fs/dlm/rcom.c

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 #include "dlm_internal.h"
  15 #include "lockspace.h"
  16 #include "member.h"
  17 #include "lowcomms.h"
  18 #include "midcomms.h"
  19 #include "rcom.h"
  20 #include "recover.h"
  21 #include "dir.h"
  22 #include "config.h"
  23 #include "memory.h"
  24 #include "lock.h"
  25 #include "util.h"
  26
  27
  28 static int rcom_response(struct dlm_ls *ls)
  29 {
  30         return test_bit(LSFL_RCOM_READY, &ls->ls_flags);
  31 }
  32
  33 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len,
  34                        struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret)
  35 {
  36         struct dlm_rcom *rc;
  37         struct dlm_mhandle *mh;
  38         char *mb;
  39         int mb_len = sizeof(struct dlm_rcom) + len;
  40
  41         mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb);
  42         if (!mh) {
  43                 log_print("create_rcom to %d type %d len %d ENOBUFS",
  44                           to_nodeid, type, len);
  45                 return -ENOBUFS;
  46         }
  47         memset(mb, 0, mb_len);
  48
  49         rc = (struct dlm_rcom *) mb;
  50
  51         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
  52         rc->rc_header.h_lockspace = ls->ls_global_id;
  53         rc->rc_header.h_nodeid = dlm_our_nodeid();
  54         rc->rc_header.h_length = mb_len;
  55         rc->rc_header.h_cmd = DLM_RCOM;
  56
  57         rc->rc_type = type;
  58
  59         spin_lock(&ls->ls_recover_lock);
  60         rc->rc_seq = ls->ls_recover_seq;
  61         spin_unlock(&ls->ls_recover_lock);
  62
  63         *mh_ret = mh;
  64         *rc_ret = rc;
  65         return 0;
  66 }
  67
  68 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh,
  69                       struct dlm_rcom *rc)
  70 {
  71         dlm_rcom_out(rc);
  72         dlm_lowcomms_commit_buffer(mh);
  73 }
  74
  75 /* When replying to a status request, a node also sends back its
  76    configuration values.  The requesting node then checks that the remote
  77    node is configured the same way as itself. */
  78
  79 static void make_config(struct dlm_ls *ls, struct rcom_config *rf)
  80 {
  81         rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen);
  82         rf->rf_lsflags = cpu_to_le32(ls->ls_exflags);
  83 }
  84
  85 static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
  86 {
  87         struct rcom_config *rf = (struct rcom_config *) rc->rc_buf;
  88
  89         if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) {
  90                 log_error(ls, "version mismatch: %x nodeid %d: %x",
  91                           DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid,
  92                           rc->rc_header.h_version);
  93                 return -EPROTO;
  94         }
  95
  96         if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen ||
  97             le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) {
  98                 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x",
  99                           ls->ls_lvblen, ls->ls_exflags, nodeid,
 100                           le32_to_cpu(rf->rf_lvblen),
 101                           le32_to_cpu(rf->rf_lsflags));
 102                 return -EPROTO;
 103         }
 104         return 0;
 105 }
 106
 107 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq)
 108 {
 109         spin_lock(&ls->ls_rcom_spin);
 110         *new_seq = ++ls->ls_rcom_seq;
 111         set_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 112         spin_unlock(&ls->ls_rcom_spin);
 113 }
 114
 115 static void disallow_sync_reply(struct dlm_ls *ls)
 116 {
 117         spin_lock(&ls->ls_rcom_spin);
 118         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 119         clear_bit(LSFL_RCOM_READY, &ls->ls_flags);
 120         spin_unlock(&ls->ls_rcom_spin);
 121 }
 122
 123 int dlm_rcom_status(struct dlm_ls *ls, int nodeid)
 124 {
 125         struct dlm_rcom *rc;
 126         struct dlm_mhandle *mh;
 127         int error = 0;
 128
 129         ls->ls_recover_nodeid = nodeid;
 130
 131         if (nodeid == dlm_our_nodeid()) {
 132                 rc = (struct dlm_rcom *) ls->ls_recover_buf;
 133                 rc->rc_result = dlm_recover_status(ls);
 134                 goto out;
 135         }
 136
 137         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh);
 138         if (error)
 139                 goto out;
 140
 141         allow_sync_reply(ls, &rc->rc_id);
 142         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
 143
 144         send_rcom(ls, mh, rc);
 145
 146         error = dlm_wait_function(ls, &rcom_response);
 147         disallow_sync_reply(ls);
 148         if (error)
 149                 goto out;
 150
 151         rc = (struct dlm_rcom *) ls->ls_recover_buf;
 152
 153         if (rc->rc_result == -ESRCH) {
 154                 /* we pretend the remote lockspace exists with 0 status */
 155                 log_debug(ls, "remote node %d not ready", nodeid);
 156                 rc->rc_result = 0;
 157         } else
 158                 error = check_config(ls, rc, nodeid);
 159         /* the caller looks at rc_result for the remote recovery status */
 160  out:
 161         return error;
 162 }
 163
 164 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 165 {
 166         struct dlm_rcom *rc;
 167         struct dlm_mhandle *mh;
 168         int error, nodeid = rc_in->rc_header.h_nodeid;
 169
 170         error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY,
 171                             sizeof(struct rcom_config), &rc, &mh);
 172         if (error)
 173                 return;
 174         rc->rc_id = rc_in->rc_id;
 175         rc->rc_seq_reply = rc_in->rc_seq;
 176         rc->rc_result = dlm_recover_status(ls);
 177         make_config(ls, (struct rcom_config *) rc->rc_buf);
 178
 179         send_rcom(ls, mh, rc);
 180 }
 181
 182 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 183 {
 184         spin_lock(&ls->ls_rcom_spin);
 185         if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) ||
 186             rc_in->rc_id != ls->ls_rcom_seq) {
 187                 log_debug(ls, "reject reply %d from %d seq %llx expect %llx",
 188                           rc_in->rc_type, rc_in->rc_header.h_nodeid,
 189                           (unsigned long long)rc_in->rc_id,
 190                           (unsigned long long)ls->ls_rcom_seq);
 191                 goto out;
 192         }
 193         memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length);
 194         set_bit(LSFL_RCOM_READY, &ls->ls_flags);
 195         clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags);
 196         wake_up(&ls->ls_wait_general);
 197  out:
 198         spin_unlock(&ls->ls_rcom_spin);
 199 }
 200
 201 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len)
 202 {
 203         struct dlm_rcom *rc;
 204         struct dlm_mhandle *mh;
 205         int error = 0, len = sizeof(struct dlm_rcom);
 206
 207         ls->ls_recover_nodeid = nodeid;
 208
 209         if (nodeid == dlm_our_nodeid()) {
 210                 dlm_copy_master_names(ls, last_name, last_len,
 211                                       ls->ls_recover_buf + len,
 212                                       dlm_config.ci_buffer_size - len, nodeid);
 213                 goto out;
 214         }
 215
 216         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh);
 217         if (error)
 218                 goto out;
 219         memcpy(rc->rc_buf, last_name, last_len);
 220
 221         allow_sync_reply(ls, &rc->rc_id);
 222         memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size);
 223
 224         send_rcom(ls, mh, rc);
 225
 226         error = dlm_wait_function(ls, &rcom_response);
 227         disallow_sync_reply(ls);
 228  out:
 229         return error;
 230 }
 231
 232 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 233 {
 234         struct dlm_rcom *rc;
 235         struct dlm_mhandle *mh;
 236         int error, inlen, outlen, nodeid;
 237
 238         nodeid = rc_in->rc_header.h_nodeid;
 239         inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
 240         outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom);
 241
 242         error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh);
 243         if (error)
 244                 return;
 245         rc->rc_id = rc_in->rc_id;
 246         rc->rc_seq_reply = rc_in->rc_seq;
 247
 248         dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen,
 249                               nodeid);
 250         send_rcom(ls, mh, rc);
 251 }
 252
 253 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid)
 254 {
 255         struct dlm_rcom *rc;
 256         struct dlm_mhandle *mh;
 257         struct dlm_ls *ls = r->res_ls;
 258         int error;
 259
 260         error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length,
 261                             &rc, &mh);
 262         if (error)
 263                 goto out;
 264         memcpy(rc->rc_buf, r->res_name, r->res_length);
 265         rc->rc_id = (unsigned long) r;
 266
 267         send_rcom(ls, mh, rc);
 268  out:
 269         return error;
 270 }
 271
 272 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 273 {
 274         struct dlm_rcom *rc;
 275         struct dlm_mhandle *mh;
 276         int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid;
 277         int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom);
 278
 279         error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh);
 280         if (error)
 281                 return;
 282
 283         error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid);
 284         if (error)
 285                 ret_nodeid = error;
 286         rc->rc_result = ret_nodeid;
 287         rc->rc_id = rc_in->rc_id;
 288         rc->rc_seq_reply = rc_in->rc_seq;
 289
 290         send_rcom(ls, mh, rc);
 291 }
 292
 293 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 294 {
 295         dlm_recover_master_reply(ls, rc_in);
 296 }
 297
 298 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb,
 299                            struct rcom_lock *rl)
 300 {
 301         memset(rl, 0, sizeof(*rl));
 302
 303         rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid);
 304         rl->rl_lkid = cpu_to_le32(lkb->lkb_id);
 305         rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags);
 306         rl->rl_flags = cpu_to_le32(lkb->lkb_flags);
 307         rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq);
 308         rl->rl_rqmode = lkb->lkb_rqmode;
 309         rl->rl_grmode = lkb->lkb_grmode;
 310         rl->rl_status = lkb->lkb_status;
 311         rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type);
 312
 313         if (lkb->lkb_bastaddr)
 314                 rl->rl_asts |= AST_BAST;
 315         if (lkb->lkb_astaddr)
 316                 rl->rl_asts |= AST_COMP;
 317
 318         rl->rl_namelen = cpu_to_le16(r->res_length);
 319         memcpy(rl->rl_name, r->res_name, r->res_length);
 320
 321         /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ?
 322            If so, receive_rcom_lock_args() won't take this copy. */
 323
 324         if (lkb->lkb_lvbptr)
 325                 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen);
 326 }
 327
 328 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb)
 329 {
 330         struct dlm_ls *ls = r->res_ls;
 331         struct dlm_rcom *rc;
 332         struct dlm_mhandle *mh;
 333         struct rcom_lock *rl;
 334         int error, len = sizeof(struct rcom_lock);
 335
 336         if (lkb->lkb_lvbptr)
 337                 len += ls->ls_lvblen;
 338
 339         error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh);
 340         if (error)
 341                 goto out;
 342
 343         rl = (struct rcom_lock *) rc->rc_buf;
 344         pack_rcom_lock(r, lkb, rl);
 345         rc->rc_id = (unsigned long) r;
 346
 347         send_rcom(ls, mh, rc);
 348  out:
 349         return error;
 350 }
 351
 352 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in)
 353 {
 354         struct dlm_rcom *rc;
 355         struct dlm_mhandle *mh;
 356         int error, nodeid = rc_in->rc_header.h_nodeid;
 357
 358         dlm_recover_master_copy(ls, rc_in);
 359
 360         error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY,
 361                             sizeof(struct rcom_lock), &rc, &mh);
 362         if (error)
 363                 return;
 364
 365         /* We send back the same rcom_lock struct we received, but
 366            dlm_recover_master_copy() has filled in rl_remid and rl_result */
 367
 368         memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock));
 369         rc->rc_id = rc_in->rc_id;
 370         rc->rc_seq_reply = rc_in->rc_seq;
 371
 372         send_rcom(ls, mh, rc);
 373 }
 374
 375 /* If the lockspace doesn't exist then still send a status message
 376    back; it's possible that it just doesn't have its global_id yet. */
 377
 378 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in)
 379 {
 380         struct dlm_rcom *rc;
 381         struct rcom_config *rf;
 382         struct dlm_mhandle *mh;
 383         char *mb;
 384         int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config);
 385
 386         mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb);
 387         if (!mh)
 388                 return -ENOBUFS;
 389         memset(mb, 0, mb_len);
 390
 391         rc = (struct dlm_rcom *) mb;
 392
 393         rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
 394         rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace;
 395         rc->rc_header.h_nodeid = dlm_our_nodeid();
 396         rc->rc_header.h_length = mb_len;
 397         rc->rc_header.h_cmd = DLM_RCOM;
 398
 399         rc->rc_type = DLM_RCOM_STATUS_REPLY;
 400         rc->rc_id = rc_in->rc_id;
 401         rc->rc_seq_reply = rc_in->rc_seq;
 402         rc->rc_result = -ESRCH;
 403
 404         rf = (struct rcom_config *) rc->rc_buf;
 405         rf->rf_lvblen = cpu_to_le32(~0U);
 406
 407         dlm_rcom_out(rc);
 408         dlm_lowcomms_commit_buffer(mh);
 409
 410         return 0;
 411 }
 412
 413 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
 414 {
 415         uint64_t seq;
 416         int rv = 0;
 417
 418         switch (rc->rc_type) {
 419         case DLM_RCOM_STATUS_REPLY:
 420         case DLM_RCOM_NAMES_REPLY:
 421         case DLM_RCOM_LOOKUP_REPLY:
 422         case DLM_RCOM_LOCK_REPLY:
 423                 spin_lock(&ls->ls_recover_lock);
 424                 seq = ls->ls_recover_seq;
 425                 spin_unlock(&ls->ls_recover_lock);
 426                 if (rc->rc_seq_reply != seq) {
 427                         log_debug(ls, "ignoring old reply %x from %d "
 428                                       "seq_reply %llx expect %llx",
 429                                       rc->rc_type, rc->rc_header.h_nodeid,
 430                                       (unsigned long long)rc->rc_seq_reply,
 431                                       (unsigned long long)seq);
 432                         rv = 1;
 433                 }
 434         }
 435         return rv;
 436 }
 437
 438 /* Called by dlm_recv; corresponds to dlm_receive_message() but special
 439    recovery-only comms are sent through here. */
 440
 441 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid)
 442 {
 443         if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) {
 444                 log_debug(ls, "ignoring recovery message %x from %d",
 445                           rc->rc_type, nodeid);
 446                 goto out;
 447         }
 448
 449         if (is_old_reply(ls, rc))
 450                 goto out;
 451
 452         switch (rc->rc_type) {
 453         case DLM_RCOM_STATUS:
 454                 receive_rcom_status(ls, rc);
 455                 break;
 456
 457         case DLM_RCOM_NAMES:
 458                 receive_rcom_names(ls, rc);
 459                 break;
 460
 461         case DLM_RCOM_LOOKUP:
 462                 receive_rcom_lookup(ls, rc);
 463                 break;
 464
 465         case DLM_RCOM_LOCK:
 466                 receive_rcom_lock(ls, rc);
 467                 break;
 468
 469         case DLM_RCOM_STATUS_REPLY:
 470                 receive_sync_reply(ls, rc);
 471                 break;
 472
 473         case DLM_RCOM_NAMES_REPLY:
 474                 receive_sync_reply(ls, rc);
 475                 break;
 476
 477         case DLM_RCOM_LOOKUP_REPLY:
 478                 receive_rcom_lookup_reply(ls, rc);
 479                 break;
 480
 481         case DLM_RCOM_LOCK_REPLY:
 482                 dlm_recover_process_copy(ls, rc);
 483                 break;
 484
 485         default:
 486                 log_error(ls, "receive_rcom bad type %d", rc->rc_type);
 487         }
 488  out:
 489         return;
 490 }
 491