lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #ifdef HAVE_UIDGID_HEADER
  50 # include <linux/uidgid.h>
  51 #endif
  52
  53 #include <lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC_PTR(sbi);
  81         if (sbi == NULL)
  82                 RETURN(NULL);
  83
  84         spin_lock_init(&sbi->ll_lock);
  85         mutex_init(&sbi->ll_lco.lco_lock);
  86         spin_lock_init(&sbi->ll_pp_extent_lock);
  87         spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         lru_page_max = pages / 2;
  93
  94         /* initialize ll_cache data */
  95         sbi->ll_cache = cl_cache_init(lru_page_max);
  96         if (sbi->ll_cache == NULL) {
  97                 OBD_FREE(sbi, sizeof(*sbi));
  98                 RETURN(NULL);
  99         }
 100
 101         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 102                                            SBI_DEFAULT_READAHEAD_MAX);
 103         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 104         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 105                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 106         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 107         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 108
 109         ll_generate_random_uuid(uuid);
 110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 112
 113         sbi->ll_flags |= LL_SBI_VERBOSE;
 114 #ifdef ENABLE_CHECKSUM
 115         sbi->ll_flags |= LL_SBI_CHECKSUM;
 116 #endif
 117
 118 #ifdef HAVE_LRU_RESIZE_SUPPORT
 119         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 120 #endif
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136
 137         /* root squash */
 138         sbi->ll_squash.rsi_uid = 0;
 139         sbi->ll_squash.rsi_gid = 0;
 140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 141         init_rwsem(&sbi->ll_squash.rsi_sem);
 142
 143         RETURN(sbi);
 144 }
 145
 146 static void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 154                 if (sbi->ll_cache != NULL) {
 155                         cl_cache_decref(sbi->ll_cache);
 156                         sbi->ll_cache = NULL;
 157                 }
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 164                                     struct vfsmount *mnt)
 165 {
 166         struct inode *root = NULL;
 167         struct ll_sb_info *sbi = ll_s2sbi(sb);
 168         struct obd_device *obd;
 169         struct obd_statfs *osfs = NULL;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         u64 valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         OBD_ALLOC_PTR(osfs);
 190         if (osfs == NULL) {
 191                 OBD_FREE_PTR(data);
 192                 RETURN(-ENOMEM);
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 204                                   OBD_CONNECT_EINPROGRESS |
 205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 207                                   OBD_CONNECT_MAX_EASIZE |
 208                                   OBD_CONNECT_FLOCK_DEAD |
 209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 210                                   OBD_CONNECT_OPEN_BY_FID |
 211                                   OBD_CONNECT_DIR_STRIPE |
 212                                   OBD_CONNECT_BULK_MBITS;
 213
 214 #ifdef HAVE_LRU_RESIZE_SUPPORT
 215         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 216                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 217 #endif
 218 #ifdef CONFIG_FS_POSIX_ACL
 219         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 220 #endif
 221
 222         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 223                 /* flag mdc connection as lightweight, only used for test
 224                  * purpose, use with care */
 225                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 226
 227         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 228         data->ocd_version = LUSTRE_VERSION_CODE;
 229
 230         if (sb->s_flags & MS_RDONLY)
 231                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 232         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 233                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 234
 235 #ifdef HAVE_MS_FLOCK_LOCK
 236         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 237         sb->s_flags |= MS_FLOCK_LOCK;
 238 #endif
 239 #ifdef MS_HAS_NEW_AOPS
 240         sb->s_flags |= MS_HAS_NEW_AOPS;
 241 #endif
 242
 243         if (sbi->ll_flags & LL_SBI_FLOCK)
 244                 sbi->ll_fop = &ll_file_operations_flock;
 245         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 246                 sbi->ll_fop = &ll_file_operations;
 247         else
 248                 sbi->ll_fop = &ll_file_operations_noflock;
 249
 250         /* real client */
 251         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 252         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 253                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 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, ",");
 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         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 407                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 408
 409         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 410                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 411                  * disabled by default, because it can still be enabled on the
 412                  * fly via /proc. As a consequence, we still need to come to an
 413                  * agreement on the supported algorithms at connect time */
 414                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 415
 416                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 417                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 418                 else
 419                         data->ocd_cksum_types = cksum_types_supported_client();
 420         }
 421
 422 #ifdef HAVE_LRU_RESIZE_SUPPORT
 423         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 424 #endif
 425         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 426                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 427
 428         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 429                "ocd_grant: %d\n", data->ocd_connect_flags,
 430                data->ocd_version, data->ocd_grant);
 431
 432         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 433         obd->obd_upcall.onu_upcall = cl_ocd_update;
 434
 435         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 436
 437         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 438                           NULL);
 439         if (err == -EBUSY) {
 440                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 441                                    "recovery, of which this client is not a "
 442                                    "part.  Please wait for recovery to "
 443                                    "complete, abort, or time out.\n", dt);
 444                 GOTO(out_md, err);
 445         } else if (err) {
 446                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 447                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 448                 GOTO(out_md, err);
 449         }
 450
 451         sbi->ll_dt_exp->exp_connect_data = *data;
 452
 453         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 454                            LUSTRE_SEQ_METADATA);
 455         if (err) {
 456                 CERROR("%s: Can't init data layer FID infrastructure, "
 457                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 458                 GOTO(out_dt, err);
 459         }
 460
 461         mutex_lock(&sbi->ll_lco.lco_lock);
 462         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 463         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 464         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 465         mutex_unlock(&sbi->ll_lco.lco_lock);
 466
 467         fid_zero(&sbi->ll_root_fid);
 468         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
 469         if (err) {
 470                 CERROR("cannot mds_connect: rc = %d\n", err);
 471                 GOTO(out_lock_cn_cb, err);
 472         }
 473         if (!fid_is_sane(&sbi->ll_root_fid)) {
 474                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 475                        sbi->ll_md_exp->exp_obd->obd_name,
 476                        PFID(&sbi->ll_root_fid));
 477                 GOTO(out_lock_cn_cb, err = -EINVAL);
 478         }
 479         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 480
 481         sb->s_op = &lustre_super_operations;
 482 #if THREAD_SIZE >= 8192 /*b=17630*/
 483         sb->s_export_op = &lustre_export_operations;
 484 #endif
 485
 486         /* make root inode
 487          * XXX: move this to after cbd setup? */
 488         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 489         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 490                 valid |= OBD_MD_FLRMTPERM;
 491         else if (sbi->ll_flags & LL_SBI_ACL)
 492                 valid |= OBD_MD_FLACL;
 493
 494         OBD_ALLOC_PTR(op_data);
 495         if (op_data == NULL)
 496                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 497
 498         op_data->op_fid1 = sbi->ll_root_fid;
 499         op_data->op_mode = 0;
 500         op_data->op_valid = valid;
 501
 502         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 503
 504         OBD_FREE_PTR(op_data);
 505         if (err) {
 506                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 507                        sbi->ll_md_exp->exp_obd->obd_name, err);
 508                 GOTO(out_lock_cn_cb, err);
 509         }
 510
 511         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 512                                sbi->ll_md_exp, &lmd);
 513         if (err) {
 514                 CERROR("failed to understand root inode md: rc = %d\n", err);
 515                 ptlrpc_req_finished(request);
 516                 GOTO(out_lock_cn_cb, err);
 517         }
 518
 519         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 520         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 521                                             sbi->ll_flags & LL_SBI_32BIT_API),
 522                        &lmd);
 523         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 524         ptlrpc_req_finished(request);
 525
 526         if (IS_ERR(root)) {
 527 #ifdef CONFIG_FS_POSIX_ACL
 528                 if (lmd.posix_acl) {
 529                         posix_acl_release(lmd.posix_acl);
 530                         lmd.posix_acl = NULL;
 531                 }
 532 #endif
 533                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 534                 root = NULL;
 535                 CERROR("lustre_lite: bad iget4 for root\n");
 536                 GOTO(out_root, err);
 537         }
 538
 539 #ifdef CONFIG_FS_POSIX_ACL
 540         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 541                 rct_init(&sbi->ll_rct);
 542                 et_init(&sbi->ll_et);
 543         }
 544 #endif
 545
 546         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 547         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 548                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 549                                  NULL);
 550         cl_sb_init(sb);
 551
 552         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 553                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 554                                  sbi->ll_cache, NULL);
 555
 556         sb->s_root = d_make_root(root);
 557         if (sb->s_root == NULL) {
 558                 CERROR("%s: can't make root dentry\n",
 559                         ll_get_fsname(sb, NULL, 0));
 560                 GOTO(out_root, err = -ENOMEM);
 561         }
 562 #ifdef HAVE_DCACHE_LOCK
 563         sb->s_root->d_op = &ll_d_ops;
 564 #endif
 565
 566         sbi->ll_sdev_orig = sb->s_dev;
 567
 568         /* We set sb->s_dev equal on all lustre clients in order to support
 569          * NFS export clustering.  NFSD requires that the FSID be the same
 570          * on all clients. */
 571         /* s_dev is also used in lt_compare() to compare two fs, but that is
 572          * only a node-local comparison. */
 573         uuid = obd_get_uuid(sbi->ll_md_exp);
 574         if (uuid != NULL)
 575                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 576
 577         if (data != NULL)
 578                 OBD_FREE_PTR(data);
 579         if (osfs != NULL)
 580                 OBD_FREE_PTR(osfs);
 581         if (proc_lustre_fs_root != NULL) {
 582                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 583                                                   dt, md);
 584                 if (err < 0) {
 585                         CERROR("%s: could not register mount in lprocfs: "
 586                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 587                         err = 0;
 588                 }
 589         }
 590
 591         RETURN(err);
 592 out_root:
 593         if (root)
 594                 iput(root);
 595 out_lock_cn_cb:
 596         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 597 out_dt:
 598         obd_disconnect(sbi->ll_dt_exp);
 599         sbi->ll_dt_exp = NULL;
 600 out_md_fid:
 601         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 602 out_md:
 603         obd_disconnect(sbi->ll_md_exp);
 604         sbi->ll_md_exp = NULL;
 605 out:
 606         if (data != NULL)
 607                 OBD_FREE_PTR(data);
 608         if (osfs != NULL)
 609                 OBD_FREE_PTR(osfs);
 610         return err;
 611 }
 612
 613 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 614 {
 615         int size, rc;
 616
 617         size = sizeof(*lmmsize);
 618         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 619                           KEY_MAX_EASIZE, &size, lmmsize);
 620         if (rc != 0) {
 621                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 622                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 623                 RETURN(rc);
 624         }
 625
 626         size = sizeof(int);
 627         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 628                           KEY_MAX_EASIZE, &size, lmmsize);
 629         if (rc)
 630                 CERROR("Get max mdsize error rc %d\n", rc);
 631
 632         RETURN(rc);
 633 }
 634
 635 /**
 636  * Get the value of the default_easize parameter.
 637  *
 638  * \see client_obd::cl_default_mds_easize
 639  *
 640  * \param[in] sbi       superblock info for this filesystem
 641  * \param[out] lmmsize  pointer to storage location for value
 642  *
 643  * \retval 0            on success
 644  * \retval negative     negated errno on failure
 645  */
 646 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 647 {
 648         int size, rc;
 649
 650         size = sizeof(int);
 651         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 652                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 653         if (rc)
 654                 CERROR("Get default mdsize error rc %d\n", rc);
 655
 656         RETURN(rc);
 657 }
 658
 659 /**
 660  * Set the default_easize parameter to the given value.
 661  *
 662  * \see client_obd::cl_default_mds_easize
 663  *
 664  * \param[in] sbi       superblock info for this filesystem
 665  * \param[in] lmmsize   the size to set
 666  *
 667  * \retval 0            on success
 668  * \retval negative     negated errno on failure
 669  */
 670 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 671 {
 672         int rc;
 673
 674         if (lmmsize < sizeof(struct lov_mds_md) ||
 675             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 676                 return -EINVAL;
 677
 678         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 679                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 680                                 sizeof(int), &lmmsize, NULL);
 681
 682         RETURN(rc);
 683 }
 684
 685 static void ll_dump_inode(struct inode *inode)
 686 {
 687         struct ll_d_hlist_node *tmp;
 688         int dentry_count = 0;
 689
 690         LASSERT(inode != NULL);
 691
 692         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 693                 dentry_count++;
 694
 695         CERROR("%s: inode %p dump: dev=%s fid="DFID
 696                " mode=%o count=%u, %d dentries\n",
 697                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 698                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 699                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 700 }
 701
 702 void lustre_dump_dentry(struct dentry *dentry, int recur)
 703 {
 704         struct list_head *tmp;
 705         int subdirs = 0;
 706
 707         LASSERT(dentry != NULL);
 708
 709         list_for_each(tmp, &dentry->d_subdirs)
 710                 subdirs++;
 711
 712         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 713                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 714                dentry->d_name.len, dentry->d_name.name,
 715                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 716                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 717                dentry->d_flags, dentry->d_fsdata, subdirs);
 718         if (dentry->d_inode != NULL)
 719                 ll_dump_inode(dentry->d_inode);
 720
 721         if (recur == 0)
 722                 return;
 723
 724         list_for_each(tmp, &dentry->d_subdirs) {
 725                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 726                 lustre_dump_dentry(d, recur - 1);
 727         }
 728 }
 729
 730 static void client_common_put_super(struct super_block *sb)
 731 {
 732         struct ll_sb_info *sbi = ll_s2sbi(sb);
 733         ENTRY;
 734
 735 #ifdef CONFIG_FS_POSIX_ACL
 736         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 737                 et_fini(&sbi->ll_et);
 738                 rct_fini(&sbi->ll_rct);
 739         }
 740 #endif
 741
 742         cl_sb_fini(sb);
 743
 744         list_del(&sbi->ll_conn_chain);
 745
 746         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 747         obd_disconnect(sbi->ll_dt_exp);
 748         sbi->ll_dt_exp = NULL;
 749
 750         lprocfs_unregister_mountpoint(sbi);
 751
 752         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 753         obd_disconnect(sbi->ll_md_exp);
 754         sbi->ll_md_exp = NULL;
 755
 756         EXIT;
 757 }
 758
 759 void ll_kill_super(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi;
 762         ENTRY;
 763
 764         /* not init sb ?*/
 765         if (!(sb->s_flags & MS_ACTIVE))
 766                 return;
 767
 768         sbi = ll_s2sbi(sb);
 769         /* we need restore s_dev from changed for clustred NFS before put_super
 770          * because new kernels have cached s_dev and change sb->s_dev in
 771          * put_super not affected real removing devices */
 772         if (sbi) {
 773                 sb->s_dev = sbi->ll_sdev_orig;
 774                 sbi->ll_umounting = 1;
 775
 776                 /* wait running statahead threads to quit */
 777                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 778                         set_current_state(TASK_UNINTERRUPTIBLE);
 779                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 780                 }
 781         }
 782
 783         EXIT;
 784 }
 785
 786 static inline int ll_set_opt(const char *opt, char *data, int fl)
 787 {
 788         if (strncmp(opt, data, strlen(opt)) != 0)
 789                 return(0);
 790         else
 791                 return(fl);
 792 }
 793
 794 /* non-client-specific mount options are parsed in lmd_parse */
 795 static int ll_options(char *options, int *flags)
 796 {
 797         int tmp;
 798         char *s1 = options, *s2;
 799         ENTRY;
 800
 801         if (!options)
 802                 RETURN(0);
 803
 804         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 805
 806         while (*s1) {
 807                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 808                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 809                 if (tmp) {
 810                         *flags |= tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("context", s1, 1);
 839                 if (tmp)
 840                         goto next;
 841                 tmp = ll_set_opt("fscontext", s1, 1);
 842                 if (tmp)
 843                         goto next;
 844                 tmp = ll_set_opt("defcontext", s1, 1);
 845                 if (tmp)
 846                         goto next;
 847                 tmp = ll_set_opt("rootcontext", s1, 1);
 848                 if (tmp)
 849                         goto next;
 850                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865
 866                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 892                 if (tmp) {
 893                         *flags &= ~tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 907                 if (tmp) {
 908                         *flags &= ~tmp;
 909                         goto next;
 910                 }
 911                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 912                                    s1);
 913                 RETURN(-EINVAL);
 914
 915 next:
 916                 /* Find next opt */
 917                 s2 = strchr(s1, ',');
 918                 if (s2 == NULL)
 919                         break;
 920                 s1 = s2 + 1;
 921         }
 922         RETURN(0);
 923 }
 924
 925 void ll_lli_init(struct ll_inode_info *lli)
 926 {
 927         lli->lli_inode_magic = LLI_INODE_MAGIC;
 928         lli->lli_flags = 0;
 929         spin_lock_init(&lli->lli_lock);
 930         lli->lli_posix_acl = NULL;
 931         lli->lli_remote_perms = NULL;
 932         mutex_init(&lli->lli_rmtperm_mutex);
 933         /* Do not set lli_fid, it has been initialized already. */
 934         fid_zero(&lli->lli_pfid);
 935         atomic_set(&lli->lli_open_count, 0);
 936         lli->lli_rmtperm_time = 0;
 937         lli->lli_mds_read_och = NULL;
 938         lli->lli_mds_write_och = NULL;
 939         lli->lli_mds_exec_och = NULL;
 940         lli->lli_open_fd_read_count = 0;
 941         lli->lli_open_fd_write_count = 0;
 942         lli->lli_open_fd_exec_count = 0;
 943         mutex_init(&lli->lli_och_mutex);
 944         spin_lock_init(&lli->lli_agl_lock);
 945         spin_lock_init(&lli->lli_layout_lock);
 946         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 947         lli->lli_clob = NULL;
 948
 949         init_rwsem(&lli->lli_xattrs_list_rwsem);
 950         mutex_init(&lli->lli_xattrs_enq_lock);
 951
 952         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 953         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 954                 mutex_init(&lli->lli_readdir_mutex);
 955                 lli->lli_opendir_key = NULL;
 956                 lli->lli_sai = NULL;
 957                 spin_lock_init(&lli->lli_sa_lock);
 958                 lli->lli_opendir_pid = 0;
 959                 lli->lli_sa_enabled = 0;
 960                 lli->lli_def_stripe_offset = -1;
 961         } else {
 962                 mutex_init(&lli->lli_size_mutex);
 963                 lli->lli_symlink_name = NULL;
 964                 init_rwsem(&lli->lli_trunc_sem);
 965                 range_lock_tree_init(&lli->lli_write_tree);
 966                 init_rwsem(&lli->lli_glimpse_sem);
 967                 lli->lli_glimpse_time = 0;
 968                 INIT_LIST_HEAD(&lli->lli_agl_list);
 969                 lli->lli_agl_index = 0;
 970                 lli->lli_async_rc = 0;
 971         }
 972         mutex_init(&lli->lli_layout_mutex);
 973 }
 974
 975 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 976 {
 977         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 978
 979         bdi->name = "lustre";
 980         return bdi_register(bdi, NULL, "lustre-%d",
 981                             atomic_inc_return(&ll_bdi_num));
 982 }
 983
 984 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 985 {
 986         struct lustre_profile *lprof = NULL;
 987         struct lustre_sb_info *lsi = s2lsi(sb);
 988         struct ll_sb_info *sbi;
 989         char  *dt = NULL, *md = NULL;
 990         char  *profilenm = get_profile_name(sb);
 991         struct config_llog_instance *cfg;
 992         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 993         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 994         int    err;
 995         ENTRY;
 996
 997         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 998
 999         OBD_ALLOC_PTR(cfg);
1000         if (cfg == NULL)
1001                 RETURN(-ENOMEM);
1002
1003         try_module_get(THIS_MODULE);
1004
1005         /* client additional sb info */
1006         lsi->lsi_llsbi = sbi = ll_init_sbi();
1007         if (!sbi) {
1008                 module_put(THIS_MODULE);
1009                 OBD_FREE_PTR(cfg);
1010                 RETURN(-ENOMEM);
1011         }
1012
1013         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1014         if (err)
1015                 GOTO(out_free, err);
1016
1017         err = bdi_init(&lsi->lsi_bdi);
1018         if (err)
1019                 GOTO(out_free, err);
1020         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1021 #ifdef HAVE_BDI_CAP_MAP_COPY
1022         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1023 #else
1024         lsi->lsi_bdi.capabilities = 0;
1025 #endif
1026         err = ll_bdi_register(&lsi->lsi_bdi);
1027         if (err)
1028                 GOTO(out_free, err);
1029
1030         sb->s_bdi = &lsi->lsi_bdi;
1031 #ifndef HAVE_DCACHE_LOCK
1032         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1033         sb->s_d_op = &ll_d_ops;
1034 #endif
1035
1036         /* Generate a string unique to this super, in case some joker tries
1037            to mount the same fs at two mount points.
1038            Use the address of the super itself.*/
1039         cfg->cfg_instance = sb;
1040         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1041         cfg->cfg_callback = class_config_llog_handler;
1042         /* set up client obds */
1043         err = lustre_process_log(sb, profilenm, cfg);
1044         if (err < 0)
1045                 GOTO(out_free, err);
1046
1047         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1048         lprof = class_get_profile(profilenm);
1049         if (lprof == NULL) {
1050                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1051                                    " read from the MGS.  Does that filesystem "
1052                                    "exist?\n", profilenm);
1053                 GOTO(out_free, err = -EINVAL);
1054         }
1055         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1056                lprof->lp_md, lprof->lp_dt);
1057
1058         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1059         if (!dt)
1060                 GOTO(out_free, err = -ENOMEM);
1061         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1062
1063         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1064         if (!md)
1065                 GOTO(out_free, err = -ENOMEM);
1066         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1067
1068         /* connections, registrations, sb setup */
1069         err = client_common_fill_super(sb, md, dt, mnt);
1070         if (err < 0)
1071                 GOTO(out_free, err);
1072
1073         sbi->ll_client_common_fill_super_succeeded = 1;
1074
1075 out_free:
1076         if (md)
1077                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1078         if (dt)
1079                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1080         if (lprof != NULL)
1081                 class_put_profile(lprof);
1082         if (err)
1083                 ll_put_super(sb);
1084         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1085                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1086
1087         OBD_FREE_PTR(cfg);
1088         RETURN(err);
1089 } /* ll_fill_super */
1090
1091 void ll_put_super(struct super_block *sb)
1092 {
1093         struct config_llog_instance cfg, params_cfg;
1094         struct obd_device *obd;
1095         struct lustre_sb_info *lsi = s2lsi(sb);
1096         struct ll_sb_info *sbi = ll_s2sbi(sb);
1097         char *profilenm = get_profile_name(sb);
1098         long ccc_count;
1099         int next, force = 1, rc = 0;
1100         ENTRY;
1101
1102         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1103
1104         cfg.cfg_instance = sb;
1105         lustre_end_log(sb, profilenm, &cfg);
1106
1107         params_cfg.cfg_instance = sb;
1108         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1109
1110         if (sbi->ll_md_exp) {
1111                 obd = class_exp2obd(sbi->ll_md_exp);
1112                 if (obd)
1113                         force = obd->obd_force;
1114         }
1115
1116         /* Wait for unstable pages to be committed to stable storage */
1117         if (force == 0) {
1118                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1119                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1120                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1121                         &lwi);
1122         }
1123
1124         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1125         if (force == 0 && rc != -EINTR)
1126                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1127
1128
1129         /* We need to set force before the lov_disconnect in
1130            lustre_common_put_super, since l_d cleans up osc's as well. */
1131         if (force) {
1132                 next = 0;
1133                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1134                                                      &next)) != NULL) {
1135                         obd->obd_force = force;
1136                 }
1137         }
1138
1139         if (sbi->ll_client_common_fill_super_succeeded) {
1140                 /* Only if client_common_fill_super succeeded */
1141                 client_common_put_super(sb);
1142         }
1143
1144         next = 0;
1145         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1146                 class_manual_cleanup(obd);
1147         }
1148
1149         if (sbi->ll_flags & LL_SBI_VERBOSE)
1150                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1151
1152         if (profilenm)
1153                 class_del_profile(profilenm);
1154
1155         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1156                 bdi_destroy(&lsi->lsi_bdi);
1157                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1158         }
1159
1160         ll_free_sbi(sb);
1161         lsi->lsi_llsbi = NULL;
1162
1163         lustre_common_put_super(sb);
1164
1165         cl_env_cache_purge(~0);
1166
1167         module_put(THIS_MODULE);
1168
1169         EXIT;
1170 } /* client_put_super */
1171
1172 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1173 {
1174         struct inode *inode = NULL;
1175
1176         /* NOTE: we depend on atomic igrab() -bzzz */
1177         lock_res_and_lock(lock);
1178         if (lock->l_resource->lr_lvb_inode) {
1179                 struct ll_inode_info * lli;
1180                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1181                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1182                         inode = igrab(lock->l_resource->lr_lvb_inode);
1183                 } else {
1184                         inode = lock->l_resource->lr_lvb_inode;
1185                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1186                                          D_WARNING, lock, "lr_lvb_inode %p is "
1187                                          "bogus: magic %08x",
1188                                          lock->l_resource->lr_lvb_inode,
1189                                          lli->lli_inode_magic);
1190                         inode = NULL;
1191                 }
1192         }
1193         unlock_res_and_lock(lock);
1194         return inode;
1195 }
1196
1197 void ll_dir_clear_lsm_md(struct inode *inode)
1198 {
1199         struct ll_inode_info *lli = ll_i2info(inode);
1200
1201         LASSERT(S_ISDIR(inode->i_mode));
1202
1203         if (lli->lli_lsm_md != NULL) {
1204                 lmv_free_memmd(lli->lli_lsm_md);
1205                 lli->lli_lsm_md = NULL;
1206         }
1207 }
1208
1209 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1210                                       const struct lu_fid *fid,
1211                                       struct lustre_md *md)
1212 {
1213         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1214         struct mdt_body         *body = md->body;
1215         struct inode            *inode;
1216         ino_t                   ino;
1217         ENTRY;
1218
1219         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1220         inode = iget_locked(sb, ino);
1221         if (inode == NULL) {
1222                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1223                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1224                 RETURN(ERR_PTR(-ENOENT));
1225         }
1226
1227         if (inode->i_state & I_NEW) {
1228                 struct ll_inode_info *lli = ll_i2info(inode);
1229                 struct lmv_stripe_md *lsm = md->lmv;
1230
1231                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1232                                 (body->mbo_mode & S_IFMT);
1233                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1234                          PFID(fid));
1235
1236                 LTIME_S(inode->i_mtime) = 0;
1237                 LTIME_S(inode->i_atime) = 0;
1238                 LTIME_S(inode->i_ctime) = 0;
1239                 inode->i_rdev = 0;
1240
1241 #ifdef HAVE_BACKING_DEV_INFO
1242                 /* initializing backing dev info. */
1243                 inode->i_mapping->backing_dev_info =
1244                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1245 #endif
1246                 inode->i_op = &ll_dir_inode_operations;
1247                 inode->i_fop = &ll_dir_operations;
1248                 lli->lli_fid = *fid;
1249                 ll_lli_init(lli);
1250
1251                 LASSERT(lsm != NULL);
1252                 /* master object FID */
1253                 lli->lli_pfid = body->mbo_fid1;
1254                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1255                        lli, PFID(fid), PFID(&lli->lli_pfid));
1256                 unlock_new_inode(inode);
1257         }
1258
1259         RETURN(inode);
1260 }
1261
1262 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1263 {
1264         struct lu_fid *fid;
1265         struct lmv_stripe_md *lsm = md->lmv;
1266         int i;
1267
1268         LASSERT(lsm != NULL);
1269         /* XXX sigh, this lsm_root initialization should be in
1270          * LMV layer, but it needs ll_iget right now, so we
1271          * put this here right now. */
1272         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1273                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1274                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1275                 /* Unfortunately ll_iget will call ll_update_inode,
1276                  * where the initialization of slave inode is slightly
1277                  * different, so it reset lsm_md to NULL to avoid
1278                  * initializing lsm for slave inode. */
1279                 /* For migrating inode, master stripe and master object will
1280                  * be same, so we only need assign this inode */
1281                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1282                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1283                 else
1284                         lsm->lsm_md_oinfo[i].lmo_root =
1285                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1286
1287                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1288                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1289
1290                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1291                         return rc;
1292                 }
1293         }
1294
1295         return 0;
1296 }
1297
1298 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1299                                 const struct lmv_stripe_md *lsm_md2)
1300 {
1301         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1302                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1303                lsm_md1->lsm_md_master_mdt_index ==
1304                                         lsm_md2->lsm_md_master_mdt_index &&
1305                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1306                lsm_md1->lsm_md_layout_version ==
1307                                         lsm_md2->lsm_md_layout_version &&
1308                strcmp(lsm_md1->lsm_md_pool_name,
1309                       lsm_md2->lsm_md_pool_name) == 0;
1310 }
1311
1312 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1313 {
1314         struct ll_inode_info *lli = ll_i2info(inode);
1315         struct lmv_stripe_md *lsm = md->lmv;
1316         int     rc;
1317         ENTRY;
1318
1319         LASSERT(S_ISDIR(inode->i_mode));
1320         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1321                PFID(ll_inode2fid(inode)));
1322
1323         /* no striped information from request. */
1324         if (lsm == NULL) {
1325                 if (lli->lli_lsm_md == NULL) {
1326                         RETURN(0);
1327                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1328                                                 LMV_HASH_FLAG_MIGRATION) {
1329                         /* migration is done, the temporay MIGRATE layout has
1330                          * been removed */
1331                         CDEBUG(D_INODE, DFID" finish migration.\n",
1332                                PFID(ll_inode2fid(inode)));
1333                         lmv_free_memmd(lli->lli_lsm_md);
1334                         lli->lli_lsm_md = NULL;
1335                         RETURN(0);
1336                 } else {
1337                         /* The lustre_md from req does not include stripeEA,
1338                          * see ll_md_setattr */
1339                         RETURN(0);
1340                 }
1341         }
1342
1343         /* set the directory layout */
1344         if (lli->lli_lsm_md == NULL) {
1345                 struct cl_attr  *attr;
1346
1347                 rc = ll_init_lsm_md(inode, md);
1348                 if (rc != 0)
1349                         RETURN(rc);
1350
1351                 /* set md->lmv to NULL, so the following free lustre_md
1352                  * will not free this lsm */
1353                 md->lmv = NULL;
1354                 lli->lli_lsm_md = lsm;
1355
1356                 OBD_ALLOC_PTR(attr);
1357                 if (attr == NULL)
1358                         RETURN(-ENOMEM);
1359
1360                 /* validate the lsm */
1361                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1362                                    ll_md_blocking_ast);
1363                 if (rc != 0) {
1364                         OBD_FREE_PTR(attr);
1365                         RETURN(rc);
1366                 }
1367
1368                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1369                         md->body->mbo_nlink = attr->cat_nlink;
1370                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1371                         md->body->mbo_size = attr->cat_size;
1372                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1373                         md->body->mbo_atime = attr->cat_atime;
1374                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1375                         md->body->mbo_ctime = attr->cat_ctime;
1376                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1377                         md->body->mbo_mtime = attr->cat_mtime;
1378
1379                 OBD_FREE_PTR(attr);
1380
1381                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1382                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1383                 RETURN(0);
1384         }
1385
1386         /* Compare the old and new stripe information */
1387         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1388                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1389                 int                     idx;
1390
1391                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1392                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1393                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1394                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1395                        inode, lsm, old_lsm,
1396                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1397                        lsm->lsm_md_stripe_count,
1398                        old_lsm->lsm_md_stripe_count,
1399                        lsm->lsm_md_master_mdt_index,
1400                        old_lsm->lsm_md_master_mdt_index,
1401                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1402                        lsm->lsm_md_layout_version,
1403                        old_lsm->lsm_md_layout_version,
1404                        lsm->lsm_md_pool_name,
1405                        old_lsm->lsm_md_pool_name);
1406
1407                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1408                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1409                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1410                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1411                 }
1412
1413                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1414                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1415                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1416                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1417                 }
1418
1419                 RETURN(-EIO);
1420         }
1421
1422         RETURN(0);
1423 }
1424
1425 void ll_clear_inode(struct inode *inode)
1426 {
1427         struct ll_inode_info *lli = ll_i2info(inode);
1428         struct ll_sb_info *sbi = ll_i2sbi(inode);
1429         ENTRY;
1430
1431         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1432                PFID(ll_inode2fid(inode)), inode);
1433
1434         if (S_ISDIR(inode->i_mode)) {
1435                 /* these should have been cleared in ll_file_release */
1436                 LASSERT(lli->lli_opendir_key == NULL);
1437                 LASSERT(lli->lli_sai == NULL);
1438                 LASSERT(lli->lli_opendir_pid == 0);
1439         }
1440
1441         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1442
1443         LASSERT(!lli->lli_open_fd_write_count);
1444         LASSERT(!lli->lli_open_fd_read_count);
1445         LASSERT(!lli->lli_open_fd_exec_count);
1446
1447         if (lli->lli_mds_write_och)
1448                 ll_md_real_close(inode, FMODE_WRITE);
1449         if (lli->lli_mds_exec_och)
1450                 ll_md_real_close(inode, FMODE_EXEC);
1451         if (lli->lli_mds_read_och)
1452                 ll_md_real_close(inode, FMODE_READ);
1453
1454         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1455                 OBD_FREE(lli->lli_symlink_name,
1456                          strlen(lli->lli_symlink_name) + 1);
1457                 lli->lli_symlink_name = NULL;
1458         }
1459
1460         ll_xattr_cache_destroy(inode);
1461
1462         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1463                 LASSERT(lli->lli_posix_acl == NULL);
1464                 if (lli->lli_remote_perms) {
1465                         free_rmtperm_hash(lli->lli_remote_perms);
1466                         lli->lli_remote_perms = NULL;
1467                 }
1468         }
1469 #ifdef CONFIG_FS_POSIX_ACL
1470         else if (lli->lli_posix_acl) {
1471                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1472                 LASSERT(lli->lli_remote_perms == NULL);
1473                 posix_acl_release(lli->lli_posix_acl);
1474                 lli->lli_posix_acl = NULL;
1475         }
1476 #endif
1477         lli->lli_inode_magic = LLI_INODE_DEAD;
1478
1479         if (S_ISDIR(inode->i_mode))
1480                 ll_dir_clear_lsm_md(inode);
1481         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1482                 LASSERT(list_empty(&lli->lli_agl_list));
1483
1484         /*
1485          * XXX This has to be done before lsm is freed below, because
1486          * cl_object still uses inode lsm.
1487          */
1488         cl_inode_fini(inode);
1489
1490         EXIT;
1491 }
1492
1493 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1494 {
1495         struct lustre_md md;
1496         struct inode *inode = dentry->d_inode;
1497         struct ll_sb_info *sbi = ll_i2sbi(inode);
1498         struct ptlrpc_request *request = NULL;
1499         int rc, ia_valid;
1500         ENTRY;
1501
1502         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1503                                      LUSTRE_OPC_ANY, NULL);
1504         if (IS_ERR(op_data))
1505                 RETURN(PTR_ERR(op_data));
1506
1507         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1508         if (rc) {
1509                 ptlrpc_req_finished(request);
1510                 if (rc == -ENOENT) {
1511                         clear_nlink(inode);
1512                         /* Unlinked special device node? Or just a race?
1513                          * Pretend we done everything. */
1514                         if (!S_ISREG(inode->i_mode) &&
1515                             !S_ISDIR(inode->i_mode)) {
1516                                 ia_valid = op_data->op_attr.ia_valid;
1517                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1518                                 rc = simple_setattr(dentry, &op_data->op_attr);
1519                                 op_data->op_attr.ia_valid = ia_valid;
1520                         }
1521                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1522                         CERROR("md_setattr fails: rc = %d\n", rc);
1523                 }
1524                 RETURN(rc);
1525         }
1526
1527         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1528                               sbi->ll_md_exp, &md);
1529         if (rc) {
1530                 ptlrpc_req_finished(request);
1531                 RETURN(rc);
1532         }
1533
1534         ia_valid = op_data->op_attr.ia_valid;
1535         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1536          * cache is not cleared yet. */
1537         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1538         if (S_ISREG(inode->i_mode))
1539                 mutex_lock(&inode->i_mutex);
1540         rc = simple_setattr(dentry, &op_data->op_attr);
1541         if (S_ISREG(inode->i_mode))
1542                 mutex_unlock(&inode->i_mutex);
1543         op_data->op_attr.ia_valid = ia_valid;
1544
1545         rc = ll_update_inode(inode, &md);
1546         ptlrpc_req_finished(request);
1547
1548         RETURN(rc);
1549 }
1550
1551 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1552  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1553  * keep these values until such a time that objects are allocated for it.
1554  * We do the MDS operations first, as it is checking permissions for us.
1555  * We don't to the MDS RPC if there is nothing that we want to store there,
1556  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1557  * going to do an RPC anyways.
1558  *
1559  * If we are doing a truncate, we will send the mtime and ctime updates
1560  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1561  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1562  * at the same time.
1563  *
1564  * In case of HSMimport, we only set attr on MDS.
1565  */
1566 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1567 {
1568         struct inode *inode = dentry->d_inode;
1569         struct ll_inode_info *lli = ll_i2info(inode);
1570         struct md_op_data *op_data = NULL;
1571         int rc = 0;
1572         ENTRY;
1573
1574         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1575                "valid %x, hsm_import %d\n",
1576                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1577                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1578                hsm_import);
1579
1580         if (attr->ia_valid & ATTR_SIZE) {
1581                 /* Check new size against VFS/VM file size limit and rlimit */
1582                 rc = inode_newsize_ok(inode, attr->ia_size);
1583                 if (rc)
1584                         RETURN(rc);
1585
1586                 /* The maximum Lustre file size is variable, based on the
1587                  * OST maximum object size and number of stripes.  This
1588                  * needs another check in addition to the VFS check above. */
1589                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1590                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1591                                PFID(&lli->lli_fid), attr->ia_size,
1592                                ll_file_maxbytes(inode));
1593                         RETURN(-EFBIG);
1594                 }
1595
1596                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1597         }
1598
1599         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1600         if (attr->ia_valid & TIMES_SET_FLAGS) {
1601                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1602                     !cfs_capable(CFS_CAP_FOWNER))
1603                         RETURN(-EPERM);
1604         }
1605
1606         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1607         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1608             (attr->ia_valid & ATTR_CTIME)) {
1609                 attr->ia_ctime = CFS_CURRENT_TIME;
1610                 attr->ia_valid |= ATTR_CTIME_SET;
1611         }
1612         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1613             (attr->ia_valid & ATTR_ATIME)) {
1614                 attr->ia_atime = CFS_CURRENT_TIME;
1615                 attr->ia_valid |= ATTR_ATIME_SET;
1616         }
1617         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1618             (attr->ia_valid & ATTR_MTIME)) {
1619                 attr->ia_mtime = CFS_CURRENT_TIME;
1620                 attr->ia_valid |= ATTR_MTIME_SET;
1621         }
1622
1623         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1624                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1625                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1626                        cfs_time_current_sec());
1627
1628         if (S_ISREG(inode->i_mode)) {
1629                 if (attr->ia_valid & ATTR_SIZE)
1630                         inode_dio_write_done(inode);
1631                 mutex_unlock(&inode->i_mutex);
1632         }
1633
1634         /* We always do an MDS RPC, even if we're only changing the size;
1635          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1636
1637         OBD_ALLOC_PTR(op_data);
1638         if (op_data == NULL)
1639                 GOTO(out, rc = -ENOMEM);
1640
1641         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1642                 /* If we are changing file size, file content is
1643                  * modified, flag it. */
1644                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1645                 op_data->op_bias |= MDS_DATA_MODIFIED;
1646                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1647         }
1648
1649         op_data->op_attr = *attr;
1650
1651         rc = ll_md_setattr(dentry, op_data);
1652         if (rc)
1653                 GOTO(out, rc);
1654
1655         if (!S_ISREG(inode->i_mode) || hsm_import)
1656                 GOTO(out, rc = 0);
1657
1658         if (attr->ia_valid & (ATTR_SIZE |
1659                               ATTR_ATIME | ATTR_ATIME_SET |
1660                               ATTR_MTIME | ATTR_MTIME_SET |
1661                               ATTR_CTIME | ATTR_CTIME_SET)) {
1662                 /* For truncate and utimes sending attributes to OSTs, setting
1663                  * mtime/atime to the past will be performed under PW [0:EOF]
1664                  * extent lock (new_size:EOF for truncate).  It may seem
1665                  * excessive to send mtime/atime updates to OSTs when not
1666                  * setting times to past, but it is necessary due to possible
1667                  * time de-synchronization between MDT inode and OST objects */
1668                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1669         }
1670
1671         /* If the file was restored, it needs to set dirty flag.
1672          *
1673          * We've already sent MDS_DATA_MODIFIED flag in
1674          * ll_md_setattr() for truncate. However, the MDT refuses to
1675          * set the HS_DIRTY flag on released files, so we have to set
1676          * it again if the file has been restored. Please check how
1677          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1678          *
1679          * Please notice that if the file is not released, the previous
1680          * MDS_DATA_MODIFIED has taken effect and usually
1681          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1682          * This way we can save an RPC for common open + trunc
1683          * operation. */
1684         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1685                 struct hsm_state_set hss = {
1686                         .hss_valid = HSS_SETMASK,
1687                         .hss_setmask = HS_DIRTY,
1688                 };
1689                 int rc2;
1690
1691                 rc2 = ll_hsm_state_set(inode, &hss);
1692                 /* truncate and write can happen at the same time, so that
1693                  * the file can be set modified even though the file is not
1694                  * restored from released state, and ll_hsm_state_set() is
1695                  * not applicable for the file, and rc2 < 0 is normal in this
1696                  * case. */
1697                 if (rc2 < 0)
1698                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1699                                PFID(ll_inode2fid(inode)), rc2);
1700         }
1701
1702         EXIT;
1703 out:
1704         if (op_data != NULL)
1705                 ll_finish_md_op_data(op_data);
1706
1707         if (S_ISREG(inode->i_mode)) {
1708                 mutex_lock(&inode->i_mutex);
1709                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1710                         inode_dio_wait(inode);
1711         }
1712
1713         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1714                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1715
1716         return rc;
1717 }
1718
1719 int ll_setattr(struct dentry *de, struct iattr *attr)
1720 {
1721         int mode = de->d_inode->i_mode;
1722
1723         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1724                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1725                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1726
1727         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1728                                (ATTR_SIZE|ATTR_MODE)) &&
1729             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1730              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1731               !(attr->ia_mode & S_ISGID))))
1732                 attr->ia_valid |= ATTR_FORCE;
1733
1734         if ((attr->ia_valid & ATTR_MODE) &&
1735             (mode & S_ISUID) &&
1736             !(attr->ia_mode & S_ISUID) &&
1737             !(attr->ia_valid & ATTR_KILL_SUID))
1738                 attr->ia_valid |= ATTR_KILL_SUID;
1739
1740         if ((attr->ia_valid & ATTR_MODE) &&
1741             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1742             !(attr->ia_mode & S_ISGID) &&
1743             !(attr->ia_valid & ATTR_KILL_SGID))
1744                 attr->ia_valid |= ATTR_KILL_SGID;
1745
1746         return ll_setattr_raw(de, attr, false);
1747 }
1748
1749 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1750                        __u64 max_age, __u32 flags)
1751 {
1752         struct ll_sb_info *sbi = ll_s2sbi(sb);
1753         struct obd_statfs obd_osfs;
1754         int rc;
1755         ENTRY;
1756
1757         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1758         if (rc) {
1759                 CERROR("md_statfs fails: rc = %d\n", rc);
1760                 RETURN(rc);
1761         }
1762
1763         osfs->os_type = sb->s_magic;
1764
1765         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1766                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1767
1768         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1769                 flags |= OBD_STATFS_NODELAY;
1770
1771         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1772         if (rc) {
1773                 CERROR("obd_statfs fails: rc = %d\n", rc);
1774                 RETURN(rc);
1775         }
1776
1777         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1778                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1779                obd_osfs.os_files);
1780
1781         osfs->os_bsize = obd_osfs.os_bsize;
1782         osfs->os_blocks = obd_osfs.os_blocks;
1783         osfs->os_bfree = obd_osfs.os_bfree;
1784         osfs->os_bavail = obd_osfs.os_bavail;
1785
1786         /* If we don't have as many objects free on the OST as inodes
1787          * on the MDS, we reduce the total number of inodes to
1788          * compensate, so that the "inodes in use" number is correct.
1789          */
1790         if (obd_osfs.os_ffree < osfs->os_ffree) {
1791                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1792                         obd_osfs.os_ffree;
1793                 osfs->os_ffree = obd_osfs.os_ffree;
1794         }
1795
1796         RETURN(rc);
1797 }
1798 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1799 {
1800         struct super_block *sb = de->d_sb;
1801         struct obd_statfs osfs;
1802         __u64 fsid = huge_encode_dev(sb->s_dev);
1803         int rc;
1804
1805         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1806         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1807
1808         /* Some amount of caching on the client is allowed */
1809         rc = ll_statfs_internal(sb, &osfs,
1810                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1811                                 0);
1812         if (rc)
1813                 return rc;
1814
1815         statfs_unpack(sfs, &osfs);
1816
1817         /* We need to downshift for all 32-bit kernels, because we can't
1818          * tell if the kernel is being called via sys_statfs64() or not.
1819          * Stop before overflowing f_bsize - in which case it is better
1820          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1821         if (sizeof(long) < 8) {
1822                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1823                         sfs->f_bsize <<= 1;
1824
1825                         osfs.os_blocks >>= 1;
1826                         osfs.os_bfree >>= 1;
1827                         osfs.os_bavail >>= 1;
1828                 }
1829         }
1830
1831         sfs->f_blocks = osfs.os_blocks;
1832         sfs->f_bfree = osfs.os_bfree;
1833         sfs->f_bavail = osfs.os_bavail;
1834         sfs->f_fsid.val[0] = (__u32)fsid;
1835         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1836         return 0;
1837 }
1838
1839 void ll_inode_size_lock(struct inode *inode)
1840 {
1841         struct ll_inode_info *lli;
1842
1843         LASSERT(!S_ISDIR(inode->i_mode));
1844
1845         lli = ll_i2info(inode);
1846         mutex_lock(&lli->lli_size_mutex);
1847 }
1848
1849 void ll_inode_size_unlock(struct inode *inode)
1850 {
1851         struct ll_inode_info *lli;
1852
1853         lli = ll_i2info(inode);
1854         mutex_unlock(&lli->lli_size_mutex);
1855 }
1856
1857 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1858 {
1859         struct ll_inode_info *lli = ll_i2info(inode);
1860         struct mdt_body *body = md->body;
1861         struct ll_sb_info *sbi = ll_i2sbi(inode);
1862
1863         if (body->mbo_valid & OBD_MD_FLEASIZE)
1864                 cl_file_inode_init(inode, md);
1865
1866         if (S_ISDIR(inode->i_mode)) {
1867                 int     rc;
1868
1869                 rc = ll_update_lsm_md(inode, md);
1870                 if (rc != 0)
1871                         return rc;
1872         }
1873
1874         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1875                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1876                         ll_update_remote_perm(inode, md->remote_perm);
1877         }
1878 #ifdef CONFIG_FS_POSIX_ACL
1879         else if (body->mbo_valid & OBD_MD_FLACL) {
1880                 spin_lock(&lli->lli_lock);
1881                 if (lli->lli_posix_acl)
1882                         posix_acl_release(lli->lli_posix_acl);
1883                 lli->lli_posix_acl = md->posix_acl;
1884                 spin_unlock(&lli->lli_lock);
1885         }
1886 #endif
1887         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1888                                         sbi->ll_flags & LL_SBI_32BIT_API);
1889         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1890
1891         if (body->mbo_valid & OBD_MD_FLATIME) {
1892                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1893                         LTIME_S(inode->i_atime) = body->mbo_atime;
1894                 lli->lli_atime = body->mbo_atime;
1895         }
1896
1897         if (body->mbo_valid & OBD_MD_FLMTIME) {
1898                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1899                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1900                                "to "LPU64"\n", inode->i_ino,
1901                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1902                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1903                 }
1904                 lli->lli_mtime = body->mbo_mtime;
1905         }
1906
1907         if (body->mbo_valid & OBD_MD_FLCTIME) {
1908                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1909                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1910                 lli->lli_ctime = body->mbo_ctime;
1911         }
1912
1913         if (body->mbo_valid & OBD_MD_FLMODE)
1914                 inode->i_mode = (inode->i_mode & S_IFMT) |
1915                                 (body->mbo_mode & ~S_IFMT);
1916
1917         if (body->mbo_valid & OBD_MD_FLTYPE)
1918                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1919                                 (body->mbo_mode & S_IFMT);
1920
1921         LASSERT(inode->i_mode != 0);
1922         if (S_ISREG(inode->i_mode))
1923                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1924                                        LL_MAX_BLKSIZE_BITS);
1925         else
1926                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1927
1928         if (body->mbo_valid & OBD_MD_FLUID)
1929                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1930         if (body->mbo_valid & OBD_MD_FLGID)
1931                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1932         if (body->mbo_valid & OBD_MD_FLFLAGS)
1933                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1934         if (body->mbo_valid & OBD_MD_FLNLINK)
1935                 set_nlink(inode, body->mbo_nlink);
1936         if (body->mbo_valid & OBD_MD_FLRDEV)
1937                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1938
1939         if (body->mbo_valid & OBD_MD_FLID) {
1940                 /* FID shouldn't be changed! */
1941                 if (fid_is_sane(&lli->lli_fid)) {
1942                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1943                                  "Trying to change FID "DFID
1944                                  " to the "DFID", inode "DFID"(%p)\n",
1945                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1946                                  PFID(ll_inode2fid(inode)), inode);
1947                 } else {
1948                         lli->lli_fid = body->mbo_fid1;
1949                 }
1950         }
1951
1952         LASSERT(fid_seq(&lli->lli_fid) != 0);
1953
1954         if (body->mbo_valid & OBD_MD_FLSIZE) {
1955                 i_size_write(inode, body->mbo_size);
1956
1957                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1958                        PFID(ll_inode2fid(inode)),
1959                        (unsigned long long)body->mbo_size);
1960
1961                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1962                         inode->i_blocks = body->mbo_blocks;
1963         }
1964
1965         if (body->mbo_valid & OBD_MD_TSTATE) {
1966                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1967                  * clear it when done to ensure to start again
1968                  * glimpsing updated attrs
1969                  */
1970                 if (body->mbo_t_state & MS_RESTORE)
1971                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1972                 else
1973                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1974         }
1975
1976         return 0;
1977 }
1978
1979 int ll_read_inode2(struct inode *inode, void *opaque)
1980 {
1981         struct lustre_md *md = opaque;
1982         struct ll_inode_info *lli = ll_i2info(inode);
1983         int     rc;
1984         ENTRY;
1985
1986         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1987                PFID(&lli->lli_fid), inode);
1988
1989         /* Core attributes from the MDS first.  This is a new inode, and
1990          * the VFS doesn't zero times in the core inode so we have to do
1991          * it ourselves.  They will be overwritten by either MDS or OST
1992          * attributes - we just need to make sure they aren't newer. */
1993         LTIME_S(inode->i_mtime) = 0;
1994         LTIME_S(inode->i_atime) = 0;
1995         LTIME_S(inode->i_ctime) = 0;
1996         inode->i_rdev = 0;
1997         rc = ll_update_inode(inode, md);
1998         if (rc != 0)
1999                 RETURN(rc);
2000
2001         /* OIDEBUG(inode); */
2002
2003 #ifdef HAVE_BACKING_DEV_INFO
2004         /* initializing backing dev info. */
2005         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2006 #endif
2007         if (S_ISREG(inode->i_mode)) {
2008                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2009                 inode->i_op = &ll_file_inode_operations;
2010                 inode->i_fop = sbi->ll_fop;
2011                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2012                 EXIT;
2013         } else if (S_ISDIR(inode->i_mode)) {
2014                 inode->i_op = &ll_dir_inode_operations;
2015                 inode->i_fop = &ll_dir_operations;
2016                 EXIT;
2017         } else if (S_ISLNK(inode->i_mode)) {
2018                 inode->i_op = &ll_fast_symlink_inode_operations;
2019                 EXIT;
2020         } else {
2021                 inode->i_op = &ll_special_inode_operations;
2022
2023                 init_special_inode(inode, inode->i_mode,
2024                                    inode->i_rdev);
2025
2026                 EXIT;
2027         }
2028
2029         return 0;
2030 }
2031
2032 void ll_delete_inode(struct inode *inode)
2033 {
2034         struct ll_inode_info *lli = ll_i2info(inode);
2035         ENTRY;
2036
2037         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2038                 /* It is last chance to write out dirty pages,
2039                  * otherwise we may lose data while umount */
2040                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2041
2042         truncate_inode_pages_final(&inode->i_data);
2043
2044         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2045                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2046                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2047
2048 #ifdef HAVE_SBOPS_EVICT_INODE
2049         ll_clear_inode(inode);
2050 #endif
2051         clear_inode(inode);
2052
2053         EXIT;
2054 }
2055
2056 int ll_iocontrol(struct inode *inode, struct file *file,
2057                  unsigned int cmd, unsigned long arg)
2058 {
2059         struct ll_sb_info *sbi = ll_i2sbi(inode);
2060         struct ptlrpc_request *req = NULL;
2061         int rc, flags = 0;
2062         ENTRY;
2063
2064         switch(cmd) {
2065         case FSFILT_IOC_GETFLAGS: {
2066                 struct mdt_body *body;
2067                 struct md_op_data *op_data;
2068
2069                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2070                                              0, 0, LUSTRE_OPC_ANY,
2071                                              NULL);
2072                 if (IS_ERR(op_data))
2073                         RETURN(PTR_ERR(op_data));
2074
2075                 op_data->op_valid = OBD_MD_FLFLAGS;
2076                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2077                 ll_finish_md_op_data(op_data);
2078                 if (rc) {
2079                         CERROR("%s: failure inode "DFID": rc = %d\n",
2080                                sbi->ll_md_exp->exp_obd->obd_name,
2081                                PFID(ll_inode2fid(inode)), rc);
2082                         RETURN(-abs(rc));
2083                 }
2084
2085                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2086
2087                 flags = body->mbo_flags;
2088
2089                 ptlrpc_req_finished(req);
2090
2091                 RETURN(put_user(flags, (int __user *)arg));
2092         }
2093         case FSFILT_IOC_SETFLAGS: {
2094                 struct iattr *attr;
2095                 struct md_op_data *op_data;
2096                 struct cl_object *obj;
2097
2098                 if (get_user(flags, (int __user *)arg))
2099                         RETURN(-EFAULT);
2100
2101                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2102                                              LUSTRE_OPC_ANY, NULL);
2103                 if (IS_ERR(op_data))
2104                         RETURN(PTR_ERR(op_data));
2105
2106                 op_data->op_attr_flags = flags;
2107                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2108                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2109                 ll_finish_md_op_data(op_data);
2110                 ptlrpc_req_finished(req);
2111                 if (rc)
2112                         RETURN(rc);
2113
2114                 inode->i_flags = ll_ext_to_inode_flags(flags);
2115
2116                 obj = ll_i2info(inode)->lli_clob;
2117                 if (obj == NULL)
2118                         RETURN(0);
2119
2120                 OBD_ALLOC_PTR(attr);
2121                 if (attr == NULL)
2122                         RETURN(-ENOMEM);
2123
2124                 attr->ia_valid = ATTR_ATTR_FLAG;
2125                 rc = cl_setattr_ost(obj, attr, flags);
2126
2127                 OBD_FREE_PTR(attr);
2128                 RETURN(rc);
2129         }
2130         default:
2131                 RETURN(-ENOSYS);
2132         }
2133
2134         RETURN(0);
2135 }
2136
2137 int ll_flush_ctx(struct inode *inode)
2138 {
2139         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2140
2141         CDEBUG(D_SEC, "flush context for user %d\n",
2142                from_kuid(&init_user_ns, current_uid()));
2143
2144         obd_set_info_async(NULL, sbi->ll_md_exp,
2145                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2146                            0, NULL, NULL);
2147         obd_set_info_async(NULL, sbi->ll_dt_exp,
2148                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2149                            0, NULL, NULL);
2150         return 0;
2151 }
2152
2153 /* umount -f client means force down, don't save state */
2154 void ll_umount_begin(struct super_block *sb)
2155 {
2156         struct ll_sb_info *sbi = ll_s2sbi(sb);
2157         struct obd_device *obd;
2158         struct obd_ioctl_data *ioc_data;
2159         ENTRY;
2160
2161         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2162                sb->s_count, atomic_read(&sb->s_active));
2163
2164         obd = class_exp2obd(sbi->ll_md_exp);
2165         if (obd == NULL) {
2166                 CERROR("Invalid MDC connection handle "LPX64"\n",
2167                        sbi->ll_md_exp->exp_handle.h_cookie);
2168                 EXIT;
2169                 return;
2170         }
2171         obd->obd_force = 1;
2172
2173         obd = class_exp2obd(sbi->ll_dt_exp);
2174         if (obd == NULL) {
2175                 CERROR("Invalid LOV connection handle "LPX64"\n",
2176                        sbi->ll_dt_exp->exp_handle.h_cookie);
2177                 EXIT;
2178                 return;
2179         }
2180         obd->obd_force = 1;
2181
2182         OBD_ALLOC_PTR(ioc_data);
2183         if (ioc_data) {
2184                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2185                               sizeof *ioc_data, ioc_data, NULL);
2186
2187                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2188                               sizeof *ioc_data, ioc_data, NULL);
2189
2190                 OBD_FREE_PTR(ioc_data);
2191         }
2192
2193         /* Really, we'd like to wait until there are no requests outstanding,
2194          * and then continue.  For now, we just invalidate the requests,
2195          * schedule() and sleep one second if needed, and hope.
2196          */
2197         schedule();
2198         EXIT;
2199 }
2200
2201 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2202 {
2203         struct ll_sb_info *sbi = ll_s2sbi(sb);
2204         char *profilenm = get_profile_name(sb);
2205         int err;
2206         __u32 read_only;
2207
2208         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2209                 read_only = *flags & MS_RDONLY;
2210                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2211                                          sizeof(KEY_READ_ONLY),
2212                                          KEY_READ_ONLY, sizeof(read_only),
2213                                          &read_only, NULL);
2214                 if (err) {
2215                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2216                                       profilenm, read_only ?
2217                                       "read-only" : "read-write", err);
2218                         return err;
2219                 }
2220
2221                 if (read_only)
2222                         sb->s_flags |= MS_RDONLY;
2223                 else
2224                         sb->s_flags &= ~MS_RDONLY;
2225
2226                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2227                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2228                                       read_only ?  "read-only" : "read-write");
2229         }
2230         return 0;
2231 }
2232
2233 /**
2234  * Cleanup the open handle that is cached on MDT-side.
2235  *
2236  * For open case, the client side open handling thread may hit error
2237  * after the MDT grant the open. Under such case, the client should
2238  * send close RPC to the MDT as cleanup; otherwise, the open handle
2239  * on the MDT will be leaked there until the client umount or evicted.
2240  *
2241  * In further, if someone unlinked the file, because the open handle
2242  * holds the reference on such file/object, then it will block the
2243  * subsequent threads that want to locate such object via FID.
2244  *
2245  * \param[in] sb        super block for this file-system
2246  * \param[in] open_req  pointer to the original open request
2247  */
2248 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2249 {
2250         struct mdt_body                 *body;
2251         struct md_op_data               *op_data;
2252         struct ptlrpc_request           *close_req = NULL;
2253         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2254         ENTRY;
2255
2256         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2257         OBD_ALLOC_PTR(op_data);
2258         if (op_data == NULL) {
2259                 CWARN("%s: cannot allocate op_data to release open handle for "
2260                       DFID"\n",
2261                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2262
2263                 RETURN_EXIT;
2264         }
2265
2266         op_data->op_fid1 = body->mbo_fid1;
2267         op_data->op_handle = body->mbo_handle;
2268         op_data->op_mod_time = cfs_time_current_sec();
2269         md_close(exp, op_data, NULL, &close_req);
2270         ptlrpc_req_finished(close_req);
2271         ll_finish_md_op_data(op_data);
2272
2273         EXIT;
2274 }
2275
2276 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2277                   struct super_block *sb, struct lookup_intent *it)
2278 {
2279         struct ll_sb_info *sbi = NULL;
2280         struct lustre_md md = { NULL };
2281         int rc;
2282         ENTRY;
2283
2284         LASSERT(*inode || sb);
2285         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2286         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2287                               sbi->ll_md_exp, &md);
2288         if (rc != 0)
2289                 GOTO(cleanup, rc);
2290
2291         if (*inode) {
2292                 rc = ll_update_inode(*inode, &md);
2293                 if (rc != 0)
2294                         GOTO(out, rc);
2295         } else {
2296                 LASSERT(sb != NULL);
2297
2298                 /*
2299                  * At this point server returns to client's same fid as client
2300                  * generated for creating. So using ->fid1 is okay here.
2301                  */
2302                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2303
2304                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2305                                              sbi->ll_flags & LL_SBI_32BIT_API),
2306                                  &md);
2307                 if (IS_ERR(*inode)) {
2308 #ifdef CONFIG_FS_POSIX_ACL
2309                         if (md.posix_acl) {
2310                                 posix_acl_release(md.posix_acl);
2311                                 md.posix_acl = NULL;
2312                         }
2313 #endif
2314                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2315                         *inode = NULL;
2316                         CERROR("new_inode -fatal: rc %d\n", rc);
2317                         GOTO(out, rc);
2318                 }
2319         }
2320
2321         /* Handling piggyback layout lock.
2322          * Layout lock can be piggybacked by getattr and open request.
2323          * The lsm can be applied to inode only if it comes with a layout lock
2324          * otherwise correct layout may be overwritten, for example:
2325          * 1. proc1: mdt returns a lsm but not granting layout
2326          * 2. layout was changed by another client
2327          * 3. proc2: refresh layout and layout lock granted
2328          * 4. proc1: to apply a stale layout */
2329         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2330                 struct lustre_handle lockh;
2331                 struct ldlm_lock *lock;
2332
2333                 lockh.cookie = it->d.lustre.it_lock_handle;
2334                 lock = ldlm_handle2lock(&lockh);
2335                 LASSERT(lock != NULL);
2336                 if (ldlm_has_layout(lock)) {
2337                         struct cl_object_conf conf;
2338
2339                         memset(&conf, 0, sizeof(conf));
2340                         conf.coc_opc = OBJECT_CONF_SET;
2341                         conf.coc_inode = *inode;
2342                         conf.coc_lock = lock;
2343                         conf.u.coc_layout = md.layout;
2344                         (void)ll_layout_conf(*inode, &conf);
2345                 }
2346                 LDLM_LOCK_PUT(lock);
2347         }
2348
2349         GOTO(out, rc = 0);
2350
2351 out:
2352         md_free_lustre_md(sbi->ll_md_exp, &md);
2353
2354 cleanup:
2355         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2356                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2357
2358         return rc;
2359 }
2360
2361 int ll_obd_statfs(struct inode *inode, void __user *arg)
2362 {
2363         struct ll_sb_info *sbi = NULL;
2364         struct obd_export *exp;
2365         char *buf = NULL;
2366         struct obd_ioctl_data *data = NULL;
2367         __u32 type;
2368         int len = 0, rc;
2369
2370         if (!inode || !(sbi = ll_i2sbi(inode)))
2371                 GOTO(out_statfs, rc = -EINVAL);
2372
2373         rc = obd_ioctl_getdata(&buf, &len, arg);
2374         if (rc)
2375                 GOTO(out_statfs, rc);
2376
2377         data = (void*)buf;
2378         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2379             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2380                 GOTO(out_statfs, rc = -EINVAL);
2381
2382         if (data->ioc_inllen1 != sizeof(__u32) ||
2383             data->ioc_inllen2 != sizeof(__u32) ||
2384             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2385             data->ioc_plen2 != sizeof(struct obd_uuid))
2386                 GOTO(out_statfs, rc = -EINVAL);
2387
2388         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2389         if (type & LL_STATFS_LMV)
2390                 exp = sbi->ll_md_exp;
2391         else if (type & LL_STATFS_LOV)
2392                 exp = sbi->ll_dt_exp;
2393         else
2394                 GOTO(out_statfs, rc = -ENODEV);
2395
2396         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2397         if (rc)
2398                 GOTO(out_statfs, rc);
2399 out_statfs:
2400         if (buf)
2401                 obd_ioctl_freedata(buf, len);
2402         return rc;
2403 }
2404
2405 int ll_process_config(struct lustre_cfg *lcfg)
2406 {
2407         struct super_block *sb;
2408         unsigned long x;
2409         int rc = 0;
2410         char *ptr;
2411
2412         /* The instance name contains the sb: lustre-client-aacfe000 */
2413         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2414         if (!ptr || !*(++ptr))
2415                 return -EINVAL;
2416         if (sscanf(ptr, "%lx", &x) != 1)
2417                 return -EINVAL;
2418         sb = (struct super_block *)x;
2419         /* This better be a real Lustre superblock! */
2420         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2421
2422         /* Note we have not called client_common_fill_super yet, so
2423            proc fns must be able to handle that! */
2424         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2425                                       lcfg, sb);
2426         if (rc > 0)
2427                 rc = 0;
2428         return rc;
2429 }
2430
2431 /* this function prepares md_op_data hint for passing it down to MD stack. */
2432 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2433                                       struct inode *i1, struct inode *i2,
2434                                       const char *name, size_t namelen,
2435                                       __u32 mode, __u32 opc, void *data)
2436 {
2437         LASSERT(i1 != NULL);
2438
2439         if (name == NULL) {
2440                 /* Do not reuse namelen for something else. */
2441                 if (namelen != 0)
2442                         return ERR_PTR(-EINVAL);
2443         } else {
2444                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2445                         return ERR_PTR(-ENAMETOOLONG);
2446
2447                 if (!lu_name_is_valid_2(name, namelen))
2448                         return ERR_PTR(-EINVAL);
2449         }
2450
2451         if (op_data == NULL)
2452                 OBD_ALLOC_PTR(op_data);
2453
2454         if (op_data == NULL)
2455                 return ERR_PTR(-ENOMEM);
2456
2457         ll_i2gids(op_data->op_suppgids, i1, i2);
2458         op_data->op_fid1 = *ll_inode2fid(i1);
2459         op_data->op_default_stripe_offset = -1;
2460         if (S_ISDIR(i1->i_mode)) {
2461                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2462                 if (opc == LUSTRE_OPC_MKDIR)
2463                         op_data->op_default_stripe_offset =
2464                                    ll_i2info(i1)->lli_def_stripe_offset;
2465         }
2466
2467         if (i2) {
2468                 op_data->op_fid2 = *ll_inode2fid(i2);
2469                 if (S_ISDIR(i2->i_mode))
2470                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2471         } else {
2472                 fid_zero(&op_data->op_fid2);
2473         }
2474
2475         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2476                 op_data->op_cli_flags |= CLI_HASH64;
2477
2478         if (ll_need_32bit_api(ll_i2sbi(i1)))
2479                 op_data->op_cli_flags |= CLI_API32;
2480
2481         op_data->op_name = name;
2482         op_data->op_namelen = namelen;
2483         op_data->op_mode = mode;
2484         op_data->op_mod_time = cfs_time_current_sec();
2485         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2486         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2487         op_data->op_cap = cfs_curproc_cap_pack();
2488         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2489              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2490                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2491         } else {
2492                 op_data->op_mds = 0;
2493         }
2494         op_data->op_data = data;
2495
2496         return op_data;
2497 }
2498
2499 void ll_finish_md_op_data(struct md_op_data *op_data)
2500 {
2501         OBD_FREE_PTR(op_data);
2502 }
2503
2504 #ifdef HAVE_SUPEROPS_USE_DENTRY
2505 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2506 #else
2507 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2508 #endif
2509 {
2510         struct ll_sb_info *sbi;
2511
2512 #ifdef HAVE_SUPEROPS_USE_DENTRY
2513         LASSERT((seq != NULL) && (dentry != NULL));
2514         sbi = ll_s2sbi(dentry->d_sb);
2515 #else
2516         LASSERT((seq != NULL) && (vfs != NULL));
2517         sbi = ll_s2sbi(vfs->mnt_sb);
2518 #endif
2519
2520         if (sbi->ll_flags & LL_SBI_NOLCK)
2521                 seq_puts(seq, ",nolock");
2522
2523         if (sbi->ll_flags & LL_SBI_FLOCK)
2524                 seq_puts(seq, ",flock");
2525
2526         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2527                 seq_puts(seq, ",localflock");
2528
2529         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2530                 seq_puts(seq, ",user_xattr");
2531
2532         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2533                 seq_puts(seq, ",lazystatfs");
2534
2535         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2536                 seq_puts(seq, ",user_fid2path");
2537
2538         RETURN(0);
2539 }
2540
2541 /**
2542  * Get obd name by cmd, and copy out to user space
2543  */
2544 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2545 {
2546         struct ll_sb_info *sbi = ll_i2sbi(inode);
2547         struct obd_device *obd;
2548         ENTRY;
2549
2550         if (cmd == OBD_IOC_GETDTNAME)
2551                 obd = class_exp2obd(sbi->ll_dt_exp);
2552         else if (cmd == OBD_IOC_GETMDNAME)
2553                 obd = class_exp2obd(sbi->ll_md_exp);
2554         else
2555                 RETURN(-EINVAL);
2556
2557         if (!obd)
2558                 RETURN(-ENOENT);
2559
2560         if (copy_to_user((void __user *)arg, obd->obd_name,
2561                          strlen(obd->obd_name) + 1))
2562                 RETURN(-EFAULT);
2563
2564         RETURN(0);
2565 }
2566
2567 /**
2568  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2569  * fsname will be returned in this buffer; otherwise, a static buffer will be
2570  * used to store the fsname and returned to caller.
2571  */
2572 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2573 {
2574         static char fsname_static[MTI_NAME_MAXLEN];
2575         struct lustre_sb_info *lsi = s2lsi(sb);
2576         char *ptr;
2577         int len;
2578
2579         if (buf == NULL) {
2580                 /* this means the caller wants to use static buffer
2581                  * and it doesn't care about race. Usually this is
2582                  * in error reporting path */
2583                 buf = fsname_static;
2584                 buflen = sizeof(fsname_static);
2585         }
2586
2587         len = strlen(lsi->lsi_lmd->lmd_profile);
2588         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2589         if (ptr && (strcmp(ptr, "-client") == 0))
2590                 len -= 7;
2591
2592         if (unlikely(len >= buflen))
2593                 len = buflen - 1;
2594         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2595         buf[len] = '\0';
2596
2597         return buf;
2598 }
2599
2600 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2601 {
2602         char *path = NULL;
2603
2604         struct path p;
2605
2606         p.dentry = dentry;
2607         p.mnt = current->fs->root.mnt;
2608         path_get(&p);
2609         path = d_path(&p, buf, bufsize);
2610         path_put(&p);
2611         return path;
2612 }
2613
2614 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2615 {
2616         char *buf, *path = NULL;
2617         struct dentry *dentry = NULL;
2618         struct inode *inode = page->mapping->host;
2619
2620         /* this can be called inside spin lock so use GFP_ATOMIC. */
2621         buf = (char *)__get_free_page(GFP_ATOMIC);
2622         if (buf != NULL) {
2623                 dentry = d_find_alias(page->mapping->host);
2624                 if (dentry != NULL)
2625                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2626         }
2627
2628         CDEBUG(D_WARNING,
2629                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2630                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2631                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2632                PFID(ll_inode2fid(inode)),
2633                (path && !IS_ERR(path)) ? path : "", ioret);
2634
2635         if (dentry != NULL)
2636                 dput(dentry);
2637
2638         if (buf != NULL)
2639                 free_page((unsigned long)buf);
2640 }
2641
2642 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2643                         struct lov_user_md **kbuf)
2644 {
2645         struct lov_user_md      lum;
2646         ssize_t                 lum_size;
2647         ENTRY;
2648
2649         if (copy_from_user(&lum, md, sizeof(lum)))
2650                 RETURN(-EFAULT);
2651
2652         lum_size = ll_lov_user_md_size(&lum);
2653         if (lum_size < 0)
2654                 RETURN(lum_size);
2655
2656         OBD_ALLOC(*kbuf, lum_size);
2657         if (*kbuf == NULL)
2658                 RETURN(-ENOMEM);
2659
2660         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2661                 OBD_FREE(*kbuf, lum_size);
2662                 RETURN(-EFAULT);
2663         }
2664
2665         RETURN(lum_size);
2666 }
2667
2668 /*
2669  * Compute llite root squash state after a change of root squash
2670  * configuration setting or add/remove of a lnet nid
2671  */
2672 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2673 {
2674         struct root_squash_info *squash = &sbi->ll_squash;
2675         int i;
2676         bool matched;
2677         lnet_process_id_t id;
2678
2679         /* Update norootsquash flag */
2680         down_write(&squash->rsi_sem);
2681         if (list_empty(&squash->rsi_nosquash_nids))
2682                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2683         else {
2684                 /* Do not apply root squash as soon as one of our NIDs is
2685                  * in the nosquash_nids list */
2686                 matched = false;
2687                 i = 0;
2688                 while (LNetGetId(i++, &id) != -ENOENT) {
2689                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2690                                 continue;
2691                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2692                                 matched = true;
2693                                 break;
2694                         }
2695                 }
2696                 if (matched)
2697                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2698                 else
2699                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2700         }
2701         up_write(&squash->rsi_sem);
2702 }
2703
2704 /**
2705  * Parse linkea content to extract information about a given hardlink
2706  *
2707  * \param[in]   ldata      - Initialized linkea data
2708  * \param[in]   linkno     - Link identifier
2709  * \param[out]  parent_fid - The entry's parent FID
2710  * \param[out]  ln         - Entry name destination buffer
2711  *
2712  * \retval 0 on success
2713  * \retval Appropriate negative error code on failure
2714  */
2715 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2716                             struct lu_fid *parent_fid, struct lu_name *ln)
2717 {
2718         unsigned int    idx;
2719         int             rc;
2720         ENTRY;
2721
2722         rc = linkea_init(ldata);
2723         if (rc < 0)
2724                 RETURN(rc);
2725
2726         if (linkno >= ldata->ld_leh->leh_reccount)
2727                 /* beyond last link */
2728                 RETURN(-ENODATA);
2729
2730         linkea_first_entry(ldata);
2731         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2732                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2733                                     parent_fid);
2734                 if (idx == linkno)
2735                         break;
2736
2737                 linkea_next_entry(ldata);
2738         }
2739
2740         if (idx < linkno)
2741                 RETURN(-ENODATA);
2742
2743         RETURN(0);
2744 }
2745
2746 /**
2747  * Get parent FID and name of an identified link. Operation is performed for
2748  * a given link number, letting the caller iterate over linkno to list one or
2749  * all links of an entry.
2750  *
2751  * \param[in]     file - File descriptor against which to perform the operation
2752  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2753  *                       on and the available size. It is eventually filled with
2754  *                       the requested information or left untouched on error
2755  *
2756  * \retval - 0 on success
2757  * \retval - Appropriate negative error code on failure
2758  */
2759 int ll_getparent(struct file *file, struct getparent __user *arg)
2760 {
2761         struct dentry           *dentry = file->f_path.dentry;
2762         struct inode            *inode = dentry->d_inode;
2763         struct linkea_data      *ldata;
2764         struct lu_buf            buf = LU_BUF_NULL;
2765         struct lu_name           ln;
2766         struct lu_fid            parent_fid;
2767         __u32                    linkno;
2768         __u32                    name_size;
2769         int                      rc;
2770
2771         ENTRY;
2772
2773         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2774             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2775                 RETURN(-EPERM);
2776
2777         if (get_user(name_size, &arg->gp_name_size))
2778                 RETURN(-EFAULT);
2779
2780         if (get_user(linkno, &arg->gp_linkno))
2781                 RETURN(-EFAULT);
2782
2783         if (name_size > PATH_MAX)
2784                 RETURN(-EINVAL);
2785
2786         OBD_ALLOC(ldata, sizeof(*ldata));
2787         if (ldata == NULL)
2788                 RETURN(-ENOMEM);
2789
2790         rc = linkea_data_new(ldata, &buf);
2791         if (rc < 0)
2792                 GOTO(ldata_free, rc);
2793
2794         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2795         if (rc < 0)
2796                 GOTO(lb_free, rc);
2797
2798         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2799         if (rc < 0)
2800                 GOTO(lb_free, rc);
2801
2802         if (ln.ln_namelen >= name_size)
2803                 GOTO(lb_free, rc = -EOVERFLOW);
2804
2805         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2806                 GOTO(lb_free, rc = -EFAULT);
2807
2808         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2809                 GOTO(lb_free, rc = -EFAULT);
2810
2811         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2812                 GOTO(lb_free, rc = -EFAULT);
2813
2814 lb_free:
2815         lu_buf_free(&buf);
2816 ldata_free:
2817         OBD_FREE(ldata, sizeof(*ldata));
2818
2819         RETURN(rc);
2820 }