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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_MDC
  34
  35 #include <linux/module.h>
  36
  37 #include <obd.h>
  38 #include <obd_class.h>
  39 #include <lustre_dlm.h>
  40 #include <lustre_fid.h>
  41 #include <lustre_intent.h>
  42 #include <lustre_mdc.h>
  43 #include <lustre_net.h>
  44 #include <lustre_req_layout.h>
  45 #include <lustre_swab.h>
  46 #include <lustre_acl.h>
  47
  48 #include "mdc_internal.h"
  49
  50 struct mdc_getattr_args {
  51         struct obd_export               *ga_exp;
  52         struct md_enqueue_info          *ga_minfo;
  53 };
  54
  55 int it_open_error(int phase, struct lookup_intent *it)
  56 {
  57         if (it_disposition(it, DISP_OPEN_LEASE)) {
  58                 if (phase >= DISP_OPEN_LEASE)
  59                         return it->it_status;
  60                 else
  61                         return 0;
  62         }
  63         if (it_disposition(it, DISP_OPEN_OPEN)) {
  64                 if (phase >= DISP_OPEN_OPEN)
  65                         return it->it_status;
  66                 else
  67                         return 0;
  68         }
  69
  70         if (it_disposition(it, DISP_OPEN_CREATE)) {
  71                 if (phase >= DISP_OPEN_CREATE)
  72                         return it->it_status;
  73                 else
  74                         return 0;
  75         }
  76
  77         if (it_disposition(it, DISP_LOOKUP_EXECD)) {
  78                 if (phase >= DISP_LOOKUP_EXECD)
  79                         return it->it_status;
  80                 else
  81                         return 0;
  82         }
  83
  84         if (it_disposition(it, DISP_IT_EXECD)) {
  85                 if (phase >= DISP_IT_EXECD)
  86                         return it->it_status;
  87                 else
  88                         return 0;
  89         }
  90
  91         CERROR("it disp: %X, status: %d\n", it->it_disposition, it->it_status);
  92         LBUG();
  93
  94         return 0;
  95 }
  96 EXPORT_SYMBOL(it_open_error);
  97
  98 /* this must be called on a lockh that is known to have a referenced lock */
  99 int mdc_set_lock_data(struct obd_export *exp, const struct lustre_handle *lockh,
 100                       void *data, __u64 *bits)
 101 {
 102         struct ldlm_lock *lock;
 103         struct inode *new_inode = data;
 104         ENTRY;
 105
 106         if(bits)
 107                 *bits = 0;
 108
 109         if (!lustre_handle_is_used(lockh))
 110                 RETURN(0);
 111
 112         lock = ldlm_handle2lock(lockh);
 113
 114         LASSERT(lock != NULL);
 115         lock_res_and_lock(lock);
 116         if (lock->l_resource->lr_lvb_inode &&
 117             lock->l_resource->lr_lvb_inode != data) {
 118                 struct inode *old_inode = lock->l_resource->lr_lvb_inode;
 119                 LASSERTF(old_inode->i_state & I_FREEING,
 120                          "Found existing inode %p/%lu/%u state %lu in lock: "
 121                          "setting data to %p/%lu/%u\n", old_inode,
 122                          old_inode->i_ino, old_inode->i_generation,
 123                          old_inode->i_state,
 124                          new_inode, new_inode->i_ino, new_inode->i_generation);
 125         }
 126         lock->l_resource->lr_lvb_inode = new_inode;
 127         if (bits)
 128                 *bits = lock->l_policy_data.l_inodebits.bits;
 129
 130         unlock_res_and_lock(lock);
 131         LDLM_LOCK_PUT(lock);
 132
 133         RETURN(0);
 134 }
 135
 136 enum ldlm_mode mdc_lock_match(struct obd_export *exp, __u64 flags,
 137                               const struct lu_fid *fid, enum ldlm_type type,
 138                               union ldlm_policy_data *policy,
 139                               enum ldlm_mode mode, struct lustre_handle *lockh)
 140 {
 141         struct ldlm_res_id res_id;
 142         enum ldlm_mode rc;
 143         ENTRY;
 144
 145         fid_build_reg_res_name(fid, &res_id);
 146         /* LU-4405: Clear bits not supported by server */
 147         policy->l_inodebits.bits &= exp_connect_ibits(exp);
 148         rc = ldlm_lock_match(class_exp2obd(exp)->obd_namespace, flags,
 149                              &res_id, type, policy, mode, lockh, 0);
 150         RETURN(rc);
 151 }
 152
 153 int mdc_cancel_unused(struct obd_export *exp, const struct lu_fid *fid,
 154                       union ldlm_policy_data *policy, enum ldlm_mode mode,
 155                       enum ldlm_cancel_flags flags, void *opaque)
 156 {
 157         struct obd_device *obd = class_exp2obd(exp);
 158         struct ldlm_res_id res_id;
 159         int rc;
 160
 161         ENTRY;
 162
 163         fid_build_reg_res_name(fid, &res_id);
 164         rc = ldlm_cli_cancel_unused_resource(obd->obd_namespace, &res_id,
 165                                              policy, mode, flags, opaque);
 166         RETURN(rc);
 167 }
 168
 169 int mdc_null_inode(struct obd_export *exp,
 170                    const struct lu_fid *fid)
 171 {
 172         struct ldlm_res_id res_id;
 173         struct ldlm_resource *res;
 174         struct ldlm_namespace *ns = class_exp2obd(exp)->obd_namespace;
 175         ENTRY;
 176
 177         LASSERTF(ns != NULL, "no namespace passed\n");
 178
 179         fid_build_reg_res_name(fid, &res_id);
 180
 181         res = ldlm_resource_get(ns, NULL, &res_id, 0, 0);
 182         if (IS_ERR(res))
 183                 RETURN(0);
 184
 185         lock_res(res);
 186         res->lr_lvb_inode = NULL;
 187         unlock_res(res);
 188
 189         ldlm_resource_putref(res);
 190         RETURN(0);
 191 }
 192
 193 static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
 194 {
 195         /* Don't hold error requests for replay. */
 196         if (req->rq_replay) {
 197                 spin_lock(&req->rq_lock);
 198                 req->rq_replay = 0;
 199                 spin_unlock(&req->rq_lock);
 200         }
 201         if (rc && req->rq_transno != 0) {
 202                 DEBUG_REQ(D_ERROR, req, "transno returned on error: rc = %d",
 203                           rc);
 204                 LBUG();
 205         }
 206 }
 207
 208 /* Save a large LOV EA into the request buffer so that it is available
 209  * for replay.  We don't do this in the initial request because the
 210  * original request doesn't need this buffer (at most it sends just the
 211  * lov_mds_md) and it is a waste of RAM/bandwidth to send the empty
 212  * buffer and may also be difficult to allocate and save a very large
 213  * request buffer for each open. (bug 5707)
 214  *
 215  * OOM here may cause recovery failure if lmm is needed (only for the
 216  * original open if the MDS crashed just when this client also OOM'd)
 217  * but this is incredibly unlikely, and questionable whether the client
 218  * could do MDS recovery under OOM anyways... */
 219 int mdc_save_lovea(struct ptlrpc_request *req,
 220                    const struct req_msg_field *field,
 221                    void *data, u32 size)
 222 {
 223         struct req_capsule *pill = &req->rq_pill;
 224         struct lov_user_md *lmm;
 225         int rc = 0;
 226
 227         if (req_capsule_get_size(pill, field, RCL_CLIENT) < size) {
 228                 rc = sptlrpc_cli_enlarge_reqbuf(req, field, size);
 229                 if (rc) {
 230                         CERROR("%s: Can't enlarge ea size to %d: rc = %d\n",
 231                                req->rq_export->exp_obd->obd_name,
 232                                size, rc);
 233                         return rc;
 234                 }
 235         } else {
 236                 req_capsule_shrink(pill, field, size, RCL_CLIENT);
 237         }
 238
 239         req_capsule_set_size(pill, field, RCL_CLIENT, size);
 240         lmm = req_capsule_client_get(pill, field);
 241         if (lmm) {
 242                 memcpy(lmm, data, size);
 243                 /* overwrite layout generation returned from the MDS */
 244                 lmm->lmm_stripe_offset =
 245                   (typeof(lmm->lmm_stripe_offset))LOV_OFFSET_DEFAULT;
 246         }
 247
 248         return rc;
 249 }
 250
 251 static struct ptlrpc_request *
 252 mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
 253                      struct md_op_data *op_data, __u32 acl_bufsize)
 254 {
 255         struct ptlrpc_request   *req;
 256         struct obd_device       *obddev = class_exp2obd(exp);
 257         struct ldlm_intent      *lit;
 258         const void              *lmm = op_data->op_data;
 259         __u32                    lmmsize = op_data->op_data_size;
 260         __u32                    mdt_md_capsule_size;
 261         LIST_HEAD(cancels);
 262         int                      count = 0;
 263         enum ldlm_mode           mode;
 264         int                      rc;
 265         int repsize, repsize_estimate;
 266
 267         ENTRY;
 268
 269         mdt_md_capsule_size = obddev->u.cli.cl_default_mds_easize;
 270
 271         it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
 272
 273         /* XXX: openlock is not cancelled for cross-refs. */
 274         /* If inode is known, cancel conflicting OPEN locks. */
 275         if (fid_is_sane(&op_data->op_fid2)) {
 276                 if (it->it_flags & MDS_OPEN_LEASE) { /* try to get lease */
 277                         if (it->it_flags & MDS_FMODE_WRITE)
 278                                 mode = LCK_EX;
 279                         else
 280                                 mode = LCK_PR;
 281                 } else {
 282                         if (it->it_flags & (MDS_FMODE_WRITE | MDS_OPEN_TRUNC))
 283                                 mode = LCK_CW;
 284 #ifdef FMODE_EXEC
 285                         else if (it->it_flags & FMODE_EXEC)
 286                                 mode = LCK_PR;
 287 #endif
 288                         else
 289                                 mode = LCK_CR;
 290                 }
 291                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
 292                                                 &cancels, mode,
 293                                                 MDS_INODELOCK_OPEN);
 294         }
 295
 296         /* If CREATE, cancel parent's UPDATE lock. */
 297         if (it->it_op & IT_CREAT)
 298                 mode = LCK_EX;
 299         else
 300                 mode = LCK_CR;
 301         count += mdc_resource_get_unused(exp, &op_data->op_fid1,
 302                                          &cancels, mode,
 303                                          MDS_INODELOCK_UPDATE);
 304
 305         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 306                                    &RQF_LDLM_INTENT_OPEN);
 307         if (req == NULL) {
 308                 ldlm_lock_list_put(&cancels, l_bl_ast, count);
 309                 RETURN(ERR_PTR(-ENOMEM));
 310         }
 311
 312         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 313                              op_data->op_namelen + 1);
 314         if (cl_is_lov_delay_create(it->it_flags)) {
 315                 /* open(O_LOV_DELAY_CREATE) won't pack lmm */
 316                 LASSERT(lmmsize == 0);
 317                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
 318         } else {
 319                 req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT,
 320                              max(lmmsize, obddev->u.cli.cl_default_mds_easize));
 321         }
 322
 323         req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
 324                              RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
 325                              op_data->op_file_secctx_name_size : 0);
 326
 327         req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
 328                              op_data->op_file_secctx_size);
 329
 330         /* get SELinux policy info if any */
 331         rc = sptlrpc_get_sepol(req);
 332         if (rc < 0) {
 333                 ptlrpc_request_free(req);
 334                 RETURN(ERR_PTR(rc));
 335         }
 336         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 337                              strlen(req->rq_sepol) ?
 338                              strlen(req->rq_sepol) + 1 : 0);
 339
 340         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 341         if (rc < 0) {
 342                 ptlrpc_request_free(req);
 343                 RETURN(ERR_PTR(rc));
 344         }
 345
 346         spin_lock(&req->rq_lock);
 347         req->rq_replay = req->rq_import->imp_replayable;
 348         spin_unlock(&req->rq_lock);
 349
 350         /* pack the intent */
 351         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 352         lit->opc = (__u64)it->it_op;
 353
 354         /* pack the intended request */
 355         mdc_open_pack(req, op_data, it->it_create_mode, 0, it->it_flags, lmm,
 356                       lmmsize);
 357
 358         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
 359                              mdt_md_capsule_size);
 360         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 361
 362         if (!(it->it_op & IT_CREAT) && it->it_op & IT_OPEN &&
 363             req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
 364                                   RCL_CLIENT) &&
 365             op_data->op_file_secctx_name_size > 0 &&
 366             op_data->op_file_secctx_name != NULL) {
 367                 char *secctx_name;
 368
 369                 secctx_name = req_capsule_client_get(&req->rq_pill,
 370                                                      &RMF_FILE_SECCTX_NAME);
 371                 memcpy(secctx_name, op_data->op_file_secctx_name,
 372                        op_data->op_file_secctx_name_size);
 373                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
 374                                      RCL_SERVER,
 375                                      obddev->u.cli.cl_max_mds_easize);
 376
 377                 CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
 378                        op_data->op_file_secctx_name_size,
 379                        op_data->op_file_secctx_name);
 380
 381         } else {
 382                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
 383                                      RCL_SERVER, 0);
 384         }
 385
 386         /**
 387          * Inline buffer for possible data from Data-on-MDT files.
 388          */
 389         req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE, RCL_SERVER,
 390                              sizeof(struct niobuf_remote));
 391         ptlrpc_request_set_replen(req);
 392
 393         /* Get real repbuf allocated size as rounded up power of 2 */
 394         repsize = size_roundup_power2(req->rq_replen +
 395                                       lustre_msg_early_size());
 396         /* Estimate free space for DoM files in repbuf */
 397         repsize_estimate = repsize - (req->rq_replen -
 398                            mdt_md_capsule_size +
 399                            sizeof(struct lov_comp_md_v1) +
 400                            sizeof(struct lov_comp_md_entry_v1) +
 401                            lov_mds_md_size(0, LOV_MAGIC_V3));
 402
 403         if (repsize_estimate < obddev->u.cli.cl_dom_min_inline_repsize) {
 404                 repsize = obddev->u.cli.cl_dom_min_inline_repsize -
 405                           repsize_estimate + sizeof(struct niobuf_remote);
 406                 req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE,
 407                                      RCL_SERVER,
 408                                      sizeof(struct niobuf_remote) + repsize);
 409                 ptlrpc_request_set_replen(req);
 410                 CDEBUG(D_INFO, "Increase repbuf by %d bytes, total: %d\n",
 411                        repsize, req->rq_replen);
 412                 repsize = size_roundup_power2(req->rq_replen +
 413                                               lustre_msg_early_size());
 414         }
 415         /* The only way to report real allocated repbuf size to the server
 416          * is the lm_repsize but it must be set prior buffer allocation itself
 417          * due to security reasons - it is part of buffer used in signature
 418          * calculation (see LU-11414). Therefore the saved size is predicted
 419          * value as rq_replen rounded to the next higher power of 2.
 420          * Such estimation is safe. Though the final allocated buffer might
 421          * be even larger, it is not possible to know that at this point.
 422          */
 423         req->rq_reqmsg->lm_repsize = repsize;
 424         RETURN(req);
 425 }
 426
 427 #define GA_DEFAULT_EA_NAME_LEN 20
 428 #define GA_DEFAULT_EA_VAL_LEN  250
 429 #define GA_DEFAULT_EA_NUM      10
 430
 431 static struct ptlrpc_request *
 432 mdc_intent_getxattr_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 ldlm_intent      *lit;
 438         int                     rc, count = 0;
 439         LIST_HEAD(cancels);
 440         u32 ea_vals_buf_size = GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM;
 441
 442         ENTRY;
 443
 444         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 445                                         &RQF_LDLM_INTENT_GETXATTR);
 446         if (req == NULL)
 447                 RETURN(ERR_PTR(-ENOMEM));
 448
 449         /* get SELinux policy info if any */
 450         rc = sptlrpc_get_sepol(req);
 451         if (rc < 0) {
 452                 ptlrpc_request_free(req);
 453                 RETURN(ERR_PTR(rc));
 454         }
 455         req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
 456                              strlen(req->rq_sepol) ?
 457                              strlen(req->rq_sepol) + 1 : 0);
 458
 459         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 460         if (rc) {
 461                 ptlrpc_request_free(req);
 462                 RETURN(ERR_PTR(rc));
 463         }
 464
 465         /* pack the intent */
 466         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 467         lit->opc = IT_GETXATTR;
 468         CDEBUG(D_INFO, "%s: get xattrs for "DFID"\n",
 469                exp->exp_obd->obd_name, PFID(&op_data->op_fid1));
 470
 471 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
 472         /* If the supplied buffer is too small then the server will
 473          * return -ERANGE and llite will fallback to using non cached
 474          * xattr operations. On servers before 2.10.1 a (non-cached)
 475          * listxattr RPC for an orphan or dead file causes an oops. So
 476          * let's try to avoid sending too small a buffer to too old a
 477          * server. This is effectively undoing the memory conservation
 478          * of LU-9417 when it would be *more* likely to crash the
 479          * server. See LU-9856. */
 480         if (exp->exp_connect_data.ocd_version < OBD_OCD_VERSION(2, 10, 1, 0))
 481                 ea_vals_buf_size = max_t(u32, ea_vals_buf_size,
 482                                          exp->exp_connect_data.ocd_max_easize);
 483 #endif
 484
 485         /* pack the intended request */
 486         mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
 487                       ea_vals_buf_size, -1, 0);
 488
 489         /* get SELinux policy info if any */
 490         mdc_file_sepol_pack(req);
 491
 492         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
 493                              GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM);
 494
 495         req_capsule_set_size(&req->rq_pill, &RMF_EAVALS, RCL_SERVER,
 496                              ea_vals_buf_size);
 497
 498         req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS, RCL_SERVER,
 499                              sizeof(u32) * GA_DEFAULT_EA_NUM);
 500
 501         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
 502
 503         ptlrpc_request_set_replen(req);
 504
 505         RETURN(req);
 506 }
 507
 508 static struct ptlrpc_request *
 509 mdc_intent_getattr_pack(struct obd_export *exp, struct lookup_intent *it,
 510                         struct md_op_data *op_data, __u32 acl_bufsize)
 511 {
 512         struct ptlrpc_request *req;
 513         struct obd_device *obddev = class_exp2obd(exp);
 514         u64 valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE |
 515                     OBD_MD_FLDIREA | OBD_MD_MEA | OBD_MD_FLACL |
 516                     OBD_MD_DEFAULT_MEA;
 517         struct ldlm_intent *lit;
 518         __u32 easize;
 519         bool have_secctx = false;
 520         int rc;
 521
 522         ENTRY;
 523
 524         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 525                                    &RQF_LDLM_INTENT_GETATTR);
 526         if (req == NULL)
 527                 RETURN(ERR_PTR(-ENOMEM));
 528
 529         /* send name of security xattr to get upon intent */
 530         if (it->it_op & (IT_LOOKUP | IT_GETATTR) &&
 531             req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
 532                                   RCL_CLIENT) &&
 533             op_data->op_file_secctx_name_size > 0 &&
 534             op_data->op_file_secctx_name != NULL) {
 535                 have_secctx = true;
 536                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
 537                                      RCL_CLIENT,
 538                                      op_data->op_file_secctx_name_size);
 539         }
 540
 541         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 542                              op_data->op_namelen + 1);
 543
 544         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 545         if (rc) {
 546                 ptlrpc_request_free(req);
 547                 RETURN(ERR_PTR(rc));
 548         }
 549
 550         /* pack the intent */
 551         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 552         lit->opc = (__u64)it->it_op;
 553
 554         easize = obddev->u.cli.cl_default_mds_easize;
 555
 556         /* pack the intended request */
 557         mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
 558
 559         req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
 560         req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 561         req_capsule_set_size(&req->rq_pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER,
 562                              sizeof(struct lmv_user_md));
 563
 564         if (have_secctx) {
 565                 char *secctx_name;
 566
 567                 secctx_name = req_capsule_client_get(&req->rq_pill,
 568                                                      &RMF_FILE_SECCTX_NAME);
 569                 memcpy(secctx_name, op_data->op_file_secctx_name,
 570                        op_data->op_file_secctx_name_size);
 571
 572                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
 573                                      RCL_SERVER, easize);
 574
 575                 CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
 576                        op_data->op_file_secctx_name_size,
 577                        op_data->op_file_secctx_name);
 578         } else {
 579                 req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
 580                                      RCL_SERVER, 0);
 581         }
 582
 583         ptlrpc_request_set_replen(req);
 584         RETURN(req);
 585 }
 586
 587 static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
 588                                                      struct lookup_intent *it,
 589                                                      struct md_op_data *op_data)
 590 {
 591         struct obd_device     *obd = class_exp2obd(exp);
 592         LIST_HEAD(cancels);
 593         struct ptlrpc_request *req;
 594         struct ldlm_intent    *lit;
 595         struct layout_intent  *layout;
 596         int count = 0, rc;
 597         ENTRY;
 598
 599         req = ptlrpc_request_alloc(class_exp2cliimp(exp),
 600                                 &RQF_LDLM_INTENT_LAYOUT);
 601         if (req == NULL)
 602                 RETURN(ERR_PTR(-ENOMEM));
 603
 604         if (fid_is_sane(&op_data->op_fid2) && (it->it_op & IT_LAYOUT) &&
 605             (it->it_flags & FMODE_WRITE)) {
 606                 count = mdc_resource_get_unused(exp, &op_data->op_fid2,
 607                                                 &cancels, LCK_EX,
 608                                                 MDS_INODELOCK_LAYOUT);
 609         }
 610
 611         req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
 612         rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
 613         if (rc) {
 614                 ptlrpc_request_free(req);
 615                 RETURN(ERR_PTR(rc));
 616         }
 617
 618         /* pack the intent */
 619         lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
 620         lit->opc = (__u64)it->it_op;
 621
 622         /* pack the layout intent request */
 623         layout = req_capsule_client_get(&req->rq_pill, &RMF_LAYOUT_INTENT);
 624         LASSERT(op_data->op_data != NULL);
 625         LASSERT(op_data->op_data_size == sizeof(*layout));
 626         memcpy(layout, op_data->op_data, sizeof(*layout));
 627
 628         req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER,
 629                              obd->u.cli.cl_default_mds_easize);
 630         ptlrpc_request_set_replen(req);
 631         RETURN(req);
 632 }
 633
 634 static struct ptlrpc_request *
 635 mdc_enqueue_pack(struct obd_export *exp, int lvb_len)
 636 {
 637         struct ptlrpc_request *req;
 638         int rc;
 639         ENTRY;
 640
 641         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LDLM_ENQUEUE);
 642         if (req == NULL)
 643                 RETURN(ERR_PTR(-ENOMEM));
 644
 645         rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
 646         if (rc) {
 647                 ptlrpc_request_free(req);
 648                 RETURN(ERR_PTR(rc));
 649         }
 650
 651         req_capsule_set_size(&req->rq_pill, &RMF_DLM_LVB, RCL_SERVER, lvb_len);
 652         ptlrpc_request_set_replen(req);
 653         RETURN(req);
 654 }
 655
 656 static int mdc_finish_enqueue(struct obd_export *exp,
 657                               struct ptlrpc_request *req,
 658                               struct ldlm_enqueue_info *einfo,
 659                               struct lookup_intent *it,
 660                               struct lustre_handle *lockh,
 661                               int rc)
 662 {
 663         struct req_capsule  *pill = &req->rq_pill;
 664         struct ldlm_request *lockreq;
 665         struct ldlm_reply   *lockrep;
 666         struct ldlm_lock    *lock;
 667         struct mdt_body     *body = NULL;
 668         void                *lvb_data = NULL;
 669         __u32                lvb_len = 0;
 670
 671         ENTRY;
 672
 673         LASSERT(rc >= 0);
 674         /* Similarly, if we're going to replay this request, we don't want to
 675          * actually get a lock, just perform the intent. */
 676         if (req->rq_transno || req->rq_replay) {
 677                 lockreq = req_capsule_client_get(pill, &RMF_DLM_REQ);
 678                 lockreq->lock_flags |= ldlm_flags_to_wire(LDLM_FL_INTENT_ONLY);
 679         }
 680
 681         if (rc == ELDLM_LOCK_ABORTED) {
 682                 einfo->ei_mode = 0;
 683                 memset(lockh, 0, sizeof(*lockh));
 684                 rc = 0;
 685         } else { /* rc = 0 */
 686                 lock = ldlm_handle2lock(lockh);
 687                 LASSERT(lock != NULL);
 688
 689                 /* If the server gave us back a different lock mode, we should
 690                  * fix up our variables. */
 691                 if (lock->l_req_mode != einfo->ei_mode) {
 692                         ldlm_lock_addref(lockh, lock->l_req_mode);
 693                         ldlm_lock_decref(lockh, einfo->ei_mode);
 694                         einfo->ei_mode = lock->l_req_mode;
 695                 }
 696                 LDLM_LOCK_PUT(lock);
 697         }
 698
 699         lockrep = req_capsule_server_get(pill, &RMF_DLM_REP);
 700         LASSERT(lockrep != NULL); /* checked by ldlm_cli_enqueue() */
 701
 702         it->it_disposition = (int)lockrep->lock_policy_res1;
 703         it->it_status = (int)lockrep->lock_policy_res2;
 704         it->it_lock_mode = einfo->ei_mode;
 705         it->it_lock_handle = lockh->cookie;
 706         it->it_request = req;
 707
 708         /* Technically speaking rq_transno must already be zero if
 709          * it_status is in error, so the check is a bit redundant */
 710         if ((!req->rq_transno || it->it_status < 0) && req->rq_replay)
 711                 mdc_clear_replay_flag(req, it->it_status);
 712
 713         /* If we're doing an IT_OPEN which did not result in an actual
 714          * successful open, then we need to remove the bit which saves
 715          * this request for unconditional replay.
 716          *
 717          * It's important that we do this first!  Otherwise we might exit the
 718          * function without doing so, and try to replay a failed create
 719          * (bug 3440) */
 720         if (it->it_op & IT_OPEN && req->rq_replay &&
 721             (!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
 722                 mdc_clear_replay_flag(req, it->it_status);
 723
 724         DEBUG_REQ(D_RPCTRACE, req, "op=%x disposition=%x, status=%d",
 725                   it->it_op, it->it_disposition, it->it_status);
 726
 727         /* We know what to expect, so we do any byte flipping required here */
 728         if (it_has_reply_body(it)) {
 729                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 730                 if (body == NULL) {
 731                         CERROR ("Can't swab mdt_body\n");
 732                         RETURN (-EPROTO);
 733                 }
 734
 735                 if (it_disposition(it, DISP_OPEN_OPEN) &&
 736                     !it_open_error(DISP_OPEN_OPEN, it)) {
 737                         /*
 738                          * If this is a successful OPEN request, we need to set
 739                          * replay handler and data early, so that if replay
 740                          * happens immediately after swabbing below, new reply
 741                          * is swabbed by that handler correctly.
 742                          */
 743                         mdc_set_open_replay_data(NULL, NULL, it);
 744                 }
 745
 746                 if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) {
 747                         void *eadata;
 748
 749                         mdc_update_max_ea_from_body(exp, body);
 750
 751                         /*
 752                          * The eadata is opaque; just check that it is there.
 753                          * Eventually, obd_unpackmd() will check the contents.
 754                          */
 755                         eadata = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
 756                                                         body->mbo_eadatasize);
 757                         if (eadata == NULL)
 758                                 RETURN(-EPROTO);
 759
 760                         /* save lvb data and length in case this is for layout
 761                          * lock */
 762                         lvb_data = eadata;
 763                         lvb_len = body->mbo_eadatasize;
 764
 765                         /*
 766                          * We save the reply LOV EA in case we have to replay a
 767                          * create for recovery.  If we didn't allocate a large
 768                          * enough request buffer above we need to reallocate it
 769                          * here to hold the actual LOV EA.
 770                          *
 771                          * To not save LOV EA if request is not going to replay
 772                          * (for example error one).
 773                          */
 774                         if ((it->it_op & IT_OPEN) && req->rq_replay) {
 775                                 rc = mdc_save_lovea(req, &RMF_EADATA, eadata,
 776                                                     body->mbo_eadatasize);
 777                                 if (rc) {
 778                                         body->mbo_valid &= ~OBD_MD_FLEASIZE;
 779                                         body->mbo_eadatasize = 0;
 780                                         rc = 0;
 781                                 }
 782                         }
 783                 }
 784         } else if (it->it_op & IT_LAYOUT) {
 785                 /* maybe the lock was granted right away and layout
 786                  * is packed into RMF_DLM_LVB of req */
 787                 lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
 788                 CDEBUG(D_INFO, "%s: layout return lvb %d transno %lld\n",
 789                        class_exp2obd(exp)->obd_name, lvb_len, req->rq_transno);
 790                 if (lvb_len > 0) {
 791                         lvb_data = req_capsule_server_sized_get(pill,
 792                                                         &RMF_DLM_LVB, lvb_len);
 793                         if (lvb_data == NULL)
 794                                 RETURN(-EPROTO);
 795
 796                         /**
 797                          * save replied layout data to the request buffer for
 798                          * recovery consideration (lest MDS reinitialize
 799                          * another set of OST objects).
 800                          */
 801                         if (req->rq_transno)
 802                                 (void)mdc_save_lovea(req, &RMF_EADATA, lvb_data,
 803                                                      lvb_len);
 804                 }
 805         }
 806
 807         /* fill in stripe data for layout lock.
 808          * LU-6581: trust layout data only if layout lock is granted. The MDT
 809          * has stopped sending layout unless the layout lock is granted. The
 810          * client still does this checking in case it's talking with an old
 811          * server. - Jinshan */
 812         lock = ldlm_handle2lock(lockh);
 813         if (lock == NULL)
 814                 RETURN(rc);
 815
 816         if (ldlm_has_layout(lock) && lvb_data != NULL &&
 817             !(lockrep->lock_flags & LDLM_FL_BLOCKED_MASK)) {
 818                 void *lmm;
 819
 820                 LDLM_DEBUG(lock, "layout lock returned by: %s, lvb_len: %d",
 821                         ldlm_it2str(it->it_op), lvb_len);
 822
 823                 OBD_ALLOC_LARGE(lmm, lvb_len);
 824                 if (lmm == NULL)
 825                         GOTO(out_lock, rc = -ENOMEM);
 826
 827                 memcpy(lmm, lvb_data, lvb_len);
 828
 829                 /* install lvb_data */
 830                 lock_res_and_lock(lock);
 831                 if (lock->l_lvb_data == NULL) {
 832                         lock->l_lvb_type = LVB_T_LAYOUT;
 833                         lock->l_lvb_data = lmm;
 834                         lock->l_lvb_len = lvb_len;
 835                         lmm = NULL;
 836                 }
 837                 unlock_res_and_lock(lock);
 838                 if (lmm != NULL)
 839                         OBD_FREE_LARGE(lmm, lvb_len);
 840         }
 841
 842         if (ldlm_has_dom(lock)) {
 843                 LASSERT(lock->l_glimpse_ast == mdc_ldlm_glimpse_ast);
 844
 845                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 846                 if (!(body->mbo_valid & OBD_MD_DOM_SIZE)) {
 847                         LDLM_ERROR(lock, "%s: DoM lock without size.",
 848                                    exp->exp_obd->obd_name);
 849                         GOTO(out_lock, rc = -EPROTO);
 850                 }
 851
 852                 LDLM_DEBUG(lock, "DoM lock is returned by: %s, size: %llu",
 853                            ldlm_it2str(it->it_op), body->mbo_dom_size);
 854
 855                 rc = mdc_fill_lvb(req, &lock->l_ost_lvb);
 856         }
 857 out_lock:
 858         LDLM_LOCK_PUT(lock);
 859
 860         RETURN(rc);
 861 }
 862
 863 static inline bool mdc_skip_mod_rpc_slot(const struct lookup_intent *it)
 864 {
 865         if (it != NULL &&
 866             (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
 867              it->it_op == IT_READDIR ||
 868              (it->it_op == IT_LAYOUT && !(it->it_flags & MDS_FMODE_WRITE))))
 869                 return true;
 870         return false;
 871 }
 872
 873 /* We always reserve enough space in the reply packet for a stripe MD, because
 874  * we don't know in advance the file type. */
 875 static int mdc_enqueue_base(struct obd_export *exp,
 876                             struct ldlm_enqueue_info *einfo,
 877                             const union ldlm_policy_data *policy,
 878                             struct lookup_intent *it,
 879                             struct md_op_data *op_data,
 880                             struct lustre_handle *lockh,
 881                             __u64 extra_lock_flags)
 882 {
 883         struct obd_device *obddev = class_exp2obd(exp);
 884         struct ptlrpc_request *req;
 885         __u64 flags, saved_flags = extra_lock_flags;
 886         struct ldlm_res_id res_id;
 887         static const union ldlm_policy_data lookup_policy = {
 888                                   .l_inodebits = { MDS_INODELOCK_LOOKUP } };
 889         static const union ldlm_policy_data update_policy = {
 890                                   .l_inodebits = { MDS_INODELOCK_UPDATE } };
 891         static const union ldlm_policy_data layout_policy = {
 892                                   .l_inodebits = { MDS_INODELOCK_LAYOUT } };
 893         static const union ldlm_policy_data getxattr_policy = {
 894                                   .l_inodebits = { MDS_INODELOCK_XATTR } };
 895         int generation, resends = 0;
 896         struct ldlm_reply *lockrep;
 897         struct obd_import *imp = class_exp2cliimp(exp);
 898         __u32 acl_bufsize;
 899         enum lvb_type lvb_type = 0;
 900         int rc;
 901         ENTRY;
 902
 903         LASSERTF(!it || einfo->ei_type == LDLM_IBITS, "lock type %d\n",
 904                  einfo->ei_type);
 905         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
 906
 907         if (it != NULL) {
 908                 LASSERT(policy == NULL);
 909
 910                 saved_flags |= LDLM_FL_HAS_INTENT;
 911                 if (it->it_op & (IT_GETATTR | IT_READDIR))
 912                         policy = &update_policy;
 913                 else if (it->it_op & IT_LAYOUT)
 914                         policy = &layout_policy;
 915                 else if (it->it_op & IT_GETXATTR)
 916                         policy = &getxattr_policy;
 917                 else
 918                         policy = &lookup_policy;
 919         }
 920
 921         generation = obddev->u.cli.cl_import->imp_generation;
 922         if (!it || (it->it_op & (IT_OPEN | IT_CREAT)))
 923                 acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
 924                                   XATTR_SIZE_MAX);
 925         else
 926                 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 927
 928 resend:
 929         flags = saved_flags;
 930         if (it == NULL) {
 931                 /* The only way right now is FLOCK. */
 932                 LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
 933                          einfo->ei_type);
 934                 res_id.name[3] = LDLM_FLOCK;
 935                 req = ldlm_enqueue_pack(exp, 0);
 936         } else if (it->it_op & IT_OPEN) {
 937                 req = mdc_intent_open_pack(exp, it, op_data, acl_bufsize);
 938         } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
 939                 req = mdc_intent_getattr_pack(exp, it, op_data, acl_bufsize);
 940         } else if (it->it_op & IT_READDIR) {
 941                 req = mdc_enqueue_pack(exp, 0);
 942         } else if (it->it_op & IT_LAYOUT) {
 943                 if (!imp_connect_lvb_type(imp))
 944                         RETURN(-EOPNOTSUPP);
 945                 req = mdc_intent_layout_pack(exp, it, op_data);
 946                 lvb_type = LVB_T_LAYOUT;
 947         } else if (it->it_op & IT_GETXATTR) {
 948                 req = mdc_intent_getxattr_pack(exp, it, op_data);
 949         } else {
 950                 LBUG();
 951                 RETURN(-EINVAL);
 952         }
 953
 954         if (IS_ERR(req))
 955                 RETURN(PTR_ERR(req));
 956
 957         if (resends) {
 958                 req->rq_generation_set = 1;
 959                 req->rq_import_generation = generation;
 960                 req->rq_sent = ktime_get_real_seconds() + resends;
 961         }
 962
 963         einfo->ei_enq_slot = !mdc_skip_mod_rpc_slot(it);
 964
 965         /* With Data-on-MDT the glimpse callback is needed too.
 966          * It is set here in advance but not in mdc_finish_enqueue()
 967          * to avoid possible races. It is safe to have glimpse handler
 968          * for non-DOM locks and costs nothing.*/
 969         if (einfo->ei_cb_gl == NULL)
 970                 einfo->ei_cb_gl = mdc_ldlm_glimpse_ast;
 971
 972         rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
 973                               0, lvb_type, lockh, 0);
 974
 975         if (!it) {
 976                 /* For flock requests we immediatelly return without further
 977                    delay and let caller deal with the rest, since rest of
 978                    this function metadata processing makes no sense for flock
 979                    requests anyway. But in case of problem during comms with
 980                    Server (ETIMEDOUT) or any signal/kill attempt (EINTR), we
 981                    can not rely on caller and this mainly for F_UNLCKs
 982                    (explicits or automatically generated by Kernel to clean
 983                    current FLocks upon exit) that can't be trashed */
 984                 if (((rc == -EINTR) || (rc == -ETIMEDOUT)) &&
 985                     (einfo->ei_type == LDLM_FLOCK) &&
 986                     (einfo->ei_mode == LCK_NL))
 987                         goto resend;
 988                 ptlrpc_req_finished(req);
 989                 RETURN(rc);
 990         }
 991
 992         if (rc < 0) {
 993                 CDEBUG(D_INFO,
 994                       "%s: ldlm_cli_enqueue "DFID":"DFID"=%s failed: rc = %d\n",
 995                       obddev->obd_name, PFID(&op_data->op_fid1),
 996                       PFID(&op_data->op_fid2), op_data->op_name ?: "", rc);
 997
 998                 mdc_clear_replay_flag(req, rc);
 999                 ptlrpc_req_finished(req);
1000                 RETURN(rc);
1001         }
1002
1003         lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1004         LASSERT(lockrep != NULL);
1005
1006         lockrep->lock_policy_res2 =
1007                 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1008
1009         /* Retry infinitely when the server returns -EINPROGRESS for the
1010          * intent operation, when server returns -EINPROGRESS for acquiring
1011          * intent lock, we'll retry in after_reply(). */
1012         if (it && (int)lockrep->lock_policy_res2 == -EINPROGRESS) {
1013                 mdc_clear_replay_flag(req, rc);
1014                 ptlrpc_req_finished(req);
1015                 if (generation == obddev->u.cli.cl_import->imp_generation) {
1016                         if (signal_pending(current))
1017                                 RETURN(-EINTR);
1018
1019                         resends++;
1020                         CDEBUG(D_HA, "%s: resend:%d op:%d "DFID"/"DFID"\n",
1021                                obddev->obd_name, resends, it->it_op,
1022                                PFID(&op_data->op_fid1),
1023                                PFID(&op_data->op_fid2));
1024                         goto resend;
1025                 } else {
1026                         CDEBUG(D_HA, "resend cross eviction\n");
1027                         RETURN(-EIO);
1028                 }
1029         }
1030
1031         if ((int)lockrep->lock_policy_res2 == -ERANGE &&
1032             it->it_op & (IT_OPEN | IT_GETATTR | IT_LOOKUP) &&
1033             acl_bufsize == LUSTRE_POSIX_ACL_MAX_SIZE_OLD) {
1034                 mdc_clear_replay_flag(req, -ERANGE);
1035                 ptlrpc_req_finished(req);
1036                 acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
1037                                   XATTR_SIZE_MAX);
1038                 goto resend;
1039         }
1040
1041         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1042         if (rc < 0) {
1043                 if (lustre_handle_is_used(lockh)) {
1044                         ldlm_lock_decref(lockh, einfo->ei_mode);
1045                         memset(lockh, 0, sizeof(*lockh));
1046                 }
1047                 ptlrpc_req_finished(req);
1048
1049                 it->it_lock_handle = 0;
1050                 it->it_lock_mode = 0;
1051                 it->it_request = NULL;
1052         }
1053
1054         RETURN(rc);
1055 }
1056
1057 int mdc_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo,
1058                 const union ldlm_policy_data *policy,
1059                 struct md_op_data *op_data,
1060                 struct lustre_handle *lockh, __u64 extra_lock_flags)
1061 {
1062         return mdc_enqueue_base(exp, einfo, policy, NULL,
1063                                 op_data, lockh, extra_lock_flags);
1064 }
1065
1066 static int mdc_finish_intent_lock(struct obd_export *exp,
1067                                   struct ptlrpc_request *request,
1068                                   struct md_op_data *op_data,
1069                                   struct lookup_intent *it,
1070                                   struct lustre_handle *lockh)
1071 {
1072         struct lustre_handle old_lock;
1073         struct ldlm_lock *lock;
1074         int rc = 0;
1075         ENTRY;
1076
1077         LASSERT(request != NULL);
1078         LASSERT(request != LP_POISON);
1079         LASSERT(request->rq_repmsg != LP_POISON);
1080
1081         if (it->it_op & IT_READDIR)
1082                 RETURN(0);
1083
1084         if (it->it_op & (IT_GETXATTR | IT_LAYOUT)) {
1085                 if (it->it_status != 0)
1086                         GOTO(out, rc = it->it_status);
1087         } else {
1088                 if (!it_disposition(it, DISP_IT_EXECD)) {
1089                         /* The server failed before it even started executing
1090                          * the intent, i.e. because it couldn't unpack the
1091                          * request.
1092                          */
1093                         LASSERT(it->it_status != 0);
1094                         GOTO(out, rc = it->it_status);
1095                 }
1096                 rc = it_open_error(DISP_IT_EXECD, it);
1097                 if (rc)
1098                         GOTO(out, rc);
1099
1100                 rc = it_open_error(DISP_LOOKUP_EXECD, it);
1101                 if (rc)
1102                         GOTO(out, rc);
1103
1104                 /* keep requests around for the multiple phases of the call
1105                  * this shows the DISP_XX must guarantee we make it into the
1106                  * call
1107                  */
1108                 if (!it_disposition(it, DISP_ENQ_CREATE_REF) &&
1109                     it_disposition(it, DISP_OPEN_CREATE) &&
1110                     !it_open_error(DISP_OPEN_CREATE, it)) {
1111                         it_set_disposition(it, DISP_ENQ_CREATE_REF);
1112                         /* balanced in ll_create_node */
1113                         ptlrpc_request_addref(request);
1114                 }
1115                 if (!it_disposition(it, DISP_ENQ_OPEN_REF) &&
1116                     it_disposition(it, DISP_OPEN_OPEN) &&
1117                     !it_open_error(DISP_OPEN_OPEN, it)) {
1118                         it_set_disposition(it, DISP_ENQ_OPEN_REF);
1119                         /* balanced in ll_file_open */
1120                         ptlrpc_request_addref(request);
1121                         /* BUG 11546 - eviction in the middle of open rpc
1122                          * processing
1123                          */
1124                         OBD_FAIL_TIMEOUT(OBD_FAIL_MDC_ENQUEUE_PAUSE,
1125                                          obd_timeout);
1126                 }
1127
1128                 if (it->it_op & IT_CREAT) {
1129                         /* XXX this belongs in ll_create_it */
1130                 } else if (it->it_op == IT_OPEN) {
1131                         LASSERT(!it_disposition(it, DISP_OPEN_CREATE));
1132                 } else {
1133                         LASSERT(it->it_op & (IT_GETATTR | IT_LOOKUP));
1134                 }
1135         }
1136
1137         /* If we already have a matching lock, then cancel the new
1138          * one.  We have to set the data here instead of in
1139          * mdc_enqueue, because we need to use the child's inode as
1140          * the l_ast_data to match, and that's not available until
1141          * intent_finish has performed the iget().) */
1142         lock = ldlm_handle2lock(lockh);
1143         if (lock) {
1144                 union ldlm_policy_data policy = lock->l_policy_data;
1145                 LDLM_DEBUG(lock, "matching against this");
1146
1147                 if (it_has_reply_body(it)) {
1148                         struct mdt_body *body;
1149
1150                         body = req_capsule_server_get(&request->rq_pill,
1151                                                       &RMF_MDT_BODY);
1152                         /* mdc_enqueue checked */
1153                         LASSERT(body != NULL);
1154                         LASSERTF(fid_res_name_eq(&body->mbo_fid1,
1155                                                  &lock->l_resource->lr_name),
1156                                  "Lock res_id: "DLDLMRES", fid: "DFID"\n",
1157                                  PLDLMRES(lock->l_resource),
1158                                  PFID(&body->mbo_fid1));
1159                 }
1160                 LDLM_LOCK_PUT(lock);
1161
1162                 memcpy(&old_lock, lockh, sizeof(*lockh));
1163                 if (ldlm_lock_match(NULL, LDLM_FL_BLOCK_GRANTED, NULL,
1164                                     LDLM_IBITS, &policy, LCK_NL, &old_lock, 0)) {
1165                         ldlm_lock_decref_and_cancel(lockh, it->it_lock_mode);
1166                         memcpy(lockh, &old_lock, sizeof(old_lock));
1167                         it->it_lock_handle = lockh->cookie;
1168                 }
1169         }
1170
1171         EXIT;
1172 out:
1173         CDEBUG(D_DENTRY,"D_IT dentry %.*s intent: %s status %d disp %x rc %d\n",
1174                 (int)op_data->op_namelen, op_data->op_name,
1175                 ldlm_it2str(it->it_op), it->it_status,
1176                 it->it_disposition, rc);
1177         return rc;
1178 }
1179
1180 int mdc_revalidate_lock(struct obd_export *exp, struct lookup_intent *it,
1181                         struct lu_fid *fid, __u64 *bits)
1182 {
1183         /* We could just return 1 immediately, but since we should only
1184          * be called in revalidate_it if we already have a lock, let's
1185          * verify that. */
1186         struct ldlm_res_id res_id;
1187         struct lustre_handle lockh;
1188         union ldlm_policy_data policy;
1189         enum ldlm_mode mode;
1190         ENTRY;
1191
1192         if (it->it_lock_handle) {
1193                 lockh.cookie = it->it_lock_handle;
1194                 mode = ldlm_revalidate_lock_handle(&lockh, bits);
1195         } else {
1196                 fid_build_reg_res_name(fid, &res_id);
1197                 switch (it->it_op) {
1198                 case IT_GETATTR:
1199                         /* File attributes are held under multiple bits:
1200                          * nlink is under lookup lock, size and times are
1201                          * under UPDATE lock and recently we've also got
1202                          * a separate permissions lock for owner/group/acl that
1203                          * were protected by lookup lock before.
1204                          * Getattr must provide all of that information,
1205                          * so we need to ensure we have all of those locks.
1206                          * Unfortunately, if the bits are split across multiple
1207                          * locks, there's no easy way to match all of them here,
1208                          * so an extra RPC would be performed to fetch all
1209                          * of those bits at once for now. */
1210                         /* For new MDTs(> 2.4), UPDATE|PERM should be enough,
1211                          * but for old MDTs (< 2.4), permission is covered
1212                          * by LOOKUP lock, so it needs to match all bits here.*/
1213                         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE |
1214                                                   MDS_INODELOCK_LOOKUP |
1215                                                   MDS_INODELOCK_PERM;
1216                         break;
1217                 case IT_READDIR:
1218                         policy.l_inodebits.bits = MDS_INODELOCK_UPDATE;
1219                         break;
1220                 case IT_LAYOUT:
1221                         policy.l_inodebits.bits = MDS_INODELOCK_LAYOUT;
1222                         break;
1223                 default:
1224                         policy.l_inodebits.bits = MDS_INODELOCK_LOOKUP;
1225                         break;
1226                 }
1227
1228                 mode = mdc_lock_match(exp, LDLM_FL_BLOCK_GRANTED, fid,
1229                                       LDLM_IBITS, &policy,
1230                                       LCK_CR | LCK_CW | LCK_PR | LCK_PW,
1231                                       &lockh);
1232         }
1233
1234         if (mode) {
1235                 it->it_lock_handle = lockh.cookie;
1236                 it->it_lock_mode = mode;
1237         } else {
1238                 it->it_lock_handle = 0;
1239                 it->it_lock_mode = 0;
1240         }
1241
1242         RETURN(!!mode);
1243 }
1244
1245 /*
1246  * This long block is all about fixing up the lock and request state
1247  * so that it is correct as of the moment _before_ the operation was
1248  * applied; that way, the VFS will think that everything is normal and
1249  * call Lustre's regular VFS methods.
1250  *
1251  * If we're performing a creation, that means that unless the creation
1252  * failed with EEXIST, we should fake up a negative dentry.
1253  *
1254  * For everything else, we want to lookup to succeed.
1255  *
1256  * One additional note: if CREATE or OPEN succeeded, we add an extra
1257  * reference to the request because we need to keep it around until
1258  * ll_create/ll_open gets called.
1259  *
1260  * The server will return to us, in it_disposition, an indication of
1261  * exactly what it_status refers to.
1262  *
1263  * If DISP_OPEN_OPEN is set, then it_status refers to the open() call,
1264  * otherwise if DISP_OPEN_CREATE is set, then it status is the
1265  * creation failure mode.  In either case, one of DISP_LOOKUP_NEG or
1266  * DISP_LOOKUP_POS will be set, indicating whether the child lookup
1267  * was successful.
1268  *
1269  * Else, if DISP_LOOKUP_EXECD then it_status is the rc of the
1270  * child lookup.
1271  */
1272 int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
1273                     struct lookup_intent *it, struct ptlrpc_request **reqp,
1274                     ldlm_blocking_callback cb_blocking, __u64 extra_lock_flags)
1275 {
1276         struct ldlm_enqueue_info einfo = {
1277                 .ei_type        = LDLM_IBITS,
1278                 .ei_mode        = it_to_lock_mode(it),
1279                 .ei_cb_bl       = cb_blocking,
1280                 .ei_cb_cp       = ldlm_completion_ast,
1281                 .ei_cb_gl       = mdc_ldlm_glimpse_ast,
1282         };
1283         struct lustre_handle lockh;
1284         int rc = 0;
1285         ENTRY;
1286         LASSERT(it);
1287
1288         CDEBUG(D_DLMTRACE, "(name: %.*s,"DFID") in obj "DFID
1289                 ", intent: %s flags %#llo\n", (int)op_data->op_namelen,
1290                 op_data->op_name, PFID(&op_data->op_fid2),
1291                 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op),
1292                 it->it_flags);
1293
1294         lockh.cookie = 0;
1295         if (fid_is_sane(&op_data->op_fid2) &&
1296             (it->it_op & (IT_LOOKUP | IT_GETATTR | IT_READDIR))) {
1297                 /* We could just return 1 immediately, but since we should only
1298                  * be called in revalidate_it if we already have a lock, let's
1299                  * verify that. */
1300                 it->it_lock_handle = 0;
1301                 rc = mdc_revalidate_lock(exp, it, &op_data->op_fid2, NULL);
1302                 /* Only return failure if it was not GETATTR by cfid
1303                    (from inode_revalidate) */
1304                 if (rc || op_data->op_namelen != 0)
1305                         RETURN(rc);
1306         }
1307
1308         /* For case if upper layer did not alloc fid, do it now. */
1309         if (!fid_is_sane(&op_data->op_fid2) && it->it_op & IT_CREAT) {
1310                 rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2, op_data);
1311                 if (rc < 0) {
1312                         CERROR("Can't alloc new fid, rc %d\n", rc);
1313                         RETURN(rc);
1314                 }
1315         }
1316
1317         rc = mdc_enqueue_base(exp, &einfo, NULL, it, op_data, &lockh,
1318                               extra_lock_flags);
1319         if (rc < 0)
1320                 RETURN(rc);
1321
1322         *reqp = it->it_request;
1323         rc = mdc_finish_intent_lock(exp, *reqp, op_data, it, &lockh);
1324         RETURN(rc);
1325 }
1326
1327 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
1328                                               struct ptlrpc_request *req,
1329                                               void *args, int rc)
1330 {
1331         struct mdc_getattr_args  *ga = args;
1332         struct obd_export *exp = ga->ga_exp;
1333         struct md_enqueue_info *minfo = ga->ga_minfo;
1334         struct ldlm_enqueue_info *einfo = &minfo->mi_einfo;
1335         struct lookup_intent     *it;
1336         struct lustre_handle     *lockh;
1337         struct ldlm_reply        *lockrep;
1338         __u64                     flags = LDLM_FL_HAS_INTENT;
1339         ENTRY;
1340
1341         it    = &minfo->mi_it;
1342         lockh = &minfo->mi_lockh;
1343
1344         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
1345                 rc = -ETIMEDOUT;
1346
1347         rc = ldlm_cli_enqueue_fini(exp, req, einfo->ei_type, 1, einfo->ei_mode,
1348                                    &flags, NULL, 0, lockh, rc);
1349         if (rc < 0) {
1350                 CERROR("ldlm_cli_enqueue_fini: %d\n", rc);
1351                 mdc_clear_replay_flag(req, rc);
1352                 GOTO(out, rc);
1353         }
1354
1355         lockrep = req_capsule_server_get(&req->rq_pill, &RMF_DLM_REP);
1356         LASSERT(lockrep != NULL);
1357
1358         lockrep->lock_policy_res2 =
1359                 ptlrpc_status_ntoh(lockrep->lock_policy_res2);
1360
1361         rc = mdc_finish_enqueue(exp, req, einfo, it, lockh, rc);
1362         if (rc)
1363                 GOTO(out, rc);
1364
1365         rc = mdc_finish_intent_lock(exp, req, &minfo->mi_data, it, lockh);
1366         EXIT;
1367
1368 out:
1369         minfo->mi_cb(req, minfo, rc);
1370         return 0;
1371 }
1372
1373 int mdc_intent_getattr_async(struct obd_export *exp,
1374                              struct md_enqueue_info *minfo)
1375 {
1376         struct md_op_data       *op_data = &minfo->mi_data;
1377         struct lookup_intent    *it = &minfo->mi_it;
1378         struct ptlrpc_request   *req;
1379         struct mdc_getattr_args *ga;
1380         struct ldlm_res_id       res_id;
1381         union ldlm_policy_data policy = {
1382                                 .l_inodebits = { MDS_INODELOCK_LOOKUP |
1383                                                  MDS_INODELOCK_UPDATE } };
1384         int                      rc = 0;
1385         __u64                    flags = LDLM_FL_HAS_INTENT;
1386         ENTRY;
1387
1388         CDEBUG(D_DLMTRACE, "name: %.*s in inode "DFID", intent: %s flags %#llo\n",
1389                 (int)op_data->op_namelen, op_data->op_name,
1390                 PFID(&op_data->op_fid1), ldlm_it2str(it->it_op), it->it_flags);
1391
1392         fid_build_reg_res_name(&op_data->op_fid1, &res_id);
1393         /* If the MDT return -ERANGE because of large ACL, then the sponsor
1394          * of the async getattr RPC will handle that by itself. */
1395         req = mdc_intent_getattr_pack(exp, it, op_data,
1396                                       LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1397         if (IS_ERR(req))
1398                 RETURN(PTR_ERR(req));
1399
1400         /* With Data-on-MDT the glimpse callback is needed too.
1401          * It is set here in advance but not in mdc_finish_enqueue()
1402          * to avoid possible races. It is safe to have glimpse handler
1403          * for non-DOM locks and costs nothing.*/
1404         if (minfo->mi_einfo.ei_cb_gl == NULL)
1405                 minfo->mi_einfo.ei_cb_gl = mdc_ldlm_glimpse_ast;
1406
1407         rc = ldlm_cli_enqueue(exp, &req, &minfo->mi_einfo, &res_id, &policy,
1408                               &flags, NULL, 0, LVB_T_NONE, &minfo->mi_lockh, 1);
1409         if (rc < 0) {
1410                 ptlrpc_req_finished(req);
1411                 RETURN(rc);
1412         }
1413
1414         ga = ptlrpc_req_async_args(ga, req);
1415         ga->ga_exp = exp;
1416         ga->ga_minfo = minfo;
1417
1418         req->rq_interpret_reply = mdc_intent_getattr_async_interpret;
1419         ptlrpcd_add_req(req);
1420
1421         RETURN(0);
1422 }