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  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_capa.c
  37  *
  38  * Lustre Metadata Target (mdt) capability key read/write/update.
  39  *
  40  * Author: Lai Siyao <lsy@clusterfs.com>
  41  */
  42
  43 #ifndef EXPORT_SYMTAB
  44 # define EXPORT_SYMTAB
  45 #endif
  46 #define DEBUG_SUBSYSTEM S_MDS
  47
  48 #include "mdt_internal.h"
  49
  50 static inline void set_capa_key_expiry(struct mdt_device *mdt)
  51 {
  52         mdt->mdt_ck_expiry = jiffies + mdt->mdt_ck_timeout * HZ;
  53 }
  54
  55 static void make_capa_key(struct lustre_capa_key *key,
  56                           mdsno_t mdsnum, int keyid)
  57 {
  58         key->lk_mdsid = mdsnum;
  59         key->lk_keyid = keyid + 1;
  60         ll_get_random_bytes(key->lk_key, sizeof(key->lk_key));
  61 }
  62
  63 enum {
  64         MDT_TXN_CAPA_KEYS_WRITE_CREDITS = 1
  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_mdsid   = cpu_to_le64(src->lk_mdsid);
  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_mdsid   = le64_to_cpu(src->lk_mdsid);
  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
  97         th = mdt_trans_start(env, mdt, MDT_TXN_CAPA_KEYS_WRITE_CREDITS);
  98         if (IS_ERR(th))
  99                 RETURN(PTR_ERR(th));
 100
 101         tmp = &mti->mti_capa_key;
 102
 103         for (i = 0; i < 2; i++) {
 104                 lck_cpu_to_le(tmp, &keys[i]);
 105
 106                 rc = mdt_record_write(env, mdt->mdt_ck_obj,
 107                                       mdt_buf_const(env, tmp, sizeof(*tmp)),
 108                                       &off, th);
 109                 if (rc)
 110                         break;
 111         }
 112
 113         mdt_trans_stop(env, mdt, th);
 114
 115         CDEBUG(D_INFO, "write capability keys rc = %d:\n", rc);
 116         return rc;
 117 }
 118
 119 static int read_capa_keys(const struct lu_env *env,
 120                           struct mdt_device *mdt,
 121                           struct lustre_capa_key *keys)
 122 {
 123         struct mdt_thread_info *mti;
 124         struct lustre_capa_key *tmp;
 125         loff_t off = 0;
 126         int i, rc;
 127
 128         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 129         tmp = &mti->mti_capa_key;
 130
 131         for (i = 0; i < 2; i++) {
 132                 rc = mdt_record_read(env, mdt->mdt_ck_obj,
 133                                      mdt_buf(env, tmp, sizeof(*tmp)), &off);
 134                 if (rc)
 135                         return rc;
 136
 137                 lck_le_to_cpu(&keys[i], tmp);
 138                 DEBUG_CAPA_KEY(D_SEC, &keys[i], "read");
 139         }
 140
 141         return 0;
 142 }
 143
 144 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt)
 145 {
 146         struct lustre_capa_key  *keys = mdt->mdt_capa_keys;
 147         struct mdt_thread_info  *mti;
 148         struct dt_object        *obj;
 149         struct lu_attr          *la;
 150         mdsno_t                  mdsnum;
 151         unsigned long            size;
 152         int                      rc;
 153         ENTRY;
 154
 155         mdsnum = mdt_md_site(mdt)->ms_node_id;
 156
 157         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
 158         LASSERT(mti != NULL);
 159         la = &mti->mti_attr.ma_attr;
 160
 161         obj = mdt->mdt_ck_obj;
 162         rc = obj->do_ops->do_attr_get(env, mdt->mdt_ck_obj, la, BYPASS_CAPA);
 163         if (rc)
 164                 RETURN(rc);
 165
 166         size = (unsigned long)la->la_size;
 167         if (size == 0) {
 168                 int i;
 169
 170                 for (i = 0; i < 2; i++) {
 171                         make_capa_key(&keys[i], mdsnum, i);
 172                         DEBUG_CAPA_KEY(D_SEC, &keys[i], "initializing");
 173                 }
 174
 175                 rc = write_capa_keys(env, mdt, keys);
 176                 if (rc) {
 177                         CERROR("error writing MDS %s: rc %d\n", CAPA_KEYS, rc);
 178                         RETURN(rc);
 179                 }
 180         } else {
 181                 rc = read_capa_keys(env, mdt, keys);
 182                 if (rc) {
 183                         CERROR("error reading MDS %s: rc %d\n", CAPA_KEYS, rc);
 184                         RETURN(rc);
 185                 }
 186         }
 187         set_capa_key_expiry(mdt);
 188         cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
 189         CDEBUG(D_SEC, "mds_ck_timer %lu\n", mdt->mdt_ck_expiry);
 190         RETURN(0);
 191 }
 192
 193 void mdt_ck_timer_callback(unsigned long castmeharder)
 194 {
 195         struct mdt_device *mdt = (struct mdt_device *)castmeharder;
 196         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 197
 198         ENTRY;
 199         thread->t_flags |= SVC_EVENT;
 200         cfs_waitq_signal(&thread->t_ctl_waitq);
 201         EXIT;
 202 }
 203
 204 static int mdt_ck_thread_main(void *args)
 205 {
 206         struct mdt_device      *mdt = args;
 207         struct ptlrpc_thread   *thread = &mdt->mdt_ck_thread;
 208         struct lustre_capa_key *bkey = &mdt->mdt_capa_keys[0],
 209                                *rkey = &mdt->mdt_capa_keys[1];
 210         struct lustre_capa_key *tmp;
 211         struct lu_env           env;
 212         struct mdt_thread_info *info;
 213         struct md_device       *next;
 214         struct l_wait_info      lwi = { 0 };
 215         mdsno_t                 mdsnum;
 216         int                     rc;
 217         ENTRY;
 218
 219         ptlrpc_daemonize("mdt_ck");
 220         cfs_block_allsigs();
 221
 222         thread->t_flags = SVC_RUNNING;
 223         cfs_waitq_signal(&thread->t_ctl_waitq);
 224
 225         rc = lu_env_init(&env, LCT_MD_THREAD|LCT_REMEMBER|LCT_NOREF);
 226         if (rc)
 227                 RETURN(rc);
 228
 229         thread->t_env = &env;
 230         env.le_ctx.lc_thread = thread;
 231         env.le_ctx.lc_cookie = 0x1;
 232
 233         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
 234         LASSERT(info != NULL);
 235
 236         tmp = &info->mti_capa_key;
 237         mdsnum = mdt_md_site(mdt)->ms_node_id;
 238         while (1) {
 239                 l_wait_event(thread->t_ctl_waitq,
 240                              thread->t_flags & (SVC_STOPPING | SVC_EVENT),
 241                              &lwi);
 242
 243                 if (thread->t_flags & SVC_STOPPING)
 244                         break;
 245                 thread->t_flags &= ~SVC_EVENT;
 246
 247                 if (cfs_time_before(cfs_time_current(), mdt->mdt_ck_expiry))
 248                         break;
 249
 250                 *tmp = *rkey;
 251                 make_capa_key(tmp, mdsnum, rkey->lk_keyid);
 252
 253                 next = mdt->mdt_child;
 254                 rc = next->md_ops->mdo_update_capa_key(&env, next, tmp);
 255                 if (!rc) {
 256                         spin_lock(&capa_lock);
 257                         *bkey = *rkey;
 258                         *rkey = *tmp;
 259                         spin_unlock(&capa_lock);
 260
 261                         rc = write_capa_keys(&env, mdt, mdt->mdt_capa_keys);
 262                         if (rc) {
 263                                 spin_lock(&capa_lock);
 264                                 *rkey = *bkey;
 265                                 memset(bkey, 0, sizeof(*bkey));
 266                                 spin_unlock(&capa_lock);
 267                         } else {
 268                                 set_capa_key_expiry(mdt);
 269                                 DEBUG_CAPA_KEY(D_SEC, rkey, "new");
 270                         }
 271                 }
 272                 if (rc) {
 273                         DEBUG_CAPA_KEY(D_ERROR, rkey, "update failed for");
 274                         /* next retry is in 300 sec */
 275                         mdt->mdt_ck_expiry = jiffies + 300 * HZ;
 276                 }
 277
 278                 cfs_timer_arm(&mdt->mdt_ck_timer, mdt->mdt_ck_expiry);
 279                 CDEBUG(D_SEC, "mdt_ck_timer %lu\n", mdt->mdt_ck_expiry);
 280         }
 281         lu_env_fini(&env);
 282
 283         thread->t_flags = SVC_STOPPED;
 284         cfs_waitq_signal(&thread->t_ctl_waitq);
 285         RETURN(0);
 286 }
 287
 288 int mdt_ck_thread_start(struct mdt_device *mdt)
 289 {
 290         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 291         int rc;
 292
 293         cfs_waitq_init(&thread->t_ctl_waitq);
 294         rc = cfs_kernel_thread(mdt_ck_thread_main, mdt,
 295                            (CLONE_VM | CLONE_FILES));
 296         if (rc < 0) {
 297                 CERROR("cannot start mdt_ck thread, rc = %d\n", rc);
 298                 return rc;
 299         }
 300
 301         cfs_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING);
 302         return 0;
 303 }
 304
 305 void mdt_ck_thread_stop(struct mdt_device *mdt)
 306 {
 307         struct ptlrpc_thread *thread = &mdt->mdt_ck_thread;
 308
 309         if (!(thread->t_flags & SVC_RUNNING))
 310                 return;
 311
 312         thread->t_flags = SVC_STOPPING;
 313         cfs_waitq_signal(&thread->t_ctl_waitq);
 314         cfs_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
 315 }