lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  *
  32  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  33  */
  34 /*
  35  * This file is part of Lustre, http://www.lustre.org/
  36  * Lustre is a trademark of Sun Microsystems, Inc.
  37  *
  38  * lustre/llite/llite_lib.c
  39  *
  40  * Lustre Light Super operations
  41  */
  42
  43 #define DEBUG_SUBSYSTEM S_LLITE
  44
  45 #include <linux/module.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49
  50 #include <lustre_lite.h>
  51 #include <lustre_ha.h>
  52 #include <lustre_dlm.h>
  53 #include <lprocfs_status.h>
  54 #include <lustre_disk.h>
  55 #include <lustre_param.h>
  56 #include <lustre_log.h>
  57 #include <cl_object.h>
  58 #include <obd_cksum.h>
  59 #include "llite_internal.h"
  60
  61 cfs_mem_cache_t *ll_file_data_slab;
  62
  63 CFS_LIST_HEAD(ll_super_blocks);
  64 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  65
  66 #ifndef MS_HAS_NEW_AOPS
  67 extern struct address_space_operations ll_aops;
  68 extern struct address_space_operations ll_dir_aops;
  69 #else
  70 extern struct address_space_operations_ext ll_aops;
  71 extern struct address_space_operations_ext ll_dir_aops;
  72 #endif
  73
  74 #ifndef log2
  75 #define log2(n) cfs_ffz(~(n))
  76 #endif
  77
  78 static struct ll_sb_info *ll_init_sbi(void)
  79 {
  80         struct ll_sb_info *sbi = NULL;
  81         unsigned long pages;
  82         struct sysinfo si;
  83         class_uuid_t uuid;
  84         int i;
  85         ENTRY;
  86
  87         OBD_ALLOC(sbi, sizeof(*sbi));
  88         if (!sbi)
  89                 RETURN(NULL);
  90
  91         cfs_spin_lock_init(&sbi->ll_lock);
  92         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  93         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  94         cfs_spin_lock_init(&sbi->ll_process_lock);
  95         sbi->ll_rw_stats_on = 0;
  96
  97         si_meminfo(&si);
  98         pages = si.totalram - si.totalhigh;
  99         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 100 #ifdef HAVE_BGL_SUPPORT
 101                 sbi->ll_async_page_max = pages / 4;
 102 #else
 103                 sbi->ll_async_page_max = pages / 2;
 104 #endif
 105         } else {
 106                 sbi->ll_async_page_max = (pages / 4) * 3;
 107         }
 108
 109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 110                                            SBI_DEFAULT_READAHEAD_MAX);
 111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 116
 117         ll_generate_random_uuid(uuid);
 118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 120
 121         cfs_spin_lock(&ll_sb_lock);
 122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 123         cfs_spin_unlock(&ll_sb_lock);
 124
 125 #ifdef ENABLE_CHECKSUM
 126         sbi->ll_flags |= LL_SBI_CHECKSUM;
 127 #endif
 128
 129 #ifdef HAVE_LRU_RESIZE_SUPPORT
 130         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 131 #endif
 132
 133         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 134                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 135                                    pp_r_hist.oh_lock);
 136                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 137                                    pp_w_hist.oh_lock);
 138         }
 139
 140         /* metadata statahead is enabled by default */
 141         sbi->ll_sa_max = LL_SA_RPC_DEF;
 142         cfs_atomic_set(&sbi->ll_sa_total, 0);
 143         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 144         cfs_atomic_set(&sbi->ll_agl_total, 0);
 145         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 146
 147         RETURN(sbi);
 148 }
 149
 150 void ll_free_sbi(struct super_block *sb)
 151 {
 152         struct ll_sb_info *sbi = ll_s2sbi(sb);
 153         ENTRY;
 154
 155         if (sbi != NULL) {
 156                 cfs_spin_lock(&ll_sb_lock);
 157                 cfs_list_del(&sbi->ll_list);
 158                 cfs_spin_unlock(&ll_sb_lock);
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static struct dentry_operations ll_d_root_ops = {
 165         .d_compare = ll_dcompare,
 166         .d_revalidate = ll_revalidate_nd,
 167 };
 168
 169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 170                                     struct vfsmount *mnt)
 171 {
 172         struct inode *root = 0;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct obd_capa *oc = NULL;
 176         struct obd_statfs *osfs = NULL;
 177         struct ptlrpc_request *request = NULL;
 178         struct obd_connect_data *data = NULL;
 179         struct obd_uuid *uuid;
 180         struct md_op_data *op_data;
 181         struct lustre_md lmd;
 182         obd_valid valid;
 183         int size, err, checksum;
 184         ENTRY;
 185
 186         obd = class_name2obd(md);
 187         if (!obd) {
 188                 CERROR("MD %s: not setup or attached\n", md);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         OBD_ALLOC_PTR(data);
 193         if (data == NULL)
 194                 RETURN(-ENOMEM);
 195
 196         OBD_ALLOC_PTR(osfs);
 197         if (osfs == NULL) {
 198                 OBD_FREE_PTR(data);
 199                 RETURN(-ENOMEM);
 200         }
 201
 202         if (proc_lustre_fs_root) {
 203                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 204                                                   dt, md);
 205                 if (err < 0)
 206                         CERROR("could not register mount in /proc/fs/lustre\n");
 207         }
 208
 209         /* indicate the features supported by this client */
 210         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 211                                   OBD_CONNECT_ATTRFID  |
 212                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 213                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 214                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 215                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 216                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 217                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH;
 218
 219         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 220                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 221
 222 #ifdef HAVE_LRU_RESIZE_SUPPORT
 223         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 224                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 225 #endif
 226 #ifdef CONFIG_FS_POSIX_ACL
 227         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 228 #endif
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237 #ifdef HAVE_MS_FLOCK_LOCK
 238         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 239         sb->s_flags |= MS_FLOCK_LOCK;
 240 #endif
 241 #ifdef MS_HAS_NEW_AOPS
 242         sb->s_flags |= MS_HAS_NEW_AOPS;
 243 #endif
 244
 245         if (sbi->ll_flags & LL_SBI_FLOCK)
 246                 sbi->ll_fop = &ll_file_operations_flock;
 247         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 248                 sbi->ll_fop = &ll_file_operations;
 249         else
 250                 sbi->ll_fop = &ll_file_operations_noflock;
 251
 252         /* real client */
 253         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 254         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 255                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 256
 257         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 258
 259         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 260         if (err == -EBUSY) {
 261                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 262                                    "recovery, of which this client is not a "
 263                                    "part. Please wait for recovery to complete,"
 264                                    " abort, or time out.\n", md);
 265                 GOTO(out, err);
 266         } else if (err) {
 267                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 268                 GOTO(out, err);
 269         }
 270
 271         err = obd_fid_init(sbi->ll_md_exp);
 272         if (err) {
 273                 CERROR("Can't init metadata layer FID infrastructure, "
 274                        "rc %d\n", err);
 275                 GOTO(out_md, err);
 276         }
 277
 278         err = obd_statfs(obd, osfs,
 279                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 280         if (err)
 281                 GOTO(out_md_fid, err);
 282
 283         /* This needs to be after statfs to ensure connect has finished.
 284          * Note that "data" does NOT contain the valid connect reply.
 285          * If connecting to a 1.8 server there will be no LMV device, so
 286          * we can access the MDC export directly and exp_connect_flags will
 287          * be non-zero, but if accessing an upgraded 2.1 server it will
 288          * have the correct flags filled in.
 289          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 290         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 291         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 292             valid != CLIENT_CONNECT_MDT_REQD) {
 293                 char *buf;
 294
 295                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 296                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 297                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 298                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 299                                    "feature(s) needed for correct operation "
 300                                    "of this client (%s). Please upgrade "
 301                                    "server or downgrade client.\n",
 302                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 303                 OBD_FREE(buf, CFS_PAGE_SIZE);
 304                 GOTO(out_md, err = -EPROTO);
 305         }
 306
 307         size = sizeof(*data);
 308         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 309                            KEY_CONN_DATA,  &size, data, NULL);
 310         if (err) {
 311                 CERROR("Get connect data failed: %d \n", err);
 312                 GOTO(out_md, err);
 313         }
 314
 315         LASSERT(osfs->os_bsize);
 316         sb->s_blocksize = osfs->os_bsize;
 317         sb->s_blocksize_bits = log2(osfs->os_bsize);
 318         sb->s_magic = LL_SUPER_MAGIC;
 319         sb->s_maxbytes = MAX_LFS_FILESIZE;
 320         sbi->ll_namelen = osfs->os_namelen;
 321         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 322
 323         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 324             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 325                 LCONSOLE_INFO("Disabling user_xattr feature because "
 326                               "it is not supported on the server\n");
 327                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 328         }
 329
 330         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 331 #ifdef MS_POSIXACL
 332                 sb->s_flags |= MS_POSIXACL;
 333 #endif
 334                 sbi->ll_flags |= LL_SBI_ACL;
 335         } else {
 336                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 337 #ifdef MS_POSIXACL
 338                 sb->s_flags &= ~MS_POSIXACL;
 339 #endif
 340                 sbi->ll_flags &= ~LL_SBI_ACL;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 344                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 345                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 346                         LCONSOLE_INFO("client is set as remote by default.\n");
 347                 }
 348         } else {
 349                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 350                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 351                         LCONSOLE_INFO("client claims to be remote, but server "
 352                                       "rejected, forced to be local.\n");
 353                 }
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 357                 LCONSOLE_INFO("client enabled MDS capability!\n");
 358                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 359         }
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 362                 LCONSOLE_INFO("client enabled OSS capability!\n");
 363                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 364         }
 365
 366         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 367                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 370                 sbi->ll_md_brw_size = data->ocd_brw_size;
 371         else
 372                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 373
 374         obd = class_name2obd(dt);
 375         if (!obd) {
 376                 CERROR("DT %s: not setup or attached\n", dt);
 377                 GOTO(out_md_fid, err = -ENODEV);
 378         }
 379
 380         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 381                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 382                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 383                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 384                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 385                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 386                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 387                                   OBD_CONNECT_MAXBYTES;
 388
 389         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 390                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 391
 392         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 393                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 394                  * disabled by default, because it can still be enabled on the
 395                  * fly via /proc. As a consequence, we still need to come to an
 396                  * agreement on the supported algorithms at connect time */
 397                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 398
 399                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 400                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 401                 else
 402                         data->ocd_cksum_types = cksum_types_supported();
 403         }
 404
 405 #ifdef HAVE_LRU_RESIZE_SUPPORT
 406         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 407 #endif
 408         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 409                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 410
 411         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 412                "ocd_grant: %d\n", data->ocd_connect_flags,
 413                data->ocd_version, data->ocd_grant);
 414
 415         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 416         obd->obd_upcall.onu_upcall = cl_ocd_update;
 417
 418         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 419
 420         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 421         if (err == -EBUSY) {
 422                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 423                                    "recovery, of which this client is not a "
 424                                    "part.  Please wait for recovery to "
 425                                    "complete, abort, or time out.\n", dt);
 426                 GOTO(out_md_fid, err);
 427         } else if (err) {
 428                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 429                 GOTO(out_md_fid, err);
 430         }
 431
 432         err = obd_fid_init(sbi->ll_dt_exp);
 433         if (err) {
 434                 CERROR("Can't init data layer FID infrastructure, "
 435                        "rc %d\n", err);
 436                 GOTO(out_dt, err);
 437         }
 438
 439         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 440         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 441         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 442         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 443         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 444
 445         fid_zero(&sbi->ll_root_fid);
 446         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 447         if (err) {
 448                 CERROR("cannot mds_connect: rc = %d\n", err);
 449                 GOTO(out_lock_cn_cb, err);
 450         }
 451         if (!fid_is_sane(&sbi->ll_root_fid)) {
 452                 CERROR("Invalid root fid during mount\n");
 453                 GOTO(out_lock_cn_cb, err = -EINVAL);
 454         }
 455         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 456
 457         sb->s_op = &lustre_super_operations;
 458 #if THREAD_SIZE >= 8192
 459         sb->s_export_op = &lustre_export_operations;
 460 #endif
 461
 462         /* make root inode
 463          * XXX: move this to after cbd setup? */
 464         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 465         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 466                 valid |= OBD_MD_FLRMTPERM;
 467         else if (sbi->ll_flags & LL_SBI_ACL)
 468                 valid |= OBD_MD_FLACL;
 469
 470         OBD_ALLOC_PTR(op_data);
 471         if (op_data == NULL)
 472                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 473
 474         op_data->op_fid1 = sbi->ll_root_fid;
 475         op_data->op_mode = 0;
 476         op_data->op_capa1 = oc;
 477         op_data->op_valid = valid;
 478
 479         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 480         if (oc)
 481                 capa_put(oc);
 482         OBD_FREE_PTR(op_data);
 483         if (err) {
 484                 CERROR("md_getattr failed for root: rc = %d\n", err);
 485                 GOTO(out_lock_cn_cb, err);
 486         }
 487         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 488                                sbi->ll_md_exp, &lmd);
 489         if (err) {
 490                 CERROR("failed to understand root inode md: rc = %d\n", err);
 491                 ptlrpc_req_finished (request);
 492                 GOTO(out_lock_cn_cb, err);
 493         }
 494
 495         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 496         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 497         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 498         ptlrpc_req_finished(request);
 499
 500         if (root == NULL || IS_ERR(root)) {
 501                 if (lmd.lsm)
 502                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 503 #ifdef CONFIG_FS_POSIX_ACL
 504                 if (lmd.posix_acl) {
 505                         posix_acl_release(lmd.posix_acl);
 506                         lmd.posix_acl = NULL;
 507                 }
 508 #endif
 509                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 510                 root = NULL;
 511                 CERROR("lustre_lite: bad iget4 for root\n");
 512                 GOTO(out_root, err);
 513         }
 514
 515         err = ll_close_thread_start(&sbi->ll_lcq);
 516         if (err) {
 517                 CERROR("cannot start close thread: rc %d\n", err);
 518                 GOTO(out_root, err);
 519         }
 520
 521 #ifdef CONFIG_FS_POSIX_ACL
 522         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 523                 rct_init(&sbi->ll_rct);
 524                 et_init(&sbi->ll_et);
 525         }
 526 #endif
 527
 528         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 529         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 530                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 531                                  NULL);
 532         cl_sb_init(sb);
 533
 534         sb->s_root = d_alloc_root(root);
 535         sb->s_root->d_op = &ll_d_root_ops;
 536
 537         sbi->ll_sdev_orig = sb->s_dev;
 538
 539         /* We set sb->s_dev equal on all lustre clients in order to support
 540          * NFS export clustering.  NFSD requires that the FSID be the same
 541          * on all clients. */
 542         /* s_dev is also used in lt_compare() to compare two fs, but that is
 543          * only a node-local comparison. */
 544         uuid = obd_get_uuid(sbi->ll_md_exp);
 545         if (uuid != NULL)
 546                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 547         sbi->ll_mnt = mnt;
 548
 549         if (data != NULL)
 550                 OBD_FREE_PTR(data);
 551         if (osfs != NULL)
 552                 OBD_FREE_PTR(osfs);
 553
 554         RETURN(err);
 555 out_root:
 556         if (root)
 557                 iput(root);
 558 out_lock_cn_cb:
 559         obd_fid_fini(sbi->ll_dt_exp);
 560 out_dt:
 561         obd_disconnect(sbi->ll_dt_exp);
 562         sbi->ll_dt_exp = NULL;
 563 out_md_fid:
 564         obd_fid_fini(sbi->ll_md_exp);
 565 out_md:
 566         obd_disconnect(sbi->ll_md_exp);
 567         sbi->ll_md_exp = NULL;
 568 out:
 569         if (data != NULL)
 570                 OBD_FREE_PTR(data);
 571         if (osfs != NULL)
 572                 OBD_FREE_PTR(osfs);
 573         lprocfs_unregister_mountpoint(sbi);
 574         return err;
 575 }
 576
 577 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 578 {
 579         int size, rc;
 580
 581         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 582         size = sizeof(int);
 583         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 584                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 585         if (rc)
 586                 CERROR("Get max mdsize error rc %d \n", rc);
 587
 588         RETURN(rc);
 589 }
 590
 591 void ll_dump_inode(struct inode *inode)
 592 {
 593         struct list_head *tmp;
 594         int dentry_count = 0;
 595
 596         LASSERT(inode != NULL);
 597
 598         list_for_each(tmp, &inode->i_dentry)
 599                 dentry_count++;
 600
 601         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 602                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 603                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 604 }
 605
 606 void lustre_dump_dentry(struct dentry *dentry, int recur)
 607 {
 608         struct list_head *tmp;
 609         int subdirs = 0;
 610
 611         LASSERT(dentry != NULL);
 612
 613         list_for_each(tmp, &dentry->d_subdirs)
 614                 subdirs++;
 615
 616         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 617                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 618                dentry->d_name.len, dentry->d_name.name,
 619                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 620                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 621                dentry->d_flags, dentry->d_fsdata, subdirs);
 622         if (dentry->d_inode != NULL)
 623                 ll_dump_inode(dentry->d_inode);
 624
 625         if (recur == 0)
 626                 return;
 627
 628        list_for_each(tmp, &dentry->d_subdirs) {
 629                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 630                 lustre_dump_dentry(d, recur - 1);
 631         }
 632 }
 633
 634 void client_common_put_super(struct super_block *sb)
 635 {
 636         struct ll_sb_info *sbi = ll_s2sbi(sb);
 637         ENTRY;
 638
 639 #ifdef CONFIG_FS_POSIX_ACL
 640         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 641                 et_fini(&sbi->ll_et);
 642                 rct_fini(&sbi->ll_rct);
 643         }
 644 #endif
 645
 646         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 647
 648         ll_close_thread_shutdown(sbi->ll_lcq);
 649
 650         cl_sb_fini(sb);
 651
 652         cfs_list_del(&sbi->ll_conn_chain);
 653
 654         obd_fid_fini(sbi->ll_dt_exp);
 655         obd_disconnect(sbi->ll_dt_exp);
 656         sbi->ll_dt_exp = NULL;
 657
 658         lprocfs_unregister_mountpoint(sbi);
 659
 660         obd_fid_fini(sbi->ll_md_exp);
 661         obd_disconnect(sbi->ll_md_exp);
 662         sbi->ll_md_exp = NULL;
 663
 664         EXIT;
 665 }
 666
 667 void ll_kill_super(struct super_block *sb)
 668 {
 669         struct ll_sb_info *sbi;
 670
 671         ENTRY;
 672
 673         /* not init sb ?*/
 674         if (!(sb->s_flags & MS_ACTIVE))
 675                 return;
 676
 677         sbi = ll_s2sbi(sb);
 678         /* we need restore s_dev from changed for clustred NFS before put_super
 679          * because new kernels have cached s_dev and change sb->s_dev in
 680          * put_super not affected real removing devices */
 681         if (sbi)
 682                 sb->s_dev = sbi->ll_sdev_orig;
 683         EXIT;
 684 }
 685
 686 char *ll_read_opt(const char *opt, char *data)
 687 {
 688         char *value;
 689         char *retval;
 690         ENTRY;
 691
 692         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 693         if (strncmp(opt, data, strlen(opt)))
 694                 RETURN(NULL);
 695         if ((value = strchr(data, '=')) == NULL)
 696                 RETURN(NULL);
 697
 698         value++;
 699         OBD_ALLOC(retval, strlen(value) + 1);
 700         if (!retval) {
 701                 CERROR("out of memory!\n");
 702                 RETURN(NULL);
 703         }
 704
 705         memcpy(retval, value, strlen(value)+1);
 706         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 707         RETURN(retval);
 708 }
 709
 710 static inline int ll_set_opt(const char *opt, char *data, int fl)
 711 {
 712         if (strncmp(opt, data, strlen(opt)) != 0)
 713                 return(0);
 714         else
 715                 return(fl);
 716 }
 717
 718 /* non-client-specific mount options are parsed in lmd_parse */
 719 static int ll_options(char *options, int *flags)
 720 {
 721         int tmp;
 722         char *s1 = options, *s2;
 723         ENTRY;
 724
 725         if (!options)
 726                 RETURN(0);
 727
 728         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 729
 730         while (*s1) {
 731                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 732                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 733                 if (tmp) {
 734                         *flags |= tmp;
 735                         goto next;
 736                 }
 737                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 738                 if (tmp) {
 739                         *flags |= tmp;
 740                         goto next;
 741                 }
 742                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 743                 if (tmp) {
 744                         *flags |= tmp;
 745                         goto next;
 746                 }
 747                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 748                 if (tmp) {
 749                         *flags &= ~tmp;
 750                         goto next;
 751                 }
 752                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 753                 if (tmp) {
 754                         *flags |= tmp;
 755                         goto next;
 756                 }
 757                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 758                 if (tmp) {
 759                         *flags &= ~tmp;
 760                         goto next;
 761                 }
 762 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 763                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 764                 if (tmp) {
 765                         /* Ignore deprecated mount option.  The client will
 766                          * always try to mount with ACL support, whether this
 767                          * is used depends on whether server supports it. */
 768                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 769                                                   "mount option 'acl'.\n");
 770                         goto next;
 771                 }
 772                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 773                 if (tmp) {
 774                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 775                                                   "mount option 'noacl'.\n");
 776                         goto next;
 777                 }
 778 #else
 779 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 780 #endif
 781                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786
 787                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 788                 if (tmp) {
 789                         *flags |= tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 793                 if (tmp) {
 794                         *flags &= ~tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 803                 if (tmp) {
 804                         *flags &= ~tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 813                 if (tmp) {
 814                         *flags &= ~tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827
 828                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 829                                    s1);
 830                 RETURN(-EINVAL);
 831
 832 next:
 833                 /* Find next opt */
 834                 s2 = strchr(s1, ',');
 835                 if (s2 == NULL)
 836                         break;
 837                 s1 = s2 + 1;
 838         }
 839         RETURN(0);
 840 }
 841
 842 void ll_lli_init(struct ll_inode_info *lli)
 843 {
 844         lli->lli_inode_magic = LLI_INODE_MAGIC;
 845         lli->lli_flags = 0;
 846         lli->lli_ioepoch = 0;
 847         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 848         cfs_spin_lock_init(&lli->lli_lock);
 849         lli->lli_posix_acl = NULL;
 850         lli->lli_remote_perms = NULL;
 851         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 852         /* Do not set lli_fid, it has been initialized already. */
 853         fid_zero(&lli->lli_pfid);
 854         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 855         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 856         cfs_atomic_set(&lli->lli_open_count, 0);
 857         lli->lli_mds_capa = NULL;
 858         lli->lli_rmtperm_time = 0;
 859         lli->lli_pending_och = NULL;
 860         lli->lli_mds_read_och = NULL;
 861         lli->lli_mds_write_och = NULL;
 862         lli->lli_mds_exec_och = NULL;
 863         lli->lli_open_fd_read_count = 0;
 864         lli->lli_open_fd_write_count = 0;
 865         lli->lli_open_fd_exec_count = 0;
 866         cfs_mutex_init(&lli->lli_och_mutex);
 867         cfs_spin_lock_init(&lli->lli_agl_lock);
 868         lli->lli_smd = NULL;
 869         lli->lli_clob = NULL;
 870
 871         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 872         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 873                 cfs_mutex_init(&lli->lli_readdir_mutex);
 874                 lli->lli_opendir_key = NULL;
 875                 lli->lli_sai = NULL;
 876                 lli->lli_sa_pos = 0;
 877                 lli->lli_def_acl = NULL;
 878                 cfs_spin_lock_init(&lli->lli_sa_lock);
 879                 lli->lli_opendir_pid = 0;
 880         } else {
 881                 cfs_sema_init(&lli->lli_size_sem, 1);
 882                 lli->lli_size_sem_owner = NULL;
 883                 lli->lli_symlink_name = NULL;
 884                 cfs_init_rwsem(&lli->lli_trunc_sem);
 885                 cfs_mutex_init(&lli->lli_write_mutex);
 886                 lli->lli_async_rc = 0;
 887                 lli->lli_write_rc = 0;
 888                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 889                 lli->lli_glimpse_time = 0;
 890                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 891                 lli->lli_agl_index = 0;
 892         }
 893 }
 894
 895 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 896 {
 897 #ifdef HAVE_BDI_REGISTER
 898         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 899
 900 #ifdef HAVE_BDI_NAME
 901         bdi->name = "lustre";
 902 #endif
 903         return bdi_register(bdi, NULL, "lustre-%d",
 904                             atomic_inc_return(&ll_bdi_num));
 905 #else
 906         return 0;
 907 #endif
 908 }
 909
 910 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 911 {
 912         struct lustre_profile *lprof = NULL;
 913         struct lustre_sb_info *lsi = s2lsi(sb);
 914         struct ll_sb_info *sbi;
 915         char  *dt = NULL, *md = NULL;
 916         char  *profilenm = get_profile_name(sb);
 917         struct config_llog_instance *cfg;
 918         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 919         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 920         int    err;
 921         ENTRY;
 922
 923         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 924
 925         OBD_ALLOC_PTR(cfg);
 926         if (cfg == NULL)
 927                 RETURN(-ENOMEM);
 928
 929         cfs_module_get();
 930
 931         /* client additional sb info */
 932         lsi->lsi_llsbi = sbi = ll_init_sbi();
 933         if (!sbi) {
 934                 cfs_module_put(THIS_MODULE);
 935                 OBD_FREE_PTR(cfg);
 936                 RETURN(-ENOMEM);
 937         }
 938
 939         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 940         if (err)
 941                 GOTO(out_free, err);
 942
 943         err = ll_bdi_init(&lsi->lsi_bdi);
 944         if (err)
 945                 GOTO(out_free, err);
 946         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 947         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 948         err = ll_bdi_register(&lsi->lsi_bdi);
 949         if (err)
 950                 GOTO(out_free, err);
 951
 952 #ifdef HAVE_SB_BDI
 953         sb->s_bdi = &lsi->lsi_bdi;
 954 #endif
 955
 956         /* Generate a string unique to this super, in case some joker tries
 957            to mount the same fs at two mount points.
 958            Use the address of the super itself.*/
 959         cfg->cfg_instance = sb;
 960         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 961
 962         /* set up client obds */
 963         err = lustre_process_log(sb, profilenm, cfg);
 964         if (err < 0) {
 965                 CERROR("Unable to process log: %d\n", err);
 966                 GOTO(out_free, err);
 967         }
 968
 969         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 970         lprof = class_get_profile(profilenm);
 971         if (lprof == NULL) {
 972                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 973                                    " read from the MGS.  Does that filesystem "
 974                                    "exist?\n", profilenm);
 975                 GOTO(out_free, err = -EINVAL);
 976         }
 977         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 978                lprof->lp_md, lprof->lp_dt);
 979
 980         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 981         if (!dt)
 982                 GOTO(out_free, err = -ENOMEM);
 983         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 984
 985         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
 986         if (!md)
 987                 GOTO(out_free, err = -ENOMEM);
 988         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
 989
 990         /* connections, registrations, sb setup */
 991         err = client_common_fill_super(sb, md, dt, mnt);
 992
 993 out_free:
 994         if (md)
 995                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
 996         if (dt)
 997                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
 998         if (err)
 999                 ll_put_super(sb);
1000         else
1001                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1002
1003         OBD_FREE_PTR(cfg);
1004         RETURN(err);
1005 } /* ll_fill_super */
1006
1007
1008 void lu_context_keys_dump(void);
1009
1010 void ll_put_super(struct super_block *sb)
1011 {
1012         struct config_llog_instance cfg;
1013         struct obd_device *obd;
1014         struct lustre_sb_info *lsi = s2lsi(sb);
1015         struct ll_sb_info *sbi = ll_s2sbi(sb);
1016         char *profilenm = get_profile_name(sb);
1017         int force = 1, next;
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1021
1022         ll_print_capa_stat(sbi);
1023
1024         cfg.cfg_instance = sb;
1025         lustre_end_log(sb, profilenm, &cfg);
1026
1027         if (sbi->ll_md_exp) {
1028                 obd = class_exp2obd(sbi->ll_md_exp);
1029                 if (obd)
1030                         force = obd->obd_force;
1031         }
1032
1033         /* We need to set force before the lov_disconnect in
1034            lustre_common_put_super, since l_d cleans up osc's as well. */
1035         if (force) {
1036                 next = 0;
1037                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1038                                                      &next)) != NULL) {
1039                         obd->obd_force = force;
1040                 }
1041         }
1042
1043         if (sbi->ll_lcq) {
1044                 /* Only if client_common_fill_super succeeded */
1045                 client_common_put_super(sb);
1046         }
1047
1048         next = 0;
1049         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1050                 class_manual_cleanup(obd);
1051         }
1052
1053         if (profilenm)
1054                 class_del_profile(profilenm);
1055
1056         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1057                 ll_bdi_destroy(&lsi->lsi_bdi);
1058                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1059         }
1060
1061         ll_free_sbi(sb);
1062         lsi->lsi_llsbi = NULL;
1063
1064         lustre_common_put_super(sb);
1065
1066         cl_env_cache_purge(~0);
1067
1068         LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1069
1070         cfs_module_put(THIS_MODULE);
1071
1072         EXIT;
1073 } /* client_put_super */
1074
1075 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1076 {
1077         struct inode *inode = NULL;
1078         /* NOTE: we depend on atomic igrab() -bzzz */
1079         lock_res_and_lock(lock);
1080         if (lock->l_ast_data) {
1081                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1082                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1083                         inode = igrab(lock->l_ast_data);
1084                 } else {
1085                         inode = lock->l_ast_data;
1086                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1087                                          D_WARNING, lock, "l_ast_data %p is "
1088                                          "bogus: magic %08x", lock->l_ast_data,
1089                                          lli->lli_inode_magic);
1090                         inode = NULL;
1091                 }
1092         }
1093         unlock_res_and_lock(lock);
1094         return inode;
1095 }
1096
1097 static int null_if_equal(struct ldlm_lock *lock, void *data)
1098 {
1099         if (data == lock->l_ast_data) {
1100                 lock->l_ast_data = NULL;
1101
1102                 if (lock->l_req_mode != lock->l_granted_mode)
1103                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1104         }
1105
1106         return LDLM_ITER_CONTINUE;
1107 }
1108
1109 void ll_clear_inode(struct inode *inode)
1110 {
1111         struct ll_inode_info *lli = ll_i2info(inode);
1112         struct ll_sb_info *sbi = ll_i2sbi(inode);
1113         ENTRY;
1114
1115         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1116                inode->i_generation, inode);
1117
1118         if (S_ISDIR(inode->i_mode)) {
1119                 /* these should have been cleared in ll_file_release */
1120                 LASSERT(lli->lli_opendir_key == NULL);
1121                 LASSERT(lli->lli_sai == NULL);
1122                 LASSERT(lli->lli_opendir_pid == 0);
1123         }
1124
1125         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1126         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1127                          null_if_equal, inode);
1128
1129         LASSERT(!lli->lli_open_fd_write_count);
1130         LASSERT(!lli->lli_open_fd_read_count);
1131         LASSERT(!lli->lli_open_fd_exec_count);
1132
1133         if (lli->lli_mds_write_och)
1134                 ll_md_real_close(inode, FMODE_WRITE);
1135         if (lli->lli_mds_exec_och)
1136                 ll_md_real_close(inode, FMODE_EXEC);
1137         if (lli->lli_mds_read_och)
1138                 ll_md_real_close(inode, FMODE_READ);
1139
1140         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1141                 OBD_FREE(lli->lli_symlink_name,
1142                          strlen(lli->lli_symlink_name) + 1);
1143                 lli->lli_symlink_name = NULL;
1144         }
1145
1146         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1147                 LASSERT(lli->lli_posix_acl == NULL);
1148                 if (lli->lli_remote_perms) {
1149                         free_rmtperm_hash(lli->lli_remote_perms);
1150                         lli->lli_remote_perms = NULL;
1151                 }
1152         }
1153 #ifdef CONFIG_FS_POSIX_ACL
1154         else if (lli->lli_posix_acl) {
1155                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1156                 LASSERT(lli->lli_remote_perms == NULL);
1157                 posix_acl_release(lli->lli_posix_acl);
1158                 lli->lli_posix_acl = NULL;
1159         }
1160 #endif
1161         lli->lli_inode_magic = LLI_INODE_DEAD;
1162
1163         ll_clear_inode_capas(inode);
1164         if (!S_ISDIR(inode->i_mode))
1165                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1166
1167         /*
1168          * XXX This has to be done before lsm is freed below, because
1169          * cl_object still uses inode lsm.
1170          */
1171         cl_inode_fini(inode);
1172
1173         if (lli->lli_smd) {
1174                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1175                 lli->lli_smd = NULL;
1176         }
1177
1178
1179         EXIT;
1180 }
1181
1182 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1183                   struct md_open_data **mod)
1184 {
1185         struct lustre_md md;
1186         struct inode *inode = dentry->d_inode;
1187         struct ll_sb_info *sbi = ll_i2sbi(inode);
1188         struct ptlrpc_request *request = NULL;
1189         int rc, ia_valid;
1190         ENTRY;
1191
1192         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1193                                      LUSTRE_OPC_ANY, NULL);
1194         if (IS_ERR(op_data))
1195                 RETURN(PTR_ERR(op_data));
1196
1197         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1198                         &request, mod);
1199         if (rc) {
1200                 ptlrpc_req_finished(request);
1201                 if (rc == -ENOENT) {
1202                         inode->i_nlink = 0;
1203                         /* Unlinked special device node? Or just a race?
1204                          * Pretend we done everything. */
1205                         if (!S_ISREG(inode->i_mode) &&
1206                             !S_ISDIR(inode->i_mode)) {
1207                                 ia_valid = op_data->op_attr.ia_valid;
1208                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1209                                 rc = simple_setattr(dentry, &op_data->op_attr);
1210                                 op_data->op_attr.ia_valid = ia_valid;
1211                         }
1212                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1213                         CERROR("md_setattr fails: rc = %d\n", rc);
1214                 }
1215                 RETURN(rc);
1216         }
1217
1218         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1219                               sbi->ll_md_exp, &md);
1220         if (rc) {
1221                 ptlrpc_req_finished(request);
1222                 RETURN(rc);
1223         }
1224
1225         /* We want to adjust timestamps.
1226          * If there is at least some data in file, we cleared ATTR_SIZE
1227          * to avoid update size, otherwise it is important to do.(SOM case)
1228          * (bug 6196) */
1229         ia_valid = op_data->op_attr.ia_valid;
1230         /* Since we set ATTR_*_SET flags above, and already done permission
1231          * check, So don't let inode_change_ok() check it again.  */
1232         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1233         rc = simple_setattr(dentry, &op_data->op_attr);
1234         op_data->op_attr.ia_valid = ia_valid;
1235
1236         /* Extract epoch data if obtained. */
1237         op_data->op_handle = md.body->handle;
1238         op_data->op_ioepoch = md.body->ioepoch;
1239
1240         ll_update_inode(inode, &md);
1241         ptlrpc_req_finished(request);
1242
1243         RETURN(rc);
1244 }
1245
1246 /* Close IO epoch and send Size-on-MDS attribute update. */
1247 static int ll_setattr_done_writing(struct inode *inode,
1248                                    struct md_op_data *op_data,
1249                                    struct md_open_data *mod)
1250 {
1251         struct ll_inode_info *lli = ll_i2info(inode);
1252         int rc = 0;
1253         ENTRY;
1254
1255         LASSERT(op_data != NULL);
1256         if (!S_ISREG(inode->i_mode))
1257                 RETURN(0);
1258
1259         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1260                op_data->op_ioepoch, PFID(&lli->lli_fid));
1261
1262         op_data->op_flags = MF_EPOCH_CLOSE;
1263         ll_done_writing_attr(inode, op_data);
1264         ll_pack_inode2opdata(inode, op_data, NULL);
1265
1266         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1267         if (rc == -EAGAIN) {
1268                 /* MDS has instructed us to obtain Size-on-MDS attribute
1269                  * from OSTs and send setattr to back to MDS. */
1270                 rc = ll_som_update(inode, op_data);
1271         } else if (rc) {
1272                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1273                        inode->i_ino, rc);
1274         }
1275         RETURN(rc);
1276 }
1277
1278 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1279 {
1280         struct obd_capa *capa;
1281         int rc;
1282
1283         if (attr->ia_valid & ATTR_SIZE)
1284                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1285         else
1286                 capa = ll_mdscapa_get(inode);
1287
1288         rc = cl_setattr_ost(inode, attr, capa);
1289
1290         if (attr->ia_valid & ATTR_SIZE)
1291                 ll_truncate_free_capa(capa);
1292         else
1293                 capa_put(capa);
1294
1295         return rc;
1296 }
1297
1298 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1299  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1300  * keep these values until such a time that objects are allocated for it.
1301  * We do the MDS operations first, as it is checking permissions for us.
1302  * We don't to the MDS RPC if there is nothing that we want to store there,
1303  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1304  * going to do an RPC anyways.
1305  *
1306  * If we are doing a truncate, we will send the mtime and ctime updates
1307  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1308  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1309  * at the same time.
1310  */
1311 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1312 {
1313         struct inode *inode = dentry->d_inode;
1314         struct ll_inode_info *lli = ll_i2info(inode);
1315         struct lov_stripe_md *lsm = lli->lli_smd;
1316         struct md_op_data *op_data = NULL;
1317         struct md_open_data *mod = NULL;
1318         int ia_valid = attr->ia_valid;
1319         int rc = 0, rc1 = 0;
1320         ENTRY;
1321
1322         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1323                attr->ia_valid);
1324
1325         if (ia_valid & ATTR_SIZE) {
1326                 /* Check new size against VFS/VM file size limit and rlimit */
1327                 rc = inode_newsize_ok(inode, attr->ia_size);
1328                 if (rc)
1329                         RETURN(rc);
1330
1331                 /* The maximum Lustre file size is variable, based on the
1332                  * OST maximum object size and number of stripes.  This
1333                  * needs another check in addition to the VFS check above. */
1334                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1335                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1336                                PFID(&lli->lli_fid), attr->ia_size,
1337                                ll_file_maxbytes(inode));
1338                         RETURN(-EFBIG);
1339                 }
1340
1341                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1342         }
1343
1344         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1345         if (ia_valid & TIMES_SET_FLAGS) {
1346                 if (cfs_curproc_fsuid() != inode->i_uid &&
1347                     !cfs_capable(CFS_CAP_FOWNER))
1348                         RETURN(-EPERM);
1349         }
1350
1351         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1352         if (attr->ia_valid & ATTR_CTIME) {
1353                 attr->ia_ctime = CFS_CURRENT_TIME;
1354                 attr->ia_valid |= ATTR_CTIME_SET;
1355         }
1356         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1357                 attr->ia_atime = CFS_CURRENT_TIME;
1358                 attr->ia_valid |= ATTR_ATIME_SET;
1359         }
1360         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1361                 attr->ia_mtime = CFS_CURRENT_TIME;
1362                 attr->ia_valid |= ATTR_MTIME_SET;
1363         }
1364
1365         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1366                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1367                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1368                        cfs_time_current_sec());
1369
1370         /* NB: ATTR_SIZE will only be set after this point if the size
1371          * resides on the MDS, ie, this file has no objects. */
1372         if (lsm)
1373                 attr->ia_valid &= ~ATTR_SIZE;
1374
1375         /* We always do an MDS RPC, even if we're only changing the size;
1376          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1377
1378         OBD_ALLOC_PTR(op_data);
1379         if (op_data == NULL)
1380                 RETURN(-ENOMEM);
1381
1382         if (!S_ISDIR(inode->i_mode)) {
1383                 if (ia_valid & ATTR_SIZE)
1384                         UP_WRITE_I_ALLOC_SEM(inode);
1385                 UNLOCK_INODE_MUTEX(inode);
1386                 cfs_down_write(&lli->lli_trunc_sem);
1387                 LOCK_INODE_MUTEX(inode);
1388                 if (ia_valid & ATTR_SIZE)
1389                         DOWN_WRITE_I_ALLOC_SEM(inode);
1390         }
1391
1392         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1393
1394         /* Open epoch for truncate. */
1395         if (exp_connect_som(ll_i2mdexp(inode)) &&
1396             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1397                 op_data->op_flags = MF_EPOCH_OPEN;
1398
1399         rc = ll_md_setattr(dentry, op_data, &mod);
1400         if (rc)
1401                 GOTO(out, rc);
1402
1403         ll_ioepoch_open(lli, op_data->op_ioepoch);
1404         if (!lsm || !S_ISREG(inode->i_mode)) {
1405                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1406                 GOTO(out, rc = 0);
1407         }
1408
1409         if (ia_valid & ATTR_SIZE)
1410                 attr->ia_valid |= ATTR_SIZE;
1411         if (ia_valid & (ATTR_SIZE |
1412                         ATTR_ATIME | ATTR_ATIME_SET |
1413                         ATTR_MTIME | ATTR_MTIME_SET))
1414                 /* For truncate and utimes sending attributes to OSTs, setting
1415                  * mtime/atime to the past will be performed under PW [0:EOF]
1416                  * extent lock (new_size:EOF for truncate).  It may seem
1417                  * excessive to send mtime/atime updates to OSTs when not
1418                  * setting times to past, but it is necessary due to possible
1419                  * time de-synchronization between MDT inode and OST objects */
1420                 rc = ll_setattr_ost(inode, attr);
1421         EXIT;
1422 out:
1423         if (op_data) {
1424                 if (op_data->op_ioepoch) {
1425                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1426                         if (!rc)
1427                                 rc = rc1;
1428                 }
1429                 ll_finish_md_op_data(op_data);
1430         }
1431         if (!S_ISDIR(inode->i_mode))
1432                 cfs_up_write(&lli->lli_trunc_sem);
1433
1434         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1435                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1436
1437         return rc;
1438 }
1439
1440 int ll_setattr(struct dentry *de, struct iattr *attr)
1441 {
1442         int mode = de->d_inode->i_mode;
1443
1444         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1445                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1446                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1447
1448         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1449                                (ATTR_SIZE|ATTR_MODE)) &&
1450             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1451              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1452               !(attr->ia_mode & S_ISGID))))
1453                 attr->ia_valid |= ATTR_FORCE;
1454
1455         if ((mode & S_ISUID) &&
1456             !(attr->ia_mode & S_ISUID) &&
1457             !(attr->ia_valid & ATTR_KILL_SUID))
1458                 attr->ia_valid |= ATTR_KILL_SUID;
1459
1460         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1461             !(attr->ia_mode & S_ISGID) &&
1462             !(attr->ia_valid & ATTR_KILL_SGID))
1463                 attr->ia_valid |= ATTR_KILL_SGID;
1464
1465         return ll_setattr_raw(de, attr);
1466 }
1467
1468 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1469                        __u64 max_age, __u32 flags)
1470 {
1471         struct ll_sb_info *sbi = ll_s2sbi(sb);
1472         struct obd_statfs obd_osfs;
1473         int rc;
1474         ENTRY;
1475
1476         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1477         if (rc) {
1478                 CERROR("md_statfs fails: rc = %d\n", rc);
1479                 RETURN(rc);
1480         }
1481
1482         osfs->os_type = sb->s_magic;
1483
1484         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1485                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1486
1487         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1488                 flags |= OBD_STATFS_NODELAY;
1489
1490         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1491                               &obd_osfs, max_age, flags);
1492         if (rc) {
1493                 CERROR("obd_statfs fails: rc = %d\n", rc);
1494                 RETURN(rc);
1495         }
1496
1497         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1498                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1499                obd_osfs.os_files);
1500
1501         osfs->os_bsize = obd_osfs.os_bsize;
1502         osfs->os_blocks = obd_osfs.os_blocks;
1503         osfs->os_bfree = obd_osfs.os_bfree;
1504         osfs->os_bavail = obd_osfs.os_bavail;
1505
1506         /* If we don't have as many objects free on the OST as inodes
1507          * on the MDS, we reduce the total number of inodes to
1508          * compensate, so that the "inodes in use" number is correct.
1509          */
1510         if (obd_osfs.os_ffree < osfs->os_ffree) {
1511                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1512                         obd_osfs.os_ffree;
1513                 osfs->os_ffree = obd_osfs.os_ffree;
1514         }
1515
1516         RETURN(rc);
1517 }
1518 #ifndef HAVE_STATFS_DENTRY_PARAM
1519 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1520 {
1521 #else
1522 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1523 {
1524         struct super_block *sb = de->d_sb;
1525 #endif
1526         struct obd_statfs osfs;
1527         int rc;
1528
1529         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1530         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1531
1532         /* Some amount of caching on the client is allowed */
1533         rc = ll_statfs_internal(sb, &osfs,
1534                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1535                                 0);
1536         if (rc)
1537                 return rc;
1538
1539         statfs_unpack(sfs, &osfs);
1540
1541         /* We need to downshift for all 32-bit kernels, because we can't
1542          * tell if the kernel is being called via sys_statfs64() or not.
1543          * Stop before overflowing f_bsize - in which case it is better
1544          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1545         if (sizeof(long) < 8) {
1546                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1547                         sfs->f_bsize <<= 1;
1548
1549                         osfs.os_blocks >>= 1;
1550                         osfs.os_bfree >>= 1;
1551                         osfs.os_bavail >>= 1;
1552                 }
1553         }
1554
1555         sfs->f_blocks = osfs.os_blocks;
1556         sfs->f_bfree = osfs.os_bfree;
1557         sfs->f_bavail = osfs.os_bavail;
1558
1559         return 0;
1560 }
1561
1562 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1563 {
1564         struct ll_inode_info *lli;
1565         struct lov_stripe_md *lsm;
1566
1567         LASSERT(!S_ISDIR(inode->i_mode));
1568
1569         lli = ll_i2info(inode);
1570         LASSERT(lli->lli_size_sem_owner != current);
1571         cfs_down(&lli->lli_size_sem);
1572         LASSERT(lli->lli_size_sem_owner == NULL);
1573         lli->lli_size_sem_owner = current;
1574         lsm = lli->lli_smd;
1575         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1576                  lsm, lock_lsm);
1577         if (lock_lsm)
1578                 lov_stripe_lock(lsm);
1579 }
1580
1581 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1582 {
1583         struct ll_inode_info *lli;
1584         struct lov_stripe_md *lsm;
1585
1586         lli = ll_i2info(inode);
1587         lsm = lli->lli_smd;
1588         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1589                  lsm, unlock_lsm);
1590         if (unlock_lsm)
1591                 lov_stripe_unlock(lsm);
1592         LASSERT(lli->lli_size_sem_owner == current);
1593         lli->lli_size_sem_owner = NULL;
1594         cfs_up(&lli->lli_size_sem);
1595 }
1596
1597 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1598 {
1599         struct ll_inode_info *lli = ll_i2info(inode);
1600         struct mdt_body *body = md->body;
1601         struct lov_stripe_md *lsm = md->lsm;
1602         struct ll_sb_info *sbi = ll_i2sbi(inode);
1603
1604         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1605         if (lsm != NULL) {
1606                 LASSERT(S_ISREG(inode->i_mode));
1607
1608                 cfs_mutex_lock(&lli->lli_och_mutex);
1609                 if (lli->lli_smd == NULL) {
1610                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1611                             lsm->lsm_magic != LOV_MAGIC_V3) {
1612                                 dump_lsm(D_ERROR, lsm);
1613                                 LBUG();
1614                         }
1615                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1616                                lsm, inode->i_ino, inode->i_generation, inode);
1617                         /* cl_file_inode_init must go before lli_smd or a race
1618                          * is possible where client thinks the file has stripes,
1619                          * but lov raid0 is not setup yet and parallel e.g.
1620                          * glimpse would try to use uninitialized lov */
1621                         cl_file_inode_init(inode, md);
1622                         cfs_spin_lock(&lli->lli_lock);
1623                         lli->lli_smd = lsm;
1624                         cfs_spin_unlock(&lli->lli_lock);
1625                         cfs_mutex_unlock(&lli->lli_och_mutex);
1626                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1627                         if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1628                                 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1629                 } else {
1630                         cfs_mutex_unlock(&lli->lli_och_mutex);
1631                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1632                                 lli->lli_smd->lsm_stripe_count ==
1633                                 lsm->lsm_stripe_count);
1634                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1635                                 CERROR("lsm mismatch for inode %ld\n",
1636                                        inode->i_ino);
1637                                 CERROR("lli_smd:\n");
1638                                 dump_lsm(D_ERROR, lli->lli_smd);
1639                                 CERROR("lsm:\n");
1640                                 dump_lsm(D_ERROR, lsm);
1641                                 LBUG();
1642                         }
1643                 }
1644                 if (lli->lli_smd != lsm)
1645                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1646         }
1647
1648         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1649                 if (body->valid & OBD_MD_FLRMTPERM)
1650                         ll_update_remote_perm(inode, md->remote_perm);
1651         }
1652 #ifdef CONFIG_FS_POSIX_ACL
1653         else if (body->valid & OBD_MD_FLACL) {
1654                 cfs_spin_lock(&lli->lli_lock);
1655                 if (lli->lli_posix_acl)
1656                         posix_acl_release(lli->lli_posix_acl);
1657                 lli->lli_posix_acl = md->posix_acl;
1658                 cfs_spin_unlock(&lli->lli_lock);
1659         }
1660 #endif
1661         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1662         inode->i_generation = cl_fid_build_gen(&body->fid1);
1663
1664         if (body->valid & OBD_MD_FLATIME) {
1665                 if (body->atime > LTIME_S(inode->i_atime))
1666                         LTIME_S(inode->i_atime) = body->atime;
1667                 lli->lli_lvb.lvb_atime = body->atime;
1668         }
1669         if (body->valid & OBD_MD_FLMTIME) {
1670                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1671                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1672                                "to "LPU64"\n", inode->i_ino,
1673                                LTIME_S(inode->i_mtime), body->mtime);
1674                         LTIME_S(inode->i_mtime) = body->mtime;
1675                 }
1676                 lli->lli_lvb.lvb_mtime = body->mtime;
1677         }
1678         if (body->valid & OBD_MD_FLCTIME) {
1679                 if (body->ctime > LTIME_S(inode->i_ctime))
1680                         LTIME_S(inode->i_ctime) = body->ctime;
1681                 lli->lli_lvb.lvb_ctime = body->ctime;
1682         }
1683         if (body->valid & OBD_MD_FLMODE)
1684                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1685         if (body->valid & OBD_MD_FLTYPE)
1686                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1687         LASSERT(inode->i_mode != 0);
1688         if (S_ISREG(inode->i_mode)) {
1689                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1690         } else {
1691                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1692         }
1693 #ifdef HAVE_INODE_BLKSIZE
1694         inode->i_blksize = 1<<inode->i_blkbits;
1695 #endif
1696         if (body->valid & OBD_MD_FLUID)
1697                 inode->i_uid = body->uid;
1698         if (body->valid & OBD_MD_FLGID)
1699                 inode->i_gid = body->gid;
1700         if (body->valid & OBD_MD_FLFLAGS)
1701                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1702         if (body->valid & OBD_MD_FLNLINK)
1703                 inode->i_nlink = body->nlink;
1704         if (body->valid & OBD_MD_FLRDEV)
1705                 inode->i_rdev = old_decode_dev(body->rdev);
1706
1707         if (body->valid & OBD_MD_FLID) {
1708                 /* FID shouldn't be changed! */
1709                 if (fid_is_sane(&lli->lli_fid)) {
1710                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1711                                  "Trying to change FID "DFID
1712                                  " to the "DFID", inode %lu/%u(%p)\n",
1713                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1714                                  inode->i_ino, inode->i_generation, inode);
1715                 } else
1716                         lli->lli_fid = body->fid1;
1717         }
1718
1719         LASSERT(fid_seq(&lli->lli_fid) != 0);
1720
1721         if (body->valid & OBD_MD_FLSIZE) {
1722                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1723                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1724                         struct lustre_handle lockh;
1725                         ldlm_mode_t mode;
1726
1727                         /* As it is possible a blocking ast has been processed
1728                          * by this time, we need to check there is an UPDATE
1729                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1730                          * it. */
1731                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1732                                                &lockh);
1733                         if (mode) {
1734                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1735                                                       LLIF_EPOCH_PENDING |
1736                                                       LLIF_SOM_DIRTY)) {
1737                                         CERROR("ino %lu flags %u still has "
1738                                                "size authority! do not trust "
1739                                                "the size got from MDS\n",
1740                                                inode->i_ino, lli->lli_flags);
1741                                 } else {
1742                                         /* Use old size assignment to avoid
1743                                          * deadlock bz14138 & bz14326 */
1744                                         i_size_write(inode, body->size);
1745                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1746                                 }
1747                                 ldlm_lock_decref(&lockh, mode);
1748                         }
1749                 } else {
1750                         /* Use old size assignment to avoid
1751                          * deadlock bz14138 & bz14326 */
1752                         i_size_write(inode, body->size);
1753
1754                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1755                                inode->i_ino, (unsigned long long)body->size);
1756                 }
1757
1758                 if (body->valid & OBD_MD_FLBLOCKS)
1759                         inode->i_blocks = body->blocks;
1760         }
1761
1762         if (body->valid & OBD_MD_FLMDSCAPA) {
1763                 LASSERT(md->mds_capa);
1764                 ll_add_capa(inode, md->mds_capa);
1765         }
1766         if (body->valid & OBD_MD_FLOSSCAPA) {
1767                 LASSERT(md->oss_capa);
1768                 ll_add_capa(inode, md->oss_capa);
1769         }
1770 }
1771
1772 void ll_read_inode2(struct inode *inode, void *opaque)
1773 {
1774         struct lustre_md *md = opaque;
1775         struct ll_inode_info *lli = ll_i2info(inode);
1776         ENTRY;
1777
1778         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1779                PFID(&lli->lli_fid), inode);
1780
1781         LASSERT(!lli->lli_smd);
1782
1783         /* Core attributes from the MDS first.  This is a new inode, and
1784          * the VFS doesn't zero times in the core inode so we have to do
1785          * it ourselves.  They will be overwritten by either MDS or OST
1786          * attributes - we just need to make sure they aren't newer. */
1787         LTIME_S(inode->i_mtime) = 0;
1788         LTIME_S(inode->i_atime) = 0;
1789         LTIME_S(inode->i_ctime) = 0;
1790         inode->i_rdev = 0;
1791         ll_update_inode(inode, md);
1792
1793         /* OIDEBUG(inode); */
1794
1795         /* initializing backing dev info. */
1796         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1797
1798
1799         if (S_ISREG(inode->i_mode)) {
1800                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1801                 inode->i_op = &ll_file_inode_operations;
1802                 inode->i_fop = sbi->ll_fop;
1803                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1804                 EXIT;
1805         } else if (S_ISDIR(inode->i_mode)) {
1806                 inode->i_op = &ll_dir_inode_operations;
1807                 inode->i_fop = &ll_dir_operations;
1808                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1809                 EXIT;
1810         } else if (S_ISLNK(inode->i_mode)) {
1811                 inode->i_op = &ll_fast_symlink_inode_operations;
1812                 EXIT;
1813         } else {
1814                 inode->i_op = &ll_special_inode_operations;
1815
1816                 init_special_inode(inode, inode->i_mode,
1817                                    kdev_t_to_nr(inode->i_rdev));
1818
1819                 EXIT;
1820         }
1821 }
1822
1823 void ll_delete_inode(struct inode *inode)
1824 {
1825         struct ll_sb_info *sbi = ll_i2sbi(inode);
1826         int rc;
1827         ENTRY;
1828
1829         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1830         if (rc)
1831                 CERROR("fid_delete() failed, rc %d\n", rc);
1832
1833         truncate_inode_pages(&inode->i_data, 0);
1834
1835         /* Workaround for LU-118 */
1836         if (inode->i_data.nrpages) {
1837                 TREE_READ_LOCK_IRQ(&inode->i_data);
1838                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1839                 LASSERTF(inode->i_data.nrpages == 0,
1840                          "inode=%lu/%u(%p) nrpages=%lu, see "
1841                          "http://jira.whamcloud.com/browse/LU-118\n",
1842                          inode->i_ino, inode->i_generation, inode,
1843                          inode->i_data.nrpages);
1844         }
1845         /* Workaround end */
1846
1847 #ifdef HAVE_SBOPS_EVICT_INODE
1848         ll_clear_inode(inode);
1849         end_writeback(inode);
1850 #else
1851         clear_inode(inode);
1852 #endif
1853
1854         EXIT;
1855 }
1856
1857 int ll_iocontrol(struct inode *inode, struct file *file,
1858                  unsigned int cmd, unsigned long arg)
1859 {
1860         struct ll_sb_info *sbi = ll_i2sbi(inode);
1861         struct ptlrpc_request *req = NULL;
1862         int rc, flags = 0;
1863         ENTRY;
1864
1865         switch(cmd) {
1866         case FSFILT_IOC_GETFLAGS: {
1867                 struct mdt_body *body;
1868                 struct md_op_data *op_data;
1869
1870                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1871                                              0, 0, LUSTRE_OPC_ANY,
1872                                              NULL);
1873                 if (IS_ERR(op_data))
1874                         RETURN(PTR_ERR(op_data));
1875
1876                 op_data->op_valid = OBD_MD_FLFLAGS;
1877                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1878                 ll_finish_md_op_data(op_data);
1879                 if (rc) {
1880                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1881                         RETURN(-abs(rc));
1882                 }
1883
1884                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1885
1886                 flags = body->flags;
1887
1888                 ptlrpc_req_finished(req);
1889
1890                 RETURN(put_user(flags, (int *)arg));
1891         }
1892         case FSFILT_IOC_SETFLAGS: {
1893                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1894                 struct obd_info oinfo = { { { 0 } } };
1895                 struct md_op_data *op_data;
1896
1897                 if (get_user(flags, (int *)arg))
1898                         RETURN(-EFAULT);
1899
1900                 oinfo.oi_md = lsm;
1901                 OBDO_ALLOC(oinfo.oi_oa);
1902                 if (!oinfo.oi_oa)
1903                         RETURN(-ENOMEM);
1904
1905                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1906                                              LUSTRE_OPC_ANY, NULL);
1907                 if (IS_ERR(op_data))
1908                         RETURN(PTR_ERR(op_data));
1909
1910                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1911                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1912                 rc = md_setattr(sbi->ll_md_exp, op_data,
1913                                 NULL, 0, NULL, 0, &req, NULL);
1914                 ll_finish_md_op_data(op_data);
1915                 ptlrpc_req_finished(req);
1916                 if (rc) {
1917                         OBDO_FREE(oinfo.oi_oa);
1918                         RETURN(rc);
1919                 }
1920
1921                 if (lsm == NULL) {
1922                         OBDO_FREE(oinfo.oi_oa);
1923                         GOTO(update_cache, rc);
1924                 }
1925
1926                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1927                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1928                 oinfo.oi_oa->o_flags = flags;
1929                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1930                                        OBD_MD_FLGROUP;
1931                 oinfo.oi_capa = ll_mdscapa_get(inode);
1932                 obdo_from_inode(oinfo.oi_oa, inode,
1933                                 &ll_i2info(inode)->lli_fid, 0);
1934                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1935                 capa_put(oinfo.oi_capa);
1936                 OBDO_FREE(oinfo.oi_oa);
1937                 if (rc) {
1938                         if (rc != -EPERM && rc != -EACCES)
1939                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1940                         RETURN(rc);
1941                 }
1942
1943                 EXIT;
1944 update_cache:
1945                 inode->i_flags = ll_ext_to_inode_flags(flags);
1946                 return 0;
1947         }
1948         default:
1949                 RETURN(-ENOSYS);
1950         }
1951
1952         RETURN(0);
1953 }
1954
1955 int ll_flush_ctx(struct inode *inode)
1956 {
1957         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1958
1959         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1960
1961         obd_set_info_async(sbi->ll_md_exp,
1962                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1963                            0, NULL, NULL);
1964         obd_set_info_async(sbi->ll_dt_exp,
1965                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1966                            0, NULL, NULL);
1967         return 0;
1968 }
1969
1970 /* umount -f client means force down, don't save state */
1971 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1972 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1973 {
1974         struct super_block *sb = vfsmnt->mnt_sb;
1975 #else
1976 void ll_umount_begin(struct super_block *sb)
1977 {
1978 #endif
1979         struct lustre_sb_info *lsi = s2lsi(sb);
1980         struct ll_sb_info *sbi = ll_s2sbi(sb);
1981         struct obd_device *obd;
1982         struct obd_ioctl_data *ioc_data;
1983         ENTRY;
1984
1985 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1986         if (!(flags & MNT_FORCE)) {
1987                 EXIT;
1988                 return;
1989         }
1990 #endif
1991
1992         /* Tell the MGC we got umount -f */
1993         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1994
1995         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1996                sb->s_count, atomic_read(&sb->s_active));
1997
1998         obd = class_exp2obd(sbi->ll_md_exp);
1999         if (obd == NULL) {
2000                 CERROR("Invalid MDC connection handle "LPX64"\n",
2001                        sbi->ll_md_exp->exp_handle.h_cookie);
2002                 EXIT;
2003                 return;
2004         }
2005         obd->obd_force = 1;
2006
2007         obd = class_exp2obd(sbi->ll_dt_exp);
2008         if (obd == NULL) {
2009                 CERROR("Invalid LOV connection handle "LPX64"\n",
2010                        sbi->ll_dt_exp->exp_handle.h_cookie);
2011                 EXIT;
2012                 return;
2013         }
2014         obd->obd_force = 1;
2015
2016         OBD_ALLOC_PTR(ioc_data);
2017         if (ioc_data) {
2018                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2019                               sizeof ioc_data, ioc_data, NULL);
2020
2021                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2022                               sizeof ioc_data, ioc_data, NULL);
2023
2024                 OBD_FREE_PTR(ioc_data);
2025         }
2026
2027
2028         /* Really, we'd like to wait until there are no requests outstanding,
2029          * and then continue.  For now, we just invalidate the requests,
2030          * schedule() and sleep one second if needed, and hope.
2031          */
2032         cfs_schedule();
2033 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2034         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2035                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2036                                                    cfs_time_seconds(1));
2037                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2038                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2039                                       "may try to umount it a bit later\n",
2040                                       atomic_read(&vfsmnt->mnt_count));
2041         }
2042 #endif
2043
2044         EXIT;
2045 }
2046
2047 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2048 {
2049         struct ll_sb_info *sbi = ll_s2sbi(sb);
2050         int err;
2051         __u32 read_only;
2052
2053         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2054                 read_only = *flags & MS_RDONLY;
2055                 err = obd_set_info_async(sbi->ll_md_exp,
2056                                          sizeof(KEY_READ_ONLY),
2057                                          KEY_READ_ONLY, sizeof(read_only),
2058                                          &read_only, NULL);
2059                 if (err) {
2060                         CERROR("Failed to change the read-only flag during "
2061                                "remount: %d\n", err);
2062                         return err;
2063                 }
2064
2065                 if (read_only)
2066                         sb->s_flags |= MS_RDONLY;
2067                 else
2068                         sb->s_flags &= ~MS_RDONLY;
2069         }
2070         return 0;
2071 }
2072
2073 int ll_prep_inode(struct inode **inode,
2074                   struct ptlrpc_request *req,
2075                   struct super_block *sb)
2076 {
2077         struct ll_sb_info *sbi = NULL;
2078         struct lustre_md md;
2079         int rc;
2080         ENTRY;
2081
2082         LASSERT(*inode || sb);
2083         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2084         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2085                               sbi->ll_md_exp, &md);
2086         if (rc)
2087                 RETURN(rc);
2088
2089         if (*inode) {
2090                 ll_update_inode(*inode, &md);
2091         } else {
2092                 LASSERT(sb != NULL);
2093
2094                 /*
2095                  * At this point server returns to client's same fid as client
2096                  * generated for creating. So using ->fid1 is okay here.
2097                  */
2098                 LASSERT(fid_is_sane(&md.body->fid1));
2099
2100                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2101                 if (*inode == NULL || IS_ERR(*inode)) {
2102                         if (md.lsm)
2103                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2104 #ifdef CONFIG_FS_POSIX_ACL
2105                         if (md.posix_acl) {
2106                                 posix_acl_release(md.posix_acl);
2107                                 md.posix_acl = NULL;
2108                         }
2109 #endif
2110                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2111                         *inode = NULL;
2112                         CERROR("new_inode -fatal: rc %d\n", rc);
2113                         GOTO(out, rc);
2114                 }
2115         }
2116
2117 out:
2118         md_free_lustre_md(sbi->ll_md_exp, &md);
2119         RETURN(rc);
2120 }
2121
2122 int ll_obd_statfs(struct inode *inode, void *arg)
2123 {
2124         struct ll_sb_info *sbi = NULL;
2125         struct obd_export *exp;
2126         char *buf = NULL;
2127         struct obd_ioctl_data *data = NULL;
2128         __u32 type;
2129         int len = 0, rc;
2130
2131         if (!inode || !(sbi = ll_i2sbi(inode)))
2132                 GOTO(out_statfs, rc = -EINVAL);
2133
2134         rc = obd_ioctl_getdata(&buf, &len, arg);
2135         if (rc)
2136                 GOTO(out_statfs, rc);
2137
2138         data = (void*)buf;
2139         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2140             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2141                 GOTO(out_statfs, rc = -EINVAL);
2142
2143         if (data->ioc_inllen1 != sizeof(__u32) ||
2144             data->ioc_inllen2 != sizeof(__u32) ||
2145             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2146             data->ioc_plen2 != sizeof(struct obd_uuid))
2147                 GOTO(out_statfs, rc = -EINVAL);
2148
2149         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2150         if (type == LL_STATFS_LMV)
2151                 exp = sbi->ll_md_exp;
2152         else if (type == LL_STATFS_LOV)
2153                 exp = sbi->ll_dt_exp;
2154         else
2155                 GOTO(out_statfs, rc = -ENODEV);
2156
2157         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2158         if (rc)
2159                 GOTO(out_statfs, rc);
2160 out_statfs:
2161         if (buf)
2162                 obd_ioctl_freedata(buf, len);
2163         return rc;
2164 }
2165
2166 int ll_process_config(struct lustre_cfg *lcfg)
2167 {
2168         char *ptr;
2169         void *sb;
2170         struct lprocfs_static_vars lvars;
2171         unsigned long x;
2172         int rc = 0;
2173
2174         lprocfs_llite_init_vars(&lvars);
2175
2176         /* The instance name contains the sb: lustre-client-aacfe000 */
2177         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2178         if (!ptr || !*(++ptr))
2179                 return -EINVAL;
2180         if (sscanf(ptr, "%lx", &x) != 1)
2181                 return -EINVAL;
2182         sb = (void *)x;
2183         /* This better be a real Lustre superblock! */
2184         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2185
2186         /* Note we have not called client_common_fill_super yet, so
2187            proc fns must be able to handle that! */
2188         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2189                                       lcfg, sb);
2190         if (rc > 0)
2191                 rc = 0;
2192         return(rc);
2193 }
2194
2195 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2196 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2197                                        struct inode *i1, struct inode *i2,
2198                                        const char *name, int namelen,
2199                                        int mode, __u32 opc, void *data)
2200 {
2201         LASSERT(i1 != NULL);
2202
2203         if (namelen > ll_i2sbi(i1)->ll_namelen)
2204                 return ERR_PTR(-ENAMETOOLONG);
2205
2206         if (op_data == NULL)
2207                 OBD_ALLOC_PTR(op_data);
2208
2209         if (op_data == NULL)
2210                 return ERR_PTR(-ENOMEM);
2211
2212         ll_i2gids(op_data->op_suppgids, i1, i2);
2213         op_data->op_fid1 = *ll_inode2fid(i1);
2214         op_data->op_capa1 = ll_mdscapa_get(i1);
2215
2216         if (i2) {
2217                 op_data->op_fid2 = *ll_inode2fid(i2);
2218                 op_data->op_capa2 = ll_mdscapa_get(i2);
2219         } else {
2220                 fid_zero(&op_data->op_fid2);
2221                 op_data->op_capa2 = NULL;
2222         }
2223
2224         op_data->op_name = name;
2225         op_data->op_namelen = namelen;
2226         op_data->op_mode = mode;
2227         op_data->op_mod_time = cfs_time_current_sec();
2228         op_data->op_fsuid = cfs_curproc_fsuid();
2229         op_data->op_fsgid = cfs_curproc_fsgid();
2230         op_data->op_cap = cfs_curproc_cap_pack();
2231         op_data->op_bias = MDS_CHECK_SPLIT;
2232         op_data->op_opc = opc;
2233         op_data->op_mds = 0;
2234         op_data->op_data = data;
2235
2236         /* If the file is being opened after mknod() (normally due to NFS)
2237          * try to use the default stripe data from parent directory for
2238          * allocating OST objects.  Try to pass the parent FID to MDS. */
2239         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2240             ll_i2info(i2)->lli_smd == NULL) {
2241                 struct ll_inode_info *lli = ll_i2info(i2);
2242
2243                 cfs_spin_lock(&lli->lli_lock);
2244                 if (likely(lli->lli_smd == NULL &&
2245                            !fid_is_zero(&lli->lli_pfid)))
2246                         op_data->op_fid1 = lli->lli_pfid;
2247                 cfs_spin_unlock(&lli->lli_lock);
2248                 /** We ignore parent's capability temporary. */
2249         }
2250
2251         return op_data;
2252 }
2253
2254 void ll_finish_md_op_data(struct md_op_data *op_data)
2255 {
2256         capa_put(op_data->op_capa1);
2257         capa_put(op_data->op_capa2);
2258         OBD_FREE_PTR(op_data);
2259 }
2260
2261 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2262 {
2263         struct ll_sb_info *sbi;
2264
2265         LASSERT((seq != NULL) && (vfs != NULL));
2266         sbi = ll_s2sbi(vfs->mnt_sb);
2267
2268         if (sbi->ll_flags & LL_SBI_NOLCK)
2269                 seq_puts(seq, ",nolock");
2270
2271         if (sbi->ll_flags & LL_SBI_FLOCK)
2272                 seq_puts(seq, ",flock");
2273
2274         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2275                 seq_puts(seq, ",localflock");
2276
2277         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2278                 seq_puts(seq, ",user_xattr");
2279
2280         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2281                 seq_puts(seq, ",lazystatfs");
2282
2283         RETURN(0);
2284 }
2285
2286 /**
2287  * Get obd name by cmd, and copy out to user space
2288  */
2289 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2290 {
2291         struct ll_sb_info *sbi = ll_i2sbi(inode);
2292         struct obd_device *obd;
2293         ENTRY;
2294
2295         if (cmd == OBD_IOC_GETDTNAME)
2296                 obd = class_exp2obd(sbi->ll_dt_exp);
2297         else if (cmd == OBD_IOC_GETMDNAME)
2298                 obd = class_exp2obd(sbi->ll_md_exp);
2299         else
2300                 RETURN(-EINVAL);
2301
2302         if (!obd)
2303                 RETURN(-ENOENT);
2304
2305         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2306                              strlen(obd->obd_name) + 1))
2307                 RETURN(-EFAULT);
2308
2309         RETURN(0);
2310 }