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 <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_PRECREATE_BATCH_DEFAULT (FILTER_SUBDIR_COUNT * 4)
53 /* on small filesystems we should not precreate too many objects in
54 * a single transaction, otherwise we can overflow transactions */
55 #define OFD_PRECREATE_SMALL_FS (1024ULL * 1024 * 1024)
56 #define OFD_PRECREATE_BATCH_SMALL 8
58 /* Limit the returned fields marked valid to those that we actually might set */
59 #define OFD_VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
60 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
62 /* per-client-per-object persistent state (LRU) */
64 cfs_list_t fmd_list; /* linked to fed_mod_list */
65 struct lu_fid fmd_fid; /* FID being written to */
66 __u64 fmd_mactime_xid; /* xid highest {m,a,c}time setattr */
67 cfs_time_t fmd_expire; /* time when the fmd should expire */
68 int fmd_refcount; /* reference counter - list holds 1 */
71 #define OFD_FMD_MAX_NUM_DEFAULT 128
72 #define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * CFS_HZ)
75 LPROC_OFD_READ_BYTES = 0,
76 LPROC_OFD_WRITE_BYTES = 1,
82 LPROC_OFD_STATS_READ = 0,
83 LPROC_OFD_STATS_WRITE = 1,
84 LPROC_OFD_STATS_SETATTR = 2,
85 LPROC_OFD_STATS_PUNCH = 3,
86 LPROC_OFD_STATS_SYNC = 4,
90 static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
91 char *jobid, long amount)
93 if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
94 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
95 lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
97 if (exp->exp_nid_stats != NULL &&
98 exp->exp_nid_stats->nid_stats != NULL) {
99 if (opcode == LPROC_OFD_STATS_READ)
100 lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
101 LPROC_OFD_READ_BYTES, amount);
102 else if (opcode == LPROC_OFD_STATS_WRITE)
103 lprocfs_counter_add(exp->exp_nid_stats->nid_stats,
104 LPROC_OFD_WRITE_BYTES, amount);
111 spinlock_t os_last_oid_lock;
112 struct mutex os_create_lock;
113 cfs_atomic_t os_refc;
114 struct dt_object *os_lastid_obj;
115 unsigned long os_destroys_in_progress:1;
119 struct dt_device ofd_dt_dev;
120 struct dt_device *ofd_osd;
121 struct obd_export *ofd_osd_exp;
122 struct dt_device_param ofd_dt_conf;
123 /* DLM name-space for meta-data locks maintained by this server */
124 struct ldlm_namespace *ofd_namespace;
126 /* transaction callbacks */
127 struct dt_txn_callback ofd_txn_cb;
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 /* sync on lock cancel */
183 ofd_sync_lock_cancel:2,
184 /* shall we grant space to clients not
185 * supporting OBD_CONNECT_GRANT_PARAM? */
186 ofd_grant_compat_disable:1;
187 struct seq_server_site ofd_seq_site;
190 static inline struct ofd_device *ofd_dev(struct lu_device *d)
192 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
195 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
197 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
200 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
202 return ofd_dev(exp->exp_obd->obd_lu_dev);
205 static inline char *ofd_name(struct ofd_device *ofd)
207 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
211 struct lu_object_header ofo_header;
212 struct dt_object ofo_obj;
216 static inline struct ofd_object *ofd_obj(struct lu_object *o)
218 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
221 static inline int ofd_object_exists(struct ofd_object *obj)
223 LASSERT(obj != NULL);
224 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
226 return lu_object_exists(&obj->ofo_obj.do_lu);
229 static inline struct dt_object *fo2dt(struct ofd_object *obj)
231 return &obj->ofo_obj;
234 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
236 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
238 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
241 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
243 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
246 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
247 const struct ofd_object *obj)
249 /* TODO: see mdd_object_capa() */
253 static inline void ofd_read_lock(const struct lu_env *env,
254 struct ofd_object *fo)
256 struct dt_object *next = ofd_object_child(fo);
258 next->do_ops->do_read_lock(env, next, 0);
261 static inline void ofd_read_unlock(const struct lu_env *env,
262 struct ofd_object *fo)
264 struct dt_object *next = ofd_object_child(fo);
266 next->do_ops->do_read_unlock(env, next);
269 static inline void ofd_write_lock(const struct lu_env *env,
270 struct ofd_object *fo)
272 struct dt_object *next = ofd_object_child(fo);
274 next->do_ops->do_write_lock(env, next, 0);
277 static inline void ofd_write_unlock(const struct lu_env *env,
278 struct ofd_object *fo)
280 struct dt_object *next = ofd_object_child(fo);
282 next->do_ops->do_write_unlock(env, next);
286 * Common data shared by obdofd-level handlers. This is allocated per-thread
287 * to reduce stack consumption.
289 struct ofd_thread_info {
290 const struct lu_env *fti_env;
292 struct obd_export *fti_exp;
295 __u64 fti_pre_version;
296 __u32 fti_has_trans:1, /* has txn already */
299 struct lu_fid fti_fid;
300 struct lu_attr fti_attr;
301 struct lu_attr fti_attr2;
302 struct ldlm_res_id fti_resid;
303 struct filter_fid fti_mds_fid;
304 struct ost_id fti_ostid;
305 struct ofd_object *fti_obj;
307 char name[64]; /* for ofd_init0() */
308 struct obd_statfs osfs; /* for obdofd_statfs() */
311 /* Ops object filename */
312 struct lu_name fti_name;
313 struct dt_object_format fti_dof;
314 struct lu_buf fti_buf;
317 /* Space used by the I/O, used by grant code */
318 unsigned long fti_used;
319 struct ost_lvb fti_lvb;
322 extern void target_recovery_fini(struct obd_device *obd);
323 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
326 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
327 int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
328 struct lustre_capa *capa, __u64 opc);
329 void ofd_free_capa_keys(struct ofd_device *ofd);
332 extern struct lu_context_key ofd_thread_key;
333 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd);
336 extern struct obd_ops ofd_obd_ops;
337 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
338 struct obd_statfs *osfs, __u64 max_age,
342 obd_id ofd_seq_last_oid(struct ofd_seq *oseq);
343 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id);
344 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
345 struct ofd_seq *oseq);
346 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
347 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq);
348 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
350 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
351 struct obd_device *obd);
352 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
353 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
354 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
356 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
359 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
360 struct obdo *oa, int objcount, struct obd_ioobj *obj,
361 struct niobuf_remote *rnb, int *nr_local,
362 struct niobuf_local *lnb, struct obd_trans_info *oti,
363 struct lustre_capa *capa);
364 int ofd_commitrw(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 npages,
367 struct niobuf_local *lnb, struct obd_trans_info *oti,
371 struct thandle *ofd_trans_create(const struct lu_env *env,
372 struct ofd_device *ofd);
373 int ofd_trans_start(const struct lu_env *env,
374 struct ofd_device *ofd, struct ofd_object *fo,
376 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
377 struct thandle *th, int rc);
378 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
383 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
384 void ofd_stats_counter_init(struct lprocfs_stats *stats);
386 static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
388 memset(lvars, 0, sizeof(*lvars));
390 static inline void ofd_stats_counter_init(struct lprocfs_stats *stats) {}
394 struct ofd_object *ofd_object_find(const struct lu_env *env,
395 struct ofd_device *ofd,
396 const struct lu_fid *fid);
397 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
398 struct ofd_device *ofd,
399 const struct lu_fid *fid,
400 struct lu_attr *attr);
401 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
402 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
403 obd_id id, struct ofd_seq *oseq, int nr, int sync);
405 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
406 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
407 struct lu_attr *la, struct filter_fid *ff);
408 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
409 __u64 start, __u64 end, struct lu_attr *la,
410 struct filter_fid *ff);
411 int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
412 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
414 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
415 struct lu_attr *la, int is_setattr);
418 #define OFD_GRANT_RATIO_SHIFT 8
419 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
421 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
424 static inline int ofd_grant_ratio_conv(int percentage)
426 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
429 static inline int ofd_grant_param_supp(struct obd_export *exp)
431 return !!(exp_connect_flags(exp) & OBD_CONNECT_GRANT_PARAM);
434 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
435 * That's 4KB=2^12 which is the biggest block size known to work whatever
436 * the client's page size is. */
437 #define COMPAT_BSIZE_SHIFT 12
438 static inline int ofd_grant_compat(struct obd_export *exp,
439 struct ofd_device *ofd)
441 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
442 * a block size > page size and consume CFS_PAGE_SIZE of grant when
443 * dirtying a page regardless of the block size */
444 return !!(ofd_obd(ofd)->obd_self_export != exp &&
445 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
446 !ofd_grant_param_supp(exp));
449 static inline int ofd_grant_prohibit(struct obd_export *exp,
450 struct ofd_device *ofd)
452 /* When ofd_grant_compat_disable is set, we don't grant any space to
453 * clients not supporting OBD_CONNECT_GRANT_PARAM.
454 * Otherwise, space granted to such a client is inflated since it
455 * consumes CFS_PAGE_SIZE of grant space per block */
456 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
459 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
460 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
461 obd_size want, bool conservative);
462 void ofd_grant_discard(struct obd_export *exp);
463 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
465 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
466 struct obdo *oa, struct niobuf_remote *rnb,
468 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
469 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
472 int ofd_fmd_init(void);
473 void ofd_fmd_exit(void);
474 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
476 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
478 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
479 void ofd_fmd_expire(struct obd_export *exp);
480 void ofd_fmd_cleanup(struct obd_export *exp);
482 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
484 #define ofd_fmd_drop(exp, fid) do {} while (0)
488 int ofd_fid_set_index(const struct lu_env *env, struct ofd_device *ofd,
490 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd);
491 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd);
494 extern struct ldlm_valblock_ops ofd_lvbo;
497 int ofd_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp,
498 void *req_cookie, ldlm_mode_t mode, __u64 flags,
501 static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
503 struct ofd_thread_info *info;
505 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
507 LASSERT(info->fti_env);
508 LASSERT(info->fti_env == env);
512 static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
513 struct obd_export *exp)
515 struct ofd_thread_info *info;
517 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
519 LASSERT(info->fti_exp == NULL);
520 LASSERT(info->fti_env == NULL);
521 LASSERT(info->fti_attr.la_valid == 0);
525 info->fti_pre_version = 0;
526 info->fti_transno = 0;
527 info->fti_has_trans = 0;
531 static inline void ofd_oti2info(struct ofd_thread_info *info,
532 struct obd_trans_info *oti)
534 info->fti_xid = oti->oti_xid;
535 info->fti_transno = oti->oti_transno;
536 info->fti_pre_version = oti->oti_pre_version;
539 static inline void ofd_info2oti(struct ofd_thread_info *info,
540 struct obd_trans_info *oti)
542 oti->oti_xid = info->fti_xid;
543 LASSERTF(ergo(oti->oti_transno > 0,
544 oti->oti_transno == info->fti_transno),
545 "Overwrite replay transno "LPX64" by "LPX64"\n",
546 oti->oti_transno, info->fti_transno);
547 oti->oti_transno = info->fti_transno;
548 oti->oti_pre_version = info->fti_pre_version;
551 /* sync on lock cancel is useless when we force a journal flush,
552 * and if we enable async journal commit, we should also turn on
553 * sync on lock cancel if it is not enabled already. */
554 static inline void ofd_slc_set(struct ofd_device *ofd)
556 if (ofd->ofd_syncjournal == 1)
557 ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
558 else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
559 ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
562 static inline void ofd_prepare_fidea(struct filter_fid *ff, struct obdo *oa)
564 if (!(oa->o_valid & OBD_MD_FLGROUP))
565 ostid_set_seq_mdt0(&oa->o_oi);
566 /* packing fid and converting it to LE for storing into EA.
567 * Here ->o_stripe_idx should be filled by LOV and rest of
568 * fields - by client. */
569 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
570 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
571 /* XXX: we are ignoring o_parent_ver here, since this should
572 * be the same for all objects in this fileset. */
573 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
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 */