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
 108         ll_generate_random_uuid(uuid);
 109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 111
 112         sbi->ll_flags |= LL_SBI_VERBOSE;
 113 #ifdef ENABLE_CHECKSUM
 114         sbi->ll_flags |= LL_SBI_CHECKSUM;
 115 #endif
 116
 117 #ifdef HAVE_LRU_RESIZE_SUPPORT
 118         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 119 #endif
 120
 121         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_r_hist.oh_lock);
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_w_hist.oh_lock);
 126         }
 127
 128         /* metadata statahead is enabled by default */
 129         sbi->ll_sa_max = LL_SA_RPC_DEF;
 130         atomic_set(&sbi->ll_sa_total, 0);
 131         atomic_set(&sbi->ll_sa_wrong, 0);
 132         atomic_set(&sbi->ll_sa_running, 0);
 133         atomic_set(&sbi->ll_agl_total, 0);
 134         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 135         sbi->ll_flags |= LL_SBI_FAST_READ;
 136
 137         /* root squash */
 138         sbi->ll_squash.rsi_uid = 0;
 139         sbi->ll_squash.rsi_gid = 0;
 140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 141         init_rwsem(&sbi->ll_squash.rsi_sem);
 142
 143         RETURN(sbi);
 144 }
 145
 146 static void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 154                 if (sbi->ll_cache != NULL) {
 155                         cl_cache_decref(sbi->ll_cache);
 156                         sbi->ll_cache = NULL;
 157                 }
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 164                                     struct vfsmount *mnt)
 165 {
 166         struct inode *root = NULL;
 167         struct ll_sb_info *sbi = ll_s2sbi(sb);
 168         struct obd_device *obd;
 169         struct obd_statfs *osfs = NULL;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         u64 valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         OBD_ALLOC_PTR(osfs);
 190         if (osfs == NULL) {
 191                 OBD_FREE_PTR(data);
 192                 RETURN(-ENOMEM);
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 204                                   OBD_CONNECT_EINPROGRESS |
 205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 207                                   OBD_CONNECT_MAX_EASIZE |
 208                                   OBD_CONNECT_FLOCK_DEAD |
 209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 210                                   OBD_CONNECT_OPEN_BY_FID |
 211                                   OBD_CONNECT_DIR_STRIPE |
 212                                   OBD_CONNECT_BULK_MBITS |
 213                                   OBD_CONNECT_SUBTREE |
 214                                   OBD_CONNECT_FLAGS2;
 215
 216         data->ocd_connect_flags2 = 0;
 217
 218 #ifdef HAVE_LRU_RESIZE_SUPPORT
 219         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 220                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 221 #endif
 222 #ifdef CONFIG_FS_POSIX_ACL
 223         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 224 #endif
 225
 226         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 227                 /* flag mdc connection as lightweight, only used for test
 228                  * purpose, use with care */
 229                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 230
 231         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 232         data->ocd_version = LUSTRE_VERSION_CODE;
 233
 234         if (sb->s_flags & MS_RDONLY)
 235                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 236         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 237                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 238
 239         if (sbi->ll_flags & LL_SBI_FLOCK)
 240                 sbi->ll_fop = &ll_file_operations_flock;
 241         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 242                 sbi->ll_fop = &ll_file_operations;
 243         else
 244                 sbi->ll_fop = &ll_file_operations_noflock;
 245
 246         /* real client */
 247         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 248         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 249                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 250
 251         /* always ping even if server suppress_pings */
 252         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 253                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 254
 255         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 256
 257         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 258         if (err == -EBUSY) {
 259                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 260                                    "recovery, of which this client is not a "
 261                                    "part. Please wait for recovery to complete,"
 262                                    " abort, or time out.\n", md);
 263                 GOTO(out, err);
 264         } else if (err) {
 265                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 266                 GOTO(out, err);
 267         }
 268
 269         sbi->ll_md_exp->exp_connect_data = *data;
 270
 271         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 272                            LUSTRE_SEQ_METADATA);
 273         if (err) {
 274                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 275                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 276                 GOTO(out_md, err);
 277         }
 278
 279         /* For mount, we only need fs info from MDT0, and also in DNE, it
 280          * can make sure the client can be mounted as long as MDT0 is
 281          * avaible */
 282         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 283                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 284                         OBD_STATFS_FOR_MDT0);
 285         if (err)
 286                 GOTO(out_md_fid, err);
 287
 288         /* This needs to be after statfs to ensure connect has finished.
 289          * Note that "data" does NOT contain the valid connect reply.
 290          * If connecting to a 1.8 server there will be no LMV device, so
 291          * we can access the MDC export directly and exp_connect_flags will
 292          * be non-zero, but if accessing an upgraded 2.1 server it will
 293          * have the correct flags filled in.
 294          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 295         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 296         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 297             valid != CLIENT_CONNECT_MDT_REQD) {
 298                 char *buf;
 299
 300                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 301                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 302                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 303                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 304                                    "feature(s) needed for correct operation "
 305                                    "of this client (%s). Please upgrade "
 306                                    "server or downgrade client.\n",
 307                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 308                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 309                 GOTO(out_md_fid, err = -EPROTO);
 310         }
 311
 312         size = sizeof(*data);
 313         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 314                            KEY_CONN_DATA,  &size, data);
 315         if (err) {
 316                 CERROR("%s: Get connect data failed: rc = %d\n",
 317                        sbi->ll_md_exp->exp_obd->obd_name, err);
 318                 GOTO(out_md_fid, err);
 319         }
 320
 321         LASSERT(osfs->os_bsize);
 322         sb->s_blocksize = osfs->os_bsize;
 323         sb->s_blocksize_bits = log2(osfs->os_bsize);
 324         sb->s_magic = LL_SUPER_MAGIC;
 325         sb->s_maxbytes = MAX_LFS_FILESIZE;
 326         sbi->ll_namelen = osfs->os_namelen;
 327
 328         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 329             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 330                 LCONSOLE_INFO("Disabling user_xattr feature because "
 331                               "it is not supported on the server\n");
 332                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 333         }
 334
 335         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 336 #ifdef MS_POSIXACL
 337                 sb->s_flags |= MS_POSIXACL;
 338 #endif
 339                 sbi->ll_flags |= LL_SBI_ACL;
 340         } else {
 341                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 342 #ifdef MS_POSIXACL
 343                 sb->s_flags &= ~MS_POSIXACL;
 344 #endif
 345                 sbi->ll_flags &= ~LL_SBI_ACL;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 349                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 350                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 351                         LCONSOLE_INFO("client is set as remote by default.\n");
 352                 }
 353         } else {
 354                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 355                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 356                         LCONSOLE_INFO("client claims to be remote, but server "
 357                                       "rejected, forced to be local.\n");
 358                 }
 359         }
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 362                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 365                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 366         else
 367                 sbi->ll_md_brw_pages = 1;
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 370                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 371
 372         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 373                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 374                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 375                                       "unknown maximum xattr size.\n", dt);
 376                 } else {
 377                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 378                         sbi->ll_xattr_cache_enabled = 1;
 379                 }
 380         }
 381
 382         obd = class_name2obd(dt);
 383         if (!obd) {
 384                 CERROR("DT %s: not setup or attached\n", dt);
 385                 GOTO(out_md_fid, err = -ENODEV);
 386         }
 387
 388         /* pass client page size via ocd_grant_blkbits, the server should report
 389          * back its backend blocksize for grant calculation purpose */
 390         data->ocd_grant_blkbits = PAGE_SHIFT;
 391
 392         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 393                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 394                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 395                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 396                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 397                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 398                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 399                                   OBD_CONNECT_MAXBYTES |
 400                                   OBD_CONNECT_EINPROGRESS |
 401                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 402                                   OBD_CONNECT_LAYOUTLOCK |
 403                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 404                                   OBD_CONNECT_BULK_MBITS;
 405
 406         data->ocd_connect_flags2 = 0;
 407
 408         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 409                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 410
 411         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 412                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 413                  * disabled by default, because it can still be enabled on the
 414                  * fly via /proc. As a consequence, we still need to come to an
 415                  * agreement on the supported algorithms at connect time */
 416                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 417
 418                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 419                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 420                 else
 421                         data->ocd_cksum_types = cksum_types_supported_client();
 422         }
 423
 424 #ifdef HAVE_LRU_RESIZE_SUPPORT
 425         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 426 #endif
 427         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 428                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 429
 430         /* always ping even if server suppress_pings */
 431         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 432                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 433
 434         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 435                "ocd_grant: %d\n", data->ocd_connect_flags,
 436                data->ocd_version, data->ocd_grant);
 437
 438         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 439         obd->obd_upcall.onu_upcall = cl_ocd_update;
 440
 441         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 442
 443         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 444                           NULL);
 445         if (err == -EBUSY) {
 446                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 447                                    "recovery, of which this client is not a "
 448                                    "part.  Please wait for recovery to "
 449                                    "complete, abort, or time out.\n", dt);
 450                 GOTO(out_md, err);
 451         } else if (err) {
 452                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 453                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 454                 GOTO(out_md, err);
 455         }
 456
 457         sbi->ll_dt_exp->exp_connect_data = *data;
 458
 459         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 460                            LUSTRE_SEQ_METADATA);
 461         if (err) {
 462                 CERROR("%s: Can't init data layer FID infrastructure, "
 463                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 464                 GOTO(out_dt, err);
 465         }
 466
 467         mutex_lock(&sbi->ll_lco.lco_lock);
 468         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 469         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 470         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 471         mutex_unlock(&sbi->ll_lco.lco_lock);
 472
 473         fid_zero(&sbi->ll_root_fid);
 474         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 475                            &sbi->ll_root_fid);
 476         if (err) {
 477                 CERROR("cannot mds_connect: rc = %d\n", err);
 478                 GOTO(out_lock_cn_cb, err);
 479         }
 480         if (!fid_is_sane(&sbi->ll_root_fid)) {
 481                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 482                        sbi->ll_md_exp->exp_obd->obd_name,
 483                        PFID(&sbi->ll_root_fid));
 484                 GOTO(out_lock_cn_cb, err = -EINVAL);
 485         }
 486         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 487
 488         sb->s_op = &lustre_super_operations;
 489 #if THREAD_SIZE >= 8192 /*b=17630*/
 490         sb->s_export_op = &lustre_export_operations;
 491 #endif
 492
 493         /* make root inode
 494          * XXX: move this to after cbd setup? */
 495         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 496         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 497                 valid |= OBD_MD_FLRMTPERM;
 498         else if (sbi->ll_flags & LL_SBI_ACL)
 499                 valid |= OBD_MD_FLACL;
 500
 501         OBD_ALLOC_PTR(op_data);
 502         if (op_data == NULL)
 503                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 504
 505         op_data->op_fid1 = sbi->ll_root_fid;
 506         op_data->op_mode = 0;
 507         op_data->op_valid = valid;
 508
 509         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 510
 511         OBD_FREE_PTR(op_data);
 512         if (err) {
 513                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 514                        sbi->ll_md_exp->exp_obd->obd_name, err);
 515                 GOTO(out_lock_cn_cb, err);
 516         }
 517
 518         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 519                                sbi->ll_md_exp, &lmd);
 520         if (err) {
 521                 CERROR("failed to understand root inode md: rc = %d\n", err);
 522                 ptlrpc_req_finished(request);
 523                 GOTO(out_lock_cn_cb, err);
 524         }
 525
 526         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 527         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 528                                             sbi->ll_flags & LL_SBI_32BIT_API),
 529                        &lmd);
 530         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 531         ptlrpc_req_finished(request);
 532
 533         if (IS_ERR(root)) {
 534 #ifdef CONFIG_FS_POSIX_ACL
 535                 if (lmd.posix_acl) {
 536                         posix_acl_release(lmd.posix_acl);
 537                         lmd.posix_acl = NULL;
 538                 }
 539 #endif
 540                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 541                 root = NULL;
 542                 CERROR("lustre_lite: bad iget4 for root\n");
 543                 GOTO(out_root, err);
 544         }
 545
 546 #ifdef CONFIG_FS_POSIX_ACL
 547         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 548                 rct_init(&sbi->ll_rct);
 549                 et_init(&sbi->ll_et);
 550         }
 551 #endif
 552
 553         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 554         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 555                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 556                                  NULL);
 557         cl_sb_init(sb);
 558
 559         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 560                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 561                                  sbi->ll_cache, NULL);
 562
 563         sb->s_root = d_make_root(root);
 564         if (sb->s_root == NULL) {
 565                 CERROR("%s: can't make root dentry\n",
 566                         ll_get_fsname(sb, NULL, 0));
 567                 GOTO(out_root, err = -ENOMEM);
 568         }
 569 #ifdef HAVE_DCACHE_LOCK
 570         sb->s_root->d_op = &ll_d_ops;
 571 #endif
 572
 573         sbi->ll_sdev_orig = sb->s_dev;
 574
 575         /* We set sb->s_dev equal on all lustre clients in order to support
 576          * NFS export clustering.  NFSD requires that the FSID be the same
 577          * on all clients. */
 578         /* s_dev is also used in lt_compare() to compare two fs, but that is
 579          * only a node-local comparison. */
 580         uuid = obd_get_uuid(sbi->ll_md_exp);
 581         if (uuid != NULL)
 582                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 583
 584         if (data != NULL)
 585                 OBD_FREE_PTR(data);
 586         if (osfs != NULL)
 587                 OBD_FREE_PTR(osfs);
 588         if (proc_lustre_fs_root != NULL) {
 589                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 590                                                   dt, md);
 591                 if (err < 0) {
 592                         CERROR("%s: could not register mount in lprocfs: "
 593                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 594                         err = 0;
 595                 }
 596         }
 597
 598         RETURN(err);
 599 out_root:
 600         if (root)
 601                 iput(root);
 602 out_lock_cn_cb:
 603         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 604 out_dt:
 605         obd_disconnect(sbi->ll_dt_exp);
 606         sbi->ll_dt_exp = NULL;
 607 out_md_fid:
 608         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 609 out_md:
 610         obd_disconnect(sbi->ll_md_exp);
 611         sbi->ll_md_exp = NULL;
 612 out:
 613         if (data != NULL)
 614                 OBD_FREE_PTR(data);
 615         if (osfs != NULL)
 616                 OBD_FREE_PTR(osfs);
 617         return err;
 618 }
 619
 620 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 621 {
 622         int size, rc;
 623
 624         size = sizeof(*lmmsize);
 625         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 626                           KEY_MAX_EASIZE, &size, lmmsize);
 627         if (rc != 0) {
 628                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 629                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 630                 RETURN(rc);
 631         }
 632
 633         size = sizeof(int);
 634         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 635                           KEY_MAX_EASIZE, &size, lmmsize);
 636         if (rc)
 637                 CERROR("Get max mdsize error rc %d\n", rc);
 638
 639         RETURN(rc);
 640 }
 641
 642 /**
 643  * Get the value of the default_easize parameter.
 644  *
 645  * \see client_obd::cl_default_mds_easize
 646  *
 647  * \param[in] sbi       superblock info for this filesystem
 648  * \param[out] lmmsize  pointer to storage location for value
 649  *
 650  * \retval 0            on success
 651  * \retval negative     negated errno on failure
 652  */
 653 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 654 {
 655         int size, rc;
 656
 657         size = sizeof(int);
 658         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 659                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 660         if (rc)
 661                 CERROR("Get default mdsize error rc %d\n", rc);
 662
 663         RETURN(rc);
 664 }
 665
 666 /**
 667  * Set the default_easize parameter to the given value.
 668  *
 669  * \see client_obd::cl_default_mds_easize
 670  *
 671  * \param[in] sbi       superblock info for this filesystem
 672  * \param[in] lmmsize   the size to set
 673  *
 674  * \retval 0            on success
 675  * \retval negative     negated errno on failure
 676  */
 677 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 678 {
 679         int rc;
 680
 681         if (lmmsize < sizeof(struct lov_mds_md) ||
 682             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 683                 return -EINVAL;
 684
 685         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 686                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 687                                 sizeof(int), &lmmsize, NULL);
 688
 689         RETURN(rc);
 690 }
 691
 692 static void ll_dump_inode(struct inode *inode)
 693 {
 694         struct ll_d_hlist_node *tmp;
 695         int dentry_count = 0;
 696
 697         LASSERT(inode != NULL);
 698
 699         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 700                 dentry_count++;
 701
 702         CERROR("%s: inode %p dump: dev=%s fid="DFID
 703                " mode=%o count=%u, %d dentries\n",
 704                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 705                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 706                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 707 }
 708
 709 void lustre_dump_dentry(struct dentry *dentry, int recur)
 710 {
 711         struct list_head *tmp;
 712         int subdirs = 0;
 713
 714         LASSERT(dentry != NULL);
 715
 716         list_for_each(tmp, &dentry->d_subdirs)
 717                 subdirs++;
 718
 719         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 720                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 721                dentry->d_name.len, dentry->d_name.name,
 722                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 723                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 724                dentry->d_flags, dentry->d_fsdata, subdirs);
 725         if (dentry->d_inode != NULL)
 726                 ll_dump_inode(dentry->d_inode);
 727
 728         if (recur == 0)
 729                 return;
 730
 731         list_for_each(tmp, &dentry->d_subdirs) {
 732                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 733                 lustre_dump_dentry(d, recur - 1);
 734         }
 735 }
 736
 737 static void client_common_put_super(struct super_block *sb)
 738 {
 739         struct ll_sb_info *sbi = ll_s2sbi(sb);
 740         ENTRY;
 741
 742 #ifdef CONFIG_FS_POSIX_ACL
 743         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 744                 et_fini(&sbi->ll_et);
 745                 rct_fini(&sbi->ll_rct);
 746         }
 747 #endif
 748
 749         cl_sb_fini(sb);
 750
 751         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 752         obd_disconnect(sbi->ll_dt_exp);
 753         sbi->ll_dt_exp = NULL;
 754
 755         lprocfs_unregister_mountpoint(sbi);
 756
 757         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 758         obd_disconnect(sbi->ll_md_exp);
 759         sbi->ll_md_exp = NULL;
 760
 761         EXIT;
 762 }
 763
 764 void ll_kill_super(struct super_block *sb)
 765 {
 766         struct ll_sb_info *sbi;
 767         ENTRY;
 768
 769         /* not init sb ?*/
 770         if (!(sb->s_flags & MS_ACTIVE))
 771                 return;
 772
 773         sbi = ll_s2sbi(sb);
 774         /* we need restore s_dev from changed for clustred NFS before put_super
 775          * because new kernels have cached s_dev and change sb->s_dev in
 776          * put_super not affected real removing devices */
 777         if (sbi) {
 778                 sb->s_dev = sbi->ll_sdev_orig;
 779                 sbi->ll_umounting = 1;
 780
 781                 /* wait running statahead threads to quit */
 782                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 783                         set_current_state(TASK_UNINTERRUPTIBLE);
 784                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 785                 }
 786         }
 787
 788         EXIT;
 789 }
 790
 791 static inline int ll_set_opt(const char *opt, char *data, int fl)
 792 {
 793         if (strncmp(opt, data, strlen(opt)) != 0)
 794                 return(0);
 795         else
 796                 return(fl);
 797 }
 798
 799 /* non-client-specific mount options are parsed in lmd_parse */
 800 static int ll_options(char *options, int *flags)
 801 {
 802         int tmp;
 803         char *s1 = options, *s2;
 804         ENTRY;
 805
 806         if (!options)
 807                 RETURN(0);
 808
 809         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 810
 811         while (*s1) {
 812                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 813                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 829                 if (tmp) {
 830                         *flags &= ~tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags |= tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 839                 if (tmp) {
 840                         *flags &= ~tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("context", s1, 1);
 844                 if (tmp)
 845                         goto next;
 846                 tmp = ll_set_opt("fscontext", s1, 1);
 847                 if (tmp)
 848                         goto next;
 849                 tmp = ll_set_opt("defcontext", s1, 1);
 850                 if (tmp)
 851                         goto next;
 852                 tmp = ll_set_opt("rootcontext", s1, 1);
 853                 if (tmp)
 854                         goto next;
 855                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags |= tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 866                 if (tmp) {
 867                         *flags &= ~tmp;
 868                         goto next;
 869                 }
 870
 871                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 872                 if (tmp) {
 873                         *flags |= tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 877                 if (tmp) {
 878                         *flags &= ~tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 882                 if (tmp) {
 883                         *flags |= tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 887                 if (tmp) {
 888                         *flags &= ~tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 897                 if (tmp) {
 898                         *flags &= ~tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 907                 if (tmp) {
 908                         *flags |= tmp;
 909                         goto next;
 910                 }
 911                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 912                 if (tmp) {
 913                         *flags &= ~tmp;
 914                         goto next;
 915                 }
 916                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 917                 if (tmp) {
 918                         *flags |= tmp;
 919                         goto next;
 920                 }
 921                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 922                                    s1);
 923                 RETURN(-EINVAL);
 924
 925 next:
 926                 /* Find next opt */
 927                 s2 = strchr(s1, ',');
 928                 if (s2 == NULL)
 929                         break;
 930                 s1 = s2 + 1;
 931         }
 932         RETURN(0);
 933 }
 934
 935 void ll_lli_init(struct ll_inode_info *lli)
 936 {
 937         lli->lli_inode_magic = LLI_INODE_MAGIC;
 938         lli->lli_flags = 0;
 939         spin_lock_init(&lli->lli_lock);
 940         lli->lli_posix_acl = NULL;
 941         lli->lli_remote_perms = NULL;
 942         mutex_init(&lli->lli_rmtperm_mutex);
 943         /* Do not set lli_fid, it has been initialized already. */
 944         fid_zero(&lli->lli_pfid);
 945         atomic_set(&lli->lli_open_count, 0);
 946         lli->lli_rmtperm_time = 0;
 947         lli->lli_mds_read_och = NULL;
 948         lli->lli_mds_write_och = NULL;
 949         lli->lli_mds_exec_och = NULL;
 950         lli->lli_open_fd_read_count = 0;
 951         lli->lli_open_fd_write_count = 0;
 952         lli->lli_open_fd_exec_count = 0;
 953         mutex_init(&lli->lli_och_mutex);
 954         spin_lock_init(&lli->lli_agl_lock);
 955         spin_lock_init(&lli->lli_layout_lock);
 956         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 957         lli->lli_clob = NULL;
 958
 959         init_rwsem(&lli->lli_xattrs_list_rwsem);
 960         mutex_init(&lli->lli_xattrs_enq_lock);
 961
 962         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 963         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 964                 mutex_init(&lli->lli_readdir_mutex);
 965                 lli->lli_opendir_key = NULL;
 966                 lli->lli_sai = NULL;
 967                 spin_lock_init(&lli->lli_sa_lock);
 968                 lli->lli_opendir_pid = 0;
 969                 lli->lli_sa_enabled = 0;
 970                 lli->lli_def_stripe_offset = -1;
 971         } else {
 972                 mutex_init(&lli->lli_size_mutex);
 973                 lli->lli_symlink_name = NULL;
 974                 init_rwsem(&lli->lli_trunc_sem);
 975                 range_lock_tree_init(&lli->lli_write_tree);
 976                 init_rwsem(&lli->lli_glimpse_sem);
 977                 lli->lli_glimpse_time = 0;
 978                 INIT_LIST_HEAD(&lli->lli_agl_list);
 979                 lli->lli_agl_index = 0;
 980                 lli->lli_async_rc = 0;
 981         }
 982         mutex_init(&lli->lli_layout_mutex);
 983 }
 984
 985 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 986 {
 987         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 988
 989         bdi->name = "lustre";
 990         return bdi_register(bdi, NULL, "lustre-%d",
 991                             atomic_inc_return(&ll_bdi_num));
 992 }
 993
 994 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 995 {
 996         struct lustre_profile *lprof = NULL;
 997         struct lustre_sb_info *lsi = s2lsi(sb);
 998         struct ll_sb_info *sbi;
 999         char  *dt = NULL, *md = NULL;
1000         char  *profilenm = get_profile_name(sb);
1001         struct config_llog_instance *cfg;
1002         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1003         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1004         int    err;
1005         ENTRY;
1006
1007         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1008
1009         OBD_ALLOC_PTR(cfg);
1010         if (cfg == NULL)
1011                 RETURN(-ENOMEM);
1012
1013         try_module_get(THIS_MODULE);
1014
1015         /* client additional sb info */
1016         lsi->lsi_llsbi = sbi = ll_init_sbi();
1017         if (!sbi) {
1018                 module_put(THIS_MODULE);
1019                 OBD_FREE_PTR(cfg);
1020                 RETURN(-ENOMEM);
1021         }
1022
1023         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1024         if (err)
1025                 GOTO(out_free, err);
1026
1027         err = bdi_init(&lsi->lsi_bdi);
1028         if (err)
1029                 GOTO(out_free, err);
1030         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1031 #ifdef HAVE_BDI_CAP_MAP_COPY
1032         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1033 #else
1034         lsi->lsi_bdi.capabilities = 0;
1035 #endif
1036         err = ll_bdi_register(&lsi->lsi_bdi);
1037         if (err)
1038                 GOTO(out_free, err);
1039
1040         sb->s_bdi = &lsi->lsi_bdi;
1041 #ifndef HAVE_DCACHE_LOCK
1042         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1043         sb->s_d_op = &ll_d_ops;
1044 #endif
1045
1046         /* Generate a string unique to this super, in case some joker tries
1047            to mount the same fs at two mount points.
1048            Use the address of the super itself.*/
1049         cfg->cfg_instance = sb;
1050         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1051         cfg->cfg_callback = class_config_llog_handler;
1052         /* set up client obds */
1053         err = lustre_process_log(sb, profilenm, cfg);
1054         if (err < 0)
1055                 GOTO(out_free, err);
1056
1057         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1058         lprof = class_get_profile(profilenm);
1059         if (lprof == NULL) {
1060                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1061                                    " read from the MGS.  Does that filesystem "
1062                                    "exist?\n", profilenm);
1063                 GOTO(out_free, err = -EINVAL);
1064         }
1065         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1066                lprof->lp_md, lprof->lp_dt);
1067
1068         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1069         if (!dt)
1070                 GOTO(out_free, err = -ENOMEM);
1071         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1072
1073         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1074         if (!md)
1075                 GOTO(out_free, err = -ENOMEM);
1076         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1077
1078         /* connections, registrations, sb setup */
1079         err = client_common_fill_super(sb, md, dt, mnt);
1080         if (err < 0)
1081                 GOTO(out_free, err);
1082
1083         sbi->ll_client_common_fill_super_succeeded = 1;
1084
1085 out_free:
1086         if (md)
1087                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1088         if (dt)
1089                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1090         if (lprof != NULL)
1091                 class_put_profile(lprof);
1092         if (err)
1093                 ll_put_super(sb);
1094         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1095                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1096
1097         OBD_FREE_PTR(cfg);
1098         RETURN(err);
1099 } /* ll_fill_super */
1100
1101 void ll_put_super(struct super_block *sb)
1102 {
1103         struct config_llog_instance cfg, params_cfg;
1104         struct obd_device *obd;
1105         struct lustre_sb_info *lsi = s2lsi(sb);
1106         struct ll_sb_info *sbi = ll_s2sbi(sb);
1107         char *profilenm = get_profile_name(sb);
1108         long ccc_count;
1109         int next, force = 1, rc = 0;
1110         ENTRY;
1111
1112         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1113
1114         cfg.cfg_instance = sb;
1115         lustre_end_log(sb, profilenm, &cfg);
1116
1117         params_cfg.cfg_instance = sb;
1118         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1119
1120         if (sbi->ll_md_exp) {
1121                 obd = class_exp2obd(sbi->ll_md_exp);
1122                 if (obd)
1123                         force = obd->obd_force;
1124         }
1125
1126         /* Wait for unstable pages to be committed to stable storage */
1127         if (force == 0) {
1128                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1129                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1130                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1131                         &lwi);
1132         }
1133
1134         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1135         if (force == 0 && rc != -EINTR)
1136                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1137
1138
1139         /* We need to set force before the lov_disconnect in
1140            lustre_common_put_super, since l_d cleans up osc's as well. */
1141         if (force) {
1142                 next = 0;
1143                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1144                                                      &next)) != NULL) {
1145                         obd->obd_force = force;
1146                 }
1147         }
1148
1149         if (sbi->ll_client_common_fill_super_succeeded) {
1150                 /* Only if client_common_fill_super succeeded */
1151                 client_common_put_super(sb);
1152         }
1153
1154         next = 0;
1155         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1156                 class_manual_cleanup(obd);
1157         }
1158
1159         if (sbi->ll_flags & LL_SBI_VERBOSE)
1160                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1161
1162         if (profilenm)
1163                 class_del_profile(profilenm);
1164
1165         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1166                 bdi_destroy(&lsi->lsi_bdi);
1167                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1168         }
1169
1170         ll_free_sbi(sb);
1171         lsi->lsi_llsbi = NULL;
1172
1173         lustre_common_put_super(sb);
1174
1175         cl_env_cache_purge(~0);
1176
1177         module_put(THIS_MODULE);
1178
1179         EXIT;
1180 } /* client_put_super */
1181
1182 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1183 {
1184         struct inode *inode = NULL;
1185
1186         /* NOTE: we depend on atomic igrab() -bzzz */
1187         lock_res_and_lock(lock);
1188         if (lock->l_resource->lr_lvb_inode) {
1189                 struct ll_inode_info * lli;
1190                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1191                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1192                         inode = igrab(lock->l_resource->lr_lvb_inode);
1193                 } else {
1194                         inode = lock->l_resource->lr_lvb_inode;
1195                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1196                                          D_WARNING, lock, "lr_lvb_inode %p is "
1197                                          "bogus: magic %08x",
1198                                          lock->l_resource->lr_lvb_inode,
1199                                          lli->lli_inode_magic);
1200                         inode = NULL;
1201                 }
1202         }
1203         unlock_res_and_lock(lock);
1204         return inode;
1205 }
1206
1207 void ll_dir_clear_lsm_md(struct inode *inode)
1208 {
1209         struct ll_inode_info *lli = ll_i2info(inode);
1210
1211         LASSERT(S_ISDIR(inode->i_mode));
1212
1213         if (lli->lli_lsm_md != NULL) {
1214                 lmv_free_memmd(lli->lli_lsm_md);
1215                 lli->lli_lsm_md = NULL;
1216         }
1217 }
1218
1219 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1220                                       const struct lu_fid *fid,
1221                                       struct lustre_md *md)
1222 {
1223         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1224         struct mdt_body         *body = md->body;
1225         struct inode            *inode;
1226         ino_t                   ino;
1227         ENTRY;
1228
1229         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1230         inode = iget_locked(sb, ino);
1231         if (inode == NULL) {
1232                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1233                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1234                 RETURN(ERR_PTR(-ENOENT));
1235         }
1236
1237         if (inode->i_state & I_NEW) {
1238                 struct ll_inode_info *lli = ll_i2info(inode);
1239                 struct lmv_stripe_md *lsm = md->lmv;
1240
1241                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1242                                 (body->mbo_mode & S_IFMT);
1243                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1244                          PFID(fid));
1245
1246                 LTIME_S(inode->i_mtime) = 0;
1247                 LTIME_S(inode->i_atime) = 0;
1248                 LTIME_S(inode->i_ctime) = 0;
1249                 inode->i_rdev = 0;
1250
1251 #ifdef HAVE_BACKING_DEV_INFO
1252                 /* initializing backing dev info. */
1253                 inode->i_mapping->backing_dev_info =
1254                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1255 #endif
1256                 inode->i_op = &ll_dir_inode_operations;
1257                 inode->i_fop = &ll_dir_operations;
1258                 lli->lli_fid = *fid;
1259                 ll_lli_init(lli);
1260
1261                 LASSERT(lsm != NULL);
1262                 /* master object FID */
1263                 lli->lli_pfid = body->mbo_fid1;
1264                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1265                        lli, PFID(fid), PFID(&lli->lli_pfid));
1266                 unlock_new_inode(inode);
1267         }
1268
1269         RETURN(inode);
1270 }
1271
1272 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1273 {
1274         struct lu_fid *fid;
1275         struct lmv_stripe_md *lsm = md->lmv;
1276         int i;
1277
1278         LASSERT(lsm != NULL);
1279         /* XXX sigh, this lsm_root initialization should be in
1280          * LMV layer, but it needs ll_iget right now, so we
1281          * put this here right now. */
1282         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1283                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1284                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1285                 /* Unfortunately ll_iget will call ll_update_inode,
1286                  * where the initialization of slave inode is slightly
1287                  * different, so it reset lsm_md to NULL to avoid
1288                  * initializing lsm for slave inode. */
1289                 /* For migrating inode, master stripe and master object will
1290                  * be same, so we only need assign this inode */
1291                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1292                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1293                 else
1294                         lsm->lsm_md_oinfo[i].lmo_root =
1295                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1296
1297                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1298                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1299
1300                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1301                         return rc;
1302                 }
1303         }
1304
1305         return 0;
1306 }
1307
1308 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1309                                 const struct lmv_stripe_md *lsm_md2)
1310 {
1311         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1312                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1313                lsm_md1->lsm_md_master_mdt_index ==
1314                                         lsm_md2->lsm_md_master_mdt_index &&
1315                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1316                lsm_md1->lsm_md_layout_version ==
1317                                         lsm_md2->lsm_md_layout_version &&
1318                strcmp(lsm_md1->lsm_md_pool_name,
1319                       lsm_md2->lsm_md_pool_name) == 0;
1320 }
1321
1322 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1323 {
1324         struct ll_inode_info *lli = ll_i2info(inode);
1325         struct lmv_stripe_md *lsm = md->lmv;
1326         int     rc;
1327         ENTRY;
1328
1329         LASSERT(S_ISDIR(inode->i_mode));
1330         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1331                PFID(ll_inode2fid(inode)));
1332
1333         /* no striped information from request. */
1334         if (lsm == NULL) {
1335                 if (lli->lli_lsm_md == NULL) {
1336                         RETURN(0);
1337                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1338                                                 LMV_HASH_FLAG_MIGRATION) {
1339                         /* migration is done, the temporay MIGRATE layout has
1340                          * been removed */
1341                         CDEBUG(D_INODE, DFID" finish migration.\n",
1342                                PFID(ll_inode2fid(inode)));
1343                         lmv_free_memmd(lli->lli_lsm_md);
1344                         lli->lli_lsm_md = NULL;
1345                         RETURN(0);
1346                 } else {
1347                         /* The lustre_md from req does not include stripeEA,
1348                          * see ll_md_setattr */
1349                         RETURN(0);
1350                 }
1351         }
1352
1353         /* set the directory layout */
1354         if (lli->lli_lsm_md == NULL) {
1355                 struct cl_attr  *attr;
1356
1357                 rc = ll_init_lsm_md(inode, md);
1358                 if (rc != 0)
1359                         RETURN(rc);
1360
1361                 /* set md->lmv to NULL, so the following free lustre_md
1362                  * will not free this lsm */
1363                 md->lmv = NULL;
1364                 lli->lli_lsm_md = lsm;
1365
1366                 OBD_ALLOC_PTR(attr);
1367                 if (attr == NULL)
1368                         RETURN(-ENOMEM);
1369
1370                 /* validate the lsm */
1371                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1372                                    ll_md_blocking_ast);
1373                 if (rc != 0) {
1374                         OBD_FREE_PTR(attr);
1375                         RETURN(rc);
1376                 }
1377
1378                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1379                         md->body->mbo_nlink = attr->cat_nlink;
1380                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1381                         md->body->mbo_size = attr->cat_size;
1382                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1383                         md->body->mbo_atime = attr->cat_atime;
1384                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1385                         md->body->mbo_ctime = attr->cat_ctime;
1386                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1387                         md->body->mbo_mtime = attr->cat_mtime;
1388
1389                 OBD_FREE_PTR(attr);
1390
1391                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1392                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1393                 RETURN(0);
1394         }
1395
1396         /* Compare the old and new stripe information */
1397         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1398                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1399                 int                     idx;
1400
1401                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1402                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1403                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1404                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1405                        inode, lsm, old_lsm,
1406                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1407                        lsm->lsm_md_stripe_count,
1408                        old_lsm->lsm_md_stripe_count,
1409                        lsm->lsm_md_master_mdt_index,
1410                        old_lsm->lsm_md_master_mdt_index,
1411                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1412                        lsm->lsm_md_layout_version,
1413                        old_lsm->lsm_md_layout_version,
1414                        lsm->lsm_md_pool_name,
1415                        old_lsm->lsm_md_pool_name);
1416
1417                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1418                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1419                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1420                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1421                 }
1422
1423                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1424                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1425                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1426                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1427                 }
1428
1429                 RETURN(-EIO);
1430         }
1431
1432         RETURN(0);
1433 }
1434
1435 void ll_clear_inode(struct inode *inode)
1436 {
1437         struct ll_inode_info *lli = ll_i2info(inode);
1438         struct ll_sb_info *sbi = ll_i2sbi(inode);
1439         ENTRY;
1440
1441         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1442                PFID(ll_inode2fid(inode)), inode);
1443
1444         if (S_ISDIR(inode->i_mode)) {
1445                 /* these should have been cleared in ll_file_release */
1446                 LASSERT(lli->lli_opendir_key == NULL);
1447                 LASSERT(lli->lli_sai == NULL);
1448                 LASSERT(lli->lli_opendir_pid == 0);
1449         }
1450
1451         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1452
1453         LASSERT(!lli->lli_open_fd_write_count);
1454         LASSERT(!lli->lli_open_fd_read_count);
1455         LASSERT(!lli->lli_open_fd_exec_count);
1456
1457         if (lli->lli_mds_write_och)
1458                 ll_md_real_close(inode, FMODE_WRITE);
1459         if (lli->lli_mds_exec_och)
1460                 ll_md_real_close(inode, FMODE_EXEC);
1461         if (lli->lli_mds_read_och)
1462                 ll_md_real_close(inode, FMODE_READ);
1463
1464         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1465                 OBD_FREE(lli->lli_symlink_name,
1466                          strlen(lli->lli_symlink_name) + 1);
1467                 lli->lli_symlink_name = NULL;
1468         }
1469
1470         ll_xattr_cache_destroy(inode);
1471
1472         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1473                 LASSERT(lli->lli_posix_acl == NULL);
1474                 if (lli->lli_remote_perms) {
1475                         free_rmtperm_hash(lli->lli_remote_perms);
1476                         lli->lli_remote_perms = NULL;
1477                 }
1478         }
1479 #ifdef CONFIG_FS_POSIX_ACL
1480         else if (lli->lli_posix_acl) {
1481                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1482                 LASSERT(lli->lli_remote_perms == NULL);
1483                 posix_acl_release(lli->lli_posix_acl);
1484                 lli->lli_posix_acl = NULL;
1485         }
1486 #endif
1487         lli->lli_inode_magic = LLI_INODE_DEAD;
1488
1489         if (S_ISDIR(inode->i_mode))
1490                 ll_dir_clear_lsm_md(inode);
1491         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1492                 LASSERT(list_empty(&lli->lli_agl_list));
1493
1494         /*
1495          * XXX This has to be done before lsm is freed below, because
1496          * cl_object still uses inode lsm.
1497          */
1498         cl_inode_fini(inode);
1499
1500         EXIT;
1501 }
1502
1503 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1504 {
1505         struct lustre_md md;
1506         struct inode *inode = dentry->d_inode;
1507         struct ll_sb_info *sbi = ll_i2sbi(inode);
1508         struct ptlrpc_request *request = NULL;
1509         int rc, ia_valid;
1510         ENTRY;
1511
1512         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1513                                      LUSTRE_OPC_ANY, NULL);
1514         if (IS_ERR(op_data))
1515                 RETURN(PTR_ERR(op_data));
1516
1517         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1518         if (rc) {
1519                 ptlrpc_req_finished(request);
1520                 if (rc == -ENOENT) {
1521                         clear_nlink(inode);
1522                         /* Unlinked special device node? Or just a race?
1523                          * Pretend we done everything. */
1524                         if (!S_ISREG(inode->i_mode) &&
1525                             !S_ISDIR(inode->i_mode)) {
1526                                 ia_valid = op_data->op_attr.ia_valid;
1527                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1528                                 rc = simple_setattr(dentry, &op_data->op_attr);
1529                                 op_data->op_attr.ia_valid = ia_valid;
1530                         }
1531                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1532                         CERROR("md_setattr fails: rc = %d\n", rc);
1533                 }
1534                 RETURN(rc);
1535         }
1536
1537         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1538                               sbi->ll_md_exp, &md);
1539         if (rc) {
1540                 ptlrpc_req_finished(request);
1541                 RETURN(rc);
1542         }
1543
1544         ia_valid = op_data->op_attr.ia_valid;
1545         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1546          * cache is not cleared yet. */
1547         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1548         if (S_ISREG(inode->i_mode))
1549                 mutex_lock(&inode->i_mutex);
1550         rc = simple_setattr(dentry, &op_data->op_attr);
1551         if (S_ISREG(inode->i_mode))
1552                 mutex_unlock(&inode->i_mutex);
1553         op_data->op_attr.ia_valid = ia_valid;
1554
1555         rc = ll_update_inode(inode, &md);
1556         ptlrpc_req_finished(request);
1557
1558         RETURN(rc);
1559 }
1560
1561 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1562  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1563  * keep these values until such a time that objects are allocated for it.
1564  * We do the MDS operations first, as it is checking permissions for us.
1565  * We don't to the MDS RPC if there is nothing that we want to store there,
1566  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1567  * going to do an RPC anyways.
1568  *
1569  * If we are doing a truncate, we will send the mtime and ctime updates
1570  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1571  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1572  * at the same time.
1573  *
1574  * In case of HSMimport, we only set attr on MDS.
1575  */
1576 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1577 {
1578         struct inode *inode = dentry->d_inode;
1579         struct ll_inode_info *lli = ll_i2info(inode);
1580         struct md_op_data *op_data = NULL;
1581         int rc = 0;
1582         ENTRY;
1583
1584         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1585                "valid %x, hsm_import %d\n",
1586                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1587                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1588                hsm_import);
1589
1590         if (attr->ia_valid & ATTR_SIZE) {
1591                 /* Check new size against VFS/VM file size limit and rlimit */
1592                 rc = inode_newsize_ok(inode, attr->ia_size);
1593                 if (rc)
1594                         RETURN(rc);
1595
1596                 /* The maximum Lustre file size is variable, based on the
1597                  * OST maximum object size and number of stripes.  This
1598                  * needs another check in addition to the VFS check above. */
1599                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1600                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1601                                PFID(&lli->lli_fid), attr->ia_size,
1602                                ll_file_maxbytes(inode));
1603                         RETURN(-EFBIG);
1604                 }
1605
1606                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1607         }
1608
1609         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1610         if (attr->ia_valid & TIMES_SET_FLAGS) {
1611                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1612                     !cfs_capable(CFS_CAP_FOWNER))
1613                         RETURN(-EPERM);
1614         }
1615
1616         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1617         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1618             (attr->ia_valid & ATTR_CTIME)) {
1619                 attr->ia_ctime = CURRENT_TIME;
1620                 attr->ia_valid |= ATTR_CTIME_SET;
1621         }
1622         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1623             (attr->ia_valid & ATTR_ATIME)) {
1624                 attr->ia_atime = CURRENT_TIME;
1625                 attr->ia_valid |= ATTR_ATIME_SET;
1626         }
1627         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1628             (attr->ia_valid & ATTR_MTIME)) {
1629                 attr->ia_mtime = CURRENT_TIME;
1630                 attr->ia_valid |= ATTR_MTIME_SET;
1631         }
1632
1633         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1634                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1635                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1636                        cfs_time_current_sec());
1637
1638         if (S_ISREG(inode->i_mode)) {
1639                 if (attr->ia_valid & ATTR_SIZE)
1640                         inode_dio_write_done(inode);
1641                 mutex_unlock(&inode->i_mutex);
1642         }
1643
1644         /* We always do an MDS RPC, even if we're only changing the size;
1645          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1646
1647         OBD_ALLOC_PTR(op_data);
1648         if (op_data == NULL)
1649                 GOTO(out, rc = -ENOMEM);
1650
1651         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1652                 /* If we are changing file size, file content is
1653                  * modified, flag it. */
1654                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1655                 op_data->op_bias |= MDS_DATA_MODIFIED;
1656                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1657         }
1658
1659         op_data->op_attr = *attr;
1660
1661         rc = ll_md_setattr(dentry, op_data);
1662         if (rc)
1663                 GOTO(out, rc);
1664
1665         if (!S_ISREG(inode->i_mode) || hsm_import)
1666                 GOTO(out, rc = 0);
1667
1668         if (attr->ia_valid & (ATTR_SIZE |
1669                               ATTR_ATIME | ATTR_ATIME_SET |
1670                               ATTR_MTIME | ATTR_MTIME_SET |
1671                               ATTR_CTIME | ATTR_CTIME_SET)) {
1672                 /* For truncate and utimes sending attributes to OSTs, setting
1673                  * mtime/atime to the past will be performed under PW [0:EOF]
1674                  * extent lock (new_size:EOF for truncate).  It may seem
1675                  * excessive to send mtime/atime updates to OSTs when not
1676                  * setting times to past, but it is necessary due to possible
1677                  * time de-synchronization between MDT inode and OST objects */
1678                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1679         }
1680
1681         /* If the file was restored, it needs to set dirty flag.
1682          *
1683          * We've already sent MDS_DATA_MODIFIED flag in
1684          * ll_md_setattr() for truncate. However, the MDT refuses to
1685          * set the HS_DIRTY flag on released files, so we have to set
1686          * it again if the file has been restored. Please check how
1687          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1688          *
1689          * Please notice that if the file is not released, the previous
1690          * MDS_DATA_MODIFIED has taken effect and usually
1691          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1692          * This way we can save an RPC for common open + trunc
1693          * operation. */
1694         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1695                 struct hsm_state_set hss = {
1696                         .hss_valid = HSS_SETMASK,
1697                         .hss_setmask = HS_DIRTY,
1698                 };
1699                 int rc2;
1700
1701                 rc2 = ll_hsm_state_set(inode, &hss);
1702                 /* truncate and write can happen at the same time, so that
1703                  * the file can be set modified even though the file is not
1704                  * restored from released state, and ll_hsm_state_set() is
1705                  * not applicable for the file, and rc2 < 0 is normal in this
1706                  * case. */
1707                 if (rc2 < 0)
1708                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1709                                PFID(ll_inode2fid(inode)), rc2);
1710         }
1711
1712         EXIT;
1713 out:
1714         if (op_data != NULL)
1715                 ll_finish_md_op_data(op_data);
1716
1717         if (S_ISREG(inode->i_mode)) {
1718                 mutex_lock(&inode->i_mutex);
1719                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1720                         inode_dio_wait(inode);
1721         }
1722
1723         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1724                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1725
1726         return rc;
1727 }
1728
1729 int ll_setattr(struct dentry *de, struct iattr *attr)
1730 {
1731         int mode = de->d_inode->i_mode;
1732
1733         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1734                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1735                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1736
1737         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1738                                (ATTR_SIZE|ATTR_MODE)) &&
1739             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1740              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1741               !(attr->ia_mode & S_ISGID))))
1742                 attr->ia_valid |= ATTR_FORCE;
1743
1744         if ((attr->ia_valid & ATTR_MODE) &&
1745             (mode & S_ISUID) &&
1746             !(attr->ia_mode & S_ISUID) &&
1747             !(attr->ia_valid & ATTR_KILL_SUID))
1748                 attr->ia_valid |= ATTR_KILL_SUID;
1749
1750         if ((attr->ia_valid & ATTR_MODE) &&
1751             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1752             !(attr->ia_mode & S_ISGID) &&
1753             !(attr->ia_valid & ATTR_KILL_SGID))
1754                 attr->ia_valid |= ATTR_KILL_SGID;
1755
1756         return ll_setattr_raw(de, attr, false);
1757 }
1758
1759 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1760                        __u64 max_age, __u32 flags)
1761 {
1762         struct ll_sb_info *sbi = ll_s2sbi(sb);
1763         struct obd_statfs obd_osfs;
1764         int rc;
1765         ENTRY;
1766
1767         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1768         if (rc) {
1769                 CERROR("md_statfs fails: rc = %d\n", rc);
1770                 RETURN(rc);
1771         }
1772
1773         osfs->os_type = sb->s_magic;
1774
1775         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1776                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1777
1778         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1779                 flags |= OBD_STATFS_NODELAY;
1780
1781         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1782         if (rc) {
1783                 CERROR("obd_statfs fails: rc = %d\n", rc);
1784                 RETURN(rc);
1785         }
1786
1787         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1788                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1789                obd_osfs.os_files);
1790
1791         osfs->os_bsize = obd_osfs.os_bsize;
1792         osfs->os_blocks = obd_osfs.os_blocks;
1793         osfs->os_bfree = obd_osfs.os_bfree;
1794         osfs->os_bavail = obd_osfs.os_bavail;
1795
1796         /* If we don't have as many objects free on the OST as inodes
1797          * on the MDS, we reduce the total number of inodes to
1798          * compensate, so that the "inodes in use" number is correct.
1799          */
1800         if (obd_osfs.os_ffree < osfs->os_ffree) {
1801                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1802                         obd_osfs.os_ffree;
1803                 osfs->os_ffree = obd_osfs.os_ffree;
1804         }
1805
1806         RETURN(rc);
1807 }
1808 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1809 {
1810         struct super_block *sb = de->d_sb;
1811         struct obd_statfs osfs;
1812         __u64 fsid = huge_encode_dev(sb->s_dev);
1813         int rc;
1814
1815         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1816         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1817
1818         /* Some amount of caching on the client is allowed */
1819         rc = ll_statfs_internal(sb, &osfs,
1820                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1821                                 0);
1822         if (rc)
1823                 return rc;
1824
1825         statfs_unpack(sfs, &osfs);
1826
1827         /* We need to downshift for all 32-bit kernels, because we can't
1828          * tell if the kernel is being called via sys_statfs64() or not.
1829          * Stop before overflowing f_bsize - in which case it is better
1830          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1831         if (sizeof(long) < 8) {
1832                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1833                         sfs->f_bsize <<= 1;
1834
1835                         osfs.os_blocks >>= 1;
1836                         osfs.os_bfree >>= 1;
1837                         osfs.os_bavail >>= 1;
1838                 }
1839         }
1840
1841         sfs->f_blocks = osfs.os_blocks;
1842         sfs->f_bfree = osfs.os_bfree;
1843         sfs->f_bavail = osfs.os_bavail;
1844         sfs->f_fsid.val[0] = (__u32)fsid;
1845         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1846         return 0;
1847 }
1848
1849 void ll_inode_size_lock(struct inode *inode)
1850 {
1851         struct ll_inode_info *lli;
1852
1853         LASSERT(!S_ISDIR(inode->i_mode));
1854
1855         lli = ll_i2info(inode);
1856         mutex_lock(&lli->lli_size_mutex);
1857 }
1858
1859 void ll_inode_size_unlock(struct inode *inode)
1860 {
1861         struct ll_inode_info *lli;
1862
1863         lli = ll_i2info(inode);
1864         mutex_unlock(&lli->lli_size_mutex);
1865 }
1866
1867 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1868 {
1869         struct ll_inode_info *lli = ll_i2info(inode);
1870         struct mdt_body *body = md->body;
1871         struct ll_sb_info *sbi = ll_i2sbi(inode);
1872
1873         if (body->mbo_valid & OBD_MD_FLEASIZE)
1874                 cl_file_inode_init(inode, md);
1875
1876         if (S_ISDIR(inode->i_mode)) {
1877                 int     rc;
1878
1879                 rc = ll_update_lsm_md(inode, md);
1880                 if (rc != 0)
1881                         return rc;
1882         }
1883
1884         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1885                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1886                         ll_update_remote_perm(inode, md->remote_perm);
1887         }
1888 #ifdef CONFIG_FS_POSIX_ACL
1889         else if (body->mbo_valid & OBD_MD_FLACL) {
1890                 spin_lock(&lli->lli_lock);
1891                 if (lli->lli_posix_acl)
1892                         posix_acl_release(lli->lli_posix_acl);
1893                 lli->lli_posix_acl = md->posix_acl;
1894                 spin_unlock(&lli->lli_lock);
1895         }
1896 #endif
1897         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1898                                         sbi->ll_flags & LL_SBI_32BIT_API);
1899         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1900
1901         if (body->mbo_valid & OBD_MD_FLATIME) {
1902                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1903                         LTIME_S(inode->i_atime) = body->mbo_atime;
1904                 lli->lli_atime = body->mbo_atime;
1905         }
1906
1907         if (body->mbo_valid & OBD_MD_FLMTIME) {
1908                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1909                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1910                                "to "LPU64"\n", inode->i_ino,
1911                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1912                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1913                 }
1914                 lli->lli_mtime = body->mbo_mtime;
1915         }
1916
1917         if (body->mbo_valid & OBD_MD_FLCTIME) {
1918                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1919                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1920                 lli->lli_ctime = body->mbo_ctime;
1921         }
1922
1923         if (body->mbo_valid & OBD_MD_FLMODE)
1924                 inode->i_mode = (inode->i_mode & S_IFMT) |
1925                                 (body->mbo_mode & ~S_IFMT);
1926
1927         if (body->mbo_valid & OBD_MD_FLTYPE)
1928                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1929                                 (body->mbo_mode & S_IFMT);
1930
1931         LASSERT(inode->i_mode != 0);
1932         if (S_ISREG(inode->i_mode))
1933                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1934                                        LL_MAX_BLKSIZE_BITS);
1935         else
1936                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1937
1938         if (body->mbo_valid & OBD_MD_FLUID)
1939                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1940         if (body->mbo_valid & OBD_MD_FLGID)
1941                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1942         if (body->mbo_valid & OBD_MD_FLFLAGS)
1943                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1944         if (body->mbo_valid & OBD_MD_FLNLINK)
1945                 set_nlink(inode, body->mbo_nlink);
1946         if (body->mbo_valid & OBD_MD_FLRDEV)
1947                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1948
1949         if (body->mbo_valid & OBD_MD_FLID) {
1950                 /* FID shouldn't be changed! */
1951                 if (fid_is_sane(&lli->lli_fid)) {
1952                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1953                                  "Trying to change FID "DFID
1954                                  " to the "DFID", inode "DFID"(%p)\n",
1955                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1956                                  PFID(ll_inode2fid(inode)), inode);
1957                 } else {
1958                         lli->lli_fid = body->mbo_fid1;
1959                 }
1960         }
1961
1962         LASSERT(fid_seq(&lli->lli_fid) != 0);
1963
1964         if (body->mbo_valid & OBD_MD_FLSIZE) {
1965                 i_size_write(inode, body->mbo_size);
1966
1967                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1968                        PFID(ll_inode2fid(inode)),
1969                        (unsigned long long)body->mbo_size);
1970
1971                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1972                         inode->i_blocks = body->mbo_blocks;
1973         }
1974
1975         if (body->mbo_valid & OBD_MD_TSTATE) {
1976                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1977                  * clear it when done to ensure to start again
1978                  * glimpsing updated attrs
1979                  */
1980                 if (body->mbo_t_state & MS_RESTORE)
1981                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1982                 else
1983                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1984         }
1985
1986         return 0;
1987 }
1988
1989 int ll_read_inode2(struct inode *inode, void *opaque)
1990 {
1991         struct lustre_md *md = opaque;
1992         struct ll_inode_info *lli = ll_i2info(inode);
1993         int     rc;
1994         ENTRY;
1995
1996         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1997                PFID(&lli->lli_fid), inode);
1998
1999         /* Core attributes from the MDS first.  This is a new inode, and
2000          * the VFS doesn't zero times in the core inode so we have to do
2001          * it ourselves.  They will be overwritten by either MDS or OST
2002          * attributes - we just need to make sure they aren't newer. */
2003         LTIME_S(inode->i_mtime) = 0;
2004         LTIME_S(inode->i_atime) = 0;
2005         LTIME_S(inode->i_ctime) = 0;
2006         inode->i_rdev = 0;
2007         rc = ll_update_inode(inode, md);
2008         if (rc != 0)
2009                 RETURN(rc);
2010
2011         /* OIDEBUG(inode); */
2012
2013 #ifdef HAVE_BACKING_DEV_INFO
2014         /* initializing backing dev info. */
2015         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2016 #endif
2017         if (S_ISREG(inode->i_mode)) {
2018                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2019                 inode->i_op = &ll_file_inode_operations;
2020                 inode->i_fop = sbi->ll_fop;
2021                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2022                 EXIT;
2023         } else if (S_ISDIR(inode->i_mode)) {
2024                 inode->i_op = &ll_dir_inode_operations;
2025                 inode->i_fop = &ll_dir_operations;
2026                 EXIT;
2027         } else if (S_ISLNK(inode->i_mode)) {
2028                 inode->i_op = &ll_fast_symlink_inode_operations;
2029                 EXIT;
2030         } else {
2031                 inode->i_op = &ll_special_inode_operations;
2032
2033                 init_special_inode(inode, inode->i_mode,
2034                                    inode->i_rdev);
2035
2036                 EXIT;
2037         }
2038
2039         return 0;
2040 }
2041
2042 void ll_delete_inode(struct inode *inode)
2043 {
2044         struct ll_inode_info *lli = ll_i2info(inode);
2045         ENTRY;
2046
2047         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2048                 /* It is last chance to write out dirty pages,
2049                  * otherwise we may lose data while umount */
2050                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2051
2052         truncate_inode_pages_final(&inode->i_data);
2053
2054         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2055                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2056                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2057
2058 #ifdef HAVE_SBOPS_EVICT_INODE
2059         ll_clear_inode(inode);
2060 #endif
2061         clear_inode(inode);
2062
2063         EXIT;
2064 }
2065
2066 int ll_iocontrol(struct inode *inode, struct file *file,
2067                  unsigned int cmd, unsigned long arg)
2068 {
2069         struct ll_sb_info *sbi = ll_i2sbi(inode);
2070         struct ptlrpc_request *req = NULL;
2071         int rc, flags = 0;
2072         ENTRY;
2073
2074         switch(cmd) {
2075         case FSFILT_IOC_GETFLAGS: {
2076                 struct mdt_body *body;
2077                 struct md_op_data *op_data;
2078
2079                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2080                                              0, 0, LUSTRE_OPC_ANY,
2081                                              NULL);
2082                 if (IS_ERR(op_data))
2083                         RETURN(PTR_ERR(op_data));
2084
2085                 op_data->op_valid = OBD_MD_FLFLAGS;
2086                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2087                 ll_finish_md_op_data(op_data);
2088                 if (rc) {
2089                         CERROR("%s: failure inode "DFID": rc = %d\n",
2090                                sbi->ll_md_exp->exp_obd->obd_name,
2091                                PFID(ll_inode2fid(inode)), rc);
2092                         RETURN(-abs(rc));
2093                 }
2094
2095                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2096
2097                 flags = body->mbo_flags;
2098
2099                 ptlrpc_req_finished(req);
2100
2101                 RETURN(put_user(flags, (int __user *)arg));
2102         }
2103         case FSFILT_IOC_SETFLAGS: {
2104                 struct iattr *attr;
2105                 struct md_op_data *op_data;
2106                 struct cl_object *obj;
2107
2108                 if (get_user(flags, (int __user *)arg))
2109                         RETURN(-EFAULT);
2110
2111                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2112                                              LUSTRE_OPC_ANY, NULL);
2113                 if (IS_ERR(op_data))
2114                         RETURN(PTR_ERR(op_data));
2115
2116                 op_data->op_attr_flags = flags;
2117                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2118                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2119                 ll_finish_md_op_data(op_data);
2120                 ptlrpc_req_finished(req);
2121                 if (rc)
2122                         RETURN(rc);
2123
2124                 inode->i_flags = ll_ext_to_inode_flags(flags);
2125
2126                 obj = ll_i2info(inode)->lli_clob;
2127                 if (obj == NULL)
2128                         RETURN(0);
2129
2130                 OBD_ALLOC_PTR(attr);
2131                 if (attr == NULL)
2132                         RETURN(-ENOMEM);
2133
2134                 attr->ia_valid = ATTR_ATTR_FLAG;
2135                 rc = cl_setattr_ost(obj, attr, flags);
2136
2137                 OBD_FREE_PTR(attr);
2138                 RETURN(rc);
2139         }
2140         default:
2141                 RETURN(-ENOSYS);
2142         }
2143
2144         RETURN(0);
2145 }
2146
2147 int ll_flush_ctx(struct inode *inode)
2148 {
2149         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2150
2151         CDEBUG(D_SEC, "flush context for user %d\n",
2152                from_kuid(&init_user_ns, current_uid()));
2153
2154         obd_set_info_async(NULL, sbi->ll_md_exp,
2155                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2156                            0, NULL, NULL);
2157         obd_set_info_async(NULL, sbi->ll_dt_exp,
2158                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2159                            0, NULL, NULL);
2160         return 0;
2161 }
2162
2163 /* umount -f client means force down, don't save state */
2164 void ll_umount_begin(struct super_block *sb)
2165 {
2166         struct ll_sb_info *sbi = ll_s2sbi(sb);
2167         struct obd_device *obd;
2168         struct obd_ioctl_data *ioc_data;
2169         ENTRY;
2170
2171         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2172                sb->s_count, atomic_read(&sb->s_active));
2173
2174         obd = class_exp2obd(sbi->ll_md_exp);
2175         if (obd == NULL) {
2176                 CERROR("Invalid MDC connection handle "LPX64"\n",
2177                        sbi->ll_md_exp->exp_handle.h_cookie);
2178                 EXIT;
2179                 return;
2180         }
2181         obd->obd_force = 1;
2182
2183         obd = class_exp2obd(sbi->ll_dt_exp);
2184         if (obd == NULL) {
2185                 CERROR("Invalid LOV connection handle "LPX64"\n",
2186                        sbi->ll_dt_exp->exp_handle.h_cookie);
2187                 EXIT;
2188                 return;
2189         }
2190         obd->obd_force = 1;
2191
2192         OBD_ALLOC_PTR(ioc_data);
2193         if (ioc_data) {
2194                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2195                               sizeof *ioc_data, ioc_data, NULL);
2196
2197                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2198                               sizeof *ioc_data, ioc_data, NULL);
2199
2200                 OBD_FREE_PTR(ioc_data);
2201         }
2202
2203         /* Really, we'd like to wait until there are no requests outstanding,
2204          * and then continue.  For now, we just invalidate the requests,
2205          * schedule() and sleep one second if needed, and hope.
2206          */
2207         schedule();
2208         EXIT;
2209 }
2210
2211 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2212 {
2213         struct ll_sb_info *sbi = ll_s2sbi(sb);
2214         char *profilenm = get_profile_name(sb);
2215         int err;
2216         __u32 read_only;
2217
2218         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2219                 read_only = *flags & MS_RDONLY;
2220                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2221                                          sizeof(KEY_READ_ONLY),
2222                                          KEY_READ_ONLY, sizeof(read_only),
2223                                          &read_only, NULL);
2224                 if (err) {
2225                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2226                                       profilenm, read_only ?
2227                                       "read-only" : "read-write", err);
2228                         return err;
2229                 }
2230
2231                 if (read_only)
2232                         sb->s_flags |= MS_RDONLY;
2233                 else
2234                         sb->s_flags &= ~MS_RDONLY;
2235
2236                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2237                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2238                                       read_only ?  "read-only" : "read-write");
2239         }
2240         return 0;
2241 }
2242
2243 /**
2244  * Cleanup the open handle that is cached on MDT-side.
2245  *
2246  * For open case, the client side open handling thread may hit error
2247  * after the MDT grant the open. Under such case, the client should
2248  * send close RPC to the MDT as cleanup; otherwise, the open handle
2249  * on the MDT will be leaked there until the client umount or evicted.
2250  *
2251  * In further, if someone unlinked the file, because the open handle
2252  * holds the reference on such file/object, then it will block the
2253  * subsequent threads that want to locate such object via FID.
2254  *
2255  * \param[in] sb        super block for this file-system
2256  * \param[in] open_req  pointer to the original open request
2257  */
2258 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2259 {
2260         struct mdt_body                 *body;
2261         struct md_op_data               *op_data;
2262         struct ptlrpc_request           *close_req = NULL;
2263         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2264         ENTRY;
2265
2266         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2267         OBD_ALLOC_PTR(op_data);
2268         if (op_data == NULL) {
2269                 CWARN("%s: cannot allocate op_data to release open handle for "
2270                       DFID"\n",
2271                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2272
2273                 RETURN_EXIT;
2274         }
2275
2276         op_data->op_fid1 = body->mbo_fid1;
2277         op_data->op_handle = body->mbo_handle;
2278         op_data->op_mod_time = cfs_time_current_sec();
2279         md_close(exp, op_data, NULL, &close_req);
2280         ptlrpc_req_finished(close_req);
2281         ll_finish_md_op_data(op_data);
2282
2283         EXIT;
2284 }
2285
2286 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2287                   struct super_block *sb, struct lookup_intent *it)
2288 {
2289         struct ll_sb_info *sbi = NULL;
2290         struct lustre_md md = { NULL };
2291         int rc;
2292         ENTRY;
2293
2294         LASSERT(*inode || sb);
2295         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2296         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2297                               sbi->ll_md_exp, &md);
2298         if (rc != 0)
2299                 GOTO(cleanup, rc);
2300
2301         if (*inode) {
2302                 rc = ll_update_inode(*inode, &md);
2303                 if (rc != 0)
2304                         GOTO(out, rc);
2305         } else {
2306                 LASSERT(sb != NULL);
2307
2308                 /*
2309                  * At this point server returns to client's same fid as client
2310                  * generated for creating. So using ->fid1 is okay here.
2311                  */
2312                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2313                         CERROR("%s: Fid is insane "DFID"\n",
2314                                 ll_get_fsname(sb, NULL, 0),
2315                                 PFID(&md.body->mbo_fid1));
2316                         GOTO(out, rc = -EINVAL);
2317                 }
2318
2319                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2320                                              sbi->ll_flags & LL_SBI_32BIT_API),
2321                                  &md);
2322                 if (IS_ERR(*inode)) {
2323 #ifdef CONFIG_FS_POSIX_ACL
2324                         if (md.posix_acl) {
2325                                 posix_acl_release(md.posix_acl);
2326                                 md.posix_acl = NULL;
2327                         }
2328 #endif
2329                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2330                         *inode = NULL;
2331                         CERROR("new_inode -fatal: rc %d\n", rc);
2332                         GOTO(out, rc);
2333                 }
2334         }
2335
2336         /* Handling piggyback layout lock.
2337          * Layout lock can be piggybacked by getattr and open request.
2338          * The lsm can be applied to inode only if it comes with a layout lock
2339          * otherwise correct layout may be overwritten, for example:
2340          * 1. proc1: mdt returns a lsm but not granting layout
2341          * 2. layout was changed by another client
2342          * 3. proc2: refresh layout and layout lock granted
2343          * 4. proc1: to apply a stale layout */
2344         if (it != NULL && it->it_lock_mode != 0) {
2345                 struct lustre_handle lockh;
2346                 struct ldlm_lock *lock;
2347
2348                 lockh.cookie = it->it_lock_handle;
2349                 lock = ldlm_handle2lock(&lockh);
2350                 LASSERT(lock != NULL);
2351                 if (ldlm_has_layout(lock)) {
2352                         struct cl_object_conf conf;
2353
2354                         memset(&conf, 0, sizeof(conf));
2355                         conf.coc_opc = OBJECT_CONF_SET;
2356                         conf.coc_inode = *inode;
2357                         conf.coc_lock = lock;
2358                         conf.u.coc_layout = md.layout;
2359                         (void)ll_layout_conf(*inode, &conf);
2360                 }
2361                 LDLM_LOCK_PUT(lock);
2362         }
2363
2364         GOTO(out, rc = 0);
2365
2366 out:
2367         md_free_lustre_md(sbi->ll_md_exp, &md);
2368
2369 cleanup:
2370         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2371                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2372
2373         return rc;
2374 }
2375
2376 int ll_obd_statfs(struct inode *inode, void __user *arg)
2377 {
2378         struct ll_sb_info *sbi = NULL;
2379         struct obd_export *exp;
2380         char *buf = NULL;
2381         struct obd_ioctl_data *data = NULL;
2382         __u32 type;
2383         int len = 0, rc;
2384
2385         if (!inode || !(sbi = ll_i2sbi(inode)))
2386                 GOTO(out_statfs, rc = -EINVAL);
2387
2388         rc = obd_ioctl_getdata(&buf, &len, arg);
2389         if (rc)
2390                 GOTO(out_statfs, rc);
2391
2392         data = (void*)buf;
2393         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2394             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2395                 GOTO(out_statfs, rc = -EINVAL);
2396
2397         if (data->ioc_inllen1 != sizeof(__u32) ||
2398             data->ioc_inllen2 != sizeof(__u32) ||
2399             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2400             data->ioc_plen2 != sizeof(struct obd_uuid))
2401                 GOTO(out_statfs, rc = -EINVAL);
2402
2403         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2404         if (type & LL_STATFS_LMV)
2405                 exp = sbi->ll_md_exp;
2406         else if (type & LL_STATFS_LOV)
2407                 exp = sbi->ll_dt_exp;
2408         else
2409                 GOTO(out_statfs, rc = -ENODEV);
2410
2411         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2412         if (rc)
2413                 GOTO(out_statfs, rc);
2414 out_statfs:
2415         if (buf)
2416                 obd_ioctl_freedata(buf, len);
2417         return rc;
2418 }
2419
2420 int ll_process_config(struct lustre_cfg *lcfg)
2421 {
2422         struct super_block *sb;
2423         unsigned long x;
2424         int rc = 0;
2425         char *ptr;
2426
2427         /* The instance name contains the sb: lustre-client-aacfe000 */
2428         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2429         if (!ptr || !*(++ptr))
2430                 return -EINVAL;
2431         if (sscanf(ptr, "%lx", &x) != 1)
2432                 return -EINVAL;
2433         sb = (struct super_block *)x;
2434         /* This better be a real Lustre superblock! */
2435         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2436
2437         /* Note we have not called client_common_fill_super yet, so
2438            proc fns must be able to handle that! */
2439         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2440                                       lcfg, sb);
2441         if (rc > 0)
2442                 rc = 0;
2443         return rc;
2444 }
2445
2446 /* this function prepares md_op_data hint for passing it down to MD stack. */
2447 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2448                                       struct inode *i1, struct inode *i2,
2449                                       const char *name, size_t namelen,
2450                                       __u32 mode, __u32 opc, void *data)
2451 {
2452         LASSERT(i1 != NULL);
2453
2454         if (name == NULL) {
2455                 /* Do not reuse namelen for something else. */
2456                 if (namelen != 0)
2457                         return ERR_PTR(-EINVAL);
2458         } else {
2459                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2460                         return ERR_PTR(-ENAMETOOLONG);
2461
2462                 if (!lu_name_is_valid_2(name, namelen))
2463                         return ERR_PTR(-EINVAL);
2464         }
2465
2466         if (op_data == NULL)
2467                 OBD_ALLOC_PTR(op_data);
2468
2469         if (op_data == NULL)
2470                 return ERR_PTR(-ENOMEM);
2471
2472         ll_i2gids(op_data->op_suppgids, i1, i2);
2473         op_data->op_fid1 = *ll_inode2fid(i1);
2474         op_data->op_default_stripe_offset = -1;
2475         if (S_ISDIR(i1->i_mode)) {
2476                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2477                 if (opc == LUSTRE_OPC_MKDIR)
2478                         op_data->op_default_stripe_offset =
2479                                    ll_i2info(i1)->lli_def_stripe_offset;
2480         }
2481
2482         if (i2) {
2483                 op_data->op_fid2 = *ll_inode2fid(i2);
2484                 if (S_ISDIR(i2->i_mode))
2485                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2486         } else {
2487                 fid_zero(&op_data->op_fid2);
2488         }
2489
2490         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2491                 op_data->op_cli_flags |= CLI_HASH64;
2492
2493         if (ll_need_32bit_api(ll_i2sbi(i1)))
2494                 op_data->op_cli_flags |= CLI_API32;
2495
2496         op_data->op_name = name;
2497         op_data->op_namelen = namelen;
2498         op_data->op_mode = mode;
2499         op_data->op_mod_time = cfs_time_current_sec();
2500         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2501         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2502         op_data->op_cap = cfs_curproc_cap_pack();
2503         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2504              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2505                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2506         } else {
2507                 op_data->op_mds = 0;
2508         }
2509         op_data->op_data = data;
2510
2511         return op_data;
2512 }
2513
2514 void ll_finish_md_op_data(struct md_op_data *op_data)
2515 {
2516         OBD_FREE_PTR(op_data);
2517 }
2518
2519 #ifdef HAVE_SUPEROPS_USE_DENTRY
2520 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2521 #else
2522 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2523 #endif
2524 {
2525         struct ll_sb_info *sbi;
2526
2527 #ifdef HAVE_SUPEROPS_USE_DENTRY
2528         LASSERT((seq != NULL) && (dentry != NULL));
2529         sbi = ll_s2sbi(dentry->d_sb);
2530 #else
2531         LASSERT((seq != NULL) && (vfs != NULL));
2532         sbi = ll_s2sbi(vfs->mnt_sb);
2533 #endif
2534
2535         if (sbi->ll_flags & LL_SBI_NOLCK)
2536                 seq_puts(seq, ",nolock");
2537
2538         if (sbi->ll_flags & LL_SBI_FLOCK)
2539                 seq_puts(seq, ",flock");
2540
2541         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2542                 seq_puts(seq, ",localflock");
2543
2544         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2545                 seq_puts(seq, ",user_xattr");
2546
2547         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2548                 seq_puts(seq, ",lazystatfs");
2549
2550         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2551                 seq_puts(seq, ",user_fid2path");
2552
2553         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2554                 seq_puts(seq, ",always_ping");
2555
2556         RETURN(0);
2557 }
2558
2559 /**
2560  * Get obd name by cmd, and copy out to user space
2561  */
2562 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2563 {
2564         struct ll_sb_info *sbi = ll_i2sbi(inode);
2565         struct obd_device *obd;
2566         ENTRY;
2567
2568         if (cmd == OBD_IOC_GETDTNAME)
2569                 obd = class_exp2obd(sbi->ll_dt_exp);
2570         else if (cmd == OBD_IOC_GETMDNAME)
2571                 obd = class_exp2obd(sbi->ll_md_exp);
2572         else
2573                 RETURN(-EINVAL);
2574
2575         if (!obd)
2576                 RETURN(-ENOENT);
2577
2578         if (copy_to_user((void __user *)arg, obd->obd_name,
2579                          strlen(obd->obd_name) + 1))
2580                 RETURN(-EFAULT);
2581
2582         RETURN(0);
2583 }
2584
2585 /**
2586  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2587  * fsname will be returned in this buffer; otherwise, a static buffer will be
2588  * used to store the fsname and returned to caller.
2589  */
2590 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2591 {
2592         static char fsname_static[MTI_NAME_MAXLEN];
2593         struct lustre_sb_info *lsi = s2lsi(sb);
2594         char *ptr;
2595         int len;
2596
2597         if (buf == NULL) {
2598                 /* this means the caller wants to use static buffer
2599                  * and it doesn't care about race. Usually this is
2600                  * in error reporting path */
2601                 buf = fsname_static;
2602                 buflen = sizeof(fsname_static);
2603         }
2604
2605         len = strlen(lsi->lsi_lmd->lmd_profile);
2606         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2607         if (ptr && (strcmp(ptr, "-client") == 0))
2608                 len -= 7;
2609
2610         if (unlikely(len >= buflen))
2611                 len = buflen - 1;
2612         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2613         buf[len] = '\0';
2614
2615         return buf;
2616 }
2617
2618 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2619 {
2620         char *path = NULL;
2621
2622         struct path p;
2623
2624         p.dentry = dentry;
2625         p.mnt = current->fs->root.mnt;
2626         path_get(&p);
2627         path = d_path(&p, buf, bufsize);
2628         path_put(&p);
2629         return path;
2630 }
2631
2632 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2633 {
2634         char *buf, *path = NULL;
2635         struct dentry *dentry = NULL;
2636         struct inode *inode = page->mapping->host;
2637
2638         /* this can be called inside spin lock so use GFP_ATOMIC. */
2639         buf = (char *)__get_free_page(GFP_ATOMIC);
2640         if (buf != NULL) {
2641                 dentry = d_find_alias(page->mapping->host);
2642                 if (dentry != NULL)
2643                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2644         }
2645
2646         CDEBUG(D_WARNING,
2647                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2648                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2649                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2650                PFID(ll_inode2fid(inode)),
2651                (path && !IS_ERR(path)) ? path : "", ioret);
2652
2653         if (dentry != NULL)
2654                 dput(dentry);
2655
2656         if (buf != NULL)
2657                 free_page((unsigned long)buf);
2658 }
2659
2660 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2661                         struct lov_user_md **kbuf)
2662 {
2663         struct lov_user_md      lum;
2664         ssize_t                 lum_size;
2665         ENTRY;
2666
2667         if (copy_from_user(&lum, md, sizeof(lum)))
2668                 RETURN(-EFAULT);
2669
2670         lum_size = ll_lov_user_md_size(&lum);
2671         if (lum_size < 0)
2672                 RETURN(lum_size);
2673
2674         OBD_ALLOC(*kbuf, lum_size);
2675         if (*kbuf == NULL)
2676                 RETURN(-ENOMEM);
2677
2678         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2679                 OBD_FREE(*kbuf, lum_size);
2680                 RETURN(-EFAULT);
2681         }
2682
2683         RETURN(lum_size);
2684 }
2685
2686 /*
2687  * Compute llite root squash state after a change of root squash
2688  * configuration setting or add/remove of a lnet nid
2689  */
2690 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2691 {
2692         struct root_squash_info *squash = &sbi->ll_squash;
2693         int i;
2694         bool matched;
2695         lnet_process_id_t id;
2696
2697         /* Update norootsquash flag */
2698         down_write(&squash->rsi_sem);
2699         if (list_empty(&squash->rsi_nosquash_nids))
2700                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2701         else {
2702                 /* Do not apply root squash as soon as one of our NIDs is
2703                  * in the nosquash_nids list */
2704                 matched = false;
2705                 i = 0;
2706                 while (LNetGetId(i++, &id) != -ENOENT) {
2707                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2708                                 continue;
2709                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2710                                 matched = true;
2711                                 break;
2712                         }
2713                 }
2714                 if (matched)
2715                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2716                 else
2717                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2718         }
2719         up_write(&squash->rsi_sem);
2720 }
2721
2722 /**
2723  * Parse linkea content to extract information about a given hardlink
2724  *
2725  * \param[in]   ldata      - Initialized linkea data
2726  * \param[in]   linkno     - Link identifier
2727  * \param[out]  parent_fid - The entry's parent FID
2728  * \param[out]  ln         - Entry name destination buffer
2729  *
2730  * \retval 0 on success
2731  * \retval Appropriate negative error code on failure
2732  */
2733 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2734                             struct lu_fid *parent_fid, struct lu_name *ln)
2735 {
2736         unsigned int    idx;
2737         int             rc;
2738         ENTRY;
2739
2740         rc = linkea_init(ldata);
2741         if (rc < 0)
2742                 RETURN(rc);
2743
2744         if (linkno >= ldata->ld_leh->leh_reccount)
2745                 /* beyond last link */
2746                 RETURN(-ENODATA);
2747
2748         linkea_first_entry(ldata);
2749         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2750                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2751                                     parent_fid);
2752                 if (idx == linkno)
2753                         break;
2754
2755                 linkea_next_entry(ldata);
2756         }
2757
2758         if (idx < linkno)
2759                 RETURN(-ENODATA);
2760
2761         RETURN(0);
2762 }
2763
2764 /**
2765  * Get parent FID and name of an identified link. Operation is performed for
2766  * a given link number, letting the caller iterate over linkno to list one or
2767  * all links of an entry.
2768  *
2769  * \param[in]     file - File descriptor against which to perform the operation
2770  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2771  *                       on and the available size. It is eventually filled with
2772  *                       the requested information or left untouched on error
2773  *
2774  * \retval - 0 on success
2775  * \retval - Appropriate negative error code on failure
2776  */
2777 int ll_getparent(struct file *file, struct getparent __user *arg)
2778 {
2779         struct dentry           *dentry = file->f_path.dentry;
2780         struct inode            *inode = dentry->d_inode;
2781         struct linkea_data      *ldata;
2782         struct lu_buf            buf = LU_BUF_NULL;
2783         struct lu_name           ln;
2784         struct lu_fid            parent_fid;
2785         __u32                    linkno;
2786         __u32                    name_size;
2787         int                      rc;
2788
2789         ENTRY;
2790
2791         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2792             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2793                 RETURN(-EPERM);
2794
2795         if (get_user(name_size, &arg->gp_name_size))
2796                 RETURN(-EFAULT);
2797
2798         if (get_user(linkno, &arg->gp_linkno))
2799                 RETURN(-EFAULT);
2800
2801         if (name_size > PATH_MAX)
2802                 RETURN(-EINVAL);
2803
2804         OBD_ALLOC(ldata, sizeof(*ldata));
2805         if (ldata == NULL)
2806                 RETURN(-ENOMEM);
2807
2808         rc = linkea_data_new(ldata, &buf);
2809         if (rc < 0)
2810                 GOTO(ldata_free, rc);
2811
2812         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2813         if (rc < 0)
2814                 GOTO(lb_free, rc);
2815
2816         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2817         if (rc < 0)
2818                 GOTO(lb_free, rc);
2819
2820         if (ln.ln_namelen >= name_size)
2821                 GOTO(lb_free, rc = -EOVERFLOW);
2822
2823         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2824                 GOTO(lb_free, rc = -EFAULT);
2825
2826         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2827                 GOTO(lb_free, rc = -EFAULT);
2828
2829         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2830                 GOTO(lb_free, rc = -EFAULT);
2831
2832 lb_free:
2833         lu_buf_free(&buf);
2834 ldata_free:
2835         OBD_FREE(ldata, sizeof(*ldata));
2836
2837         RETURN(rc);
2838 }