1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (c) 2002 Cluster File Systems, Inc. <info@clusterfs.com>
6 * This file is part of Lustre, http://www.lustre.org.
8 * Lustre is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Lustre is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Lustre; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 #define DEBUG_SUBSYSTEM S_CMOBD
24 #include <linux/version.h>
25 #include <linux/init.h>
26 #include <linux/obd_support.h>
27 #include <linux/lustre_lib.h>
28 #include <linux/lustre_net.h>
29 #include <linux/lustre_idl.h>
30 #include <linux/obd_class.h>
31 #include <linux/lustre_log.h>
32 #include <linux/lustre_cmobd.h>
33 #include <linux/lustre_fsfilt.h>
34 #include <linux/lustre_smfs.h>
36 #include "cmobd_internal.h"
38 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count, int pattern);
39 void lov_free_memmd(struct lov_stripe_md **lsmp);
40 int smfs_rec_unpack(struct smfs_proc_args *args, char *record, char **pbuf,
43 /* helper functions for cmobd to construct pseudo lsm */
44 int cmobd_dummy_lsm(struct lov_stripe_md **lsmp, int stripe_cnt,
45 struct obdo *oa, __u32 stripe_size)
50 rc = lov_alloc_memmd(lsmp, stripe_cnt, LOV_PATTERN_CMOBD);
54 for (i = 0; i < stripe_cnt; i++) {
55 (*lsmp)->lsm_oinfo[i].loi_id = oa->o_id;
56 (*lsmp)->lsm_object_id = oa->o_id;
57 if (oa->o_valid & OBD_MD_FLGROUP) {
58 (*lsmp)->lsm_oinfo[i].loi_gr = oa->o_gr;
59 (*lsmp)->lsm_object_gr = oa->o_gr;
61 (*lsmp)->lsm_oinfo[i].loi_ost_idx = i;
62 (*lsmp)->lsm_stripe_size = stripe_size;
67 void cmobd_free_lsm(struct lov_stripe_md **lsmp)
74 /* reintegration functions */
75 int cmobd_reint_setattr(struct obd_device *obd, void *rec)
77 struct cache_manager_obd *cmobd = &obd->u.cmobd;
78 struct lov_obd *lov = &cmobd->cm_master_obd->u.lov;
79 struct obd_export *exp = cmobd->cm_master_exp;
80 struct lov_stripe_md *lsm;
81 struct obdo *oa = (struct obdo*)rec;
85 rc = cmobd_dummy_lsm(&lsm, lov->desc.ld_tgt_count, oa,
86 (__u32)lov->desc.ld_default_stripe_size);
90 rc = obd_setattr(exp, oa, lsm, NULL);
97 int cmobd_reint_create(struct obd_device *obd, void *rec)
99 struct cache_manager_obd *cmobd = &obd->u.cmobd;
100 struct lov_obd *lov = &cmobd->cm_master_obd->u.lov;
101 struct obd_export *exp = cmobd->cm_master_exp;
102 struct lov_stripe_md *lsm;
103 struct obd_trans_info oti = { 0 };
104 struct obdo *oa=(struct obdo*)rec;
108 rc = cmobd_dummy_lsm(&lsm, lov->desc.ld_tgt_count, oa,
109 (__u32)lov->desc.ld_default_stripe_size);
112 if (cmobd->cm_master_group != oa->o_gr) {
113 int group = oa->o_gr;
114 int valsize = sizeof(group);
115 rc = obd_set_info(exp, strlen("mds_conn"), "mds_conn",
118 GOTO(out, rc = -EINVAL);
119 cmobd->cm_master_group = oa->o_gr;
121 rc = obd_create(exp, oa, &lsm, &oti);
123 cmobd_free_lsm(&lsm);
128 /* direct cut-n-paste of filter_blocking_ast() */
129 static int cache_blocking_ast(struct ldlm_lock *lock,
130 struct ldlm_lock_desc *desc,
131 void *data, int flag)
136 if (flag == LDLM_CB_CANCELING) {
137 /* Don't need to do anything here. */
141 /* XXX layering violation! -phil */
142 l_lock(&lock->l_resource->lr_namespace->ns_lock);
143 /* Get this: if filter_blocking_ast is racing with ldlm_intent_policy,
144 * such that filter_blocking_ast is called just before l_i_p takes the
145 * ns_lock, then by the time we get the lock, we might not be the
146 * correct blocking function anymore. So check, and return early, if
148 if (lock->l_blocking_ast != cache_blocking_ast) {
149 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
153 lock->l_flags |= LDLM_FL_CBPENDING;
154 do_ast = (!lock->l_readers && !lock->l_writers);
155 l_unlock(&lock->l_resource->lr_namespace->ns_lock);
158 struct lustre_handle lockh;
161 LDLM_DEBUG(lock, "already unused, calling ldlm_cli_cancel");
162 ldlm_lock2handle(lock, &lockh);
163 rc = ldlm_cli_cancel(&lockh);
165 CERROR("ldlm_cli_cancel: %d\n", rc);
167 LDLM_DEBUG(lock, "Lock still has references, will be "
173 static int master_blocking_ast(struct ldlm_lock *lock,
174 struct ldlm_lock_desc *desc,
175 void *data, int flag)
178 struct lustre_handle lockh;
182 case LDLM_CB_BLOCKING:
183 ldlm_lock2handle(lock, &lockh);
184 rc = ldlm_cli_cancel(&lockh);
186 CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
190 case LDLM_CB_CANCELING:
191 /* do nothing here by now */
199 static int cmobd_write_extents(struct obd_device *obd, struct obdo *oa,
200 struct ldlm_extent *extent)
202 struct cache_manager_obd *cmobd = &obd->u.cmobd;
203 struct obd_device *cache = cmobd->cm_cache_obd;
204 struct lov_obd *lov = &cmobd->cm_master_obd->u.lov;
205 struct ldlm_res_id res_id;
206 ldlm_policy_data_t policy;
207 struct lustre_handle lockh_src = { 0 };
208 struct lustre_handle lockh_dst = { 0 };
209 struct lov_stripe_md *lsm;
210 int flags = 0, err, rc = 0;
213 /* XXX for debug write replay without smfs and kml */
214 res_id.name[0]= oa->o_id;
215 res_id.name[1]= oa->o_gr;
216 policy.l_extent.start = extent->start;
217 policy.l_extent.end = extent->end;
219 /* get extent read lock on the source replay file */
220 rc = ldlm_cli_enqueue(NULL, NULL, cache->obd_namespace, res_id,
221 LDLM_EXTENT, &policy, LCK_PR,
222 &flags, cache_blocking_ast, ldlm_completion_ast,
223 NULL, NULL, NULL, 0, NULL, &lockh_src);
227 /* construct the pseudo lsm */
228 rc = cmobd_dummy_lsm(&lsm, lov->desc.ld_tgt_count, oa,
229 (__u32)lov->desc.ld_default_stripe_size);
233 rc = obd_enqueue(cmobd->cm_master_exp, lsm, LDLM_EXTENT, &policy,
234 LCK_PW, &flags, master_blocking_ast,
235 ldlm_completion_ast, NULL,
236 NULL, 0, NULL, &lockh_dst);
240 err = cmobd_replay_write(obd, oa, &policy.l_extent);
242 rc = obd_cancel(cmobd->cm_master_exp, lsm, LCK_PW, &lockh_dst);
245 /* XXX in fact, I just want to cancel the only lockh_dst
247 rc = obd_cancel_unused(cmobd->cm_master_exp, lsm, 0, NULL);
251 cmobd_free_lsm(&lsm);
253 ldlm_lock_decref(&lockh_src, LCK_PR);
256 int cmobd_reint_write(struct obd_device *obd, void *rec)
258 struct cache_manager_obd *cmobd = &obd->u.cmobd;
259 struct obd_device *cache = cmobd->cm_cache_obd;
260 struct obdo *oa = (struct obdo *)rec;
261 struct ldlm_extent *extent = NULL;
262 unsigned long csb, ino;
263 char *extents_buf = NULL;
264 int size = 0, rc = 0, ext_num = 0;
269 obd_get_info(cmobd->cm_cache_exp, strlen("cache_sb") + 1,
270 "cache_sb", &size, &csb);
272 ino = *(int*)(&oa->o_inline[0]);
273 rc = fsfilt_get_ino_write_extents(cache, (struct super_block *)csb, ino,
274 &extents_buf, &ext_num);
277 extent = (struct ldlm_extent *)extents_buf;
279 while (extent && size --) {
280 rc = cmobd_write_extents(obd, oa, extent);
287 fsfilt_free_write_extents(cache, (struct super_block *)csb, ino,
288 extents_buf, ext_num);