1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre_log.h
38 * Generic infrastructure for managing a collection of logs.
39 * These logs are used for:
41 * - orphan recovery: OST adds record on create
42 * - mtime/size consistency: the OST adds a record on first write
43 * - open/unlinked objects: OST adds a record on destroy
45 * - mds unlink log: the MDS adds an entry upon delete
47 * - raid1 replication log between OST's
48 * - MDS replication logs
54 #if defined(__linux__)
55 #include <linux/lustre_log.h>
56 #elif defined(__APPLE__)
57 #include <darwin/lustre_log.h>
58 #elif defined(__WINNT__)
59 #include <winnt/lustre_log.h>
61 #error Unsupported operating system.
66 #include <lustre/lustre_idl.h>
68 #define LOG_NAME_LIMIT(logname, name) \
69 snprintf(logname, sizeof(logname), "LOGS/%s", name)
70 #define LLOG_EEMPTY 4711
72 struct plain_handle_data {
73 struct list_head phd_entry;
74 struct llog_handle *phd_cat_handle;
75 struct llog_cookie phd_cookie; /* cookie of this log in its cat */
79 struct cat_handle_data {
80 struct list_head chd_head;
81 struct llog_handle *chd_current_log; /* currently open log */
84 /* In-memory descriptor for a log object or log catalog */
86 struct rw_semaphore lgh_lock;
87 struct llog_logid lgh_id; /* id of this log */
88 struct llog_log_hdr *lgh_hdr;
89 struct file *lgh_file;
91 int lgh_cur_idx; /* used during llog_process */
92 __u64 lgh_cur_offset; /* used during llog_process */
93 struct llog_ctxt *lgh_ctxt;
95 struct plain_handle_data phd;
96 struct cat_handle_data chd;
100 /* llog.c - general API */
101 typedef int (*llog_cb_t)(struct llog_handle *, struct llog_rec_hdr *, void *);
102 typedef int (*llog_fill_rec_cb_t)(struct llog_rec_hdr *rec, void *data);
103 extern struct llog_handle *llog_alloc_handle(void);
104 int llog_init_handle(struct llog_handle *handle, int flags,
105 struct obd_uuid *uuid);
106 extern void llog_free_handle(struct llog_handle *handle);
107 int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
108 void *data, void *catdata);
109 int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
110 void *data, void *catdata);
111 extern int llog_cancel_rec(struct llog_handle *loghandle, int index);
112 extern int llog_close(struct llog_handle *cathandle);
113 extern int llog_get_size(struct llog_handle *loghandle);
115 /* llog_cat.c - catalog api */
116 struct llog_process_data {
118 * Any useful data needed while processing catalog. This is
119 * passed later to process callback.
123 * Catalog process callback function, called for each record
128 * Start processing the catalog from startcat/startidx
134 struct llog_process_cat_data {
136 * Temporary stored first_idx while scanning log.
140 * Temporary stored last_idx while scanning log.
145 struct llog_process_cat_args {
147 * Llog context used in recovery thread on OST (recov_thread.c)
149 struct llog_ctxt *lpca_ctxt;
151 * Llog callback used in recovery thread on OST (recov_thread.c)
155 * Data pointer for llog callback.
160 int llog_cat_put(struct llog_handle *cathandle);
161 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
162 struct llog_cookie *reccookie, void *buf);
163 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
164 struct llog_cookie *cookies);
165 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
166 int startcat, int startidx);
167 int llog_cat_process_thread(void *data);
168 int llog_cat_reverse_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data);
169 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
172 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
173 int index, struct obd_device *disk_obd, int count,
174 struct llog_logid *logid, const char *logname,
175 struct llog_operations *op);
176 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
177 struct obd_device *disk_obd, int count, struct llog_logid *logid,
178 struct llog_operations *op);
179 int __llog_ctxt_put(struct llog_ctxt *ctxt);
180 int llog_cleanup(struct llog_ctxt *);
181 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp);
182 int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
183 struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
185 int llog_cancel(struct llog_ctxt *, struct lov_stripe_md *lsm,
186 int count, struct llog_cookie *cookies, int flags);
188 int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
189 int index, struct obd_device *disk_obd, int count,
190 struct llog_logid *logid, const char *name);
191 int llog_obd_origin_cleanup(struct llog_ctxt *ctxt);
192 int llog_obd_origin_add(struct llog_ctxt *ctxt,
193 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
194 struct llog_cookie *logcookies, int numcookies);
196 int llog_cat_initialize(struct obd_device *obd, struct obd_llog_group *olg,
197 int idx, struct obd_uuid *uuid);
198 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
199 struct obd_device *disk_obd, int count,
200 struct llog_catid *logid, struct obd_uuid *uuid);
202 int obd_llog_finish(struct obd_device *obd, int count);
205 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
206 int llog_catalog_list(struct obd_device *obd, int count,
207 struct obd_ioctl_data *data);
210 int llog_initiator_connect(struct llog_ctxt *ctxt);
211 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
212 int llog_origin_connect(struct llog_ctxt *ctxt,
213 struct llog_logid *logid, struct llog_gen *gen,
214 struct obd_uuid *uuid);
215 int llog_handle_connect(struct ptlrpc_request *req);
218 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
219 struct lov_stripe_md *lsm, int count,
220 struct llog_cookie *cookies, int flags);
221 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp);
222 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
223 struct llog_logid *logid, struct llog_gen *gen,
224 struct obd_uuid *uuid);
226 struct llog_operations {
227 int (*lop_write_rec)(struct llog_handle *loghandle,
228 struct llog_rec_hdr *rec,
229 struct llog_cookie *logcookies, int numcookies,
231 int (*lop_destroy)(struct llog_handle *handle);
232 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
233 int next_idx, __u64 *offset, void *buf, int len);
234 int (*lop_prev_block)(struct llog_handle *h,
235 int prev_idx, void *buf, int len);
236 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
237 struct llog_logid *logid, char *name);
238 int (*lop_close)(struct llog_handle *handle);
239 int (*lop_read_header)(struct llog_handle *handle);
241 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
242 int ctxt_idx, struct obd_device *disk_obd, int count,
243 struct llog_logid *logid, const char *name);
244 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp);
245 int (*lop_cleanup)(struct llog_ctxt *ctxt);
246 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
247 struct lov_stripe_md *lsm,
248 struct llog_cookie *logcookies, int numcookies);
249 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
250 int count, struct llog_cookie *cookies, int flags);
251 int (*lop_connect)(struct llog_ctxt *ctxt,
252 struct llog_logid *logid, struct llog_gen *gen,
253 struct obd_uuid *uuid);
254 /* XXX add 2 more: commit callbacks and llog recovery functions */
258 extern struct llog_operations llog_lvfs_ops;
259 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
260 char *name, int idx, int count,
261 struct llog_catid *idarray);
263 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
264 char *name, int idx, int count, struct llog_catid *idarray);
267 int loc_idx; /* my index the obd array of ctxt's */
268 struct llog_gen loc_gen;
269 struct obd_device *loc_obd; /* points back to the containing obd*/
270 struct obd_llog_group *loc_olg; /* group containing that ctxt */
271 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
272 struct obd_import *loc_imp; /* to use in RPC's: can be backward
274 struct llog_operations *loc_logops;
275 struct llog_handle *loc_handle;
276 struct llog_commit_master *loc_lcm;
277 struct llog_canceld_ctxt *loc_llcd;
278 struct semaphore loc_sem; /* protects loc_llcd and loc_imp */
279 atomic_t loc_refcount;
283 #define LCM_NAME_SIZE 64
285 struct llog_commit_master {
287 * Thread control flags (start, stop, etc.)
291 * Number of llcds onthis lcm.
295 * Thread control structure. Used for control commit thread.
297 struct ptlrpcd_ctl lcm_pc;
299 * Lock protecting list of llcds.
303 * Llcds in flight for debugging purposes.
305 struct list_head lcm_llcds;
307 * Commit thread name buffer. Only used for thread start.
309 char lcm_name[LCM_NAME_SIZE];
312 struct llog_canceld_ctxt {
314 * Llog context this llcd is attached to. Used for accessing
315 * ->loc_import and others in process of canceling cookies
316 * gathered in this llcd.
318 struct llog_ctxt *llcd_ctxt;
320 * Cancel thread control stucture pointer. Used for accessing
321 * it to see if should stop processing and other needs.
323 struct llog_commit_master *llcd_lcm;
325 * Maximal llcd size. Used in calculations on how much of room
326 * left in llcd to cookie comming cookies.
330 * Link to lcm llcds list.
332 struct list_head llcd_list;
334 * Current llcd size while gathering cookies. This should not be
335 * more than ->llcd_size. Used for determining if we need to
336 * send this llcd (if full) and allocate new one. This is also
337 * used for copying new cookie at the end of buffer.
339 int llcd_cookiebytes;
341 * Pointer to the start of cookies buffer.
343 struct llog_cookie llcd_cookies[0];
346 /* ptlrpc/recov_thread.c */
347 extern struct llog_commit_master *llog_recov_thread_init(char *name);
348 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
350 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
351 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
354 static inline void llog_gen_init(struct llog_ctxt *ctxt)
356 struct obd_device *obd = ctxt->loc_exp->exp_obd;
358 if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
359 ctxt->loc_gen.mnt_cnt = obd->u.mds.mds_mount_count;
360 else if (!strstr(obd->obd_type->typ_name, LUSTRE_OST_NAME))
361 ctxt->loc_gen.mnt_cnt = obd->u.filter.fo_mount_count;
363 ctxt->loc_gen.mnt_cnt = 0;
366 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
368 if (a.mnt_cnt < b.mnt_cnt)
370 if (a.mnt_cnt > b.mnt_cnt)
372 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
375 #define LLOG_GEN_INC(gen) ((gen).conn_cnt ++)
376 #define LLOG_PROC_BREAK 0x0001
377 #define LLOG_DEL_RECORD 0x0002
379 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
380 struct llog_operations **lop)
385 *lop = ctxt->loc_logops;
392 static inline int llog_handle2ops(struct llog_handle *loghandle,
393 struct llog_operations **lop)
395 if (loghandle == NULL)
398 return llog_obd2ops(loghandle->lgh_ctxt, lop);
401 static inline int llog_data_len(int len)
403 return size_round(len);
406 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
408 LASSERT(atomic_read(&ctxt->loc_refcount) > 0);
409 atomic_inc(&ctxt->loc_refcount);
410 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
411 atomic_read(&ctxt->loc_refcount));
415 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
419 LASSERT(atomic_read(&ctxt->loc_refcount) > 0);
420 LASSERT(atomic_read(&ctxt->loc_refcount) < 0x5a5a5a);
421 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
422 atomic_read(&ctxt->loc_refcount) - 1);
423 __llog_ctxt_put(ctxt);
426 static inline void llog_group_init(struct obd_llog_group *olg, int group)
428 cfs_waitq_init(&olg->olg_waitq);
429 spin_lock_init(&olg->olg_lock);
430 sema_init(&olg->olg_cat_processing, 1);
431 olg->olg_group = group;
434 static inline void llog_group_set_export(struct obd_llog_group *olg,
435 struct obd_export *exp)
437 LASSERT(exp != NULL);
439 spin_lock(&olg->olg_lock);
440 if (olg->olg_exp != NULL && olg->olg_exp != exp)
441 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
442 exp->exp_obd->obd_name, olg->olg_group,
445 spin_unlock(&olg->olg_lock);
448 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
449 struct llog_ctxt *ctxt, int index)
451 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
453 spin_lock(&olg->olg_lock);
454 if (olg->olg_ctxts[index] != NULL) {
455 spin_unlock(&olg->olg_lock);
458 olg->olg_ctxts[index] = ctxt;
459 spin_unlock(&olg->olg_lock);
463 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
466 struct llog_ctxt *ctxt;
468 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
470 spin_lock(&olg->olg_lock);
471 if (olg->olg_ctxts[index] == NULL) {
474 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
476 spin_unlock(&olg->olg_lock);
480 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
483 return llog_group_get_ctxt(&obd->obd_olg, index);
486 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
488 return (olg->olg_ctxts[index] == NULL);
491 static inline int llog_ctxt_null(struct obd_device *obd, int index)
493 return (llog_group_ctxt_null(&obd->obd_olg, index));
496 static inline int llog_write_rec(struct llog_handle *handle,
497 struct llog_rec_hdr *rec,
498 struct llog_cookie *logcookies,
499 int numcookies, void *buf, int idx)
501 struct llog_operations *lop;
502 int raised, rc, buflen;
505 rc = llog_handle2ops(handle, &lop);
509 if (lop->lop_write_rec == NULL)
513 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
514 + sizeof(struct llog_rec_tail);
516 buflen = rec->lrh_len;
517 LASSERT(size_round(buflen) == buflen);
519 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
521 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
522 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
524 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
528 static inline int llog_read_header(struct llog_handle *handle)
530 struct llog_operations *lop;
534 rc = llog_handle2ops(handle, &lop);
537 if (lop->lop_read_header == NULL)
540 rc = lop->lop_read_header(handle);
544 static inline int llog_destroy(struct llog_handle *handle)
546 struct llog_operations *lop;
550 rc = llog_handle2ops(handle, &lop);
553 if (lop->lop_destroy == NULL)
556 rc = lop->lop_destroy(handle);
561 static inline int llog_cancel(struct obd_export *exp,
562 struct lov_stripe_md *lsm, int count,
563 struct llog_cookie *cookies, int flags)
565 struct llog_operations *lop;
569 rc = llog_handle2ops(loghandle, &lop);
572 if (lop->lop_cancel == NULL)
575 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
580 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
581 int next_idx, __u64 *cur_offset, void *buf,
584 struct llog_operations *lop;
588 rc = llog_handle2ops(loghandle, &lop);
591 if (lop->lop_next_block == NULL)
594 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
599 static inline int llog_prev_block(struct llog_handle *loghandle,
600 int prev_idx, void *buf, int len)
602 struct llog_operations *lop;
606 rc = llog_handle2ops(loghandle, &lop);
609 if (lop->lop_prev_block == NULL)
612 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
616 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
617 struct llog_logid *logid, char *name)
619 struct llog_operations *lop;
623 rc = llog_obd2ops(ctxt, &lop);
626 if (lop->lop_create == NULL)
629 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
631 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
632 rc = lop->lop_create(ctxt, res, logid, name);
634 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
638 static inline int llog_connect(struct llog_ctxt *ctxt,
639 struct llog_logid *logid, struct llog_gen *gen,
640 struct obd_uuid *uuid)
642 struct llog_operations *lop;
646 rc = llog_obd2ops(ctxt, &lop);
649 if (lop->lop_connect == NULL)
652 rc = lop->lop_connect(ctxt, logid, gen, uuid);