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, 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->exp_connect_flags & 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);
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_seq_count_file;
133 struct dt_object *ofd_health_check_file;
135 int ofd_subdir_count;
137 cfs_list_t ofd_seq_list;
138 rwlock_t ofd_seq_list_lock;
140 int ofd_precreate_batch;
141 spinlock_t ofd_batch_lock;
143 /* protect all statfs-related counters */
144 spinlock_t ofd_osfs_lock;
145 /* statfs optimization: we cache a bit */
146 struct obd_statfs ofd_osfs;
149 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
150 obd_size ofd_osfs_unstable;
152 /* counters used during statfs update, protected by ofd_osfs_lock.
153 * record when some statfs refresh are in progress */
154 int ofd_statfs_inflight;
155 /* track writes completed while statfs refresh is underway.
156 * tracking is only effective when ofd_statfs_inflight > 1 */
157 obd_size ofd_osfs_inflight;
159 /* grants: all values in bytes */
160 /* grant lock to protect all grant counters */
161 spinlock_t ofd_grant_lock;
162 /* total amount of dirty data reported by clients in incoming obdo */
163 obd_size ofd_tot_dirty;
164 /* sum of filesystem space granted to clients for async writes */
165 obd_size ofd_tot_granted;
166 /* grant used by I/Os in progress (between prepare and commit) */
167 obd_size ofd_tot_pending;
168 /* free space threshold over which we stop granting space to clients
169 * ofd_grant_ratio is stored as a fixed-point fraction using
170 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
173 /* number of clients using grants */
174 int ofd_tot_granted_clients;
176 /* ofd mod data: ofd_device wide values */
177 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
178 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
180 spinlock_t ofd_flags_lock;
181 unsigned long ofd_raid_degraded:1,
182 /* sync journal on writes */
184 /* sync on lock cancel */
185 ofd_sync_lock_cancel:2,
186 /* shall we grant space to clients not
187 * supporting OBD_CONNECT_GRANT_PARAM? */
188 ofd_grant_compat_disable:1;
191 static inline struct ofd_device *ofd_dev(struct lu_device *d)
193 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
196 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
198 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
201 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
203 return ofd_dev(exp->exp_obd->obd_lu_dev);
206 static inline char *ofd_name(struct ofd_device *ofd)
208 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
212 struct lu_object_header ofo_header;
213 struct dt_object ofo_obj;
217 static inline struct ofd_object *ofd_obj(struct lu_object *o)
219 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
222 static inline int ofd_object_exists(struct ofd_object *obj)
224 LASSERT(obj != NULL);
225 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
227 return lu_object_exists(&obj->ofo_obj.do_lu);
230 static inline struct dt_object *fo2dt(struct ofd_object *obj)
232 return &obj->ofo_obj;
235 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
237 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
239 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
242 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
244 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
247 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
248 const struct ofd_object *obj)
250 /* TODO: see mdd_object_capa() */
254 static inline void ofd_read_lock(const struct lu_env *env,
255 struct ofd_object *fo)
257 struct dt_object *next = ofd_object_child(fo);
259 next->do_ops->do_read_lock(env, next, 0);
262 static inline void ofd_read_unlock(const struct lu_env *env,
263 struct ofd_object *fo)
265 struct dt_object *next = ofd_object_child(fo);
267 next->do_ops->do_read_unlock(env, next);
270 static inline void ofd_write_lock(const struct lu_env *env,
271 struct ofd_object *fo)
273 struct dt_object *next = ofd_object_child(fo);
275 next->do_ops->do_write_lock(env, next, 0);
278 static inline void ofd_write_unlock(const struct lu_env *env,
279 struct ofd_object *fo)
281 struct dt_object *next = ofd_object_child(fo);
283 next->do_ops->do_write_unlock(env, next);
287 * Common data shared by obdofd-level handlers. This is allocated per-thread
288 * to reduce stack consumption.
290 struct ofd_thread_info {
291 const struct lu_env *fti_env;
293 struct obd_export *fti_exp;
296 __u64 fti_pre_version;
297 __u32 fti_has_trans:1, /* has txn already */
300 struct lu_fid fti_fid;
301 struct lu_attr fti_attr;
302 struct lu_attr fti_attr2;
303 struct ldlm_res_id fti_resid;
304 struct filter_fid fti_mds_fid;
305 struct filter_fid fti_mds_fid2;
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, struct lu_fid *fid, obd_seq seq,
330 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);
338 extern struct obd_ops ofd_obd_ops;
339 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
340 struct obd_statfs *osfs, __u64 max_age,
344 obd_id ofd_seq_last_oid(struct ofd_seq *oseq);
345 void ofd_seq_last_oid_set(struct ofd_seq *oseq, obd_id id);
346 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
347 struct ofd_seq *oseq);
348 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
349 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, obd_seq seq);
350 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
352 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
353 struct obd_device *obd);
354 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
355 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
356 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
358 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
361 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
362 struct obdo *oa, int objcount, struct obd_ioobj *obj,
363 struct niobuf_remote *rnb, int *nr_local,
364 struct niobuf_local *lnb, struct obd_trans_info *oti,
365 struct lustre_capa *capa);
366 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
367 struct obdo *oa, int objcount, struct obd_ioobj *obj,
368 struct niobuf_remote *rnb, int npages,
369 struct niobuf_local *lnb, struct obd_trans_info *oti,
373 struct thandle *ofd_trans_create(const struct lu_env *env,
374 struct ofd_device *ofd);
375 int ofd_trans_start(const struct lu_env *env,
376 struct ofd_device *ofd, struct ofd_object *fo,
378 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
379 struct thandle *th, int rc);
380 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
385 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
386 int lproc_ofd_attach_seqstat(struct obd_device *dev);
387 extern struct file_operations ofd_per_nid_stats_fops;
388 void ofd_stats_counter_init(struct lprocfs_stats *stats);
390 static void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars)
392 memset(lvars, 0, sizeof(*lvars));
394 static inline int lproc_ofd_attach_seqstat(struct obd_device *dev) {}
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);
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 #define OFD_GRANT_RATIO_SHIFT 8
424 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
426 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
429 static inline int ofd_grant_ratio_conv(int percentage)
431 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
434 static inline int ofd_grant_param_supp(struct obd_export *exp)
436 return !!(exp->exp_connect_flags & OBD_CONNECT_GRANT_PARAM);
439 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
440 * That's 4KB=2^12 which is the biggest block size known to work whatever
441 * the client's page size is. */
442 #define COMPAT_BSIZE_SHIFT 12
443 static inline int ofd_grant_compat(struct obd_export *exp,
444 struct ofd_device *ofd)
446 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
447 * a block size > page size and consume CFS_PAGE_SIZE of grant when
448 * dirtying a page regardless of the block size */
449 return !!(ofd_obd(ofd)->obd_self_export != exp &&
450 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
451 !ofd_grant_param_supp(exp));
454 static inline int ofd_grant_prohibit(struct obd_export *exp,
455 struct ofd_device *ofd)
457 /* When ofd_grant_compat_disable is set, we don't grant any space to
458 * clients not supporting OBD_CONNECT_GRANT_PARAM.
459 * Otherwise, space granted to such a client is inflated since it
460 * consumes CFS_PAGE_SIZE of grant space per block */
461 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
464 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
465 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
467 void ofd_grant_discard(struct obd_export *exp);
468 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
470 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
471 struct obdo *oa, struct niobuf_remote *rnb,
473 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
474 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
477 int ofd_fmd_init(void);
478 void ofd_fmd_exit(void);
479 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
481 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
483 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
484 void ofd_fmd_expire(struct obd_export *exp);
485 void ofd_fmd_cleanup(struct obd_export *exp);
487 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
489 #define ofd_fmd_drop(exp, fid) do {} while (0)
493 extern struct ldlm_valblock_ops ofd_lvbo;
496 int ofd_intent_policy(struct ldlm_namespace *ns, struct ldlm_lock **lockp,
497 void *req_cookie, ldlm_mode_t mode, __u64 flags,
500 static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
502 struct ofd_thread_info *info;
504 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
506 LASSERT(info->fti_env);
507 LASSERT(info->fti_env == env);
511 static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
512 struct obd_export *exp)
514 struct ofd_thread_info *info;
516 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
518 LASSERT(info->fti_exp == NULL);
519 LASSERT(info->fti_env == NULL);
520 LASSERT(info->fti_attr.la_valid == 0);
524 info->fti_pre_version = 0;
525 info->fti_transno = 0;
526 info->fti_has_trans = 0;
530 /* The same as osc_build_res_name() */
531 static inline void ofd_build_resid(const struct lu_fid *fid,
532 struct ldlm_res_id *resname)
534 if (fid_is_idif(fid)) {
535 /* get id/seq like ostid_idif_pack() does */
536 osc_build_res_name(fid_idif_id(fid_seq(fid), fid_oid(fid),
538 FID_SEQ_OST_MDT0, resname);
540 /* In the future, where OSTs have FID sequences allocated. */
541 fid_build_reg_res_name(fid, resname);
545 static inline void ofd_fid_from_resid(struct lu_fid *fid,
546 const struct ldlm_res_id *name)
548 /* if seq is FID_SEQ_OST_MDT0 then we have IDIF and resid was built
549 * using osc_build_res_name function. */
550 if (fid_seq_is_mdt0(name->name[LUSTRE_RES_ID_VER_OID_OFF])) {
553 ostid.oi_id = name->name[LUSTRE_RES_ID_SEQ_OFF];
554 ostid.oi_seq = name->name[LUSTRE_RES_ID_VER_OID_OFF];
555 fid_ostid_unpack(fid, &ostid, 0);
557 fid->f_seq = name->name[LUSTRE_RES_ID_SEQ_OFF];
558 fid->f_oid = (__u32)name->name[LUSTRE_RES_ID_VER_OID_OFF];
559 fid->f_ver = name->name[LUSTRE_RES_ID_VER_OID_OFF] >> 32;
563 static inline void ofd_oti2info(struct ofd_thread_info *info,
564 struct obd_trans_info *oti)
566 info->fti_xid = oti->oti_xid;
567 info->fti_transno = oti->oti_transno;
568 info->fti_pre_version = oti->oti_pre_version;
571 static inline void ofd_info2oti(struct ofd_thread_info *info,
572 struct obd_trans_info *oti)
574 oti->oti_xid = info->fti_xid;
575 LASSERTF(ergo(oti->oti_transno > 0,
576 oti->oti_transno == info->fti_transno),
577 "Overwrite replay transno "LPX64" by "LPX64"\n",
578 oti->oti_transno, info->fti_transno);
579 oti->oti_transno = info->fti_transno;
580 oti->oti_pre_version = info->fti_pre_version;
583 /* sync on lock cancel is useless when we force a journal flush,
584 * and if we enable async journal commit, we should also turn on
585 * sync on lock cancel if it is not enabled already. */
586 static inline void ofd_slc_set(struct ofd_device *ofd)
588 if (ofd->ofd_syncjournal == 1)
589 ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
590 else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
591 ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
594 static inline void ofd_prepare_fidea(struct filter_fid *ff, struct obdo *oa)
596 if (!(oa->o_valid & OBD_MD_FLGROUP))
598 /* packing fid and converting it to LE for storing into EA.
599 * Here ->o_stripe_idx should be filled by LOV and rest of
600 * fields - by client. */
601 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
602 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
603 /* XXX: we are ignoring o_parent_ver here, since this should
604 * be the same for all objects in this fileset. */
605 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
606 ff->ff_objid = cpu_to_le64(oa->o_id);
607 ff->ff_seq = cpu_to_le64(oa->o_seq);
610 /* niobuf_remote has no rnb_ prefix in master */
611 #define rnb_offset offset
612 #define rnb_flags flags
614 /* the same for niobuf_local */
615 #define lnb_flags flags
618 #endif /* _OFD_INTERNAL_H */