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, 2015, 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 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC_PTR(sbi);
  81         if (sbi == NULL)
  82                 RETURN(NULL);
  83
  84         spin_lock_init(&sbi->ll_lock);
  85         mutex_init(&sbi->ll_lco.lco_lock);
  86         spin_lock_init(&sbi->ll_pp_extent_lock);
  87         spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         lru_page_max = pages / 2;
  93
  94         /* initialize ll_cache data */
  95         sbi->ll_cache = cl_cache_init(lru_page_max);
  96         if (sbi->ll_cache == NULL) {
  97                 OBD_FREE(sbi, sizeof(*sbi));
  98                 RETURN(NULL);
  99         }
 100
 101         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 102                                            SBI_DEFAULT_READAHEAD_MAX);
 103         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 104         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 105                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 106         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 107         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 108
 109         ll_generate_random_uuid(uuid);
 110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 112
 113         sbi->ll_flags |= LL_SBI_VERBOSE;
 114 #ifdef ENABLE_CHECKSUM
 115         sbi->ll_flags |= LL_SBI_CHECKSUM;
 116 #endif
 117
 118 #ifdef HAVE_LRU_RESIZE_SUPPORT
 119         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 120 #endif
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136
 137         /* root squash */
 138         sbi->ll_squash.rsi_uid = 0;
 139         sbi->ll_squash.rsi_gid = 0;
 140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 141         init_rwsem(&sbi->ll_squash.rsi_sem);
 142
 143         RETURN(sbi);
 144 }
 145
 146 static void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 154                 if (sbi->ll_cache != NULL) {
 155                         cl_cache_decref(sbi->ll_cache);
 156                         sbi->ll_cache = NULL;
 157                 }
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 164                                     struct vfsmount *mnt)
 165 {
 166         struct inode *root = NULL;
 167         struct ll_sb_info *sbi = ll_s2sbi(sb);
 168         struct obd_device *obd;
 169         struct obd_statfs *osfs = NULL;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         u64 valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         OBD_ALLOC_PTR(osfs);
 190         if (osfs == NULL) {
 191                 OBD_FREE_PTR(data);
 192                 RETURN(-ENOMEM);
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 204                                   OBD_CONNECT_EINPROGRESS |
 205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 207                                   OBD_CONNECT_MAX_EASIZE |
 208                                   OBD_CONNECT_FLOCK_DEAD |
 209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 210                                   OBD_CONNECT_OPEN_BY_FID |
 211                                   OBD_CONNECT_DIR_STRIPE |
 212                                   OBD_CONNECT_BULK_MBITS |
 213                                   OBD_CONNECT_SUBTREE;
 214
 215 #ifdef HAVE_LRU_RESIZE_SUPPORT
 216         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 217                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 218 #endif
 219 #ifdef CONFIG_FS_POSIX_ACL
 220         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 221 #endif
 222
 223         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 224                 /* flag mdc connection as lightweight, only used for test
 225                  * purpose, use with care */
 226                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 227
 228         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 229         data->ocd_version = LUSTRE_VERSION_CODE;
 230
 231         if (sb->s_flags & MS_RDONLY)
 232                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 233         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 234                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 235
 236 #ifdef HAVE_MS_FLOCK_LOCK
 237         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 238         sb->s_flags |= MS_FLOCK_LOCK;
 239 #endif
 240 #ifdef MS_HAS_NEW_AOPS
 241         sb->s_flags |= MS_HAS_NEW_AOPS;
 242 #endif
 243
 244         if (sbi->ll_flags & LL_SBI_FLOCK)
 245                 sbi->ll_fop = &ll_file_operations_flock;
 246         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 247                 sbi->ll_fop = &ll_file_operations;
 248         else
 249                 sbi->ll_fop = &ll_file_operations_noflock;
 250
 251         /* real client */
 252         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 253         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 254                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 255
 256         /* always ping even if server suppress_pings */
 257         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 258                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 259
 260         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 261
 262         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 263         if (err == -EBUSY) {
 264                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 265                                    "recovery, of which this client is not a "
 266                                    "part. Please wait for recovery to complete,"
 267                                    " abort, or time out.\n", md);
 268                 GOTO(out, err);
 269         } else if (err) {
 270                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 271                 GOTO(out, err);
 272         }
 273
 274         sbi->ll_md_exp->exp_connect_data = *data;
 275
 276         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 277                            LUSTRE_SEQ_METADATA);
 278         if (err) {
 279                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 280                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 281                 GOTO(out_md, err);
 282         }
 283
 284         /* For mount, we only need fs info from MDT0, and also in DNE, it
 285          * can make sure the client can be mounted as long as MDT0 is
 286          * avaible */
 287         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 288                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 289                         OBD_STATFS_FOR_MDT0);
 290         if (err)
 291                 GOTO(out_md_fid, err);
 292
 293         /* This needs to be after statfs to ensure connect has finished.
 294          * Note that "data" does NOT contain the valid connect reply.
 295          * If connecting to a 1.8 server there will be no LMV device, so
 296          * we can access the MDC export directly and exp_connect_flags will
 297          * be non-zero, but if accessing an upgraded 2.1 server it will
 298          * have the correct flags filled in.
 299          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 300         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 301         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 302             valid != CLIENT_CONNECT_MDT_REQD) {
 303                 char *buf;
 304
 305                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 306                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 307                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 308                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 309                                    "feature(s) needed for correct operation "
 310                                    "of this client (%s). Please upgrade "
 311                                    "server or downgrade client.\n",
 312                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 313                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 314                 GOTO(out_md_fid, err = -EPROTO);
 315         }
 316
 317         size = sizeof(*data);
 318         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 319                            KEY_CONN_DATA,  &size, data);
 320         if (err) {
 321                 CERROR("%s: Get connect data failed: rc = %d\n",
 322                        sbi->ll_md_exp->exp_obd->obd_name, err);
 323                 GOTO(out_md_fid, err);
 324         }
 325
 326         LASSERT(osfs->os_bsize);
 327         sb->s_blocksize = osfs->os_bsize;
 328         sb->s_blocksize_bits = log2(osfs->os_bsize);
 329         sb->s_magic = LL_SUPER_MAGIC;
 330         sb->s_maxbytes = MAX_LFS_FILESIZE;
 331         sbi->ll_namelen = osfs->os_namelen;
 332
 333         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 334             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 335                 LCONSOLE_INFO("Disabling user_xattr feature because "
 336                               "it is not supported on the server\n");
 337                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 338         }
 339
 340         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 341 #ifdef MS_POSIXACL
 342                 sb->s_flags |= MS_POSIXACL;
 343 #endif
 344                 sbi->ll_flags |= LL_SBI_ACL;
 345         } else {
 346                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags &= ~MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags &= ~LL_SBI_ACL;
 351         }
 352
 353         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 354                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 355                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 356                         LCONSOLE_INFO("client is set as remote by default.\n");
 357                 }
 358         } else {
 359                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 360                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client claims to be remote, but server "
 362                                       "rejected, forced to be local.\n");
 363                 }
 364         }
 365
 366         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 367                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 370                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 371         else
 372                 sbi->ll_md_brw_pages = 1;
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 375                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 376
 377         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 378                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 379                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 380                                       "unknown maximum xattr size.\n", dt);
 381                 } else {
 382                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 383                         sbi->ll_xattr_cache_enabled = 1;
 384                 }
 385         }
 386
 387         obd = class_name2obd(dt);
 388         if (!obd) {
 389                 CERROR("DT %s: not setup or attached\n", dt);
 390                 GOTO(out_md_fid, err = -ENODEV);
 391         }
 392
 393         /* pass client page size via ocd_grant_blkbits, the server should report
 394          * back its backend blocksize for grant calculation purpose */
 395         data->ocd_grant_blkbits = PAGE_SHIFT;
 396
 397         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 398                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 399                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 400                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 401                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 402                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 403                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 404                                   OBD_CONNECT_MAXBYTES |
 405                                   OBD_CONNECT_EINPROGRESS |
 406                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 407                                   OBD_CONNECT_LAYOUTLOCK |
 408                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 409                                   OBD_CONNECT_BULK_MBITS;
 410
 411         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 412                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 413
 414         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 415                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 416                  * disabled by default, because it can still be enabled on the
 417                  * fly via /proc. As a consequence, we still need to come to an
 418                  * agreement on the supported algorithms at connect time */
 419                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 420
 421                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 422                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 423                 else
 424                         data->ocd_cksum_types = cksum_types_supported_client();
 425         }
 426
 427 #ifdef HAVE_LRU_RESIZE_SUPPORT
 428         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 429 #endif
 430         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 431                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 432
 433         /* always ping even if server suppress_pings */
 434         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 435                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 436
 437         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 438                "ocd_grant: %d\n", data->ocd_connect_flags,
 439                data->ocd_version, data->ocd_grant);
 440
 441         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 442         obd->obd_upcall.onu_upcall = cl_ocd_update;
 443
 444         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 445
 446         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 447                           NULL);
 448         if (err == -EBUSY) {
 449                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 450                                    "recovery, of which this client is not a "
 451                                    "part.  Please wait for recovery to "
 452                                    "complete, abort, or time out.\n", dt);
 453                 GOTO(out_md, err);
 454         } else if (err) {
 455                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 456                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 457                 GOTO(out_md, err);
 458         }
 459
 460         sbi->ll_dt_exp->exp_connect_data = *data;
 461
 462         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 463                            LUSTRE_SEQ_METADATA);
 464         if (err) {
 465                 CERROR("%s: Can't init data layer FID infrastructure, "
 466                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 467                 GOTO(out_dt, err);
 468         }
 469
 470         mutex_lock(&sbi->ll_lco.lco_lock);
 471         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 472         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 473         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 474         mutex_unlock(&sbi->ll_lco.lco_lock);
 475
 476         fid_zero(&sbi->ll_root_fid);
 477         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 478                            &sbi->ll_root_fid);
 479         if (err) {
 480                 CERROR("cannot mds_connect: rc = %d\n", err);
 481                 GOTO(out_lock_cn_cb, err);
 482         }
 483         if (!fid_is_sane(&sbi->ll_root_fid)) {
 484                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 485                        sbi->ll_md_exp->exp_obd->obd_name,
 486                        PFID(&sbi->ll_root_fid));
 487                 GOTO(out_lock_cn_cb, err = -EINVAL);
 488         }
 489         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 490
 491         sb->s_op = &lustre_super_operations;
 492 #if THREAD_SIZE >= 8192 /*b=17630*/
 493         sb->s_export_op = &lustre_export_operations;
 494 #endif
 495
 496         /* make root inode
 497          * XXX: move this to after cbd setup? */
 498         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 499         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 500                 valid |= OBD_MD_FLRMTPERM;
 501         else if (sbi->ll_flags & LL_SBI_ACL)
 502                 valid |= OBD_MD_FLACL;
 503
 504         OBD_ALLOC_PTR(op_data);
 505         if (op_data == NULL)
 506                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 507
 508         op_data->op_fid1 = sbi->ll_root_fid;
 509         op_data->op_mode = 0;
 510         op_data->op_valid = valid;
 511
 512         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 513
 514         OBD_FREE_PTR(op_data);
 515         if (err) {
 516                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 517                        sbi->ll_md_exp->exp_obd->obd_name, err);
 518                 GOTO(out_lock_cn_cb, err);
 519         }
 520
 521         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 522                                sbi->ll_md_exp, &lmd);
 523         if (err) {
 524                 CERROR("failed to understand root inode md: rc = %d\n", err);
 525                 ptlrpc_req_finished(request);
 526                 GOTO(out_lock_cn_cb, err);
 527         }
 528
 529         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 530         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 531                                             sbi->ll_flags & LL_SBI_32BIT_API),
 532                        &lmd);
 533         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 534         ptlrpc_req_finished(request);
 535
 536         if (IS_ERR(root)) {
 537 #ifdef CONFIG_FS_POSIX_ACL
 538                 if (lmd.posix_acl) {
 539                         posix_acl_release(lmd.posix_acl);
 540                         lmd.posix_acl = NULL;
 541                 }
 542 #endif
 543                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 544                 root = NULL;
 545                 CERROR("lustre_lite: bad iget4 for root\n");
 546                 GOTO(out_root, err);
 547         }
 548
 549 #ifdef CONFIG_FS_POSIX_ACL
 550         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 551                 rct_init(&sbi->ll_rct);
 552                 et_init(&sbi->ll_et);
 553         }
 554 #endif
 555
 556         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 557         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 558                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 559                                  NULL);
 560         cl_sb_init(sb);
 561
 562         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 563                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 564                                  sbi->ll_cache, NULL);
 565
 566         sb->s_root = d_make_root(root);
 567         if (sb->s_root == NULL) {
 568                 CERROR("%s: can't make root dentry\n",
 569                         ll_get_fsname(sb, NULL, 0));
 570                 GOTO(out_root, err = -ENOMEM);
 571         }
 572 #ifdef HAVE_DCACHE_LOCK
 573         sb->s_root->d_op = &ll_d_ops;
 574 #endif
 575
 576         sbi->ll_sdev_orig = sb->s_dev;
 577
 578         /* We set sb->s_dev equal on all lustre clients in order to support
 579          * NFS export clustering.  NFSD requires that the FSID be the same
 580          * on all clients. */
 581         /* s_dev is also used in lt_compare() to compare two fs, but that is
 582          * only a node-local comparison. */
 583         uuid = obd_get_uuid(sbi->ll_md_exp);
 584         if (uuid != NULL)
 585                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 586
 587         if (data != NULL)
 588                 OBD_FREE_PTR(data);
 589         if (osfs != NULL)
 590                 OBD_FREE_PTR(osfs);
 591         if (proc_lustre_fs_root != NULL) {
 592                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 593                                                   dt, md);
 594                 if (err < 0) {
 595                         CERROR("%s: could not register mount in lprocfs: "
 596                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 597                         err = 0;
 598                 }
 599         }
 600
 601         RETURN(err);
 602 out_root:
 603         if (root)
 604                 iput(root);
 605 out_lock_cn_cb:
 606         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 607 out_dt:
 608         obd_disconnect(sbi->ll_dt_exp);
 609         sbi->ll_dt_exp = NULL;
 610 out_md_fid:
 611         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 612 out_md:
 613         obd_disconnect(sbi->ll_md_exp);
 614         sbi->ll_md_exp = NULL;
 615 out:
 616         if (data != NULL)
 617                 OBD_FREE_PTR(data);
 618         if (osfs != NULL)
 619                 OBD_FREE_PTR(osfs);
 620         return err;
 621 }
 622
 623 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 624 {
 625         int size, rc;
 626
 627         size = sizeof(*lmmsize);
 628         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 629                           KEY_MAX_EASIZE, &size, lmmsize);
 630         if (rc != 0) {
 631                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 632                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 633                 RETURN(rc);
 634         }
 635
 636         size = sizeof(int);
 637         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 638                           KEY_MAX_EASIZE, &size, lmmsize);
 639         if (rc)
 640                 CERROR("Get max mdsize error rc %d\n", rc);
 641
 642         RETURN(rc);
 643 }
 644
 645 /**
 646  * Get the value of the default_easize parameter.
 647  *
 648  * \see client_obd::cl_default_mds_easize
 649  *
 650  * \param[in] sbi       superblock info for this filesystem
 651  * \param[out] lmmsize  pointer to storage location for value
 652  *
 653  * \retval 0            on success
 654  * \retval negative     negated errno on failure
 655  */
 656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 657 {
 658         int size, rc;
 659
 660         size = sizeof(int);
 661         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 662                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 663         if (rc)
 664                 CERROR("Get default mdsize error rc %d\n", rc);
 665
 666         RETURN(rc);
 667 }
 668
 669 /**
 670  * Set the default_easize parameter to the given value.
 671  *
 672  * \see client_obd::cl_default_mds_easize
 673  *
 674  * \param[in] sbi       superblock info for this filesystem
 675  * \param[in] lmmsize   the size to set
 676  *
 677  * \retval 0            on success
 678  * \retval negative     negated errno on failure
 679  */
 680 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 681 {
 682         int rc;
 683
 684         if (lmmsize < sizeof(struct lov_mds_md) ||
 685             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 686                 return -EINVAL;
 687
 688         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 689                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 690                                 sizeof(int), &lmmsize, NULL);
 691
 692         RETURN(rc);
 693 }
 694
 695 static void ll_dump_inode(struct inode *inode)
 696 {
 697         struct ll_d_hlist_node *tmp;
 698         int dentry_count = 0;
 699
 700         LASSERT(inode != NULL);
 701
 702         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 703                 dentry_count++;
 704
 705         CERROR("%s: inode %p dump: dev=%s fid="DFID
 706                " mode=%o count=%u, %d dentries\n",
 707                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 708                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 709                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 710 }
 711
 712 void lustre_dump_dentry(struct dentry *dentry, int recur)
 713 {
 714         struct list_head *tmp;
 715         int subdirs = 0;
 716
 717         LASSERT(dentry != NULL);
 718
 719         list_for_each(tmp, &dentry->d_subdirs)
 720                 subdirs++;
 721
 722         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 723                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 724                dentry->d_name.len, dentry->d_name.name,
 725                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 726                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 727                dentry->d_flags, dentry->d_fsdata, subdirs);
 728         if (dentry->d_inode != NULL)
 729                 ll_dump_inode(dentry->d_inode);
 730
 731         if (recur == 0)
 732                 return;
 733
 734         list_for_each(tmp, &dentry->d_subdirs) {
 735                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 736                 lustre_dump_dentry(d, recur - 1);
 737         }
 738 }
 739
 740 static void client_common_put_super(struct super_block *sb)
 741 {
 742         struct ll_sb_info *sbi = ll_s2sbi(sb);
 743         ENTRY;
 744
 745 #ifdef CONFIG_FS_POSIX_ACL
 746         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 747                 et_fini(&sbi->ll_et);
 748                 rct_fini(&sbi->ll_rct);
 749         }
 750 #endif
 751
 752         cl_sb_fini(sb);
 753
 754         list_del(&sbi->ll_conn_chain);
 755
 756         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 757         obd_disconnect(sbi->ll_dt_exp);
 758         sbi->ll_dt_exp = NULL;
 759
 760         lprocfs_unregister_mountpoint(sbi);
 761
 762         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 763         obd_disconnect(sbi->ll_md_exp);
 764         sbi->ll_md_exp = NULL;
 765
 766         EXIT;
 767 }
 768
 769 void ll_kill_super(struct super_block *sb)
 770 {
 771         struct ll_sb_info *sbi;
 772         ENTRY;
 773
 774         /* not init sb ?*/
 775         if (!(sb->s_flags & MS_ACTIVE))
 776                 return;
 777
 778         sbi = ll_s2sbi(sb);
 779         /* we need restore s_dev from changed for clustred NFS before put_super
 780          * because new kernels have cached s_dev and change sb->s_dev in
 781          * put_super not affected real removing devices */
 782         if (sbi) {
 783                 sb->s_dev = sbi->ll_sdev_orig;
 784                 sbi->ll_umounting = 1;
 785
 786                 /* wait running statahead threads to quit */
 787                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 788                         set_current_state(TASK_UNINTERRUPTIBLE);
 789                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 790                 }
 791         }
 792
 793         EXIT;
 794 }
 795
 796 static inline int ll_set_opt(const char *opt, char *data, int fl)
 797 {
 798         if (strncmp(opt, data, strlen(opt)) != 0)
 799                 return(0);
 800         else
 801                 return(fl);
 802 }
 803
 804 /* non-client-specific mount options are parsed in lmd_parse */
 805 static int ll_options(char *options, int *flags)
 806 {
 807         int tmp;
 808         char *s1 = options, *s2;
 809         ENTRY;
 810
 811         if (!options)
 812                 RETURN(0);
 813
 814         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 815
 816         while (*s1) {
 817                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 818                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("context", s1, 1);
 849                 if (tmp)
 850                         goto next;
 851                 tmp = ll_set_opt("fscontext", s1, 1);
 852                 if (tmp)
 853                         goto next;
 854                 tmp = ll_set_opt("defcontext", s1, 1);
 855                 if (tmp)
 856                         goto next;
 857                 tmp = ll_set_opt("rootcontext", s1, 1);
 858                 if (tmp)
 859                         goto next;
 860                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 861                 if (tmp) {
 862                         *flags |= tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 871                 if (tmp) {
 872                         *flags &= ~tmp;
 873                         goto next;
 874                 }
 875
 876                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 892                 if (tmp) {
 893                         *flags &= ~tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 902                 if (tmp) {
 903                         *flags &= ~tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 907                 if (tmp) {
 908                         *flags |= tmp;
 909                         goto next;
 910                 }
 911                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 912                 if (tmp) {
 913                         *flags |= tmp;
 914                         goto next;
 915                 }
 916                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 917                 if (tmp) {
 918                         *flags &= ~tmp;
 919                         goto next;
 920                 }
 921                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 922                 if (tmp) {
 923                         *flags |= tmp;
 924                         goto next;
 925                 }
 926                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 927                                    s1);
 928                 RETURN(-EINVAL);
 929
 930 next:
 931                 /* Find next opt */
 932                 s2 = strchr(s1, ',');
 933                 if (s2 == NULL)
 934                         break;
 935                 s1 = s2 + 1;
 936         }
 937         RETURN(0);
 938 }
 939
 940 void ll_lli_init(struct ll_inode_info *lli)
 941 {
 942         lli->lli_inode_magic = LLI_INODE_MAGIC;
 943         lli->lli_flags = 0;
 944         spin_lock_init(&lli->lli_lock);
 945         lli->lli_posix_acl = NULL;
 946         lli->lli_remote_perms = NULL;
 947         mutex_init(&lli->lli_rmtperm_mutex);
 948         /* Do not set lli_fid, it has been initialized already. */
 949         fid_zero(&lli->lli_pfid);
 950         atomic_set(&lli->lli_open_count, 0);
 951         lli->lli_rmtperm_time = 0;
 952         lli->lli_mds_read_och = NULL;
 953         lli->lli_mds_write_och = NULL;
 954         lli->lli_mds_exec_och = NULL;
 955         lli->lli_open_fd_read_count = 0;
 956         lli->lli_open_fd_write_count = 0;
 957         lli->lli_open_fd_exec_count = 0;
 958         mutex_init(&lli->lli_och_mutex);
 959         spin_lock_init(&lli->lli_agl_lock);
 960         spin_lock_init(&lli->lli_layout_lock);
 961         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 962         lli->lli_clob = NULL;
 963
 964         init_rwsem(&lli->lli_xattrs_list_rwsem);
 965         mutex_init(&lli->lli_xattrs_enq_lock);
 966
 967         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 968         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 969                 mutex_init(&lli->lli_readdir_mutex);
 970                 lli->lli_opendir_key = NULL;
 971                 lli->lli_sai = NULL;
 972                 spin_lock_init(&lli->lli_sa_lock);
 973                 lli->lli_opendir_pid = 0;
 974                 lli->lli_sa_enabled = 0;
 975                 lli->lli_def_stripe_offset = -1;
 976         } else {
 977                 mutex_init(&lli->lli_size_mutex);
 978                 lli->lli_symlink_name = NULL;
 979                 init_rwsem(&lli->lli_trunc_sem);
 980                 range_lock_tree_init(&lli->lli_write_tree);
 981                 init_rwsem(&lli->lli_glimpse_sem);
 982                 lli->lli_glimpse_time = 0;
 983                 INIT_LIST_HEAD(&lli->lli_agl_list);
 984                 lli->lli_agl_index = 0;
 985                 lli->lli_async_rc = 0;
 986         }
 987         mutex_init(&lli->lli_layout_mutex);
 988 }
 989
 990 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 991 {
 992         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 993
 994         bdi->name = "lustre";
 995         return bdi_register(bdi, NULL, "lustre-%d",
 996                             atomic_inc_return(&ll_bdi_num));
 997 }
 998
 999 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1000 {
1001         struct lustre_profile *lprof = NULL;
1002         struct lustre_sb_info *lsi = s2lsi(sb);
1003         struct ll_sb_info *sbi;
1004         char  *dt = NULL, *md = NULL;
1005         char  *profilenm = get_profile_name(sb);
1006         struct config_llog_instance *cfg;
1007         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1008         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1009         int    err;
1010         ENTRY;
1011
1012         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1013
1014         OBD_ALLOC_PTR(cfg);
1015         if (cfg == NULL)
1016                 RETURN(-ENOMEM);
1017
1018         try_module_get(THIS_MODULE);
1019
1020         /* client additional sb info */
1021         lsi->lsi_llsbi = sbi = ll_init_sbi();
1022         if (!sbi) {
1023                 module_put(THIS_MODULE);
1024                 OBD_FREE_PTR(cfg);
1025                 RETURN(-ENOMEM);
1026         }
1027
1028         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1029         if (err)
1030                 GOTO(out_free, err);
1031
1032         err = bdi_init(&lsi->lsi_bdi);
1033         if (err)
1034                 GOTO(out_free, err);
1035         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1036 #ifdef HAVE_BDI_CAP_MAP_COPY
1037         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1038 #else
1039         lsi->lsi_bdi.capabilities = 0;
1040 #endif
1041         err = ll_bdi_register(&lsi->lsi_bdi);
1042         if (err)
1043                 GOTO(out_free, err);
1044
1045         sb->s_bdi = &lsi->lsi_bdi;
1046 #ifndef HAVE_DCACHE_LOCK
1047         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1048         sb->s_d_op = &ll_d_ops;
1049 #endif
1050
1051         /* Generate a string unique to this super, in case some joker tries
1052            to mount the same fs at two mount points.
1053            Use the address of the super itself.*/
1054         cfg->cfg_instance = sb;
1055         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1056         cfg->cfg_callback = class_config_llog_handler;
1057         /* set up client obds */
1058         err = lustre_process_log(sb, profilenm, cfg);
1059         if (err < 0)
1060                 GOTO(out_free, err);
1061
1062         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1063         lprof = class_get_profile(profilenm);
1064         if (lprof == NULL) {
1065                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1066                                    " read from the MGS.  Does that filesystem "
1067                                    "exist?\n", profilenm);
1068                 GOTO(out_free, err = -EINVAL);
1069         }
1070         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1071                lprof->lp_md, lprof->lp_dt);
1072
1073         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1074         if (!dt)
1075                 GOTO(out_free, err = -ENOMEM);
1076         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1077
1078         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1079         if (!md)
1080                 GOTO(out_free, err = -ENOMEM);
1081         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1082
1083         /* connections, registrations, sb setup */
1084         err = client_common_fill_super(sb, md, dt, mnt);
1085         if (err < 0)
1086                 GOTO(out_free, err);
1087
1088         sbi->ll_client_common_fill_super_succeeded = 1;
1089
1090 out_free:
1091         if (md)
1092                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1093         if (dt)
1094                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1095         if (lprof != NULL)
1096                 class_put_profile(lprof);
1097         if (err)
1098                 ll_put_super(sb);
1099         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1100                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1101
1102         OBD_FREE_PTR(cfg);
1103         RETURN(err);
1104 } /* ll_fill_super */
1105
1106 void ll_put_super(struct super_block *sb)
1107 {
1108         struct config_llog_instance cfg, params_cfg;
1109         struct obd_device *obd;
1110         struct lustre_sb_info *lsi = s2lsi(sb);
1111         struct ll_sb_info *sbi = ll_s2sbi(sb);
1112         char *profilenm = get_profile_name(sb);
1113         long ccc_count;
1114         int next, force = 1, rc = 0;
1115         ENTRY;
1116
1117         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1118
1119         cfg.cfg_instance = sb;
1120         lustre_end_log(sb, profilenm, &cfg);
1121
1122         params_cfg.cfg_instance = sb;
1123         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1124
1125         if (sbi->ll_md_exp) {
1126                 obd = class_exp2obd(sbi->ll_md_exp);
1127                 if (obd)
1128                         force = obd->obd_force;
1129         }
1130
1131         /* Wait for unstable pages to be committed to stable storage */
1132         if (force == 0) {
1133                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1134                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1135                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1136                         &lwi);
1137         }
1138
1139         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1140         if (force == 0 && rc != -EINTR)
1141                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1142
1143
1144         /* We need to set force before the lov_disconnect in
1145            lustre_common_put_super, since l_d cleans up osc's as well. */
1146         if (force) {
1147                 next = 0;
1148                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1149                                                      &next)) != NULL) {
1150                         obd->obd_force = force;
1151                 }
1152         }
1153
1154         if (sbi->ll_client_common_fill_super_succeeded) {
1155                 /* Only if client_common_fill_super succeeded */
1156                 client_common_put_super(sb);
1157         }
1158
1159         next = 0;
1160         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1161                 class_manual_cleanup(obd);
1162         }
1163
1164         if (sbi->ll_flags & LL_SBI_VERBOSE)
1165                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1166
1167         if (profilenm)
1168                 class_del_profile(profilenm);
1169
1170         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1171                 bdi_destroy(&lsi->lsi_bdi);
1172                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1173         }
1174
1175         ll_free_sbi(sb);
1176         lsi->lsi_llsbi = NULL;
1177
1178         lustre_common_put_super(sb);
1179
1180         cl_env_cache_purge(~0);
1181
1182         module_put(THIS_MODULE);
1183
1184         EXIT;
1185 } /* client_put_super */
1186
1187 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1188 {
1189         struct inode *inode = NULL;
1190
1191         /* NOTE: we depend on atomic igrab() -bzzz */
1192         lock_res_and_lock(lock);
1193         if (lock->l_resource->lr_lvb_inode) {
1194                 struct ll_inode_info * lli;
1195                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1196                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1197                         inode = igrab(lock->l_resource->lr_lvb_inode);
1198                 } else {
1199                         inode = lock->l_resource->lr_lvb_inode;
1200                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1201                                          D_WARNING, lock, "lr_lvb_inode %p is "
1202                                          "bogus: magic %08x",
1203                                          lock->l_resource->lr_lvb_inode,
1204                                          lli->lli_inode_magic);
1205                         inode = NULL;
1206                 }
1207         }
1208         unlock_res_and_lock(lock);
1209         return inode;
1210 }
1211
1212 void ll_dir_clear_lsm_md(struct inode *inode)
1213 {
1214         struct ll_inode_info *lli = ll_i2info(inode);
1215
1216         LASSERT(S_ISDIR(inode->i_mode));
1217
1218         if (lli->lli_lsm_md != NULL) {
1219                 lmv_free_memmd(lli->lli_lsm_md);
1220                 lli->lli_lsm_md = NULL;
1221         }
1222 }
1223
1224 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1225                                       const struct lu_fid *fid,
1226                                       struct lustre_md *md)
1227 {
1228         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1229         struct mdt_body         *body = md->body;
1230         struct inode            *inode;
1231         ino_t                   ino;
1232         ENTRY;
1233
1234         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1235         inode = iget_locked(sb, ino);
1236         if (inode == NULL) {
1237                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1238                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1239                 RETURN(ERR_PTR(-ENOENT));
1240         }
1241
1242         if (inode->i_state & I_NEW) {
1243                 struct ll_inode_info *lli = ll_i2info(inode);
1244                 struct lmv_stripe_md *lsm = md->lmv;
1245
1246                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1247                                 (body->mbo_mode & S_IFMT);
1248                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1249                          PFID(fid));
1250
1251                 LTIME_S(inode->i_mtime) = 0;
1252                 LTIME_S(inode->i_atime) = 0;
1253                 LTIME_S(inode->i_ctime) = 0;
1254                 inode->i_rdev = 0;
1255
1256 #ifdef HAVE_BACKING_DEV_INFO
1257                 /* initializing backing dev info. */
1258                 inode->i_mapping->backing_dev_info =
1259                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1260 #endif
1261                 inode->i_op = &ll_dir_inode_operations;
1262                 inode->i_fop = &ll_dir_operations;
1263                 lli->lli_fid = *fid;
1264                 ll_lli_init(lli);
1265
1266                 LASSERT(lsm != NULL);
1267                 /* master object FID */
1268                 lli->lli_pfid = body->mbo_fid1;
1269                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1270                        lli, PFID(fid), PFID(&lli->lli_pfid));
1271                 unlock_new_inode(inode);
1272         }
1273
1274         RETURN(inode);
1275 }
1276
1277 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1278 {
1279         struct lu_fid *fid;
1280         struct lmv_stripe_md *lsm = md->lmv;
1281         int i;
1282
1283         LASSERT(lsm != NULL);
1284         /* XXX sigh, this lsm_root initialization should be in
1285          * LMV layer, but it needs ll_iget right now, so we
1286          * put this here right now. */
1287         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1288                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1289                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1290                 /* Unfortunately ll_iget will call ll_update_inode,
1291                  * where the initialization of slave inode is slightly
1292                  * different, so it reset lsm_md to NULL to avoid
1293                  * initializing lsm for slave inode. */
1294                 /* For migrating inode, master stripe and master object will
1295                  * be same, so we only need assign this inode */
1296                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1297                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1298                 else
1299                         lsm->lsm_md_oinfo[i].lmo_root =
1300                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1301
1302                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1303                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1304
1305                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1306                         return rc;
1307                 }
1308         }
1309
1310         return 0;
1311 }
1312
1313 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1314                                 const struct lmv_stripe_md *lsm_md2)
1315 {
1316         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1317                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1318                lsm_md1->lsm_md_master_mdt_index ==
1319                                         lsm_md2->lsm_md_master_mdt_index &&
1320                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1321                lsm_md1->lsm_md_layout_version ==
1322                                         lsm_md2->lsm_md_layout_version &&
1323                strcmp(lsm_md1->lsm_md_pool_name,
1324                       lsm_md2->lsm_md_pool_name) == 0;
1325 }
1326
1327 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1328 {
1329         struct ll_inode_info *lli = ll_i2info(inode);
1330         struct lmv_stripe_md *lsm = md->lmv;
1331         int     rc;
1332         ENTRY;
1333
1334         LASSERT(S_ISDIR(inode->i_mode));
1335         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1336                PFID(ll_inode2fid(inode)));
1337
1338         /* no striped information from request. */
1339         if (lsm == NULL) {
1340                 if (lli->lli_lsm_md == NULL) {
1341                         RETURN(0);
1342                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1343                                                 LMV_HASH_FLAG_MIGRATION) {
1344                         /* migration is done, the temporay MIGRATE layout has
1345                          * been removed */
1346                         CDEBUG(D_INODE, DFID" finish migration.\n",
1347                                PFID(ll_inode2fid(inode)));
1348                         lmv_free_memmd(lli->lli_lsm_md);
1349                         lli->lli_lsm_md = NULL;
1350                         RETURN(0);
1351                 } else {
1352                         /* The lustre_md from req does not include stripeEA,
1353                          * see ll_md_setattr */
1354                         RETURN(0);
1355                 }
1356         }
1357
1358         /* set the directory layout */
1359         if (lli->lli_lsm_md == NULL) {
1360                 struct cl_attr  *attr;
1361
1362                 rc = ll_init_lsm_md(inode, md);
1363                 if (rc != 0)
1364                         RETURN(rc);
1365
1366                 /* set md->lmv to NULL, so the following free lustre_md
1367                  * will not free this lsm */
1368                 md->lmv = NULL;
1369                 lli->lli_lsm_md = lsm;
1370
1371                 OBD_ALLOC_PTR(attr);
1372                 if (attr == NULL)
1373                         RETURN(-ENOMEM);
1374
1375                 /* validate the lsm */
1376                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1377                                    ll_md_blocking_ast);
1378                 if (rc != 0) {
1379                         OBD_FREE_PTR(attr);
1380                         RETURN(rc);
1381                 }
1382
1383                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1384                         md->body->mbo_nlink = attr->cat_nlink;
1385                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1386                         md->body->mbo_size = attr->cat_size;
1387                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1388                         md->body->mbo_atime = attr->cat_atime;
1389                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1390                         md->body->mbo_ctime = attr->cat_ctime;
1391                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1392                         md->body->mbo_mtime = attr->cat_mtime;
1393
1394                 OBD_FREE_PTR(attr);
1395
1396                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1397                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1398                 RETURN(0);
1399         }
1400
1401         /* Compare the old and new stripe information */
1402         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1403                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1404                 int                     idx;
1405
1406                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1407                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1408                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1409                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1410                        inode, lsm, old_lsm,
1411                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1412                        lsm->lsm_md_stripe_count,
1413                        old_lsm->lsm_md_stripe_count,
1414                        lsm->lsm_md_master_mdt_index,
1415                        old_lsm->lsm_md_master_mdt_index,
1416                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1417                        lsm->lsm_md_layout_version,
1418                        old_lsm->lsm_md_layout_version,
1419                        lsm->lsm_md_pool_name,
1420                        old_lsm->lsm_md_pool_name);
1421
1422                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1423                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1424                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1425                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1426                 }
1427
1428                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1429                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1430                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1431                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1432                 }
1433
1434                 RETURN(-EIO);
1435         }
1436
1437         RETURN(0);
1438 }
1439
1440 void ll_clear_inode(struct inode *inode)
1441 {
1442         struct ll_inode_info *lli = ll_i2info(inode);
1443         struct ll_sb_info *sbi = ll_i2sbi(inode);
1444         ENTRY;
1445
1446         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1447                PFID(ll_inode2fid(inode)), inode);
1448
1449         if (S_ISDIR(inode->i_mode)) {
1450                 /* these should have been cleared in ll_file_release */
1451                 LASSERT(lli->lli_opendir_key == NULL);
1452                 LASSERT(lli->lli_sai == NULL);
1453                 LASSERT(lli->lli_opendir_pid == 0);
1454         }
1455
1456         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1457
1458         LASSERT(!lli->lli_open_fd_write_count);
1459         LASSERT(!lli->lli_open_fd_read_count);
1460         LASSERT(!lli->lli_open_fd_exec_count);
1461
1462         if (lli->lli_mds_write_och)
1463                 ll_md_real_close(inode, FMODE_WRITE);
1464         if (lli->lli_mds_exec_och)
1465                 ll_md_real_close(inode, FMODE_EXEC);
1466         if (lli->lli_mds_read_och)
1467                 ll_md_real_close(inode, FMODE_READ);
1468
1469         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1470                 OBD_FREE(lli->lli_symlink_name,
1471                          strlen(lli->lli_symlink_name) + 1);
1472                 lli->lli_symlink_name = NULL;
1473         }
1474
1475         ll_xattr_cache_destroy(inode);
1476
1477         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1478                 LASSERT(lli->lli_posix_acl == NULL);
1479                 if (lli->lli_remote_perms) {
1480                         free_rmtperm_hash(lli->lli_remote_perms);
1481                         lli->lli_remote_perms = NULL;
1482                 }
1483         }
1484 #ifdef CONFIG_FS_POSIX_ACL
1485         else if (lli->lli_posix_acl) {
1486                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1487                 LASSERT(lli->lli_remote_perms == NULL);
1488                 posix_acl_release(lli->lli_posix_acl);
1489                 lli->lli_posix_acl = NULL;
1490         }
1491 #endif
1492         lli->lli_inode_magic = LLI_INODE_DEAD;
1493
1494         if (S_ISDIR(inode->i_mode))
1495                 ll_dir_clear_lsm_md(inode);
1496         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1497                 LASSERT(list_empty(&lli->lli_agl_list));
1498
1499         /*
1500          * XXX This has to be done before lsm is freed below, because
1501          * cl_object still uses inode lsm.
1502          */
1503         cl_inode_fini(inode);
1504
1505         EXIT;
1506 }
1507
1508 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1509 {
1510         struct lustre_md md;
1511         struct inode *inode = dentry->d_inode;
1512         struct ll_sb_info *sbi = ll_i2sbi(inode);
1513         struct ptlrpc_request *request = NULL;
1514         int rc, ia_valid;
1515         ENTRY;
1516
1517         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1518                                      LUSTRE_OPC_ANY, NULL);
1519         if (IS_ERR(op_data))
1520                 RETURN(PTR_ERR(op_data));
1521
1522         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1523         if (rc) {
1524                 ptlrpc_req_finished(request);
1525                 if (rc == -ENOENT) {
1526                         clear_nlink(inode);
1527                         /* Unlinked special device node? Or just a race?
1528                          * Pretend we done everything. */
1529                         if (!S_ISREG(inode->i_mode) &&
1530                             !S_ISDIR(inode->i_mode)) {
1531                                 ia_valid = op_data->op_attr.ia_valid;
1532                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1533                                 rc = simple_setattr(dentry, &op_data->op_attr);
1534                                 op_data->op_attr.ia_valid = ia_valid;
1535                         }
1536                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1537                         CERROR("md_setattr fails: rc = %d\n", rc);
1538                 }
1539                 RETURN(rc);
1540         }
1541
1542         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1543                               sbi->ll_md_exp, &md);
1544         if (rc) {
1545                 ptlrpc_req_finished(request);
1546                 RETURN(rc);
1547         }
1548
1549         ia_valid = op_data->op_attr.ia_valid;
1550         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1551          * cache is not cleared yet. */
1552         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1553         if (S_ISREG(inode->i_mode))
1554                 mutex_lock(&inode->i_mutex);
1555         rc = simple_setattr(dentry, &op_data->op_attr);
1556         if (S_ISREG(inode->i_mode))
1557                 mutex_unlock(&inode->i_mutex);
1558         op_data->op_attr.ia_valid = ia_valid;
1559
1560         rc = ll_update_inode(inode, &md);
1561         ptlrpc_req_finished(request);
1562
1563         RETURN(rc);
1564 }
1565
1566 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1567  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1568  * keep these values until such a time that objects are allocated for it.
1569  * We do the MDS operations first, as it is checking permissions for us.
1570  * We don't to the MDS RPC if there is nothing that we want to store there,
1571  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1572  * going to do an RPC anyways.
1573  *
1574  * If we are doing a truncate, we will send the mtime and ctime updates
1575  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1576  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1577  * at the same time.
1578  *
1579  * In case of HSMimport, we only set attr on MDS.
1580  */
1581 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1582 {
1583         struct inode *inode = dentry->d_inode;
1584         struct ll_inode_info *lli = ll_i2info(inode);
1585         struct md_op_data *op_data = NULL;
1586         int rc = 0;
1587         ENTRY;
1588
1589         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1590                "valid %x, hsm_import %d\n",
1591                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1592                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1593                hsm_import);
1594
1595         if (attr->ia_valid & ATTR_SIZE) {
1596                 /* Check new size against VFS/VM file size limit and rlimit */
1597                 rc = inode_newsize_ok(inode, attr->ia_size);
1598                 if (rc)
1599                         RETURN(rc);
1600
1601                 /* The maximum Lustre file size is variable, based on the
1602                  * OST maximum object size and number of stripes.  This
1603                  * needs another check in addition to the VFS check above. */
1604                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1605                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1606                                PFID(&lli->lli_fid), attr->ia_size,
1607                                ll_file_maxbytes(inode));
1608                         RETURN(-EFBIG);
1609                 }
1610
1611                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1612         }
1613
1614         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1615         if (attr->ia_valid & TIMES_SET_FLAGS) {
1616                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1617                     !cfs_capable(CFS_CAP_FOWNER))
1618                         RETURN(-EPERM);
1619         }
1620
1621         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1622         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1623             (attr->ia_valid & ATTR_CTIME)) {
1624                 attr->ia_ctime = CFS_CURRENT_TIME;
1625                 attr->ia_valid |= ATTR_CTIME_SET;
1626         }
1627         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1628             (attr->ia_valid & ATTR_ATIME)) {
1629                 attr->ia_atime = CFS_CURRENT_TIME;
1630                 attr->ia_valid |= ATTR_ATIME_SET;
1631         }
1632         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1633             (attr->ia_valid & ATTR_MTIME)) {
1634                 attr->ia_mtime = CFS_CURRENT_TIME;
1635                 attr->ia_valid |= ATTR_MTIME_SET;
1636         }
1637
1638         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1639                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1640                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1641                        cfs_time_current_sec());
1642
1643         if (S_ISREG(inode->i_mode)) {
1644                 if (attr->ia_valid & ATTR_SIZE)
1645                         inode_dio_write_done(inode);
1646                 mutex_unlock(&inode->i_mutex);
1647         }
1648
1649         /* We always do an MDS RPC, even if we're only changing the size;
1650          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1651
1652         OBD_ALLOC_PTR(op_data);
1653         if (op_data == NULL)
1654                 GOTO(out, rc = -ENOMEM);
1655
1656         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1657                 /* If we are changing file size, file content is
1658                  * modified, flag it. */
1659                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1660                 op_data->op_bias |= MDS_DATA_MODIFIED;
1661                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1662         }
1663
1664         op_data->op_attr = *attr;
1665
1666         rc = ll_md_setattr(dentry, op_data);
1667         if (rc)
1668                 GOTO(out, rc);
1669
1670         if (!S_ISREG(inode->i_mode) || hsm_import)
1671                 GOTO(out, rc = 0);
1672
1673         if (attr->ia_valid & (ATTR_SIZE |
1674                               ATTR_ATIME | ATTR_ATIME_SET |
1675                               ATTR_MTIME | ATTR_MTIME_SET |
1676                               ATTR_CTIME | ATTR_CTIME_SET)) {
1677                 /* For truncate and utimes sending attributes to OSTs, setting
1678                  * mtime/atime to the past will be performed under PW [0:EOF]
1679                  * extent lock (new_size:EOF for truncate).  It may seem
1680                  * excessive to send mtime/atime updates to OSTs when not
1681                  * setting times to past, but it is necessary due to possible
1682                  * time de-synchronization between MDT inode and OST objects */
1683                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1684         }
1685
1686         /* If the file was restored, it needs to set dirty flag.
1687          *
1688          * We've already sent MDS_DATA_MODIFIED flag in
1689          * ll_md_setattr() for truncate. However, the MDT refuses to
1690          * set the HS_DIRTY flag on released files, so we have to set
1691          * it again if the file has been restored. Please check how
1692          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1693          *
1694          * Please notice that if the file is not released, the previous
1695          * MDS_DATA_MODIFIED has taken effect and usually
1696          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1697          * This way we can save an RPC for common open + trunc
1698          * operation. */
1699         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1700                 struct hsm_state_set hss = {
1701                         .hss_valid = HSS_SETMASK,
1702                         .hss_setmask = HS_DIRTY,
1703                 };
1704                 int rc2;
1705
1706                 rc2 = ll_hsm_state_set(inode, &hss);
1707                 /* truncate and write can happen at the same time, so that
1708                  * the file can be set modified even though the file is not
1709                  * restored from released state, and ll_hsm_state_set() is
1710                  * not applicable for the file, and rc2 < 0 is normal in this
1711                  * case. */
1712                 if (rc2 < 0)
1713                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1714                                PFID(ll_inode2fid(inode)), rc2);
1715         }
1716
1717         EXIT;
1718 out:
1719         if (op_data != NULL)
1720                 ll_finish_md_op_data(op_data);
1721
1722         if (S_ISREG(inode->i_mode)) {
1723                 mutex_lock(&inode->i_mutex);
1724                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1725                         inode_dio_wait(inode);
1726         }
1727
1728         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1729                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1730
1731         return rc;
1732 }
1733
1734 int ll_setattr(struct dentry *de, struct iattr *attr)
1735 {
1736         int mode = de->d_inode->i_mode;
1737
1738         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1739                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1740                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1741
1742         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1743                                (ATTR_SIZE|ATTR_MODE)) &&
1744             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1745              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1746               !(attr->ia_mode & S_ISGID))))
1747                 attr->ia_valid |= ATTR_FORCE;
1748
1749         if ((attr->ia_valid & ATTR_MODE) &&
1750             (mode & S_ISUID) &&
1751             !(attr->ia_mode & S_ISUID) &&
1752             !(attr->ia_valid & ATTR_KILL_SUID))
1753                 attr->ia_valid |= ATTR_KILL_SUID;
1754
1755         if ((attr->ia_valid & ATTR_MODE) &&
1756             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1757             !(attr->ia_mode & S_ISGID) &&
1758             !(attr->ia_valid & ATTR_KILL_SGID))
1759                 attr->ia_valid |= ATTR_KILL_SGID;
1760
1761         return ll_setattr_raw(de, attr, false);
1762 }
1763
1764 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1765                        __u64 max_age, __u32 flags)
1766 {
1767         struct ll_sb_info *sbi = ll_s2sbi(sb);
1768         struct obd_statfs obd_osfs;
1769         int rc;
1770         ENTRY;
1771
1772         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1773         if (rc) {
1774                 CERROR("md_statfs fails: rc = %d\n", rc);
1775                 RETURN(rc);
1776         }
1777
1778         osfs->os_type = sb->s_magic;
1779
1780         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1781                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1782
1783         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1784                 flags |= OBD_STATFS_NODELAY;
1785
1786         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1787         if (rc) {
1788                 CERROR("obd_statfs fails: rc = %d\n", rc);
1789                 RETURN(rc);
1790         }
1791
1792         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1793                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1794                obd_osfs.os_files);
1795
1796         osfs->os_bsize = obd_osfs.os_bsize;
1797         osfs->os_blocks = obd_osfs.os_blocks;
1798         osfs->os_bfree = obd_osfs.os_bfree;
1799         osfs->os_bavail = obd_osfs.os_bavail;
1800
1801         /* If we don't have as many objects free on the OST as inodes
1802          * on the MDS, we reduce the total number of inodes to
1803          * compensate, so that the "inodes in use" number is correct.
1804          */
1805         if (obd_osfs.os_ffree < osfs->os_ffree) {
1806                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1807                         obd_osfs.os_ffree;
1808                 osfs->os_ffree = obd_osfs.os_ffree;
1809         }
1810
1811         RETURN(rc);
1812 }
1813 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1814 {
1815         struct super_block *sb = de->d_sb;
1816         struct obd_statfs osfs;
1817         __u64 fsid = huge_encode_dev(sb->s_dev);
1818         int rc;
1819
1820         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1821         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1822
1823         /* Some amount of caching on the client is allowed */
1824         rc = ll_statfs_internal(sb, &osfs,
1825                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1826                                 0);
1827         if (rc)
1828                 return rc;
1829
1830         statfs_unpack(sfs, &osfs);
1831
1832         /* We need to downshift for all 32-bit kernels, because we can't
1833          * tell if the kernel is being called via sys_statfs64() or not.
1834          * Stop before overflowing f_bsize - in which case it is better
1835          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1836         if (sizeof(long) < 8) {
1837                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1838                         sfs->f_bsize <<= 1;
1839
1840                         osfs.os_blocks >>= 1;
1841                         osfs.os_bfree >>= 1;
1842                         osfs.os_bavail >>= 1;
1843                 }
1844         }
1845
1846         sfs->f_blocks = osfs.os_blocks;
1847         sfs->f_bfree = osfs.os_bfree;
1848         sfs->f_bavail = osfs.os_bavail;
1849         sfs->f_fsid.val[0] = (__u32)fsid;
1850         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1851         return 0;
1852 }
1853
1854 void ll_inode_size_lock(struct inode *inode)
1855 {
1856         struct ll_inode_info *lli;
1857
1858         LASSERT(!S_ISDIR(inode->i_mode));
1859
1860         lli = ll_i2info(inode);
1861         mutex_lock(&lli->lli_size_mutex);
1862 }
1863
1864 void ll_inode_size_unlock(struct inode *inode)
1865 {
1866         struct ll_inode_info *lli;
1867
1868         lli = ll_i2info(inode);
1869         mutex_unlock(&lli->lli_size_mutex);
1870 }
1871
1872 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1873 {
1874         struct ll_inode_info *lli = ll_i2info(inode);
1875         struct mdt_body *body = md->body;
1876         struct ll_sb_info *sbi = ll_i2sbi(inode);
1877
1878         if (body->mbo_valid & OBD_MD_FLEASIZE)
1879                 cl_file_inode_init(inode, md);
1880
1881         if (S_ISDIR(inode->i_mode)) {
1882                 int     rc;
1883
1884                 rc = ll_update_lsm_md(inode, md);
1885                 if (rc != 0)
1886                         return rc;
1887         }
1888
1889         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1890                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1891                         ll_update_remote_perm(inode, md->remote_perm);
1892         }
1893 #ifdef CONFIG_FS_POSIX_ACL
1894         else if (body->mbo_valid & OBD_MD_FLACL) {
1895                 spin_lock(&lli->lli_lock);
1896                 if (lli->lli_posix_acl)
1897                         posix_acl_release(lli->lli_posix_acl);
1898                 lli->lli_posix_acl = md->posix_acl;
1899                 spin_unlock(&lli->lli_lock);
1900         }
1901 #endif
1902         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1903                                         sbi->ll_flags & LL_SBI_32BIT_API);
1904         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1905
1906         if (body->mbo_valid & OBD_MD_FLATIME) {
1907                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1908                         LTIME_S(inode->i_atime) = body->mbo_atime;
1909                 lli->lli_atime = body->mbo_atime;
1910         }
1911
1912         if (body->mbo_valid & OBD_MD_FLMTIME) {
1913                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1914                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1915                                "to "LPU64"\n", inode->i_ino,
1916                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1917                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1918                 }
1919                 lli->lli_mtime = body->mbo_mtime;
1920         }
1921
1922         if (body->mbo_valid & OBD_MD_FLCTIME) {
1923                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1924                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1925                 lli->lli_ctime = body->mbo_ctime;
1926         }
1927
1928         if (body->mbo_valid & OBD_MD_FLMODE)
1929                 inode->i_mode = (inode->i_mode & S_IFMT) |
1930                                 (body->mbo_mode & ~S_IFMT);
1931
1932         if (body->mbo_valid & OBD_MD_FLTYPE)
1933                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1934                                 (body->mbo_mode & S_IFMT);
1935
1936         LASSERT(inode->i_mode != 0);
1937         if (S_ISREG(inode->i_mode))
1938                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1939                                        LL_MAX_BLKSIZE_BITS);
1940         else
1941                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1942
1943         if (body->mbo_valid & OBD_MD_FLUID)
1944                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1945         if (body->mbo_valid & OBD_MD_FLGID)
1946                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1947         if (body->mbo_valid & OBD_MD_FLFLAGS)
1948                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1949         if (body->mbo_valid & OBD_MD_FLNLINK)
1950                 set_nlink(inode, body->mbo_nlink);
1951         if (body->mbo_valid & OBD_MD_FLRDEV)
1952                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1953
1954         if (body->mbo_valid & OBD_MD_FLID) {
1955                 /* FID shouldn't be changed! */
1956                 if (fid_is_sane(&lli->lli_fid)) {
1957                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1958                                  "Trying to change FID "DFID
1959                                  " to the "DFID", inode "DFID"(%p)\n",
1960                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1961                                  PFID(ll_inode2fid(inode)), inode);
1962                 } else {
1963                         lli->lli_fid = body->mbo_fid1;
1964                 }
1965         }
1966
1967         LASSERT(fid_seq(&lli->lli_fid) != 0);
1968
1969         if (body->mbo_valid & OBD_MD_FLSIZE) {
1970                 i_size_write(inode, body->mbo_size);
1971
1972                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1973                        PFID(ll_inode2fid(inode)),
1974                        (unsigned long long)body->mbo_size);
1975
1976                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1977                         inode->i_blocks = body->mbo_blocks;
1978         }
1979
1980         if (body->mbo_valid & OBD_MD_TSTATE) {
1981                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1982                  * clear it when done to ensure to start again
1983                  * glimpsing updated attrs
1984                  */
1985                 if (body->mbo_t_state & MS_RESTORE)
1986                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1987                 else
1988                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1989         }
1990
1991         return 0;
1992 }
1993
1994 int ll_read_inode2(struct inode *inode, void *opaque)
1995 {
1996         struct lustre_md *md = opaque;
1997         struct ll_inode_info *lli = ll_i2info(inode);
1998         int     rc;
1999         ENTRY;
2000
2001         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2002                PFID(&lli->lli_fid), inode);
2003
2004         /* Core attributes from the MDS first.  This is a new inode, and
2005          * the VFS doesn't zero times in the core inode so we have to do
2006          * it ourselves.  They will be overwritten by either MDS or OST
2007          * attributes - we just need to make sure they aren't newer. */
2008         LTIME_S(inode->i_mtime) = 0;
2009         LTIME_S(inode->i_atime) = 0;
2010         LTIME_S(inode->i_ctime) = 0;
2011         inode->i_rdev = 0;
2012         rc = ll_update_inode(inode, md);
2013         if (rc != 0)
2014                 RETURN(rc);
2015
2016         /* OIDEBUG(inode); */
2017
2018 #ifdef HAVE_BACKING_DEV_INFO
2019         /* initializing backing dev info. */
2020         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2021 #endif
2022         if (S_ISREG(inode->i_mode)) {
2023                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2024                 inode->i_op = &ll_file_inode_operations;
2025                 inode->i_fop = sbi->ll_fop;
2026                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2027                 EXIT;
2028         } else if (S_ISDIR(inode->i_mode)) {
2029                 inode->i_op = &ll_dir_inode_operations;
2030                 inode->i_fop = &ll_dir_operations;
2031                 EXIT;
2032         } else if (S_ISLNK(inode->i_mode)) {
2033                 inode->i_op = &ll_fast_symlink_inode_operations;
2034                 EXIT;
2035         } else {
2036                 inode->i_op = &ll_special_inode_operations;
2037
2038                 init_special_inode(inode, inode->i_mode,
2039                                    inode->i_rdev);
2040
2041                 EXIT;
2042         }
2043
2044         return 0;
2045 }
2046
2047 void ll_delete_inode(struct inode *inode)
2048 {
2049         struct ll_inode_info *lli = ll_i2info(inode);
2050         ENTRY;
2051
2052         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2053                 /* It is last chance to write out dirty pages,
2054                  * otherwise we may lose data while umount */
2055                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2056
2057         truncate_inode_pages_final(&inode->i_data);
2058
2059         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2060                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2061                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2062
2063 #ifdef HAVE_SBOPS_EVICT_INODE
2064         ll_clear_inode(inode);
2065 #endif
2066         clear_inode(inode);
2067
2068         EXIT;
2069 }
2070
2071 int ll_iocontrol(struct inode *inode, struct file *file,
2072                  unsigned int cmd, unsigned long arg)
2073 {
2074         struct ll_sb_info *sbi = ll_i2sbi(inode);
2075         struct ptlrpc_request *req = NULL;
2076         int rc, flags = 0;
2077         ENTRY;
2078
2079         switch(cmd) {
2080         case FSFILT_IOC_GETFLAGS: {
2081                 struct mdt_body *body;
2082                 struct md_op_data *op_data;
2083
2084                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2085                                              0, 0, LUSTRE_OPC_ANY,
2086                                              NULL);
2087                 if (IS_ERR(op_data))
2088                         RETURN(PTR_ERR(op_data));
2089
2090                 op_data->op_valid = OBD_MD_FLFLAGS;
2091                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2092                 ll_finish_md_op_data(op_data);
2093                 if (rc) {
2094                         CERROR("%s: failure inode "DFID": rc = %d\n",
2095                                sbi->ll_md_exp->exp_obd->obd_name,
2096                                PFID(ll_inode2fid(inode)), rc);
2097                         RETURN(-abs(rc));
2098                 }
2099
2100                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2101
2102                 flags = body->mbo_flags;
2103
2104                 ptlrpc_req_finished(req);
2105
2106                 RETURN(put_user(flags, (int __user *)arg));
2107         }
2108         case FSFILT_IOC_SETFLAGS: {
2109                 struct iattr *attr;
2110                 struct md_op_data *op_data;
2111                 struct cl_object *obj;
2112
2113                 if (get_user(flags, (int __user *)arg))
2114                         RETURN(-EFAULT);
2115
2116                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2117                                              LUSTRE_OPC_ANY, NULL);
2118                 if (IS_ERR(op_data))
2119                         RETURN(PTR_ERR(op_data));
2120
2121                 op_data->op_attr_flags = flags;
2122                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2123                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2124                 ll_finish_md_op_data(op_data);
2125                 ptlrpc_req_finished(req);
2126                 if (rc)
2127                         RETURN(rc);
2128
2129                 inode->i_flags = ll_ext_to_inode_flags(flags);
2130
2131                 obj = ll_i2info(inode)->lli_clob;
2132                 if (obj == NULL)
2133                         RETURN(0);
2134
2135                 OBD_ALLOC_PTR(attr);
2136                 if (attr == NULL)
2137                         RETURN(-ENOMEM);
2138
2139                 attr->ia_valid = ATTR_ATTR_FLAG;
2140                 rc = cl_setattr_ost(obj, attr, flags);
2141
2142                 OBD_FREE_PTR(attr);
2143                 RETURN(rc);
2144         }
2145         default:
2146                 RETURN(-ENOSYS);
2147         }
2148
2149         RETURN(0);
2150 }
2151
2152 int ll_flush_ctx(struct inode *inode)
2153 {
2154         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2155
2156         CDEBUG(D_SEC, "flush context for user %d\n",
2157                from_kuid(&init_user_ns, current_uid()));
2158
2159         obd_set_info_async(NULL, sbi->ll_md_exp,
2160                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2161                            0, NULL, NULL);
2162         obd_set_info_async(NULL, sbi->ll_dt_exp,
2163                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2164                            0, NULL, NULL);
2165         return 0;
2166 }
2167
2168 /* umount -f client means force down, don't save state */
2169 void ll_umount_begin(struct super_block *sb)
2170 {
2171         struct ll_sb_info *sbi = ll_s2sbi(sb);
2172         struct obd_device *obd;
2173         struct obd_ioctl_data *ioc_data;
2174         ENTRY;
2175
2176         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2177                sb->s_count, atomic_read(&sb->s_active));
2178
2179         obd = class_exp2obd(sbi->ll_md_exp);
2180         if (obd == NULL) {
2181                 CERROR("Invalid MDC connection handle "LPX64"\n",
2182                        sbi->ll_md_exp->exp_handle.h_cookie);
2183                 EXIT;
2184                 return;
2185         }
2186         obd->obd_force = 1;
2187
2188         obd = class_exp2obd(sbi->ll_dt_exp);
2189         if (obd == NULL) {
2190                 CERROR("Invalid LOV connection handle "LPX64"\n",
2191                        sbi->ll_dt_exp->exp_handle.h_cookie);
2192                 EXIT;
2193                 return;
2194         }
2195         obd->obd_force = 1;
2196
2197         OBD_ALLOC_PTR(ioc_data);
2198         if (ioc_data) {
2199                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2200                               sizeof *ioc_data, ioc_data, NULL);
2201
2202                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2203                               sizeof *ioc_data, ioc_data, NULL);
2204
2205                 OBD_FREE_PTR(ioc_data);
2206         }
2207
2208         /* Really, we'd like to wait until there are no requests outstanding,
2209          * and then continue.  For now, we just invalidate the requests,
2210          * schedule() and sleep one second if needed, and hope.
2211          */
2212         schedule();
2213         EXIT;
2214 }
2215
2216 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2217 {
2218         struct ll_sb_info *sbi = ll_s2sbi(sb);
2219         char *profilenm = get_profile_name(sb);
2220         int err;
2221         __u32 read_only;
2222
2223         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2224                 read_only = *flags & MS_RDONLY;
2225                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2226                                          sizeof(KEY_READ_ONLY),
2227                                          KEY_READ_ONLY, sizeof(read_only),
2228                                          &read_only, NULL);
2229                 if (err) {
2230                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2231                                       profilenm, read_only ?
2232                                       "read-only" : "read-write", err);
2233                         return err;
2234                 }
2235
2236                 if (read_only)
2237                         sb->s_flags |= MS_RDONLY;
2238                 else
2239                         sb->s_flags &= ~MS_RDONLY;
2240
2241                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2242                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2243                                       read_only ?  "read-only" : "read-write");
2244         }
2245         return 0;
2246 }
2247
2248 /**
2249  * Cleanup the open handle that is cached on MDT-side.
2250  *
2251  * For open case, the client side open handling thread may hit error
2252  * after the MDT grant the open. Under such case, the client should
2253  * send close RPC to the MDT as cleanup; otherwise, the open handle
2254  * on the MDT will be leaked there until the client umount or evicted.
2255  *
2256  * In further, if someone unlinked the file, because the open handle
2257  * holds the reference on such file/object, then it will block the
2258  * subsequent threads that want to locate such object via FID.
2259  *
2260  * \param[in] sb        super block for this file-system
2261  * \param[in] open_req  pointer to the original open request
2262  */
2263 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2264 {
2265         struct mdt_body                 *body;
2266         struct md_op_data               *op_data;
2267         struct ptlrpc_request           *close_req = NULL;
2268         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2269         ENTRY;
2270
2271         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2272         OBD_ALLOC_PTR(op_data);
2273         if (op_data == NULL) {
2274                 CWARN("%s: cannot allocate op_data to release open handle for "
2275                       DFID"\n",
2276                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2277
2278                 RETURN_EXIT;
2279         }
2280
2281         op_data->op_fid1 = body->mbo_fid1;
2282         op_data->op_handle = body->mbo_handle;
2283         op_data->op_mod_time = cfs_time_current_sec();
2284         md_close(exp, op_data, NULL, &close_req);
2285         ptlrpc_req_finished(close_req);
2286         ll_finish_md_op_data(op_data);
2287
2288         EXIT;
2289 }
2290
2291 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2292                   struct super_block *sb, struct lookup_intent *it)
2293 {
2294         struct ll_sb_info *sbi = NULL;
2295         struct lustre_md md = { NULL };
2296         int rc;
2297         ENTRY;
2298
2299         LASSERT(*inode || sb);
2300         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2301         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2302                               sbi->ll_md_exp, &md);
2303         if (rc != 0)
2304                 GOTO(cleanup, rc);
2305
2306         if (*inode) {
2307                 rc = ll_update_inode(*inode, &md);
2308                 if (rc != 0)
2309                         GOTO(out, rc);
2310         } else {
2311                 LASSERT(sb != NULL);
2312
2313                 /*
2314                  * At this point server returns to client's same fid as client
2315                  * generated for creating. So using ->fid1 is okay here.
2316                  */
2317                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2318
2319                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2320                                              sbi->ll_flags & LL_SBI_32BIT_API),
2321                                  &md);
2322                 if (IS_ERR(*inode)) {
2323 #ifdef CONFIG_FS_POSIX_ACL
2324                         if (md.posix_acl) {
2325                                 posix_acl_release(md.posix_acl);
2326                                 md.posix_acl = NULL;
2327                         }
2328 #endif
2329                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2330                         *inode = NULL;
2331                         CERROR("new_inode -fatal: rc %d\n", rc);
2332                         GOTO(out, rc);
2333                 }
2334         }
2335
2336         /* Handling piggyback layout lock.
2337          * Layout lock can be piggybacked by getattr and open request.
2338          * The lsm can be applied to inode only if it comes with a layout lock
2339          * otherwise correct layout may be overwritten, for example:
2340          * 1. proc1: mdt returns a lsm but not granting layout
2341          * 2. layout was changed by another client
2342          * 3. proc2: refresh layout and layout lock granted
2343          * 4. proc1: to apply a stale layout */
2344         if (it != NULL && it->it_lock_mode != 0) {
2345                 struct lustre_handle lockh;
2346                 struct ldlm_lock *lock;
2347
2348                 lockh.cookie = it->it_lock_handle;
2349                 lock = ldlm_handle2lock(&lockh);
2350                 LASSERT(lock != NULL);
2351                 if (ldlm_has_layout(lock)) {
2352                         struct cl_object_conf conf;
2353
2354                         memset(&conf, 0, sizeof(conf));
2355                         conf.coc_opc = OBJECT_CONF_SET;
2356                         conf.coc_inode = *inode;
2357                         conf.coc_lock = lock;
2358                         conf.u.coc_layout = md.layout;
2359                         (void)ll_layout_conf(*inode, &conf);
2360                 }
2361                 LDLM_LOCK_PUT(lock);
2362         }
2363
2364         GOTO(out, rc = 0);
2365
2366 out:
2367         md_free_lustre_md(sbi->ll_md_exp, &md);
2368
2369 cleanup:
2370         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2371                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2372
2373         return rc;
2374 }
2375
2376 int ll_obd_statfs(struct inode *inode, void __user *arg)
2377 {
2378         struct ll_sb_info *sbi = NULL;
2379         struct obd_export *exp;
2380         char *buf = NULL;
2381         struct obd_ioctl_data *data = NULL;
2382         __u32 type;
2383         int len = 0, rc;
2384
2385         if (!inode || !(sbi = ll_i2sbi(inode)))
2386                 GOTO(out_statfs, rc = -EINVAL);
2387
2388         rc = obd_ioctl_getdata(&buf, &len, arg);
2389         if (rc)
2390                 GOTO(out_statfs, rc);
2391
2392         data = (void*)buf;
2393         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2394             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2395                 GOTO(out_statfs, rc = -EINVAL);
2396
2397         if (data->ioc_inllen1 != sizeof(__u32) ||
2398             data->ioc_inllen2 != sizeof(__u32) ||
2399             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2400             data->ioc_plen2 != sizeof(struct obd_uuid))
2401                 GOTO(out_statfs, rc = -EINVAL);
2402
2403         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2404         if (type & LL_STATFS_LMV)
2405                 exp = sbi->ll_md_exp;
2406         else if (type & LL_STATFS_LOV)
2407                 exp = sbi->ll_dt_exp;
2408         else
2409                 GOTO(out_statfs, rc = -ENODEV);
2410
2411         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2412         if (rc)
2413                 GOTO(out_statfs, rc);
2414 out_statfs:
2415         if (buf)
2416                 obd_ioctl_freedata(buf, len);
2417         return rc;
2418 }
2419
2420 int ll_process_config(struct lustre_cfg *lcfg)
2421 {
2422         struct super_block *sb;
2423         unsigned long x;
2424         int rc = 0;
2425         char *ptr;
2426
2427         /* The instance name contains the sb: lustre-client-aacfe000 */
2428         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2429         if (!ptr || !*(++ptr))
2430                 return -EINVAL;
2431         if (sscanf(ptr, "%lx", &x) != 1)
2432                 return -EINVAL;
2433         sb = (struct super_block *)x;
2434         /* This better be a real Lustre superblock! */
2435         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2436
2437         /* Note we have not called client_common_fill_super yet, so
2438            proc fns must be able to handle that! */
2439         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2440                                       lcfg, sb);
2441         if (rc > 0)
2442                 rc = 0;
2443         return rc;
2444 }
2445
2446 /* this function prepares md_op_data hint for passing it down to MD stack. */
2447 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2448                                       struct inode *i1, struct inode *i2,
2449                                       const char *name, size_t namelen,
2450                                       __u32 mode, __u32 opc, void *data)
2451 {
2452         LASSERT(i1 != NULL);
2453
2454         if (name == NULL) {
2455                 /* Do not reuse namelen for something else. */
2456                 if (namelen != 0)
2457                         return ERR_PTR(-EINVAL);
2458         } else {
2459                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2460                         return ERR_PTR(-ENAMETOOLONG);
2461
2462                 if (!lu_name_is_valid_2(name, namelen))
2463                         return ERR_PTR(-EINVAL);
2464         }
2465
2466         if (op_data == NULL)
2467                 OBD_ALLOC_PTR(op_data);
2468
2469         if (op_data == NULL)
2470                 return ERR_PTR(-ENOMEM);
2471
2472         ll_i2gids(op_data->op_suppgids, i1, i2);
2473         op_data->op_fid1 = *ll_inode2fid(i1);
2474         op_data->op_default_stripe_offset = -1;
2475         if (S_ISDIR(i1->i_mode)) {
2476                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2477                 if (opc == LUSTRE_OPC_MKDIR)
2478                         op_data->op_default_stripe_offset =
2479                                    ll_i2info(i1)->lli_def_stripe_offset;
2480         }
2481
2482         if (i2) {
2483                 op_data->op_fid2 = *ll_inode2fid(i2);
2484                 if (S_ISDIR(i2->i_mode))
2485                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2486         } else {
2487                 fid_zero(&op_data->op_fid2);
2488         }
2489
2490         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2491                 op_data->op_cli_flags |= CLI_HASH64;
2492
2493         if (ll_need_32bit_api(ll_i2sbi(i1)))
2494                 op_data->op_cli_flags |= CLI_API32;
2495
2496         op_data->op_name = name;
2497         op_data->op_namelen = namelen;
2498         op_data->op_mode = mode;
2499         op_data->op_mod_time = cfs_time_current_sec();
2500         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2501         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2502         op_data->op_cap = cfs_curproc_cap_pack();
2503         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2504              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2505                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2506         } else {
2507                 op_data->op_mds = 0;
2508         }
2509         op_data->op_data = data;
2510
2511         return op_data;
2512 }
2513
2514 void ll_finish_md_op_data(struct md_op_data *op_data)
2515 {
2516         OBD_FREE_PTR(op_data);
2517 }
2518
2519 #ifdef HAVE_SUPEROPS_USE_DENTRY
2520 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2521 #else
2522 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2523 #endif
2524 {
2525         struct ll_sb_info *sbi;
2526
2527 #ifdef HAVE_SUPEROPS_USE_DENTRY
2528         LASSERT((seq != NULL) && (dentry != NULL));
2529         sbi = ll_s2sbi(dentry->d_sb);
2530 #else
2531         LASSERT((seq != NULL) && (vfs != NULL));
2532         sbi = ll_s2sbi(vfs->mnt_sb);
2533 #endif
2534
2535         if (sbi->ll_flags & LL_SBI_NOLCK)
2536                 seq_puts(seq, ",nolock");
2537
2538         if (sbi->ll_flags & LL_SBI_FLOCK)
2539                 seq_puts(seq, ",flock");
2540
2541         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2542                 seq_puts(seq, ",localflock");
2543
2544         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2545                 seq_puts(seq, ",user_xattr");
2546
2547         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2548                 seq_puts(seq, ",lazystatfs");
2549
2550         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2551                 seq_puts(seq, ",user_fid2path");
2552
2553         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2554                 seq_puts(seq, ",always_ping");
2555
2556         RETURN(0);
2557 }
2558
2559 /**
2560  * Get obd name by cmd, and copy out to user space
2561  */
2562 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2563 {
2564         struct ll_sb_info *sbi = ll_i2sbi(inode);
2565         struct obd_device *obd;
2566         ENTRY;
2567
2568         if (cmd == OBD_IOC_GETDTNAME)
2569                 obd = class_exp2obd(sbi->ll_dt_exp);
2570         else if (cmd == OBD_IOC_GETMDNAME)
2571                 obd = class_exp2obd(sbi->ll_md_exp);
2572         else
2573                 RETURN(-EINVAL);
2574
2575         if (!obd)
2576                 RETURN(-ENOENT);
2577
2578         if (copy_to_user((void __user *)arg, obd->obd_name,
2579                          strlen(obd->obd_name) + 1))
2580                 RETURN(-EFAULT);
2581
2582         RETURN(0);
2583 }
2584
2585 /**
2586  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2587  * fsname will be returned in this buffer; otherwise, a static buffer will be
2588  * used to store the fsname and returned to caller.
2589  */
2590 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2591 {
2592         static char fsname_static[MTI_NAME_MAXLEN];
2593         struct lustre_sb_info *lsi = s2lsi(sb);
2594         char *ptr;
2595         int len;
2596
2597         if (buf == NULL) {
2598                 /* this means the caller wants to use static buffer
2599                  * and it doesn't care about race. Usually this is
2600                  * in error reporting path */
2601                 buf = fsname_static;
2602                 buflen = sizeof(fsname_static);
2603         }
2604
2605         len = strlen(lsi->lsi_lmd->lmd_profile);
2606         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2607         if (ptr && (strcmp(ptr, "-client") == 0))
2608                 len -= 7;
2609
2610         if (unlikely(len >= buflen))
2611                 len = buflen - 1;
2612         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2613         buf[len] = '\0';
2614
2615         return buf;
2616 }
2617
2618 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2619 {
2620         char *path = NULL;
2621
2622         struct path p;
2623
2624         p.dentry = dentry;
2625         p.mnt = current->fs->root.mnt;
2626         path_get(&p);
2627         path = d_path(&p, buf, bufsize);
2628         path_put(&p);
2629         return path;
2630 }
2631
2632 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2633 {
2634         char *buf, *path = NULL;
2635         struct dentry *dentry = NULL;
2636         struct inode *inode = page->mapping->host;
2637
2638         /* this can be called inside spin lock so use GFP_ATOMIC. */
2639         buf = (char *)__get_free_page(GFP_ATOMIC);
2640         if (buf != NULL) {
2641                 dentry = d_find_alias(page->mapping->host);
2642                 if (dentry != NULL)
2643                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2644         }
2645
2646         CDEBUG(D_WARNING,
2647                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2648                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2649                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2650                PFID(ll_inode2fid(inode)),
2651                (path && !IS_ERR(path)) ? path : "", ioret);
2652
2653         if (dentry != NULL)
2654                 dput(dentry);
2655
2656         if (buf != NULL)
2657                 free_page((unsigned long)buf);
2658 }
2659
2660 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2661                         struct lov_user_md **kbuf)
2662 {
2663         struct lov_user_md      lum;
2664         ssize_t                 lum_size;
2665         ENTRY;
2666
2667         if (copy_from_user(&lum, md, sizeof(lum)))
2668                 RETURN(-EFAULT);
2669
2670         lum_size = ll_lov_user_md_size(&lum);
2671         if (lum_size < 0)
2672                 RETURN(lum_size);
2673
2674         OBD_ALLOC(*kbuf, lum_size);
2675         if (*kbuf == NULL)
2676                 RETURN(-ENOMEM);
2677
2678         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2679                 OBD_FREE(*kbuf, lum_size);
2680                 RETURN(-EFAULT);
2681         }
2682
2683         RETURN(lum_size);
2684 }
2685
2686 /*
2687  * Compute llite root squash state after a change of root squash
2688  * configuration setting or add/remove of a lnet nid
2689  */
2690 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2691 {
2692         struct root_squash_info *squash = &sbi->ll_squash;
2693         int i;
2694         bool matched;
2695         lnet_process_id_t id;
2696
2697         /* Update norootsquash flag */
2698         down_write(&squash->rsi_sem);
2699         if (list_empty(&squash->rsi_nosquash_nids))
2700                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2701         else {
2702                 /* Do not apply root squash as soon as one of our NIDs is
2703                  * in the nosquash_nids list */
2704                 matched = false;
2705                 i = 0;
2706                 while (LNetGetId(i++, &id) != -ENOENT) {
2707                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2708                                 continue;
2709                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2710                                 matched = true;
2711                                 break;
2712                         }
2713                 }
2714                 if (matched)
2715                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2716                 else
2717                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2718         }
2719         up_write(&squash->rsi_sem);
2720 }
2721
2722 /**
2723  * Parse linkea content to extract information about a given hardlink
2724  *
2725  * \param[in]   ldata      - Initialized linkea data
2726  * \param[in]   linkno     - Link identifier
2727  * \param[out]  parent_fid - The entry's parent FID
2728  * \param[out]  ln         - Entry name destination buffer
2729  *
2730  * \retval 0 on success
2731  * \retval Appropriate negative error code on failure
2732  */
2733 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2734                             struct lu_fid *parent_fid, struct lu_name *ln)
2735 {
2736         unsigned int    idx;
2737         int             rc;
2738         ENTRY;
2739
2740         rc = linkea_init(ldata);
2741         if (rc < 0)
2742                 RETURN(rc);
2743
2744         if (linkno >= ldata->ld_leh->leh_reccount)
2745                 /* beyond last link */
2746                 RETURN(-ENODATA);
2747
2748         linkea_first_entry(ldata);
2749         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2750                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2751                                     parent_fid);
2752                 if (idx == linkno)
2753                         break;
2754
2755                 linkea_next_entry(ldata);
2756         }
2757
2758         if (idx < linkno)
2759                 RETURN(-ENODATA);
2760
2761         RETURN(0);
2762 }
2763
2764 /**
2765  * Get parent FID and name of an identified link. Operation is performed for
2766  * a given link number, letting the caller iterate over linkno to list one or
2767  * all links of an entry.
2768  *
2769  * \param[in]     file - File descriptor against which to perform the operation
2770  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2771  *                       on and the available size. It is eventually filled with
2772  *                       the requested information or left untouched on error
2773  *
2774  * \retval - 0 on success
2775  * \retval - Appropriate negative error code on failure
2776  */
2777 int ll_getparent(struct file *file, struct getparent __user *arg)
2778 {
2779         struct dentry           *dentry = file->f_path.dentry;
2780         struct inode            *inode = dentry->d_inode;
2781         struct linkea_data      *ldata;
2782         struct lu_buf            buf = LU_BUF_NULL;
2783         struct lu_name           ln;
2784         struct lu_fid            parent_fid;
2785         __u32                    linkno;
2786         __u32                    name_size;
2787         int                      rc;
2788
2789         ENTRY;
2790
2791         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2792             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2793                 RETURN(-EPERM);
2794
2795         if (get_user(name_size, &arg->gp_name_size))
2796                 RETURN(-EFAULT);
2797
2798         if (get_user(linkno, &arg->gp_linkno))
2799                 RETURN(-EFAULT);
2800
2801         if (name_size > PATH_MAX)
2802                 RETURN(-EINVAL);
2803
2804         OBD_ALLOC(ldata, sizeof(*ldata));
2805         if (ldata == NULL)
2806                 RETURN(-ENOMEM);
2807
2808         rc = linkea_data_new(ldata, &buf);
2809         if (rc < 0)
2810                 GOTO(ldata_free, rc);
2811
2812         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2813         if (rc < 0)
2814                 GOTO(lb_free, rc);
2815
2816         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2817         if (rc < 0)
2818                 GOTO(lb_free, rc);
2819
2820         if (ln.ln_namelen >= name_size)
2821                 GOTO(lb_free, rc = -EOVERFLOW);
2822
2823         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2824                 GOTO(lb_free, rc = -EFAULT);
2825
2826         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2827                 GOTO(lb_free, rc = -EFAULT);
2828
2829         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2830                 GOTO(lb_free, rc = -EFAULT);
2831
2832 lb_free:
2833         lu_buf_free(&buf);
2834 ldata_free:
2835         OBD_FREE(ldata, sizeof(*ldata));
2836
2837         RETURN(rc);
2838 }