lustre/mdc/mdc_reint.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
  37 #ifndef EXPORT_SYMTAB
  38 # define EXPORT_SYMTAB
  39 #endif
  40 #define DEBUG_SUBSYSTEM S_MDC
  41
  42 #ifdef __KERNEL__
  43 #ifndef AUTOCONF_INCLUDED
  44 # include <linux/config.h>
  45 #endif
  46 # include <linux/module.h>
  47 # include <linux/kernel.h>
  48 #else
  49 # include <liblustre.h>
  50 #endif
  51
  52 #include <obd_class.h>
  53 #include "mdc_internal.h"
  54 #include <lustre_fid.h>
  55
  56 /* mdc_setattr does its own semaphore handling */
  57 static int mdc_reint(struct ptlrpc_request *request,
  58                      struct mdc_rpc_lock *rpc_lock,
  59                      int level)
  60 {
  61         int rc;
  62
  63         request->rq_send_state = level;
  64
  65         mdc_get_rpc_lock(rpc_lock, NULL);
  66         rc = ptlrpc_queue_wait(request);
  67         mdc_put_rpc_lock(rpc_lock, NULL);
  68         if (rc)
  69                 CDEBUG(D_INFO, "error in handling %d\n", rc);
  70         else if (!req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY)) {
  71                 rc = -EPROTO;
  72         }
  73         return rc;
  74 }
  75
  76 /* Find and cancel locally locks matched by inode @bits & @mode in the resource
  77  * found by @fid. Found locks are added into @cancel list. Returns the amount of
  78  * locks added to @cancels list. */
  79 int mdc_resource_get_unused(struct obd_export *exp, struct lu_fid *fid,
  80                             struct list_head *cancels, ldlm_mode_t mode,
  81                             __u64 bits)
  82 {
  83         ldlm_policy_data_t policy = {{0}};
  84         struct ldlm_res_id res_id;
  85         struct ldlm_resource *res;
  86         int count;
  87         ENTRY;
  88
  89         fid_build_reg_res_name(fid, &res_id);
  90         res = ldlm_resource_get(exp->exp_obd->obd_namespace,
  91                                 NULL, &res_id, 0, 0);
  92         if (res == NULL)
  93                 RETURN(0);
  94         LDLM_RESOURCE_ADDREF(res);
  95         /* Initialize ibits lock policy. */
  96         policy.l_inodebits.bits = bits;
  97         count = ldlm_cancel_resource_local(res, cancels, &policy,
  98                                            mode, 0, 0, NULL);
  99         LDLM_RESOURCE_DELREF(res);
 100         ldlm_resource_putref(res);
 101         RETURN(count);
 102 }
 103
 104 static int mdc_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
 105                             struct list_head *cancels, int count)
 106 {
 107         return ldlm_prep_elc_req(exp, req, LUSTRE_MDS_VERSION, MDS_REINT,
 108                                  0, cancels, count);
 109 }
 110
 111 /* If mdc_setattr is called with an 'iattr', then it is a normal RPC that
 112  * should take the normal semaphore and go to the normal portal.
 113  *
 114  * If it is called with iattr->ia_valid & ATTR_FROM_OPEN, then it is a
 115  * magic open-path setattr that should take the setattr semaphore and
 116  * go to the setattr portal. */
 117 int mdc_setattr(struct obd_export *exp, struct md_op_data *op_data,
 118                 void *ea, int ealen, void *ea2, int ea2len,
 119                 struct ptlrpc_request **request, struct md_open_data **mod)
 120 {
 121         CFS_LIST_HEAD(cancels);
 122         struct ptlrpc_request *req;
 123         struct mdc_rpc_lock *rpc_lock;
 124         struct obd_device *obd = exp->exp_obd;
 125         int count = 0, rc;
 126         __u64 bits;
 127         ENTRY;
 128
 129         LASSERT(op_data != NULL);
 130
 131         bits = MDS_INODELOCK_UPDATE;
 132         if (op_data->op_attr.ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID))
 133                 bits |= MDS_INODELOCK_LOOKUP;
 134         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
 135             (fid_is_sane(&op_data->op_fid1)))
 136                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
 137                                                 &cancels, LCK_EX, bits);
 138         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 139                                    &RQF_MDS_REINT_SETATTR);
 140         if (req == NULL) {
 141                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 142                 RETURN(-ENOMEM);
 143         }
 144         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 145         if ((op_data->op_flags & (MF_SOM_CHANGE | MF_EPOCH_OPEN)) == 0)
 146                 req_capsule_set_size(&req->rq_pill, &RMF_MDT_EPOCH, RCL_CLIENT,
 147                                      0);
 148         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, ealen);
 149         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_CLIENT,
 150                              ea2len);
 151
 152         rc = mdc_prep_elc_req(exp, req, &cancels, count);
 153         if (rc) {
 154                 ptlrpc_request_free(req);
 155                 RETURN(rc);
 156         }
 157
 158         if (op_data->op_attr.ia_valid & ATTR_FROM_OPEN) {
 159                 req->rq_request_portal = MDS_SETATTR_PORTAL;
 160                 ptlrpc_at_set_req_timeout(req);
 161                 rpc_lock = obd->u.cli.cl_setattr_lock;
 162         } else {
 163                 rpc_lock = obd->u.cli.cl_rpc_lock;
 164         }
 165
 166         if (op_data->op_attr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
 167                 CDEBUG(D_INODE, "setting mtime "CFS_TIME_T
 168                        ", ctime "CFS_TIME_T"\n",
 169                        LTIME_S(op_data->op_attr.ia_mtime),
 170                        LTIME_S(op_data->op_attr.ia_ctime));
 171         mdc_setattr_pack(req, op_data, ea, ealen, ea2, ea2len);
 172
 173         ptlrpc_request_set_replen(req);
 174         if (mod && (op_data->op_flags & MF_EPOCH_OPEN) &&
 175             req->rq_import->imp_replayable)
 176         {
 177                 LASSERT(*mod == NULL);
 178
 179                 OBD_ALLOC_PTR(*mod);
 180                 if (*mod == NULL) {
 181                         DEBUG_REQ(D_ERROR, req, "Can't allocate "
 182                                   "md_open_data");
 183                 } else {
 184                         req->rq_replay = 1;
 185                         req->rq_cb_data = *mod;
 186                         (*mod)->mod_open_req = req;
 187                         req->rq_commit_cb = mdc_commit_open;
 188                 }
 189         }
 190
 191         rc = mdc_reint(req, rpc_lock, LUSTRE_IMP_FULL);
 192
 193         /* Save the obtained info in the original RPC for the replay case. */
 194         if (rc == 0 && (op_data->op_flags & MF_EPOCH_OPEN)) {
 195                 struct mdt_epoch *epoch;
 196                 struct mdt_body  *body;
 197
 198                 epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
 199                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 200                 LASSERT(epoch != NULL);
 201                 LASSERT(body != NULL);
 202                 epoch->handle = body->handle;
 203                 epoch->ioepoch = body->ioepoch;
 204                 req->rq_replay_cb = mdc_replay_open;
 205         }
 206         *request = req;
 207         if (rc == -ERESTARTSYS)
 208                 rc = 0;
 209         if (rc && req->rq_commit_cb)
 210                 req->rq_commit_cb(req);
 211         RETURN(rc);
 212 }
 213
 214 int mdc_create(struct obd_export *exp, struct md_op_data *op_data,
 215                const void *data, int datalen, int mode, __u32 uid, __u32 gid,
 216                cfs_cap_t cap_effective, __u64 rdev,
 217                struct ptlrpc_request **request)
 218 {
 219         struct ptlrpc_request *req;
 220         int level, rc;
 221         int count = 0;
 222         CFS_LIST_HEAD(cancels);
 223         ENTRY;
 224
 225         /* For case if upper layer did not alloc fid, do it now. */
 226         if (!fid_is_sane(&op_data->op_fid2)) {
 227                 /*
 228                  * mdc_fid_alloc() may return errno 1 in case of switch to new
 229                  * sequence, handle this.
 230                  */
 231                 rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 232                 if (rc < 0) {
 233                         CERROR("Can't alloc new fid, rc %d\n", rc);
 234                         RETURN(rc);
 235                 }
 236         }
 237
 238         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
 239             (fid_is_sane(&op_data->op_fid1)))
 240                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
 241                                                 &cancels, LCK_EX,
 242                                                 MDS_INODELOCK_UPDATE);
 243
 244         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 245                                    &RQF_MDS_REINT_CREATE_RMT_ACL);
 246         if (req == NULL) {
 247                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 248                 RETURN(-ENOMEM);
 249         }
 250         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 251         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 252                              op_data->op_namelen + 1);
 253         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 254                              data && datalen ? datalen : 0);
 255
 256         rc = mdc_prep_elc_req(exp, req, &cancels, count);
 257         if (rc) {
 258                 ptlrpc_request_free(req);
 259                 RETURN(rc);
 260         }
 261
 262         /*
 263          * mdc_create_pack() fills msg->bufs[1] with name and msg->bufs[2] with
 264          * tgt, for symlinks or lov MD data.
 265          */
 266         mdc_create_pack(req, op_data, data, datalen, mode, uid,
 267                         gid, cap_effective, rdev);
 268
 269         ptlrpc_request_set_replen(req);
 270
 271         level = LUSTRE_IMP_FULL;
 272  resend:
 273         rc = mdc_reint(req, exp->exp_obd->u.cli.cl_rpc_lock, level);
 274
 275         /* Resend if we were told to. */
 276         if (rc == -ERESTARTSYS) {
 277                 level = LUSTRE_IMP_RECOVER;
 278                 goto resend;
 279         } else if (rc == 0) {
 280                 struct mdt_body *body;
 281                 struct lustre_capa *capa;
 282
 283                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 284                 LASSERT(body);
 285                 if (body->valid & OBD_MD_FLMDSCAPA) {
 286                         capa = req_capsule_server_get(&req->rq_pill,
 287                                                       &RMF_CAPA1);
 288                         if (capa == NULL)
 289                                 rc = -EPROTO;
 290                 }
 291         }
 292
 293         *request = req;
 294         RETURN(rc);
 295 }
 296
 297 int mdc_unlink(struct obd_export *exp, struct md_op_data *op_data,
 298                struct ptlrpc_request **request)
 299 {
 300         CFS_LIST_HEAD(cancels);
 301         struct obd_device *obd = class_exp2obd(exp);
 302         struct ptlrpc_request *req = *request;
 303         int count = 0, rc;
 304         ENTRY;
 305
 306         LASSERT(req == NULL);
 307
 308         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
 309             (fid_is_sane(&op_data->op_fid1)))
 310                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
 311                                                 &cancels, LCK_EX,
 312                                                 MDS_INODELOCK_UPDATE);
 313         if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
 314             (fid_is_sane(&op_data->op_fid3)))
 315                 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
 316                                                  &cancels, LCK_EX,
 317                                                  MDS_INODELOCK_FULL);
 318         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 319                                    &RQF_MDS_REINT_UNLINK);
 320         if (req == NULL) {
 321                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 322                 RETURN(-ENOMEM);
 323         }
 324         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 325         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 326                              op_data->op_namelen + 1);
 327
 328         rc = mdc_prep_elc_req(exp, req, &cancels, count);
 329         if (rc) {
 330                 ptlrpc_request_free(req);
 331                 RETURN(rc);
 332         }
 333
 334         mdc_unlink_pack(req, op_data);
 335
 336         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 337                              obd->u.cli.cl_max_mds_easize);
 338         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 339                              obd->u.cli.cl_max_mds_cookiesize);
 340         ptlrpc_request_set_replen(req);
 341
 342         *request = req;
 343
 344         rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
 345         if (rc == -ERESTARTSYS)
 346                 rc = 0;
 347         RETURN(rc);
 348 }
 349
 350 int mdc_link(struct obd_export *exp, struct md_op_data *op_data,
 351              struct ptlrpc_request **request)
 352 {
 353         CFS_LIST_HEAD(cancels);
 354         struct obd_device *obd = exp->exp_obd;
 355         struct ptlrpc_request *req;
 356         int count = 0, rc;
 357         ENTRY;
 358
 359         if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
 360             (fid_is_sane(&op_data->op_fid2)))
 361                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
 362                                                 &cancels, LCK_EX,
 363                                                 MDS_INODELOCK_UPDATE);
 364         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
 365             (fid_is_sane(&op_data->op_fid1)))
 366                 count += mdc_resource_get_unused(exp, &op_data->op_fid1,
 367                                                  &cancels, LCK_EX,
 368                                                  MDS_INODELOCK_UPDATE);
 369
 370         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_REINT_LINK);
 371         if (req == NULL) {
 372                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 373                 RETURN(-ENOMEM);
 374         }
 375         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 376         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
 377         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 378                              op_data->op_namelen + 1);
 379
 380         rc = mdc_prep_elc_req(exp, req, &cancels, count);
 381         if (rc) {
 382                 ptlrpc_request_free(req);
 383                 RETURN(rc);
 384         }
 385
 386         mdc_link_pack(req, op_data);
 387         ptlrpc_request_set_replen(req);
 388
 389         rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
 390         *request = req;
 391         if (rc == -ERESTARTSYS)
 392                 rc = 0;
 393
 394         RETURN(rc);
 395 }
 396
 397 int mdc_rename(struct obd_export *exp, struct md_op_data *op_data,
 398                const char *old, int oldlen, const char *new, int newlen,
 399                struct ptlrpc_request **request)
 400 {
 401         CFS_LIST_HEAD(cancels);
 402         struct obd_device *obd = exp->exp_obd;
 403         struct ptlrpc_request *req;
 404         int count = 0, rc;
 405         ENTRY;
 406
 407         if ((op_data->op_flags & MF_MDC_CANCEL_FID1) &&
 408             (fid_is_sane(&op_data->op_fid1)))
 409                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
 410                                                 &cancels, LCK_EX,
 411                                                 MDS_INODELOCK_UPDATE);
 412         if ((op_data->op_flags & MF_MDC_CANCEL_FID2) &&
 413             (fid_is_sane(&op_data->op_fid2)))
 414                 count += mdc_resource_get_unused(exp, &op_data->op_fid2,
 415                                                  &cancels, LCK_EX,
 416                                                  MDS_INODELOCK_UPDATE);
 417         if ((op_data->op_flags & MF_MDC_CANCEL_FID3) &&
 418             (fid_is_sane(&op_data->op_fid3)))
 419                 count += mdc_resource_get_unused(exp, &op_data->op_fid3,
 420                                                  &cancels, LCK_EX,
 421                                                  MDS_INODELOCK_LOOKUP);
 422         if ((op_data->op_flags & MF_MDC_CANCEL_FID4) &&
 423              (fid_is_sane(&op_data->op_fid4)))
 424                 count += mdc_resource_get_unused(exp, &op_data->op_fid4,
 425                                                  &cancels, LCK_EX,
 426                                                  MDS_INODELOCK_FULL);
 427
 428         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 429                                    &RQF_MDS_REINT_RENAME);
 430         if (req == NULL) {
 431                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 432                 RETURN(-ENOMEM);
 433         }
 434
 435         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 436         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
 437         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT, oldlen + 1);
 438         req_capsule_set_size(&req->rq_pill, &RMF_SYMTGT, RCL_CLIENT, newlen+1);
 439
 440         rc = mdc_prep_elc_req(exp, req, &cancels, count);
 441         if (rc) {
 442                 ptlrpc_request_free(req);
 443                 RETURN(rc);
 444         }
 445
 446         if (exp_connect_cancelset(exp) && req)
 447                 ldlm_cli_cancel_list(&cancels, count, req, 0);
 448
 449         mdc_rename_pack(req, op_data, old, oldlen, new, newlen);
 450
 451         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 452                              obd->u.cli.cl_max_mds_easize);
 453         req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
 454                              obd->u.cli.cl_max_mds_cookiesize);
 455         ptlrpc_request_set_replen(req);
 456
 457         rc = mdc_reint(req, obd->u.cli.cl_rpc_lock, LUSTRE_IMP_FULL);
 458         *request = req;
 459         if (rc == -ERESTARTSYS)
 460                 rc = 0;
 461
 462         RETURN(rc);
 463 }