists(&obj->ofo_obj.do_lu);
205 static inline struct dt_object *fo2dt(struct ofd_object *obj)
207 return &obj->ofo_obj;
210 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
212 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
214 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
217 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
219 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
222 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
223 const struct ofd_object *obj)
225 /* TODO: see mdd_object_capa() */
229 static inline void ofd_read_lock(const struct lu_env *env,
230 struct ofd_object *fo)
232 struct dt_object *next = ofd_object_child(fo);
234 next->do_ops->do_read_lock(env, next, 0);
237 static inline void ofd_read_unlock(const struct lu_env *env,
238 struct ofd_object *fo)
240 struct dt_object *next = ofd_object_child(fo);
242 next->do_ops->do_read_unlock(env, next);
245 static inline void ofd_write_lock(const struct lu_env *env,
246 struct ofd_object *fo)
248 struct dt_object *next = ofd_object_child(fo);
250 next->do_ops->do_write_lock(env, next, 0);
253 static inline void ofd_write_unlock(const struct lu_env *env,
254 struct ofd_object *fo)
256 struct dt_object *next = ofd_object_child(fo);
258 next->do_ops->do_write_unlock(env, next);
262 * Common data shared by obdofd-level handlers. This is allocated per-thread
263 * to reduce stack consumption.
265 struct ofd_thread_info {
266 const struct lu_env *fti_env;
268 struct obd_export *fti_exp;
271 __u64 fti_pre_version;
272 __u32 fti_has_trans:1, /* has txn already */
275 struct lu_fid fti_fid;
276 struct lu_attr fti_attr;
277 struct lu_attr fti_attr2;
278 struct ldlm_res_id fti_resid;
279 struct filter_fid fti_mds_fid;
280 struct filter_fid fti_mds_fid2;
281 struct ost_id fti_ostid;
282 struct ofd_object *fti_obj;
284 char name[64]; /* for ofd_init0() */
285 struct obd_statfs osfs; /* for obdofd_statfs() */
288 /* Ops object filename */
289 struct lu_name fti_name;
290 struct dt_object_format fti_dof;
291 struct lu_buf fti_buf;
294 /* Space used by the I/O, used by grant code */
295 unsigned long fti_used;
298 extern void target_recovery_fini(struct obd_device *obd);
299 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
302 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
303 int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
304 struct lustre_capa *capa, __u64 opc);
305 void ofd_free_capa_keys(struct ofd_device *ofd);
308 extern struct lu_context_key ofd_thread_key;
311 extern struct obd_ops ofd_obd_ops;
312 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
313 struct obd_statfs *osfs, __u64 max_age,
317 obd_id ofd_last_id(struct ofd_device *ofd, obd_seq seq);
318 void ofd_last_id_set(struct ofd_device *ofd, obd_id id, obd_seq seq);
319 int ofd_last_id_write(const struct lu_env *env, struct ofd_device *ofd,
321 int ofd_group_load(const struct lu_env *env, struct ofd_device *ofd, int);
322 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
323 struct obd_device *obd);
324 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
327 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
328 struct obdo *oa, int objcount, struct obd_ioobj *obj,
329 struct niobuf_remote *rnb, int *nr_local,
330 struct niobuf_local *lnb, struct obd_trans_info *oti,
331 struct lustre_capa *capa);
332 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
333 struct obdo *oa, int objcount, struct obd_ioobj *obj,
334 struct niobuf_remote *rnb, int npages,
335 struct niobuf_local *lnb, struct obd_trans_info *oti,
339 struct thandle *ofd_trans_create(const struct lu_env *env,
340 struct ofd_device *ofd);
341 int ofd_trans_start(const struct lu_env *env,
342 struct ofd_device *ofd, struct ofd_object *fo,
344 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
345 struct thandle *th, int rc);
346 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
351 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
352 int lproc_ofd_attach_seqstat(struct obd_device *dev);
353 extern struct file_operations ofd_per_nid_stats_fops;
354 void ofd_stats_counter_init(struct lprocfs_stats *stats);
356 static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
358 memset(lvars, 0, sizeof(*lvars));
360 static inline int lproc_ofd_attach_seqstat(struct obd_device *dev) {}
361 static inline void ofd_stats_counter_init(struct lprocfs_stats *stats) {}
365 struct ofd_object *ofd_object_find(const struct lu_env *env,
366 struct ofd_device *ofd,
367 const struct lu_fid *fid);
368 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
369 struct ofd_device *ofd,
370 const struct lu_fid *fid,
371 struct lu_attr *attr);
372 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
373 int ofd_precreate_object(const struct lu_env *env, struct ofd_device *ofd,
374 obd_id id, obd_seq seq);
376 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
377 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
378 struct lu_attr *la, struct filter_fid *ff);
379 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
380 __u64 start, __u64 end, struct lu_attr *la,
381 struct filter_fid *ff);
382 int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
383 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
385 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
386 struct lu_attr *la, int is_setattr);
389 #define OFD_GRANT_RATIO_SHIFT 8
390 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
392 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
395 static inline int ofd_grant_ratio_conv(int percentage)
397 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
400 static inline int ofd_grant_param_supp(struct obd_export *exp)
402 return !!(exp->exp_connect_flags & OBD_CONNECT_GRANT_PARAM);
405 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
406 * That's 4KB=2^12 which is the biggest block size known to work whatever
407 * the client's page size is. */
408 #define COMPAT_BSIZE_SHIFT 12
409 static inline int ofd_grant_compat(struct obd_export *exp,
410 struct ofd_device *ofd)
412 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
413 * a block size > page size and consume CFS_PAGE_SIZE of grant when
414 * dirtying a page regardless of the block size */
415 return !!(ofd_obd(ofd)->obd_self_export != exp &&
416 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
417 !ofd_grant_param_supp(exp));
420 static inline int ofd_grant_prohibit(struct obd_export *exp,
421 struct ofd_device *ofd)
423 /* When ofd_grant_compat_disable is set, we don't grant any space to
424 * clients not supporting OBD_CONNECT_GRANT_PARAM.
425 * Otherwise, space granted to such a client is inflated since it
426 * consumes CFS_PAGE_SIZE of grant space per block */
427 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
430 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
431 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
433 void ofd_grant_discard(struct obd_export *exp);
434 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
436 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
437 struct obdo *oa, struct niobuf_remote *rnb,
439 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
440 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
443 int ofd_fmd_init(void);
444 void ofd_fmd_exit(void);
445 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
447 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
449 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
450 void ofd_fmd_expire(struct obd_export *exp);
451 void ofd_fmd_cleanup(struct obd_export *exp);
453 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
455 #define ofd_fmd_drop(exp, fid) do {} while (0)
459 extern struct ldlm_valblock_ops ofd_lvbo;
462 int ofd_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp,
463 void *req_cookie, ldlm_mode_t mode, int flags,
466 static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
468 struct ofd_thread_info *info;
470 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
472 LASSERT(info->fti_env);
473 LASSERT(info->fti_env == env);
477 static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
478 struct obd_export *exp)
480 struct ofd_thread_info *info;
482 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
484 LASSERT(info->fti_exp == NULL);
485 LASSERT(info->fti_env == NULL);
486 LASSERT(info->fti_attr.la_valid == 0);
490 info->fti_pre_version = 0;
491 info->fti_transno = 0;
492 info->fti_has_trans = 0;
496 /* The same as osc_build_res_name() */
497 static inline void ofd_build_resid(const struct lu_fid *fid,
498 struct ldlm_res_id *resname)
500 if (fid_is_idif(fid)) {
501 /* get id/seq like ostid_idif_pack() does */
502 osc_build_res_name(fid_idif_id(fid_seq(fid), fid_oid(fid),
504 FID_SEQ_OST_MDT0, resname);
506 /* In the future, where OSTs have FID sequences allocated. */
507 fid_build_reg_res_name(fid, resname);
511 static inline void ofd_fid_from_resid(struct lu_fid *fid,
512 const struct ldlm_res_id *name)
514 /* if seq is FID_SEQ_OST_MDT0 then we have IDIF and resid was built
515 * using osc_build_res_name function. */
516 if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
519 ostid.oi_id = name->name[LUSTRE_RES_ID_SEQ_OFF];
520 ostid.oi_seq = name->name[LUSTRE_RES_ID_VER_OID_OFF];
521 fid_ostid_unpack(fid, &ostid, 0);
523 fid->f_seq = name->name[LUSTRE_RES_ID_SEQ_OFF];
524 fid->f_oid = (__u32)name->name[LUSTRE_RES_ID_VER_OID_OFF];
525 fid->f_ver = name->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32;
529 static inline void ofd_oti2info(struct ofd_thread_info *info,
530 struct obd_trans_info *oti)
532 info->fti_xid = oti->oti_xid;
533 info->fti_transno = oti->oti_transno;
534 info->fti_pre_version = oti->oti_pre_version;
537 static inline void ofd_info2oti(struct ofd_thread_info *info,
538 struct obd_trans_info *oti)
540 oti->oti_xid = info->fti_xid;
541 LASSERTF(ergo(oti->oti_transno > 0,
542 oti->oti_transno == info->fti_transno),
543 "Overwrite replay transno "LPX64" by "LPX64"\n",
544 oti->oti_transno, info->fti_transno);
545 oti->oti_transno = info->fti_transno;
546 oti->oti_pre_version = info->fti_pre_version;
549 /* sync on lock cancel is useless when we force a journal flush,
550 * and if we enable async journal commit, we should also turn on
551 * sync on lock cancel if it is not enabled already. */
552 static inline void ofd_slc_set(struct ofd_device *ofd)
554 if (ofd->ofd_syncjournal == 1)
555 ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
556 else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
557 ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
560 static inline void ofd_prepare_fidea(struct filter_fid *ff, struct obdo *oa)
562 if (!(oa->o_valid & OBD_MD_FLGROUP))
564 /* packing fid and converting it to LE for storing into EA.
565 * Here ->o_stripe_idx should be filled by LOV and rest of
566 * fields - by client. */
567 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
568 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
569 /* XXX: we are ignoring o_parent_ver here, since this should
570 * be the same for all objects in this fileset. */
571 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
572 ff->ff_objid = cpu_to_le64(oa->o_id);
573 ff->ff_seq = cpu_to_le64(oa->o_seq);
576 /* niobuf_remote has no rnb_ prefix in master */
577 #define rnb_offset offset
578 #define rnb_flags flags
580 /* the same for niobuf_local */
581 #define lnb_flags flags
584 #endif /* _OFD_INTERNAL_H */