lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #ifdef HAVE_UIDGID_HEADER
  50 # include <linux/uidgid.h>
  51 #endif
  52
  53 #include <lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  67 static DEFINE_SPINLOCK(ll_sb_lock);
  68
  69 #ifndef log2
  70 #define log2(n) ffz(~(n))
  71 #endif
  72
  73 static struct ll_sb_info *ll_init_sbi(void)
  74 {
  75         struct ll_sb_info *sbi = NULL;
  76         unsigned long pages;
  77         unsigned long lru_page_max;
  78         struct sysinfo si;
  79         class_uuid_t uuid;
  80         int i;
  81         ENTRY;
  82
  83         OBD_ALLOC_PTR(sbi);
  84         if (sbi == NULL)
  85                 RETURN(NULL);
  86
  87         spin_lock_init(&sbi->ll_lock);
  88         mutex_init(&sbi->ll_lco.lco_lock);
  89         spin_lock_init(&sbi->ll_pp_extent_lock);
  90         spin_lock_init(&sbi->ll_process_lock);
  91         sbi->ll_rw_stats_on = 0;
  92
  93         si_meminfo(&si);
  94         pages = si.totalram - si.totalhigh;
  95         lru_page_max = pages / 2;
  96
  97         /* initialize ll_cache data */
  98         sbi->ll_cache = cl_cache_init(lru_page_max);
  99         if (sbi->ll_cache == NULL) {
 100                 OBD_FREE(sbi, sizeof(*sbi));
 101                 RETURN(NULL);
 102         }
 103
 104         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 105                                            SBI_DEFAULT_READAHEAD_MAX);
 106         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 107         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 108                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 109         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 110         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 111
 112         ll_generate_random_uuid(uuid);
 113         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 114         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 115
 116         spin_lock(&ll_sb_lock);
 117         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 118         spin_unlock(&ll_sb_lock);
 119
 120         sbi->ll_flags |= LL_SBI_VERBOSE;
 121 #ifdef ENABLE_CHECKSUM
 122         sbi->ll_flags |= LL_SBI_CHECKSUM;
 123 #endif
 124
 125 #ifdef HAVE_LRU_RESIZE_SUPPORT
 126         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 127 #endif
 128
 129         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 130                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 131                                pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 133                                pp_w_hist.oh_lock);
 134         }
 135
 136         /* metadata statahead is enabled by default */
 137         sbi->ll_sa_max = LL_SA_RPC_DEF;
 138         atomic_set(&sbi->ll_sa_total, 0);
 139         atomic_set(&sbi->ll_sa_wrong, 0);
 140         atomic_set(&sbi->ll_sa_running, 0);
 141         atomic_set(&sbi->ll_agl_total, 0);
 142         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 spin_lock(&ll_sb_lock);
 160                 list_del(&sbi->ll_list);
 161                 spin_unlock(&ll_sb_lock);
 162                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 163                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 164                 if (sbi->ll_cache != NULL) {
 165                         cl_cache_decref(sbi->ll_cache);
 166                         sbi->ll_cache = NULL;
 167                 }
 168                 OBD_FREE(sbi, sizeof(*sbi));
 169         }
 170         EXIT;
 171 }
 172
 173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 174                                     struct vfsmount *mnt)
 175 {
 176         struct inode *root = NULL;
 177         struct ll_sb_info *sbi = ll_s2sbi(sb);
 178         struct obd_device *obd;
 179         struct obd_capa *oc = NULL;
 180         struct obd_statfs *osfs = NULL;
 181         struct ptlrpc_request *request = NULL;
 182         struct obd_connect_data *data = NULL;
 183         struct obd_uuid *uuid;
 184         struct md_op_data *op_data;
 185         struct lustre_md lmd;
 186         u64 valid;
 187         int size, err, checksum;
 188         ENTRY;
 189
 190         obd = class_name2obd(md);
 191         if (!obd) {
 192                 CERROR("MD %s: not setup or attached\n", md);
 193                 RETURN(-EINVAL);
 194         }
 195
 196         OBD_ALLOC_PTR(data);
 197         if (data == NULL)
 198                 RETURN(-ENOMEM);
 199
 200         OBD_ALLOC_PTR(osfs);
 201         if (osfs == NULL) {
 202                 OBD_FREE_PTR(data);
 203                 RETURN(-ENOMEM);
 204         }
 205
 206         /* indicate the features supported by this client */
 207         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 208                                   OBD_CONNECT_ATTRFID  |
 209                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 210                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 211                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 212                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 213                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 214                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 215                                   OBD_CONNECT_EINPROGRESS |
 216                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 217                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 218                                   OBD_CONNECT_MAX_EASIZE |
 219                                   OBD_CONNECT_FLOCK_DEAD |
 220                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 221                                   OBD_CONNECT_OPEN_BY_FID |
 222                                   OBD_CONNECT_DIR_STRIPE;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 230 #endif
 231
 232         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 233                 /* flag mdc connection as lightweight, only used for test
 234                  * purpose, use with care */
 235                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 236
 237         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 238         data->ocd_version = LUSTRE_VERSION_CODE;
 239
 240         if (sb->s_flags & MS_RDONLY)
 241                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 242         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 243                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 244
 245 #ifdef HAVE_MS_FLOCK_LOCK
 246         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 247         sb->s_flags |= MS_FLOCK_LOCK;
 248 #endif
 249 #ifdef MS_HAS_NEW_AOPS
 250         sb->s_flags |= MS_HAS_NEW_AOPS;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* real client */
 261         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 262         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 263                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 338
 339         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 340             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 341                 LCONSOLE_INFO("Disabling user_xattr feature because "
 342                               "it is not supported on the server\n");
 343                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags |= MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags |= LL_SBI_ACL;
 351         } else {
 352                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 353 #ifdef MS_POSIXACL
 354                 sb->s_flags &= ~MS_POSIXACL;
 355 #endif
 356                 sbi->ll_flags &= ~LL_SBI_ACL;
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 360                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 361                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client is set as remote by default.\n");
 363                 }
 364         } else {
 365                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 366                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 367                         LCONSOLE_INFO("client claims to be remote, but server "
 368                                       "rejected, forced to be local.\n");
 369                 }
 370         }
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 373                 LCONSOLE_INFO("client enabled MDS capability!\n");
 374                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 378                 LCONSOLE_INFO("client enabled OSS capability!\n");
 379                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 380         }
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 383                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 384
 385         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 386                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 387         else
 388                 sbi->ll_md_brw_pages = 1;
 389
 390         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 391                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 392
 393         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 394                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 395                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 396                                       "unknown maximum xattr size.\n", dt);
 397                 } else {
 398                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 399                         sbi->ll_xattr_cache_enabled = 1;
 400                 }
 401         }
 402
 403         obd = class_name2obd(dt);
 404         if (!obd) {
 405                 CERROR("DT %s: not setup or attached\n", dt);
 406                 GOTO(out_md_fid, err = -ENODEV);
 407         }
 408
 409         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 410                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 411                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 412                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 413                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 414                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 415                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 416                                   OBD_CONNECT_MAXBYTES |
 417                                   OBD_CONNECT_EINPROGRESS |
 418                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 419                                   OBD_CONNECT_LAYOUTLOCK |
 420                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 421
 422         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 423                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 424                  * disabled by default, because it can still be enabled on the
 425                  * fly via /proc. As a consequence, we still need to come to an
 426                  * agreement on the supported algorithms at connect time */
 427                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 428
 429                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 430                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 431                 else
 432                         data->ocd_cksum_types = cksum_types_supported_client();
 433         }
 434
 435 #ifdef HAVE_LRU_RESIZE_SUPPORT
 436         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 437 #endif
 438         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 439                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 440
 441         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 442                "ocd_grant: %d\n", data->ocd_connect_flags,
 443                data->ocd_version, data->ocd_grant);
 444
 445         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 446         obd->obd_upcall.onu_upcall = cl_ocd_update;
 447
 448         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 449
 450         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 451                           NULL);
 452         if (err == -EBUSY) {
 453                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 454                                    "recovery, of which this client is not a "
 455                                    "part.  Please wait for recovery to "
 456                                    "complete, abort, or time out.\n", dt);
 457                 GOTO(out_md, err);
 458         } else if (err) {
 459                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 460                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 461                 GOTO(out_md, err);
 462         }
 463
 464         sbi->ll_dt_exp->exp_connect_data = *data;
 465
 466         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 467                            LUSTRE_SEQ_METADATA);
 468         if (err) {
 469                 CERROR("%s: Can't init data layer FID infrastructure, "
 470                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 471                 GOTO(out_dt, err);
 472         }
 473
 474         mutex_lock(&sbi->ll_lco.lco_lock);
 475         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 476         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 477         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 478         mutex_unlock(&sbi->ll_lco.lco_lock);
 479
 480         fid_zero(&sbi->ll_root_fid);
 481         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 482         if (err) {
 483                 CERROR("cannot mds_connect: rc = %d\n", err);
 484                 GOTO(out_lock_cn_cb, err);
 485         }
 486         if (!fid_is_sane(&sbi->ll_root_fid)) {
 487                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 488                        sbi->ll_md_exp->exp_obd->obd_name,
 489                        PFID(&sbi->ll_root_fid));
 490                 GOTO(out_lock_cn_cb, err = -EINVAL);
 491         }
 492         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 493
 494         sb->s_op = &lustre_super_operations;
 495 #if THREAD_SIZE >= 8192 /*b=17630*/
 496         sb->s_export_op = &lustre_export_operations;
 497 #endif
 498
 499         /* make root inode
 500          * XXX: move this to after cbd setup? */
 501         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 502                 OBD_MD_FLMODEASIZE;
 503         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 504                 valid |= OBD_MD_FLRMTPERM;
 505         else if (sbi->ll_flags & LL_SBI_ACL)
 506                 valid |= OBD_MD_FLACL;
 507
 508         OBD_ALLOC_PTR(op_data);
 509         if (op_data == NULL)
 510                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 511
 512         op_data->op_fid1 = sbi->ll_root_fid;
 513         op_data->op_mode = 0;
 514         op_data->op_capa1 = oc;
 515         op_data->op_valid = valid;
 516
 517         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 518         if (oc)
 519                 capa_put(oc);
 520         OBD_FREE_PTR(op_data);
 521         if (err) {
 522                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 523                        sbi->ll_md_exp->exp_obd->obd_name, err);
 524                 GOTO(out_lock_cn_cb, err);
 525         }
 526
 527         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 528                                sbi->ll_md_exp, &lmd);
 529         if (err) {
 530                 CERROR("failed to understand root inode md: rc = %d\n", err);
 531                 ptlrpc_req_finished(request);
 532                 GOTO(out_lock_cn_cb, err);
 533         }
 534
 535         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 536         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 537                                             sbi->ll_flags & LL_SBI_32BIT_API),
 538                        &lmd);
 539         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 540         ptlrpc_req_finished(request);
 541
 542         if (IS_ERR(root)) {
 543                 if (lmd.lsm)
 544                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 545 #ifdef CONFIG_FS_POSIX_ACL
 546                 if (lmd.posix_acl) {
 547                         posix_acl_release(lmd.posix_acl);
 548                         lmd.posix_acl = NULL;
 549                 }
 550 #endif
 551                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 552                 root = NULL;
 553                 CERROR("lustre_lite: bad iget4 for root\n");
 554                 GOTO(out_root, err);
 555         }
 556
 557 #ifdef CONFIG_FS_POSIX_ACL
 558         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 559                 rct_init(&sbi->ll_rct);
 560                 et_init(&sbi->ll_et);
 561         }
 562 #endif
 563
 564         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 565         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 566                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 567                                  NULL);
 568         cl_sb_init(sb);
 569
 570         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 571                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 572                                  sbi->ll_cache, NULL);
 573
 574         sb->s_root = d_make_root(root);
 575         if (sb->s_root == NULL) {
 576                 CERROR("%s: can't make root dentry\n",
 577                         ll_get_fsname(sb, NULL, 0));
 578                 GOTO(out_root, err = -ENOMEM);
 579         }
 580 #ifdef HAVE_DCACHE_LOCK
 581         sb->s_root->d_op = &ll_d_ops;
 582 #endif
 583
 584         sbi->ll_sdev_orig = sb->s_dev;
 585
 586         /* We set sb->s_dev equal on all lustre clients in order to support
 587          * NFS export clustering.  NFSD requires that the FSID be the same
 588          * on all clients. */
 589         /* s_dev is also used in lt_compare() to compare two fs, but that is
 590          * only a node-local comparison. */
 591         uuid = obd_get_uuid(sbi->ll_md_exp);
 592         if (uuid != NULL)
 593                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 594
 595         if (data != NULL)
 596                 OBD_FREE_PTR(data);
 597         if (osfs != NULL)
 598                 OBD_FREE_PTR(osfs);
 599         if (proc_lustre_fs_root != NULL) {
 600                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 601                                                   dt, md);
 602                 if (err < 0) {
 603                         CERROR("%s: could not register mount in lprocfs: "
 604                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 605                         err = 0;
 606                 }
 607         }
 608
 609         RETURN(err);
 610 out_root:
 611         if (root)
 612                 iput(root);
 613 out_lock_cn_cb:
 614         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 615 out_dt:
 616         obd_disconnect(sbi->ll_dt_exp);
 617         sbi->ll_dt_exp = NULL;
 618 out_md_fid:
 619         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 620 out_md:
 621         obd_disconnect(sbi->ll_md_exp);
 622         sbi->ll_md_exp = NULL;
 623 out:
 624         if (data != NULL)
 625                 OBD_FREE_PTR(data);
 626         if (osfs != NULL)
 627                 OBD_FREE_PTR(osfs);
 628         return err;
 629 }
 630
 631 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 632 {
 633         int size, rc;
 634
 635         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 636         size = sizeof(int);
 637         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 638                           KEY_MAX_EASIZE, &size, lmmsize);
 639         if (rc)
 640                 CERROR("Get max mdsize error rc %d\n", rc);
 641
 642         RETURN(rc);
 643 }
 644
 645 /**
 646  * Get the value of the default_easize parameter.
 647  *
 648  * \see client_obd::cl_default_mds_easize
 649  *
 650  * \param[in] sbi       superblock info for this filesystem
 651  * \param[out] lmmsize  pointer to storage location for value
 652  *
 653  * \retval 0            on success
 654  * \retval negative     negated errno on failure
 655  */
 656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 657 {
 658         int size, rc;
 659
 660         size = sizeof(int);
 661         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 662                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 663         if (rc)
 664                 CERROR("Get default mdsize error rc %d\n", rc);
 665
 666         RETURN(rc);
 667 }
 668
 669 /**
 670  * Set the default_easize parameter to the given value.
 671  *
 672  * \see client_obd::cl_default_mds_easize
 673  *
 674  * \param[in] sbi       superblock info for this filesystem
 675  * \param[in] lmmsize   the size to set
 676  *
 677  * \retval 0            on success
 678  * \retval negative     negated errno on failure
 679  */
 680 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 681 {
 682         int rc;
 683
 684         if (lmmsize < sizeof(struct lov_mds_md) ||
 685             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 686                 return -EINVAL;
 687
 688         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 689                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 690                                 sizeof(int), &lmmsize, NULL);
 691
 692         RETURN(rc);
 693 }
 694
 695 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 696 {
 697         int size, rc;
 698
 699         size = sizeof(int);
 700         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 701                           KEY_MAX_COOKIESIZE, &size, lmmsize);
 702         if (rc)
 703                 CERROR("Get max cookiesize error rc %d\n", rc);
 704
 705         RETURN(rc);
 706 }
 707
 708 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 709 {
 710         int size, rc;
 711
 712         size = sizeof(int);
 713         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 714                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize);
 715         if (rc)
 716                 CERROR("Get default cookiesize error rc %d\n", rc);
 717
 718         RETURN(rc);
 719 }
 720
 721 static void ll_dump_inode(struct inode *inode)
 722 {
 723         struct ll_d_hlist_node *tmp;
 724         int dentry_count = 0;
 725
 726         LASSERT(inode != NULL);
 727
 728         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 729                 dentry_count++;
 730
 731         CERROR("%s: inode %p dump: dev=%s fid="DFID
 732                " mode=%o count=%u, %d dentries\n",
 733                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 734                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 735                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 736 }
 737
 738 void lustre_dump_dentry(struct dentry *dentry, int recur)
 739 {
 740         struct list_head *tmp;
 741         int subdirs = 0;
 742
 743         LASSERT(dentry != NULL);
 744
 745         list_for_each(tmp, &dentry->d_subdirs)
 746                 subdirs++;
 747
 748         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 749                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 750                dentry->d_name.len, dentry->d_name.name,
 751                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 752                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 753                dentry->d_flags, dentry->d_fsdata, subdirs);
 754         if (dentry->d_inode != NULL)
 755                 ll_dump_inode(dentry->d_inode);
 756
 757         if (recur == 0)
 758                 return;
 759
 760         list_for_each(tmp, &dentry->d_subdirs) {
 761                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 762                 lustre_dump_dentry(d, recur - 1);
 763         }
 764 }
 765
 766 static void client_common_put_super(struct super_block *sb)
 767 {
 768         struct ll_sb_info *sbi = ll_s2sbi(sb);
 769         ENTRY;
 770
 771 #ifdef CONFIG_FS_POSIX_ACL
 772         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 773                 et_fini(&sbi->ll_et);
 774                 rct_fini(&sbi->ll_rct);
 775         }
 776 #endif
 777
 778         cl_sb_fini(sb);
 779
 780         list_del(&sbi->ll_conn_chain);
 781
 782         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 783         obd_disconnect(sbi->ll_dt_exp);
 784         sbi->ll_dt_exp = NULL;
 785
 786         lprocfs_unregister_mountpoint(sbi);
 787
 788         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 789         obd_disconnect(sbi->ll_md_exp);
 790         sbi->ll_md_exp = NULL;
 791
 792         EXIT;
 793 }
 794
 795 void ll_kill_super(struct super_block *sb)
 796 {
 797         struct ll_sb_info *sbi;
 798         ENTRY;
 799
 800         /* not init sb ?*/
 801         if (!(sb->s_flags & MS_ACTIVE))
 802                 return;
 803
 804         sbi = ll_s2sbi(sb);
 805         /* we need restore s_dev from changed for clustred NFS before put_super
 806          * because new kernels have cached s_dev and change sb->s_dev in
 807          * put_super not affected real removing devices */
 808         if (sbi) {
 809                 sb->s_dev = sbi->ll_sdev_orig;
 810                 sbi->ll_umounting = 1;
 811
 812                 /* wait running statahead threads to quit */
 813                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 814                         set_current_state(TASK_UNINTERRUPTIBLE);
 815                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 816                 }
 817         }
 818
 819         EXIT;
 820 }
 821
 822 static inline int ll_set_opt(const char *opt, char *data, int fl)
 823 {
 824         if (strncmp(opt, data, strlen(opt)) != 0)
 825                 return(0);
 826         else
 827                 return(fl);
 828 }
 829
 830 /* non-client-specific mount options are parsed in lmd_parse */
 831 static int ll_options(char *options, int *flags)
 832 {
 833         int tmp;
 834         char *s1 = options, *s2;
 835         ENTRY;
 836
 837         if (!options)
 838                 RETURN(0);
 839
 840         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 841
 842         while (*s1) {
 843                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 844                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 845                 if (tmp) {
 846                         *flags |= tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 855                 if (tmp) {
 856                         *flags |= tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 860                 if (tmp) {
 861                         *flags &= ~tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 865                 if (tmp) {
 866                         *flags |= tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 870                 if (tmp) {
 871                         *flags &= ~tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 875                 if (tmp) {
 876                         *flags |= tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889
 890                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 891                 if (tmp) {
 892                         *flags |= tmp;
 893                         goto next;
 894                 }
 895                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 896                 if (tmp) {
 897                         *flags &= ~tmp;
 898                         goto next;
 899                 }
 900                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 901                 if (tmp) {
 902                         *flags |= tmp;
 903                         goto next;
 904                 }
 905                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 906                 if (tmp) {
 907                         *flags &= ~tmp;
 908                         goto next;
 909                 }
 910                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 911                 if (tmp) {
 912                         *flags |= tmp;
 913                         goto next;
 914                 }
 915                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 916                 if (tmp) {
 917                         *flags &= ~tmp;
 918                         goto next;
 919                 }
 920                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 921                 if (tmp) {
 922                         *flags |= tmp;
 923                         goto next;
 924                 }
 925                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 926                 if (tmp) {
 927                         *flags |= tmp;
 928                         goto next;
 929                 }
 930                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 931                 if (tmp) {
 932                         *flags &= ~tmp;
 933                         goto next;
 934                 }
 935                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 936                                    s1);
 937                 RETURN(-EINVAL);
 938
 939 next:
 940                 /* Find next opt */
 941                 s2 = strchr(s1, ',');
 942                 if (s2 == NULL)
 943                         break;
 944                 s1 = s2 + 1;
 945         }
 946         RETURN(0);
 947 }
 948
 949 void ll_lli_init(struct ll_inode_info *lli)
 950 {
 951         lli->lli_inode_magic = LLI_INODE_MAGIC;
 952         lli->lli_flags = 0;
 953         spin_lock_init(&lli->lli_lock);
 954         lli->lli_posix_acl = NULL;
 955         lli->lli_remote_perms = NULL;
 956         mutex_init(&lli->lli_rmtperm_mutex);
 957         /* Do not set lli_fid, it has been initialized already. */
 958         fid_zero(&lli->lli_pfid);
 959         INIT_LIST_HEAD(&lli->lli_oss_capas);
 960         atomic_set(&lli->lli_open_count, 0);
 961         lli->lli_mds_capa = NULL;
 962         lli->lli_rmtperm_time = 0;
 963         lli->lli_mds_read_och = NULL;
 964         lli->lli_mds_write_och = NULL;
 965         lli->lli_mds_exec_och = NULL;
 966         lli->lli_open_fd_read_count = 0;
 967         lli->lli_open_fd_write_count = 0;
 968         lli->lli_open_fd_exec_count = 0;
 969         mutex_init(&lli->lli_och_mutex);
 970         spin_lock_init(&lli->lli_agl_lock);
 971         spin_lock_init(&lli->lli_layout_lock);
 972         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 973         lli->lli_clob = NULL;
 974
 975         init_rwsem(&lli->lli_xattrs_list_rwsem);
 976         mutex_init(&lli->lli_xattrs_enq_lock);
 977
 978         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 979         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 980                 mutex_init(&lli->lli_readdir_mutex);
 981                 lli->lli_opendir_key = NULL;
 982                 lli->lli_sai = NULL;
 983                 spin_lock_init(&lli->lli_sa_lock);
 984                 lli->lli_opendir_pid = 0;
 985                 lli->lli_sa_enabled = 0;
 986                 lli->lli_def_stripe_offset = -1;
 987         } else {
 988                 mutex_init(&lli->lli_size_mutex);
 989                 lli->lli_symlink_name = NULL;
 990                 init_rwsem(&lli->lli_trunc_sem);
 991                 range_lock_tree_init(&lli->lli_write_tree);
 992                 init_rwsem(&lli->lli_glimpse_sem);
 993                 lli->lli_glimpse_time = 0;
 994                 INIT_LIST_HEAD(&lli->lli_agl_list);
 995                 lli->lli_agl_index = 0;
 996                 lli->lli_async_rc = 0;
 997         }
 998         mutex_init(&lli->lli_layout_mutex);
 999 }
1000
1001 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1002 {
1003         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1004
1005         bdi->name = "lustre";
1006         return bdi_register(bdi, NULL, "lustre-%d",
1007                             atomic_inc_return(&ll_bdi_num));
1008 }
1009
1010 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1011 {
1012         struct lustre_profile *lprof = NULL;
1013         struct lustre_sb_info *lsi = s2lsi(sb);
1014         struct ll_sb_info *sbi;
1015         char  *dt = NULL, *md = NULL;
1016         char  *profilenm = get_profile_name(sb);
1017         struct config_llog_instance *cfg;
1018         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1019         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1020         int    err;
1021         ENTRY;
1022
1023         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1024
1025         OBD_ALLOC_PTR(cfg);
1026         if (cfg == NULL)
1027                 RETURN(-ENOMEM);
1028
1029         try_module_get(THIS_MODULE);
1030
1031         /* client additional sb info */
1032         lsi->lsi_llsbi = sbi = ll_init_sbi();
1033         if (!sbi) {
1034                 module_put(THIS_MODULE);
1035                 OBD_FREE_PTR(cfg);
1036                 RETURN(-ENOMEM);
1037         }
1038
1039         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1040         if (err)
1041                 GOTO(out_free, err);
1042
1043         err = bdi_init(&lsi->lsi_bdi);
1044         if (err)
1045                 GOTO(out_free, err);
1046         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1047         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1048         err = ll_bdi_register(&lsi->lsi_bdi);
1049         if (err)
1050                 GOTO(out_free, err);
1051
1052         sb->s_bdi = &lsi->lsi_bdi;
1053 #ifndef HAVE_DCACHE_LOCK
1054         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1055         sb->s_d_op = &ll_d_ops;
1056 #endif
1057
1058         /* Generate a string unique to this super, in case some joker tries
1059            to mount the same fs at two mount points.
1060            Use the address of the super itself.*/
1061         cfg->cfg_instance = sb;
1062         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1063         cfg->cfg_callback = class_config_llog_handler;
1064         /* set up client obds */
1065         err = lustre_process_log(sb, profilenm, cfg);
1066         if (err < 0)
1067                 GOTO(out_free, err);
1068
1069         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1070         lprof = class_get_profile(profilenm);
1071         if (lprof == NULL) {
1072                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1073                                    " read from the MGS.  Does that filesystem "
1074                                    "exist?\n", profilenm);
1075                 GOTO(out_free, err = -EINVAL);
1076         }
1077         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1078                lprof->lp_md, lprof->lp_dt);
1079
1080         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1081         if (!dt)
1082                 GOTO(out_free, err = -ENOMEM);
1083         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1084
1085         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1086         if (!md)
1087                 GOTO(out_free, err = -ENOMEM);
1088         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1089
1090         /* connections, registrations, sb setup */
1091         err = client_common_fill_super(sb, md, dt, mnt);
1092         if (err < 0)
1093                 GOTO(out_free, err);
1094
1095         sbi->ll_client_common_fill_super_succeeded = 1;
1096
1097 out_free:
1098         if (md)
1099                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1100         if (dt)
1101                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1102         if (err)
1103                 ll_put_super(sb);
1104         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1106
1107         OBD_FREE_PTR(cfg);
1108         RETURN(err);
1109 } /* ll_fill_super */
1110
1111 void ll_put_super(struct super_block *sb)
1112 {
1113         struct config_llog_instance cfg, params_cfg;
1114         struct obd_device *obd;
1115         struct lustre_sb_info *lsi = s2lsi(sb);
1116         struct ll_sb_info *sbi = ll_s2sbi(sb);
1117         char *profilenm = get_profile_name(sb);
1118         long ccc_count;
1119         int next, force = 1, rc = 0;
1120         ENTRY;
1121
1122         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1123
1124         ll_print_capa_stat(sbi);
1125
1126         cfg.cfg_instance = sb;
1127         lustre_end_log(sb, profilenm, &cfg);
1128
1129         params_cfg.cfg_instance = sb;
1130         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1131
1132         if (sbi->ll_md_exp) {
1133                 obd = class_exp2obd(sbi->ll_md_exp);
1134                 if (obd)
1135                         force = obd->obd_force;
1136         }
1137
1138         /* Wait for unstable pages to be committed to stable storage */
1139         if (force == 0) {
1140                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1141                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1142                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1143                         &lwi);
1144         }
1145
1146         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1147         if (force == 0 && rc != -EINTR)
1148                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1149
1150
1151         /* We need to set force before the lov_disconnect in
1152            lustre_common_put_super, since l_d cleans up osc's as well. */
1153         if (force) {
1154                 next = 0;
1155                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1156                                                      &next)) != NULL) {
1157                         obd->obd_force = force;
1158                 }
1159         }
1160
1161         if (sbi->ll_client_common_fill_super_succeeded) {
1162                 /* Only if client_common_fill_super succeeded */
1163                 client_common_put_super(sb);
1164         }
1165
1166         next = 0;
1167         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1168                 class_manual_cleanup(obd);
1169         }
1170
1171         if (sbi->ll_flags & LL_SBI_VERBOSE)
1172                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1173
1174         if (profilenm)
1175                 class_del_profile(profilenm);
1176
1177         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1178                 bdi_destroy(&lsi->lsi_bdi);
1179                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1180         }
1181
1182         ll_free_sbi(sb);
1183         lsi->lsi_llsbi = NULL;
1184
1185         lustre_common_put_super(sb);
1186
1187         cl_env_cache_purge(~0);
1188
1189         module_put(THIS_MODULE);
1190
1191         EXIT;
1192 } /* client_put_super */
1193
1194 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1195 {
1196         struct inode *inode = NULL;
1197
1198         /* NOTE: we depend on atomic igrab() -bzzz */
1199         lock_res_and_lock(lock);
1200         if (lock->l_resource->lr_lvb_inode) {
1201                 struct ll_inode_info * lli;
1202                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1203                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1204                         inode = igrab(lock->l_resource->lr_lvb_inode);
1205                 } else {
1206                         inode = lock->l_resource->lr_lvb_inode;
1207                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1208                                          D_WARNING, lock, "lr_lvb_inode %p is "
1209                                          "bogus: magic %08x",
1210                                          lock->l_resource->lr_lvb_inode,
1211                                          lli->lli_inode_magic);
1212                         inode = NULL;
1213                 }
1214         }
1215         unlock_res_and_lock(lock);
1216         return inode;
1217 }
1218
1219 static void ll_dir_clear_lsm_md(struct inode *inode)
1220 {
1221         struct ll_inode_info *lli = ll_i2info(inode);
1222
1223         LASSERT(S_ISDIR(inode->i_mode));
1224
1225         if (lli->lli_lsm_md != NULL) {
1226                 lmv_free_memmd(lli->lli_lsm_md);
1227                 lli->lli_lsm_md = NULL;
1228         }
1229 }
1230
1231 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1232                                       const struct lu_fid *fid,
1233                                       struct lustre_md *md)
1234 {
1235         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1236         struct mdt_body         *body = md->body;
1237         struct inode            *inode;
1238         ino_t                   ino;
1239         ENTRY;
1240
1241         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1242         inode = iget_locked(sb, ino);
1243         if (inode == NULL) {
1244                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1245                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1246                 RETURN(ERR_PTR(-ENOENT));
1247         }
1248
1249         if (inode->i_state & I_NEW) {
1250                 struct ll_inode_info *lli = ll_i2info(inode);
1251                 struct lmv_stripe_md *lsm = md->lmv;
1252
1253                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1254                                 (body->mbo_mode & S_IFMT);
1255                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1256                          PFID(fid));
1257
1258                 LTIME_S(inode->i_mtime) = 0;
1259                 LTIME_S(inode->i_atime) = 0;
1260                 LTIME_S(inode->i_ctime) = 0;
1261                 inode->i_rdev = 0;
1262
1263                 /* initializing backing dev info. */
1264                 inode->i_mapping->backing_dev_info =
1265                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1266                 inode->i_op = &ll_dir_inode_operations;
1267                 inode->i_fop = &ll_dir_operations;
1268                 lli->lli_fid = *fid;
1269                 ll_lli_init(lli);
1270
1271                 LASSERT(lsm != NULL);
1272                 /* master object FID */
1273                 lli->lli_pfid = body->mbo_fid1;
1274                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1275                        lli, PFID(fid), PFID(&lli->lli_pfid));
1276                 unlock_new_inode(inode);
1277         }
1278
1279         RETURN(inode);
1280 }
1281
1282 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1283 {
1284         struct lu_fid *fid;
1285         struct lmv_stripe_md *lsm = md->lmv;
1286         int i;
1287
1288         LASSERT(lsm != NULL);
1289         /* XXX sigh, this lsm_root initialization should be in
1290          * LMV layer, but it needs ll_iget right now, so we
1291          * put this here right now. */
1292         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1293                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1294                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1295                 /* Unfortunately ll_iget will call ll_update_inode,
1296                  * where the initialization of slave inode is slightly
1297                  * different, so it reset lsm_md to NULL to avoid
1298                  * initializing lsm for slave inode. */
1299                 /* For migrating inode, master stripe and master object will
1300                  * be same, so we only need assign this inode */
1301                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1302                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1303                 else
1304                         lsm->lsm_md_oinfo[i].lmo_root =
1305                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1306
1307                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1308                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1309
1310                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1311                         return rc;
1312                 }
1313         }
1314
1315         return 0;
1316 }
1317
1318 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1319                                 const struct lmv_stripe_md *lsm_md2)
1320 {
1321         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1322                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1323                lsm_md1->lsm_md_master_mdt_index ==
1324                                         lsm_md2->lsm_md_master_mdt_index &&
1325                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1326                lsm_md1->lsm_md_layout_version ==
1327                                         lsm_md2->lsm_md_layout_version &&
1328                strcmp(lsm_md1->lsm_md_pool_name,
1329                       lsm_md2->lsm_md_pool_name) == 0;
1330 }
1331
1332 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1333 {
1334         struct ll_inode_info *lli = ll_i2info(inode);
1335         struct lmv_stripe_md *lsm = md->lmv;
1336         int     rc;
1337         ENTRY;
1338
1339         LASSERT(S_ISDIR(inode->i_mode));
1340         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1341                PFID(ll_inode2fid(inode)));
1342
1343         /* no striped information from request. */
1344         if (lsm == NULL) {
1345                 if (lli->lli_lsm_md == NULL) {
1346                         RETURN(0);
1347                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1348                                                 LMV_HASH_FLAG_MIGRATION) {
1349                         /* migration is done, the temporay MIGRATE layout has
1350                          * been removed */
1351                         CDEBUG(D_INODE, DFID" finish migration.\n",
1352                                PFID(ll_inode2fid(inode)));
1353                         lmv_free_memmd(lli->lli_lsm_md);
1354                         lli->lli_lsm_md = NULL;
1355                         RETURN(0);
1356                 } else {
1357                         /* The lustre_md from req does not include stripeEA,
1358                          * see ll_md_setattr */
1359                         RETURN(0);
1360                 }
1361         }
1362
1363         /* set the directory layout */
1364         if (lli->lli_lsm_md == NULL) {
1365
1366                 rc = ll_init_lsm_md(inode, md);
1367                 if (rc != 0)
1368                         RETURN(rc);
1369
1370                 lli->lli_lsm_md = lsm;
1371                 /* set lsm_md to NULL, so the following free lustre_md
1372                  * will not free this lsm */
1373                 md->lmv = NULL;
1374                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1375                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1376                 RETURN(0);
1377         }
1378
1379         /* Compare the old and new stripe information */
1380         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1381                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1382                 int                     idx;
1383
1384                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1385                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1386                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1387                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1388                        inode, lsm, old_lsm,
1389                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1390                        lsm->lsm_md_stripe_count,
1391                        old_lsm->lsm_md_stripe_count,
1392                        lsm->lsm_md_master_mdt_index,
1393                        old_lsm->lsm_md_master_mdt_index,
1394                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1395                        lsm->lsm_md_layout_version,
1396                        old_lsm->lsm_md_layout_version,
1397                        lsm->lsm_md_pool_name,
1398                        old_lsm->lsm_md_pool_name);
1399
1400                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1401                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1402                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1403                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1404                 }
1405
1406                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1407                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1408                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1409                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1410                 }
1411
1412                 RETURN(-EIO);
1413         }
1414
1415         RETURN(0);
1416 }
1417
1418 void ll_clear_inode(struct inode *inode)
1419 {
1420         struct ll_inode_info *lli = ll_i2info(inode);
1421         struct ll_sb_info *sbi = ll_i2sbi(inode);
1422         ENTRY;
1423
1424         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1425                PFID(ll_inode2fid(inode)), inode);
1426
1427         if (S_ISDIR(inode->i_mode)) {
1428                 /* these should have been cleared in ll_file_release */
1429                 LASSERT(lli->lli_opendir_key == NULL);
1430                 LASSERT(lli->lli_sai == NULL);
1431                 LASSERT(lli->lli_opendir_pid == 0);
1432         }
1433
1434         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1435
1436         LASSERT(!lli->lli_open_fd_write_count);
1437         LASSERT(!lli->lli_open_fd_read_count);
1438         LASSERT(!lli->lli_open_fd_exec_count);
1439
1440         if (lli->lli_mds_write_och)
1441                 ll_md_real_close(inode, FMODE_WRITE);
1442         if (lli->lli_mds_exec_och)
1443                 ll_md_real_close(inode, FMODE_EXEC);
1444         if (lli->lli_mds_read_och)
1445                 ll_md_real_close(inode, FMODE_READ);
1446
1447         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1448                 OBD_FREE(lli->lli_symlink_name,
1449                          strlen(lli->lli_symlink_name) + 1);
1450                 lli->lli_symlink_name = NULL;
1451         }
1452
1453         ll_xattr_cache_destroy(inode);
1454
1455         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1456                 LASSERT(lli->lli_posix_acl == NULL);
1457                 if (lli->lli_remote_perms) {
1458                         free_rmtperm_hash(lli->lli_remote_perms);
1459                         lli->lli_remote_perms = NULL;
1460                 }
1461         }
1462 #ifdef CONFIG_FS_POSIX_ACL
1463         else if (lli->lli_posix_acl) {
1464                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1465                 LASSERT(lli->lli_remote_perms == NULL);
1466                 posix_acl_release(lli->lli_posix_acl);
1467                 lli->lli_posix_acl = NULL;
1468         }
1469 #endif
1470         lli->lli_inode_magic = LLI_INODE_DEAD;
1471
1472         ll_clear_inode_capas(inode);
1473         if (S_ISDIR(inode->i_mode))
1474                 ll_dir_clear_lsm_md(inode);
1475         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1476                 LASSERT(list_empty(&lli->lli_agl_list));
1477
1478         /*
1479          * XXX This has to be done before lsm is freed below, because
1480          * cl_object still uses inode lsm.
1481          */
1482         cl_inode_fini(inode);
1483
1484         EXIT;
1485 }
1486
1487 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1488 {
1489         struct lustre_md md;
1490         struct inode *inode = dentry->d_inode;
1491         struct ll_sb_info *sbi = ll_i2sbi(inode);
1492         struct ptlrpc_request *request = NULL;
1493         int rc, ia_valid;
1494         ENTRY;
1495
1496         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1497                                      LUSTRE_OPC_ANY, NULL);
1498         if (IS_ERR(op_data))
1499                 RETURN(PTR_ERR(op_data));
1500
1501         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1502         if (rc) {
1503                 ptlrpc_req_finished(request);
1504                 if (rc == -ENOENT) {
1505                         clear_nlink(inode);
1506                         /* Unlinked special device node? Or just a race?
1507                          * Pretend we done everything. */
1508                         if (!S_ISREG(inode->i_mode) &&
1509                             !S_ISDIR(inode->i_mode)) {
1510                                 ia_valid = op_data->op_attr.ia_valid;
1511                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1512                                 rc = simple_setattr(dentry, &op_data->op_attr);
1513                                 op_data->op_attr.ia_valid = ia_valid;
1514                         }
1515                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1516                         CERROR("md_setattr fails: rc = %d\n", rc);
1517                 }
1518                 RETURN(rc);
1519         }
1520
1521         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1522                               sbi->ll_md_exp, &md);
1523         if (rc) {
1524                 ptlrpc_req_finished(request);
1525                 RETURN(rc);
1526         }
1527
1528         ia_valid = op_data->op_attr.ia_valid;
1529         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1530          * cache is not cleared yet. */
1531         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1532         rc = simple_setattr(dentry, &op_data->op_attr);
1533         op_data->op_attr.ia_valid = ia_valid;
1534
1535         rc = ll_update_inode(inode, &md);
1536         ptlrpc_req_finished(request);
1537
1538         RETURN(rc);
1539 }
1540
1541 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1542 {
1543         struct obd_capa *capa;
1544         int rc;
1545
1546         if (attr->ia_valid & ATTR_SIZE)
1547                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1548         else
1549                 capa = ll_mdscapa_get(inode);
1550
1551         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1552
1553         if (attr->ia_valid & ATTR_SIZE)
1554                 ll_truncate_free_capa(capa);
1555         else
1556                 capa_put(capa);
1557
1558         return rc;
1559 }
1560
1561 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1562  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1563  * keep these values until such a time that objects are allocated for it.
1564  * We do the MDS operations first, as it is checking permissions for us.
1565  * We don't to the MDS RPC if there is nothing that we want to store there,
1566  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1567  * going to do an RPC anyways.
1568  *
1569  * If we are doing a truncate, we will send the mtime and ctime updates
1570  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1571  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1572  * at the same time.
1573  *
1574  * In case of HSMimport, we only set attr on MDS.
1575  */
1576 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1577 {
1578         struct inode *inode = dentry->d_inode;
1579         struct ll_inode_info *lli = ll_i2info(inode);
1580         struct md_op_data *op_data = NULL;
1581         bool file_is_released = false;
1582         int rc = 0;
1583         ENTRY;
1584
1585         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1586                "valid %x, hsm_import %d\n",
1587                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1588                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1589                hsm_import);
1590
1591         if (attr->ia_valid & ATTR_SIZE) {
1592                 /* Check new size against VFS/VM file size limit and rlimit */
1593                 rc = inode_newsize_ok(inode, attr->ia_size);
1594                 if (rc)
1595                         RETURN(rc);
1596
1597                 /* The maximum Lustre file size is variable, based on the
1598                  * OST maximum object size and number of stripes.  This
1599                  * needs another check in addition to the VFS check above. */
1600                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1601                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1602                                PFID(&lli->lli_fid), attr->ia_size,
1603                                ll_file_maxbytes(inode));
1604                         RETURN(-EFBIG);
1605                 }
1606
1607                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1608         }
1609
1610         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1611         if (attr->ia_valid & TIMES_SET_FLAGS) {
1612                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1613                     !cfs_capable(CFS_CAP_FOWNER))
1614                         RETURN(-EPERM);
1615         }
1616
1617         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1618         if (attr->ia_valid & ATTR_CTIME) {
1619                 attr->ia_ctime = CFS_CURRENT_TIME;
1620                 attr->ia_valid |= ATTR_CTIME_SET;
1621         }
1622         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1623             (attr->ia_valid & ATTR_ATIME)) {
1624                 attr->ia_atime = CFS_CURRENT_TIME;
1625                 attr->ia_valid |= ATTR_ATIME_SET;
1626         }
1627         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1628             (attr->ia_valid & ATTR_MTIME)) {
1629                 attr->ia_mtime = CFS_CURRENT_TIME;
1630                 attr->ia_valid |= ATTR_MTIME_SET;
1631         }
1632
1633         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1634                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1635                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1636                        cfs_time_current_sec());
1637
1638         /* We always do an MDS RPC, even if we're only changing the size;
1639          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1640
1641         OBD_ALLOC_PTR(op_data);
1642         if (op_data == NULL)
1643                 RETURN(-ENOMEM);
1644
1645         if (!S_ISDIR(inode->i_mode)) {
1646                 if (attr->ia_valid & ATTR_SIZE)
1647                         inode_dio_write_done(inode);
1648                 mutex_unlock(&inode->i_mutex);
1649         }
1650
1651         /* truncate on a released file must failed with -ENODATA,
1652          * so size must not be set on MDS for released file
1653          * but other attributes must be set
1654          */
1655         if (S_ISREG(inode->i_mode)) {
1656                 struct cl_layout cl = {
1657                         .cl_is_released = false,
1658                 };
1659                 struct lu_env *env;
1660                 int refcheck;
1661                 __u32 gen;
1662
1663                 rc = ll_layout_refresh(inode, &gen);
1664                 if (rc < 0)
1665                         GOTO(out, rc);
1666
1667                 /* XXX: the only place we need to know the layout type,
1668                  * this will be removed by a later patch. -Jinshan */
1669                 env = cl_env_get(&refcheck);
1670                 if (IS_ERR(env))
1671                         GOTO(out, rc = PTR_ERR(env));
1672
1673                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1674                 cl_env_put(env, &refcheck);
1675                 if (rc < 0)
1676                         GOTO(out, rc);
1677
1678                 file_is_released = cl.cl_is_released;
1679
1680                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1681                         if (file_is_released) {
1682                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1683                                 if (rc < 0)
1684                                         GOTO(out, rc);
1685
1686                                 file_is_released = false;
1687                                 ll_layout_refresh(inode, &gen);
1688                         }
1689
1690                         /* If we are changing file size, file content is
1691                          * modified, flag it. */
1692                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1693                         spin_lock(&lli->lli_lock);
1694                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1695                         spin_unlock(&lli->lli_lock);
1696                         op_data->op_bias |= MDS_DATA_MODIFIED;
1697                 }
1698         }
1699
1700         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1701
1702         rc = ll_md_setattr(dentry, op_data);
1703         if (rc)
1704                 GOTO(out, rc);
1705
1706         /* RPC to MDT is sent, cancel data modification flag */
1707         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1708                 spin_lock(&lli->lli_lock);
1709                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1710                 spin_unlock(&lli->lli_lock);
1711         }
1712
1713         if (!S_ISREG(inode->i_mode) || file_is_released)
1714                 GOTO(out, rc = 0);
1715
1716         if (attr->ia_valid & (ATTR_SIZE |
1717                               ATTR_ATIME | ATTR_ATIME_SET |
1718                               ATTR_MTIME | ATTR_MTIME_SET)) {
1719                 /* For truncate and utimes sending attributes to OSTs, setting
1720                  * mtime/atime to the past will be performed under PW [0:EOF]
1721                  * extent lock (new_size:EOF for truncate).  It may seem
1722                  * excessive to send mtime/atime updates to OSTs when not
1723                  * setting times to past, but it is necessary due to possible
1724                  * time de-synchronization between MDT inode and OST objects */
1725                 if (attr->ia_valid & ATTR_SIZE)
1726                         down_write(&lli->lli_trunc_sem);
1727                 rc = ll_setattr_ost(inode, attr);
1728                 if (attr->ia_valid & ATTR_SIZE)
1729                         up_write(&lli->lli_trunc_sem);
1730         }
1731         EXIT;
1732 out:
1733         if (op_data != NULL)
1734                 ll_finish_md_op_data(op_data);
1735
1736         if (!S_ISDIR(inode->i_mode)) {
1737                 mutex_lock(&inode->i_mutex);
1738                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1739                         inode_dio_wait(inode);
1740         }
1741
1742         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1743                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1744
1745         return rc;
1746 }
1747
1748 int ll_setattr(struct dentry *de, struct iattr *attr)
1749 {
1750         int mode = de->d_inode->i_mode;
1751
1752         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1753                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1754                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1755
1756         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1757                                (ATTR_SIZE|ATTR_MODE)) &&
1758             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1759              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1760               !(attr->ia_mode & S_ISGID))))
1761                 attr->ia_valid |= ATTR_FORCE;
1762
1763         if ((attr->ia_valid & ATTR_MODE) &&
1764             (mode & S_ISUID) &&
1765             !(attr->ia_mode & S_ISUID) &&
1766             !(attr->ia_valid & ATTR_KILL_SUID))
1767                 attr->ia_valid |= ATTR_KILL_SUID;
1768
1769         if ((attr->ia_valid & ATTR_MODE) &&
1770             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1771             !(attr->ia_mode & S_ISGID) &&
1772             !(attr->ia_valid & ATTR_KILL_SGID))
1773                 attr->ia_valid |= ATTR_KILL_SGID;
1774
1775         return ll_setattr_raw(de, attr, false);
1776 }
1777
1778 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1779                        __u64 max_age, __u32 flags)
1780 {
1781         struct ll_sb_info *sbi = ll_s2sbi(sb);
1782         struct obd_statfs obd_osfs;
1783         int rc;
1784         ENTRY;
1785
1786         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1787         if (rc) {
1788                 CERROR("md_statfs fails: rc = %d\n", rc);
1789                 RETURN(rc);
1790         }
1791
1792         osfs->os_type = sb->s_magic;
1793
1794         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1795                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1796
1797         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1798                 flags |= OBD_STATFS_NODELAY;
1799
1800         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1801         if (rc) {
1802                 CERROR("obd_statfs fails: rc = %d\n", rc);
1803                 RETURN(rc);
1804         }
1805
1806         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1807                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1808                obd_osfs.os_files);
1809
1810         osfs->os_bsize = obd_osfs.os_bsize;
1811         osfs->os_blocks = obd_osfs.os_blocks;
1812         osfs->os_bfree = obd_osfs.os_bfree;
1813         osfs->os_bavail = obd_osfs.os_bavail;
1814
1815         /* If we don't have as many objects free on the OST as inodes
1816          * on the MDS, we reduce the total number of inodes to
1817          * compensate, so that the "inodes in use" number is correct.
1818          */
1819         if (obd_osfs.os_ffree < osfs->os_ffree) {
1820                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1821                         obd_osfs.os_ffree;
1822                 osfs->os_ffree = obd_osfs.os_ffree;
1823         }
1824
1825         RETURN(rc);
1826 }
1827 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1828 {
1829         struct super_block *sb = de->d_sb;
1830         struct obd_statfs osfs;
1831         __u64 fsid = huge_encode_dev(sb->s_dev);
1832         int rc;
1833
1834         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1835         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1836
1837         /* Some amount of caching on the client is allowed */
1838         rc = ll_statfs_internal(sb, &osfs,
1839                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1840                                 0);
1841         if (rc)
1842                 return rc;
1843
1844         statfs_unpack(sfs, &osfs);
1845
1846         /* We need to downshift for all 32-bit kernels, because we can't
1847          * tell if the kernel is being called via sys_statfs64() or not.
1848          * Stop before overflowing f_bsize - in which case it is better
1849          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1850         if (sizeof(long) < 8) {
1851                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1852                         sfs->f_bsize <<= 1;
1853
1854                         osfs.os_blocks >>= 1;
1855                         osfs.os_bfree >>= 1;
1856                         osfs.os_bavail >>= 1;
1857                 }
1858         }
1859
1860         sfs->f_blocks = osfs.os_blocks;
1861         sfs->f_bfree = osfs.os_bfree;
1862         sfs->f_bavail = osfs.os_bavail;
1863         sfs->f_fsid.val[0] = (__u32)fsid;
1864         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1865         return 0;
1866 }
1867
1868 void ll_inode_size_lock(struct inode *inode)
1869 {
1870         struct ll_inode_info *lli;
1871
1872         LASSERT(!S_ISDIR(inode->i_mode));
1873
1874         lli = ll_i2info(inode);
1875         mutex_lock(&lli->lli_size_mutex);
1876 }
1877
1878 void ll_inode_size_unlock(struct inode *inode)
1879 {
1880         struct ll_inode_info *lli;
1881
1882         lli = ll_i2info(inode);
1883         mutex_unlock(&lli->lli_size_mutex);
1884 }
1885
1886 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1887 {
1888         struct ll_inode_info *lli = ll_i2info(inode);
1889         struct mdt_body *body = md->body;
1890         struct lov_stripe_md *lsm = md->lsm;
1891         struct ll_sb_info *sbi = ll_i2sbi(inode);
1892
1893         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1894         if (lsm != NULL)
1895                 cl_file_inode_init(inode, md);
1896
1897         if (S_ISDIR(inode->i_mode)) {
1898                 int     rc;
1899
1900                 rc = ll_update_lsm_md(inode, md);
1901                 if (rc != 0)
1902                         return rc;
1903         }
1904
1905         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1906                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1907                         ll_update_remote_perm(inode, md->remote_perm);
1908         }
1909 #ifdef CONFIG_FS_POSIX_ACL
1910         else if (body->mbo_valid & OBD_MD_FLACL) {
1911                 spin_lock(&lli->lli_lock);
1912                 if (lli->lli_posix_acl)
1913                         posix_acl_release(lli->lli_posix_acl);
1914                 lli->lli_posix_acl = md->posix_acl;
1915                 spin_unlock(&lli->lli_lock);
1916         }
1917 #endif
1918         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1919                                         sbi->ll_flags & LL_SBI_32BIT_API);
1920         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1921
1922         if (body->mbo_valid & OBD_MD_FLATIME) {
1923                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1924                         LTIME_S(inode->i_atime) = body->mbo_atime;
1925                 lli->lli_atime = body->mbo_atime;
1926         }
1927
1928         if (body->mbo_valid & OBD_MD_FLMTIME) {
1929                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1930                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1931                                "to "LPU64"\n", inode->i_ino,
1932                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1933                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1934                 }
1935                 lli->lli_mtime = body->mbo_mtime;
1936         }
1937
1938         if (body->mbo_valid & OBD_MD_FLCTIME) {
1939                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1940                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1941                 lli->lli_ctime = body->mbo_ctime;
1942         }
1943
1944         if (body->mbo_valid & OBD_MD_FLMODE)
1945                 inode->i_mode = (inode->i_mode & S_IFMT) |
1946                                 (body->mbo_mode & ~S_IFMT);
1947
1948         if (body->mbo_valid & OBD_MD_FLTYPE)
1949                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1950                                 (body->mbo_mode & S_IFMT);
1951
1952         LASSERT(inode->i_mode != 0);
1953         if (S_ISREG(inode->i_mode))
1954                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1955                                        LL_MAX_BLKSIZE_BITS);
1956         else
1957                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1958
1959         if (body->mbo_valid & OBD_MD_FLUID)
1960                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1961         if (body->mbo_valid & OBD_MD_FLGID)
1962                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1963         if (body->mbo_valid & OBD_MD_FLFLAGS)
1964                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1965         if (body->mbo_valid & OBD_MD_FLNLINK)
1966                 set_nlink(inode, body->mbo_nlink);
1967         if (body->mbo_valid & OBD_MD_FLRDEV)
1968                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1969
1970         if (body->mbo_valid & OBD_MD_FLID) {
1971                 /* FID shouldn't be changed! */
1972                 if (fid_is_sane(&lli->lli_fid)) {
1973                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1974                                  "Trying to change FID "DFID
1975                                  " to the "DFID", inode "DFID"(%p)\n",
1976                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1977                                  PFID(ll_inode2fid(inode)), inode);
1978                 } else {
1979                         lli->lli_fid = body->mbo_fid1;
1980                 }
1981         }
1982
1983         LASSERT(fid_seq(&lli->lli_fid) != 0);
1984
1985         if (body->mbo_valid & OBD_MD_FLSIZE) {
1986                 i_size_write(inode, body->mbo_size);
1987
1988                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1989                        PFID(ll_inode2fid(inode)),
1990                        (unsigned long long)body->mbo_size);
1991
1992                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1993                         inode->i_blocks = body->mbo_blocks;
1994         }
1995
1996         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1997                 LASSERT(md->mds_capa);
1998                 ll_add_capa(inode, md->mds_capa);
1999         }
2000
2001         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2002                 LASSERT(md->oss_capa);
2003                 ll_add_capa(inode, md->oss_capa);
2004         }
2005
2006         if (body->mbo_valid & OBD_MD_TSTATE) {
2007                 if (body->mbo_t_state & MS_RESTORE)
2008                         lli->lli_flags |= LLIF_FILE_RESTORING;
2009         }
2010
2011         return 0;
2012 }
2013
2014 int ll_read_inode2(struct inode *inode, void *opaque)
2015 {
2016         struct lustre_md *md = opaque;
2017         struct ll_inode_info *lli = ll_i2info(inode);
2018         int     rc;
2019         ENTRY;
2020
2021         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2022                PFID(&lli->lli_fid), inode);
2023
2024         /* Core attributes from the MDS first.  This is a new inode, and
2025          * the VFS doesn't zero times in the core inode so we have to do
2026          * it ourselves.  They will be overwritten by either MDS or OST
2027          * attributes - we just need to make sure they aren't newer. */
2028         LTIME_S(inode->i_mtime) = 0;
2029         LTIME_S(inode->i_atime) = 0;
2030         LTIME_S(inode->i_ctime) = 0;
2031         inode->i_rdev = 0;
2032         rc = ll_update_inode(inode, md);
2033         if (rc != 0)
2034                 RETURN(rc);
2035
2036         /* OIDEBUG(inode); */
2037
2038         /* initializing backing dev info. */
2039         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2040
2041
2042         if (S_ISREG(inode->i_mode)) {
2043                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2044                 inode->i_op = &ll_file_inode_operations;
2045                 inode->i_fop = sbi->ll_fop;
2046                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2047                 EXIT;
2048         } else if (S_ISDIR(inode->i_mode)) {
2049                 inode->i_op = &ll_dir_inode_operations;
2050                 inode->i_fop = &ll_dir_operations;
2051                 EXIT;
2052         } else if (S_ISLNK(inode->i_mode)) {
2053                 inode->i_op = &ll_fast_symlink_inode_operations;
2054                 EXIT;
2055         } else {
2056                 inode->i_op = &ll_special_inode_operations;
2057
2058                 init_special_inode(inode, inode->i_mode,
2059                                    inode->i_rdev);
2060
2061                 EXIT;
2062         }
2063
2064         return 0;
2065 }
2066
2067 void ll_delete_inode(struct inode *inode)
2068 {
2069         struct ll_inode_info *lli = ll_i2info(inode);
2070         ENTRY;
2071
2072         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2073                 /* It is last chance to write out dirty pages,
2074                  * otherwise we may lose data while umount */
2075                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2076
2077         truncate_inode_pages_final(&inode->i_data);
2078
2079         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2080                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2081                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2082
2083 #ifdef HAVE_SBOPS_EVICT_INODE
2084         ll_clear_inode(inode);
2085 #endif
2086         clear_inode(inode);
2087
2088         EXIT;
2089 }
2090
2091 int ll_iocontrol(struct inode *inode, struct file *file,
2092                  unsigned int cmd, unsigned long arg)
2093 {
2094         struct ll_sb_info *sbi = ll_i2sbi(inode);
2095         struct ptlrpc_request *req = NULL;
2096         int rc, flags = 0;
2097         ENTRY;
2098
2099         switch(cmd) {
2100         case FSFILT_IOC_GETFLAGS: {
2101                 struct mdt_body *body;
2102                 struct md_op_data *op_data;
2103
2104                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2105                                              0, 0, LUSTRE_OPC_ANY,
2106                                              NULL);
2107                 if (IS_ERR(op_data))
2108                         RETURN(PTR_ERR(op_data));
2109
2110                 op_data->op_valid = OBD_MD_FLFLAGS;
2111                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2112                 ll_finish_md_op_data(op_data);
2113                 if (rc) {
2114                         CERROR("%s: failure inode "DFID": rc = %d\n",
2115                                sbi->ll_md_exp->exp_obd->obd_name,
2116                                PFID(ll_inode2fid(inode)), rc);
2117                         RETURN(-abs(rc));
2118                 }
2119
2120                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2121
2122                 flags = body->mbo_flags;
2123
2124                 ptlrpc_req_finished(req);
2125
2126                 RETURN(put_user(flags, (int __user *)arg));
2127         }
2128         case FSFILT_IOC_SETFLAGS: {
2129                 struct iattr *attr;
2130                 struct md_op_data *op_data;
2131                 struct cl_object *obj;
2132                 struct obd_capa *capa;
2133
2134                 if (get_user(flags, (int __user *)arg))
2135                         RETURN(-EFAULT);
2136
2137                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2138                                              LUSTRE_OPC_ANY, NULL);
2139                 if (IS_ERR(op_data))
2140                         RETURN(PTR_ERR(op_data));
2141
2142                 op_data->op_attr_flags = flags;
2143                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2144                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2145                 ll_finish_md_op_data(op_data);
2146                 ptlrpc_req_finished(req);
2147                 if (rc)
2148                         RETURN(rc);
2149
2150                 inode->i_flags = ll_ext_to_inode_flags(flags);
2151
2152                 obj = ll_i2info(inode)->lli_clob;
2153                 if (obj == NULL)
2154                         RETURN(0);
2155
2156                 OBD_ALLOC_PTR(attr);
2157                 if (attr == NULL)
2158                         RETURN(-ENOMEM);
2159
2160                 attr->ia_valid = ATTR_ATTR_FLAG;
2161
2162                 capa = ll_mdscapa_get(inode);
2163                 rc = cl_setattr_ost(obj, attr, flags, capa);
2164                 capa_put(capa);
2165
2166                 OBD_FREE_PTR(attr);
2167                 RETURN(rc);
2168         }
2169         default:
2170                 RETURN(-ENOSYS);
2171         }
2172
2173         RETURN(0);
2174 }
2175
2176 int ll_flush_ctx(struct inode *inode)
2177 {
2178         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2179
2180         CDEBUG(D_SEC, "flush context for user %d\n",
2181                from_kuid(&init_user_ns, current_uid()));
2182
2183         obd_set_info_async(NULL, sbi->ll_md_exp,
2184                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2185                            0, NULL, NULL);
2186         obd_set_info_async(NULL, sbi->ll_dt_exp,
2187                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2188                            0, NULL, NULL);
2189         return 0;
2190 }
2191
2192 /* umount -f client means force down, don't save state */
2193 void ll_umount_begin(struct super_block *sb)
2194 {
2195         struct ll_sb_info *sbi = ll_s2sbi(sb);
2196         struct obd_device *obd;
2197         struct obd_ioctl_data *ioc_data;
2198         ENTRY;
2199
2200         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2201                sb->s_count, atomic_read(&sb->s_active));
2202
2203         obd = class_exp2obd(sbi->ll_md_exp);
2204         if (obd == NULL) {
2205                 CERROR("Invalid MDC connection handle "LPX64"\n",
2206                        sbi->ll_md_exp->exp_handle.h_cookie);
2207                 EXIT;
2208                 return;
2209         }
2210         obd->obd_force = 1;
2211
2212         obd = class_exp2obd(sbi->ll_dt_exp);
2213         if (obd == NULL) {
2214                 CERROR("Invalid LOV connection handle "LPX64"\n",
2215                        sbi->ll_dt_exp->exp_handle.h_cookie);
2216                 EXIT;
2217                 return;
2218         }
2219         obd->obd_force = 1;
2220
2221         OBD_ALLOC_PTR(ioc_data);
2222         if (ioc_data) {
2223                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2224                               sizeof *ioc_data, ioc_data, NULL);
2225
2226                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2227                               sizeof *ioc_data, ioc_data, NULL);
2228
2229                 OBD_FREE_PTR(ioc_data);
2230         }
2231
2232         /* Really, we'd like to wait until there are no requests outstanding,
2233          * and then continue.  For now, we just invalidate the requests,
2234          * schedule() and sleep one second if needed, and hope.
2235          */
2236         schedule();
2237         EXIT;
2238 }
2239
2240 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2241 {
2242         struct ll_sb_info *sbi = ll_s2sbi(sb);
2243         char *profilenm = get_profile_name(sb);
2244         int err;
2245         __u32 read_only;
2246
2247         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2248                 read_only = *flags & MS_RDONLY;
2249                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2250                                          sizeof(KEY_READ_ONLY),
2251                                          KEY_READ_ONLY, sizeof(read_only),
2252                                          &read_only, NULL);
2253                 if (err) {
2254                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2255                                       profilenm, read_only ?
2256                                       "read-only" : "read-write", err);
2257                         return err;
2258                 }
2259
2260                 if (read_only)
2261                         sb->s_flags |= MS_RDONLY;
2262                 else
2263                         sb->s_flags &= ~MS_RDONLY;
2264
2265                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2266                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2267                                       read_only ?  "read-only" : "read-write");
2268         }
2269         return 0;
2270 }
2271
2272 /**
2273  * Cleanup the open handle that is cached on MDT-side.
2274  *
2275  * For open case, the client side open handling thread may hit error
2276  * after the MDT grant the open. Under such case, the client should
2277  * send close RPC to the MDT as cleanup; otherwise, the open handle
2278  * on the MDT will be leaked there until the client umount or evicted.
2279  *
2280  * In further, if someone unlinked the file, because the open handle
2281  * holds the reference on such file/object, then it will block the
2282  * subsequent threads that want to locate such object via FID.
2283  *
2284  * \param[in] sb        super block for this file-system
2285  * \param[in] open_req  pointer to the original open request
2286  */
2287 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2288 {
2289         struct mdt_body                 *body;
2290         struct md_op_data               *op_data;
2291         struct ptlrpc_request           *close_req = NULL;
2292         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2293         ENTRY;
2294
2295         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2296         OBD_ALLOC_PTR(op_data);
2297         if (op_data == NULL) {
2298                 CWARN("%s: cannot allocate op_data to release open handle for "
2299                       DFID"\n",
2300                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2301
2302                 RETURN_EXIT;
2303         }
2304
2305         op_data->op_fid1 = body->mbo_fid1;
2306         op_data->op_handle = body->mbo_handle;
2307         op_data->op_mod_time = cfs_time_current_sec();
2308         md_close(exp, op_data, NULL, &close_req);
2309         ptlrpc_req_finished(close_req);
2310         ll_finish_md_op_data(op_data);
2311
2312         EXIT;
2313 }
2314
2315 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2316                   struct super_block *sb, struct lookup_intent *it)
2317 {
2318         struct ll_sb_info *sbi = NULL;
2319         struct lustre_md md = { NULL };
2320         int rc;
2321         ENTRY;
2322
2323         LASSERT(*inode || sb);
2324         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2325         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2326                               sbi->ll_md_exp, &md);
2327         if (rc != 0)
2328                 GOTO(cleanup, rc);
2329
2330         if (*inode) {
2331                 rc = ll_update_inode(*inode, &md);
2332                 if (rc != 0)
2333                         GOTO(out, rc);
2334         } else {
2335                 LASSERT(sb != NULL);
2336
2337                 /*
2338                  * At this point server returns to client's same fid as client
2339                  * generated for creating. So using ->fid1 is okay here.
2340                  */
2341                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2342
2343                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2344                                              sbi->ll_flags & LL_SBI_32BIT_API),
2345                                  &md);
2346                 if (IS_ERR(*inode)) {
2347 #ifdef CONFIG_FS_POSIX_ACL
2348                         if (md.posix_acl) {
2349                                 posix_acl_release(md.posix_acl);
2350                                 md.posix_acl = NULL;
2351                         }
2352 #endif
2353                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2354                         *inode = NULL;
2355                         CERROR("new_inode -fatal: rc %d\n", rc);
2356                         GOTO(out, rc);
2357                 }
2358         }
2359
2360         /* Handling piggyback layout lock.
2361          * Layout lock can be piggybacked by getattr and open request.
2362          * The lsm can be applied to inode only if it comes with a layout lock
2363          * otherwise correct layout may be overwritten, for example:
2364          * 1. proc1: mdt returns a lsm but not granting layout
2365          * 2. layout was changed by another client
2366          * 3. proc2: refresh layout and layout lock granted
2367          * 4. proc1: to apply a stale layout */
2368         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2369                 struct lustre_handle lockh;
2370                 struct ldlm_lock *lock;
2371
2372                 lockh.cookie = it->d.lustre.it_lock_handle;
2373                 lock = ldlm_handle2lock(&lockh);
2374                 LASSERT(lock != NULL);
2375                 if (ldlm_has_layout(lock)) {
2376                         struct cl_object_conf conf;
2377
2378                         memset(&conf, 0, sizeof(conf));
2379                         conf.coc_opc = OBJECT_CONF_SET;
2380                         conf.coc_inode = *inode;
2381                         conf.coc_lock = lock;
2382                         conf.u.coc_md = &md;
2383                         (void)ll_layout_conf(*inode, &conf);
2384                 }
2385                 LDLM_LOCK_PUT(lock);
2386         }
2387
2388         GOTO(out, rc = 0);
2389
2390 out:
2391         if (md.lsm != NULL)
2392                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2393         md_free_lustre_md(sbi->ll_md_exp, &md);
2394
2395 cleanup:
2396         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2397                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2398
2399         return rc;
2400 }
2401
2402 int ll_obd_statfs(struct inode *inode, void __user *arg)
2403 {
2404         struct ll_sb_info *sbi = NULL;
2405         struct obd_export *exp;
2406         char *buf = NULL;
2407         struct obd_ioctl_data *data = NULL;
2408         __u32 type;
2409         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2410         int len = 0, rc;
2411
2412         if (!inode || !(sbi = ll_i2sbi(inode)))
2413                 GOTO(out_statfs, rc = -EINVAL);
2414
2415         rc = obd_ioctl_getdata(&buf, &len, arg);
2416         if (rc)
2417                 GOTO(out_statfs, rc);
2418
2419         data = (void*)buf;
2420         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2421             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2422                 GOTO(out_statfs, rc = -EINVAL);
2423
2424         if (data->ioc_inllen1 != sizeof(__u32) ||
2425             data->ioc_inllen2 != sizeof(__u32) ||
2426             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2427             data->ioc_plen2 != sizeof(struct obd_uuid))
2428                 GOTO(out_statfs, rc = -EINVAL);
2429
2430         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2431         if (type & LL_STATFS_LMV)
2432                 exp = sbi->ll_md_exp;
2433         else if (type & LL_STATFS_LOV)
2434                 exp = sbi->ll_dt_exp;
2435         else
2436                 GOTO(out_statfs, rc = -ENODEV);
2437
2438         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2439         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2440         if (rc)
2441                 GOTO(out_statfs, rc);
2442 out_statfs:
2443         if (buf)
2444                 obd_ioctl_freedata(buf, len);
2445         return rc;
2446 }
2447
2448 int ll_process_config(struct lustre_cfg *lcfg)
2449 {
2450         struct super_block *sb;
2451         unsigned long x;
2452         int rc = 0;
2453         char *ptr;
2454
2455         /* The instance name contains the sb: lustre-client-aacfe000 */
2456         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2457         if (!ptr || !*(++ptr))
2458                 return -EINVAL;
2459         if (sscanf(ptr, "%lx", &x) != 1)
2460                 return -EINVAL;
2461         sb = (struct super_block *)x;
2462         /* This better be a real Lustre superblock! */
2463         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2464
2465         /* Note we have not called client_common_fill_super yet, so
2466            proc fns must be able to handle that! */
2467         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2468                                       lcfg, sb);
2469         if (rc > 0)
2470                 rc = 0;
2471         return rc;
2472 }
2473
2474 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2475 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2476                                        struct inode *i1, struct inode *i2,
2477                                        const char *name, size_t namelen,
2478                                        __u32 mode, __u32 opc, void *data)
2479 {
2480         LASSERT(i1 != NULL);
2481
2482         if (name == NULL) {
2483                 /* Do not reuse namelen for something else. */
2484                 if (namelen != 0)
2485                         return ERR_PTR(-EINVAL);
2486         } else {
2487                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2488                         return ERR_PTR(-ENAMETOOLONG);
2489
2490                 if (!lu_name_is_valid_2(name, namelen))
2491                         return ERR_PTR(-EINVAL);
2492         }
2493
2494         if (op_data == NULL)
2495                 OBD_ALLOC_PTR(op_data);
2496
2497         if (op_data == NULL)
2498                 return ERR_PTR(-ENOMEM);
2499
2500         ll_i2gids(op_data->op_suppgids, i1, i2);
2501         op_data->op_fid1 = *ll_inode2fid(i1);
2502         op_data->op_capa1 = ll_mdscapa_get(i1);
2503         op_data->op_default_stripe_offset = -1;
2504         if (S_ISDIR(i1->i_mode)) {
2505                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2506                 if (opc == LUSTRE_OPC_MKDIR)
2507                         op_data->op_default_stripe_offset =
2508                                    ll_i2info(i1)->lli_def_stripe_offset;
2509         }
2510
2511         if (i2) {
2512                 op_data->op_fid2 = *ll_inode2fid(i2);
2513                 op_data->op_capa2 = ll_mdscapa_get(i2);
2514                 if (S_ISDIR(i2->i_mode))
2515                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2516         } else {
2517                 fid_zero(&op_data->op_fid2);
2518                 op_data->op_capa2 = NULL;
2519         }
2520
2521         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2522                 op_data->op_cli_flags |= CLI_HASH64;
2523
2524         if (ll_need_32bit_api(ll_i2sbi(i1)))
2525                 op_data->op_cli_flags |= CLI_API32;
2526
2527         op_data->op_name = name;
2528         op_data->op_namelen = namelen;
2529         op_data->op_mode = mode;
2530         op_data->op_mod_time = cfs_time_current_sec();
2531         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2532         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2533         op_data->op_cap = cfs_curproc_cap_pack();
2534         op_data->op_bias = 0;
2535         op_data->op_cli_flags = 0;
2536         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2537              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2538                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2539         } else {
2540                 op_data->op_mds = 0;
2541         }
2542         op_data->op_data = data;
2543
2544         /* When called by ll_setattr_raw, file is i1. */
2545         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2546                 op_data->op_bias |= MDS_DATA_MODIFIED;
2547
2548         return op_data;
2549 }
2550
2551 void ll_finish_md_op_data(struct md_op_data *op_data)
2552 {
2553         capa_put(op_data->op_capa1);
2554         capa_put(op_data->op_capa2);
2555         OBD_FREE_PTR(op_data);
2556 }
2557
2558 #ifdef HAVE_SUPEROPS_USE_DENTRY
2559 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2560 #else
2561 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2562 #endif
2563 {
2564         struct ll_sb_info *sbi;
2565
2566 #ifdef HAVE_SUPEROPS_USE_DENTRY
2567         LASSERT((seq != NULL) && (dentry != NULL));
2568         sbi = ll_s2sbi(dentry->d_sb);
2569 #else
2570         LASSERT((seq != NULL) && (vfs != NULL));
2571         sbi = ll_s2sbi(vfs->mnt_sb);
2572 #endif
2573
2574         if (sbi->ll_flags & LL_SBI_NOLCK)
2575                 seq_puts(seq, ",nolock");
2576
2577         if (sbi->ll_flags & LL_SBI_FLOCK)
2578                 seq_puts(seq, ",flock");
2579
2580         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2581                 seq_puts(seq, ",localflock");
2582
2583         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2584                 seq_puts(seq, ",user_xattr");
2585
2586         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2587                 seq_puts(seq, ",lazystatfs");
2588
2589         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2590                 seq_puts(seq, ",user_fid2path");
2591
2592         RETURN(0);
2593 }
2594
2595 /**
2596  * Get obd name by cmd, and copy out to user space
2597  */
2598 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2599 {
2600         struct ll_sb_info *sbi = ll_i2sbi(inode);
2601         struct obd_device *obd;
2602         ENTRY;
2603
2604         if (cmd == OBD_IOC_GETDTNAME)
2605                 obd = class_exp2obd(sbi->ll_dt_exp);
2606         else if (cmd == OBD_IOC_GETMDNAME)
2607                 obd = class_exp2obd(sbi->ll_md_exp);
2608         else
2609                 RETURN(-EINVAL);
2610
2611         if (!obd)
2612                 RETURN(-ENOENT);
2613
2614         if (copy_to_user((void __user *)arg, obd->obd_name,
2615                          strlen(obd->obd_name) + 1))
2616                 RETURN(-EFAULT);
2617
2618         RETURN(0);
2619 }
2620
2621 /**
2622  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2623  * fsname will be returned in this buffer; otherwise, a static buffer will be
2624  * used to store the fsname and returned to caller.
2625  */
2626 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2627 {
2628         static char fsname_static[MTI_NAME_MAXLEN];
2629         struct lustre_sb_info *lsi = s2lsi(sb);
2630         char *ptr;
2631         int len;
2632
2633         if (buf == NULL) {
2634                 /* this means the caller wants to use static buffer
2635                  * and it doesn't care about race. Usually this is
2636                  * in error reporting path */
2637                 buf = fsname_static;
2638                 buflen = sizeof(fsname_static);
2639         }
2640
2641         len = strlen(lsi->lsi_lmd->lmd_profile);
2642         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2643         if (ptr && (strcmp(ptr, "-client") == 0))
2644                 len -= 7;
2645
2646         if (unlikely(len >= buflen))
2647                 len = buflen - 1;
2648         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2649         buf[len] = '\0';
2650
2651         return buf;
2652 }
2653
2654 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2655 {
2656         char *path = NULL;
2657
2658         struct path p;
2659
2660         p.dentry = dentry;
2661         p.mnt = current->fs->root.mnt;
2662         path_get(&p);
2663         path = d_path(&p, buf, bufsize);
2664         path_put(&p);
2665         return path;
2666 }
2667
2668 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2669 {
2670         char *buf, *path = NULL;
2671         struct dentry *dentry = NULL;
2672         struct inode *inode = page->mapping->host;
2673
2674         /* this can be called inside spin lock so use GFP_ATOMIC. */
2675         buf = (char *)__get_free_page(GFP_ATOMIC);
2676         if (buf != NULL) {
2677                 dentry = d_find_alias(page->mapping->host);
2678                 if (dentry != NULL)
2679                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2680         }
2681
2682         CDEBUG(D_WARNING,
2683                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2684                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2685                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2686                PFID(ll_inode2fid(inode)),
2687                (path && !IS_ERR(path)) ? path : "", ioret);
2688
2689         if (dentry != NULL)
2690                 dput(dentry);
2691
2692         if (buf != NULL)
2693                 free_page((unsigned long)buf);
2694 }
2695
2696 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2697                         struct lov_user_md **kbuf)
2698 {
2699         struct lov_user_md      lum;
2700         ssize_t                 lum_size;
2701         ENTRY;
2702
2703         if (copy_from_user(&lum, md, sizeof(lum)))
2704                 RETURN(-EFAULT);
2705
2706         lum_size = ll_lov_user_md_size(&lum);
2707         if (lum_size < 0)
2708                 RETURN(lum_size);
2709
2710         OBD_ALLOC(*kbuf, lum_size);
2711         if (*kbuf == NULL)
2712                 RETURN(-ENOMEM);
2713
2714         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2715                 OBD_FREE(*kbuf, lum_size);
2716                 RETURN(-EFAULT);
2717         }
2718
2719         RETURN(lum_size);
2720 }
2721
2722 /*
2723  * Compute llite root squash state after a change of root squash
2724  * configuration setting or add/remove of a lnet nid
2725  */
2726 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2727 {
2728         struct root_squash_info *squash = &sbi->ll_squash;
2729         int i;
2730         bool matched;
2731         lnet_process_id_t id;
2732
2733         /* Update norootsquash flag */
2734         down_write(&squash->rsi_sem);
2735         if (list_empty(&squash->rsi_nosquash_nids))
2736                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2737         else {
2738                 /* Do not apply root squash as soon as one of our NIDs is
2739                  * in the nosquash_nids list */
2740                 matched = false;
2741                 i = 0;
2742                 while (LNetGetId(i++, &id) != -ENOENT) {
2743                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2744                                 continue;
2745                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2746                                 matched = true;
2747                                 break;
2748                         }
2749                 }
2750                 if (matched)
2751                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2752                 else
2753                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2754         }
2755         up_write(&squash->rsi_sem);
2756 }
2757
2758 /**
2759  * Parse linkea content to extract information about a given hardlink
2760  *
2761  * \param[in]   ldata      - Initialized linkea data
2762  * \param[in]   linkno     - Link identifier
2763  * \param[out]  parent_fid - The entry's parent FID
2764  * \param[out]  ln         - Entry name destination buffer
2765  *
2766  * \retval 0 on success
2767  * \retval Appropriate negative error code on failure
2768  */
2769 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2770                             struct lu_fid *parent_fid, struct lu_name *ln)
2771 {
2772         unsigned int    idx;
2773         int             rc;
2774         ENTRY;
2775
2776         rc = linkea_init(ldata);
2777         if (rc < 0)
2778                 RETURN(rc);
2779
2780         if (linkno >= ldata->ld_leh->leh_reccount)
2781                 /* beyond last link */
2782                 RETURN(-ENODATA);
2783
2784         linkea_first_entry(ldata);
2785         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2786                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2787                                     parent_fid);
2788                 if (idx == linkno)
2789                         break;
2790
2791                 linkea_next_entry(ldata);
2792         }
2793
2794         if (idx < linkno)
2795                 RETURN(-ENODATA);
2796
2797         RETURN(0);
2798 }
2799
2800 /**
2801  * Get parent FID and name of an identified link. Operation is performed for
2802  * a given link number, letting the caller iterate over linkno to list one or
2803  * all links of an entry.
2804  *
2805  * \param[in]     file - File descriptor against which to perform the operation
2806  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2807  *                       on and the available size. It is eventually filled with
2808  *                       the requested information or left untouched on error
2809  *
2810  * \retval - 0 on success
2811  * \retval - Appropriate negative error code on failure
2812  */
2813 int ll_getparent(struct file *file, struct getparent __user *arg)
2814 {
2815         struct dentry           *dentry = file->f_dentry;
2816         struct inode            *inode = file->f_dentry->d_inode;
2817         struct linkea_data      *ldata;
2818         struct lu_buf            buf = LU_BUF_NULL;
2819         struct lu_name           ln;
2820         struct lu_fid            parent_fid;
2821         __u32                    linkno;
2822         __u32                    name_size;
2823         int                      rc;
2824
2825         ENTRY;
2826
2827         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2828             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2829                 RETURN(-EPERM);
2830
2831         if (get_user(name_size, &arg->gp_name_size))
2832                 RETURN(-EFAULT);
2833
2834         if (get_user(linkno, &arg->gp_linkno))
2835                 RETURN(-EFAULT);
2836
2837         if (name_size > PATH_MAX)
2838                 RETURN(-EINVAL);
2839
2840         OBD_ALLOC(ldata, sizeof(*ldata));
2841         if (ldata == NULL)
2842                 RETURN(-ENOMEM);
2843
2844         rc = linkea_data_new(ldata, &buf);
2845         if (rc < 0)
2846                 GOTO(ldata_free, rc);
2847
2848         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2849         if (rc < 0)
2850                 GOTO(lb_free, rc);
2851
2852         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2853         if (rc < 0)
2854                 GOTO(lb_free, rc);
2855
2856         if (ln.ln_namelen >= name_size)
2857                 GOTO(lb_free, rc = -EOVERFLOW);
2858
2859         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2860                 GOTO(lb_free, rc = -EFAULT);
2861
2862         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2863                 GOTO(lb_free, rc = -EFAULT);
2864
2865         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2866                 GOTO(lb_free, rc = -EFAULT);
2867
2868 lb_free:
2869         lu_buf_free(&buf);
2870 ldata_free:
2871         OBD_FREE(ldata, sizeof(*ldata));
2872
2873         RETURN(rc);
2874 }