lustre/mdt/mdt_capa.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   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.
  11  *
  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).
  17  *
  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
  21  *
  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
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  * Copyright (c) 2011 Whamcloud, Inc.
  32  */
  33 /*
  34  * Copyright (c) 2011 Whamcloud, Inc.
  35  */
  36 /*
  37  * This file is part of Lustre, http://www.lustre.org/
  38  * Lustre is a trademark of Sun Microsystems, Inc.
  39  *
  40  * lustre/mdt/mdt_capa.c
  41  *
  42  * Lustre Metadata Target (mdt) capability key read/write/update.
  43  *
  44  * Author: Lai Siyao <lsy@clusterfs.com>
  45  */
  46
  47 #ifndef EXPORT_SYMTAB
  48 # define EXPORT_SYMTAB
  49 #endif
  50 #define DEBUG_SUBSYSTEM S_MDS
  51
  52 #include "mdt_internal.h"
  53
  54 static inline void set_capa_key_expiry(struct mdt_device *mdt)
  55 {
  56         mdt->mdt_ck_expiry = jiffies + mdt->mdt_ck_timeout * CFS_HZ;
  57 }
  58
  59 static void make_capa_key(struct lustre_capa_key *key,
  60                           mdsno_t mdsnum, int keyid)
  61 {
  62         key->lk_seq = mdsnum;
  63         key->lk_keyid = keyid + 1;
  64         cfs_get_random_bytes(key->lk_key, sizeof(key->lk_key));
  65 }
  66
  67 static inline void lck_cpu_to_le(struct lustre_capa_key *tgt,
  68                                  struct lustre_capa_key *src)
  69 {
  70         tgt->lk_seq   = cpu_to_le64(src->lk_seq);
  71         tgt->lk_keyid   = cpu_to_le32(src->lk_keyid);
  72         tgt->lk_padding = cpu_to_le32(src->lk_padding);
  73         memcpy(tgt->lk_key, src->lk_key, sizeof(src->lk_key));
  74 }
  75
  76 static inline void lck_le_to_cpu(struct lustre_capa_key *tgt,
  77                                  struct lustre_capa_key *src)
  78 {
  79         tgt->lk_seq   = le64_to_cpu(src->lk_seq);
  80         tgt->lk_keyid   = le32_to_cpu(src->lk_keyid);
  81         tgt->lk_padding = le32_to_cpu(src->lk_padding);
  82         memcpy(tgt->lk_key, src->lk_key, sizeof(src->lk_key));
  83 }
  84
  85 static int write_capa_keys(const struct lu_env *env,
  86                            struct mdt_device *mdt,
  87                            struct lustre_capa_key *keys)
  88 {
  89         struct mdt_thread_info *mti;
  90         struct lustre_capa_key *tmp;
  91         struct thandle *th;
  92         loff_t off = 0;
  93         int i, rc;
  94
  95         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
  96         th = mdt_trans_create(env, mdt);
  97         if (IS_ERR(th))
  98                 RETURN(PTR_ERR(th));
  99
 100         rc = dt_declare_record_write(env, mdt->mdt_ck_obj,
 101                                      sizeof(*tmp) * 3, 0, th);
 102         if (rc)
 103                 goto stop;
 104
 105         rc = mdt_trans_start(env, mdt, th);
 106         if (rc)
 107                 goto stop;
 108
 109         tmp = &mti->mti_capa_key;
 110
 111         for (i = 0; i < 2; i++) {
 112                 lck_cpu_to_le(tmp, &keys[i]);
 113
 114                 rc = dt_record_write(env, mdt->mdt_ck_obj,
 115                                      mdt_buf_const(env, tmp, sizeof(*tmp)),
 116                                      &off, th);
 117                 if (rc)
 118                         break;
 119         }
 120
 121 stop:
 122         mdt_trans_stop(env, mdt, th);
 123
 124         CDEBUG(D_INFO, "write capability keys rc = %d:\n", rc);
 125         return rc;
 126 }
 127
 128 static int read_capa_keys(const struct lu_env *env,
 129                           struct mdt_device *mdt,
 130                           struct lustre_capa_key *keys)
 131 {
 132         struct mdt_thread_info *mti;
 133         struct lustre_capa_key *tmp;
 134         loff_t off = 0;
 135         int i, rc;
 136
 137         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 138         tmp = &mti->mti_capa_key;
 139
 140         for (i = 0; i < 2; i++) {
 141                 rc = dt_record_read(env, mdt->mdt_ck_obj,
 142                                     mdt_buf(env, tmp, sizeof(*tmp)), &off);
 143                 if (rc)
 144                         return rc;
 145
 146                 lck_le_to_cpu(&keys[i], tmp);
 147                 DEBUG_CAPA_KEY(D_SEC, &keys[i], "read");
 148         }
 149
 150         return 0;
 151 }
 152
 153 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt)
 154 {
 155         struct lustre_capa_key  *keys = mdt->mdt_capa_keys;
 156         struct mdt_thread_info  *mti;
 157         struct dt_object        *obj;
 158         struct lu_attr          *la;
 159         mdsno_t                  mdsnum;
 160         unsigned long            size;
 161         int                      rc;
 162         ENTRY;
 163
 164         mdsnum = mdt_md_site(mdt)->ms_node_id;
 165
 166         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 167         LASSERT(mti != NULL);
 168         la = &mti->mti_attr.ma_attr;
 169
 170         obj = mdt->mdt_ck_obj;
 171         rc = obj->do_ops->do_attr_get(env, mdt->mdt_ck_obj, la, BYPASS_CAPA);
 172         if (rc)
 173                 RETURN(rc);
 174
 175         size = (unsigned long)la->la_size;
 176         if (size == 0) {
 177                 int i;
 178
 179                 for (i = 0; i < 2; i++) {
 180                         make_capa_key(&keys[i], mdsnum, i);
 181                         DEBUG_CAPA_KEY(D_SEC, &keys[i], "initializing");
 182                 }
 183
 184                 rc = write_capa_keys(env, mdt, keys);
 185                 if (rc) {
 186                         CERROR("error writing MDS %s: rc %d\n", CAPA_KEYS, rc);
 187                         RETURN(rc);
 188                 }
 189         } else {
 190                 rc = read_capa_keys(env, mdt, keys);
 191                 if (rc) {
 192                         CERROR("error reading MDS %s: rc %d\n", CAPA_KEYS, rc);
 193                         RETURN(rc);
 194                 }
 195         }
 196         set_capa_key_expiry(mdt);
 197         cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
 198         CDEBUG(D_SEC, "mds_ck_timer %lu\n", mdt->mdt_ck_expiry);
 199         RETURN(0);
 200 }
 201
 202 void mdt_ck_timer_callback(unsigned long castmeharder)
 203 {
 204         struct mdt_device *mdt = (struct mdt_device *)castmeharder;
 205         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 206
 207         ENTRY;
 208         thread_add_flags(thread, SVC_EVENT);
 209         cfs_waitq_signal(&thread->t_ctl_waitq);
 210         EXIT;
 211 }
 212
 213 static int mdt_ck_thread_main(void *args)
 214 {
 215         struct mdt_device      *mdt = args;
 216         struct ptlrpc_thread   *thread = &mdt->mdt_ck_thread;
 217         struct lustre_capa_key *bkey = &mdt->mdt_capa_keys[0],
 218                                *rkey = &mdt->mdt_capa_keys[1];
 219         struct lustre_capa_key *tmp;
 220         struct lu_env           env;
 221         struct mdt_thread_info *info;
 222         struct md_device       *next;
 223         struct l_wait_info      lwi = { 0 };
 224         mdsno_t                 mdsnum;
 225         int                     rc;
 226         ENTRY;
 227
 228         cfs_daemonize_ctxt("mdt_ck");
 229         cfs_block_allsigs();
 230
 231         thread_set_flags(thread, SVC_RUNNING);
 232         cfs_waitq_signal(&thread->t_ctl_waitq);
 233
 234         rc = lu_env_init(&env, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
 235         if (rc)
 236                 RETURN(rc);
 237
 238         thread->t_env = &env;
 239         env.le_ctx.lc_thread = thread;
 240         env.le_ctx.lc_cookie = 0x1;
 241
 242         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
 243         LASSERT(info != NULL);
 244
 245         tmp = &info->mti_capa_key;
 246         mdsnum = mdt_md_site(mdt)->ms_node_id;
 247         while (1) {
 248                 l_wait_event(thread->t_ctl_waitq,
 249                              thread_is_stopping(thread) ||
 250                              thread_is_event(thread),
 251                              &lwi);
 252
 253                 if (thread_is_stopping(thread))
 254                         break;
 255                 thread_clear_flags(thread, SVC_EVENT);
 256
 257                 if (cfs_time_before(cfs_time_current(), mdt->mdt_ck_expiry))
 258                         break;
 259
 260                 *tmp = *rkey;
 261                 make_capa_key(tmp, mdsnum, rkey->lk_keyid);
 262
 263                 next = mdt->mdt_child;
 264                 rc = next->md_ops->mdo_update_capa_key(&env, next, tmp);
 265                 if (!rc) {
 266                         cfs_spin_lock(&capa_lock);
 267                         *bkey = *rkey;
 268                         *rkey = *tmp;
 269                         cfs_spin_unlock(&capa_lock);
 270
 271                         rc = write_capa_keys(&env, mdt, mdt->mdt_capa_keys);
 272                         if (rc) {
 273                                 cfs_spin_lock(&capa_lock);
 274                                 *rkey = *bkey;
 275                                 memset(bkey, 0, sizeof(*bkey));
 276                                 cfs_spin_unlock(&capa_lock);
 277                         } else {
 278                                 set_capa_key_expiry(mdt);
 279                                 DEBUG_CAPA_KEY(D_SEC, rkey, "new");
 280                         }
 281                 }
 282                 if (rc) {
 283                         DEBUG_CAPA_KEY(D_ERROR, rkey, "update failed for");
 284                         /* next retry is in 300 sec */
 285                         mdt->mdt_ck_expiry = jiffies + 300 * CFS_HZ;
 286                 }
 287
 288                 cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
 289                 CDEBUG(D_SEC, "mdt_ck_timer %lu\n", mdt->mdt_ck_expiry);
 290         }
 291         lu_env_fini(&env);
 292
 293         thread_set_flags(thread, SVC_STOPPED);
 294         cfs_waitq_signal(&thread->t_ctl_waitq);
 295         RETURN(0);
 296 }
 297
 298 int mdt_ck_thread_start(struct mdt_device *mdt)
 299 {
 300         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 301         int rc;
 302
 303         cfs_waitq_init(&thread->t_ctl_waitq);
 304         rc = cfs_create_thread(mdt_ck_thread_main, mdt, CFS_DAEMON_FLAGS);
 305         if (rc < 0) {
 306                 CERROR("cannot start mdt_ck thread, rc = %d\n", rc);
 307                 return rc;
 308         }
 309
 310         l_wait_condition(thread->t_ctl_waitq, thread_is_running(thread));
 311         return 0;
 312 }
 313
 314 void mdt_ck_thread_stop(struct mdt_device *mdt)
 315 {
 316         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 317
 318         if (!thread_is_running(thread))
 319                 return;
 320
 321         thread_set_flags(thread, SVC_STOPPING);
 322         cfs_waitq_signal(&thread->t_ctl_waitq);
 323         l_wait_condition(thread->t_ctl_waitq, thread_is_stopped(thread));
 324 }