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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <lustre_ioctl.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 struct kmem_cache *ll_file_data_slab;
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC_PTR(sbi);
  79         if (sbi == NULL)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         sbi->ll_cache = cl_cache_init(lru_page_max);
  94         if (sbi->ll_cache == NULL) {
  95                 OBD_FREE(sbi, sizeof(*sbi));
  96                 RETURN(NULL);
  97         }
  98
  99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 100                                            SBI_DEFAULT_READAHEAD_MAX);
 101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 103                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 104
 105         ll_generate_random_uuid(uuid);
 106         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 107         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 108
 109         sbi->ll_flags |= LL_SBI_VERBOSE;
 110 #ifdef ENABLE_CHECKSUM
 111         sbi->ll_flags |= LL_SBI_CHECKSUM;
 112 #endif
 113
 114 #ifdef HAVE_LRU_RESIZE_SUPPORT
 115         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 116 #endif
 117         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 118
 119         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 120                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 121                                pp_r_hist.oh_lock);
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_w_hist.oh_lock);
 124         }
 125
 126         /* metadata statahead is enabled by default */
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134
 135         /* root squash */
 136         sbi->ll_squash.rsi_uid = 0;
 137         sbi->ll_squash.rsi_gid = 0;
 138         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 139         init_rwsem(&sbi->ll_squash.rsi_sem);
 140
 141         RETURN(sbi);
 142 }
 143
 144 static void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 151                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 152                 if (sbi->ll_cache != NULL) {
 153                         cl_cache_decref(sbi->ll_cache);
 154                         sbi->ll_cache = NULL;
 155                 }
 156                 OBD_FREE(sbi, sizeof(*sbi));
 157         }
 158         EXIT;
 159 }
 160
 161 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 162 {
 163         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 164                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 165 }
 166
 167 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 168                                     struct vfsmount *mnt)
 169 {
 170         struct inode *root = NULL;
 171         struct ll_sb_info *sbi = ll_s2sbi(sb);
 172         struct obd_device *obd;
 173         struct obd_statfs *osfs = NULL;
 174         struct ptlrpc_request *request = NULL;
 175         struct obd_connect_data *data = NULL;
 176         struct obd_uuid *uuid;
 177         struct md_op_data *op_data;
 178         struct lustre_md lmd;
 179         u64 valid;
 180         int size, err, checksum;
 181         ENTRY;
 182
 183         obd = class_name2obd(md);
 184         if (!obd) {
 185                 CERROR("MD %s: not setup or attached\n", md);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         OBD_ALLOC_PTR(data);
 190         if (data == NULL)
 191                 RETURN(-ENOMEM);
 192
 193         OBD_ALLOC_PTR(osfs);
 194         if (osfs == NULL) {
 195                 OBD_FREE_PTR(data);
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* indicate the features supported by this client */
 200         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 201                                   OBD_CONNECT_ATTRFID  |
 202                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 204                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 205                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 206                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 207                                   OBD_CONNECT_64BITHASH |
 208                                   OBD_CONNECT_EINPROGRESS |
 209                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 210                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 211                                   OBD_CONNECT_MAX_EASIZE |
 212                                   OBD_CONNECT_FLOCK_DEAD |
 213                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 214                                   OBD_CONNECT_OPEN_BY_FID |
 215                                   OBD_CONNECT_DIR_STRIPE |
 216                                   OBD_CONNECT_BULK_MBITS |
 217                                   OBD_CONNECT_SUBTREE |
 218                                   OBD_CONNECT_FLAGS2;
 219
 220         data->ocd_connect_flags2 = 0;
 221
 222 #ifdef HAVE_LRU_RESIZE_SUPPORT
 223         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 224                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 225 #endif
 226 #ifdef CONFIG_FS_POSIX_ACL
 227         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 228 #endif
 229
 230         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 231                 /* flag mdc connection as lightweight, only used for test
 232                  * purpose, use with care */
 233                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 234
 235         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 236         data->ocd_version = LUSTRE_VERSION_CODE;
 237
 238         if (sb->s_flags & MS_RDONLY)
 239                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 240         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 241                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 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         /* always ping even if server suppress_pings */
 251         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 252                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 253
 254 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
 255         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 256 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 257
 258         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 259
 260         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 261         if (err == -EBUSY) {
 262                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 263                                    "recovery, of which this client is not a "
 264                                    "part. Please wait for recovery to complete,"
 265                                    " abort, or time out.\n", md);
 266                 GOTO(out, err);
 267         } else if (err) {
 268                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 269                 GOTO(out, err);
 270         }
 271
 272         sbi->ll_md_exp->exp_connect_data = *data;
 273
 274         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 275                            LUSTRE_SEQ_METADATA);
 276         if (err) {
 277                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 278                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 279                 GOTO(out_md, err);
 280         }
 281
 282         /* For mount, we only need fs info from MDT0, and also in DNE, it
 283          * can make sure the client can be mounted as long as MDT0 is
 284          * avaible */
 285         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 286                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 287                         OBD_STATFS_FOR_MDT0);
 288         if (err)
 289                 GOTO(out_md_fid, err);
 290
 291         /* This needs to be after statfs to ensure connect has finished.
 292          * Note that "data" does NOT contain the valid connect reply.
 293          * If connecting to a 1.8 server there will be no LMV device, so
 294          * we can access the MDC export directly and exp_connect_flags will
 295          * be non-zero, but if accessing an upgraded 2.1 server it will
 296          * have the correct flags filled in.
 297          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 298         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 299         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 300             valid != CLIENT_CONNECT_MDT_REQD) {
 301                 char *buf;
 302
 303                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 304                 obd_connect_flags2str(buf, PAGE_SIZE,
 305                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 306                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 307                                    "feature(s) needed for correct operation "
 308                                    "of this client (%s). Please upgrade "
 309                                    "server or downgrade client.\n",
 310                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 311                 OBD_FREE(buf, PAGE_SIZE);
 312                 GOTO(out_md_fid, err = -EPROTO);
 313         }
 314
 315         size = sizeof(*data);
 316         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 317                            KEY_CONN_DATA,  &size, data);
 318         if (err) {
 319                 CERROR("%s: Get connect data failed: rc = %d\n",
 320                        sbi->ll_md_exp->exp_obd->obd_name, err);
 321                 GOTO(out_md_fid, err);
 322         }
 323
 324         LASSERT(osfs->os_bsize);
 325         sb->s_blocksize = osfs->os_bsize;
 326         sb->s_blocksize_bits = log2(osfs->os_bsize);
 327         sb->s_magic = LL_SUPER_MAGIC;
 328         sb->s_maxbytes = MAX_LFS_FILESIZE;
 329         sbi->ll_namelen = osfs->os_namelen;
 330         sbi->ll_mnt.mnt = current->fs->root.mnt;
 331
 332         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 333             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 334                 LCONSOLE_INFO("Disabling user_xattr feature because "
 335                               "it is not supported on the server\n");
 336                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 337         }
 338
 339         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 340 #ifdef MS_POSIXACL
 341                 sb->s_flags |= MS_POSIXACL;
 342 #endif
 343                 sbi->ll_flags |= LL_SBI_ACL;
 344         } else {
 345                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 346 #ifdef MS_POSIXACL
 347                 sb->s_flags &= ~MS_POSIXACL;
 348 #endif
 349                 sbi->ll_flags &= ~LL_SBI_ACL;
 350         }
 351
 352         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 353                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 356                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
 357         else
 358                 sbi->ll_md_brw_pages = 1;
 359
 360         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 361                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 362
 363         if (obd_connect_has_secctx(data))
 364                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 365
 366         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 367                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 368                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 369                                       "unknown maximum xattr size.\n", dt);
 370                 } else {
 371                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 372                         sbi->ll_xattr_cache_enabled = 1;
 373                 }
 374         }
 375
 376         obd = class_name2obd(dt);
 377         if (!obd) {
 378                 CERROR("DT %s: not setup or attached\n", dt);
 379                 GOTO(out_md_fid, err = -ENODEV);
 380         }
 381
 382         /* pass client page size via ocd_grant_blkbits, the server should report
 383          * back its backend blocksize for grant calculation purpose */
 384         data->ocd_grant_blkbits = PAGE_SHIFT;
 385
 386         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 387                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 388                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 389                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 390                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 391                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 392                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 393                                   OBD_CONNECT_EINPROGRESS |
 394                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 395                                   OBD_CONNECT_LAYOUTLOCK |
 396                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 397                                   OBD_CONNECT_BULK_MBITS;
 398
 399         data->ocd_connect_flags2 = 0;
 400
 401         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 402                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 403
 404         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 405                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 406                  * disabled by default, because it can still be enabled on the
 407                  * fly via /proc. As a consequence, we still need to come to an
 408                  * agreement on the supported algorithms at connect time */
 409                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 410
 411                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 412                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 413                 else
 414                         data->ocd_cksum_types = cksum_types_supported_client();
 415         }
 416
 417 #ifdef HAVE_LRU_RESIZE_SUPPORT
 418         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 419 #endif
 420         /* always ping even if server suppress_pings */
 421         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 422                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 423
 424         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 425                "ocd_grant: %d\n", data->ocd_connect_flags,
 426                data->ocd_version, data->ocd_grant);
 427
 428         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 429         obd->obd_upcall.onu_upcall = cl_ocd_update;
 430
 431         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 432
 433         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 434                           NULL);
 435         if (err == -EBUSY) {
 436                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 437                                    "recovery, of which this client is not a "
 438                                    "part.  Please wait for recovery to "
 439                                    "complete, abort, or time out.\n", dt);
 440                 GOTO(out_md, err);
 441         } else if (err) {
 442                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 443                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 444                 GOTO(out_md, err);
 445         }
 446
 447         sbi->ll_dt_exp->exp_connect_data = *data;
 448
 449         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 450                            LUSTRE_SEQ_METADATA);
 451         if (err) {
 452                 CERROR("%s: Can't init data layer FID infrastructure, "
 453                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 454                 GOTO(out_dt, err);
 455         }
 456
 457         mutex_lock(&sbi->ll_lco.lco_lock);
 458         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 459         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 460         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 461         mutex_unlock(&sbi->ll_lco.lco_lock);
 462
 463         fid_zero(&sbi->ll_root_fid);
 464         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 465                            &sbi->ll_root_fid);
 466         if (err) {
 467                 CERROR("cannot mds_connect: rc = %d\n", err);
 468                 GOTO(out_lock_cn_cb, err);
 469         }
 470         if (!fid_is_sane(&sbi->ll_root_fid)) {
 471                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 472                        sbi->ll_md_exp->exp_obd->obd_name,
 473                        PFID(&sbi->ll_root_fid));
 474                 GOTO(out_lock_cn_cb, err = -EINVAL);
 475         }
 476         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 477
 478         sb->s_op = &lustre_super_operations;
 479 #if THREAD_SIZE >= 8192 /*b=17630*/
 480         sb->s_export_op = &lustre_export_operations;
 481 #endif
 482
 483         /* make root inode
 484          * XXX: move this to after cbd setup? */
 485         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 486         if (sbi->ll_flags & LL_SBI_ACL)
 487                 valid |= OBD_MD_FLACL;
 488
 489         OBD_ALLOC_PTR(op_data);
 490         if (op_data == NULL)
 491                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 492
 493         op_data->op_fid1 = sbi->ll_root_fid;
 494         op_data->op_mode = 0;
 495         op_data->op_valid = valid;
 496
 497         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 498
 499         OBD_FREE_PTR(op_data);
 500         if (err) {
 501                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 502                        sbi->ll_md_exp->exp_obd->obd_name, err);
 503                 GOTO(out_lock_cn_cb, err);
 504         }
 505
 506         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 507                                sbi->ll_md_exp, &lmd);
 508         if (err) {
 509                 CERROR("failed to understand root inode md: rc = %d\n", err);
 510                 ptlrpc_req_finished(request);
 511                 GOTO(out_lock_cn_cb, err);
 512         }
 513
 514         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 515         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 516                                             sbi->ll_flags & LL_SBI_32BIT_API),
 517                        &lmd);
 518         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 519         ptlrpc_req_finished(request);
 520
 521         if (IS_ERR(root)) {
 522 #ifdef CONFIG_FS_POSIX_ACL
 523                 if (lmd.posix_acl) {
 524                         posix_acl_release(lmd.posix_acl);
 525                         lmd.posix_acl = NULL;
 526                 }
 527 #endif
 528                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 529                 root = NULL;
 530                 CERROR("lustre_lite: bad iget4 for root\n");
 531                 GOTO(out_root, err);
 532         }
 533
 534         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 535         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 536                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 537                                  NULL);
 538         if (err) {
 539                 CERROR("%s: Set checksum failed: rc = %d\n",
 540                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 541                 GOTO(out_root, err);
 542         }
 543         cl_sb_init(sb);
 544
 545         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 546                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 547                                  sbi->ll_cache, NULL);
 548         if (err) {
 549                 CERROR("%s: Set cache_set failed: rc = %d\n",
 550                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 551                 GOTO(out_root, err);
 552         }
 553
 554         sb->s_root = d_make_root(root);
 555         if (sb->s_root == NULL) {
 556                 CERROR("%s: can't make root dentry\n",
 557                         ll_get_fsname(sb, NULL, 0));
 558                 GOTO(out_root, err = -ENOMEM);
 559         }
 560 #ifdef HAVE_DCACHE_LOCK
 561         sb->s_root->d_op = &ll_d_ops;
 562 #endif
 563
 564         sbi->ll_sdev_orig = sb->s_dev;
 565
 566         /* We set sb->s_dev equal on all lustre clients in order to support
 567          * NFS export clustering.  NFSD requires that the FSID be the same
 568          * on all clients. */
 569         /* s_dev is also used in lt_compare() to compare two fs, but that is
 570          * only a node-local comparison. */
 571         uuid = obd_get_uuid(sbi->ll_md_exp);
 572         if (uuid != NULL)
 573                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 574
 575         if (data != NULL)
 576                 OBD_FREE_PTR(data);
 577         if (osfs != NULL)
 578                 OBD_FREE_PTR(osfs);
 579         if (proc_lustre_fs_root != NULL) {
 580                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 581                                                   dt, md);
 582                 if (err < 0) {
 583                         CERROR("%s: could not register mount in lprocfs: "
 584                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 585                         err = 0;
 586                 }
 587         }
 588
 589         RETURN(err);
 590 out_root:
 591         if (root)
 592                 iput(root);
 593 out_lock_cn_cb:
 594         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 595 out_dt:
 596         obd_disconnect(sbi->ll_dt_exp);
 597         sbi->ll_dt_exp = NULL;
 598 out_md_fid:
 599         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 600 out_md:
 601         obd_disconnect(sbi->ll_md_exp);
 602         sbi->ll_md_exp = NULL;
 603 out:
 604         if (data != NULL)
 605                 OBD_FREE_PTR(data);
 606         if (osfs != NULL)
 607                 OBD_FREE_PTR(osfs);
 608         return err;
 609 }
 610
 611 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 612 {
 613         int size, rc;
 614
 615         size = sizeof(*lmmsize);
 616         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 617                           KEY_MAX_EASIZE, &size, lmmsize);
 618         if (rc != 0) {
 619                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 620                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 621                 RETURN(rc);
 622         }
 623
 624         size = sizeof(int);
 625         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 626                           KEY_MAX_EASIZE, &size, lmmsize);
 627         if (rc)
 628                 CERROR("Get max mdsize error rc %d\n", rc);
 629
 630         RETURN(rc);
 631 }
 632
 633 /**
 634  * Get the value of the default_easize parameter.
 635  *
 636  * \see client_obd::cl_default_mds_easize
 637  *
 638  * \param[in] sbi       superblock info for this filesystem
 639  * \param[out] lmmsize  pointer to storage location for value
 640  *
 641  * \retval 0            on success
 642  * \retval negative     negated errno on failure
 643  */
 644 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 645 {
 646         int size, rc;
 647
 648         size = sizeof(int);
 649         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 650                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 651         if (rc)
 652                 CERROR("Get default mdsize error rc %d\n", rc);
 653
 654         RETURN(rc);
 655 }
 656
 657 /**
 658  * Set the default_easize parameter to the given value.
 659  *
 660  * \see client_obd::cl_default_mds_easize
 661  *
 662  * \param[in] sbi       superblock info for this filesystem
 663  * \param[in] lmmsize   the size to set
 664  *
 665  * \retval 0            on success
 666  * \retval negative     negated errno on failure
 667  */
 668 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 669 {
 670         int rc;
 671
 672         if (lmmsize < sizeof(struct lov_mds_md) ||
 673             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 674                 return -EINVAL;
 675
 676         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 677                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 678                                 sizeof(int), &lmmsize, NULL);
 679
 680         RETURN(rc);
 681 }
 682
 683 static void ll_dump_inode(struct inode *inode)
 684 {
 685         struct ll_d_hlist_node *tmp;
 686         int dentry_count = 0;
 687
 688         LASSERT(inode != NULL);
 689
 690         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 691                 dentry_count++;
 692
 693         CERROR("%s: inode %p dump: dev=%s fid="DFID
 694                " mode=%o count=%u, %d dentries\n",
 695                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 696                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 697                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 698 }
 699
 700 void lustre_dump_dentry(struct dentry *dentry, int recur)
 701 {
 702         struct list_head *tmp;
 703         int subdirs = 0;
 704
 705         LASSERT(dentry != NULL);
 706
 707         list_for_each(tmp, &dentry->d_subdirs)
 708                 subdirs++;
 709
 710         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 711                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 712                dentry->d_name.len, dentry->d_name.name,
 713                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 714                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 715                dentry->d_flags, dentry->d_fsdata, subdirs);
 716         if (dentry->d_inode != NULL)
 717                 ll_dump_inode(dentry->d_inode);
 718
 719         if (recur == 0)
 720                 return;
 721
 722         list_for_each(tmp, &dentry->d_subdirs) {
 723                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 724                 lustre_dump_dentry(d, recur - 1);
 725         }
 726 }
 727
 728 static void client_common_put_super(struct super_block *sb)
 729 {
 730         struct ll_sb_info *sbi = ll_s2sbi(sb);
 731         ENTRY;
 732
 733         cl_sb_fini(sb);
 734
 735         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 736         obd_disconnect(sbi->ll_dt_exp);
 737         sbi->ll_dt_exp = NULL;
 738
 739         lprocfs_unregister_mountpoint(sbi);
 740
 741         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 742         obd_disconnect(sbi->ll_md_exp);
 743         sbi->ll_md_exp = NULL;
 744
 745         EXIT;
 746 }
 747
 748 void ll_kill_super(struct super_block *sb)
 749 {
 750         struct ll_sb_info *sbi;
 751         ENTRY;
 752
 753         /* not init sb ?*/
 754         if (!(sb->s_flags & MS_ACTIVE))
 755                 return;
 756
 757         sbi = ll_s2sbi(sb);
 758         /* we need restore s_dev from changed for clustred NFS before put_super
 759          * because new kernels have cached s_dev and change sb->s_dev in
 760          * put_super not affected real removing devices */
 761         if (sbi) {
 762                 sb->s_dev = sbi->ll_sdev_orig;
 763                 sbi->ll_umounting = 1;
 764
 765                 /* wait running statahead threads to quit */
 766                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 767                         set_current_state(TASK_UNINTERRUPTIBLE);
 768                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 769                 }
 770         }
 771
 772         EXIT;
 773 }
 774
 775 static inline int ll_set_opt(const char *opt, char *data, int fl)
 776 {
 777         if (strncmp(opt, data, strlen(opt)) != 0)
 778                 return(0);
 779         else
 780                 return(fl);
 781 }
 782
 783 /* non-client-specific mount options are parsed in lmd_parse */
 784 static int ll_options(char *options, int *flags)
 785 {
 786         int tmp;
 787         char *s1 = options, *s2;
 788         ENTRY;
 789
 790         if (!options)
 791                 RETURN(0);
 792
 793         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 794
 795         while (*s1) {
 796                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 797                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 813                 if (tmp) {
 814                         *flags &= ~tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 823                 if (tmp) {
 824                         *flags &= ~tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("context", s1, 1);
 828                 if (tmp)
 829                         goto next;
 830                 tmp = ll_set_opt("fscontext", s1, 1);
 831                 if (tmp)
 832                         goto next;
 833                 tmp = ll_set_opt("defcontext", s1, 1);
 834                 if (tmp)
 835                         goto next;
 836                 tmp = ll_set_opt("rootcontext", s1, 1);
 837                 if (tmp)
 838                         goto next;
 839                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 840                 if (tmp) {
 841                         *flags |= tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 845                 if (tmp) {
 846                         *flags &= ~tmp;
 847                         goto next;
 848                 }
 849
 850                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 856                 if (tmp) {
 857                         *flags &= ~tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 861                 if (tmp) {
 862                         *flags |= tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 866                 if (tmp) {
 867                         *flags &= ~tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 871                 if (tmp) {
 872                         *flags |= tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 876                 if (tmp) {
 877                         *flags &= ~tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 881                 if (tmp) {
 882                         *flags |= tmp;
 883                         goto next;
 884                 }
 885                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 886                 if (tmp) {
 887                         *flags |= tmp;
 888                         goto next;
 889                 }
 890                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 891                 if (tmp) {
 892                         *flags &= ~tmp;
 893                         goto next;
 894                 }
 895                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 896                 if (tmp) {
 897                         *flags |= tmp;
 898                         goto next;
 899                 }
 900                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 901                                    s1);
 902                 RETURN(-EINVAL);
 903
 904 next:
 905                 /* Find next opt */
 906                 s2 = strchr(s1, ',');
 907                 if (s2 == NULL)
 908                         break;
 909                 s1 = s2 + 1;
 910         }
 911         RETURN(0);
 912 }
 913
 914 void ll_lli_init(struct ll_inode_info *lli)
 915 {
 916         lli->lli_inode_magic = LLI_INODE_MAGIC;
 917         lli->lli_flags = 0;
 918         spin_lock_init(&lli->lli_lock);
 919         lli->lli_posix_acl = NULL;
 920         /* Do not set lli_fid, it has been initialized already. */
 921         fid_zero(&lli->lli_pfid);
 922         lli->lli_mds_read_och = NULL;
 923         lli->lli_mds_write_och = NULL;
 924         lli->lli_mds_exec_och = NULL;
 925         lli->lli_open_fd_read_count = 0;
 926         lli->lli_open_fd_write_count = 0;
 927         lli->lli_open_fd_exec_count = 0;
 928         mutex_init(&lli->lli_och_mutex);
 929         spin_lock_init(&lli->lli_agl_lock);
 930         spin_lock_init(&lli->lli_layout_lock);
 931         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 932         lli->lli_clob = NULL;
 933
 934         init_rwsem(&lli->lli_xattrs_list_rwsem);
 935         mutex_init(&lli->lli_xattrs_enq_lock);
 936
 937         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 938         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 939                 mutex_init(&lli->lli_readdir_mutex);
 940                 lli->lli_opendir_key = NULL;
 941                 lli->lli_sai = NULL;
 942                 spin_lock_init(&lli->lli_sa_lock);
 943                 lli->lli_opendir_pid = 0;
 944                 lli->lli_sa_enabled = 0;
 945                 lli->lli_def_stripe_offset = -1;
 946         } else {
 947                 mutex_init(&lli->lli_size_mutex);
 948                 lli->lli_symlink_name = NULL;
 949                 init_rwsem(&lli->lli_trunc_sem);
 950                 range_lock_tree_init(&lli->lli_write_tree);
 951                 init_rwsem(&lli->lli_glimpse_sem);
 952                 lli->lli_glimpse_time = 0;
 953                 INIT_LIST_HEAD(&lli->lli_agl_list);
 954                 lli->lli_agl_index = 0;
 955                 lli->lli_async_rc = 0;
 956         }
 957         mutex_init(&lli->lli_layout_mutex);
 958         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 959 }
 960
 961 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 962 {
 963         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 964
 965         bdi->name = "lustre";
 966         return bdi_register(bdi, NULL, "lustre-%d",
 967                             atomic_inc_return(&ll_bdi_num));
 968 }
 969
 970 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 971 {
 972         struct lustre_profile *lprof = NULL;
 973         struct lustre_sb_info *lsi = s2lsi(sb);
 974         struct ll_sb_info *sbi;
 975         char  *dt = NULL, *md = NULL;
 976         char  *profilenm = get_profile_name(sb);
 977         struct config_llog_instance *cfg;
 978         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 979         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 980         int    err;
 981         ENTRY;
 982
 983         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 984
 985         OBD_ALLOC_PTR(cfg);
 986         if (cfg == NULL)
 987                 RETURN(-ENOMEM);
 988
 989         try_module_get(THIS_MODULE);
 990
 991         /* client additional sb info */
 992         lsi->lsi_llsbi = sbi = ll_init_sbi();
 993         if (!sbi) {
 994                 module_put(THIS_MODULE);
 995                 OBD_FREE_PTR(cfg);
 996                 RETURN(-ENOMEM);
 997         }
 998
 999         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003         err = bdi_init(&lsi->lsi_bdi);
1004         if (err)
1005                 GOTO(out_free, err);
1006         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1007 #ifdef HAVE_BDI_CAP_MAP_COPY
1008         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1009 #else
1010         lsi->lsi_bdi.capabilities = 0;
1011 #endif
1012         err = ll_bdi_register(&lsi->lsi_bdi);
1013         if (err)
1014                 GOTO(out_free, err);
1015
1016         sb->s_bdi = &lsi->lsi_bdi;
1017 #ifndef HAVE_DCACHE_LOCK
1018         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1019         sb->s_d_op = &ll_d_ops;
1020 #endif
1021
1022         /* Generate a string unique to this super, in case some joker tries
1023            to mount the same fs at two mount points.
1024            Use the address of the super itself.*/
1025         cfg->cfg_instance = sb;
1026         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1027         cfg->cfg_callback = class_config_llog_handler;
1028         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1029         /* set up client obds */
1030         err = lustre_process_log(sb, profilenm, cfg);
1031         if (err < 0)
1032                 GOTO(out_free, err);
1033
1034         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1035         lprof = class_get_profile(profilenm);
1036         if (lprof == NULL) {
1037                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1038                                    " read from the MGS.  Does that filesystem "
1039                                    "exist?\n", profilenm);
1040                 GOTO(out_free, err = -EINVAL);
1041         }
1042         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1043                lprof->lp_md, lprof->lp_dt);
1044
1045         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1046         if (!dt)
1047                 GOTO(out_free, err = -ENOMEM);
1048         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1049
1050         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1051         if (!md)
1052                 GOTO(out_free, err = -ENOMEM);
1053         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1054
1055         /* connections, registrations, sb setup */
1056         err = client_common_fill_super(sb, md, dt, mnt);
1057         if (err < 0)
1058                 GOTO(out_free, err);
1059
1060         sbi->ll_client_common_fill_super_succeeded = 1;
1061
1062 out_free:
1063         if (md)
1064                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1065         if (dt)
1066                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1067         if (lprof != NULL)
1068                 class_put_profile(lprof);
1069         if (err)
1070                 ll_put_super(sb);
1071         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1072                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1073
1074         OBD_FREE_PTR(cfg);
1075         RETURN(err);
1076 } /* ll_fill_super */
1077
1078 void ll_put_super(struct super_block *sb)
1079 {
1080         struct config_llog_instance cfg, params_cfg;
1081         struct obd_device *obd;
1082         struct lustre_sb_info *lsi = s2lsi(sb);
1083         struct ll_sb_info *sbi = ll_s2sbi(sb);
1084         char *profilenm = get_profile_name(sb);
1085         long ccc_count;
1086         int next, force = 1, rc = 0;
1087         ENTRY;
1088
1089         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1090
1091         cfg.cfg_instance = sb;
1092         lustre_end_log(sb, profilenm, &cfg);
1093
1094         params_cfg.cfg_instance = sb;
1095         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1096
1097         if (sbi->ll_md_exp) {
1098                 obd = class_exp2obd(sbi->ll_md_exp);
1099                 if (obd)
1100                         force = obd->obd_force;
1101         }
1102
1103         /* Wait for unstable pages to be committed to stable storage */
1104         if (force == 0) {
1105                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1106                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1107                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1108                         &lwi);
1109         }
1110
1111         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1112         if (force == 0 && rc != -EINTR)
1113                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1114
1115
1116         /* We need to set force before the lov_disconnect in
1117            lustre_common_put_super, since l_d cleans up osc's as well. */
1118         if (force) {
1119                 next = 0;
1120                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1121                                                      &next)) != NULL) {
1122                         obd->obd_force = force;
1123                 }
1124         }
1125
1126         if (sbi->ll_client_common_fill_super_succeeded) {
1127                 /* Only if client_common_fill_super succeeded */
1128                 client_common_put_super(sb);
1129         }
1130
1131         next = 0;
1132         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1133                 class_manual_cleanup(obd);
1134         }
1135
1136         if (sbi->ll_flags & LL_SBI_VERBOSE)
1137                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1138
1139         if (profilenm)
1140                 class_del_profile(profilenm);
1141
1142         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1143                 bdi_destroy(&lsi->lsi_bdi);
1144                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1145         }
1146
1147         ll_free_sbi(sb);
1148         lsi->lsi_llsbi = NULL;
1149
1150         lustre_common_put_super(sb);
1151
1152         cl_env_cache_purge(~0);
1153
1154         module_put(THIS_MODULE);
1155
1156         EXIT;
1157 } /* client_put_super */
1158
1159 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1160 {
1161         struct inode *inode = NULL;
1162
1163         /* NOTE: we depend on atomic igrab() -bzzz */
1164         lock_res_and_lock(lock);
1165         if (lock->l_resource->lr_lvb_inode) {
1166                 struct ll_inode_info * lli;
1167                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1168                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1169                         inode = igrab(lock->l_resource->lr_lvb_inode);
1170                 } else {
1171                         inode = lock->l_resource->lr_lvb_inode;
1172                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1173                                          D_WARNING, lock, "lr_lvb_inode %p is "
1174                                          "bogus: magic %08x",
1175                                          lock->l_resource->lr_lvb_inode,
1176                                          lli->lli_inode_magic);
1177                         inode = NULL;
1178                 }
1179         }
1180         unlock_res_and_lock(lock);
1181         return inode;
1182 }
1183
1184 void ll_dir_clear_lsm_md(struct inode *inode)
1185 {
1186         struct ll_inode_info *lli = ll_i2info(inode);
1187
1188         LASSERT(S_ISDIR(inode->i_mode));
1189
1190         if (lli->lli_lsm_md != NULL) {
1191                 lmv_free_memmd(lli->lli_lsm_md);
1192                 lli->lli_lsm_md = NULL;
1193         }
1194 }
1195
1196 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1197                                       const struct lu_fid *fid,
1198                                       struct lustre_md *md)
1199 {
1200         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1201         struct mdt_body         *body = md->body;
1202         struct inode            *inode;
1203         ino_t                   ino;
1204         ENTRY;
1205
1206         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1207         inode = iget_locked(sb, ino);
1208         if (inode == NULL) {
1209                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1210                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1211                 RETURN(ERR_PTR(-ENOENT));
1212         }
1213
1214         if (inode->i_state & I_NEW) {
1215                 struct ll_inode_info *lli = ll_i2info(inode);
1216                 struct lmv_stripe_md *lsm = md->lmv;
1217
1218                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1219                                 (body->mbo_mode & S_IFMT);
1220                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1221                          PFID(fid));
1222
1223                 LTIME_S(inode->i_mtime) = 0;
1224                 LTIME_S(inode->i_atime) = 0;
1225                 LTIME_S(inode->i_ctime) = 0;
1226                 inode->i_rdev = 0;
1227
1228 #ifdef HAVE_BACKING_DEV_INFO
1229                 /* initializing backing dev info. */
1230                 inode->i_mapping->backing_dev_info =
1231                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1232 #endif
1233                 inode->i_op = &ll_dir_inode_operations;
1234                 inode->i_fop = &ll_dir_operations;
1235                 lli->lli_fid = *fid;
1236                 ll_lli_init(lli);
1237
1238                 LASSERT(lsm != NULL);
1239                 /* master object FID */
1240                 lli->lli_pfid = body->mbo_fid1;
1241                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1242                        lli, PFID(fid), PFID(&lli->lli_pfid));
1243                 unlock_new_inode(inode);
1244         }
1245
1246         RETURN(inode);
1247 }
1248
1249 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1250 {
1251         struct lu_fid *fid;
1252         struct lmv_stripe_md *lsm = md->lmv;
1253         int i;
1254
1255         LASSERT(lsm != NULL);
1256         /* XXX sigh, this lsm_root initialization should be in
1257          * LMV layer, but it needs ll_iget right now, so we
1258          * put this here right now. */
1259         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1260                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1261                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1262                 /* Unfortunately ll_iget will call ll_update_inode,
1263                  * where the initialization of slave inode is slightly
1264                  * different, so it reset lsm_md to NULL to avoid
1265                  * initializing lsm for slave inode. */
1266                 /* For migrating inode, master stripe and master object will
1267                  * be same, so we only need assign this inode */
1268                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1269                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1270                 else
1271                         lsm->lsm_md_oinfo[i].lmo_root =
1272                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1273
1274                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1275                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1276
1277                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1278                         return rc;
1279                 }
1280         }
1281
1282         return 0;
1283 }
1284
1285 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1286                                 const struct lmv_stripe_md *lsm_md2)
1287 {
1288         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1289                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1290                lsm_md1->lsm_md_master_mdt_index ==
1291                                         lsm_md2->lsm_md_master_mdt_index &&
1292                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1293                lsm_md1->lsm_md_layout_version ==
1294                                         lsm_md2->lsm_md_layout_version &&
1295                strcmp(lsm_md1->lsm_md_pool_name,
1296                       lsm_md2->lsm_md_pool_name) == 0;
1297 }
1298
1299 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1300 {
1301         struct ll_inode_info *lli = ll_i2info(inode);
1302         struct lmv_stripe_md *lsm = md->lmv;
1303         int     rc;
1304         ENTRY;
1305
1306         LASSERT(S_ISDIR(inode->i_mode));
1307         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1308                PFID(ll_inode2fid(inode)));
1309
1310         /* no striped information from request. */
1311         if (lsm == NULL) {
1312                 if (lli->lli_lsm_md == NULL) {
1313                         RETURN(0);
1314                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1315                                                 LMV_HASH_FLAG_MIGRATION) {
1316                         /* migration is done, the temporay MIGRATE layout has
1317                          * been removed */
1318                         CDEBUG(D_INODE, DFID" finish migration.\n",
1319                                PFID(ll_inode2fid(inode)));
1320                         lmv_free_memmd(lli->lli_lsm_md);
1321                         lli->lli_lsm_md = NULL;
1322                         RETURN(0);
1323                 } else {
1324                         /* The lustre_md from req does not include stripeEA,
1325                          * see ll_md_setattr */
1326                         RETURN(0);
1327                 }
1328         }
1329
1330         /* set the directory layout */
1331         if (lli->lli_lsm_md == NULL) {
1332                 struct cl_attr  *attr;
1333
1334                 rc = ll_init_lsm_md(inode, md);
1335                 if (rc != 0)
1336                         RETURN(rc);
1337
1338                 /* set md->lmv to NULL, so the following free lustre_md
1339                  * will not free this lsm */
1340                 md->lmv = NULL;
1341                 lli->lli_lsm_md = lsm;
1342
1343                 OBD_ALLOC_PTR(attr);
1344                 if (attr == NULL)
1345                         RETURN(-ENOMEM);
1346
1347                 /* validate the lsm */
1348                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1349                                    ll_md_blocking_ast);
1350                 if (rc != 0) {
1351                         OBD_FREE_PTR(attr);
1352                         RETURN(rc);
1353                 }
1354
1355                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1356                         md->body->mbo_nlink = attr->cat_nlink;
1357                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1358                         md->body->mbo_size = attr->cat_size;
1359                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1360                         md->body->mbo_atime = attr->cat_atime;
1361                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1362                         md->body->mbo_ctime = attr->cat_ctime;
1363                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1364                         md->body->mbo_mtime = attr->cat_mtime;
1365
1366                 OBD_FREE_PTR(attr);
1367
1368                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1369                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1370                 RETURN(0);
1371         }
1372
1373         /* Compare the old and new stripe information */
1374         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1375                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1376                 int                     idx;
1377
1378                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1379                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1380                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1381                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1382                        inode, lsm, old_lsm,
1383                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1384                        lsm->lsm_md_stripe_count,
1385                        old_lsm->lsm_md_stripe_count,
1386                        lsm->lsm_md_master_mdt_index,
1387                        old_lsm->lsm_md_master_mdt_index,
1388                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1389                        lsm->lsm_md_layout_version,
1390                        old_lsm->lsm_md_layout_version,
1391                        lsm->lsm_md_pool_name,
1392                        old_lsm->lsm_md_pool_name);
1393
1394                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1395                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1396                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1397                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1398                 }
1399
1400                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1401                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1402                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1404                 }
1405
1406                 RETURN(-EIO);
1407         }
1408
1409         RETURN(0);
1410 }
1411
1412 void ll_clear_inode(struct inode *inode)
1413 {
1414         struct ll_inode_info *lli = ll_i2info(inode);
1415         struct ll_sb_info *sbi = ll_i2sbi(inode);
1416         ENTRY;
1417
1418         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1419                PFID(ll_inode2fid(inode)), inode);
1420
1421         if (S_ISDIR(inode->i_mode)) {
1422                 /* these should have been cleared in ll_file_release */
1423                 LASSERT(lli->lli_opendir_key == NULL);
1424                 LASSERT(lli->lli_sai == NULL);
1425                 LASSERT(lli->lli_opendir_pid == 0);
1426         }
1427
1428         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1429
1430         LASSERT(!lli->lli_open_fd_write_count);
1431         LASSERT(!lli->lli_open_fd_read_count);
1432         LASSERT(!lli->lli_open_fd_exec_count);
1433
1434         if (lli->lli_mds_write_och)
1435                 ll_md_real_close(inode, FMODE_WRITE);
1436         if (lli->lli_mds_exec_och)
1437                 ll_md_real_close(inode, FMODE_EXEC);
1438         if (lli->lli_mds_read_och)
1439                 ll_md_real_close(inode, FMODE_READ);
1440
1441         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1442                 OBD_FREE(lli->lli_symlink_name,
1443                          strlen(lli->lli_symlink_name) + 1);
1444                 lli->lli_symlink_name = NULL;
1445         }
1446
1447         ll_xattr_cache_destroy(inode);
1448
1449 #ifdef CONFIG_FS_POSIX_ACL
1450         if (lli->lli_posix_acl) {
1451                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1452                 posix_acl_release(lli->lli_posix_acl);
1453                 lli->lli_posix_acl = NULL;
1454         }
1455 #endif
1456         lli->lli_inode_magic = LLI_INODE_DEAD;
1457
1458         if (S_ISDIR(inode->i_mode))
1459                 ll_dir_clear_lsm_md(inode);
1460         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1461                 LASSERT(list_empty(&lli->lli_agl_list));
1462
1463         /*
1464          * XXX This has to be done before lsm is freed below, because
1465          * cl_object still uses inode lsm.
1466          */
1467         cl_inode_fini(inode);
1468
1469         EXIT;
1470 }
1471
1472 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1473 {
1474         struct lustre_md md;
1475         struct inode *inode = dentry->d_inode;
1476         struct ll_sb_info *sbi = ll_i2sbi(inode);
1477         struct ptlrpc_request *request = NULL;
1478         int rc, ia_valid;
1479         ENTRY;
1480
1481         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1482                                      LUSTRE_OPC_ANY, NULL);
1483         if (IS_ERR(op_data))
1484                 RETURN(PTR_ERR(op_data));
1485
1486         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1487         if (rc) {
1488                 ptlrpc_req_finished(request);
1489                 if (rc == -ENOENT) {
1490                         clear_nlink(inode);
1491                         /* Unlinked special device node? Or just a race?
1492                          * Pretend we done everything. */
1493                         if (!S_ISREG(inode->i_mode) &&
1494                             !S_ISDIR(inode->i_mode)) {
1495                                 ia_valid = op_data->op_attr.ia_valid;
1496                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1497                                 rc = simple_setattr(dentry, &op_data->op_attr);
1498                                 op_data->op_attr.ia_valid = ia_valid;
1499                         }
1500                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1501                         CERROR("md_setattr fails: rc = %d\n", rc);
1502                 }
1503                 RETURN(rc);
1504         }
1505
1506         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1507                               sbi->ll_md_exp, &md);
1508         if (rc) {
1509                 ptlrpc_req_finished(request);
1510                 RETURN(rc);
1511         }
1512
1513         ia_valid = op_data->op_attr.ia_valid;
1514         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1515          * cache is not cleared yet. */
1516         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1517         if (S_ISREG(inode->i_mode))
1518                 inode_lock(inode);
1519         rc = simple_setattr(dentry, &op_data->op_attr);
1520         if (S_ISREG(inode->i_mode))
1521                 inode_unlock(inode);
1522         op_data->op_attr.ia_valid = ia_valid;
1523
1524         rc = ll_update_inode(inode, &md);
1525         ptlrpc_req_finished(request);
1526
1527         RETURN(rc);
1528 }
1529
1530 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1531  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1532  * keep these values until such a time that objects are allocated for it.
1533  * We do the MDS operations first, as it is checking permissions for us.
1534  * We don't to the MDS RPC if there is nothing that we want to store there,
1535  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1536  * going to do an RPC anyways.
1537  *
1538  * If we are doing a truncate, we will send the mtime and ctime updates
1539  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1540  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1541  * at the same time.
1542  *
1543  * In case of HSMimport, we only set attr on MDS.
1544  */
1545 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1546 {
1547         struct inode *inode = dentry->d_inode;
1548         struct ll_inode_info *lli = ll_i2info(inode);
1549         struct md_op_data *op_data = NULL;
1550         int rc = 0;
1551         ENTRY;
1552
1553         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1554                "valid %x, hsm_import %d\n",
1555                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1556                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1557                hsm_import);
1558
1559         if (attr->ia_valid & ATTR_SIZE) {
1560                 /* Check new size against VFS/VM file size limit and rlimit */
1561                 rc = inode_newsize_ok(inode, attr->ia_size);
1562                 if (rc)
1563                         RETURN(rc);
1564
1565                 /* The maximum Lustre file size is variable, based on the
1566                  * OST maximum object size and number of stripes.  This
1567                  * needs another check in addition to the VFS check above. */
1568                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1569                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1570                                PFID(&lli->lli_fid), attr->ia_size,
1571                                ll_file_maxbytes(inode));
1572                         RETURN(-EFBIG);
1573                 }
1574
1575                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1576         }
1577
1578         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1579         if (attr->ia_valid & TIMES_SET_FLAGS) {
1580                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1581                     !cfs_capable(CFS_CAP_FOWNER))
1582                         RETURN(-EPERM);
1583         }
1584
1585         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1586         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1587             (attr->ia_valid & ATTR_CTIME)) {
1588                 attr->ia_ctime = CURRENT_TIME;
1589                 attr->ia_valid |= ATTR_CTIME_SET;
1590         }
1591         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1592             (attr->ia_valid & ATTR_ATIME)) {
1593                 attr->ia_atime = CURRENT_TIME;
1594                 attr->ia_valid |= ATTR_ATIME_SET;
1595         }
1596         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1597             (attr->ia_valid & ATTR_MTIME)) {
1598                 attr->ia_mtime = CURRENT_TIME;
1599                 attr->ia_valid |= ATTR_MTIME_SET;
1600         }
1601
1602         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1603                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1604                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1605                        cfs_time_current_sec());
1606
1607         if (S_ISREG(inode->i_mode)) {
1608                 if (attr->ia_valid & ATTR_SIZE)
1609                         inode_dio_write_done(inode);
1610                 inode_unlock(inode);
1611         }
1612
1613         /* We always do an MDS RPC, even if we're only changing the size;
1614          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1615
1616         OBD_ALLOC_PTR(op_data);
1617         if (op_data == NULL)
1618                 GOTO(out, rc = -ENOMEM);
1619
1620         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1621                 /* If we are changing file size, file content is
1622                  * modified, flag it. */
1623                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1624                 op_data->op_bias |= MDS_DATA_MODIFIED;
1625                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1626         }
1627
1628         op_data->op_attr = *attr;
1629
1630         rc = ll_md_setattr(dentry, op_data);
1631         if (rc)
1632                 GOTO(out, rc);
1633
1634         if (!S_ISREG(inode->i_mode) || hsm_import)
1635                 GOTO(out, rc = 0);
1636
1637         if (attr->ia_valid & (ATTR_SIZE |
1638                               ATTR_ATIME | ATTR_ATIME_SET |
1639                               ATTR_MTIME | ATTR_MTIME_SET |
1640                               ATTR_CTIME | ATTR_CTIME_SET)) {
1641                 /* For truncate and utimes sending attributes to OSTs, setting
1642                  * mtime/atime to the past will be performed under PW [0:EOF]
1643                  * extent lock (new_size:EOF for truncate).  It may seem
1644                  * excessive to send mtime/atime updates to OSTs when not
1645                  * setting times to past, but it is necessary due to possible
1646                  * time de-synchronization between MDT inode and OST objects */
1647                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1648         }
1649
1650         /* If the file was restored, it needs to set dirty flag.
1651          *
1652          * We've already sent MDS_DATA_MODIFIED flag in
1653          * ll_md_setattr() for truncate. However, the MDT refuses to
1654          * set the HS_DIRTY flag on released files, so we have to set
1655          * it again if the file has been restored. Please check how
1656          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1657          *
1658          * Please notice that if the file is not released, the previous
1659          * MDS_DATA_MODIFIED has taken effect and usually
1660          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1661          * This way we can save an RPC for common open + trunc
1662          * operation. */
1663         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1664                 struct hsm_state_set hss = {
1665                         .hss_valid = HSS_SETMASK,
1666                         .hss_setmask = HS_DIRTY,
1667                 };
1668                 int rc2;
1669
1670                 rc2 = ll_hsm_state_set(inode, &hss);
1671                 /* truncate and write can happen at the same time, so that
1672                  * the file can be set modified even though the file is not
1673                  * restored from released state, and ll_hsm_state_set() is
1674                  * not applicable for the file, and rc2 < 0 is normal in this
1675                  * case. */
1676                 if (rc2 < 0)
1677                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1678                                PFID(ll_inode2fid(inode)), rc2);
1679         }
1680
1681         EXIT;
1682 out:
1683         if (op_data != NULL)
1684                 ll_finish_md_op_data(op_data);
1685
1686         if (S_ISREG(inode->i_mode)) {
1687                 inode_lock(inode);
1688                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1689                         inode_dio_wait(inode);
1690         }
1691
1692         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1693                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1694
1695         return rc;
1696 }
1697
1698 int ll_setattr(struct dentry *de, struct iattr *attr)
1699 {
1700         int mode = de->d_inode->i_mode;
1701
1702         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1703                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1704                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1705
1706         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1707                                (ATTR_SIZE|ATTR_MODE)) &&
1708             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1709              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1710               !(attr->ia_mode & S_ISGID))))
1711                 attr->ia_valid |= ATTR_FORCE;
1712
1713         if ((attr->ia_valid & ATTR_MODE) &&
1714             (mode & S_ISUID) &&
1715             !(attr->ia_mode & S_ISUID) &&
1716             !(attr->ia_valid & ATTR_KILL_SUID))
1717                 attr->ia_valid |= ATTR_KILL_SUID;
1718
1719         if ((attr->ia_valid & ATTR_MODE) &&
1720             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1721             !(attr->ia_mode & S_ISGID) &&
1722             !(attr->ia_valid & ATTR_KILL_SGID))
1723                 attr->ia_valid |= ATTR_KILL_SGID;
1724
1725         return ll_setattr_raw(de, attr, false);
1726 }
1727
1728 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1729                        __u64 max_age, __u32 flags)
1730 {
1731         struct ll_sb_info *sbi = ll_s2sbi(sb);
1732         struct obd_statfs obd_osfs;
1733         int rc;
1734         ENTRY;
1735
1736         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1737         if (rc) {
1738                 CERROR("md_statfs fails: rc = %d\n", rc);
1739                 RETURN(rc);
1740         }
1741
1742         osfs->os_type = sb->s_magic;
1743
1744         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1745                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1746
1747         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1748                 flags |= OBD_STATFS_NODELAY;
1749
1750         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1751         if (rc) {
1752                 CERROR("obd_statfs fails: rc = %d\n", rc);
1753                 RETURN(rc);
1754         }
1755
1756         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1757                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1758                obd_osfs.os_files);
1759
1760         osfs->os_bsize = obd_osfs.os_bsize;
1761         osfs->os_blocks = obd_osfs.os_blocks;
1762         osfs->os_bfree = obd_osfs.os_bfree;
1763         osfs->os_bavail = obd_osfs.os_bavail;
1764
1765         /* If we don't have as many objects free on the OST as inodes
1766          * on the MDS, we reduce the total number of inodes to
1767          * compensate, so that the "inodes in use" number is correct.
1768          */
1769         if (obd_osfs.os_ffree < osfs->os_ffree) {
1770                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1771                         obd_osfs.os_ffree;
1772                 osfs->os_ffree = obd_osfs.os_ffree;
1773         }
1774
1775         RETURN(rc);
1776 }
1777 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1778 {
1779         struct super_block *sb = de->d_sb;
1780         struct obd_statfs osfs;
1781         __u64 fsid = huge_encode_dev(sb->s_dev);
1782         int rc;
1783
1784         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1785         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1786
1787         /* Some amount of caching on the client is allowed */
1788         rc = ll_statfs_internal(sb, &osfs,
1789                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1790                                 0);
1791         if (rc)
1792                 return rc;
1793
1794         statfs_unpack(sfs, &osfs);
1795
1796         /* We need to downshift for all 32-bit kernels, because we can't
1797          * tell if the kernel is being called via sys_statfs64() or not.
1798          * Stop before overflowing f_bsize - in which case it is better
1799          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1800         if (sizeof(long) < 8) {
1801                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1802                         sfs->f_bsize <<= 1;
1803
1804                         osfs.os_blocks >>= 1;
1805                         osfs.os_bfree >>= 1;
1806                         osfs.os_bavail >>= 1;
1807                 }
1808         }
1809
1810         sfs->f_blocks = osfs.os_blocks;
1811         sfs->f_bfree = osfs.os_bfree;
1812         sfs->f_bavail = osfs.os_bavail;
1813         sfs->f_fsid.val[0] = (__u32)fsid;
1814         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1815         return 0;
1816 }
1817
1818 void ll_inode_size_lock(struct inode *inode)
1819 {
1820         struct ll_inode_info *lli;
1821
1822         LASSERT(!S_ISDIR(inode->i_mode));
1823
1824         lli = ll_i2info(inode);
1825         mutex_lock(&lli->lli_size_mutex);
1826 }
1827
1828 void ll_inode_size_unlock(struct inode *inode)
1829 {
1830         struct ll_inode_info *lli;
1831
1832         lli = ll_i2info(inode);
1833         mutex_unlock(&lli->lli_size_mutex);
1834 }
1835
1836 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1837 {
1838         struct ll_inode_info *lli = ll_i2info(inode);
1839         struct mdt_body *body = md->body;
1840         struct ll_sb_info *sbi = ll_i2sbi(inode);
1841
1842         if (body->mbo_valid & OBD_MD_FLEASIZE)
1843                 cl_file_inode_init(inode, md);
1844
1845         if (S_ISDIR(inode->i_mode)) {
1846                 int     rc;
1847
1848                 rc = ll_update_lsm_md(inode, md);
1849                 if (rc != 0)
1850                         return rc;
1851         }
1852
1853 #ifdef CONFIG_FS_POSIX_ACL
1854         if (body->mbo_valid & OBD_MD_FLACL) {
1855                 spin_lock(&lli->lli_lock);
1856                 if (lli->lli_posix_acl)
1857                         posix_acl_release(lli->lli_posix_acl);
1858                 lli->lli_posix_acl = md->posix_acl;
1859                 spin_unlock(&lli->lli_lock);
1860         }
1861 #endif
1862         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1863                                         sbi->ll_flags & LL_SBI_32BIT_API);
1864         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1865
1866         if (body->mbo_valid & OBD_MD_FLATIME) {
1867                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1868                         LTIME_S(inode->i_atime) = body->mbo_atime;
1869                 lli->lli_atime = body->mbo_atime;
1870         }
1871
1872         if (body->mbo_valid & OBD_MD_FLMTIME) {
1873                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1874                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1875                                "to %llu\n", inode->i_ino,
1876                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1877                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1878                 }
1879                 lli->lli_mtime = body->mbo_mtime;
1880         }
1881
1882         if (body->mbo_valid & OBD_MD_FLCTIME) {
1883                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1884                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1885                 lli->lli_ctime = body->mbo_ctime;
1886         }
1887
1888         if (body->mbo_valid & OBD_MD_FLMODE)
1889                 inode->i_mode = (inode->i_mode & S_IFMT) |
1890                                 (body->mbo_mode & ~S_IFMT);
1891
1892         if (body->mbo_valid & OBD_MD_FLTYPE)
1893                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1894                                 (body->mbo_mode & S_IFMT);
1895
1896         LASSERT(inode->i_mode != 0);
1897         if (S_ISREG(inode->i_mode))
1898                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1899                                        LL_MAX_BLKSIZE_BITS);
1900         else
1901                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1902
1903         if (body->mbo_valid & OBD_MD_FLUID)
1904                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1905         if (body->mbo_valid & OBD_MD_FLGID)
1906                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1907         if (body->mbo_valid & OBD_MD_FLFLAGS)
1908                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1909         if (body->mbo_valid & OBD_MD_FLNLINK)
1910                 set_nlink(inode, body->mbo_nlink);
1911         if (body->mbo_valid & OBD_MD_FLRDEV)
1912                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1913
1914         if (body->mbo_valid & OBD_MD_FLID) {
1915                 /* FID shouldn't be changed! */
1916                 if (fid_is_sane(&lli->lli_fid)) {
1917                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1918                                  "Trying to change FID "DFID
1919                                  " to the "DFID", inode "DFID"(%p)\n",
1920                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1921                                  PFID(ll_inode2fid(inode)), inode);
1922                 } else {
1923                         lli->lli_fid = body->mbo_fid1;
1924                 }
1925         }
1926
1927         LASSERT(fid_seq(&lli->lli_fid) != 0);
1928
1929         if (body->mbo_valid & OBD_MD_FLSIZE) {
1930                 i_size_write(inode, body->mbo_size);
1931
1932                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1933                        PFID(ll_inode2fid(inode)),
1934                        (unsigned long long)body->mbo_size);
1935
1936                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1937                         inode->i_blocks = body->mbo_blocks;
1938         }
1939
1940         if (body->mbo_valid & OBD_MD_TSTATE) {
1941                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1942                  * clear it when done to ensure to start again
1943                  * glimpsing updated attrs
1944                  */
1945                 if (body->mbo_t_state & MS_RESTORE)
1946                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1947                 else
1948                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1949         }
1950
1951         return 0;
1952 }
1953
1954 int ll_read_inode2(struct inode *inode, void *opaque)
1955 {
1956         struct lustre_md *md = opaque;
1957         struct ll_inode_info *lli = ll_i2info(inode);
1958         int     rc;
1959         ENTRY;
1960
1961         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1962                PFID(&lli->lli_fid), inode);
1963
1964         /* Core attributes from the MDS first.  This is a new inode, and
1965          * the VFS doesn't zero times in the core inode so we have to do
1966          * it ourselves.  They will be overwritten by either MDS or OST
1967          * attributes - we just need to make sure they aren't newer. */
1968         LTIME_S(inode->i_mtime) = 0;
1969         LTIME_S(inode->i_atime) = 0;
1970         LTIME_S(inode->i_ctime) = 0;
1971         inode->i_rdev = 0;
1972         rc = ll_update_inode(inode, md);
1973         if (rc != 0)
1974                 RETURN(rc);
1975
1976         /* OIDEBUG(inode); */
1977
1978 #ifdef HAVE_BACKING_DEV_INFO
1979         /* initializing backing dev info. */
1980         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1981 #endif
1982         if (S_ISREG(inode->i_mode)) {
1983                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1984                 inode->i_op = &ll_file_inode_operations;
1985                 inode->i_fop = sbi->ll_fop;
1986                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1987                 EXIT;
1988         } else if (S_ISDIR(inode->i_mode)) {
1989                 inode->i_op = &ll_dir_inode_operations;
1990                 inode->i_fop = &ll_dir_operations;
1991                 EXIT;
1992         } else if (S_ISLNK(inode->i_mode)) {
1993                 inode->i_op = &ll_fast_symlink_inode_operations;
1994                 EXIT;
1995         } else {
1996                 inode->i_op = &ll_special_inode_operations;
1997
1998                 init_special_inode(inode, inode->i_mode,
1999                                    inode->i_rdev);
2000
2001                 EXIT;
2002         }
2003
2004         return 0;
2005 }
2006
2007 void ll_delete_inode(struct inode *inode)
2008 {
2009         struct ll_inode_info *lli = ll_i2info(inode);
2010         ENTRY;
2011
2012         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2013                 /* It is last chance to write out dirty pages,
2014                  * otherwise we may lose data while umount */
2015                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2016
2017         truncate_inode_pages_final(&inode->i_data);
2018
2019         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2020                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2021                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2022
2023 #ifdef HAVE_SBOPS_EVICT_INODE
2024         ll_clear_inode(inode);
2025 #endif
2026         clear_inode(inode);
2027
2028         EXIT;
2029 }
2030
2031 int ll_iocontrol(struct inode *inode, struct file *file,
2032                  unsigned int cmd, unsigned long arg)
2033 {
2034         struct ll_sb_info *sbi = ll_i2sbi(inode);
2035         struct ptlrpc_request *req = NULL;
2036         int rc, flags = 0;
2037         ENTRY;
2038
2039         switch(cmd) {
2040         case FSFILT_IOC_GETFLAGS: {
2041                 struct mdt_body *body;
2042                 struct md_op_data *op_data;
2043
2044                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2045                                              0, 0, LUSTRE_OPC_ANY,
2046                                              NULL);
2047                 if (IS_ERR(op_data))
2048                         RETURN(PTR_ERR(op_data));
2049
2050                 op_data->op_valid = OBD_MD_FLFLAGS;
2051                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2052                 ll_finish_md_op_data(op_data);
2053                 if (rc) {
2054                         CERROR("%s: failure inode "DFID": rc = %d\n",
2055                                sbi->ll_md_exp->exp_obd->obd_name,
2056                                PFID(ll_inode2fid(inode)), rc);
2057                         RETURN(-abs(rc));
2058                 }
2059
2060                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2061
2062                 flags = body->mbo_flags;
2063
2064                 ptlrpc_req_finished(req);
2065
2066                 RETURN(put_user(flags, (int __user *)arg));
2067         }
2068         case FSFILT_IOC_SETFLAGS: {
2069                 struct iattr *attr;
2070                 struct md_op_data *op_data;
2071                 struct cl_object *obj;
2072
2073                 if (get_user(flags, (int __user *)arg))
2074                         RETURN(-EFAULT);
2075
2076                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2077                                              LUSTRE_OPC_ANY, NULL);
2078                 if (IS_ERR(op_data))
2079                         RETURN(PTR_ERR(op_data));
2080
2081                 op_data->op_attr_flags = flags;
2082                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2083                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2084                 ll_finish_md_op_data(op_data);
2085                 ptlrpc_req_finished(req);
2086                 if (rc)
2087                         RETURN(rc);
2088
2089                 inode->i_flags = ll_ext_to_inode_flags(flags);
2090
2091                 obj = ll_i2info(inode)->lli_clob;
2092                 if (obj == NULL)
2093                         RETURN(0);
2094
2095                 OBD_ALLOC_PTR(attr);
2096                 if (attr == NULL)
2097                         RETURN(-ENOMEM);
2098
2099                 attr->ia_valid = ATTR_ATTR_FLAG;
2100                 rc = cl_setattr_ost(obj, attr, flags);
2101
2102                 OBD_FREE_PTR(attr);
2103                 RETURN(rc);
2104         }
2105         default:
2106                 RETURN(-ENOSYS);
2107         }
2108
2109         RETURN(0);
2110 }
2111
2112 int ll_flush_ctx(struct inode *inode)
2113 {
2114         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2115
2116         CDEBUG(D_SEC, "flush context for user %d\n",
2117                from_kuid(&init_user_ns, current_uid()));
2118
2119         obd_set_info_async(NULL, sbi->ll_md_exp,
2120                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2121                            0, NULL, NULL);
2122         obd_set_info_async(NULL, sbi->ll_dt_exp,
2123                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2124                            0, NULL, NULL);
2125         return 0;
2126 }
2127
2128 /* umount -f client means force down, don't save state */
2129 void ll_umount_begin(struct super_block *sb)
2130 {
2131         struct ll_sb_info *sbi = ll_s2sbi(sb);
2132         struct obd_device *obd;
2133         struct obd_ioctl_data *ioc_data;
2134         struct l_wait_info lwi;
2135         wait_queue_head_t waitq;
2136         ENTRY;
2137
2138         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2139                sb->s_count, atomic_read(&sb->s_active));
2140
2141         obd = class_exp2obd(sbi->ll_md_exp);
2142         if (obd == NULL) {
2143                 CERROR("Invalid MDC connection handle %#llx\n",
2144                        sbi->ll_md_exp->exp_handle.h_cookie);
2145                 EXIT;
2146                 return;
2147         }
2148         obd->obd_force = 1;
2149
2150         obd = class_exp2obd(sbi->ll_dt_exp);
2151         if (obd == NULL) {
2152                 CERROR("Invalid LOV connection handle %#llx\n",
2153                        sbi->ll_dt_exp->exp_handle.h_cookie);
2154                 EXIT;
2155                 return;
2156         }
2157         obd->obd_force = 1;
2158
2159         OBD_ALLOC_PTR(ioc_data);
2160         if (ioc_data) {
2161                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2162                               sizeof *ioc_data, ioc_data, NULL);
2163
2164                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2165                               sizeof *ioc_data, ioc_data, NULL);
2166
2167                 OBD_FREE_PTR(ioc_data);
2168         }
2169
2170         /* Really, we'd like to wait until there are no requests outstanding,
2171          * and then continue.  For now, we just periodically checking for vfs
2172          * to decrement mnt_cnt and hope to finish it within 10sec.
2173          */
2174         init_waitqueue_head(&waitq);
2175         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2176                                    cfs_time_seconds(1), NULL, NULL);
2177         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2178
2179         EXIT;
2180 }
2181
2182 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2183 {
2184         struct ll_sb_info *sbi = ll_s2sbi(sb);
2185         char *profilenm = get_profile_name(sb);
2186         int err;
2187         __u32 read_only;
2188
2189         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2190                 read_only = *flags & MS_RDONLY;
2191                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2192                                          sizeof(KEY_READ_ONLY),
2193                                          KEY_READ_ONLY, sizeof(read_only),
2194                                          &read_only, NULL);
2195                 if (err) {
2196                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2197                                       profilenm, read_only ?
2198                                       "read-only" : "read-write", err);
2199                         return err;
2200                 }
2201
2202                 if (read_only)
2203                         sb->s_flags |= MS_RDONLY;
2204                 else
2205                         sb->s_flags &= ~MS_RDONLY;
2206
2207                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2208                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2209                                       read_only ?  "read-only" : "read-write");
2210         }
2211         return 0;
2212 }
2213
2214 /**
2215  * Cleanup the open handle that is cached on MDT-side.
2216  *
2217  * For open case, the client side open handling thread may hit error
2218  * after the MDT grant the open. Under such case, the client should
2219  * send close RPC to the MDT as cleanup; otherwise, the open handle
2220  * on the MDT will be leaked there until the client umount or evicted.
2221  *
2222  * In further, if someone unlinked the file, because the open handle
2223  * holds the reference on such file/object, then it will block the
2224  * subsequent threads that want to locate such object via FID.
2225  *
2226  * \param[in] sb        super block for this file-system
2227  * \param[in] open_req  pointer to the original open request
2228  */
2229 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2230 {
2231         struct mdt_body                 *body;
2232         struct md_op_data               *op_data;
2233         struct ptlrpc_request           *close_req = NULL;
2234         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2235         ENTRY;
2236
2237         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2238         OBD_ALLOC_PTR(op_data);
2239         if (op_data == NULL) {
2240                 CWARN("%s: cannot allocate op_data to release open handle for "
2241                       DFID"\n",
2242                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2243
2244                 RETURN_EXIT;
2245         }
2246
2247         op_data->op_fid1 = body->mbo_fid1;
2248         op_data->op_handle = body->mbo_handle;
2249         op_data->op_mod_time = ktime_get_real_seconds();
2250         md_close(exp, op_data, NULL, &close_req);
2251         ptlrpc_req_finished(close_req);
2252         ll_finish_md_op_data(op_data);
2253
2254         EXIT;
2255 }
2256
2257 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2258                   struct super_block *sb, struct lookup_intent *it)
2259 {
2260         struct ll_sb_info *sbi = NULL;
2261         struct lustre_md md = { NULL };
2262         int rc;
2263         ENTRY;
2264
2265         LASSERT(*inode || sb);
2266         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2267         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2268                               sbi->ll_md_exp, &md);
2269         if (rc != 0)
2270                 GOTO(cleanup, rc);
2271
2272         if (*inode) {
2273                 rc = ll_update_inode(*inode, &md);
2274                 if (rc != 0)
2275                         GOTO(out, rc);
2276         } else {
2277                 LASSERT(sb != NULL);
2278
2279                 /*
2280                  * At this point server returns to client's same fid as client
2281                  * generated for creating. So using ->fid1 is okay here.
2282                  */
2283                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2284                         CERROR("%s: Fid is insane "DFID"\n",
2285                                 ll_get_fsname(sb, NULL, 0),
2286                                 PFID(&md.body->mbo_fid1));
2287                         GOTO(out, rc = -EINVAL);
2288                 }
2289
2290                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2291                                              sbi->ll_flags & LL_SBI_32BIT_API),
2292                                  &md);
2293                 if (IS_ERR(*inode)) {
2294 #ifdef CONFIG_FS_POSIX_ACL
2295                         if (md.posix_acl) {
2296                                 posix_acl_release(md.posix_acl);
2297                                 md.posix_acl = NULL;
2298                         }
2299 #endif
2300                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2301                         *inode = NULL;
2302                         CERROR("new_inode -fatal: rc %d\n", rc);
2303                         GOTO(out, rc);
2304                 }
2305         }
2306
2307         /* Handling piggyback layout lock.
2308          * Layout lock can be piggybacked by getattr and open request.
2309          * The lsm can be applied to inode only if it comes with a layout lock
2310          * otherwise correct layout may be overwritten, for example:
2311          * 1. proc1: mdt returns a lsm but not granting layout
2312          * 2. layout was changed by another client
2313          * 3. proc2: refresh layout and layout lock granted
2314          * 4. proc1: to apply a stale layout */
2315         if (it != NULL && it->it_lock_mode != 0) {
2316                 struct lustre_handle lockh;
2317                 struct ldlm_lock *lock;
2318
2319                 lockh.cookie = it->it_lock_handle;
2320                 lock = ldlm_handle2lock(&lockh);
2321                 LASSERT(lock != NULL);
2322                 if (ldlm_has_layout(lock)) {
2323                         struct cl_object_conf conf;
2324
2325                         memset(&conf, 0, sizeof(conf));
2326                         conf.coc_opc = OBJECT_CONF_SET;
2327                         conf.coc_inode = *inode;
2328                         conf.coc_lock = lock;
2329                         conf.u.coc_layout = md.layout;
2330                         (void)ll_layout_conf(*inode, &conf);
2331                 }
2332                 LDLM_LOCK_PUT(lock);
2333         }
2334
2335         GOTO(out, rc = 0);
2336
2337 out:
2338         md_free_lustre_md(sbi->ll_md_exp, &md);
2339
2340 cleanup:
2341         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2342                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2343
2344         return rc;
2345 }
2346
2347 int ll_obd_statfs(struct inode *inode, void __user *arg)
2348 {
2349         struct ll_sb_info *sbi = NULL;
2350         struct obd_export *exp;
2351         char *buf = NULL;
2352         struct obd_ioctl_data *data = NULL;
2353         __u32 type;
2354         int len = 0, rc;
2355
2356         if (!inode || !(sbi = ll_i2sbi(inode)))
2357                 GOTO(out_statfs, rc = -EINVAL);
2358
2359         rc = obd_ioctl_getdata(&buf, &len, arg);
2360         if (rc)
2361                 GOTO(out_statfs, rc);
2362
2363         data = (void*)buf;
2364         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2365             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2366                 GOTO(out_statfs, rc = -EINVAL);
2367
2368         if (data->ioc_inllen1 != sizeof(__u32) ||
2369             data->ioc_inllen2 != sizeof(__u32) ||
2370             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2371             data->ioc_plen2 != sizeof(struct obd_uuid))
2372                 GOTO(out_statfs, rc = -EINVAL);
2373
2374         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2375         if (type & LL_STATFS_LMV)
2376                 exp = sbi->ll_md_exp;
2377         else if (type & LL_STATFS_LOV)
2378                 exp = sbi->ll_dt_exp;
2379         else
2380                 GOTO(out_statfs, rc = -ENODEV);
2381
2382         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2383         if (rc)
2384                 GOTO(out_statfs, rc);
2385 out_statfs:
2386         if (buf)
2387                 obd_ioctl_freedata(buf, len);
2388         return rc;
2389 }
2390
2391 int ll_process_config(struct lustre_cfg *lcfg)
2392 {
2393         struct super_block *sb;
2394         unsigned long x;
2395         int rc = 0;
2396         char *ptr;
2397
2398         /* The instance name contains the sb: lustre-client-aacfe000 */
2399         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2400         if (!ptr || !*(++ptr))
2401                 return -EINVAL;
2402         if (sscanf(ptr, "%lx", &x) != 1)
2403                 return -EINVAL;
2404         sb = (struct super_block *)x;
2405         /* This better be a real Lustre superblock! */
2406         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2407
2408         /* Note we have not called client_common_fill_super yet, so
2409            proc fns must be able to handle that! */
2410         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2411                                       lcfg, sb);
2412         if (rc > 0)
2413                 rc = 0;
2414         return rc;
2415 }
2416
2417 /* this function prepares md_op_data hint for passing it down to MD stack. */
2418 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2419                                       struct inode *i1, struct inode *i2,
2420                                       const char *name, size_t namelen,
2421                                       __u32 mode, __u32 opc, void *data)
2422 {
2423         LASSERT(i1 != NULL);
2424
2425         if (name == NULL) {
2426                 /* Do not reuse namelen for something else. */
2427                 if (namelen != 0)
2428                         return ERR_PTR(-EINVAL);
2429         } else {
2430                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2431                         return ERR_PTR(-ENAMETOOLONG);
2432
2433                 if (!lu_name_is_valid_2(name, namelen))
2434                         return ERR_PTR(-EINVAL);
2435         }
2436
2437         if (op_data == NULL)
2438                 OBD_ALLOC_PTR(op_data);
2439
2440         if (op_data == NULL)
2441                 return ERR_PTR(-ENOMEM);
2442
2443         ll_i2gids(op_data->op_suppgids, i1, i2);
2444         op_data->op_fid1 = *ll_inode2fid(i1);
2445         op_data->op_default_stripe_offset = -1;
2446         if (S_ISDIR(i1->i_mode)) {
2447                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2448                 if (opc == LUSTRE_OPC_MKDIR)
2449                         op_data->op_default_stripe_offset =
2450                                    ll_i2info(i1)->lli_def_stripe_offset;
2451         }
2452
2453         if (i2) {
2454                 op_data->op_fid2 = *ll_inode2fid(i2);
2455                 if (S_ISDIR(i2->i_mode))
2456                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2457         } else {
2458                 fid_zero(&op_data->op_fid2);
2459         }
2460
2461         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2462                 op_data->op_cli_flags |= CLI_HASH64;
2463
2464         if (ll_need_32bit_api(ll_i2sbi(i1)))
2465                 op_data->op_cli_flags |= CLI_API32;
2466
2467         op_data->op_name = name;
2468         op_data->op_namelen = namelen;
2469         op_data->op_mode = mode;
2470         op_data->op_mod_time = cfs_time_current_sec();
2471         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2472         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2473         op_data->op_cap = cfs_curproc_cap_pack();
2474         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2475              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2476                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2477         } else {
2478                 op_data->op_mds = 0;
2479         }
2480         op_data->op_data = data;
2481
2482         return op_data;
2483 }
2484
2485 void ll_finish_md_op_data(struct md_op_data *op_data)
2486 {
2487         security_release_secctx(op_data->op_file_secctx,
2488                                 op_data->op_file_secctx_size);
2489         OBD_FREE_PTR(op_data);
2490 }
2491
2492 #ifdef HAVE_SUPEROPS_USE_DENTRY
2493 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2494 #else
2495 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2496 #endif
2497 {
2498         struct ll_sb_info *sbi;
2499
2500 #ifdef HAVE_SUPEROPS_USE_DENTRY
2501         LASSERT((seq != NULL) && (dentry != NULL));
2502         sbi = ll_s2sbi(dentry->d_sb);
2503 #else
2504         LASSERT((seq != NULL) && (vfs != NULL));
2505         sbi = ll_s2sbi(vfs->mnt_sb);
2506 #endif
2507
2508         if (sbi->ll_flags & LL_SBI_NOLCK)
2509                 seq_puts(seq, ",nolock");
2510
2511         if (sbi->ll_flags & LL_SBI_FLOCK)
2512                 seq_puts(seq, ",flock");
2513
2514         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2515                 seq_puts(seq, ",localflock");
2516
2517         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2518                 seq_puts(seq, ",user_xattr");
2519
2520         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2521                 seq_puts(seq, ",lazystatfs");
2522
2523         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2524                 seq_puts(seq, ",user_fid2path");
2525
2526         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2527                 seq_puts(seq, ",always_ping");
2528
2529         RETURN(0);
2530 }
2531
2532 /**
2533  * Get obd name by cmd, and copy out to user space
2534  */
2535 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2536 {
2537         struct ll_sb_info *sbi = ll_i2sbi(inode);
2538         struct obd_device *obd;
2539         ENTRY;
2540
2541         if (cmd == OBD_IOC_GETDTNAME)
2542                 obd = class_exp2obd(sbi->ll_dt_exp);
2543         else if (cmd == OBD_IOC_GETMDNAME)
2544                 obd = class_exp2obd(sbi->ll_md_exp);
2545         else
2546                 RETURN(-EINVAL);
2547
2548         if (!obd)
2549                 RETURN(-ENOENT);
2550
2551         if (copy_to_user((void __user *)arg, obd->obd_name,
2552                          strlen(obd->obd_name) + 1))
2553                 RETURN(-EFAULT);
2554
2555         RETURN(0);
2556 }
2557
2558 /**
2559  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2560  * fsname will be returned in this buffer; otherwise, a static buffer will be
2561  * used to store the fsname and returned to caller.
2562  */
2563 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2564 {
2565         static char fsname_static[MTI_NAME_MAXLEN];
2566         struct lustre_sb_info *lsi = s2lsi(sb);
2567         char *ptr;
2568         int len;
2569
2570         if (buf == NULL) {
2571                 /* this means the caller wants to use static buffer
2572                  * and it doesn't care about race. Usually this is
2573                  * in error reporting path */
2574                 buf = fsname_static;
2575                 buflen = sizeof(fsname_static);
2576         }
2577
2578         len = strlen(lsi->lsi_lmd->lmd_profile);
2579         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2580         if (ptr && (strcmp(ptr, "-client") == 0))
2581                 len -= 7;
2582
2583         if (unlikely(len >= buflen))
2584                 len = buflen - 1;
2585         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2586         buf[len] = '\0';
2587
2588         return buf;
2589 }
2590
2591 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2592 {
2593         char *path = NULL;
2594
2595         struct path p;
2596
2597         p.dentry = dentry;
2598         p.mnt = current->fs->root.mnt;
2599         path_get(&p);
2600         path = d_path(&p, buf, bufsize);
2601         path_put(&p);
2602         return path;
2603 }
2604
2605 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2606 {
2607         char *buf, *path = NULL;
2608         struct dentry *dentry = NULL;
2609         struct inode *inode = page->mapping->host;
2610
2611         /* this can be called inside spin lock so use GFP_ATOMIC. */
2612         buf = (char *)__get_free_page(GFP_ATOMIC);
2613         if (buf != NULL) {
2614                 dentry = d_find_alias(page->mapping->host);
2615                 if (dentry != NULL)
2616                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2617         }
2618
2619         CDEBUG(D_WARNING,
2620                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2621                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2622                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2623                PFID(ll_inode2fid(inode)),
2624                (path && !IS_ERR(path)) ? path : "", ioret);
2625
2626         if (dentry != NULL)
2627                 dput(dentry);
2628
2629         if (buf != NULL)
2630                 free_page((unsigned long)buf);
2631 }
2632
2633 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2634                         struct lov_user_md **kbuf)
2635 {
2636         struct lov_user_md      lum;
2637         ssize_t                 lum_size;
2638         ENTRY;
2639
2640         if (copy_from_user(&lum, md, sizeof(lum)))
2641                 RETURN(-EFAULT);
2642
2643         lum_size = ll_lov_user_md_size(&lum);
2644         if (lum_size < 0)
2645                 RETURN(lum_size);
2646
2647         OBD_ALLOC(*kbuf, lum_size);
2648         if (*kbuf == NULL)
2649                 RETURN(-ENOMEM);
2650
2651         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2652                 OBD_FREE(*kbuf, lum_size);
2653                 RETURN(-EFAULT);
2654         }
2655
2656         RETURN(lum_size);
2657 }
2658
2659 /*
2660  * Compute llite root squash state after a change of root squash
2661  * configuration setting or add/remove of a lnet nid
2662  */
2663 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2664 {
2665         struct root_squash_info *squash = &sbi->ll_squash;
2666         int i;
2667         bool matched;
2668         struct lnet_process_id id;
2669
2670         /* Update norootsquash flag */
2671         down_write(&squash->rsi_sem);
2672         if (list_empty(&squash->rsi_nosquash_nids))
2673                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2674         else {
2675                 /* Do not apply root squash as soon as one of our NIDs is
2676                  * in the nosquash_nids list */
2677                 matched = false;
2678                 i = 0;
2679                 while (LNetGetId(i++, &id) != -ENOENT) {
2680                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2681                                 continue;
2682                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2683                                 matched = true;
2684                                 break;
2685                         }
2686                 }
2687                 if (matched)
2688                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2689                 else
2690                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2691         }
2692         up_write(&squash->rsi_sem);
2693 }
2694
2695 /**
2696  * Parse linkea content to extract information about a given hardlink
2697  *
2698  * \param[in]   ldata      - Initialized linkea data
2699  * \param[in]   linkno     - Link identifier
2700  * \param[out]  parent_fid - The entry's parent FID
2701  * \param[out]  ln         - Entry name destination buffer
2702  *
2703  * \retval 0 on success
2704  * \retval Appropriate negative error code on failure
2705  */
2706 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2707                             struct lu_fid *parent_fid, struct lu_name *ln)
2708 {
2709         unsigned int    idx;
2710         int             rc;
2711         ENTRY;
2712
2713         rc = linkea_init_with_rec(ldata);
2714         if (rc < 0)
2715                 RETURN(rc);
2716
2717         if (linkno >= ldata->ld_leh->leh_reccount)
2718                 /* beyond last link */
2719                 RETURN(-ENODATA);
2720
2721         linkea_first_entry(ldata);
2722         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2723                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2724                                     parent_fid);
2725                 if (idx == linkno)
2726                         break;
2727
2728                 linkea_next_entry(ldata);
2729         }
2730
2731         if (idx < linkno)
2732                 RETURN(-ENODATA);
2733
2734         RETURN(0);
2735 }
2736
2737 /**
2738  * Get parent FID and name of an identified link. Operation is performed for
2739  * a given link number, letting the caller iterate over linkno to list one or
2740  * all links of an entry.
2741  *
2742  * \param[in]     file - File descriptor against which to perform the operation
2743  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2744  *                       on and the available size. It is eventually filled with
2745  *                       the requested information or left untouched on error
2746  *
2747  * \retval - 0 on success
2748  * \retval - Appropriate negative error code on failure
2749  */
2750 int ll_getparent(struct file *file, struct getparent __user *arg)
2751 {
2752         struct dentry           *dentry = file_dentry(file);
2753         struct inode            *inode = file_inode(file);
2754         struct linkea_data      *ldata;
2755         struct lu_buf            buf = LU_BUF_NULL;
2756         struct lu_name           ln;
2757         struct lu_fid            parent_fid;
2758         __u32                    linkno;
2759         __u32                    name_size;
2760         int                      rc;
2761
2762         ENTRY;
2763
2764         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2765             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2766                 RETURN(-EPERM);
2767
2768         if (get_user(name_size, &arg->gp_name_size))
2769                 RETURN(-EFAULT);
2770
2771         if (get_user(linkno, &arg->gp_linkno))
2772                 RETURN(-EFAULT);
2773
2774         if (name_size > PATH_MAX)
2775                 RETURN(-EINVAL);
2776
2777         OBD_ALLOC(ldata, sizeof(*ldata));
2778         if (ldata == NULL)
2779                 RETURN(-ENOMEM);
2780
2781         rc = linkea_data_new(ldata, &buf);
2782         if (rc < 0)
2783                 GOTO(ldata_free, rc);
2784
2785         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2786         if (rc < 0)
2787                 GOTO(lb_free, rc);
2788
2789         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2790         if (rc < 0)
2791                 GOTO(lb_free, rc);
2792
2793         if (ln.ln_namelen >= name_size)
2794                 GOTO(lb_free, rc = -EOVERFLOW);
2795
2796         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2797                 GOTO(lb_free, rc = -EFAULT);
2798
2799         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2800                 GOTO(lb_free, rc = -EFAULT);
2801
2802         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2803                 GOTO(lb_free, rc = -EFAULT);
2804
2805 lb_free:
2806         lu_buf_free(&buf);
2807 ldata_free:
2808         OBD_FREE(ldata, sizeof(*ldata));
2809
2810         RETURN(rc);
2811 }