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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
32 #ifndef _OFD_INTERNAL_H
33 #define _OFD_INTERNAL_H
36 #include <obd_class.h>
37 #include <dt_object.h>
38 #include <md_object.h>
39 #include <lustre_fid.h>
41 #define OFD_INIT_OBJID 0
42 #define OFD_PRECREATE_BATCH_DEFAULT (OBJ_SUBDIR_COUNT * 4)
44 /* on small filesystems we should not precreate too many objects in
45 * a single transaction, otherwise we can overflow transactions */
46 #define OFD_PRECREATE_SMALL_FS (1024ULL * 1024 * 1024)
47 #define OFD_PRECREATE_BATCH_SMALL 8
49 /* Limit the returned fields marked valid to those that we actually might set */
50 #define OFD_VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
51 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
53 #define OFD_SOFT_SYNC_LIMIT_DEFAULT 16
56 * update atime if on-disk value older than client's one
57 * by OFD_ATIME_DIFF or more
59 #define OFD_DEF_ATIME_DIFF 0 /* disabled */
63 LPROC_OFD_STATS_READ_BYTES = 0,
64 LPROC_OFD_STATS_WRITE_BYTES,
66 LPROC_OFD_STATS_WRITE,
67 LPROC_OFD_STATS_GETATTR,
68 LPROC_OFD_STATS_SETATTR,
69 LPROC_OFD_STATS_PUNCH,
71 LPROC_OFD_STATS_DESTROY,
72 LPROC_OFD_STATS_CREATE,
73 LPROC_OFD_STATS_STATFS,
74 LPROC_OFD_STATS_GET_INFO,
75 LPROC_OFD_STATS_SET_INFO,
76 LPROC_OFD_STATS_QUOTACTL,
77 LPROC_OFD_STATS_PREALLOC,
81 static inline void ofd_counter_incr(struct obd_export *exp, int opcode,
82 char *jobid, long amount)
84 if (exp->exp_obd && exp->exp_obd->obd_stats)
85 lprocfs_counter_add(exp->exp_obd->obd_stats, opcode, amount);
87 if (exp->exp_obd && obd2obt(exp->exp_obd)->obt_jobstats.ojs_hash &&
88 (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS))
89 lprocfs_job_stats_log(exp->exp_obd, jobid, opcode, amount);
91 if (exp->exp_nid_stats != NULL &&
92 exp->exp_nid_stats->nid_stats != NULL) {
93 lprocfs_counter_add(exp->exp_nid_stats->nid_stats, opcode,
99 struct list_head os_list;
101 spinlock_t os_last_oid_lock;
102 struct mutex os_create_lock;
104 atomic_t os_precreate_in_progress;
105 struct dt_object *os_lastid_obj;
106 unsigned long os_destroys_in_progress:1,
111 struct dt_device ofd_dt_dev;
112 struct dt_device *ofd_osd;
113 struct obd_export *ofd_osd_exp;
114 /* DLM name-space for meta-data locks maintained by this server */
115 struct ldlm_namespace *ofd_namespace;
118 struct lu_target ofd_lut;
119 struct dt_object *ofd_health_check_file;
120 struct local_oid_storage *ofd_los;
122 __u64 ofd_inconsistency_self_detected;
123 __u64 ofd_inconsistency_self_repaired;
125 struct ofd_access_log *ofd_access_log;
126 unsigned int ofd_access_log_size;
127 unsigned int ofd_access_log_mask;
129 struct list_head ofd_seq_list;
130 rwlock_t ofd_seq_list_lock;
132 int ofd_precreate_batch;
133 spinlock_t ofd_batch_lock;
135 /* preferred BRW size, decided by storage type and capability */
137 spinlock_t ofd_flags_lock;
138 unsigned long ofd_raid_degraded:1,
139 /* sync journal on writes */
141 /* Protected by ofd_lastid_rwsem. */
142 ofd_lastid_rebuilding:1,
143 ofd_record_fid_accessed:1,
144 ofd_lfsck_verify_pfid:1,
146 struct seq_server_site ofd_seq_site;
147 /* the limit of SOFT_SYNC RPCs that will trigger a soft sync */
148 unsigned int ofd_soft_sync_limit;
149 /* Protect ::ofd_lastid_rebuilding */
150 struct rw_semaphore ofd_lastid_rwsem;
151 __u64 ofd_lastid_gen;
152 struct task_struct *ofd_inconsistency_task;
153 struct list_head ofd_inconsistency_list;
154 spinlock_t ofd_inconsistency_lock;
155 /* Backwards compatibility */
156 struct attribute *ofd_read_cache_enable;
157 struct attribute *ofd_read_cache_max_filesize;
158 struct attribute *ofd_write_cache_enable;
159 time64_t ofd_atime_diff;
162 static inline struct ofd_device *ofd_dev(struct lu_device *d)
164 return container_of_safe(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
167 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
169 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
172 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
174 return ofd_dev(exp->exp_obd->obd_lu_dev);
177 static inline char *ofd_name(struct ofd_device *ofd)
179 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
183 * for compatibility, filter_fid could occupy more space in newer version and
184 * downgraded Lustre would fail reading it with -ERANGE, so it can read it
185 * again with more space to hold it.
187 #define FILTER_FID_EXTRA_SIZE 32
190 struct lu_object_header ofo_header;
191 struct dt_object ofo_obj;
192 struct filter_fid ofo_ff;
193 time64_t ofo_atime_ondisk;
194 unsigned int ofo_pfid_checking:1,
198 static inline struct ofd_object *ofd_obj(struct lu_object *o)
200 return container_of_safe(o, struct ofd_object, ofo_obj.do_lu);
203 static inline int ofd_object_exists(struct ofd_object *obj)
205 LASSERT(obj != NULL);
206 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
208 return lu_object_exists(&obj->ofo_obj.do_lu);
211 static inline struct dt_object *fo2dt(struct ofd_object *obj)
213 return &obj->ofo_obj;
216 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
218 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
220 return container_of(lu_object_next(lu), struct dt_object, do_lu);
223 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
225 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
228 static inline void ofd_read_lock(const struct lu_env *env,
229 struct ofd_object *fo)
231 struct dt_object *next = ofd_object_child(fo);
233 next->do_ops->do_read_lock(env, next, 0);
236 static inline void ofd_read_unlock(const struct lu_env *env,
237 struct ofd_object *fo)
239 struct dt_object *next = ofd_object_child(fo);
241 next->do_ops->do_read_unlock(env, next);
244 static inline void ofd_write_lock(const struct lu_env *env,
245 struct ofd_object *fo)
247 struct dt_object *next = ofd_object_child(fo);
249 next->do_ops->do_write_lock(env, next, 0);
252 static inline void ofd_write_unlock(const struct lu_env *env,
253 struct ofd_object *fo)
255 struct dt_object *next = ofd_object_child(fo);
257 next->do_ops->do_write_unlock(env, next);
261 * Common data shared by obdofd-level handlers. This is allocated per-thread
262 * to reduce stack consumption.
264 struct ofd_thread_info {
265 const struct lu_env *fti_env;
267 struct obd_export *fti_exp;
269 __u64 fti_pre_version;
271 struct lu_fid fti_fid;
272 struct lu_attr fti_attr;
273 struct lu_attr fti_attr2;
274 struct ldlm_res_id fti_resid;
275 struct filter_fid fti_mds_fid;
276 struct ost_id fti_ostid;
277 struct ofd_object *fti_obj;
279 char name[64]; /* for ofd_init0() */
280 struct obd_statfs osfs; /* for obdofd_statfs() */
283 /* Ops object filename */
284 struct lu_name fti_name;
285 struct dt_object_format fti_dof;
286 struct lu_buf fti_buf;
289 struct ost_lvb fti_lvb;
291 struct lfsck_req_local fti_lrl;
292 struct obd_connect_data fti_ocd;
296 extern void target_recovery_fini(struct obd_device *obd);
297 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
299 /* ofd_access_log.c */
300 bool ofd_access_log_size_is_valid(unsigned int size);
301 int ofd_access_log_module_init(void);
302 void ofd_access_log_module_exit(void);
304 struct ofd_access_log;
305 struct ofd_access_log *ofd_access_log_create(const char *ofd_name, size_t size);
306 void ofd_access_log_delete(struct ofd_access_log *oal);
307 void ofd_access(const struct lu_env *env, struct ofd_device *m,
308 const struct lu_fid *parent_fid, __u64 begin, __u64 end,
309 unsigned int size, unsigned int segment_count, int rw);
312 extern struct lu_context_key ofd_thread_key;
313 int ofd_postrecov(const struct lu_env *env, struct ofd_device *ofd);
314 int ofd_fiemap_get(const struct lu_env *env, struct ofd_device *ofd,
315 struct lu_fid *fid, struct fiemap *fiemap);
318 extern const struct obd_ops ofd_obd_ops;
319 int ofd_destroy_by_fid(const struct lu_env *env, struct ofd_device *ofd,
320 const struct lu_fid *fid, int orphan);
321 int ofd_statfs(const struct lu_env *env, struct obd_export *exp,
322 struct obd_statfs *osfs, time64_t max_age, __u32 flags);
323 int ofd_obd_disconnect(struct obd_export *exp);
326 u64 ofd_seq_last_oid(struct ofd_seq *oseq);
327 void ofd_seq_last_oid_set(struct ofd_seq *oseq, u64 id);
328 int ofd_seq_last_oid_write(const struct lu_env *env, struct ofd_device *ofd,
329 struct ofd_seq *oseq);
330 int ofd_seqs_init(const struct lu_env *env, struct ofd_device *ofd);
331 struct ofd_seq *ofd_seq_get(struct ofd_device *ofd, u64 seq);
332 void ofd_seq_put(const struct lu_env *env, struct ofd_seq *oseq);
334 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
335 struct obd_device *obd);
336 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
337 int ofd_precreate_batch(struct ofd_device *ofd, int batch);
338 struct ofd_seq *ofd_seq_load(const struct lu_env *env, struct ofd_device *ofd,
340 void ofd_seqs_fini(const struct lu_env *env, struct ofd_device *ofd);
341 void ofd_seqs_free(const struct lu_env *env, struct ofd_device *ofd);
344 int ofd_start_inconsistency_verification_thread(struct ofd_device *ofd);
345 int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd);
346 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
348 int ofd_verify_layout_version(const struct lu_env *env,
349 struct ofd_object *fo, const struct obdo *oa);
350 int ofd_preprw(const struct lu_env *env,int cmd, struct obd_export *exp,
351 struct obdo *oa, int objcount, struct obd_ioobj *obj,
352 struct niobuf_remote *rnb, int *nr_local,
353 struct niobuf_local *lnb);
354 int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
355 struct obdo *oa, int objcount, struct obd_ioobj *obj,
356 struct niobuf_remote *rnb, int npages,
357 struct niobuf_local *lnb, int old_rc, int nob,
361 struct thandle *ofd_trans_create(const struct lu_env *env,
362 struct ofd_device *ofd);
363 int ofd_trans_start(const struct lu_env *env,
364 struct ofd_device *ofd, struct ofd_object *fo,
366 int ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
367 struct thandle *th, int rc);
368 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
372 int ofd_tunables_init(struct ofd_device *ofd);
373 #ifdef CONFIG_PROC_FS
374 void ofd_stats_counter_init(struct lprocfs_stats *stats, unsigned int offset,
375 enum lprocfs_counter_config cntr_umask);
378 ofd_stats_counter_init(struct lprocfs_stats *stats,
380 enum lprocfs_counter_config cntr_umask) {}
384 struct ofd_object *ofd_object_find(const struct lu_env *env,
385 struct ofd_device *ofd,
386 const struct lu_fid *fid);
387 int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo);
388 int ofd_object_ff_update(const struct lu_env *env, struct ofd_object *fo,
389 const struct obdo *oa, struct filter_fid *ff);
390 int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
391 u64 id, struct ofd_seq *oseq, int nr, int sync,
394 static inline void ofd_object_put(const struct lu_env *env,
395 struct ofd_object *fo)
397 dt_object_put(env, &fo->ofo_obj);
399 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
400 struct lu_attr *la, struct obdo *oa);
401 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
402 __u64 start, __u64 end, struct lu_attr *la,
404 int ofd_object_fallocate(const struct lu_env *env, struct ofd_object *fo,
405 __u64 start, __u64 end, int mode, struct lu_attr *la,
407 int ofd_destroy(const struct lu_env *, struct ofd_object *, int);
408 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
410 int ofd_attr_handle_id(const struct lu_env *env, struct ofd_object *fo,
411 struct lu_attr *la, int is_setattr);
414 struct ofd_object *ofd_object_find_exists(const struct lu_env *env,
415 struct ofd_device *ofd,
416 const struct lu_fid *fid)
418 struct ofd_object *fo;
420 fo = ofd_object_find(env, ofd, fid);
421 if (!IS_ERR(fo) && !ofd_object_exists(fo)) {
422 ofd_object_put(env, fo);
423 fo = ERR_PTR(-ENOENT);
429 int ofd_fid_set_index(const struct lu_env *env, struct ofd_device *ofd,
431 int ofd_fid_init(const struct lu_env *env, struct ofd_device *ofd);
432 int ofd_fid_fini(const struct lu_env *env, struct ofd_device *ofd);
435 extern struct ldlm_valblock_ops ofd_lvbo;
438 extern struct kmem_cache *ldlm_glimpse_work_kmem;
440 int ofd_intent_policy(const struct lu_env *env, struct ldlm_namespace *ns,
441 struct ldlm_lock **lockp, void *req_cookie,
442 enum ldlm_mode mode, __u64 flags, void *data);
444 static inline struct ofd_thread_info *ofd_info(const struct lu_env *env)
446 return lu_env_info(env, &ofd_thread_key);
449 static inline struct ofd_thread_info *ofd_info_init(const struct lu_env *env,
450 struct obd_export *exp)
452 struct ofd_thread_info *info;
454 info = ofd_info(env);
455 LASSERT(info->fti_exp == NULL);
456 LASSERT(info->fti_env == NULL);
457 LASSERT(info->fti_attr.la_valid == 0);
461 info->fti_pre_version = 0;
465 static inline struct ofd_thread_info *tsi2ofd_info(struct tgt_session_info *tsi)
467 struct ptlrpc_request *req = tgt_ses_req(tsi);
468 struct ofd_thread_info *info;
470 info = ofd_info(tsi->tsi_env);
471 LASSERT(info->fti_exp == NULL);
472 LASSERT(info->fti_env == NULL);
473 LASSERT(info->fti_attr.la_valid == 0);
475 info->fti_env = tsi->tsi_env;
476 info->fti_exp = tsi->tsi_exp;
478 info->fti_xid = req->rq_xid;
479 /** VBR: take versions from request */
480 if (req->rq_reqmsg != NULL &&
481 lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) {
482 __u64 *pre_version = lustre_msg_get_versions(req->rq_reqmsg);
484 info->fti_pre_version = pre_version ? pre_version[0] : 0;
489 /* sync on lock cancel is useless when we force a journal flush,
490 * and if we enable async journal commit, we should also turn on
491 * sync on lock cancel if it is not enabled already. */
492 static inline void ofd_slc_set(struct ofd_device *ofd)
494 if (ofd->ofd_sync_journal == 1)
495 ofd->ofd_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
496 else if (ofd->ofd_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
497 ofd->ofd_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_ALWAYS;
500 static inline int ofd_validate_seq(struct obd_export *exp, __u64 seq)
502 struct filter_export_data *fed = &exp->exp_filter_data;
504 if (unlikely(seq == FID_SEQ_OST_MDT0 && fed->fed_group != 0)) {
505 /* IDIF request only operates on MDT0 group */
506 CERROR("%s: Invalid sequence %#llx for group %u\n",
507 exp->exp_obd->obd_name, seq, fed->fed_group);
515 * whether the requestion IO contains older layout version than that on the
518 static inline bool ofd_layout_version_less(__u32 req_version,
519 __u32 ondisk_version)
521 __u32 req = req_version & ~LU_LAYOUT_RESYNC;
522 __u32 ondisk = ondisk_version & ~LU_LAYOUT_RESYNC;
525 * request layout version could be circularly increased to the samllest
526 * value, in that case @req < @ondisk but @req does not have the high
527 * end bit set while @ondisk does.
529 return (req < ondisk) &&
530 ((req & LU_LAYOUT_HIGEN) == (ondisk & LU_LAYOUT_HIGEN));
533 static inline int ofd_seq_is_exhausted(struct ofd_device *ofd,
534 const struct obdo *oa)
536 struct seq_server_site *ss = &ofd->ofd_seq_site;
537 __u64 seq_width = ss->ss_client_seq->lcs_width;
538 __u64 seq = ostid_seq(&oa->o_oi);
539 __u64 oid = ostid_id(&oa->o_oi);
541 if (fid_seq_is_norm(seq))
542 return oid >= min(seq_width, OBIF_MAX_OID);
543 if (fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq))
544 return oid >= min(seq_width, IDIF_MAX_OID);
545 CERROR("%s : invalid o_seq "DOSTID"\n",
546 ofd_name(ofd), POSTID(&oa->o_oi));
550 #endif /* _OFD_INTERNAL_H */