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>
44 #include <lustre_capa.h>
46 #define OFD_INIT_OBJID 0
47 #define OFD_ROCOMPAT_SUPP (0)
48 #define OFD_INCOMPAT_SUPP (OBD_INCOMPAT_GROUPS | OBD_INCOMPAT_OST | \
49 OBD_INCOMPAT_COMMON_LR)
50 #define OFD_MAX_GROUPS 256
52 /* per-client-per-object persistent state (LRU) */
54 cfs_list_t fmd_list; /* linked to fed_mod_list */
55 struct lu_fid fmd_fid; /* FID being written to */
56 __u64 fmd_mactime_xid; /* xid highest {m,a,c}time setattr */
57 cfs_time_t fmd_expire; /* time when the fmd should expire */
58 int fmd_refcount; /* reference counter - list holds 1 */
61 #define OFD_FMD_MAX_NUM_DEFAULT 128
62 #define OFD_FMD_MAX_AGE_DEFAULT ((obd_timeout + 10) * CFS_HZ)
65 LPROC_OFD_READ_BYTES = 0,
66 LPROC_OFD_WRITE_BYTES = 1,
71 struct dt_device ofd_dt_dev;
72 struct dt_device *ofd_osd;
73 struct dt_device_param ofd_dt_conf;
74 /* DLM name-space for meta-data locks maintained by this server */
75 struct ldlm_namespace *ofd_namespace;
77 /* transaction callbacks */
78 struct dt_txn_callback ofd_txn_cb;
81 struct lu_target ofd_lut;
82 struct dt_object *ofd_last_group_file;
83 struct dt_object *ofd_health_check_file;
88 obd_id ofd_last_objids[OFD_MAX_GROUPS];
89 cfs_mutex_t ofd_create_locks[OFD_MAX_GROUPS];
90 struct dt_object *ofd_lastid_obj[OFD_MAX_GROUPS];
91 cfs_spinlock_t ofd_objid_lock;
93 /* protect all statfs-related counters */
94 cfs_spinlock_t ofd_osfs_lock;
95 /* statfs optimization: we cache a bit */
96 struct obd_statfs ofd_osfs;
99 /* writes which might be be accounted twice in ofd_osfs.os_bavail */
100 obd_size ofd_osfs_unstable;
102 /* counters used during statfs update, protected by ofd_osfs_lock.
103 * record when some statfs refresh are in progress */
104 int ofd_statfs_inflight;
105 /* track writes completed while statfs refresh is underway.
106 * tracking is only effective when ofd_statfs_inflight > 1 */
107 obd_size ofd_osfs_inflight;
109 /* grants: all values in bytes */
110 /* grant lock to protect all grant counters */
111 cfs_spinlock_t ofd_grant_lock;
112 /* total amount of dirty data reported by clients in incoming obdo */
113 obd_size ofd_tot_dirty;
114 /* sum of filesystem space granted to clients for async writes */
115 obd_size ofd_tot_granted;
116 /* grant used by I/Os in progress (between prepare and commit) */
117 obd_size ofd_tot_pending;
118 /* free space threshold over which we stop granting space to clients
119 * ofd_grant_ratio is stored as a fixed-point fraction using
120 * OFD_GRANT_RATIO_SHIFT of the remaining free space, not in percentage
123 /* number of clients using grants */
124 int ofd_tot_granted_clients;
126 /* ofd mod data: ofd_device wide values */
127 int ofd_fmd_max_num; /* per ofd ofd_mod_data */
128 cfs_duration_t ofd_fmd_max_age; /* time to fmd expiry */
130 cfs_spinlock_t ofd_flags_lock;
131 unsigned long ofd_raid_degraded:1,
132 /* sync journal on writes */
134 /* sync on lock cancel */
135 ofd_sync_lock_cancel:2,
136 /* shall we grant space to clients not
137 * supporting OBD_CONNECT_GRANT_PARAM? */
138 ofd_grant_compat_disable:1;
140 struct lu_site ofd_site;
143 static inline struct ofd_device *ofd_dev(struct lu_device *d)
145 return container_of0(d, struct ofd_device, ofd_dt_dev.dd_lu_dev);
148 static inline struct obd_device *ofd_obd(struct ofd_device *ofd)
150 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd;
153 static inline struct ofd_device *ofd_exp(struct obd_export *exp)
155 return ofd_dev(exp->exp_obd->obd_lu_dev);
158 static inline char *ofd_name(struct ofd_device *ofd)
160 return ofd->ofd_dt_dev.dd_lu_dev.ld_obd->obd_name;
164 struct lu_object_header ofo_header;
165 struct dt_object ofo_obj;
168 static inline struct ofd_object *ofd_obj(struct lu_object *o)
170 return container_of0(o, struct ofd_object, ofo_obj.do_lu);
173 static inline int ofd_object_exists(struct ofd_object *obj)
175 LASSERT(obj != NULL);
176 if (lu_object_is_dying(obj->ofo_obj.do_lu.lo_header))
178 return lu_object_exists(&obj->ofo_obj.do_lu);
181 static inline struct dt_object *fo2dt(struct ofd_object *obj)
183 return &obj->ofo_obj;
186 static inline struct dt_object *ofd_object_child(struct ofd_object *_obj)
188 struct lu_object *lu = &(_obj)->ofo_obj.do_lu;
190 return container_of0(lu_object_next(lu), struct dt_object, do_lu);
194 * Common data shared by obdofd-level handlers. This is allocated per-thread
195 * to reduce stack consumption.
197 struct ofd_thread_info {
198 const struct lu_env *fti_env;
200 struct obd_export *fti_exp;
203 __u64 fti_pre_version;
204 __u32 fti_has_trans:1, /* has txn already */
207 struct lu_fid fti_fid;
208 struct lu_attr fti_attr;
209 struct ofd_object *fti_obj;
211 char name[64]; /* for ofd_init0() */
212 struct obd_statfs osfs; /* for obdofd_statfs() */
215 struct dt_object_format fti_dof;
216 struct lu_buf fti_buf;
219 /* Space used by the I/O, used by grant code */
220 unsigned long fti_used;
223 extern void target_recovery_fini(struct obd_device *obd);
224 extern void target_recovery_init(struct lu_target *lut, svc_handler_t handler);
227 int ofd_update_capa_key(struct ofd_device *ofd, struct lustre_capa_key *key);
228 int ofd_auth_capa(struct obd_export *exp, struct lu_fid *fid, obd_seq seq,
229 struct lustre_capa *capa, __u64 opc);
230 void ofd_free_capa_keys(struct ofd_device *ofd);
233 extern struct lu_context_key ofd_thread_key;
236 extern struct obd_ops ofd_obd_ops;
237 int ofd_statfs_internal(const struct lu_env *env, struct ofd_device *ofd,
238 struct obd_statfs *osfs, __u64 max_age,
242 obd_id ofd_last_id(struct ofd_device *ofd, obd_seq seq);
243 int ofd_group_load(const struct lu_env *env, struct ofd_device *ofd, int);
244 int ofd_fs_setup(const struct lu_env *env, struct ofd_device *ofd,
245 struct obd_device *obd);
246 void ofd_fs_cleanup(const struct lu_env *env, struct ofd_device *ofd);
249 struct thandle *ofd_trans_create(const struct lu_env *env,
250 struct ofd_device *ofd);
251 int ofd_trans_start(const struct lu_env *env,
252 struct ofd_device *ofd, struct ofd_object *fo,
254 void ofd_trans_stop(const struct lu_env *env, struct ofd_device *ofd,
255 struct thandle *th, int rc);
256 int ofd_txn_stop_cb(const struct lu_env *env, struct thandle *txn,
260 void lprocfs_ofd_init_vars(struct lprocfs_static_vars *lvars);
261 int lproc_ofd_attach_seqstat(struct obd_device *dev);
262 extern struct file_operations ofd_per_nid_stats_fops;
265 #define OFD_GRANT_RATIO_SHIFT 8
266 static inline __u64 ofd_grant_reserved(struct ofd_device *ofd, obd_size bavail)
268 return (bavail * ofd->ofd_grant_ratio) >> OFD_GRANT_RATIO_SHIFT;
271 static inline int ofd_grant_ratio_conv(int percentage)
273 return (percentage << OFD_GRANT_RATIO_SHIFT) / 100;
276 static inline int ofd_grant_param_supp(struct obd_export *exp)
278 return !!(exp->exp_connect_flags & OBD_CONNECT_GRANT_PARAM);
281 /* Blocksize used for client not supporting OBD_CONNECT_GRANT_PARAM.
282 * That's 4KB=2^12 which is the biggest block size known to work whatever
283 * the client's page size is. */
284 #define COMPAT_BSIZE_SHIFT 12
285 static inline int ofd_grant_compat(struct obd_export *exp,
286 struct ofd_device *ofd)
288 /* Clients which don't support OBD_CONNECT_GRANT_PARAM cannot handle
289 * a block size > page size and consume CFS_PAGE_SIZE of grant when
290 * dirtying a page regardless of the block size */
291 return !!(ofd_obd(ofd)->obd_self_export != exp &&
292 ofd->ofd_blockbits > COMPAT_BSIZE_SHIFT &&
293 !ofd_grant_param_supp(exp));
296 static inline int ofd_grant_prohibit(struct obd_export *exp,
297 struct ofd_device *ofd)
299 /* When ofd_grant_compat_disable is set, we don't grant any space to
300 * clients not supporting OBD_CONNECT_GRANT_PARAM.
301 * Otherwise, space granted to such a client is inflated since it
302 * consumes CFS_PAGE_SIZE of grant space per block */
303 return !!(ofd_grant_compat(exp, ofd) && ofd->ofd_grant_compat_disable);
306 void ofd_grant_sanity_check(struct obd_device *obd, const char *func);
307 long ofd_grant_connect(const struct lu_env *env, struct obd_export *exp,
309 void ofd_grant_discard(struct obd_export *exp);
310 void ofd_grant_prepare_read(const struct lu_env *env, struct obd_export *exp,
312 void ofd_grant_prepare_write(const struct lu_env *env, struct obd_export *exp,
313 struct obdo *oa, struct niobuf_remote *rnb,
315 void ofd_grant_commit(const struct lu_env *env, struct obd_export *exp, int rc);
316 int ofd_grant_create(const struct lu_env *env, struct obd_export *exp, int *nr);
319 int ofd_fmd_init(void);
320 void ofd_fmd_exit(void);
321 struct ofd_mod_data *ofd_fmd_find(struct obd_export *exp,
323 struct ofd_mod_data *ofd_fmd_get(struct obd_export *exp,
325 void ofd_fmd_put(struct obd_export *exp, struct ofd_mod_data *fmd);
326 void ofd_fmd_expire(struct obd_export *exp);
327 void ofd_fmd_cleanup(struct obd_export *exp);
329 void ofd_fmd_drop(struct obd_export *exp, struct lu_fid *fid);
331 #define ofd_fmd_drop(exp, fid) do {} while (0)
334 static inline struct ofd_thread_info * ofd_info(const struct lu_env *env)
336 struct ofd_thread_info *info;
338 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
340 LASSERT(info->fti_env);
341 LASSERT(info->fti_env == env);
345 static inline struct ofd_thread_info * ofd_info_init(const struct lu_env *env,
346 struct obd_export *exp)
348 struct ofd_thread_info *info;
350 info = lu_context_key_get(&env->le_ctx, &ofd_thread_key);
352 LASSERT(info->fti_exp == NULL);
353 LASSERT(info->fti_env == NULL);
360 /* sync on lock cancel is useless when we force a journal flush,
361 * and if we enable async journal commit, we should also turn on
362 * sync on lock cancel if it is not enabled already. */
363 static inline void ofd_slc_set(struct ofd_device *ofd)
365 if (ofd->ofd_syncjournal == 1)
366 ofd->ofd_sync_lock_cancel = NEVER_SYNC_ON_CANCEL;
367 else if (ofd->ofd_sync_lock_cancel == NEVER_SYNC_ON_CANCEL)
368 ofd->ofd_sync_lock_cancel = ALWAYS_SYNC_ON_CANCEL;
371 /* niobuf_local has no rnb_ prefix in master */
372 #define rnb_offset offset
373 #define rnb_flags flags
376 #endif /* _OFD_INTERNAL_H */