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 /* per-client-per-object persistent state (LRU) */
55 cfs_list_t fmd_list; /* linked to fed_mod_list */
56 struct lu_fid fmd_fid; /* FID being written to */
57 __u64 fmd_mactime_xid; /* xid highest {m,a,c}time setattr */
58 cfs_time_t fmd_expire; /* time when the fmd should expire */
59 int fmd_refcount; /* reference counter - list holds 1 */
62 #define OFD_FMD_MAX_NUM_DEFAULT 128
63 #define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * CFS_HZ)
66 LPROC_OFD_READ_BYTES = 0,
67 LPROC_OFD_WRITE_BYTES = 1,
72 struct dt_device ofd_dt_dev;
73 struct dt_device *ofd_osd;
74 struct dt_device_param ofd_dt_conf;
75 /* DLM name-space for meta-data locks maintained by this server */
76 struct ldlm_namespace *ofd_namespace;
78 /* transaction callbacks */
79 struct dt_txn_callback ofd_txn_cb;
82 struct lu_target ofd_lut;
83 struct dt_object *ofd_last_group_file;
84 struct dt_object *ofd_health_check_file;
89 obd_id ofd_last_objids[OFD_MAX_GROUPS];
90 cfs_mutex_t ofd_create_locks[OFD_MAX_GROUPS];
91 struct dt_object *ofd_lastid_obj[OFD_MAX_GROUPS];
92 cfs_spinlock_t ofd_objid_lock;
94 /* protect all statfs-related counters */
95 cfs_spinlock_t ofd_osfs_lock;
96 /* statfs optimization: we cache a bit */
97 struct obd_statfs ofd_osfs;
100 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
101 obd_size ofd_osfs_unstable;
103 /* counters used during statfs update, protected by ofd_osfs_lock.
104 * record when some statfs refresh are in progress */
105 int ofd_statfs_inflight;
106 /* track writes completed while statfs refresh is underway.
107 * tracking is only effective when ofd_statfs_inflight > 1 */
108 obd_size ofd_osfs_inflight;
110 /* grants: all values in bytes */
111 /* grant lock to protect all grant counters */
112 cfs_spinlock_t ofd_grant_lock;
113 /* total amount of dirty data reported by clients in incoming obdo */
114 obd_size ofd_tot_dirty;
115 /* sum of filesystem space granted to clients for async writes */
116 obd_size ofd_tot_granted;
117 /* grant used by I/Os in progress (between prepare and commit) */
118 obd_size ofd_tot_pending;
119 /* free space threshold over which we stop granting space to clients
120 * ofd_grant_ratio is stored as a fixed-point fraction using
121 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
124 /* number of clients using grants */
125 int ofd_tot_granted_clients;
127 /* ofd mod data: ofd_device wide values */
128 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
129 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
131 cfs_spinlock_t ofd_flags_lock;
132 unsigned long ofd_raid_degraded:1,
133 /* sync journal on writes */
135 /* sync on lock cancel */
136 ofd_sync_lock_cancel:2,
137 /* shall we grant space to clients not
138 * supporting OBD_CONNECT_GRANT_PARAM? */
139 ofd_grant_compat_disable:1;
141 struct lu_site ofd_site;
144 static inline struct ofd_device *ofd_dev(struct lu_device *d)
146 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
149 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
151 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
154 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
156 return ofd_dev(exp->exp_obd->obd_lu_dev);
159 static inline char *ofd_name(struct ofd_device *ofd)
161 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
165 struct lu_object_header ofo_header;
166 struct dt_object ofo_obj;
170 static inline struct ofd_object *ofd_obj(struct lu_object *o)
172 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
175 static inline int ofd_object_exists(struct ofd_object *obj)
177 LASSERT(obj != NULL);
178 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
180 return lu_object_exists(&obj->ofo_obj.do_lu);
183 static inline struct dt_object *fo2dt(struct ofd_object *obj)
185 return &obj->ofo_obj;
188 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
190 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
192 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
195 static inline struct ofd_device *ofd_obj2dev(const struct ofd_object *fo)
197 return ofd_dev(fo->ofo_obj.do_lu.lo_dev);
200 static inline struct lustre_capa *ofd_object_capa(const struct lu_env *env,
201 const struct ofd_object *obj)
203 /* TODO: see mdd_object_capa() */
207 static inline void ofd_read_lock(const struct lu_env *env,
208 struct ofd_object *fo)
210 struct dt_object *next = ofd_object_child(fo);
212 next->do_ops->do_read_lock(env, next, 0);
215 static inline void ofd_read_unlock(const struct lu_env *env,
216 struct ofd_object *fo)
218 struct dt_object *next = ofd_object_child(fo);
220 next->do_ops->do_read_unlock(env, next);
223 static inline void ofd_write_lock(const struct lu_env *env,
224 struct ofd_object *fo)
226 struct dt_object *next = ofd_object_child(fo);
228 next->do_ops->do_write_lock(env, next, 0);
231 static inline void ofd_write_unlock(const struct lu_env *env,
232 struct ofd_object *fo)
234 struct dt_object *next = ofd_object_child(fo);
236 next->do_ops->do_write_unlock(env, next);
240 * Common data shared by obdofd-level handlers. This is allocated per-thread
241 * to reduce stack consumption.
243 struct ofd_thread_info {
244 const struct lu_env *fti_env;
246 struct obd_export *fti_exp;
249 __u64 fti_pre_version;
250 __u32 fti_has_trans:1, /* has txn already */
253 struct lu_fid fti_fid;
254 struct lu_attr fti_attr;
255 struct lu_attr fti_attr2;
256 struct filter_fid fti_mds_fid2;
257 struct ost_id fti_ostid;
258 struct ofd_object *fti_obj;
260 char name[64]; /* for ofd_init0() */
261 struct obd_statfs osfs; /* for obdofd_statfs() */
264 /* Ops object filename */
265 struct lu_name fti_name;
266 struct dt_object_format fti_dof;
267 struct lu_buf fti_buf;
270 /* Space used by the I/O, used by grant code */
271 unsigned long fti_used;
274 extern void target_recovery_fini(struct obd_device *obd);
275 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
278 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
279 int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
280 struct lustre_capa *capa, __u64 opc);
281 void ofd_free_capa_keys(struct ofd_device *ofd);
284 extern struct lu_context_key ofd_thread_key;
287 extern struct obd_ops ofd_obd_ops;
288 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
289 struct obd_statfs *osfs, __u64 max_age,
293 obd_id ofd_last_id(struct ofd_device *ofd, obd_seq seq);
294 void ofd_last_id_set(struct ofd_device *ofd, obd_id id, obd_seq seq);
295 int ofd_group_load(const struct lu_env *env, struct ofd_device *ofd, int);
296 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
297 struct obd_device *obd);
298 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
301 struct thandle *ofd_trans_create(const struct lu_env *env,
302 struct ofd_device *ofd);
303 int ofd_trans_start(const struct lu_env *env,
304 struct ofd_device *ofd, struct ofd_object *fo,
306 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
307 struct thandle *th, int rc);
308 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
312 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
313 int lproc_ofd_attach_seqstat(struct obd_device *dev);
314 extern struct file_operations ofd_per_nid_stats_fops;
317 struct ofd_object *ofd_object_find(const struct lu_env *env,
318 struct ofd_device *ofd,
319 const struct lu_fid *fid);
320 struct ofd_object *ofd_object_find_or_create(const struct lu_env *env,
321 struct ofd_device *ofd,
322 const struct lu_fid *fid,
323 struct lu_attr *attr);
324 int ofd_object_ff_check(const struct lu_env *env, struct ofd_object *fo);
325 int ofd_precreate_object(const struct lu_env *env, struct ofd_device *ofd,
326 obd_id id, obd_seq seq);
328 void ofd_object_put(const struct lu_env *env, struct ofd_object *fo);
329 int ofd_attr_set(const struct lu_env *env, struct ofd_object *fo,
330 struct lu_attr *la, struct filter_fid *ff);
331 int ofd_object_punch(const struct lu_env *env, struct ofd_object *fo,
332 __u64 start, __u64 end, struct lu_attr *la,
333 struct filter_fid *ff);
334 int ofd_object_destroy(const struct lu_env *, struct ofd_object *, int);
335 int ofd_attr_get(const struct lu_env *env, struct ofd_object *fo,
337 int ofd_attr_handle_ugid(const struct lu_env *env, struct ofd_object *fo,
338 struct lu_attr *la, int is_setattr);
341 #define OFD_GRANT_RATIO_SHIFT 8
342 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
344 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
347 static inline int ofd_grant_ratio_conv(int percentage)
349 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
352 static inline int ofd_grant_param_supp(struct obd_export *exp)
354 return !!(exp->exp_connect_flags & OBD_CONNECT_GRANT_PARAM);
357 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
358 * That's 4KB=2^12 which is the biggest block size known to work whatever
359 * the client's page size is. */
360 #define COMPAT_BSIZE_SHIFT 12
361 static inline int ofd_grant_compat(struct obd_export *exp,
362 struct ofd_device *ofd)
364 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
365 * a block size > page size and consume CFS_PAGE_SIZE of grant when
366 * dirtying a page regardless of the block size */
367 return !!(ofd_obd(ofd)->obd_self_export != exp &&
368 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
369 !ofd_grant_param_supp(exp));
372 static inline int ofd_grant_prohibit(struct obd_export *exp,
373 struct ofd_device *ofd)
375 /* When ofd_grant_compat_disable is set, we don't grant any space to
376 * clients not supporting OBD_CONNECT_GRANT_PARAM.
377 * Otherwise, space granted to such a client is inflated since it
378 * consumes CFS_PAGE_SIZE of grant space per block */
379 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
382 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
383 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
385 void ofd_grant_discard(struct obd_export *exp);
386 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
388 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
389 struct obdo *oa, struct niobuf_remote *rnb,
391 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
392 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
395 int ofd_fmd_init(void);
396 void ofd_fmd_exit(void);
397 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
399 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
401 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
402 void ofd_fmd_expire(struct obd_export *exp);
403 void ofd_fmd_cleanup(struct obd_export *exp);
405 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
407 #define ofd_fmd_drop(exp, fid) do {} while (0)
410 static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
412 struct ofd_thread_info *info;
414 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
416 LASSERT(info->fti_env);
417 LASSERT(info->fti_env == env);
421 static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
422 struct obd_export *exp)
424 struct ofd_thread_info *info;
426 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
428 LASSERT(info->fti_exp == NULL);
429 LASSERT(info->fti_env == NULL);
430 LASSERT(info->fti_attr.la_valid == 0);
434 info->fti_pre_version = 0;
435 info->fti_transno = 0;
436 info->fti_has_trans = 0;
440 /* sync on lock cancel is useless when we force a journal flush,
441 * and if we enable async journal commit, we should also turn on
442 * sync on lock cancel if it is not enabled already. */
443 static inline void ofd_slc_set(struct ofd_device *ofd)
445 if (ofd->ofd_syncjournal == 1)
446 ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
447 else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
448 ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
451 static inline void ofd_prepare_fidea(struct filter_fid *ff, struct obdo *oa)
453 if (!(oa->o_valid & OBD_MD_FLGROUP))
455 /* packing fid and converting it to LE for storing into EA.
456 * Here ->o_stripe_idx should be filled by LOV and rest of
457 * fields - by client. */
458 ff->ff_parent.f_seq = cpu_to_le64(oa->o_parent_seq);
459 ff->ff_parent.f_oid = cpu_to_le32(oa->o_parent_oid);
460 /* XXX: we are ignoring o_parent_ver here, since this should
461 * be the same for all objects in this fileset. */
462 ff->ff_parent.f_ver = cpu_to_le32(oa->o_stripe_idx);
463 ff->ff_objid = cpu_to_le64(oa->o_id);
464 ff->ff_seq = cpu_to_le64(oa->o_seq);
467 /* niobuf_local has no rnb_ prefix in master */
468 #define rnb_offset offset
469 #define rnb_flags flags
472 #endif /* _OFD_INTERNAL_H */