lustre/mdc/mdc_locks.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2001-2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of the Lustre file system, http://www.lustre.org
   7  *   Lustre is a trademark of Cluster File Systems, Inc.
   8  *
   9  *   You may have signed or agreed to another license before downloading
  10  *   this software.  If so, you are bound by the terms and conditions
  11  *   of that agreement, and the following does not apply to you.  See the
  12  *   LICENSE file included with this distribution for more information.
  13  *
  14  *   If you did not agree to a different license, then this copy of Lustre
  15  *   is open source software; you can redistribute it and/or modify it
  16  *   under the terms of version 2 of the GNU General Public License as
  17  *   published by the Free Software Foundation.
  18  *
  19  *   In either case, Lustre is distributed in the hope that it will be
  20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   license text for more details.
  23  */
  24
  25 #ifndef EXPORT_SYMTAB
  26 # define EXPORT_SYMTAB
  27 #endif
  28 #define DEBUG_SUBSYSTEM S_MDC
  29
  30 #ifdef __KERNEL__
  31 # include <linux/module.h>
  32 # include <linux/pagemap.h>
  33 # include <linux/miscdevice.h>
  34 # include <linux/init.h>
  35 #else
  36 # include <liblustre.h>
  37 #endif
  38
  39 #include <linux/lustre_acl.h>
  40 #include <obd_class.h>
  41 #include <lustre_dlm.h>
  42 /* fid_res_name_eq() */
  43 #include <lustre_fid.h>
  44 #include <lprocfs_status.h>
  45 #include "mdc_internal.h"
  46
  47 int it_disposition(struct lookup_intent *it, int flag)
  48 {
  49         return it->d.lustre.it_disposition & flag;
  50 }
  51 EXPORT_SYMBOL(it_disposition);
  52
  53 void it_set_disposition(struct lookup_intent *it, int flag)
  54 {
  55         it->d.lustre.it_disposition |= flag;
  56 }
  57 EXPORT_SYMBOL(it_set_disposition);
  58
  59 void it_clear_disposition(struct lookup_intent *it, int flag)
  60 {
  61         it->d.lustre.it_disposition &= ~flag;
  62 }
  63 EXPORT_SYMBOL(it_clear_disposition);
  64
  65 int it_open_error(int phase, struct lookup_intent *it)
  66 {
  67         if (it_disposition(it, DISP_OPEN_OPEN)) {
  68                 if (phase >= DISP_OPEN_OPEN)
  69                         return it->d.lustre.it_status;
  70                 else
  71                         return 0;
  72         }
  73
  74         if (it_disposition(it, DISP_OPEN_CREATE)) {
  75                 if (phase >= DISP_OPEN_CREATE)
  76                         return it->d.lustre.it_status;
  77                 else
  78                         return 0;
  79         }
  80
  81         if (it_disposition(it, DISP_LOOKUP_EXECD)) {
  82                 if (phase >= DISP_LOOKUP_EXECD)
  83                         return it->d.lustre.it_status;
  84                 else
  85                         return 0;
  86         }
  87
  88         if (it_disposition(it, DISP_IT_EXECD)) {
  89                 if (phase >= DISP_IT_EXECD)
  90                         return it->d.lustre.it_status;
  91                 else
  92                         return 0;
  93         }
  94         CERROR("it disp: %X, status: %d\n", it->d.lustre.it_disposition,
  95                it->d.lustre.it_status);
  96         LBUG();
  97         return 0;
  98 }
  99 EXPORT_SYMBOL(it_open_error);
 100
 101 /* this must be called on a lockh that is known to have a referenced lock */
 102 int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data)
 103 {
 104         struct ldlm_lock *lock;
 105         ENTRY;
 106
 107         if (!*lockh) {
 108                 EXIT;
 109                 RETURN(0);
 110         }
 111
 112         lock = ldlm_handle2lock((struct lustre_handle *)lockh);
 113
 114         LASSERT(lock != NULL);
 115         lock_res_and_lock(lock);
 116 #ifdef __KERNEL__
 117         if (lock->l_ast_data && lock->l_ast_data != data) {
 118                 struct inode *new_inode = data;
 119                 struct inode *old_inode = lock->l_ast_data;
 120                 LASSERTF(old_inode->i_state & I_FREEING,
 121                          "Found existing inode %p/%lu/%u state %lu in lock: "
 122                          "setting data to %p/%lu/%u\n", old_inode,
 123                          old_inode->i_ino, old_inode->i_generation,
 124                          old_inode->i_state,
 125                          new_inode, new_inode->i_ino, new_inode->i_generation);
 126         }
 127 #endif
 128         lock->l_ast_data = data;
 129         unlock_res_and_lock(lock);
 130         LDLM_LOCK_PUT(lock);
 131
 132         RETURN(0);
 133 }
 134
 135 ldlm_mode_t mdc_lock_match(struct obd_export *exp, int flags,
 136                            const struct lu_fid *fid, ldlm_type_t type,
 137                            ldlm_policy_data_t *policy, ldlm_mode_t mode,
 138                            struct lustre_handle *lockh)
 139 {
 140         struct ldlm_res_id res_id;
 141         ldlm_mode_t rc;
 142         ENTRY;
 143
 144         fid_build_reg_res_name(fid, &res_id);
 145         rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
 146                              &res_id, type, policy, mode, lockh);
 147         RETURN(rc);
 148 }
 149
 150 int mdc_cancel_unused(struct obd_export *exp,
 151                       const struct lu_fid *fid,
 152                       ldlm_policy_data_t *policy,
 153                       ldlm_mode_t mode, int flags, void *opaque)
 154 {
 155         struct ldlm_res_id res_id;
 156         struct obd_device *obd = class_exp2obd(exp);
 157         int rc;
 158
 159         ENTRY;
 160
 161         fid_build_reg_res_name(fid, &res_id);
 162         rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
 163                                              policy, mode, flags, opaque);
 164         RETURN(rc);
 165 }
 166
 167 int mdc_change_cbdata(struct obd_export *exp,
 168                       const struct lu_fid *fid,
 169                       ldlm_iterator_t it, void *data)
 170 {
 171         struct ldlm_res_id res_id;
 172         ENTRY;
 173
 174         fid_build_reg_res_name(fid, &res_id);
 175         ldlm_resource_iterate(class_exp2obd(exp)->obd_namespace,
 176                               &res_id, it, data);
 177
 178         EXIT;
 179         return 0;
 180 }
 181
 182 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
 183 {
 184         /* Don't hold error requests for replay. */
 185         if (req->rq_replay) {
 186                 spin_lock(&req->rq_lock);
 187                 req->rq_replay = 0;
 188                 spin_unlock(&req->rq_lock);
 189         }
 190         if (rc && req->rq_transno != 0) {
 191                 DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
 192                 LBUG();
 193         }
 194 }
 195
 196 /* Save a large LOV EA into the request buffer so that it is available
 197  * for replay.  We don't do this in the initial request because the
 198  * original request doesn't need this buffer (at most it sends just the
 199  * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
 200  * buffer and may also be difficult to allocate and save a very large
 201  * request buffer for each open. (bug 5707)
 202  *
 203  * OOM here may cause recovery failure if lmm is needed (only for the
 204  * original open if the MDS crashed just when this client also OOM'd)
 205  * but this is incredibly unlikely, and questionable whether the client
 206  * could do MDS recovery under OOM anyways... */
 207 static void mdc_realloc_openmsg(struct ptlrpc_request *req,
 208                                 struct mdt_body *body)
 209 {
 210         int     rc;
 211
 212         /* FIXME: remove this explicit offset. */
 213         rc = sptlrpc_cli_enlarge_reqbuf(req, DLM_INTENT_REC_OFF + 4,
 214                                         body->eadatasize);
 215         if (rc) {
 216                 CERROR("Can't enlarge segment %d size to %d\n",
 217                        DLM_INTENT_REC_OFF + 4, body->eadatasize);
 218                 body->valid &= ~OBD_MD_FLEASIZE;
 219                 body->eadatasize = 0;
 220         }
 221 }
 222
 223 static struct ptlrpc_request *mdc_intent_open_pack(struct obd_export *exp,
 224                                                    struct lookup_intent *it,
 225                                                    struct md_op_data *op_data,
 226                                                    void *lmm, int lmmsize,
 227                                                    void *cb_data)
 228 {
 229         struct ptlrpc_request *req;
 230         struct obd_device     *obddev = class_exp2obd(exp);
 231         struct ldlm_intent    *lit;
 232         int                    joinfile = !!((it->it_flags & O_JOIN_FILE) &&
 233                                               op_data->op_data);
 234         CFS_LIST_HEAD(cancels);
 235         int                    count = 0;
 236         int                    mode;
 237         int                    rc;
 238         ENTRY;
 239
 240         it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
 241
 242         /* XXX: openlock is not cancelled for cross-refs. */
 243         /* If inode is known, cancel conflicting OPEN locks. */
 244         if (fid_is_sane(&op_data->op_fid2)) {
 245                 if (it->it_flags & (FMODE_WRITE|MDS_OPEN_TRUNC))
 246                         mode = LCK_CW;
 247 #ifdef FMODE_EXEC
 248                 else if (it->it_flags & FMODE_EXEC)
 249                         mode = LCK_PR;
 250 #endif
 251                 else
 252                         mode = LCK_CR;
 253                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
 254                                                 &cancels, mode,
 255                                                 MDS_INODELOCK_OPEN);
 256         }
 257
 258         /* If CREATE or JOIN_FILE, cancel parent's UPDATE lock. */
 259         if (it->it_op & IT_CREAT || joinfile)
 260                 mode = LCK_EX;
 261         else
 262                 mode = LCK_CR;
 263         count += mdc_resource_get_unused(exp, &op_data->op_fid1,
 264                                          &cancels, mode,
 265                                          MDS_INODELOCK_UPDATE);
 266
 267         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 268                                    &RQF_LDLM_INTENT_OPEN);
 269         if (req == NULL) {
 270                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 271                 RETURN(ERR_PTR(-ENOMEM));
 272         }
 273
 274         /* parent capability */
 275         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 276         /* child capability, reserve the size according to parent capa, it will
 277          * be filled after we get the reply */
 278         mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa1);
 279
 280         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 281                              op_data->op_namelen + 1);
 282         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 283                              max(lmmsize, obddev->u.cli.cl_default_mds_easize));
 284         if (!joinfile) {
 285                 req_capsule_set_size(&req->rq_pill, &RMF_REC_JOINFILE,
 286                                      RCL_CLIENT, 0);
 287         }
 288
 289         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 290         if (rc) {
 291                 ptlrpc_request_free(req);
 292                 return NULL;
 293         }
 294
 295         if (joinfile) {
 296                 __u64 head_size = *(__u64 *)op_data->op_data;
 297                 mdc_join_pack(req, op_data, head_size);
 298         }
 299
 300         spin_lock(&req->rq_lock);
 301         req->rq_replay = 1;
 302         spin_unlock(&req->rq_lock);
 303
 304         /* pack the intent */
 305         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 306         lit->opc = (__u64)it->it_op;
 307
 308         /* pack the intended request */
 309         mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
 310                       lmmsize);
 311
 312         /* for remote client, fetch remote perm for current user */
 313         if (client_is_remote(exp))
 314                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 315                                      sizeof(struct mdt_remote_perm));
 316         ptlrpc_request_set_replen(req);
 317         return req;
 318 }
 319
 320 static struct ptlrpc_request *mdc_intent_unlink_pack(struct obd_export *exp,
 321                                                      struct lookup_intent *it,
 322                                                      struct md_op_data *op_data)
 323 {
 324         struct ptlrpc_request *req;
 325         struct obd_device     *obddev = class_exp2obd(exp);
 326         struct ldlm_intent    *lit;
 327         int                    rc;
 328         ENTRY;
 329
 330         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 331                                    &RQF_LDLM_INTENT_UNLINK);
 332         if (req == NULL)
 333                 RETURN(ERR_PTR(-ENOMEM));
 334
 335         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 336         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 337                              op_data->op_namelen + 1);
 338
 339         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 340         if (rc) {
 341                 ptlrpc_request_free(req);
 342                 RETURN(ERR_PTR(rc));
 343         }
 344
 345         /* pack the intent */
 346         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 347         lit->opc = (__u64)it->it_op;
 348
 349         /* pack the intended request */
 350         mdc_unlink_pack(req, op_data);
 351
 352         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 353                              obddev->u.cli.cl_max_mds_easize);
 354         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 355                              obddev->u.cli.cl_max_mds_cookiesize);
 356         ptlrpc_request_set_replen(req);
 357         RETURN(req);
 358 }
 359
 360 static struct ptlrpc_request *mdc_intent_getattr_pack(struct obd_export *exp,
 361                                                       struct lookup_intent *it,
 362                                                       struct md_op_data *op_data)
 363 {
 364         struct ptlrpc_request *req;
 365         struct obd_device     *obddev = class_exp2obd(exp);
 366         obd_valid              valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
 367                                        OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
 368                                        OBD_MD_FLMDSCAPA | OBD_MD_MEA |
 369                                        (client_is_remote(exp) ?
 370                                                OBD_MD_FLRMTPERM : OBD_MD_FLACL);
 371         struct ldlm_intent    *lit;
 372         int                    rc;
 373         ENTRY;
 374
 375         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 376                                    &RQF_LDLM_INTENT_GETATTR);
 377         if (req == NULL)
 378                 RETURN(ERR_PTR(-ENOMEM));
 379
 380         mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 381         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 382                              op_data->op_namelen + 1);
 383
 384         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 385         if (rc) {
 386                 ptlrpc_request_free(req);
 387                 RETURN(ERR_PTR(rc));
 388         }
 389
 390         /* pack the intent */
 391         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 392         lit->opc = (__u64)it->it_op;
 393
 394         /* pack the intended request */
 395         mdc_getattr_pack(req, valid, it->it_flags, op_data);
 396
 397         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 398                              obddev->u.cli.cl_max_mds_easize);
 399         if (client_is_remote(exp))
 400                 req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
 401                                      sizeof(struct mdt_remote_perm));
 402         ptlrpc_request_set_replen(req);
 403         RETURN(req);
 404 }
 405
 406 static struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp)
 407 {
 408         struct ptlrpc_request *req;
 409         int rc;
 410         ENTRY;
 411
 412         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
 413         if (req == NULL)
 414                 RETURN(ERR_PTR(-ENOMEM));
 415
 416         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 417         if (rc) {
 418                 ptlrpc_request_free(req);
 419                 RETURN(ERR_PTR(rc));
 420         }
 421
 422         ptlrpc_request_set_replen(req);
 423         RETURN(req);
 424 }
 425
 426 static int mdc_finish_enqueue(struct obd_export *exp,
 427                               struct ptlrpc_request *req,
 428                               struct ldlm_enqueue_info *einfo,
 429                               struct lookup_intent *it,
 430                               struct lustre_handle *lockh,
 431                               int rc)
 432 {
 433         struct req_capsule  *pill = &req->rq_pill;
 434         struct ldlm_request *lockreq;
 435         struct ldlm_reply   *lockrep;
 436         ENTRY;
 437
 438         LASSERT(rc >= 0);
 439         /* Similarly, if we're going to replay this request, we don't want to
 440          * actually get a lock, just perform the intent. */
 441         if (req->rq_transno || req->rq_replay) {
 442                 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
 443                 lockreq->lock_flags |= LDLM_FL_INTENT_ONLY;
 444         }
 445
 446         if (rc == ELDLM_LOCK_ABORTED) {
 447                 einfo->ei_mode = 0;
 448                 memset(lockh, 0, sizeof(*lockh));
 449                 rc = 0;
 450         } else { /* rc = 0 */
 451                 struct ldlm_lock *lock = ldlm_handle2lock(lockh);
 452                 LASSERT(lock);
 453
 454                 /* If the server gave us back a different lock mode, we should
 455                  * fix up our variables. */
 456                 if (lock->l_req_mode != einfo->ei_mode) {
 457                         ldlm_lock_addref(lockh, lock->l_req_mode);
 458                         ldlm_lock_decref(lockh, einfo->ei_mode);
 459                         einfo->ei_mode = lock->l_req_mode;
 460                 }
 461                 LDLM_LOCK_PUT(lock);
 462         }
 463
 464         lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
 465         LASSERT(lockrep != NULL);                 /* checked by ldlm_cli_enqueue() */
 466
 467         it->d.lustre.it_disposition = (int)lockrep->lock_policy_res1;
 468         it->d.lustre.it_status = (int)lockrep->lock_policy_res2;
 469         it->d.lustre.it_lock_mode = einfo->ei_mode;
 470         it->d.lustre.it_data = req;
 471
 472         if (it->d.lustre.it_status < 0 && req->rq_replay)
 473                 mdc_clear_replay_flag(req, it->d.lustre.it_status);
 474
 475         /* If we're doing an IT_OPEN which did not result in an actual
 476          * successful open, then we need to remove the bit which saves
 477          * this request for unconditional replay.
 478          *
 479          * It's important that we do this first!  Otherwise we might exit the
 480          * function without doing so, and try to replay a failed create
 481          * (bug 3440) */
 482         if (it->it_op & IT_OPEN && req->rq_replay &&
 483             (!it_disposition(it, DISP_OPEN_OPEN) ||it->d.lustre.it_status != 0))
 484                 mdc_clear_replay_flag(req, it->d.lustre.it_status);
 485
 486         DEBUG_REQ(D_RPCTRACE, req, "op: %d disposition: %x, status: %d",
 487                   it->it_op,it->d.lustre.it_disposition,it->d.lustre.it_status);
 488
 489         /* We know what to expect, so we do any byte flipping required here */
 490         if (it->it_op & (IT_OPEN | IT_UNLINK | IT_LOOKUP | IT_GETATTR)) {
 491                 struct mdt_body *body;
 492
 493                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 494                 if (body == NULL) {
 495                         CERROR ("Can't swab mdt_body\n");
 496                         RETURN (-EPROTO);
 497                 }
 498
 499                 if (it_disposition(it, DISP_OPEN_OPEN) &&
 500                     !it_open_error(DISP_OPEN_OPEN, it)) {
 501                         /*
 502                          * If this is a successful OPEN request, we need to set
 503                          * replay handler and data early, so that if replay
 504                          * happens immediately after swabbing below, new reply
 505                          * is swabbed by that handler correctly.
 506                          */
 507                         mdc_set_open_replay_data(NULL, NULL, req);
 508                 }
 509
 510                 if ((body->valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) != 0) {
 511                         void *eadata;
 512
 513                         /*
 514                          * The eadata is opaque; just check that it is there.
 515                          * Eventually, obd_unpackmd() will check the contents.
 516                          */
 517                         eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 518                                                               body->eadatasize);
 519                         if (eadata == NULL)
 520                                 RETURN(-EPROTO);
 521
 522                         if (body->valid & OBD_MD_FLMODEASIZE) {
 523                                 struct obd_device *obddev = class_exp2obd(exp);
 524
 525                                 if (obddev->u.cli.cl_max_mds_easize <
 526                                     body->max_mdsize) {
 527                                         obddev->u.cli.cl_max_mds_easize =
 528                                                 body->max_mdsize;
 529                                         CDEBUG(D_INFO, "maxeasize become %d\n",
 530                                                body->max_mdsize);
 531                                 }
 532                                 if (obddev->u.cli.cl_max_mds_cookiesize <
 533                                     body->max_cookiesize) {
 534                                         obddev->u.cli.cl_max_mds_cookiesize =
 535                                                 body->max_cookiesize;
 536                                         CDEBUG(D_INFO, "cookiesize become %d\n",
 537                                                body->max_cookiesize);
 538                                 }
 539                         }
 540
 541                         /*
 542                          * We save the reply LOV EA in case we have to replay a
 543                          * create for recovery.  If we didn't allocate a large
 544                          * enough request buffer above we need to reallocate it
 545                          * here to hold the actual LOV EA.
 546                          *
 547                          * To not save LOV EA if request is not going to replay
 548                          * (for example error one).
 549                          */
 550                         if ((it->it_op & IT_OPEN) && req->rq_replay) {
 551                                 void *lmm;
 552                                 if (req_capsule_get_size(pill, &RMF_EADATA,
 553                                                          RCL_CLIENT) <
 554                                     body->eadatasize) {
 555                                         mdc_realloc_openmsg(req, body);
 556                                         req_capsule_set_size(pill, &RMF_EADATA,
 557                                                              RCL_CLIENT,
 558                                                              body->eadatasize);
 559                                 }
 560                                 lmm = req_capsule_client_get(pill, &RMF_EADATA);
 561                                 if (lmm)
 562                                         memcpy(lmm, eadata, body->eadatasize);
 563                         }
 564                 }
 565
 566                 if (body->valid & OBD_MD_FLRMTPERM) {
 567                         struct mdt_remote_perm *perm;
 568
 569                         LASSERT(client_is_remote(exp));
 570                         perm = req_capsule_server_swab_get(pill, &RMF_ACL,
 571                                                 lustre_swab_mdt_remote_perm);
 572                         if (perm == NULL)
 573                                 RETURN(-EPROTO);
 574                 }
 575                 if (body->valid & OBD_MD_FLMDSCAPA) {
 576                         struct lustre_capa *capa, *p;
 577
 578                         capa = req_capsule_server_get(pill, &RMF_CAPA1);
 579                         if (capa == NULL)
 580                                 RETURN(-EPROTO);
 581
 582                         if (it->it_op & IT_OPEN) {
 583                                 /* client fid capa will be checked in replay */
 584                                 p = req_capsule_client_get(pill, &RMF_CAPA2);
 585                                 LASSERT(p);
 586                                 *p = *capa;
 587                         }
 588                 }
 589                 if (body->valid & OBD_MD_FLOSSCAPA) {
 590                         struct lustre_capa *capa;
 591
 592                         capa = req_capsule_server_get(pill, &RMF_CAPA2);
 593                         if (capa == NULL)
 594                                 RETURN(-EPROTO);
 595                 }
 596         }
 597
 598         RETURN(rc);
 599 }
 600
 601 /* We always reserve enough space in the reply packet for a stripe MD, because
 602  * we don't know in advance the file type. */
 603 int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
 604                 struct lookup_intent *it, struct md_op_data *op_data,
 605                 struct lustre_handle *lockh, void *lmm, int lmmsize,
 606                 int extra_lock_flags)
 607 {
 608         struct obd_device     *obddev = class_exp2obd(exp);
 609         struct ptlrpc_request *req;
 610         struct req_capsule    *pill;
 611         int                    flags = extra_lock_flags | LDLM_FL_HAS_INTENT;
 612         int                    rc;
 613         struct ldlm_res_id res_id;
 614         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_LOOKUP } };
 615         ENTRY;
 616
 617         LASSERTF(einfo->ei_type == LDLM_IBITS, "lock type %d\n", einfo->ei_type);
 618
 619         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
 620
 621         if (it->it_op & (IT_UNLINK | IT_GETATTR | IT_READDIR))
 622                 policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
 623
 624         if (it->it_op & IT_OPEN) {
 625                 int joinfile = !!((it->it_flags & O_JOIN_FILE) &&
 626                                               op_data->op_data);
 627
 628                 req = mdc_intent_open_pack(exp, it, op_data, lmm, lmmsize,
 629                                            einfo->ei_cbdata);
 630                 if (!joinfile) {
 631                         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
 632                         einfo->ei_cbdata = NULL;
 633                         lmm = NULL;
 634                 } else
 635                         it->it_flags &= ~O_JOIN_FILE;
 636         } else if (it->it_op & IT_UNLINK)
 637                 req = mdc_intent_unlink_pack(exp, it, op_data);
 638         else if (it->it_op & (IT_GETATTR | IT_LOOKUP))
 639                 req = mdc_intent_getattr_pack(exp, it, op_data);
 640         else if (it->it_op == IT_READDIR)
 641                 req = ldlm_enqueue_pack(exp);
 642         else {
 643                 LBUG();
 644                 RETURN(-EINVAL);
 645         }
 646
 647         if (IS_ERR(req))
 648                 RETURN(PTR_ERR(req));
 649         pill = &req->rq_pill;
 650
 651         /* It is important to obtain rpc_lock first (if applicable), so that
 652          * threads that are serialised with rpc_lock are not polluting our
 653          * rpcs in flight counter */
 654         mdc_get_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
 655         mdc_enter_request(&obddev->u.cli);
 656         rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
 657                               0, NULL, lockh, 0);
 658         mdc_exit_request(&obddev->u.cli);
 659         mdc_put_rpc_lock(obddev->u.cli.cl_rpc_lock, it);
 660         if (rc < 0) {
 661                 CERROR("ldlm_cli_enqueue: %d\n", rc);
 662                 mdc_clear_replay_flag(req, rc);
 663                 ptlrpc_req_finished(req);
 664                 RETURN(rc);
 665         }
 666         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
 667
 668         RETURN(rc);
 669 }
 670
 671 static int mdc_finish_intent_lock(struct obd_export *exp,
 672                                   struct ptlrpc_request *request,
 673                                   struct md_op_data *op_data,
 674                                   struct lookup_intent *it,
 675                                   struct lustre_handle *lockh)
 676 {
 677         struct lustre_handle old_lock;
 678         struct mdt_body *mdt_body;
 679         struct ldlm_lock *lock;
 680         int rc;
 681
 682
 683         LASSERT(request != NULL);
 684         LASSERT(request != LP_POISON);
 685         LASSERT(request->rq_repmsg != LP_POISON);
 686
 687         if (!it_disposition(it, DISP_IT_EXECD)) {
 688                 /* The server failed before it even started executing the
 689                  * intent, i.e. because it couldn't unpack the request. */
 690                 LASSERT(it->d.lustre.it_status != 0);
 691                 RETURN(it->d.lustre.it_status);
 692         }
 693         rc = it_open_error(DISP_IT_EXECD, it);
 694         if (rc)
 695                 RETURN(rc);
 696
 697         mdt_body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 698         LASSERT(mdt_body != NULL);      /* mdc_enqueue checked */
 699
 700         /* If we were revalidating a fid/name pair, mark the intent in
 701          * case we fail and get called again from lookup */
 702         if (fid_is_sane(&op_data->op_fid2) &&
 703             it->it_flags & O_CHECK_STALE &&
 704             it->it_op != IT_GETATTR) {
 705                 it_set_disposition(it, DISP_ENQ_COMPLETE);
 706
 707                 /* Also: did we find the same inode? */
 708                 /* sever can return one of two fids:
 709                  * op_fid2 - new allocated fid - if file is created.
 710                  * op_fid3 - existent fid - if file only open.
 711                  * op_fid3 is saved in lmv_intent_open */
 712                 if ((!lu_fid_eq(&op_data->op_fid2, &mdt_body->fid1)) &&
 713                     (!lu_fid_eq(&op_data->op_fid3, &mdt_body->fid1))) {
 714                         CDEBUG(D_DENTRY, "Found stale data "DFID"("DFID")/"DFID
 715                                "\n", PFID(&op_data->op_fid2),
 716                                PFID(&op_data->op_fid2), PFID(&mdt_body->fid1));
 717                         RETURN(-ESTALE);
 718                 }
 719         }
 720
 721         rc = it_open_error(DISP_LOOKUP_EXECD, it);
 722         if (rc)
 723                 RETURN(rc);
 724
 725         /* keep requests around for the multiple phases of the call
 726          * this shows the DISP_XX must guarantee we make it into the call
 727          */
 728         if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
 729             it_disposition(it, DISP_OPEN_CREATE) &&
 730             !it_open_error(DISP_OPEN_CREATE, it)) {
 731                 it_set_disposition(it, DISP_ENQ_CREATE_REF);
 732                 ptlrpc_request_addref(request); /* balanced in ll_create_node */
 733         }
 734         if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
 735             it_disposition(it, DISP_OPEN_OPEN) &&
 736             !it_open_error(DISP_OPEN_OPEN, it)) {
 737                 it_set_disposition(it, DISP_ENQ_OPEN_REF);
 738                 ptlrpc_request_addref(request); /* balanced in ll_file_open */
 739                 /* BUG 11546 - eviction in the middle of open rpc processing */
 740                 OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE, obd_timeout);
 741         }
 742
 743         if (it->it_op & IT_CREAT) {
 744                 /* XXX this belongs in ll_create_it */
 745         } else if (it->it_op == IT_OPEN) {
 746                 LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
 747         } else {
 748                 LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
 749         }
 750
 751         /* If we already have a matching lock, then cancel the new
 752          * one.  We have to set the data here instead of in
 753          * mdc_enqueue, because we need to use the child's inode as
 754          * the l_ast_data to match, and that's not available until
 755          * intent_finish has performed the iget().) */
 756         lock = ldlm_handle2lock(lockh);
 757         if (lock) {
 758                 ldlm_policy_data_t policy = lock->l_policy_data;
 759                 LDLM_DEBUG(lock, "matching against this");
 760
 761                 LASSERTF(fid_res_name_eq(&mdt_body->fid1,
 762                                          &lock->l_resource->lr_name),
 763                          "Lock res_id: %lu/%lu/%lu, fid: %lu/%lu/%lu.\n",
 764                          (unsigned long)lock->l_resource->lr_name.name[0],
 765                          (unsigned long)lock->l_resource->lr_name.name[1],
 766                          (unsigned long)lock->l_resource->lr_name.name[2],
 767                          (unsigned long)fid_seq(&mdt_body->fid1),
 768                          (unsigned long)fid_oid(&mdt_body->fid1),
 769                          (unsigned long)fid_ver(&mdt_body->fid1));
 770                 LDLM_LOCK_PUT(lock);
 771
 772                 memcpy(&old_lock, lockh, sizeof(*lockh));
 773                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
 774                                     LDLM_IBITS, &policy, LCK_NL, &old_lock)) {
 775                         ldlm_lock_decref_and_cancel(lockh,
 776                                                     it->d.lustre.it_lock_mode);
 777                         memcpy(lockh, &old_lock, sizeof(old_lock));
 778                         it->d.lustre.it_lock_handle = lockh->cookie;
 779                 }
 780         }
 781         CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
 782                op_data->op_namelen, op_data->op_name, ldlm_it2str(it->it_op),
 783                it->d.lustre.it_status, it->d.lustre.it_disposition, rc);
 784         RETURN(rc);
 785 }
 786
 787 /*
 788  * This long block is all about fixing up the lock and request state
 789  * so that it is correct as of the moment _before_ the operation was
 790  * applied; that way, the VFS will think that everything is normal and
 791  * call Lustre's regular VFS methods.
 792  *
 793  * If we're performing a creation, that means that unless the creation
 794  * failed with EEXIST, we should fake up a negative dentry.
 795  *
 796  * For everything else, we want to lookup to succeed.
 797  *
 798  * One additional note: if CREATE or OPEN succeeded, we add an extra
 799  * reference to the request because we need to keep it around until
 800  * ll_create/ll_open gets called.
 801  *
 802  * The server will return to us, in it_disposition, an indication of
 803  * exactly what d.lustre.it_status refers to.
 804  *
 805  * If DISP_OPEN_OPEN is set, then d.lustre.it_status refers to the open() call,
 806  * otherwise if DISP_OPEN_CREATE is set, then it status is the
 807  * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
 808  * DISP_LOOKUP_POS will be set, indicating whether the child lookup
 809  * was successful.
 810  *
 811  * Else, if DISP_LOOKUP_EXECD then d.lustre.it_status is the rc of the
 812  * child lookup.
 813  */
 814 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
 815                     void *lmm, int lmmsize, struct lookup_intent *it,
 816                     int lookup_flags, struct ptlrpc_request **reqp,
 817                     ldlm_blocking_callback cb_blocking,
 818                     int extra_lock_flags)
 819 {
 820         struct lustre_handle lockh;
 821         int rc = 0;
 822         ENTRY;
 823         LASSERT(it);
 824
 825         CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
 826                ", intent: %s flags %#o\n", op_data->op_namelen,
 827                op_data->op_name, PFID(&op_data->op_fid2),
 828                PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
 829                it->it_flags);
 830
 831         lockh.cookie = 0;
 832         if (fid_is_sane(&op_data->op_fid2) &&
 833             (it->it_op & (IT_LOOKUP | IT_GETATTR))) {
 834                 /* We could just return 1 immediately, but since we should only
 835                  * be called in revalidate_it if we already have a lock, let's
 836                  * verify that. */
 837                 ldlm_policy_data_t policy;
 838                 ldlm_mode_t mode;
 839
 840                 /* As not all attributes are kept under update lock, e.g.
 841                    owner/group/acls are under lookup lock, we need both
 842                    ibits for GETATTR. */
 843
 844                 /* For CMD, UPDATE lock and LOOKUP lock can not be got
 845                  * at the same for cross-object, so we can not match
 846                  * the 2 lock at the same time FIXME: but how to handle
 847                  * the above situation */
 848                 policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
 849                         MDS_INODELOCK_UPDATE : MDS_INODELOCK_LOOKUP;
 850
 851                 mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED,
 852                                       &op_data->op_fid2, LDLM_IBITS, &policy,
 853                                       LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh);
 854                 if (mode) {
 855                         it->d.lustre.it_lock_handle = lockh.cookie;
 856                         it->d.lustre.it_lock_mode = mode;
 857                 }
 858
 859                 /* Only return failure if it was not GETATTR by cfid
 860                    (from inode_revalidate) */
 861                 if (mode || op_data->op_namelen != 0)
 862                         RETURN(!!mode);
 863         }
 864
 865         /* lookup_it may be called only after revalidate_it has run, because
 866          * revalidate_it cannot return errors, only zero.  Returning zero causes
 867          * this call to lookup, which *can* return an error.
 868          *
 869          * We only want to execute the request associated with the intent one
 870          * time, however, so don't send the request again.  Instead, skip past
 871          * this and use the request from revalidate.  In this case, revalidate
 872          * never dropped its reference, so the refcounts are all OK */
 873         if (!it_disposition(it, DISP_ENQ_COMPLETE)) {
 874                 struct ldlm_enqueue_info einfo =
 875                         { LDLM_IBITS, it_to_lock_mode(it), cb_blocking,
 876                           ldlm_completion_ast, NULL, NULL };
 877
 878                 /* For case if upper layer did not alloc fid, do it now. */
 879                 if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
 880                         rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
 881                         if (rc < 0) {
 882                                 CERROR("Can't alloc new fid, rc %d\n", rc);
 883                                 RETURN(rc);
 884                         }
 885                 }
 886                 rc = mdc_enqueue(exp, &einfo, it, op_data, &lockh,
 887                                  lmm, lmmsize, extra_lock_flags);
 888                 if (rc < 0)
 889                         RETURN(rc);
 890                 it->d.lustre.it_lock_handle = lockh.cookie;
 891         } else if (!fid_is_sane(&op_data->op_fid2) ||
 892                    !(it->it_flags & O_CHECK_STALE)) {
 893                 /* DISP_ENQ_COMPLETE set means there is extra reference on
 894                  * request referenced from this intent, saved for subsequent
 895                  * lookup.  This path is executed when we proceed to this
 896                  * lookup, so we clear DISP_ENQ_COMPLETE */
 897                 it_clear_disposition(it, DISP_ENQ_COMPLETE);
 898         }
 899         *reqp = it->d.lustre.it_data;
 900         rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
 901         RETURN(rc);
 902 }
 903
 904 static int mdc_intent_getattr_async_interpret(struct ptlrpc_request *req,
 905                                               void *unused, int rc)
 906 {
 907         struct obd_export        *exp = req->rq_async_args.pointer_arg[0];
 908         struct md_enqueue_info   *minfo = req->rq_async_args.pointer_arg[1];
 909         struct ldlm_enqueue_info *einfo = req->rq_async_args.pointer_arg[2];
 910         struct lookup_intent     *it;
 911         struct lustre_handle     *lockh;
 912         struct obd_device        *obddev;
 913         int                       flags = LDLM_FL_HAS_INTENT;
 914         ENTRY;
 915
 916         it    = &minfo->mi_it;
 917         lockh = &minfo->mi_lockh;
 918
 919         obddev = class_exp2obd(exp);
 920
 921         mdc_exit_request(&obddev->u.cli);
 922         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
 923                 rc = -ETIMEDOUT;
 924
 925         rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
 926                                    &flags, NULL, 0, NULL, lockh, rc);
 927         if (rc < 0) {
 928                 CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
 929                 mdc_clear_replay_flag(req, rc);
 930                 GOTO(out, rc);
 931         }
 932
 933         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
 934         if (rc)
 935                 GOTO(out, rc);
 936
 937         it->d.lustre.it_lock_handle = lockh->cookie;
 938
 939         rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
 940         EXIT;
 941
 942 out:
 943         OBD_FREE_PTR(einfo);
 944         minfo->mi_cb(req, minfo, rc);
 945         return 0;
 946 }
 947
 948 int mdc_intent_getattr_async(struct obd_export *exp,
 949                              struct md_enqueue_info *minfo,
 950                              struct ldlm_enqueue_info *einfo)
 951 {
 952         struct md_op_data       *op_data = &minfo->mi_data;
 953         struct lookup_intent    *it = &minfo->mi_it;
 954         struct ptlrpc_request   *req;
 955         struct obd_device       *obddev = class_exp2obd(exp);
 956         struct ldlm_res_id       res_id;
 957         ldlm_policy_data_t       policy = {
 958                                         .l_inodebits = { MDS_INODELOCK_LOOKUP }
 959                                  };
 960         int                      rc;
 961         int                      flags = LDLM_FL_HAS_INTENT;
 962         ENTRY;
 963
 964         CDEBUG(D_DLMTRACE,"name: %.*s in inode "DFID", intent: %s flags %#o\n",
 965                op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1),
 966                ldlm_it2str(it->it_op), it->it_flags);
 967
 968         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
 969         req = mdc_intent_getattr_pack(exp, it, op_data);
 970         if (!req)
 971                 RETURN(-ENOMEM);
 972
 973         mdc_enter_request(&obddev->u.cli);
 974         rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, &policy, &flags, NULL,
 975                               0, NULL, &minfo->mi_lockh, 1);
 976         if (rc < 0) {
 977                 mdc_exit_request(&obddev->u.cli);
 978                 RETURN(rc);
 979         }
 980
 981         req->rq_async_args.pointer_arg[0] = exp;
 982         req->rq_async_args.pointer_arg[1] = minfo;
 983         req->rq_async_args.pointer_arg[2] = einfo;
 984         req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
 985         ptlrpcd_add_req(req);
 986
 987         RETURN(0);
 988 }
 989
 990 int mdc_revalidate_lock(struct obd_export *exp,
 991                         struct lookup_intent *it,
 992                         struct lu_fid *fid)
 993 {
 994         /* We could just return 1 immediately, but since we should only
 995          * be called in revalidate_it if we already have a lock, let's
 996          * verify that. */
 997         struct ldlm_res_id res_id;
 998         struct lustre_handle lockh;
 999         ldlm_policy_data_t policy;
1000         ldlm_mode_t mode;
1001         ENTRY;
1002
1003         fid_build_reg_res_name(fid, &res_id);
1004         /* As not all attributes are kept under update lock, e.g.
1005            owner/group/acls are under lookup lock, we need both
1006            ibits for GETATTR. */
1007         policy.l_inodebits.bits = (it->it_op == IT_GETATTR) ?
1008                 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP :
1009                 MDS_INODELOCK_LOOKUP;
1010
1011         mode = ldlm_lock_match(exp->exp_obd->obd_namespace,
1012                                LDLM_FL_BLOCK_GRANTED, &res_id, LDLM_IBITS,
1013                                &policy, LCK_CR|LCK_CW|LCK_PR|LCK_PW, &lockh);
1014         if (mode) {
1015                 it->d.lustre.it_lock_handle = lockh.cookie;
1016                 it->d.lustre.it_lock_mode = mode;
1017         }
1018
1019         RETURN(!!mode);
1020 }