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/time.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49 #include <linux/user_namespace.h>
  50 #ifdef HAVE_UIDGID_HEADER
  51 # include <linux/uidgid.h>
  52 #endif
  53
  54 #include <lustre_ioctl.h>
  55 #include <lustre_ha.h>
  56 #include <lustre_dlm.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_disk.h>
  59 #include <lustre_param.h>
  60 #include <lustre_log.h>
  61 #include <cl_object.h>
  62 #include <obd_cksum.h>
  63 #include "llite_internal.h"
  64
  65 struct kmem_cache *ll_file_data_slab;
  66
  67 #ifndef log2
  68 #define log2(n) ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         unsigned long lru_page_max;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         mutex_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         lru_page_max = pages / 2;
  94
  95         /* initialize ll_cache data */
  96         sbi->ll_cache = cl_cache_init(lru_page_max);
  97         if (sbi->ll_cache == NULL) {
  98                 OBD_FREE(sbi, sizeof(*sbi));
  99                 RETURN(NULL);
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         sbi->ll_flags |= LL_SBI_VERBOSE;
 115 #ifdef ENABLE_CHECKSUM
 116         sbi->ll_flags |= LL_SBI_CHECKSUM;
 117 #endif
 118
 119 #ifdef HAVE_LRU_RESIZE_SUPPORT
 120         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 121 #endif
 122
 123         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_r_hist.oh_lock);
 126                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 127                                pp_w_hist.oh_lock);
 128         }
 129
 130         /* metadata statahead is enabled by default */
 131         sbi->ll_sa_max = LL_SA_RPC_DEF;
 132         atomic_set(&sbi->ll_sa_total, 0);
 133         atomic_set(&sbi->ll_sa_wrong, 0);
 134         atomic_set(&sbi->ll_sa_running, 0);
 135         atomic_set(&sbi->ll_agl_total, 0);
 136         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 137
 138         /* root squash */
 139         sbi->ll_squash.rsi_uid = 0;
 140         sbi->ll_squash.rsi_gid = 0;
 141         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 142         init_rwsem(&sbi->ll_squash.rsi_sem);
 143
 144         RETURN(sbi);
 145 }
 146
 147 static void ll_free_sbi(struct super_block *sb)
 148 {
 149         struct ll_sb_info *sbi = ll_s2sbi(sb);
 150         ENTRY;
 151
 152         if (sbi != NULL) {
 153                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 154                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 155                 if (sbi->ll_cache != NULL) {
 156                         cl_cache_decref(sbi->ll_cache);
 157                         sbi->ll_cache = NULL;
 158                 }
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = NULL;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178         ENTRY;
 179
 180         obd = class_name2obd(md);
 181         if (!obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* indicate the features supported by this client */
 197         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 198                                   OBD_CONNECT_ATTRFID  |
 199                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 200                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 201                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 202                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 203                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 204                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 205                                   OBD_CONNECT_EINPROGRESS |
 206                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 207                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 208                                   OBD_CONNECT_MAX_EASIZE |
 209                                   OBD_CONNECT_FLOCK_DEAD |
 210                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 211                                   OBD_CONNECT_OPEN_BY_FID |
 212                                   OBD_CONNECT_DIR_STRIPE |
 213                                   OBD_CONNECT_BULK_MBITS |
 214                                   OBD_CONNECT_SUBTREE;
 215
 216 #ifdef HAVE_LRU_RESIZE_SUPPORT
 217         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 218                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 219 #endif
 220 #ifdef CONFIG_FS_POSIX_ACL
 221         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 222 #endif
 223
 224         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 225                 /* flag mdc connection as lightweight, only used for test
 226                  * purpose, use with care */
 227                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 228
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237 #ifdef HAVE_MS_FLOCK_LOCK
 238         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 239         sb->s_flags |= MS_FLOCK_LOCK;
 240 #endif
 241 #ifdef MS_HAS_NEW_AOPS
 242         sb->s_flags |= MS_HAS_NEW_AOPS;
 243 #endif
 244
 245         if (sbi->ll_flags & LL_SBI_FLOCK)
 246                 sbi->ll_fop = &ll_file_operations_flock;
 247         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 248                 sbi->ll_fop = &ll_file_operations;
 249         else
 250                 sbi->ll_fop = &ll_file_operations_noflock;
 251
 252         /* real client */
 253         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 254         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 255                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 256
 257         /* always ping even if server suppress_pings */
 258         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 259                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 260
 261         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 262
 263         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 264         if (err == -EBUSY) {
 265                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 266                                    "recovery, of which this client is not a "
 267                                    "part. Please wait for recovery to complete,"
 268                                    " abort, or time out.\n", md);
 269                 GOTO(out, err);
 270         } else if (err) {
 271                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 272                 GOTO(out, err);
 273         }
 274
 275         sbi->ll_md_exp->exp_connect_data = *data;
 276
 277         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 278                            LUSTRE_SEQ_METADATA);
 279         if (err) {
 280                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 281                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 282                 GOTO(out_md, err);
 283         }
 284
 285         /* For mount, we only need fs info from MDT0, and also in DNE, it
 286          * can make sure the client can be mounted as long as MDT0 is
 287          * avaible */
 288         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 289                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 290                         OBD_STATFS_FOR_MDT0);
 291         if (err)
 292                 GOTO(out_md_fid, err);
 293
 294         /* This needs to be after statfs to ensure connect has finished.
 295          * Note that "data" does NOT contain the valid connect reply.
 296          * If connecting to a 1.8 server there will be no LMV device, so
 297          * we can access the MDC export directly and exp_connect_flags will
 298          * be non-zero, but if accessing an upgraded 2.1 server it will
 299          * have the correct flags filled in.
 300          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 301         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 302         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 303             valid != CLIENT_CONNECT_MDT_REQD) {
 304                 char *buf;
 305
 306                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 307                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 308                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 309                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 310                                    "feature(s) needed for correct operation "
 311                                    "of this client (%s). Please upgrade "
 312                                    "server or downgrade client.\n",
 313                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 314                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 315                 GOTO(out_md_fid, err = -EPROTO);
 316         }
 317
 318         size = sizeof(*data);
 319         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 320                            KEY_CONN_DATA,  &size, data);
 321         if (err) {
 322                 CERROR("%s: Get connect data failed: rc = %d\n",
 323                        sbi->ll_md_exp->exp_obd->obd_name, err);
 324                 GOTO(out_md_fid, err);
 325         }
 326
 327         LASSERT(osfs->os_bsize);
 328         sb->s_blocksize = osfs->os_bsize;
 329         sb->s_blocksize_bits = log2(osfs->os_bsize);
 330         sb->s_magic = LL_SUPER_MAGIC;
 331         sb->s_maxbytes = MAX_LFS_FILESIZE;
 332         sbi->ll_namelen = osfs->os_namelen;
 333
 334         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 335             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 336                 LCONSOLE_INFO("Disabling user_xattr feature because "
 337                               "it is not supported on the server\n");
 338                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 339         }
 340
 341         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 342 #ifdef MS_POSIXACL
 343                 sb->s_flags |= MS_POSIXACL;
 344 #endif
 345                 sbi->ll_flags |= LL_SBI_ACL;
 346         } else {
 347                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 348 #ifdef MS_POSIXACL
 349                 sb->s_flags &= ~MS_POSIXACL;
 350 #endif
 351                 sbi->ll_flags &= ~LL_SBI_ACL;
 352         }
 353
 354         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 355                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 356                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 357                         LCONSOLE_INFO("client is set as remote by default.\n");
 358                 }
 359         } else {
 360                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 361                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client claims to be remote, but server "
 363                                       "rejected, forced to be local.\n");
 364                 }
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 368                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 371                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 372         else
 373                 sbi->ll_md_brw_pages = 1;
 374
 375         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 376                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 377
 378         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 379                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 380                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 381                                       "unknown maximum xattr size.\n", dt);
 382                 } else {
 383                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 384                         sbi->ll_xattr_cache_enabled = 1;
 385                 }
 386         }
 387
 388         obd = class_name2obd(dt);
 389         if (!obd) {
 390                 CERROR("DT %s: not setup or attached\n", dt);
 391                 GOTO(out_md_fid, err = -ENODEV);
 392         }
 393
 394         /* pass client page size via ocd_grant_blkbits, the server should report
 395          * back its backend blocksize for grant calculation purpose */
 396         data->ocd_grant_blkbits = PAGE_SHIFT;
 397
 398         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 399                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 400                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 401                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 402                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 403                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 404                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 405                                   OBD_CONNECT_MAXBYTES |
 406                                   OBD_CONNECT_EINPROGRESS |
 407                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 408                                   OBD_CONNECT_LAYOUTLOCK |
 409                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 410                                   OBD_CONNECT_BULK_MBITS;
 411
 412         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 413                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 414
 415         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 416                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 417                  * disabled by default, because it can still be enabled on the
 418                  * fly via /proc. As a consequence, we still need to come to an
 419                  * agreement on the supported algorithms at connect time */
 420                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 421
 422                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 423                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 424                 else
 425                         data->ocd_cksum_types = cksum_types_supported_client();
 426         }
 427
 428 #ifdef HAVE_LRU_RESIZE_SUPPORT
 429         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 430 #endif
 431         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 432                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 433
 434         /* always ping even if server suppress_pings */
 435         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 436                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 437
 438         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 439                "ocd_grant: %d\n", data->ocd_connect_flags,
 440                data->ocd_version, data->ocd_grant);
 441
 442         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 443         obd->obd_upcall.onu_upcall = cl_ocd_update;
 444
 445         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 446
 447         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 448                           NULL);
 449         if (err == -EBUSY) {
 450                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 451                                    "recovery, of which this client is not a "
 452                                    "part.  Please wait for recovery to "
 453                                    "complete, abort, or time out.\n", dt);
 454                 GOTO(out_md, err);
 455         } else if (err) {
 456                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 457                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 458                 GOTO(out_md, err);
 459         }
 460
 461         sbi->ll_dt_exp->exp_connect_data = *data;
 462
 463         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 464                            LUSTRE_SEQ_METADATA);
 465         if (err) {
 466                 CERROR("%s: Can't init data layer FID infrastructure, "
 467                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 468                 GOTO(out_dt, err);
 469         }
 470
 471         mutex_lock(&sbi->ll_lco.lco_lock);
 472         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 473         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 474         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 475         mutex_unlock(&sbi->ll_lco.lco_lock);
 476
 477         fid_zero(&sbi->ll_root_fid);
 478         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 479                            &sbi->ll_root_fid);
 480         if (err) {
 481                 CERROR("cannot mds_connect: rc = %d\n", err);
 482                 GOTO(out_lock_cn_cb, err);
 483         }
 484         if (!fid_is_sane(&sbi->ll_root_fid)) {
 485                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 486                        sbi->ll_md_exp->exp_obd->obd_name,
 487                        PFID(&sbi->ll_root_fid));
 488                 GOTO(out_lock_cn_cb, err = -EINVAL);
 489         }
 490         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 491
 492         sb->s_op = &lustre_super_operations;
 493 #if THREAD_SIZE >= 8192 /*b=17630*/
 494         sb->s_export_op = &lustre_export_operations;
 495 #endif
 496
 497         /* make root inode
 498          * XXX: move this to after cbd setup? */
 499         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 500         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 501                 valid |= OBD_MD_FLRMTPERM;
 502         else if (sbi->ll_flags & LL_SBI_ACL)
 503                 valid |= OBD_MD_FLACL;
 504
 505         OBD_ALLOC_PTR(op_data);
 506         if (op_data == NULL)
 507                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 508
 509         op_data->op_fid1 = sbi->ll_root_fid;
 510         op_data->op_mode = 0;
 511         op_data->op_valid = valid;
 512
 513         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 514
 515         OBD_FREE_PTR(op_data);
 516         if (err) {
 517                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 518                        sbi->ll_md_exp->exp_obd->obd_name, err);
 519                 GOTO(out_lock_cn_cb, err);
 520         }
 521
 522         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 523                                sbi->ll_md_exp, &lmd);
 524         if (err) {
 525                 CERROR("failed to understand root inode md: rc = %d\n", err);
 526                 ptlrpc_req_finished(request);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 531         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 532                                             sbi->ll_flags & LL_SBI_32BIT_API),
 533                        &lmd);
 534         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 535         ptlrpc_req_finished(request);
 536
 537         if (IS_ERR(root)) {
 538 #ifdef CONFIG_FS_POSIX_ACL
 539                 if (lmd.posix_acl) {
 540                         posix_acl_release(lmd.posix_acl);
 541                         lmd.posix_acl = NULL;
 542                 }
 543 #endif
 544                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 545                 root = NULL;
 546                 CERROR("lustre_lite: bad iget4 for root\n");
 547                 GOTO(out_root, err);
 548         }
 549
 550 #ifdef CONFIG_FS_POSIX_ACL
 551         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 552                 rct_init(&sbi->ll_rct);
 553                 et_init(&sbi->ll_et);
 554         }
 555 #endif
 556
 557         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 558         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 559                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 560                                  NULL);
 561         cl_sb_init(sb);
 562
 563         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 564                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 565                                  sbi->ll_cache, NULL);
 566
 567         sb->s_root = d_make_root(root);
 568         if (sb->s_root == NULL) {
 569                 CERROR("%s: can't make root dentry\n",
 570                         ll_get_fsname(sb, NULL, 0));
 571                 GOTO(out_root, err = -ENOMEM);
 572         }
 573 #ifdef HAVE_DCACHE_LOCK
 574         sb->s_root->d_op = &ll_d_ops;
 575 #endif
 576
 577         sbi->ll_sdev_orig = sb->s_dev;
 578
 579         /* We set sb->s_dev equal on all lustre clients in order to support
 580          * NFS export clustering.  NFSD requires that the FSID be the same
 581          * on all clients. */
 582         /* s_dev is also used in lt_compare() to compare two fs, but that is
 583          * only a node-local comparison. */
 584         uuid = obd_get_uuid(sbi->ll_md_exp);
 585         if (uuid != NULL)
 586                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 587
 588         if (data != NULL)
 589                 OBD_FREE_PTR(data);
 590         if (osfs != NULL)
 591                 OBD_FREE_PTR(osfs);
 592         if (proc_lustre_fs_root != NULL) {
 593                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 594                                                   dt, md);
 595                 if (err < 0) {
 596                         CERROR("%s: could not register mount in lprocfs: "
 597                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 598                         err = 0;
 599                 }
 600         }
 601
 602         RETURN(err);
 603 out_root:
 604         if (root)
 605                 iput(root);
 606 out_lock_cn_cb:
 607         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 608 out_dt:
 609         obd_disconnect(sbi->ll_dt_exp);
 610         sbi->ll_dt_exp = NULL;
 611 out_md_fid:
 612         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 613 out_md:
 614         obd_disconnect(sbi->ll_md_exp);
 615         sbi->ll_md_exp = NULL;
 616 out:
 617         if (data != NULL)
 618                 OBD_FREE_PTR(data);
 619         if (osfs != NULL)
 620                 OBD_FREE_PTR(osfs);
 621         return err;
 622 }
 623
 624 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 625 {
 626         int size, rc;
 627
 628         size = sizeof(*lmmsize);
 629         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 630                           KEY_MAX_EASIZE, &size, lmmsize);
 631         if (rc != 0) {
 632                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 633                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 634                 RETURN(rc);
 635         }
 636
 637         size = sizeof(int);
 638         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 639                           KEY_MAX_EASIZE, &size, lmmsize);
 640         if (rc)
 641                 CERROR("Get max mdsize error rc %d\n", rc);
 642
 643         RETURN(rc);
 644 }
 645
 646 /**
 647  * Get the value of the default_easize parameter.
 648  *
 649  * \see client_obd::cl_default_mds_easize
 650  *
 651  * \param[in] sbi       superblock info for this filesystem
 652  * \param[out] lmmsize  pointer to storage location for value
 653  *
 654  * \retval 0            on success
 655  * \retval negative     negated errno on failure
 656  */
 657 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 658 {
 659         int size, rc;
 660
 661         size = sizeof(int);
 662         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 663                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 664         if (rc)
 665                 CERROR("Get default mdsize error rc %d\n", rc);
 666
 667         RETURN(rc);
 668 }
 669
 670 /**
 671  * Set the default_easize parameter to the given value.
 672  *
 673  * \see client_obd::cl_default_mds_easize
 674  *
 675  * \param[in] sbi       superblock info for this filesystem
 676  * \param[in] lmmsize   the size to set
 677  *
 678  * \retval 0            on success
 679  * \retval negative     negated errno on failure
 680  */
 681 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 682 {
 683         int rc;
 684
 685         if (lmmsize < sizeof(struct lov_mds_md) ||
 686             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 687                 return -EINVAL;
 688
 689         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 690                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 691                                 sizeof(int), &lmmsize, NULL);
 692
 693         RETURN(rc);
 694 }
 695
 696 static void ll_dump_inode(struct inode *inode)
 697 {
 698         struct ll_d_hlist_node *tmp;
 699         int dentry_count = 0;
 700
 701         LASSERT(inode != NULL);
 702
 703         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 704                 dentry_count++;
 705
 706         CERROR("%s: inode %p dump: dev=%s fid="DFID
 707                " mode=%o count=%u, %d dentries\n",
 708                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 709                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 710                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 711 }
 712
 713 void lustre_dump_dentry(struct dentry *dentry, int recur)
 714 {
 715         struct list_head *tmp;
 716         int subdirs = 0;
 717
 718         LASSERT(dentry != NULL);
 719
 720         list_for_each(tmp, &dentry->d_subdirs)
 721                 subdirs++;
 722
 723         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 724                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 725                dentry->d_name.len, dentry->d_name.name,
 726                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 727                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 728                dentry->d_flags, dentry->d_fsdata, subdirs);
 729         if (dentry->d_inode != NULL)
 730                 ll_dump_inode(dentry->d_inode);
 731
 732         if (recur == 0)
 733                 return;
 734
 735         list_for_each(tmp, &dentry->d_subdirs) {
 736                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 737                 lustre_dump_dentry(d, recur - 1);
 738         }
 739 }
 740
 741 static void client_common_put_super(struct super_block *sb)
 742 {
 743         struct ll_sb_info *sbi = ll_s2sbi(sb);
 744         ENTRY;
 745
 746 #ifdef CONFIG_FS_POSIX_ACL
 747         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 748                 et_fini(&sbi->ll_et);
 749                 rct_fini(&sbi->ll_rct);
 750         }
 751 #endif
 752
 753         cl_sb_fini(sb);
 754
 755         list_del(&sbi->ll_conn_chain);
 756
 757         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 758         obd_disconnect(sbi->ll_dt_exp);
 759         sbi->ll_dt_exp = NULL;
 760
 761         lprocfs_unregister_mountpoint(sbi);
 762
 763         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 764         obd_disconnect(sbi->ll_md_exp);
 765         sbi->ll_md_exp = NULL;
 766
 767         EXIT;
 768 }
 769
 770 void ll_kill_super(struct super_block *sb)
 771 {
 772         struct ll_sb_info *sbi;
 773         ENTRY;
 774
 775         /* not init sb ?*/
 776         if (!(sb->s_flags & MS_ACTIVE))
 777                 return;
 778
 779         sbi = ll_s2sbi(sb);
 780         /* we need restore s_dev from changed for clustred NFS before put_super
 781          * because new kernels have cached s_dev and change sb->s_dev in
 782          * put_super not affected real removing devices */
 783         if (sbi) {
 784                 sb->s_dev = sbi->ll_sdev_orig;
 785                 sbi->ll_umounting = 1;
 786
 787                 /* wait running statahead threads to quit */
 788                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 789                         set_current_state(TASK_UNINTERRUPTIBLE);
 790                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 791                 }
 792         }
 793
 794         EXIT;
 795 }
 796
 797 static inline int ll_set_opt(const char *opt, char *data, int fl)
 798 {
 799         if (strncmp(opt, data, strlen(opt)) != 0)
 800                 return(0);
 801         else
 802                 return(fl);
 803 }
 804
 805 /* non-client-specific mount options are parsed in lmd_parse */
 806 static int ll_options(char *options, int *flags)
 807 {
 808         int tmp;
 809         char *s1 = options, *s2;
 810         ENTRY;
 811
 812         if (!options)
 813                 RETURN(0);
 814
 815         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 816
 817         while (*s1) {
 818                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 819                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 835                 if (tmp) {
 836                         *flags &= ~tmp;
 837                         goto next;
 838                 }
 839                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 840                 if (tmp) {
 841                         *flags |= tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 845                 if (tmp) {
 846                         *flags &= ~tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("context", s1, 1);
 850                 if (tmp)
 851                         goto next;
 852                 tmp = ll_set_opt("fscontext", s1, 1);
 853                 if (tmp)
 854                         goto next;
 855                 tmp = ll_set_opt("defcontext", s1, 1);
 856                 if (tmp)
 857                         goto next;
 858                 tmp = ll_set_opt("rootcontext", s1, 1);
 859                 if (tmp)
 860                         goto next;
 861                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876
 877                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 893                 if (tmp) {
 894                         *flags &= ~tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 908                 if (tmp) {
 909                         *flags |= tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 913                 if (tmp) {
 914                         *flags |= tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 918                 if (tmp) {
 919                         *flags &= ~tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 923                 if (tmp) {
 924                         *flags |= tmp;
 925                         goto next;
 926                 }
 927                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 928                                    s1);
 929                 RETURN(-EINVAL);
 930
 931 next:
 932                 /* Find next opt */
 933                 s2 = strchr(s1, ',');
 934                 if (s2 == NULL)
 935                         break;
 936                 s1 = s2 + 1;
 937         }
 938         RETURN(0);
 939 }
 940
 941 void ll_lli_init(struct ll_inode_info *lli)
 942 {
 943         lli->lli_inode_magic = LLI_INODE_MAGIC;
 944         lli->lli_flags = 0;
 945         spin_lock_init(&lli->lli_lock);
 946         lli->lli_posix_acl = NULL;
 947         lli->lli_remote_perms = NULL;
 948         mutex_init(&lli->lli_rmtperm_mutex);
 949         /* Do not set lli_fid, it has been initialized already. */
 950         fid_zero(&lli->lli_pfid);
 951         atomic_set(&lli->lli_open_count, 0);
 952         lli->lli_rmtperm_time = 0;
 953         lli->lli_mds_read_och = NULL;
 954         lli->lli_mds_write_och = NULL;
 955         lli->lli_mds_exec_och = NULL;
 956         lli->lli_open_fd_read_count = 0;
 957         lli->lli_open_fd_write_count = 0;
 958         lli->lli_open_fd_exec_count = 0;
 959         mutex_init(&lli->lli_och_mutex);
 960         spin_lock_init(&lli->lli_agl_lock);
 961         spin_lock_init(&lli->lli_layout_lock);
 962         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 963         lli->lli_clob = NULL;
 964
 965         init_rwsem(&lli->lli_xattrs_list_rwsem);
 966         mutex_init(&lli->lli_xattrs_enq_lock);
 967
 968         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 969         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 970                 mutex_init(&lli->lli_readdir_mutex);
 971                 lli->lli_opendir_key = NULL;
 972                 lli->lli_sai = NULL;
 973                 spin_lock_init(&lli->lli_sa_lock);
 974                 lli->lli_opendir_pid = 0;
 975                 lli->lli_sa_enabled = 0;
 976                 lli->lli_def_stripe_offset = -1;
 977         } else {
 978                 mutex_init(&lli->lli_size_mutex);
 979                 lli->lli_symlink_name = NULL;
 980                 init_rwsem(&lli->lli_trunc_sem);
 981                 range_lock_tree_init(&lli->lli_write_tree);
 982                 init_rwsem(&lli->lli_glimpse_sem);
 983                 lli->lli_glimpse_time = 0;
 984                 INIT_LIST_HEAD(&lli->lli_agl_list);
 985                 lli->lli_agl_index = 0;
 986                 lli->lli_async_rc = 0;
 987         }
 988         mutex_init(&lli->lli_layout_mutex);
 989 }
 990
 991 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 992 {
 993         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 994
 995         bdi->name = "lustre";
 996         return bdi_register(bdi, NULL, "lustre-%d",
 997                             atomic_inc_return(&ll_bdi_num));
 998 }
 999
1000 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1001 {
1002         struct lustre_profile *lprof = NULL;
1003         struct lustre_sb_info *lsi = s2lsi(sb);
1004         struct ll_sb_info *sbi;
1005         char  *dt = NULL, *md = NULL;
1006         char  *profilenm = get_profile_name(sb);
1007         struct config_llog_instance *cfg;
1008         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1009         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1010         int    err;
1011         ENTRY;
1012
1013         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1014
1015         OBD_ALLOC_PTR(cfg);
1016         if (cfg == NULL)
1017                 RETURN(-ENOMEM);
1018
1019         try_module_get(THIS_MODULE);
1020
1021         /* client additional sb info */
1022         lsi->lsi_llsbi = sbi = ll_init_sbi();
1023         if (!sbi) {
1024                 module_put(THIS_MODULE);
1025                 OBD_FREE_PTR(cfg);
1026                 RETURN(-ENOMEM);
1027         }
1028
1029         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1030         if (err)
1031                 GOTO(out_free, err);
1032
1033         err = bdi_init(&lsi->lsi_bdi);
1034         if (err)
1035                 GOTO(out_free, err);
1036         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1037 #ifdef HAVE_BDI_CAP_MAP_COPY
1038         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1039 #else
1040         lsi->lsi_bdi.capabilities = 0;
1041 #endif
1042         err = ll_bdi_register(&lsi->lsi_bdi);
1043         if (err)
1044                 GOTO(out_free, err);
1045
1046         sb->s_bdi = &lsi->lsi_bdi;
1047 #ifndef HAVE_DCACHE_LOCK
1048         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1049         sb->s_d_op = &ll_d_ops;
1050 #endif
1051
1052         /* Generate a string unique to this super, in case some joker tries
1053            to mount the same fs at two mount points.
1054            Use the address of the super itself.*/
1055         cfg->cfg_instance = sb;
1056         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1057         cfg->cfg_callback = class_config_llog_handler;
1058         /* set up client obds */
1059         err = lustre_process_log(sb, profilenm, cfg);
1060         if (err < 0)
1061                 GOTO(out_free, err);
1062
1063         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1064         lprof = class_get_profile(profilenm);
1065         if (lprof == NULL) {
1066                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1067                                    " read from the MGS.  Does that filesystem "
1068                                    "exist?\n", profilenm);
1069                 GOTO(out_free, err = -EINVAL);
1070         }
1071         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1072                lprof->lp_md, lprof->lp_dt);
1073
1074         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1075         if (!dt)
1076                 GOTO(out_free, err = -ENOMEM);
1077         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1078
1079         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1080         if (!md)
1081                 GOTO(out_free, err = -ENOMEM);
1082         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1083
1084         /* connections, registrations, sb setup */
1085         err = client_common_fill_super(sb, md, dt, mnt);
1086         if (err < 0)
1087                 GOTO(out_free, err);
1088
1089         sbi->ll_client_common_fill_super_succeeded = 1;
1090
1091 out_free:
1092         if (md)
1093                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1094         if (dt)
1095                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1096         if (lprof != NULL)
1097                 class_put_profile(lprof);
1098         if (err)
1099                 ll_put_super(sb);
1100         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1101                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1102
1103         OBD_FREE_PTR(cfg);
1104         RETURN(err);
1105 } /* ll_fill_super */
1106
1107 void ll_put_super(struct super_block *sb)
1108 {
1109         struct config_llog_instance cfg, params_cfg;
1110         struct obd_device *obd;
1111         struct lustre_sb_info *lsi = s2lsi(sb);
1112         struct ll_sb_info *sbi = ll_s2sbi(sb);
1113         char *profilenm = get_profile_name(sb);
1114         long ccc_count;
1115         int next, force = 1, rc = 0;
1116         ENTRY;
1117
1118         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1119
1120         cfg.cfg_instance = sb;
1121         lustre_end_log(sb, profilenm, &cfg);
1122
1123         params_cfg.cfg_instance = sb;
1124         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1125
1126         if (sbi->ll_md_exp) {
1127                 obd = class_exp2obd(sbi->ll_md_exp);
1128                 if (obd)
1129                         force = obd->obd_force;
1130         }
1131
1132         /* Wait for unstable pages to be committed to stable storage */
1133         if (force == 0) {
1134                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1135                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1136                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1137                         &lwi);
1138         }
1139
1140         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1141         if (force == 0 && rc != -EINTR)
1142                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1143
1144
1145         /* We need to set force before the lov_disconnect in
1146            lustre_common_put_super, since l_d cleans up osc's as well. */
1147         if (force) {
1148                 next = 0;
1149                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1150                                                      &next)) != NULL) {
1151                         obd->obd_force = force;
1152                 }
1153         }
1154
1155         if (sbi->ll_client_common_fill_super_succeeded) {
1156                 /* Only if client_common_fill_super succeeded */
1157                 client_common_put_super(sb);
1158         }
1159
1160         next = 0;
1161         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1162                 class_manual_cleanup(obd);
1163         }
1164
1165         if (sbi->ll_flags & LL_SBI_VERBOSE)
1166                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1167
1168         if (profilenm)
1169                 class_del_profile(profilenm);
1170
1171         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1172                 bdi_destroy(&lsi->lsi_bdi);
1173                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1174         }
1175
1176         ll_free_sbi(sb);
1177         lsi->lsi_llsbi = NULL;
1178
1179         lustre_common_put_super(sb);
1180
1181         cl_env_cache_purge(~0);
1182
1183         module_put(THIS_MODULE);
1184
1185         EXIT;
1186 } /* client_put_super */
1187
1188 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1189 {
1190         struct inode *inode = NULL;
1191
1192         /* NOTE: we depend on atomic igrab() -bzzz */
1193         lock_res_and_lock(lock);
1194         if (lock->l_resource->lr_lvb_inode) {
1195                 struct ll_inode_info * lli;
1196                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1197                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1198                         inode = igrab(lock->l_resource->lr_lvb_inode);
1199                 } else {
1200                         inode = lock->l_resource->lr_lvb_inode;
1201                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1202                                          D_WARNING, lock, "lr_lvb_inode %p is "
1203                                          "bogus: magic %08x",
1204                                          lock->l_resource->lr_lvb_inode,
1205                                          lli->lli_inode_magic);
1206                         inode = NULL;
1207                 }
1208         }
1209         unlock_res_and_lock(lock);
1210         return inode;
1211 }
1212
1213 void ll_dir_clear_lsm_md(struct inode *inode)
1214 {
1215         struct ll_inode_info *lli = ll_i2info(inode);
1216
1217         LASSERT(S_ISDIR(inode->i_mode));
1218
1219         if (lli->lli_lsm_md != NULL) {
1220                 lmv_free_memmd(lli->lli_lsm_md);
1221                 lli->lli_lsm_md = NULL;
1222         }
1223 }
1224
1225 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1226                                       const struct lu_fid *fid,
1227                                       struct lustre_md *md)
1228 {
1229         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1230         struct mdt_body         *body = md->body;
1231         struct inode            *inode;
1232         ino_t                   ino;
1233         ENTRY;
1234
1235         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1236         inode = iget_locked(sb, ino);
1237         if (inode == NULL) {
1238                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1239                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1240                 RETURN(ERR_PTR(-ENOENT));
1241         }
1242
1243         if (inode->i_state & I_NEW) {
1244                 struct ll_inode_info *lli = ll_i2info(inode);
1245                 struct lmv_stripe_md *lsm = md->lmv;
1246
1247                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1248                                 (body->mbo_mode & S_IFMT);
1249                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1250                          PFID(fid));
1251
1252                 LTIME_S(inode->i_mtime) = 0;
1253                 LTIME_S(inode->i_atime) = 0;
1254                 LTIME_S(inode->i_ctime) = 0;
1255                 inode->i_rdev = 0;
1256
1257 #ifdef HAVE_BACKING_DEV_INFO
1258                 /* initializing backing dev info. */
1259                 inode->i_mapping->backing_dev_info =
1260                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1261 #endif
1262                 inode->i_op = &ll_dir_inode_operations;
1263                 inode->i_fop = &ll_dir_operations;
1264                 lli->lli_fid = *fid;
1265                 ll_lli_init(lli);
1266
1267                 LASSERT(lsm != NULL);
1268                 /* master object FID */
1269                 lli->lli_pfid = body->mbo_fid1;
1270                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1271                        lli, PFID(fid), PFID(&lli->lli_pfid));
1272                 unlock_new_inode(inode);
1273         }
1274
1275         RETURN(inode);
1276 }
1277
1278 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1279 {
1280         struct lu_fid *fid;
1281         struct lmv_stripe_md *lsm = md->lmv;
1282         int i;
1283
1284         LASSERT(lsm != NULL);
1285         /* XXX sigh, this lsm_root initialization should be in
1286          * LMV layer, but it needs ll_iget right now, so we
1287          * put this here right now. */
1288         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1289                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1290                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1291                 /* Unfortunately ll_iget will call ll_update_inode,
1292                  * where the initialization of slave inode is slightly
1293                  * different, so it reset lsm_md to NULL to avoid
1294                  * initializing lsm for slave inode. */
1295                 /* For migrating inode, master stripe and master object will
1296                  * be same, so we only need assign this inode */
1297                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1298                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1299                 else
1300                         lsm->lsm_md_oinfo[i].lmo_root =
1301                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1302
1303                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1304                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1305
1306                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1307                         return rc;
1308                 }
1309         }
1310
1311         return 0;
1312 }
1313
1314 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1315                                 const struct lmv_stripe_md *lsm_md2)
1316 {
1317         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1318                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1319                lsm_md1->lsm_md_master_mdt_index ==
1320                                         lsm_md2->lsm_md_master_mdt_index &&
1321                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1322                lsm_md1->lsm_md_layout_version ==
1323                                         lsm_md2->lsm_md_layout_version &&
1324                strcmp(lsm_md1->lsm_md_pool_name,
1325                       lsm_md2->lsm_md_pool_name) == 0;
1326 }
1327
1328 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1329 {
1330         struct ll_inode_info *lli = ll_i2info(inode);
1331         struct lmv_stripe_md *lsm = md->lmv;
1332         int     rc;
1333         ENTRY;
1334
1335         LASSERT(S_ISDIR(inode->i_mode));
1336         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1337                PFID(ll_inode2fid(inode)));
1338
1339         /* no striped information from request. */
1340         if (lsm == NULL) {
1341                 if (lli->lli_lsm_md == NULL) {
1342                         RETURN(0);
1343                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1344                                                 LMV_HASH_FLAG_MIGRATION) {
1345                         /* migration is done, the temporay MIGRATE layout has
1346                          * been removed */
1347                         CDEBUG(D_INODE, DFID" finish migration.\n",
1348                                PFID(ll_inode2fid(inode)));
1349                         lmv_free_memmd(lli->lli_lsm_md);
1350                         lli->lli_lsm_md = NULL;
1351                         RETURN(0);
1352                 } else {
1353                         /* The lustre_md from req does not include stripeEA,
1354                          * see ll_md_setattr */
1355                         RETURN(0);
1356                 }
1357         }
1358
1359         /* set the directory layout */
1360         if (lli->lli_lsm_md == NULL) {
1361                 struct cl_attr  *attr;
1362
1363                 rc = ll_init_lsm_md(inode, md);
1364                 if (rc != 0)
1365                         RETURN(rc);
1366
1367                 /* set md->lmv to NULL, so the following free lustre_md
1368                  * will not free this lsm */
1369                 md->lmv = NULL;
1370                 lli->lli_lsm_md = lsm;
1371
1372                 OBD_ALLOC_PTR(attr);
1373                 if (attr == NULL)
1374                         RETURN(-ENOMEM);
1375
1376                 /* validate the lsm */
1377                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1378                                    ll_md_blocking_ast);
1379                 if (rc != 0) {
1380                         OBD_FREE_PTR(attr);
1381                         RETURN(rc);
1382                 }
1383
1384                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1385                         md->body->mbo_nlink = attr->cat_nlink;
1386                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1387                         md->body->mbo_size = attr->cat_size;
1388                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1389                         md->body->mbo_atime = attr->cat_atime;
1390                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1391                         md->body->mbo_ctime = attr->cat_ctime;
1392                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1393                         md->body->mbo_mtime = attr->cat_mtime;
1394
1395                 OBD_FREE_PTR(attr);
1396
1397                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1398                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1399                 RETURN(0);
1400         }
1401
1402         /* Compare the old and new stripe information */
1403         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1404                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1405                 int                     idx;
1406
1407                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1408                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1409                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1410                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1411                        inode, lsm, old_lsm,
1412                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1413                        lsm->lsm_md_stripe_count,
1414                        old_lsm->lsm_md_stripe_count,
1415                        lsm->lsm_md_master_mdt_index,
1416                        old_lsm->lsm_md_master_mdt_index,
1417                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1418                        lsm->lsm_md_layout_version,
1419                        old_lsm->lsm_md_layout_version,
1420                        lsm->lsm_md_pool_name,
1421                        old_lsm->lsm_md_pool_name);
1422
1423                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1424                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1425                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1426                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1427                 }
1428
1429                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1430                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1431                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1432                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1433                 }
1434
1435                 RETURN(-EIO);
1436         }
1437
1438         RETURN(0);
1439 }
1440
1441 void ll_clear_inode(struct inode *inode)
1442 {
1443         struct ll_inode_info *lli = ll_i2info(inode);
1444         struct ll_sb_info *sbi = ll_i2sbi(inode);
1445         ENTRY;
1446
1447         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1448                PFID(ll_inode2fid(inode)), inode);
1449
1450         if (S_ISDIR(inode->i_mode)) {
1451                 /* these should have been cleared in ll_file_release */
1452                 LASSERT(lli->lli_opendir_key == NULL);
1453                 LASSERT(lli->lli_sai == NULL);
1454                 LASSERT(lli->lli_opendir_pid == 0);
1455         }
1456
1457         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1458
1459         LASSERT(!lli->lli_open_fd_write_count);
1460         LASSERT(!lli->lli_open_fd_read_count);
1461         LASSERT(!lli->lli_open_fd_exec_count);
1462
1463         if (lli->lli_mds_write_och)
1464                 ll_md_real_close(inode, FMODE_WRITE);
1465         if (lli->lli_mds_exec_och)
1466                 ll_md_real_close(inode, FMODE_EXEC);
1467         if (lli->lli_mds_read_och)
1468                 ll_md_real_close(inode, FMODE_READ);
1469
1470         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1471                 OBD_FREE(lli->lli_symlink_name,
1472                          strlen(lli->lli_symlink_name) + 1);
1473                 lli->lli_symlink_name = NULL;
1474         }
1475
1476         ll_xattr_cache_destroy(inode);
1477
1478         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1479                 LASSERT(lli->lli_posix_acl == NULL);
1480                 if (lli->lli_remote_perms) {
1481                         free_rmtperm_hash(lli->lli_remote_perms);
1482                         lli->lli_remote_perms = NULL;
1483                 }
1484         }
1485 #ifdef CONFIG_FS_POSIX_ACL
1486         else if (lli->lli_posix_acl) {
1487                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1488                 LASSERT(lli->lli_remote_perms == NULL);
1489                 posix_acl_release(lli->lli_posix_acl);
1490                 lli->lli_posix_acl = NULL;
1491         }
1492 #endif
1493         lli->lli_inode_magic = LLI_INODE_DEAD;
1494
1495         if (S_ISDIR(inode->i_mode))
1496                 ll_dir_clear_lsm_md(inode);
1497         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1498                 LASSERT(list_empty(&lli->lli_agl_list));
1499
1500         /*
1501          * XXX This has to be done before lsm is freed below, because
1502          * cl_object still uses inode lsm.
1503          */
1504         cl_inode_fini(inode);
1505
1506         EXIT;
1507 }
1508
1509 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1510 {
1511         struct lustre_md md;
1512         struct inode *inode = dentry->d_inode;
1513         struct ll_sb_info *sbi = ll_i2sbi(inode);
1514         struct ptlrpc_request *request = NULL;
1515         int rc, ia_valid;
1516         ENTRY;
1517
1518         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1519                                      LUSTRE_OPC_ANY, NULL);
1520         if (IS_ERR(op_data))
1521                 RETURN(PTR_ERR(op_data));
1522
1523         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1524         if (rc) {
1525                 ptlrpc_req_finished(request);
1526                 if (rc == -ENOENT) {
1527                         clear_nlink(inode);
1528                         /* Unlinked special device node? Or just a race?
1529                          * Pretend we done everything. */
1530                         if (!S_ISREG(inode->i_mode) &&
1531                             !S_ISDIR(inode->i_mode)) {
1532                                 ia_valid = op_data->op_attr.ia_valid;
1533                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1534                                 rc = simple_setattr(dentry, &op_data->op_attr);
1535                                 op_data->op_attr.ia_valid = ia_valid;
1536                         }
1537                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1538                         CERROR("md_setattr fails: rc = %d\n", rc);
1539                 }
1540                 RETURN(rc);
1541         }
1542
1543         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1544                               sbi->ll_md_exp, &md);
1545         if (rc) {
1546                 ptlrpc_req_finished(request);
1547                 RETURN(rc);
1548         }
1549
1550         ia_valid = op_data->op_attr.ia_valid;
1551         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1552          * cache is not cleared yet. */
1553         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1554         if (S_ISREG(inode->i_mode))
1555                 mutex_lock(&inode->i_mutex);
1556         rc = simple_setattr(dentry, &op_data->op_attr);
1557         if (S_ISREG(inode->i_mode))
1558                 mutex_unlock(&inode->i_mutex);
1559         op_data->op_attr.ia_valid = ia_valid;
1560
1561         rc = ll_update_inode(inode, &md);
1562         ptlrpc_req_finished(request);
1563
1564         RETURN(rc);
1565 }
1566
1567 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1568  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1569  * keep these values until such a time that objects are allocated for it.
1570  * We do the MDS operations first, as it is checking permissions for us.
1571  * We don't to the MDS RPC if there is nothing that we want to store there,
1572  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1573  * going to do an RPC anyways.
1574  *
1575  * If we are doing a truncate, we will send the mtime and ctime updates
1576  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1577  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1578  * at the same time.
1579  *
1580  * In case of HSMimport, we only set attr on MDS.
1581  */
1582 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1583 {
1584         struct inode *inode = dentry->d_inode;
1585         struct ll_inode_info *lli = ll_i2info(inode);
1586         struct md_op_data *op_data = NULL;
1587         int rc = 0;
1588         ENTRY;
1589
1590         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1591                "valid %x, hsm_import %d\n",
1592                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1593                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1594                hsm_import);
1595
1596         if (attr->ia_valid & ATTR_SIZE) {
1597                 /* Check new size against VFS/VM file size limit and rlimit */
1598                 rc = inode_newsize_ok(inode, attr->ia_size);
1599                 if (rc)
1600                         RETURN(rc);
1601
1602                 /* The maximum Lustre file size is variable, based on the
1603                  * OST maximum object size and number of stripes.  This
1604                  * needs another check in addition to the VFS check above. */
1605                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1606                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1607                                PFID(&lli->lli_fid), attr->ia_size,
1608                                ll_file_maxbytes(inode));
1609                         RETURN(-EFBIG);
1610                 }
1611
1612                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1613         }
1614
1615         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1616         if (attr->ia_valid & TIMES_SET_FLAGS) {
1617                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1618                     !cfs_capable(CFS_CAP_FOWNER))
1619                         RETURN(-EPERM);
1620         }
1621
1622         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1623         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1624             (attr->ia_valid & ATTR_CTIME)) {
1625                 attr->ia_ctime = CURRENT_TIME;
1626                 attr->ia_valid |= ATTR_CTIME_SET;
1627         }
1628         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1629             (attr->ia_valid & ATTR_ATIME)) {
1630                 attr->ia_atime = CURRENT_TIME;
1631                 attr->ia_valid |= ATTR_ATIME_SET;
1632         }
1633         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1634             (attr->ia_valid & ATTR_MTIME)) {
1635                 attr->ia_mtime = CURRENT_TIME;
1636                 attr->ia_valid |= ATTR_MTIME_SET;
1637         }
1638
1639         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1640                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1641                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1642                        cfs_time_current_sec());
1643
1644         if (S_ISREG(inode->i_mode)) {
1645                 if (attr->ia_valid & ATTR_SIZE)
1646                         inode_dio_write_done(inode);
1647                 mutex_unlock(&inode->i_mutex);
1648         }
1649
1650         /* We always do an MDS RPC, even if we're only changing the size;
1651          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1652
1653         OBD_ALLOC_PTR(op_data);
1654         if (op_data == NULL)
1655                 GOTO(out, rc = -ENOMEM);
1656
1657         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1658                 /* If we are changing file size, file content is
1659                  * modified, flag it. */
1660                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1661                 op_data->op_bias |= MDS_DATA_MODIFIED;
1662                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1663         }
1664
1665         op_data->op_attr = *attr;
1666
1667         rc = ll_md_setattr(dentry, op_data);
1668         if (rc)
1669                 GOTO(out, rc);
1670
1671         if (!S_ISREG(inode->i_mode) || hsm_import)
1672                 GOTO(out, rc = 0);
1673
1674         if (attr->ia_valid & (ATTR_SIZE |
1675                               ATTR_ATIME | ATTR_ATIME_SET |
1676                               ATTR_MTIME | ATTR_MTIME_SET |
1677                               ATTR_CTIME | ATTR_CTIME_SET)) {
1678                 /* For truncate and utimes sending attributes to OSTs, setting
1679                  * mtime/atime to the past will be performed under PW [0:EOF]
1680                  * extent lock (new_size:EOF for truncate).  It may seem
1681                  * excessive to send mtime/atime updates to OSTs when not
1682                  * setting times to past, but it is necessary due to possible
1683                  * time de-synchronization between MDT inode and OST objects */
1684                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1685         }
1686
1687         /* If the file was restored, it needs to set dirty flag.
1688          *
1689          * We've already sent MDS_DATA_MODIFIED flag in
1690          * ll_md_setattr() for truncate. However, the MDT refuses to
1691          * set the HS_DIRTY flag on released files, so we have to set
1692          * it again if the file has been restored. Please check how
1693          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1694          *
1695          * Please notice that if the file is not released, the previous
1696          * MDS_DATA_MODIFIED has taken effect and usually
1697          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1698          * This way we can save an RPC for common open + trunc
1699          * operation. */
1700         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1701                 struct hsm_state_set hss = {
1702                         .hss_valid = HSS_SETMASK,
1703                         .hss_setmask = HS_DIRTY,
1704                 };
1705                 int rc2;
1706
1707                 rc2 = ll_hsm_state_set(inode, &hss);
1708                 /* truncate and write can happen at the same time, so that
1709                  * the file can be set modified even though the file is not
1710                  * restored from released state, and ll_hsm_state_set() is
1711                  * not applicable for the file, and rc2 < 0 is normal in this
1712                  * case. */
1713                 if (rc2 < 0)
1714                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1715                                PFID(ll_inode2fid(inode)), rc2);
1716         }
1717
1718         EXIT;
1719 out:
1720         if (op_data != NULL)
1721                 ll_finish_md_op_data(op_data);
1722
1723         if (S_ISREG(inode->i_mode)) {
1724                 mutex_lock(&inode->i_mutex);
1725                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1726                         inode_dio_wait(inode);
1727         }
1728
1729         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1730                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1731
1732         return rc;
1733 }
1734
1735 int ll_setattr(struct dentry *de, struct iattr *attr)
1736 {
1737         int mode = de->d_inode->i_mode;
1738
1739         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1740                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1741                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1742
1743         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1744                                (ATTR_SIZE|ATTR_MODE)) &&
1745             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1746              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1747               !(attr->ia_mode & S_ISGID))))
1748                 attr->ia_valid |= ATTR_FORCE;
1749
1750         if ((attr->ia_valid & ATTR_MODE) &&
1751             (mode & S_ISUID) &&
1752             !(attr->ia_mode & S_ISUID) &&
1753             !(attr->ia_valid & ATTR_KILL_SUID))
1754                 attr->ia_valid |= ATTR_KILL_SUID;
1755
1756         if ((attr->ia_valid & ATTR_MODE) &&
1757             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1758             !(attr->ia_mode & S_ISGID) &&
1759             !(attr->ia_valid & ATTR_KILL_SGID))
1760                 attr->ia_valid |= ATTR_KILL_SGID;
1761
1762         return ll_setattr_raw(de, attr, false);
1763 }
1764
1765 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1766                        __u64 max_age, __u32 flags)
1767 {
1768         struct ll_sb_info *sbi = ll_s2sbi(sb);
1769         struct obd_statfs obd_osfs;
1770         int rc;
1771         ENTRY;
1772
1773         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1774         if (rc) {
1775                 CERROR("md_statfs fails: rc = %d\n", rc);
1776                 RETURN(rc);
1777         }
1778
1779         osfs->os_type = sb->s_magic;
1780
1781         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1782                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1783
1784         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1785                 flags |= OBD_STATFS_NODELAY;
1786
1787         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1788         if (rc) {
1789                 CERROR("obd_statfs fails: rc = %d\n", rc);
1790                 RETURN(rc);
1791         }
1792
1793         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1794                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1795                obd_osfs.os_files);
1796
1797         osfs->os_bsize = obd_osfs.os_bsize;
1798         osfs->os_blocks = obd_osfs.os_blocks;
1799         osfs->os_bfree = obd_osfs.os_bfree;
1800         osfs->os_bavail = obd_osfs.os_bavail;
1801
1802         /* If we don't have as many objects free on the OST as inodes
1803          * on the MDS, we reduce the total number of inodes to
1804          * compensate, so that the "inodes in use" number is correct.
1805          */
1806         if (obd_osfs.os_ffree < osfs->os_ffree) {
1807                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1808                         obd_osfs.os_ffree;
1809                 osfs->os_ffree = obd_osfs.os_ffree;
1810         }
1811
1812         RETURN(rc);
1813 }
1814 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1815 {
1816         struct super_block *sb = de->d_sb;
1817         struct obd_statfs osfs;
1818         __u64 fsid = huge_encode_dev(sb->s_dev);
1819         int rc;
1820
1821         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1822         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1823
1824         /* Some amount of caching on the client is allowed */
1825         rc = ll_statfs_internal(sb, &osfs,
1826                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1827                                 0);
1828         if (rc)
1829                 return rc;
1830
1831         statfs_unpack(sfs, &osfs);
1832
1833         /* We need to downshift for all 32-bit kernels, because we can't
1834          * tell if the kernel is being called via sys_statfs64() or not.
1835          * Stop before overflowing f_bsize - in which case it is better
1836          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1837         if (sizeof(long) < 8) {
1838                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1839                         sfs->f_bsize <<= 1;
1840
1841                         osfs.os_blocks >>= 1;
1842                         osfs.os_bfree >>= 1;
1843                         osfs.os_bavail >>= 1;
1844                 }
1845         }
1846
1847         sfs->f_blocks = osfs.os_blocks;
1848         sfs->f_bfree = osfs.os_bfree;
1849         sfs->f_bavail = osfs.os_bavail;
1850         sfs->f_fsid.val[0] = (__u32)fsid;
1851         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1852         return 0;
1853 }
1854
1855 void ll_inode_size_lock(struct inode *inode)
1856 {
1857         struct ll_inode_info *lli;
1858
1859         LASSERT(!S_ISDIR(inode->i_mode));
1860
1861         lli = ll_i2info(inode);
1862         mutex_lock(&lli->lli_size_mutex);
1863 }
1864
1865 void ll_inode_size_unlock(struct inode *inode)
1866 {
1867         struct ll_inode_info *lli;
1868
1869         lli = ll_i2info(inode);
1870         mutex_unlock(&lli->lli_size_mutex);
1871 }
1872
1873 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1874 {
1875         struct ll_inode_info *lli = ll_i2info(inode);
1876         struct mdt_body *body = md->body;
1877         struct ll_sb_info *sbi = ll_i2sbi(inode);
1878
1879         if (body->mbo_valid & OBD_MD_FLEASIZE)
1880                 cl_file_inode_init(inode, md);
1881
1882         if (S_ISDIR(inode->i_mode)) {
1883                 int     rc;
1884
1885                 rc = ll_update_lsm_md(inode, md);
1886                 if (rc != 0)
1887                         return rc;
1888         }
1889
1890         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1891                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1892                         ll_update_remote_perm(inode, md->remote_perm);
1893         }
1894 #ifdef CONFIG_FS_POSIX_ACL
1895         else if (body->mbo_valid & OBD_MD_FLACL) {
1896                 spin_lock(&lli->lli_lock);
1897                 if (lli->lli_posix_acl)
1898                         posix_acl_release(lli->lli_posix_acl);
1899                 lli->lli_posix_acl = md->posix_acl;
1900                 spin_unlock(&lli->lli_lock);
1901         }
1902 #endif
1903         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1904                                         sbi->ll_flags & LL_SBI_32BIT_API);
1905         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1906
1907         if (body->mbo_valid & OBD_MD_FLATIME) {
1908                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1909                         LTIME_S(inode->i_atime) = body->mbo_atime;
1910                 lli->lli_atime = body->mbo_atime;
1911         }
1912
1913         if (body->mbo_valid & OBD_MD_FLMTIME) {
1914                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1915                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1916                                "to "LPU64"\n", inode->i_ino,
1917                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1918                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1919                 }
1920                 lli->lli_mtime = body->mbo_mtime;
1921         }
1922
1923         if (body->mbo_valid & OBD_MD_FLCTIME) {
1924                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1925                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1926                 lli->lli_ctime = body->mbo_ctime;
1927         }
1928
1929         if (body->mbo_valid & OBD_MD_FLMODE)
1930                 inode->i_mode = (inode->i_mode & S_IFMT) |
1931                                 (body->mbo_mode & ~S_IFMT);
1932
1933         if (body->mbo_valid & OBD_MD_FLTYPE)
1934                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1935                                 (body->mbo_mode & S_IFMT);
1936
1937         LASSERT(inode->i_mode != 0);
1938         if (S_ISREG(inode->i_mode))
1939                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1940                                        LL_MAX_BLKSIZE_BITS);
1941         else
1942                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1943
1944         if (body->mbo_valid & OBD_MD_FLUID)
1945                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1946         if (body->mbo_valid & OBD_MD_FLGID)
1947                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1948         if (body->mbo_valid & OBD_MD_FLFLAGS)
1949                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1950         if (body->mbo_valid & OBD_MD_FLNLINK)
1951                 set_nlink(inode, body->mbo_nlink);
1952         if (body->mbo_valid & OBD_MD_FLRDEV)
1953                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1954
1955         if (body->mbo_valid & OBD_MD_FLID) {
1956                 /* FID shouldn't be changed! */
1957                 if (fid_is_sane(&lli->lli_fid)) {
1958                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1959                                  "Trying to change FID "DFID
1960                                  " to the "DFID", inode "DFID"(%p)\n",
1961                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1962                                  PFID(ll_inode2fid(inode)), inode);
1963                 } else {
1964                         lli->lli_fid = body->mbo_fid1;
1965                 }
1966         }
1967
1968         LASSERT(fid_seq(&lli->lli_fid) != 0);
1969
1970         if (body->mbo_valid & OBD_MD_FLSIZE) {
1971                 i_size_write(inode, body->mbo_size);
1972
1973                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1974                        PFID(ll_inode2fid(inode)),
1975                        (unsigned long long)body->mbo_size);
1976
1977                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1978                         inode->i_blocks = body->mbo_blocks;
1979         }
1980
1981         if (body->mbo_valid & OBD_MD_TSTATE) {
1982                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1983                  * clear it when done to ensure to start again
1984                  * glimpsing updated attrs
1985                  */
1986                 if (body->mbo_t_state & MS_RESTORE)
1987                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1988                 else
1989                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1990         }
1991
1992         return 0;
1993 }
1994
1995 int ll_read_inode2(struct inode *inode, void *opaque)
1996 {
1997         struct lustre_md *md = opaque;
1998         struct ll_inode_info *lli = ll_i2info(inode);
1999         int     rc;
2000         ENTRY;
2001
2002         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2003                PFID(&lli->lli_fid), inode);
2004
2005         /* Core attributes from the MDS first.  This is a new inode, and
2006          * the VFS doesn't zero times in the core inode so we have to do
2007          * it ourselves.  They will be overwritten by either MDS or OST
2008          * attributes - we just need to make sure they aren't newer. */
2009         LTIME_S(inode->i_mtime) = 0;
2010         LTIME_S(inode->i_atime) = 0;
2011         LTIME_S(inode->i_ctime) = 0;
2012         inode->i_rdev = 0;
2013         rc = ll_update_inode(inode, md);
2014         if (rc != 0)
2015                 RETURN(rc);
2016
2017         /* OIDEBUG(inode); */
2018
2019 #ifdef HAVE_BACKING_DEV_INFO
2020         /* initializing backing dev info. */
2021         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2022 #endif
2023         if (S_ISREG(inode->i_mode)) {
2024                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2025                 inode->i_op = &ll_file_inode_operations;
2026                 inode->i_fop = sbi->ll_fop;
2027                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2028                 EXIT;
2029         } else if (S_ISDIR(inode->i_mode)) {
2030                 inode->i_op = &ll_dir_inode_operations;
2031                 inode->i_fop = &ll_dir_operations;
2032                 EXIT;
2033         } else if (S_ISLNK(inode->i_mode)) {
2034                 inode->i_op = &ll_fast_symlink_inode_operations;
2035                 EXIT;
2036         } else {
2037                 inode->i_op = &ll_special_inode_operations;
2038
2039                 init_special_inode(inode, inode->i_mode,
2040                                    inode->i_rdev);
2041
2042                 EXIT;
2043         }
2044
2045         return 0;
2046 }
2047
2048 void ll_delete_inode(struct inode *inode)
2049 {
2050         struct ll_inode_info *lli = ll_i2info(inode);
2051         ENTRY;
2052
2053         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2054                 /* It is last chance to write out dirty pages,
2055                  * otherwise we may lose data while umount */
2056                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2057
2058         truncate_inode_pages_final(&inode->i_data);
2059
2060         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2061                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2062                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2063
2064 #ifdef HAVE_SBOPS_EVICT_INODE
2065         ll_clear_inode(inode);
2066 #endif
2067         clear_inode(inode);
2068
2069         EXIT;
2070 }
2071
2072 int ll_iocontrol(struct inode *inode, struct file *file,
2073                  unsigned int cmd, unsigned long arg)
2074 {
2075         struct ll_sb_info *sbi = ll_i2sbi(inode);
2076         struct ptlrpc_request *req = NULL;
2077         int rc, flags = 0;
2078         ENTRY;
2079
2080         switch(cmd) {
2081         case FSFILT_IOC_GETFLAGS: {
2082                 struct mdt_body *body;
2083                 struct md_op_data *op_data;
2084
2085                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2086                                              0, 0, LUSTRE_OPC_ANY,
2087                                              NULL);
2088                 if (IS_ERR(op_data))
2089                         RETURN(PTR_ERR(op_data));
2090
2091                 op_data->op_valid = OBD_MD_FLFLAGS;
2092                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2093                 ll_finish_md_op_data(op_data);
2094                 if (rc) {
2095                         CERROR("%s: failure inode "DFID": rc = %d\n",
2096                                sbi->ll_md_exp->exp_obd->obd_name,
2097                                PFID(ll_inode2fid(inode)), rc);
2098                         RETURN(-abs(rc));
2099                 }
2100
2101                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2102
2103                 flags = body->mbo_flags;
2104
2105                 ptlrpc_req_finished(req);
2106
2107                 RETURN(put_user(flags, (int __user *)arg));
2108         }
2109         case FSFILT_IOC_SETFLAGS: {
2110                 struct iattr *attr;
2111                 struct md_op_data *op_data;
2112                 struct cl_object *obj;
2113
2114                 if (get_user(flags, (int __user *)arg))
2115                         RETURN(-EFAULT);
2116
2117                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2118                                              LUSTRE_OPC_ANY, NULL);
2119                 if (IS_ERR(op_data))
2120                         RETURN(PTR_ERR(op_data));
2121
2122                 op_data->op_attr_flags = flags;
2123                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2124                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2125                 ll_finish_md_op_data(op_data);
2126                 ptlrpc_req_finished(req);
2127                 if (rc)
2128                         RETURN(rc);
2129
2130                 inode->i_flags = ll_ext_to_inode_flags(flags);
2131
2132                 obj = ll_i2info(inode)->lli_clob;
2133                 if (obj == NULL)
2134                         RETURN(0);
2135
2136                 OBD_ALLOC_PTR(attr);
2137                 if (attr == NULL)
2138                         RETURN(-ENOMEM);
2139
2140                 attr->ia_valid = ATTR_ATTR_FLAG;
2141                 rc = cl_setattr_ost(obj, attr, flags);
2142
2143                 OBD_FREE_PTR(attr);
2144                 RETURN(rc);
2145         }
2146         default:
2147                 RETURN(-ENOSYS);
2148         }
2149
2150         RETURN(0);
2151 }
2152
2153 int ll_flush_ctx(struct inode *inode)
2154 {
2155         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2156
2157         CDEBUG(D_SEC, "flush context for user %d\n",
2158                from_kuid(&init_user_ns, current_uid()));
2159
2160         obd_set_info_async(NULL, sbi->ll_md_exp,
2161                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2162                            0, NULL, NULL);
2163         obd_set_info_async(NULL, sbi->ll_dt_exp,
2164                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2165                            0, NULL, NULL);
2166         return 0;
2167 }
2168
2169 /* umount -f client means force down, don't save state */
2170 void ll_umount_begin(struct super_block *sb)
2171 {
2172         struct ll_sb_info *sbi = ll_s2sbi(sb);
2173         struct obd_device *obd;
2174         struct obd_ioctl_data *ioc_data;
2175         ENTRY;
2176
2177         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2178                sb->s_count, atomic_read(&sb->s_active));
2179
2180         obd = class_exp2obd(sbi->ll_md_exp);
2181         if (obd == NULL) {
2182                 CERROR("Invalid MDC connection handle "LPX64"\n",
2183                        sbi->ll_md_exp->exp_handle.h_cookie);
2184                 EXIT;
2185                 return;
2186         }
2187         obd->obd_force = 1;
2188
2189         obd = class_exp2obd(sbi->ll_dt_exp);
2190         if (obd == NULL) {
2191                 CERROR("Invalid LOV connection handle "LPX64"\n",
2192                        sbi->ll_dt_exp->exp_handle.h_cookie);
2193                 EXIT;
2194                 return;
2195         }
2196         obd->obd_force = 1;
2197
2198         OBD_ALLOC_PTR(ioc_data);
2199         if (ioc_data) {
2200                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2201                               sizeof *ioc_data, ioc_data, NULL);
2202
2203                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2204                               sizeof *ioc_data, ioc_data, NULL);
2205
2206                 OBD_FREE_PTR(ioc_data);
2207         }
2208
2209         /* Really, we'd like to wait until there are no requests outstanding,
2210          * and then continue.  For now, we just invalidate the requests,
2211          * schedule() and sleep one second if needed, and hope.
2212          */
2213         schedule();
2214         EXIT;
2215 }
2216
2217 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2218 {
2219         struct ll_sb_info *sbi = ll_s2sbi(sb);
2220         char *profilenm = get_profile_name(sb);
2221         int err;
2222         __u32 read_only;
2223
2224         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2225                 read_only = *flags & MS_RDONLY;
2226                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2227                                          sizeof(KEY_READ_ONLY),
2228                                          KEY_READ_ONLY, sizeof(read_only),
2229                                          &read_only, NULL);
2230                 if (err) {
2231                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2232                                       profilenm, read_only ?
2233                                       "read-only" : "read-write", err);
2234                         return err;
2235                 }
2236
2237                 if (read_only)
2238                         sb->s_flags |= MS_RDONLY;
2239                 else
2240                         sb->s_flags &= ~MS_RDONLY;
2241
2242                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2243                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2244                                       read_only ?  "read-only" : "read-write");
2245         }
2246         return 0;
2247 }
2248
2249 /**
2250  * Cleanup the open handle that is cached on MDT-side.
2251  *
2252  * For open case, the client side open handling thread may hit error
2253  * after the MDT grant the open. Under such case, the client should
2254  * send close RPC to the MDT as cleanup; otherwise, the open handle
2255  * on the MDT will be leaked there until the client umount or evicted.
2256  *
2257  * In further, if someone unlinked the file, because the open handle
2258  * holds the reference on such file/object, then it will block the
2259  * subsequent threads that want to locate such object via FID.
2260  *
2261  * \param[in] sb        super block for this file-system
2262  * \param[in] open_req  pointer to the original open request
2263  */
2264 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2265 {
2266         struct mdt_body                 *body;
2267         struct md_op_data               *op_data;
2268         struct ptlrpc_request           *close_req = NULL;
2269         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2270         ENTRY;
2271
2272         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2273         OBD_ALLOC_PTR(op_data);
2274         if (op_data == NULL) {
2275                 CWARN("%s: cannot allocate op_data to release open handle for "
2276                       DFID"\n",
2277                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2278
2279                 RETURN_EXIT;
2280         }
2281
2282         op_data->op_fid1 = body->mbo_fid1;
2283         op_data->op_handle = body->mbo_handle;
2284         op_data->op_mod_time = cfs_time_current_sec();
2285         md_close(exp, op_data, NULL, &close_req);
2286         ptlrpc_req_finished(close_req);
2287         ll_finish_md_op_data(op_data);
2288
2289         EXIT;
2290 }
2291
2292 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2293                   struct super_block *sb, struct lookup_intent *it)
2294 {
2295         struct ll_sb_info *sbi = NULL;
2296         struct lustre_md md = { NULL };
2297         int rc;
2298         ENTRY;
2299
2300         LASSERT(*inode || sb);
2301         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2302         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2303                               sbi->ll_md_exp, &md);
2304         if (rc != 0)
2305                 GOTO(cleanup, rc);
2306
2307         if (*inode) {
2308                 rc = ll_update_inode(*inode, &md);
2309                 if (rc != 0)
2310                         GOTO(out, rc);
2311         } else {
2312                 LASSERT(sb != NULL);
2313
2314                 /*
2315                  * At this point server returns to client's same fid as client
2316                  * generated for creating. So using ->fid1 is okay here.
2317                  */
2318                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2319
2320                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2321                                              sbi->ll_flags & LL_SBI_32BIT_API),
2322                                  &md);
2323                 if (IS_ERR(*inode)) {
2324 #ifdef CONFIG_FS_POSIX_ACL
2325                         if (md.posix_acl) {
2326                                 posix_acl_release(md.posix_acl);
2327                                 md.posix_acl = NULL;
2328                         }
2329 #endif
2330                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2331                         *inode = NULL;
2332                         CERROR("new_inode -fatal: rc %d\n", rc);
2333                         GOTO(out, rc);
2334                 }
2335         }
2336
2337         /* Handling piggyback layout lock.
2338          * Layout lock can be piggybacked by getattr and open request.
2339          * The lsm can be applied to inode only if it comes with a layout lock
2340          * otherwise correct layout may be overwritten, for example:
2341          * 1. proc1: mdt returns a lsm but not granting layout
2342          * 2. layout was changed by another client
2343          * 3. proc2: refresh layout and layout lock granted
2344          * 4. proc1: to apply a stale layout */
2345         if (it != NULL && it->it_lock_mode != 0) {
2346                 struct lustre_handle lockh;
2347                 struct ldlm_lock *lock;
2348
2349                 lockh.cookie = it->it_lock_handle;
2350                 lock = ldlm_handle2lock(&lockh);
2351                 LASSERT(lock != NULL);
2352                 if (ldlm_has_layout(lock)) {
2353                         struct cl_object_conf conf;
2354
2355                         memset(&conf, 0, sizeof(conf));
2356                         conf.coc_opc = OBJECT_CONF_SET;
2357                         conf.coc_inode = *inode;
2358                         conf.coc_lock = lock;
2359                         conf.u.coc_layout = md.layout;
2360                         (void)ll_layout_conf(*inode, &conf);
2361                 }
2362                 LDLM_LOCK_PUT(lock);
2363         }
2364
2365         GOTO(out, rc = 0);
2366
2367 out:
2368         md_free_lustre_md(sbi->ll_md_exp, &md);
2369
2370 cleanup:
2371         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2372                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2373
2374         return rc;
2375 }
2376
2377 int ll_obd_statfs(struct inode *inode, void __user *arg)
2378 {
2379         struct ll_sb_info *sbi = NULL;
2380         struct obd_export *exp;
2381         char *buf = NULL;
2382         struct obd_ioctl_data *data = NULL;
2383         __u32 type;
2384         int len = 0, rc;
2385
2386         if (!inode || !(sbi = ll_i2sbi(inode)))
2387                 GOTO(out_statfs, rc = -EINVAL);
2388
2389         rc = obd_ioctl_getdata(&buf, &len, arg);
2390         if (rc)
2391                 GOTO(out_statfs, rc);
2392
2393         data = (void*)buf;
2394         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2395             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2396                 GOTO(out_statfs, rc = -EINVAL);
2397
2398         if (data->ioc_inllen1 != sizeof(__u32) ||
2399             data->ioc_inllen2 != sizeof(__u32) ||
2400             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2401             data->ioc_plen2 != sizeof(struct obd_uuid))
2402                 GOTO(out_statfs, rc = -EINVAL);
2403
2404         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2405         if (type & LL_STATFS_LMV)
2406                 exp = sbi->ll_md_exp;
2407         else if (type & LL_STATFS_LOV)
2408                 exp = sbi->ll_dt_exp;
2409         else
2410                 GOTO(out_statfs, rc = -ENODEV);
2411
2412         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2413         if (rc)
2414                 GOTO(out_statfs, rc);
2415 out_statfs:
2416         if (buf)
2417                 obd_ioctl_freedata(buf, len);
2418         return rc;
2419 }
2420
2421 int ll_process_config(struct lustre_cfg *lcfg)
2422 {
2423         struct super_block *sb;
2424         unsigned long x;
2425         int rc = 0;
2426         char *ptr;
2427
2428         /* The instance name contains the sb: lustre-client-aacfe000 */
2429         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2430         if (!ptr || !*(++ptr))
2431                 return -EINVAL;
2432         if (sscanf(ptr, "%lx", &x) != 1)
2433                 return -EINVAL;
2434         sb = (struct super_block *)x;
2435         /* This better be a real Lustre superblock! */
2436         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2437
2438         /* Note we have not called client_common_fill_super yet, so
2439            proc fns must be able to handle that! */
2440         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2441                                       lcfg, sb);
2442         if (rc > 0)
2443                 rc = 0;
2444         return rc;
2445 }
2446
2447 /* this function prepares md_op_data hint for passing it down to MD stack. */
2448 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2449                                       struct inode *i1, struct inode *i2,
2450                                       const char *name, size_t namelen,
2451                                       __u32 mode, __u32 opc, void *data)
2452 {
2453         LASSERT(i1 != NULL);
2454
2455         if (name == NULL) {
2456                 /* Do not reuse namelen for something else. */
2457                 if (namelen != 0)
2458                         return ERR_PTR(-EINVAL);
2459         } else {
2460                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2461                         return ERR_PTR(-ENAMETOOLONG);
2462
2463                 if (!lu_name_is_valid_2(name, namelen))
2464                         return ERR_PTR(-EINVAL);
2465         }
2466
2467         if (op_data == NULL)
2468                 OBD_ALLOC_PTR(op_data);
2469
2470         if (op_data == NULL)
2471                 return ERR_PTR(-ENOMEM);
2472
2473         ll_i2gids(op_data->op_suppgids, i1, i2);
2474         op_data->op_fid1 = *ll_inode2fid(i1);
2475         op_data->op_default_stripe_offset = -1;
2476         if (S_ISDIR(i1->i_mode)) {
2477                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2478                 if (opc == LUSTRE_OPC_MKDIR)
2479                         op_data->op_default_stripe_offset =
2480                                    ll_i2info(i1)->lli_def_stripe_offset;
2481         }
2482
2483         if (i2) {
2484                 op_data->op_fid2 = *ll_inode2fid(i2);
2485                 if (S_ISDIR(i2->i_mode))
2486                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2487         } else {
2488                 fid_zero(&op_data->op_fid2);
2489         }
2490
2491         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2492                 op_data->op_cli_flags |= CLI_HASH64;
2493
2494         if (ll_need_32bit_api(ll_i2sbi(i1)))
2495                 op_data->op_cli_flags |= CLI_API32;
2496
2497         op_data->op_name = name;
2498         op_data->op_namelen = namelen;
2499         op_data->op_mode = mode;
2500         op_data->op_mod_time = cfs_time_current_sec();
2501         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2502         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2503         op_data->op_cap = cfs_curproc_cap_pack();
2504         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2505              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2506                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2507         } else {
2508                 op_data->op_mds = 0;
2509         }
2510         op_data->op_data = data;
2511
2512         return op_data;
2513 }
2514
2515 void ll_finish_md_op_data(struct md_op_data *op_data)
2516 {
2517         OBD_FREE_PTR(op_data);
2518 }
2519
2520 #ifdef HAVE_SUPEROPS_USE_DENTRY
2521 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2522 #else
2523 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2524 #endif
2525 {
2526         struct ll_sb_info *sbi;
2527
2528 #ifdef HAVE_SUPEROPS_USE_DENTRY
2529         LASSERT((seq != NULL) && (dentry != NULL));
2530         sbi = ll_s2sbi(dentry->d_sb);
2531 #else
2532         LASSERT((seq != NULL) && (vfs != NULL));
2533         sbi = ll_s2sbi(vfs->mnt_sb);
2534 #endif
2535
2536         if (sbi->ll_flags & LL_SBI_NOLCK)
2537                 seq_puts(seq, ",nolock");
2538
2539         if (sbi->ll_flags & LL_SBI_FLOCK)
2540                 seq_puts(seq, ",flock");
2541
2542         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2543                 seq_puts(seq, ",localflock");
2544
2545         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2546                 seq_puts(seq, ",user_xattr");
2547
2548         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2549                 seq_puts(seq, ",lazystatfs");
2550
2551         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2552                 seq_puts(seq, ",user_fid2path");
2553
2554         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2555                 seq_puts(seq, ",always_ping");
2556
2557         RETURN(0);
2558 }
2559
2560 /**
2561  * Get obd name by cmd, and copy out to user space
2562  */
2563 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2564 {
2565         struct ll_sb_info *sbi = ll_i2sbi(inode);
2566         struct obd_device *obd;
2567         ENTRY;
2568
2569         if (cmd == OBD_IOC_GETDTNAME)
2570                 obd = class_exp2obd(sbi->ll_dt_exp);
2571         else if (cmd == OBD_IOC_GETMDNAME)
2572                 obd = class_exp2obd(sbi->ll_md_exp);
2573         else
2574                 RETURN(-EINVAL);
2575
2576         if (!obd)
2577                 RETURN(-ENOENT);
2578
2579         if (copy_to_user((void __user *)arg, obd->obd_name,
2580                          strlen(obd->obd_name) + 1))
2581                 RETURN(-EFAULT);
2582
2583         RETURN(0);
2584 }
2585
2586 /**
2587  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2588  * fsname will be returned in this buffer; otherwise, a static buffer will be
2589  * used to store the fsname and returned to caller.
2590  */
2591 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2592 {
2593         static char fsname_static[MTI_NAME_MAXLEN];
2594         struct lustre_sb_info *lsi = s2lsi(sb);
2595         char *ptr;
2596         int len;
2597
2598         if (buf == NULL) {
2599                 /* this means the caller wants to use static buffer
2600                  * and it doesn't care about race. Usually this is
2601                  * in error reporting path */
2602                 buf = fsname_static;
2603                 buflen = sizeof(fsname_static);
2604         }
2605
2606         len = strlen(lsi->lsi_lmd->lmd_profile);
2607         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2608         if (ptr && (strcmp(ptr, "-client") == 0))
2609                 len -= 7;
2610
2611         if (unlikely(len >= buflen))
2612                 len = buflen - 1;
2613         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2614         buf[len] = '\0';
2615
2616         return buf;
2617 }
2618
2619 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2620 {
2621         char *path = NULL;
2622
2623         struct path p;
2624
2625         p.dentry = dentry;
2626         p.mnt = current->fs->root.mnt;
2627         path_get(&p);
2628         path = d_path(&p, buf, bufsize);
2629         path_put(&p);
2630         return path;
2631 }
2632
2633 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2634 {
2635         char *buf, *path = NULL;
2636         struct dentry *dentry = NULL;
2637         struct inode *inode = page->mapping->host;
2638
2639         /* this can be called inside spin lock so use GFP_ATOMIC. */
2640         buf = (char *)__get_free_page(GFP_ATOMIC);
2641         if (buf != NULL) {
2642                 dentry = d_find_alias(page->mapping->host);
2643                 if (dentry != NULL)
2644                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2645         }
2646
2647         CDEBUG(D_WARNING,
2648                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2649                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2650                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2651                PFID(ll_inode2fid(inode)),
2652                (path && !IS_ERR(path)) ? path : "", ioret);
2653
2654         if (dentry != NULL)
2655                 dput(dentry);
2656
2657         if (buf != NULL)
2658                 free_page((unsigned long)buf);
2659 }
2660
2661 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2662                         struct lov_user_md **kbuf)
2663 {
2664         struct lov_user_md      lum;
2665         ssize_t                 lum_size;
2666         ENTRY;
2667
2668         if (copy_from_user(&lum, md, sizeof(lum)))
2669                 RETURN(-EFAULT);
2670
2671         lum_size = ll_lov_user_md_size(&lum);
2672         if (lum_size < 0)
2673                 RETURN(lum_size);
2674
2675         OBD_ALLOC(*kbuf, lum_size);
2676         if (*kbuf == NULL)
2677                 RETURN(-ENOMEM);
2678
2679         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2680                 OBD_FREE(*kbuf, lum_size);
2681                 RETURN(-EFAULT);
2682         }
2683
2684         RETURN(lum_size);
2685 }
2686
2687 /*
2688  * Compute llite root squash state after a change of root squash
2689  * configuration setting or add/remove of a lnet nid
2690  */
2691 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2692 {
2693         struct root_squash_info *squash = &sbi->ll_squash;
2694         int i;
2695         bool matched;
2696         lnet_process_id_t id;
2697
2698         /* Update norootsquash flag */
2699         down_write(&squash->rsi_sem);
2700         if (list_empty(&squash->rsi_nosquash_nids))
2701                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2702         else {
2703                 /* Do not apply root squash as soon as one of our NIDs is
2704                  * in the nosquash_nids list */
2705                 matched = false;
2706                 i = 0;
2707                 while (LNetGetId(i++, &id) != -ENOENT) {
2708                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2709                                 continue;
2710                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2711                                 matched = true;
2712                                 break;
2713                         }
2714                 }
2715                 if (matched)
2716                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2717                 else
2718                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2719         }
2720         up_write(&squash->rsi_sem);
2721 }
2722
2723 /**
2724  * Parse linkea content to extract information about a given hardlink
2725  *
2726  * \param[in]   ldata      - Initialized linkea data
2727  * \param[in]   linkno     - Link identifier
2728  * \param[out]  parent_fid - The entry's parent FID
2729  * \param[out]  ln         - Entry name destination buffer
2730  *
2731  * \retval 0 on success
2732  * \retval Appropriate negative error code on failure
2733  */
2734 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2735                             struct lu_fid *parent_fid, struct lu_name *ln)
2736 {
2737         unsigned int    idx;
2738         int             rc;
2739         ENTRY;
2740
2741         rc = linkea_init(ldata);
2742         if (rc < 0)
2743                 RETURN(rc);
2744
2745         if (linkno >= ldata->ld_leh->leh_reccount)
2746                 /* beyond last link */
2747                 RETURN(-ENODATA);
2748
2749         linkea_first_entry(ldata);
2750         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2751                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2752                                     parent_fid);
2753                 if (idx == linkno)
2754                         break;
2755
2756                 linkea_next_entry(ldata);
2757         }
2758
2759         if (idx < linkno)
2760                 RETURN(-ENODATA);
2761
2762         RETURN(0);
2763 }
2764
2765 /**
2766  * Get parent FID and name of an identified link. Operation is performed for
2767  * a given link number, letting the caller iterate over linkno to list one or
2768  * all links of an entry.
2769  *
2770  * \param[in]     file - File descriptor against which to perform the operation
2771  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2772  *                       on and the available size. It is eventually filled with
2773  *                       the requested information or left untouched on error
2774  *
2775  * \retval - 0 on success
2776  * \retval - Appropriate negative error code on failure
2777  */
2778 int ll_getparent(struct file *file, struct getparent __user *arg)
2779 {
2780         struct dentry           *dentry = file->f_path.dentry;
2781         struct inode            *inode = dentry->d_inode;
2782         struct linkea_data      *ldata;
2783         struct lu_buf            buf = LU_BUF_NULL;
2784         struct lu_name           ln;
2785         struct lu_fid            parent_fid;
2786         __u32                    linkno;
2787         __u32                    name_size;
2788         int                      rc;
2789
2790         ENTRY;
2791
2792         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2793             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2794                 RETURN(-EPERM);
2795
2796         if (get_user(name_size, &arg->gp_name_size))
2797                 RETURN(-EFAULT);
2798
2799         if (get_user(linkno, &arg->gp_linkno))
2800                 RETURN(-EFAULT);
2801
2802         if (name_size > PATH_MAX)
2803                 RETURN(-EINVAL);
2804
2805         OBD_ALLOC(ldata, sizeof(*ldata));
2806         if (ldata == NULL)
2807                 RETURN(-ENOMEM);
2808
2809         rc = linkea_data_new(ldata, &buf);
2810         if (rc < 0)
2811                 GOTO(ldata_free, rc);
2812
2813         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2814         if (rc < 0)
2815                 GOTO(lb_free, rc);
2816
2817         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2818         if (rc < 0)
2819                 GOTO(lb_free, rc);
2820
2821         if (ln.ln_namelen >= name_size)
2822                 GOTO(lb_free, rc = -EOVERFLOW);
2823
2824         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2825                 GOTO(lb_free, rc = -EFAULT);
2826
2827         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2828                 GOTO(lb_free, rc = -EFAULT);
2829
2830         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2831                 GOTO(lb_free, rc = -EFAULT);
2832
2833 lb_free:
2834         lu_buf_free(&buf);
2835 ldata_free:
2836         OBD_FREE(ldata, sizeof(*ldata));
2837
2838         RETURN(rc);
2839 }