lustre/mdc/mdc_locks.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  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 #define DEBUG_SUBSYSTEM S_MDC
  38
  39 #ifdef __KERNEL__
  40 # include <linux/module.h>
  41 # include <linux/pagemap.h>
  42 # include <linux/miscdevice.h>
  43 # include <linux/init.h>
  44 #else
  45 # include <liblustre.h>
  46 #endif
  47
  48 #include <lustre_acl.h>
  49 #include <obd_class.h>
  50 #include <lustre_dlm.h>
  51 /* fid_res_name_eq() */
  52 #include <lustre_fid.h>
  53 #include <lprocfs_status.h>
  54 #include "mdc_internal.h"
  55
  56 struct mdc_getattr_args {
  57         struct obd_export           *ga_exp;
  58         struct md_enqueue_info      *ga_minfo;
  59         struct ldlm_enqueue_info    *ga_einfo;
  60 };
  61
  62 int it_disposition(struct lookup_intent *it, int flag)
  63 {
  64         return it->d.lustre.it_disposition & flag;
  65 }
  66 EXPORT_SYMBOL(it_disposition);
  67
  68 void it_set_disposition(struct lookup_intent *it, int flag)
  69 {
  70         it->d.lustre.it_disposition |= flag;
  71 }
  72 EXPORT_SYMBOL(it_set_disposition);
  73
  74 void it_clear_disposition(struct lookup_intent *it, int flag)
  75 {
  76         it->d.lustre.it_disposition &= ~flag;
  77 }
  78 EXPORT_SYMBOL(it_clear_disposition);
  79
  80 int it_open_error(int phase, struct lookup_intent *it)
  81 {
  82         if (it_disposition(it, DISP_OPEN_LEASE)) {
  83                 if (phase >= DISP_OPEN_LEASE)
  84                         return it->d.lustre.it_status;
  85                 else
  86                         return 0;
  87         }
  88         if (it_disposition(it, DISP_OPEN_OPEN)) {
  89                 if (phase >= DISP_OPEN_OPEN)
  90                         return it->d.lustre.it_status;
  91                 else
  92                         return 0;
  93         }
  94
  95         if (it_disposition(it, DISP_OPEN_CREATE)) {
  96                 if (phase >= DISP_OPEN_CREATE)
  97                         return it->d.lustre.it_status;
  98                 else
  99                         return 0;
 100         }
 101
 102         if (it_disposition(it, DISP_LOOKUP_EXECD)) {
 103                 if (phase >= DISP_LOOKUP_EXECD)
 104                         return it->d.lustre.it_status;
 105                 else
 106                         return 0;
 107         }
 108
 109         if (it_disposition(it, DISP_IT_EXECD)) {
 110                 if (phase >= DISP_IT_EXECD)
 111                         return it->d.lustre.it_status;
 112                 else
 113                         return 0;
 114         }
 115         CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
 116                it->d.lustre.it_status);
 117         LBUG();
 118         return 0;
 119 }
 120 EXPORT_SYMBOL(it_open_error);
 121
 122 /* this must be called on a lockh that is known to have a referenced lock */
 123 int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data,
 124                       __u64 *bits)
 125 {
 126         struct ldlm_lock *lock;
 127         struct inode *new_inode = data;
 128         ENTRY;
 129
 130         if(bits)
 131                 *bits = 0;
 132
 133         if (!*lockh)
 134                 RETURN(0);
 135
 136         lock = ldlm_handle2lock((struct lustre_handle *)lockh);
 137
 138         LASSERT(lock != NULL);
 139         lock_res_and_lock(lock);
 140 #ifdef __KERNEL__
 141         if (lock->l_resource->lr_lvb_inode &&
 142             lock->l_resource->lr_lvb_inode != data) {
 143                 struct inode *old_inode = lock->l_resource->lr_lvb_inode;
 144                 LASSERTF(old_inode->i_state & I_FREEING,
 145                          "Found existing inode %p/%lu/%u state %lu in lock: "
 146                          "setting data to %p/%lu/%u\n", old_inode,
 147                          old_inode->i_ino, old_inode->i_generation,
 148                          old_inode->i_state,
 149                          new_inode, new_inode->i_ino, new_inode->i_generation);
 150         }
 151 #endif
 152         lock->l_resource->lr_lvb_inode = new_inode;
 153         if (bits)
 154                 *bits = lock->l_policy_data.l_inodebits.bits;
 155
 156         unlock_res_and_lock(lock);
 157         LDLM_LOCK_PUT(lock);
 158
 159         RETURN(0);
 160 }
 161
 162 ldlm_mode_t mdc_lock_match(struct obd_export *exp, __u64 flags,
 163                            const struct lu_fid *fid, ldlm_type_t type,
 164                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
 165                            struct lustre_handle *lockh)
 166 {
 167         struct ldlm_res_id res_id;
 168         ldlm_mode_t rc;
 169         ENTRY;
 170
 171         fid_build_reg_res_name(fid, &res_id);
 172         rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
 173                              &res_id, type, policy, mode, lockh, 0);
 174         RETURN(rc);
 175 }
 176
 177 int mdc_cancel_unused(struct obd_export *exp,
 178                       const struct lu_fid *fid,
 179                       ldlm_policy_data_t *policy,
 180                       ldlm_mode_t mode,
 181                       ldlm_cancel_flags_t flags,
 182                       void *opaque)
 183 {
 184         struct ldlm_res_id res_id;
 185         struct obd_device *obd = class_exp2obd(exp);
 186         int rc;
 187
 188         ENTRY;
 189
 190         fid_build_reg_res_name(fid, &res_id);
 191         rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
 192                                              policy, mode, flags, opaque);
 193         RETURN(rc);
 194 }
 195
 196 int mdc_null_inode(struct obd_export *exp,
 197                    const struct lu_fid *fid)
 198 {
 199         struct ldlm_res_id res_id;
 200         struct ldlm_resource *res;
 201         struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
 202         ENTRY;
 203
 204         LASSERTF(ns != NULL, "no namespace passed\n");
 205
 206         fid_build_reg_res_name(fid, &res_id);
 207
 208         res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
 209         if(res == NULL)
 210                 RETURN(0);
 211
 212         lock_res(res);
 213         res->lr_lvb_inode = NULL;
 214         unlock_res(res);
 215
 216         ldlm_resource_putref(res);
 217         RETURN(0);
 218 }
 219
 220 /* find any ldlm lock of the inode in mdc
 221  * return 0    not find
 222  *        1    find one
 223  *      < 0    error */
 224 int mdc_find_cbdata(struct obd_export *exp,
 225                     const struct lu_fid *fid,
 226                     ldlm_iterator_t it, void *data)
 227 {
 228         struct ldlm_res_id res_id;
 229         int rc = 0;
 230         ENTRY;
 231
 232         fid_build_reg_res_name((struct lu_fid*)fid, &res_id);
 233         rc = ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace, &res_id,
 234                                    it, data);
 235         if (rc == LDLM_ITER_STOP)
 236                 RETURN(1);
 237         else if (rc == LDLM_ITER_CONTINUE)
 238                 RETURN(0);
 239         RETURN(rc);
 240 }
 241
 242 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
 243 {
 244         /* Don't hold error requests for replay. */
 245         if (req->rq_replay) {
 246                 spin_lock(&req->rq_lock);
 247                 req->rq_replay = 0;
 248                 spin_unlock(&req->rq_lock);
 249         }
 250         if (rc && req->rq_transno != 0) {
 251                 DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
 252                 LBUG();
 253         }
 254 }
 255
 256 /* Save a large LOV EA into the request buffer so that it is available
 257  * for replay.  We don't do this in the initial request because the
 258  * original request doesn't need this buffer (at most it sends just the
 259  * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
 260  * buffer and may also be difficult to allocate and save a very large
 261  * request buffer for each open. (bug 5707)
 262  *
 263  * OOM here may cause recovery failure if lmm is needed (only for the
 264  * original open if the MDS crashed just when this client also OOM'd)
 265  * but this is incredibly unlikely, and questionable whether the client
 266  * could do MDS recovery under OOM anyways... */
 267 static void mdc_realloc_openmsg(struct ptlrpc_request *req,
 268                                 struct mdt_body *body)
 269 {
 270         int     rc;
 271
 272         /* FIXME: remove this explicit offset. */
 273         rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
 274                                         body->eadatasize);
 275         if (rc) {
 276                 CERROR("Can't enlarge segment %d size to %d\n",
 277                        DLM_INTENT_REC_OFF + 4, body->eadatasize);
 278                 body->valid &= ~OBD_MD_FLEASIZE;
 279                 body->eadatasize = 0;
 280         }
 281 }
 282
 283 static struct ptlrpc_request *mdc_intent_open_pack(struct obd_export *exp,
 284                                                    struct lookup_intent *it,
 285                                                    struct md_op_data *op_data,
 286                                                    void *lmm, int lmmsize,
 287                                                    void *cb_data)
 288 {
 289         struct ptlrpc_request *req;
 290         struct obd_device     *obddev = class_exp2obd(exp);
 291         struct ldlm_intent    *lit;
 292         CFS_LIST_HEAD(cancels);
 293         int                    count = 0;
 294         int                    mode;
 295         int                    rc;
 296         ENTRY;
 297
 298         it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
 299
 300         /* XXX: openlock is not cancelled for cross-refs. */
 301         /* If inode is known, cancel conflicting OPEN locks. */
 302         if (fid_is_sane(&op_data->op_fid2)) {
 303                 if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
 304                         if (it->it_flags & FMODE_WRITE)
 305                                 mode = LCK_EX;
 306                         else
 307                                 mode = LCK_PR;
 308                 } else {
 309                         if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
 310                                 mode = LCK_CW;
 311 #ifdef FMODE_EXEC
 312                         else if (it->it_flags & FMODE_EXEC)
 313                                 mode = LCK_PR;
 314 #endif
 315                         else
 316                                 mode = LCK_CR;
 317                 }
 318                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
 319                                                 &cancels, mode,
 320                                                 MDS_INODELOCK_OPEN);
 321         }
 322
 323         /* If CREATE, cancel parent's UPDATE lock. */
 324         if (it->it_op & IT_CREAT)
 325                 mode = LCK_EX;
 326         else
 327                 mode = LCK_CR;
 328         count += mdc_resource_get_unused(exp, &op_data->op_fid1,
 329                                          &cancels, mode,
 330                                          MDS_INODELOCK_UPDATE);
 331
 332         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 333                                    &RQF_LDLM_INTENT_OPEN);
 334         if (req == NULL) {
 335                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 336                 RETURN(ERR_PTR(-ENOMEM));
 337         }
 338
 339         /* parent capability */
 340         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 341         /* child capability, reserve the size according to parent capa, it will
 342          * be filled after we get the reply */
 343         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa1);
 344
 345         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 346                              op_data->op_namelen + 1);
 347         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 348                              max(lmmsize, obddev->u.cli.cl_default_mds_easize));
 349
 350         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 351         if (rc) {
 352                 ptlrpc_request_free(req);
 353                 return NULL;
 354         }
 355
 356         spin_lock(&req->rq_lock);
 357         req->rq_replay = req->rq_import->imp_replayable;
 358         spin_unlock(&req->rq_lock);
 359
 360         /* pack the intent */
 361         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 362         lit->opc = (__u64)it->it_op;
 363
 364         /* pack the intended request */
 365         mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
 366                       lmmsize);
 367
 368         /* for remote client, fetch remote perm for current user */
 369         if (client_is_remote(exp))
 370                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 371                                      sizeof(struct mdt_remote_perm));
 372         ptlrpc_request_set_replen(req);
 373         return req;
 374 }
 375
 376 static struct ptlrpc_request *
 377 mdc_intent_getxattr_pack(struct obd_export *exp,
 378                          struct lookup_intent *it,
 379                          struct md_op_data *op_data)
 380 {
 381         struct ptlrpc_request   *req;
 382         struct ldlm_intent      *lit;
 383         int                     rc, count = 0, maxdata;
 384         CFS_LIST_HEAD(cancels);
 385
 386         ENTRY;
 387
 388         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 389                                         &RQF_LDLM_INTENT_GETXATTR);
 390         if (req == NULL)
 391                 RETURN(ERR_PTR(-ENOMEM));
 392
 393         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 394
 395         if (it->it_op == IT_SETXATTR)
 396                 /* If we want to upgrade to LCK_PW, let's cancel LCK_PR
 397                  * locks now. This avoids unnecessary ASTs. */
 398                 count = mdc_resource_get_unused(exp, &op_data->op_fid1,
 399                                                 &cancels, LCK_PW,
 400                                                 MDS_INODELOCK_XATTR);
 401
 402         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 403         if (rc) {
 404                 ptlrpc_request_free(req);
 405                 RETURN(ERR_PTR(rc));
 406         }
 407
 408         /* pack the intent */
 409         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 410         lit->opc = IT_GETXATTR;
 411
 412         maxdata = class_exp2cliimp(exp)->imp_connect_data.ocd_max_easize;
 413
 414         /* pack the intended request */
 415         mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
 416                         op_data->op_valid, maxdata, -1, 0);
 417
 418         req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
 419                                 RCL_SERVER, maxdata);
 420
 421         req_capsule_set_size(&req->rq_pill, &RMF_EAVALS,
 422                                 RCL_SERVER, maxdata);
 423
 424         req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS,
 425                                 RCL_SERVER, maxdata);
 426
 427         ptlrpc_request_set_replen(req);
 428
 429         RETURN(req);
 430 }
 431
 432 static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
 433                                                      struct lookup_intent *it,
 434                                                      struct md_op_data *op_data)
 435 {
 436         struct ptlrpc_request *req;
 437         struct obd_device     *obddev = class_exp2obd(exp);
 438         struct ldlm_intent    *lit;
 439         int                    rc;
 440         ENTRY;
 441
 442         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 443                                    &RQF_LDLM_INTENT_UNLINK);
 444         if (req == NULL)
 445                 RETURN(ERR_PTR(-ENOMEM));
 446
 447         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 448         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 449                              op_data->op_namelen + 1);
 450
 451         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 452         if (rc) {
 453                 ptlrpc_request_free(req);
 454                 RETURN(ERR_PTR(rc));
 455         }
 456
 457         /* pack the intent */
 458         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 459         lit->opc = (__u64)it->it_op;
 460
 461         /* pack the intended request */
 462         mdc_unlink_pack(req, op_data);
 463
 464         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 465                              obddev->u.cli.cl_max_mds_easize);
 466         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 467                              obddev->u.cli.cl_max_mds_cookiesize);
 468         ptlrpc_request_set_replen(req);
 469         RETURN(req);
 470 }
 471
 472 static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
 473                                                       struct lookup_intent *it,
 474                                                       struct md_op_data *op_data)
 475 {
 476         struct ptlrpc_request *req;
 477         struct obd_device     *obddev = class_exp2obd(exp);
 478         obd_valid              valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
 479                                        OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
 480                                        OBD_MD_FLMDSCAPA | OBD_MD_MEA |
 481                                        (client_is_remote(exp) ?
 482                                                OBD_MD_FLRMTPERM : OBD_MD_FLACL);
 483         struct ldlm_intent    *lit;
 484         int                    rc;
 485         ENTRY;
 486
 487         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 488                                    &RQF_LDLM_INTENT_GETATTR);
 489         if (req == NULL)
 490                 RETURN(ERR_PTR(-ENOMEM));
 491
 492         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 493         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 494                              op_data->op_namelen + 1);
 495
 496         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 497         if (rc) {
 498                 ptlrpc_request_free(req);
 499                 RETURN(ERR_PTR(rc));
 500         }
 501
 502         /* pack the intent */
 503         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 504         lit->opc = (__u64)it->it_op;
 505
 506         /* pack the intended request */
 507         mdc_getattr_pack(req, valid, it->it_flags, op_data,
 508                          obddev->u.cli.cl_max_mds_easize);
 509
 510         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 511                              obddev->u.cli.cl_max_mds_easize);
 512         if (client_is_remote(exp))
 513                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 514                                      sizeof(struct mdt_remote_perm));
 515         ptlrpc_request_set_replen(req);
 516         RETURN(req);
 517 }
 518
 519 static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
 520                                                      struct lookup_intent *it,
 521                                                      struct md_op_data *unused)
 522 {
 523         struct obd_device     *obd = class_exp2obd(exp);
 524         struct ptlrpc_request *req;
 525         struct ldlm_intent    *lit;
 526         struct layout_intent  *layout;
 527         int rc;
 528         ENTRY;
 529
 530         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 531                                 &RQF_LDLM_INTENT_LAYOUT);
 532         if (req == NULL)
 533                 RETURN(ERR_PTR(-ENOMEM));
 534
 535         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
 536         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 537         if (rc) {
 538                 ptlrpc_request_free(req);
 539                 RETURN(ERR_PTR(rc));
 540         }
 541
 542         /* pack the intent */
 543         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 544         lit->opc = (__u64)it->it_op;
 545
 546         /* pack the layout intent request */
 547         layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
 548         /* LAYOUT_INTENT_ACCESS is generic, specific operation will be
 549          * set for replication */
 550         layout->li_opc = LAYOUT_INTENT_ACCESS;
 551
 552         req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
 553                         obd->u.cli.cl_max_mds_easize);
 554         ptlrpc_request_set_replen(req);
 555         RETURN(req);
 556 }
 557
 558 static struct ptlrpc_request *
 559 mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
 560 {
 561         struct ptlrpc_request *req;
 562         int rc;
 563         ENTRY;
 564
 565         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
 566         if (req == NULL)
 567                 RETURN(ERR_PTR(-ENOMEM));
 568
 569         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 570         if (rc) {
 571                 ptlrpc_request_free(req);
 572                 RETURN(ERR_PTR(rc));
 573         }
 574
 575         req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
 576         ptlrpc_request_set_replen(req);
 577         RETURN(req);
 578 }
 579
 580 static int mdc_finish_enqueue(struct obd_export *exp,
 581                               struct ptlrpc_request *req,
 582                               struct ldlm_enqueue_info *einfo,
 583                               struct lookup_intent *it,
 584                               struct lustre_handle *lockh,
 585                               int rc)
 586 {
 587         struct req_capsule  *pill = &req->rq_pill;
 588         struct ldlm_request *lockreq;
 589         struct ldlm_reply   *lockrep;
 590         struct lustre_intent_data *intent = &it->d.lustre;
 591         struct ldlm_lock    *lock;
 592         void                *lvb_data = NULL;
 593         int                  lvb_len = 0;
 594         ENTRY;
 595
 596         LASSERT(rc >= 0);
 597         /* Similarly, if we're going to replay this request, we don't want to
 598          * actually get a lock, just perform the intent. */
 599         if (req->rq_transno || req->rq_replay) {
 600                 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
 601                 lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
 602         }
 603
 604         if (rc == ELDLM_LOCK_ABORTED) {
 605                 einfo->ei_mode = 0;
 606                 memset(lockh, 0, sizeof(*lockh));
 607                 rc = 0;
 608         } else { /* rc = 0 */
 609                 lock = ldlm_handle2lock(lockh);
 610                 LASSERT(lock != NULL);
 611
 612                 /* If the server gave us back a different lock mode, we should
 613                  * fix up our variables. */
 614                 if (lock->l_req_mode != einfo->ei_mode) {
 615                         ldlm_lock_addref(lockh, lock->l_req_mode);
 616                         ldlm_lock_decref(lockh, einfo->ei_mode);
 617                         einfo->ei_mode = lock->l_req_mode;
 618                 }
 619                 LDLM_LOCK_PUT(lock);
 620         }
 621
 622         lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
 623         LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
 624
 625         intent->it_disposition = (int)lockrep->lock_policy_res1;
 626         intent->it_status = (int)lockrep->lock_policy_res2;
 627         intent->it_lock_mode = einfo->ei_mode;
 628         intent->it_lock_handle = lockh->cookie;
 629         intent->it_data = req;
 630
 631         /* Technically speaking rq_transno must already be zero if
 632          * it_status is in error, so the check is a bit redundant */
 633         if ((!req->rq_transno || intent->it_status < 0) && req->rq_replay)
 634                 mdc_clear_replay_flag(req, intent->it_status);
 635
 636         /* If we're doing an IT_OPEN which did not result in an actual
 637          * successful open, then we need to remove the bit which saves
 638          * this request for unconditional replay.
 639          *
 640          * It's important that we do this first!  Otherwise we might exit the
 641          * function without doing so, and try to replay a failed create
 642          * (bug 3440) */
 643         if (it->it_op & IT_OPEN && req->rq_replay &&
 644             (!it_disposition(it, DISP_OPEN_OPEN) ||intent->it_status != 0))
 645                 mdc_clear_replay_flag(req, intent->it_status);
 646
 647         DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
 648                   it->it_op, intent->it_disposition, intent->it_status);
 649
 650         /* We know what to expect, so we do any byte flipping required here */
 651         if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
 652                 struct mdt_body *body;
 653
 654                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 655                 if (body == NULL) {
 656                         CERROR ("Can't swab mdt_body\n");
 657                         RETURN (-EPROTO);
 658                 }
 659
 660                 if (it_disposition(it, DISP_OPEN_OPEN) &&
 661                     !it_open_error(DISP_OPEN_OPEN, it)) {
 662                         /*
 663                          * If this is a successful OPEN request, we need to set
 664                          * replay handler and data early, so that if replay
 665                          * happens immediately after swabbing below, new reply
 666                          * is swabbed by that handler correctly.
 667                          */
 668                         mdc_set_open_replay_data(NULL, NULL, req);
 669                 }
 670
 671                 if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
 672                         void *eadata;
 673
 674                         mdc_update_max_ea_from_body(exp, body);
 675
 676                         /*
 677                          * The eadata is opaque; just check that it is there.
 678                          * Eventually, obd_unpackmd() will check the contents.
 679                          */
 680                         eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 681                                                               body->eadatasize);
 682                         if (eadata == NULL)
 683                                 RETURN(-EPROTO);
 684
 685                         /* save lvb data and length in case this is for layout
 686                          * lock */
 687                         lvb_data = eadata;
 688                         lvb_len = body->eadatasize;
 689
 690                         /*
 691                          * We save the reply LOV EA in case we have to replay a
 692                          * create for recovery.  If we didn't allocate a large
 693                          * enough request buffer above we need to reallocate it
 694                          * here to hold the actual LOV EA.
 695                          *
 696                          * To not save LOV EA if request is not going to replay
 697                          * (for example error one).
 698                          */
 699                         if ((it->it_op & IT_OPEN) && req->rq_replay) {
 700                                 void *lmm;
 701                                 if (req_capsule_get_size(pill, &RMF_EADATA,
 702                                                          RCL_CLIENT) <
 703                                     body->eadatasize)
 704                                         mdc_realloc_openmsg(req, body);
 705                                 else
 706                                         req_capsule_shrink(pill, &RMF_EADATA,
 707                                                            body->eadatasize,
 708                                                            RCL_CLIENT);
 709
 710                                 req_capsule_set_size(pill, &RMF_EADATA,
 711                                                      RCL_CLIENT,
 712                                                      body->eadatasize);
 713
 714                                 lmm = req_capsule_client_get(pill, &RMF_EADATA);
 715                                 if (lmm)
 716                                         memcpy(lmm, eadata, body->eadatasize);
 717                         }
 718                 }
 719
 720                 if (body->valid & OBD_MD_FLRMTPERM) {
 721                         struct mdt_remote_perm *perm;
 722
 723                         LASSERT(client_is_remote(exp));
 724                         perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 725                                                 lustre_swab_mdt_remote_perm);
 726                         if (perm == NULL)
 727                                 RETURN(-EPROTO);
 728                 }
 729                 if (body->valid & OBD_MD_FLMDSCAPA) {
 730                         struct lustre_capa *capa, *p;
 731
 732                         capa = req_capsule_server_get(pill, &RMF_CAPA1);
 733                         if (capa == NULL)
 734                                 RETURN(-EPROTO);
 735
 736                         if (it->it_op & IT_OPEN) {
 737                                 /* client fid capa will be checked in replay */
 738                                 p = req_capsule_client_get(pill, &RMF_CAPA2);
 739                                 LASSERT(p);
 740                                 *p = *capa;
 741                         }
 742                 }
 743                 if (body->valid & OBD_MD_FLOSSCAPA) {
 744                         struct lustre_capa *capa;
 745
 746                         capa = req_capsule_server_get(pill, &RMF_CAPA2);
 747                         if (capa == NULL)
 748                                 RETURN(-EPROTO);
 749                 }
 750         } else if (it->it_op & IT_LAYOUT) {
 751                 /* maybe the lock was granted right away and layout
 752                  * is packed into RMF_DLM_LVB of req */
 753                 lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
 754                 if (lvb_len > 0) {
 755                         lvb_data = req_capsule_server_sized_get(pill,
 756                                                         &RMF_DLM_LVB, lvb_len);
 757                         if (lvb_data == NULL)
 758                                 RETURN(-EPROTO);
 759                 }
 760         }
 761
 762         /* fill in stripe data for layout lock */
 763         lock = ldlm_handle2lock(lockh);
 764         if (lock != NULL && ldlm_has_layout(lock) && lvb_data != NULL) {
 765                 void *lmm;
 766
 767                 LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d\n",
 768                         ldlm_it2str(it->it_op), lvb_len);
 769
 770                 OBD_ALLOC_LARGE(lmm, lvb_len);
 771                 if (lmm == NULL) {
 772                         LDLM_LOCK_PUT(lock);
 773                         RETURN(-ENOMEM);
 774                 }
 775                 memcpy(lmm, lvb_data, lvb_len);
 776
 777                 /* install lvb_data */
 778                 lock_res_and_lock(lock);
 779                 if (lock->l_lvb_data == NULL) {
 780                         lock->l_lvb_data = lmm;
 781                         lock->l_lvb_len = lvb_len;
 782                         lmm = NULL;
 783                 }
 784                 unlock_res_and_lock(lock);
 785                 if (lmm != NULL)
 786                         OBD_FREE_LARGE(lmm, lvb_len);
 787         }
 788         if (lock != NULL)
 789                 LDLM_LOCK_PUT(lock);
 790
 791         RETURN(rc);
 792 }
 793
 794 /* We always reserve enough space in the reply packet for a stripe MD, because
 795  * we don't know in advance the file type. */
 796 int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
 797                 struct lookup_intent *it, struct md_op_data *op_data,
 798                 struct lustre_handle *lockh, void *lmm, int lmmsize,
 799                 struct ptlrpc_request **reqp, __u64 extra_lock_flags)
 800 {
 801         struct obd_device     *obddev = class_exp2obd(exp);
 802         struct ptlrpc_request *req = NULL;
 803         __u64                  flags, saved_flags = extra_lock_flags;
 804         int                    rc;
 805         struct ldlm_res_id res_id;
 806         static const ldlm_policy_data_t lookup_policy =
 807                             { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
 808         static const ldlm_policy_data_t update_policy =
 809                             { .l_inodebits = { MDS_INODELOCK_UPDATE } };
 810         static const ldlm_policy_data_t layout_policy =
 811                             { .l_inodebits = { MDS_INODELOCK_LAYOUT } };
 812         static const ldlm_policy_data_t getxattr_policy = {
 813                               .l_inodebits = { MDS_INODELOCK_XATTR } };
 814         ldlm_policy_data_t const *policy = &lookup_policy;
 815         int                    generation, resends = 0;
 816         struct ldlm_reply     *lockrep;
 817         enum lvb_type          lvb_type = 0;
 818         ENTRY;
 819
 820         LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
 821                  einfo->ei_type);
 822
 823         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
 824
 825         if (it) {
 826                 saved_flags |= LDLM_FL_HAS_INTENT;
 827                 if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
 828                         policy = &update_policy;
 829                 else if (it->it_op & IT_LAYOUT)
 830                         policy = &layout_policy;
 831                 else if (it->it_op & (IT_GETXATTR | IT_SETXATTR))
 832                         policy = &getxattr_policy;
 833         }
 834
 835         LASSERT(reqp == NULL);
 836
 837         generation = obddev->u.cli.cl_import->imp_generation;
 838 resend:
 839         flags = saved_flags;
 840         if (!it) {
 841                 /* The only way right now is FLOCK, in this case we hide flock
 842                    policy as lmm, but lmmsize is 0 */
 843                 LASSERT(lmm && lmmsize == 0);
 844                 LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
 845                          einfo->ei_type);
 846                 policy = (ldlm_policy_data_t *)lmm;
 847                 res_id.name[3] = LDLM_FLOCK;
 848         } else if (it->it_op & IT_OPEN) {
 849                 req = mdc_intent_open_pack(exp, it, op_data, lmm, lmmsize,
 850                                            einfo->ei_cbdata);
 851                 policy = &update_policy;
 852                 einfo->ei_cbdata = NULL;
 853                 lmm = NULL;
 854         } else if (it->it_op & IT_UNLINK) {
 855                 req = mdc_intent_unlink_pack(exp, it, op_data);
 856         } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
 857                 req = mdc_intent_getattr_pack(exp, it, op_data);
 858         } else if (it->it_op & IT_READDIR) {
 859                 req = mdc_enqueue_pack(exp, 0);
 860         } else if (it->it_op & IT_LAYOUT) {
 861                 if (!imp_connect_lvb_type(class_exp2cliimp(exp)))
 862                         RETURN(-EOPNOTSUPP);
 863                 req = mdc_intent_layout_pack(exp, it, op_data);
 864                 lvb_type = LVB_T_LAYOUT;
 865         } else if (it->it_op & (IT_GETXATTR | IT_SETXATTR)) {
 866                 req = mdc_intent_getxattr_pack(exp, it, op_data);
 867         } else {
 868                 LBUG();
 869                 RETURN(-EINVAL);
 870         }
 871
 872         if (IS_ERR(req))
 873                 RETURN(PTR_ERR(req));
 874
 875         if (req != NULL && it && it->it_op & IT_CREAT)
 876                 /* ask ptlrpc not to resend on EINPROGRESS since we have our own
 877                  * retry logic */
 878                 req->rq_no_retry_einprogress = 1;
 879
 880         if (resends) {
 881                 req->rq_generation_set = 1;
 882                 req->rq_import_generation = generation;
 883                 req->rq_sent = cfs_time_current_sec() + resends;
 884         }
 885
 886         /* It is important to obtain rpc_lock first (if applicable), so that
 887          * threads that are serialised with rpc_lock are not polluting our
 888          * rpcs in flight counter. We do not do flock request limiting, though*/
 889         if (it) {
 890                 mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
 891                 rc = mdc_enter_request(&obddev->u.cli);
 892                 if (rc != 0) {
 893                         mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
 894                         mdc_clear_replay_flag(req, 0);
 895                         ptlrpc_req_finished(req);
 896                         RETURN(rc);
 897                 }
 898         }
 899
 900         rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
 901                               0, lvb_type, lockh, 0);
 902         if (!it) {
 903                 /* For flock requests we immediatelly return without further
 904                    delay and let caller deal with the rest, since rest of
 905                    this function metadata processing makes no sense for flock
 906                    requests anyway. But in case of problem during comms with
 907                    Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
 908                    can not rely on caller and this mainly for F_UNLCKs
 909                    (explicits or automatically generated by Kernel to clean
 910                    current FLocks upon exit) that can't be trashed */
 911                 if ((rc == -EINTR) || (rc == -ETIMEDOUT))
 912                         goto resend;
 913                 RETURN(rc);
 914         }
 915
 916         mdc_exit_request(&obddev->u.cli);
 917         mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
 918
 919         if (rc < 0) {
 920                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 921                 mdc_clear_replay_flag(req, rc);
 922                 ptlrpc_req_finished(req);
 923                 RETURN(rc);
 924         }
 925
 926         lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
 927         LASSERT(lockrep != NULL);
 928
 929         lockrep->lock_policy_res2 =
 930                 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
 931
 932         /* Retry the create infinitely when we get -EINPROGRESS from
 933          * server. This is required by the new quota design. */
 934         if (it && it->it_op & IT_CREAT &&
 935             (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
 936                 mdc_clear_replay_flag(req, rc);
 937                 ptlrpc_req_finished(req);
 938                 resends++;
 939
 940                 CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
 941                        obddev->obd_name, resends, it->it_op,
 942                        PFID(&op_data->op_fid1), PFID(&op_data->op_fid2));
 943
 944                 if (generation == obddev->u.cli.cl_import->imp_generation) {
 945                         goto resend;
 946                 } else {
 947                         CDEBUG(D_HA, "resend cross eviction\n");
 948                         RETURN(-EIO);
 949                 }
 950         }
 951
 952         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
 953         if (rc < 0) {
 954                 if (lustre_handle_is_used(lockh)) {
 955                         ldlm_lock_decref(lockh, einfo->ei_mode);
 956                         memset(lockh, 0, sizeof(*lockh));
 957                 }
 958                 ptlrpc_req_finished(req);
 959         }
 960         RETURN(rc);
 961 }
 962
 963 static int mdc_finish_intent_lock(struct obd_export *exp,
 964                                   struct ptlrpc_request *request,
 965                                   struct md_op_data *op_data,
 966                                   struct lookup_intent *it,
 967                                   struct lustre_handle *lockh)
 968 {
 969         struct lustre_handle old_lock;
 970         struct mdt_body *mdt_body;
 971         struct ldlm_lock *lock;
 972         int rc;
 973         ENTRY;
 974
 975         LASSERT(request != NULL);
 976         LASSERT(request != LP_POISON);
 977         LASSERT(request->rq_repmsg != LP_POISON);
 978
 979         if (!it_disposition(it, DISP_IT_EXECD)) {
 980                 /* The server failed before it even started executing the
 981                  * intent, i.e. because it couldn't unpack the request. */
 982                 LASSERT(it->d.lustre.it_status != 0);
 983                 RETURN(it->d.lustre.it_status);
 984         }
 985         rc = it_open_error(DISP_IT_EXECD, it);
 986         if (rc)
 987                 RETURN(rc);
 988
 989         mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 990         LASSERT(mdt_body != NULL);      /* mdc_enqueue checked */
 991
 992         /* If we were revalidating a fid/name pair, mark the intent in
 993          * case we fail and get called again from lookup */
 994         if (fid_is_sane(&op_data->op_fid2) &&
 995             it->it_create_mode & M_CHECK_STALE &&
 996             it->it_op != IT_GETATTR) {
 997                 it_set_disposition(it, DISP_ENQ_COMPLETE);
 998
 999                 /* Also: did we find the same inode? */
1000                 /* sever can return one of two fids:
1001                  * op_fid2 - new allocated fid - if file is created.
1002                  * op_fid3 - existent fid - if file only open.
1003                  * op_fid3 is saved in lmv_intent_open */
1004                 if ((!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1)) &&
1005                     (!lu_fid_eq(&op_data->op_fid3, &mdt_body->fid1))) {
1006                         CDEBUG(D_DENTRY, "Found stale data "DFID"("DFID")/"DFID
1007                                "\n", PFID(&op_data->op_fid2),
1008                                PFID(&op_data->op_fid2), PFID(&mdt_body->fid1));
1009                         RETURN(-ESTALE);
1010                 }
1011         }
1012
1013         rc = it_open_error(DISP_LOOKUP_EXECD, it);
1014         if (rc)
1015                 RETURN(rc);
1016
1017         /* keep requests around for the multiple phases of the call
1018          * this shows the DISP_XX must guarantee we make it into the call
1019          */
1020         if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
1021             it_disposition(it, DISP_OPEN_CREATE) &&
1022             !it_open_error(DISP_OPEN_CREATE, it)) {
1023                 it_set_disposition(it, DISP_ENQ_CREATE_REF);
1024                 ptlrpc_request_addref(request); /* balanced in ll_create_node */
1025         }
1026         if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
1027             it_disposition(it, DISP_OPEN_OPEN) &&
1028             !it_open_error(DISP_OPEN_OPEN, it)) {
1029                 it_set_disposition(it, DISP_ENQ_OPEN_REF);
1030                 ptlrpc_request_addref(request); /* balanced in ll_file_open */
1031                 /* BUG 11546 - eviction in the middle of open rpc processing */
1032                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
1033         }
1034
1035         if (it->it_op & IT_CREAT) {
1036                 /* XXX this belongs in ll_create_it */
1037         } else if (it->it_op == IT_OPEN) {
1038                 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
1039         } else {
1040                 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP | IT_LAYOUT));
1041         }
1042
1043         /* If we already have a matching lock, then cancel the new
1044          * one.  We have to set the data here instead of in
1045          * mdc_enqueue, because we need to use the child's inode as
1046          * the l_ast_data to match, and that's not available until
1047          * intent_finish has performed the iget().) */
1048         lock = ldlm_handle2lock(lockh);
1049         if (lock) {
1050                 ldlm_policy_data_t policy = lock->l_policy_data;
1051                 LDLM_DEBUG(lock, "matching against this");
1052
1053                 LASSERTF(fid_res_name_eq(&mdt_body->fid1,
1054                                          &lock->l_resource->lr_name),
1055                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1056                          PLDLMRES(lock->l_resource), PFID(&mdt_body->fid1));
1057                 LDLM_LOCK_PUT(lock);
1058
1059                 memcpy(&old_lock, lockh, sizeof(*lockh));
1060                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
1061                                     LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
1062                         ldlm_lock_decref_and_cancel(lockh,
1063                                                     it->d.lustre.it_lock_mode);
1064                         memcpy(lockh, &old_lock, sizeof(old_lock));
1065                         it->d.lustre.it_lock_handle = lockh->cookie;
1066                 }
1067         }
1068         CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
1069                op_data->op_namelen, op_data->op_name, ldlm_it2str(it->it_op),
1070                it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
1071         RETURN(rc);
1072 }
1073
1074 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
1075                         struct lu_fid *fid, __u64 *bits)
1076 {
1077         /* We could just return 1 immediately, but since we should only
1078          * be called in revalidate_it if we already have a lock, let's
1079          * verify that. */
1080         struct ldlm_res_id res_id;
1081         struct lustre_handle lockh;
1082         ldlm_policy_data_t policy;
1083         ldlm_mode_t mode;
1084         ENTRY;
1085
1086         if (it->d.lustre.it_lock_handle) {
1087                 lockh.cookie = it->d.lustre.it_lock_handle;
1088                 mode = ldlm_revalidate_lock_handle(&lockh, bits);
1089         } else {
1090                 fid_build_reg_res_name(fid, &res_id);
1091                 switch (it->it_op) {
1092                 case IT_GETATTR:
1093                         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1094                         break;
1095                 case IT_LAYOUT:
1096                         policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
1097                         break;
1098                 default:
1099                         policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1100                         break;
1101                 }
1102                 mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
1103                                        LDLM_FL_BLOCK_GRANTED, &res_id,
1104                                        LDLM_IBITS, &policy,
1105                                        LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh, 0);
1106         }
1107
1108         if (mode) {
1109                 it->d.lustre.it_lock_handle = lockh.cookie;
1110                 it->d.lustre.it_lock_mode = mode;
1111         } else {
1112                 it->d.lustre.it_lock_handle = 0;
1113                 it->d.lustre.it_lock_mode = 0;
1114         }
1115
1116         RETURN(!!mode);
1117 }
1118
1119 /*
1120  * This long block is all about fixing up the lock and request state
1121  * so that it is correct as of the moment _before_ the operation was
1122  * applied; that way, the VFS will think that everything is normal and
1123  * call Lustre's regular VFS methods.
1124  *
1125  * If we're performing a creation, that means that unless the creation
1126  * failed with EEXIST, we should fake up a negative dentry.
1127  *
1128  * For everything else, we want to lookup to succeed.
1129  *
1130  * One additional note: if CREATE or OPEN succeeded, we add an extra
1131  * reference to the request because we need to keep it around until
1132  * ll_create/ll_open gets called.
1133  *
1134  * The server will return to us, in it_disposition, an indication of
1135  * exactly what d.lustre.it_status refers to.
1136  *
1137  * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
1138  * otherwise if DISP_OPEN_CREATE is set, then it status is the
1139  * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
1140  * DISP_LOOKUP_POS will be set, indicating whether the child lookup
1141  * was successful.
1142  *
1143  * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
1144  * child lookup.
1145  */
1146 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
1147                     void *lmm, int lmmsize, struct lookup_intent *it,
1148                     int lookup_flags, struct ptlrpc_request **reqp,
1149                     ldlm_blocking_callback cb_blocking,
1150                     __u64 extra_lock_flags)
1151 {
1152         struct lustre_handle lockh;
1153         int rc = 0;
1154         ENTRY;
1155         LASSERT(it);
1156
1157         CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
1158                 ", intent: %s flags %#Lo\n", op_data->op_namelen,
1159                 op_data->op_name, PFID(&op_data->op_fid2),
1160                 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
1161                 it->it_flags);
1162
1163         lockh.cookie = 0;
1164         if (fid_is_sane(&op_data->op_fid2) &&
1165             (it->it_op & (IT_LOOKUP | IT_GETATTR))) {
1166                 /* We could just return 1 immediately, but since we should only
1167                  * be called in revalidate_it if we already have a lock, let's
1168                  * verify that. */
1169                 it->d.lustre.it_lock_handle = 0;
1170                 rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
1171                 /* Only return failure if it was not GETATTR by cfid
1172                    (from inode_revalidate) */
1173                 if (rc || op_data->op_namelen != 0)
1174                         RETURN(rc);
1175         }
1176
1177         /* lookup_it may be called only after revalidate_it has run, because
1178          * revalidate_it cannot return errors, only zero.  Returning zero causes
1179          * this call to lookup, which *can* return an error.
1180          *
1181          * We only want to execute the request associated with the intent one
1182          * time, however, so don't send the request again.  Instead, skip past
1183          * this and use the request from revalidate.  In this case, revalidate
1184          * never dropped its reference, so the refcounts are all OK */
1185         if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
1186                 struct ldlm_enqueue_info einfo = {
1187                         .ei_type        = LDLM_IBITS,
1188                         .ei_mode        = it_to_lock_mode(it),
1189                         .ei_cb_bl       = cb_blocking,
1190                         .ei_cb_cp       = ldlm_completion_ast,
1191                 };
1192
1193                 /* For case if upper layer did not alloc fid, do it now. */
1194                 if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
1195                         rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
1196                         if (rc < 0) {
1197                                 CERROR("Can't alloc new fid, rc %d\n", rc);
1198                                 RETURN(rc);
1199                         }
1200                 }
1201                 rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh,
1202                                  lmm, lmmsize, NULL, extra_lock_flags);
1203                 if (rc < 0)
1204                         RETURN(rc);
1205         } else if (!fid_is_sane(&op_data->op_fid2) ||
1206                    !(it->it_create_mode & M_CHECK_STALE)) {
1207                 /* DISP_ENQ_COMPLETE set means there is extra reference on
1208                  * request referenced from this intent, saved for subsequent
1209                  * lookup.  This path is executed when we proceed to this
1210                  * lookup, so we clear DISP_ENQ_COMPLETE */
1211                 it_clear_disposition(it, DISP_ENQ_COMPLETE);
1212         }
1213         *reqp = it->d.lustre.it_data;
1214         rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
1215         RETURN(rc);
1216 }
1217
1218 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
1219                                               struct ptlrpc_request *req,
1220                                               void *args, int rc)
1221 {
1222         struct mdc_getattr_args  *ga = args;
1223         struct obd_export        *exp = ga->ga_exp;
1224         struct md_enqueue_info   *minfo = ga->ga_minfo;
1225         struct ldlm_enqueue_info *einfo = ga->ga_einfo;
1226         struct lookup_intent     *it;
1227         struct lustre_handle     *lockh;
1228         struct obd_device        *obddev;
1229         struct ldlm_reply        *lockrep;
1230         __u64                     flags = LDLM_FL_HAS_INTENT;
1231         ENTRY;
1232
1233         it    = &minfo->mi_it;
1234         lockh = &minfo->mi_lockh;
1235
1236         obddev = class_exp2obd(exp);
1237
1238         mdc_exit_request(&obddev->u.cli);
1239         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
1240                 rc = -ETIMEDOUT;
1241
1242         rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
1243                                    &flags, NULL, 0, lockh, rc);
1244         if (rc < 0) {
1245                 CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
1246                 mdc_clear_replay_flag(req, rc);
1247                 GOTO(out, rc);
1248         }
1249
1250         lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1251         LASSERT(lockrep != NULL);
1252
1253         lockrep->lock_policy_res2 =
1254                 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1255
1256         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1257         if (rc)
1258                 GOTO(out, rc);
1259
1260         rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
1261         EXIT;
1262
1263 out:
1264         OBD_FREE_PTR(einfo);
1265         minfo->mi_cb(req, minfo, rc);
1266         return 0;
1267 }
1268
1269 int mdc_intent_getattr_async(struct obd_export *exp,
1270                              struct md_enqueue_info *minfo,
1271                              struct ldlm_enqueue_info *einfo)
1272 {
1273         struct md_op_data       *op_data = &minfo->mi_data;
1274         struct lookup_intent    *it = &minfo->mi_it;
1275         struct ptlrpc_request   *req;
1276         struct mdc_getattr_args *ga;
1277         struct obd_device       *obddev = class_exp2obd(exp);
1278         struct ldlm_res_id       res_id;
1279         /*XXX: Both MDS_INODELOCK_LOOKUP and MDS_INODELOCK_UPDATE are needed
1280          *     for statahead currently. Consider CMD in future, such two bits
1281          *     maybe managed by different MDS, should be adjusted then. */
1282         ldlm_policy_data_t       policy = {
1283                                         .l_inodebits = { MDS_INODELOCK_LOOKUP |
1284                                                          MDS_INODELOCK_UPDATE }
1285                                  };
1286         int                      rc = 0;
1287         __u64                    flags = LDLM_FL_HAS_INTENT;
1288         ENTRY;
1289
1290         CDEBUG(D_DLMTRACE,"name: %.*s in inode "DFID", intent: %s flags %#Lo\n",
1291                 op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
1292                 ldlm_it2str(it->it_op), it->it_flags);
1293
1294         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
1295         req = mdc_intent_getattr_pack(exp, it, op_data);
1296         if (!req)
1297                 RETURN(-ENOMEM);
1298
1299         rc = mdc_enter_request(&obddev->u.cli);
1300         if (rc != 0) {
1301                 ptlrpc_req_finished(req);
1302                 RETURN(rc);
1303         }
1304
1305         rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
1306                               0, LVB_T_NONE, &minfo->mi_lockh, 1);
1307         if (rc < 0) {
1308                 mdc_exit_request(&obddev->u.cli);
1309                 ptlrpc_req_finished(req);
1310                 RETURN(rc);
1311         }
1312
1313         CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
1314         ga = ptlrpc_req_async_args(req);
1315         ga->ga_exp = exp;
1316         ga->ga_minfo = minfo;
1317         ga->ga_einfo = einfo;
1318
1319         req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
1320         ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
1321
1322         RETURN(0);
1323 }