lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #ifdef HAVE_UIDGID_HEADER
  50 # include <linux/uidgid.h>
  51 #endif
  52
  53 #include <lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  67 static DEFINE_SPINLOCK(ll_sb_lock);
  68
  69 #ifndef log2
  70 #define log2(n) ffz(~(n))
  71 #endif
  72
  73 static struct ll_sb_info *ll_init_sbi(void)
  74 {
  75         struct ll_sb_info *sbi = NULL;
  76         unsigned long pages;
  77         unsigned long lru_page_max;
  78         struct sysinfo si;
  79         class_uuid_t uuid;
  80         int i;
  81         ENTRY;
  82
  83         OBD_ALLOC_PTR(sbi);
  84         if (sbi == NULL)
  85                 RETURN(NULL);
  86
  87         spin_lock_init(&sbi->ll_lock);
  88         mutex_init(&sbi->ll_lco.lco_lock);
  89         spin_lock_init(&sbi->ll_pp_extent_lock);
  90         spin_lock_init(&sbi->ll_process_lock);
  91         sbi->ll_rw_stats_on = 0;
  92
  93         si_meminfo(&si);
  94         pages = si.totalram - si.totalhigh;
  95         lru_page_max = pages / 2;
  96
  97         /* initialize ll_cache data */
  98         sbi->ll_cache = cl_cache_init(lru_page_max);
  99         if (sbi->ll_cache == NULL) {
 100                 OBD_FREE(sbi, sizeof(*sbi));
 101                 RETURN(NULL);
 102         }
 103
 104         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 105                                            SBI_DEFAULT_READAHEAD_MAX);
 106         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 107         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 108                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 109         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 110         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 111
 112         ll_generate_random_uuid(uuid);
 113         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 114         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 115
 116         spin_lock(&ll_sb_lock);
 117         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 118         spin_unlock(&ll_sb_lock);
 119
 120         sbi->ll_flags |= LL_SBI_VERBOSE;
 121 #ifdef ENABLE_CHECKSUM
 122         sbi->ll_flags |= LL_SBI_CHECKSUM;
 123 #endif
 124
 125 #ifdef HAVE_LRU_RESIZE_SUPPORT
 126         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 127 #endif
 128
 129         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 130                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 131                                pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 133                                pp_w_hist.oh_lock);
 134         }
 135
 136         /* metadata statahead is enabled by default */
 137         sbi->ll_sa_max = LL_SA_RPC_DEF;
 138         atomic_set(&sbi->ll_sa_total, 0);
 139         atomic_set(&sbi->ll_sa_wrong, 0);
 140         atomic_set(&sbi->ll_sa_running, 0);
 141         atomic_set(&sbi->ll_agl_total, 0);
 142         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 spin_lock(&ll_sb_lock);
 160                 list_del(&sbi->ll_list);
 161                 spin_unlock(&ll_sb_lock);
 162                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 163                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 164                 if (sbi->ll_cache != NULL) {
 165                         cl_cache_decref(sbi->ll_cache);
 166                         sbi->ll_cache = NULL;
 167                 }
 168                 OBD_FREE(sbi, sizeof(*sbi));
 169         }
 170         EXIT;
 171 }
 172
 173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 174                                     struct vfsmount *mnt)
 175 {
 176         struct inode *root = NULL;
 177         struct ll_sb_info *sbi = ll_s2sbi(sb);
 178         struct obd_device *obd;
 179         struct obd_capa *oc = NULL;
 180         struct obd_statfs *osfs = NULL;
 181         struct ptlrpc_request *request = NULL;
 182         struct obd_connect_data *data = NULL;
 183         struct obd_uuid *uuid;
 184         struct md_op_data *op_data;
 185         struct lustre_md lmd;
 186         u64 valid;
 187         int size, err, checksum;
 188         ENTRY;
 189
 190         obd = class_name2obd(md);
 191         if (!obd) {
 192                 CERROR("MD %s: not setup or attached\n", md);
 193                 RETURN(-EINVAL);
 194         }
 195
 196         OBD_ALLOC_PTR(data);
 197         if (data == NULL)
 198                 RETURN(-ENOMEM);
 199
 200         OBD_ALLOC_PTR(osfs);
 201         if (osfs == NULL) {
 202                 OBD_FREE_PTR(data);
 203                 RETURN(-ENOMEM);
 204         }
 205
 206         /* indicate the features supported by this client */
 207         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 208                                   OBD_CONNECT_ATTRFID  |
 209                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 210                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 211                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 212                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 213                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 214                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 215                                   OBD_CONNECT_EINPROGRESS |
 216                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 217                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 218                                   OBD_CONNECT_MAX_EASIZE |
 219                                   OBD_CONNECT_FLOCK_DEAD |
 220                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 221                                   OBD_CONNECT_OPEN_BY_FID |
 222                                   OBD_CONNECT_DIR_STRIPE;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 230 #endif
 231
 232         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 233                 /* flag mdc connection as lightweight, only used for test
 234                  * purpose, use with care */
 235                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 236
 237         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 238         data->ocd_version = LUSTRE_VERSION_CODE;
 239
 240         if (sb->s_flags & MS_RDONLY)
 241                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 242         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 243                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 244
 245 #ifdef HAVE_MS_FLOCK_LOCK
 246         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 247         sb->s_flags |= MS_FLOCK_LOCK;
 248 #endif
 249 #ifdef MS_HAS_NEW_AOPS
 250         sb->s_flags |= MS_HAS_NEW_AOPS;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* real client */
 261         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 262         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 263                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337
 338         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 339             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 340                 LCONSOLE_INFO("Disabling user_xattr feature because "
 341                               "it is not supported on the server\n");
 342                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 346 #ifdef MS_POSIXACL
 347                 sb->s_flags |= MS_POSIXACL;
 348 #endif
 349                 sbi->ll_flags |= LL_SBI_ACL;
 350         } else {
 351                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 352 #ifdef MS_POSIXACL
 353                 sb->s_flags &= ~MS_POSIXACL;
 354 #endif
 355                 sbi->ll_flags &= ~LL_SBI_ACL;
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 359                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 360                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client is set as remote by default.\n");
 362                 }
 363         } else {
 364                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 365                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 366                         LCONSOLE_INFO("client claims to be remote, but server "
 367                                       "rejected, forced to be local.\n");
 368                 }
 369         }
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 372                 LCONSOLE_INFO("client enabled MDS capability!\n");
 373                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 377                 LCONSOLE_INFO("client enabled OSS capability!\n");
 378                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 379         }
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 382                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 385                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 386         else
 387                 sbi->ll_md_brw_pages = 1;
 388
 389         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391
 392         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 393                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 394                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 395                                       "unknown maximum xattr size.\n", dt);
 396                 } else {
 397                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 398                         sbi->ll_xattr_cache_enabled = 1;
 399                 }
 400         }
 401
 402         obd = class_name2obd(dt);
 403         if (!obd) {
 404                 CERROR("DT %s: not setup or attached\n", dt);
 405                 GOTO(out_md_fid, err = -ENODEV);
 406         }
 407
 408         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 409                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 410                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 411                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 412                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 413                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 414                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 415                                   OBD_CONNECT_MAXBYTES |
 416                                   OBD_CONNECT_EINPROGRESS |
 417                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 418                                   OBD_CONNECT_LAYOUTLOCK |
 419                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 420
 421         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 422                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 423                  * disabled by default, because it can still be enabled on the
 424                  * fly via /proc. As a consequence, we still need to come to an
 425                  * agreement on the supported algorithms at connect time */
 426                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 427
 428                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 429                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 430                 else
 431                         data->ocd_cksum_types = cksum_types_supported_client();
 432         }
 433
 434 #ifdef HAVE_LRU_RESIZE_SUPPORT
 435         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 436 #endif
 437         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 438                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 439
 440         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 441                "ocd_grant: %d\n", data->ocd_connect_flags,
 442                data->ocd_version, data->ocd_grant);
 443
 444         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 445         obd->obd_upcall.onu_upcall = cl_ocd_update;
 446
 447         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 448
 449         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 450                           NULL);
 451         if (err == -EBUSY) {
 452                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 453                                    "recovery, of which this client is not a "
 454                                    "part.  Please wait for recovery to "
 455                                    "complete, abort, or time out.\n", dt);
 456                 GOTO(out_md, err);
 457         } else if (err) {
 458                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 459                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 460                 GOTO(out_md, err);
 461         }
 462
 463         sbi->ll_dt_exp->exp_connect_data = *data;
 464
 465         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 466                            LUSTRE_SEQ_METADATA);
 467         if (err) {
 468                 CERROR("%s: Can't init data layer FID infrastructure, "
 469                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 470                 GOTO(out_dt, err);
 471         }
 472
 473         mutex_lock(&sbi->ll_lco.lco_lock);
 474         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 475         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 476         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 477         mutex_unlock(&sbi->ll_lco.lco_lock);
 478
 479         fid_zero(&sbi->ll_root_fid);
 480         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 481         if (err) {
 482                 CERROR("cannot mds_connect: rc = %d\n", err);
 483                 GOTO(out_lock_cn_cb, err);
 484         }
 485         if (!fid_is_sane(&sbi->ll_root_fid)) {
 486                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 487                        sbi->ll_md_exp->exp_obd->obd_name,
 488                        PFID(&sbi->ll_root_fid));
 489                 GOTO(out_lock_cn_cb, err = -EINVAL);
 490         }
 491         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 492
 493         sb->s_op = &lustre_super_operations;
 494 #if THREAD_SIZE >= 8192 /*b=17630*/
 495         sb->s_export_op = &lustre_export_operations;
 496 #endif
 497
 498         /* make root inode
 499          * XXX: move this to after cbd setup? */
 500         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 501                 OBD_MD_FLMODEASIZE;
 502         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 503                 valid |= OBD_MD_FLRMTPERM;
 504         else if (sbi->ll_flags & LL_SBI_ACL)
 505                 valid |= OBD_MD_FLACL;
 506
 507         OBD_ALLOC_PTR(op_data);
 508         if (op_data == NULL)
 509                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 510
 511         op_data->op_fid1 = sbi->ll_root_fid;
 512         op_data->op_mode = 0;
 513         op_data->op_capa1 = oc;
 514         op_data->op_valid = valid;
 515
 516         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 517         if (oc)
 518                 capa_put(oc);
 519         OBD_FREE_PTR(op_data);
 520         if (err) {
 521                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 522                        sbi->ll_md_exp->exp_obd->obd_name, err);
 523                 GOTO(out_lock_cn_cb, err);
 524         }
 525
 526         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 527                                sbi->ll_md_exp, &lmd);
 528         if (err) {
 529                 CERROR("failed to understand root inode md: rc = %d\n", err);
 530                 ptlrpc_req_finished(request);
 531                 GOTO(out_lock_cn_cb, err);
 532         }
 533
 534         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 535         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 536                                             sbi->ll_flags & LL_SBI_32BIT_API),
 537                        &lmd);
 538         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 539         ptlrpc_req_finished(request);
 540
 541         if (IS_ERR(root)) {
 542                 if (lmd.lsm)
 543                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 544 #ifdef CONFIG_FS_POSIX_ACL
 545                 if (lmd.posix_acl) {
 546                         posix_acl_release(lmd.posix_acl);
 547                         lmd.posix_acl = NULL;
 548                 }
 549 #endif
 550                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 551                 root = NULL;
 552                 CERROR("lustre_lite: bad iget4 for root\n");
 553                 GOTO(out_root, err);
 554         }
 555
 556 #ifdef CONFIG_FS_POSIX_ACL
 557         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 558                 rct_init(&sbi->ll_rct);
 559                 et_init(&sbi->ll_et);
 560         }
 561 #endif
 562
 563         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 564         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 565                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 566                                  NULL);
 567         cl_sb_init(sb);
 568
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 570                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 571                                  sbi->ll_cache, NULL);
 572
 573         sb->s_root = d_make_root(root);
 574         if (sb->s_root == NULL) {
 575                 CERROR("%s: can't make root dentry\n",
 576                         ll_get_fsname(sb, NULL, 0));
 577                 GOTO(out_root, err = -ENOMEM);
 578         }
 579 #ifdef HAVE_DCACHE_LOCK
 580         sb->s_root->d_op = &ll_d_ops;
 581 #endif
 582
 583         sbi->ll_sdev_orig = sb->s_dev;
 584
 585         /* We set sb->s_dev equal on all lustre clients in order to support
 586          * NFS export clustering.  NFSD requires that the FSID be the same
 587          * on all clients. */
 588         /* s_dev is also used in lt_compare() to compare two fs, but that is
 589          * only a node-local comparison. */
 590         uuid = obd_get_uuid(sbi->ll_md_exp);
 591         if (uuid != NULL)
 592                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 593
 594         if (data != NULL)
 595                 OBD_FREE_PTR(data);
 596         if (osfs != NULL)
 597                 OBD_FREE_PTR(osfs);
 598         if (proc_lustre_fs_root != NULL) {
 599                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 600                                                   dt, md);
 601                 if (err < 0) {
 602                         CERROR("%s: could not register mount in lprocfs: "
 603                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 604                         err = 0;
 605                 }
 606         }
 607
 608         RETURN(err);
 609 out_root:
 610         if (root)
 611                 iput(root);
 612 out_lock_cn_cb:
 613         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 614 out_dt:
 615         obd_disconnect(sbi->ll_dt_exp);
 616         sbi->ll_dt_exp = NULL;
 617 out_md_fid:
 618         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 619 out_md:
 620         obd_disconnect(sbi->ll_md_exp);
 621         sbi->ll_md_exp = NULL;
 622 out:
 623         if (data != NULL)
 624                 OBD_FREE_PTR(data);
 625         if (osfs != NULL)
 626                 OBD_FREE_PTR(osfs);
 627         return err;
 628 }
 629
 630 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 631 {
 632         int size, rc;
 633
 634         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 635         size = sizeof(int);
 636         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 637                           KEY_MAX_EASIZE, &size, lmmsize);
 638         if (rc)
 639                 CERROR("Get max mdsize error rc %d\n", rc);
 640
 641         RETURN(rc);
 642 }
 643
 644 /**
 645  * Get the value of the default_easize parameter.
 646  *
 647  * \see client_obd::cl_default_mds_easize
 648  *
 649  * \param[in] sbi       superblock info for this filesystem
 650  * \param[out] lmmsize  pointer to storage location for value
 651  *
 652  * \retval 0            on success
 653  * \retval negative     negated errno on failure
 654  */
 655 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 656 {
 657         int size, rc;
 658
 659         size = sizeof(int);
 660         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 661                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 662         if (rc)
 663                 CERROR("Get default mdsize error rc %d\n", rc);
 664
 665         RETURN(rc);
 666 }
 667
 668 /**
 669  * Set the default_easize parameter to the given value.
 670  *
 671  * \see client_obd::cl_default_mds_easize
 672  *
 673  * \param[in] sbi       superblock info for this filesystem
 674  * \param[in] lmmsize   the size to set
 675  *
 676  * \retval 0            on success
 677  * \retval negative     negated errno on failure
 678  */
 679 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 680 {
 681         int rc;
 682
 683         if (lmmsize < sizeof(struct lov_mds_md) ||
 684             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 685                 return -EINVAL;
 686
 687         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 688                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 689                                 sizeof(int), &lmmsize, NULL);
 690
 691         RETURN(rc);
 692 }
 693
 694 static void ll_dump_inode(struct inode *inode)
 695 {
 696         struct ll_d_hlist_node *tmp;
 697         int dentry_count = 0;
 698
 699         LASSERT(inode != NULL);
 700
 701         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 702                 dentry_count++;
 703
 704         CERROR("%s: inode %p dump: dev=%s fid="DFID
 705                " mode=%o count=%u, %d dentries\n",
 706                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 707                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 708                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 709 }
 710
 711 void lustre_dump_dentry(struct dentry *dentry, int recur)
 712 {
 713         struct list_head *tmp;
 714         int subdirs = 0;
 715
 716         LASSERT(dentry != NULL);
 717
 718         list_for_each(tmp, &dentry->d_subdirs)
 719                 subdirs++;
 720
 721         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 722                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 723                dentry->d_name.len, dentry->d_name.name,
 724                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 725                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 726                dentry->d_flags, dentry->d_fsdata, subdirs);
 727         if (dentry->d_inode != NULL)
 728                 ll_dump_inode(dentry->d_inode);
 729
 730         if (recur == 0)
 731                 return;
 732
 733         list_for_each(tmp, &dentry->d_subdirs) {
 734                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 735                 lustre_dump_dentry(d, recur - 1);
 736         }
 737 }
 738
 739 static void client_common_put_super(struct super_block *sb)
 740 {
 741         struct ll_sb_info *sbi = ll_s2sbi(sb);
 742         ENTRY;
 743
 744 #ifdef CONFIG_FS_POSIX_ACL
 745         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 746                 et_fini(&sbi->ll_et);
 747                 rct_fini(&sbi->ll_rct);
 748         }
 749 #endif
 750
 751         cl_sb_fini(sb);
 752
 753         list_del(&sbi->ll_conn_chain);
 754
 755         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 756         obd_disconnect(sbi->ll_dt_exp);
 757         sbi->ll_dt_exp = NULL;
 758
 759         lprocfs_unregister_mountpoint(sbi);
 760
 761         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 762         obd_disconnect(sbi->ll_md_exp);
 763         sbi->ll_md_exp = NULL;
 764
 765         EXIT;
 766 }
 767
 768 void ll_kill_super(struct super_block *sb)
 769 {
 770         struct ll_sb_info *sbi;
 771         ENTRY;
 772
 773         /* not init sb ?*/
 774         if (!(sb->s_flags & MS_ACTIVE))
 775                 return;
 776
 777         sbi = ll_s2sbi(sb);
 778         /* we need restore s_dev from changed for clustred NFS before put_super
 779          * because new kernels have cached s_dev and change sb->s_dev in
 780          * put_super not affected real removing devices */
 781         if (sbi) {
 782                 sb->s_dev = sbi->ll_sdev_orig;
 783                 sbi->ll_umounting = 1;
 784
 785                 /* wait running statahead threads to quit */
 786                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 787                         set_current_state(TASK_UNINTERRUPTIBLE);
 788                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 789                 }
 790         }
 791
 792         EXIT;
 793 }
 794
 795 static inline int ll_set_opt(const char *opt, char *data, int fl)
 796 {
 797         if (strncmp(opt, data, strlen(opt)) != 0)
 798                 return(0);
 799         else
 800                 return(fl);
 801 }
 802
 803 /* non-client-specific mount options are parsed in lmd_parse */
 804 static int ll_options(char *options, int *flags)
 805 {
 806         int tmp;
 807         char *s1 = options, *s2;
 808         ENTRY;
 809
 810         if (!options)
 811                 RETURN(0);
 812
 813         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 814
 815         while (*s1) {
 816                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 817                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 828                 if (tmp) {
 829                         *flags |= tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 833                 if (tmp) {
 834                         *flags &= ~tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862
 863                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 869                 if (tmp) {
 870                         *flags &= ~tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 874                 if (tmp) {
 875                         *flags |= tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags &= ~tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 884                 if (tmp) {
 885                         *flags |= tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags &= ~tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 894                 if (tmp) {
 895                         *flags |= tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 904                 if (tmp) {
 905                         *flags &= ~tmp;
 906                         goto next;
 907                 }
 908                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 909                                    s1);
 910                 RETURN(-EINVAL);
 911
 912 next:
 913                 /* Find next opt */
 914                 s2 = strchr(s1, ',');
 915                 if (s2 == NULL)
 916                         break;
 917                 s1 = s2 + 1;
 918         }
 919         RETURN(0);
 920 }
 921
 922 void ll_lli_init(struct ll_inode_info *lli)
 923 {
 924         lli->lli_inode_magic = LLI_INODE_MAGIC;
 925         lli->lli_flags = 0;
 926         spin_lock_init(&lli->lli_lock);
 927         lli->lli_posix_acl = NULL;
 928         lli->lli_remote_perms = NULL;
 929         mutex_init(&lli->lli_rmtperm_mutex);
 930         /* Do not set lli_fid, it has been initialized already. */
 931         fid_zero(&lli->lli_pfid);
 932         INIT_LIST_HEAD(&lli->lli_oss_capas);
 933         atomic_set(&lli->lli_open_count, 0);
 934         lli->lli_mds_capa = NULL;
 935         lli->lli_rmtperm_time = 0;
 936         lli->lli_mds_read_och = NULL;
 937         lli->lli_mds_write_och = NULL;
 938         lli->lli_mds_exec_och = NULL;
 939         lli->lli_open_fd_read_count = 0;
 940         lli->lli_open_fd_write_count = 0;
 941         lli->lli_open_fd_exec_count = 0;
 942         mutex_init(&lli->lli_och_mutex);
 943         spin_lock_init(&lli->lli_agl_lock);
 944         spin_lock_init(&lli->lli_layout_lock);
 945         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 946         lli->lli_clob = NULL;
 947
 948         init_rwsem(&lli->lli_xattrs_list_rwsem);
 949         mutex_init(&lli->lli_xattrs_enq_lock);
 950
 951         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 952         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 953                 mutex_init(&lli->lli_readdir_mutex);
 954                 lli->lli_opendir_key = NULL;
 955                 lli->lli_sai = NULL;
 956                 spin_lock_init(&lli->lli_sa_lock);
 957                 lli->lli_opendir_pid = 0;
 958                 lli->lli_sa_enabled = 0;
 959                 lli->lli_def_stripe_offset = -1;
 960         } else {
 961                 mutex_init(&lli->lli_size_mutex);
 962                 lli->lli_symlink_name = NULL;
 963                 init_rwsem(&lli->lli_trunc_sem);
 964                 range_lock_tree_init(&lli->lli_write_tree);
 965                 init_rwsem(&lli->lli_glimpse_sem);
 966                 lli->lli_glimpse_time = 0;
 967                 INIT_LIST_HEAD(&lli->lli_agl_list);
 968                 lli->lli_agl_index = 0;
 969                 lli->lli_async_rc = 0;
 970         }
 971         mutex_init(&lli->lli_layout_mutex);
 972 }
 973
 974 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 975 {
 976         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 977
 978         bdi->name = "lustre";
 979         return bdi_register(bdi, NULL, "lustre-%d",
 980                             atomic_inc_return(&ll_bdi_num));
 981 }
 982
 983 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 984 {
 985         struct lustre_profile *lprof = NULL;
 986         struct lustre_sb_info *lsi = s2lsi(sb);
 987         struct ll_sb_info *sbi;
 988         char  *dt = NULL, *md = NULL;
 989         char  *profilenm = get_profile_name(sb);
 990         struct config_llog_instance *cfg;
 991         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 992         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 993         int    err;
 994         ENTRY;
 995
 996         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 997
 998         OBD_ALLOC_PTR(cfg);
 999         if (cfg == NULL)
1000                 RETURN(-ENOMEM);
1001
1002         try_module_get(THIS_MODULE);
1003
1004         /* client additional sb info */
1005         lsi->lsi_llsbi = sbi = ll_init_sbi();
1006         if (!sbi) {
1007                 module_put(THIS_MODULE);
1008                 OBD_FREE_PTR(cfg);
1009                 RETURN(-ENOMEM);
1010         }
1011
1012         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1013         if (err)
1014                 GOTO(out_free, err);
1015
1016         err = bdi_init(&lsi->lsi_bdi);
1017         if (err)
1018                 GOTO(out_free, err);
1019         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1020         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1021         err = ll_bdi_register(&lsi->lsi_bdi);
1022         if (err)
1023                 GOTO(out_free, err);
1024
1025         sb->s_bdi = &lsi->lsi_bdi;
1026 #ifndef HAVE_DCACHE_LOCK
1027         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1028         sb->s_d_op = &ll_d_ops;
1029 #endif
1030
1031         /* Generate a string unique to this super, in case some joker tries
1032            to mount the same fs at two mount points.
1033            Use the address of the super itself.*/
1034         cfg->cfg_instance = sb;
1035         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1036         cfg->cfg_callback = class_config_llog_handler;
1037         /* set up client obds */
1038         err = lustre_process_log(sb, profilenm, cfg);
1039         if (err < 0)
1040                 GOTO(out_free, err);
1041
1042         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1043         lprof = class_get_profile(profilenm);
1044         if (lprof == NULL) {
1045                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1046                                    " read from the MGS.  Does that filesystem "
1047                                    "exist?\n", profilenm);
1048                 GOTO(out_free, err = -EINVAL);
1049         }
1050         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1051                lprof->lp_md, lprof->lp_dt);
1052
1053         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1054         if (!dt)
1055                 GOTO(out_free, err = -ENOMEM);
1056         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1057
1058         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1059         if (!md)
1060                 GOTO(out_free, err = -ENOMEM);
1061         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1062
1063         /* connections, registrations, sb setup */
1064         err = client_common_fill_super(sb, md, dt, mnt);
1065         if (err < 0)
1066                 GOTO(out_free, err);
1067
1068         sbi->ll_client_common_fill_super_succeeded = 1;
1069
1070 out_free:
1071         if (md)
1072                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1073         if (dt)
1074                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1075         if (err)
1076                 ll_put_super(sb);
1077         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1078                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1079
1080         OBD_FREE_PTR(cfg);
1081         RETURN(err);
1082 } /* ll_fill_super */
1083
1084 void ll_put_super(struct super_block *sb)
1085 {
1086         struct config_llog_instance cfg, params_cfg;
1087         struct obd_device *obd;
1088         struct lustre_sb_info *lsi = s2lsi(sb);
1089         struct ll_sb_info *sbi = ll_s2sbi(sb);
1090         char *profilenm = get_profile_name(sb);
1091         long ccc_count;
1092         int next, force = 1, rc = 0;
1093         ENTRY;
1094
1095         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1096
1097         ll_print_capa_stat(sbi);
1098
1099         cfg.cfg_instance = sb;
1100         lustre_end_log(sb, profilenm, &cfg);
1101
1102         params_cfg.cfg_instance = sb;
1103         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1104
1105         if (sbi->ll_md_exp) {
1106                 obd = class_exp2obd(sbi->ll_md_exp);
1107                 if (obd)
1108                         force = obd->obd_force;
1109         }
1110
1111         /* Wait for unstable pages to be committed to stable storage */
1112         if (force == 0) {
1113                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1114                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1115                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1116                         &lwi);
1117         }
1118
1119         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1120         if (force == 0 && rc != -EINTR)
1121                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1122
1123
1124         /* We need to set force before the lov_disconnect in
1125            lustre_common_put_super, since l_d cleans up osc's as well. */
1126         if (force) {
1127                 next = 0;
1128                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1129                                                      &next)) != NULL) {
1130                         obd->obd_force = force;
1131                 }
1132         }
1133
1134         if (sbi->ll_client_common_fill_super_succeeded) {
1135                 /* Only if client_common_fill_super succeeded */
1136                 client_common_put_super(sb);
1137         }
1138
1139         next = 0;
1140         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1141                 class_manual_cleanup(obd);
1142         }
1143
1144         if (sbi->ll_flags & LL_SBI_VERBOSE)
1145                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1146
1147         if (profilenm)
1148                 class_del_profile(profilenm);
1149
1150         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1151                 bdi_destroy(&lsi->lsi_bdi);
1152                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1153         }
1154
1155         ll_free_sbi(sb);
1156         lsi->lsi_llsbi = NULL;
1157
1158         lustre_common_put_super(sb);
1159
1160         cl_env_cache_purge(~0);
1161
1162         module_put(THIS_MODULE);
1163
1164         EXIT;
1165 } /* client_put_super */
1166
1167 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1168 {
1169         struct inode *inode = NULL;
1170
1171         /* NOTE: we depend on atomic igrab() -bzzz */
1172         lock_res_and_lock(lock);
1173         if (lock->l_resource->lr_lvb_inode) {
1174                 struct ll_inode_info * lli;
1175                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1176                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1177                         inode = igrab(lock->l_resource->lr_lvb_inode);
1178                 } else {
1179                         inode = lock->l_resource->lr_lvb_inode;
1180                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1181                                          D_WARNING, lock, "lr_lvb_inode %p is "
1182                                          "bogus: magic %08x",
1183                                          lock->l_resource->lr_lvb_inode,
1184                                          lli->lli_inode_magic);
1185                         inode = NULL;
1186                 }
1187         }
1188         unlock_res_and_lock(lock);
1189         return inode;
1190 }
1191
1192 static void ll_dir_clear_lsm_md(struct inode *inode)
1193 {
1194         struct ll_inode_info *lli = ll_i2info(inode);
1195
1196         LASSERT(S_ISDIR(inode->i_mode));
1197
1198         if (lli->lli_lsm_md != NULL) {
1199                 lmv_free_memmd(lli->lli_lsm_md);
1200                 lli->lli_lsm_md = NULL;
1201         }
1202 }
1203
1204 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1205                                       const struct lu_fid *fid,
1206                                       struct lustre_md *md)
1207 {
1208         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1209         struct mdt_body         *body = md->body;
1210         struct inode            *inode;
1211         ino_t                   ino;
1212         ENTRY;
1213
1214         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1215         inode = iget_locked(sb, ino);
1216         if (inode == NULL) {
1217                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1218                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1219                 RETURN(ERR_PTR(-ENOENT));
1220         }
1221
1222         if (inode->i_state & I_NEW) {
1223                 struct ll_inode_info *lli = ll_i2info(inode);
1224                 struct lmv_stripe_md *lsm = md->lmv;
1225
1226                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1227                                 (body->mbo_mode & S_IFMT);
1228                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1229                          PFID(fid));
1230
1231                 LTIME_S(inode->i_mtime) = 0;
1232                 LTIME_S(inode->i_atime) = 0;
1233                 LTIME_S(inode->i_ctime) = 0;
1234                 inode->i_rdev = 0;
1235
1236                 /* initializing backing dev info. */
1237                 inode->i_mapping->backing_dev_info =
1238                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1239                 inode->i_op = &ll_dir_inode_operations;
1240                 inode->i_fop = &ll_dir_operations;
1241                 lli->lli_fid = *fid;
1242                 ll_lli_init(lli);
1243
1244                 LASSERT(lsm != NULL);
1245                 /* master object FID */
1246                 lli->lli_pfid = body->mbo_fid1;
1247                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1248                        lli, PFID(fid), PFID(&lli->lli_pfid));
1249                 unlock_new_inode(inode);
1250         }
1251
1252         RETURN(inode);
1253 }
1254
1255 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1256 {
1257         struct lu_fid *fid;
1258         struct lmv_stripe_md *lsm = md->lmv;
1259         int i;
1260
1261         LASSERT(lsm != NULL);
1262         /* XXX sigh, this lsm_root initialization should be in
1263          * LMV layer, but it needs ll_iget right now, so we
1264          * put this here right now. */
1265         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1266                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1267                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1268                 /* Unfortunately ll_iget will call ll_update_inode,
1269                  * where the initialization of slave inode is slightly
1270                  * different, so it reset lsm_md to NULL to avoid
1271                  * initializing lsm for slave inode. */
1272                 /* For migrating inode, master stripe and master object will
1273                  * be same, so we only need assign this inode */
1274                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1275                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1276                 else
1277                         lsm->lsm_md_oinfo[i].lmo_root =
1278                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1279
1280                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1281                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1282
1283                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1284                         return rc;
1285                 }
1286         }
1287
1288         return 0;
1289 }
1290
1291 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1292                                 const struct lmv_stripe_md *lsm_md2)
1293 {
1294         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1295                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1296                lsm_md1->lsm_md_master_mdt_index ==
1297                                         lsm_md2->lsm_md_master_mdt_index &&
1298                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1299                lsm_md1->lsm_md_layout_version ==
1300                                         lsm_md2->lsm_md_layout_version &&
1301                strcmp(lsm_md1->lsm_md_pool_name,
1302                       lsm_md2->lsm_md_pool_name) == 0;
1303 }
1304
1305 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1306 {
1307         struct ll_inode_info *lli = ll_i2info(inode);
1308         struct lmv_stripe_md *lsm = md->lmv;
1309         int     rc;
1310         ENTRY;
1311
1312         LASSERT(S_ISDIR(inode->i_mode));
1313         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1314                PFID(ll_inode2fid(inode)));
1315
1316         /* no striped information from request. */
1317         if (lsm == NULL) {
1318                 if (lli->lli_lsm_md == NULL) {
1319                         RETURN(0);
1320                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1321                                                 LMV_HASH_FLAG_MIGRATION) {
1322                         /* migration is done, the temporay MIGRATE layout has
1323                          * been removed */
1324                         CDEBUG(D_INODE, DFID" finish migration.\n",
1325                                PFID(ll_inode2fid(inode)));
1326                         lmv_free_memmd(lli->lli_lsm_md);
1327                         lli->lli_lsm_md = NULL;
1328                         RETURN(0);
1329                 } else {
1330                         /* The lustre_md from req does not include stripeEA,
1331                          * see ll_md_setattr */
1332                         RETURN(0);
1333                 }
1334         }
1335
1336         /* set the directory layout */
1337         if (lli->lli_lsm_md == NULL) {
1338
1339                 rc = ll_init_lsm_md(inode, md);
1340                 if (rc != 0)
1341                         RETURN(rc);
1342
1343                 lli->lli_lsm_md = lsm;
1344                 /* set lsm_md to NULL, so the following free lustre_md
1345                  * will not free this lsm */
1346                 md->lmv = NULL;
1347                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1348                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1349                 RETURN(0);
1350         }
1351
1352         /* Compare the old and new stripe information */
1353         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1354                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1355                 int                     idx;
1356
1357                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1358                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1359                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1360                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1361                        inode, lsm, old_lsm,
1362                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1363                        lsm->lsm_md_stripe_count,
1364                        old_lsm->lsm_md_stripe_count,
1365                        lsm->lsm_md_master_mdt_index,
1366                        old_lsm->lsm_md_master_mdt_index,
1367                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1368                        lsm->lsm_md_layout_version,
1369                        old_lsm->lsm_md_layout_version,
1370                        lsm->lsm_md_pool_name,
1371                        old_lsm->lsm_md_pool_name);
1372
1373                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1374                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1375                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1376                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1377                 }
1378
1379                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1380                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1381                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1382                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1383                 }
1384
1385                 RETURN(-EIO);
1386         }
1387
1388         RETURN(0);
1389 }
1390
1391 void ll_clear_inode(struct inode *inode)
1392 {
1393         struct ll_inode_info *lli = ll_i2info(inode);
1394         struct ll_sb_info *sbi = ll_i2sbi(inode);
1395         ENTRY;
1396
1397         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1398                PFID(ll_inode2fid(inode)), inode);
1399
1400         if (S_ISDIR(inode->i_mode)) {
1401                 /* these should have been cleared in ll_file_release */
1402                 LASSERT(lli->lli_opendir_key == NULL);
1403                 LASSERT(lli->lli_sai == NULL);
1404                 LASSERT(lli->lli_opendir_pid == 0);
1405         }
1406
1407         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1408
1409         LASSERT(!lli->lli_open_fd_write_count);
1410         LASSERT(!lli->lli_open_fd_read_count);
1411         LASSERT(!lli->lli_open_fd_exec_count);
1412
1413         if (lli->lli_mds_write_och)
1414                 ll_md_real_close(inode, FMODE_WRITE);
1415         if (lli->lli_mds_exec_och)
1416                 ll_md_real_close(inode, FMODE_EXEC);
1417         if (lli->lli_mds_read_och)
1418                 ll_md_real_close(inode, FMODE_READ);
1419
1420         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1421                 OBD_FREE(lli->lli_symlink_name,
1422                          strlen(lli->lli_symlink_name) + 1);
1423                 lli->lli_symlink_name = NULL;
1424         }
1425
1426         ll_xattr_cache_destroy(inode);
1427
1428         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1429                 LASSERT(lli->lli_posix_acl == NULL);
1430                 if (lli->lli_remote_perms) {
1431                         free_rmtperm_hash(lli->lli_remote_perms);
1432                         lli->lli_remote_perms = NULL;
1433                 }
1434         }
1435 #ifdef CONFIG_FS_POSIX_ACL
1436         else if (lli->lli_posix_acl) {
1437                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1438                 LASSERT(lli->lli_remote_perms == NULL);
1439                 posix_acl_release(lli->lli_posix_acl);
1440                 lli->lli_posix_acl = NULL;
1441         }
1442 #endif
1443         lli->lli_inode_magic = LLI_INODE_DEAD;
1444
1445         ll_clear_inode_capas(inode);
1446         if (S_ISDIR(inode->i_mode))
1447                 ll_dir_clear_lsm_md(inode);
1448         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1449                 LASSERT(list_empty(&lli->lli_agl_list));
1450
1451         /*
1452          * XXX This has to be done before lsm is freed below, because
1453          * cl_object still uses inode lsm.
1454          */
1455         cl_inode_fini(inode);
1456
1457         EXIT;
1458 }
1459
1460 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1461 {
1462         struct lustre_md md;
1463         struct inode *inode = dentry->d_inode;
1464         struct ll_sb_info *sbi = ll_i2sbi(inode);
1465         struct ptlrpc_request *request = NULL;
1466         int rc, ia_valid;
1467         ENTRY;
1468
1469         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1470                                      LUSTRE_OPC_ANY, NULL);
1471         if (IS_ERR(op_data))
1472                 RETURN(PTR_ERR(op_data));
1473
1474         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1475         if (rc) {
1476                 ptlrpc_req_finished(request);
1477                 if (rc == -ENOENT) {
1478                         clear_nlink(inode);
1479                         /* Unlinked special device node? Or just a race?
1480                          * Pretend we done everything. */
1481                         if (!S_ISREG(inode->i_mode) &&
1482                             !S_ISDIR(inode->i_mode)) {
1483                                 ia_valid = op_data->op_attr.ia_valid;
1484                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1485                                 rc = simple_setattr(dentry, &op_data->op_attr);
1486                                 op_data->op_attr.ia_valid = ia_valid;
1487                         }
1488                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1489                         CERROR("md_setattr fails: rc = %d\n", rc);
1490                 }
1491                 RETURN(rc);
1492         }
1493
1494         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1495                               sbi->ll_md_exp, &md);
1496         if (rc) {
1497                 ptlrpc_req_finished(request);
1498                 RETURN(rc);
1499         }
1500
1501         ia_valid = op_data->op_attr.ia_valid;
1502         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1503          * cache is not cleared yet. */
1504         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1505         rc = simple_setattr(dentry, &op_data->op_attr);
1506         op_data->op_attr.ia_valid = ia_valid;
1507
1508         rc = ll_update_inode(inode, &md);
1509         ptlrpc_req_finished(request);
1510
1511         RETURN(rc);
1512 }
1513
1514 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1515 {
1516         struct obd_capa *capa;
1517         int rc;
1518
1519         if (attr->ia_valid & ATTR_SIZE)
1520                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1521         else
1522                 capa = ll_mdscapa_get(inode);
1523
1524         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1525
1526         if (attr->ia_valid & ATTR_SIZE)
1527                 ll_truncate_free_capa(capa);
1528         else
1529                 capa_put(capa);
1530
1531         return rc;
1532 }
1533
1534 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1535  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1536  * keep these values until such a time that objects are allocated for it.
1537  * We do the MDS operations first, as it is checking permissions for us.
1538  * We don't to the MDS RPC if there is nothing that we want to store there,
1539  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1540  * going to do an RPC anyways.
1541  *
1542  * If we are doing a truncate, we will send the mtime and ctime updates
1543  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1544  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1545  * at the same time.
1546  *
1547  * In case of HSMimport, we only set attr on MDS.
1548  */
1549 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1550 {
1551         struct inode *inode = dentry->d_inode;
1552         struct ll_inode_info *lli = ll_i2info(inode);
1553         struct md_op_data *op_data = NULL;
1554         bool file_is_released = false;
1555         int rc = 0;
1556         ENTRY;
1557
1558         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1559                "valid %x, hsm_import %d\n",
1560                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1561                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1562                hsm_import);
1563
1564         if (attr->ia_valid & ATTR_SIZE) {
1565                 /* Check new size against VFS/VM file size limit and rlimit */
1566                 rc = inode_newsize_ok(inode, attr->ia_size);
1567                 if (rc)
1568                         RETURN(rc);
1569
1570                 /* The maximum Lustre file size is variable, based on the
1571                  * OST maximum object size and number of stripes.  This
1572                  * needs another check in addition to the VFS check above. */
1573                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1574                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1575                                PFID(&lli->lli_fid), attr->ia_size,
1576                                ll_file_maxbytes(inode));
1577                         RETURN(-EFBIG);
1578                 }
1579
1580                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1581         }
1582
1583         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1584         if (attr->ia_valid & TIMES_SET_FLAGS) {
1585                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1586                     !cfs_capable(CFS_CAP_FOWNER))
1587                         RETURN(-EPERM);
1588         }
1589
1590         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1591         if (attr->ia_valid & ATTR_CTIME) {
1592                 attr->ia_ctime = CFS_CURRENT_TIME;
1593                 attr->ia_valid |= ATTR_CTIME_SET;
1594         }
1595         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1596             (attr->ia_valid & ATTR_ATIME)) {
1597                 attr->ia_atime = CFS_CURRENT_TIME;
1598                 attr->ia_valid |= ATTR_ATIME_SET;
1599         }
1600         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1601             (attr->ia_valid & ATTR_MTIME)) {
1602                 attr->ia_mtime = CFS_CURRENT_TIME;
1603                 attr->ia_valid |= ATTR_MTIME_SET;
1604         }
1605
1606         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1607                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1608                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1609                        cfs_time_current_sec());
1610
1611         /* We always do an MDS RPC, even if we're only changing the size;
1612          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1613
1614         OBD_ALLOC_PTR(op_data);
1615         if (op_data == NULL)
1616                 RETURN(-ENOMEM);
1617
1618         if (!S_ISDIR(inode->i_mode)) {
1619                 if (attr->ia_valid & ATTR_SIZE)
1620                         inode_dio_write_done(inode);
1621                 mutex_unlock(&inode->i_mutex);
1622         }
1623
1624         /* truncate on a released file must failed with -ENODATA,
1625          * so size must not be set on MDS for released file
1626          * but other attributes must be set
1627          */
1628         if (S_ISREG(inode->i_mode)) {
1629                 struct cl_layout cl = {
1630                         .cl_is_released = false,
1631                 };
1632                 struct lu_env *env;
1633                 int refcheck;
1634                 __u32 gen;
1635
1636                 rc = ll_layout_refresh(inode, &gen);
1637                 if (rc < 0)
1638                         GOTO(out, rc);
1639
1640                 /* XXX: the only place we need to know the layout type,
1641                  * this will be removed by a later patch. -Jinshan */
1642                 env = cl_env_get(&refcheck);
1643                 if (IS_ERR(env))
1644                         GOTO(out, rc = PTR_ERR(env));
1645
1646                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1647                 cl_env_put(env, &refcheck);
1648                 if (rc < 0)
1649                         GOTO(out, rc);
1650
1651                 file_is_released = cl.cl_is_released;
1652
1653                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1654                         if (file_is_released) {
1655                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1656                                 if (rc < 0)
1657                                         GOTO(out, rc);
1658
1659                                 file_is_released = false;
1660                                 ll_layout_refresh(inode, &gen);
1661                         }
1662
1663                         /* If we are changing file size, file content is
1664                          * modified, flag it. */
1665                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1666                         spin_lock(&lli->lli_lock);
1667                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1668                         spin_unlock(&lli->lli_lock);
1669                         op_data->op_bias |= MDS_DATA_MODIFIED;
1670                 }
1671         }
1672
1673         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1674
1675         rc = ll_md_setattr(dentry, op_data);
1676         if (rc)
1677                 GOTO(out, rc);
1678
1679         /* RPC to MDT is sent, cancel data modification flag */
1680         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1681                 spin_lock(&lli->lli_lock);
1682                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1683                 spin_unlock(&lli->lli_lock);
1684         }
1685
1686         if (!S_ISREG(inode->i_mode) || file_is_released)
1687                 GOTO(out, rc = 0);
1688
1689         if (attr->ia_valid & (ATTR_SIZE |
1690                               ATTR_ATIME | ATTR_ATIME_SET |
1691                               ATTR_MTIME | ATTR_MTIME_SET)) {
1692                 /* For truncate and utimes sending attributes to OSTs, setting
1693                  * mtime/atime to the past will be performed under PW [0:EOF]
1694                  * extent lock (new_size:EOF for truncate).  It may seem
1695                  * excessive to send mtime/atime updates to OSTs when not
1696                  * setting times to past, but it is necessary due to possible
1697                  * time de-synchronization between MDT inode and OST objects */
1698                 if (attr->ia_valid & ATTR_SIZE)
1699                         down_write(&lli->lli_trunc_sem);
1700                 rc = ll_setattr_ost(inode, attr);
1701                 if (attr->ia_valid & ATTR_SIZE)
1702                         up_write(&lli->lli_trunc_sem);
1703         }
1704         EXIT;
1705 out:
1706         if (op_data != NULL)
1707                 ll_finish_md_op_data(op_data);
1708
1709         if (!S_ISDIR(inode->i_mode)) {
1710                 mutex_lock(&inode->i_mutex);
1711                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1712                         inode_dio_wait(inode);
1713         }
1714
1715         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1716                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1717
1718         return rc;
1719 }
1720
1721 int ll_setattr(struct dentry *de, struct iattr *attr)
1722 {
1723         int mode = de->d_inode->i_mode;
1724
1725         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1726                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1727                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1728
1729         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1730                                (ATTR_SIZE|ATTR_MODE)) &&
1731             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1732              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1733               !(attr->ia_mode & S_ISGID))))
1734                 attr->ia_valid |= ATTR_FORCE;
1735
1736         if ((attr->ia_valid & ATTR_MODE) &&
1737             (mode & S_ISUID) &&
1738             !(attr->ia_mode & S_ISUID) &&
1739             !(attr->ia_valid & ATTR_KILL_SUID))
1740                 attr->ia_valid |= ATTR_KILL_SUID;
1741
1742         if ((attr->ia_valid & ATTR_MODE) &&
1743             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1744             !(attr->ia_mode & S_ISGID) &&
1745             !(attr->ia_valid & ATTR_KILL_SGID))
1746                 attr->ia_valid |= ATTR_KILL_SGID;
1747
1748         return ll_setattr_raw(de, attr, false);
1749 }
1750
1751 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1752                        __u64 max_age, __u32 flags)
1753 {
1754         struct ll_sb_info *sbi = ll_s2sbi(sb);
1755         struct obd_statfs obd_osfs;
1756         int rc;
1757         ENTRY;
1758
1759         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1760         if (rc) {
1761                 CERROR("md_statfs fails: rc = %d\n", rc);
1762                 RETURN(rc);
1763         }
1764
1765         osfs->os_type = sb->s_magic;
1766
1767         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1768                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1769
1770         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1771                 flags |= OBD_STATFS_NODELAY;
1772
1773         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1774         if (rc) {
1775                 CERROR("obd_statfs fails: rc = %d\n", rc);
1776                 RETURN(rc);
1777         }
1778
1779         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1780                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1781                obd_osfs.os_files);
1782
1783         osfs->os_bsize = obd_osfs.os_bsize;
1784         osfs->os_blocks = obd_osfs.os_blocks;
1785         osfs->os_bfree = obd_osfs.os_bfree;
1786         osfs->os_bavail = obd_osfs.os_bavail;
1787
1788         /* If we don't have as many objects free on the OST as inodes
1789          * on the MDS, we reduce the total number of inodes to
1790          * compensate, so that the "inodes in use" number is correct.
1791          */
1792         if (obd_osfs.os_ffree < osfs->os_ffree) {
1793                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1794                         obd_osfs.os_ffree;
1795                 osfs->os_ffree = obd_osfs.os_ffree;
1796         }
1797
1798         RETURN(rc);
1799 }
1800 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1801 {
1802         struct super_block *sb = de->d_sb;
1803         struct obd_statfs osfs;
1804         __u64 fsid = huge_encode_dev(sb->s_dev);
1805         int rc;
1806
1807         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1808         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1809
1810         /* Some amount of caching on the client is allowed */
1811         rc = ll_statfs_internal(sb, &osfs,
1812                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1813                                 0);
1814         if (rc)
1815                 return rc;
1816
1817         statfs_unpack(sfs, &osfs);
1818
1819         /* We need to downshift for all 32-bit kernels, because we can't
1820          * tell if the kernel is being called via sys_statfs64() or not.
1821          * Stop before overflowing f_bsize - in which case it is better
1822          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1823         if (sizeof(long) < 8) {
1824                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1825                         sfs->f_bsize <<= 1;
1826
1827                         osfs.os_blocks >>= 1;
1828                         osfs.os_bfree >>= 1;
1829                         osfs.os_bavail >>= 1;
1830                 }
1831         }
1832
1833         sfs->f_blocks = osfs.os_blocks;
1834         sfs->f_bfree = osfs.os_bfree;
1835         sfs->f_bavail = osfs.os_bavail;
1836         sfs->f_fsid.val[0] = (__u32)fsid;
1837         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1838         return 0;
1839 }
1840
1841 void ll_inode_size_lock(struct inode *inode)
1842 {
1843         struct ll_inode_info *lli;
1844
1845         LASSERT(!S_ISDIR(inode->i_mode));
1846
1847         lli = ll_i2info(inode);
1848         mutex_lock(&lli->lli_size_mutex);
1849 }
1850
1851 void ll_inode_size_unlock(struct inode *inode)
1852 {
1853         struct ll_inode_info *lli;
1854
1855         lli = ll_i2info(inode);
1856         mutex_unlock(&lli->lli_size_mutex);
1857 }
1858
1859 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1860 {
1861         struct ll_inode_info *lli = ll_i2info(inode);
1862         struct mdt_body *body = md->body;
1863         struct lov_stripe_md *lsm = md->lsm;
1864         struct ll_sb_info *sbi = ll_i2sbi(inode);
1865
1866         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1867         if (lsm != NULL)
1868                 cl_file_inode_init(inode, md);
1869
1870         if (S_ISDIR(inode->i_mode)) {
1871                 int     rc;
1872
1873                 rc = ll_update_lsm_md(inode, md);
1874                 if (rc != 0)
1875                         return rc;
1876         }
1877
1878         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1879                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1880                         ll_update_remote_perm(inode, md->remote_perm);
1881         }
1882 #ifdef CONFIG_FS_POSIX_ACL
1883         else if (body->mbo_valid & OBD_MD_FLACL) {
1884                 spin_lock(&lli->lli_lock);
1885                 if (lli->lli_posix_acl)
1886                         posix_acl_release(lli->lli_posix_acl);
1887                 lli->lli_posix_acl = md->posix_acl;
1888                 spin_unlock(&lli->lli_lock);
1889         }
1890 #endif
1891         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1892                                         sbi->ll_flags & LL_SBI_32BIT_API);
1893         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1894
1895         if (body->mbo_valid & OBD_MD_FLATIME) {
1896                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1897                         LTIME_S(inode->i_atime) = body->mbo_atime;
1898                 lli->lli_atime = body->mbo_atime;
1899         }
1900
1901         if (body->mbo_valid & OBD_MD_FLMTIME) {
1902                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1903                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1904                                "to "LPU64"\n", inode->i_ino,
1905                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1906                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1907                 }
1908                 lli->lli_mtime = body->mbo_mtime;
1909         }
1910
1911         if (body->mbo_valid & OBD_MD_FLCTIME) {
1912                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1913                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1914                 lli->lli_ctime = body->mbo_ctime;
1915         }
1916
1917         if (body->mbo_valid & OBD_MD_FLMODE)
1918                 inode->i_mode = (inode->i_mode & S_IFMT) |
1919                                 (body->mbo_mode & ~S_IFMT);
1920
1921         if (body->mbo_valid & OBD_MD_FLTYPE)
1922                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1923                                 (body->mbo_mode & S_IFMT);
1924
1925         LASSERT(inode->i_mode != 0);
1926         if (S_ISREG(inode->i_mode))
1927                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1928                                        LL_MAX_BLKSIZE_BITS);
1929         else
1930                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1931
1932         if (body->mbo_valid & OBD_MD_FLUID)
1933                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1934         if (body->mbo_valid & OBD_MD_FLGID)
1935                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1936         if (body->mbo_valid & OBD_MD_FLFLAGS)
1937                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1938         if (body->mbo_valid & OBD_MD_FLNLINK)
1939                 set_nlink(inode, body->mbo_nlink);
1940         if (body->mbo_valid & OBD_MD_FLRDEV)
1941                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1942
1943         if (body->mbo_valid & OBD_MD_FLID) {
1944                 /* FID shouldn't be changed! */
1945                 if (fid_is_sane(&lli->lli_fid)) {
1946                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1947                                  "Trying to change FID "DFID
1948                                  " to the "DFID", inode "DFID"(%p)\n",
1949                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1950                                  PFID(ll_inode2fid(inode)), inode);
1951                 } else {
1952                         lli->lli_fid = body->mbo_fid1;
1953                 }
1954         }
1955
1956         LASSERT(fid_seq(&lli->lli_fid) != 0);
1957
1958         if (body->mbo_valid & OBD_MD_FLSIZE) {
1959                 i_size_write(inode, body->mbo_size);
1960
1961                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1962                        PFID(ll_inode2fid(inode)),
1963                        (unsigned long long)body->mbo_size);
1964
1965                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1966                         inode->i_blocks = body->mbo_blocks;
1967         }
1968
1969         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1970                 LASSERT(md->mds_capa);
1971                 ll_add_capa(inode, md->mds_capa);
1972         }
1973
1974         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1975                 LASSERT(md->oss_capa);
1976                 ll_add_capa(inode, md->oss_capa);
1977         }
1978
1979         if (body->mbo_valid & OBD_MD_TSTATE) {
1980                 if (body->mbo_t_state & MS_RESTORE)
1981                         lli->lli_flags |= LLIF_FILE_RESTORING;
1982         }
1983
1984         return 0;
1985 }
1986
1987 int ll_read_inode2(struct inode *inode, void *opaque)
1988 {
1989         struct lustre_md *md = opaque;
1990         struct ll_inode_info *lli = ll_i2info(inode);
1991         int     rc;
1992         ENTRY;
1993
1994         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1995                PFID(&lli->lli_fid), inode);
1996
1997         /* Core attributes from the MDS first.  This is a new inode, and
1998          * the VFS doesn't zero times in the core inode so we have to do
1999          * it ourselves.  They will be overwritten by either MDS or OST
2000          * attributes - we just need to make sure they aren't newer. */
2001         LTIME_S(inode->i_mtime) = 0;
2002         LTIME_S(inode->i_atime) = 0;
2003         LTIME_S(inode->i_ctime) = 0;
2004         inode->i_rdev = 0;
2005         rc = ll_update_inode(inode, md);
2006         if (rc != 0)
2007                 RETURN(rc);
2008
2009         /* OIDEBUG(inode); */
2010
2011         /* initializing backing dev info. */
2012         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2013
2014
2015         if (S_ISREG(inode->i_mode)) {
2016                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2017                 inode->i_op = &ll_file_inode_operations;
2018                 inode->i_fop = sbi->ll_fop;
2019                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2020                 EXIT;
2021         } else if (S_ISDIR(inode->i_mode)) {
2022                 inode->i_op = &ll_dir_inode_operations;
2023                 inode->i_fop = &ll_dir_operations;
2024                 EXIT;
2025         } else if (S_ISLNK(inode->i_mode)) {
2026                 inode->i_op = &ll_fast_symlink_inode_operations;
2027                 EXIT;
2028         } else {
2029                 inode->i_op = &ll_special_inode_operations;
2030
2031                 init_special_inode(inode, inode->i_mode,
2032                                    inode->i_rdev);
2033
2034                 EXIT;
2035         }
2036
2037         return 0;
2038 }
2039
2040 void ll_delete_inode(struct inode *inode)
2041 {
2042         struct ll_inode_info *lli = ll_i2info(inode);
2043         ENTRY;
2044
2045         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2046                 /* It is last chance to write out dirty pages,
2047                  * otherwise we may lose data while umount */
2048                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2049
2050         truncate_inode_pages_final(&inode->i_data);
2051
2052         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2053                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2054                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2055
2056 #ifdef HAVE_SBOPS_EVICT_INODE
2057         ll_clear_inode(inode);
2058 #endif
2059         clear_inode(inode);
2060
2061         EXIT;
2062 }
2063
2064 int ll_iocontrol(struct inode *inode, struct file *file,
2065                  unsigned int cmd, unsigned long arg)
2066 {
2067         struct ll_sb_info *sbi = ll_i2sbi(inode);
2068         struct ptlrpc_request *req = NULL;
2069         int rc, flags = 0;
2070         ENTRY;
2071
2072         switch(cmd) {
2073         case FSFILT_IOC_GETFLAGS: {
2074                 struct mdt_body *body;
2075                 struct md_op_data *op_data;
2076
2077                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2078                                              0, 0, LUSTRE_OPC_ANY,
2079                                              NULL);
2080                 if (IS_ERR(op_data))
2081                         RETURN(PTR_ERR(op_data));
2082
2083                 op_data->op_valid = OBD_MD_FLFLAGS;
2084                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2085                 ll_finish_md_op_data(op_data);
2086                 if (rc) {
2087                         CERROR("%s: failure inode "DFID": rc = %d\n",
2088                                sbi->ll_md_exp->exp_obd->obd_name,
2089                                PFID(ll_inode2fid(inode)), rc);
2090                         RETURN(-abs(rc));
2091                 }
2092
2093                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2094
2095                 flags = body->mbo_flags;
2096
2097                 ptlrpc_req_finished(req);
2098
2099                 RETURN(put_user(flags, (int __user *)arg));
2100         }
2101         case FSFILT_IOC_SETFLAGS: {
2102                 struct iattr *attr;
2103                 struct md_op_data *op_data;
2104                 struct cl_object *obj;
2105                 struct obd_capa *capa;
2106
2107                 if (get_user(flags, (int __user *)arg))
2108                         RETURN(-EFAULT);
2109
2110                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2111                                              LUSTRE_OPC_ANY, NULL);
2112                 if (IS_ERR(op_data))
2113                         RETURN(PTR_ERR(op_data));
2114
2115                 op_data->op_attr_flags = flags;
2116                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2117                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2118                 ll_finish_md_op_data(op_data);
2119                 ptlrpc_req_finished(req);
2120                 if (rc)
2121                         RETURN(rc);
2122
2123                 inode->i_flags = ll_ext_to_inode_flags(flags);
2124
2125                 obj = ll_i2info(inode)->lli_clob;
2126                 if (obj == NULL)
2127                         RETURN(0);
2128
2129                 OBD_ALLOC_PTR(attr);
2130                 if (attr == NULL)
2131                         RETURN(-ENOMEM);
2132
2133                 attr->ia_valid = ATTR_ATTR_FLAG;
2134
2135                 capa = ll_mdscapa_get(inode);
2136                 rc = cl_setattr_ost(obj, attr, flags, capa);
2137                 capa_put(capa);
2138
2139                 OBD_FREE_PTR(attr);
2140                 RETURN(rc);
2141         }
2142         default:
2143                 RETURN(-ENOSYS);
2144         }
2145
2146         RETURN(0);
2147 }
2148
2149 int ll_flush_ctx(struct inode *inode)
2150 {
2151         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2152
2153         CDEBUG(D_SEC, "flush context for user %d\n",
2154                from_kuid(&init_user_ns, current_uid()));
2155
2156         obd_set_info_async(NULL, sbi->ll_md_exp,
2157                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2158                            0, NULL, NULL);
2159         obd_set_info_async(NULL, sbi->ll_dt_exp,
2160                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2161                            0, NULL, NULL);
2162         return 0;
2163 }
2164
2165 /* umount -f client means force down, don't save state */
2166 void ll_umount_begin(struct super_block *sb)
2167 {
2168         struct ll_sb_info *sbi = ll_s2sbi(sb);
2169         struct obd_device *obd;
2170         struct obd_ioctl_data *ioc_data;
2171         ENTRY;
2172
2173         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2174                sb->s_count, atomic_read(&sb->s_active));
2175
2176         obd = class_exp2obd(sbi->ll_md_exp);
2177         if (obd == NULL) {
2178                 CERROR("Invalid MDC connection handle "LPX64"\n",
2179                        sbi->ll_md_exp->exp_handle.h_cookie);
2180                 EXIT;
2181                 return;
2182         }
2183         obd->obd_force = 1;
2184
2185         obd = class_exp2obd(sbi->ll_dt_exp);
2186         if (obd == NULL) {
2187                 CERROR("Invalid LOV connection handle "LPX64"\n",
2188                        sbi->ll_dt_exp->exp_handle.h_cookie);
2189                 EXIT;
2190                 return;
2191         }
2192         obd->obd_force = 1;
2193
2194         OBD_ALLOC_PTR(ioc_data);
2195         if (ioc_data) {
2196                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2197                               sizeof *ioc_data, ioc_data, NULL);
2198
2199                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2200                               sizeof *ioc_data, ioc_data, NULL);
2201
2202                 OBD_FREE_PTR(ioc_data);
2203         }
2204
2205         /* Really, we'd like to wait until there are no requests outstanding,
2206          * and then continue.  For now, we just invalidate the requests,
2207          * schedule() and sleep one second if needed, and hope.
2208          */
2209         schedule();
2210         EXIT;
2211 }
2212
2213 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2214 {
2215         struct ll_sb_info *sbi = ll_s2sbi(sb);
2216         char *profilenm = get_profile_name(sb);
2217         int err;
2218         __u32 read_only;
2219
2220         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2221                 read_only = *flags & MS_RDONLY;
2222                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2223                                          sizeof(KEY_READ_ONLY),
2224                                          KEY_READ_ONLY, sizeof(read_only),
2225                                          &read_only, NULL);
2226                 if (err) {
2227                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2228                                       profilenm, read_only ?
2229                                       "read-only" : "read-write", err);
2230                         return err;
2231                 }
2232
2233                 if (read_only)
2234                         sb->s_flags |= MS_RDONLY;
2235                 else
2236                         sb->s_flags &= ~MS_RDONLY;
2237
2238                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2239                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2240                                       read_only ?  "read-only" : "read-write");
2241         }
2242         return 0;
2243 }
2244
2245 /**
2246  * Cleanup the open handle that is cached on MDT-side.
2247  *
2248  * For open case, the client side open handling thread may hit error
2249  * after the MDT grant the open. Under such case, the client should
2250  * send close RPC to the MDT as cleanup; otherwise, the open handle
2251  * on the MDT will be leaked there until the client umount or evicted.
2252  *
2253  * In further, if someone unlinked the file, because the open handle
2254  * holds the reference on such file/object, then it will block the
2255  * subsequent threads that want to locate such object via FID.
2256  *
2257  * \param[in] sb        super block for this file-system
2258  * \param[in] open_req  pointer to the original open request
2259  */
2260 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2261 {
2262         struct mdt_body                 *body;
2263         struct md_op_data               *op_data;
2264         struct ptlrpc_request           *close_req = NULL;
2265         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2266         ENTRY;
2267
2268         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2269         OBD_ALLOC_PTR(op_data);
2270         if (op_data == NULL) {
2271                 CWARN("%s: cannot allocate op_data to release open handle for "
2272                       DFID"\n",
2273                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2274
2275                 RETURN_EXIT;
2276         }
2277
2278         op_data->op_fid1 = body->mbo_fid1;
2279         op_data->op_handle = body->mbo_handle;
2280         op_data->op_mod_time = cfs_time_current_sec();
2281         md_close(exp, op_data, NULL, &close_req);
2282         ptlrpc_req_finished(close_req);
2283         ll_finish_md_op_data(op_data);
2284
2285         EXIT;
2286 }
2287
2288 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2289                   struct super_block *sb, struct lookup_intent *it)
2290 {
2291         struct ll_sb_info *sbi = NULL;
2292         struct lustre_md md = { NULL };
2293         int rc;
2294         ENTRY;
2295
2296         LASSERT(*inode || sb);
2297         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2298         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2299                               sbi->ll_md_exp, &md);
2300         if (rc != 0)
2301                 GOTO(cleanup, rc);
2302
2303         if (*inode) {
2304                 rc = ll_update_inode(*inode, &md);
2305                 if (rc != 0)
2306                         GOTO(out, rc);
2307         } else {
2308                 LASSERT(sb != NULL);
2309
2310                 /*
2311                  * At this point server returns to client's same fid as client
2312                  * generated for creating. So using ->fid1 is okay here.
2313                  */
2314                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2315
2316                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2317                                              sbi->ll_flags & LL_SBI_32BIT_API),
2318                                  &md);
2319                 if (IS_ERR(*inode)) {
2320 #ifdef CONFIG_FS_POSIX_ACL
2321                         if (md.posix_acl) {
2322                                 posix_acl_release(md.posix_acl);
2323                                 md.posix_acl = NULL;
2324                         }
2325 #endif
2326                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2327                         *inode = NULL;
2328                         CERROR("new_inode -fatal: rc %d\n", rc);
2329                         GOTO(out, rc);
2330                 }
2331         }
2332
2333         /* Handling piggyback layout lock.
2334          * Layout lock can be piggybacked by getattr and open request.
2335          * The lsm can be applied to inode only if it comes with a layout lock
2336          * otherwise correct layout may be overwritten, for example:
2337          * 1. proc1: mdt returns a lsm but not granting layout
2338          * 2. layout was changed by another client
2339          * 3. proc2: refresh layout and layout lock granted
2340          * 4. proc1: to apply a stale layout */
2341         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2342                 struct lustre_handle lockh;
2343                 struct ldlm_lock *lock;
2344
2345                 lockh.cookie = it->d.lustre.it_lock_handle;
2346                 lock = ldlm_handle2lock(&lockh);
2347                 LASSERT(lock != NULL);
2348                 if (ldlm_has_layout(lock)) {
2349                         struct cl_object_conf conf;
2350
2351                         memset(&conf, 0, sizeof(conf));
2352                         conf.coc_opc = OBJECT_CONF_SET;
2353                         conf.coc_inode = *inode;
2354                         conf.coc_lock = lock;
2355                         conf.u.coc_md = &md;
2356                         (void)ll_layout_conf(*inode, &conf);
2357                 }
2358                 LDLM_LOCK_PUT(lock);
2359         }
2360
2361         GOTO(out, rc = 0);
2362
2363 out:
2364         if (md.lsm != NULL)
2365                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2366         md_free_lustre_md(sbi->ll_md_exp, &md);
2367
2368 cleanup:
2369         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2370                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2371
2372         return rc;
2373 }
2374
2375 int ll_obd_statfs(struct inode *inode, void __user *arg)
2376 {
2377         struct ll_sb_info *sbi = NULL;
2378         struct obd_export *exp;
2379         char *buf = NULL;
2380         struct obd_ioctl_data *data = NULL;
2381         __u32 type;
2382         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2383         int len = 0, rc;
2384
2385         if (!inode || !(sbi = ll_i2sbi(inode)))
2386                 GOTO(out_statfs, rc = -EINVAL);
2387
2388         rc = obd_ioctl_getdata(&buf, &len, arg);
2389         if (rc)
2390                 GOTO(out_statfs, rc);
2391
2392         data = (void*)buf;
2393         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2394             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2395                 GOTO(out_statfs, rc = -EINVAL);
2396
2397         if (data->ioc_inllen1 != sizeof(__u32) ||
2398             data->ioc_inllen2 != sizeof(__u32) ||
2399             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2400             data->ioc_plen2 != sizeof(struct obd_uuid))
2401                 GOTO(out_statfs, rc = -EINVAL);
2402
2403         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2404         if (type & LL_STATFS_LMV)
2405                 exp = sbi->ll_md_exp;
2406         else if (type & LL_STATFS_LOV)
2407                 exp = sbi->ll_dt_exp;
2408         else
2409                 GOTO(out_statfs, rc = -ENODEV);
2410
2411         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2412         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2413         if (rc)
2414                 GOTO(out_statfs, rc);
2415 out_statfs:
2416         if (buf)
2417                 obd_ioctl_freedata(buf, len);
2418         return rc;
2419 }
2420
2421 int ll_process_config(struct lustre_cfg *lcfg)
2422 {
2423         struct super_block *sb;
2424         unsigned long x;
2425         int rc = 0;
2426         char *ptr;
2427
2428         /* The instance name contains the sb: lustre-client-aacfe000 */
2429         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2430         if (!ptr || !*(++ptr))
2431                 return -EINVAL;
2432         if (sscanf(ptr, "%lx", &x) != 1)
2433                 return -EINVAL;
2434         sb = (struct super_block *)x;
2435         /* This better be a real Lustre superblock! */
2436         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2437
2438         /* Note we have not called client_common_fill_super yet, so
2439            proc fns must be able to handle that! */
2440         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2441                                       lcfg, sb);
2442         if (rc > 0)
2443                 rc = 0;
2444         return rc;
2445 }
2446
2447 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2448 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2449                                        struct inode *i1, struct inode *i2,
2450                                        const char *name, size_t namelen,
2451                                        __u32 mode, __u32 opc, void *data)
2452 {
2453         LASSERT(i1 != NULL);
2454
2455         if (name == NULL) {
2456                 /* Do not reuse namelen for something else. */
2457                 if (namelen != 0)
2458                         return ERR_PTR(-EINVAL);
2459         } else {
2460                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2461                         return ERR_PTR(-ENAMETOOLONG);
2462
2463                 if (!lu_name_is_valid_2(name, namelen))
2464                         return ERR_PTR(-EINVAL);
2465         }
2466
2467         if (op_data == NULL)
2468                 OBD_ALLOC_PTR(op_data);
2469
2470         if (op_data == NULL)
2471                 return ERR_PTR(-ENOMEM);
2472
2473         ll_i2gids(op_data->op_suppgids, i1, i2);
2474         op_data->op_fid1 = *ll_inode2fid(i1);
2475         op_data->op_capa1 = ll_mdscapa_get(i1);
2476         op_data->op_default_stripe_offset = -1;
2477         if (S_ISDIR(i1->i_mode)) {
2478                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2479                 if (opc == LUSTRE_OPC_MKDIR)
2480                         op_data->op_default_stripe_offset =
2481                                    ll_i2info(i1)->lli_def_stripe_offset;
2482         }
2483
2484         if (i2) {
2485                 op_data->op_fid2 = *ll_inode2fid(i2);
2486                 op_data->op_capa2 = ll_mdscapa_get(i2);
2487                 if (S_ISDIR(i2->i_mode))
2488                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2489         } else {
2490                 fid_zero(&op_data->op_fid2);
2491                 op_data->op_capa2 = NULL;
2492         }
2493
2494         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2495                 op_data->op_cli_flags |= CLI_HASH64;
2496
2497         if (ll_need_32bit_api(ll_i2sbi(i1)))
2498                 op_data->op_cli_flags |= CLI_API32;
2499
2500         op_data->op_name = name;
2501         op_data->op_namelen = namelen;
2502         op_data->op_mode = mode;
2503         op_data->op_mod_time = cfs_time_current_sec();
2504         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2505         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2506         op_data->op_cap = cfs_curproc_cap_pack();
2507         op_data->op_bias = 0;
2508         op_data->op_cli_flags = 0;
2509         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2510              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2511                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2512         } else {
2513                 op_data->op_mds = 0;
2514         }
2515         op_data->op_data = data;
2516
2517         /* When called by ll_setattr_raw, file is i1. */
2518         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2519                 op_data->op_bias |= MDS_DATA_MODIFIED;
2520
2521         return op_data;
2522 }
2523
2524 void ll_finish_md_op_data(struct md_op_data *op_data)
2525 {
2526         capa_put(op_data->op_capa1);
2527         capa_put(op_data->op_capa2);
2528         OBD_FREE_PTR(op_data);
2529 }
2530
2531 #ifdef HAVE_SUPEROPS_USE_DENTRY
2532 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2533 #else
2534 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2535 #endif
2536 {
2537         struct ll_sb_info *sbi;
2538
2539 #ifdef HAVE_SUPEROPS_USE_DENTRY
2540         LASSERT((seq != NULL) && (dentry != NULL));
2541         sbi = ll_s2sbi(dentry->d_sb);
2542 #else
2543         LASSERT((seq != NULL) && (vfs != NULL));
2544         sbi = ll_s2sbi(vfs->mnt_sb);
2545 #endif
2546
2547         if (sbi->ll_flags & LL_SBI_NOLCK)
2548                 seq_puts(seq, ",nolock");
2549
2550         if (sbi->ll_flags & LL_SBI_FLOCK)
2551                 seq_puts(seq, ",flock");
2552
2553         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2554                 seq_puts(seq, ",localflock");
2555
2556         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2557                 seq_puts(seq, ",user_xattr");
2558
2559         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2560                 seq_puts(seq, ",lazystatfs");
2561
2562         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2563                 seq_puts(seq, ",user_fid2path");
2564
2565         RETURN(0);
2566 }
2567
2568 /**
2569  * Get obd name by cmd, and copy out to user space
2570  */
2571 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2572 {
2573         struct ll_sb_info *sbi = ll_i2sbi(inode);
2574         struct obd_device *obd;
2575         ENTRY;
2576
2577         if (cmd == OBD_IOC_GETDTNAME)
2578                 obd = class_exp2obd(sbi->ll_dt_exp);
2579         else if (cmd == OBD_IOC_GETMDNAME)
2580                 obd = class_exp2obd(sbi->ll_md_exp);
2581         else
2582                 RETURN(-EINVAL);
2583
2584         if (!obd)
2585                 RETURN(-ENOENT);
2586
2587         if (copy_to_user((void __user *)arg, obd->obd_name,
2588                          strlen(obd->obd_name) + 1))
2589                 RETURN(-EFAULT);
2590
2591         RETURN(0);
2592 }
2593
2594 /**
2595  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2596  * fsname will be returned in this buffer; otherwise, a static buffer will be
2597  * used to store the fsname and returned to caller.
2598  */
2599 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2600 {
2601         static char fsname_static[MTI_NAME_MAXLEN];
2602         struct lustre_sb_info *lsi = s2lsi(sb);
2603         char *ptr;
2604         int len;
2605
2606         if (buf == NULL) {
2607                 /* this means the caller wants to use static buffer
2608                  * and it doesn't care about race. Usually this is
2609                  * in error reporting path */
2610                 buf = fsname_static;
2611                 buflen = sizeof(fsname_static);
2612         }
2613
2614         len = strlen(lsi->lsi_lmd->lmd_profile);
2615         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2616         if (ptr && (strcmp(ptr, "-client") == 0))
2617                 len -= 7;
2618
2619         if (unlikely(len >= buflen))
2620                 len = buflen - 1;
2621         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2622         buf[len] = '\0';
2623
2624         return buf;
2625 }
2626
2627 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2628 {
2629         char *path = NULL;
2630
2631         struct path p;
2632
2633         p.dentry = dentry;
2634         p.mnt = current->fs->root.mnt;
2635         path_get(&p);
2636         path = d_path(&p, buf, bufsize);
2637         path_put(&p);
2638         return path;
2639 }
2640
2641 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2642 {
2643         char *buf, *path = NULL;
2644         struct dentry *dentry = NULL;
2645         struct inode *inode = page->mapping->host;
2646
2647         /* this can be called inside spin lock so use GFP_ATOMIC. */
2648         buf = (char *)__get_free_page(GFP_ATOMIC);
2649         if (buf != NULL) {
2650                 dentry = d_find_alias(page->mapping->host);
2651                 if (dentry != NULL)
2652                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2653         }
2654
2655         CDEBUG(D_WARNING,
2656                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2657                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2658                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2659                PFID(ll_inode2fid(inode)),
2660                (path && !IS_ERR(path)) ? path : "", ioret);
2661
2662         if (dentry != NULL)
2663                 dput(dentry);
2664
2665         if (buf != NULL)
2666                 free_page((unsigned long)buf);
2667 }
2668
2669 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2670                         struct lov_user_md **kbuf)
2671 {
2672         struct lov_user_md      lum;
2673         ssize_t                 lum_size;
2674         ENTRY;
2675
2676         if (copy_from_user(&lum, md, sizeof(lum)))
2677                 RETURN(-EFAULT);
2678
2679         lum_size = ll_lov_user_md_size(&lum);
2680         if (lum_size < 0)
2681                 RETURN(lum_size);
2682
2683         OBD_ALLOC(*kbuf, lum_size);
2684         if (*kbuf == NULL)
2685                 RETURN(-ENOMEM);
2686
2687         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2688                 OBD_FREE(*kbuf, lum_size);
2689                 RETURN(-EFAULT);
2690         }
2691
2692         RETURN(lum_size);
2693 }
2694
2695 /*
2696  * Compute llite root squash state after a change of root squash
2697  * configuration setting or add/remove of a lnet nid
2698  */
2699 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2700 {
2701         struct root_squash_info *squash = &sbi->ll_squash;
2702         int i;
2703         bool matched;
2704         lnet_process_id_t id;
2705
2706         /* Update norootsquash flag */
2707         down_write(&squash->rsi_sem);
2708         if (list_empty(&squash->rsi_nosquash_nids))
2709                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2710         else {
2711                 /* Do not apply root squash as soon as one of our NIDs is
2712                  * in the nosquash_nids list */
2713                 matched = false;
2714                 i = 0;
2715                 while (LNetGetId(i++, &id) != -ENOENT) {
2716                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2717                                 continue;
2718                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2719                                 matched = true;
2720                                 break;
2721                         }
2722                 }
2723                 if (matched)
2724                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2725                 else
2726                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2727         }
2728         up_write(&squash->rsi_sem);
2729 }
2730
2731 /**
2732  * Parse linkea content to extract information about a given hardlink
2733  *
2734  * \param[in]   ldata      - Initialized linkea data
2735  * \param[in]   linkno     - Link identifier
2736  * \param[out]  parent_fid - The entry's parent FID
2737  * \param[out]  ln         - Entry name destination buffer
2738  *
2739  * \retval 0 on success
2740  * \retval Appropriate negative error code on failure
2741  */
2742 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2743                             struct lu_fid *parent_fid, struct lu_name *ln)
2744 {
2745         unsigned int    idx;
2746         int             rc;
2747         ENTRY;
2748
2749         rc = linkea_init(ldata);
2750         if (rc < 0)
2751                 RETURN(rc);
2752
2753         if (linkno >= ldata->ld_leh->leh_reccount)
2754                 /* beyond last link */
2755                 RETURN(-ENODATA);
2756
2757         linkea_first_entry(ldata);
2758         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2759                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2760                                     parent_fid);
2761                 if (idx == linkno)
2762                         break;
2763
2764                 linkea_next_entry(ldata);
2765         }
2766
2767         if (idx < linkno)
2768                 RETURN(-ENODATA);
2769
2770         RETURN(0);
2771 }
2772
2773 /**
2774  * Get parent FID and name of an identified link. Operation is performed for
2775  * a given link number, letting the caller iterate over linkno to list one or
2776  * all links of an entry.
2777  *
2778  * \param[in]     file - File descriptor against which to perform the operation
2779  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2780  *                       on and the available size. It is eventually filled with
2781  *                       the requested information or left untouched on error
2782  *
2783  * \retval - 0 on success
2784  * \retval - Appropriate negative error code on failure
2785  */
2786 int ll_getparent(struct file *file, struct getparent __user *arg)
2787 {
2788         struct dentry           *dentry = file->f_dentry;
2789         struct inode            *inode = file->f_dentry->d_inode;
2790         struct linkea_data      *ldata;
2791         struct lu_buf            buf = LU_BUF_NULL;
2792         struct lu_name           ln;
2793         struct lu_fid            parent_fid;
2794         __u32                    linkno;
2795         __u32                    name_size;
2796         int                      rc;
2797
2798         ENTRY;
2799
2800         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2801             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2802                 RETURN(-EPERM);
2803
2804         if (get_user(name_size, &arg->gp_name_size))
2805                 RETURN(-EFAULT);
2806
2807         if (get_user(linkno, &arg->gp_linkno))
2808                 RETURN(-EFAULT);
2809
2810         if (name_size > PATH_MAX)
2811                 RETURN(-EINVAL);
2812
2813         OBD_ALLOC(ldata, sizeof(*ldata));
2814         if (ldata == NULL)
2815                 RETURN(-ENOMEM);
2816
2817         rc = linkea_data_new(ldata, &buf);
2818         if (rc < 0)
2819                 GOTO(ldata_free, rc);
2820
2821         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2822         if (rc < 0)
2823                 GOTO(lb_free, rc);
2824
2825         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2826         if (rc < 0)
2827                 GOTO(lb_free, rc);
2828
2829         if (ln.ln_namelen >= name_size)
2830                 GOTO(lb_free, rc = -EOVERFLOW);
2831
2832         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2833                 GOTO(lb_free, rc = -EFAULT);
2834
2835         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2836                 GOTO(lb_free, rc = -EFAULT);
2837
2838         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2839                 GOTO(lb_free, rc = -EFAULT);
2840
2841 lb_free:
2842         lu_buf_free(&buf);
2843 ldata_free:
2844         OBD_FREE(ldata, sizeof(*ldata));
2845
2846         RETURN(rc);
2847 }