lustre/mdt/mdt_handler.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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2010, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/mdt/mdt_handler.c
  32  *
  33  * Lustre Metadata Target (mdt) request handler
  34  *
  35  * Author: Peter Braam <braam@clusterfs.com>
  36  * Author: Andreas Dilger <adilger@clusterfs.com>
  37  * Author: Phil Schwan <phil@clusterfs.com>
  38  * Author: Mike Shaver <shaver@clusterfs.com>
  39  * Author: Nikita Danilov <nikita@clusterfs.com>
  40  * Author: Huang Hua <huanghua@clusterfs.com>
  41  * Author: Yury Umanets <umka@clusterfs.com>
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_MDS
  45
  46 #include <linux/module.h>
  47 #include <linux/pagemap.h>
  48
  49 #include <dt_object.h>
  50 #include <lustre_acl.h>
  51 #include <lustre_export.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <lustre_lfsck.h>
  54 #include <lustre_log.h>
  55 #include <lustre_nodemap.h>
  56 #include <lustre_mds.h>
  57 #include <uapi/linux/lustre/lustre_param.h>
  58 #include <lustre_quota.h>
  59 #include <lustre_swab.h>
  60 #include <lustre_lmv.h>
  61 #include <obd.h>
  62 #include <obd_support.h>
  63 #include <lustre_barrier.h>
  64 #include <obd_cksum.h>
  65 #include <llog_swab.h>
  66 #include <lustre_crypto.h>
  67
  68 #include "mdt_internal.h"
  69
  70 #if OBD_OCD_VERSION(3, 0, 53, 0) > LUSTRE_VERSION_CODE
  71 static int mdt_max_mod_rpcs_per_client_set(const char *val,
  72                                 cfs_kernel_param_arg_t *kp)
  73 {
  74         unsigned int num;
  75         int rc;
  76
  77         rc = kstrtouint(val, 0, &num);
  78         if (rc < 0)
  79                 return rc;
  80
  81         if (num < 1 || num > OBD_MAX_RIF_MAX)
  82                 return -EINVAL;
  83
  84         CWARN("max_mod_rpcs_per_client is deprecated, set mdt.*.max_mod_rpcs_in_flight parameter instead\n");
  85
  86         max_mod_rpcs_per_client = num;
  87         return 0;
  88 }
  89 static const struct kernel_param_ops
  90                                 param_ops_max_mod_rpcs_per_client = {
  91         .set = mdt_max_mod_rpcs_per_client_set,
  92         .get = param_get_uint,
  93 };
  94
  95 #define param_check_max_mod_rpcs_per_client(name, p) \
  96                 __param_check(name, p, unsigned int)
  97
  98 module_param_cb(max_mod_rpcs_per_client,
  99                                 &param_ops_max_mod_rpcs_per_client,
 100                                 &max_mod_rpcs_per_client, 0644);
 101
 102 MODULE_PARM_DESC(max_mod_rpcs_per_client,
 103         "maximum number of modify RPCs in flight allowed per client (Deprecated)");
 104 #endif
 105
 106 mdl_mode_t mdt_mdl_lock_modes[] = {
 107         [LCK_MINMODE] = MDL_MINMODE,
 108         [LCK_EX]      = MDL_EX,
 109         [LCK_PW]      = MDL_PW,
 110         [LCK_PR]      = MDL_PR,
 111         [LCK_CW]      = MDL_CW,
 112         [LCK_CR]      = MDL_CR,
 113         [LCK_NL]      = MDL_NL,
 114         [LCK_GROUP]   = MDL_GROUP
 115 };
 116
 117 enum ldlm_mode mdt_dlm_lock_modes[] = {
 118         [MDL_MINMODE]   = LCK_MINMODE,
 119         [MDL_EX]        = LCK_EX,
 120         [MDL_PW]        = LCK_PW,
 121         [MDL_PR]        = LCK_PR,
 122         [MDL_CW]        = LCK_CW,
 123         [MDL_CR]        = LCK_CR,
 124         [MDL_NL]        = LCK_NL,
 125         [MDL_GROUP]     = LCK_GROUP
 126 };
 127
 128 static struct mdt_device *mdt_dev(struct lu_device *d);
 129
 130 static const struct lu_object_operations mdt_obj_ops;
 131
 132 /* Slab for MDT object allocation */
 133 static struct kmem_cache *mdt_object_kmem;
 134
 135 /* For HSM restore handles */
 136 struct kmem_cache *mdt_hsm_cdt_kmem;
 137
 138 /* For HSM request handles */
 139 struct kmem_cache *mdt_hsm_car_kmem;
 140
 141 static struct lu_kmem_descr mdt_caches[] = {
 142         {
 143                 .ckd_cache = &mdt_object_kmem,
 144                 .ckd_name  = "mdt_obj",
 145                 .ckd_size  = sizeof(struct mdt_object)
 146         },
 147         {
 148                 .ckd_cache      = &mdt_hsm_cdt_kmem,
 149                 .ckd_name       = "mdt_cdt_restore_handle",
 150                 .ckd_size       = sizeof(struct cdt_restore_handle)
 151         },
 152         {
 153                 .ckd_cache      = &mdt_hsm_car_kmem,
 154                 .ckd_name       = "mdt_cdt_agent_req",
 155                 .ckd_size       = sizeof(struct cdt_agent_req)
 156         },
 157         {
 158                 .ckd_cache = NULL
 159         }
 160 };
 161
 162 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag)
 163 {
 164         if (!rep)
 165                 return 0;
 166         return rep->lock_policy_res1 & op_flag;
 167 }
 168
 169 void mdt_clear_disposition(struct mdt_thread_info *info,
 170                            struct ldlm_reply *rep, __u64 op_flag)
 171 {
 172         if (info) {
 173                 info->mti_opdata &= ~op_flag;
 174                 tgt_opdata_clear(info->mti_env, op_flag);
 175         }
 176         if (rep)
 177                 rep->lock_policy_res1 &= ~op_flag;
 178 }
 179
 180 void mdt_set_disposition(struct mdt_thread_info *info,
 181                          struct ldlm_reply *rep, __u64 op_flag)
 182 {
 183         if (info) {
 184                 info->mti_opdata |= op_flag;
 185                 tgt_opdata_set(info->mti_env, op_flag);
 186         }
 187         if (rep)
 188                 rep->lock_policy_res1 |= op_flag;
 189 }
 190
 191 /* assert lock is unlocked before reuse */
 192 static inline void mdt_lock_handle_assert(struct mdt_lock_handle *lh)
 193 {
 194         LASSERT(!lustre_handle_is_used(&lh->mlh_reg_lh));
 195         LASSERT(!lustre_handle_is_used(&lh->mlh_pdo_lh));
 196         LASSERT(!lustre_handle_is_used(&lh->mlh_rreg_lh));
 197 }
 198
 199 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm)
 200 {
 201         mdt_lock_handle_assert(lh);
 202         lh->mlh_pdo_hash = 0;
 203         lh->mlh_reg_mode = lm;
 204         lh->mlh_rreg_mode = lm;
 205         lh->mlh_type = MDT_REG_LOCK;
 206 }
 207
 208 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock)
 209 {
 210         mdt_lock_reg_init(lh, lock->l_req_mode);
 211         if (lock->l_req_mode == LCK_GROUP)
 212                 lh->mlh_gid = lock->l_policy_data.l_inodebits.li_gid;
 213 }
 214
 215 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
 216                        const struct lu_name *lname)
 217 {
 218         mdt_lock_handle_assert(lh);
 219         lh->mlh_reg_mode = lock_mode;
 220         lh->mlh_pdo_mode = LCK_MINMODE;
 221         lh->mlh_rreg_mode = lock_mode;
 222         lh->mlh_type = MDT_PDO_LOCK;
 223
 224         if (lu_name_is_valid(lname)) {
 225                 lh->mlh_pdo_hash = ll_full_name_hash(NULL, lname->ln_name,
 226                                                      lname->ln_namelen);
 227                 /* XXX Workaround for LU-2856
 228                  *
 229                  * Zero is a valid return value of full_name_hash, but several
 230                  * users of mlh_pdo_hash assume a non-zero hash value. We
 231                  * therefore map zero onto an arbitrary, but consistent
 232                  * value (1) to avoid problems further down the road.
 233                  */
 234                 if (unlikely(lh->mlh_pdo_hash == 0))
 235                         lh->mlh_pdo_hash = 1;
 236         } else {
 237                 lh->mlh_pdo_hash = 0;
 238         }
 239 }
 240
 241 static void mdt_lock_pdo_mode(struct mdt_thread_info *info, struct mdt_object *o,
 242                               struct mdt_lock_handle *lh)
 243 {
 244         mdl_mode_t mode;
 245
 246         ENTRY;
 247
 248         /*
 249          * Any dir access needs couple of locks:
 250          *
 251          * 1) on part of dir we gonna take lookup/modify;
 252          *
 253          * 2) on whole dir to protect it from concurrent splitting and/or to
 254          * flush client's cache for readdir().
 255          *
 256          * so, for a given mode and object this routine decides what lock mode
 257          * to use for lock #2:
 258          *
 259          * 1) if caller's gonna lookup in dir then we need to protect dir from
 260          * being splitted only - LCK_CR
 261          *
 262          * 2) if caller's gonna modify dir then we need to protect dir from
 263          * being splitted and to flush cache - LCK_CW
 264          *
 265          * 3) if caller's gonna modify dir and that dir seems ready for
 266          * splitting then we need to protect it from any type of access
 267          * (lookup/modify/split) - LCK_EX --bzzz
 268          */
 269
 270         LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
 271         LASSERT(lh->mlh_pdo_mode == LCK_MINMODE);
 272
 273         /*
 274          * Ask underlaying level its opinion about preferable PDO lock mode
 275          * having access type passed as regular lock mode:
 276          *
 277          * - MDL_MINMODE means that lower layer does not want to specify lock
 278          * mode;
 279          *
 280          * - MDL_NL means that no PDO lock should be taken. This is used in some
 281          * cases. Say, for non-splittable directories no need to use PDO locks
 282          * at all.
 283          */
 284         mode = mdo_lock_mode(info->mti_env, mdt_object_child(o),
 285                              mdt_dlm_mode2mdl_mode(lh->mlh_reg_mode));
 286
 287         if (mode != MDL_MINMODE) {
 288                 lh->mlh_pdo_mode = mdt_mdl_mode2dlm_mode(mode);
 289         } else {
 290                 /*
 291                  * Lower layer does not want to specify locking mode. We do it
 292                  * our selves. No special protection is needed, just flush
 293                  * client's cache on modification and allow concurrent
 294                  * mondification.
 295                  */
 296                 switch (lh->mlh_reg_mode) {
 297                 case LCK_EX:
 298                         lh->mlh_pdo_mode = LCK_EX;
 299                         break;
 300                 case LCK_PR:
 301                         lh->mlh_pdo_mode = LCK_CR;
 302                         break;
 303                 case LCK_PW:
 304                         lh->mlh_pdo_mode = LCK_CW;
 305                         break;
 306                 default:
 307                         CERROR("Not expected lock type (0x%x)\n",
 308                                (int)lh->mlh_reg_mode);
 309                         LBUG();
 310                 }
 311         }
 312
 313         LASSERT(lh->mlh_pdo_mode != LCK_MINMODE);
 314         EXIT;
 315 }
 316
 317 /**
 318  * Check whether \a o is directory stripe object.
 319  *
 320  * \param[in]  info     thread environment
 321  * \param[in]  o        MDT object
 322  *
 323  * \retval 1    is directory stripe.
 324  * \retval 0    isn't directory stripe.
 325  * \retval < 1  error code
 326  */
 327 static int mdt_is_dir_stripe(struct mdt_thread_info *info,
 328                                 struct mdt_object *o)
 329 {
 330         struct md_attr *ma = &info->mti_attr;
 331         struct lmv_mds_md_v1 *lmv;
 332         int rc;
 333
 334         rc = mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
 335         if (rc < 0)
 336                 return rc;
 337
 338         if (!(ma->ma_valid & MA_LMV))
 339                 return 0;
 340
 341         lmv = &ma->ma_lmv->lmv_md_v1;
 342
 343         if (!lmv_is_sane2(lmv))
 344                 return -EBADF;
 345
 346         if (le32_to_cpu(lmv->lmv_magic) == LMV_MAGIC_STRIPE)
 347                 return 1;
 348
 349         return 0;
 350 }
 351
 352 static int mdt_lookup_fileset(struct mdt_thread_info *info, const char *fileset,
 353                               struct lu_fid *fid)
 354 {
 355         struct mdt_device *mdt = info->mti_mdt;
 356         struct lu_name *lname = &info->mti_name;
 357         const char *start = fileset;
 358         char *filename = info->mti_filename;
 359         struct mdt_object *obj;
 360         int rc = 0;
 361
 362         LASSERT(!info->mti_cross_ref);
 363
 364         /*
 365          * We may want to allow this to mount a completely separate
 366          * fileset from the MDT in the future, but keeping it to
 367          * ROOT/ only for now avoid potential security issues.
 368          */
 369         *fid = mdt->mdt_md_root_fid;
 370
 371         while (rc == 0 && start != NULL && *start != '\0') {
 372                 const char *s1 = start;
 373                 const char *s2;
 374
 375                 while (*++s1 == '/')
 376                         ;
 377                 s2 = s1;
 378                 while (*s2 != '/' && *s2 != '\0')
 379                         s2++;
 380
 381                 if (s2 == s1)
 382                         break;
 383
 384                 start = s2;
 385
 386                 lname->ln_namelen = s2 - s1;
 387                 if (lname->ln_namelen > NAME_MAX) {
 388                         rc = -EINVAL;
 389                         break;
 390                 }
 391
 392                 /* reject .. as a path component */
 393                 if (lname->ln_namelen == 2 && strncmp(s1, "..", 2) == 0) {
 394                         rc = -EINVAL;
 395                         break;
 396                 }
 397
 398                 strncpy(filename, s1, lname->ln_namelen);
 399                 filename[lname->ln_namelen] = '\0';
 400                 lname->ln_name = filename;
 401
 402                 obj = mdt_object_find(info->mti_env, mdt, fid);
 403                 if (IS_ERR(obj)) {
 404                         rc = PTR_ERR(obj);
 405                         break;
 406                 }
 407                 /* Only got the fid of this obj by name */
 408                 fid_zero(fid);
 409                 rc = mdo_lookup(info->mti_env, mdt_object_child(obj), lname,
 410                                 fid, &info->mti_spec);
 411                 if (!rc && !S_ISDIR(lu_object_attr(&obj->mot_obj)))
 412                         rc = -ENOTDIR;
 413                 mdt_object_put(info->mti_env, obj);
 414         }
 415
 416         return rc;
 417 }
 418
 419 static int mdt_get_root(struct tgt_session_info *tsi)
 420 {
 421         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
 422         struct mdt_device       *mdt = info->mti_mdt;
 423         struct mdt_body         *repbody;
 424         char                    *fileset = NULL, *buffer = NULL;
 425         int                      rc;
 426         struct obd_export       *exp = info->mti_exp;
 427         char                    *nodemap_fileset;
 428
 429         ENTRY;
 430
 431         rc = mdt_check_ucred(info);
 432         if (rc)
 433                 GOTO(out, rc = err_serious(rc));
 434
 435         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_GET_ROOT_PACK))
 436                 GOTO(out, rc = err_serious(-ENOMEM));
 437
 438         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 439         if (req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) > 0) {
 440                 fileset = req_capsule_client_get(info->mti_pill, &RMF_NAME);
 441                 if (fileset == NULL)
 442                         GOTO(out, rc = err_serious(-EFAULT));
 443         }
 444
 445         nodemap_fileset = nodemap_get_fileset(exp->exp_target_data.ted_nodemap);
 446         if (nodemap_fileset && nodemap_fileset[0]) {
 447                 CDEBUG(D_INFO, "nodemap fileset is %s\n", nodemap_fileset);
 448                 if (fileset) {
 449                         /* consider fileset from client as a sub-fileset
 450                          * of the nodemap one
 451                          */
 452                         OBD_ALLOC(buffer, PATH_MAX + 1);
 453                         if (buffer == NULL)
 454                                 GOTO(out, rc = err_serious(-ENOMEM));
 455                         if (snprintf(buffer, PATH_MAX + 1, "%s/%s",
 456                                      nodemap_fileset, fileset) >= PATH_MAX + 1)
 457                                 GOTO(out, rc = err_serious(-EINVAL));
 458                         fileset = buffer;
 459                 } else {
 460                         /* enforce fileset as specified in the nodemap */
 461                         fileset = nodemap_fileset;
 462                 }
 463         }
 464
 465         if (fileset) {
 466                 CDEBUG(D_INFO, "Getting fileset %s\n", fileset);
 467                 rc = mdt_lookup_fileset(info, fileset, &repbody->mbo_fid1);
 468                 if (rc < 0)
 469                         GOTO(out, rc = err_serious(rc));
 470         } else {
 471                 repbody->mbo_fid1 = mdt->mdt_md_root_fid;
 472         }
 473         exp->exp_root_fid = repbody->mbo_fid1;
 474         repbody->mbo_valid |= OBD_MD_FLID;
 475
 476         EXIT;
 477 out:
 478         mdt_thread_info_fini(info);
 479         if (buffer)
 480                 OBD_FREE(buffer, PATH_MAX+1);
 481         return rc;
 482 }
 483
 484 static int mdt_statfs(struct tgt_session_info *tsi)
 485 {
 486         struct ptlrpc_request *req = tgt_ses_req(tsi);
 487         struct mdt_thread_info *info = tsi2mdt_info(tsi);
 488         struct mdt_device *mdt = info->mti_mdt;
 489         struct tg_grants_data *tgd = &mdt->mdt_lut.lut_tgd;
 490         struct md_device *next = mdt->mdt_child;
 491         struct ptlrpc_service_part *svcpt;
 492         struct obd_statfs *osfs;
 493         struct mdt_body *reqbody = NULL;
 494         struct mdt_statfs_cache *msf;
 495         ktime_t kstart = ktime_get();
 496         int current_blockbits;
 497         int rc;
 498         timeout_t at_est;
 499
 500         ENTRY;
 501
 502         svcpt = req->rq_rqbd->rqbd_svcpt;
 503
 504         /* This will trigger a watchdog timeout */
 505         at_est = obd_at_get(mdt->mdt_lu_dev.ld_obd, &svcpt->scp_at_estimate);
 506         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
 507                          (MDT_SERVICE_WATCHDOG_FACTOR * at_est) + 1);
 508
 509         rc = mdt_check_ucred(info);
 510         if (rc)
 511                 GOTO(out, rc = err_serious(rc));
 512
 513         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK))
 514                 GOTO(out, rc = err_serious(-ENOMEM));
 515
 516         osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS);
 517         if (!osfs)
 518                 GOTO(out, rc = -EPROTO);
 519
 520         if (mdt_is_sum_statfs_client(req->rq_export) &&
 521                 lustre_packed_msg_size(req->rq_reqmsg) ==
 522                 req_capsule_fmt_size(req->rq_reqmsg->lm_magic,
 523                                      &RQF_MDS_STATFS_NEW, RCL_CLIENT)) {
 524                 req_capsule_extend(info->mti_pill, &RQF_MDS_STATFS_NEW);
 525                 reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
 526         }
 527
 528         if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 529                 msf = &mdt->mdt_sum_osfs;
 530         else
 531                 msf = &mdt->mdt_osfs;
 532
 533         if (msf->msf_age + OBD_STATFS_CACHE_SECONDS <= ktime_get_seconds()) {
 534                 /** statfs data is too old, get up-to-date one */
 535                 if (reqbody && reqbody->mbo_valid & OBD_MD_FLAGSTATFS)
 536                         rc = next->md_ops->mdo_statfs(info->mti_env, next,
 537                                                       osfs);
 538                 else
 539                         rc = dt_statfs(info->mti_env, mdt->mdt_bottom, osfs);
 540                 if (rc)
 541                         GOTO(out, rc);
 542                 spin_lock(&mdt->mdt_lock);
 543                 msf->msf_osfs = *osfs;
 544                 msf->msf_age = ktime_get_seconds();
 545                 spin_unlock(&mdt->mdt_lock);
 546         } else {
 547                 /** use cached statfs data */
 548                 spin_lock(&mdt->mdt_lock);
 549                 *osfs = msf->msf_osfs;
 550                 spin_unlock(&mdt->mdt_lock);
 551         }
 552
 553         /* tgd_blockbit is recordsize bits set during mkfs.
 554          * This once set does not change. However, 'zfs set'
 555          * can be used to change the MDT blocksize. Instead
 556          * of using cached value of 'tgd_blockbit' always
 557          * calculate the blocksize bits which may have
 558          * changed.
 559          */
 560         current_blockbits = fls64(osfs->os_bsize) - 1;
 561
 562         /* Account for cached pages. its still racy and might be under-reporting
 563          * if clients haven't announced their caches with brw recently
 564          */
 565         CDEBUG(D_SUPER | D_CACHE, "blocks cached %llu granted %llu pending %llu free %llu avail %llu\n",
 566                tgd->tgd_tot_dirty, tgd->tgd_tot_granted,
 567                tgd->tgd_tot_pending,
 568                osfs->os_bfree << current_blockbits,
 569                osfs->os_bavail << current_blockbits);
 570
 571         osfs->os_bavail -= min_t(u64, osfs->os_bavail,
 572                                  ((tgd->tgd_tot_dirty + tgd->tgd_tot_pending +
 573                                    osfs->os_bsize - 1) >> current_blockbits));
 574
 575         tgt_grant_sanity_check(mdt->mdt_lu_dev.ld_obd, __func__);
 576         if (mdt->mdt_lut.lut_no_create)
 577                 osfs->os_state |= OS_STATFS_NOCREATE;
 578         CDEBUG(D_CACHE, "%llu blocks: %llu free, %llu avail; "
 579                "%llu objects: %llu free; state %x\n",
 580                osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
 581                osfs->os_files, osfs->os_ffree, osfs->os_state);
 582
 583         if (!exp_grant_param_supp(tsi->tsi_exp) &&
 584             current_blockbits > COMPAT_BSIZE_SHIFT) {
 585                 /* clients which don't support OBD_CONNECT_GRANT_PARAM
 586                  * should not see a block size > page size, otherwise
 587                  * cl_lost_grant goes mad. Therefore, we emulate a 4KB (=2^12)
 588                  * block size which is the biggest block size known to work
 589                  * with all client's page size.
 590                  */
 591                 osfs->os_blocks <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 592                 osfs->os_bfree  <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 593                 osfs->os_bavail <<= current_blockbits - COMPAT_BSIZE_SHIFT;
 594                 osfs->os_bsize = 1 << COMPAT_BSIZE_SHIFT;
 595         }
 596         if (rc == 0)
 597                 mdt_counter_incr(req, LPROC_MDT_STATFS,
 598                                  ktime_us_delta(ktime_get(), kstart));
 599 out:
 600         mdt_thread_info_fini(info);
 601         RETURN(rc);
 602 }
 603
 604 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *is_dom_only)
 605 {
 606         struct lov_comp_md_v1 *comp_v1;
 607         struct lov_mds_md *v1;
 608         __u32 off;
 609         __u32 dom_stripesize = 0;
 610         int i;
 611         bool has_ost_stripes = false;
 612
 613         ENTRY;
 614
 615         if (is_dom_only)
 616                 *is_dom_only = 0;
 617
 618         if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
 619                 RETURN(0);
 620
 621         comp_v1 = (struct lov_comp_md_v1 *)lmm;
 622         off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
 623         v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 624
 625         /* Fast check for DoM entry with no mirroring, should be the first */
 626         if (le16_to_cpu(comp_v1->lcm_mirror_count) == 0 &&
 627             !(lov_pattern(le32_to_cpu(v1->lmm_pattern)) & LOV_PATTERN_MDT))
 628                 RETURN(0);
 629
 630         /* check all entries otherwise */
 631         for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
 632                 struct lov_comp_md_entry_v1 *lcme;
 633
 634                 lcme = &comp_v1->lcm_entries[i];
 635                 if (!(le32_to_cpu(lcme->lcme_flags) & LCME_FL_INIT))
 636                         continue;
 637
 638                 off = le32_to_cpu(lcme->lcme_offset);
 639                 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
 640
 641                 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) &
 642                     LOV_PATTERN_MDT)
 643                         dom_stripesize = le32_to_cpu(v1->lmm_stripe_size);
 644                 else
 645                         has_ost_stripes = true;
 646
 647                 if (dom_stripesize && has_ost_stripes)
 648                         RETURN(dom_stripesize);
 649         }
 650         /* DoM-only case exits here */
 651         if (is_dom_only && dom_stripesize)
 652                 *is_dom_only = 1;
 653         RETURN(dom_stripesize);
 654 }
 655
 656 /* Pack size attributes into the reply. */
 657 int mdt_pack_size2body(struct mdt_thread_info *info,
 658                         const struct lu_fid *fid, struct lustre_handle *lh)
 659 {
 660         struct mdt_body *b;
 661         struct md_attr *ma = &info->mti_attr;
 662         __u32 dom_stripe;
 663         bool dom_lock = false;
 664
 665         ENTRY;
 666
 667         LASSERT(ma->ma_attr.la_valid & LA_MODE);
 668
 669         if (!S_ISREG(ma->ma_attr.la_mode) ||
 670             !(ma->ma_valid & MA_LOV && ma->ma_lmm != NULL))
 671                 RETURN(-ENODATA);
 672
 673         dom_stripe = mdt_lmm_dom_stripesize(ma->ma_lmm);
 674         /* no DoM stripe, no size in reply */
 675         if (!dom_stripe)
 676                 RETURN(-ENOENT);
 677
 678         if (lustre_handle_is_used(lh)) {
 679                 struct ldlm_lock *lock;
 680
 681                 lock = ldlm_handle2lock(lh);
 682                 if (lock != NULL) {
 683                         dom_lock = ldlm_has_dom(lock);
 684                         LDLM_LOCK_PUT(lock);
 685                 }
 686         }
 687
 688         /* no DoM lock, no size in reply */
 689         if (!dom_lock)
 690                 RETURN(0);
 691
 692         /* Either DoM lock exists or LMM has only DoM stripe then
 693          * return size on body.
 694          */
 695         b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 696
 697         mdt_dom_object_size(info->mti_env, info->mti_mdt, fid, b, dom_lock);
 698         RETURN(0);
 699 }
 700
 701 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
 702 /*
 703  * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
 704  *
 705  * \param       info    thread info object
 706  * \param       repbody reply to pack ACLs into
 707  * \param       o       mdt object of file to examine
 708  * \param       nodemap nodemap of client to reply to
 709  * \retval      0       success
 710  * \retval      -errno  error getting or parsing ACL from disk
 711  */
 712 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
 713                       struct mdt_object *o, struct lu_nodemap *nodemap)
 714 {
 715         const struct lu_env     *env = info->mti_env;
 716         struct md_object        *next = mdt_object_child(o);
 717         struct lu_buf           *buf = &info->mti_buf;
 718         struct mdt_device       *mdt = info->mti_mdt;
 719         struct req_capsule *pill = info->mti_pill;
 720         int rc;
 721
 722         ENTRY;
 723
 724         buf->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
 725         buf->lb_len = req_capsule_get_size(pill, &RMF_ACL, RCL_SERVER);
 726         if (buf->lb_len == 0)
 727                 RETURN(0);
 728
 729         LASSERT(!info->mti_big_acl_used);
 730 again:
 731         rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
 732         if (rc < 0) {
 733                 if (rc == -ENODATA) {
 734                         repbody->mbo_aclsize = 0;
 735                         repbody->mbo_valid |= OBD_MD_FLACL;
 736                         rc = 0;
 737                 } else if (rc == -EOPNOTSUPP) {
 738                         rc = 0;
 739                 } else if (rc == -ERANGE) {
 740                         if (exp_connect_large_acl(info->mti_exp) &&
 741                             !info->mti_big_acl_used) {
 742                                 if (info->mti_big_acl == NULL) {
 743                                         info->mti_big_aclsize =
 744                                                         min_t(unsigned int,
 745                                                               mdt->mdt_max_ea_size,
 746                                                               XATTR_SIZE_MAX);
 747                                         OBD_ALLOC_LARGE(info->mti_big_acl,
 748                                                         info->mti_big_aclsize);
 749                                         if (info->mti_big_acl == NULL) {
 750                                                 info->mti_big_aclsize = 0;
 751                                                 CERROR("%s: unable to grow "
 752                                                        DFID" ACL buffer\n",
 753                                                        mdt_obd_name(mdt),
 754                                                        PFID(mdt_object_fid(o)));
 755                                                 RETURN(-ENOMEM);
 756                                         }
 757                                 }
 758
 759                                 CDEBUG(D_INODE, "%s: grow the "DFID
 760                                        " ACL buffer to size %d\n",
 761                                        mdt_obd_name(mdt),
 762                                        PFID(mdt_object_fid(o)),
 763                                        info->mti_big_aclsize);
 764
 765                                 buf->lb_buf = info->mti_big_acl;
 766                                 buf->lb_len = info->mti_big_aclsize;
 767                                 info->mti_big_acl_used = 1;
 768                                 goto again;
 769                         }
 770                         /* FS has ACL bigger that our limits */
 771                         CDEBUG(D_INODE, "%s: "DFID" ACL can't fit into %d\n",
 772                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)),
 773                                info->mti_big_aclsize);
 774                         rc = -E2BIG;
 775                 } else {
 776                         CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
 777                                mdt_obd_name(mdt), PFID(mdt_object_fid(o)), rc);
 778                 }
 779         } else {
 780                 rc = nodemap_map_acl(nodemap, buf->lb_buf,
 781                                      rc, NODEMAP_FS_TO_CLIENT);
 782                 /* if all ACLs mapped out, rc is still >= 0 */
 783                 if (rc < 0) {
 784                         CERROR("%s: nodemap_map_acl unable to parse "DFID
 785                                " ACL: rc = %d\n", mdt_obd_name(mdt),
 786                                PFID(mdt_object_fid(o)), rc);
 787                         repbody->mbo_aclsize = 0;
 788                         repbody->mbo_valid &= ~OBD_MD_FLACL;
 789                 } else {
 790                         repbody->mbo_aclsize = rc;
 791                         repbody->mbo_valid |= OBD_MD_FLACL;
 792                         rc = 0;
 793                 }
 794         }
 795
 796         RETURN(rc);
 797 }
 798 #endif
 799
 800 /* XXX Look into layout in MDT layer. */
 801 static inline bool mdt_hsm_is_released(struct lov_mds_md *lmm)
 802 {
 803         struct lov_comp_md_v1   *comp_v1;
 804         struct lov_mds_md       *v1;
 805         int                      i;
 806
 807         if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
 808                 comp_v1 = (struct lov_comp_md_v1 *)lmm;
 809
 810                 for (i = 0; i < comp_v1->lcm_entry_count; i++) {
 811                         v1 = (struct lov_mds_md *)((char *)comp_v1 +
 812                                 comp_v1->lcm_entries[i].lcme_offset);
 813                         /* We don't support partial release for now */
 814                         if (!(v1->lmm_pattern & LOV_PATTERN_F_RELEASED))
 815                                 return false;
 816                 }
 817                 return true;
 818         } else {
 819                 return (lmm->lmm_pattern & LOV_PATTERN_F_RELEASED) ?
 820                         true : false;
 821         }
 822 }
 823
 824 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
 825                         const struct lu_attr *attr, const struct lu_fid *fid)
 826 {
 827         struct mdt_device *mdt = info->mti_mdt;
 828         struct obd_export *exp = info->mti_exp;
 829         struct md_attr *ma = &info->mti_attr;
 830         struct lu_nodemap *nodemap = NULL;
 831
 832         LASSERT(ma->ma_valid & MA_INODE);
 833
 834         if (attr->la_valid & LA_ATIME) {
 835                 b->mbo_atime = attr->la_atime;
 836                 b->mbo_valid |= OBD_MD_FLATIME;
 837         }
 838         if (attr->la_valid & LA_MTIME) {
 839                 b->mbo_mtime = attr->la_mtime;
 840                 b->mbo_valid |= OBD_MD_FLMTIME;
 841         }
 842         if (attr->la_valid & LA_CTIME) {
 843                 b->mbo_ctime = attr->la_ctime;
 844                 b->mbo_valid |= OBD_MD_FLCTIME;
 845         }
 846         if (attr->la_valid & LA_BTIME) {
 847                 b->mbo_btime = attr->la_btime;
 848                 b->mbo_valid |= OBD_MD_FLBTIME;
 849         }
 850         if (attr->la_valid & LA_FLAGS) {
 851                 b->mbo_flags = attr->la_flags;
 852                 b->mbo_valid |= OBD_MD_FLFLAGS;
 853         }
 854         if (attr->la_valid & LA_NLINK) {
 855                 b->mbo_nlink = attr->la_nlink;
 856                 b->mbo_valid |= OBD_MD_FLNLINK;
 857         }
 858         if (attr->la_valid & (LA_UID|LA_GID|LA_PROJID)) {
 859                 nodemap = nodemap_get_from_exp(exp);
 860                 if (IS_ERR(nodemap))
 861                         goto out;
 862         }
 863         if (attr->la_valid & LA_UID) {
 864                 b->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 865                                             NODEMAP_FS_TO_CLIENT,
 866                                             attr->la_uid);
 867                 b->mbo_valid |= OBD_MD_FLUID;
 868         }
 869         if (attr->la_valid & LA_GID) {
 870                 b->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 871                                             NODEMAP_FS_TO_CLIENT,
 872                                             attr->la_gid);
 873                 b->mbo_valid |= OBD_MD_FLGID;
 874         }
 875
 876         if (attr->la_valid & LA_PROJID) {
 877                 b->mbo_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
 878                                                NODEMAP_FS_TO_CLIENT,
 879                                                attr->la_projid);
 880                 b->mbo_valid |= OBD_MD_FLPROJID;
 881         }
 882
 883         b->mbo_mode = attr->la_mode;
 884         if (attr->la_valid & LA_MODE)
 885                 b->mbo_valid |= OBD_MD_FLMODE;
 886         if (attr->la_valid & LA_TYPE)
 887                 b->mbo_valid |= OBD_MD_FLTYPE;
 888
 889         if (fid != NULL) {
 890                 b->mbo_fid1 = *fid;
 891                 b->mbo_valid |= OBD_MD_FLID;
 892                 CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, valid=%#llx\n",
 893                        PFID(fid), b->mbo_nlink, b->mbo_mode, b->mbo_valid);
 894         }
 895
 896         if (!(attr->la_valid & LA_TYPE))
 897                 return;
 898
 899         b->mbo_rdev   = attr->la_rdev;
 900         b->mbo_size   = attr->la_size;
 901         b->mbo_blocks = attr->la_blocks;
 902
 903         if (!S_ISREG(attr->la_mode)) {
 904                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV;
 905         } else if (ma->ma_need & MA_LOV && !(ma->ma_valid & MA_LOV)) {
 906                 /* means no objects are allocated on osts. */
 907                 LASSERT(!(ma->ma_valid & MA_LOV));
 908                 /* just ignore blocks occupied by extend attributes on MDS */
 909                 b->mbo_blocks = 0;
 910                 /* if no object is allocated on osts, the size on mds is valid.
 911                  * b=22272
 912                  */
 913                 b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 914         } else if ((ma->ma_valid & MA_LOV) && ma->ma_lmm != NULL) {
 915                 if (mdt_hsm_is_released(ma->ma_lmm)) {
 916                         /* A released file stores its size on MDS. */
 917                         /* But return 1 block for released file, unless tools
 918                          * like tar will consider it fully sparse. (LU-3864)
 919                          */
 920                         if (unlikely(b->mbo_size == 0))
 921                                 b->mbo_blocks = 0;
 922                         else
 923                                 b->mbo_blocks = 1;
 924                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 925                 } else if (info->mti_som_strict && mdt->mdt_enable_strict_som) {
 926                         /* use SOM for size*/
 927                         b->mbo_valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 928                 } else if (ma->ma_valid & MA_SOM) { /* lsom is valid */
 929                         b->mbo_valid |= OBD_MD_FLLAZYSIZE | OBD_MD_FLLAZYBLOCKS;
 930                         b->mbo_size = ma->ma_som.ms_size;
 931                         b->mbo_blocks = ma->ma_som.ms_blocks;
 932                 }
 933         }
 934
 935         if (fid != NULL && (b->mbo_valid & OBD_MD_FLSIZE ||
 936                             b->mbo_valid & OBD_MD_FLLAZYSIZE))
 937                 CDEBUG(D_VFSTRACE, DFID": returning size %llu\n",
 938                        PFID(fid), (unsigned long long)b->mbo_size);
 939
 940 out:
 941         if (!IS_ERR_OR_NULL(nodemap))
 942                 nodemap_putref(nodemap);
 943 }
 944
 945 static inline int mdt_body_has_lov(const struct lu_attr *la,
 946                                    const struct mdt_body *body)
 947 {
 948         return (S_ISREG(la->la_mode) && (body->mbo_valid & OBD_MD_FLEASIZE)) ||
 949                (S_ISDIR(la->la_mode) && (body->mbo_valid & OBD_MD_FLDIREA));
 950 }
 951
 952 void mdt_client_compatibility(struct mdt_thread_info *info)
 953 {
 954         struct mdt_body       *body;
 955         struct ptlrpc_request *req = mdt_info_req(info);
 956         struct obd_export     *exp = req->rq_export;
 957         struct md_attr        *ma = &info->mti_attr;
 958         struct lu_attr        *la = &ma->ma_attr;
 959
 960         ENTRY;
 961
 962         if (exp_connect_layout(exp))
 963                 /* the client can deal with 16-bit lmm_stripe_count */
 964                 RETURN_EXIT;
 965
 966         body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 967
 968         if (!mdt_body_has_lov(la, body))
 969                 RETURN_EXIT;
 970
 971         /* now we have a reply with a lov for a client not compatible with the
 972          * layout lock so we have to clean the layout generation number
 973          */
 974         if (S_ISREG(la->la_mode))
 975                 ma->ma_lmm->lmm_layout_gen = 0;
 976         EXIT;
 977 }
 978
 979 static int mdt_attr_get_eabuf_size(struct mdt_thread_info *info,
 980                                    struct mdt_object *o)
 981 {
 982         const struct lu_env *env = info->mti_env;
 983         int rc, rc2;
 984
 985         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 986                           XATTR_NAME_LOV);
 987
 988         if (rc == -ENODATA)
 989                 rc = 0;
 990
 991         if (rc < 0)
 992                 goto out;
 993
 994         /* Is it a directory? Let's check for the LMV as well */
 995         if (S_ISDIR(lu_object_attr(&mdt_object_child(o)->mo_lu))) {
 996                 rc2 = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL,
 997                                    XATTR_NAME_LMV);
 998
 999                 if (rc2 == -ENODATA)
1000                         rc2 = mo_xattr_get(env, mdt_object_child(o),
1001                                            &LU_BUF_NULL,
1002                                            XATTR_NAME_DEFAULT_LMV);
1003
1004                 if ((rc2 < 0 && rc2 != -ENODATA) || (rc2 > rc))
1005                         rc = rc2;
1006         }
1007
1008 out:
1009         return rc;
1010 }
1011
1012 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
1013                       const char *name)
1014 {
1015         const struct lu_env *env = info->mti_env;
1016         int rc;
1017
1018         ENTRY;
1019
1020         LASSERT(info->mti_big_lmm_used == 0);
1021         rc = mo_xattr_get(env, mdt_object_child(o), &LU_BUF_NULL, name);
1022         if (rc < 0)
1023                 RETURN(rc);
1024
1025         /* big_lmm may need to be grown */
1026         if (info->mti_big_lmmsize < rc) {
1027                 int size = size_roundup_power2(rc);
1028
1029                 if (info->mti_big_lmmsize > 0) {
1030                         /* free old buffer */
1031                         LASSERT(info->mti_big_lmm);
1032                         OBD_FREE_LARGE(info->mti_big_lmm,
1033                                        info->mti_big_lmmsize);
1034                         info->mti_big_lmm = NULL;
1035                         info->mti_big_lmmsize = 0;
1036                 }
1037
1038                 OBD_ALLOC_LARGE(info->mti_big_lmm, size);
1039                 if (info->mti_big_lmm == NULL)
1040                         RETURN(-ENOMEM);
1041                 info->mti_big_lmmsize = size;
1042         }
1043         LASSERT(info->mti_big_lmmsize >= rc);
1044
1045         info->mti_buf.lb_buf = info->mti_big_lmm;
1046         info->mti_buf.lb_len = info->mti_big_lmmsize;
1047         rc = mo_xattr_get(env, mdt_object_child(o), &info->mti_buf, name);
1048
1049         RETURN(rc);
1050 }
1051
1052 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1053                      struct md_attr *ma, const char *name)
1054 {
1055         struct md_object *next = mdt_object_child(o);
1056         struct lu_buf    *buf = &info->mti_buf;
1057         int rc;
1058
1059         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1060                 buf->lb_buf = ma->ma_lmm;
1061                 buf->lb_len = ma->ma_lmm_size;
1062                 LASSERT(!(ma->ma_valid & MA_LOV));
1063         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1064                 buf->lb_buf = ma->ma_lmv;
1065                 buf->lb_len = ma->ma_lmv_size;
1066                 LASSERT(!(ma->ma_valid & MA_LMV));
1067         } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1068                 buf->lb_buf = ma->ma_default_lmv;
1069                 buf->lb_len = ma->ma_default_lmv_size;
1070                 LASSERT(!(ma->ma_valid & MA_LMV_DEF));
1071         } else {
1072                 return -EINVAL;
1073         }
1074
1075         LASSERT(buf->lb_buf);
1076
1077         if (!mdt_object_exists(o))
1078                 return -ENOENT;
1079
1080         if (mdt_object_remote(o) && S_ISDIR(lu_object_attr(&o->mot_obj)))
1081                 /* force reload layout for remote dir in case layout changed */
1082                 mo_invalidate(info->mti_env, mdt_object_child(o));
1083
1084         rc = mo_xattr_get(info->mti_env, next, buf, name);
1085         if (rc > 0) {
1086
1087 got:
1088                 if (strcmp(name, XATTR_NAME_LOV) == 0) {
1089                         if (info->mti_big_lmm_used)
1090                                 ma->ma_lmm = info->mti_big_lmm;
1091
1092                         /* NOT return LOV EA with hole to old client. */
1093                         if (unlikely(le32_to_cpu(ma->ma_lmm->lmm_pattern) &
1094                                      LOV_PATTERN_F_HOLE) &&
1095                             !(exp_connect_flags(info->mti_exp) &
1096                               OBD_CONNECT_LFSCK)) {
1097                                 return -EIO;
1098                         }
1099                         ma->ma_lmm_size = rc;
1100                         ma->ma_valid |= MA_LOV;
1101                 } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1102                         if (info->mti_big_lmm_used)
1103                                 ma->ma_lmv = info->mti_big_lmm;
1104
1105                         ma->ma_lmv_size = rc;
1106                         ma->ma_valid |= MA_LMV;
1107                 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1108                         ma->ma_default_lmv_size = rc;
1109                         ma->ma_valid |= MA_LMV_DEF;
1110                 }
1111
1112                 /* Update mdt_max_mdsize so all clients will be aware that */
1113                 if (info->mti_mdt->mdt_max_mdsize < rc)
1114                         info->mti_mdt->mdt_max_mdsize = rc;
1115
1116                 rc = 0;
1117         } else if (rc == -ENODATA) {
1118                 /* no LOV EA */
1119                 rc = 0;
1120         } else if (rc == -ERANGE) {
1121                 /* Default LMV has fixed size, so it must be able to fit
1122                  * in the original buffer
1123                  */
1124                 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
1125                         return rc;
1126                 rc = mdt_big_xattr_get(info, o, name);
1127                 if (rc > 0) {
1128                         info->mti_big_lmm_used = 1;
1129                         goto got;
1130                 }
1131         }
1132
1133         return rc;
1134 }
1135
1136 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
1137                    struct md_attr *ma, const char *name)
1138 {
1139         int rc;
1140
1141         if (!info->mti_big_lmm) {
1142                 OBD_ALLOC(info->mti_big_lmm, PAGE_SIZE);
1143                 if (!info->mti_big_lmm)
1144                         return -ENOMEM;
1145                 info->mti_big_lmmsize = PAGE_SIZE;
1146         }
1147
1148         if (strcmp(name, XATTR_NAME_LOV) == 0) {
1149                 ma->ma_lmm = info->mti_big_lmm;
1150                 ma->ma_lmm_size = info->mti_big_lmmsize;
1151                 ma->ma_valid &= ~MA_LOV;
1152         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
1153                 ma->ma_lmv = info->mti_big_lmm;
1154                 ma->ma_lmv_size = info->mti_big_lmmsize;
1155                 ma->ma_valid &= ~MA_LMV;
1156         } else {
1157                 LBUG();
1158         }
1159
1160         LASSERT(!info->mti_big_lmm_used);
1161         rc = __mdt_stripe_get(info, o, ma, name);
1162         /* since big_lmm is always used here, clear 'used' flag to avoid
1163          * assertion in mdt_big_xattr_get().
1164          */
1165         info->mti_big_lmm_used = 0;
1166
1167         return rc;
1168 }
1169
1170 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
1171                       struct lu_fid *pfid)
1172 {
1173         struct lu_buf           *buf = &info->mti_buf;
1174         struct link_ea_header   *leh;
1175         struct link_ea_entry    *lee;
1176         int                      rc;
1177
1178         ENTRY;
1179
1180         buf->lb_buf = info->mti_big_lmm;
1181         buf->lb_len = info->mti_big_lmmsize;
1182         rc = mo_xattr_get(info->mti_env, mdt_object_child(o),
1183                           buf, XATTR_NAME_LINK);
1184         /* ignore errors, MA_PFID won't be set and it is
1185          * up to the caller to treat this as an error
1186          */
1187         if (rc == -ERANGE || buf->lb_len == 0) {
1188                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1189                 buf->lb_buf = info->mti_big_lmm;
1190                 buf->lb_len = info->mti_big_lmmsize;
1191         }
1192
1193         if (rc < 0)
1194                 RETURN(rc);
1195         if (rc < sizeof(*leh)) {
1196                 CERROR("short LinkEA on "DFID": rc = %d\n",
1197                        PFID(mdt_object_fid(o)), rc);
1198                 RETURN(-ENODATA);
1199         }
1200
1201         leh = (struct link_ea_header *) buf->lb_buf;
1202         lee = (struct link_ea_entry *)(leh + 1);
1203         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1204                 leh->leh_magic = LINK_EA_MAGIC;
1205                 leh->leh_reccount = __swab32(leh->leh_reccount);
1206                 leh->leh_len = __swab64(leh->leh_len);
1207         }
1208         if (leh->leh_magic != LINK_EA_MAGIC)
1209                 RETURN(-EINVAL);
1210         if (leh->leh_reccount == 0)
1211                 RETURN(-ENODATA);
1212
1213         memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
1214         fid_be_to_cpu(pfid, pfid);
1215
1216         RETURN(0);
1217 }
1218
1219 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
1220                            struct lu_fid *pfid, struct lu_name *lname)
1221 {
1222         struct lu_buf *buf = &info->mti_buf;
1223         struct link_ea_header *leh;
1224         struct link_ea_entry *lee;
1225         int reclen;
1226         int rc;
1227
1228         buf->lb_buf = info->mti_xattr_buf;
1229         buf->lb_len = sizeof(info->mti_xattr_buf);
1230         rc = mo_xattr_get(info->mti_env, mdt_object_child(o), buf,
1231                           XATTR_NAME_LINK);
1232         if (rc == -ERANGE) {
1233                 rc = mdt_big_xattr_get(info, o, XATTR_NAME_LINK);
1234                 buf->lb_buf = info->mti_big_lmm;
1235                 buf->lb_len = info->mti_big_lmmsize;
1236         }
1237         if (rc < 0)
1238                 return rc;
1239
1240         if (rc < sizeof(*leh)) {
1241                 CERROR("short LinkEA on "DFID": rc = %d\n",
1242                        PFID(mdt_object_fid(o)), rc);
1243                 return -ENODATA;
1244         }
1245
1246         leh = (struct link_ea_header *)buf->lb_buf;
1247         lee = (struct link_ea_entry *)(leh + 1);
1248         if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) {
1249                 leh->leh_magic = LINK_EA_MAGIC;
1250                 leh->leh_reccount = __swab32(leh->leh_reccount);
1251                 leh->leh_len = __swab64(leh->leh_len);
1252         }
1253         if (leh->leh_magic != LINK_EA_MAGIC)
1254                 return -EINVAL;
1255
1256         if (leh->leh_reccount == 0)
1257                 return -ENODATA;
1258
1259         linkea_entry_unpack(lee, &reclen, lname, pfid);
1260
1261         return 0;
1262 }
1263
1264 int mdt_attr_get_complex(struct mdt_thread_info *info,
1265                          struct mdt_object *o, struct md_attr *ma)
1266 {
1267         const struct lu_env *env = info->mti_env;
1268         struct md_object    *next = mdt_object_child(o);
1269         struct lu_buf       *buf = &info->mti_buf;
1270         int                  need = ma->ma_need;
1271         int                  rc = 0, rc2;
1272         u32                  mode;
1273
1274         ENTRY;
1275
1276         ma->ma_valid = 0;
1277
1278         if (mdt_object_exists(o) == 0)
1279                 GOTO(out, rc = -ENOENT);
1280         mode = lu_object_attr(&next->mo_lu);
1281
1282         if (need & MA_INODE) {
1283                 ma->ma_need = MA_INODE;
1284                 if (need & MA_DIRENT_CNT)
1285                         ma->ma_attr.la_valid |= LA_DIRENT_CNT;
1286                 else
1287                         ma->ma_attr.la_valid &= ~LA_DIRENT_CNT;
1288                 rc = mo_attr_get(env, next, ma);
1289                 if (rc)
1290                         GOTO(out, rc);
1291
1292                 if (S_ISREG(mode))
1293                         (void) mdt_get_som(info, o, ma);
1294                 ma->ma_valid |= MA_INODE;
1295         }
1296
1297         if (need & MA_PFID) {
1298                 rc = mdt_attr_get_pfid(info, o, &ma->ma_pfid);
1299                 if (rc == 0)
1300                         ma->ma_valid |= MA_PFID;
1301                 /* ignore this error, parent fid is not mandatory */
1302                 rc = 0;
1303         }
1304
1305         if (need & MA_LOV && (S_ISREG(mode) || S_ISDIR(mode))) {
1306                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LOV);
1307                 if (rc)
1308                         GOTO(out, rc);
1309         }
1310
1311         if (need & MA_LMV && S_ISDIR(mode)) {
1312                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_LMV);
1313                 if (rc != 0)
1314                         GOTO(out, rc);
1315         }
1316
1317         if (need & MA_LMV_DEF && S_ISDIR(mode)) {
1318                 rc = __mdt_stripe_get(info, o, ma, XATTR_NAME_DEFAULT_LMV);
1319                 if (rc != 0)
1320                         GOTO(out, rc);
1321         }
1322
1323         /* In the handle of MA_INODE, we may already get the SOM attr.  */
1324         if (need & MA_SOM && S_ISREG(mode) && !(ma->ma_valid & MA_SOM)) {
1325                 rc = mdt_get_som(info, o, ma);
1326                 if (rc != 0)
1327                         GOTO(out, rc);
1328         }
1329
1330         if (need & MA_HSM && S_ISREG(mode)) {
1331                 buf->lb_buf = info->mti_xattr_buf;
1332                 buf->lb_len = sizeof(info->mti_xattr_buf);
1333                 BUILD_BUG_ON(sizeof(struct hsm_attrs) >
1334                              sizeof(info->mti_xattr_buf));
1335                 rc2 = mo_xattr_get(info->mti_env, next, buf, XATTR_NAME_HSM);
1336                 rc2 = lustre_buf2hsm(info->mti_xattr_buf, rc2, &ma->ma_hsm);
1337                 if (rc2 == 0)
1338                         ma->ma_valid |= MA_HSM;
1339                 else if (rc2 < 0 && rc2 != -ENODATA)
1340                         GOTO(out, rc = rc2);
1341         }
1342
1343 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1344         if (need & MA_ACL_DEF && S_ISDIR(mode)) {
1345                 buf->lb_buf = ma->ma_acl;
1346                 buf->lb_len = ma->ma_acl_size;
1347                 rc2 = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_DEFAULT);
1348                 if (rc2 > 0) {
1349                         ma->ma_acl_size = rc2;
1350                         ma->ma_valid |= MA_ACL_DEF;
1351                 } else if (rc2 == -ENODATA) {
1352                         /* no ACLs */
1353                         ma->ma_acl_size = 0;
1354                 } else
1355                         GOTO(out, rc = rc2);
1356         }
1357 #endif
1358 out:
1359         ma->ma_need = need;
1360         CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = %#llx ma_lmm=%p\n",
1361                rc, ma->ma_valid, ma->ma_lmm);
1362         RETURN(rc);
1363 }
1364
1365 static void mdt_preset_encctx_size(struct mdt_thread_info *info)
1366 {
1367         struct req_capsule *pill = info->mti_pill;
1368
1369         ENTRY;
1370         if (req_capsule_has_field(pill, &RMF_FILE_ENCCTX,
1371                                   RCL_SERVER))
1372                 /* pre-set size in server part with max size */
1373                 req_capsule_set_size(pill, &RMF_FILE_ENCCTX,
1374                                      RCL_SERVER,
1375                                      info->mti_mdt->mdt_max_mdsize);
1376         EXIT;
1377 }
1378
1379 static int mdt_getattr_internal(struct mdt_thread_info *info,
1380                                 struct mdt_object *o, int ma_need)
1381 {
1382         struct mdt_device *mdt = info->mti_mdt;
1383         struct md_object *next = mdt_object_child(o);
1384         const struct mdt_body *reqbody = info->mti_body;
1385         struct ptlrpc_request *req = mdt_info_req(info);
1386         struct md_attr *ma = &info->mti_attr;
1387         struct lu_attr *la = &ma->ma_attr;
1388         struct req_capsule *pill = info->mti_pill;
1389         const struct lu_env *env = info->mti_env;
1390         struct mdt_body *repbody;
1391         struct lu_buf *buffer = &info->mti_buf;
1392         struct obd_export *exp = info->mti_exp;
1393         ktime_t kstart = ktime_get();
1394         int rc;
1395
1396         ENTRY;
1397
1398         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
1399                 RETURN(err_serious(-ENOMEM));
1400
1401         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1402
1403         ma->ma_valid = 0;
1404
1405         if (mdt_object_remote(o)) {
1406                 /* obj is located on remote node Return -ENOTSUPP(old client) */
1407                 if (!mdt_is_dne_client(req->rq_export))
1408                         GOTO(out, rc = -ENOTSUPP);
1409
1410                 repbody->mbo_fid1 = *mdt_object_fid(o);
1411                 repbody->mbo_valid = OBD_MD_FLID | OBD_MD_MDS;
1412                 GOTO(out, rc = 0);
1413         }
1414
1415         if (reqbody->mbo_eadatasize > 0) {
1416                 buffer->lb_buf = req_capsule_server_get(pill, &RMF_MDT_MD);
1417                 if (buffer->lb_buf == NULL)
1418                         GOTO(out, rc = -EPROTO);
1419                 buffer->lb_len = req_capsule_get_size(pill, &RMF_MDT_MD,
1420                                                       RCL_SERVER);
1421         } else {
1422                 buffer->lb_buf = NULL;
1423                 buffer->lb_len = 0;
1424                 ma_need &= ~(MA_LOV | MA_LMV);
1425                 CDEBUG(D_INFO, "%s: RPC from %s: does not need LOVEA.\n",
1426                        mdt_obd_name(info->mti_mdt),
1427                        req->rq_export->exp_client_uuid.uuid);
1428         }
1429
1430         /* from 2.12.58 intent_getattr pack default LMV in reply */
1431         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1432             ((reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) ==
1433                     (OBD_MD_MEA | OBD_MD_DEFAULT_MEA)) &&
1434             req_capsule_has_field(&req->rq_pill, &RMF_DEFAULT_MDT_MD,
1435                                   RCL_SERVER)) {
1436                 ma->ma_lmv = buffer->lb_buf;
1437                 ma->ma_lmv_size = buffer->lb_len;
1438                 ma->ma_default_lmv = req_capsule_server_get(pill,
1439                                                 &RMF_DEFAULT_MDT_MD);
1440                 ma->ma_default_lmv_size = req_capsule_get_size(pill,
1441                                                 &RMF_DEFAULT_MDT_MD,
1442                                                 RCL_SERVER);
1443                 ma->ma_need = MA_INODE;
1444                 if (ma->ma_lmv_size > 0)
1445                         ma->ma_need |= MA_LMV;
1446                 if (ma->ma_default_lmv_size > 0)
1447                         ma->ma_need |= MA_LMV_DEF;
1448         } else if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1449                    (reqbody->mbo_valid & (OBD_MD_MEA | OBD_MD_DEFAULT_MEA))) {
1450                 /* If it is dir and client require MEA, then we got MEA */
1451                 /* Assumption: MDT_MD size is enough for lmv size. */
1452                 ma->ma_lmv = buffer->lb_buf;
1453                 ma->ma_lmv_size = buffer->lb_len;
1454                 ma->ma_need = MA_INODE;
1455                 if (ma->ma_lmv_size > 0) {
1456                         if (reqbody->mbo_valid & OBD_MD_MEA) {
1457                                 ma->ma_need |= MA_LMV;
1458                         } else if (reqbody->mbo_valid & OBD_MD_DEFAULT_MEA) {
1459                                 ma->ma_need |= MA_LMV_DEF;
1460                                 ma->ma_default_lmv = buffer->lb_buf;
1461                                 ma->ma_lmv = NULL;
1462                                 ma->ma_default_lmv_size = buffer->lb_len;
1463                                 ma->ma_lmv_size = 0;
1464                         }
1465                 }
1466         } else {
1467                 ma->ma_lmm = buffer->lb_buf;
1468                 ma->ma_lmm_size = buffer->lb_len;
1469                 ma->ma_need = MA_INODE | MA_HSM;
1470                 if (ma->ma_lmm_size > 0) {
1471                         ma->ma_need |= MA_LOV;
1472                         /* Older clients may crash if they getattr overstriped
1473                          * files
1474                          */
1475                         if (!exp_connect_overstriping(exp) &&
1476                             mdt_lmm_is_overstriping(ma->ma_lmm))
1477                                 RETURN(-EOPNOTSUPP);
1478                 }
1479         }
1480
1481         if (S_ISDIR(lu_object_attr(&next->mo_lu)) &&
1482             reqbody->mbo_valid & OBD_MD_FLDIREA  &&
1483             lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) {
1484                 /* get default stripe info for this dir. */
1485                 ma->ma_need |= MA_LOV_DEF;
1486         }
1487         ma->ma_need |= ma_need;
1488
1489         rc = mdt_attr_get_complex(info, o, ma);
1490         if (unlikely(rc)) {
1491                 CDEBUG_LIMIT(rc == -ENOENT ? D_OTHER : D_ERROR,
1492                              "%s: getattr error for "DFID": rc = %d\n",
1493                              mdt_obd_name(info->mti_mdt),
1494                              PFID(mdt_object_fid(o)), rc);
1495                 RETURN(rc);
1496         }
1497
1498         /* return immutable attr on fscrypt metadata files
1499          * if fscrypt admin is not permitted
1500          */
1501         if (o->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD &&
1502             !mdt_ucred(info)->uc_rbac_fscrypt_admin)
1503                 la->la_flags |= LUSTRE_IMMUTABLE_FL;
1504
1505         /* if file is released, check if a restore is running */
1506         if (ma->ma_valid & MA_HSM) {
1507                 repbody->mbo_valid |= OBD_MD_TSTATE;
1508                 if ((ma->ma_hsm.mh_flags & HS_RELEASED) &&
1509                     mdt_hsm_restore_is_running(info, mdt_object_fid(o)))
1510                         repbody->mbo_t_state = MS_RESTORE;
1511         }
1512
1513         if (unlikely(!(ma->ma_valid & MA_INODE)))
1514                 RETURN(-EFAULT);
1515
1516         mdt_pack_attr2body(info, repbody, la, mdt_object_fid(o));
1517
1518         if (mdt_body_has_lov(la, reqbody)) {
1519                 u32 stripe_count = 1;
1520                 bool fixed_layout = false;
1521
1522                 if (ma->ma_valid & MA_LOV) {
1523                         LASSERT(ma->ma_lmm_size);
1524                         repbody->mbo_eadatasize = ma->ma_lmm_size;
1525                         if (S_ISDIR(la->la_mode))
1526                                 repbody->mbo_valid |= OBD_MD_FLDIREA;
1527                         else
1528                                 repbody->mbo_valid |= OBD_MD_FLEASIZE;
1529                         mdt_dump_lmm(D_INFO, ma->ma_lmm, repbody->mbo_valid);
1530                 }
1531                 if (ma->ma_valid & MA_LMV) {
1532                         struct lmv_mds_md_v1 *lmv = &ma->ma_lmv->lmv_md_v1;
1533                         u32 magic = le32_to_cpu(lmv->lmv_magic);
1534
1535                         /* Return -ENOTSUPP for old client */
1536                         if (!mdt_is_striped_client(req->rq_export))
1537                                 RETURN(-ENOTSUPP);
1538
1539                         LASSERT(S_ISDIR(la->la_mode));
1540                         mdt_dump_lmv(D_INFO, ma->ma_lmv);
1541                         repbody->mbo_eadatasize = ma->ma_lmv_size;
1542                         repbody->mbo_valid |= (OBD_MD_FLDIREA|OBD_MD_MEA);
1543
1544                         stripe_count = le32_to_cpu(lmv->lmv_stripe_count);
1545                         fixed_layout = lmv_is_fixed(lmv);
1546                         if (magic == LMV_MAGIC_STRIPE && lmv_is_restriping(lmv))
1547                                 mdt_restripe_migrate_add(info, o);
1548                         else if (magic == LMV_MAGIC_V1 &&
1549                                  lmv_is_restriping(lmv))
1550                                 mdt_restripe_update_add(info, o);
1551                 }
1552                 if (ma->ma_valid & MA_LMV_DEF) {
1553                         /* Return -ENOTSUPP for old client */
1554                         if (!mdt_is_striped_client(req->rq_export))
1555                                 RETURN(-ENOTSUPP);
1556                         LASSERT(S_ISDIR(la->la_mode));
1557                         /*
1558                          * when ll_dir_getstripe() gets default LMV, it
1559                          * checks mbo_eadatasize.
1560                          */
1561                         if (!(ma->ma_valid & MA_LMV))
1562                                 repbody->mbo_eadatasize =
1563                                         ma->ma_default_lmv_size;
1564                         repbody->mbo_valid |= (OBD_MD_FLDIREA |
1565                                                OBD_MD_DEFAULT_MEA);
1566                 }
1567                 CDEBUG(D_VFSTRACE,
1568                        "dirent count %llu stripe count %u MDT count %d\n",
1569                        ma->ma_attr.la_dirent_count, stripe_count,
1570                        atomic_read(&mdt->mdt_mds_mds_conns) + 1);
1571                 if (ma->ma_attr.la_dirent_count != LU_DIRENT_COUNT_UNSET &&
1572                     ma->ma_attr.la_dirent_count >
1573                         mdt->mdt_restriper.mdr_dir_split_count &&
1574                     !fid_is_root(mdt_object_fid(o)) &&
1575                     mdt->mdt_enable_dir_auto_split &&
1576                     !o->mot_restriping &&
1577                     stripe_count < atomic_read(&mdt->mdt_mds_mds_conns) + 1 &&
1578                     !fixed_layout)
1579                         mdt_auto_split_add(info, o);
1580         } else if (S_ISLNK(la->la_mode) &&
1581                    reqbody->mbo_valid & OBD_MD_LINKNAME) {
1582                 buffer->lb_buf = ma->ma_lmm;
1583                 /* eadatasize from client includes NULL-terminator, so
1584                  * there is no need to read it
1585                  */
1586                 buffer->lb_len = reqbody->mbo_eadatasize - 1;
1587                 rc = mo_readlink(env, next, buffer);
1588                 if (unlikely(rc <= 0)) {
1589                         CERROR("%s: readlink failed for "DFID": rc = %d\n",
1590                                mdt_obd_name(info->mti_mdt),
1591                                PFID(mdt_object_fid(o)), rc);
1592                         rc = -EFAULT;
1593                 } else {
1594                         int print_limit = min_t(int, PAGE_SIZE - 128, rc);
1595
1596                         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
1597                                 rc -= 2;
1598                         repbody->mbo_valid |= OBD_MD_LINKNAME;
1599                         /* we need to report back size with NULL-terminator
1600                          * because client expects that
1601                          */
1602                         repbody->mbo_eadatasize = rc + 1;
1603                         if (repbody->mbo_eadatasize != reqbody->mbo_eadatasize)
1604                                 CDEBUG(D_INODE, "%s: Read shorter symlink %d on "
1605                                        DFID ", expected %d\n",
1606                                        mdt_obd_name(info->mti_mdt),
1607                                        rc, PFID(mdt_object_fid(o)),
1608                                        reqbody->mbo_eadatasize - 1);
1609                         /* NULL terminate */
1610                         ((char *)ma->ma_lmm)[rc] = 0;
1611
1612                         /* If the total CDEBUG() size is larger than a page, it
1613                          * will print a warning to the console, avoid this by
1614                          * printing just the last part of the symlink.
1615                          */
1616                         CDEBUG(D_INODE, "symlink dest %s%.*s, len = %d\n",
1617                                print_limit < rc ? "..." : "", print_limit,
1618                                (char *)ma->ma_lmm + rc - print_limit, rc);
1619                         rc = 0;
1620                 }
1621         }
1622
1623         if (reqbody->mbo_valid & OBD_MD_FLMODEASIZE) {
1624                 repbody->mbo_max_mdsize = info->mti_mdt->mdt_max_mdsize;
1625                 repbody->mbo_valid |= OBD_MD_FLMODEASIZE;
1626                 CDEBUG(D_INODE, "changing the max MD size to %u\n",
1627                        repbody->mbo_max_mdsize);
1628         }
1629
1630 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
1631         if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
1632                  (reqbody->mbo_valid & OBD_MD_FLACL)) {
1633                 struct lu_nodemap *nodemap = nodemap_get_from_exp(exp);
1634
1635                 if (IS_ERR(nodemap))
1636                         RETURN(PTR_ERR(nodemap));
1637
1638                 rc = mdt_pack_acl2body(info, repbody, o, nodemap);
1639                 nodemap_putref(nodemap);
1640         }
1641 #endif
1642
1643 out:
1644         if (rc == 0)
1645                 mdt_counter_incr(req, LPROC_MDT_GETATTR,
1646                                  ktime_us_delta(ktime_get(), kstart));
1647
1648         RETURN(rc);
1649 }
1650
1651 static int mdt_getattr(struct tgt_session_info *tsi)
1652 {
1653         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
1654         struct mdt_object       *obj = info->mti_object;
1655         struct req_capsule      *pill = info->mti_pill;
1656         struct mdt_body         *reqbody;
1657         struct mdt_body         *repbody;
1658         int rc, rc2;
1659
1660         ENTRY;
1661
1662         if (unlikely(info->mti_object == NULL))
1663                 RETURN(-EPROTO);
1664
1665         reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
1666         LASSERT(reqbody);
1667         LASSERT(lu_object_assert_exists(&obj->mot_obj));
1668
1669         /* Special case for Data-on-MDT files to get data version */
1670         if (unlikely(reqbody->mbo_valid & OBD_MD_FLDATAVERSION)) {
1671                 rc = mdt_data_version_get(tsi);
1672                 GOTO(out, rc);
1673         }
1674
1675         /* Unlike intent case where we need to pre-fill out buffers early on
1676          * in intent policy for ldlm reasons, here we can have a much better
1677          * guess at EA size by just reading it from disk.
1678          * Exceptions are readdir and (missing) directory striping
1679          */
1680         if (reqbody->mbo_valid & OBD_MD_LINKNAME) { /* Readlink */
1681                 /* No easy way to know how long is the symlink, but it cannot
1682                  * be more than PATH_MAX, so we allocate +1
1683                  */
1684                 rc = PATH_MAX + 1;
1685         /* A special case for fs ROOT: getattr there might fetch
1686          * default EA for entire fs, not just for this dir!
1687          */
1688         } else if (lu_fid_eq(mdt_object_fid(obj),
1689                              &info->mti_mdt->mdt_md_root_fid) &&
1690                    (reqbody->mbo_valid & OBD_MD_FLDIREA) &&
1691                    (lustre_msg_get_opc(mdt_info_req(info)->rq_reqmsg) ==
1692                                                                  MDS_GETATTR)) {
1693                 /* Should the default strping be bigger, mdt_fix_reply
1694                  * will reallocate
1695                  */
1696                 rc = DEF_REP_MD_SIZE;
1697         } else {
1698                 /* Read the actual EA size from disk */
1699                 rc = mdt_attr_get_eabuf_size(info, obj);
1700         }
1701
1702         if (rc < 0)
1703                 GOTO(out, rc = err_serious(rc));
1704
1705         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, rc);
1706
1707         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
1708          * by default. If the target object has more ACL entries, then
1709          * enlarge the buffer when necessary.
1710          */
1711         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
1712                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
1713         mdt_preset_encctx_size(info);
1714
1715         rc = req_capsule_server_pack(pill);
1716         if (unlikely(rc != 0))
1717                 GOTO(out, rc = err_serious(rc));
1718
1719         repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
1720         LASSERT(repbody != NULL);
1721         repbody->mbo_eadatasize = 0;
1722         repbody->mbo_aclsize = 0;
1723
1724         rc = mdt_check_ucred(info);
1725         if (unlikely(rc))
1726                 GOTO(out_shrink, rc);
1727
1728         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
1729
1730         rc = mdt_init_ucred(info, reqbody);
1731         if (rc)
1732                 GOTO(out_shrink, rc);
1733
1734         rc = mdt_getattr_internal(info, obj, 0);
1735         if (unlikely(rc))
1736                 GOTO(out_ucred, rc);
1737
1738         rc = mdt_pack_encctx_in_reply(info, obj);
1739         EXIT;
1740 out_ucred:
1741         mdt_exit_ucred(info);
1742 out_shrink:
1743         mdt_client_compatibility(info);
1744         rc2 = mdt_fix_reply(info);
1745         if (rc == 0)
1746                 rc = rc2;
1747 out:
1748         mdt_thread_info_fini(info);
1749         return rc;
1750 }
1751
1752 /**
1753  * Handler of layout intent RPC requiring the layout modification
1754  *
1755  * \param[in]  info     thread environment
1756  * \param[in]  obj      object
1757  * \param[out] lhc      object ldlm lock handle
1758  * \param[in]  layout   layout change descriptor
1759  *
1760  * \retval 0    on success
1761  * \retval < 0  error code
1762  */
1763 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
1764                       struct mdt_lock_handle *lhc,
1765                       struct md_layout_change *layout)
1766 {
1767         int rc;
1768
1769         ENTRY;
1770
1771         if (!mdt_object_exists(obj))
1772                 RETURN(-ENOENT);
1773
1774         if (!S_ISREG(lu_object_attr(&obj->mot_obj)))
1775                 RETURN(-EINVAL);
1776
1777         rc = mo_permission(info->mti_env, NULL, mdt_object_child(obj), NULL,
1778                            MAY_WRITE);
1779         if (rc)
1780                 RETURN(rc);
1781
1782         rc = mdt_check_resent_lock(info, obj, lhc);
1783         if (rc < 0)
1784                 RETURN(rc);
1785
1786         if (rc > 0) {
1787                 /* not resent */
1788                 __u64 lockpart = MDS_INODELOCK_LAYOUT;
1789
1790                 /* take layout lock to prepare layout change */
1791                 if (layout->mlc_opc == MD_LAYOUT_WRITE)
1792                         lockpart |= MDS_INODELOCK_UPDATE;
1793
1794                 rc = mdt_object_lock(info, obj, lhc, lockpart, LCK_EX);
1795                 if (rc)
1796                         RETURN(rc);
1797         }
1798
1799         mutex_lock(&obj->mot_som_mutex);
1800         rc = mo_layout_change(info->mti_env, mdt_object_child(obj), layout);
1801         mutex_unlock(&obj->mot_som_mutex);
1802
1803         if (rc)
1804                 mdt_object_unlock(info, obj, lhc, 1);
1805
1806         RETURN(rc);
1807 }
1808
1809 /**
1810  * Exchange MOF_LOV_CREATED flags between two objects after a
1811  * layout swap. No assumption is made on whether o1 or o2 have
1812  * created objects or not.
1813  *
1814  * \param[in,out] o1    First swap layout object
1815  * \param[in,out] o2    Second swap layout object
1816  */
1817 static void mdt_swap_lov_flag(struct mdt_object *o1, struct mdt_object *o2)
1818 {
1819         unsigned int o1_lov_created = o1->mot_lov_created;
1820
1821         mutex_lock(&o1->mot_lov_mutex);
1822         mutex_lock(&o2->mot_lov_mutex);
1823
1824         o1->mot_lov_created = o2->mot_lov_created;
1825         o2->mot_lov_created = o1_lov_created;
1826
1827         mutex_unlock(&o2->mot_lov_mutex);
1828         mutex_unlock(&o1->mot_lov_mutex);
1829 }
1830
1831 static int mdt_swap_layouts(struct tgt_session_info *tsi)
1832 {
1833         struct mdt_thread_info  *info;
1834         struct ptlrpc_request   *req = tgt_ses_req(tsi);
1835         struct obd_export       *exp = req->rq_export;
1836         struct mdt_object       *o1, *o2, *o;
1837         struct mdt_lock_handle  *lh1, *lh2;
1838         struct mdc_swap_layouts *msl;
1839         int rc;
1840
1841         ENTRY;
1842
1843         /* client does not support layout lock, so layout swaping
1844          * is disabled.
1845          * FIXME: there is a problem for old clients which don't support
1846          * layout lock yet. If those clients have already opened the file
1847          * they won't be notified at all so that old layout may still be
1848          * used to do IO. This can be fixed after file release is landed by
1849          * doing exclusive open and taking full EX ibits lock. - Jinshan
1850          */
1851         if (!exp_connect_layout(exp))
1852                 RETURN(-EOPNOTSUPP);
1853
1854         info = tsi2mdt_info(tsi);
1855         if (unlikely(info->mti_object == NULL))
1856                 RETURN(-EPROTO);
1857
1858         if (info->mti_dlm_req != NULL)
1859                 ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
1860
1861         o1 = info->mti_object;
1862         o = o2 = mdt_object_find(info->mti_env, info->mti_mdt,
1863                                 &info->mti_body->mbo_fid2);
1864         if (IS_ERR(o))
1865                 GOTO(out, rc = PTR_ERR(o));
1866
1867         if (mdt_object_remote(o) || !mdt_object_exists(o)) /* remote object */
1868                 GOTO(put, rc = -ENOENT);
1869
1870         rc = lu_fid_cmp(&info->mti_body->mbo_fid1, &info->mti_body->mbo_fid2);
1871         if (unlikely(rc == 0)) /* same file, you kidding me? no-op. */
1872                 GOTO(put, rc);
1873
1874         if (rc < 0)
1875                 swap(o1, o2);
1876
1877         /* permission check. Make sure the calling process having permission
1878          * to write both files.
1879          */
1880         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o1), NULL,
1881                            MAY_WRITE);
1882         if (rc < 0)
1883                 GOTO(put, rc);
1884
1885         rc = mo_permission(info->mti_env, NULL, mdt_object_child(o2), NULL,
1886                            MAY_WRITE);
1887         if (rc < 0)
1888                 GOTO(put, rc);
1889
1890         msl = req_capsule_client_get(info->mti_pill, &RMF_SWAP_LAYOUTS);
1891         if (msl == NULL)
1892                 GOTO(put, rc = -EPROTO);
1893
1894         lh1 = &info->mti_lh[MDT_LH_NEW];
1895         lh2 = &info->mti_lh[MDT_LH_OLD];
1896         rc = mdt_object_lock(info, o1, lh1, MDS_INODELOCK_LAYOUT |
1897                              MDS_INODELOCK_XATTR, LCK_EX);
1898         if (rc < 0)
1899                 GOTO(put, rc);
1900
1901         rc = mdt_object_lock(info, o2, lh2, MDS_INODELOCK_LAYOUT |
1902                              MDS_INODELOCK_XATTR, LCK_EX);
1903         if (rc < 0)
1904                 GOTO(unlock1, rc);
1905
1906         rc = mo_swap_layouts(info->mti_env, mdt_object_child(o1),
1907                              mdt_object_child(o2), 0, 0, msl->msl_flags);
1908         if (rc < 0)
1909                 GOTO(unlock2, rc);
1910
1911         mdt_swap_lov_flag(o1, o2);
1912
1913 unlock2:
1914         mdt_object_unlock(info, o2, lh2, rc);
1915 unlock1:
1916         mdt_object_unlock(info, o1, lh1, rc);
1917 put:
1918         mdt_object_put(info->mti_env, o);
1919 out:
1920         mdt_thread_info_fini(info);
1921         RETURN(rc);
1922 }
1923
1924 static int mdt_raw_lookup(struct mdt_thread_info *info,
1925                           struct mdt_object *parent,
1926                           const struct lu_name *lname)
1927 {
1928         struct lu_fid *fid = &info->mti_tmp_fid1;
1929         struct mdt_body *repbody;
1930         bool is_dotdot = false;
1931         bool is_old_parent_stripe = false;
1932         bool is_new_parent_checked = false;
1933         int rc;
1934
1935         ENTRY;
1936
1937         LASSERT(!info->mti_cross_ref);
1938         /* Always allow to lookup ".." */
1939         if (lname->ln_namelen == 2 &&
1940             lname->ln_name[0] == '.' && lname->ln_name[1] == '.') {
1941                 info->mti_spec.sp_permitted = 1;
1942                 is_dotdot = true;
1943                 if (mdt_is_dir_stripe(info, parent) == 1)
1944                         is_old_parent_stripe = true;
1945         }
1946
1947         mdt_object_get(info->mti_env, parent);
1948 lookup:
1949         /* Only got the fid of this obj by name */
1950         fid_zero(fid);
1951         rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, fid,
1952                         &info->mti_spec);
1953         mdt_object_put(info->mti_env, parent);
1954         if (rc)
1955                 RETURN(rc);
1956
1957         /* getattr_name("..") should return master object FID for striped dir */
1958         if (is_dotdot && (is_old_parent_stripe || !is_new_parent_checked)) {
1959                 parent = mdt_object_find(info->mti_env, info->mti_mdt, fid);
1960                 if (IS_ERR(parent))
1961                         RETURN(PTR_ERR(parent));
1962
1963                 /* old client getattr_name("..") with stripe FID */
1964                 if (unlikely(is_old_parent_stripe)) {
1965                         is_old_parent_stripe = false;
1966                         goto lookup;
1967                 }
1968
1969                 /* ".." may be a stripe */
1970                 if (unlikely(mdt_is_dir_stripe(info, parent) == 1)) {
1971                         is_new_parent_checked = true;
1972                         goto lookup;
1973                 }
1974
1975                 mdt_object_put(info->mti_env, parent);
1976         }
1977
1978         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
1979         repbody->mbo_fid1 = *fid;
1980         repbody->mbo_valid = OBD_MD_FLID;
1981
1982         RETURN(rc);
1983 }
1984
1985 /**
1986  * Find name matching hash
1987  *
1988  * We search \a child LinkEA for a name whose hash matches \a lname
1989  * (it contains an encoded hash).
1990  *
1991  * \param info mdt thread info
1992  * \param lname encoded hash to find
1993  * \param parent parent object
1994  * \param child object to search with LinkEA
1995  *
1996  * \retval 1 match found
1997  * \retval 0 no match found
1998  * \retval -ev negative errno upon error
1999  */
2000 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
2001                             struct mdt_object *parent, struct mdt_object *child)
2002 {
2003         /* Here, lname is an encoded hash of on-disk name, and
2004          * client is doing access without encryption key.
2005          * So we need to get LinkEA, check parent fid is correct and
2006          * compare name hash with the one in the request.
2007          */
2008         struct lu_buf *buf = &info->mti_big_buf;
2009         struct lu_name name;
2010         struct lu_fid pfid;
2011         struct linkea_data ldata = { NULL };
2012         struct link_ea_header *leh;
2013         struct link_ea_entry *lee;
2014         struct lu_buf link = { 0 };
2015         char *hash;
2016         int reclen, count, rc;
2017
2018         ENTRY;
2019         if (lname->ln_namelen < LL_CRYPTO_BLOCK_SIZE)
2020                 RETURN(-EINVAL);
2021
2022         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2023         if (!buf->lb_buf)
2024                 RETURN(-ENOMEM);
2025
2026         ldata.ld_buf = buf;
2027         rc = mdt_links_read(info, child, &ldata);
2028         if (rc < 0)
2029                 RETURN(rc);
2030
2031         hash = kmalloc(lname->ln_namelen, GFP_NOFS);
2032         if (!hash)
2033                 RETURN(-ENOMEM);
2034         rc = critical_decode(lname->ln_name, lname->ln_namelen, hash);
2035
2036         leh = buf->lb_buf;
2037         lee = (struct link_ea_entry *)(leh + 1);
2038         for (count = 0; count < leh->leh_reccount; count++) {
2039                 linkea_entry_unpack(lee, &reclen, &name, &pfid);
2040                 if (!parent || lu_fid_eq(&pfid, mdt_object_fid(parent))) {
2041                         lu_buf_check_and_alloc(&link, name.ln_namelen);
2042                         if (!link.lb_buf)
2043                                 GOTO(out_match, rc = -ENOMEM);
2044                         rc = critical_decode(name.ln_name, name.ln_namelen,
2045                                              link.lb_buf);
2046
2047                         if (memcmp(LLCRYPT_EXTRACT_DIGEST(link.lb_buf, rc),
2048                                    hash, LL_CRYPTO_BLOCK_SIZE) == 0) {
2049                                 *lname = name;
2050                                 break;
2051                         }
2052                 }
2053                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2054         }
2055         if (count == leh->leh_reccount)
2056                 rc = 0;
2057         else
2058                 rc = 1;
2059
2060 out_match:
2061         lu_buf_free(&link);
2062         kfree(hash);
2063
2064         RETURN(rc);
2065 }
2066
2067 /*
2068  * UPDATE lock should be taken against parent, and be released before exit;
2069  * child_bits lock should be taken against child, and be returned back:
2070  *            (1)normal request should release the child lock;
2071  *            (2)intent request will grant the lock to client.
2072  */
2073 static int mdt_getattr_name_lock(struct mdt_thread_info *info,
2074                                  struct mdt_lock_handle *lhc,
2075                                  __u64 child_bits,
2076                                  struct ldlm_reply *ldlm_rep)
2077 {
2078         struct ptlrpc_request *req = mdt_info_req(info);
2079         struct mdt_body *reqbody = NULL;
2080         struct mdt_object *parent = info->mti_object;
2081         struct mdt_object *child = NULL;
2082         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2083         struct lu_name *lname = NULL;
2084         struct mdt_lock_handle *lhp = NULL;
2085         struct ldlm_lock *lock;
2086         struct req_capsule *pill = info->mti_pill;
2087         bool fscrypt_md = false;
2088         __u64 try_bits = 0;
2089         bool is_resent;
2090         int ma_need = 0;
2091         int rc;
2092
2093         ENTRY;
2094
2095         is_resent = lustre_handle_is_used(&lhc->mlh_reg_lh);
2096         LASSERT(ergo(is_resent,
2097                      lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
2098
2099         if (parent == NULL)
2100                 RETURN(-ENOENT);
2101
2102         if (info->mti_mdt->mdt_enable_dir_auto_split)
2103                 ma_need |= MA_DIRENT_CNT;
2104
2105         if (info->mti_cross_ref) {
2106                 /* Only getattr on the child. Parent is on another node. */
2107                 mdt_set_disposition(info, ldlm_rep,
2108                                     DISP_LOOKUP_EXECD | DISP_LOOKUP_POS);
2109                 child = parent;
2110                 CDEBUG(D_INODE, "partial getattr_name child_fid = "DFID", ldlm_rep = %p\n",
2111                        PFID(mdt_object_fid(child)), ldlm_rep);
2112
2113                 rc = mdt_check_resent_lock(info, child, lhc);
2114                 if (rc < 0) {
2115                         RETURN(rc);
2116                 } else if (rc > 0) {
2117                         /*
2118                          * Object's name entry is on another MDS, it will
2119                          * request PERM lock only because LOOKUP lock is owned
2120                          * by the MDS where name entry resides.
2121                          *
2122                          * TODO: it should try layout lock too. - Jinshan
2123                          */
2124                         child_bits &= ~(MDS_INODELOCK_LOOKUP |
2125                                         MDS_INODELOCK_LAYOUT);
2126                         child_bits |= MDS_INODELOCK_PERM;
2127                         rc = mdt_object_lock(info, child, lhc, child_bits,
2128                                              LCK_PR);
2129                         if (rc < 0)
2130                                 RETURN(rc);
2131                 }
2132
2133                 /* Finally, we can get attr for child. */
2134                 if (!mdt_object_exists(child)) {
2135                         LU_OBJECT_DEBUG(D_INFO, info->mti_env,
2136                                         &child->mot_obj,
2137                                         "remote object doesn't exist.");
2138                         mdt_object_unlock(info, child, lhc, 1);
2139                         RETURN(-ENOENT);
2140                 }
2141
2142                 rc = mdt_getattr_internal(info, child, ma_need);
2143                 if (unlikely(rc != 0)) {
2144                         mdt_object_unlock(info, child, lhc, 1);
2145                         RETURN(rc);
2146                 }
2147
2148                 rc = mdt_pack_secctx_in_reply(info, child);
2149                 if (unlikely(rc)) {
2150                         mdt_object_unlock(info, child, lhc, 1);
2151                         RETURN(rc);
2152                 }
2153
2154                 rc = mdt_pack_encctx_in_reply(info, child);
2155                 if (unlikely(rc))
2156                         mdt_object_unlock(info, child, lhc, 1);
2157                 RETURN(rc);
2158         }
2159
2160         lname = &info->mti_name;
2161         mdt_name_unpack(pill, &RMF_NAME, lname, MNF_FIX_ANON);
2162
2163         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2164                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2165                 if (unlikely(reqbody == NULL))
2166                         RETURN(err_serious(-EPROTO));
2167
2168                 *child_fid = reqbody->mbo_fid2;
2169                 if (unlikely(!fid_is_sane(child_fid)))
2170                         RETURN(err_serious(-EINVAL));
2171
2172                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2173                         mdt_object_get(info->mti_env, parent);
2174                         child = parent;
2175                 } else {
2176                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2177                                                 child_fid);
2178                         if (IS_ERR(child))
2179                                 RETURN(PTR_ERR(child));
2180                 }
2181
2182                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", ldlm_rep = %p\n",
2183                        PFID(mdt_object_fid(parent)),
2184                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2185         } else if (lu_name_is_valid(lname)) {
2186                 if (mdt_object_remote(parent)) {
2187                         CERROR("%s: parent "DFID" is on remote target\n",
2188                                mdt_obd_name(info->mti_mdt),
2189                                PFID(mdt_object_fid(parent)));
2190                         RETURN(-EPROTO);
2191                 }
2192
2193                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DNAME", ldlm_rep = %p\n",
2194                        PFID(mdt_object_fid(parent)),
2195                        PNAME(lname), ldlm_rep);
2196
2197                 if (parent->mot_obj.lo_header->loh_attr & LOHA_FSCRYPT_MD ||
2198                     (fid_is_root(mdt_object_fid(parent)) &&
2199                      lname->ln_namelen == strlen(dot_fscrypt_name) &&
2200                      strncmp(lname->ln_name, dot_fscrypt_name,
2201                              lname->ln_namelen) == 0))
2202                         fscrypt_md = true;
2203         } else {
2204                 reqbody = req_capsule_client_get(pill, &RMF_MDT_BODY);
2205                 if (unlikely(reqbody == NULL))
2206                         RETURN(err_serious(-EPROTO));
2207
2208                 *child_fid = reqbody->mbo_fid2;
2209                 if (unlikely(!fid_is_sane(child_fid)))
2210                         RETURN(err_serious(-EINVAL));
2211
2212                 if (lu_fid_eq(mdt_object_fid(parent), child_fid)) {
2213                         mdt_object_get(info->mti_env, parent);
2214                         child = parent;
2215                 } else {
2216                         child = mdt_object_find(info->mti_env, info->mti_mdt,
2217                                                 child_fid);
2218                         if (IS_ERR(child))
2219                                 RETURN(PTR_ERR(child));
2220                 }
2221
2222                 if (mdt_object_remote(child)) {
2223                         CERROR("%s: child "DFID" is on remote target\n",
2224                                mdt_obd_name(info->mti_mdt),
2225                                PFID(mdt_object_fid(child)));
2226                         GOTO(out_child, rc = -EPROTO);
2227                 }
2228
2229                 /* don't fetch LOOKUP lock if it's remote object */
2230                 rc = mdt_is_remote_object(info, parent, child);
2231                 if (rc < 0)
2232                         GOTO(out_child, rc);
2233                 if (rc)
2234                         child_bits &= ~MDS_INODELOCK_LOOKUP;
2235
2236                 CDEBUG(D_INODE, "getattr with lock for "DFID"/"DFID", ldlm_rep = %p\n",
2237                        PFID(mdt_object_fid(parent)),
2238                        PFID(&reqbody->mbo_fid2), ldlm_rep);
2239         }
2240
2241         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_EXECD);
2242
2243         if (unlikely(!mdt_object_exists(parent)) &&
2244             !(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2245             lu_name_is_valid(lname)) {
2246                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2247                                 &parent->mot_obj,
2248                                 "Parent doesn't exist!");
2249                 GOTO(out_child, rc = -ESTALE);
2250         }
2251
2252         if (!child && is_resent) {
2253                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2254                 if (lock == NULL) {
2255                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
2256                          * a resend case, however, it could be already destroyed
2257                          * due to client eviction or a raced cancel RPC.
2258                          */
2259                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
2260                                           lhc->mlh_reg_lh.cookie);
2261                         RETURN(-ESTALE);
2262                 }
2263                 fid_extract_from_res_name(child_fid,
2264                                           &lock->l_resource->lr_name);
2265                 LDLM_LOCK_PUT(lock);
2266                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2267                                         child_fid);
2268                 if (IS_ERR(child))
2269                         RETURN(PTR_ERR(child));
2270         } else if (!(info->mti_body->mbo_valid & OBD_MD_NAMEHASH) &&
2271             lu_name_is_valid(lname)) {
2272                 if (info->mti_body->mbo_valid == OBD_MD_FLID) {
2273                         rc = mdt_raw_lookup(info, parent, lname);
2274
2275                         RETURN(rc);
2276                 }
2277
2278                 /* step 1: lock parent only if parent is a directory */
2279                 if (S_ISDIR(lu_object_attr(&parent->mot_obj))) {
2280                         lhp = &info->mti_lh[MDT_LH_PARENT];
2281                         rc = mdt_parent_lock(info, parent, lhp, lname, LCK_PR);
2282                         if (unlikely(rc != 0))
2283                                 RETURN(rc);
2284                 }
2285
2286                 /* step 2: lookup child's fid by name */
2287                 fid_zero(child_fid);
2288                 rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
2289                                 child_fid, &info->mti_spec);
2290                 if (rc == -ENOENT)
2291                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2292
2293                 if (rc != 0)
2294                         GOTO(unlock_parent, rc);
2295
2296                 child = mdt_object_find(info->mti_env, info->mti_mdt,
2297                                         child_fid);
2298                 if (unlikely(IS_ERR(child)))
2299                         GOTO(unlock_parent, rc = PTR_ERR(child));
2300         }
2301
2302         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2303
2304         /* step 3: lock child regardless if it is local or remote. */
2305         LASSERT(child);
2306
2307         if (info->mti_body->mbo_valid & OBD_MD_NAMEHASH) {
2308                 /* Here, lname is an encoded hash of on-disk name, and
2309                  * client is doing access without encryption key.
2310                  * So we need to compare name hash with the one in the request.
2311                  */
2312                 if (!find_name_matching_hash(info, lname, parent,
2313                                              child)) {
2314                         mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
2315                         mdt_clear_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
2316                         GOTO(out_child, rc = -ENOENT);
2317                 }
2318         }
2319
2320         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout * 2);
2321         if (!mdt_object_exists(child)) {
2322                 LU_OBJECT_DEBUG(D_INODE, info->mti_env,
2323                                 &child->mot_obj,
2324                                 "Object doesn't exist!");
2325                 GOTO(out_child, rc = -ENOENT);
2326         }
2327
2328         rc = mdt_check_resent_lock(info, child, lhc);
2329         if (rc < 0) {
2330                 GOTO(out_child, rc);
2331         } else if (rc > 0) {
2332                 if (!(child_bits & MDS_INODELOCK_UPDATE) &&
2333                     !mdt_object_remote(child)) {
2334                         struct md_attr *ma = &info->mti_attr;
2335
2336                         ma->ma_valid = 0;
2337                         ma->ma_need = MA_INODE;
2338                         rc = mdt_attr_get_complex(info, child, ma);
2339                         if (unlikely(rc != 0))
2340                                 GOTO(out_child, rc);
2341
2342                         /* If the file has not been changed for some time, we
2343                          * return not only a LOOKUP lock, but also an UPDATE
2344                          * lock and this might save us RPC on later STAT. For
2345                          * directories, it also let negative dentry cache start
2346                          * working for this dir.
2347                          */
2348                         if (ma->ma_valid & MA_INODE &&
2349                             ma->ma_attr.la_valid & LA_CTIME &&
2350                             info->mti_mdt->mdt_namespace->ns_ctime_age_limit +
2351                             ma->ma_attr.la_ctime < ktime_get_real_seconds())
2352                                 child_bits |= MDS_INODELOCK_UPDATE;
2353                 }
2354
2355                 /* layout lock must be granted in a best-effort way
2356                  * for IT operations
2357                  */
2358                 LASSERT(!(child_bits & MDS_INODELOCK_LAYOUT));
2359                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2360                     !mdt_object_remote(child) && ldlm_rep != NULL) {
2361                         if (!CFS_FAIL_CHECK(OBD_FAIL_MDS_NO_LL_GETATTR) &&
2362                             exp_connect_layout(info->mti_exp)) {
2363                                 /* try to grant layout lock for regular file. */
2364                                 try_bits = MDS_INODELOCK_LAYOUT;
2365                         }
2366                         /* Acquire DOM lock in advance for data-on-mdt file */
2367                         if (child != parent)
2368                                 try_bits |= MDS_INODELOCK_DOM;
2369                 }
2370
2371                 /*
2372                  * To avoid possible deadlock between batched statahead RPC
2373                  * and rename()/migrate() operation, it should use trylock to
2374                  * obtain the DLM PR ibits lock for file attributes in a
2375                  * batched statahead RPC. A failed trylock means that other
2376                  * users maybe modify the directory simultaneously as in current
2377                  * Lustre design the server only grants read lock to a client.
2378                  *
2379                  * When a trylock failed, the MDT reports the conflict with
2380                  * error code -EBUSY, and stops statahead immediately.
2381                  */
2382                 if (info->mti_batch_env) {
2383                         /*
2384                          * This is a sub stat-ahead request in a batched RPC.
2385                          * However, the @child is a remote object, we just
2386                          * return -EREMOTE here to forbid stat-ahead on it.
2387                          */
2388                         if (mdt_object_remote(child))
2389                                 GOTO(out_child, rc = -EREMOTE);
2390
2391                         try_bits |= child_bits;
2392                         child_bits = 0;
2393                 }
2394
2395                 if (try_bits != 0) {
2396                         /* try layout lock, it may fail to be granted due to
2397                          * contention at LOOKUP or UPDATE
2398                          */
2399                         rc = mdt_object_lock_try(info, child, lhc, &child_bits,
2400                                                  try_bits, LCK_PR);
2401                         if (child_bits & MDS_INODELOCK_LAYOUT)
2402                                 ma_need |= MA_LOV;
2403                 } else {
2404                         /* Do not enqueue the UPDATE lock from MDT(cross-MDT),
2405                          * client will enqueue the lock to the remote MDT
2406                          */
2407                         if (mdt_object_remote(child))
2408                                 rc = mdt_object_lookup_lock(info, NULL, child,
2409                                                             lhc, LCK_PR);
2410                         else
2411                                 rc = mdt_object_lock(info, child, lhc,
2412                                                      child_bits, LCK_PR);
2413                 }
2414                 if (unlikely(rc != 0))
2415                         GOTO(out_child, rc);
2416                 if (info->mti_batch_env && child_bits == 0) {
2417                         if (!is_resent)
2418                                 mdt_object_unlock(info, child, lhc, 1);
2419                         GOTO(out_child, rc = -EBUSY);
2420                 }
2421         }
2422
2423         if (fscrypt_md)
2424                 child->mot_obj.lo_header->loh_attr |= LOHA_FSCRYPT_MD;
2425
2426         /* finally, we can get attr for child. */
2427         rc = mdt_getattr_internal(info, child, ma_need);
2428         if (unlikely(rc != 0)) {
2429                 if (!is_resent)
2430                         mdt_object_unlock(info, child, lhc, 1);
2431                 GOTO(out_child, rc);
2432         }
2433
2434         rc = mdt_pack_secctx_in_reply(info, child);
2435         if (unlikely(rc)) {
2436                 if (!is_resent)
2437                         mdt_object_unlock(info, child, lhc, 1);
2438                 GOTO(out_child, rc);
2439         }
2440
2441         rc = mdt_pack_encctx_in_reply(info, child);
2442         if (unlikely(rc)) {
2443                 if (!is_resent)
2444                         mdt_object_unlock(info, child, lhc, 1);
2445                 GOTO(out_child, rc);
2446         }
2447
2448         lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
2449         if (lock) {
2450                 /* Debugging code. */
2451                 LDLM_DEBUG(lock, "Returning lock to client");
2452                 LASSERTF(fid_res_name_eq(mdt_object_fid(child),
2453                                          &lock->l_resource->lr_name),
2454                          "Lock res_id: "DLDLMRES", fid: "DFID"\n",
2455                          PLDLMRES(lock->l_resource),
2456                          PFID(mdt_object_fid(child)));
2457
2458                 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_ENQ_RESEND))) {
2459                         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
2460                                 CFS_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_ENQ_RESEND,
2461                                                  req->rq_deadline -
2462                                                  req->rq_arrival_time.tv_sec +
2463                                                  cfs_fail_val ?: 3);
2464                         /* Put the lock to the waiting list and force the cancel */
2465                         ldlm_set_ast_sent(lock);
2466                 }
2467
2468                 /*
2469                  * check whether the object is remote as we can't
2470                  * really check attributes w/o explicit check for
2471                  * object's existence first.
2472                  */
2473                 if (!mdt_object_remote(child) && child != parent &&
2474                     S_ISREG(lu_object_attr(&child->mot_obj))) {
2475                         mdt_object_put(info->mti_env, child);
2476                         rc = mdt_pack_size2body(info, child_fid,
2477                                                 &lhc->mlh_reg_lh);
2478                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2479                                 /* DOM lock was taken in advance but this is
2480                                  * not DoM file. Drop the lock.
2481                                  */
2482                                 lock_res_and_lock(lock);
2483                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2484                                 unlock_res_and_lock(lock);
2485                         }
2486                         LDLM_LOCK_PUT(lock);
2487                         GOTO(unlock_parent, rc = 0);
2488                 }
2489                 LDLM_LOCK_PUT(lock);
2490         }
2491
2492         EXIT;
2493 out_child:
2494         if (child)
2495                 mdt_object_put(info->mti_env, child);
2496 unlock_parent:
2497         if (lhp)
2498                 mdt_object_unlock(info, parent, lhp, 1);
2499         if (rc == -ENOENT) {
2500                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2501                  * client if trying to access an encrypted file
2502                  */
2503                 int rc2 = mdt_check_enc(info, parent);
2504
2505                 if (rc2)
2506                         rc = rc2;
2507         }
2508         return rc;
2509 }
2510
2511 /* normal handler: should release the child lock */
2512 static int mdt_getattr_name(struct tgt_session_info *tsi)
2513 {
2514         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2515         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2516         struct mdt_body *reqbody;
2517         struct mdt_body *repbody;
2518         int rc, rc2;
2519
2520         ENTRY;
2521
2522         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2523         LASSERT(reqbody != NULL);
2524         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2525         LASSERT(repbody != NULL);
2526
2527         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2528         repbody->mbo_eadatasize = 0;
2529         repbody->mbo_aclsize = 0;
2530
2531         rc = mdt_init_ucred(info, reqbody);
2532         if (unlikely(rc))
2533                 GOTO(out_shrink, rc);
2534
2535         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2536         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2537                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2538                 lhc->mlh_reg_lh.cookie = 0;
2539         }
2540         mdt_exit_ucred(info);
2541         EXIT;
2542 out_shrink:
2543         mdt_client_compatibility(info);
2544         rc2 = mdt_fix_reply(info);
2545         if (rc == 0)
2546                 rc = rc2;
2547         mdt_thread_info_fini(info);
2548         return rc;
2549 }
2550
2551 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2552                             const struct lu_fid *pfid,
2553                             const struct lu_name *name,
2554                             struct mdt_object *obj, s64 ctime)
2555 {
2556         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2557         struct ldlm_enqueue_info *einfo = &info->mti_einfo;
2558         struct mdt_device *mdt = info->mti_mdt;
2559         struct md_attr *ma = &info->mti_attr;
2560         struct mdt_lock_handle *parent_lh;
2561         struct mdt_lock_handle *child_lh;
2562         struct mdt_object *pobj;
2563         int rc;
2564
2565         ENTRY;
2566
2567         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2568         if (IS_ERR(pobj))
2569                 GOTO(out, rc = PTR_ERR(pobj));
2570
2571         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2572         rc = mdt_parent_lock(info, pobj, parent_lh, name, LCK_PW);
2573         if (rc != 0)
2574                 GOTO(put_parent, rc);
2575
2576         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2577                         name, child_fid, &info->mti_spec);
2578         if (rc != 0)
2579                 GOTO(unlock_parent, rc);
2580
2581         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2582                 GOTO(unlock_parent, rc = -EREMCHG);
2583
2584         child_lh = &info->mti_lh[MDT_LH_CHILD];
2585         rc = mdt_object_stripes_lock(info, pobj, obj, child_lh, einfo,
2586                                      MDS_INODELOCK_LOOKUP |
2587                                      MDS_INODELOCK_UPDATE, LCK_EX);
2588         if (rc != 0)
2589                 GOTO(unlock_parent, rc);
2590
2591         if (atomic_read(&obj->mot_open_count)) {
2592                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2593                        PFID(mdt_object_fid(obj)));
2594                 GOTO(unlock_child, rc = -EBUSY);
2595         }
2596
2597         ma->ma_need = 0;
2598         ma->ma_valid = MA_INODE;
2599         ma->ma_attr.la_valid = LA_CTIME;
2600         ma->ma_attr.la_ctime = ctime;
2601
2602         mutex_lock(&obj->mot_lov_mutex);
2603
2604         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2605                         mdt_object_child(obj), name, ma, 0);
2606
2607         mutex_unlock(&obj->mot_lov_mutex);
2608
2609 unlock_child:
2610         mdt_object_stripes_unlock(info, obj, child_lh, einfo, 1);
2611 unlock_parent:
2612         mdt_object_unlock(info, pobj, parent_lh, 1);
2613 put_parent:
2614         mdt_object_put(info->mti_env, pobj);
2615 out:
2616         RETURN(rc);
2617 }
2618
2619 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2620                                         struct mdt_object *obj)
2621 {
2622         struct lu_ucred *uc = lu_ucred(info->mti_env);
2623         struct md_attr *ma = &info->mti_attr;
2624         struct lu_attr *la = &ma->ma_attr;
2625         int rc = 0;
2626
2627         ENTRY;
2628
2629         ma->ma_need = MA_INODE;
2630         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2631         if (rc)
2632                 GOTO(out, rc);
2633
2634         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2635                 rc = -EACCES;
2636
2637         /* we want rbac roles to have precedence over any other
2638          * permission or capability checks
2639          */
2640         if (!uc->uc_rbac_byfid_ops)
2641                 RETURN(-EPERM);
2642         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2643                 RETURN(0);
2644         if (uc->uc_fsuid == la->la_uid) {
2645                 if ((la->la_mode & 0200) == 0)
2646                         rc = -EACCES;
2647         } else if (uc->uc_fsgid == la->la_gid) {
2648                 if ((la->la_mode & 0020) == 0)
2649                         rc = -EACCES;
2650         } else if ((la->la_mode & 0002) == 0) {
2651                         rc = -EACCES;
2652         }
2653
2654 out:
2655         RETURN(rc);
2656 }
2657
2658 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2659                          s64 ctime)
2660 {
2661         struct mdt_device *mdt = info->mti_mdt;
2662         struct mdt_object *obj = NULL;
2663         struct linkea_data ldata = { NULL };
2664         struct lu_buf *buf = &info->mti_big_buf;
2665         struct lu_name *name = &info->mti_name;
2666         struct lu_fid *pfid = &info->mti_tmp_fid1;
2667         struct link_ea_header *leh;
2668         struct link_ea_entry *lee;
2669         int reclen, count, rc = 0;
2670
2671         ENTRY;
2672
2673         if (!fid_is_sane(fid))
2674                 GOTO(out, rc = -EINVAL);
2675
2676         if (!fid_is_namespace_visible(fid))
2677                 GOTO(out, rc = -EINVAL);
2678
2679         obj = mdt_object_find(info->mti_env, mdt, fid);
2680         if (IS_ERR(obj))
2681                 GOTO(out, rc = PTR_ERR(obj));
2682
2683         if (mdt_object_remote(obj))
2684                 GOTO(out, rc = -EREMOTE);
2685         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2686                 GOTO(out, rc = -ENOENT);
2687
2688         rc = mdt_rmfid_check_permission(info, obj);
2689         if (rc)
2690                 GOTO(out, rc);
2691
2692         /* take LinkEA */
2693         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2694         if (!buf->lb_buf)
2695                 GOTO(out, rc = -ENOMEM);
2696
2697         ldata.ld_buf = buf;
2698         rc = mdt_links_read(info, obj, &ldata);
2699         if (rc)
2700                 GOTO(out, rc);
2701
2702         leh = buf->lb_buf;
2703         lee = (struct link_ea_entry *)(leh + 1);
2704         for (count = 0; count < leh->leh_reccount; count++) {
2705                 /* remove every hardlink */
2706                 linkea_entry_unpack(lee, &reclen, name, pfid);
2707                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2708                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2709                 if (rc)
2710                         break;
2711         }
2712
2713 out:
2714         if (obj && !IS_ERR(obj))
2715                 mdt_object_put(info->mti_env, obj);
2716         if (info->mti_big_buf.lb_buf)
2717                 lu_buf_free(&info->mti_big_buf);
2718
2719         RETURN(rc);
2720 }
2721
2722 static int mdt_rmfid(struct tgt_session_info *tsi)
2723 {
2724         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2725         struct mdt_body *reqbody;
2726         struct lu_fid *fids, *rfids;
2727         int bufsize, rc;
2728         __u32 *rcs;
2729         int i, nr;
2730
2731         ENTRY;
2732
2733         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2734         if (reqbody == NULL)
2735                 RETURN(-EPROTO);
2736         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2737                                        RCL_CLIENT);
2738         nr = bufsize / sizeof(struct lu_fid);
2739         if (nr * sizeof(struct lu_fid) != bufsize)
2740                 RETURN(-EINVAL);
2741         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2742                              RCL_SERVER, nr * sizeof(__u32));
2743         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2744                              RCL_SERVER, nr * sizeof(struct lu_fid));
2745         rc = req_capsule_server_pack(tsi->tsi_pill);
2746         if (rc)
2747                 GOTO(out, rc = err_serious(rc));
2748         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2749         if (fids == NULL)
2750                 RETURN(-EPROTO);
2751         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2752         LASSERT(rcs);
2753         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2754         LASSERT(rfids);
2755
2756         mdt_init_ucred(mti, reqbody);
2757         for (i = 0; i < nr; i++) {
2758                 rfids[i] = fids[i];
2759                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2760         }
2761         mdt_exit_ucred(mti);
2762
2763 out:
2764         RETURN(rc);
2765 }
2766
2767 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2768                          void *karg, void __user *uarg);
2769
2770 int mdt_io_set_info(struct tgt_session_info *tsi)
2771 {
2772         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2773         struct ost_body         *body = NULL, *repbody;
2774         void                    *key, *val = NULL;
2775         int                      keylen, vallen, rc = 0;
2776         bool                     is_grant_shrink;
2777
2778         ENTRY;
2779
2780         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2781         if (key == NULL) {
2782                 DEBUG_REQ(D_HA, req, "no set_info key");
2783                 RETURN(err_serious(-EFAULT));
2784         }
2785         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2786                                       RCL_CLIENT);
2787
2788         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2789         if (val == NULL) {
2790                 DEBUG_REQ(D_HA, req, "no set_info val");
2791                 RETURN(err_serious(-EFAULT));
2792         }
2793         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2794                                       RCL_CLIENT);
2795
2796         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2797         if (is_grant_shrink)
2798                 /* In this case the value is actually an RMF_OST_BODY, so we
2799                  * transmutate the type of this PTLRPC
2800                  */
2801                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2802
2803         rc = req_capsule_server_pack(tsi->tsi_pill);
2804         if (rc < 0)
2805                 RETURN(rc);
2806
2807         if (is_grant_shrink) {
2808                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2809
2810                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2811                 *repbody = *body;
2812
2813                 /** handle grant shrink, similar to a read request */
2814                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2815                                        &repbody->oa);
2816         } else {
2817                 CERROR("%s: Unsupported key %s\n",
2818                        tgt_name(tsi->tsi_tgt), (char *)key);
2819                 rc = -EOPNOTSUPP;
2820         }
2821
2822         RETURN(rc);
2823 }
2824
2825
2826 static int mdt_set_info(struct tgt_session_info *tsi)
2827 {
2828         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2829         char                    *key;
2830         void                    *val;
2831         int                      keylen, vallen, rc = 0;
2832
2833         ENTRY;
2834
2835         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2836         if (key == NULL) {
2837                 DEBUG_REQ(D_HA, req, "no set_info key");
2838                 RETURN(err_serious(-EFAULT));
2839         }
2840
2841         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2842                                       RCL_CLIENT);
2843
2844         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2845         if (val == NULL) {
2846                 DEBUG_REQ(D_HA, req, "no set_info val");
2847                 RETURN(err_serious(-EFAULT));
2848         }
2849
2850         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2851                                       RCL_CLIENT);
2852
2853         /* Swab any part of val you need to here */
2854         if (KEY_IS(KEY_READ_ONLY)) {
2855                 /* If client wants rw, make sure nodemap does not enforce ro. */
2856                 if (!*(__u32 *)val) {
2857                         struct lu_nodemap *nm = NULL;
2858                         bool readonly = false;
2859
2860                         if (req->rq_export)
2861                                 nm = nodemap_get_from_exp(req->rq_export);
2862
2863                         if (!IS_ERR_OR_NULL(nm)) {
2864                                 readonly = nm->nmf_readonly_mount;
2865                                 nodemap_putref(nm);
2866                         }
2867
2868                         if (unlikely(readonly))
2869                                 RETURN(-EROFS);
2870                 }
2871                 spin_lock(&req->rq_export->exp_lock);
2872                 if (*(__u32 *)val)
2873                         *exp_connect_flags_ptr(req->rq_export) |=
2874                                 OBD_CONNECT_RDONLY;
2875                 else
2876                         *exp_connect_flags_ptr(req->rq_export) &=
2877                                 ~OBD_CONNECT_RDONLY;
2878                 spin_unlock(&req->rq_export->exp_lock);
2879         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2880                 struct changelog_setinfo *cs = val;
2881
2882                 if (vallen != sizeof(*cs)) {
2883                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2884                                tgt_name(tsi->tsi_tgt), vallen);
2885                         RETURN(-EINVAL);
2886                 }
2887                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2888                         __swab64s(&cs->cs_recno);
2889                         __swab32s(&cs->cs_id);
2890                 }
2891
2892                 if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
2893                         RETURN(-EACCES);
2894                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2895                                    vallen, val, NULL);
2896         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2897                 if (vallen > 0)
2898                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2899         } else {
2900                 RETURN(-EINVAL);
2901         }
2902         RETURN(rc);
2903 }
2904
2905 static int mdt_readpage(struct tgt_session_info *tsi)
2906 {
2907         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2908         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2909         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2910         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2911         struct mdt_body         *repbody;
2912         int                      rc;
2913         int                      i;
2914
2915         ENTRY;
2916
2917         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2918                 RETURN(err_serious(-ENOMEM));
2919
2920         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2921         if (repbody == NULL || reqbody == NULL)
2922                 RETURN(err_serious(-EFAULT));
2923
2924         /*
2925          * prepare @rdpg before calling lower layers and transfer itself. Here
2926          * reqbody->size contains offset of where to start to read and
2927          * reqbody->nlink contains number bytes to read.
2928          */
2929         rdpg->rp_hash = reqbody->mbo_size;
2930         if (rdpg->rp_hash != reqbody->mbo_size) {
2931                 CERROR("Invalid hash: %#llx != %#llx\n",
2932                        rdpg->rp_hash, reqbody->mbo_size);
2933                 RETURN(-EFAULT);
2934         }
2935
2936         rdpg->rp_attrs = reqbody->mbo_mode;
2937         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2938                 rdpg->rp_attrs |= LUDA_64BITHASH;
2939         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2940                                 exp_max_brw_size(tsi->tsi_exp));
2941         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2942                           PAGE_SHIFT;
2943         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2944         if (rdpg->rp_pages == NULL)
2945                 RETURN(-ENOMEM);
2946
2947         for (i = 0; i < rdpg->rp_npages; ++i) {
2948                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2949                 if (rdpg->rp_pages[i] == NULL)
2950                         GOTO(free_rdpg, rc = -ENOMEM);
2951         }
2952
2953         /* call lower layers to fill allocated pages with directory data */
2954         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2955         if (rc < 0)
2956                 GOTO(free_rdpg, rc);
2957
2958         /* send pages to client */
2959         rc = tgt_sendpage(tsi, rdpg, rc);
2960
2961         EXIT;
2962 free_rdpg:
2963
2964         for (i = 0; i < rdpg->rp_npages; i++)
2965                 if (rdpg->rp_pages[i] != NULL)
2966                         __free_page(rdpg->rp_pages[i]);
2967         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2968
2969         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2970                 RETURN(0);
2971
2972         return rc;
2973 }
2974
2975 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2976 {
2977         struct lu_ucred *uc = mdt_ucred_check(info);
2978         struct lu_attr *attr = &info->mti_attr.ma_attr;
2979
2980         if (uc == NULL)
2981                 return -EINVAL;
2982
2983         if (op != REINT_SETATTR) {
2984                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2985                         attr->la_uid = uc->uc_fsuid;
2986                 /* for S_ISGID, inherit gid from his parent, such work will be
2987                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid.
2988                  */
2989                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2990                         attr->la_gid = uc->uc_fsgid;
2991         }
2992
2993         return 0;
2994 }
2995
2996 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2997 {
2998         return op == REINT_OPEN &&
2999              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
3000 }
3001
3002 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
3003 {
3004         struct req_capsule *pill = info->mti_pill;
3005
3006         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
3007                                   RCL_SERVER) &&
3008             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
3009                                   RCL_CLIENT)) {
3010                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
3011                                          RCL_CLIENT) != 0)
3012                         /* pre-set size in server part with max size */
3013                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
3014                                              RCL_SERVER,
3015                                              req_capsule_ptlreq(pill) ?
3016                                              OBD_MAX_DEFAULT_EA_SIZE :
3017                                              MAX_MD_SIZE_OLD);
3018                 else
3019                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
3020                                              RCL_SERVER, 0);
3021         }
3022 }
3023
3024 static int mdt_reint_internal(struct mdt_thread_info *info,
3025                               struct mdt_lock_handle *lhc,
3026                               __u32 op)
3027 {
3028         struct req_capsule      *pill = info->mti_pill;
3029         struct mdt_body         *repbody;
3030         int                      rc = 0, rc2;
3031
3032         ENTRY;
3033
3034         rc = mdt_reint_unpack(info, op);
3035         if (rc != 0) {
3036                 CERROR("Can't unpack reint, rc %d\n", rc);
3037                 RETURN(err_serious(rc));
3038         }
3039
3040
3041         /* check if the file system is set to readonly. O_RDONLY open
3042          * is still allowed even the file system is set to readonly mode
3043          */
3044         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
3045                 RETURN(err_serious(-EROFS));
3046
3047         /* for replay (no_create) lmm is not needed, client has it already */
3048         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3049                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3050                                      DEF_REP_MD_SIZE);
3051
3052         /* llog cookies are always 0, the field is kept for compatibility */
3053         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3054                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
3055
3056         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3057          * by default. If the target object has more ACL entries, then
3058          * enlarge the buffer when necessary.
3059          */
3060         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3061                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3062                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3063
3064         mdt_preset_secctx_size(info);
3065         mdt_preset_encctx_size(info);
3066
3067         rc = req_capsule_server_pack(pill);
3068         if (rc != 0) {
3069                 CERROR("Can't pack response, rc %d\n", rc);
3070                 RETURN(err_serious(rc));
3071         }
3072
3073         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
3074                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
3075                 LASSERT(repbody);
3076                 repbody->mbo_eadatasize = 0;
3077                 repbody->mbo_aclsize = 0;
3078         }
3079
3080         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
3081
3082         /* for replay no cookkie / lmm need, because client have this already */
3083         if (info->mti_spec.no_create)
3084                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3085                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
3086
3087         rc = mdt_init_ucred_reint(info);
3088         if (rc)
3089                 GOTO(out_shrink, rc);
3090
3091         rc = mdt_fix_attr_ucred(info, op);
3092         if (rc != 0)
3093                 GOTO(out_ucred, rc = err_serious(rc));
3094
3095         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
3096         if (rc < 0) {
3097                 GOTO(out_ucred, rc);
3098         } else if (rc == 1) {
3099                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
3100                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
3101                 GOTO(out_ucred, rc);
3102         }
3103         rc = mdt_reint_rec(info, lhc);
3104         EXIT;
3105 out_ucred:
3106         mdt_exit_ucred(info);
3107 out_shrink:
3108         mdt_client_compatibility(info);
3109
3110         rc2 = mdt_fix_reply(info);
3111         if (rc == 0)
3112                 rc = rc2;
3113
3114         /*
3115          * Data-on-MDT optimization - read data along with OPEN and return it
3116          * in reply when possible.
3117          */
3118         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
3119                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
3120                                           &lhc->mlh_reg_lh);
3121
3122         return rc;
3123 }
3124
3125 static long mdt_reint_opcode(struct ptlrpc_request *req,
3126                              const struct req_format **fmt)
3127 {
3128         struct mdt_device       *mdt;
3129         struct mdt_rec_reint    *rec;
3130         long                     opc;
3131
3132         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3133         if (rec != NULL) {
3134                 opc = rec->rr_opcode;
3135                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3136                 if (opc < REINT_MAX && fmt[opc] != NULL)
3137                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3138                 else {
3139                         mdt = mdt_exp2dev(req->rq_export);
3140                         CERROR("%s: Unsupported opcode '%ld' from client '%s': rc = %d\n",
3141                                req->rq_export->exp_obd->obd_name,
3142                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3143                         opc = err_serious(-EFAULT);
3144                 }
3145         } else {
3146                 opc = err_serious(-EFAULT);
3147         }
3148         return opc;
3149 }
3150
3151 static int mdt_reint(struct tgt_session_info *tsi)
3152 {
3153         long opc;
3154         int  rc;
3155         static const struct req_format *reint_fmts[REINT_MAX] = {
3156                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3157                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3158                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3159                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3160                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3161                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3162                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3163                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3164                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3165                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3166         };
3167
3168         ENTRY;
3169
3170         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3171         if (opc >= 0) {
3172                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3173                 /*
3174                  * No lock possible here from client to pass it to reint code
3175                  * path.
3176                  */
3177                 rc = mdt_reint_internal(info, NULL, opc);
3178                 mdt_thread_info_fini(info);
3179         } else {
3180                 rc = opc;
3181         }
3182
3183         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3184         RETURN(rc);
3185 }
3186
3187 /* this should sync the whole device */
3188 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3189 {
3190         struct dt_device *dt = mdt->mdt_bottom;
3191         int rc;
3192
3193         ENTRY;
3194
3195         rc = dt->dd_ops->dt_sync(env, dt);
3196         RETURN(rc);
3197 }
3198
3199 /* this should sync this object */
3200 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3201                            struct mdt_object *mo)
3202 {
3203         int rc = 0;
3204
3205         ENTRY;
3206
3207         if (!mdt_object_exists(mo)) {
3208                 CWARN("%s: non existing object "DFID": rc = %d\n",
3209                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3210                       -ESTALE);
3211                 RETURN(-ESTALE);
3212         }
3213
3214         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3215                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3216                 dt_obj_version_t version;
3217
3218                 version = dt_version_get(env, mdt_obj2dt(mo));
3219                 if (version > tgt->lut_obd->obd_last_committed)
3220                         rc = mo_object_sync(env, mdt_object_child(mo));
3221         } else {
3222                 rc = mo_object_sync(env, mdt_object_child(mo));
3223         }
3224
3225         RETURN(rc);
3226 }
3227
3228 static int mdt_sync(struct tgt_session_info *tsi)
3229 {
3230         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3231         struct req_capsule      *pill = tsi->tsi_pill;
3232         struct mdt_body         *body;
3233         ktime_t                  kstart = ktime_get();
3234         int                      rc;
3235
3236         ENTRY;
3237
3238         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3239                 RETURN(err_serious(-ENOMEM));
3240
3241         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3242                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3243         } else {
3244                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3245
3246                 if (unlikely(info->mti_object == NULL))
3247                         RETURN(-EPROTO);
3248
3249                 /* sync an object */
3250                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3251                                      info->mti_object);
3252                 if (rc == 0) {
3253                         const struct lu_fid *fid;
3254                         struct lu_attr *la = &info->mti_attr.ma_attr;
3255
3256                         info->mti_attr.ma_need = MA_INODE;
3257                         info->mti_attr.ma_valid = 0;
3258                         rc = mdt_attr_get_complex(info, info->mti_object,
3259                                                   &info->mti_attr);
3260                         if (rc == 0) {
3261                                 body = req_capsule_server_get(pill,
3262                                                               &RMF_MDT_BODY);
3263                                 fid = mdt_object_fid(info->mti_object);
3264                                 mdt_pack_attr2body(info, body, la, fid);
3265                         }
3266                 }
3267                 mdt_thread_info_fini(info);
3268         }
3269         if (rc == 0)
3270                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3271                                  ktime_us_delta(ktime_get(), kstart));
3272
3273         RETURN(rc);
3274 }
3275
3276 static int mdt_data_sync(struct tgt_session_info *tsi)
3277 {
3278         struct mdt_thread_info *info;
3279         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3280         struct ost_body *body = tsi->tsi_ost_body;
3281         struct ost_body *repbody;
3282         struct mdt_object *mo = NULL;
3283         struct md_attr *ma;
3284         int rc = 0;
3285
3286         ENTRY;
3287
3288         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3289
3290          /* device sync is done via MDS_SYNC. NOOP if no fid is specified */
3291         if (fid_is_zero(&tsi->tsi_fid))
3292                 RETURN(0);
3293
3294         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3295         if (IS_ERR(mo))
3296                 RETURN(PTR_ERR(mo));
3297
3298         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3299         if (rc)
3300                 GOTO(put, rc);
3301
3302         repbody->oa.o_oi = body->oa.o_oi;
3303         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3304
3305         info = tsi2mdt_info(tsi);
3306         ma = &info->mti_attr;
3307         ma->ma_need = MA_INODE;
3308         ma->ma_valid = 0;
3309         rc = mdt_attr_get_complex(info, mo, ma);
3310         if (rc == 0)
3311                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3312         else
3313                 rc = 0;
3314         mdt_thread_info_fini(info);
3315
3316         EXIT;
3317 put:
3318         if (mo != NULL)
3319                 mdt_object_put(tsi->tsi_env, mo);
3320         return rc;
3321 }
3322
3323 /*
3324  * Handle quota control requests to consult current usage/limit, but also
3325  * to configure quota enforcement
3326  */
3327 static int mdt_quotactl(struct tgt_session_info *tsi)
3328 {
3329         struct obd_export *exp  = tsi->tsi_exp;
3330         struct req_capsule *pill = tsi->tsi_pill;
3331         struct obd_quotactl *oqctl, *repoqc;
3332         struct mdt_device *mdt = mdt_exp2dev(exp);
3333         struct lu_device *qmt = mdt->mdt_qmt_dev;
3334         struct lu_nodemap *nodemap;
3335         char *buffer = NULL;
3336         int id, rc;
3337
3338         ENTRY;
3339
3340         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3341         if (!oqctl)
3342                 RETURN(err_serious(-EPROTO));
3343
3344         if (oqctl->qc_cmd == LUSTRE_Q_ITERQUOTA ||
3345             oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA)
3346                 req_capsule_set_size(pill, &RMF_OBD_QUOTA_ITER, RCL_SERVER,
3347                                      LQUOTA_ITER_BUFLEN);
3348         else
3349                 req_capsule_set_size(pill, &RMF_OBD_QUOTA_ITER, RCL_SERVER, 0);
3350
3351         rc = req_capsule_server_pack(pill);
3352         if (rc)
3353                 RETURN(err_serious(rc));
3354
3355         nodemap = nodemap_get_from_exp(exp);
3356         if (IS_ERR(nodemap))
3357                 RETURN(PTR_ERR(nodemap));
3358
3359         switch (oqctl->qc_cmd) {
3360                 /* master quotactl */
3361         case Q_SETINFO:
3362         case Q_SETQUOTA:
3363         case LUSTRE_Q_SETDEFAULT:
3364         case LUSTRE_Q_SETQUOTAPOOL:
3365         case LUSTRE_Q_SETINFOPOOL:
3366         case LUSTRE_Q_SETDEFAULT_POOL:
3367         case LUSTRE_Q_DELETEQID:
3368         case LUSTRE_Q_RESETQID:
3369                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3370                                           oqctl->qc_id))
3371                         GOTO(out_nodemap, rc = -EPERM);
3372                 fallthrough;
3373         case Q_GETINFO:
3374         case Q_GETQUOTA:
3375         case LUSTRE_Q_GETDEFAULT:
3376         case LUSTRE_Q_GETQUOTAPOOL:
3377         case LUSTRE_Q_GETINFOPOOL:
3378         case LUSTRE_Q_GETDEFAULT_POOL:
3379         case LUSTRE_Q_ITERQUOTA:
3380                 if (qmt == NULL)
3381                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3382                 /* slave quotactl */
3383                 fallthrough;
3384         case Q_GETOINFO:
3385         case Q_GETOQUOTA:
3386         case LUSTRE_Q_ITEROQUOTA:
3387                 break;
3388         default:
3389                 rc = -EFAULT;
3390                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3391                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3392                 GOTO(out_nodemap, rc);
3393         }
3394
3395         id = oqctl->qc_id;
3396         switch (oqctl->qc_type) {
3397         case USRQUOTA:
3398                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3399                                     NODEMAP_CLIENT_TO_FS, id);
3400                 break;
3401         case GRPQUOTA:
3402                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3403                                     NODEMAP_CLIENT_TO_FS, id);
3404                 break;
3405         case PRJQUOTA:
3406                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3407                                     NODEMAP_CLIENT_TO_FS, id);
3408                 break;
3409         default:
3410                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3411         }
3412         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3413         if (repoqc == NULL)
3414                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3415
3416         if (oqctl->qc_cmd == LUSTRE_Q_ITERQUOTA ||
3417             oqctl->qc_cmd == LUSTRE_Q_ITEROQUOTA) {
3418                 buffer = req_capsule_server_get(pill, &RMF_OBD_QUOTA_ITER);
3419                 if (buffer == NULL)
3420                         GOTO(out_nodemap, rc = err_serious(-EFAULT));
3421         }
3422
3423         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3424                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3425
3426         if (oqctl->qc_id != id)
3427                 swap(oqctl->qc_id, id);
3428
3429         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3430                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3431                         RETURN(-EINPROGRESS);
3432         }
3433
3434         switch (oqctl->qc_cmd) {
3435
3436         case Q_GETINFO:
3437         case Q_SETINFO:
3438         case Q_SETQUOTA:
3439         case Q_GETQUOTA:
3440         case LUSTRE_Q_SETDEFAULT:
3441         case LUSTRE_Q_GETDEFAULT:
3442         case LUSTRE_Q_SETQUOTAPOOL:
3443         case LUSTRE_Q_GETQUOTAPOOL:
3444         case LUSTRE_Q_SETINFOPOOL:
3445         case LUSTRE_Q_GETINFOPOOL:
3446         case LUSTRE_Q_SETDEFAULT_POOL:
3447         case LUSTRE_Q_GETDEFAULT_POOL:
3448         case LUSTRE_Q_DELETEQID:
3449         case LUSTRE_Q_RESETQID:
3450         case LUSTRE_Q_ITERQUOTA:
3451                 /* forward quotactl request to QMT */
3452                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl, buffer,
3453                                             buffer == NULL ? 0 :
3454                                                         LQUOTA_ITER_BUFLEN);
3455                 break;
3456
3457         case Q_GETOINFO:
3458         case Q_GETOQUOTA:
3459         case LUSTRE_Q_ITEROQUOTA:
3460                 /* slave quotactl */
3461                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3462                                    oqctl, buffer,
3463                                    buffer == NULL ? 0 : LQUOTA_ITER_BUFLEN);
3464                 break;
3465
3466         default:
3467                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3468                 GOTO(out_nodemap, rc = -EFAULT);
3469         }
3470
3471         if (oqctl->qc_id != id)
3472                 swap(oqctl->qc_id, id);
3473
3474         QCTL_COPY_NO_PNAME(repoqc, oqctl);
3475         EXIT;
3476
3477 out_nodemap:
3478         nodemap_putref(nodemap);
3479
3480         return rc;
3481 }
3482
3483 /** clone llog ctxt from child (mdd)
3484  * This allows remote llog (replicator) access.
3485  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3486  * context was originally set up, or we can handle them directly.
3487  * I choose the latter, but that means I need any llog
3488  * contexts set up by child to be accessable by the mdt.  So we clone the
3489  * context into our context list here.
3490  */
3491 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3492                                int idx)
3493 {
3494         struct md_device  *next = mdt->mdt_child;
3495         struct llog_ctxt *ctxt;
3496         int rc;
3497
3498         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3499                 return 0;
3500
3501         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3502         if (rc || ctxt == NULL)
3503                 return 0;
3504
3505         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3506         if (rc)
3507                 CERROR("Can't set mdt ctxt %d\n", rc);
3508
3509         return rc;
3510 }
3511
3512 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3513                                  struct mdt_device *mdt, int idx)
3514 {
3515         struct llog_ctxt *ctxt;
3516
3517         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3518         if (ctxt == NULL)
3519                 return 0;
3520         /* Put once for the get we just did, and once for the clone */
3521         llog_ctxt_put(ctxt);
3522         llog_ctxt_put(ctxt);
3523         return 0;
3524 }
3525
3526 /*
3527  * sec context handlers
3528  */
3529 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3530 {
3531         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3532
3533         return 0;
3534 }
3535
3536 /*
3537  * quota request handlers
3538  */
3539 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3540 {
3541         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3542         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3543         int                      rc;
3544
3545         ENTRY;
3546
3547         if (qmt == NULL)
3548                 RETURN(err_serious(-EOPNOTSUPP));
3549
3550         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3551         RETURN(rc);
3552 }
3553
3554 struct mdt_object *mdt_object_new(const struct lu_env *env,
3555                                   struct mdt_device *d,
3556                                   const struct lu_fid *f)
3557 {
3558         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3559         struct lu_object *o;
3560         struct mdt_object *m;
3561
3562         ENTRY;
3563
3564         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3565         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3566         if (unlikely(IS_ERR(o)))
3567                 m = (struct mdt_object *)o;
3568         else
3569                 m = mdt_obj(o);
3570         RETURN(m);
3571 }
3572
3573 struct mdt_object *mdt_object_find(const struct lu_env *env,
3574                                    struct mdt_device *d,
3575                                    const struct lu_fid *f)
3576 {
3577         struct lu_object *o;
3578         struct mdt_object *m;
3579
3580         ENTRY;
3581
3582         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3583         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3584         if (unlikely(IS_ERR(o)))
3585                 m = (struct mdt_object *)o;
3586         else
3587                 m = mdt_obj(o);
3588
3589         RETURN(m);
3590 }
3591
3592 /**
3593  * Asyncronous commit for mdt device.
3594  *
3595  * Pass asynchonous commit call down the MDS stack.
3596  *
3597  * \param env environment
3598  * \param mdt the mdt device
3599  */
3600 static void mdt_device_commit_async(const struct lu_env *env,
3601                                     struct mdt_device *mdt)
3602 {
3603         struct dt_device *dt = mdt->mdt_bottom;
3604         int rc;
3605
3606         ENTRY;
3607
3608         rc = dt->dd_ops->dt_commit_async(env, dt);
3609         if (unlikely(rc != 0))
3610                 CWARN("%s: async commit start failed: rc = %d\n",
3611                       mdt_obd_name(mdt), rc);
3612         atomic_inc(&mdt->mdt_async_commit_count);
3613         EXIT;
3614 }
3615
3616 /**
3617  * Mark the lock as "synchonous".
3618  *
3619  * Mark the lock to deffer transaction commit to the unlock time.
3620  *
3621  * \param lock the lock to mark as "synchonous"
3622  *
3623  * \see mdt_is_lock_sync
3624  * \see mdt_save_lock
3625  */
3626 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3627 {
3628         lock->l_ast_data = (void *)1;
3629 }
3630
3631 /**
3632  * Check whehter the lock "synchonous" or not.
3633  *
3634  * \param lock the lock to check
3635  * \retval 1 the lock is "synchonous"
3636  * \retval 0 the lock isn't "synchronous"
3637  *
3638  * \see mdt_set_lock_sync
3639  * \see mdt_save_lock
3640  */
3641 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3642 {
3643         return lock->l_ast_data != NULL;
3644 }
3645
3646 /**
3647  * Blocking AST for mdt locks.
3648  *
3649  * Starts transaction commit if in case of COS lock conflict or
3650  * deffers such a commit to the mdt_save_lock.
3651  *
3652  * \param lock the lock which blocks a request or cancelling lock
3653  * \param desc unused
3654  * \param data unused
3655  * \param flag indicates whether this cancelling or blocking callback
3656  * \retval 0
3657  * \see ldlm_blocking_ast_nocheck
3658  */
3659 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3660                      void *data, int flag)
3661 {
3662         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3663         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3664         struct ldlm_cb_set_arg *arg = data;
3665         bool commit_async = false;
3666         int rc;
3667
3668         ENTRY;
3669         if (flag == LDLM_CB_CANCELING)
3670                 RETURN(0);
3671
3672         lock_res_and_lock(lock);
3673         if (lock->l_blocking_ast != mdt_blocking_ast) {
3674                 unlock_res_and_lock(lock);
3675                 RETURN(0);
3676         }
3677
3678         /* A blocking ast may be sent from ldlm_lock_decref_internal
3679          * when the last reference to a local lock was released and
3680          * during blocking event from ldlm_work_bl_ast_lock().
3681          * The 'data' parameter is l_ast_data in the first case and
3682          * callback arguments in the second one. Distinguish them by that.
3683          */
3684         if (data && data != lock->l_ast_data && arg->bl_desc) {
3685                 if (lock->l_req_mode & (LCK_COS | LCK_TXN))
3686                         commit_async = true;
3687                 else if ((lock->l_req_mode & (LCK_PW | LCK_EX)) &&
3688                          ((mdt_cos_is_enabled(mdt) &&
3689                              !arg->bl_desc->bl_same_client) ||
3690                           (mdt_slc_is_enabled(mdt) &&
3691                              arg->bl_desc->bl_txn_dependent)))
3692                         mdt_set_lock_sync(lock);
3693         }
3694
3695         rc = ldlm_blocking_ast_nocheck(lock);
3696
3697         if (commit_async) {
3698                 struct lu_env env;
3699
3700                 rc = lu_env_init(&env, LCT_LOCAL);
3701                 if (unlikely(rc != 0))
3702                         CWARN("%s: lu_env initialization failed, cannot start asynchronous commit: rc = %d\n",
3703                               obd->obd_name, rc);
3704                 else
3705                         mdt_device_commit_async(&env, mdt);
3706                 lu_env_fini(&env);
3707         }
3708         RETURN(rc);
3709 }
3710
3711 /*
3712  * Blocking AST for cross-MDT lock
3713  *
3714  * Discard lock from uncommitted_slc_locks and cancel it.
3715  *
3716  * \param lock  the lock which blocks a request or cancelling lock
3717  * \param desc  unused
3718  * \param data  unused
3719  * \param flag  indicates whether this cancelling or blocking callback
3720  * \retval      0 on success
3721  * \retval      negative number on error
3722  */
3723 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3724                             void *data, int flag)
3725 {
3726         int rc = 0;
3727
3728         ENTRY;
3729
3730         switch (flag) {
3731         case LDLM_CB_BLOCKING: {
3732                 struct lustre_handle lockh;
3733
3734                 ldlm_lock2handle(lock, &lockh);
3735                 rc = ldlm_cli_cancel(&lockh,
3736                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3737                 if (rc < 0) {
3738                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3739                         RETURN(rc);
3740                 }
3741                 break;
3742         }
3743         case LDLM_CB_CANCELING: {
3744                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3745                 struct mdt_device *mdt =
3746                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3747
3748                 LDLM_DEBUG(lock, "Revoke remote lock");
3749
3750                 /* discard slc lock here so that it can be cleaned anytime,
3751                  * especially for cleanup_resource()
3752                  */
3753                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3754
3755                 /* once we cache lock, l_ast_data is set to mdt_object */
3756                 if (lock->l_ast_data != NULL) {
3757                         struct mdt_object *mo = lock->l_ast_data;
3758                         struct lu_env env;
3759
3760                         rc = lu_env_init(&env, LCT_MD_THREAD);
3761                         if (unlikely(rc != 0)) {
3762                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3763                                       obd->obd_name, mo,
3764                                       PFID(mdt_object_fid(mo)), rc);
3765                                 RETURN(rc);
3766                         }
3767
3768                         if (lock->l_policy_data.l_inodebits.bits &
3769                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3770                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3771                                 mo->mot_cache_attr = 0;
3772                         }
3773                         mdt_object_put(&env, mo);
3774                         lu_env_fini(&env);
3775                 }
3776                 break;
3777         }
3778         default:
3779                 LBUG();
3780         }
3781
3782         RETURN(rc);
3783 }
3784
3785 int mdt_check_resent_lock(struct mdt_thread_info *info,
3786                           struct mdt_object *mo,
3787                           struct mdt_lock_handle *lhc)
3788 {
3789         /* the lock might already be gotten in ldlm_handle_enqueue() */
3790         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3791                 struct ptlrpc_request *req = mdt_info_req(info);
3792                 struct ldlm_lock      *lock;
3793
3794                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3795                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3796                 if (lock == NULL) {
3797                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3798                          * a resend case, however, it could be already destroyed
3799                          * due to client eviction or a raced cancel RPC.
3800                          */
3801                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3802                                           lhc->mlh_reg_lh.cookie);
3803                         RETURN(-ESTALE);
3804                 }
3805
3806                 if (!fid_res_name_eq(mdt_object_fid(mo),
3807                                      &lock->l_resource->lr_name)) {
3808                         CWARN("%s: Although resent, but still not get child lock:"
3809                               DFID"\n",
3810                               info->mti_exp->exp_obd->obd_name,
3811                               PFID(mdt_object_fid(mo)));
3812                         LDLM_LOCK_PUT(lock);
3813                         RETURN(-EPROTO);
3814                 }
3815                 LDLM_LOCK_PUT(lock);
3816                 return 0;
3817         }
3818         return 1;
3819 }
3820
3821 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3822 {
3823         mdt_object_get(NULL, lock->l_ast_data);
3824 }
3825
3826 static int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3827                                       struct mdt_object *obj,
3828                                       struct lustre_handle *lh,
3829                                       enum ldlm_mode mode,
3830                                       union ldlm_policy_data *policy,
3831                                       struct ldlm_res_id *res_id,
3832                                       bool cache)
3833 {
3834         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3835         int rc;
3836
3837         LASSERT(mdt_object_remote(obj));
3838
3839         memset(einfo, 0, sizeof(*einfo));
3840         einfo->ei_type = LDLM_IBITS;
3841         einfo->ei_mode = mode;
3842         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3843         einfo->ei_cb_cp = ldlm_completion_ast;
3844         einfo->ei_enq_slave = 0;
3845         einfo->ei_res_id = res_id;
3846         einfo->ei_req_slot = 1;
3847         if (cache) {
3848                 /*
3849                  * if we cache lock, couple lock with mdt_object, so that object
3850                  * can be easily found in lock ASTs.
3851                  */
3852                 einfo->ei_cbdata = obj;
3853                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3854         }
3855
3856         rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), lh, einfo,
3857                             policy);
3858         if (rc) {
3859                 lh->cookie = 0ull;
3860                 return rc;
3861         }
3862
3863         /* other components like LFSCK can use lockless access
3864          * and populate cache, so we better invalidate it
3865          */
3866         if (policy->l_inodebits.bits &
3867             (MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR))
3868                 mo_invalidate(mti->mti_env, mdt_object_child(obj));
3869
3870         return 0;
3871 }
3872
3873 /*
3874  * Helper function to take PDO and hash lock.
3875  *
3876  * if \a pdo_lock is false, don't take PDO lock, this is case in rename.
3877  */
3878 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3879                         struct mdt_lock_handle *lh, const struct lu_name *name,
3880                         enum ldlm_mode mode, bool pdo_lock)
3881 {
3882         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3883         union ldlm_policy_data *policy = &info->mti_policy;
3884         struct ldlm_res_id *res_id = &info->mti_res_id;
3885         /*
3886          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it is never going to
3887          * be sent to client and we do not want it slowed down due to possible
3888          * cancels.
3889          */
3890         __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3891         __u64 *cookie = NULL;
3892         int rc;
3893
3894         LASSERT(obj);
3895
3896         /* check for exists after object is locked */
3897         if (!mdt_object_exists(obj))
3898                 /* Non-existent object shouldn't have PDO lock */
3899                 return -ESTALE;
3900
3901         /* Non-dir object shouldn't have PDO lock */
3902         if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
3903                 return -ENOTDIR;
3904
3905         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
3906         policy->l_inodebits.try_bits = 0;
3907         policy->l_inodebits.li_gid = 0;
3908         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3909         fid_build_reg_res_name(mdt_object_fid(obj), res_id);
3910         if (info->mti_exp)
3911                 cookie = &info->mti_exp->exp_handle.h_cookie;
3912
3913         mdt_lock_pdo_init(lh, mode, name);
3914         mdt_lock_pdo_mode(info, obj, lh);
3915         if (lh->mlh_pdo_mode != LCK_NL) {
3916                 if (pdo_lock) {
3917                         if (mdt_object_remote(obj)) {
3918                                 rc = mdt_remote_object_lock_try(info, obj,
3919                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3920                                         policy, res_id, false);
3921                                 lh->mlh_pdo_remote = 1;
3922                         } else {
3923                                 rc = mdt_fid_lock(info->mti_env, ns,
3924                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3925                                         policy, res_id, dlmflags, cookie);
3926                         }
3927                         if (rc) {
3928                                 mdt_object_unlock(info, obj, lh, 1);
3929                                 return rc;
3930                         }
3931                 }
3932                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3933         }
3934
3935         if (mdt_object_remote(obj))
3936                 rc = mdt_remote_object_lock_try(info, obj, &lh->mlh_rreg_lh,
3937                         lh->mlh_rreg_mode, policy, res_id, false);
3938         else
3939                 rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh,
3940                         lh->mlh_reg_mode, policy, res_id, dlmflags, cookie);
3941         if (rc)
3942                 mdt_object_unlock(info, obj, lh, 1);
3943         else if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK) &&
3944                    lh->mlh_pdo_hash != 0 &&
3945                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3946                 CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3947
3948         return rc;
3949 }
3950
3951 int mdt_object_lock_internal(struct mdt_thread_info *info,
3952                              struct mdt_object *obj, const struct lu_fid *fid,
3953                              struct mdt_lock_handle *lh, __u64 *ibits,
3954                              __u64 trybits, bool cache)
3955 {
3956         union ldlm_policy_data *policy = &info->mti_policy;
3957         struct ldlm_res_id *res_id = &info->mti_res_id;
3958         struct lustre_handle *handle;
3959         int rc;
3960
3961         policy->l_inodebits.bits = *ibits;
3962         policy->l_inodebits.try_bits = trybits;
3963         policy->l_inodebits.li_gid = lh->mlh_gid;
3964         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3965         fid_build_reg_res_name(fid, res_id);
3966
3967         if (obj && mdt_object_remote(obj)) {
3968                 handle = &lh->mlh_rreg_lh;
3969                 LASSERT(!lustre_handle_is_used(handle));
3970                 LASSERT(lh->mlh_rreg_mode != LCK_MINMODE);
3971                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3972                 rc = mdt_remote_object_lock_try(info, obj, handle,
3973                                                 lh->mlh_rreg_mode, policy,
3974                                                 res_id, cache);
3975         } else {
3976                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3977                 /*
3978                  * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if
3979                  * it is going to be sent to client. If it is -
3980                  * mdt_intent_policy() path will fix it up and turn FL_LOCAL
3981                  * flag off.
3982                  */
3983                 __u64 dlmflags = LDLM_FL_ATOMIC_CB | LDLM_FL_LOCAL_ONLY;
3984                 __u64 *cookie = NULL;
3985
3986                 handle = &lh->mlh_reg_lh;
3987                 LASSERT(!lustre_handle_is_used(handle));
3988                 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3989                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3990
3991                 /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3992                 if (lh->mlh_type == MDT_REG_LOCK &&
3993                     lh->mlh_reg_mode == LCK_EX && *ibits == MDS_INODELOCK_OPEN)
3994                         dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3995
3996
3997                 if (info->mti_exp)
3998                         cookie = &info->mti_exp->exp_handle.h_cookie;
3999
4000                 rc = mdt_fid_lock(info->mti_env, ns, handle, lh->mlh_reg_mode,
4001                                   policy, res_id, dlmflags, cookie);
4002                 if (rc)
4003                         mdt_object_unlock(info, obj, lh, 1);
4004         }
4005
4006         if (rc == 0) {
4007                 struct ldlm_lock *lock;
4008
4009                 /* Return successfully acquired bits to a caller */
4010                 lock = ldlm_handle2lock(handle);
4011                 LASSERT(lock);
4012                 *ibits = lock->l_policy_data.l_inodebits.bits;
4013                 LDLM_LOCK_PUT(lock);
4014         }
4015
4016         return rc;
4017 }
4018
4019 /*
4020  * MDT object locking functions:
4021  * mdt_object_lock(): lock object, this is used in most places, and normally
4022  *      lock ibits doesn't contain LOOKUP, unless the caller knows it's not
4023  *      remote object.
4024  * mdt_object_check_lock(): lock object with LOOKUP and other ibits, it needs
4025  *      to check whether parent is on remote MDT, if so, take LOOKUP on parent
4026  *      MDT separately, and then lock other ibits on child object.
4027  * mdt_parent_lock(): take parent UPDATE lock with specific mode, if parent is
4028  *      local, take PDO lock by name hash, otherwise take regular lock.
4029  * mdt_object_stripes_lock(): lock object which should be local, and if it's a
4030  *      striped directory, lock its stripes, this is called in operations which
4031  *      modify both object and stripes.
4032  * mdt_object_lock_try(): lock object with trybits, the trybits contains
4033  *      optional inode lock bits that can be granted. This is called by
4034  *      getattr/open to fetch more inode lock bits to client, and is also called
4035  *      by dir migration to lock link parent in non-block mode to avoid
4036  *      deadlock.
4037  */
4038
4039 /**
4040  * lock object
4041  *
4042  * this is used to lock object in most places, and normally lock ibits doesn't
4043  * contain LOOKUP, unless the caller knows it's not remote object.
4044  *
4045  * \param info          struct mdt_thread_info
4046  * \param obj           object
4047  * \param lh            lock handle
4048  * \param ibits         MDS inode lock bits
4049  * \param mode          lock mode
4050  *
4051  * \retval              0 on success, -ev on error.
4052  */
4053 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4054                     struct mdt_lock_handle *lh, __u64 ibits,
4055                     enum ldlm_mode mode)
4056 {
4057         int rc;
4058
4059         ENTRY;
4060         mdt_lock_reg_init(lh, mode);
4061         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh,
4062                                       &ibits, 0, false);
4063         RETURN(rc);
4064 }
4065
4066 /**
4067  * lock object with LOOKUP and other ibits
4068  *
4069  * it will check whether parent and child are on different MDTs, if so, take
4070  * LOOKUP lock on parent MDT, and lock other ibits on child MDT, otherwise lock
4071  * all ibits on child MDT. Note, parent and child shouldn't be both on remote
4072  * MDTs, in which case specific lock function should be used, and it's in
4073  * rename and migrate only.
4074  *
4075  * \param info          struct mdt_thread_info
4076  * \param parent        parent object
4077  * \param child         child object
4078  * \param lh            lock handle
4079  * \param ibits         MDS inode lock bits
4080  * \param mode          lock mode
4081  *
4082  * \retval              0 on success, -ev on error.
4083  */
4084 int mdt_object_check_lock(struct mdt_thread_info *info,
4085                           struct mdt_object *parent, struct mdt_object *child,
4086                           struct mdt_lock_handle *lh, __u64 ibits,
4087                           enum ldlm_mode mode)
4088 {
4089         int rc;
4090
4091         ENTRY;
4092         /* if LOOKUP ibit is not set, use mdt_object_lock() */
4093         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
4094         /* if only LOOKUP ibit is needed, use mdt_object_lookup_lock() */
4095         LASSERT(ibits != MDS_INODELOCK_LOOKUP);
4096         LASSERT(parent);
4097         /* @parent and @child shouldn't both be on remote MDTs */
4098         LASSERT(!(mdt_object_remote(parent) && mdt_object_remote(child)));
4099
4100         mdt_lock_reg_init(lh, mode);
4101         if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
4102                 __u64 lookup_ibits = MDS_INODELOCK_LOOKUP;
4103
4104                 rc = mdt_object_lock_internal(info, parent,
4105                                               mdt_object_fid(child), lh,
4106                                               &lookup_ibits, 0, false);
4107                 if (rc)
4108                         RETURN(rc);
4109
4110                 ibits &= ~MDS_INODELOCK_LOOKUP;
4111         }
4112
4113         rc = mdt_object_lock_internal(info, child, mdt_object_fid(child), lh,
4114                                       &ibits, 0, false);
4115         if (rc && !(ibits & MDS_INODELOCK_LOOKUP))
4116                 mdt_object_unlock(info, NULL, lh, 1);
4117
4118         RETURN(rc);
4119 }
4120
4121 /**
4122  * take parent UPDATE lock
4123  *
4124  * if parent is local or mode is LCK_PW, take PDO lock, otherwise take regular
4125  * lock.
4126  *
4127  * \param info  struct mdt_thread_info
4128  * \param obj   parent object
4129  * \param lh    lock handle
4130  * \param lname child name
4131  * \param mode  lock mode
4132  *
4133  * \retval      0 on success, -ev on error.
4134  */
4135 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4136                     struct mdt_lock_handle *lh, const struct lu_name *lname,
4137                     enum ldlm_mode mode)
4138 {
4139         int rc;
4140
4141         ENTRY;
4142         LASSERT(obj && lname);
4143         LASSERT(mode == LCK_PW || mode == LCK_PR);
4144         if (mdt_object_remote(obj) && mode == LCK_PR) {
4145                 __u64 ibits = MDS_INODELOCK_UPDATE;
4146
4147                 mdt_lock_reg_init(lh, mode);
4148                 rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj),
4149                                               lh, &ibits, 0, false);
4150         } else {
4151                 rc = mdt_object_pdo_lock(info, obj, lh, lname, mode, true);
4152         }
4153         RETURN(rc);
4154 }
4155
4156 /**
4157  * lock object with trybits
4158  *
4159  * the trybits contains optional inode lock bits that can be granted. This is
4160  * called by getattr/open to fetch more inode lock bits to client, and is also
4161  * called by dir migration to lock link parent in non-block mode to avoid
4162  * deadlock.
4163  *
4164  * \param info          struct mdt_thread_info
4165  * \param obj           object
4166  * \param lh            lock handle
4167  * \param ibits         MDS inode lock bits
4168  * \param trybits       optional inode lock bits
4169  * \param mode          lock mode
4170  *
4171  * \retval              0 on success, -ev on error.
4172  */
4173 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *obj,
4174                         struct mdt_lock_handle *lh, __u64 *ibits,
4175                         __u64 trybits, enum ldlm_mode mode)
4176 {
4177         bool trylock_only = *ibits == 0;
4178         int rc;
4179
4180         ENTRY;
4181         LASSERT(!(*ibits & trybits));
4182         mdt_lock_reg_init(lh, mode);
4183         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh, ibits,
4184                                       trybits, false);
4185         if (rc && trylock_only) { /* clear error for try ibits lock only */
4186                 LASSERT(*ibits == 0);
4187                 rc = 0;
4188         }
4189         RETURN(rc);
4190 }
4191
4192 /*
4193  * Helper function to take \a obj LOOKUP lock.
4194  *
4195  * Both \a pobj and \a obj may be located on remote MDTs.
4196  */
4197 int mdt_object_lookup_lock(struct mdt_thread_info *info,
4198                            struct mdt_object *pobj, struct mdt_object *obj,
4199                            struct mdt_lock_handle *lh, enum ldlm_mode mode)
4200 {
4201         __u64 ibits = MDS_INODELOCK_LOOKUP;
4202         int rc;
4203
4204         ENTRY;
4205         /* if @parent is NULL, it's on local MDT, and @child is remote,
4206          * this is case in getattr/unlink/open by name.
4207          */
4208         LASSERT(ergo(!pobj, mdt_object_remote(obj)));
4209         mdt_lock_reg_init(lh, mode);
4210         rc = mdt_object_lock_internal(info, pobj, mdt_object_fid(obj), lh,
4211                                       &ibits, 0, false);
4212         RETURN(rc);
4213 }
4214
4215 /**
4216  * Save a lock within request object.
4217  *
4218  * Keep the lock referenced until whether client ACK or transaction
4219  * commit happens or release the lock immediately depending on input
4220  * parameters. If COS is ON, a write lock is converted to COS lock
4221  * before saving.
4222  *
4223  * \param info thead info object
4224  * \param h lock handle
4225  * \param mode lock mode
4226  * \param decref force immediate lock releasing
4227  */
4228 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
4229                           enum ldlm_mode mode, int decref)
4230 {
4231         struct tgt_session_info *tsi = info->mti_env->le_ses ?
4232                                        tgt_ses_info(info->mti_env) : NULL;
4233
4234         ENTRY;
4235
4236         if (lustre_handle_is_used(h)) {
4237                 bool has_trans = tsi && tsi->tsi_has_trans;
4238
4239                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
4240                         mdt_fid_unlock(h, mode);
4241                 } else {
4242                         struct mdt_device *mdt = info->mti_mdt;
4243                         struct ldlm_lock *lock = ldlm_handle2lock(h);
4244                         struct ptlrpc_request *req = mdt_info_req(info);
4245                         bool no_ack = false;
4246
4247                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
4248                                  h->cookie);
4249
4250                         /* there is no request if mdt_object_unlock() is called
4251                          * from mdt_export_cleanup()->mdt_add_dirty_flag()
4252                          */
4253                         if (likely(req != NULL)) {
4254                                 LDLM_DEBUG(lock, "save lock request %p reply state %p transno %lld",
4255                                         req,
4256                                         req->rq_reply_state, req->rq_transno);
4257                                 if (mdt_cos_is_enabled(mdt)) {
4258                                         mode = LCK_COS;
4259                                         no_ack = true;
4260                                         ldlm_lock_mode_downgrade(lock, mode);
4261                                 } else if (mdt_slc_is_enabled(mdt)) {
4262                                         no_ack = true;
4263                                         if (mode != LCK_TXN) {
4264                                                 mode = LCK_TXN;
4265                                                 ldlm_lock_mode_downgrade(lock,
4266                                                                          mode);
4267                                         }
4268                                 }
4269                                 if (req->rq_export->exp_disconnected)
4270                                         mdt_fid_unlock(h, mode);
4271                                 else
4272                                         ptlrpc_save_lock(req, h, no_ack);
4273                         } else {
4274                                 mdt_fid_unlock(h, mode);
4275                         }
4276
4277                         if (mdt_is_lock_sync(lock)) {
4278                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4279                                 mdt_device_commit_async(info->mti_env, mdt);
4280                         }
4281                         LDLM_LOCK_PUT(lock);
4282                 }
4283                 h->cookie = 0ull;
4284         }
4285
4286         EXIT;
4287 }
4288
4289 /**
4290  * Save cross-MDT lock in uncommitted_slc_locks
4291  *
4292  * Keep the lock referenced until transaction commit happens or release the lock
4293  * immediately depending on input parameters.
4294  *
4295  * \param info thead info object
4296  * \param h lock handle
4297  * \param mode lock mode
4298  * \param decref force immediate lock releasing
4299  */
4300 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4301                                  struct mdt_object *o, struct lustre_handle *h,
4302                                  enum ldlm_mode mode, int decref)
4303 {
4304         ENTRY;
4305
4306         if (lustre_handle_is_used(h)) {
4307                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4308                 struct ptlrpc_request *req = mdt_info_req(info);
4309
4310                 if (o != NULL &&
4311                     (lock->l_policy_data.l_inodebits.bits &
4312                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4313                         mo_invalidate(info->mti_env, mdt_object_child(o));
4314
4315                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4316                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4317                         ldlm_lock_decref_and_cancel(h, mode);
4318                         LDLM_LOCK_PUT(lock);
4319                 } else {
4320                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4321                                           req->rq_transno);
4322                         ldlm_lock_decref(h, mode);
4323                 }
4324                 h->cookie = 0ull;
4325         }
4326
4327         EXIT;
4328 }
4329
4330 /**
4331  * Unlock mdt object.
4332  *
4333  * Immeditely release the regular lock and the PDO lock or save the
4334  * lock in request and keep them referenced until client ACK or
4335  * transaction commit.
4336  *
4337  * \param info thread info object
4338  * \param o mdt object
4339  * \param lh mdt lock handle referencing regular and PDO locks
4340  * \param decref force immediate lock releasing
4341  *
4342  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4343  */
4344 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4345                        struct mdt_lock_handle *lh, int decref)
4346 {
4347         ENTRY;
4348
4349         if (lh->mlh_pdo_remote)
4350                 mdt_save_remote_lock(info, o, &lh->mlh_pdo_lh,
4351                                      lh->mlh_pdo_mode, decref);
4352         else
4353                 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4354         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4355         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4356                              decref);
4357
4358         EXIT;
4359 }
4360
4361 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4362                                         const struct lu_fid *f,
4363                                         struct mdt_lock_handle *lh,
4364                                         __u64 ibits, enum ldlm_mode mode)
4365 {
4366         struct mdt_object *o;
4367
4368         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4369         if (!IS_ERR(o)) {
4370                 int rc;
4371
4372                 rc = mdt_object_lock(info, o, lh, ibits, mode);
4373                 if (rc != 0) {
4374                         mdt_object_put(info->mti_env, o);
4375                         o = ERR_PTR(rc);
4376                 }
4377         }
4378         return o;
4379 }
4380
4381 void mdt_object_unlock_put(struct mdt_thread_info *info,
4382                            struct mdt_object *o,
4383                            struct mdt_lock_handle *lh,
4384                            int decref)
4385 {
4386         mdt_object_unlock(info, o, lh, decref);
4387         mdt_object_put(info->mti_env, o);
4388 }
4389
4390 /*
4391  * Generic code handling requests that have struct mdt_body passed in:
4392  *
4393  *  - extract mdt_body from request and save it in @info, if present;
4394  *
4395  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4396  *  @info;
4397  *
4398  *  - if HAS_BODY flag is set for this request type check whether object
4399  *  actually exists on storage (lu_object_exists()).
4400  *
4401  */
4402 static int mdt_body_unpack(struct mdt_thread_info *info,
4403                            enum tgt_handler_flags flags)
4404 {
4405         const struct mdt_body    *body;
4406         struct mdt_object        *obj;
4407         const struct lu_env      *env;
4408         struct req_capsule       *pill;
4409         int                       rc;
4410
4411         ENTRY;
4412
4413         env = info->mti_env;
4414         pill = info->mti_pill;
4415
4416         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4417         if (body == NULL)
4418                 RETURN(-EFAULT);
4419
4420         if (!(body->mbo_valid & OBD_MD_FLID))
4421                 RETURN(0);
4422
4423         if (!fid_is_sane(&body->mbo_fid1)) {
4424                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4425                 RETURN(-EINVAL);
4426         }
4427
4428         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4429         if (!IS_ERR(obj)) {
4430                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4431                         mdt_object_put(env, obj);
4432                         rc = -ENOENT;
4433                 } else {
4434                         info->mti_object = obj;
4435                         rc = 0;
4436                 }
4437         } else
4438                 rc = PTR_ERR(obj);
4439
4440         RETURN(rc);
4441 }
4442
4443 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4444                                    enum tgt_handler_flags flags)
4445 {
4446         struct req_capsule *pill = info->mti_pill;
4447         int rc;
4448
4449         ENTRY;
4450
4451         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4452                 rc = mdt_body_unpack(info, flags);
4453         else
4454                 rc = 0;
4455
4456         if (rc == 0 && (flags & HAS_REPLY)) {
4457                 /* Pack reply. */
4458                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4459                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4460                                              req_capsule_ptlreq(pill) ?
4461                                              DEF_REP_MD_SIZE : MAX_MD_SIZE_OLD);
4462
4463                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4464                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4465                                              RCL_SERVER, 0);
4466
4467                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4468                  * by default. If the target object has more ACL entries, then
4469                  * enlarge the buffer when necessary.
4470                  */
4471                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4472                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4473                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4474
4475                 mdt_preset_secctx_size(info);
4476                 mdt_preset_encctx_size(info);
4477
4478                 rc = req_capsule_server_pack(pill);
4479                 if (rc)
4480                         CWARN("%s: cannot pack response: rc = %d\n",
4481                                       mdt_obd_name(info->mti_mdt), rc);
4482         }
4483         RETURN(rc);
4484 }
4485
4486 void mdt_thread_info_reset(struct mdt_thread_info *info)
4487 {
4488         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4489         info->mti_body = NULL;
4490         info->mti_dlm_req = NULL;
4491         info->mti_cross_ref = 0;
4492         info->mti_opdata = 0;
4493         info->mti_big_lmm_used = 0;
4494         info->mti_big_acl_used = 0;
4495         info->mti_som_strict = 0;
4496
4497         info->mti_spec.no_create = 0;
4498         info->mti_spec.sp_rm_entry = 0;
4499         info->mti_spec.sp_permitted = 0;
4500
4501         info->mti_spec.u.sp_ea.eadata = NULL;
4502         info->mti_spec.u.sp_ea.eadatalen = 0;
4503
4504         if (info->mti_batch_env && info->mti_object != NULL) {
4505                 mdt_object_put(info->mti_env, info->mti_object);
4506                 info->mti_object = NULL;
4507         }
4508 }
4509
4510 /*
4511  * Initialize fields of struct mdt_thread_info. Other fields are left in
4512  * uninitialized state, because it's too expensive to zero out whole
4513  * mdt_thread_info (> 1K) on each request arrival.
4514  */
4515 void mdt_thread_info_init(struct ptlrpc_request *req,
4516                           struct mdt_thread_info *info)
4517 {
4518         info->mti_pill = &req->rq_pill;
4519
4520         /* mdt device: it can be NULL while CONNECT */
4521         if (req->rq_export) {
4522                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4523                 info->mti_exp = req->rq_export;
4524         } else
4525                 info->mti_mdt = NULL;
4526         info->mti_env = req->rq_svc_thread->t_env;
4527         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4528         info->mti_big_buf = LU_BUF_NULL;
4529         info->mti_batch_env = 0;
4530         info->mti_object = NULL;
4531
4532         mdt_thread_info_reset(info);
4533 }
4534
4535 void mdt_thread_info_fini(struct mdt_thread_info *info)
4536 {
4537         int i;
4538
4539         if (info->mti_object != NULL) {
4540                 mdt_object_put(info->mti_env, info->mti_object);
4541                 info->mti_object = NULL;
4542         }
4543
4544         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4545                 mdt_lock_handle_assert(&info->mti_lh[i]);
4546         info->mti_env = NULL;
4547         info->mti_pill = NULL;
4548         info->mti_exp = NULL;
4549         info->mti_mdt = NULL;
4550
4551         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4552                 lu_buf_free(&info->mti_big_buf);
4553 }
4554
4555 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4556 {
4557         struct mdt_thread_info  *mti;
4558
4559         mti = mdt_th_info(tsi->tsi_env);
4560         LASSERT(mti != NULL);
4561
4562         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4563         if (tsi->tsi_corpus != NULL) {
4564                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4565                 lu_object_get(tsi->tsi_corpus);
4566         }
4567         mti->mti_body = tsi->tsi_mdt_body;
4568         mti->mti_dlm_req = tsi->tsi_dlm_req;
4569
4570         return mti;
4571 }
4572
4573 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4574 {
4575         if (CFS_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4576             cfs_fail_val ==
4577             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4578                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4579
4580         return tgt_connect(tsi);
4581 }
4582
4583 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4584                               struct mdt_thread_info *info,
4585                               struct ldlm_lock **lockp, __u64 flags)
4586 {
4587         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4588                                    lockp, flags);
4589 }
4590 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4591                           struct mdt_thread_info *info,
4592                           struct ldlm_lock **lockp, __u64 flags)
4593 {
4594         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4595                                lockp, flags);
4596 }
4597
4598 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4599                             struct ldlm_lock **lockp,
4600                             struct mdt_lock_handle *lh,
4601                             __u64 flags, int result)
4602 {
4603         struct ptlrpc_request  *req = mdt_info_req(info);
4604         struct ldlm_lock       *lock = *lockp;
4605         struct ldlm_lock       *new_lock;
4606
4607         /* If possible resent found a lock, @lh is set to its handle */
4608         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4609
4610         if (new_lock == NULL) {
4611                 if (flags & LDLM_FL_INTENT_ONLY) {
4612                         result = 0;
4613                 } else if (flags & LDLM_FL_RESENT) {
4614                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4615                          * resend case, however, it could be already destroyed
4616                          * due to client eviction or a raced cancel RPC.
4617                          */
4618                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4619                                           lh->mlh_reg_lh.cookie);
4620                         result = -ESTALE;
4621                 } else {
4622                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4623                                mdt_obd_name(info->mti_mdt),
4624                                lh->mlh_reg_lh.cookie, flags,
4625                                PFID(&info->mti_tmp_fid1),
4626                                PFID(&info->mti_tmp_fid2), result);
4627                         result = -ESTALE;
4628                 }
4629                 lh->mlh_reg_lh.cookie = 0;
4630                 RETURN(result);
4631         }
4632
4633         /*
4634          * If we've already given this lock to a client once, then we should
4635          * have no readers or writers.  Otherwise, we should have one reader
4636          * _or_ writer ref (which will be zeroed below) before returning the
4637          * lock to a client.
4638          */
4639         if (new_lock->l_export == req->rq_export) {
4640                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4641         } else {
4642                 LASSERT(new_lock->l_export == NULL);
4643                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4644         }
4645
4646         *lockp = new_lock;
4647
4648         if (new_lock->l_export == req->rq_export) {
4649                 /*
4650                  * Already gave this to the client, which means that we
4651                  * reconstructed a reply.
4652                  */
4653                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4654                         MSG_RESENT);
4655
4656                 LDLM_LOCK_RELEASE(new_lock);
4657                 lh->mlh_reg_lh.cookie = 0;
4658                 RETURN(ELDLM_LOCK_REPLACED);
4659         }
4660
4661         /*
4662          * Fixup the lock to be given to the client.
4663          */
4664         lock_res_and_lock(new_lock);
4665         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4666          * possible blocking AST.
4667          */
4668         while (new_lock->l_readers > 0) {
4669                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4670                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4671                 new_lock->l_readers--;
4672         }
4673         while (new_lock->l_writers > 0) {
4674                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4675                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4676                 new_lock->l_writers--;
4677         }
4678
4679         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4680         new_lock->l_blocking_ast = lock->l_blocking_ast;
4681         new_lock->l_completion_ast = lock->l_completion_ast;
4682         if (ldlm_has_dom(new_lock))
4683                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4684         new_lock->l_remote_handle = lock->l_remote_handle;
4685         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4686
4687         unlock_res_and_lock(new_lock);
4688
4689         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4690                      &new_lock->l_remote_handle,
4691                      &new_lock->l_exp_hash);
4692
4693         LDLM_LOCK_RELEASE(new_lock);
4694         lh->mlh_reg_lh.cookie = 0;
4695
4696         RETURN(ELDLM_LOCK_REPLACED);
4697 }
4698
4699 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4700                              struct ldlm_lock *new_lock,
4701                              struct mdt_lock_handle *lh, __u64 flags)
4702 {
4703         struct ptlrpc_request  *req = mdt_info_req(info);
4704         struct ldlm_request    *dlmreq;
4705
4706         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4707                 return;
4708
4709         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4710
4711         /* if this is a resend case (MSG_RESENT is set on RPC) and a lock was
4712          * found by ldlm_handle_enqueue(); if so @lh must be initialized.
4713          */
4714         if (flags & LDLM_FL_RESENT) {
4715                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4716                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4717
4718                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4719                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4720                           lh->mlh_reg_lh.cookie);
4721                 return;
4722         }
4723
4724         /*
4725          * If the xid matches, then we know this is a resent request, and allow
4726          * it. (It's probably an OPEN, for which we don't send a lock.
4727          */
4728         if (req_can_reconstruct(req, NULL) != 0)
4729                 return;
4730
4731         /*
4732          * This remote handle isn't enqueued, so we never received or processed
4733          * this request.  Clear MSG_RESENT, because it can be handled like any
4734          * normal request now.
4735          */
4736         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4737
4738         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4739                   dlmreq->lock_handle[0].cookie);
4740 }
4741
4742 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4743                                struct mdt_thread_info *info,
4744                                struct ldlm_lock **lockp,
4745                                __u64 flags)
4746 {
4747         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4748         struct ldlm_reply      *ldlm_rep = NULL;
4749         int rc;
4750
4751         ENTRY;
4752
4753         /*
4754          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4755          * (for the resend case) or a new lock. Below we will use it to
4756          * replace the original lock.
4757          */
4758         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4759         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4760                 rc = mdt_object_lock(info, info->mti_object, lhc,
4761                                      MDS_INODELOCK_XATTR, (*lockp)->l_req_mode);
4762                 if (rc)
4763                         return rc;
4764         }
4765
4766         rc = mdt_getxattr(info);
4767
4768         if (mdt_info_req(info)->rq_repmsg != NULL)
4769                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4770
4771         if (ldlm_rep == NULL ||
4772             CFS_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4773                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4774                 if (is_serious(rc))
4775                         RETURN(rc);
4776                 else
4777                         RETURN(err_serious(-EFAULT));
4778         }
4779
4780         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4781
4782         /* This is for interop instead of adding a new interop flag. LU-7433 */
4783 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4784         if (ldlm_rep->lock_policy_res2) {
4785                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4786                 RETURN(ELDLM_LOCK_ABORTED);
4787         }
4788 #endif
4789
4790         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4791         RETURN(rc);
4792 }
4793
4794 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4795                               struct mdt_thread_info *info,
4796                               struct ldlm_lock **lockp,
4797                               __u64 flags)
4798 {
4799         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4800         __u64                   child_bits;
4801         struct ldlm_reply      *ldlm_rep;
4802         struct mdt_body        *reqbody;
4803         struct mdt_body        *repbody;
4804         int                     rc, rc2;
4805
4806         ENTRY;
4807
4808         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4809         LASSERT(reqbody);
4810
4811         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4812         LASSERT(repbody);
4813
4814         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4815         repbody->mbo_eadatasize = 0;
4816         repbody->mbo_aclsize = 0;
4817
4818         switch (it_opc) {
4819         case IT_LOOKUP:
4820                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4821                 break;
4822         case IT_GETATTR:
4823                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4824                              MDS_INODELOCK_PERM;
4825                 break;
4826         default:
4827                 CERROR("%s: unsupported intent %#x\n",
4828                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4829                 GOTO(out_shrink, rc = -EINVAL);
4830         }
4831
4832         rc = mdt_init_ucred(info, reqbody);
4833         if (rc)
4834                 GOTO(out_shrink, rc);
4835
4836         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4837         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4838
4839         /* Get lock from request for possible resent case. */
4840         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4841
4842         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4843         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4844
4845         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4846             ldlm_rep->lock_policy_res2 != -ENOKEY)
4847                 ldlm_rep->lock_policy_res2 = 0;
4848         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4849             ldlm_rep->lock_policy_res2) {
4850                 lhc->mlh_reg_lh.cookie = 0ull;
4851                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4852         }
4853
4854         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4855         EXIT;
4856 out_ucred:
4857         mdt_exit_ucred(info);
4858 out_shrink:
4859         mdt_client_compatibility(info);
4860         rc2 = mdt_fix_reply(info);
4861         if (rc == 0)
4862                 rc = rc2;
4863         return rc;
4864 }
4865
4866 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4867                              struct mdt_thread_info *info,
4868                              struct ldlm_lock **lockp,
4869                              __u64 flags)
4870 {
4871         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4872         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4873         struct layout_intent *intent;
4874         struct ldlm_reply *ldlm_rep;
4875         struct lu_fid *fid = &info->mti_tmp_fid2;
4876         struct mdt_object *obj = NULL;
4877         int layout_size = 0;
4878         struct lu_buf *buf = &layout.mlc_buf;
4879         int rc = 0;
4880
4881         ENTRY;
4882
4883         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4884
4885         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4886         if (intent == NULL)
4887                 RETURN(-EPROTO);
4888
4889         CDEBUG(D_INFO, DFID "got layout change request from client: opc:%u flags:%#x extent "
4890                DEXT"\n",
4891                PFID(fid), intent->lai_opc, intent->lai_flags,
4892                PEXT(&intent->lai_extent));
4893
4894         switch (intent->lai_opc) {
4895         case LAYOUT_INTENT_TRUNC:
4896         case LAYOUT_INTENT_WRITE:
4897         case LAYOUT_INTENT_PCCRO_SET:
4898         case LAYOUT_INTENT_PCCRO_CLEAR:
4899                 layout.mlc_opc = MD_LAYOUT_WRITE;
4900                 layout.mlc_intent = intent;
4901                 break;
4902         case LAYOUT_INTENT_ACCESS:
4903                 break;
4904         case LAYOUT_INTENT_READ:
4905         case LAYOUT_INTENT_GLIMPSE:
4906         case LAYOUT_INTENT_RELEASE:
4907         case LAYOUT_INTENT_RESTORE:
4908                 CERROR("%s: Unsupported layout intent opc %d\n",
4909                        mdt_obd_name(info->mti_mdt), intent->lai_opc);
4910                 RETURN(-ENOTSUPP);
4911         default:
4912                 CERROR("%s: Unknown layout intent opc %d\n",
4913                        mdt_obd_name(info->mti_mdt), intent->lai_opc);
4914                 RETURN(-EINVAL);
4915         }
4916
4917         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4918         if (IS_ERR(obj))
4919                 RETURN(PTR_ERR(obj));
4920
4921         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4922                 /* if layout is going to be changed don't use the current EA
4923                  * size but the maximum one. That buffer will be shrinked
4924                  * to the actual size in req_capsule_shrink() before reply.
4925                  */
4926                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4927                         layout_size = info->mti_mdt->mdt_max_mdsize;
4928                 } else {
4929                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4930                         if (layout_size < 0)
4931                                 GOTO(out, rc = layout_size);
4932
4933                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4934                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4935                 }
4936                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4937                        mdt_obd_name(info->mti_mdt), layout_size);
4938         }
4939
4940         /*
4941          * set reply buffer size, so that ldlm_handle_enqueue0()->
4942          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4943          */
4944         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4945                              layout_size);
4946         rc = req_capsule_server_pack(info->mti_pill);
4947         if (rc)
4948                 GOTO(out, rc);
4949
4950         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4951         if (!ldlm_rep)
4952                 GOTO(out, rc = -EPROTO);
4953
4954         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4955
4956         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4957         if (layout.mlc_opc == MD_LAYOUT_NOP)
4958                 GOTO(out, rc = 0);
4959
4960         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4961         if (rc < 0)
4962                 GOTO(out, rc);
4963         if (rc == 1) {
4964                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4965                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4966                 GOTO(out, rc);
4967         }
4968
4969         buf->lb_buf = NULL;
4970         buf->lb_len = 0;
4971         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4972                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4973                                                      &RMF_EADATA);
4974                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4975                                                      &RMF_EADATA, RCL_CLIENT);
4976                 /*
4977                  * If it's a replay of layout write intent RPC, the client has
4978                  * saved the extended lovea when it get reply then.
4979                  */
4980                 if (buf->lb_len > 0)
4981                         mdt_fix_lov_magic(info, buf->lb_buf);
4982         }
4983
4984         /* Get lock from request for possible resent case. */
4985         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4986         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4987
4988         /*
4989          * Instantiate some layout components, if @buf contains lovea, then it's
4990          * a replay of the layout intent write RPC.
4991          */
4992         rc = mdt_layout_change(info, obj, lhc, &layout);
4993         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4994
4995         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4996                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4997                 if (rc == ELDLM_LOCK_REPLACED &&
4998                     (*lockp)->l_granted_mode == LCK_EX)
4999                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
5000         }
5001
5002         EXIT;
5003 out:
5004         mdt_object_put(info->mti_env, obj);
5005         return rc;
5006 }
5007
5008 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
5009                            struct mdt_thread_info *info,
5010                            struct ldlm_lock **lockp,
5011                            __u64 flags)
5012 {
5013         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
5014         struct ldlm_reply      *rep = NULL;
5015         long                    opc;
5016         int                     rc;
5017         struct ptlrpc_request  *req = mdt_info_req(info);
5018
5019         static const struct req_format *intent_fmts[REINT_MAX] = {
5020                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
5021                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
5022         };
5023
5024         ENTRY;
5025
5026         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
5027         if (opc < 0)
5028                 RETURN(opc);
5029
5030         /* Get lock from request for possible resent case. */
5031         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
5032
5033         rc = mdt_reint_internal(info, lhc, opc);
5034
5035         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
5036                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
5037
5038         /* Check whether the reply has been packed successfully. */
5039         if (mdt_info_req(info)->rq_repmsg != NULL)
5040                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5041         if (rep == NULL) {
5042                 if (is_serious(rc))
5043                         RETURN(rc);
5044                 else
5045                         RETURN(err_serious(-EFAULT));
5046         }
5047
5048         /* MDC expects this in any case */
5049         if (rc != 0)
5050                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
5051
5052         /* the open lock or the lock for cross-ref object should be
5053          * returned to the client
5054          */
5055         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
5056             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
5057                 rep->lock_policy_res2 = 0;
5058                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
5059                 RETURN(rc);
5060         }
5061
5062         rep->lock_policy_res2 = clear_serious(rc);
5063
5064         if (rep->lock_policy_res2 == -ENOENT &&
5065             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
5066             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
5067                 rep->lock_policy_res2 = 0;
5068
5069         lhc->mlh_reg_lh.cookie = 0ull;
5070         if (rc == -ENOTCONN || rc == -ENODEV ||
5071             rc == -EOVERFLOW) { /**< if VBR failure then return error */
5072                 /*
5073                  * If it is the disconnect error (ENODEV & ENOCONN), the error
5074                  * will be returned by rq_status, and client at ptlrpc layer
5075                  * will detect this, then disconnect, reconnect the import
5076                  * immediately, instead of impacting the following the rpc.
5077                  */
5078                 RETURN(rc);
5079         }
5080         /*
5081          * For other cases, the error will be returned by intent, and client
5082          * will retrieve the result from intent.
5083          */
5084         RETURN(ELDLM_LOCK_ABORTED);
5085 }
5086
5087 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
5088                           struct mdt_thread_info *info,
5089                           struct ldlm_lock **lockp,
5090                           u64 flags /* LDLM_FL_* */)
5091 {
5092         struct req_capsule *pill = info->mti_pill;
5093         struct ptlrpc_request *req = mdt_info_req(info);
5094         const struct req_format *it_format;
5095         int (*it_handler)(enum ldlm_intent_flags,
5096                           struct mdt_thread_info *,
5097                           struct ldlm_lock **,
5098                           u64);
5099         enum tgt_handler_flags it_handler_flags = 0;
5100         struct ldlm_reply *rep;
5101         bool check_mdt_object = false;
5102         int rc;
5103
5104         ENTRY;
5105
5106         switch (it_opc) {
5107         case IT_OPEN:
5108         case IT_OPEN|IT_CREAT:
5109                 /*
5110                  * OCREAT is not a IS_MUTABLE request since the file may
5111                  * already exist. We do the extra check of
5112                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
5113                  * really need to create the object.
5114                  */
5115                 it_format = &RQF_LDLM_INTENT;
5116                 it_handler = &mdt_intent_open;
5117                 break;
5118         case IT_GETATTR:
5119                 check_mdt_object = true;
5120                 fallthrough;
5121         case IT_LOOKUP:
5122                 it_format = &RQF_LDLM_INTENT_GETATTR;
5123                 it_handler = &mdt_intent_getattr;
5124                 it_handler_flags = HAS_REPLY;
5125                 break;
5126         case IT_GETXATTR:
5127                 check_mdt_object = true;
5128                 it_format = &RQF_LDLM_INTENT_GETXATTR;
5129                 it_handler = &mdt_intent_getxattr;
5130                 it_handler_flags = HAS_BODY;
5131                 break;
5132         case IT_LAYOUT:
5133                 it_format = &RQF_LDLM_INTENT_LAYOUT;
5134                 it_handler = &mdt_intent_layout;
5135                 break;
5136         case IT_GLIMPSE:
5137                 it_format = &RQF_LDLM_INTENT;
5138                 it_handler = &mdt_intent_glimpse;
5139                 break;
5140         case IT_BRW:
5141                 it_format = &RQF_LDLM_INTENT;
5142                 it_handler = &mdt_intent_brw;
5143                 break;
5144         case IT_QUOTA_DQACQ:
5145         case IT_QUOTA_CONN: {
5146                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
5147
5148                 if (qmt == NULL)
5149                         RETURN(-EOPNOTSUPP);
5150
5151                 if (mdt_rdonly(req->rq_export))
5152                         RETURN(-EROFS);
5153
5154                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
5155                 /* pass the request to quota master */
5156                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
5157                                                  mdt_info_req(info), lockp,
5158                                                  flags);
5159                 RETURN(rc);
5160         }
5161         default:
5162                 CERROR("%s: unknown intent code %#x\n",
5163                        mdt_obd_name(info->mti_mdt), it_opc);
5164                 RETURN(-EPROTO);
5165         }
5166
5167         if (!info->mti_batch_env)
5168                 req_capsule_extend(pill, it_format);
5169
5170         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
5171         if (rc < 0)
5172                 RETURN(rc);
5173
5174         if (unlikely(info->mti_object == NULL && check_mdt_object))
5175                 RETURN(-EPROTO);
5176
5177         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
5178                 RETURN(-EROFS);
5179
5180         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
5181
5182         /* execute policy */
5183         rc = (*it_handler)(it_opc, info, lockp, flags);
5184
5185         /* Check whether the reply has been packed successfully. */
5186         if (info->mti_batch_env || req->rq_repmsg != NULL) {
5187                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5188                 rep->lock_policy_res2 =
5189                         ptlrpc_status_hton(rep->lock_policy_res2);
5190         }
5191
5192         RETURN(rc);
5193 }
5194
5195 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
5196                                     enum ldlm_intent_flags it_opc)
5197 {
5198         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
5199
5200         /* update stats when IT code is known */
5201         if (srv_stats != NULL)
5202                 lprocfs_counter_incr(srv_stats,
5203                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
5204                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
5205 }
5206
5207 static int mdt_intent_policy(const struct lu_env *env,
5208                              struct ldlm_namespace *ns,
5209                              struct ldlm_lock **lockp,
5210                              void *req_cookie,
5211                              enum ldlm_mode mode,
5212                              __u64 flags, void *data)
5213 {
5214         struct tgt_session_info *tsi;
5215         struct mdt_thread_info  *info;
5216         struct ptlrpc_request   *req  =  req_cookie;
5217         struct ldlm_intent      *it;
5218         struct req_capsule      *pill;
5219         const struct ldlm_lock_desc *ldesc;
5220         int rc;
5221
5222         ENTRY;
5223
5224         LASSERT(req != NULL);
5225
5226         tsi = tgt_ses_info(env);
5227
5228         info = mdt_th_info(env);
5229         LASSERT(info != NULL);
5230
5231         /* Check whether it is a sub request processing in a batch request */
5232         if (info->mti_batch_env) {
5233                 pill = info->mti_pill;
5234                 LASSERT(pill == &info->mti_sub_pill);
5235         } else {
5236                 info = tsi2mdt_info(tsi);
5237                 pill = info->mti_pill;
5238         }
5239
5240         LASSERT(pill->rc_req == req);
5241         ldesc = &info->mti_dlm_req->lock_desc;
5242
5243         if (info->mti_batch_env ||
5244             req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
5245                 /*
5246                  * For batch processing environment, the request format has
5247                  * already been set.
5248                  */
5249                 if (!info->mti_batch_env)
5250                         req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
5251
5252                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
5253                 if (it != NULL) {
5254                         mdt_ptlrpc_stats_update(req, it->opc);
5255                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
5256                         if (rc == 0)
5257                                 rc = ELDLM_OK;
5258
5259                         /* Lock without inodebits makes no sense and will oops
5260                          * later in ldlm. Let's check it now to see if we have
5261                          * ibits corrupted somewhere in mdt_intent_opc().
5262                          * The case for client miss to set ibits has been
5263                          * processed by others.
5264                          */
5265                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
5266                                 ldesc->l_policy_data.l_inodebits.bits != 0));
5267                 } else {
5268                         rc = err_serious(-EFAULT);
5269                 }
5270         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
5271                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
5272                 struct ldlm_reply *rep;
5273
5274                 /* No intent was provided but INTENT flag is set along with
5275                  * DOM bit, this is considered as GLIMPSE request.
5276                  * This logic is common for MDT and OST glimpse
5277                  */
5278                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
5279                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
5280                 /* Check whether the reply has been packed successfully. */
5281                 if (req->rq_repmsg != NULL) {
5282                         rep = req_capsule_server_get(info->mti_pill,
5283                                                      &RMF_DLM_REP);
5284                         rep->lock_policy_res2 =
5285                                 ptlrpc_status_hton(rep->lock_policy_res2);
5286                 }
5287         } else {
5288                 /* No intent was provided */
5289                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5290                 rc = req_capsule_server_pack(pill);
5291                 if (rc)
5292                         rc = err_serious(rc);
5293         }
5294
5295         if (!info->mti_batch_env)
5296                 mdt_thread_info_fini(info);
5297         RETURN(rc);
5298 }
5299
5300 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5301 {
5302         struct seq_server_site  *ss = mdt_seq_site(mdt);
5303
5304         if (ss->ss_node_id == 0)
5305                 return;
5306
5307         if (ss->ss_client_seq != NULL) {
5308                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5309                 ss->ss_client_seq->lcs_exp = NULL;
5310         }
5311
5312         if (ss->ss_server_fld != NULL) {
5313                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5314                 ss->ss_server_fld->lsf_control_exp = NULL;
5315         }
5316 }
5317
5318 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5319 {
5320         struct seq_server_site *ss = mdt_seq_site(mdt);
5321
5322         if (ss == NULL)
5323                 return;
5324
5325         if (ss->ss_server_seq != NULL)
5326                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5327 }
5328
5329 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5330 {
5331         mdt_seq_fini_cli(mdt);
5332         mdt_deregister_seq_exp(mdt);
5333
5334         return seq_site_fini(env, mdt_seq_site(mdt));
5335 }
5336
5337 /**
5338  * It will retrieve its FLDB entries from MDT0, and it only happens
5339  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5340  * and it needs to refresh FLDB from the MDT0.
5341  **/
5342 static int mdt_register_lwp_callback(void *data)
5343 {
5344         struct lu_env           env;
5345         struct mdt_device       *mdt = data;
5346         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5347         int                     rc;
5348
5349         ENTRY;
5350
5351         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5352
5353         rc = lu_env_init(&env, LCT_MD_THREAD);
5354         if (rc < 0) {
5355                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5356                 RETURN(rc);
5357         }
5358
5359         /* Allocate new sequence now to avoid creating local transaction
5360          * in the normal transaction process
5361          */
5362         rc = seq_server_check_and_alloc_super(&env,
5363                                               mdt_seq_site(mdt)->ss_server_seq);
5364         if (rc < 0)
5365                 GOTO(out, rc);
5366
5367         if (fld->lsf_new) {
5368                 rc = fld_update_from_controller(&env, fld);
5369                 if (rc != 0) {
5370                         CERROR("%s: cannot update controller: rc = %d\n",
5371                                mdt_obd_name(mdt), rc);
5372                         GOTO(out, rc);
5373                 }
5374         }
5375 out:
5376         lu_env_fini(&env);
5377         RETURN(rc);
5378 }
5379
5380 static int mdt_register_seq_exp(struct mdt_device *mdt)
5381 {
5382         struct seq_server_site  *ss = mdt_seq_site(mdt);
5383         char                    *lwp_name = NULL;
5384         int                     rc;
5385
5386         if (ss->ss_node_id == 0)
5387                 return 0;
5388
5389         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5390         if (lwp_name == NULL)
5391                 GOTO(out_free, rc = -ENOMEM);
5392
5393         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5394         if (rc != 0)
5395                 GOTO(out_free, rc);
5396
5397         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5398                                       NULL, NULL);
5399         if (rc != 0)
5400                 GOTO(out_free, rc);
5401
5402         rc = lustre_register_lwp_item(lwp_name,
5403                                       &ss->ss_server_fld->lsf_control_exp,
5404                                       mdt_register_lwp_callback, mdt);
5405         if (rc != 0) {
5406                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5407                 ss->ss_client_seq->lcs_exp = NULL;
5408                 GOTO(out_free, rc);
5409         }
5410 out_free:
5411         if (lwp_name != NULL)
5412                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5413
5414         return rc;
5415 }
5416
5417 /*
5418  * Init client sequence manager which is used by local MDS to talk to sequence
5419  * controller on remote node.
5420  */
5421 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5422 {
5423         struct seq_server_site *ss = mdt_seq_site(mdt);
5424         char *prefix;
5425
5426         ENTRY;
5427
5428         /* check if this is adding the first MDC and controller is not yet
5429          * initialized.
5430          */
5431         OBD_ALLOC_PTR(ss->ss_client_seq);
5432         if (ss->ss_client_seq == NULL)
5433                 RETURN(-ENOMEM);
5434
5435         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5436         if (prefix == NULL) {
5437                 OBD_FREE_PTR(ss->ss_client_seq);
5438                 ss->ss_client_seq = NULL;
5439                 RETURN(-ENOMEM);
5440         }
5441
5442         /* Note: seq_client_fini will be called in seq_site_fini */
5443         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5444         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5445                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5446                                                             NULL);
5447         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5448
5449         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5450 }
5451
5452 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5453 {
5454         struct seq_server_site  *ss;
5455         int                     rc;
5456
5457         ENTRY;
5458
5459         ss = mdt_seq_site(mdt);
5460         /* init sequence controller server(MDT0) */
5461         if (ss->ss_node_id == 0) {
5462                 OBD_ALLOC_PTR(ss->ss_control_seq);
5463                 if (ss->ss_control_seq == NULL)
5464                         RETURN(-ENOMEM);
5465
5466                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5467                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5468                                      ss);
5469                 if (rc)
5470                         GOTO(out_seq_fini, rc);
5471         }
5472
5473         /* Init normal sequence server */
5474         OBD_ALLOC_PTR(ss->ss_server_seq);
5475         if (ss->ss_server_seq == NULL)
5476                 GOTO(out_seq_fini, rc = -ENOMEM);
5477
5478         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5479                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5480         if (rc)
5481                 GOTO(out_seq_fini, rc);
5482
5483         /* init seq client for seq server to talk to seq controller(MDT0) */
5484         rc = mdt_seq_init_cli(env, mdt);
5485         if (rc != 0)
5486                 GOTO(out_seq_fini, rc);
5487
5488         if (ss->ss_node_id != 0)
5489                 /* register controller export through lwp */
5490                 rc = mdt_register_seq_exp(mdt);
5491
5492         EXIT;
5493 out_seq_fini:
5494         if (rc)
5495                 mdt_seq_fini(env, mdt);
5496
5497         return rc;
5498 }
5499
5500 /*
5501  * FLD wrappers
5502  */
5503 static int mdt_fld_fini(const struct lu_env *env,
5504                         struct mdt_device *m)
5505 {
5506         struct seq_server_site *ss = mdt_seq_site(m);
5507
5508         ENTRY;
5509
5510         if (ss && ss->ss_server_fld) {
5511                 fld_server_fini(env, ss->ss_server_fld);
5512                 OBD_FREE_PTR(ss->ss_server_fld);
5513                 ss->ss_server_fld = NULL;
5514         }
5515
5516         RETURN(0);
5517 }
5518
5519 static int mdt_fld_init(const struct lu_env *env,
5520                         const char *uuid,
5521                         struct mdt_device *m)
5522 {
5523         struct seq_server_site *ss;
5524         int rc;
5525
5526         ENTRY;
5527
5528         ss = mdt_seq_site(m);
5529
5530         OBD_ALLOC_PTR(ss->ss_server_fld);
5531         if (ss->ss_server_fld == NULL)
5532                 RETURN(rc = -ENOMEM);
5533
5534         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5535                              LU_SEQ_RANGE_MDT);
5536         if (rc) {
5537                 OBD_FREE_PTR(ss->ss_server_fld);
5538                 ss->ss_server_fld = NULL;
5539                 RETURN(rc);
5540         }
5541
5542         RETURN(0);
5543 }
5544
5545 static void mdt_stack_pre_fini(const struct lu_env *env,
5546                            struct mdt_device *m, struct lu_device *top)
5547 {
5548         struct lustre_cfg_bufs  *bufs;
5549         struct lustre_cfg       *lcfg;
5550         struct mdt_thread_info  *info;
5551
5552         ENTRY;
5553
5554         LASSERT(top);
5555
5556         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5557         LASSERT(info != NULL);
5558
5559         bufs = &info->mti_u.bufs;
5560
5561         LASSERT(m->mdt_child_exp);
5562         LASSERT(m->mdt_child_exp->exp_obd);
5563
5564         /* process cleanup, pass mdt obd name to get obd umount flags */
5565         /* XXX: this is needed because all layers are referenced by
5566          * objects (some of them are pinned by osd, for example *
5567          * the proper solution should be a model where object used
5568          * by osd only doesn't have mdt/mdd slices -bzzz
5569          */
5570         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5571         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5572         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5573         if (!lcfg)
5574                 RETURN_EXIT;
5575         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5576
5577         top->ld_ops->ldo_process_config(env, top, lcfg);
5578         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5579         EXIT;
5580 }
5581
5582 static void mdt_stack_fini(const struct lu_env *env,
5583                            struct mdt_device *m, struct lu_device *top)
5584 {
5585         struct obd_device       *obd = mdt2obd_dev(m);
5586         struct lustre_cfg_bufs  *bufs;
5587         struct lustre_cfg       *lcfg;
5588         struct mdt_thread_info  *info;
5589         char                     flags[3] = "";
5590
5591         ENTRY;
5592
5593         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5594         LASSERT(info != NULL);
5595
5596         lu_dev_del_linkage(top->ld_site, top);
5597
5598         lu_site_purge(env, top->ld_site, -1);
5599
5600         bufs = &info->mti_u.bufs;
5601         /* process cleanup, pass mdt obd name to get obd umount flags */
5602         /* another purpose is to let all layers to release their objects */
5603         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5604         if (obd->obd_force)
5605                 strcat(flags, "F");
5606         if (obd->obd_fail)
5607                 strcat(flags, "A");
5608         lustre_cfg_bufs_set_string(bufs, 1, flags);
5609         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5610         if (!lcfg)
5611                 RETURN_EXIT;
5612         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5613
5614         LASSERT(top);
5615         top->ld_ops->ldo_process_config(env, top, lcfg);
5616         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5617
5618         lu_site_purge(env, top->ld_site, -1);
5619
5620         m->mdt_child = NULL;
5621         m->mdt_bottom = NULL;
5622
5623         obd_disconnect(m->mdt_child_exp);
5624         m->mdt_child_exp = NULL;
5625
5626         obd_disconnect(m->mdt_bottom_exp);
5627         m->mdt_child_exp = NULL;
5628 }
5629
5630 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5631                                const char *next, struct obd_export **exp)
5632 {
5633         struct obd_connect_data *data = NULL;
5634         struct obd_device       *obd;
5635         int                      rc;
5636
5637         ENTRY;
5638
5639         OBD_ALLOC_PTR(data);
5640         if (data == NULL)
5641                 GOTO(out, rc = -ENOMEM);
5642
5643         obd = class_name2obd(next);
5644         if (obd == NULL) {
5645                 CERROR("%s: can't locate next device: %s\n",
5646                        mdt_obd_name(m), next);
5647                 GOTO(out, rc = -ENOTCONN);
5648         }
5649
5650         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5651         data->ocd_version = LUSTRE_VERSION_CODE;
5652
5653         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5654         if (rc) {
5655                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5656                        mdt_obd_name(m), next, rc);
5657                 GOTO(out, rc);
5658         }
5659
5660 out:
5661         if (data)
5662                 OBD_FREE_PTR(data);
5663         RETURN(rc);
5664 }
5665
5666 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5667                           struct lustre_cfg *cfg)
5668 {
5669         char                   *dev = lustre_cfg_string(cfg, 0);
5670         int                     rc, name_size, uuid_size;
5671         char                   *name, *uuid, *p;
5672         struct lustre_cfg_bufs *bufs;
5673         struct lustre_cfg      *lcfg;
5674         struct obd_device      *obd;
5675         struct lustre_profile  *lprof;
5676         struct lu_site         *site;
5677
5678         ENTRY;
5679
5680         /* in 1.8 we had the only device in the stack - MDS.
5681          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5682          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5683          * to generate names and setup MDT, MDD. MDT will be using
5684          * generated name to connect to MDD. for MDD the next device
5685          * will be LOD with name taken from so called "profile" which
5686          * is generated by mount_option line
5687          *
5688          * 1.8 MGS generates config. commands like this:
5689          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5690          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5691          * 2.0 MGS generates config. commands like this:
5692          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5693          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5694          *                    3:lustre-MDT0000-mdtlov  4:f
5695          *
5696          * we generate MDD name from MDT one, just replacing T with D
5697          *
5698          * after all the preparations, the logical equivalent will be
5699          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5700          *                    3:lustre-MDT0000-mdtlov  4:f
5701          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5702          *                    3:lustre-MDD0000  4:f
5703          *
5704          *  notice we build the stack from down to top: MDD first, then MDT
5705          */
5706
5707         name_size = MAX_OBD_NAME;
5708         uuid_size = MAX_OBD_NAME;
5709
5710         OBD_ALLOC(name, name_size);
5711         OBD_ALLOC(uuid, uuid_size);
5712         if (name == NULL || uuid == NULL)
5713                 GOTO(cleanup_mem, rc = -ENOMEM);
5714
5715         OBD_ALLOC_PTR(bufs);
5716         if (!bufs)
5717                 GOTO(cleanup_mem, rc = -ENOMEM);
5718
5719         strcpy(name, dev);
5720         p = strstr(name, "-MDT");
5721         if (p == NULL)
5722                 GOTO(free_bufs, rc = -ENOMEM);
5723         p[3] = 'D';
5724
5725         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5726
5727         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5728         if (lprof == NULL || lprof->lp_dt == NULL) {
5729                 CERROR("can't find the profile: %s\n",
5730                        lustre_cfg_string(cfg, 0));
5731                 GOTO(free_bufs, rc = -EINVAL);
5732         }
5733
5734         lustre_cfg_bufs_reset(bufs, name);
5735         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5736         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5737         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5738
5739         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5740         if (!lcfg)
5741                 GOTO(put_profile, rc = -ENOMEM);
5742         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5743
5744         rc = class_attach(lcfg);
5745         if (rc)
5746                 GOTO(lcfg_cleanup, rc);
5747
5748         obd = class_name2obd(name);
5749         if (!obd) {
5750                 CERROR("Can not find obd %s (%s in config)\n",
5751                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5752                 GOTO(lcfg_cleanup, rc = -EINVAL);
5753         }
5754
5755         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5756
5757         lustre_cfg_bufs_reset(bufs, name);
5758         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5759         lustre_cfg_bufs_set_string(bufs, 2, dev);
5760         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5761
5762         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5763         if (!lcfg)
5764                 GOTO(class_detach, rc = -ENOMEM);
5765         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5766
5767         rc = class_setup(obd, lcfg);
5768         if (rc)
5769                 GOTO(class_detach, rc);
5770
5771         /* connect to MDD we just setup */
5772         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5773         if (rc)
5774                 GOTO(class_detach, rc);
5775
5776         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5777         LASSERT(site);
5778         LASSERT(mdt_lu_site(mdt) == NULL);
5779         mdt->mdt_lu_dev.ld_site = site;
5780         site->ls_top_dev = &mdt->mdt_lu_dev;
5781         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5782
5783         /* now connect to bottom OSD */
5784         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5785         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5786         if (rc)
5787                 GOTO(class_detach, rc);
5788         mdt->mdt_bottom =
5789                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5790
5791         rc = lu_env_refill((struct lu_env *)env);
5792         if (rc != 0)
5793                 CERROR("Failure to refill session: '%d'\n", rc);
5794
5795         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5796
5797         EXIT;
5798 class_detach:
5799         if (rc)
5800                 class_detach(obd, lcfg);
5801 lcfg_cleanup:
5802         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5803 put_profile:
5804         class_put_profile(lprof);
5805 free_bufs:
5806         OBD_FREE_PTR(bufs);
5807 cleanup_mem:
5808         if (name)
5809                 OBD_FREE(name, name_size);
5810         if (uuid)
5811                 OBD_FREE(uuid, uuid_size);
5812         RETURN(rc);
5813 }
5814
5815 /* setup quota master target on MDT0 */
5816 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5817                           struct lustre_cfg *cfg)
5818 {
5819         struct obd_device       *obd;
5820         char                    *dev = lustre_cfg_string(cfg, 0);
5821         char                    *qmtname, *uuid, *p;
5822         struct lustre_cfg_bufs  *bufs;
5823         struct lustre_cfg       *lcfg;
5824         struct lustre_profile   *lprof;
5825         struct obd_connect_data *data;
5826         int                      rc;
5827
5828         ENTRY;
5829
5830         LASSERT(mdt->mdt_qmt_exp == NULL);
5831         LASSERT(mdt->mdt_qmt_dev == NULL);
5832
5833         /* quota master is on MDT0 only for now */
5834         if (mdt->mdt_seq_site.ss_node_id != 0)
5835                 RETURN(0);
5836
5837         /* MGS generates config commands which look as follows:
5838          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5839          *                    3:lustre-MDT0000-mdtlov  4:f
5840          *
5841          * We generate the QMT name from the MDT one, just replacing MD with QM
5842          * after all the preparations, the logical equivalent will be:
5843          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5844          *                    3:lustre-MDT0000-osd  4:f
5845          */
5846         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5847         OBD_ALLOC(uuid, UUID_MAX);
5848         OBD_ALLOC_PTR(bufs);
5849         OBD_ALLOC_PTR(data);
5850         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5851                 GOTO(cleanup_mem, rc = -ENOMEM);
5852
5853         strcpy(qmtname, dev);
5854         p = strstr(qmtname, "-MDT");
5855         if (p == NULL)
5856                 GOTO(cleanup_mem, rc = -ENOMEM);
5857         /* replace MD with QM */
5858         p[1] = 'Q';
5859         p[2] = 'M';
5860
5861         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5862
5863         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5864         if (lprof == NULL || lprof->lp_dt == NULL) {
5865                 CERROR("can't find profile for %s\n",
5866                        lustre_cfg_string(cfg, 0));
5867                 GOTO(cleanup_mem, rc = -EINVAL);
5868         }
5869
5870         lustre_cfg_bufs_reset(bufs, qmtname);
5871         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5872         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5873         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5874
5875         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5876         if (!lcfg)
5877                 GOTO(put_profile, rc = -ENOMEM);
5878         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5879
5880         rc = class_attach(lcfg);
5881         if (rc)
5882                 GOTO(lcfg_cleanup, rc);
5883
5884         obd = class_name2obd(qmtname);
5885         if (!obd) {
5886                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5887                        lustre_cfg_string(cfg, 0));
5888                 GOTO(lcfg_cleanup, rc = -EINVAL);
5889         }
5890
5891         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5892
5893         lustre_cfg_bufs_reset(bufs, qmtname);
5894         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5895         lustre_cfg_bufs_set_string(bufs, 2, dev);
5896
5897         /* for quota, the next device should be the OSD device */
5898         lustre_cfg_bufs_set_string(bufs, 3,
5899                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5900
5901         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5902         if (!lcfg)
5903                 GOTO(class_detach, rc = -ENOMEM);
5904         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5905
5906         rc = class_setup(obd, lcfg);
5907         if (rc)
5908                 GOTO(class_detach, rc);
5909
5910         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5911
5912         /* configure local quota objects */
5913         if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5914                 rc = -EBADF;
5915         else
5916                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5917                                                            &mdt->mdt_lu_dev,
5918                                                            mdt->mdt_qmt_dev);
5919         if (rc)
5920                 GOTO(class_cleanup, rc);
5921
5922         /* connect to quota master target */
5923         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5924         data->ocd_version = LUSTRE_VERSION_CODE;
5925         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5926                          data, NULL);
5927         if (rc) {
5928                 CERROR("cannot connect to quota master device %s (%d)\n",
5929                        qmtname, rc);
5930                 GOTO(class_cleanup, rc);
5931         }
5932
5933         EXIT;
5934 class_cleanup:
5935         if (rc) {
5936                 class_manual_cleanup(obd);
5937                 mdt->mdt_qmt_dev = NULL;
5938                 GOTO(lcfg_cleanup, rc);
5939         }
5940 class_detach:
5941         if (rc)
5942                 class_detach(obd, lcfg);
5943 lcfg_cleanup:
5944         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5945 put_profile:
5946         class_put_profile(lprof);
5947 cleanup_mem:
5948         if (bufs)
5949                 OBD_FREE_PTR(bufs);
5950         if (qmtname)
5951                 OBD_FREE(qmtname, MAX_OBD_NAME);
5952         if (uuid)
5953                 OBD_FREE(uuid, UUID_MAX);
5954         if (data)
5955                 OBD_FREE_PTR(data);
5956         return rc;
5957 }
5958
5959 /* Shutdown quota master target associated with mdt */
5960 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5961 {
5962         ENTRY;
5963
5964         if (mdt->mdt_qmt_exp == NULL)
5965                 RETURN_EXIT;
5966         LASSERT(mdt->mdt_qmt_dev != NULL);
5967
5968         /* the qmt automatically shuts down when the mdt disconnects */
5969         obd_disconnect(mdt->mdt_qmt_exp);
5970         mdt->mdt_qmt_exp = NULL;
5971         mdt->mdt_qmt_dev = NULL;
5972         EXIT;
5973 }
5974
5975 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5976  * for now. This will be removed along with converting rest of MDT code
5977  * to use tgt_session_info
5978  */
5979 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5980 {
5981         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5982         int                      rc;
5983
5984         if (unlikely(info->mti_object == NULL))
5985                 return -EPROTO;
5986
5987         rc = mdt_getxattr(info);
5988
5989         mdt_thread_info_fini(info);
5990         return rc;
5991 }
5992
5993 static int mdt_llog_open(struct tgt_session_info *tsi)
5994 {
5995         ENTRY;
5996
5997         if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
5998                 RETURN(err_serious(-EACCES));
5999
6000         RETURN(tgt_llog_open(tsi));
6001 }
6002
6003 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
6004 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
6005 #define OST_BRW_READ    OST_READ
6006 #define OST_BRW_WRITE   OST_WRITE
6007
6008 static struct tgt_handler mdt_tgt_handlers[] = {
6009 TGT_RPC_HANDLER(MDS_FIRST_OPC,
6010                 0,                      MDS_CONNECT,    mdt_tgt_connect,
6011                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
6012 TGT_RPC_HANDLER(MDS_FIRST_OPC,
6013                 0,                      MDS_DISCONNECT, tgt_disconnect,
6014                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
6015 TGT_RPC_HANDLER(MDS_FIRST_OPC,
6016                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
6017                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
6018 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
6019 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
6020 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
6021 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
6022                                                         mdt_getattr_name),
6023 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
6024 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
6025 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
6026 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
6027 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
6028 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
6029 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
6030 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
6031                                                         mdt_hsm_progress),
6032 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
6033                                                         mdt_hsm_ct_register),
6034 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
6035                                                         mdt_hsm_ct_unregister),
6036 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
6037                                                         mdt_hsm_state_get),
6038 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
6039                                                         mdt_hsm_state_set),
6040 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
6041 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
6042                                                         mdt_hsm_request),
6043 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
6044             MDS_SWAP_LAYOUTS,
6045             mdt_swap_layouts),
6046 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
6047 TGT_MDT_HDL(IS_MUTABLE,         MDS_BATCH,      mdt_batch),
6048 };
6049
6050 static struct tgt_handler mdt_io_ops[] = {
6051 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
6052                                                         mdt_hp_brw),
6053 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
6054                                                         mdt_hp_brw),
6055 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
6056                                          OST_PUNCH,     mdt_punch_hdl,
6057                                                         mdt_hp_punch),
6058 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
6059 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
6060                                                         mdt_fallocate_hdl),
6061 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
6062 TGT_RPC_HANDLER(OST_FIRST_OPC,
6063                 0,                      OST_SET_INFO,   mdt_io_set_info,
6064                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
6065 };
6066
6067 static struct tgt_handler mdt_sec_ctx_ops[] = {
6068 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
6069 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT, mdt_sec_ctx_handle),
6070 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
6071 };
6072
6073 static struct tgt_handler mdt_quota_ops[] = {
6074 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
6075 };
6076
6077 static struct tgt_handler mdt_llog_handlers[] = {
6078         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
6079         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
6080         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
6081         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
6082 };
6083
6084 static struct tgt_opc_slice mdt_common_slice[] = {
6085         {
6086                 .tos_opc_start  = MDS_FIRST_OPC,
6087                 .tos_opc_end    = MDS_LAST_OPC,
6088                 .tos_hs         = mdt_tgt_handlers
6089         },
6090         {
6091                 .tos_opc_start  = OBD_FIRST_OPC,
6092                 .tos_opc_end    = OBD_LAST_OPC,
6093                 .tos_hs         = tgt_obd_handlers
6094         },
6095         {
6096                 .tos_opc_start  = LDLM_FIRST_OPC,
6097                 .tos_opc_end    = LDLM_LAST_OPC,
6098                 .tos_hs         = tgt_dlm_handlers
6099         },
6100         {
6101                 .tos_opc_start  = SEC_FIRST_OPC,
6102                 .tos_opc_end    = SEC_LAST_OPC,
6103                 .tos_hs         = mdt_sec_ctx_ops
6104         },
6105         {
6106                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
6107                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
6108                 .tos_hs         = tgt_out_handlers
6109         },
6110         {
6111                 .tos_opc_start  = FLD_FIRST_OPC,
6112                 .tos_opc_end    = FLD_LAST_OPC,
6113                 .tos_hs         = fld_handlers
6114         },
6115         {
6116                 .tos_opc_start  = SEQ_FIRST_OPC,
6117                 .tos_opc_end    = SEQ_LAST_OPC,
6118                 .tos_hs         = seq_handlers
6119         },
6120         {
6121                 .tos_opc_start  = QUOTA_DQACQ,
6122                 .tos_opc_end    = QUOTA_LAST_OPC,
6123                 .tos_hs         = mdt_quota_ops
6124         },
6125         {
6126                 .tos_opc_start  = LLOG_FIRST_OPC,
6127                 .tos_opc_end    = LLOG_LAST_OPC,
6128                 .tos_hs         = mdt_llog_handlers
6129         },
6130         {
6131                 .tos_opc_start  = LFSCK_FIRST_OPC,
6132                 .tos_opc_end    = LFSCK_LAST_OPC,
6133                 .tos_hs         = tgt_lfsck_handlers
6134         },
6135         {
6136                 .tos_opc_start  = OST_FIRST_OPC,
6137                 .tos_opc_end    = OST_LAST_OPC,
6138                 .tos_hs         = mdt_io_ops
6139         },
6140         {
6141                 .tos_hs         = NULL
6142         }
6143 };
6144
6145 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
6146 {
6147         struct md_device *next = m->mdt_child;
6148         struct lu_device *d = &m->mdt_lu_dev;
6149         struct obd_device *obd = mdt2obd_dev(m);
6150         struct lfsck_stop stop;
6151
6152         ENTRY;
6153         stop.ls_status = LS_PAUSED;
6154         stop.ls_flags = 0;
6155         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
6156
6157         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
6158
6159         mdt_restriper_stop(m);
6160         ping_evictor_stop();
6161
6162         /* Remove the HSM /proc entry so the coordinator cannot be
6163          * restarted by a user while it's shutting down.
6164          */
6165         mdt_hsm_cdt_stop(m);
6166
6167         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
6168         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
6169
6170         if (m->mdt_namespace != NULL)
6171                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
6172                                           d->ld_obd->obd_force);
6173
6174         obd_exports_barrier(obd);
6175         obd_zombie_barrier();
6176
6177         mdt_quota_fini(env, m);
6178
6179         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
6180
6181         /* Calling the cleanup functions in the same order as in the mdt_init0
6182          * error path
6183          */
6184         mdt_tunables_fini(m);
6185
6186         target_recovery_fini(obd);
6187         upcall_cache_cleanup(m->mdt_identity_cache);
6188         m->mdt_identity_cache = NULL;
6189
6190         tgt_fini(env, &m->mdt_lut);
6191
6192         mdt_hsm_cdt_fini(m);
6193
6194         if (m->mdt_los != NULL) {
6195                 local_oid_storage_fini(env, m->mdt_los);
6196                 m->mdt_los = NULL;
6197         }
6198
6199         if (m->mdt_namespace != NULL) {
6200                 ldlm_namespace_free_post(m->mdt_namespace);
6201                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
6202         }
6203
6204         if (m->mdt_md_root != NULL) {
6205                 mdt_object_put(env, m->mdt_md_root);
6206                 m->mdt_md_root = NULL;
6207         }
6208
6209         mdt_seq_fini(env, m);
6210
6211         mdt_fld_fini(env, m);
6212
6213         /*
6214          * Finish the stack
6215          */
6216         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6217
6218         LASSERT(atomic_read(&d->ld_ref) == 0);
6219
6220         server_put_mount(mdt_obd_name(m), true);
6221
6222         EXIT;
6223 }
6224
6225 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
6226
6227 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
6228                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
6229 {
6230         const struct dt_device_param *dt_conf;
6231         struct mdt_thread_info *info;
6232         struct obd_device *obd;
6233         const char *dev = lustre_cfg_string(cfg, 0);
6234         const char *num = lustre_cfg_string(cfg, 2);
6235         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
6236         struct lustre_mount_info *lmi = NULL;
6237         struct lustre_sb_info *lsi;
6238         struct lu_site *s;
6239         struct seq_server_site *ss_site;
6240         const char *identity_upcall = "NONE";
6241         struct md_device *next;
6242         struct lu_fid fid;
6243         int rc;
6244         long node_id;
6245         mntopt_t mntopts;
6246
6247         ENTRY;
6248
6249         lu_device_init(&m->mdt_lu_dev, ldt);
6250         /*
6251          * Environment (env) might be missing mdt_thread_key values at that
6252          * point, if device is allocated when mdt_thread_key is in QUIESCENT
6253          * mode.
6254          *
6255          * Usually device allocation path doesn't use module key values, but
6256          * mdt has to do a lot of work here, so allocate key value.
6257          */
6258         rc = lu_env_refill((struct lu_env *)env);
6259         if (rc != 0)
6260                 RETURN(rc);
6261
6262         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
6263         LASSERT(info != NULL);
6264
6265         obd = class_name2obd(dev);
6266         LASSERT(obd != NULL);
6267
6268         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
6269         m->mdt_evict_tgt_nids = 1;
6270         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
6271
6272         lmi = server_get_mount(dev);
6273         if (lmi == NULL) {
6274                 CERROR("Cannot get mount info for %s!\n", dev);
6275                 RETURN(-EFAULT);
6276         } else {
6277                 lsi = s2lsi(lmi->lmi_sb);
6278                 LASSERT(lsi->lsi_lmd);
6279                 /* CMD is supported only in IAM mode */
6280                 LASSERT(num);
6281                 rc = kstrtol(num, 10, &node_id);
6282                 if (rc)
6283                         RETURN(rc);
6284
6285                 obd_obt_init(obd);
6286                 if (test_bit(LMD_FLG_SKIP_LFSCK, lsi->lsi_lmd->lmd_flags))
6287                         m->mdt_skip_lfsck = 1;
6288                 if (test_bit(LMD_FLG_NO_CREATE, lsi->lsi_lmd->lmd_flags))
6289                         m->mdt_lut.lut_no_create = 1;
6290         }
6291
6292         /* Just try to get a DoM lock by default. Otherwise, having a group
6293          * lock granted, it may get blocked for a long time.
6294          */
6295         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
6296         /* DoM files are read at open and data is packed in the reply */
6297         m->mdt_dom_read_open = 1;
6298
6299         m->mdt_squash.rsi_uid = 0;
6300         m->mdt_squash.rsi_gid = 0;
6301         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
6302         spin_lock_init(&m->mdt_squash.rsi_lock);
6303         spin_lock_init(&m->mdt_lock);
6304         m->mdt_enable_chprojid_gid = 0;
6305         m->mdt_enable_dir_migration = 1;
6306         m->mdt_enable_dir_restripe = 0;
6307         m->mdt_enable_dir_auto_split = 0;
6308         m->mdt_enable_parallel_rename_dir = 1;
6309         m->mdt_enable_parallel_rename_file = 1;
6310         m->mdt_enable_parallel_rename_crossdir = 1;
6311         m->mdt_enable_remote_dir = 1;
6312         m->mdt_enable_remote_dir_gid = 0;
6313         m->mdt_enable_remote_rename = 1;
6314         m->mdt_enable_striped_dir = 1;
6315         m->mdt_enable_dmv_implicit_inherit = 1;
6316         m->mdt_dir_restripe_nsonly = 1;
6317         m->mdt_max_mod_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
6318
6319         atomic_set(&m->mdt_mds_mds_conns, 0);
6320         atomic_set(&m->mdt_async_commit_count, 0);
6321         atomic_set(&m->mdt_dmv_old_client_count, 0);
6322
6323         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6324         m->mdt_lu_dev.ld_obd = obd;
6325         /* Set this lu_device to obd for error handling purposes. */
6326         obd->obd_lu_dev = &m->mdt_lu_dev;
6327
6328         strncpy(m->mdt_job_xattr, XATTR_NAME_JOB_DEFAULT, XATTR_JOB_MAX_LEN);
6329
6330         /* init the stack */
6331         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6332         if (rc) {
6333                 CERROR("%s: Can't init device stack, rc %d\n",
6334                        mdt_obd_name(m), rc);
6335                 GOTO(err_lmi, rc);
6336         }
6337
6338         s = mdt_lu_site(m);
6339         ss_site = mdt_seq_site(m);
6340         s->ld_seq_site = ss_site;
6341         ss_site->ss_lu = s;
6342
6343         /* set server index */
6344         ss_site->ss_node_id = node_id;
6345
6346         /* failover is the default
6347          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6348          * assumed whose ss_node_id == 0 XXX
6349          */
6350         obd->obd_replayable = 1;
6351         /* No connection accepted until configurations will finish */
6352         obd->obd_no_conn = 1;
6353
6354         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6355                 char *str = lustre_cfg_string(cfg, 4);
6356
6357                 if (strchr(str, 'n')) {
6358                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6359                         obd->obd_replayable = 0;
6360                 }
6361         }
6362
6363         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6364         if (rc)
6365                 GOTO(err_fini_stack, rc);
6366
6367         rc = mdt_seq_init(env, m);
6368         if (rc)
6369                 GOTO(err_fini_fld, rc);
6370
6371         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6372                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6373         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6374                                               LDLM_NAMESPACE_SERVER,
6375                                               LDLM_NAMESPACE_GREEDY,
6376                                               LDLM_NS_TYPE_MDT);
6377         if (IS_ERR(m->mdt_namespace)) {
6378                 rc = PTR_ERR(m->mdt_namespace);
6379                 CERROR("%s: unable to create server namespace: rc = %d\n",
6380                        obd->obd_name, rc);
6381                 m->mdt_namespace = NULL;
6382                 GOTO(err_fini_seq, rc);
6383         }
6384
6385         m->mdt_namespace->ns_lvbp = m;
6386         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6387
6388         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6389         /* set obd_namespace for compatibility with old code */
6390         obd->obd_namespace = m->mdt_namespace;
6391
6392         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6393                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6394                       OBD_FAIL_MDS_ALL_REPLY_NET);
6395         if (rc)
6396                 GOTO(err_free_ns, rc);
6397
6398         /* Amount of available space excluded from granting and reserved
6399          * for metadata. It is a percentage of the total MDT size.
6400          */
6401         tgd->tgd_reserved_pcnt = 10;
6402
6403         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6404                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6405         else
6406                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6407
6408         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6409                 GOTO(err_tgt, rc = -ENOENT);
6410
6411         fid.f_seq = FID_SEQ_LOCAL_NAME;
6412         fid.f_oid = 1;
6413         fid.f_ver = 0;
6414         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6415         if (rc != 0)
6416                 GOTO(err_tgt, rc);
6417
6418         rc = mdt_hsm_cdt_init(m);
6419         if (rc != 0) {
6420                 CERROR("%s: error initializing coordinator, rc %d\n",
6421                        mdt_obd_name(m), rc);
6422                 GOTO(err_los_fini, rc);
6423         }
6424
6425         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6426
6427         next = m->mdt_child;
6428         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6429
6430         mntopts = dt_conf->ddp_mntopts;
6431
6432         if (mntopts & MNTOPT_USERXATTR)
6433                 m->mdt_opts.mo_user_xattr = 1;
6434         else
6435                 m->mdt_opts.mo_user_xattr = 0;
6436
6437         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6438
6439         if (mntopts & MNTOPT_ACL)
6440                 m->mdt_opts.mo_acl = 1;
6441         else
6442                 m->mdt_opts.mo_acl = 0;
6443
6444         m->mdt_enable_strict_som = 1;
6445
6446         /* XXX: to support suppgid for ACL, we enable identity_upcall
6447          * by default, otherwise, maybe got unexpected -EACCESS.
6448          */
6449         if (m->mdt_opts.mo_acl)
6450                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6451
6452         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6453                                                 identity_upcall,
6454                                                 UC_IDCACHE_HASH_SIZE,
6455                                                 1200, /* entry expire: 20 mn */
6456                                                 30, /* acquire expire: 30 s */
6457                                                 true, /* acquire can replay */
6458                                                 &mdt_identity_upcall_cache_ops);
6459         if (IS_ERR(m->mdt_identity_cache)) {
6460                 rc = PTR_ERR(m->mdt_identity_cache);
6461                 m->mdt_identity_cache = NULL;
6462                 GOTO(err_free_hsm, rc);
6463         }
6464
6465         rc = mdt_tunables_init(m, dev);
6466         if (rc) {
6467                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6468                 GOTO(err_recovery, rc);
6469         }
6470
6471         rc = mdt_quota_init(env, m, cfg);
6472         if (rc)
6473                 GOTO(err_procfs, rc);
6474
6475         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6476         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6477                            "mdt_ldlm_client", m->mdt_ldlm_client);
6478
6479         ping_evictor_start();
6480
6481         /* recovery will be started upon mdt_prepare() when the whole stack is
6482          * complete and ready to serve the requests
6483          */
6484
6485         /* Reduce the initial timeout on an MDS because it doesn't need such
6486          * a long timeout as an OST does. Adaptive timeouts will adjust this
6487          * value appropriately.
6488          */
6489         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6490                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6491
6492         if (test_bit(LMD_FLG_LOCAL_RECOV, lsi->lsi_lmd->lmd_flags))
6493                 m->mdt_lut.lut_local_recovery = 1;
6494
6495         rc = mdt_restriper_start(m);
6496         if (rc)
6497                 GOTO(err_ping_evictor, rc);
6498
6499         RETURN(0);
6500
6501 err_ping_evictor:
6502         ping_evictor_stop();
6503 err_procfs:
6504         mdt_tunables_fini(m);
6505 err_recovery:
6506         upcall_cache_cleanup(m->mdt_identity_cache);
6507         m->mdt_identity_cache = NULL;
6508 err_free_hsm:
6509         mdt_hsm_cdt_fini(m);
6510 err_los_fini:
6511         local_oid_storage_fini(env, m->mdt_los);
6512         m->mdt_los = NULL;
6513 err_tgt:
6514         /* keep recoverable clients */
6515         obd->obd_fail = 1;
6516         target_recovery_fini(obd);
6517         obd_exports_barrier(obd);
6518         obd_zombie_barrier();
6519         tgt_fini(env, &m->mdt_lut);
6520 err_free_ns:
6521         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6522         obd->obd_namespace = m->mdt_namespace = NULL;
6523 err_fini_seq:
6524         mdt_seq_fini(env, m);
6525 err_fini_fld:
6526         mdt_fld_fini(env, m);
6527 err_fini_stack:
6528         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6529 err_lmi:
6530         if (lmi)
6531                 server_put_mount(dev, true);
6532         return rc;
6533 }
6534
6535 /* For interoperability, the left element is old parameter, the right one
6536  * is the new version of the parameter, if some parameter is deprecated,
6537  * the new version should be set as NULL.
6538  */
6539 static struct cfg_interop_param mdt_interop_param[] = {
6540         { "mdt.group_upcall",   NULL },
6541         { "mdt.quota_type",     NULL },
6542         { "mdd.quota_type",     NULL },
6543         { "mdt.som",            NULL },
6544         { "mdt.rootsquash",     "mdt.root_squash" },
6545         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6546         { NULL }
6547 };
6548
6549 /* used by MGS to process specific configurations */
6550 static int mdt_process_config(const struct lu_env *env,
6551                               struct lu_device *d, struct lustre_cfg *cfg)
6552 {
6553         struct mdt_device *m = mdt_dev(d);
6554         struct md_device *md_next = m->mdt_child;
6555         struct lu_device *next = md2lu_dev(md_next);
6556         int rc;
6557
6558         ENTRY;
6559
6560         switch (cfg->lcfg_command) {
6561         case LCFG_PARAM: {
6562                 struct obd_device *obd = d->ld_obd;
6563                 /* For interoperability */
6564                 struct cfg_interop_param *ptr = NULL;
6565                 struct lustre_cfg *old_cfg = NULL;
6566                 char *param = NULL;
6567                 ssize_t count;
6568
6569                 param = lustre_cfg_string(cfg, 1);
6570                 if (param == NULL) {
6571                         CERROR("param is empty\n");
6572                         rc = -EINVAL;
6573                         break;
6574                 }
6575
6576                 ptr = class_find_old_param(param, mdt_interop_param);
6577                 if (ptr != NULL) {
6578                         if (ptr->new_param == NULL) {
6579                                 rc = 0;
6580                                 CWARN("For interoperability, skip this %s. It is obsolete.\n",
6581                                       ptr->old_param);
6582                                 break;
6583                         }
6584
6585                         CWARN("Found old param %s, changed it to %s.\n",
6586                               ptr->old_param, ptr->new_param);
6587
6588                         old_cfg = cfg;
6589                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6590                         if (IS_ERR(cfg)) {
6591                                 rc = PTR_ERR(cfg);
6592                                 break;
6593                         }
6594                 }
6595
6596                 count = class_modify_config(cfg, PARAM_MDT,
6597                                             &obd->obd_kset.kobj);
6598                 if (count < 0) {
6599                         struct coordinator *cdt = &m->mdt_coordinator;
6600
6601                         /* is it an HSM var ? */
6602                         count = class_modify_config(cfg, PARAM_HSM,
6603                                                     &cdt->cdt_hsm_kobj);
6604                         if (count < 0)
6605                                 /* we don't understand; pass it on */
6606                                 rc = next->ld_ops->ldo_process_config(env, next,
6607                                                                       cfg);
6608                         else
6609                                 rc = count > 0 ? 0 : count;
6610                 } else {
6611                         rc = count > 0 ? 0 : count;
6612                 }
6613
6614                 if (old_cfg)
6615                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6616                                                      cfg->lcfg_buflens));
6617                 break;
6618         }
6619         default:
6620                 /* others are passed further */
6621                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6622                 break;
6623         }
6624         RETURN(rc);
6625 }
6626
6627 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6628                                           const struct lu_object_header *hdr,
6629                                           struct lu_device *d)
6630 {
6631         struct mdt_object *mo;
6632
6633         ENTRY;
6634
6635         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6636         if (mo != NULL) {
6637                 struct lu_object *o;
6638                 struct lu_object_header *h;
6639
6640                 o = &mo->mot_obj;
6641                 h = &mo->mot_header;
6642                 lu_object_header_init(h);
6643                 lu_object_init(o, h, d);
6644                 lu_object_add_top(h, o);
6645                 o->lo_ops = &mdt_obj_ops;
6646                 spin_lock_init(&mo->mot_write_lock);
6647                 mutex_init(&mo->mot_som_mutex);
6648                 mutex_init(&mo->mot_lov_mutex);
6649                 init_rwsem(&mo->mot_dom_sem);
6650                 init_rwsem(&mo->mot_open_sem);
6651                 atomic_set(&mo->mot_open_count, 0);
6652                 mo->mot_restripe_offset = 0;
6653                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6654                 mo->mot_lsom_size = 0;
6655                 mo->mot_lsom_blocks = 0;
6656                 mo->mot_lsom_inited = false;
6657                 RETURN(o);
6658         }
6659         RETURN(NULL);
6660 }
6661
6662 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6663                            const struct lu_object_conf *unused)
6664 {
6665         struct mdt_device *d = mdt_dev(o->lo_dev);
6666         struct lu_device  *under;
6667         struct lu_object  *below;
6668         int                rc = 0;
6669
6670         ENTRY;
6671
6672         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6673                PFID(lu_object_fid(o)));
6674
6675         under = &d->mdt_child->md_lu_dev;
6676         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6677         if (below != NULL)
6678                 lu_object_add(o, below);
6679         else
6680                 rc = -ENOMEM;
6681
6682         RETURN(rc);
6683 }
6684
6685 static void mdt_object_free_rcu(struct rcu_head *head)
6686 {
6687         struct mdt_object *mo = container_of(head, struct mdt_object,
6688                                              mot_header.loh_rcu);
6689
6690         kmem_cache_free(mdt_object_kmem, mo);
6691 }
6692
6693 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6694 {
6695         struct mdt_object *mo = mdt_obj(o);
6696         struct lu_object_header *h;
6697
6698         ENTRY;
6699
6700         h = o->lo_header;
6701         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6702                PFID(lu_object_fid(o)));
6703
6704         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6705         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6706
6707         lu_object_fini(o);
6708         lu_object_header_fini(h);
6709         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6710         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6711
6712         EXIT;
6713 }
6714
6715 static int mdt_object_print(const struct lu_env *env, void *cookie,
6716                             lu_printer_t p, const struct lu_object *o)
6717 {
6718         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6719
6720         return (*p)(env, cookie,
6721                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6722                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6723                     mdto->mot_cache_attr ? "cache_attr" : "",
6724                     mdto->mot_write_count);
6725 }
6726
6727 static int mdt_prepare(const struct lu_env *env,
6728                 struct lu_device *pdev,
6729                 struct lu_device *cdev)
6730 {
6731         struct mdt_device *mdt = mdt_dev(cdev);
6732         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6733         struct obd_device *obd = cdev->ld_obd;
6734         int rc;
6735
6736         ENTRY;
6737
6738         LASSERT(obd);
6739
6740         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6741         if (rc)
6742                 RETURN(rc);
6743
6744         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6745         if (rc)
6746                 RETURN(rc);
6747
6748         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6749         if (rc)
6750                 RETURN(rc);
6751
6752         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6753         /* The LFSCK instance is registered just now, so it must be there when
6754          * register the namespace to such instance.
6755          */
6756         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6757
6758         if (mdt->mdt_seq_site.ss_node_id == 0) {
6759                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6760                                                          &mdt->mdt_md_root_fid);
6761                 if (rc)
6762                         RETURN(rc);
6763         }
6764
6765         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6766
6767         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6768         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6769         LASSERT(obd->obd_no_conn);
6770         spin_lock(&obd->obd_dev_lock);
6771         obd->obd_no_conn = 0;
6772         spin_unlock(&obd->obd_dev_lock);
6773
6774         if (obd->obd_recovering == 0)
6775                 mdt_postrecov(env, mdt);
6776
6777         RETURN(rc);
6778 }
6779
6780 const struct lu_device_operations mdt_lu_ops = {
6781         .ldo_object_alloc   = mdt_object_alloc,
6782         .ldo_process_config = mdt_process_config,
6783         .ldo_prepare        = mdt_prepare,
6784 };
6785
6786 static const struct lu_object_operations mdt_obj_ops = {
6787         .loo_object_init    = mdt_object_init,
6788         .loo_object_free    = mdt_object_free,
6789         .loo_object_print   = mdt_object_print
6790 };
6791
6792 static int mdt_obd_set_info_async(const struct lu_env *env,
6793                                   struct obd_export *exp,
6794                                   __u32 keylen, void *key,
6795                                   __u32 vallen, void *val,
6796                                   struct ptlrpc_request_set *set)
6797 {
6798         int rc;
6799
6800         ENTRY;
6801
6802         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6803                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6804                 RETURN(rc);
6805         }
6806
6807         RETURN(0);
6808 }
6809
6810 static inline void mdt_enable_slc(struct mdt_device *mdt)
6811 {
6812         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6813                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6814 }
6815
6816 static inline void mdt_disable_slc(struct mdt_device *mdt)
6817 {
6818         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6819                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6820 }
6821
6822 /**
6823  * Match client and server connection feature flags.
6824  *
6825  * Compute the compatibility flags for a connection request based on
6826  * features mutually supported by client and server.
6827  *
6828  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6829  * must not be updated here, otherwise a partially initialized value may
6830  * be exposed. After the connection request is successfully processed,
6831  * the top-level MDT connect request handler atomically updates the export
6832  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6833  * reply. \see mdt_connect().
6834  *
6835  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6836  * full struct obd_connect_data. So care must be taken when accessing fields
6837  * that are not present in struct obd_connect_data_v1. See LU-16.
6838  *
6839  * \param exp   the obd_export associated with this client/target pair
6840  * \param mdt   the target device for the connection
6841  * \param data  stores data for this connect request
6842  *
6843  * \retval 0       success
6844  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6845  * \retval -EBADE  client and server feature requirements are incompatible
6846  */
6847 static int mdt_connect_internal(const struct lu_env *env,
6848                                 struct obd_export *exp,
6849                                 struct mdt_device *mdt,
6850                                 struct obd_connect_data *data, bool reconnect)
6851 {
6852         const char *obd_name = mdt_obd_name(mdt);
6853
6854         LASSERT(data != NULL);
6855
6856         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6857
6858         if (mdt->mdt_bottom->dd_rdonly &&
6859             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6860             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6861                 RETURN(-EACCES);
6862
6863         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6864                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6865
6866         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6867
6868         if (!mdt->mdt_opts.mo_acl)
6869                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6870
6871         if (!mdt->mdt_opts.mo_user_xattr)
6872                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6873
6874         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6875                 data->ocd_brw_size = min(data->ocd_brw_size,
6876                                          mdt->mdt_brw_size);
6877                 if (data->ocd_brw_size == 0) {
6878                         CERROR("%s: cli %s/%p ocd_connect_flags: %#llx ocd_version: %x ocd_grant: %d ocd_index: %u ocd_brw_size unexpectedly zero, network data corruption? Refusing to connect this client\n", obd_name, exp->exp_client_uuid.uuid,
6879                                exp, data->ocd_connect_flags, data->ocd_version,
6880                                data->ocd_grant, data->ocd_index);
6881                         return -EPROTO;
6882                 }
6883         }
6884
6885         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6886                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6887
6888                 /* client is reporting its page size, for future use */
6889                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6890                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6891                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6892                  * it's LDISKFS_DIR_REC_LEN(20) = 28.
6893                  */
6894                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6895                 /* ocd_grant_tax_kb is in 1K byte blocks */
6896                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6897                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6898         }
6899
6900         /* Save connect_data we have so far because tgt_grant_connect()
6901          * uses it to calculate grant, and we want to save the client
6902          * version before it is overwritten by LUSTRE_VERSION_CODE.
6903          */
6904         exp->exp_connect_data = *data;
6905         if (OCD_HAS_FLAG(data, GRANT))
6906                 tgt_grant_connect(env, exp, data, !reconnect);
6907
6908         if (OCD_HAS_FLAG(data, MAXBYTES))
6909                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6910
6911         /* NB: Disregard the rule against updating
6912          * exp_connect_data.ocd_connect_flags in this case, since
6913          * tgt_client_new() needs to know if this is a lightweight
6914          * connection, and it is safe to expose this flag before
6915          * connection processing completes.
6916          */
6917         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6918                 spin_lock(&exp->exp_lock);
6919                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6920                 spin_unlock(&exp->exp_lock);
6921         }
6922
6923         data->ocd_version = LUSTRE_VERSION_CODE;
6924
6925         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6926                 CWARN("%s: MDS requires FID support, but client not\n",
6927                       obd_name);
6928                 return -EBADE;
6929         }
6930
6931         if (OCD_HAS_FLAG(data, PINGLESS)) {
6932                 if (ptlrpc_pinger_suppress_pings()) {
6933                         spin_lock(&exp->exp_obd->obd_dev_lock);
6934                         list_del_init(&exp->exp_obd_chain_timed);
6935                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6936                 } else {
6937                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6938                 }
6939         }
6940
6941         data->ocd_max_easize = mdt->mdt_max_ea_size;
6942
6943         /* NB: Disregard the rule against updating
6944          * exp_connect_data.ocd_connect_flags in this case, since
6945          * tgt_client_new() needs to know if this is client supports
6946          * multiple modify RPCs, and it is safe to expose this flag before
6947          * connection processing completes.
6948          */
6949         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6950                 if (mdt_max_mod_rpcs_changed(mdt))
6951                         /* The new mdt.*.max_mod_rpcs_in_flight parameter
6952                          * has not changed since initialization, but the
6953                          * deprecated module parameter was changed,
6954                          * so use that instead.
6955                          */
6956                         data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6957                 else
6958                         data->ocd_maxmodrpcs = mdt->mdt_max_mod_rpcs_in_flight;
6959                 spin_lock(&exp->exp_lock);
6960                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6961                 spin_unlock(&exp->exp_lock);
6962         }
6963
6964         if (OCD_HAS_FLAG(data, CKSUM)) {
6965                 __u32 cksum_types = data->ocd_cksum_types;
6966
6967                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6968
6969                 if (unlikely(data->ocd_cksum_types == 0)) {
6970                         CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n",
6971                                exp->exp_obd->obd_name);
6972                         RETURN(-EPROTO);
6973                 }
6974
6975                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return %x\n",
6976                        exp->exp_obd->obd_name, obd_export_nid2str(exp),
6977                        cksum_types, data->ocd_cksum_types);
6978         } else {
6979                 /* Client not support OBD_CONNECT_CKSUM? fall back to CRC32 */
6980                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support OBD_CONNECT_CKSUM, CRC32 will be used\n",
6981                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6982         }
6983
6984         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6985             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6986                 atomic_inc(&mdt->mdt_mds_mds_conns);
6987                 mdt_enable_slc(mdt);
6988         }
6989
6990         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6991                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6992
6993         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
6994             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT)) {
6995                 atomic_inc(&mdt->mdt_dmv_old_client_count);
6996                 mdt->mdt_enable_dmv_implicit_inherit = 0;
6997         }
6998
6999         return 0;
7000 }
7001
7002 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
7003                                    struct mdt_thread_info *info,
7004                                    struct mdt_file_data *mfd)
7005 {
7006         struct lu_context ses;
7007         int rc;
7008
7009         ENTRY;
7010
7011         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
7012         if (rc)
7013                 RETURN(rc);
7014
7015         env->le_ses = &ses;
7016         lu_context_enter(&ses);
7017
7018         mdt_ucred(info)->uc_valid = UCRED_OLD;
7019         /* do not let rbac interfere with dirty flag internal system event */
7020         mdt_ucred(info)->uc_rbac_file_perms = 1;
7021         mdt_ucred(info)->uc_rbac_dne_ops = 1;
7022         mdt_ucred(info)->uc_rbac_quota_ops = 1;
7023         mdt_ucred(info)->uc_rbac_byfid_ops = 1;
7024         mdt_ucred(info)->uc_rbac_chlg_ops = 1;
7025         mdt_ucred(info)->uc_rbac_fscrypt_admin = 1;
7026         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
7027
7028         lu_context_exit(&ses);
7029         lu_context_fini(&ses);
7030         env->le_ses = NULL;
7031
7032         RETURN(rc);
7033 }
7034
7035 static int mdt_export_cleanup(struct obd_export *exp)
7036 {
7037         LIST_HEAD(closing_list);
7038         struct mdt_export_data  *med = &exp->exp_mdt_data;
7039         struct obd_device       *obd = exp->exp_obd;
7040         struct mdt_device       *mdt;
7041         struct mdt_thread_info  *info;
7042         struct lu_env            env;
7043         struct mdt_file_data    *mfd, *n;
7044         int rc = 0;
7045
7046         ENTRY;
7047
7048         spin_lock(&med->med_open_lock);
7049         while (!list_empty(&med->med_open_head)) {
7050                 struct list_head *tmp = med->med_open_head.next;
7051
7052                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
7053
7054                 /* Remove mfd handle so it can't be found again.
7055                  * We are consuming the mfd_list reference here.
7056                  */
7057                 class_handle_unhash(&mfd->mfd_open_handle);
7058                 list_move_tail(&mfd->mfd_list, &closing_list);
7059         }
7060         spin_unlock(&med->med_open_lock);
7061         mdt = mdt_dev(obd->obd_lu_dev);
7062         LASSERT(mdt != NULL);
7063
7064         rc = lu_env_init(&env, LCT_MD_THREAD);
7065         if (rc)
7066                 RETURN(rc);
7067
7068         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7069         LASSERT(info != NULL);
7070         memset(info, 0, sizeof(*info));
7071         info->mti_env = &env;
7072         info->mti_mdt = mdt;
7073         info->mti_exp = exp;
7074
7075         if (!list_empty(&closing_list)) {
7076                 struct md_attr *ma = &info->mti_attr;
7077
7078                 /* Close any open files (which may cause orphan unlinking). */
7079                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
7080                         list_del_init(&mfd->mfd_list);
7081                         ma->ma_need = ma->ma_valid = 0;
7082
7083                         /* This file is being closed due to an eviction, it
7084                          * could have been modified and now dirty regarding to
7085                          * HSM archive, check this!
7086                          * The logic here is to mark a file dirty if there's a
7087                          * chance it was dirtied before the client was evicted,
7088                          * so that we don't have to wait for a release attempt
7089                          * before finding out the file was actually dirty and
7090                          * fail the release. Aggressively marking it dirty here
7091                          * will cause the policy engine to attempt to
7092                          * re-archive it; when rearchiving, we can compare the
7093                          * current version to the HSM data_version and make the
7094                          * archive request into a noop if it's not actually
7095                          * dirty.
7096                          */
7097                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
7098                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
7099
7100                         /* Don't unlink orphan on failover umount, LU-184 */
7101                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
7102                             exp->exp_obd->obd_stopping) {
7103                                 ma->ma_valid = MA_FLAGS;
7104                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
7105                         }
7106                         ma->ma_valid |= MA_FORCE_LOG;
7107                         mdt_mfd_close(info, mfd);
7108                 }
7109         }
7110         info->mti_mdt = NULL;
7111         /* cleanup client slot early */
7112         /* Do not erase record for recoverable client. */
7113         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
7114                 tgt_client_del(&env, exp);
7115         lu_env_fini(&env);
7116
7117         RETURN(rc);
7118 }
7119
7120 static int mdt_obd_disconnect(struct obd_export *exp)
7121 {
7122         struct obd_connect_data *data = &exp->exp_connect_data;
7123         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
7124         int rc;
7125
7126         ENTRY;
7127
7128         LASSERT(exp);
7129         class_export_get(exp);
7130
7131         if (OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7132             atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
7133                 mdt_disable_slc(mdt);
7134
7135         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7136             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT) &&
7137             atomic_dec_and_test(&mdt->mdt_dmv_old_client_count))
7138                 mdt->mdt_enable_dmv_implicit_inherit = 1;
7139
7140         rc = server_disconnect_export(exp);
7141         if (rc != 0)
7142                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
7143
7144         tgt_grant_discard(exp);
7145
7146         if (!(exp->exp_flags & OBD_OPT_FORCE))
7147                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7148
7149         rc = mdt_export_cleanup(exp);
7150         nodemap_del_member(exp);
7151         class_export_put(exp);
7152         RETURN(rc);
7153 }
7154
7155 /* mds_connect copy */
7156 static int mdt_obd_connect(const struct lu_env *env,
7157                            struct obd_export **exp, struct obd_device *obd,
7158                            struct obd_uuid *cluuid,
7159                            struct obd_connect_data *data,
7160                            void *localdata)
7161 {
7162         struct obd_export       *lexp;
7163         struct lustre_handle    conn = { 0 };
7164         struct mdt_device       *mdt;
7165         int                      rc;
7166         struct lnet_nid         *client_nid = localdata;
7167
7168         ENTRY;
7169
7170         LASSERT(env != NULL);
7171         LASSERT(data != NULL);
7172
7173         if (!exp || !obd || !cluuid)
7174                 RETURN(-EINVAL);
7175
7176         mdt = mdt_dev(obd->obd_lu_dev);
7177
7178         /*
7179          * first, check whether the stack is ready to handle requests
7180          * XXX: probably not very appropriate method is used now
7181          *      at some point we should find a better one
7182          */
7183         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
7184             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
7185             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
7186                 rc = obd_get_info(env, mdt->mdt_child_exp,
7187                                   sizeof(KEY_OSP_CONNECTED),
7188                                   KEY_OSP_CONNECTED, NULL, NULL);
7189                 if (rc)
7190                         RETURN(-EAGAIN);
7191                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
7192         }
7193
7194         rc = class_connect(&conn, obd, cluuid);
7195         if (rc)
7196                 RETURN(rc);
7197
7198         lexp = class_conn2export(&conn);
7199         LASSERT(lexp != NULL);
7200
7201         rc = nodemap_add_member(client_nid, lexp);
7202         if (rc != 0 && rc != -EEXIST)
7203                 GOTO(out, rc);
7204
7205         rc = mdt_connect_internal(env, lexp, mdt, data, false);
7206         if (rc == 0) {
7207                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
7208
7209                 LASSERT(lcd);
7210                 memcpy(lcd->lcd_uuid, cluuid, sizeof(lcd->lcd_uuid));
7211                 rc = tgt_client_new(env, lexp);
7212                 if (rc == 0)
7213                         mdt_export_stats_init(obd, lexp, localdata);
7214         }
7215 out:
7216         if (rc != 0) {
7217                 class_disconnect(lexp);
7218                 nodemap_del_member(lexp);
7219                 *exp = NULL;
7220         } else {
7221                 *exp = lexp;
7222                 /* Because we do not want this export to be evicted by pinger,
7223                  * let's not add this export to the timed chain list.
7224                  */
7225                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
7226                         spin_lock(&lexp->exp_obd->obd_dev_lock);
7227                         list_del_init(&lexp->exp_obd_chain_timed);
7228                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
7229                 }
7230         }
7231
7232         RETURN(rc);
7233 }
7234
7235 static int mdt_obd_reconnect(const struct lu_env *env,
7236                              struct obd_export *exp, struct obd_device *obd,
7237                              struct obd_uuid *cluuid,
7238                              struct obd_connect_data *data,
7239                              void *localdata)
7240 {
7241         struct lnet_nid *client_nid = localdata;
7242         int rc;
7243
7244         ENTRY;
7245
7246         if (exp == NULL || obd == NULL || cluuid == NULL)
7247                 RETURN(-EINVAL);
7248
7249         rc = nodemap_add_member(client_nid, exp);
7250         if (rc != 0 && rc != -EEXIST)
7251                 RETURN(rc);
7252
7253         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
7254                                   true);
7255         if (rc == 0)
7256                 mdt_export_stats_init(obd, exp, localdata);
7257         else
7258                 nodemap_del_member(exp);
7259
7260         RETURN(rc);
7261 }
7262
7263 /* FIXME: Can we avoid using these two interfaces? */
7264 static int mdt_init_export(struct obd_export *exp)
7265 {
7266         struct mdt_export_data *med = &exp->exp_mdt_data;
7267         int                     rc;
7268
7269         ENTRY;
7270
7271         INIT_LIST_HEAD(&med->med_open_head);
7272         spin_lock_init(&med->med_open_lock);
7273         spin_lock(&exp->exp_lock);
7274         exp->exp_connecting = 1;
7275         spin_unlock(&exp->exp_lock);
7276
7277         OBD_ALLOC(exp->exp_used_slots,
7278                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7279         if (exp->exp_used_slots == NULL)
7280                 RETURN(-ENOMEM);
7281
7282         /* self-export doesn't need client data and ldlm initialization */
7283         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7284                                      &exp->exp_client_uuid)))
7285                 RETURN(0);
7286
7287         rc = tgt_client_alloc(exp);
7288         if (rc)
7289                 GOTO(err, rc);
7290
7291         rc = ldlm_init_export(exp);
7292         if (rc)
7293                 GOTO(err_free, rc);
7294
7295         RETURN(rc);
7296
7297 err_free:
7298         tgt_client_free(exp);
7299 err:
7300         OBD_FREE(exp->exp_used_slots,
7301                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7302         exp->exp_used_slots = NULL;
7303
7304         CERROR("%s: Failed to initialize export: rc = %d\n",
7305                exp->exp_obd->obd_name, rc);
7306         return rc;
7307 }
7308
7309 static int mdt_destroy_export(struct obd_export *exp)
7310 {
7311         ENTRY;
7312
7313         target_destroy_export(exp);
7314         if (exp->exp_used_slots)
7315                 OBD_FREE(exp->exp_used_slots,
7316                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7317
7318         /* destroy can be called from failed obd_setup, so
7319          * checking uuid is safer than obd_self_export
7320          */
7321         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7322                                      &exp->exp_client_uuid)))
7323                 RETURN(0);
7324
7325         ldlm_destroy_export(exp);
7326         tgt_client_free(exp);
7327
7328         LASSERT(list_empty(&exp->exp_outstanding_replies));
7329         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
7330
7331         /*
7332          * discard grants once we're sure no more
7333          * interaction with the client is possible
7334          */
7335         tgt_grant_discard(exp);
7336         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
7337                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
7338
7339         if (!(exp->exp_flags & OBD_OPT_FORCE))
7340                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7341
7342         RETURN(0);
7343 }
7344
7345 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
7346                    struct linkea_data *ldata)
7347 {
7348         int rc;
7349
7350         LASSERT(ldata->ld_buf->lb_buf != NULL);
7351
7352         if (!mdt_object_exists(mdt_obj))
7353                 return -ENODATA;
7354
7355         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7356                           ldata->ld_buf, XATTR_NAME_LINK);
7357         if (rc == -ERANGE) {
7358                 /* Buf was too small, figure out what we need. */
7359                 lu_buf_free(ldata->ld_buf);
7360                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7361                                   ldata->ld_buf, XATTR_NAME_LINK);
7362                 if (rc < 0)
7363                         return rc;
7364                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7365                 if (ldata->ld_buf->lb_buf == NULL)
7366                         return -ENOMEM;
7367                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7368                                   ldata->ld_buf, XATTR_NAME_LINK);
7369         }
7370         if (rc < 0)
7371                 return rc;
7372
7373         return linkea_init_with_rec(ldata);
7374 }
7375
7376 /**
7377  * Given an MDT object, try to look up the full path to the object.
7378  * Part of the MDT layer implementation of lfs fid2path.
7379  *
7380  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7381  * \param[in]     obj   Object to do path lookup of
7382  * \param[in,out] fp    User-provided struct to store path information
7383  * \param[in]     root_fid Root FID of current path should reach
7384  *
7385  * \retval 0 Lookup successful, path information stored in fp
7386  * \retval -EAGAIN Lookup failed, usually because object is being moved
7387  * \retval negative errno if there was a problem
7388  */
7389 static int mdt_path_current(struct mdt_thread_info *info,
7390                             struct mdt_object *obj,
7391                             struct getinfo_fid2path *fp,
7392                             struct lu_fid *root_fid)
7393 {
7394         struct mdt_device *mdt = info->mti_mdt;
7395         struct lu_name *tmpname = &info->mti_name;
7396         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7397         struct lu_buf *buf = &info->mti_big_buf;
7398         struct linkea_data ldata = { NULL };
7399         bool first = true;
7400         struct mdt_object *mdt_obj;
7401         struct link_ea_header *leh;
7402         struct link_ea_entry *lee;
7403         bool worthchecking = true;
7404         bool needsfid = false;
7405         bool supported = false;
7406         int isenc = -1;
7407         char *ptr;
7408         int reclen;
7409         int rc = 0;
7410
7411         ENTRY;
7412
7413         /* temp buffer for path element, the buffer will be finally freed
7414          * in mdt_thread_info_fini
7415          */
7416         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7417         if (buf->lb_buf == NULL)
7418                 RETURN(-ENOMEM);
7419
7420         ldata.ld_buf = buf;
7421         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7422         *ptr = 0;
7423         --ptr;
7424         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7425
7426         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7427                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7428                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7429                         GOTO(out, rc = -ENOENT);
7430
7431                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7432                         mdt_obj = obj;
7433                         mdt_object_get(info->mti_env, mdt_obj);
7434                 } else {
7435                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7436                         if (IS_ERR(mdt_obj))
7437                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7438                 }
7439
7440                 if (!mdt_object_exists(mdt_obj)) {
7441                         mdt_object_put(info->mti_env, mdt_obj);
7442                         GOTO(out, rc = -ENOENT);
7443                 }
7444
7445                 if (mdt_object_remote(mdt_obj)) {
7446                         mdt_object_put(info->mti_env, mdt_obj);
7447                         GOTO(remote_out, rc = -EREMOTE);
7448                 }
7449
7450                 if (worthchecking) {
7451                         /* need to know if FID being looked up is encrypted */
7452                         struct lu_attr la = { 0 };
7453                         struct dt_object *dt = mdt_obj2dt(mdt_obj);
7454
7455                         if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7456                                 dt_attr_get(info->mti_env, dt, &la);
7457                         if (la.la_valid & LA_FLAGS &&
7458                             la.la_flags & LUSTRE_ENCRYPT_FL) {
7459                                 if (!supported && mdt_info_req(info) &&
7460                                     !exp_connect_encrypt_fid2path(
7461                                             mdt_info_req(info)->rq_export)) {
7462                                         /* client does not support fid2path
7463                                          * for encrypted files
7464                                          */
7465                                         mdt_object_put(info->mti_env, mdt_obj);
7466                                         GOTO(out, rc = -ENODATA);
7467                                 } else {
7468                                         supported = true;
7469                                 }
7470                                 needsfid = true;
7471                                 if (isenc == -1)
7472                                         isenc = 1;
7473                         } else {
7474                                 worthchecking = false;
7475                                 needsfid = false;
7476                                 if (isenc == -1)
7477                                         isenc = 0;
7478                         }
7479                 }
7480
7481                 rc = mdt_links_read(info, mdt_obj, &ldata);
7482                 if (rc != 0) {
7483                         mdt_object_put(info->mti_env, mdt_obj);
7484                         GOTO(out, rc);
7485                 }
7486
7487                 leh = buf->lb_buf;
7488                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7489                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7490                 /* If set, use link #linkno for path lookup, otherwise use
7491                  *  link #0.  Only do this for the final path ement.
7492                  */
7493                 if (first && fp->gf_linkno < leh->leh_reccount) {
7494                         int count;
7495
7496                         for (count = 0; count < fp->gf_linkno; count++) {
7497                                 lee = (struct link_ea_entry *)
7498                                      ((char *)lee + reclen);
7499                                 linkea_entry_unpack(lee, &reclen, tmpname,
7500                                                     tmpfid);
7501                         }
7502                         if (fp->gf_linkno < leh->leh_reccount - 1)
7503                                 /* indicate to user there are more links */
7504                                 fp->gf_linkno++;
7505                 }
7506
7507                 /* Check if it is slave stripes */
7508                 rc = mdt_is_dir_stripe(info, mdt_obj);
7509                 mdt_object_put(info->mti_env, mdt_obj);
7510                 if (rc < 0)
7511                         GOTO(out, rc);
7512                 if (rc == 1) {
7513                         fp->gf_fid = *tmpfid;
7514                         continue;
7515                 }
7516
7517                 /* Pack the name in the end of the buffer */
7518                 ptr -= tmpname->ln_namelen;
7519                 if (ptr - 1 <= fp->gf_u.gf_path)
7520                         GOTO(out, rc = -ENAMETOOLONG);
7521                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7522                 if (needsfid) {
7523                         /* Pack FID before file name, so that client can build
7524                          * encoded/digested form.
7525                          */
7526                         char fidstr[FID_LEN + 1];
7527
7528                         snprintf(fidstr, sizeof(fidstr), DFID,
7529                                  PFID(&fp->gf_fid));
7530                         ptr -= strlen(fidstr);
7531                         if (ptr - 1 <= fp->gf_u.gf_path)
7532                                 GOTO(out, rc = -ENAMETOOLONG);
7533                         strncpy(ptr, fidstr, strlen(fidstr));
7534                 }
7535                 *(--ptr) = '/';
7536
7537                 /* keep the last resolved fid to the client, so the client will
7538                  * build the left path on another MDT for remote object
7539                  */
7540                 fp->gf_fid = *tmpfid;
7541
7542                 first = false;
7543         }
7544
7545         /* non-zero will be treated as an error */
7546         rc = 0;
7547
7548 remote_out:
7549         if (isenc != 1)
7550                 ptr++; /* skip leading / unless this is an encrypted file */
7551         memmove(fp->gf_u.gf_path, ptr,
7552                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7553
7554 out:
7555         RETURN(rc);
7556 }
7557
7558 /**
7559  * Given an MDT object, use mdt_path_current to get the path.
7560  * Essentially a wrapper to retry mdt_path_current a set number of times
7561  * if -EAGAIN is returned (usually because an object is being moved).
7562  *
7563  * Part of the MDT layer implementation of lfs fid2path.
7564  *
7565  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7566  * \param[in]     obj   Object to do path lookup of
7567  * \param[in,out] fp    User-provided struct for arguments and to store path
7568  *                      information
7569  *
7570  * \retval 0 Lookup successful, path information stored in fp
7571  * \retval negative errno if there was a problem
7572  */
7573 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7574                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7575 {
7576         struct mdt_device       *mdt = info->mti_mdt;
7577         int                     tries = 3;
7578         int                     rc = -EAGAIN;
7579
7580         ENTRY;
7581
7582         if (fp->gf_pathlen < 3)
7583                 RETURN(-EOVERFLOW);
7584
7585         if (root_fid == NULL)
7586                 root_fid = &mdt->mdt_md_root_fid;
7587
7588         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7589                 fp->gf_u.gf_path[0] = '\0';
7590                 RETURN(0);
7591         }
7592
7593         /* Retry multiple times in case file is being moved */
7594         while (tries-- && rc == -EAGAIN)
7595                 rc = mdt_path_current(info, obj, fp, root_fid);
7596
7597         RETURN(rc);
7598 }
7599
7600 /**
7601  * Get the full path of the provided FID, as of changelog record recno.
7602  *
7603  * This checks sanity and looks up object for user provided FID
7604  * before calling the actual path lookup code.
7605  *
7606  * Part of the MDT layer implementation of lfs fid2path.
7607  *
7608  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7609  * \param[in,out] fp    User-provided struct for arguments and to store path
7610  *                      information
7611  *
7612  * \retval 0 Lookup successful, path information and recno stored in fp
7613  * \retval -ENOENT, object does not exist
7614  * \retval negative errno if there was a problem
7615  */
7616 static int mdt_fid2path(struct mdt_thread_info *info,
7617                         struct lu_fid *root_fid,
7618                         struct getinfo_fid2path *fp)
7619 {
7620         struct mdt_device *mdt = info->mti_mdt;
7621         struct mdt_object *obj;
7622         int    rc;
7623
7624         ENTRY;
7625
7626         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7627                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7628
7629         if (!fid_is_sane(&fp->gf_fid))
7630                 RETURN(-EINVAL);
7631
7632         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7633                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx, or f_oid != 0, or f_ver == 0\n",
7634                        mdt_obd_name(mdt),
7635                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7636                 RETURN(-EINVAL);
7637         }
7638
7639         /* return error if client-provided root fid is not the one stored in
7640          * the export
7641          */
7642         if (root_fid && !fid_is_zero(&info->mti_exp->exp_root_fid) &&
7643             !lu_fid_eq(root_fid, &info->mti_exp->exp_root_fid)) {
7644                 CDEBUG(D_INFO,
7645                        "%s: root fid from client "DFID" but "DFID" stored in export\n",
7646                        mdt_obd_name(mdt), PFID(root_fid),
7647                        PFID(&info->mti_exp->exp_root_fid));
7648                 RETURN(-EXDEV);
7649         }
7650
7651         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7652         if (IS_ERR(obj)) {
7653                 rc = PTR_ERR(obj);
7654                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7655                        PFID(&fp->gf_fid), rc);
7656                 RETURN(rc);
7657         }
7658
7659         if (mdt_object_remote(obj))
7660                 rc = -EREMOTE;
7661         else if (!mdt_object_exists(obj))
7662                 rc = -ENOENT;
7663         else
7664                 rc = 0;
7665
7666         if (rc < 0) {
7667                 mdt_object_put(info->mti_env, obj);
7668                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7669                        PFID(&fp->gf_fid), rc);
7670                 RETURN(rc);
7671         }
7672
7673         rc = mdt_path(info, obj, fp, root_fid);
7674
7675         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7676                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7677                fp->gf_recno, fp->gf_linkno);
7678
7679         mdt_object_put(info->mti_env, obj);
7680
7681         RETURN(rc);
7682 }
7683
7684 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7685                             void *val, int vallen)
7686 {
7687         struct getinfo_fid2path *fpout, *fpin;
7688         struct lu_fid *root_fid = NULL;
7689         int rc = 0;
7690
7691         fpin = key + round_up(sizeof(KEY_FID2PATH), 8);
7692         fpout = val;
7693
7694         if (req_capsule_req_need_swab(info->mti_pill))
7695                 lustre_swab_fid2path(fpin);
7696
7697         memcpy(fpout, fpin, sizeof(*fpin));
7698         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7699                 RETURN(-EINVAL);
7700
7701         if (keylen >= round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*fpin) +
7702                       sizeof(struct lu_fid)) {
7703                 /* client sent its root FID, which is normally fileset FID */
7704                 root_fid = fpin->gf_u.gf_root_fid;
7705                 if (req_capsule_req_need_swab(info->mti_pill))
7706                         lustre_swab_lu_fid(root_fid);
7707
7708                 if (root_fid != NULL && !fid_is_sane(root_fid))
7709                         RETURN(-EINVAL);
7710         }
7711
7712         rc = mdt_fid2path(info, root_fid, fpout);
7713         RETURN(rc);
7714 }
7715
7716 int mdt_get_info(struct tgt_session_info *tsi)
7717 {
7718         char    *key;
7719         int      keylen;
7720         __u32   *vallen;
7721         void    *valout;
7722         int      rc;
7723
7724         ENTRY;
7725
7726         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7727         if (key == NULL) {
7728                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7729                 RETURN(err_serious(-EFAULT));
7730         }
7731         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7732                                       RCL_CLIENT);
7733
7734         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7735         if (vallen == NULL) {
7736                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7737                                 tgt_name(tsi->tsi_tgt));
7738                 RETURN(err_serious(-EFAULT));
7739         }
7740
7741         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7742                              *vallen);
7743         rc = req_capsule_server_pack(tsi->tsi_pill);
7744         if (rc)
7745                 RETURN(err_serious(rc));
7746
7747         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7748         if (valout == NULL) {
7749                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7750                                 tgt_name(tsi->tsi_tgt));
7751                 RETURN(err_serious(-EFAULT));
7752         }
7753
7754         if (KEY_IS(KEY_FID2PATH)) {
7755                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7756
7757                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7758                 mdt_thread_info_fini(info);
7759         } else {
7760                 rc = -EINVAL;
7761         }
7762         RETURN(rc);
7763 }
7764
7765 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7766 {
7767         struct obd_ioctl_data *data = karg;
7768         struct lu_fid *fid;
7769         __u64 version;
7770         struct mdt_object *obj;
7771         struct mdt_lock_handle  *lh;
7772         int rc;
7773
7774         ENTRY;
7775
7776         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7777             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7778                 RETURN(-EINVAL);
7779
7780         fid = (struct lu_fid *)data->ioc_inlbuf1;
7781
7782         if (!fid_is_sane(fid))
7783                 RETURN(-EINVAL);
7784
7785         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7786
7787         lh = &mti->mti_lh[MDT_LH_PARENT];
7788         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE, LCK_CR);
7789         if (IS_ERR(obj))
7790                 RETURN(PTR_ERR(obj));
7791
7792         if (mdt_object_remote(obj)) {
7793                 rc = -EREMOTE;
7794                 /**
7795                  * before calling version get the correct MDS should be
7796                  * fid, this is error to find remote object here
7797                  */
7798                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7799         } else if (!mdt_object_exists(obj)) {
7800                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7801                 rc = -ENOENT;
7802         } else {
7803                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7804                *(__u64 *)data->ioc_inlbuf2 = version;
7805                 rc = 0;
7806         }
7807         mdt_object_unlock_put(mti, obj, lh, 1);
7808         RETURN(rc);
7809 }
7810
7811 /* ioctls on obd dev */
7812 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7813                          void *karg, void __user *uarg)
7814 {
7815         struct obd_device *obd = exp->exp_obd;
7816         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7817         struct dt_device *dt = mdt->mdt_bottom;
7818         struct obd_ioctl_data *data;
7819         struct lu_env env;
7820         int rc;
7821
7822         ENTRY;
7823         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
7824                obd->obd_name, cmd, len, karg, uarg);
7825
7826         rc = lu_env_init(&env, LCT_MD_THREAD);
7827         if (rc)
7828                 RETURN(rc);
7829
7830         /* handle commands that don't use @karg first */
7831         switch (cmd) {
7832         case OBD_IOC_SYNC:
7833                 rc = mdt_device_sync(&env, mdt);
7834                 GOTO(out, rc);
7835         case OBD_IOC_SET_READONLY:
7836                 rc = dt_sync(&env, dt);
7837                 if (rc == 0)
7838                         rc = dt_ro(&env, dt);
7839                 GOTO(out, rc);
7840         }
7841
7842         if (unlikely(karg == NULL)) {
7843                 OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc = -EINVAL);
7844                 GOTO(out, rc);
7845         }
7846         data = karg;
7847
7848         switch (cmd) {
7849         case OBD_IOC_ABORT_RECOVERY: {
7850                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7851                         LCONSOLE_WARN("%s: Aborting MDT recovery\n",
7852                                       obd->obd_name);
7853                         obd->obd_abort_mdt_recovery = 1;
7854                         wake_up(&obd->obd_next_transno_waitq);
7855                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7856                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7857                         LCONSOLE_WARN("%s: Aborting client recovery\n",
7858                                       obd->obd_name);
7859                         obd->obd_abort_recovery = 1;
7860                         target_stop_recovery_thread(obd);
7861                 }
7862                 rc = 0;
7863                 break;
7864         }
7865         case OBD_IOC_CHANGELOG_REG:
7866         case OBD_IOC_CHANGELOG_DEREG:
7867         case OBD_IOC_CHANGELOG_CLEAR:
7868         case OBD_IOC_LLOG_PRINT:
7869         case OBD_IOC_LLOG_CANCEL:
7870                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env, mdt->mdt_child,
7871                                                            cmd, len, karg);
7872                 break;
7873         case OBD_IOC_START_LFSCK: {
7874                 struct md_device *next = mdt->mdt_child;
7875                 struct lfsck_start_param lsp;
7876
7877                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7878                 lsp.lsp_index_valid = 0;
7879                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7880                 break;
7881         }
7882         case OBD_IOC_STOP_LFSCK: {
7883                 struct md_device *next = mdt->mdt_child;
7884                 struct lfsck_stop stop;
7885
7886                 stop.ls_status = LS_STOPPED;
7887                 /* Old lfsck utils may pass NULL @stop. */
7888                 if (data->ioc_inlbuf1 == NULL)
7889                         stop.ls_flags = 0;
7890                 else
7891                         stop.ls_flags =
7892                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7893
7894                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7895                 break;
7896         }
7897         case OBD_IOC_QUERY_LFSCK: {
7898                 struct md_device *next = mdt->mdt_child;
7899
7900                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7901                                                  data->ioc_inlbuf1);
7902                 break;
7903         }
7904         case OBD_IOC_GET_OBJ_VERSION: {
7905                 struct mdt_thread_info *mti;
7906
7907                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7908                 memset(mti, 0, sizeof(*mti));
7909                 mti->mti_env = &env;
7910                 mti->mti_mdt = mdt;
7911                 mti->mti_exp = exp;
7912
7913                 rc = mdt_ioc_version_get(mti, karg);
7914                 break;
7915         }
7916         case OBD_IOC_CATLOGLIST: {
7917                 struct mdt_thread_info *mti;
7918
7919                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7920                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7921                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7922                                        &mti->mti_tmp_fid1);
7923                 break;
7924         }
7925         default:
7926                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
7927                 break;
7928         }
7929 out:
7930         lu_env_fini(&env);
7931         RETURN(rc);
7932 }
7933
7934 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7935 {
7936         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7937         int rc;
7938
7939         ENTRY;
7940
7941         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7942                 struct lfsck_start_param lsp;
7943
7944                 lsp.lsp_start = NULL;
7945                 lsp.lsp_index_valid = 0;
7946                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7947                                                            OBD_IOC_START_LFSCK,
7948                                                            0, &lsp);
7949                 if (rc != 0 && rc != -EALREADY)
7950                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7951                               mdt_obd_name(mdt), rc);
7952         }
7953
7954         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7955         RETURN(rc);
7956 }
7957
7958 static int mdt_obd_postrecov(struct obd_device *obd)
7959 {
7960         struct lu_env env;
7961         int rc;
7962
7963         rc = lu_env_init(&env, LCT_MD_THREAD);
7964         if (rc)
7965                 RETURN(rc);
7966         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7967         lu_env_fini(&env);
7968         return rc;
7969 }
7970
7971 static const struct obd_ops mdt_obd_device_ops = {
7972         .o_owner          = THIS_MODULE,
7973         .o_set_info_async = mdt_obd_set_info_async,
7974         .o_connect        = mdt_obd_connect,
7975         .o_reconnect      = mdt_obd_reconnect,
7976         .o_disconnect     = mdt_obd_disconnect,
7977         .o_init_export    = mdt_init_export,
7978         .o_destroy_export = mdt_destroy_export,
7979         .o_iocontrol      = mdt_iocontrol,
7980         .o_postrecov      = mdt_obd_postrecov,
7981         /* Data-on-MDT IO methods */
7982         .o_preprw         = mdt_obd_preprw,
7983         .o_commitrw       = mdt_obd_commitrw,
7984 };
7985
7986 static struct lu_device *mdt_device_fini(const struct lu_env *env,
7987                                          struct lu_device *d)
7988 {
7989         struct mdt_device *m = mdt_dev(d);
7990
7991         ENTRY;
7992
7993         mdt_fini(env, m);
7994         RETURN(NULL);
7995 }
7996
7997 static struct lu_device *mdt_device_free(const struct lu_env *env,
7998                                          struct lu_device *d)
7999 {
8000         struct mdt_device *m = mdt_dev(d);
8001
8002         ENTRY;
8003
8004         lu_device_fini(&m->mdt_lu_dev);
8005         OBD_FREE_PTR(m);
8006
8007         RETURN(NULL);
8008 }
8009
8010 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
8011                                           struct lu_device_type *t,
8012                                           struct lustre_cfg *cfg)
8013 {
8014         struct lu_device  *l;
8015         struct mdt_device *m;
8016
8017         OBD_ALLOC_PTR(m);
8018         if (m != NULL) {
8019                 int rc;
8020
8021                 l = &m->mdt_lu_dev;
8022                 rc = mdt_init0(env, m, t, cfg);
8023                 if (rc != 0) {
8024                         mdt_device_free(env, l);
8025                         l = ERR_PTR(rc);
8026                         return l;
8027                 }
8028         } else
8029                 l = ERR_PTR(-ENOMEM);
8030         return l;
8031 }
8032
8033 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
8034 LU_KEY_INIT(mdt, struct mdt_thread_info);
8035
8036 static void mdt_key_fini(const struct lu_context *ctx,
8037                          struct lu_context_key *key, void *data)
8038 {
8039         struct mdt_thread_info *info = data;
8040
8041         if (info->mti_big_lmm) {
8042                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
8043                 info->mti_big_lmm = NULL;
8044                 info->mti_big_lmmsize = 0;
8045         }
8046
8047         if (info->mti_big_acl) {
8048                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
8049                 info->mti_big_acl = NULL;
8050                 info->mti_big_aclsize = 0;
8051         }
8052
8053         OBD_FREE_PTR(info);
8054 }
8055
8056 /* context key: mdt_thread_key */
8057 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
8058
8059 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
8060 {
8061         return lu_ucred(info->mti_env);
8062 }
8063
8064 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
8065 {
8066         return lu_ucred_check(info->mti_env);
8067 }
8068
8069 /**
8070  * Enable/disable COS (Commit On Sharing).
8071  *
8072  * Set/Clear the COS flag in mdt options.
8073  *
8074  * \param mdt mdt device
8075  * \param val 0 disables COS, other values enable COS
8076  */
8077 void mdt_enable_cos(struct mdt_device *mdt, bool val)
8078 {
8079         struct lu_env env;
8080         int rc;
8081
8082         mdt->mdt_opts.mo_cos = val;
8083         rc = lu_env_init(&env, LCT_LOCAL);
8084         if (unlikely(rc != 0)) {
8085                 CWARN("%s: lu_env initialization failed, cannot sync: rc = %d\n",
8086                       mdt_obd_name(mdt), rc);
8087                 return;
8088         }
8089         mdt_device_sync(&env, mdt);
8090         lu_env_fini(&env);
8091 }
8092
8093 /**
8094  * Check COS (Commit On Sharing) status.
8095  *
8096  * Return COS flag status.
8097  *
8098  * \param mdt mdt device
8099  */
8100 int mdt_cos_is_enabled(struct mdt_device *mdt)
8101 {
8102         return mdt->mdt_opts.mo_cos != 0;
8103 }
8104
8105 static const struct lu_device_type_operations mdt_device_type_ops = {
8106         .ldto_device_alloc = mdt_device_alloc,
8107         .ldto_device_free  = mdt_device_free,
8108         .ldto_device_fini  = mdt_device_fini
8109 };
8110
8111 static struct lu_device_type mdt_device_type = {
8112         .ldt_tags     = LU_DEVICE_MD,
8113         .ldt_name     = LUSTRE_MDT_NAME,
8114         .ldt_ops      = &mdt_device_type_ops,
8115         .ldt_ctx_tags = LCT_MD_THREAD
8116 };
8117
8118 static int __init mdt_init(void)
8119 {
8120         int rc;
8121
8122         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
8123                      FID_NOBRACE_LEN + 1);
8124         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
8125                      FID_LEN + 1);
8126
8127         rc = libcfs_setup();
8128         if (rc)
8129                 return rc;
8130
8131         rc = lu_kmem_init(mdt_caches);
8132         if (rc)
8133                 return rc;
8134
8135         rc = mds_mod_init();
8136         if (rc)
8137                 GOTO(lu_fini, rc);
8138
8139         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
8140                                  LUSTRE_MDT_NAME, &mdt_device_type);
8141         if (rc)
8142                 GOTO(mds_fini, rc);
8143 lu_fini:
8144         if (rc)
8145                 lu_kmem_fini(mdt_caches);
8146 mds_fini:
8147         if (rc)
8148                 mds_mod_exit();
8149         return rc;
8150 }
8151
8152 static void __exit mdt_exit(void)
8153 {
8154         class_unregister_type(LUSTRE_MDT_NAME);
8155         mds_mod_exit();
8156         lu_kmem_fini(mdt_caches);
8157 }
8158
8159 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8160 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
8161 MODULE_VERSION(LUSTRE_VERSION_STRING);
8162 MODULE_LICENSE("GPL");
8163
8164 module_init(mdt_init);
8165 module_exit(mdt_exit);