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) 2007, 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.
36 * lustre/osp/osp_internal.h
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #ifndef _OSP_INTERNAL_H
42 #define _OSP_INTERNAL_H
45 #include <obd_class.h>
46 #include <dt_object.h>
47 #include <md_object.h>
48 #include <lustre_fid.h>
49 #include <lustre_update.h>
50 #include <lu_target.h>
53 * Infrastructure to support tracking of last committed llog record
55 struct osp_id_tracker {
58 __u32 otr_committed_id;
59 /* callback is register once per diskfs -- that's the whole point */
60 struct dt_txn_callback otr_tx_cb;
61 /* single node can run many clusters */
62 struct list_head otr_wakeup_list;
63 struct list_head otr_list;
64 /* underlying shared device */
65 struct dt_device *otr_dev;
66 /* how many users of this tracker */
67 atomic_t otr_refcount;
70 struct osp_precreate {
74 spinlock_t osp_pre_lock;
76 /* last fid to assign in creation */
77 struct lu_fid osp_pre_used_fid;
78 /* last created id OST reported, next-created - available id's */
79 struct lu_fid osp_pre_last_created_fid;
80 /* how many ids are reserved in declare, we shouldn't block in create */
81 __u64 osp_pre_reserved;
82 /* consumers (who needs new ids) wait here */
83 wait_queue_head_t osp_pre_user_waitq;
84 /* current precreation status: working, failed, stopping? */
86 /* how many to precreate next time */
87 int osp_pre_grow_count;
88 int osp_pre_min_grow_count;
89 int osp_pre_max_grow_count;
90 /* whether to grow precreation window next time or not */
91 int osp_pre_grow_slow;
92 /* cleaning up orphans or recreating missing objects */
93 int osp_pre_recovering;
97 struct dt_device opd_dt_dev;
98 /* corresponded OST index */
101 /* corrsponded MDT index, which will be used when connecting to OST
102 * for validating the connection (see ofd_parse_connect_data) */
104 /* device used to store persistent state (llogs, last ids) */
105 struct obd_export *opd_storage_exp;
106 struct dt_device *opd_storage;
107 struct dt_object *opd_last_used_oid_file;
108 struct dt_object *opd_last_used_seq_file;
110 /* stored persistently in LE format, updated directly to/from disk
111 * and required le64_to_cpu() conversion before use.
112 * Protected by opd_pre_lock */
113 struct lu_fid opd_last_used_fid;
114 struct lu_fid opd_gap_start_fid;
116 /* connection to OST */
117 struct obd_device *opd_obd;
118 struct obd_export *opd_exp;
119 struct obd_uuid opd_cluuid;
120 struct obd_connect_data *opd_connect_data;
122 cfs_proc_dir_entry_t *opd_proc_entry;
123 struct lprocfs_stats *opd_stats;
124 /* connection status. */
125 unsigned int opd_new_connection:1,
126 opd_got_disconnected:1,
129 opd_imp_seen_connected:1,
132 /* whether local recovery is completed:
133 * reported via ->ldo_recovery_complete() */
134 int opd_recovery_completed;
136 /* precreate structure for OSP */
137 struct osp_precreate *opd_pre;
138 /* dedicate precreate thread */
139 struct ptlrpc_thread opd_pre_thread;
140 /* thread waits for signals about pool going empty */
141 wait_queue_head_t opd_pre_waitq;
144 * OST synchronization
146 spinlock_t opd_syn_lock;
147 /* unique generation, to recognize start of new records in the llog */
148 struct llog_gen opd_syn_generation;
149 /* number of changes to sync, used to wake up sync thread */
150 unsigned long opd_syn_changes;
151 /* processing of changes from previous mount is done? */
152 int opd_syn_prev_done;
154 struct ptlrpc_thread opd_syn_thread;
155 wait_queue_head_t opd_syn_waitq;
156 /* list of remotely committed rpc */
157 struct list_head opd_syn_committed_there;
158 /* number of changes being under sync */
159 int opd_syn_sync_in_progress;
160 /* number of RPCs in flight - flow control */
161 int opd_syn_rpc_in_flight;
162 int opd_syn_max_rpc_in_flight;
163 /* number of RPC in processing (including non-committed by OST) */
164 int opd_syn_rpc_in_progress;
165 int opd_syn_max_rpc_in_progress;
166 /* osd api's commit cb control structure */
167 struct dt_txn_callback opd_syn_txn_cb;
168 /* last used change number -- semantically similar to transno */
169 unsigned long opd_syn_last_used_id;
170 /* last committed change number -- semantically similar to
172 unsigned long opd_syn_last_committed_id;
173 /* last processed (taken from llog) id */
174 unsigned long opd_syn_last_processed_id;
175 struct osp_id_tracker *opd_syn_tracker;
176 struct list_head opd_syn_ontrack;
179 * statfs related fields: OSP maintains it on its own
181 struct obd_statfs opd_statfs;
182 cfs_time_t opd_statfs_fresh_till;
183 struct timer_list opd_statfs_timer;
184 int opd_statfs_update_in_progress;
185 /* how often to update statfs data */
186 int opd_statfs_maxage;
188 cfs_proc_dir_entry_t *opd_symlink;
190 /* If the caller wants to do some idempotent async operations on
191 * remote server, it can append the async remote requests on the
192 * osp_device::opd_async_requests via declare() functions, these
193 * requests can be packed together and sent to the remote server
194 * via single OUT RPC later. */
195 struct dt_update_request *opd_async_requests;
196 /* Protect current operations on opd_async_requests. */
197 struct mutex opd_async_requests_mutex;
200 #define opd_pre_lock opd_pre->osp_pre_lock
201 #define opd_pre_used_fid opd_pre->osp_pre_used_fid
202 #define opd_pre_last_created_fid opd_pre->osp_pre_last_created_fid
203 #define opd_pre_reserved opd_pre->osp_pre_reserved
204 #define opd_pre_user_waitq opd_pre->osp_pre_user_waitq
205 #define opd_pre_status opd_pre->osp_pre_status
206 #define opd_pre_grow_count opd_pre->osp_pre_grow_count
207 #define opd_pre_min_grow_count opd_pre->osp_pre_min_grow_count
208 #define opd_pre_max_grow_count opd_pre->osp_pre_max_grow_count
209 #define opd_pre_grow_slow opd_pre->osp_pre_grow_slow
210 #define opd_pre_recovering opd_pre->osp_pre_recovering
212 extern struct kmem_cache *osp_object_kmem;
214 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
215 * The left part is for value, binary mode. */
216 struct osp_xattr_entry {
217 struct list_head oxe_list;
223 unsigned int oxe_exist:1,
228 struct osp_object_attr {
229 struct lu_attr ooa_attr;
230 struct list_head ooa_xattr_list;
233 /* this is a top object */
235 struct lu_object_header opo_header;
236 struct dt_object opo_obj;
237 unsigned int opo_reserved:1,
241 /* read/write lock for md osp object */
242 struct rw_semaphore opo_sem;
243 const struct lu_env *opo_owner;
244 struct osp_object_attr *opo_ooa;
245 /* Protect opo_ooa. */
249 extern struct lu_object_operations osp_lu_obj_ops;
250 extern const struct dt_device_operations osp_dt_ops;
251 extern struct dt_object_operations osp_md_obj_ops;
252 extern struct dt_body_operations osp_md_body_ops;
254 struct osp_thread_info {
255 struct lu_buf osi_lb;
256 struct lu_buf osi_lb2;
257 struct lu_fid osi_fid;
258 struct lu_attr osi_attr;
259 struct ost_id osi_oi;
260 struct ost_id osi_oi2;
264 struct llog_rec_hdr osi_hdr;
265 struct llog_unlink64_rec osi_unlink;
266 struct llog_setattr64_rec osi_setattr;
267 struct llog_gen_rec osi_gen;
269 struct llog_cookie osi_cookie;
270 struct llog_catid osi_cid;
271 struct lu_seq_range osi_seq;
272 struct ldlm_res_id osi_resid;
273 struct obdo osi_obdo;
276 /* Iterator for OSP */
279 __u32 ooi_pos_lu_page;
281 int ooi_total_npages;
282 int ooi_valid_npages;
283 unsigned int ooi_swab:1;
285 struct dt_object *ooi_obj;
287 struct page *ooi_cur_page;
288 struct lu_idxpage *ooi_cur_idxpage;
289 struct page **ooi_pages;
292 /* The transaction only include the updates on the remote node, and
293 * no local updates at all */
294 static inline bool is_only_remote_trans(struct thandle *th)
296 return th->th_dev != NULL && th->th_dev->dd_ops == &osp_dt_ops;
299 static inline void osp_objid_buf_prep(struct lu_buf *buf, loff_t *off,
300 __u32 *id, int index)
302 buf->lb_buf = (void *)id;
303 buf->lb_len = sizeof(obd_id);
304 *off = sizeof(obd_id) * index;
307 static inline void osp_objseq_buf_prep(struct lu_buf *buf, loff_t *off,
308 __u64 *seq, int index)
310 buf->lb_buf = (void *)seq;
311 buf->lb_len = sizeof(obd_id);
312 *off = sizeof(obd_id) * index;
315 static inline void osp_buf_prep(struct lu_buf *lb, void *buf, int buf_len)
318 lb->lb_len = buf_len;
321 extern struct lu_context_key osp_thread_key;
323 static inline struct osp_thread_info *osp_env_info(const struct lu_env *env)
325 struct osp_thread_info *info;
327 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
329 lu_env_refill((struct lu_env *)env);
330 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
336 struct osp_txn_info {
337 __u32 oti_current_id;
340 extern struct lu_context_key osp_txn_key;
342 static inline struct osp_txn_info *osp_txn_info(struct lu_context *ctx)
344 struct osp_txn_info *info;
346 info = lu_context_key_get(ctx, &osp_txn_key);
350 extern const struct lu_device_operations osp_lu_ops;
352 static inline int lu_device_is_osp(struct lu_device *d)
354 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osp_lu_ops);
357 static inline struct osp_device *lu2osp_dev(struct lu_device *d)
359 LASSERT(lu_device_is_osp(d));
360 return container_of0(d, struct osp_device, opd_dt_dev.dd_lu_dev);
363 static inline struct lu_device *osp2lu_dev(struct osp_device *d)
365 return &d->opd_dt_dev.dd_lu_dev;
368 static inline struct osp_device *dt2osp_dev(struct dt_device *d)
370 LASSERT(lu_device_is_osp(&d->dd_lu_dev));
371 return container_of0(d, struct osp_device, opd_dt_dev);
374 static inline struct osp_object *lu2osp_obj(struct lu_object *o)
376 LASSERT(ergo(o != NULL, lu_device_is_osp(o->lo_dev)));
377 return container_of0(o, struct osp_object, opo_obj.do_lu);
380 static inline struct lu_object *osp2lu_obj(struct osp_object *obj)
382 return &obj->opo_obj.do_lu;
385 static inline struct osp_object *osp_obj(const struct lu_object *o)
387 LASSERT(lu_device_is_osp(o->lo_dev));
388 return container_of0(o, struct osp_object, opo_obj.do_lu);
391 static inline struct osp_object *dt2osp_obj(const struct dt_object *d)
393 return osp_obj(&d->do_lu);
396 static inline struct dt_object *osp_object_child(struct osp_object *o)
398 return container_of0(lu_object_next(osp2lu_obj(o)),
399 struct dt_object, do_lu);
402 static inline struct seq_server_site *osp_seq_site(struct osp_device *osp)
404 return osp->opd_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
407 #define osp_init_rpc_lock(lck) mdc_init_rpc_lock(lck)
408 #define osp_get_rpc_lock(lck, it) mdc_get_rpc_lock(lck, it)
409 #define osp_put_rpc_lock(lck, it) mdc_put_rpc_lock(lck, it)
411 static inline int osp_fid_diff(const struct lu_fid *fid1,
412 const struct lu_fid *fid2)
414 /* In 2.6+ ost_idx is packed into IDIF FID, while in 2.4 and 2.5 IDIF
415 * is always FID_SEQ_IDIF(0x100000000ULL), which does not include OST
416 * index in the seq. So we can not compare IDIF FID seq here */
417 if (fid_is_idif(fid1) && fid_is_idif(fid2)) {
418 __u32 ost_idx1 = fid_idif_ost_idx(fid1);
419 __u32 ost_idx2 = fid_idif_ost_idx(fid2);
421 LASSERTF(ost_idx1 == 0 || ost_idx2 == 0 || ost_idx1 == ost_idx2,
422 "fid1: "DFID", fid2: "DFID"\n", PFID(fid1),
425 return fid_idif_id(fid1->f_seq, fid1->f_oid, 0) -
426 fid_idif_id(fid2->f_seq, fid2->f_oid, 0);
429 LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID
430 ", fid2:"DFID"\n", PFID(fid1), PFID(fid2));
432 return fid_oid(fid1) - fid_oid(fid2);
436 static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
438 int diff = osp_fid_diff(fid, &d->opd_last_used_fid);
440 * we might have lost precreated objects due to VBR and precreate
441 * orphans, the gap in objid can be calculated properly only here
445 d->opd_gap_start_fid = d->opd_last_used_fid;
446 d->opd_gap_start_fid.f_oid++;
447 d->opd_gap_count = diff - 1;
448 CDEBUG(D_HA, "Gap in objids: start="DFID", count =%d\n",
449 PFID(&d->opd_gap_start_fid), d->opd_gap_count);
451 d->opd_last_used_fid = *fid;
455 static int osp_fid_end_seq(const struct lu_env *env, struct lu_fid *fid)
457 if (fid_is_idif(fid)) {
458 struct osp_thread_info *info = osp_env_info(env);
459 struct ost_id *oi = &info->osi_oi;
461 fid_to_ostid(fid, oi);
462 return ostid_id(oi) == IDIF_MAX_OID;
464 return fid_oid(fid) == LUSTRE_DATA_SEQ_MAX_WIDTH;
468 static inline int osp_precreate_end_seq_nolock(const struct lu_env *env,
469 struct osp_device *osp)
471 struct lu_fid *fid = &osp->opd_pre_last_created_fid;
473 return osp_fid_end_seq(env, fid);
476 static inline int osp_precreate_end_seq(const struct lu_env *env,
477 struct osp_device *osp)
481 spin_lock(&osp->opd_pre_lock);
482 rc = osp_precreate_end_seq_nolock(env, osp);
483 spin_unlock(&osp->opd_pre_lock);
487 static inline int osp_is_fid_client(struct osp_device *osp)
489 struct obd_import *imp = osp->opd_obd->u.cli.cl_import;
491 return imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_FID;
494 typedef int (*osp_async_update_interpterer_t)(const struct lu_env *env,
495 struct object_update_reply *reply,
496 struct ptlrpc_request *req,
497 struct osp_object *obj,
498 void *data, int index, int rc);
501 void osp_update_last_id(struct osp_device *d, obd_id objid);
502 extern struct llog_operations osp_mds_ost_orig_logops;
505 struct dt_update_request *
506 osp_find_or_create_async_update_request(struct osp_device *osp);
507 int osp_insert_async_update(const struct lu_env *env,
508 struct dt_update_request *update, int op,
509 struct osp_object *obj, int count,
510 int *lens, const char **bufs, void *data,
511 osp_async_update_interpterer_t interpterer);
512 int osp_unplug_async_update(const struct lu_env *env,
513 struct osp_device *osp,
514 struct dt_update_request *update);
515 struct thandle *osp_trans_create(const struct lu_env *env,
516 struct dt_device *d);
517 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
521 int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
522 struct lu_attr *attr, struct lustre_capa *capa);
523 int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
524 struct lu_buf *buf, const char *name,
525 struct lustre_capa *capa);
526 int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
527 const struct lu_buf *buf, const char *name,
528 int flag, struct thandle *th);
529 int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
530 const struct lu_buf *buf, const char *name, int fl,
531 struct thandle *th, struct lustre_capa *capa);
532 int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
533 const char *name, struct thandle *th);
534 int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
535 const char *name, struct thandle *th,
536 struct lustre_capa *capa);
538 int osp_declare_object_destroy(const struct lu_env *env,
539 struct dt_object *dt, struct thandle *th);
540 int osp_object_destroy(const struct lu_env *env, struct dt_object *dt,
543 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
546 struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
547 __u32 attr, struct lustre_capa *capa);
548 void osp_it_fini(const struct lu_env *env, struct dt_it *di);
549 int osp_it_get(const struct lu_env *env, struct dt_it *di,
550 const struct dt_key *key);
551 void osp_it_put(const struct lu_env *env, struct dt_it *di);
552 __u64 osp_it_store(const struct lu_env *env, const struct dt_it *di);
553 int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
555 int osp_it_next_page(const struct lu_env *env, struct dt_it *di);
556 /* osp_md_object.c */
557 int osp_md_declare_object_create(const struct lu_env *env,
558 struct dt_object *dt,
559 struct lu_attr *attr,
560 struct dt_allocation_hint *hint,
561 struct dt_object_format *dof,
563 int osp_md_object_create(const struct lu_env *env, struct dt_object *dt,
564 struct lu_attr *attr, struct dt_allocation_hint *hint,
565 struct dt_object_format *dof, struct thandle *th);
566 int osp_md_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
567 const struct lu_attr *attr, struct thandle *th);
568 int osp_md_attr_set(const struct lu_env *env, struct dt_object *dt,
569 const struct lu_attr *attr, struct thandle *th,
570 struct lustre_capa *capa);
571 extern const struct dt_index_operations osp_md_index_ops;
573 /* osp_precreate.c */
574 int osp_init_precreate(struct osp_device *d);
575 int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d);
576 __u64 osp_precreate_get_id(struct osp_device *d);
577 int osp_precreate_get_fid(const struct lu_env *env, struct osp_device *d,
579 void osp_precreate_fini(struct osp_device *d);
580 int osp_object_truncate(const struct lu_env *env, struct dt_object *dt, __u64);
581 void osp_pre_update_status(struct osp_device *d, int rc);
582 void osp_statfs_need_now(struct osp_device *d);
583 int osp_reset_last_used(const struct lu_env *env, struct osp_device *osp);
584 int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
585 struct lu_fid *fid, int sync);
586 int osp_init_pre_fid(struct osp_device *osp);
589 void lprocfs_osp_init_vars(struct lprocfs_static_vars *lvars);
590 void osp_lprocfs_init(struct osp_device *osp);
593 int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
594 llog_op_type type, struct thandle *th);
595 int osp_sync_add(const struct lu_env *env, struct osp_object *o,
596 llog_op_type type, struct thandle *th,
597 const struct lu_attr *attr);
598 int osp_sync_init(const struct lu_env *env, struct osp_device *d);
599 int osp_sync_fini(struct osp_device *d);
600 void __osp_sync_check_for_work(struct osp_device *d);
603 void lprocfs_lwp_init_vars(struct lprocfs_static_vars *lvars);
604 extern struct obd_ops lwp_obd_device_ops;
605 extern struct lu_device_type lwp_device_type;