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