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), 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                 if (S_ISREG(lu_object_attr(&child->mot_obj)) &&
2469                     !mdt_object_remote(child) && child != parent) {
2470                         mdt_object_put(info->mti_env, child);
2471                         rc = mdt_pack_size2body(info, child_fid,
2472                                                 &lhc->mlh_reg_lh);
2473                         if (rc != 0 && child_bits & MDS_INODELOCK_DOM) {
2474                                 /* DOM lock was taken in advance but this is
2475                                  * not DoM file. Drop the lock.
2476                                  */
2477                                 lock_res_and_lock(lock);
2478                                 ldlm_inodebits_drop(lock, MDS_INODELOCK_DOM);
2479                                 unlock_res_and_lock(lock);
2480                         }
2481                         LDLM_LOCK_PUT(lock);
2482                         GOTO(unlock_parent, rc = 0);
2483                 }
2484                 LDLM_LOCK_PUT(lock);
2485         }
2486
2487         EXIT;
2488 out_child:
2489         if (child)
2490                 mdt_object_put(info->mti_env, child);
2491 unlock_parent:
2492         if (lhp)
2493                 mdt_object_unlock(info, parent, lhp, 1);
2494         if (rc == -ENOENT) {
2495                 /* return -ENOKEY instead of -ENOENT to encryption-unaware
2496                  * client if trying to access an encrypted file
2497                  */
2498                 int rc2 = mdt_check_enc(info, parent);
2499
2500                 if (rc2)
2501                         rc = rc2;
2502         }
2503         return rc;
2504 }
2505
2506 /* normal handler: should release the child lock */
2507 static int mdt_getattr_name(struct tgt_session_info *tsi)
2508 {
2509         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
2510         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_CHILD];
2511         struct mdt_body *reqbody;
2512         struct mdt_body *repbody;
2513         int rc, rc2;
2514
2515         ENTRY;
2516
2517         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
2518         LASSERT(reqbody != NULL);
2519         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
2520         LASSERT(repbody != NULL);
2521
2522         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
2523         repbody->mbo_eadatasize = 0;
2524         repbody->mbo_aclsize = 0;
2525
2526         rc = mdt_init_ucred(info, reqbody);
2527         if (unlikely(rc))
2528                 GOTO(out_shrink, rc);
2529
2530         rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL);
2531         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
2532                 ldlm_lock_decref(&lhc->mlh_reg_lh, lhc->mlh_reg_mode);
2533                 lhc->mlh_reg_lh.cookie = 0;
2534         }
2535         mdt_exit_ucred(info);
2536         EXIT;
2537 out_shrink:
2538         mdt_client_compatibility(info);
2539         rc2 = mdt_fix_reply(info);
2540         if (rc == 0)
2541                 rc = rc2;
2542         mdt_thread_info_fini(info);
2543         return rc;
2544 }
2545
2546 static int mdt_rmfid_unlink(struct mdt_thread_info *info,
2547                             const struct lu_fid *pfid,
2548                             const struct lu_name *name,
2549                             struct mdt_object *obj, s64 ctime)
2550 {
2551         struct lu_fid *child_fid = &info->mti_tmp_fid1;
2552         struct ldlm_enqueue_info *einfo = &info->mti_einfo;
2553         struct mdt_device *mdt = info->mti_mdt;
2554         struct md_attr *ma = &info->mti_attr;
2555         struct mdt_lock_handle *parent_lh;
2556         struct mdt_lock_handle *child_lh;
2557         struct mdt_object *pobj;
2558         int rc;
2559
2560         ENTRY;
2561
2562         pobj = mdt_object_find(info->mti_env, mdt, pfid);
2563         if (IS_ERR(pobj))
2564                 GOTO(out, rc = PTR_ERR(pobj));
2565
2566         parent_lh = &info->mti_lh[MDT_LH_PARENT];
2567         rc = mdt_parent_lock(info, pobj, parent_lh, name, LCK_PW);
2568         if (rc != 0)
2569                 GOTO(put_parent, rc);
2570
2571         rc = mdo_lookup(info->mti_env, mdt_object_child(pobj),
2572                         name, child_fid, &info->mti_spec);
2573         if (rc != 0)
2574                 GOTO(unlock_parent, rc);
2575
2576         if (!lu_fid_eq(child_fid, mdt_object_fid(obj)))
2577                 GOTO(unlock_parent, rc = -EREMCHG);
2578
2579         child_lh = &info->mti_lh[MDT_LH_CHILD];
2580         rc = mdt_object_stripes_lock(info, pobj, obj, child_lh, einfo,
2581                                      MDS_INODELOCK_LOOKUP |
2582                                      MDS_INODELOCK_UPDATE, LCK_EX);
2583         if (rc != 0)
2584                 GOTO(unlock_parent, rc);
2585
2586         if (atomic_read(&obj->mot_open_count)) {
2587                 CDEBUG(D_OTHER, "object "DFID" open, skip\n",
2588                        PFID(mdt_object_fid(obj)));
2589                 GOTO(unlock_child, rc = -EBUSY);
2590         }
2591
2592         ma->ma_need = 0;
2593         ma->ma_valid = MA_INODE;
2594         ma->ma_attr.la_valid = LA_CTIME;
2595         ma->ma_attr.la_ctime = ctime;
2596
2597         mutex_lock(&obj->mot_lov_mutex);
2598
2599         rc = mdo_unlink(info->mti_env, mdt_object_child(pobj),
2600                         mdt_object_child(obj), name, ma, 0);
2601
2602         mutex_unlock(&obj->mot_lov_mutex);
2603
2604 unlock_child:
2605         mdt_object_stripes_unlock(info, obj, child_lh, einfo, 1);
2606 unlock_parent:
2607         mdt_object_unlock(info, pobj, parent_lh, 1);
2608 put_parent:
2609         mdt_object_put(info->mti_env, pobj);
2610 out:
2611         RETURN(rc);
2612 }
2613
2614 static int mdt_rmfid_check_permission(struct mdt_thread_info *info,
2615                                         struct mdt_object *obj)
2616 {
2617         struct lu_ucred *uc = lu_ucred(info->mti_env);
2618         struct md_attr *ma = &info->mti_attr;
2619         struct lu_attr *la = &ma->ma_attr;
2620         int rc = 0;
2621
2622         ENTRY;
2623
2624         ma->ma_need = MA_INODE;
2625         rc = mo_attr_get(info->mti_env, mdt_object_child(obj), ma);
2626         if (rc)
2627                 GOTO(out, rc);
2628
2629         if (la->la_flags & LUSTRE_IMMUTABLE_FL)
2630                 rc = -EACCES;
2631
2632         /* we want rbac roles to have precedence over any other
2633          * permission or capability checks
2634          */
2635         if (!uc->uc_rbac_byfid_ops)
2636                 RETURN(-EPERM);
2637         if (cap_raised(uc->uc_cap, CAP_DAC_OVERRIDE))
2638                 RETURN(0);
2639         if (uc->uc_fsuid == la->la_uid) {
2640                 if ((la->la_mode & 0200) == 0)
2641                         rc = -EACCES;
2642         } else if (uc->uc_fsgid == la->la_gid) {
2643                 if ((la->la_mode & 0020) == 0)
2644                         rc = -EACCES;
2645         } else if ((la->la_mode & 0002) == 0) {
2646                         rc = -EACCES;
2647         }
2648
2649 out:
2650         RETURN(rc);
2651 }
2652
2653 static int mdt_rmfid_one(struct mdt_thread_info *info, struct lu_fid *fid,
2654                          s64 ctime)
2655 {
2656         struct mdt_device *mdt = info->mti_mdt;
2657         struct mdt_object *obj = NULL;
2658         struct linkea_data ldata = { NULL };
2659         struct lu_buf *buf = &info->mti_big_buf;
2660         struct lu_name *name = &info->mti_name;
2661         struct lu_fid *pfid = &info->mti_tmp_fid1;
2662         struct link_ea_header *leh;
2663         struct link_ea_entry *lee;
2664         int reclen, count, rc = 0;
2665
2666         ENTRY;
2667
2668         if (!fid_is_sane(fid))
2669                 GOTO(out, rc = -EINVAL);
2670
2671         if (!fid_is_namespace_visible(fid))
2672                 GOTO(out, rc = -EINVAL);
2673
2674         obj = mdt_object_find(info->mti_env, mdt, fid);
2675         if (IS_ERR(obj))
2676                 GOTO(out, rc = PTR_ERR(obj));
2677
2678         if (mdt_object_remote(obj))
2679                 GOTO(out, rc = -EREMOTE);
2680         if (!mdt_object_exists(obj) || lu_object_is_dying(&obj->mot_header))
2681                 GOTO(out, rc = -ENOENT);
2682
2683         rc = mdt_rmfid_check_permission(info, obj);
2684         if (rc)
2685                 GOTO(out, rc);
2686
2687         /* take LinkEA */
2688         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
2689         if (!buf->lb_buf)
2690                 GOTO(out, rc = -ENOMEM);
2691
2692         ldata.ld_buf = buf;
2693         rc = mdt_links_read(info, obj, &ldata);
2694         if (rc)
2695                 GOTO(out, rc);
2696
2697         leh = buf->lb_buf;
2698         lee = (struct link_ea_entry *)(leh + 1);
2699         for (count = 0; count < leh->leh_reccount; count++) {
2700                 /* remove every hardlink */
2701                 linkea_entry_unpack(lee, &reclen, name, pfid);
2702                 lee = (struct link_ea_entry *) ((char *)lee + reclen);
2703                 rc = mdt_rmfid_unlink(info, pfid, name, obj, ctime);
2704                 if (rc)
2705                         break;
2706         }
2707
2708 out:
2709         if (obj && !IS_ERR(obj))
2710                 mdt_object_put(info->mti_env, obj);
2711         if (info->mti_big_buf.lb_buf)
2712                 lu_buf_free(&info->mti_big_buf);
2713
2714         RETURN(rc);
2715 }
2716
2717 static int mdt_rmfid(struct tgt_session_info *tsi)
2718 {
2719         struct mdt_thread_info *mti = tsi2mdt_info(tsi);
2720         struct mdt_body *reqbody;
2721         struct lu_fid *fids, *rfids;
2722         int bufsize, rc;
2723         __u32 *rcs;
2724         int i, nr;
2725
2726         ENTRY;
2727
2728         reqbody = req_capsule_client_get(tsi->tsi_pill, &RMF_MDT_BODY);
2729         if (reqbody == NULL)
2730                 RETURN(-EPROTO);
2731         bufsize = req_capsule_get_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2732                                        RCL_CLIENT);
2733         nr = bufsize / sizeof(struct lu_fid);
2734         if (nr * sizeof(struct lu_fid) != bufsize)
2735                 RETURN(-EINVAL);
2736         req_capsule_set_size(tsi->tsi_pill, &RMF_RCS,
2737                              RCL_SERVER, nr * sizeof(__u32));
2738         req_capsule_set_size(tsi->tsi_pill, &RMF_FID_ARRAY,
2739                              RCL_SERVER, nr * sizeof(struct lu_fid));
2740         rc = req_capsule_server_pack(tsi->tsi_pill);
2741         if (rc)
2742                 GOTO(out, rc = err_serious(rc));
2743         fids = req_capsule_client_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2744         if (fids == NULL)
2745                 RETURN(-EPROTO);
2746         rcs = req_capsule_server_get(tsi->tsi_pill, &RMF_RCS);
2747         LASSERT(rcs);
2748         rfids = req_capsule_server_get(tsi->tsi_pill, &RMF_FID_ARRAY);
2749         LASSERT(rfids);
2750
2751         mdt_init_ucred(mti, reqbody);
2752         for (i = 0; i < nr; i++) {
2753                 rfids[i] = fids[i];
2754                 rcs[i] = mdt_rmfid_one(mti, fids + i, reqbody->mbo_ctime);
2755         }
2756         mdt_exit_ucred(mti);
2757
2758 out:
2759         RETURN(rc);
2760 }
2761
2762 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
2763                          void *karg, void __user *uarg);
2764
2765 int mdt_io_set_info(struct tgt_session_info *tsi)
2766 {
2767         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2768         struct ost_body         *body = NULL, *repbody;
2769         void                    *key, *val = NULL;
2770         int                      keylen, vallen, rc = 0;
2771         bool                     is_grant_shrink;
2772
2773         ENTRY;
2774
2775         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2776         if (key == NULL) {
2777                 DEBUG_REQ(D_HA, req, "no set_info key");
2778                 RETURN(err_serious(-EFAULT));
2779         }
2780         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2781                                       RCL_CLIENT);
2782
2783         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2784         if (val == NULL) {
2785                 DEBUG_REQ(D_HA, req, "no set_info val");
2786                 RETURN(err_serious(-EFAULT));
2787         }
2788         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2789                                       RCL_CLIENT);
2790
2791         is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK);
2792         if (is_grant_shrink)
2793                 /* In this case the value is actually an RMF_OST_BODY, so we
2794                  * transmutate the type of this PTLRPC
2795                  */
2796                 req_capsule_extend(tsi->tsi_pill, &RQF_OST_SET_GRANT_INFO);
2797
2798         rc = req_capsule_server_pack(tsi->tsi_pill);
2799         if (rc < 0)
2800                 RETURN(rc);
2801
2802         if (is_grant_shrink) {
2803                 body = req_capsule_client_get(tsi->tsi_pill, &RMF_OST_BODY);
2804
2805                 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2806                 *repbody = *body;
2807
2808                 /** handle grant shrink, similar to a read request */
2809                 tgt_grant_prepare_read(tsi->tsi_env, tsi->tsi_exp,
2810                                        &repbody->oa);
2811         } else {
2812                 CERROR("%s: Unsupported key %s\n",
2813                        tgt_name(tsi->tsi_tgt), (char *)key);
2814                 rc = -EOPNOTSUPP;
2815         }
2816
2817         RETURN(rc);
2818 }
2819
2820
2821 static int mdt_set_info(struct tgt_session_info *tsi)
2822 {
2823         struct ptlrpc_request   *req = tgt_ses_req(tsi);
2824         char                    *key;
2825         void                    *val;
2826         int                      keylen, vallen, rc = 0;
2827
2828         ENTRY;
2829
2830         key = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_KEY);
2831         if (key == NULL) {
2832                 DEBUG_REQ(D_HA, req, "no set_info key");
2833                 RETURN(err_serious(-EFAULT));
2834         }
2835
2836         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_KEY,
2837                                       RCL_CLIENT);
2838
2839         val = req_capsule_client_get(tsi->tsi_pill, &RMF_SETINFO_VAL);
2840         if (val == NULL) {
2841                 DEBUG_REQ(D_HA, req, "no set_info val");
2842                 RETURN(err_serious(-EFAULT));
2843         }
2844
2845         vallen = req_capsule_get_size(tsi->tsi_pill, &RMF_SETINFO_VAL,
2846                                       RCL_CLIENT);
2847
2848         /* Swab any part of val you need to here */
2849         if (KEY_IS(KEY_READ_ONLY)) {
2850                 spin_lock(&req->rq_export->exp_lock);
2851                 if (*(__u32 *)val)
2852                         *exp_connect_flags_ptr(req->rq_export) |=
2853                                 OBD_CONNECT_RDONLY;
2854                 else
2855                         *exp_connect_flags_ptr(req->rq_export) &=
2856                                 ~OBD_CONNECT_RDONLY;
2857                 spin_unlock(&req->rq_export->exp_lock);
2858         } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
2859                 struct changelog_setinfo *cs = val;
2860
2861                 if (vallen != sizeof(*cs)) {
2862                         CERROR("%s: bad changelog_clear setinfo size %d\n",
2863                                tgt_name(tsi->tsi_tgt), vallen);
2864                         RETURN(-EINVAL);
2865                 }
2866                 if (req_capsule_req_need_swab(&req->rq_pill)) {
2867                         __swab64s(&cs->cs_recno);
2868                         __swab32s(&cs->cs_id);
2869                 }
2870
2871                 if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
2872                         RETURN(-EACCES);
2873                 rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, req->rq_export,
2874                                    vallen, val, NULL);
2875         } else if (KEY_IS(KEY_EVICT_BY_NID)) {
2876                 if (vallen > 0)
2877                         obd_export_evict_by_nid(req->rq_export->exp_obd, val);
2878         } else {
2879                 RETURN(-EINVAL);
2880         }
2881         RETURN(rc);
2882 }
2883
2884 static int mdt_readpage(struct tgt_session_info *tsi)
2885 {
2886         struct mdt_thread_info  *info = mdt_th_info(tsi->tsi_env);
2887         struct mdt_object       *object = mdt_obj(tsi->tsi_corpus);
2888         struct lu_rdpg          *rdpg = &info->mti_u.rdpg.mti_rdpg;
2889         const struct mdt_body   *reqbody = tsi->tsi_mdt_body;
2890         struct mdt_body         *repbody;
2891         int                      rc;
2892         int                      i;
2893
2894         ENTRY;
2895
2896         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK))
2897                 RETURN(err_serious(-ENOMEM));
2898
2899         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_MDT_BODY);
2900         if (repbody == NULL || reqbody == NULL)
2901                 RETURN(err_serious(-EFAULT));
2902
2903         /*
2904          * prepare @rdpg before calling lower layers and transfer itself. Here
2905          * reqbody->size contains offset of where to start to read and
2906          * reqbody->nlink contains number bytes to read.
2907          */
2908         rdpg->rp_hash = reqbody->mbo_size;
2909         if (rdpg->rp_hash != reqbody->mbo_size) {
2910                 CERROR("Invalid hash: %#llx != %#llx\n",
2911                        rdpg->rp_hash, reqbody->mbo_size);
2912                 RETURN(-EFAULT);
2913         }
2914
2915         rdpg->rp_attrs = reqbody->mbo_mode;
2916         if (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_64BITHASH)
2917                 rdpg->rp_attrs |= LUDA_64BITHASH;
2918         rdpg->rp_count  = min_t(unsigned int, reqbody->mbo_nlink,
2919                                 exp_max_brw_size(tsi->tsi_exp));
2920         rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >>
2921                           PAGE_SHIFT;
2922         OBD_ALLOC_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2923         if (rdpg->rp_pages == NULL)
2924                 RETURN(-ENOMEM);
2925
2926         for (i = 0; i < rdpg->rp_npages; ++i) {
2927                 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
2928                 if (rdpg->rp_pages[i] == NULL)
2929                         GOTO(free_rdpg, rc = -ENOMEM);
2930         }
2931
2932         /* call lower layers to fill allocated pages with directory data */
2933         rc = mo_readpage(tsi->tsi_env, mdt_object_child(object), rdpg);
2934         if (rc < 0)
2935                 GOTO(free_rdpg, rc);
2936
2937         /* send pages to client */
2938         rc = tgt_sendpage(tsi, rdpg, rc);
2939
2940         EXIT;
2941 free_rdpg:
2942
2943         for (i = 0; i < rdpg->rp_npages; i++)
2944                 if (rdpg->rp_pages[i] != NULL)
2945                         __free_page(rdpg->rp_pages[i]);
2946         OBD_FREE_PTR_ARRAY_LARGE(rdpg->rp_pages, rdpg->rp_npages);
2947
2948         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE))
2949                 RETURN(0);
2950
2951         return rc;
2952 }
2953
2954 static int mdt_fix_attr_ucred(struct mdt_thread_info *info, __u32 op)
2955 {
2956         struct lu_ucred *uc = mdt_ucred_check(info);
2957         struct lu_attr *attr = &info->mti_attr.ma_attr;
2958
2959         if (uc == NULL)
2960                 return -EINVAL;
2961
2962         if (op != REINT_SETATTR) {
2963                 if ((attr->la_valid & LA_UID) && (attr->la_uid != -1))
2964                         attr->la_uid = uc->uc_fsuid;
2965                 /* for S_ISGID, inherit gid from his parent, such work will be
2966                  * done in cmm/mdd layer, here set all cases as uc->uc_fsgid.
2967                  */
2968                 if ((attr->la_valid & LA_GID) && (attr->la_gid != -1))
2969                         attr->la_gid = uc->uc_fsgid;
2970         }
2971
2972         return 0;
2973 }
2974
2975 static inline bool mdt_is_readonly_open(struct mdt_thread_info *info, __u32 op)
2976 {
2977         return op == REINT_OPEN &&
2978              !(info->mti_spec.sp_cr_flags & (MDS_FMODE_WRITE | MDS_OPEN_CREAT));
2979 }
2980
2981 static void mdt_preset_secctx_size(struct mdt_thread_info *info)
2982 {
2983         struct req_capsule *pill = info->mti_pill;
2984
2985         if (req_capsule_has_field(pill, &RMF_FILE_SECCTX,
2986                                   RCL_SERVER) &&
2987             req_capsule_has_field(pill, &RMF_FILE_SECCTX_NAME,
2988                                   RCL_CLIENT)) {
2989                 if (req_capsule_get_size(pill, &RMF_FILE_SECCTX_NAME,
2990                                          RCL_CLIENT) != 0)
2991                         /* pre-set size in server part with max size */
2992                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2993                                              RCL_SERVER,
2994                                              req_capsule_ptlreq(pill) ?
2995                                              OBD_MAX_DEFAULT_EA_SIZE :
2996                                              MAX_MD_SIZE_OLD);
2997                 else
2998                         req_capsule_set_size(pill, &RMF_FILE_SECCTX,
2999                                              RCL_SERVER, 0);
3000         }
3001 }
3002
3003 static int mdt_reint_internal(struct mdt_thread_info *info,
3004                               struct mdt_lock_handle *lhc,
3005                               __u32 op)
3006 {
3007         struct req_capsule      *pill = info->mti_pill;
3008         struct mdt_body         *repbody;
3009         int                      rc = 0, rc2;
3010
3011         ENTRY;
3012
3013         rc = mdt_reint_unpack(info, op);
3014         if (rc != 0) {
3015                 CERROR("Can't unpack reint, rc %d\n", rc);
3016                 RETURN(err_serious(rc));
3017         }
3018
3019
3020         /* check if the file system is set to readonly. O_RDONLY open
3021          * is still allowed even the file system is set to readonly mode
3022          */
3023         if (mdt_rdonly(info->mti_exp) && !mdt_is_readonly_open(info, op))
3024                 RETURN(err_serious(-EROFS));
3025
3026         /* for replay (no_create) lmm is not needed, client has it already */
3027         if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3028                 req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
3029                                      DEF_REP_MD_SIZE);
3030
3031         /* llog cookies are always 0, the field is kept for compatibility */
3032         if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
3033                 req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, 0);
3034
3035         /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
3036          * by default. If the target object has more ACL entries, then
3037          * enlarge the buffer when necessary.
3038          */
3039         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
3040                 req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
3041                                      LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
3042
3043         mdt_preset_secctx_size(info);
3044         mdt_preset_encctx_size(info);
3045
3046         rc = req_capsule_server_pack(pill);
3047         if (rc != 0) {
3048                 CERROR("Can't pack response, rc %d\n", rc);
3049                 RETURN(err_serious(rc));
3050         }
3051
3052         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_SERVER)) {
3053                 repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
3054                 LASSERT(repbody);
3055                 repbody->mbo_eadatasize = 0;
3056                 repbody->mbo_aclsize = 0;
3057         }
3058
3059         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
3060
3061         /* for replay no cookkie / lmm need, because client have this already */
3062         if (info->mti_spec.no_create)
3063                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
3064                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0);
3065
3066         rc = mdt_init_ucred_reint(info);
3067         if (rc)
3068                 GOTO(out_shrink, rc);
3069
3070         rc = mdt_fix_attr_ucred(info, op);
3071         if (rc != 0)
3072                 GOTO(out_ucred, rc = err_serious(rc));
3073
3074         rc = mdt_check_resent(info, mdt_reconstruct, lhc);
3075         if (rc < 0) {
3076                 GOTO(out_ucred, rc);
3077         } else if (rc == 1) {
3078                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt");
3079                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
3080                 GOTO(out_ucred, rc);
3081         }
3082         rc = mdt_reint_rec(info, lhc);
3083         EXIT;
3084 out_ucred:
3085         mdt_exit_ucred(info);
3086 out_shrink:
3087         mdt_client_compatibility(info);
3088
3089         rc2 = mdt_fix_reply(info);
3090         if (rc == 0)
3091                 rc = rc2;
3092
3093         /*
3094          * Data-on-MDT optimization - read data along with OPEN and return it
3095          * in reply when possible.
3096          */
3097         if (rc == 0 && op == REINT_OPEN && !req_is_replay(pill->rc_req))
3098                 rc = mdt_dom_read_on_open(info, info->mti_mdt,
3099                                           &lhc->mlh_reg_lh);
3100
3101         return rc;
3102 }
3103
3104 static long mdt_reint_opcode(struct ptlrpc_request *req,
3105                              const struct req_format **fmt)
3106 {
3107         struct mdt_device       *mdt;
3108         struct mdt_rec_reint    *rec;
3109         long                     opc;
3110
3111         rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
3112         if (rec != NULL) {
3113                 opc = rec->rr_opcode;
3114                 DEBUG_REQ(D_INODE, req, "reint opt = %ld", opc);
3115                 if (opc < REINT_MAX && fmt[opc] != NULL)
3116                         req_capsule_extend(&req->rq_pill, fmt[opc]);
3117                 else {
3118                         mdt = mdt_exp2dev(req->rq_export);
3119                         CERROR("%s: Unsupported opcode '%ld' from client '%s': rc = %d\n",
3120                                req->rq_export->exp_obd->obd_name,
3121                                opc, mdt->mdt_ldlm_client->cli_name, -EFAULT);
3122                         opc = err_serious(-EFAULT);
3123                 }
3124         } else {
3125                 opc = err_serious(-EFAULT);
3126         }
3127         return opc;
3128 }
3129
3130 static int mdt_reint(struct tgt_session_info *tsi)
3131 {
3132         long opc;
3133         int  rc;
3134         static const struct req_format *reint_fmts[REINT_MAX] = {
3135                 [REINT_SETATTR]  = &RQF_MDS_REINT_SETATTR,
3136                 [REINT_CREATE]   = &RQF_MDS_REINT_CREATE,
3137                 [REINT_LINK]     = &RQF_MDS_REINT_LINK,
3138                 [REINT_UNLINK]   = &RQF_MDS_REINT_UNLINK,
3139                 [REINT_RENAME]   = &RQF_MDS_REINT_RENAME,
3140                 [REINT_OPEN]     = &RQF_MDS_REINT_OPEN,
3141                 [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR,
3142                 [REINT_RMENTRY]  = &RQF_MDS_REINT_UNLINK,
3143                 [REINT_MIGRATE]  = &RQF_MDS_REINT_MIGRATE,
3144                 [REINT_RESYNC]   = &RQF_MDS_REINT_RESYNC,
3145         };
3146
3147         ENTRY;
3148
3149         opc = mdt_reint_opcode(tgt_ses_req(tsi), reint_fmts);
3150         if (opc >= 0) {
3151                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3152                 /*
3153                  * No lock possible here from client to pass it to reint code
3154                  * path.
3155                  */
3156                 rc = mdt_reint_internal(info, NULL, opc);
3157                 mdt_thread_info_fini(info);
3158         } else {
3159                 rc = opc;
3160         }
3161
3162         tsi->tsi_reply_fail_id = OBD_FAIL_MDS_REINT_NET_REP;
3163         RETURN(rc);
3164 }
3165
3166 /* this should sync the whole device */
3167 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt)
3168 {
3169         struct dt_device *dt = mdt->mdt_bottom;
3170         int rc;
3171
3172         ENTRY;
3173
3174         rc = dt->dd_ops->dt_sync(env, dt);
3175         RETURN(rc);
3176 }
3177
3178 /* this should sync this object */
3179 static int mdt_object_sync(const struct lu_env *env, struct obd_export *exp,
3180                            struct mdt_object *mo)
3181 {
3182         int rc = 0;
3183
3184         ENTRY;
3185
3186         if (!mdt_object_exists(mo)) {
3187                 CWARN("%s: non existing object "DFID": rc = %d\n",
3188                       exp->exp_obd->obd_name, PFID(mdt_object_fid(mo)),
3189                       -ESTALE);
3190                 RETURN(-ESTALE);
3191         }
3192
3193         if (S_ISREG(lu_object_attr(&mo->mot_obj))) {
3194                 struct lu_target *tgt = tgt_ses_info(env)->tsi_tgt;
3195                 dt_obj_version_t version;
3196
3197                 version = dt_version_get(env, mdt_obj2dt(mo));
3198                 if (version > tgt->lut_obd->obd_last_committed)
3199                         rc = mo_object_sync(env, mdt_object_child(mo));
3200         } else {
3201                 rc = mo_object_sync(env, mdt_object_child(mo));
3202         }
3203
3204         RETURN(rc);
3205 }
3206
3207 static int mdt_sync(struct tgt_session_info *tsi)
3208 {
3209         struct ptlrpc_request   *req = tgt_ses_req(tsi);
3210         struct req_capsule      *pill = tsi->tsi_pill;
3211         struct mdt_body         *body;
3212         ktime_t                  kstart = ktime_get();
3213         int                      rc;
3214
3215         ENTRY;
3216
3217         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK))
3218                 RETURN(err_serious(-ENOMEM));
3219
3220         if (fid_seq(&tsi->tsi_mdt_body->mbo_fid1) == 0) {
3221                 rc = mdt_device_sync(tsi->tsi_env, mdt_exp2dev(tsi->tsi_exp));
3222         } else {
3223                 struct mdt_thread_info *info = tsi2mdt_info(tsi);
3224
3225                 if (unlikely(info->mti_object == NULL))
3226                         RETURN(-EPROTO);
3227
3228                 /* sync an object */
3229                 rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp,
3230                                      info->mti_object);
3231                 if (rc == 0) {
3232                         const struct lu_fid *fid;
3233                         struct lu_attr *la = &info->mti_attr.ma_attr;
3234
3235                         info->mti_attr.ma_need = MA_INODE;
3236                         info->mti_attr.ma_valid = 0;
3237                         rc = mdt_attr_get_complex(info, info->mti_object,
3238                                                   &info->mti_attr);
3239                         if (rc == 0) {
3240                                 body = req_capsule_server_get(pill,
3241                                                               &RMF_MDT_BODY);
3242                                 fid = mdt_object_fid(info->mti_object);
3243                                 mdt_pack_attr2body(info, body, la, fid);
3244                         }
3245                 }
3246                 mdt_thread_info_fini(info);
3247         }
3248         if (rc == 0)
3249                 mdt_counter_incr(req, LPROC_MDT_SYNC,
3250                                  ktime_us_delta(ktime_get(), kstart));
3251
3252         RETURN(rc);
3253 }
3254
3255 static int mdt_data_sync(struct tgt_session_info *tsi)
3256 {
3257         struct mdt_thread_info *info;
3258         struct mdt_device *mdt = mdt_exp2dev(tsi->tsi_exp);
3259         struct ost_body *body = tsi->tsi_ost_body;
3260         struct ost_body *repbody;
3261         struct mdt_object *mo = NULL;
3262         struct md_attr *ma;
3263         int rc = 0;
3264
3265         ENTRY;
3266
3267         repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
3268
3269          /* device sync is done via MDS_SYNC. NOOP if no fid is specified */
3270         if (fid_is_zero(&tsi->tsi_fid))
3271                 RETURN(0);
3272
3273         mo = mdt_object_find(tsi->tsi_env, mdt, &tsi->tsi_fid);
3274         if (IS_ERR(mo))
3275                 RETURN(PTR_ERR(mo));
3276
3277         rc = mdt_object_sync(tsi->tsi_env, tsi->tsi_exp, mo);
3278         if (rc)
3279                 GOTO(put, rc);
3280
3281         repbody->oa.o_oi = body->oa.o_oi;
3282         repbody->oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
3283
3284         info = tsi2mdt_info(tsi);
3285         ma = &info->mti_attr;
3286         ma->ma_need = MA_INODE;
3287         ma->ma_valid = 0;
3288         rc = mdt_attr_get_complex(info, mo, ma);
3289         if (rc == 0)
3290                 obdo_from_la(&repbody->oa, &ma->ma_attr, VALID_FLAGS);
3291         else
3292                 rc = 0;
3293         mdt_thread_info_fini(info);
3294
3295         EXIT;
3296 put:
3297         if (mo != NULL)
3298                 mdt_object_put(tsi->tsi_env, mo);
3299         return rc;
3300 }
3301
3302 /*
3303  * Handle quota control requests to consult current usage/limit, but also
3304  * to configure quota enforcement
3305  */
3306 static int mdt_quotactl(struct tgt_session_info *tsi)
3307 {
3308         struct obd_export *exp  = tsi->tsi_exp;
3309         struct req_capsule *pill = tsi->tsi_pill;
3310         struct obd_quotactl *oqctl, *repoqc;
3311         int id, rc;
3312         struct mdt_device *mdt = mdt_exp2dev(exp);
3313         struct lu_device *qmt = mdt->mdt_qmt_dev;
3314         struct lu_nodemap *nodemap;
3315
3316         ENTRY;
3317
3318         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
3319         if (!oqctl)
3320                 RETURN(err_serious(-EPROTO));
3321
3322         rc = req_capsule_server_pack(pill);
3323         if (rc)
3324                 RETURN(err_serious(rc));
3325
3326         nodemap = nodemap_get_from_exp(exp);
3327         if (IS_ERR(nodemap))
3328                 RETURN(PTR_ERR(nodemap));
3329
3330         switch (oqctl->qc_cmd) {
3331                 /* master quotactl */
3332         case Q_SETINFO:
3333         case Q_SETQUOTA:
3334         case LUSTRE_Q_SETDEFAULT:
3335         case LUSTRE_Q_SETQUOTAPOOL:
3336         case LUSTRE_Q_SETINFOPOOL:
3337         case LUSTRE_Q_SETDEFAULT_POOL:
3338         case LUSTRE_Q_DELETEQID:
3339         case LUSTRE_Q_RESETQID:
3340                 if (!nodemap_can_setquota(nodemap, oqctl->qc_type,
3341                                           oqctl->qc_id))
3342                         GOTO(out_nodemap, rc = -EPERM);
3343                 fallthrough;
3344         case Q_GETINFO:
3345         case Q_GETQUOTA:
3346         case LUSTRE_Q_GETDEFAULT:
3347         case LUSTRE_Q_GETQUOTAPOOL:
3348         case LUSTRE_Q_GETINFOPOOL:
3349         case LUSTRE_Q_GETDEFAULT_POOL:
3350                 if (qmt == NULL)
3351                         GOTO(out_nodemap, rc = -EOPNOTSUPP);
3352                 /* slave quotactl */
3353                 fallthrough;
3354         case Q_GETOINFO:
3355         case Q_GETOQUOTA:
3356                 break;
3357         default:
3358                 rc = -EFAULT;
3359                 CERROR("%s: unsupported quotactl command %d: rc = %d\n",
3360                        mdt_obd_name(mdt), oqctl->qc_cmd, rc);
3361                 GOTO(out_nodemap, rc);
3362         }
3363
3364         id = oqctl->qc_id;
3365         switch (oqctl->qc_type) {
3366         case USRQUOTA:
3367                 id = nodemap_map_id(nodemap, NODEMAP_UID,
3368                                     NODEMAP_CLIENT_TO_FS, id);
3369                 break;
3370         case GRPQUOTA:
3371                 id = nodemap_map_id(nodemap, NODEMAP_GID,
3372                                     NODEMAP_CLIENT_TO_FS, id);
3373                 break;
3374         case PRJQUOTA:
3375                 id = nodemap_map_id(nodemap, NODEMAP_PROJID,
3376                                     NODEMAP_CLIENT_TO_FS, id);
3377                 break;
3378         default:
3379                 GOTO(out_nodemap, rc = -EOPNOTSUPP);
3380         }
3381         repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
3382         if (repoqc == NULL)
3383                 GOTO(out_nodemap, rc = err_serious(-EFAULT));
3384
3385         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA)
3386                 barrier_exit(tsi->tsi_tgt->lut_bottom);
3387
3388         if (oqctl->qc_id != id)
3389                 swap(oqctl->qc_id, id);
3390
3391         if (oqctl->qc_cmd == Q_SETINFO || oqctl->qc_cmd == Q_SETQUOTA) {
3392                 if (unlikely(!barrier_entry(tsi->tsi_tgt->lut_bottom)))
3393                         RETURN(-EINPROGRESS);
3394         }
3395
3396         switch (oqctl->qc_cmd) {
3397
3398         case Q_GETINFO:
3399         case Q_SETINFO:
3400         case Q_SETQUOTA:
3401         case Q_GETQUOTA:
3402         case LUSTRE_Q_SETDEFAULT:
3403         case LUSTRE_Q_GETDEFAULT:
3404         case LUSTRE_Q_SETQUOTAPOOL:
3405         case LUSTRE_Q_GETQUOTAPOOL:
3406         case LUSTRE_Q_SETINFOPOOL:
3407         case LUSTRE_Q_GETINFOPOOL:
3408         case LUSTRE_Q_SETDEFAULT_POOL:
3409         case LUSTRE_Q_GETDEFAULT_POOL:
3410         case LUSTRE_Q_DELETEQID:
3411         case LUSTRE_Q_RESETQID:
3412                 /* forward quotactl request to QMT */
3413                 rc = qmt_hdls.qmth_quotactl(tsi->tsi_env, qmt, oqctl);
3414                 break;
3415
3416         case Q_GETOINFO:
3417         case Q_GETOQUOTA:
3418                 /* slave quotactl */
3419                 rc = lquotactl_slv(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
3420                                    oqctl);
3421                 break;
3422
3423         default:
3424                 CERROR("Unsupported quotactl command: %d\n", oqctl->qc_cmd);
3425                 GOTO(out_nodemap, rc = -EFAULT);
3426         }
3427
3428         if (oqctl->qc_id != id)
3429                 swap(oqctl->qc_id, id);
3430
3431         QCTL_COPY_NO_PNAME(repoqc, oqctl);
3432         EXIT;
3433
3434 out_nodemap:
3435         nodemap_putref(nodemap);
3436
3437         return rc;
3438 }
3439
3440 /** clone llog ctxt from child (mdd)
3441  * This allows remote llog (replicator) access.
3442  * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
3443  * context was originally set up, or we can handle them directly.
3444  * I choose the latter, but that means I need any llog
3445  * contexts set up by child to be accessable by the mdt.  So we clone the
3446  * context into our context list here.
3447  */
3448 static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
3449                                int idx)
3450 {
3451         struct md_device  *next = mdt->mdt_child;
3452         struct llog_ctxt *ctxt;
3453         int rc;
3454
3455         if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
3456                 return 0;
3457
3458         rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
3459         if (rc || ctxt == NULL)
3460                 return 0;
3461
3462         rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
3463         if (rc)
3464                 CERROR("Can't set mdt ctxt %d\n", rc);
3465
3466         return rc;
3467 }
3468
3469 static int mdt_llog_ctxt_unclone(const struct lu_env *env,
3470                                  struct mdt_device *mdt, int idx)
3471 {
3472         struct llog_ctxt *ctxt;
3473
3474         ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
3475         if (ctxt == NULL)
3476                 return 0;
3477         /* Put once for the get we just did, and once for the clone */
3478         llog_ctxt_put(ctxt);
3479         llog_ctxt_put(ctxt);
3480         return 0;
3481 }
3482
3483 /*
3484  * sec context handlers
3485  */
3486 static int mdt_sec_ctx_handle(struct tgt_session_info *tsi)
3487 {
3488         CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
3489
3490         return 0;
3491 }
3492
3493 /*
3494  * quota request handlers
3495  */
3496 static int mdt_quota_dqacq(struct tgt_session_info *tsi)
3497 {
3498         struct mdt_device       *mdt = mdt_exp2dev(tsi->tsi_exp);
3499         struct lu_device        *qmt = mdt->mdt_qmt_dev;
3500         int                      rc;
3501
3502         ENTRY;
3503
3504         if (qmt == NULL)
3505                 RETURN(err_serious(-EOPNOTSUPP));
3506
3507         rc = qmt_hdls.qmth_dqacq(tsi->tsi_env, qmt, tgt_ses_req(tsi));
3508         RETURN(rc);
3509 }
3510
3511 struct mdt_object *mdt_object_new(const struct lu_env *env,
3512                                   struct mdt_device *d,
3513                                   const struct lu_fid *f)
3514 {
3515         struct lu_object_conf conf = { .loc_flags = LOC_F_NEW };
3516         struct lu_object *o;
3517         struct mdt_object *m;
3518
3519         ENTRY;
3520
3521         CDEBUG(D_INFO, "Allocate object for "DFID"\n", PFID(f));
3522         o = lu_object_find(env, &d->mdt_lu_dev, f, &conf);
3523         if (unlikely(IS_ERR(o)))
3524                 m = (struct mdt_object *)o;
3525         else
3526                 m = mdt_obj(o);
3527         RETURN(m);
3528 }
3529
3530 struct mdt_object *mdt_object_find(const struct lu_env *env,
3531                                    struct mdt_device *d,
3532                                    const struct lu_fid *f)
3533 {
3534         struct lu_object *o;
3535         struct mdt_object *m;
3536
3537         ENTRY;
3538
3539         CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f));
3540         o = lu_object_find(env, &d->mdt_lu_dev, f, NULL);
3541         if (unlikely(IS_ERR(o)))
3542                 m = (struct mdt_object *)o;
3543         else
3544                 m = mdt_obj(o);
3545
3546         RETURN(m);
3547 }
3548
3549 /**
3550  * Asyncronous commit for mdt device.
3551  *
3552  * Pass asynchonous commit call down the MDS stack.
3553  *
3554  * \param env environment
3555  * \param mdt the mdt device
3556  */
3557 static void mdt_device_commit_async(const struct lu_env *env,
3558                                     struct mdt_device *mdt)
3559 {
3560         struct dt_device *dt = mdt->mdt_bottom;
3561         int rc;
3562
3563         ENTRY;
3564
3565         rc = dt->dd_ops->dt_commit_async(env, dt);
3566         if (unlikely(rc != 0))
3567                 CWARN("%s: async commit start failed: rc = %d\n",
3568                       mdt_obd_name(mdt), rc);
3569         atomic_inc(&mdt->mdt_async_commit_count);
3570         EXIT;
3571 }
3572
3573 /**
3574  * Mark the lock as "synchonous".
3575  *
3576  * Mark the lock to deffer transaction commit to the unlock time.
3577  *
3578  * \param lock the lock to mark as "synchonous"
3579  *
3580  * \see mdt_is_lock_sync
3581  * \see mdt_save_lock
3582  */
3583 static inline void mdt_set_lock_sync(struct ldlm_lock *lock)
3584 {
3585         lock->l_ast_data = (void *)1;
3586 }
3587
3588 /**
3589  * Check whehter the lock "synchonous" or not.
3590  *
3591  * \param lock the lock to check
3592  * \retval 1 the lock is "synchonous"
3593  * \retval 0 the lock isn't "synchronous"
3594  *
3595  * \see mdt_set_lock_sync
3596  * \see mdt_save_lock
3597  */
3598 static inline int mdt_is_lock_sync(struct ldlm_lock *lock)
3599 {
3600         return lock->l_ast_data != NULL;
3601 }
3602
3603 /**
3604  * Blocking AST for mdt locks.
3605  *
3606  * Starts transaction commit if in case of COS lock conflict or
3607  * deffers such a commit to the mdt_save_lock.
3608  *
3609  * \param lock the lock which blocks a request or cancelling lock
3610  * \param desc unused
3611  * \param data unused
3612  * \param flag indicates whether this cancelling or blocking callback
3613  * \retval 0
3614  * \see ldlm_blocking_ast_nocheck
3615  */
3616 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3617                      void *data, int flag)
3618 {
3619         struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3620         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
3621         struct ldlm_cb_set_arg *arg = data;
3622         bool commit_async = false;
3623         int rc;
3624
3625         ENTRY;
3626         if (flag == LDLM_CB_CANCELING)
3627                 RETURN(0);
3628
3629         lock_res_and_lock(lock);
3630         if (lock->l_blocking_ast != mdt_blocking_ast) {
3631                 unlock_res_and_lock(lock);
3632                 RETURN(0);
3633         }
3634
3635         /* A blocking ast may be sent from ldlm_lock_decref_internal
3636          * when the last reference to a local lock was released and
3637          * during blocking event from ldlm_work_bl_ast_lock().
3638          * The 'data' parameter is l_ast_data in the first case and
3639          * callback arguments in the second one. Distinguish them by that.
3640          */
3641         if (data && data != lock->l_ast_data && arg->bl_desc) {
3642                 if (lock->l_req_mode & (LCK_COS | LCK_TXN))
3643                         commit_async = true;
3644                 else if ((lock->l_req_mode & (LCK_PW | LCK_EX)) &&
3645                          ((mdt_cos_is_enabled(mdt) &&
3646                              !arg->bl_desc->bl_same_client) ||
3647                           (mdt_slc_is_enabled(mdt) &&
3648                              arg->bl_desc->bl_txn_dependent)))
3649                         mdt_set_lock_sync(lock);
3650         }
3651
3652         rc = ldlm_blocking_ast_nocheck(lock);
3653
3654         if (commit_async) {
3655                 struct lu_env env;
3656
3657                 rc = lu_env_init(&env, LCT_LOCAL);
3658                 if (unlikely(rc != 0))
3659                         CWARN("%s: lu_env initialization failed, cannot start asynchronous commit: rc = %d\n",
3660                               obd->obd_name, rc);
3661                 else
3662                         mdt_device_commit_async(&env, mdt);
3663                 lu_env_fini(&env);
3664         }
3665         RETURN(rc);
3666 }
3667
3668 /*
3669  * Blocking AST for cross-MDT lock
3670  *
3671  * Discard lock from uncommitted_slc_locks and cancel it.
3672  *
3673  * \param lock  the lock which blocks a request or cancelling lock
3674  * \param desc  unused
3675  * \param data  unused
3676  * \param flag  indicates whether this cancelling or blocking callback
3677  * \retval      0 on success
3678  * \retval      negative number on error
3679  */
3680 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
3681                             void *data, int flag)
3682 {
3683         int rc = 0;
3684
3685         ENTRY;
3686
3687         switch (flag) {
3688         case LDLM_CB_BLOCKING: {
3689                 struct lustre_handle lockh;
3690
3691                 ldlm_lock2handle(lock, &lockh);
3692                 rc = ldlm_cli_cancel(&lockh,
3693                         ldlm_is_atomic_cb(lock) ? 0 : LCF_ASYNC);
3694                 if (rc < 0) {
3695                         CDEBUG(D_INODE, "ldlm_cli_cancel: %d\n", rc);
3696                         RETURN(rc);
3697                 }
3698                 break;
3699         }
3700         case LDLM_CB_CANCELING: {
3701                 struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd;
3702                 struct mdt_device *mdt =
3703                                 mdt_dev(obd->obd_lu_dev->ld_site->ls_top_dev);
3704
3705                 LDLM_DEBUG(lock, "Revoke remote lock");
3706
3707                 /* discard slc lock here so that it can be cleaned anytime,
3708                  * especially for cleanup_resource()
3709                  */
3710                 tgt_discard_slc_lock(&mdt->mdt_lut, lock);
3711
3712                 /* once we cache lock, l_ast_data is set to mdt_object */
3713                 if (lock->l_ast_data != NULL) {
3714                         struct mdt_object *mo = lock->l_ast_data;
3715                         struct lu_env env;
3716
3717                         rc = lu_env_init(&env, LCT_MD_THREAD);
3718                         if (unlikely(rc != 0)) {
3719                                 CWARN("%s: lu_env initialization failed, object %p "DFID" is leaked!: rc = %d\n",
3720                                       obd->obd_name, mo,
3721                                       PFID(mdt_object_fid(mo)), rc);
3722                                 RETURN(rc);
3723                         }
3724
3725                         if (lock->l_policy_data.l_inodebits.bits &
3726                             (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)) {
3727                                 rc = mo_invalidate(&env, mdt_object_child(mo));
3728                                 mo->mot_cache_attr = 0;
3729                         }
3730                         mdt_object_put(&env, mo);
3731                         lu_env_fini(&env);
3732                 }
3733                 break;
3734         }
3735         default:
3736                 LBUG();
3737         }
3738
3739         RETURN(rc);
3740 }
3741
3742 int mdt_check_resent_lock(struct mdt_thread_info *info,
3743                           struct mdt_object *mo,
3744                           struct mdt_lock_handle *lhc)
3745 {
3746         /* the lock might already be gotten in ldlm_handle_enqueue() */
3747         if (unlikely(lustre_handle_is_used(&lhc->mlh_reg_lh))) {
3748                 struct ptlrpc_request *req = mdt_info_req(info);
3749                 struct ldlm_lock      *lock;
3750
3751                 lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
3752                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
3753                 if (lock == NULL) {
3754                         /* Lock is pinned by ldlm_handle_enqueue0() as it is
3755                          * a resend case, however, it could be already destroyed
3756                          * due to client eviction or a raced cancel RPC.
3757                          */
3758                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx",
3759                                           lhc->mlh_reg_lh.cookie);
3760                         RETURN(-ESTALE);
3761                 }
3762
3763                 if (!fid_res_name_eq(mdt_object_fid(mo),
3764                                      &lock->l_resource->lr_name)) {
3765                         CWARN("%s: Although resent, but still not get child lock:"
3766                               DFID"\n",
3767                               info->mti_exp->exp_obd->obd_name,
3768                               PFID(mdt_object_fid(mo)));
3769                         LDLM_LOCK_PUT(lock);
3770                         RETURN(-EPROTO);
3771                 }
3772                 LDLM_LOCK_PUT(lock);
3773                 return 0;
3774         }
3775         return 1;
3776 }
3777
3778 static void mdt_remote_object_lock_created_cb(struct ldlm_lock *lock)
3779 {
3780         mdt_object_get(NULL, lock->l_ast_data);
3781 }
3782
3783 static int mdt_remote_object_lock_try(struct mdt_thread_info *mti,
3784                                       struct mdt_object *obj,
3785                                       struct lustre_handle *lh,
3786                                       enum ldlm_mode mode,
3787                                       union ldlm_policy_data *policy,
3788                                       struct ldlm_res_id *res_id,
3789                                       bool cache)
3790 {
3791         struct ldlm_enqueue_info *einfo = &mti->mti_remote_einfo;
3792         int rc;
3793
3794         LASSERT(mdt_object_remote(obj));
3795
3796         memset(einfo, 0, sizeof(*einfo));
3797         einfo->ei_type = LDLM_IBITS;
3798         einfo->ei_mode = mode;
3799         einfo->ei_cb_bl = mdt_remote_blocking_ast;
3800         einfo->ei_cb_cp = ldlm_completion_ast;
3801         einfo->ei_enq_slave = 0;
3802         einfo->ei_res_id = res_id;
3803         einfo->ei_req_slot = 1;
3804         if (cache) {
3805                 /*
3806                  * if we cache lock, couple lock with mdt_object, so that object
3807                  * can be easily found in lock ASTs.
3808                  */
3809                 einfo->ei_cbdata = obj;
3810                 einfo->ei_cb_created = mdt_remote_object_lock_created_cb;
3811         }
3812
3813         rc = mo_object_lock(mti->mti_env, mdt_object_child(obj), lh, einfo,
3814                             policy);
3815         if (rc) {
3816                 lh->cookie = 0ull;
3817                 return rc;
3818         }
3819
3820         /* other components like LFSCK can use lockless access
3821          * and populate cache, so we better invalidate it
3822          */
3823         if (policy->l_inodebits.bits &
3824             (MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR))
3825                 mo_invalidate(mti->mti_env, mdt_object_child(obj));
3826
3827         return 0;
3828 }
3829
3830 /*
3831  * Helper function to take PDO and hash lock.
3832  *
3833  * if \a pdo_lock is false, don't take PDO lock, this is case in rename.
3834  */
3835 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
3836                         struct mdt_lock_handle *lh, const struct lu_name *name,
3837                         enum ldlm_mode mode, bool pdo_lock)
3838 {
3839         struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3840         union ldlm_policy_data *policy = &info->mti_policy;
3841         struct ldlm_res_id *res_id = &info->mti_res_id;
3842         /*
3843          * Do not use LDLM_FL_LOCAL_ONLY for parallel lock, it is never going to
3844          * be sent to client and we do not want it slowed down due to possible
3845          * cancels.
3846          */
3847         __u64 dlmflags = LDLM_FL_ATOMIC_CB;
3848         __u64 *cookie = NULL;
3849         int rc;
3850
3851         LASSERT(obj);
3852
3853         /* check for exists after object is locked */
3854         if (!mdt_object_exists(obj))
3855                 /* Non-existent object shouldn't have PDO lock */
3856                 return -ESTALE;
3857
3858         /* Non-dir object shouldn't have PDO lock */
3859         if (!S_ISDIR(lu_object_attr(&obj->mot_obj)))
3860                 return -ENOTDIR;
3861
3862         policy->l_inodebits.bits = MDS_INODELOCK_UPDATE;
3863         policy->l_inodebits.try_bits = 0;
3864         policy->l_inodebits.li_gid = 0;
3865         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3866         fid_build_reg_res_name(mdt_object_fid(obj), res_id);
3867         if (info->mti_exp)
3868                 cookie = &info->mti_exp->exp_handle.h_cookie;
3869
3870         mdt_lock_pdo_init(lh, mode, name);
3871         mdt_lock_pdo_mode(info, obj, lh);
3872         if (lh->mlh_pdo_mode != LCK_NL) {
3873                 if (pdo_lock) {
3874                         if (mdt_object_remote(obj)) {
3875                                 rc = mdt_remote_object_lock_try(info, obj,
3876                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3877                                         policy, res_id, false);
3878                                 lh->mlh_pdo_remote = 1;
3879                         } else {
3880                                 rc = mdt_fid_lock(info->mti_env, ns,
3881                                         &lh->mlh_pdo_lh, lh->mlh_pdo_mode,
3882                                         policy, res_id, dlmflags, cookie);
3883                         }
3884                         if (rc) {
3885                                 mdt_object_unlock(info, obj, lh, 1);
3886                                 return rc;
3887                         }
3888                 }
3889                 res_id->name[LUSTRE_RES_ID_HSH_OFF] = lh->mlh_pdo_hash;
3890         }
3891
3892         if (mdt_object_remote(obj))
3893                 rc = mdt_remote_object_lock_try(info, obj, &lh->mlh_rreg_lh,
3894                         lh->mlh_rreg_mode, policy, res_id, false);
3895         else
3896                 rc = mdt_fid_lock(info->mti_env, ns, &lh->mlh_reg_lh,
3897                         lh->mlh_reg_mode, policy, res_id, dlmflags, cookie);
3898         if (rc)
3899                 mdt_object_unlock(info, obj, lh, 1);
3900         else if (CFS_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK) &&
3901                    lh->mlh_pdo_hash != 0 &&
3902                    (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX))
3903                 CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 15);
3904
3905         return rc;
3906 }
3907
3908 int mdt_object_lock_internal(struct mdt_thread_info *info,
3909                              struct mdt_object *obj, const struct lu_fid *fid,
3910                              struct mdt_lock_handle *lh, __u64 *ibits,
3911                              __u64 trybits, bool cache)
3912 {
3913         union ldlm_policy_data *policy = &info->mti_policy;
3914         struct ldlm_res_id *res_id = &info->mti_res_id;
3915         struct lustre_handle *handle;
3916         int rc;
3917
3918         policy->l_inodebits.bits = *ibits;
3919         policy->l_inodebits.try_bits = trybits;
3920         policy->l_inodebits.li_gid = lh->mlh_gid;
3921         policy->l_inodebits.li_initiator_id = mdt_node_id(info->mti_mdt);
3922         fid_build_reg_res_name(fid, res_id);
3923
3924         if (obj && mdt_object_remote(obj)) {
3925                 handle = &lh->mlh_rreg_lh;
3926                 LASSERT(!lustre_handle_is_used(handle));
3927                 LASSERT(lh->mlh_rreg_mode != LCK_MINMODE);
3928                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3929                 rc = mdt_remote_object_lock_try(info, obj, handle,
3930                                                 lh->mlh_rreg_mode, policy,
3931                                                 res_id, cache);
3932         } else {
3933                 struct ldlm_namespace *ns = info->mti_mdt->mdt_namespace;
3934                 /*
3935                  * Use LDLM_FL_LOCAL_ONLY for this lock. We do not know yet if
3936                  * it is going to be sent to client. If it is -
3937                  * mdt_intent_policy() path will fix it up and turn FL_LOCAL
3938                  * flag off.
3939                  */
3940                 __u64 dlmflags = LDLM_FL_ATOMIC_CB | LDLM_FL_LOCAL_ONLY;
3941                 __u64 *cookie = NULL;
3942
3943                 handle = &lh->mlh_reg_lh;
3944                 LASSERT(!lustre_handle_is_used(handle));
3945                 LASSERT(lh->mlh_reg_mode != LCK_MINMODE);
3946                 LASSERT(lh->mlh_type != MDT_NUL_LOCK);
3947
3948                 /* Lease lock are granted with LDLM_FL_CANCEL_ON_BLOCK */
3949                 if (lh->mlh_type == MDT_REG_LOCK &&
3950                     lh->mlh_reg_mode == LCK_EX && *ibits == MDS_INODELOCK_OPEN)
3951                         dlmflags |= LDLM_FL_CANCEL_ON_BLOCK;
3952
3953
3954                 if (info->mti_exp)
3955                         cookie = &info->mti_exp->exp_handle.h_cookie;
3956
3957                 rc = mdt_fid_lock(info->mti_env, ns, handle, lh->mlh_reg_mode,
3958                                   policy, res_id, dlmflags, cookie);
3959                 if (rc)
3960                         mdt_object_unlock(info, obj, lh, 1);
3961         }
3962
3963         if (rc == 0) {
3964                 struct ldlm_lock *lock;
3965
3966                 /* Return successfully acquired bits to a caller */
3967                 lock = ldlm_handle2lock(handle);
3968                 LASSERT(lock);
3969                 *ibits = lock->l_policy_data.l_inodebits.bits;
3970                 LDLM_LOCK_PUT(lock);
3971         }
3972
3973         return rc;
3974 }
3975
3976 /*
3977  * MDT object locking functions:
3978  * mdt_object_lock(): lock object, this is used in most places, and normally
3979  *      lock ibits doesn't contain LOOKUP, unless the caller knows it's not
3980  *      remote object.
3981  * mdt_object_check_lock(): lock object with LOOKUP and other ibits, it needs
3982  *      to check whether parent is on remote MDT, if so, take LOOKUP on parent
3983  *      MDT separately, and then lock other ibits on child object.
3984  * mdt_parent_lock(): take parent UPDATE lock with specific mode, if parent is
3985  *      local, take PDO lock by name hash, otherwise take regular lock.
3986  * mdt_object_stripes_lock(): lock object which should be local, and if it's a
3987  *      striped directory, lock its stripes, this is called in operations which
3988  *      modify both object and stripes.
3989  * mdt_object_lock_try(): lock object with trybits, the trybits contains
3990  *      optional inode lock bits that can be granted. This is called by
3991  *      getattr/open to fetch more inode lock bits to client, and is also called
3992  *      by dir migration to lock link parent in non-block mode to avoid
3993  *      deadlock.
3994  */
3995
3996 /**
3997  * lock object
3998  *
3999  * this is used to lock object in most places, and normally lock ibits doesn't
4000  * contain LOOKUP, unless the caller knows it's not remote object.
4001  *
4002  * \param info          struct mdt_thread_info
4003  * \param obj           object
4004  * \param lh            lock handle
4005  * \param ibits         MDS inode lock bits
4006  * \param mode          lock mode
4007  *
4008  * \retval              0 on success, -ev on error.
4009  */
4010 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4011                     struct mdt_lock_handle *lh, __u64 ibits,
4012                     enum ldlm_mode mode)
4013 {
4014         int rc;
4015
4016         ENTRY;
4017         mdt_lock_reg_init(lh, mode);
4018         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh,
4019                                       &ibits, 0, false);
4020         RETURN(rc);
4021 }
4022
4023 /**
4024  * lock object with LOOKUP and other ibits
4025  *
4026  * it will check whether parent and child are on different MDTs, if so, take
4027  * LOOKUP lock on parent MDT, and lock other ibits on child MDT, otherwise lock
4028  * all ibits on child MDT. Note, parent and child shouldn't be both on remote
4029  * MDTs, in which case specific lock function should be used, and it's in
4030  * rename and migrate only.
4031  *
4032  * \param info          struct mdt_thread_info
4033  * \param parent        parent object
4034  * \param child         child object
4035  * \param lh            lock handle
4036  * \param ibits         MDS inode lock bits
4037  * \param mode          lock mode
4038  *
4039  * \retval              0 on success, -ev on error.
4040  */
4041 int mdt_object_check_lock(struct mdt_thread_info *info,
4042                           struct mdt_object *parent, struct mdt_object *child,
4043                           struct mdt_lock_handle *lh, __u64 ibits,
4044                           enum ldlm_mode mode)
4045 {
4046         int rc;
4047
4048         ENTRY;
4049         /* if LOOKUP ibit is not set, use mdt_object_lock() */
4050         LASSERT(ibits & MDS_INODELOCK_LOOKUP);
4051         /* if only LOOKUP ibit is needed, use mdt_object_lookup_lock() */
4052         LASSERT(ibits != MDS_INODELOCK_LOOKUP);
4053         LASSERT(parent);
4054         /* @parent and @child shouldn't both be on remote MDTs */
4055         LASSERT(!(mdt_object_remote(parent) && mdt_object_remote(child)));
4056
4057         mdt_lock_reg_init(lh, mode);
4058         if (mdt_object_remote(parent) ^ mdt_object_remote(child)) {
4059                 __u64 lookup_ibits = MDS_INODELOCK_LOOKUP;
4060
4061                 rc = mdt_object_lock_internal(info, parent,
4062                                               mdt_object_fid(child), lh,
4063                                               &lookup_ibits, 0, false);
4064                 if (rc)
4065                         RETURN(rc);
4066
4067                 ibits &= ~MDS_INODELOCK_LOOKUP;
4068         }
4069
4070         rc = mdt_object_lock_internal(info, child, mdt_object_fid(child), lh,
4071                                       &ibits, 0, false);
4072         if (rc && !(ibits & MDS_INODELOCK_LOOKUP))
4073                 mdt_object_unlock(info, NULL, lh, 1);
4074
4075         RETURN(rc);
4076 }
4077
4078 /**
4079  * take parent UPDATE lock
4080  *
4081  * if parent is local or mode is LCK_PW, take PDO lock, otherwise take regular
4082  * lock.
4083  *
4084  * \param info  struct mdt_thread_info
4085  * \param obj   parent object
4086  * \param lh    lock handle
4087  * \param lname child name
4088  * \param mode  lock mode
4089  *
4090  * \retval      0 on success, -ev on error.
4091  */
4092 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *obj,
4093                     struct mdt_lock_handle *lh, const struct lu_name *lname,
4094                     enum ldlm_mode mode)
4095 {
4096         int rc;
4097
4098         ENTRY;
4099         LASSERT(obj && lname);
4100         LASSERT(mode == LCK_PW || mode == LCK_PR);
4101         if (mdt_object_remote(obj) && mode == LCK_PR) {
4102                 __u64 ibits = MDS_INODELOCK_UPDATE;
4103
4104                 mdt_lock_reg_init(lh, mode);
4105                 rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj),
4106                                               lh, &ibits, 0, false);
4107         } else {
4108                 rc = mdt_object_pdo_lock(info, obj, lh, lname, mode, true);
4109         }
4110         RETURN(rc);
4111 }
4112
4113 /**
4114  * lock object with trybits
4115  *
4116  * the trybits contains optional inode lock bits that can be granted. This is
4117  * called by getattr/open to fetch more inode lock bits to client, and is also
4118  * called by dir migration to lock link parent in non-block mode to avoid
4119  * deadlock.
4120  *
4121  * \param info          struct mdt_thread_info
4122  * \param obj           object
4123  * \param lh            lock handle
4124  * \param ibits         MDS inode lock bits
4125  * \param trybits       optional inode lock bits
4126  * \param mode          lock mode
4127  *
4128  * \retval              0 on success, -ev on error.
4129  */
4130 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *obj,
4131                         struct mdt_lock_handle *lh, __u64 *ibits,
4132                         __u64 trybits, enum ldlm_mode mode)
4133 {
4134         bool trylock_only = *ibits == 0;
4135         int rc;
4136
4137         ENTRY;
4138         LASSERT(!(*ibits & trybits));
4139         mdt_lock_reg_init(lh, mode);
4140         rc = mdt_object_lock_internal(info, obj, mdt_object_fid(obj), lh, ibits,
4141                                       trybits, false);
4142         if (rc && trylock_only) { /* clear error for try ibits lock only */
4143                 LASSERT(*ibits == 0);
4144                 rc = 0;
4145         }
4146         RETURN(rc);
4147 }
4148
4149 /*
4150  * Helper function to take \a obj LOOKUP lock.
4151  *
4152  * Both \a pobj and \a obj may be located on remote MDTs.
4153  */
4154 int mdt_object_lookup_lock(struct mdt_thread_info *info,
4155                            struct mdt_object *pobj, struct mdt_object *obj,
4156                            struct mdt_lock_handle *lh, enum ldlm_mode mode)
4157 {
4158         __u64 ibits = MDS_INODELOCK_LOOKUP;
4159         int rc;
4160
4161         ENTRY;
4162         /* if @parent is NULL, it's on local MDT, and @child is remote,
4163          * this is case in getattr/unlink/open by name.
4164          */
4165         LASSERT(ergo(!pobj, mdt_object_remote(obj)));
4166         mdt_lock_reg_init(lh, mode);
4167         rc = mdt_object_lock_internal(info, pobj, mdt_object_fid(obj), lh,
4168                                       &ibits, 0, false);
4169         RETURN(rc);
4170 }
4171
4172 /**
4173  * Save a lock within request object.
4174  *
4175  * Keep the lock referenced until whether client ACK or transaction
4176  * commit happens or release the lock immediately depending on input
4177  * parameters. If COS is ON, a write lock is converted to COS lock
4178  * before saving.
4179  *
4180  * \param info thead info object
4181  * \param h lock handle
4182  * \param mode lock mode
4183  * \param decref force immediate lock releasing
4184  */
4185 static void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
4186                           enum ldlm_mode mode, int decref)
4187 {
4188         struct tgt_session_info *tsi = info->mti_env->le_ses ?
4189                                        tgt_ses_info(info->mti_env) : NULL;
4190
4191         ENTRY;
4192
4193         if (lustre_handle_is_used(h)) {
4194                 bool has_trans = tsi && tsi->tsi_has_trans;
4195
4196                 if (decref || !has_trans || !(mode & (LCK_PW | LCK_EX))) {
4197                         mdt_fid_unlock(h, mode);
4198                 } else {
4199                         struct mdt_device *mdt = info->mti_mdt;
4200                         struct ldlm_lock *lock = ldlm_handle2lock(h);
4201                         struct ptlrpc_request *req = mdt_info_req(info);
4202                         bool no_ack = false;
4203
4204                         LASSERTF(lock != NULL, "no lock for cookie %#llx\n",
4205                                  h->cookie);
4206
4207                         /* there is no request if mdt_object_unlock() is called
4208                          * from mdt_export_cleanup()->mdt_add_dirty_flag()
4209                          */
4210                         if (likely(req != NULL)) {
4211                                 LDLM_DEBUG(lock, "save lock request %p reply state %p transno %lld",
4212                                         req,
4213                                         req->rq_reply_state, req->rq_transno);
4214                                 if (mdt_cos_is_enabled(mdt)) {
4215                                         mode = LCK_COS;
4216                                         no_ack = true;
4217                                         ldlm_lock_mode_downgrade(lock, mode);
4218                                 } else if (mdt_slc_is_enabled(mdt)) {
4219                                         no_ack = true;
4220                                         if (mode != LCK_TXN) {
4221                                                 mode = LCK_TXN;
4222                                                 ldlm_lock_mode_downgrade(lock,
4223                                                                          mode);
4224                                         }
4225                                 }
4226                                 if (req->rq_export->exp_disconnected)
4227                                         mdt_fid_unlock(h, mode);
4228                                 else
4229                                         ptlrpc_save_lock(req, h, mode, no_ack);
4230                         } else {
4231                                 mdt_fid_unlock(h, mode);
4232                         }
4233
4234                         if (mdt_is_lock_sync(lock)) {
4235                                 CDEBUG(D_HA, "sync_lock, do async commit\n");
4236                                 mdt_device_commit_async(info->mti_env, mdt);
4237                         }
4238                         LDLM_LOCK_PUT(lock);
4239                 }
4240                 h->cookie = 0ull;
4241         }
4242
4243         EXIT;
4244 }
4245
4246 /**
4247  * Save cross-MDT lock in uncommitted_slc_locks
4248  *
4249  * Keep the lock referenced until transaction commit happens or release the lock
4250  * immediately depending on input parameters.
4251  *
4252  * \param info thead info object
4253  * \param h lock handle
4254  * \param mode lock mode
4255  * \param decref force immediate lock releasing
4256  */
4257 static void mdt_save_remote_lock(struct mdt_thread_info *info,
4258                                  struct mdt_object *o, struct lustre_handle *h,
4259                                  enum ldlm_mode mode, int decref)
4260 {
4261         ENTRY;
4262
4263         if (lustre_handle_is_used(h)) {
4264                 struct ldlm_lock *lock = ldlm_handle2lock(h);
4265                 struct ptlrpc_request *req = mdt_info_req(info);
4266
4267                 if (o != NULL &&
4268                     (lock->l_policy_data.l_inodebits.bits &
4269                      (MDS_INODELOCK_XATTR | MDS_INODELOCK_UPDATE)))
4270                         mo_invalidate(info->mti_env, mdt_object_child(o));
4271
4272                 if (decref || !req || !(mode & (LCK_PW | LCK_EX)) ||
4273                     !tgt_ses_info(info->mti_env)->tsi_has_trans) {
4274                         ldlm_lock_decref_and_cancel(h, mode);
4275                         LDLM_LOCK_PUT(lock);
4276                 } else {
4277                         tgt_save_slc_lock(&info->mti_mdt->mdt_lut, lock,
4278                                           req->rq_transno);
4279                         ldlm_lock_decref(h, mode);
4280                 }
4281                 h->cookie = 0ull;
4282         }
4283
4284         EXIT;
4285 }
4286
4287 /**
4288  * Unlock mdt object.
4289  *
4290  * Immeditely release the regular lock and the PDO lock or save the
4291  * lock in request and keep them referenced until client ACK or
4292  * transaction commit.
4293  *
4294  * \param info thread info object
4295  * \param o mdt object
4296  * \param lh mdt lock handle referencing regular and PDO locks
4297  * \param decref force immediate lock releasing
4298  *
4299  * XXX o is not used and may be NULL, see hsm_cdt_request_completed().
4300  */
4301 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
4302                        struct mdt_lock_handle *lh, int decref)
4303 {
4304         ENTRY;
4305
4306         if (lh->mlh_pdo_remote)
4307                 mdt_save_remote_lock(info, o, &lh->mlh_pdo_lh,
4308                                      lh->mlh_pdo_mode, decref);
4309         else
4310                 mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref);
4311         mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref);
4312         mdt_save_remote_lock(info, o, &lh->mlh_rreg_lh, lh->mlh_rreg_mode,
4313                              decref);
4314
4315         EXIT;
4316 }
4317
4318 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
4319                                         const struct lu_fid *f,
4320                                         struct mdt_lock_handle *lh,
4321                                         __u64 ibits, enum ldlm_mode mode)
4322 {
4323         struct mdt_object *o;
4324
4325         o = mdt_object_find(info->mti_env, info->mti_mdt, f);
4326         if (!IS_ERR(o)) {
4327                 int rc;
4328
4329                 rc = mdt_object_lock(info, o, lh, ibits, mode);
4330                 if (rc != 0) {
4331                         mdt_object_put(info->mti_env, o);
4332                         o = ERR_PTR(rc);
4333                 }
4334         }
4335         return o;
4336 }
4337
4338 void mdt_object_unlock_put(struct mdt_thread_info *info,
4339                            struct mdt_object *o,
4340                            struct mdt_lock_handle *lh,
4341                            int decref)
4342 {
4343         mdt_object_unlock(info, o, lh, decref);
4344         mdt_object_put(info->mti_env, o);
4345 }
4346
4347 /*
4348  * Generic code handling requests that have struct mdt_body passed in:
4349  *
4350  *  - extract mdt_body from request and save it in @info, if present;
4351  *
4352  *  - create lu_object, corresponding to the fid in mdt_body, and save it in
4353  *  @info;
4354  *
4355  *  - if HAS_BODY flag is set for this request type check whether object
4356  *  actually exists on storage (lu_object_exists()).
4357  *
4358  */
4359 static int mdt_body_unpack(struct mdt_thread_info *info,
4360                            enum tgt_handler_flags flags)
4361 {
4362         const struct mdt_body    *body;
4363         struct mdt_object        *obj;
4364         const struct lu_env      *env;
4365         struct req_capsule       *pill;
4366         int                       rc;
4367
4368         ENTRY;
4369
4370         env = info->mti_env;
4371         pill = info->mti_pill;
4372
4373         body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY);
4374         if (body == NULL)
4375                 RETURN(-EFAULT);
4376
4377         if (!(body->mbo_valid & OBD_MD_FLID))
4378                 RETURN(0);
4379
4380         if (!fid_is_sane(&body->mbo_fid1)) {
4381                 CERROR("Invalid fid: "DFID"\n", PFID(&body->mbo_fid1));
4382                 RETURN(-EINVAL);
4383         }
4384
4385         obj = mdt_object_find(env, info->mti_mdt, &body->mbo_fid1);
4386         if (!IS_ERR(obj)) {
4387                 if ((flags & HAS_BODY) && !mdt_object_exists(obj)) {
4388                         mdt_object_put(env, obj);
4389                         rc = -ENOENT;
4390                 } else {
4391                         info->mti_object = obj;
4392                         rc = 0;
4393                 }
4394         } else
4395                 rc = PTR_ERR(obj);
4396
4397         RETURN(rc);
4398 }
4399
4400 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info,
4401                                    enum tgt_handler_flags flags)
4402 {
4403         struct req_capsule *pill = info->mti_pill;
4404         int rc;
4405
4406         ENTRY;
4407
4408         if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT))
4409                 rc = mdt_body_unpack(info, flags);
4410         else
4411                 rc = 0;
4412
4413         if (rc == 0 && (flags & HAS_REPLY)) {
4414                 /* Pack reply. */
4415                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
4416                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
4417                                              req_capsule_ptlreq(pill) ?
4418                                              DEF_REP_MD_SIZE : MAX_MD_SIZE_OLD);
4419
4420                 if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
4421                         req_capsule_set_size(pill, &RMF_LOGCOOKIES,
4422                                              RCL_SERVER, 0);
4423
4424                 /* Set ACL reply buffer size as LUSTRE_POSIX_ACL_MAX_SIZE_OLD
4425                  * by default. If the target object has more ACL entries, then
4426                  * enlarge the buffer when necessary.
4427                  */
4428                 if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
4429                         req_capsule_set_size(pill, &RMF_ACL, RCL_SERVER,
4430                                              LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
4431
4432                 mdt_preset_secctx_size(info);
4433                 mdt_preset_encctx_size(info);
4434
4435                 rc = req_capsule_server_pack(pill);
4436                 if (rc)
4437                         CWARN("%s: cannot pack response: rc = %d\n",
4438                                       mdt_obd_name(info->mti_mdt), rc);
4439         }
4440         RETURN(rc);
4441 }
4442
4443 void mdt_thread_info_reset(struct mdt_thread_info *info)
4444 {
4445         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
4446         info->mti_body = NULL;
4447         info->mti_dlm_req = NULL;
4448         info->mti_cross_ref = 0;
4449         info->mti_opdata = 0;
4450         info->mti_big_lmm_used = 0;
4451         info->mti_big_acl_used = 0;
4452         info->mti_som_strict = 0;
4453
4454         info->mti_spec.no_create = 0;
4455         info->mti_spec.sp_rm_entry = 0;
4456         info->mti_spec.sp_permitted = 0;
4457
4458         info->mti_spec.u.sp_ea.eadata = NULL;
4459         info->mti_spec.u.sp_ea.eadatalen = 0;
4460
4461         if (info->mti_batch_env && info->mti_object != NULL) {
4462                 mdt_object_put(info->mti_env, info->mti_object);
4463                 info->mti_object = NULL;
4464         }
4465 }
4466
4467 /*
4468  * Initialize fields of struct mdt_thread_info. Other fields are left in
4469  * uninitialized state, because it's too expensive to zero out whole
4470  * mdt_thread_info (> 1K) on each request arrival.
4471  */
4472 void mdt_thread_info_init(struct ptlrpc_request *req,
4473                           struct mdt_thread_info *info)
4474 {
4475         info->mti_pill = &req->rq_pill;
4476
4477         /* mdt device: it can be NULL while CONNECT */
4478         if (req->rq_export) {
4479                 info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev);
4480                 info->mti_exp = req->rq_export;
4481         } else
4482                 info->mti_mdt = NULL;
4483         info->mti_env = req->rq_svc_thread->t_env;
4484         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
4485         info->mti_big_buf = LU_BUF_NULL;
4486         info->mti_batch_env = 0;
4487         info->mti_object = NULL;
4488
4489         mdt_thread_info_reset(info);
4490 }
4491
4492 void mdt_thread_info_fini(struct mdt_thread_info *info)
4493 {
4494         int i;
4495
4496         if (info->mti_object != NULL) {
4497                 mdt_object_put(info->mti_env, info->mti_object);
4498                 info->mti_object = NULL;
4499         }
4500
4501         for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++)
4502                 mdt_lock_handle_assert(&info->mti_lh[i]);
4503         info->mti_env = NULL;
4504         info->mti_pill = NULL;
4505         info->mti_exp = NULL;
4506         info->mti_mdt = NULL;
4507
4508         if (unlikely(info->mti_big_buf.lb_buf != NULL))
4509                 lu_buf_free(&info->mti_big_buf);
4510 }
4511
4512 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi)
4513 {
4514         struct mdt_thread_info  *mti;
4515
4516         mti = mdt_th_info(tsi->tsi_env);
4517         LASSERT(mti != NULL);
4518
4519         mdt_thread_info_init(tgt_ses_req(tsi), mti);
4520         if (tsi->tsi_corpus != NULL) {
4521                 mti->mti_object = mdt_obj(tsi->tsi_corpus);
4522                 lu_object_get(tsi->tsi_corpus);
4523         }
4524         mti->mti_body = tsi->tsi_mdt_body;
4525         mti->mti_dlm_req = tsi->tsi_dlm_req;
4526
4527         return mti;
4528 }
4529
4530 static int mdt_tgt_connect(struct tgt_session_info *tsi)
4531 {
4532         if (CFS_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
4533             cfs_fail_val ==
4534             tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
4535                 schedule_timeout_uninterruptible(cfs_time_seconds(3));
4536
4537         return tgt_connect(tsi);
4538 }
4539
4540 static int mdt_intent_glimpse(enum ldlm_intent_flags it_opc,
4541                               struct mdt_thread_info *info,
4542                               struct ldlm_lock **lockp, __u64 flags)
4543 {
4544         return mdt_glimpse_enqueue(info, info->mti_mdt->mdt_namespace,
4545                                    lockp, flags);
4546 }
4547 static int mdt_intent_brw(enum ldlm_intent_flags it_opc,
4548                           struct mdt_thread_info *info,
4549                           struct ldlm_lock **lockp, __u64 flags)
4550 {
4551         return mdt_brw_enqueue(info, info->mti_mdt->mdt_namespace,
4552                                lockp, flags);
4553 }
4554
4555 int mdt_intent_lock_replace(struct mdt_thread_info *info,
4556                             struct ldlm_lock **lockp,
4557                             struct mdt_lock_handle *lh,
4558                             __u64 flags, int result)
4559 {
4560         struct ptlrpc_request  *req = mdt_info_req(info);
4561         struct ldlm_lock       *lock = *lockp;
4562         struct ldlm_lock       *new_lock;
4563
4564         /* If possible resent found a lock, @lh is set to its handle */
4565         new_lock = ldlm_handle2lock_long(&lh->mlh_reg_lh, 0);
4566
4567         if (new_lock == NULL) {
4568                 if (flags & LDLM_FL_INTENT_ONLY) {
4569                         result = 0;
4570                 } else if (flags & LDLM_FL_RESENT) {
4571                         /* Lock is pinned by ldlm_handle_enqueue0() as it is a
4572                          * resend case, however, it could be already destroyed
4573                          * due to client eviction or a raced cancel RPC.
4574                          */
4575                         LDLM_DEBUG_NOLOCK("Invalid lock handle %#llx\n",
4576                                           lh->mlh_reg_lh.cookie);
4577                         result = -ESTALE;
4578                 } else {
4579                         CERROR("%s: Invalid lockh=%#llx flags=%#llx fid1="DFID" fid2="DFID": rc = %d\n",
4580                                mdt_obd_name(info->mti_mdt),
4581                                lh->mlh_reg_lh.cookie, flags,
4582                                PFID(&info->mti_tmp_fid1),
4583                                PFID(&info->mti_tmp_fid2), result);
4584                         result = -ESTALE;
4585                 }
4586                 lh->mlh_reg_lh.cookie = 0;
4587                 RETURN(result);
4588         }
4589
4590         /*
4591          * If we've already given this lock to a client once, then we should
4592          * have no readers or writers.  Otherwise, we should have one reader
4593          * _or_ writer ref (which will be zeroed below) before returning the
4594          * lock to a client.
4595          */
4596         if (new_lock->l_export == req->rq_export) {
4597                 LASSERT(new_lock->l_readers + new_lock->l_writers == 0);
4598         } else {
4599                 LASSERT(new_lock->l_export == NULL);
4600                 LASSERT(new_lock->l_readers + new_lock->l_writers == 1);
4601         }
4602
4603         *lockp = new_lock;
4604
4605         if (new_lock->l_export == req->rq_export) {
4606                 /*
4607                  * Already gave this to the client, which means that we
4608                  * reconstructed a reply.
4609                  */
4610                 LASSERT(lustre_msg_get_flags(req->rq_reqmsg) &
4611                         MSG_RESENT);
4612
4613                 LDLM_LOCK_RELEASE(new_lock);
4614                 lh->mlh_reg_lh.cookie = 0;
4615                 RETURN(ELDLM_LOCK_REPLACED);
4616         }
4617
4618         /*
4619          * Fixup the lock to be given to the client.
4620          */
4621         lock_res_and_lock(new_lock);
4622         /* Zero new_lock->l_readers and new_lock->l_writers without triggering
4623          * possible blocking AST.
4624          */
4625         while (new_lock->l_readers > 0) {
4626                 lu_ref_del(&new_lock->l_reference, "reader", new_lock);
4627                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4628                 new_lock->l_readers--;
4629         }
4630         while (new_lock->l_writers > 0) {
4631                 lu_ref_del(&new_lock->l_reference, "writer", new_lock);
4632                 lu_ref_del(&new_lock->l_reference, "user", new_lock);
4633                 new_lock->l_writers--;
4634         }
4635
4636         new_lock->l_export = class_export_lock_get(req->rq_export, new_lock);
4637         new_lock->l_blocking_ast = lock->l_blocking_ast;
4638         new_lock->l_completion_ast = lock->l_completion_ast;
4639         if (ldlm_has_dom(new_lock))
4640                 new_lock->l_glimpse_ast = ldlm_server_glimpse_ast;
4641         new_lock->l_remote_handle = lock->l_remote_handle;
4642         new_lock->l_flags &= ~LDLM_FL_LOCAL;
4643
4644         unlock_res_and_lock(new_lock);
4645
4646         cfs_hash_add(new_lock->l_export->exp_lock_hash,
4647                      &new_lock->l_remote_handle,
4648                      &new_lock->l_exp_hash);
4649
4650         LDLM_LOCK_RELEASE(new_lock);
4651         lh->mlh_reg_lh.cookie = 0;
4652
4653         RETURN(ELDLM_LOCK_REPLACED);
4654 }
4655
4656 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
4657                              struct ldlm_lock *new_lock,
4658                              struct mdt_lock_handle *lh, __u64 flags)
4659 {
4660         struct ptlrpc_request  *req = mdt_info_req(info);
4661         struct ldlm_request    *dlmreq;
4662
4663         if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT))
4664                 return;
4665
4666         dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ);
4667
4668         /* if this is a resend case (MSG_RESENT is set on RPC) and a lock was
4669          * found by ldlm_handle_enqueue(); if so @lh must be initialized.
4670          */
4671         if (flags & LDLM_FL_RESENT) {
4672                 lh->mlh_reg_lh.cookie = new_lock->l_handle.h_cookie;
4673                 lh->mlh_reg_mode = new_lock->l_granted_mode;
4674
4675                 LDLM_DEBUG(new_lock, "Restoring lock cookie");
4676                 DEBUG_REQ(D_DLMTRACE, req, "restoring lock cookie %#llx",
4677                           lh->mlh_reg_lh.cookie);
4678                 return;
4679         }
4680
4681         /*
4682          * If the xid matches, then we know this is a resent request, and allow
4683          * it. (It's probably an OPEN, for which we don't send a lock.
4684          */
4685         if (req_can_reconstruct(req, NULL) != 0)
4686                 return;
4687
4688         /*
4689          * This remote handle isn't enqueued, so we never received or processed
4690          * this request.  Clear MSG_RESENT, because it can be handled like any
4691          * normal request now.
4692          */
4693         lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT);
4694
4695         DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle %#llx",
4696                   dlmreq->lock_handle[0].cookie);
4697 }
4698
4699 static int mdt_intent_getxattr(enum ldlm_intent_flags it_opc,
4700                                struct mdt_thread_info *info,
4701                                struct ldlm_lock **lockp,
4702                                __u64 flags)
4703 {
4704         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4705         struct ldlm_reply      *ldlm_rep = NULL;
4706         int rc;
4707
4708         ENTRY;
4709
4710         /*
4711          * Initialize lhc->mlh_reg_lh either from a previously granted lock
4712          * (for the resend case) or a new lock. Below we will use it to
4713          * replace the original lock.
4714          */
4715         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4716         if (!lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4717                 rc = mdt_object_lock(info, info->mti_object, lhc,
4718                                      MDS_INODELOCK_XATTR, (*lockp)->l_req_mode);
4719                 if (rc)
4720                         return rc;
4721         }
4722
4723         rc = mdt_getxattr(info);
4724
4725         if (mdt_info_req(info)->rq_repmsg != NULL)
4726                 ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4727
4728         if (ldlm_rep == NULL ||
4729             CFS_FAIL_CHECK(OBD_FAIL_MDS_XATTR_REP)) {
4730                 mdt_object_unlock(info,  info->mti_object, lhc, 1);
4731                 if (is_serious(rc))
4732                         RETURN(rc);
4733                 else
4734                         RETURN(err_serious(-EFAULT));
4735         }
4736
4737         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4738
4739         /* This is for interop instead of adding a new interop flag. LU-7433 */
4740 #if LUSTRE_VERSION_CODE > OBD_OCD_VERSION(3, 0, 0, 0)
4741         if (ldlm_rep->lock_policy_res2) {
4742                 mdt_object_unlock(info, info->mti_object, lhc, 1);
4743                 RETURN(ELDLM_LOCK_ABORTED);
4744         }
4745 #endif
4746
4747         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4748         RETURN(rc);
4749 }
4750
4751 static int mdt_intent_getattr(enum ldlm_intent_flags it_opc,
4752                               struct mdt_thread_info *info,
4753                               struct ldlm_lock **lockp,
4754                               __u64 flags)
4755 {
4756         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4757         __u64                   child_bits;
4758         struct ldlm_reply      *ldlm_rep;
4759         struct mdt_body        *reqbody;
4760         struct mdt_body        *repbody;
4761         int                     rc, rc2;
4762
4763         ENTRY;
4764
4765         reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY);
4766         LASSERT(reqbody);
4767
4768         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
4769         LASSERT(repbody);
4770
4771         info->mti_cross_ref = !!(reqbody->mbo_valid & OBD_MD_FLCROSSREF);
4772         repbody->mbo_eadatasize = 0;
4773         repbody->mbo_aclsize = 0;
4774
4775         switch (it_opc) {
4776         case IT_LOOKUP:
4777                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_PERM;
4778                 break;
4779         case IT_GETATTR:
4780                 child_bits = MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
4781                              MDS_INODELOCK_PERM;
4782                 break;
4783         default:
4784                 CERROR("%s: unsupported intent %#x\n",
4785                        mdt_obd_name(info->mti_mdt), (unsigned int)it_opc);
4786                 GOTO(out_shrink, rc = -EINVAL);
4787         }
4788
4789         rc = mdt_init_ucred(info, reqbody);
4790         if (rc)
4791                 GOTO(out_shrink, rc);
4792
4793         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4794         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4795
4796         /* Get lock from request for possible resent case. */
4797         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4798
4799         rc = mdt_getattr_name_lock(info, lhc, child_bits, ldlm_rep);
4800         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4801
4802         if (mdt_get_disposition(ldlm_rep, DISP_LOOKUP_NEG) &&
4803             ldlm_rep->lock_policy_res2 != -ENOKEY)
4804                 ldlm_rep->lock_policy_res2 = 0;
4805         if (!mdt_get_disposition(ldlm_rep, DISP_LOOKUP_POS) ||
4806             ldlm_rep->lock_policy_res2) {
4807                 lhc->mlh_reg_lh.cookie = 0ull;
4808                 GOTO(out_ucred, rc = ELDLM_LOCK_ABORTED);
4809         }
4810
4811         rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4812         EXIT;
4813 out_ucred:
4814         mdt_exit_ucred(info);
4815 out_shrink:
4816         mdt_client_compatibility(info);
4817         rc2 = mdt_fix_reply(info);
4818         if (rc == 0)
4819                 rc = rc2;
4820         return rc;
4821 }
4822
4823 static int mdt_intent_layout(enum ldlm_intent_flags it_opc,
4824                              struct mdt_thread_info *info,
4825                              struct ldlm_lock **lockp,
4826                              __u64 flags)
4827 {
4828         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4829         struct md_layout_change layout = { .mlc_opc = MD_LAYOUT_NOP };
4830         struct layout_intent *intent;
4831         struct ldlm_reply *ldlm_rep;
4832         struct lu_fid *fid = &info->mti_tmp_fid2;
4833         struct mdt_object *obj = NULL;
4834         int layout_size = 0;
4835         struct lu_buf *buf = &layout.mlc_buf;
4836         int rc = 0;
4837
4838         ENTRY;
4839
4840         fid_extract_from_res_name(fid, &(*lockp)->l_resource->lr_name);
4841
4842         intent = req_capsule_client_get(info->mti_pill, &RMF_LAYOUT_INTENT);
4843         if (intent == NULL)
4844                 RETURN(-EPROTO);
4845
4846         CDEBUG(D_INFO, DFID "got layout change request from client: opc:%u flags:%#x extent "
4847                DEXT"\n",
4848                PFID(fid), intent->lai_opc, intent->lai_flags,
4849                PEXT(&intent->lai_extent));
4850
4851         switch (intent->lai_opc) {
4852         case LAYOUT_INTENT_TRUNC:
4853         case LAYOUT_INTENT_WRITE:
4854         case LAYOUT_INTENT_PCCRO_SET:
4855         case LAYOUT_INTENT_PCCRO_CLEAR:
4856                 layout.mlc_opc = MD_LAYOUT_WRITE;
4857                 layout.mlc_intent = intent;
4858                 break;
4859         case LAYOUT_INTENT_ACCESS:
4860                 break;
4861         case LAYOUT_INTENT_READ:
4862         case LAYOUT_INTENT_GLIMPSE:
4863         case LAYOUT_INTENT_RELEASE:
4864         case LAYOUT_INTENT_RESTORE:
4865                 CERROR("%s: Unsupported layout intent opc %d\n",
4866                        mdt_obd_name(info->mti_mdt), intent->lai_opc);
4867                 RETURN(-ENOTSUPP);
4868         default:
4869                 CERROR("%s: Unknown layout intent opc %d\n",
4870                        mdt_obd_name(info->mti_mdt), intent->lai_opc);
4871                 RETURN(-EINVAL);
4872         }
4873
4874         obj = mdt_object_find(info->mti_env, info->mti_mdt, fid);
4875         if (IS_ERR(obj))
4876                 RETURN(PTR_ERR(obj));
4877
4878         if (mdt_object_exists(obj) && !mdt_object_remote(obj)) {
4879                 /* if layout is going to be changed don't use the current EA
4880                  * size but the maximum one. That buffer will be shrinked
4881                  * to the actual size in req_capsule_shrink() before reply.
4882                  */
4883                 if (layout.mlc_opc == MD_LAYOUT_WRITE) {
4884                         layout_size = info->mti_mdt->mdt_max_mdsize;
4885                 } else {
4886                         layout_size = mdt_attr_get_eabuf_size(info, obj);
4887                         if (layout_size < 0)
4888                                 GOTO(out, rc = layout_size);
4889
4890                         if (layout_size > info->mti_mdt->mdt_max_mdsize)
4891                                 info->mti_mdt->mdt_max_mdsize = layout_size;
4892                 }
4893                 CDEBUG(D_INFO, "%s: layout_size %d\n",
4894                        mdt_obd_name(info->mti_mdt), layout_size);
4895         }
4896
4897         /*
4898          * set reply buffer size, so that ldlm_handle_enqueue0()->
4899          * ldlm_lvbo_fill() will fill the reply buffer with lovea.
4900          */
4901         req_capsule_set_size(info->mti_pill, &RMF_DLM_LVB, RCL_SERVER,
4902                              layout_size);
4903         rc = req_capsule_server_pack(info->mti_pill);
4904         if (rc)
4905                 GOTO(out, rc);
4906
4907         ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4908         if (!ldlm_rep)
4909                 GOTO(out, rc = -EPROTO);
4910
4911         mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD);
4912
4913         /* take lock in ldlm_lock_enqueue() for LAYOUT_INTENT_ACCESS */
4914         if (layout.mlc_opc == MD_LAYOUT_NOP)
4915                 GOTO(out, rc = 0);
4916
4917         rc = mdt_check_resent(info, mdt_reconstruct_generic, lhc);
4918         if (rc < 0)
4919                 GOTO(out, rc);
4920         if (rc == 1) {
4921                 DEBUG_REQ(D_INODE, mdt_info_req(info), "resent opt.");
4922                 rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg);
4923                 GOTO(out, rc);
4924         }
4925
4926         buf->lb_buf = NULL;
4927         buf->lb_len = 0;
4928         if (unlikely(req_is_replay(mdt_info_req(info)))) {
4929                 buf->lb_buf = req_capsule_client_get(info->mti_pill,
4930                                                      &RMF_EADATA);
4931                 buf->lb_len = req_capsule_get_size(info->mti_pill,
4932                                                      &RMF_EADATA, RCL_CLIENT);
4933                 /*
4934                  * If it's a replay of layout write intent RPC, the client has
4935                  * saved the extended lovea when it get reply then.
4936                  */
4937                 if (buf->lb_len > 0)
4938                         mdt_fix_lov_magic(info, buf->lb_buf);
4939         }
4940
4941         /* Get lock from request for possible resent case. */
4942         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4943         (*lockp)->l_lvb_type = LVB_T_LAYOUT;
4944
4945         /*
4946          * Instantiate some layout components, if @buf contains lovea, then it's
4947          * a replay of the layout intent write RPC.
4948          */
4949         rc = mdt_layout_change(info, obj, lhc, &layout);
4950         ldlm_rep->lock_policy_res2 = clear_serious(rc);
4951
4952         if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
4953                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
4954                 if (rc == ELDLM_LOCK_REPLACED &&
4955                     (*lockp)->l_granted_mode == LCK_EX)
4956                         ldlm_lock_mode_downgrade(*lockp, LCK_CR);
4957         }
4958
4959         EXIT;
4960 out:
4961         mdt_object_put(info->mti_env, obj);
4962         return rc;
4963 }
4964
4965 static int mdt_intent_open(enum ldlm_intent_flags it_opc,
4966                            struct mdt_thread_info *info,
4967                            struct ldlm_lock **lockp,
4968                            __u64 flags)
4969 {
4970         struct mdt_lock_handle *lhc = &info->mti_lh[MDT_LH_RMT];
4971         struct ldlm_reply      *rep = NULL;
4972         long                    opc;
4973         int                     rc;
4974         struct ptlrpc_request  *req = mdt_info_req(info);
4975
4976         static const struct req_format *intent_fmts[REINT_MAX] = {
4977                 [REINT_CREATE]  = &RQF_LDLM_INTENT_CREATE,
4978                 [REINT_OPEN]    = &RQF_LDLM_INTENT_OPEN
4979         };
4980
4981         ENTRY;
4982
4983         opc = mdt_reint_opcode(mdt_info_req(info), intent_fmts);
4984         if (opc < 0)
4985                 RETURN(opc);
4986
4987         /* Get lock from request for possible resent case. */
4988         mdt_intent_fixup_resent(info, *lockp, lhc, flags);
4989
4990         rc = mdt_reint_internal(info, lhc, opc);
4991
4992         if (rc < 0 && lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY)
4993                 DEBUG_REQ(D_ERROR, req, "Replay open failed with %d", rc);
4994
4995         /* Check whether the reply has been packed successfully. */
4996         if (mdt_info_req(info)->rq_repmsg != NULL)
4997                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
4998         if (rep == NULL) {
4999                 if (is_serious(rc))
5000                         RETURN(rc);
5001                 else
5002                         RETURN(err_serious(-EFAULT));
5003         }
5004
5005         /* MDC expects this in any case */
5006         if (rc != 0)
5007                 mdt_set_disposition(info, rep, DISP_LOOKUP_EXECD);
5008
5009         /* the open lock or the lock for cross-ref object should be
5010          * returned to the client
5011          */
5012         if (lustre_handle_is_used(&lhc->mlh_reg_lh) &&
5013             (rc == 0 || rc == -MDT_EREMOTE_OPEN)) {
5014                 rep->lock_policy_res2 = 0;
5015                 rc = mdt_intent_lock_replace(info, lockp, lhc, flags, rc);
5016                 RETURN(rc);
5017         }
5018
5019         rep->lock_policy_res2 = clear_serious(rc);
5020
5021         if (rep->lock_policy_res2 == -ENOENT &&
5022             mdt_get_disposition(rep, DISP_LOOKUP_NEG) &&
5023             !mdt_get_disposition(rep, DISP_OPEN_CREATE))
5024                 rep->lock_policy_res2 = 0;
5025
5026         lhc->mlh_reg_lh.cookie = 0ull;
5027         if (rc == -ENOTCONN || rc == -ENODEV ||
5028             rc == -EOVERFLOW) { /**< if VBR failure then return error */
5029                 /*
5030                  * If it is the disconnect error (ENODEV & ENOCONN), the error
5031                  * will be returned by rq_status, and client at ptlrpc layer
5032                  * will detect this, then disconnect, reconnect the import
5033                  * immediately, instead of impacting the following the rpc.
5034                  */
5035                 RETURN(rc);
5036         }
5037         /*
5038          * For other cases, the error will be returned by intent, and client
5039          * will retrieve the result from intent.
5040          */
5041         RETURN(ELDLM_LOCK_ABORTED);
5042 }
5043
5044 static int mdt_intent_opc(enum ldlm_intent_flags it_opc,
5045                           struct mdt_thread_info *info,
5046                           struct ldlm_lock **lockp,
5047                           u64 flags /* LDLM_FL_* */)
5048 {
5049         struct req_capsule *pill = info->mti_pill;
5050         struct ptlrpc_request *req = mdt_info_req(info);
5051         const struct req_format *it_format;
5052         int (*it_handler)(enum ldlm_intent_flags,
5053                           struct mdt_thread_info *,
5054                           struct ldlm_lock **,
5055                           u64);
5056         enum tgt_handler_flags it_handler_flags = 0;
5057         struct ldlm_reply *rep;
5058         bool check_mdt_object = false;
5059         int rc;
5060
5061         ENTRY;
5062
5063         switch (it_opc) {
5064         case IT_OPEN:
5065         case IT_OPEN|IT_CREAT:
5066                 /*
5067                  * OCREAT is not a IS_MUTABLE request since the file may
5068                  * already exist. We do the extra check of
5069                  * OBD_CONNECT_RDONLY in mdt_reint_open() when we
5070                  * really need to create the object.
5071                  */
5072                 it_format = &RQF_LDLM_INTENT;
5073                 it_handler = &mdt_intent_open;
5074                 break;
5075         case IT_GETATTR:
5076                 check_mdt_object = true;
5077                 fallthrough;
5078         case IT_LOOKUP:
5079                 it_format = &RQF_LDLM_INTENT_GETATTR;
5080                 it_handler = &mdt_intent_getattr;
5081                 it_handler_flags = HAS_REPLY;
5082                 break;
5083         case IT_GETXATTR:
5084                 check_mdt_object = true;
5085                 it_format = &RQF_LDLM_INTENT_GETXATTR;
5086                 it_handler = &mdt_intent_getxattr;
5087                 it_handler_flags = HAS_BODY;
5088                 break;
5089         case IT_LAYOUT:
5090                 it_format = &RQF_LDLM_INTENT_LAYOUT;
5091                 it_handler = &mdt_intent_layout;
5092                 break;
5093         case IT_GLIMPSE:
5094                 it_format = &RQF_LDLM_INTENT;
5095                 it_handler = &mdt_intent_glimpse;
5096                 break;
5097         case IT_BRW:
5098                 it_format = &RQF_LDLM_INTENT;
5099                 it_handler = &mdt_intent_brw;
5100                 break;
5101         case IT_QUOTA_DQACQ:
5102         case IT_QUOTA_CONN: {
5103                 struct lu_device *qmt = info->mti_mdt->mdt_qmt_dev;
5104
5105                 if (qmt == NULL)
5106                         RETURN(-EOPNOTSUPP);
5107
5108                 if (mdt_rdonly(req->rq_export))
5109                         RETURN(-EROFS);
5110
5111                 (*lockp)->l_lvb_type = LVB_T_LQUOTA;
5112                 /* pass the request to quota master */
5113                 rc = qmt_hdls.qmth_intent_policy(info->mti_env, qmt,
5114                                                  mdt_info_req(info), lockp,
5115                                                  flags);
5116                 RETURN(rc);
5117         }
5118         default:
5119                 CERROR("%s: unknown intent code %#x\n",
5120                        mdt_obd_name(info->mti_mdt), it_opc);
5121                 RETURN(-EPROTO);
5122         }
5123
5124         if (!info->mti_batch_env)
5125                 req_capsule_extend(pill, it_format);
5126
5127         rc = mdt_unpack_req_pack_rep(info, it_handler_flags);
5128         if (rc < 0)
5129                 RETURN(rc);
5130
5131         if (unlikely(info->mti_object == NULL && check_mdt_object))
5132                 RETURN(-EPROTO);
5133
5134         if (it_handler_flags & IS_MUTABLE && mdt_rdonly(req->rq_export))
5135                 RETURN(-EROFS);
5136
5137         CFS_FAIL_TIMEOUT(OBD_FAIL_MDS_INTENT_DELAY, 10);
5138
5139         /* execute policy */
5140         rc = (*it_handler)(it_opc, info, lockp, flags);
5141
5142         /* Check whether the reply has been packed successfully. */
5143         if (info->mti_batch_env || req->rq_repmsg != NULL) {
5144                 rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP);
5145                 rep->lock_policy_res2 =
5146                         ptlrpc_status_hton(rep->lock_policy_res2);
5147         }
5148
5149         RETURN(rc);
5150 }
5151
5152 static void mdt_ptlrpc_stats_update(struct ptlrpc_request *req,
5153                                     enum ldlm_intent_flags it_opc)
5154 {
5155         struct lprocfs_stats *srv_stats = ptlrpc_req2svc(req)->srv_stats;
5156
5157         /* update stats when IT code is known */
5158         if (srv_stats != NULL)
5159                 lprocfs_counter_incr(srv_stats,
5160                                 PTLRPC_LAST_CNTR + (it_opc == IT_GLIMPSE ?
5161                                 LDLM_GLIMPSE_ENQUEUE : LDLM_IBITS_ENQUEUE));
5162 }
5163
5164 static int mdt_intent_policy(const struct lu_env *env,
5165                              struct ldlm_namespace *ns,
5166                              struct ldlm_lock **lockp,
5167                              void *req_cookie,
5168                              enum ldlm_mode mode,
5169                              __u64 flags, void *data)
5170 {
5171         struct tgt_session_info *tsi;
5172         struct mdt_thread_info  *info;
5173         struct ptlrpc_request   *req  =  req_cookie;
5174         struct ldlm_intent      *it;
5175         struct req_capsule      *pill;
5176         const struct ldlm_lock_desc *ldesc;
5177         int rc;
5178
5179         ENTRY;
5180
5181         LASSERT(req != NULL);
5182
5183         tsi = tgt_ses_info(env);
5184
5185         info = mdt_th_info(env);
5186         LASSERT(info != NULL);
5187
5188         /* Check whether it is a sub request processing in a batch request */
5189         if (info->mti_batch_env) {
5190                 pill = info->mti_pill;
5191                 LASSERT(pill == &info->mti_sub_pill);
5192         } else {
5193                 info = tsi2mdt_info(tsi);
5194                 pill = info->mti_pill;
5195         }
5196
5197         LASSERT(pill->rc_req == req);
5198         ldesc = &info->mti_dlm_req->lock_desc;
5199
5200         if (info->mti_batch_env ||
5201             req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) {
5202                 /*
5203                  * For batch processing environment, the request format has
5204                  * already been set.
5205                  */
5206                 if (!info->mti_batch_env)
5207                         req_capsule_extend(pill, &RQF_LDLM_INTENT_BASIC);
5208
5209                 it = req_capsule_client_get(pill, &RMF_LDLM_INTENT);
5210                 if (it != NULL) {
5211                         mdt_ptlrpc_stats_update(req, it->opc);
5212                         rc = mdt_intent_opc(it->opc, info, lockp, flags);
5213                         if (rc == 0)
5214                                 rc = ELDLM_OK;
5215
5216                         /* Lock without inodebits makes no sense and will oops
5217                          * later in ldlm. Let's check it now to see if we have
5218                          * ibits corrupted somewhere in mdt_intent_opc().
5219                          * The case for client miss to set ibits has been
5220                          * processed by others.
5221                          */
5222                         LASSERT(ergo(ldesc->l_resource.lr_type == LDLM_IBITS,
5223                                 ldesc->l_policy_data.l_inodebits.bits != 0));
5224                 } else {
5225                         rc = err_serious(-EFAULT);
5226                 }
5227         } else if (ldesc->l_resource.lr_type == LDLM_IBITS &&
5228                    ldesc->l_policy_data.l_inodebits.bits == MDS_INODELOCK_DOM) {
5229                 struct ldlm_reply *rep;
5230
5231                 /* No intent was provided but INTENT flag is set along with
5232                  * DOM bit, this is considered as GLIMPSE request.
5233                  * This logic is common for MDT and OST glimpse
5234                  */
5235                 mdt_ptlrpc_stats_update(req, IT_GLIMPSE);
5236                 rc = mdt_glimpse_enqueue(info, ns, lockp, flags);
5237                 /* Check whether the reply has been packed successfully. */
5238                 if (req->rq_repmsg != NULL) {
5239                         rep = req_capsule_server_get(info->mti_pill,
5240                                                      &RMF_DLM_REP);
5241                         rep->lock_policy_res2 =
5242                                 ptlrpc_status_hton(rep->lock_policy_res2);
5243                 }
5244         } else {
5245                 /* No intent was provided */
5246                 req_capsule_set_size(pill, &RMF_DLM_LVB, RCL_SERVER, 0);
5247                 rc = req_capsule_server_pack(pill);
5248                 if (rc)
5249                         rc = err_serious(rc);
5250         }
5251
5252         if (!info->mti_batch_env)
5253                 mdt_thread_info_fini(info);
5254         RETURN(rc);
5255 }
5256
5257 static void mdt_deregister_seq_exp(struct mdt_device *mdt)
5258 {
5259         struct seq_server_site  *ss = mdt_seq_site(mdt);
5260
5261         if (ss->ss_node_id == 0)
5262                 return;
5263
5264         if (ss->ss_client_seq != NULL) {
5265                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5266                 ss->ss_client_seq->lcs_exp = NULL;
5267         }
5268
5269         if (ss->ss_server_fld != NULL) {
5270                 lustre_deregister_lwp_item(&ss->ss_server_fld->lsf_control_exp);
5271                 ss->ss_server_fld->lsf_control_exp = NULL;
5272         }
5273 }
5274
5275 static void mdt_seq_fini_cli(struct mdt_device *mdt)
5276 {
5277         struct seq_server_site *ss = mdt_seq_site(mdt);
5278
5279         if (ss == NULL)
5280                 return;
5281
5282         if (ss->ss_server_seq != NULL)
5283                 seq_server_set_cli(NULL, ss->ss_server_seq, NULL);
5284 }
5285
5286 static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *mdt)
5287 {
5288         mdt_seq_fini_cli(mdt);
5289         mdt_deregister_seq_exp(mdt);
5290
5291         return seq_site_fini(env, mdt_seq_site(mdt));
5292 }
5293
5294 /**
5295  * It will retrieve its FLDB entries from MDT0, and it only happens
5296  * when upgrading existent FS to 2.6 or when local FLDB is corrupted,
5297  * and it needs to refresh FLDB from the MDT0.
5298  **/
5299 static int mdt_register_lwp_callback(void *data)
5300 {
5301         struct lu_env           env;
5302         struct mdt_device       *mdt = data;
5303         struct lu_server_fld    *fld = mdt_seq_site(mdt)->ss_server_fld;
5304         int                     rc;
5305
5306         ENTRY;
5307
5308         LASSERT(mdt_seq_site(mdt)->ss_node_id != 0);
5309
5310         rc = lu_env_init(&env, LCT_MD_THREAD);
5311         if (rc < 0) {
5312                 CERROR("%s: cannot init env: rc = %d\n", mdt_obd_name(mdt), rc);
5313                 RETURN(rc);
5314         }
5315
5316         /* Allocate new sequence now to avoid creating local transaction
5317          * in the normal transaction process
5318          */
5319         rc = seq_server_check_and_alloc_super(&env,
5320                                               mdt_seq_site(mdt)->ss_server_seq);
5321         if (rc < 0)
5322                 GOTO(out, rc);
5323
5324         if (fld->lsf_new) {
5325                 rc = fld_update_from_controller(&env, fld);
5326                 if (rc != 0) {
5327                         CERROR("%s: cannot update controller: rc = %d\n",
5328                                mdt_obd_name(mdt), rc);
5329                         GOTO(out, rc);
5330                 }
5331         }
5332 out:
5333         lu_env_fini(&env);
5334         RETURN(rc);
5335 }
5336
5337 static int mdt_register_seq_exp(struct mdt_device *mdt)
5338 {
5339         struct seq_server_site  *ss = mdt_seq_site(mdt);
5340         char                    *lwp_name = NULL;
5341         int                     rc;
5342
5343         if (ss->ss_node_id == 0)
5344                 return 0;
5345
5346         OBD_ALLOC(lwp_name, MAX_OBD_NAME);
5347         if (lwp_name == NULL)
5348                 GOTO(out_free, rc = -ENOMEM);
5349
5350         rc = tgt_name2lwp_name(mdt_obd_name(mdt), lwp_name, MAX_OBD_NAME, 0);
5351         if (rc != 0)
5352                 GOTO(out_free, rc);
5353
5354         rc = lustre_register_lwp_item(lwp_name, &ss->ss_client_seq->lcs_exp,
5355                                       NULL, NULL);
5356         if (rc != 0)
5357                 GOTO(out_free, rc);
5358
5359         rc = lustre_register_lwp_item(lwp_name,
5360                                       &ss->ss_server_fld->lsf_control_exp,
5361                                       mdt_register_lwp_callback, mdt);
5362         if (rc != 0) {
5363                 lustre_deregister_lwp_item(&ss->ss_client_seq->lcs_exp);
5364                 ss->ss_client_seq->lcs_exp = NULL;
5365                 GOTO(out_free, rc);
5366         }
5367 out_free:
5368         if (lwp_name != NULL)
5369                 OBD_FREE(lwp_name, MAX_OBD_NAME);
5370
5371         return rc;
5372 }
5373
5374 /*
5375  * Init client sequence manager which is used by local MDS to talk to sequence
5376  * controller on remote node.
5377  */
5378 static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *mdt)
5379 {
5380         struct seq_server_site *ss = mdt_seq_site(mdt);
5381         char *prefix;
5382
5383         ENTRY;
5384
5385         /* check if this is adding the first MDC and controller is not yet
5386          * initialized.
5387          */
5388         OBD_ALLOC_PTR(ss->ss_client_seq);
5389         if (ss->ss_client_seq == NULL)
5390                 RETURN(-ENOMEM);
5391
5392         OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
5393         if (prefix == NULL) {
5394                 OBD_FREE_PTR(ss->ss_client_seq);
5395                 ss->ss_client_seq = NULL;
5396                 RETURN(-ENOMEM);
5397         }
5398
5399         /* Note: seq_client_fini will be called in seq_site_fini */
5400         snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdt_obd_name(mdt));
5401         seq_client_init(ss->ss_client_seq, NULL, LUSTRE_SEQ_METADATA,
5402                         prefix, ss->ss_node_id == 0 ?  ss->ss_control_seq :
5403                                                             NULL);
5404         OBD_FREE(prefix, MAX_OBD_NAME + 5);
5405
5406         RETURN(seq_server_set_cli(env, ss->ss_server_seq, ss->ss_client_seq));
5407 }
5408
5409 static int mdt_seq_init(const struct lu_env *env, struct mdt_device *mdt)
5410 {
5411         struct seq_server_site  *ss;
5412         int                     rc;
5413
5414         ENTRY;
5415
5416         ss = mdt_seq_site(mdt);
5417         /* init sequence controller server(MDT0) */
5418         if (ss->ss_node_id == 0) {
5419                 OBD_ALLOC_PTR(ss->ss_control_seq);
5420                 if (ss->ss_control_seq == NULL)
5421                         RETURN(-ENOMEM);
5422
5423                 rc = seq_server_init(env, ss->ss_control_seq, mdt->mdt_bottom,
5424                                      mdt_obd_name(mdt), LUSTRE_SEQ_CONTROLLER,
5425                                      ss);
5426                 if (rc)
5427                         GOTO(out_seq_fini, rc);
5428         }
5429
5430         /* Init normal sequence server */
5431         OBD_ALLOC_PTR(ss->ss_server_seq);
5432         if (ss->ss_server_seq == NULL)
5433                 GOTO(out_seq_fini, rc = -ENOMEM);
5434
5435         rc = seq_server_init(env, ss->ss_server_seq, mdt->mdt_bottom,
5436                              mdt_obd_name(mdt), LUSTRE_SEQ_SERVER, ss);
5437         if (rc)
5438                 GOTO(out_seq_fini, rc);
5439
5440         /* init seq client for seq server to talk to seq controller(MDT0) */
5441         rc = mdt_seq_init_cli(env, mdt);
5442         if (rc != 0)
5443                 GOTO(out_seq_fini, rc);
5444
5445         if (ss->ss_node_id != 0)
5446                 /* register controller export through lwp */
5447                 rc = mdt_register_seq_exp(mdt);
5448
5449         EXIT;
5450 out_seq_fini:
5451         if (rc)
5452                 mdt_seq_fini(env, mdt);
5453
5454         return rc;
5455 }
5456
5457 /*
5458  * FLD wrappers
5459  */
5460 static int mdt_fld_fini(const struct lu_env *env,
5461                         struct mdt_device *m)
5462 {
5463         struct seq_server_site *ss = mdt_seq_site(m);
5464
5465         ENTRY;
5466
5467         if (ss && ss->ss_server_fld) {
5468                 fld_server_fini(env, ss->ss_server_fld);
5469                 OBD_FREE_PTR(ss->ss_server_fld);
5470                 ss->ss_server_fld = NULL;
5471         }
5472
5473         RETURN(0);
5474 }
5475
5476 static int mdt_fld_init(const struct lu_env *env,
5477                         const char *uuid,
5478                         struct mdt_device *m)
5479 {
5480         struct seq_server_site *ss;
5481         int rc;
5482
5483         ENTRY;
5484
5485         ss = mdt_seq_site(m);
5486
5487         OBD_ALLOC_PTR(ss->ss_server_fld);
5488         if (ss->ss_server_fld == NULL)
5489                 RETURN(rc = -ENOMEM);
5490
5491         rc = fld_server_init(env, ss->ss_server_fld, m->mdt_bottom, uuid,
5492                              LU_SEQ_RANGE_MDT);
5493         if (rc) {
5494                 OBD_FREE_PTR(ss->ss_server_fld);
5495                 ss->ss_server_fld = NULL;
5496                 RETURN(rc);
5497         }
5498
5499         RETURN(0);
5500 }
5501
5502 static void mdt_stack_pre_fini(const struct lu_env *env,
5503                            struct mdt_device *m, struct lu_device *top)
5504 {
5505         struct lustre_cfg_bufs  *bufs;
5506         struct lustre_cfg       *lcfg;
5507         struct mdt_thread_info  *info;
5508
5509         ENTRY;
5510
5511         LASSERT(top);
5512
5513         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5514         LASSERT(info != NULL);
5515
5516         bufs = &info->mti_u.bufs;
5517
5518         LASSERT(m->mdt_child_exp);
5519         LASSERT(m->mdt_child_exp->exp_obd);
5520
5521         /* process cleanup, pass mdt obd name to get obd umount flags */
5522         /* XXX: this is needed because all layers are referenced by
5523          * objects (some of them are pinned by osd, for example *
5524          * the proper solution should be a model where object used
5525          * by osd only doesn't have mdt/mdd slices -bzzz
5526          */
5527         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5528         lustre_cfg_bufs_set_string(bufs, 1, NULL);
5529         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5530         if (!lcfg)
5531                 RETURN_EXIT;
5532         lustre_cfg_init(lcfg, LCFG_PRE_CLEANUP, bufs);
5533
5534         top->ld_ops->ldo_process_config(env, top, lcfg);
5535         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5536         EXIT;
5537 }
5538
5539 static void mdt_stack_fini(const struct lu_env *env,
5540                            struct mdt_device *m, struct lu_device *top)
5541 {
5542         struct obd_device       *obd = mdt2obd_dev(m);
5543         struct lustre_cfg_bufs  *bufs;
5544         struct lustre_cfg       *lcfg;
5545         struct mdt_thread_info  *info;
5546         char                     flags[3] = "";
5547
5548         ENTRY;
5549
5550         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
5551         LASSERT(info != NULL);
5552
5553         lu_dev_del_linkage(top->ld_site, top);
5554
5555         lu_site_purge(env, top->ld_site, -1);
5556
5557         bufs = &info->mti_u.bufs;
5558         /* process cleanup, pass mdt obd name to get obd umount flags */
5559         /* another purpose is to let all layers to release their objects */
5560         lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
5561         if (obd->obd_force)
5562                 strcat(flags, "F");
5563         if (obd->obd_fail)
5564                 strcat(flags, "A");
5565         lustre_cfg_bufs_set_string(bufs, 1, flags);
5566         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5567         if (!lcfg)
5568                 RETURN_EXIT;
5569         lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
5570
5571         LASSERT(top);
5572         top->ld_ops->ldo_process_config(env, top, lcfg);
5573         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5574
5575         lu_site_purge(env, top->ld_site, -1);
5576
5577         m->mdt_child = NULL;
5578         m->mdt_bottom = NULL;
5579
5580         obd_disconnect(m->mdt_child_exp);
5581         m->mdt_child_exp = NULL;
5582
5583         obd_disconnect(m->mdt_bottom_exp);
5584         m->mdt_child_exp = NULL;
5585 }
5586
5587 static int mdt_connect_to_next(const struct lu_env *env, struct mdt_device *m,
5588                                const char *next, struct obd_export **exp)
5589 {
5590         struct obd_connect_data *data = NULL;
5591         struct obd_device       *obd;
5592         int                      rc;
5593
5594         ENTRY;
5595
5596         OBD_ALLOC_PTR(data);
5597         if (data == NULL)
5598                 GOTO(out, rc = -ENOMEM);
5599
5600         obd = class_name2obd(next);
5601         if (obd == NULL) {
5602                 CERROR("%s: can't locate next device: %s\n",
5603                        mdt_obd_name(m), next);
5604                 GOTO(out, rc = -ENOTCONN);
5605         }
5606
5607         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5608         data->ocd_version = LUSTRE_VERSION_CODE;
5609
5610         rc = obd_connect(NULL, exp, obd, &obd->obd_uuid, data, NULL);
5611         if (rc) {
5612                 CERROR("%s: cannot connect to next dev %s (%d)\n",
5613                        mdt_obd_name(m), next, rc);
5614                 GOTO(out, rc);
5615         }
5616
5617 out:
5618         if (data)
5619                 OBD_FREE_PTR(data);
5620         RETURN(rc);
5621 }
5622
5623 static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
5624                           struct lustre_cfg *cfg)
5625 {
5626         char                   *dev = lustre_cfg_string(cfg, 0);
5627         int                     rc, name_size, uuid_size;
5628         char                   *name, *uuid, *p;
5629         struct lustre_cfg_bufs *bufs;
5630         struct lustre_cfg      *lcfg;
5631         struct obd_device      *obd;
5632         struct lustre_profile  *lprof;
5633         struct lu_site         *site;
5634
5635         ENTRY;
5636
5637         /* in 1.8 we had the only device in the stack - MDS.
5638          * 2.0 introduces MDT, MDD, OSD; MDT starts others internally.
5639          * in 2.3 OSD is instantiated by obd_mount.c, so we need
5640          * to generate names and setup MDT, MDD. MDT will be using
5641          * generated name to connect to MDD. for MDD the next device
5642          * will be LOD with name taken from so called "profile" which
5643          * is generated by mount_option line
5644          *
5645          * 1.8 MGS generates config. commands like this:
5646          *   #06 (104)mount_option 0:  1:lustre-MDT0000  2:lustre-mdtlov
5647          *   #08 (120)setup   0:lustre-MDT0000  1:dev 2:type 3:lustre-MDT0000
5648          * 2.0 MGS generates config. commands like this:
5649          *   #07 (112)mount_option 0:  1:lustre-MDT0000  2:lustre-MDT0000-mdtlov
5650          *   #08 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5651          *                    3:lustre-MDT0000-mdtlov  4:f
5652          *
5653          * we generate MDD name from MDT one, just replacing T with D
5654          *
5655          * after all the preparations, the logical equivalent will be
5656          *   #01 (160)setup   0:lustre-MDD0000  1:lustre-MDD0000_UUID  2:0
5657          *                    3:lustre-MDT0000-mdtlov  4:f
5658          *   #02 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5659          *                    3:lustre-MDD0000  4:f
5660          *
5661          *  notice we build the stack from down to top: MDD first, then MDT
5662          */
5663
5664         name_size = MAX_OBD_NAME;
5665         uuid_size = MAX_OBD_NAME;
5666
5667         OBD_ALLOC(name, name_size);
5668         OBD_ALLOC(uuid, uuid_size);
5669         if (name == NULL || uuid == NULL)
5670                 GOTO(cleanup_mem, rc = -ENOMEM);
5671
5672         OBD_ALLOC_PTR(bufs);
5673         if (!bufs)
5674                 GOTO(cleanup_mem, rc = -ENOMEM);
5675
5676         strcpy(name, dev);
5677         p = strstr(name, "-MDT");
5678         if (p == NULL)
5679                 GOTO(free_bufs, rc = -ENOMEM);
5680         p[3] = 'D';
5681
5682         snprintf(uuid, MAX_OBD_NAME, "%s_UUID", name);
5683
5684         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5685         if (lprof == NULL || lprof->lp_dt == NULL) {
5686                 CERROR("can't find the profile: %s\n",
5687                        lustre_cfg_string(cfg, 0));
5688                 GOTO(free_bufs, rc = -EINVAL);
5689         }
5690
5691         lustre_cfg_bufs_reset(bufs, name);
5692         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_MDD_NAME);
5693         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5694         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5695
5696         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5697         if (!lcfg)
5698                 GOTO(put_profile, rc = -ENOMEM);
5699         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5700
5701         rc = class_attach(lcfg);
5702         if (rc)
5703                 GOTO(lcfg_cleanup, rc);
5704
5705         obd = class_name2obd(name);
5706         if (!obd) {
5707                 CERROR("Can not find obd %s (%s in config)\n",
5708                        MDD_OBD_NAME, lustre_cfg_string(cfg, 0));
5709                 GOTO(lcfg_cleanup, rc = -EINVAL);
5710         }
5711
5712         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5713
5714         lustre_cfg_bufs_reset(bufs, name);
5715         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5716         lustre_cfg_bufs_set_string(bufs, 2, dev);
5717         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5718
5719         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5720         if (!lcfg)
5721                 GOTO(class_detach, rc = -ENOMEM);
5722         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5723
5724         rc = class_setup(obd, lcfg);
5725         if (rc)
5726                 GOTO(class_detach, rc);
5727
5728         /* connect to MDD we just setup */
5729         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_child_exp);
5730         if (rc)
5731                 GOTO(class_detach, rc);
5732
5733         site = mdt->mdt_child_exp->exp_obd->obd_lu_dev->ld_site;
5734         LASSERT(site);
5735         LASSERT(mdt_lu_site(mdt) == NULL);
5736         mdt->mdt_lu_dev.ld_site = site;
5737         site->ls_top_dev = &mdt->mdt_lu_dev;
5738         mdt->mdt_child = lu2md_dev(mdt->mdt_child_exp->exp_obd->obd_lu_dev);
5739
5740         /* now connect to bottom OSD */
5741         snprintf(name, MAX_OBD_NAME, "%s-osd", dev);
5742         rc = mdt_connect_to_next(env, mdt, name, &mdt->mdt_bottom_exp);
5743         if (rc)
5744                 GOTO(class_detach, rc);
5745         mdt->mdt_bottom =
5746                 lu2dt_dev(mdt->mdt_bottom_exp->exp_obd->obd_lu_dev);
5747
5748         rc = lu_env_refill((struct lu_env *)env);
5749         if (rc != 0)
5750                 CERROR("Failure to refill session: '%d'\n", rc);
5751
5752         lu_dev_add_linkage(site, &mdt->mdt_lu_dev);
5753
5754         EXIT;
5755 class_detach:
5756         if (rc)
5757                 class_detach(obd, lcfg);
5758 lcfg_cleanup:
5759         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5760 put_profile:
5761         class_put_profile(lprof);
5762 free_bufs:
5763         OBD_FREE_PTR(bufs);
5764 cleanup_mem:
5765         if (name)
5766                 OBD_FREE(name, name_size);
5767         if (uuid)
5768                 OBD_FREE(uuid, uuid_size);
5769         RETURN(rc);
5770 }
5771
5772 /* setup quota master target on MDT0 */
5773 static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
5774                           struct lustre_cfg *cfg)
5775 {
5776         struct obd_device       *obd;
5777         char                    *dev = lustre_cfg_string(cfg, 0);
5778         char                    *qmtname, *uuid, *p;
5779         struct lustre_cfg_bufs  *bufs;
5780         struct lustre_cfg       *lcfg;
5781         struct lustre_profile   *lprof;
5782         struct obd_connect_data *data;
5783         int                      rc;
5784
5785         ENTRY;
5786
5787         LASSERT(mdt->mdt_qmt_exp == NULL);
5788         LASSERT(mdt->mdt_qmt_dev == NULL);
5789
5790         /* quota master is on MDT0 only for now */
5791         if (mdt->mdt_seq_site.ss_node_id != 0)
5792                 RETURN(0);
5793
5794         /* MGS generates config commands which look as follows:
5795          *   #01 (160)setup   0:lustre-MDT0000  1:lustre-MDT0000_UUID  2:0
5796          *                    3:lustre-MDT0000-mdtlov  4:f
5797          *
5798          * We generate the QMT name from the MDT one, just replacing MD with QM
5799          * after all the preparations, the logical equivalent will be:
5800          *   #01 (160)setup   0:lustre-QMT0000  1:lustre-QMT0000_UUID  2:0
5801          *                    3:lustre-MDT0000-osd  4:f
5802          */
5803         OBD_ALLOC(qmtname, MAX_OBD_NAME);
5804         OBD_ALLOC(uuid, UUID_MAX);
5805         OBD_ALLOC_PTR(bufs);
5806         OBD_ALLOC_PTR(data);
5807         if (qmtname == NULL || uuid == NULL || bufs == NULL || data == NULL)
5808                 GOTO(cleanup_mem, rc = -ENOMEM);
5809
5810         strcpy(qmtname, dev);
5811         p = strstr(qmtname, "-MDT");
5812         if (p == NULL)
5813                 GOTO(cleanup_mem, rc = -ENOMEM);
5814         /* replace MD with QM */
5815         p[1] = 'Q';
5816         p[2] = 'M';
5817
5818         snprintf(uuid, UUID_MAX, "%s_UUID", qmtname);
5819
5820         lprof = class_get_profile(lustre_cfg_string(cfg, 0));
5821         if (lprof == NULL || lprof->lp_dt == NULL) {
5822                 CERROR("can't find profile for %s\n",
5823                        lustre_cfg_string(cfg, 0));
5824                 GOTO(cleanup_mem, rc = -EINVAL);
5825         }
5826
5827         lustre_cfg_bufs_reset(bufs, qmtname);
5828         lustre_cfg_bufs_set_string(bufs, 1, LUSTRE_QMT_NAME);
5829         lustre_cfg_bufs_set_string(bufs, 2, uuid);
5830         lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
5831
5832         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5833         if (!lcfg)
5834                 GOTO(put_profile, rc = -ENOMEM);
5835         lustre_cfg_init(lcfg, LCFG_ATTACH, bufs);
5836
5837         rc = class_attach(lcfg);
5838         if (rc)
5839                 GOTO(lcfg_cleanup, rc);
5840
5841         obd = class_name2obd(qmtname);
5842         if (!obd) {
5843                 CERROR("Can not find obd %s (%s in config)\n", qmtname,
5844                        lustre_cfg_string(cfg, 0));
5845                 GOTO(lcfg_cleanup, rc = -EINVAL);
5846         }
5847
5848         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5849
5850         lustre_cfg_bufs_reset(bufs, qmtname);
5851         lustre_cfg_bufs_set_string(bufs, 1, uuid);
5852         lustre_cfg_bufs_set_string(bufs, 2, dev);
5853
5854         /* for quota, the next device should be the OSD device */
5855         lustre_cfg_bufs_set_string(bufs, 3,
5856                                    mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
5857
5858         OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
5859         if (!lcfg)
5860                 GOTO(class_detach, rc = -ENOMEM);
5861         lustre_cfg_init(lcfg, LCFG_SETUP, bufs);
5862
5863         rc = class_setup(obd, lcfg);
5864         if (rc)
5865                 GOTO(class_detach, rc);
5866
5867         mdt->mdt_qmt_dev = obd->obd_lu_dev;
5868
5869         /* configure local quota objects */
5870         if (CFS_FAIL_CHECK(OBD_FAIL_QUOTA_INIT))
5871                 rc = -EBADF;
5872         else
5873                 rc = mdt->mdt_qmt_dev->ld_ops->ldo_prepare(env,
5874                                                            &mdt->mdt_lu_dev,
5875                                                            mdt->mdt_qmt_dev);
5876         if (rc)
5877                 GOTO(class_cleanup, rc);
5878
5879         /* connect to quota master target */
5880         data->ocd_connect_flags = OBD_CONNECT_VERSION;
5881         data->ocd_version = LUSTRE_VERSION_CODE;
5882         rc = obd_connect(NULL, &mdt->mdt_qmt_exp, obd, &obd->obd_uuid,
5883                          data, NULL);
5884         if (rc) {
5885                 CERROR("cannot connect to quota master device %s (%d)\n",
5886                        qmtname, rc);
5887                 GOTO(class_cleanup, rc);
5888         }
5889
5890         EXIT;
5891 class_cleanup:
5892         if (rc) {
5893                 class_manual_cleanup(obd);
5894                 mdt->mdt_qmt_dev = NULL;
5895                 GOTO(lcfg_cleanup, rc);
5896         }
5897 class_detach:
5898         if (rc)
5899                 class_detach(obd, lcfg);
5900 lcfg_cleanup:
5901         OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
5902 put_profile:
5903         class_put_profile(lprof);
5904 cleanup_mem:
5905         if (bufs)
5906                 OBD_FREE_PTR(bufs);
5907         if (qmtname)
5908                 OBD_FREE(qmtname, MAX_OBD_NAME);
5909         if (uuid)
5910                 OBD_FREE(uuid, UUID_MAX);
5911         if (data)
5912                 OBD_FREE_PTR(data);
5913         return rc;
5914 }
5915
5916 /* Shutdown quota master target associated with mdt */
5917 static void mdt_quota_fini(const struct lu_env *env, struct mdt_device *mdt)
5918 {
5919         ENTRY;
5920
5921         if (mdt->mdt_qmt_exp == NULL)
5922                 RETURN_EXIT;
5923         LASSERT(mdt->mdt_qmt_dev != NULL);
5924
5925         /* the qmt automatically shuts down when the mdt disconnects */
5926         obd_disconnect(mdt->mdt_qmt_exp);
5927         mdt->mdt_qmt_exp = NULL;
5928         mdt->mdt_qmt_dev = NULL;
5929         EXIT;
5930 }
5931
5932 /* mdt_getxattr() is used from mdt_intent_getxattr(), use this wrapper
5933  * for now. This will be removed along with converting rest of MDT code
5934  * to use tgt_session_info
5935  */
5936 static int mdt_tgt_getxattr(struct tgt_session_info *tsi)
5937 {
5938         struct mdt_thread_info  *info = tsi2mdt_info(tsi);
5939         int                      rc;
5940
5941         if (unlikely(info->mti_object == NULL))
5942                 return -EPROTO;
5943
5944         rc = mdt_getxattr(info);
5945
5946         mdt_thread_info_fini(info);
5947         return rc;
5948 }
5949
5950 static int mdt_llog_open(struct tgt_session_info *tsi)
5951 {
5952         ENTRY;
5953
5954         if (!mdt_changelog_allow(tsi2mdt_info(tsi)))
5955                 RETURN(err_serious(-EACCES));
5956
5957         RETURN(tgt_llog_open(tsi));
5958 }
5959
5960 #define OBD_FAIL_OST_READ_NET   OBD_FAIL_OST_BRW_NET
5961 #define OBD_FAIL_OST_WRITE_NET  OBD_FAIL_OST_BRW_NET
5962 #define OST_BRW_READ    OST_READ
5963 #define OST_BRW_WRITE   OST_WRITE
5964
5965 static struct tgt_handler mdt_tgt_handlers[] = {
5966 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5967                 0,                      MDS_CONNECT,    mdt_tgt_connect,
5968                 &RQF_CONNECT, LUSTRE_OBD_VERSION),
5969 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5970                 0,                      MDS_DISCONNECT, tgt_disconnect,
5971                 &RQF_MDS_DISCONNECT, LUSTRE_OBD_VERSION),
5972 TGT_RPC_HANDLER(MDS_FIRST_OPC,
5973                 HAS_REPLY,              MDS_SET_INFO,   mdt_set_info,
5974                 &RQF_MDT_SET_INFO, LUSTRE_MDS_VERSION),
5975 TGT_MDT_HDL(0,                          MDS_GET_INFO,   mdt_get_info),
5976 TGT_MDT_HDL(HAS_REPLY,          MDS_GET_ROOT,   mdt_get_root),
5977 TGT_MDT_HDL(HAS_BODY,           MDS_GETATTR,    mdt_getattr),
5978 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_GETATTR_NAME,
5979                                                         mdt_getattr_name),
5980 TGT_MDT_HDL(HAS_BODY,           MDS_GETXATTR,   mdt_tgt_getxattr),
5981 TGT_MDT_HDL(HAS_REPLY,          MDS_STATFS,     mdt_statfs),
5982 TGT_MDT_HDL(IS_MUTABLE,         MDS_REINT,      mdt_reint),
5983 TGT_MDT_HDL(HAS_BODY,           MDS_CLOSE,      mdt_close),
5984 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_READPAGE,   mdt_readpage),
5985 TGT_MDT_HDL(HAS_BODY | HAS_REPLY,       MDS_SYNC,       mdt_sync),
5986 TGT_MDT_HDL(0,                          MDS_QUOTACTL,   mdt_quotactl),
5987 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_PROGRESS,
5988                                                         mdt_hsm_progress),
5989 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_REGISTER,
5990                                                         mdt_hsm_ct_register),
5991 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_CT_UNREGISTER,
5992                                                         mdt_hsm_ct_unregister),
5993 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_STATE_GET,
5994                                                         mdt_hsm_state_get),
5995 TGT_MDT_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, MDS_HSM_STATE_SET,
5996                                                         mdt_hsm_state_set),
5997 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_ACTION,       mdt_hsm_action),
5998 TGT_MDT_HDL(HAS_BODY | HAS_REPLY, MDS_HSM_REQUEST,
5999                                                         mdt_hsm_request),
6000 TGT_MDT_HDL(HAS_KEY | HAS_BODY | HAS_REPLY | IS_MUTABLE,
6001             MDS_SWAP_LAYOUTS,
6002             mdt_swap_layouts),
6003 TGT_MDT_HDL(IS_MUTABLE,         MDS_RMFID,      mdt_rmfid),
6004 TGT_MDT_HDL(IS_MUTABLE,         MDS_BATCH,      mdt_batch),
6005 };
6006
6007 static struct tgt_handler mdt_io_ops[] = {
6008 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY, OST_BRW_READ, tgt_brw_read,
6009                                                         mdt_hp_brw),
6010 TGT_OST_HDL_HP(HAS_BODY | IS_MUTABLE,    OST_BRW_WRITE, tgt_brw_write,
6011                                                         mdt_hp_brw),
6012 TGT_OST_HDL_HP(HAS_BODY | HAS_REPLY | IS_MUTABLE,
6013                                          OST_PUNCH,     mdt_punch_hdl,
6014                                                         mdt_hp_punch),
6015 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SYNC,     mdt_data_sync),
6016 TGT_OST_HDL(HAS_BODY | HAS_REPLY | IS_MUTABLE, OST_FALLOCATE,
6017                                                         mdt_fallocate_hdl),
6018 TGT_OST_HDL(HAS_BODY | HAS_REPLY, OST_SEEK, tgt_lseek),
6019 TGT_RPC_HANDLER(OST_FIRST_OPC,
6020                 0,                      OST_SET_INFO,   mdt_io_set_info,
6021                 &RQF_OBD_SET_INFO, LUSTRE_OST_VERSION),
6022 };
6023
6024 static struct tgt_handler mdt_sec_ctx_ops[] = {
6025 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT,     mdt_sec_ctx_handle),
6026 TGT_SEC_HDL_VAR(0,                      SEC_CTX_INIT_CONT, mdt_sec_ctx_handle),
6027 TGT_SEC_HDL_VAR(0,                      SEC_CTX_FINI,     mdt_sec_ctx_handle)
6028 };
6029
6030 static struct tgt_handler mdt_quota_ops[] = {
6031 TGT_QUOTA_HDL(HAS_REPLY,                QUOTA_DQACQ,      mdt_quota_dqacq),
6032 };
6033
6034 static struct tgt_handler mdt_llog_handlers[] = {
6035         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_CREATE,      mdt_llog_open),
6036         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK,  tgt_llog_next_block),
6037         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
6038         TGT_LLOG_HDL(0, LLOG_ORIGIN_HANDLE_PREV_BLOCK,  tgt_llog_prev_block),
6039 };
6040
6041 static struct tgt_opc_slice mdt_common_slice[] = {
6042         {
6043                 .tos_opc_start  = MDS_FIRST_OPC,
6044                 .tos_opc_end    = MDS_LAST_OPC,
6045                 .tos_hs         = mdt_tgt_handlers
6046         },
6047         {
6048                 .tos_opc_start  = OBD_FIRST_OPC,
6049                 .tos_opc_end    = OBD_LAST_OPC,
6050                 .tos_hs         = tgt_obd_handlers
6051         },
6052         {
6053                 .tos_opc_start  = LDLM_FIRST_OPC,
6054                 .tos_opc_end    = LDLM_LAST_OPC,
6055                 .tos_hs         = tgt_dlm_handlers
6056         },
6057         {
6058                 .tos_opc_start  = SEC_FIRST_OPC,
6059                 .tos_opc_end    = SEC_LAST_OPC,
6060                 .tos_hs         = mdt_sec_ctx_ops
6061         },
6062         {
6063                 .tos_opc_start  = OUT_UPDATE_FIRST_OPC,
6064                 .tos_opc_end    = OUT_UPDATE_LAST_OPC,
6065                 .tos_hs         = tgt_out_handlers
6066         },
6067         {
6068                 .tos_opc_start  = FLD_FIRST_OPC,
6069                 .tos_opc_end    = FLD_LAST_OPC,
6070                 .tos_hs         = fld_handlers
6071         },
6072         {
6073                 .tos_opc_start  = SEQ_FIRST_OPC,
6074                 .tos_opc_end    = SEQ_LAST_OPC,
6075                 .tos_hs         = seq_handlers
6076         },
6077         {
6078                 .tos_opc_start  = QUOTA_DQACQ,
6079                 .tos_opc_end    = QUOTA_LAST_OPC,
6080                 .tos_hs         = mdt_quota_ops
6081         },
6082         {
6083                 .tos_opc_start  = LLOG_FIRST_OPC,
6084                 .tos_opc_end    = LLOG_LAST_OPC,
6085                 .tos_hs         = mdt_llog_handlers
6086         },
6087         {
6088                 .tos_opc_start  = LFSCK_FIRST_OPC,
6089                 .tos_opc_end    = LFSCK_LAST_OPC,
6090                 .tos_hs         = tgt_lfsck_handlers
6091         },
6092         {
6093                 .tos_opc_start  = OST_FIRST_OPC,
6094                 .tos_opc_end    = OST_LAST_OPC,
6095                 .tos_hs         = mdt_io_ops
6096         },
6097         {
6098                 .tos_hs         = NULL
6099         }
6100 };
6101
6102 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
6103 {
6104         struct md_device *next = m->mdt_child;
6105         struct lu_device *d = &m->mdt_lu_dev;
6106         struct obd_device *obd = mdt2obd_dev(m);
6107         struct lfsck_stop stop;
6108
6109         ENTRY;
6110         stop.ls_status = LS_PAUSED;
6111         stop.ls_flags = 0;
6112         next->md_ops->mdo_iocontrol(env, next, OBD_IOC_STOP_LFSCK, 0, &stop);
6113
6114         mdt_stack_pre_fini(env, m, md2lu_dev(m->mdt_child));
6115
6116         mdt_restriper_stop(m);
6117         ping_evictor_stop();
6118
6119         /* Remove the HSM /proc entry so the coordinator cannot be
6120          * restarted by a user while it's shutting down.
6121          */
6122         mdt_hsm_cdt_stop(m);
6123
6124         mdt_llog_ctxt_unclone(env, m, LLOG_AGENT_ORIG_CTXT);
6125         mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
6126
6127         if (m->mdt_namespace != NULL)
6128                 ldlm_namespace_free_prior(m->mdt_namespace, NULL,
6129                                           d->ld_obd->obd_force);
6130
6131         obd_exports_barrier(obd);
6132         obd_zombie_barrier();
6133
6134         mdt_quota_fini(env, m);
6135
6136         cfs_free_nidlist(&m->mdt_squash.rsi_nosquash_nids);
6137
6138         /* Calling the cleanup functions in the same order as in the mdt_init0
6139          * error path
6140          */
6141         mdt_tunables_fini(m);
6142
6143         target_recovery_fini(obd);
6144         upcall_cache_cleanup(m->mdt_identity_cache);
6145         m->mdt_identity_cache = NULL;
6146
6147         tgt_fini(env, &m->mdt_lut);
6148
6149         mdt_hsm_cdt_fini(m);
6150
6151         if (m->mdt_los != NULL) {
6152                 local_oid_storage_fini(env, m->mdt_los);
6153                 m->mdt_los = NULL;
6154         }
6155
6156         if (m->mdt_namespace != NULL) {
6157                 ldlm_namespace_free_post(m->mdt_namespace);
6158                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
6159         }
6160
6161         if (m->mdt_md_root != NULL) {
6162                 mdt_object_put(env, m->mdt_md_root);
6163                 m->mdt_md_root = NULL;
6164         }
6165
6166         mdt_seq_fini(env, m);
6167
6168         mdt_fld_fini(env, m);
6169
6170         /*
6171          * Finish the stack
6172          */
6173         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6174
6175         LASSERT(atomic_read(&d->ld_ref) == 0);
6176
6177         server_put_mount(mdt_obd_name(m), true);
6178
6179         EXIT;
6180 }
6181
6182 static int mdt_postrecov(const struct lu_env *, struct mdt_device *);
6183
6184 static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
6185                      struct lu_device_type *ldt, struct lustre_cfg *cfg)
6186 {
6187         const struct dt_device_param *dt_conf;
6188         struct mdt_thread_info *info;
6189         struct obd_device *obd;
6190         const char *dev = lustre_cfg_string(cfg, 0);
6191         const char *num = lustre_cfg_string(cfg, 2);
6192         struct tg_grants_data *tgd = &m->mdt_lut.lut_tgd;
6193         struct lustre_mount_info *lmi = NULL;
6194         struct lustre_sb_info *lsi;
6195         struct lu_site *s;
6196         struct seq_server_site *ss_site;
6197         const char *identity_upcall = "NONE";
6198         struct md_device *next;
6199         struct lu_fid fid;
6200         int rc;
6201         long node_id;
6202         mntopt_t mntopts;
6203
6204         ENTRY;
6205
6206         lu_device_init(&m->mdt_lu_dev, ldt);
6207         /*
6208          * Environment (env) might be missing mdt_thread_key values at that
6209          * point, if device is allocated when mdt_thread_key is in QUIESCENT
6210          * mode.
6211          *
6212          * Usually device allocation path doesn't use module key values, but
6213          * mdt has to do a lot of work here, so allocate key value.
6214          */
6215         rc = lu_env_refill((struct lu_env *)env);
6216         if (rc != 0)
6217                 RETURN(rc);
6218
6219         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
6220         LASSERT(info != NULL);
6221
6222         obd = class_name2obd(dev);
6223         LASSERT(obd != NULL);
6224
6225         m->mdt_max_mdsize = MAX_MD_SIZE_OLD;
6226         m->mdt_evict_tgt_nids = 1;
6227         m->mdt_opts.mo_cos = MDT_COS_DEFAULT;
6228
6229         lmi = server_get_mount(dev);
6230         if (lmi == NULL) {
6231                 CERROR("Cannot get mount info for %s!\n", dev);
6232                 RETURN(-EFAULT);
6233         } else {
6234                 lsi = s2lsi(lmi->lmi_sb);
6235                 LASSERT(lsi->lsi_lmd);
6236                 /* CMD is supported only in IAM mode */
6237                 LASSERT(num);
6238                 rc = kstrtol(num, 10, &node_id);
6239                 if (rc)
6240                         RETURN(rc);
6241
6242                 obd_obt_init(obd);
6243                 if (test_bit(LMD_FLG_SKIP_LFSCK, lsi->lsi_lmd->lmd_flags))
6244                         m->mdt_skip_lfsck = 1;
6245                 if (test_bit(LMD_FLG_NO_CREATE, lsi->lsi_lmd->lmd_flags))
6246                         m->mdt_lut.lut_no_create = 1;
6247         }
6248
6249         /* Just try to get a DoM lock by default. Otherwise, having a group
6250          * lock granted, it may get blocked for a long time.
6251          */
6252         m->mdt_opts.mo_dom_lock = TRYLOCK_DOM_ON_OPEN;
6253         /* DoM files are read at open and data is packed in the reply */
6254         m->mdt_dom_read_open = 1;
6255
6256         m->mdt_squash.rsi_uid = 0;
6257         m->mdt_squash.rsi_gid = 0;
6258         INIT_LIST_HEAD(&m->mdt_squash.rsi_nosquash_nids);
6259         spin_lock_init(&m->mdt_squash.rsi_lock);
6260         spin_lock_init(&m->mdt_lock);
6261         m->mdt_enable_chprojid_gid = 0;
6262         m->mdt_enable_dir_migration = 1;
6263         m->mdt_enable_dir_restripe = 0;
6264         m->mdt_enable_dir_auto_split = 0;
6265         m->mdt_enable_parallel_rename_dir = 1;
6266         m->mdt_enable_parallel_rename_file = 1;
6267         m->mdt_enable_parallel_rename_crossdir = 1;
6268         m->mdt_enable_remote_dir = 1;
6269         m->mdt_enable_remote_dir_gid = 0;
6270         m->mdt_enable_remote_rename = 1;
6271         m->mdt_enable_striped_dir = 1;
6272         m->mdt_enable_dmv_implicit_inherit = 1;
6273         m->mdt_dir_restripe_nsonly = 1;
6274         m->mdt_max_mod_rpcs_in_flight = OBD_MAX_RIF_DEFAULT;
6275
6276         atomic_set(&m->mdt_mds_mds_conns, 0);
6277         atomic_set(&m->mdt_async_commit_count, 0);
6278         atomic_set(&m->mdt_dmv_old_client_count, 0);
6279
6280         m->mdt_lu_dev.ld_ops = &mdt_lu_ops;
6281         m->mdt_lu_dev.ld_obd = obd;
6282         /* Set this lu_device to obd for error handling purposes. */
6283         obd->obd_lu_dev = &m->mdt_lu_dev;
6284
6285         strncpy(m->mdt_job_xattr, XATTR_NAME_JOB_DEFAULT, XATTR_JOB_MAX_LEN);
6286
6287         /* init the stack */
6288         rc = mdt_stack_init((struct lu_env *)env, m, cfg);
6289         if (rc) {
6290                 CERROR("%s: Can't init device stack, rc %d\n",
6291                        mdt_obd_name(m), rc);
6292                 GOTO(err_lmi, rc);
6293         }
6294
6295         s = mdt_lu_site(m);
6296         ss_site = mdt_seq_site(m);
6297         s->ld_seq_site = ss_site;
6298         ss_site->ss_lu = s;
6299
6300         /* set server index */
6301         ss_site->ss_node_id = node_id;
6302
6303         /* failover is the default
6304          * FIXME: we do not failout mds0/mgs, which may cause some problems.
6305          * assumed whose ss_node_id == 0 XXX
6306          */
6307         obd->obd_replayable = 1;
6308         /* No connection accepted until configurations will finish */
6309         obd->obd_no_conn = 1;
6310
6311         if (cfg->lcfg_bufcount > 4 && LUSTRE_CFG_BUFLEN(cfg, 4) > 0) {
6312                 char *str = lustre_cfg_string(cfg, 4);
6313
6314                 if (strchr(str, 'n')) {
6315                         CWARN("%s: recovery disabled\n", mdt_obd_name(m));
6316                         obd->obd_replayable = 0;
6317                 }
6318         }
6319
6320         rc = mdt_fld_init(env, mdt_obd_name(m), m);
6321         if (rc)
6322                 GOTO(err_fini_stack, rc);
6323
6324         rc = mdt_seq_init(env, m);
6325         if (rc)
6326                 GOTO(err_fini_fld, rc);
6327
6328         snprintf(info->mti_u.ns_name, sizeof(info->mti_u.ns_name), "%s-%s",
6329                  LUSTRE_MDT_NAME, obd->obd_uuid.uuid);
6330         m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name,
6331                                               LDLM_NAMESPACE_SERVER,
6332                                               LDLM_NAMESPACE_GREEDY,
6333                                               LDLM_NS_TYPE_MDT);
6334         if (IS_ERR(m->mdt_namespace)) {
6335                 rc = PTR_ERR(m->mdt_namespace);
6336                 CERROR("%s: unable to create server namespace: rc = %d\n",
6337                        obd->obd_name, rc);
6338                 m->mdt_namespace = NULL;
6339                 GOTO(err_fini_seq, rc);
6340         }
6341
6342         m->mdt_namespace->ns_lvbp = m;
6343         m->mdt_namespace->ns_lvbo = &mdt_lvbo;
6344
6345         ldlm_register_intent(m->mdt_namespace, mdt_intent_policy);
6346         /* set obd_namespace for compatibility with old code */
6347         obd->obd_namespace = m->mdt_namespace;
6348
6349         rc = tgt_init(env, &m->mdt_lut, obd, m->mdt_bottom, mdt_common_slice,
6350                       OBD_FAIL_MDS_ALL_REQUEST_NET,
6351                       OBD_FAIL_MDS_ALL_REPLY_NET);
6352         if (rc)
6353                 GOTO(err_free_ns, rc);
6354
6355         /* Amount of available space excluded from granting and reserved
6356          * for metadata. It is a percentage of the total MDT size.
6357          */
6358         tgd->tgd_reserved_pcnt = 10;
6359
6360         if (ONE_MB_BRW_SIZE < (1U << tgd->tgd_blockbits))
6361                 m->mdt_brw_size = 1U << tgd->tgd_blockbits;
6362         else
6363                 m->mdt_brw_size = ONE_MB_BRW_SIZE;
6364
6365         if (CFS_FAIL_CHECK(OBD_FAIL_MDS_FS_SETUP))
6366                 GOTO(err_tgt, rc = -ENOENT);
6367
6368         fid.f_seq = FID_SEQ_LOCAL_NAME;
6369         fid.f_oid = 1;
6370         fid.f_ver = 0;
6371         rc = local_oid_storage_init(env, m->mdt_bottom, &fid, &m->mdt_los);
6372         if (rc != 0)
6373                 GOTO(err_tgt, rc);
6374
6375         rc = mdt_hsm_cdt_init(m);
6376         if (rc != 0) {
6377                 CERROR("%s: error initializing coordinator, rc %d\n",
6378                        mdt_obd_name(m), rc);
6379                 GOTO(err_los_fini, rc);
6380         }
6381
6382         tgt_adapt_sptlrpc_conf(&m->mdt_lut);
6383
6384         next = m->mdt_child;
6385         dt_conf = next->md_ops->mdo_dtconf_get(env, next);
6386
6387         mntopts = dt_conf->ddp_mntopts;
6388
6389         if (mntopts & MNTOPT_USERXATTR)
6390                 m->mdt_opts.mo_user_xattr = 1;
6391         else
6392                 m->mdt_opts.mo_user_xattr = 0;
6393
6394         m->mdt_max_ea_size = dt_conf->ddp_max_ea_size;
6395
6396         if (mntopts & MNTOPT_ACL)
6397                 m->mdt_opts.mo_acl = 1;
6398         else
6399                 m->mdt_opts.mo_acl = 0;
6400
6401         m->mdt_enable_strict_som = 1;
6402
6403         /* XXX: to support suppgid for ACL, we enable identity_upcall
6404          * by default, otherwise, maybe got unexpected -EACCESS.
6405          */
6406         if (m->mdt_opts.mo_acl)
6407                 identity_upcall = MDT_IDENTITY_UPCALL_PATH;
6408
6409         m->mdt_identity_cache = upcall_cache_init(mdt_obd_name(m),
6410                                                 identity_upcall,
6411                                                 UC_IDCACHE_HASH_SIZE,
6412                                                 1200, /* entry expire: 20 mn */
6413                                                 30, /* acquire expire: 30 s */
6414                                                 true, /* acquire can replay */
6415                                                 &mdt_identity_upcall_cache_ops);
6416         if (IS_ERR(m->mdt_identity_cache)) {
6417                 rc = PTR_ERR(m->mdt_identity_cache);
6418                 m->mdt_identity_cache = NULL;
6419                 GOTO(err_free_hsm, rc);
6420         }
6421
6422         rc = mdt_tunables_init(m, dev);
6423         if (rc) {
6424                 CERROR("Can't init MDT lprocfs, rc %d\n", rc);
6425                 GOTO(err_recovery, rc);
6426         }
6427
6428         rc = mdt_quota_init(env, m, cfg);
6429         if (rc)
6430                 GOTO(err_procfs, rc);
6431
6432         m->mdt_ldlm_client = &mdt2obd_dev(m)->obd_ldlm_client;
6433         ptlrpc_init_client(LDLM_CB_REQUEST_PORTAL, LDLM_CB_REPLY_PORTAL,
6434                            "mdt_ldlm_client", m->mdt_ldlm_client);
6435
6436         ping_evictor_start();
6437
6438         /* recovery will be started upon mdt_prepare() when the whole stack is
6439          * complete and ready to serve the requests
6440          */
6441
6442         /* Reduce the initial timeout on an MDS because it doesn't need such
6443          * a long timeout as an OST does. Adaptive timeouts will adjust this
6444          * value appropriately.
6445          */
6446         if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT)
6447                 ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT;
6448
6449         if (test_bit(LMD_FLG_LOCAL_RECOV, lsi->lsi_lmd->lmd_flags))
6450                 m->mdt_lut.lut_local_recovery = 1;
6451
6452         rc = mdt_restriper_start(m);
6453         if (rc)
6454                 GOTO(err_ping_evictor, rc);
6455
6456         RETURN(0);
6457
6458 err_ping_evictor:
6459         ping_evictor_stop();
6460 err_procfs:
6461         mdt_tunables_fini(m);
6462 err_recovery:
6463         upcall_cache_cleanup(m->mdt_identity_cache);
6464         m->mdt_identity_cache = NULL;
6465 err_free_hsm:
6466         mdt_hsm_cdt_fini(m);
6467 err_los_fini:
6468         local_oid_storage_fini(env, m->mdt_los);
6469         m->mdt_los = NULL;
6470 err_tgt:
6471         /* keep recoverable clients */
6472         obd->obd_fail = 1;
6473         target_recovery_fini(obd);
6474         obd_exports_barrier(obd);
6475         obd_zombie_barrier();
6476         tgt_fini(env, &m->mdt_lut);
6477 err_free_ns:
6478         ldlm_namespace_free(m->mdt_namespace, NULL, 0);
6479         obd->obd_namespace = m->mdt_namespace = NULL;
6480 err_fini_seq:
6481         mdt_seq_fini(env, m);
6482 err_fini_fld:
6483         mdt_fld_fini(env, m);
6484 err_fini_stack:
6485         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
6486 err_lmi:
6487         if (lmi)
6488                 server_put_mount(dev, true);
6489         return rc;
6490 }
6491
6492 /* For interoperability, the left element is old parameter, the right one
6493  * is the new version of the parameter, if some parameter is deprecated,
6494  * the new version should be set as NULL.
6495  */
6496 static struct cfg_interop_param mdt_interop_param[] = {
6497         { "mdt.group_upcall",   NULL },
6498         { "mdt.quota_type",     NULL },
6499         { "mdd.quota_type",     NULL },
6500         { "mdt.som",            NULL },
6501         { "mdt.rootsquash",     "mdt.root_squash" },
6502         { "mdt.nosquash_nid",   "mdt.nosquash_nids" },
6503         { NULL }
6504 };
6505
6506 /* used by MGS to process specific configurations */
6507 static int mdt_process_config(const struct lu_env *env,
6508                               struct lu_device *d, struct lustre_cfg *cfg)
6509 {
6510         struct mdt_device *m = mdt_dev(d);
6511         struct md_device *md_next = m->mdt_child;
6512         struct lu_device *next = md2lu_dev(md_next);
6513         int rc;
6514
6515         ENTRY;
6516
6517         switch (cfg->lcfg_command) {
6518         case LCFG_PARAM: {
6519                 struct obd_device *obd = d->ld_obd;
6520                 /* For interoperability */
6521                 struct cfg_interop_param *ptr = NULL;
6522                 struct lustre_cfg *old_cfg = NULL;
6523                 char *param = NULL;
6524                 ssize_t count;
6525
6526                 param = lustre_cfg_string(cfg, 1);
6527                 if (param == NULL) {
6528                         CERROR("param is empty\n");
6529                         rc = -EINVAL;
6530                         break;
6531                 }
6532
6533                 ptr = class_find_old_param(param, mdt_interop_param);
6534                 if (ptr != NULL) {
6535                         if (ptr->new_param == NULL) {
6536                                 rc = 0;
6537                                 CWARN("For interoperability, skip this %s. It is obsolete.\n",
6538                                       ptr->old_param);
6539                                 break;
6540                         }
6541
6542                         CWARN("Found old param %s, changed it to %s.\n",
6543                               ptr->old_param, ptr->new_param);
6544
6545                         old_cfg = cfg;
6546                         cfg = lustre_cfg_rename(old_cfg, ptr->new_param);
6547                         if (IS_ERR(cfg)) {
6548                                 rc = PTR_ERR(cfg);
6549                                 break;
6550                         }
6551                 }
6552
6553                 count = class_modify_config(cfg, PARAM_MDT,
6554                                             &obd->obd_kset.kobj);
6555                 if (count < 0) {
6556                         struct coordinator *cdt = &m->mdt_coordinator;
6557
6558                         /* is it an HSM var ? */
6559                         count = class_modify_config(cfg, PARAM_HSM,
6560                                                     &cdt->cdt_hsm_kobj);
6561                         if (count < 0)
6562                                 /* we don't understand; pass it on */
6563                                 rc = next->ld_ops->ldo_process_config(env, next,
6564                                                                       cfg);
6565                         else
6566                                 rc = count > 0 ? 0 : count;
6567                 } else {
6568                         rc = count > 0 ? 0 : count;
6569                 }
6570
6571                 if (old_cfg)
6572                         OBD_FREE(cfg, lustre_cfg_len(cfg->lcfg_bufcount,
6573                                                      cfg->lcfg_buflens));
6574                 break;
6575         }
6576         default:
6577                 /* others are passed further */
6578                 rc = next->ld_ops->ldo_process_config(env, next, cfg);
6579                 break;
6580         }
6581         RETURN(rc);
6582 }
6583
6584 static struct lu_object *mdt_object_alloc(const struct lu_env *env,
6585                                           const struct lu_object_header *hdr,
6586                                           struct lu_device *d)
6587 {
6588         struct mdt_object *mo;
6589
6590         ENTRY;
6591
6592         OBD_SLAB_ALLOC_PTR_GFP(mo, mdt_object_kmem, GFP_NOFS);
6593         if (mo != NULL) {
6594                 struct lu_object *o;
6595                 struct lu_object_header *h;
6596
6597                 o = &mo->mot_obj;
6598                 h = &mo->mot_header;
6599                 lu_object_header_init(h);
6600                 lu_object_init(o, h, d);
6601                 lu_object_add_top(h, o);
6602                 o->lo_ops = &mdt_obj_ops;
6603                 spin_lock_init(&mo->mot_write_lock);
6604                 mutex_init(&mo->mot_som_mutex);
6605                 mutex_init(&mo->mot_lov_mutex);
6606                 init_rwsem(&mo->mot_dom_sem);
6607                 init_rwsem(&mo->mot_open_sem);
6608                 atomic_set(&mo->mot_open_count, 0);
6609                 mo->mot_restripe_offset = 0;
6610                 INIT_LIST_HEAD(&mo->mot_restripe_linkage);
6611                 mo->mot_lsom_size = 0;
6612                 mo->mot_lsom_blocks = 0;
6613                 mo->mot_lsom_inited = false;
6614                 RETURN(o);
6615         }
6616         RETURN(NULL);
6617 }
6618
6619 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
6620                            const struct lu_object_conf *unused)
6621 {
6622         struct mdt_device *d = mdt_dev(o->lo_dev);
6623         struct lu_device  *under;
6624         struct lu_object  *below;
6625         int                rc = 0;
6626
6627         ENTRY;
6628
6629         CDEBUG(D_INFO, "object init, fid = "DFID"\n",
6630                PFID(lu_object_fid(o)));
6631
6632         under = &d->mdt_child->md_lu_dev;
6633         below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
6634         if (below != NULL)
6635                 lu_object_add(o, below);
6636         else
6637                 rc = -ENOMEM;
6638
6639         RETURN(rc);
6640 }
6641
6642 static void mdt_object_free_rcu(struct rcu_head *head)
6643 {
6644         struct mdt_object *mo = container_of(head, struct mdt_object,
6645                                              mot_header.loh_rcu);
6646
6647         kmem_cache_free(mdt_object_kmem, mo);
6648 }
6649
6650 static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
6651 {
6652         struct mdt_object *mo = mdt_obj(o);
6653         struct lu_object_header *h;
6654
6655         ENTRY;
6656
6657         h = o->lo_header;
6658         CDEBUG(D_INFO, "object free, fid = "DFID"\n",
6659                PFID(lu_object_fid(o)));
6660
6661         LASSERT(atomic_read(&mo->mot_open_count) == 0);
6662         LASSERT(atomic_read(&mo->mot_lease_count) == 0);
6663
6664         lu_object_fini(o);
6665         lu_object_header_fini(h);
6666         OBD_FREE_PRE(mo, sizeof(*mo), "slab-freed");
6667         call_rcu(&mo->mot_header.loh_rcu, mdt_object_free_rcu);
6668
6669         EXIT;
6670 }
6671
6672 static int mdt_object_print(const struct lu_env *env, void *cookie,
6673                             lu_printer_t p, const struct lu_object *o)
6674 {
6675         struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
6676
6677         return (*p)(env, cookie,
6678                     LUSTRE_MDT_NAME"-object@%p(%s %s, writecount=%d)",
6679                     mdto, mdto->mot_lov_created ? "lov_created" : "",
6680                     mdto->mot_cache_attr ? "cache_attr" : "",
6681                     mdto->mot_write_count);
6682 }
6683
6684 static int mdt_prepare(const struct lu_env *env,
6685                 struct lu_device *pdev,
6686                 struct lu_device *cdev)
6687 {
6688         struct mdt_device *mdt = mdt_dev(cdev);
6689         struct lu_device *next = &mdt->mdt_child->md_lu_dev;
6690         struct obd_device *obd = cdev->ld_obd;
6691         int rc;
6692
6693         ENTRY;
6694
6695         LASSERT(obd);
6696
6697         rc = next->ld_ops->ldo_prepare(env, cdev, next);
6698         if (rc)
6699                 RETURN(rc);
6700
6701         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_CHANGELOG_ORIG_CTXT);
6702         if (rc)
6703                 RETURN(rc);
6704
6705         rc = mdt_llog_ctxt_clone(env, mdt, LLOG_AGENT_ORIG_CTXT);
6706         if (rc)
6707                 RETURN(rc);
6708
6709         rc = lfsck_register_namespace(env, mdt->mdt_bottom, mdt->mdt_namespace);
6710         /* The LFSCK instance is registered just now, so it must be there when
6711          * register the namespace to such instance.
6712          */
6713         LASSERTF(rc == 0, "register namespace failed: rc = %d\n", rc);
6714
6715         if (mdt->mdt_seq_site.ss_node_id == 0) {
6716                 rc = mdt->mdt_child->md_ops->mdo_root_get(env, mdt->mdt_child,
6717                                                          &mdt->mdt_md_root_fid);
6718                 if (rc)
6719                         RETURN(rc);
6720         }
6721
6722         LASSERT(!test_bit(MDT_FL_CFGLOG, &mdt->mdt_state));
6723
6724         target_recovery_init(&mdt->mdt_lut, tgt_request_handle);
6725         set_bit(MDT_FL_CFGLOG, &mdt->mdt_state);
6726         LASSERT(obd->obd_no_conn);
6727         spin_lock(&obd->obd_dev_lock);
6728         obd->obd_no_conn = 0;
6729         spin_unlock(&obd->obd_dev_lock);
6730
6731         if (obd->obd_recovering == 0)
6732                 mdt_postrecov(env, mdt);
6733
6734         RETURN(rc);
6735 }
6736
6737 const struct lu_device_operations mdt_lu_ops = {
6738         .ldo_object_alloc   = mdt_object_alloc,
6739         .ldo_process_config = mdt_process_config,
6740         .ldo_prepare        = mdt_prepare,
6741 };
6742
6743 static const struct lu_object_operations mdt_obj_ops = {
6744         .loo_object_init    = mdt_object_init,
6745         .loo_object_free    = mdt_object_free,
6746         .loo_object_print   = mdt_object_print
6747 };
6748
6749 static int mdt_obd_set_info_async(const struct lu_env *env,
6750                                   struct obd_export *exp,
6751                                   __u32 keylen, void *key,
6752                                   __u32 vallen, void *val,
6753                                   struct ptlrpc_request_set *set)
6754 {
6755         int rc;
6756
6757         ENTRY;
6758
6759         if (KEY_IS(KEY_SPTLRPC_CONF)) {
6760                 rc = tgt_adapt_sptlrpc_conf(class_exp2tgt(exp));
6761                 RETURN(rc);
6762         }
6763
6764         RETURN(0);
6765 }
6766
6767 static inline void mdt_enable_slc(struct mdt_device *mdt)
6768 {
6769         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_NEVER)
6770                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_BLOCKING;
6771 }
6772
6773 static inline void mdt_disable_slc(struct mdt_device *mdt)
6774 {
6775         if (mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING)
6776                 mdt->mdt_lut.lut_sync_lock_cancel = SYNC_LOCK_CANCEL_NEVER;
6777 }
6778
6779 /**
6780  * Match client and server connection feature flags.
6781  *
6782  * Compute the compatibility flags for a connection request based on
6783  * features mutually supported by client and server.
6784  *
6785  * The obd_export::exp_connect_data.ocd_connect_flags field in \a exp
6786  * must not be updated here, otherwise a partially initialized value may
6787  * be exposed. After the connection request is successfully processed,
6788  * the top-level MDT connect request handler atomically updates the export
6789  * connect flags from the obd_connect_data::ocd_connect_flags field of the
6790  * reply. \see mdt_connect().
6791  *
6792  * Before 2.7.50 clients will send a struct obd_connect_data_v1 rather than a
6793  * full struct obd_connect_data. So care must be taken when accessing fields
6794  * that are not present in struct obd_connect_data_v1. See LU-16.
6795  *
6796  * \param exp   the obd_export associated with this client/target pair
6797  * \param mdt   the target device for the connection
6798  * \param data  stores data for this connect request
6799  *
6800  * \retval 0       success
6801  * \retval -EPROTO \a data unexpectedly has zero obd_connect_data::ocd_brw_size
6802  * \retval -EBADE  client and server feature requirements are incompatible
6803  */
6804 static int mdt_connect_internal(const struct lu_env *env,
6805                                 struct obd_export *exp,
6806                                 struct mdt_device *mdt,
6807                                 struct obd_connect_data *data, bool reconnect)
6808 {
6809         const char *obd_name = mdt_obd_name(mdt);
6810
6811         LASSERT(data != NULL);
6812
6813         data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED;
6814
6815         if (mdt->mdt_bottom->dd_rdonly &&
6816             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6817             !(data->ocd_connect_flags & OBD_CONNECT_RDONLY))
6818                 RETURN(-EACCES);
6819
6820         if (data->ocd_connect_flags & OBD_CONNECT_FLAGS2)
6821                 data->ocd_connect_flags2 &= MDT_CONNECT_SUPPORTED2;
6822
6823         data->ocd_ibits_known &= MDS_INODELOCK_FULL;
6824
6825         if (!mdt->mdt_opts.mo_acl)
6826                 data->ocd_connect_flags &= ~OBD_CONNECT_ACL;
6827
6828         if (!mdt->mdt_opts.mo_user_xattr)
6829                 data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
6830
6831         if (OCD_HAS_FLAG(data, BRW_SIZE)) {
6832                 data->ocd_brw_size = min(data->ocd_brw_size,
6833                                          mdt->mdt_brw_size);
6834                 if (data->ocd_brw_size == 0) {
6835                         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,
6836                                exp, data->ocd_connect_flags, data->ocd_version,
6837                                data->ocd_grant, data->ocd_index);
6838                         return -EPROTO;
6839                 }
6840         }
6841
6842         if (OCD_HAS_FLAG(data, GRANT_PARAM)) {
6843                 struct dt_device_param *ddp = &mdt->mdt_lut.lut_dt_conf;
6844
6845                 /* client is reporting its page size, for future use */
6846                 exp->exp_target_data.ted_pagebits = data->ocd_grant_blkbits;
6847                 data->ocd_grant_blkbits  = mdt->mdt_lut.lut_tgd.tgd_blockbits;
6848                 /* ddp_inodespace may not be power-of-two value, eg. for ldiskfs
6849                  * it's LDISKFS_DIR_REC_LEN(20) = 28.
6850                  */
6851                 data->ocd_grant_inobits = fls(ddp->ddp_inodespace - 1);
6852                 /* ocd_grant_tax_kb is in 1K byte blocks */
6853                 data->ocd_grant_tax_kb = ddp->ddp_extent_tax >> 10;
6854                 data->ocd_grant_max_blks = ddp->ddp_max_extent_blks;
6855         }
6856
6857         /* Save connect_data we have so far because tgt_grant_connect()
6858          * uses it to calculate grant, and we want to save the client
6859          * version before it is overwritten by LUSTRE_VERSION_CODE.
6860          */
6861         exp->exp_connect_data = *data;
6862         if (OCD_HAS_FLAG(data, GRANT))
6863                 tgt_grant_connect(env, exp, data, !reconnect);
6864
6865         if (OCD_HAS_FLAG(data, MAXBYTES))
6866                 data->ocd_maxbytes = mdt->mdt_lut.lut_dt_conf.ddp_maxbytes;
6867
6868         /* NB: Disregard the rule against updating
6869          * exp_connect_data.ocd_connect_flags in this case, since
6870          * tgt_client_new() needs to know if this is a lightweight
6871          * connection, and it is safe to expose this flag before
6872          * connection processing completes.
6873          */
6874         if (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) {
6875                 spin_lock(&exp->exp_lock);
6876                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_LIGHTWEIGHT;
6877                 spin_unlock(&exp->exp_lock);
6878         }
6879
6880         data->ocd_version = LUSTRE_VERSION_CODE;
6881
6882         if ((data->ocd_connect_flags & OBD_CONNECT_FID) == 0) {
6883                 CWARN("%s: MDS requires FID support, but client not\n",
6884                       obd_name);
6885                 return -EBADE;
6886         }
6887
6888         if (OCD_HAS_FLAG(data, PINGLESS)) {
6889                 if (ptlrpc_pinger_suppress_pings()) {
6890                         spin_lock(&exp->exp_obd->obd_dev_lock);
6891                         list_del_init(&exp->exp_obd_chain_timed);
6892                         spin_unlock(&exp->exp_obd->obd_dev_lock);
6893                 } else {
6894                         data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
6895                 }
6896         }
6897
6898         data->ocd_max_easize = mdt->mdt_max_ea_size;
6899
6900         /* NB: Disregard the rule against updating
6901          * exp_connect_data.ocd_connect_flags in this case, since
6902          * tgt_client_new() needs to know if this is client supports
6903          * multiple modify RPCs, and it is safe to expose this flag before
6904          * connection processing completes.
6905          */
6906         if (data->ocd_connect_flags & OBD_CONNECT_MULTIMODRPCS) {
6907                 if (mdt_max_mod_rpcs_changed(mdt))
6908                         /* The new mdt.*.max_mod_rpcs_in_flight parameter
6909                          * has not changed since initialization, but the
6910                          * deprecated module parameter was changed,
6911                          * so use that instead.
6912                          */
6913                         data->ocd_maxmodrpcs = max_mod_rpcs_per_client;
6914                 else
6915                         data->ocd_maxmodrpcs = mdt->mdt_max_mod_rpcs_in_flight;
6916                 spin_lock(&exp->exp_lock);
6917                 *exp_connect_flags_ptr(exp) |= OBD_CONNECT_MULTIMODRPCS;
6918                 spin_unlock(&exp->exp_lock);
6919         }
6920
6921         if (OCD_HAS_FLAG(data, CKSUM)) {
6922                 __u32 cksum_types = data->ocd_cksum_types;
6923
6924                 tgt_mask_cksum_types(&mdt->mdt_lut, &data->ocd_cksum_types);
6925
6926                 if (unlikely(data->ocd_cksum_types == 0)) {
6927                         CERROR("%s: Connect with checksum support but no ocd_cksum_types is set\n",
6928                                exp->exp_obd->obd_name);
6929                         RETURN(-EPROTO);
6930                 }
6931
6932                 CDEBUG(D_RPCTRACE, "%s: cli %s supports cksum type %x, return %x\n",
6933                        exp->exp_obd->obd_name, obd_export_nid2str(exp),
6934                        cksum_types, data->ocd_cksum_types);
6935         } else {
6936                 /* Client not support OBD_CONNECT_CKSUM? fall back to CRC32 */
6937                 CDEBUG(D_RPCTRACE, "%s: cli %s does not support OBD_CONNECT_CKSUM, CRC32 will be used\n",
6938                        exp->exp_obd->obd_name, obd_export_nid2str(exp));
6939         }
6940
6941         if ((data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
6942             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
6943                 atomic_inc(&mdt->mdt_mds_mds_conns);
6944                 mdt_enable_slc(mdt);
6945         }
6946
6947         if (!mdt->mdt_lut.lut_dt_conf.ddp_has_lseek_data_hole)
6948                 data->ocd_connect_flags2 &= ~OBD_CONNECT2_LSEEK;
6949
6950         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
6951             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT)) {
6952                 atomic_inc(&mdt->mdt_dmv_old_client_count);
6953                 mdt->mdt_enable_dmv_implicit_inherit = 0;
6954         }
6955
6956         return 0;
6957 }
6958
6959 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
6960                                    struct mdt_thread_info *info,
6961                                    struct mdt_file_data *mfd)
6962 {
6963         struct lu_context ses;
6964         int rc;
6965
6966         ENTRY;
6967
6968         rc = lu_context_init(&ses, LCT_SERVER_SESSION);
6969         if (rc)
6970                 RETURN(rc);
6971
6972         env->le_ses = &ses;
6973         lu_context_enter(&ses);
6974
6975         mdt_ucred(info)->uc_valid = UCRED_OLD;
6976         /* do not let rbac interfere with dirty flag internal system event */
6977         mdt_ucred(info)->uc_rbac_file_perms = 1;
6978         mdt_ucred(info)->uc_rbac_dne_ops = 1;
6979         mdt_ucred(info)->uc_rbac_quota_ops = 1;
6980         mdt_ucred(info)->uc_rbac_byfid_ops = 1;
6981         mdt_ucred(info)->uc_rbac_chlg_ops = 1;
6982         mdt_ucred(info)->uc_rbac_fscrypt_admin = 1;
6983         rc = mdt_add_dirty_flag(info, mfd->mfd_object, &info->mti_attr);
6984
6985         lu_context_exit(&ses);
6986         lu_context_fini(&ses);
6987         env->le_ses = NULL;
6988
6989         RETURN(rc);
6990 }
6991
6992 static int mdt_export_cleanup(struct obd_export *exp)
6993 {
6994         LIST_HEAD(closing_list);
6995         struct mdt_export_data  *med = &exp->exp_mdt_data;
6996         struct obd_device       *obd = exp->exp_obd;
6997         struct mdt_device       *mdt;
6998         struct mdt_thread_info  *info;
6999         struct lu_env            env;
7000         struct mdt_file_data    *mfd, *n;
7001         int rc = 0;
7002
7003         ENTRY;
7004
7005         spin_lock(&med->med_open_lock);
7006         while (!list_empty(&med->med_open_head)) {
7007                 struct list_head *tmp = med->med_open_head.next;
7008
7009                 mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
7010
7011                 /* Remove mfd handle so it can't be found again.
7012                  * We are consuming the mfd_list reference here.
7013                  */
7014                 class_handle_unhash(&mfd->mfd_open_handle);
7015                 list_move_tail(&mfd->mfd_list, &closing_list);
7016         }
7017         spin_unlock(&med->med_open_lock);
7018         mdt = mdt_dev(obd->obd_lu_dev);
7019         LASSERT(mdt != NULL);
7020
7021         rc = lu_env_init(&env, LCT_MD_THREAD);
7022         if (rc)
7023                 RETURN(rc);
7024
7025         info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7026         LASSERT(info != NULL);
7027         memset(info, 0, sizeof(*info));
7028         info->mti_env = &env;
7029         info->mti_mdt = mdt;
7030         info->mti_exp = exp;
7031
7032         if (!list_empty(&closing_list)) {
7033                 struct md_attr *ma = &info->mti_attr;
7034
7035                 /* Close any open files (which may cause orphan unlinking). */
7036                 list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
7037                         list_del_init(&mfd->mfd_list);
7038                         ma->ma_need = ma->ma_valid = 0;
7039
7040                         /* This file is being closed due to an eviction, it
7041                          * could have been modified and now dirty regarding to
7042                          * HSM archive, check this!
7043                          * The logic here is to mark a file dirty if there's a
7044                          * chance it was dirtied before the client was evicted,
7045                          * so that we don't have to wait for a release attempt
7046                          * before finding out the file was actually dirty and
7047                          * fail the release. Aggressively marking it dirty here
7048                          * will cause the policy engine to attempt to
7049                          * re-archive it; when rearchiving, we can compare the
7050                          * current version to the HSM data_version and make the
7051                          * archive request into a noop if it's not actually
7052                          * dirty.
7053                          */
7054                         if (mfd->mfd_open_flags & MDS_FMODE_WRITE)
7055                                 rc = mdt_ctxt_add_dirty_flag(&env, info, mfd);
7056
7057                         /* Don't unlink orphan on failover umount, LU-184 */
7058                         if (exp->exp_flags & OBD_OPT_FAILOVER ||
7059                             exp->exp_obd->obd_stopping) {
7060                                 ma->ma_valid = MA_FLAGS;
7061                                 ma->ma_attr_flags |= MDS_KEEP_ORPHAN;
7062                         }
7063                         ma->ma_valid |= MA_FORCE_LOG;
7064                         mdt_mfd_close(info, mfd);
7065                 }
7066         }
7067         info->mti_mdt = NULL;
7068         /* cleanup client slot early */
7069         /* Do not erase record for recoverable client. */
7070         if (!(exp->exp_flags & OBD_OPT_FAILOVER) || exp->exp_failed)
7071                 tgt_client_del(&env, exp);
7072         lu_env_fini(&env);
7073
7074         RETURN(rc);
7075 }
7076
7077 static int mdt_obd_disconnect(struct obd_export *exp)
7078 {
7079         struct obd_connect_data *data = &exp->exp_connect_data;
7080         struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev);
7081         int rc;
7082
7083         ENTRY;
7084
7085         LASSERT(exp);
7086         class_export_get(exp);
7087
7088         if (OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7089             atomic_dec_and_test(&mdt->mdt_mds_mds_conns))
7090                 mdt_disable_slc(mdt);
7091
7092         if (!OCD_HAS_FLAG(data, MDS_MDS) && !OCD_HAS_FLAG(data, LIGHTWEIGHT) &&
7093             !OCD_HAS_FLAG2(data, DMV_IMP_INHERIT) &&
7094             atomic_dec_and_test(&mdt->mdt_dmv_old_client_count))
7095                 mdt->mdt_enable_dmv_implicit_inherit = 1;
7096
7097         rc = server_disconnect_export(exp);
7098         if (rc != 0)
7099                 CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
7100
7101         tgt_grant_discard(exp);
7102
7103         if (!(exp->exp_flags & OBD_OPT_FORCE))
7104                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7105
7106         rc = mdt_export_cleanup(exp);
7107         nodemap_del_member(exp);
7108         class_export_put(exp);
7109         RETURN(rc);
7110 }
7111
7112 /* mds_connect copy */
7113 static int mdt_obd_connect(const struct lu_env *env,
7114                            struct obd_export **exp, struct obd_device *obd,
7115                            struct obd_uuid *cluuid,
7116                            struct obd_connect_data *data,
7117                            void *localdata)
7118 {
7119         struct obd_export       *lexp;
7120         struct lustre_handle    conn = { 0 };
7121         struct mdt_device       *mdt;
7122         int                      rc;
7123         struct lnet_nid         *client_nid = localdata;
7124
7125         ENTRY;
7126
7127         LASSERT(env != NULL);
7128         LASSERT(data != NULL);
7129
7130         if (!exp || !obd || !cluuid)
7131                 RETURN(-EINVAL);
7132
7133         mdt = mdt_dev(obd->obd_lu_dev);
7134
7135         /*
7136          * first, check whether the stack is ready to handle requests
7137          * XXX: probably not very appropriate method is used now
7138          *      at some point we should find a better one
7139          */
7140         if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) &&
7141             !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) &&
7142             !(data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
7143                 rc = obd_get_info(env, mdt->mdt_child_exp,
7144                                   sizeof(KEY_OSP_CONNECTED),
7145                                   KEY_OSP_CONNECTED, NULL, NULL);
7146                 if (rc)
7147                         RETURN(-EAGAIN);
7148                 set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
7149         }
7150
7151         rc = class_connect(&conn, obd, cluuid);
7152         if (rc)
7153                 RETURN(rc);
7154
7155         lexp = class_conn2export(&conn);
7156         LASSERT(lexp != NULL);
7157
7158         rc = nodemap_add_member(client_nid, lexp);
7159         if (rc != 0 && rc != -EEXIST)
7160                 GOTO(out, rc);
7161
7162         rc = mdt_connect_internal(env, lexp, mdt, data, false);
7163         if (rc == 0) {
7164                 struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
7165
7166                 LASSERT(lcd);
7167                 memcpy(lcd->lcd_uuid, cluuid, sizeof(lcd->lcd_uuid));
7168                 rc = tgt_client_new(env, lexp);
7169                 if (rc == 0)
7170                         mdt_export_stats_init(obd, lexp, localdata);
7171         }
7172 out:
7173         if (rc != 0) {
7174                 class_disconnect(lexp);
7175                 nodemap_del_member(lexp);
7176                 *exp = NULL;
7177         } else {
7178                 *exp = lexp;
7179                 /* Because we do not want this export to be evicted by pinger,
7180                  * let's not add this export to the timed chain list.
7181                  */
7182                 if (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
7183                         spin_lock(&lexp->exp_obd->obd_dev_lock);
7184                         list_del_init(&lexp->exp_obd_chain_timed);
7185                         spin_unlock(&lexp->exp_obd->obd_dev_lock);
7186                 }
7187         }
7188
7189         RETURN(rc);
7190 }
7191
7192 static int mdt_obd_reconnect(const struct lu_env *env,
7193                              struct obd_export *exp, struct obd_device *obd,
7194                              struct obd_uuid *cluuid,
7195                              struct obd_connect_data *data,
7196                              void *localdata)
7197 {
7198         struct lnet_nid *client_nid = localdata;
7199         int rc;
7200
7201         ENTRY;
7202
7203         if (exp == NULL || obd == NULL || cluuid == NULL)
7204                 RETURN(-EINVAL);
7205
7206         rc = nodemap_add_member(client_nid, exp);
7207         if (rc != 0 && rc != -EEXIST)
7208                 RETURN(rc);
7209
7210         rc = mdt_connect_internal(env, exp, mdt_dev(obd->obd_lu_dev), data,
7211                                   true);
7212         if (rc == 0)
7213                 mdt_export_stats_init(obd, exp, localdata);
7214         else
7215                 nodemap_del_member(exp);
7216
7217         RETURN(rc);
7218 }
7219
7220 /* FIXME: Can we avoid using these two interfaces? */
7221 static int mdt_init_export(struct obd_export *exp)
7222 {
7223         struct mdt_export_data *med = &exp->exp_mdt_data;
7224         int                     rc;
7225
7226         ENTRY;
7227
7228         INIT_LIST_HEAD(&med->med_open_head);
7229         spin_lock_init(&med->med_open_lock);
7230         spin_lock(&exp->exp_lock);
7231         exp->exp_connecting = 1;
7232         spin_unlock(&exp->exp_lock);
7233
7234         OBD_ALLOC(exp->exp_used_slots,
7235                   BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7236         if (exp->exp_used_slots == NULL)
7237                 RETURN(-ENOMEM);
7238
7239         /* self-export doesn't need client data and ldlm initialization */
7240         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7241                                      &exp->exp_client_uuid)))
7242                 RETURN(0);
7243
7244         rc = tgt_client_alloc(exp);
7245         if (rc)
7246                 GOTO(err, rc);
7247
7248         rc = ldlm_init_export(exp);
7249         if (rc)
7250                 GOTO(err_free, rc);
7251
7252         RETURN(rc);
7253
7254 err_free:
7255         tgt_client_free(exp);
7256 err:
7257         OBD_FREE(exp->exp_used_slots,
7258                  BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7259         exp->exp_used_slots = NULL;
7260
7261         CERROR("%s: Failed to initialize export: rc = %d\n",
7262                exp->exp_obd->obd_name, rc);
7263         return rc;
7264 }
7265
7266 static int mdt_destroy_export(struct obd_export *exp)
7267 {
7268         ENTRY;
7269
7270         target_destroy_export(exp);
7271         if (exp->exp_used_slots)
7272                 OBD_FREE(exp->exp_used_slots,
7273                          BITS_TO_LONGS(OBD_MAX_RIF_MAX) * sizeof(long));
7274
7275         /* destroy can be called from failed obd_setup, so
7276          * checking uuid is safer than obd_self_export
7277          */
7278         if (unlikely(obd_uuid_equals(&exp->exp_obd->obd_uuid,
7279                                      &exp->exp_client_uuid)))
7280                 RETURN(0);
7281
7282         ldlm_destroy_export(exp);
7283         tgt_client_free(exp);
7284
7285         LASSERT(list_empty(&exp->exp_outstanding_replies));
7286         LASSERT(list_empty(&exp->exp_mdt_data.med_open_head));
7287
7288         /*
7289          * discard grants once we're sure no more
7290          * interaction with the client is possible
7291          */
7292         tgt_grant_discard(exp);
7293         if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
7294                 obd2obt(exp->exp_obd)->obt_lut->lut_tgd.tgd_tot_granted_clients--;
7295
7296         if (!(exp->exp_flags & OBD_OPT_FORCE))
7297                 tgt_grant_sanity_check(exp->exp_obd, __func__);
7298
7299         RETURN(0);
7300 }
7301
7302 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
7303                    struct linkea_data *ldata)
7304 {
7305         int rc;
7306
7307         LASSERT(ldata->ld_buf->lb_buf != NULL);
7308
7309         if (!mdt_object_exists(mdt_obj))
7310                 return -ENODATA;
7311
7312         rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7313                           ldata->ld_buf, XATTR_NAME_LINK);
7314         if (rc == -ERANGE) {
7315                 /* Buf was too small, figure out what we need. */
7316                 lu_buf_free(ldata->ld_buf);
7317                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7318                                   ldata->ld_buf, XATTR_NAME_LINK);
7319                 if (rc < 0)
7320                         return rc;
7321                 ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc);
7322                 if (ldata->ld_buf->lb_buf == NULL)
7323                         return -ENOMEM;
7324                 rc = mo_xattr_get(info->mti_env, mdt_object_child(mdt_obj),
7325                                   ldata->ld_buf, XATTR_NAME_LINK);
7326         }
7327         if (rc < 0)
7328                 return rc;
7329
7330         return linkea_init_with_rec(ldata);
7331 }
7332
7333 /**
7334  * Given an MDT object, try to look up the full path to the object.
7335  * Part of the MDT layer implementation of lfs fid2path.
7336  *
7337  * \param[in]     info  Per-thread common data shared by MDT level handlers.
7338  * \param[in]     obj   Object to do path lookup of
7339  * \param[in,out] fp    User-provided struct to store path information
7340  * \param[in]     root_fid Root FID of current path should reach
7341  *
7342  * \retval 0 Lookup successful, path information stored in fp
7343  * \retval -EAGAIN Lookup failed, usually because object is being moved
7344  * \retval negative errno if there was a problem
7345  */
7346 static int mdt_path_current(struct mdt_thread_info *info,
7347                             struct mdt_object *obj,
7348                             struct getinfo_fid2path *fp,
7349                             struct lu_fid *root_fid)
7350 {
7351         struct mdt_device *mdt = info->mti_mdt;
7352         struct lu_name *tmpname = &info->mti_name;
7353         struct lu_fid *tmpfid = &info->mti_tmp_fid1;
7354         struct lu_buf *buf = &info->mti_big_buf;
7355         struct linkea_data ldata = { NULL };
7356         bool first = true;
7357         struct mdt_object *mdt_obj;
7358         struct link_ea_header *leh;
7359         struct link_ea_entry *lee;
7360         bool worthchecking = true;
7361         bool needsfid = false;
7362         bool supported = false;
7363         int isenc = -1;
7364         char *ptr;
7365         int reclen;
7366         int rc = 0;
7367
7368         ENTRY;
7369
7370         /* temp buffer for path element, the buffer will be finally freed
7371          * in mdt_thread_info_fini
7372          */
7373         buf = lu_buf_check_and_alloc(buf, PATH_MAX);
7374         if (buf->lb_buf == NULL)
7375                 RETURN(-ENOMEM);
7376
7377         ldata.ld_buf = buf;
7378         ptr = fp->gf_u.gf_path + fp->gf_pathlen - 1;
7379         *ptr = 0;
7380         --ptr;
7381         *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
7382
7383         while (!lu_fid_eq(root_fid, &fp->gf_fid)) {
7384                 if (!lu_fid_eq(root_fid, &mdt->mdt_md_root_fid) &&
7385                     lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid))
7386                         GOTO(out, rc = -ENOENT);
7387
7388                 if (lu_fid_eq(mdt_object_fid(obj), tmpfid)) {
7389                         mdt_obj = obj;
7390                         mdt_object_get(info->mti_env, mdt_obj);
7391                 } else {
7392                         mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
7393                         if (IS_ERR(mdt_obj))
7394                                 GOTO(out, rc = PTR_ERR(mdt_obj));
7395                 }
7396
7397                 if (!mdt_object_exists(mdt_obj)) {
7398                         mdt_object_put(info->mti_env, mdt_obj);
7399                         GOTO(out, rc = -ENOENT);
7400                 }
7401
7402                 if (mdt_object_remote(mdt_obj)) {
7403                         mdt_object_put(info->mti_env, mdt_obj);
7404                         GOTO(remote_out, rc = -EREMOTE);
7405                 }
7406
7407                 if (worthchecking) {
7408                         /* need to know if FID being looked up is encrypted */
7409                         struct lu_attr la = { 0 };
7410                         struct dt_object *dt = mdt_obj2dt(mdt_obj);
7411
7412                         if (dt && dt->do_ops && dt->do_ops->do_attr_get)
7413                                 dt_attr_get(info->mti_env, dt, &la);
7414                         if (la.la_valid & LA_FLAGS &&
7415                             la.la_flags & LUSTRE_ENCRYPT_FL) {
7416                                 if (!supported && mdt_info_req(info) &&
7417                                     !exp_connect_encrypt_fid2path(
7418                                             mdt_info_req(info)->rq_export)) {
7419                                         /* client does not support fid2path
7420                                          * for encrypted files
7421                                          */
7422                                         mdt_object_put(info->mti_env, mdt_obj);
7423                                         GOTO(out, rc = -ENODATA);
7424                                 } else {
7425                                         supported = true;
7426                                 }
7427                                 needsfid = true;
7428                                 if (isenc == -1)
7429                                         isenc = 1;
7430                         } else {
7431                                 worthchecking = false;
7432                                 needsfid = false;
7433                                 if (isenc == -1)
7434                                         isenc = 0;
7435                         }
7436                 }
7437
7438                 rc = mdt_links_read(info, mdt_obj, &ldata);
7439                 if (rc != 0) {
7440                         mdt_object_put(info->mti_env, mdt_obj);
7441                         GOTO(out, rc);
7442                 }
7443
7444                 leh = buf->lb_buf;
7445                 lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
7446                 linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
7447                 /* If set, use link #linkno for path lookup, otherwise use
7448                  *  link #0.  Only do this for the final path ement.
7449                  */
7450                 if (first && fp->gf_linkno < leh->leh_reccount) {
7451                         int count;
7452
7453                         for (count = 0; count < fp->gf_linkno; count++) {
7454                                 lee = (struct link_ea_entry *)
7455                                      ((char *)lee + reclen);
7456                                 linkea_entry_unpack(lee, &reclen, tmpname,
7457                                                     tmpfid);
7458                         }
7459                         if (fp->gf_linkno < leh->leh_reccount - 1)
7460                                 /* indicate to user there are more links */
7461                                 fp->gf_linkno++;
7462                 }
7463
7464                 /* Check if it is slave stripes */
7465                 rc = mdt_is_dir_stripe(info, mdt_obj);
7466                 mdt_object_put(info->mti_env, mdt_obj);
7467                 if (rc < 0)
7468                         GOTO(out, rc);
7469                 if (rc == 1) {
7470                         fp->gf_fid = *tmpfid;
7471                         continue;
7472                 }
7473
7474                 /* Pack the name in the end of the buffer */
7475                 ptr -= tmpname->ln_namelen;
7476                 if (ptr - 1 <= fp->gf_u.gf_path)
7477                         GOTO(out, rc = -ENAMETOOLONG);
7478                 strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
7479                 if (needsfid) {
7480                         /* Pack FID before file name, so that client can build
7481                          * encoded/digested form.
7482                          */
7483                         char fidstr[FID_LEN + 1];
7484
7485                         snprintf(fidstr, sizeof(fidstr), DFID,
7486                                  PFID(&fp->gf_fid));
7487                         ptr -= strlen(fidstr);
7488                         if (ptr - 1 <= fp->gf_u.gf_path)
7489                                 GOTO(out, rc = -ENAMETOOLONG);
7490                         strncpy(ptr, fidstr, strlen(fidstr));
7491                 }
7492                 *(--ptr) = '/';
7493
7494                 /* keep the last resolved fid to the client, so the client will
7495                  * build the left path on another MDT for remote object
7496                  */
7497                 fp->gf_fid = *tmpfid;
7498
7499                 first = false;
7500         }
7501
7502         /* non-zero will be treated as an error */
7503         rc = 0;
7504
7505 remote_out:
7506         if (isenc != 1)
7507                 ptr++; /* skip leading / unless this is an encrypted file */
7508         memmove(fp->gf_u.gf_path, ptr,
7509                 fp->gf_u.gf_path + fp->gf_pathlen - ptr);
7510
7511 out:
7512         RETURN(rc);
7513 }
7514
7515 /**
7516  * Given an MDT object, use mdt_path_current to get the path.
7517  * Essentially a wrapper to retry mdt_path_current a set number of times
7518  * if -EAGAIN is returned (usually because an object is being moved).
7519  *
7520  * Part of the MDT layer implementation of lfs fid2path.
7521  *
7522  * \param[in]     info  Per-thread common data shared by mdt level handlers.
7523  * \param[in]     obj   Object to do path lookup of
7524  * \param[in,out] fp    User-provided struct for arguments and to store path
7525  *                      information
7526  *
7527  * \retval 0 Lookup successful, path information stored in fp
7528  * \retval negative errno if there was a problem
7529  */
7530 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
7531                     struct getinfo_fid2path *fp, struct lu_fid *root_fid)
7532 {
7533         struct mdt_device       *mdt = info->mti_mdt;
7534         int                     tries = 3;
7535         int                     rc = -EAGAIN;
7536
7537         ENTRY;
7538
7539         if (fp->gf_pathlen < 3)
7540                 RETURN(-EOVERFLOW);
7541
7542         if (root_fid == NULL)
7543                 root_fid = &mdt->mdt_md_root_fid;
7544
7545         if (lu_fid_eq(root_fid, mdt_object_fid(obj))) {
7546                 fp->gf_u.gf_path[0] = '\0';
7547                 RETURN(0);
7548         }
7549
7550         /* Retry multiple times in case file is being moved */
7551         while (tries-- && rc == -EAGAIN)
7552                 rc = mdt_path_current(info, obj, fp, root_fid);
7553
7554         RETURN(rc);
7555 }
7556
7557 /**
7558  * Get the full path of the provided FID, as of changelog record recno.
7559  *
7560  * This checks sanity and looks up object for user provided FID
7561  * before calling the actual path lookup code.
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,out] fp    User-provided struct for arguments and to store path
7567  *                      information
7568  *
7569  * \retval 0 Lookup successful, path information and recno stored in fp
7570  * \retval -ENOENT, object does not exist
7571  * \retval negative errno if there was a problem
7572  */
7573 static int mdt_fid2path(struct mdt_thread_info *info,
7574                         struct lu_fid *root_fid,
7575                         struct getinfo_fid2path *fp)
7576 {
7577         struct mdt_device *mdt = info->mti_mdt;
7578         struct mdt_object *obj;
7579         int    rc;
7580
7581         ENTRY;
7582
7583         CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
7584                 PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
7585
7586         if (!fid_is_sane(&fp->gf_fid))
7587                 RETURN(-EINVAL);
7588
7589         if (!fid_is_namespace_visible(&fp->gf_fid)) {
7590                 CDEBUG(D_INFO, "%s: "DFID" is invalid, f_seq should be >= %#llx, or f_oid != 0, or f_ver == 0\n",
7591                        mdt_obd_name(mdt),
7592                        PFID(&fp->gf_fid), (__u64)FID_SEQ_NORMAL);
7593                 RETURN(-EINVAL);
7594         }
7595
7596         /* return error if client-provided root fid is not the one stored in
7597          * the export
7598          */
7599         if (root_fid && !fid_is_zero(&info->mti_exp->exp_root_fid) &&
7600             !lu_fid_eq(root_fid, &info->mti_exp->exp_root_fid)) {
7601                 CDEBUG(D_INFO,
7602                        "%s: root fid from client "DFID" but "DFID" stored in export\n",
7603                        mdt_obd_name(mdt), PFID(root_fid),
7604                        PFID(&info->mti_exp->exp_root_fid));
7605                 RETURN(-EXDEV);
7606         }
7607
7608         obj = mdt_object_find(info->mti_env, mdt, &fp->gf_fid);
7609         if (IS_ERR(obj)) {
7610                 rc = PTR_ERR(obj);
7611                 CDEBUG(D_IOCTL, "cannot find "DFID": rc = %d\n",
7612                        PFID(&fp->gf_fid), rc);
7613                 RETURN(rc);
7614         }
7615
7616         if (mdt_object_remote(obj))
7617                 rc = -EREMOTE;
7618         else if (!mdt_object_exists(obj))
7619                 rc = -ENOENT;
7620         else
7621                 rc = 0;
7622
7623         if (rc < 0) {
7624                 mdt_object_put(info->mti_env, obj);
7625                 CDEBUG(D_IOCTL, "nonlocal object "DFID": rc = %d\n",
7626                        PFID(&fp->gf_fid), rc);
7627                 RETURN(rc);
7628         }
7629
7630         rc = mdt_path(info, obj, fp, root_fid);
7631
7632         CDEBUG(D_INFO, "fid "DFID", path %s recno %#llx linkno %u\n",
7633                PFID(&fp->gf_fid), fp->gf_u.gf_path,
7634                fp->gf_recno, fp->gf_linkno);
7635
7636         mdt_object_put(info->mti_env, obj);
7637
7638         RETURN(rc);
7639 }
7640
7641 static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, int keylen,
7642                             void *val, int vallen)
7643 {
7644         struct getinfo_fid2path *fpout, *fpin;
7645         struct lu_fid *root_fid = NULL;
7646         int rc = 0;
7647
7648         fpin = key + round_up(sizeof(KEY_FID2PATH), 8);
7649         fpout = val;
7650
7651         if (req_capsule_req_need_swab(info->mti_pill))
7652                 lustre_swab_fid2path(fpin);
7653
7654         memcpy(fpout, fpin, sizeof(*fpin));
7655         if (fpout->gf_pathlen != vallen - sizeof(*fpin))
7656                 RETURN(-EINVAL);
7657
7658         if (keylen >= round_up(sizeof(KEY_FID2PATH), 8) + sizeof(*fpin) +
7659                       sizeof(struct lu_fid)) {
7660                 /* client sent its root FID, which is normally fileset FID */
7661                 root_fid = fpin->gf_u.gf_root_fid;
7662                 if (req_capsule_req_need_swab(info->mti_pill))
7663                         lustre_swab_lu_fid(root_fid);
7664
7665                 if (root_fid != NULL && !fid_is_sane(root_fid))
7666                         RETURN(-EINVAL);
7667         }
7668
7669         rc = mdt_fid2path(info, root_fid, fpout);
7670         RETURN(rc);
7671 }
7672
7673 int mdt_get_info(struct tgt_session_info *tsi)
7674 {
7675         char    *key;
7676         int      keylen;
7677         __u32   *vallen;
7678         void    *valout;
7679         int      rc;
7680
7681         ENTRY;
7682
7683         key = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_KEY);
7684         if (key == NULL) {
7685                 CDEBUG(D_IOCTL, "No GETINFO key\n");
7686                 RETURN(err_serious(-EFAULT));
7687         }
7688         keylen = req_capsule_get_size(tsi->tsi_pill, &RMF_GETINFO_KEY,
7689                                       RCL_CLIENT);
7690
7691         vallen = req_capsule_client_get(tsi->tsi_pill, &RMF_GETINFO_VALLEN);
7692         if (vallen == NULL) {
7693                 CDEBUG(D_IOCTL, "%s: cannot get RMF_GETINFO_VALLEN buffer\n",
7694                                 tgt_name(tsi->tsi_tgt));
7695                 RETURN(err_serious(-EFAULT));
7696         }
7697
7698         req_capsule_set_size(tsi->tsi_pill, &RMF_GETINFO_VAL, RCL_SERVER,
7699                              *vallen);
7700         rc = req_capsule_server_pack(tsi->tsi_pill);
7701         if (rc)
7702                 RETURN(err_serious(rc));
7703
7704         valout = req_capsule_server_get(tsi->tsi_pill, &RMF_GETINFO_VAL);
7705         if (valout == NULL) {
7706                 CDEBUG(D_IOCTL, "%s: cannot get get-info RPC out buffer\n",
7707                                 tgt_name(tsi->tsi_tgt));
7708                 RETURN(err_serious(-EFAULT));
7709         }
7710
7711         if (KEY_IS(KEY_FID2PATH)) {
7712                 struct mdt_thread_info  *info = tsi2mdt_info(tsi);
7713
7714                 rc = mdt_rpc_fid2path(info, key, keylen, valout, *vallen);
7715                 mdt_thread_info_fini(info);
7716         } else {
7717                 rc = -EINVAL;
7718         }
7719         RETURN(rc);
7720 }
7721
7722 static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
7723 {
7724         struct obd_ioctl_data *data = karg;
7725         struct lu_fid *fid;
7726         __u64 version;
7727         struct mdt_object *obj;
7728         struct mdt_lock_handle  *lh;
7729         int rc;
7730
7731         ENTRY;
7732
7733         if (data->ioc_inlbuf1 == NULL || data->ioc_inllen1 != sizeof(*fid) ||
7734             data->ioc_inlbuf2 == NULL || data->ioc_inllen2 != sizeof(version))
7735                 RETURN(-EINVAL);
7736
7737         fid = (struct lu_fid *)data->ioc_inlbuf1;
7738
7739         if (!fid_is_sane(fid))
7740                 RETURN(-EINVAL);
7741
7742         CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
7743
7744         lh = &mti->mti_lh[MDT_LH_PARENT];
7745         obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE, LCK_CR);
7746         if (IS_ERR(obj))
7747                 RETURN(PTR_ERR(obj));
7748
7749         if (mdt_object_remote(obj)) {
7750                 rc = -EREMOTE;
7751                 /**
7752                  * before calling version get the correct MDS should be
7753                  * fid, this is error to find remote object here
7754                  */
7755                 CERROR("nonlocal object "DFID"\n", PFID(fid));
7756         } else if (!mdt_object_exists(obj)) {
7757                 *(__u64 *)data->ioc_inlbuf2 = ENOENT_VERSION;
7758                 rc = -ENOENT;
7759         } else {
7760                 version = dt_version_get(mti->mti_env, mdt_obj2dt(obj));
7761                *(__u64 *)data->ioc_inlbuf2 = version;
7762                 rc = 0;
7763         }
7764         mdt_object_unlock_put(mti, obj, lh, 1);
7765         RETURN(rc);
7766 }
7767
7768 /* ioctls on obd dev */
7769 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
7770                          void *karg, void __user *uarg)
7771 {
7772         struct obd_device *obd = exp->exp_obd;
7773         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
7774         struct dt_device *dt = mdt->mdt_bottom;
7775         struct obd_ioctl_data *data;
7776         struct lu_env env;
7777         int rc;
7778
7779         ENTRY;
7780         CDEBUG(D_IOCTL, "%s: cmd=%x len=%u karg=%pK uarg=%pK\n",
7781                obd->obd_name, cmd, len, karg, uarg);
7782
7783         rc = lu_env_init(&env, LCT_MD_THREAD);
7784         if (rc)
7785                 RETURN(rc);
7786
7787         /* handle commands that don't use @karg first */
7788         switch (cmd) {
7789         case OBD_IOC_SYNC:
7790                 rc = mdt_device_sync(&env, mdt);
7791                 GOTO(out, rc);
7792         case OBD_IOC_SET_READONLY:
7793                 rc = dt_sync(&env, dt);
7794                 if (rc == 0)
7795                         rc = dt_ro(&env, dt);
7796                 GOTO(out, rc);
7797         }
7798
7799         if (unlikely(karg == NULL)) {
7800                 OBD_IOC_ERROR(obd->obd_name, cmd, "karg=NULL", rc = -EINVAL);
7801                 GOTO(out, rc);
7802         }
7803         data = karg;
7804
7805         switch (cmd) {
7806         case OBD_IOC_ABORT_RECOVERY: {
7807                 if (data->ioc_type & OBD_FLG_ABORT_RECOV_MDT) {
7808                         LCONSOLE_WARN("%s: Aborting MDT recovery\n",
7809                                       obd->obd_name);
7810                         obd->obd_abort_mdt_recovery = 1;
7811                         wake_up(&obd->obd_next_transno_waitq);
7812                 } else { /* if (data->ioc_type & OBD_FLG_ABORT_RECOV_OST) */
7813                         /* lctl didn't set OBD_FLG_ABORT_RECOV_OST < 2.13.57 */
7814                         LCONSOLE_WARN("%s: Aborting client recovery\n",
7815                                       obd->obd_name);
7816                         obd->obd_abort_recovery = 1;
7817                         target_stop_recovery_thread(obd);
7818                 }
7819                 rc = 0;
7820                 break;
7821         }
7822         case OBD_IOC_CHANGELOG_REG:
7823         case OBD_IOC_CHANGELOG_DEREG:
7824         case OBD_IOC_CHANGELOG_CLEAR:
7825         case OBD_IOC_LLOG_PRINT:
7826         case OBD_IOC_LLOG_CANCEL:
7827                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(&env, mdt->mdt_child,
7828                                                            cmd, len, karg);
7829                 break;
7830         case OBD_IOC_START_LFSCK: {
7831                 struct md_device *next = mdt->mdt_child;
7832                 struct lfsck_start_param lsp;
7833
7834                 lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
7835                 lsp.lsp_index_valid = 0;
7836                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &lsp);
7837                 break;
7838         }
7839         case OBD_IOC_STOP_LFSCK: {
7840                 struct md_device *next = mdt->mdt_child;
7841                 struct lfsck_stop stop;
7842
7843                 stop.ls_status = LS_STOPPED;
7844                 /* Old lfsck utils may pass NULL @stop. */
7845                 if (data->ioc_inlbuf1 == NULL)
7846                         stop.ls_flags = 0;
7847                 else
7848                         stop.ls_flags =
7849                         ((struct lfsck_stop *)(data->ioc_inlbuf1))->ls_flags;
7850
7851                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0, &stop);
7852                 break;
7853         }
7854         case OBD_IOC_QUERY_LFSCK: {
7855                 struct md_device *next = mdt->mdt_child;
7856
7857                 rc = next->md_ops->mdo_iocontrol(&env, next, cmd, 0,
7858                                                  data->ioc_inlbuf1);
7859                 break;
7860         }
7861         case OBD_IOC_GET_OBJ_VERSION: {
7862                 struct mdt_thread_info *mti;
7863
7864                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7865                 memset(mti, 0, sizeof(*mti));
7866                 mti->mti_env = &env;
7867                 mti->mti_mdt = mdt;
7868                 mti->mti_exp = exp;
7869
7870                 rc = mdt_ioc_version_get(mti, karg);
7871                 break;
7872         }
7873         case OBD_IOC_CATLOGLIST: {
7874                 struct mdt_thread_info *mti;
7875
7876                 mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
7877                 lu_local_obj_fid(&mti->mti_tmp_fid1, LLOG_CATALOGS_OID);
7878                 rc = llog_catalog_list(&env, mdt->mdt_bottom, 0, karg,
7879                                        &mti->mti_tmp_fid1);
7880                 break;
7881         }
7882         default:
7883                 rc = OBD_IOC_ERROR(obd->obd_name, cmd, "unrecognized", -ENOTTY);
7884                 break;
7885         }
7886 out:
7887         lu_env_fini(&env);
7888         RETURN(rc);
7889 }
7890
7891 static int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
7892 {
7893         struct lu_device *ld = md2lu_dev(mdt->mdt_child);
7894         int rc;
7895
7896         ENTRY;
7897
7898         if (!mdt->mdt_skip_lfsck && !mdt->mdt_bottom->dd_rdonly) {
7899                 struct lfsck_start_param lsp;
7900
7901                 lsp.lsp_start = NULL;
7902                 lsp.lsp_index_valid = 0;
7903                 rc = mdt->mdt_child->md_ops->mdo_iocontrol(env, mdt->mdt_child,
7904                                                            OBD_IOC_START_LFSCK,
7905                                                            0, &lsp);
7906                 if (rc != 0 && rc != -EALREADY)
7907                         CWARN("%s: auto trigger paused LFSCK failed: rc = %d\n",
7908                               mdt_obd_name(mdt), rc);
7909         }
7910
7911         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
7912         RETURN(rc);
7913 }
7914
7915 static int mdt_obd_postrecov(struct obd_device *obd)
7916 {
7917         struct lu_env env;
7918         int rc;
7919
7920         rc = lu_env_init(&env, LCT_MD_THREAD);
7921         if (rc)
7922                 RETURN(rc);
7923         rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev));
7924         lu_env_fini(&env);
7925         return rc;
7926 }
7927
7928 static const struct obd_ops mdt_obd_device_ops = {
7929         .o_owner          = THIS_MODULE,
7930         .o_set_info_async = mdt_obd_set_info_async,
7931         .o_connect        = mdt_obd_connect,
7932         .o_reconnect      = mdt_obd_reconnect,
7933         .o_disconnect     = mdt_obd_disconnect,
7934         .o_init_export    = mdt_init_export,
7935         .o_destroy_export = mdt_destroy_export,
7936         .o_iocontrol      = mdt_iocontrol,
7937         .o_postrecov      = mdt_obd_postrecov,
7938         /* Data-on-MDT IO methods */
7939         .o_preprw         = mdt_obd_preprw,
7940         .o_commitrw       = mdt_obd_commitrw,
7941 };
7942
7943 static struct lu_device *mdt_device_fini(const struct lu_env *env,
7944                                          struct lu_device *d)
7945 {
7946         struct mdt_device *m = mdt_dev(d);
7947
7948         ENTRY;
7949
7950         mdt_fini(env, m);
7951         RETURN(NULL);
7952 }
7953
7954 static struct lu_device *mdt_device_free(const struct lu_env *env,
7955                                          struct lu_device *d)
7956 {
7957         struct mdt_device *m = mdt_dev(d);
7958
7959         ENTRY;
7960
7961         lu_device_fini(&m->mdt_lu_dev);
7962         OBD_FREE_PTR(m);
7963
7964         RETURN(NULL);
7965 }
7966
7967 static struct lu_device *mdt_device_alloc(const struct lu_env *env,
7968                                           struct lu_device_type *t,
7969                                           struct lustre_cfg *cfg)
7970 {
7971         struct lu_device  *l;
7972         struct mdt_device *m;
7973
7974         OBD_ALLOC_PTR(m);
7975         if (m != NULL) {
7976                 int rc;
7977
7978                 l = &m->mdt_lu_dev;
7979                 rc = mdt_init0(env, m, t, cfg);
7980                 if (rc != 0) {
7981                         mdt_device_free(env, l);
7982                         l = ERR_PTR(rc);
7983                         return l;
7984                 }
7985         } else
7986                 l = ERR_PTR(-ENOMEM);
7987         return l;
7988 }
7989
7990 /* context key constructor/destructor: mdt_key_init, mdt_key_fini */
7991 LU_KEY_INIT(mdt, struct mdt_thread_info);
7992
7993 static void mdt_key_fini(const struct lu_context *ctx,
7994                          struct lu_context_key *key, void *data)
7995 {
7996         struct mdt_thread_info *info = data;
7997
7998         if (info->mti_big_lmm) {
7999                 OBD_FREE_LARGE(info->mti_big_lmm, info->mti_big_lmmsize);
8000                 info->mti_big_lmm = NULL;
8001                 info->mti_big_lmmsize = 0;
8002         }
8003
8004         if (info->mti_big_acl) {
8005                 OBD_FREE_LARGE(info->mti_big_acl, info->mti_big_aclsize);
8006                 info->mti_big_acl = NULL;
8007                 info->mti_big_aclsize = 0;
8008         }
8009
8010         OBD_FREE_PTR(info);
8011 }
8012
8013 /* context key: mdt_thread_key */
8014 LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD);
8015
8016 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info)
8017 {
8018         return lu_ucred(info->mti_env);
8019 }
8020
8021 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info)
8022 {
8023         return lu_ucred_check(info->mti_env);
8024 }
8025
8026 /**
8027  * Enable/disable COS (Commit On Sharing).
8028  *
8029  * Set/Clear the COS flag in mdt options.
8030  *
8031  * \param mdt mdt device
8032  * \param val 0 disables COS, other values enable COS
8033  */
8034 void mdt_enable_cos(struct mdt_device *mdt, bool val)
8035 {
8036         struct lu_env env;
8037         int rc;
8038
8039         mdt->mdt_opts.mo_cos = val;
8040         rc = lu_env_init(&env, LCT_LOCAL);
8041         if (unlikely(rc != 0)) {
8042                 CWARN("%s: lu_env initialization failed, cannot sync: rc = %d\n",
8043                       mdt_obd_name(mdt), rc);
8044                 return;
8045         }
8046         mdt_device_sync(&env, mdt);
8047         lu_env_fini(&env);
8048 }
8049
8050 /**
8051  * Check COS (Commit On Sharing) status.
8052  *
8053  * Return COS flag status.
8054  *
8055  * \param mdt mdt device
8056  */
8057 int mdt_cos_is_enabled(struct mdt_device *mdt)
8058 {
8059         return mdt->mdt_opts.mo_cos != 0;
8060 }
8061
8062 static const struct lu_device_type_operations mdt_device_type_ops = {
8063         .ldto_device_alloc = mdt_device_alloc,
8064         .ldto_device_free  = mdt_device_free,
8065         .ldto_device_fini  = mdt_device_fini
8066 };
8067
8068 static struct lu_device_type mdt_device_type = {
8069         .ldt_tags     = LU_DEVICE_MD,
8070         .ldt_name     = LUSTRE_MDT_NAME,
8071         .ldt_ops      = &mdt_device_type_ops,
8072         .ldt_ctx_tags = LCT_MD_THREAD
8073 };
8074
8075 static int __init mdt_init(void)
8076 {
8077         int rc;
8078
8079         BUILD_BUG_ON(sizeof("0x0123456789ABCDEF:0x01234567:0x01234567") !=
8080                      FID_NOBRACE_LEN + 1);
8081         BUILD_BUG_ON(sizeof("[0x0123456789ABCDEF:0x01234567:0x01234567]") !=
8082                      FID_LEN + 1);
8083
8084         rc = libcfs_setup();
8085         if (rc)
8086                 return rc;
8087
8088         rc = lu_kmem_init(mdt_caches);
8089         if (rc)
8090                 return rc;
8091
8092         rc = mds_mod_init();
8093         if (rc)
8094                 GOTO(lu_fini, rc);
8095
8096         rc = class_register_type(&mdt_obd_device_ops, NULL, true,
8097                                  LUSTRE_MDT_NAME, &mdt_device_type);
8098         if (rc)
8099                 GOTO(mds_fini, rc);
8100 lu_fini:
8101         if (rc)
8102                 lu_kmem_fini(mdt_caches);
8103 mds_fini:
8104         if (rc)
8105                 mds_mod_exit();
8106         return rc;
8107 }
8108
8109 static void __exit mdt_exit(void)
8110 {
8111         class_unregister_type(LUSTRE_MDT_NAME);
8112         mds_mod_exit();
8113         lu_kmem_fini(mdt_caches);
8114 }
8115
8116 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8117 MODULE_DESCRIPTION("Lustre Metadata Target ("LUSTRE_MDT_NAME")");
8118 MODULE_VERSION(LUSTRE_VERSION_STRING);
8119 MODULE_LICENSE("GPL");
8120
8121 module_init(mdt_init);
8122 module_exit(mdt_exit);