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         sbi->ll_flags |= LL_SBI_VERBOSE;
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_r_hist.oh_lock);
 137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 138                                    pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145         cfs_atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147
 148         RETURN(sbi);
 149 }
 150
 151 void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 cfs_spin_lock(&ll_sb_lock);
 158                 cfs_list_del(&sbi->ll_list);
 159                 cfs_spin_unlock(&ll_sb_lock);
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static struct dentry_operations ll_d_root_ops = {
 166         .d_compare = ll_dcompare,
 167         .d_revalidate = ll_revalidate_nd,
 168 };
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = 0;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         if (proc_lustre_fs_root) {
 204                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 205                                                   dt, md);
 206                 if (err < 0)
 207                         CERROR("could not register mount in /proc/fs/lustre\n");
 208         }
 209
 210         /* indicate the features supported by this client */
 211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 212                                   OBD_CONNECT_ATTRFID  |
 213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 214                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 215                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 216                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 217                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 218                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 219                                   OBD_CONNECT_EINPROGRESS;
 220
 221         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 222                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 230 #endif
 231         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 232         data->ocd_version = LUSTRE_VERSION_CODE;
 233
 234         if (sb->s_flags & MS_RDONLY)
 235                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 236         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 237                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 238
 239 #ifdef HAVE_MS_FLOCK_LOCK
 240         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 241         sb->s_flags |= MS_FLOCK_LOCK;
 242 #endif
 243 #ifdef MS_HAS_NEW_AOPS
 244         sb->s_flags |= MS_HAS_NEW_AOPS;
 245 #endif
 246
 247         if (sbi->ll_flags & LL_SBI_FLOCK)
 248                 sbi->ll_fop = &ll_file_operations_flock;
 249         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 250                 sbi->ll_fop = &ll_file_operations;
 251         else
 252                 sbi->ll_fop = &ll_file_operations_noflock;
 253
 254         /* real client */
 255         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 256         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 257                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 258
 259         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 260
 261         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 262         if (err == -EBUSY) {
 263                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 264                                    "recovery, of which this client is not a "
 265                                    "part. Please wait for recovery to complete,"
 266                                    " abort, or time out.\n", md);
 267                 GOTO(out, err);
 268         } else if (err) {
 269                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 270                 GOTO(out, err);
 271         }
 272
 273         err = obd_fid_init(sbi->ll_md_exp);
 274         if (err) {
 275                 CERROR("Can't init metadata layer FID infrastructure, "
 276                        "rc %d\n", err);
 277                 GOTO(out_md, err);
 278         }
 279
 280         err = obd_statfs(obd, osfs,
 281                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 282         if (err)
 283                 GOTO(out_md_fid, err);
 284
 285         /* This needs to be after statfs to ensure connect has finished.
 286          * Note that "data" does NOT contain the valid connect reply.
 287          * If connecting to a 1.8 server there will be no LMV device, so
 288          * we can access the MDC export directly and exp_connect_flags will
 289          * be non-zero, but if accessing an upgraded 2.1 server it will
 290          * have the correct flags filled in.
 291          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 292         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 293         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 294             valid != CLIENT_CONNECT_MDT_REQD) {
 295                 char *buf;
 296
 297                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 298                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 299                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 300                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 301                                    "feature(s) needed for correct operation "
 302                                    "of this client (%s). Please upgrade "
 303                                    "server or downgrade client.\n",
 304                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 305                 OBD_FREE(buf, CFS_PAGE_SIZE);
 306                 GOTO(out_md, err = -EPROTO);
 307         }
 308
 309         size = sizeof(*data);
 310         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 311                            KEY_CONN_DATA,  &size, data, NULL);
 312         if (err) {
 313                 CERROR("Get connect data failed: %d \n", err);
 314                 GOTO(out_md, err);
 315         }
 316
 317         LASSERT(osfs->os_bsize);
 318         sb->s_blocksize = osfs->os_bsize;
 319         sb->s_blocksize_bits = log2(osfs->os_bsize);
 320         sb->s_magic = LL_SUPER_MAGIC;
 321         sb->s_maxbytes = MAX_LFS_FILESIZE;
 322         sbi->ll_namelen = osfs->os_namelen;
 323         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 324
 325         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 326             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 327                 LCONSOLE_INFO("Disabling user_xattr feature because "
 328                               "it is not supported on the server\n");
 329                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 330         }
 331
 332         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 333 #ifdef MS_POSIXACL
 334                 sb->s_flags |= MS_POSIXACL;
 335 #endif
 336                 sbi->ll_flags |= LL_SBI_ACL;
 337         } else {
 338                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags &= ~MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags &= ~LL_SBI_ACL;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 346                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 347                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 348                         LCONSOLE_INFO("client is set as remote by default.\n");
 349                 }
 350         } else {
 351                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 352                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 353                         LCONSOLE_INFO("client claims to be remote, but server "
 354                                       "rejected, forced to be local.\n");
 355                 }
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 359                 LCONSOLE_INFO("client enabled MDS capability!\n");
 360                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 364                 LCONSOLE_INFO("client enabled OSS capability!\n");
 365                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 369                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 372                 sbi->ll_md_brw_size = data->ocd_brw_size;
 373         else
 374                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 375
 376         obd = class_name2obd(dt);
 377         if (!obd) {
 378                 CERROR("DT %s: not setup or attached\n", dt);
 379                 GOTO(out_md_fid, err = -ENODEV);
 380         }
 381
 382         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 383                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 384                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 385                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 386                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 387                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 388                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 389                                   OBD_CONNECT_MAXBYTES |
 390                                   OBD_CONNECT_EINPROGRESS;
 391
 392         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 393                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 394
 395         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 396                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 397                  * disabled by default, because it can still be enabled on the
 398                  * fly via /proc. As a consequence, we still need to come to an
 399                  * agreement on the supported algorithms at connect time */
 400                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 401
 402                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 403                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 404                 else
 405                         data->ocd_cksum_types = cksum_types_supported();
 406         }
 407
 408 #ifdef HAVE_LRU_RESIZE_SUPPORT
 409         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 410 #endif
 411         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 412                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 413
 414         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 415                "ocd_grant: %d\n", data->ocd_connect_flags,
 416                data->ocd_version, data->ocd_grant);
 417
 418         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 419         obd->obd_upcall.onu_upcall = cl_ocd_update;
 420
 421         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 422
 423         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 424         if (err == -EBUSY) {
 425                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 426                                    "recovery, of which this client is not a "
 427                                    "part.  Please wait for recovery to "
 428                                    "complete, abort, or time out.\n", dt);
 429                 GOTO(out_md_fid, err);
 430         } else if (err) {
 431                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 432                 GOTO(out_md_fid, err);
 433         }
 434
 435         err = obd_fid_init(sbi->ll_dt_exp);
 436         if (err) {
 437                 CERROR("Can't init data layer FID infrastructure, "
 438                        "rc %d\n", err);
 439                 GOTO(out_dt, err);
 440         }
 441
 442         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 443         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 444         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 445         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 446         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 447
 448         fid_zero(&sbi->ll_root_fid);
 449         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 450         if (err) {
 451                 CERROR("cannot mds_connect: rc = %d\n", err);
 452                 GOTO(out_lock_cn_cb, err);
 453         }
 454         if (!fid_is_sane(&sbi->ll_root_fid)) {
 455                 CERROR("Invalid root fid during mount\n");
 456                 GOTO(out_lock_cn_cb, err = -EINVAL);
 457         }
 458         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 459
 460         sb->s_op = &lustre_super_operations;
 461 #if THREAD_SIZE >= 8192
 462         sb->s_export_op = &lustre_export_operations;
 463 #endif
 464
 465         /* make root inode
 466          * XXX: move this to after cbd setup? */
 467         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 468         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 469                 valid |= OBD_MD_FLRMTPERM;
 470         else if (sbi->ll_flags & LL_SBI_ACL)
 471                 valid |= OBD_MD_FLACL;
 472
 473         OBD_ALLOC_PTR(op_data);
 474         if (op_data == NULL)
 475                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 476
 477         op_data->op_fid1 = sbi->ll_root_fid;
 478         op_data->op_mode = 0;
 479         op_data->op_capa1 = oc;
 480         op_data->op_valid = valid;
 481
 482         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 483         if (oc)
 484                 capa_put(oc);
 485         OBD_FREE_PTR(op_data);
 486         if (err) {
 487                 CERROR("md_getattr failed for root: rc = %d\n", err);
 488                 GOTO(out_lock_cn_cb, err);
 489         }
 490         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 491                                sbi->ll_md_exp, &lmd);
 492         if (err) {
 493                 CERROR("failed to understand root inode md: rc = %d\n", err);
 494                 ptlrpc_req_finished (request);
 495                 GOTO(out_lock_cn_cb, err);
 496         }
 497
 498         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 499         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 500         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 501         ptlrpc_req_finished(request);
 502
 503         if (root == NULL || IS_ERR(root)) {
 504                 if (lmd.lsm)
 505                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 506 #ifdef CONFIG_FS_POSIX_ACL
 507                 if (lmd.posix_acl) {
 508                         posix_acl_release(lmd.posix_acl);
 509                         lmd.posix_acl = NULL;
 510                 }
 511 #endif
 512                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 513                 root = NULL;
 514                 CERROR("lustre_lite: bad iget4 for root\n");
 515                 GOTO(out_root, err);
 516         }
 517
 518         err = ll_close_thread_start(&sbi->ll_lcq);
 519         if (err) {
 520                 CERROR("cannot start close thread: rc %d\n", err);
 521                 GOTO(out_root, err);
 522         }
 523
 524 #ifdef CONFIG_FS_POSIX_ACL
 525         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 526                 rct_init(&sbi->ll_rct);
 527                 et_init(&sbi->ll_et);
 528         }
 529 #endif
 530
 531         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 532         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 533                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 534                                  NULL);
 535         cl_sb_init(sb);
 536
 537         sb->s_root = d_alloc_root(root);
 538         sb->s_root->d_op = &ll_d_root_ops;
 539
 540         sbi->ll_sdev_orig = sb->s_dev;
 541
 542         /* We set sb->s_dev equal on all lustre clients in order to support
 543          * NFS export clustering.  NFSD requires that the FSID be the same
 544          * on all clients. */
 545         /* s_dev is also used in lt_compare() to compare two fs, but that is
 546          * only a node-local comparison. */
 547         uuid = obd_get_uuid(sbi->ll_md_exp);
 548         if (uuid != NULL)
 549                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 550         sbi->ll_mnt = mnt;
 551
 552         if (data != NULL)
 553                 OBD_FREE_PTR(data);
 554         if (osfs != NULL)
 555                 OBD_FREE_PTR(osfs);
 556
 557         RETURN(err);
 558 out_root:
 559         if (root)
 560                 iput(root);
 561 out_lock_cn_cb:
 562         obd_fid_fini(sbi->ll_dt_exp);
 563 out_dt:
 564         obd_disconnect(sbi->ll_dt_exp);
 565         sbi->ll_dt_exp = NULL;
 566 out_md_fid:
 567         obd_fid_fini(sbi->ll_md_exp);
 568 out_md:
 569         obd_disconnect(sbi->ll_md_exp);
 570         sbi->ll_md_exp = NULL;
 571 out:
 572         if (data != NULL)
 573                 OBD_FREE_PTR(data);
 574         if (osfs != NULL)
 575                 OBD_FREE_PTR(osfs);
 576         lprocfs_unregister_mountpoint(sbi);
 577         return err;
 578 }
 579
 580 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 581 {
 582         int size, rc;
 583
 584         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 585         size = sizeof(int);
 586         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 587                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 588         if (rc)
 589                 CERROR("Get max mdsize error rc %d \n", rc);
 590
 591         RETURN(rc);
 592 }
 593
 594 void ll_dump_inode(struct inode *inode)
 595 {
 596         struct list_head *tmp;
 597         int dentry_count = 0;
 598
 599         LASSERT(inode != NULL);
 600
 601         list_for_each(tmp, &inode->i_dentry)
 602                 dentry_count++;
 603
 604         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 605                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 606                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 607 }
 608
 609 void lustre_dump_dentry(struct dentry *dentry, int recur)
 610 {
 611         struct list_head *tmp;
 612         int subdirs = 0;
 613
 614         LASSERT(dentry != NULL);
 615
 616         list_for_each(tmp, &dentry->d_subdirs)
 617                 subdirs++;
 618
 619         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 620                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 621                dentry->d_name.len, dentry->d_name.name,
 622                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 623                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 624                dentry->d_flags, dentry->d_fsdata, subdirs);
 625         if (dentry->d_inode != NULL)
 626                 ll_dump_inode(dentry->d_inode);
 627
 628         if (recur == 0)
 629                 return;
 630
 631        list_for_each(tmp, &dentry->d_subdirs) {
 632                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 633                 lustre_dump_dentry(d, recur - 1);
 634         }
 635 }
 636
 637 void client_common_put_super(struct super_block *sb)
 638 {
 639         struct ll_sb_info *sbi = ll_s2sbi(sb);
 640         ENTRY;
 641
 642 #ifdef CONFIG_FS_POSIX_ACL
 643         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 644                 et_fini(&sbi->ll_et);
 645                 rct_fini(&sbi->ll_rct);
 646         }
 647 #endif
 648
 649         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 650
 651         ll_close_thread_shutdown(sbi->ll_lcq);
 652
 653         cl_sb_fini(sb);
 654
 655         cfs_list_del(&sbi->ll_conn_chain);
 656
 657         obd_fid_fini(sbi->ll_dt_exp);
 658         obd_disconnect(sbi->ll_dt_exp);
 659         sbi->ll_dt_exp = NULL;
 660
 661         lprocfs_unregister_mountpoint(sbi);
 662
 663         obd_fid_fini(sbi->ll_md_exp);
 664         obd_disconnect(sbi->ll_md_exp);
 665         sbi->ll_md_exp = NULL;
 666
 667         EXIT;
 668 }
 669
 670 void ll_kill_super(struct super_block *sb)
 671 {
 672         struct ll_sb_info *sbi;
 673
 674         ENTRY;
 675
 676         /* not init sb ?*/
 677         if (!(sb->s_flags & MS_ACTIVE))
 678                 return;
 679
 680         sbi = ll_s2sbi(sb);
 681         /* we need restore s_dev from changed for clustred NFS before put_super
 682          * because new kernels have cached s_dev and change sb->s_dev in
 683          * put_super not affected real removing devices */
 684         if (sbi)
 685                 sb->s_dev = sbi->ll_sdev_orig;
 686         EXIT;
 687 }
 688
 689 char *ll_read_opt(const char *opt, char *data)
 690 {
 691         char *value;
 692         char *retval;
 693         ENTRY;
 694
 695         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 696         if (strncmp(opt, data, strlen(opt)))
 697                 RETURN(NULL);
 698         if ((value = strchr(data, '=')) == NULL)
 699                 RETURN(NULL);
 700
 701         value++;
 702         OBD_ALLOC(retval, strlen(value) + 1);
 703         if (!retval) {
 704                 CERROR("out of memory!\n");
 705                 RETURN(NULL);
 706         }
 707
 708         memcpy(retval, value, strlen(value)+1);
 709         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 710         RETURN(retval);
 711 }
 712
 713 static inline int ll_set_opt(const char *opt, char *data, int fl)
 714 {
 715         if (strncmp(opt, data, strlen(opt)) != 0)
 716                 return(0);
 717         else
 718                 return(fl);
 719 }
 720
 721 /* non-client-specific mount options are parsed in lmd_parse */
 722 static int ll_options(char *options, int *flags)
 723 {
 724         int tmp;
 725         char *s1 = options, *s2;
 726         ENTRY;
 727
 728         if (!options)
 729                 RETURN(0);
 730
 731         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 732
 733         while (*s1) {
 734                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 735                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 736                 if (tmp) {
 737                         *flags |= tmp;
 738                         goto next;
 739                 }
 740                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 741                 if (tmp) {
 742                         *flags |= tmp;
 743                         goto next;
 744                 }
 745                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 746                 if (tmp) {
 747                         *flags |= tmp;
 748                         goto next;
 749                 }
 750                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 751                 if (tmp) {
 752                         *flags &= ~tmp;
 753                         goto next;
 754                 }
 755                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 756                 if (tmp) {
 757                         *flags |= tmp;
 758                         goto next;
 759                 }
 760                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 761                 if (tmp) {
 762                         *flags &= ~tmp;
 763                         goto next;
 764                 }
 765 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 766                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 767                 if (tmp) {
 768                         /* Ignore deprecated mount option.  The client will
 769                          * always try to mount with ACL support, whether this
 770                          * is used depends on whether server supports it. */
 771                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 772                                                   "mount option 'acl'.\n");
 773                         goto next;
 774                 }
 775                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 776                 if (tmp) {
 777                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 778                                                   "mount option 'noacl'.\n");
 779                         goto next;
 780                 }
 781 #else
 782 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 783 #endif
 784                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 785                 if (tmp) {
 786                         *flags |= tmp;
 787                         goto next;
 788                 }
 789
 790                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 791                 if (tmp) {
 792                         *flags |= tmp;
 793                         goto next;
 794                 }
 795                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 796                 if (tmp) {
 797                         *flags &= ~tmp;
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 806                 if (tmp) {
 807                         *flags &= ~tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 816                 if (tmp) {
 817                         *flags &= ~tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 821                 if (tmp) {
 822                         *flags |= tmp;
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 826                 if (tmp) {
 827                         *flags |= tmp;
 828                         goto next;
 829                 }
 830                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 836                 if (tmp) {
 837                         *flags &= ~tmp;
 838                         goto next;
 839                 }
 840                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 841                                    s1);
 842                 RETURN(-EINVAL);
 843
 844 next:
 845                 /* Find next opt */
 846                 s2 = strchr(s1, ',');
 847                 if (s2 == NULL)
 848                         break;
 849                 s1 = s2 + 1;
 850         }
 851         RETURN(0);
 852 }
 853
 854 void ll_lli_init(struct ll_inode_info *lli)
 855 {
 856         lli->lli_inode_magic = LLI_INODE_MAGIC;
 857         lli->lli_flags = 0;
 858         lli->lli_ioepoch = 0;
 859         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 860         cfs_spin_lock_init(&lli->lli_lock);
 861         lli->lli_posix_acl = NULL;
 862         lli->lli_remote_perms = NULL;
 863         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 864         /* Do not set lli_fid, it has been initialized already. */
 865         fid_zero(&lli->lli_pfid);
 866         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 867         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 868         cfs_atomic_set(&lli->lli_open_count, 0);
 869         lli->lli_mds_capa = NULL;
 870         lli->lli_rmtperm_time = 0;
 871         lli->lli_pending_och = NULL;
 872         lli->lli_mds_read_och = NULL;
 873         lli->lli_mds_write_och = NULL;
 874         lli->lli_mds_exec_och = NULL;
 875         lli->lli_open_fd_read_count = 0;
 876         lli->lli_open_fd_write_count = 0;
 877         lli->lli_open_fd_exec_count = 0;
 878         cfs_mutex_init(&lli->lli_och_mutex);
 879         cfs_spin_lock_init(&lli->lli_agl_lock);
 880         lli->lli_smd = NULL;
 881         lli->lli_clob = NULL;
 882
 883         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 884         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 885                 cfs_mutex_init(&lli->lli_readdir_mutex);
 886                 lli->lli_opendir_key = NULL;
 887                 lli->lli_sai = NULL;
 888                 lli->lli_sa_pos = 0;
 889                 lli->lli_def_acl = NULL;
 890                 cfs_spin_lock_init(&lli->lli_sa_lock);
 891                 lli->lli_opendir_pid = 0;
 892         } else {
 893                 cfs_sema_init(&lli->lli_size_sem, 1);
 894                 lli->lli_size_sem_owner = NULL;
 895                 lli->lli_symlink_name = NULL;
 896                 cfs_init_rwsem(&lli->lli_trunc_sem);
 897                 cfs_mutex_init(&lli->lli_write_mutex);
 898                 lli->lli_async_rc = 0;
 899                 lli->lli_write_rc = 0;
 900                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 901                 lli->lli_glimpse_time = 0;
 902                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 903                 lli->lli_agl_index = 0;
 904         }
 905 }
 906
 907 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 908 {
 909 #ifdef HAVE_BDI_REGISTER
 910         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 911
 912 #ifdef HAVE_BDI_NAME
 913         bdi->name = "lustre";
 914 #endif
 915         return bdi_register(bdi, NULL, "lustre-%d",
 916                             atomic_inc_return(&ll_bdi_num));
 917 #else
 918         return 0;
 919 #endif
 920 }
 921
 922 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 923 {
 924         struct lustre_profile *lprof = NULL;
 925         struct lustre_sb_info *lsi = s2lsi(sb);
 926         struct ll_sb_info *sbi;
 927         char  *dt = NULL, *md = NULL;
 928         char  *profilenm = get_profile_name(sb);
 929         struct config_llog_instance *cfg;
 930         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 931         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 932         int    err;
 933         ENTRY;
 934
 935         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 936
 937         OBD_ALLOC_PTR(cfg);
 938         if (cfg == NULL)
 939                 RETURN(-ENOMEM);
 940
 941         cfs_module_get();
 942
 943         /* client additional sb info */
 944         lsi->lsi_llsbi = sbi = ll_init_sbi();
 945         if (!sbi) {
 946                 cfs_module_put(THIS_MODULE);
 947                 OBD_FREE_PTR(cfg);
 948                 RETURN(-ENOMEM);
 949         }
 950
 951         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 952         if (err)
 953                 GOTO(out_free, err);
 954
 955         err = ll_bdi_init(&lsi->lsi_bdi);
 956         if (err)
 957                 GOTO(out_free, err);
 958         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 959         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 960         err = ll_bdi_register(&lsi->lsi_bdi);
 961         if (err)
 962                 GOTO(out_free, err);
 963
 964 #ifdef HAVE_SB_BDI
 965         sb->s_bdi = &lsi->lsi_bdi;
 966 #endif
 967
 968         /* Generate a string unique to this super, in case some joker tries
 969            to mount the same fs at two mount points.
 970            Use the address of the super itself.*/
 971         cfg->cfg_instance = sb;
 972         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 973
 974         /* set up client obds */
 975         err = lustre_process_log(sb, profilenm, cfg);
 976         if (err < 0) {
 977                 CERROR("Unable to process log: %d\n", err);
 978                 GOTO(out_free, err);
 979         }
 980
 981         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 982         lprof = class_get_profile(profilenm);
 983         if (lprof == NULL) {
 984                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 985                                    " read from the MGS.  Does that filesystem "
 986                                    "exist?\n", profilenm);
 987                 GOTO(out_free, err = -EINVAL);
 988         }
 989         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 990                lprof->lp_md, lprof->lp_dt);
 991
 992         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 993         if (!dt)
 994                 GOTO(out_free, err = -ENOMEM);
 995         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 996
 997         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
 998         if (!md)
 999                 GOTO(out_free, err = -ENOMEM);
1000         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1001
1002         /* connections, registrations, sb setup */
1003         err = client_common_fill_super(sb, md, dt, mnt);
1004
1005 out_free:
1006         if (md)
1007                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1008         if (dt)
1009                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1010         if (err)
1011                 ll_put_super(sb);
1012         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1013                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1014
1015         OBD_FREE_PTR(cfg);
1016         RETURN(err);
1017 } /* ll_fill_super */
1018
1019
1020 void lu_context_keys_dump(void);
1021
1022 void ll_put_super(struct super_block *sb)
1023 {
1024         struct config_llog_instance cfg;
1025         struct obd_device *obd;
1026         struct lustre_sb_info *lsi = s2lsi(sb);
1027         struct ll_sb_info *sbi = ll_s2sbi(sb);
1028         char *profilenm = get_profile_name(sb);
1029         int force = 1, next;
1030         ENTRY;
1031
1032         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1033
1034         ll_print_capa_stat(sbi);
1035
1036         cfg.cfg_instance = sb;
1037         lustre_end_log(sb, profilenm, &cfg);
1038
1039         if (sbi->ll_md_exp) {
1040                 obd = class_exp2obd(sbi->ll_md_exp);
1041                 if (obd)
1042                         force = obd->obd_force;
1043         }
1044
1045         /* We need to set force before the lov_disconnect in
1046            lustre_common_put_super, since l_d cleans up osc's as well. */
1047         if (force) {
1048                 next = 0;
1049                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1050                                                      &next)) != NULL) {
1051                         obd->obd_force = force;
1052                 }
1053         }
1054
1055         if (sbi->ll_lcq) {
1056                 /* Only if client_common_fill_super succeeded */
1057                 client_common_put_super(sb);
1058         }
1059
1060         next = 0;
1061         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1062                 class_manual_cleanup(obd);
1063         }
1064
1065         if (sbi->ll_flags & LL_SBI_VERBOSE)
1066                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1067
1068         if (profilenm)
1069                 class_del_profile(profilenm);
1070
1071         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1072                 ll_bdi_destroy(&lsi->lsi_bdi);
1073                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1074         }
1075
1076         ll_free_sbi(sb);
1077         lsi->lsi_llsbi = NULL;
1078
1079         lustre_common_put_super(sb);
1080
1081         cl_env_cache_purge(~0);
1082
1083         cfs_module_put(THIS_MODULE);
1084
1085         EXIT;
1086 } /* client_put_super */
1087
1088 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1089 {
1090         struct inode *inode = NULL;
1091         /* NOTE: we depend on atomic igrab() -bzzz */
1092         lock_res_and_lock(lock);
1093         if (lock->l_ast_data) {
1094                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1095                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1096                         inode = igrab(lock->l_ast_data);
1097                 } else {
1098                         inode = lock->l_ast_data;
1099                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1100                                          D_WARNING, lock, "l_ast_data %p is "
1101                                          "bogus: magic %08x", lock->l_ast_data,
1102                                          lli->lli_inode_magic);
1103                         inode = NULL;
1104                 }
1105         }
1106         unlock_res_and_lock(lock);
1107         return inode;
1108 }
1109
1110 static int null_if_equal(struct ldlm_lock *lock, void *data)
1111 {
1112         if (data == lock->l_ast_data) {
1113                 lock->l_ast_data = NULL;
1114
1115                 if (lock->l_req_mode != lock->l_granted_mode)
1116                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1117         }
1118
1119         return LDLM_ITER_CONTINUE;
1120 }
1121
1122 void ll_clear_inode(struct inode *inode)
1123 {
1124         struct ll_inode_info *lli = ll_i2info(inode);
1125         struct ll_sb_info *sbi = ll_i2sbi(inode);
1126         ENTRY;
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1129                inode->i_generation, inode);
1130
1131         if (S_ISDIR(inode->i_mode)) {
1132                 /* these should have been cleared in ll_file_release */
1133                 LASSERT(lli->lli_opendir_key == NULL);
1134                 LASSERT(lli->lli_sai == NULL);
1135                 LASSERT(lli->lli_opendir_pid == 0);
1136         }
1137
1138         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1139         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1140                          null_if_equal, inode);
1141
1142         LASSERT(!lli->lli_open_fd_write_count);
1143         LASSERT(!lli->lli_open_fd_read_count);
1144         LASSERT(!lli->lli_open_fd_exec_count);
1145
1146         if (lli->lli_mds_write_och)
1147                 ll_md_real_close(inode, FMODE_WRITE);
1148         if (lli->lli_mds_exec_och)
1149                 ll_md_real_close(inode, FMODE_EXEC);
1150         if (lli->lli_mds_read_och)
1151                 ll_md_real_close(inode, FMODE_READ);
1152
1153         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1154                 OBD_FREE(lli->lli_symlink_name,
1155                          strlen(lli->lli_symlink_name) + 1);
1156                 lli->lli_symlink_name = NULL;
1157         }
1158
1159         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1160                 LASSERT(lli->lli_posix_acl == NULL);
1161                 if (lli->lli_remote_perms) {
1162                         free_rmtperm_hash(lli->lli_remote_perms);
1163                         lli->lli_remote_perms = NULL;
1164                 }
1165         }
1166 #ifdef CONFIG_FS_POSIX_ACL
1167         else if (lli->lli_posix_acl) {
1168                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1169                 LASSERT(lli->lli_remote_perms == NULL);
1170                 posix_acl_release(lli->lli_posix_acl);
1171                 lli->lli_posix_acl = NULL;
1172         }
1173 #endif
1174         lli->lli_inode_magic = LLI_INODE_DEAD;
1175
1176         ll_clear_inode_capas(inode);
1177         if (!S_ISDIR(inode->i_mode))
1178                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1179
1180         /*
1181          * XXX This has to be done before lsm is freed below, because
1182          * cl_object still uses inode lsm.
1183          */
1184         cl_inode_fini(inode);
1185
1186         if (lli->lli_smd) {
1187                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1188                 lli->lli_smd = NULL;
1189         }
1190
1191
1192         EXIT;
1193 }
1194
1195 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1196                   struct md_open_data **mod)
1197 {
1198         struct lustre_md md;
1199         struct inode *inode = dentry->d_inode;
1200         struct ll_sb_info *sbi = ll_i2sbi(inode);
1201         struct ptlrpc_request *request = NULL;
1202         int rc, ia_valid;
1203         ENTRY;
1204
1205         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1206                                      LUSTRE_OPC_ANY, NULL);
1207         if (IS_ERR(op_data))
1208                 RETURN(PTR_ERR(op_data));
1209
1210         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1211                         &request, mod);
1212         if (rc) {
1213                 ptlrpc_req_finished(request);
1214                 if (rc == -ENOENT) {
1215                         inode->i_nlink = 0;
1216                         /* Unlinked special device node? Or just a race?
1217                          * Pretend we done everything. */
1218                         if (!S_ISREG(inode->i_mode) &&
1219                             !S_ISDIR(inode->i_mode)) {
1220                                 ia_valid = op_data->op_attr.ia_valid;
1221                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1222                                 rc = simple_setattr(dentry, &op_data->op_attr);
1223                                 op_data->op_attr.ia_valid = ia_valid;
1224                         }
1225                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1226                         CERROR("md_setattr fails: rc = %d\n", rc);
1227                 }
1228                 RETURN(rc);
1229         }
1230
1231         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1232                               sbi->ll_md_exp, &md);
1233         if (rc) {
1234                 ptlrpc_req_finished(request);
1235                 RETURN(rc);
1236         }
1237
1238         /* We want to adjust timestamps.
1239          * If there is at least some data in file, we cleared ATTR_SIZE
1240          * to avoid update size, otherwise it is important to do.(SOM case)
1241          * (bug 6196) */
1242         ia_valid = op_data->op_attr.ia_valid;
1243         /* Since we set ATTR_*_SET flags above, and already done permission
1244          * check, So don't let inode_change_ok() check it again.  */
1245         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1246         rc = simple_setattr(dentry, &op_data->op_attr);
1247         op_data->op_attr.ia_valid = ia_valid;
1248
1249         /* Extract epoch data if obtained. */
1250         op_data->op_handle = md.body->handle;
1251         op_data->op_ioepoch = md.body->ioepoch;
1252
1253         ll_update_inode(inode, &md);
1254         ptlrpc_req_finished(request);
1255
1256         RETURN(rc);
1257 }
1258
1259 /* Close IO epoch and send Size-on-MDS attribute update. */
1260 static int ll_setattr_done_writing(struct inode *inode,
1261                                    struct md_op_data *op_data,
1262                                    struct md_open_data *mod)
1263 {
1264         struct ll_inode_info *lli = ll_i2info(inode);
1265         int rc = 0;
1266         ENTRY;
1267
1268         LASSERT(op_data != NULL);
1269         if (!S_ISREG(inode->i_mode))
1270                 RETURN(0);
1271
1272         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1273                op_data->op_ioepoch, PFID(&lli->lli_fid));
1274
1275         op_data->op_flags = MF_EPOCH_CLOSE;
1276         ll_done_writing_attr(inode, op_data);
1277         ll_pack_inode2opdata(inode, op_data, NULL);
1278
1279         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1280         if (rc == -EAGAIN) {
1281                 /* MDS has instructed us to obtain Size-on-MDS attribute
1282                  * from OSTs and send setattr to back to MDS. */
1283                 rc = ll_som_update(inode, op_data);
1284         } else if (rc) {
1285                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1286                        inode->i_ino, rc);
1287         }
1288         RETURN(rc);
1289 }
1290
1291 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1292 {
1293         struct obd_capa *capa;
1294         int rc;
1295
1296         if (attr->ia_valid & ATTR_SIZE)
1297                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1298         else
1299                 capa = ll_mdscapa_get(inode);
1300
1301         rc = cl_setattr_ost(inode, attr, capa);
1302
1303         if (attr->ia_valid & ATTR_SIZE)
1304                 ll_truncate_free_capa(capa);
1305         else
1306                 capa_put(capa);
1307
1308         return rc;
1309 }
1310
1311 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1312  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1313  * keep these values until such a time that objects are allocated for it.
1314  * We do the MDS operations first, as it is checking permissions for us.
1315  * We don't to the MDS RPC if there is nothing that we want to store there,
1316  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1317  * going to do an RPC anyways.
1318  *
1319  * If we are doing a truncate, we will send the mtime and ctime updates
1320  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1321  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1322  * at the same time.
1323  */
1324 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1325 {
1326         struct inode *inode = dentry->d_inode;
1327         struct ll_inode_info *lli = ll_i2info(inode);
1328         struct lov_stripe_md *lsm = lli->lli_smd;
1329         struct md_op_data *op_data = NULL;
1330         struct md_open_data *mod = NULL;
1331         int ia_valid = attr->ia_valid;
1332         int rc = 0, rc1 = 0;
1333         ENTRY;
1334
1335         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1336                attr->ia_valid);
1337
1338         if (ia_valid & ATTR_SIZE) {
1339                 /* Check new size against VFS/VM file size limit and rlimit */
1340                 rc = inode_newsize_ok(inode, attr->ia_size);
1341                 if (rc)
1342                         RETURN(rc);
1343
1344                 /* The maximum Lustre file size is variable, based on the
1345                  * OST maximum object size and number of stripes.  This
1346                  * needs another check in addition to the VFS check above. */
1347                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1348                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1349                                PFID(&lli->lli_fid), attr->ia_size,
1350                                ll_file_maxbytes(inode));
1351                         RETURN(-EFBIG);
1352                 }
1353
1354                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1355         }
1356
1357         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1358         if (ia_valid & TIMES_SET_FLAGS) {
1359                 if (cfs_curproc_fsuid() != inode->i_uid &&
1360                     !cfs_capable(CFS_CAP_FOWNER))
1361                         RETURN(-EPERM);
1362         }
1363
1364         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1365         if (attr->ia_valid & ATTR_CTIME) {
1366                 attr->ia_ctime = CFS_CURRENT_TIME;
1367                 attr->ia_valid |= ATTR_CTIME_SET;
1368         }
1369         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1370                 attr->ia_atime = CFS_CURRENT_TIME;
1371                 attr->ia_valid |= ATTR_ATIME_SET;
1372         }
1373         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1374                 attr->ia_mtime = CFS_CURRENT_TIME;
1375                 attr->ia_valid |= ATTR_MTIME_SET;
1376         }
1377
1378         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1379                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1380                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1381                        cfs_time_current_sec());
1382
1383         /* NB: ATTR_SIZE will only be set after this point if the size
1384          * resides on the MDS, ie, this file has no objects. */
1385         if (lsm)
1386                 attr->ia_valid &= ~ATTR_SIZE;
1387
1388         /* We always do an MDS RPC, even if we're only changing the size;
1389          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1390
1391         OBD_ALLOC_PTR(op_data);
1392         if (op_data == NULL)
1393                 RETURN(-ENOMEM);
1394
1395         if (!S_ISDIR(inode->i_mode)) {
1396                 if (ia_valid & ATTR_SIZE)
1397                         UP_WRITE_I_ALLOC_SEM(inode);
1398                 UNLOCK_INODE_MUTEX(inode);
1399                 cfs_down_write(&lli->lli_trunc_sem);
1400                 LOCK_INODE_MUTEX(inode);
1401                 if (ia_valid & ATTR_SIZE)
1402                         DOWN_WRITE_I_ALLOC_SEM(inode);
1403         }
1404
1405         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1406
1407         /* Open epoch for truncate. */
1408         if (exp_connect_som(ll_i2mdexp(inode)) &&
1409             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1410                 op_data->op_flags = MF_EPOCH_OPEN;
1411
1412         rc = ll_md_setattr(dentry, op_data, &mod);
1413         if (rc)
1414                 GOTO(out, rc);
1415
1416         ll_ioepoch_open(lli, op_data->op_ioepoch);
1417         if (!lsm || !S_ISREG(inode->i_mode)) {
1418                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1419                 GOTO(out, rc = 0);
1420         }
1421
1422         if (ia_valid & ATTR_SIZE)
1423                 attr->ia_valid |= ATTR_SIZE;
1424         if (ia_valid & (ATTR_SIZE |
1425                         ATTR_ATIME | ATTR_ATIME_SET |
1426                         ATTR_MTIME | ATTR_MTIME_SET))
1427                 /* For truncate and utimes sending attributes to OSTs, setting
1428                  * mtime/atime to the past will be performed under PW [0:EOF]
1429                  * extent lock (new_size:EOF for truncate).  It may seem
1430                  * excessive to send mtime/atime updates to OSTs when not
1431                  * setting times to past, but it is necessary due to possible
1432                  * time de-synchronization between MDT inode and OST objects */
1433                 rc = ll_setattr_ost(inode, attr);
1434         EXIT;
1435 out:
1436         if (op_data) {
1437                 if (op_data->op_ioepoch) {
1438                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1439                         if (!rc)
1440                                 rc = rc1;
1441                 }
1442                 ll_finish_md_op_data(op_data);
1443         }
1444         if (!S_ISDIR(inode->i_mode))
1445                 cfs_up_write(&lli->lli_trunc_sem);
1446
1447         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1448                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1449
1450         return rc;
1451 }
1452
1453 int ll_setattr(struct dentry *de, struct iattr *attr)
1454 {
1455         int mode = de->d_inode->i_mode;
1456
1457         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1458                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1459                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1460
1461         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1462                                (ATTR_SIZE|ATTR_MODE)) &&
1463             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1464              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1465               !(attr->ia_mode & S_ISGID))))
1466                 attr->ia_valid |= ATTR_FORCE;
1467
1468         if ((mode & S_ISUID) &&
1469             !(attr->ia_mode & S_ISUID) &&
1470             !(attr->ia_valid & ATTR_KILL_SUID))
1471                 attr->ia_valid |= ATTR_KILL_SUID;
1472
1473         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1474             !(attr->ia_mode & S_ISGID) &&
1475             !(attr->ia_valid & ATTR_KILL_SGID))
1476                 attr->ia_valid |= ATTR_KILL_SGID;
1477
1478         return ll_setattr_raw(de, attr);
1479 }
1480
1481 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1482                        __u64 max_age, __u32 flags)
1483 {
1484         struct ll_sb_info *sbi = ll_s2sbi(sb);
1485         struct obd_statfs obd_osfs;
1486         int rc;
1487         ENTRY;
1488
1489         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1490         if (rc) {
1491                 CERROR("md_statfs fails: rc = %d\n", rc);
1492                 RETURN(rc);
1493         }
1494
1495         osfs->os_type = sb->s_magic;
1496
1497         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1498                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1499
1500         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1501                 flags |= OBD_STATFS_NODELAY;
1502
1503         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1504                               &obd_osfs, max_age, flags);
1505         if (rc) {
1506                 CERROR("obd_statfs fails: rc = %d\n", rc);
1507                 RETURN(rc);
1508         }
1509
1510         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1511                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1512                obd_osfs.os_files);
1513
1514         osfs->os_bsize = obd_osfs.os_bsize;
1515         osfs->os_blocks = obd_osfs.os_blocks;
1516         osfs->os_bfree = obd_osfs.os_bfree;
1517         osfs->os_bavail = obd_osfs.os_bavail;
1518
1519         /* If we don't have as many objects free on the OST as inodes
1520          * on the MDS, we reduce the total number of inodes to
1521          * compensate, so that the "inodes in use" number is correct.
1522          */
1523         if (obd_osfs.os_ffree < osfs->os_ffree) {
1524                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1525                         obd_osfs.os_ffree;
1526                 osfs->os_ffree = obd_osfs.os_ffree;
1527         }
1528
1529         RETURN(rc);
1530 }
1531 #ifndef HAVE_STATFS_DENTRY_PARAM
1532 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1533 {
1534 #else
1535 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1536 {
1537         struct super_block *sb = de->d_sb;
1538 #endif
1539         struct obd_statfs osfs;
1540         int rc;
1541
1542         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1543         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1544
1545         /* Some amount of caching on the client is allowed */
1546         rc = ll_statfs_internal(sb, &osfs,
1547                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1548                                 0);
1549         if (rc)
1550                 return rc;
1551
1552         statfs_unpack(sfs, &osfs);
1553
1554         /* We need to downshift for all 32-bit kernels, because we can't
1555          * tell if the kernel is being called via sys_statfs64() or not.
1556          * Stop before overflowing f_bsize - in which case it is better
1557          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1558         if (sizeof(long) < 8) {
1559                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1560                         sfs->f_bsize <<= 1;
1561
1562                         osfs.os_blocks >>= 1;
1563                         osfs.os_bfree >>= 1;
1564                         osfs.os_bavail >>= 1;
1565                 }
1566         }
1567
1568         sfs->f_blocks = osfs.os_blocks;
1569         sfs->f_bfree = osfs.os_bfree;
1570         sfs->f_bavail = osfs.os_bavail;
1571
1572         return 0;
1573 }
1574
1575 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1576 {
1577         struct ll_inode_info *lli;
1578         struct lov_stripe_md *lsm;
1579
1580         LASSERT(!S_ISDIR(inode->i_mode));
1581
1582         lli = ll_i2info(inode);
1583         LASSERT(lli->lli_size_sem_owner != current);
1584         cfs_down(&lli->lli_size_sem);
1585         LASSERT(lli->lli_size_sem_owner == NULL);
1586         lli->lli_size_sem_owner = current;
1587         lsm = lli->lli_smd;
1588         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1589                  lsm, lock_lsm);
1590         if (lock_lsm)
1591                 lov_stripe_lock(lsm);
1592 }
1593
1594 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1595 {
1596         struct ll_inode_info *lli;
1597         struct lov_stripe_md *lsm;
1598
1599         lli = ll_i2info(inode);
1600         lsm = lli->lli_smd;
1601         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1602                  lsm, unlock_lsm);
1603         if (unlock_lsm)
1604                 lov_stripe_unlock(lsm);
1605         LASSERT(lli->lli_size_sem_owner == current);
1606         lli->lli_size_sem_owner = NULL;
1607         cfs_up(&lli->lli_size_sem);
1608 }
1609
1610 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1611 {
1612         struct ll_inode_info *lli = ll_i2info(inode);
1613         struct mdt_body *body = md->body;
1614         struct lov_stripe_md *lsm = md->lsm;
1615         struct ll_sb_info *sbi = ll_i2sbi(inode);
1616
1617         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1618         if (lsm != NULL) {
1619                 LASSERT(S_ISREG(inode->i_mode));
1620
1621                 cfs_mutex_lock(&lli->lli_och_mutex);
1622                 if (lli->lli_smd == NULL) {
1623                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1624                             lsm->lsm_magic != LOV_MAGIC_V3) {
1625                                 dump_lsm(D_ERROR, lsm);
1626                                 LBUG();
1627                         }
1628                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1629                                lsm, inode->i_ino, inode->i_generation, inode);
1630                         /* cl_file_inode_init must go before lli_smd or a race
1631                          * is possible where client thinks the file has stripes,
1632                          * but lov raid0 is not setup yet and parallel e.g.
1633                          * glimpse would try to use uninitialized lov */
1634                         cl_file_inode_init(inode, md);
1635                         cfs_spin_lock(&lli->lli_lock);
1636                         lli->lli_smd = lsm;
1637                         cfs_spin_unlock(&lli->lli_lock);
1638                         cfs_mutex_unlock(&lli->lli_och_mutex);
1639                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1640                         if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1641                                 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1642                 } else {
1643                         cfs_mutex_unlock(&lli->lli_och_mutex);
1644                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1645                                 lli->lli_smd->lsm_stripe_count ==
1646                                 lsm->lsm_stripe_count);
1647                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1648                                 CERROR("lsm mismatch for inode %ld\n",
1649                                        inode->i_ino);
1650                                 CERROR("lli_smd:\n");
1651                                 dump_lsm(D_ERROR, lli->lli_smd);
1652                                 CERROR("lsm:\n");
1653                                 dump_lsm(D_ERROR, lsm);
1654                                 LBUG();
1655                         }
1656                 }
1657                 if (lli->lli_smd != lsm)
1658                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1659         }
1660
1661         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1662                 if (body->valid & OBD_MD_FLRMTPERM)
1663                         ll_update_remote_perm(inode, md->remote_perm);
1664         }
1665 #ifdef CONFIG_FS_POSIX_ACL
1666         else if (body->valid & OBD_MD_FLACL) {
1667                 cfs_spin_lock(&lli->lli_lock);
1668                 if (lli->lli_posix_acl)
1669                         posix_acl_release(lli->lli_posix_acl);
1670                 lli->lli_posix_acl = md->posix_acl;
1671                 cfs_spin_unlock(&lli->lli_lock);
1672         }
1673 #endif
1674         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1675         inode->i_generation = cl_fid_build_gen(&body->fid1);
1676
1677         if (body->valid & OBD_MD_FLATIME) {
1678                 if (body->atime > LTIME_S(inode->i_atime))
1679                         LTIME_S(inode->i_atime) = body->atime;
1680                 lli->lli_lvb.lvb_atime = body->atime;
1681         }
1682         if (body->valid & OBD_MD_FLMTIME) {
1683                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1684                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1685                                "to "LPU64"\n", inode->i_ino,
1686                                LTIME_S(inode->i_mtime), body->mtime);
1687                         LTIME_S(inode->i_mtime) = body->mtime;
1688                 }
1689                 lli->lli_lvb.lvb_mtime = body->mtime;
1690         }
1691         if (body->valid & OBD_MD_FLCTIME) {
1692                 if (body->ctime > LTIME_S(inode->i_ctime))
1693                         LTIME_S(inode->i_ctime) = body->ctime;
1694                 lli->lli_lvb.lvb_ctime = body->ctime;
1695         }
1696         if (body->valid & OBD_MD_FLMODE)
1697                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1698         if (body->valid & OBD_MD_FLTYPE)
1699                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1700         LASSERT(inode->i_mode != 0);
1701         if (S_ISREG(inode->i_mode)) {
1702                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1703         } else {
1704                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1705         }
1706 #ifdef HAVE_INODE_BLKSIZE
1707         inode->i_blksize = 1<<inode->i_blkbits;
1708 #endif
1709         if (body->valid & OBD_MD_FLUID)
1710                 inode->i_uid = body->uid;
1711         if (body->valid & OBD_MD_FLGID)
1712                 inode->i_gid = body->gid;
1713         if (body->valid & OBD_MD_FLFLAGS)
1714                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1715         if (body->valid & OBD_MD_FLNLINK)
1716                 inode->i_nlink = body->nlink;
1717         if (body->valid & OBD_MD_FLRDEV)
1718                 inode->i_rdev = old_decode_dev(body->rdev);
1719
1720         if (body->valid & OBD_MD_FLID) {
1721                 /* FID shouldn't be changed! */
1722                 if (fid_is_sane(&lli->lli_fid)) {
1723                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1724                                  "Trying to change FID "DFID
1725                                  " to the "DFID", inode %lu/%u(%p)\n",
1726                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1727                                  inode->i_ino, inode->i_generation, inode);
1728                 } else
1729                         lli->lli_fid = body->fid1;
1730         }
1731
1732         LASSERT(fid_seq(&lli->lli_fid) != 0);
1733
1734         if (body->valid & OBD_MD_FLSIZE) {
1735                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1736                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1737                         struct lustre_handle lockh;
1738                         ldlm_mode_t mode;
1739
1740                         /* As it is possible a blocking ast has been processed
1741                          * by this time, we need to check there is an UPDATE
1742                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1743                          * it. */
1744                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1745                                                &lockh);
1746                         if (mode) {
1747                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1748                                                       LLIF_EPOCH_PENDING |
1749                                                       LLIF_SOM_DIRTY)) {
1750                                         CERROR("ino %lu flags %u still has "
1751                                                "size authority! do not trust "
1752                                                "the size got from MDS\n",
1753                                                inode->i_ino, lli->lli_flags);
1754                                 } else {
1755                                         /* Use old size assignment to avoid
1756                                          * deadlock bz14138 & bz14326 */
1757                                         i_size_write(inode, body->size);
1758                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1759                                 }
1760                                 ldlm_lock_decref(&lockh, mode);
1761                         }
1762                 } else {
1763                         /* Use old size assignment to avoid
1764                          * deadlock bz14138 & bz14326 */
1765                         i_size_write(inode, body->size);
1766
1767                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1768                                inode->i_ino, (unsigned long long)body->size);
1769                 }
1770
1771                 if (body->valid & OBD_MD_FLBLOCKS)
1772                         inode->i_blocks = body->blocks;
1773         }
1774
1775         if (body->valid & OBD_MD_FLMDSCAPA) {
1776                 LASSERT(md->mds_capa);
1777                 ll_add_capa(inode, md->mds_capa);
1778         }
1779         if (body->valid & OBD_MD_FLOSSCAPA) {
1780                 LASSERT(md->oss_capa);
1781                 ll_add_capa(inode, md->oss_capa);
1782         }
1783 }
1784
1785 void ll_read_inode2(struct inode *inode, void *opaque)
1786 {
1787         struct lustre_md *md = opaque;
1788         struct ll_inode_info *lli = ll_i2info(inode);
1789         ENTRY;
1790
1791         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1792                PFID(&lli->lli_fid), inode);
1793
1794         LASSERT(!lli->lli_smd);
1795
1796         /* Core attributes from the MDS first.  This is a new inode, and
1797          * the VFS doesn't zero times in the core inode so we have to do
1798          * it ourselves.  They will be overwritten by either MDS or OST
1799          * attributes - we just need to make sure they aren't newer. */
1800         LTIME_S(inode->i_mtime) = 0;
1801         LTIME_S(inode->i_atime) = 0;
1802         LTIME_S(inode->i_ctime) = 0;
1803         inode->i_rdev = 0;
1804         ll_update_inode(inode, md);
1805
1806         /* OIDEBUG(inode); */
1807
1808         /* initializing backing dev info. */
1809         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1810
1811
1812         if (S_ISREG(inode->i_mode)) {
1813                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1814                 inode->i_op = &ll_file_inode_operations;
1815                 inode->i_fop = sbi->ll_fop;
1816                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1817                 EXIT;
1818         } else if (S_ISDIR(inode->i_mode)) {
1819                 inode->i_op = &ll_dir_inode_operations;
1820                 inode->i_fop = &ll_dir_operations;
1821                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1822                 EXIT;
1823         } else if (S_ISLNK(inode->i_mode)) {
1824                 inode->i_op = &ll_fast_symlink_inode_operations;
1825                 EXIT;
1826         } else {
1827                 inode->i_op = &ll_special_inode_operations;
1828
1829                 init_special_inode(inode, inode->i_mode,
1830                                    kdev_t_to_nr(inode->i_rdev));
1831
1832                 EXIT;
1833         }
1834 }
1835
1836 void ll_delete_inode(struct inode *inode)
1837 {
1838         struct ll_sb_info *sbi = ll_i2sbi(inode);
1839         int rc;
1840         ENTRY;
1841
1842         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1843         if (rc)
1844                 CERROR("fid_delete() failed, rc %d\n", rc);
1845
1846         truncate_inode_pages(&inode->i_data, 0);
1847
1848         /* Workaround for LU-118 */
1849         if (inode->i_data.nrpages) {
1850                 TREE_READ_LOCK_IRQ(&inode->i_data);
1851                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1852                 LASSERTF(inode->i_data.nrpages == 0,
1853                          "inode=%lu/%u(%p) nrpages=%lu, see "
1854                          "http://jira.whamcloud.com/browse/LU-118\n",
1855                          inode->i_ino, inode->i_generation, inode,
1856                          inode->i_data.nrpages);
1857         }
1858         /* Workaround end */
1859
1860 #ifdef HAVE_SBOPS_EVICT_INODE
1861         ll_clear_inode(inode);
1862         end_writeback(inode);
1863 #else
1864         clear_inode(inode);
1865 #endif
1866
1867         EXIT;
1868 }
1869
1870 int ll_iocontrol(struct inode *inode, struct file *file,
1871                  unsigned int cmd, unsigned long arg)
1872 {
1873         struct ll_sb_info *sbi = ll_i2sbi(inode);
1874         struct ptlrpc_request *req = NULL;
1875         int rc, flags = 0;
1876         ENTRY;
1877
1878         switch(cmd) {
1879         case FSFILT_IOC_GETFLAGS: {
1880                 struct mdt_body *body;
1881                 struct md_op_data *op_data;
1882
1883                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1884                                              0, 0, LUSTRE_OPC_ANY,
1885                                              NULL);
1886                 if (IS_ERR(op_data))
1887                         RETURN(PTR_ERR(op_data));
1888
1889                 op_data->op_valid = OBD_MD_FLFLAGS;
1890                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1891                 ll_finish_md_op_data(op_data);
1892                 if (rc) {
1893                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1894                         RETURN(-abs(rc));
1895                 }
1896
1897                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1898
1899                 flags = body->flags;
1900
1901                 ptlrpc_req_finished(req);
1902
1903                 RETURN(put_user(flags, (int *)arg));
1904         }
1905         case FSFILT_IOC_SETFLAGS: {
1906                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1907                 struct obd_info oinfo = { { { 0 } } };
1908                 struct md_op_data *op_data;
1909
1910                 if (get_user(flags, (int *)arg))
1911                         RETURN(-EFAULT);
1912
1913                 oinfo.oi_md = lsm;
1914                 OBDO_ALLOC(oinfo.oi_oa);
1915                 if (!oinfo.oi_oa)
1916                         RETURN(-ENOMEM);
1917
1918                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1919                                              LUSTRE_OPC_ANY, NULL);
1920                 if (IS_ERR(op_data))
1921                         RETURN(PTR_ERR(op_data));
1922
1923                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1924                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1925                 rc = md_setattr(sbi->ll_md_exp, op_data,
1926                                 NULL, 0, NULL, 0, &req, NULL);
1927                 ll_finish_md_op_data(op_data);
1928                 ptlrpc_req_finished(req);
1929                 if (rc) {
1930                         OBDO_FREE(oinfo.oi_oa);
1931                         RETURN(rc);
1932                 }
1933
1934                 if (lsm == NULL) {
1935                         OBDO_FREE(oinfo.oi_oa);
1936                         GOTO(update_cache, rc);
1937                 }
1938
1939                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1940                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1941                 oinfo.oi_oa->o_flags = flags;
1942                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1943                                        OBD_MD_FLGROUP;
1944                 oinfo.oi_capa = ll_mdscapa_get(inode);
1945                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1946                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1947                 capa_put(oinfo.oi_capa);
1948                 OBDO_FREE(oinfo.oi_oa);
1949                 if (rc) {
1950                         if (rc != -EPERM && rc != -EACCES)
1951                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1952                         RETURN(rc);
1953                 }
1954
1955                 EXIT;
1956 update_cache:
1957                 inode->i_flags = ll_ext_to_inode_flags(flags);
1958                 return 0;
1959         }
1960         default:
1961                 RETURN(-ENOSYS);
1962         }
1963
1964         RETURN(0);
1965 }
1966
1967 int ll_flush_ctx(struct inode *inode)
1968 {
1969         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1970
1971         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1972
1973         obd_set_info_async(sbi->ll_md_exp,
1974                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1975                            0, NULL, NULL);
1976         obd_set_info_async(sbi->ll_dt_exp,
1977                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1978                            0, NULL, NULL);
1979         return 0;
1980 }
1981
1982 /* umount -f client means force down, don't save state */
1983 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1984 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1985 {
1986         struct super_block *sb = vfsmnt->mnt_sb;
1987 #else
1988 void ll_umount_begin(struct super_block *sb)
1989 {
1990 #endif
1991         struct lustre_sb_info *lsi = s2lsi(sb);
1992         struct ll_sb_info *sbi = ll_s2sbi(sb);
1993         struct obd_device *obd;
1994         struct obd_ioctl_data *ioc_data;
1995         ENTRY;
1996
1997 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1998         if (!(flags & MNT_FORCE)) {
1999                 EXIT;
2000                 return;
2001         }
2002 #endif
2003
2004         /* Tell the MGC we got umount -f */
2005         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2006
2007         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2008                sb->s_count, atomic_read(&sb->s_active));
2009
2010         obd = class_exp2obd(sbi->ll_md_exp);
2011         if (obd == NULL) {
2012                 CERROR("Invalid MDC connection handle "LPX64"\n",
2013                        sbi->ll_md_exp->exp_handle.h_cookie);
2014                 EXIT;
2015                 return;
2016         }
2017         obd->obd_force = 1;
2018
2019         obd = class_exp2obd(sbi->ll_dt_exp);
2020         if (obd == NULL) {
2021                 CERROR("Invalid LOV connection handle "LPX64"\n",
2022                        sbi->ll_dt_exp->exp_handle.h_cookie);
2023                 EXIT;
2024                 return;
2025         }
2026         obd->obd_force = 1;
2027
2028         OBD_ALLOC_PTR(ioc_data);
2029         if (ioc_data) {
2030                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2031                               sizeof ioc_data, ioc_data, NULL);
2032
2033                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2034                               sizeof ioc_data, ioc_data, NULL);
2035
2036                 OBD_FREE_PTR(ioc_data);
2037         }
2038
2039
2040         /* Really, we'd like to wait until there are no requests outstanding,
2041          * and then continue.  For now, we just invalidate the requests,
2042          * schedule() and sleep one second if needed, and hope.
2043          */
2044         cfs_schedule();
2045 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2046         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2047                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2048                                                    cfs_time_seconds(1));
2049                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2050                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2051                                       "may try to umount it a bit later\n",
2052                                       atomic_read(&vfsmnt->mnt_count));
2053         }
2054 #endif
2055
2056         EXIT;
2057 }
2058
2059 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2060 {
2061         struct ll_sb_info *sbi = ll_s2sbi(sb);
2062         char *profilenm = get_profile_name(sb);
2063         int err;
2064         __u32 read_only;
2065
2066         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2067                 read_only = *flags & MS_RDONLY;
2068                 err = obd_set_info_async(sbi->ll_md_exp,
2069                                          sizeof(KEY_READ_ONLY),
2070                                          KEY_READ_ONLY, sizeof(read_only),
2071                                          &read_only, NULL);
2072                 if (err) {
2073                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2074                                       profilenm, read_only ?
2075                                       "read-only" : "read-write", err);
2076                         return err;
2077                 }
2078
2079                 if (read_only)
2080                         sb->s_flags |= MS_RDONLY;
2081                 else
2082                         sb->s_flags &= ~MS_RDONLY;
2083
2084                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2085                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2086                                       read_only ?  "read-only" : "read-write");
2087         }
2088         return 0;
2089 }
2090
2091 int ll_prep_inode(struct inode **inode,
2092                   struct ptlrpc_request *req,
2093                   struct super_block *sb)
2094 {
2095         struct ll_sb_info *sbi = NULL;
2096         struct lustre_md md;
2097         int rc;
2098         ENTRY;
2099
2100         LASSERT(*inode || sb);
2101         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2102         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2103                               sbi->ll_md_exp, &md);
2104         if (rc)
2105                 RETURN(rc);
2106
2107         if (*inode) {
2108                 ll_update_inode(*inode, &md);
2109         } else {
2110                 LASSERT(sb != NULL);
2111
2112                 /*
2113                  * At this point server returns to client's same fid as client
2114                  * generated for creating. So using ->fid1 is okay here.
2115                  */
2116                 LASSERT(fid_is_sane(&md.body->fid1));
2117
2118                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2119                 if (*inode == NULL || IS_ERR(*inode)) {
2120                         if (md.lsm)
2121                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2122 #ifdef CONFIG_FS_POSIX_ACL
2123                         if (md.posix_acl) {
2124                                 posix_acl_release(md.posix_acl);
2125                                 md.posix_acl = NULL;
2126                         }
2127 #endif
2128                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2129                         *inode = NULL;
2130                         CERROR("new_inode -fatal: rc %d\n", rc);
2131                         GOTO(out, rc);
2132                 }
2133         }
2134
2135 out:
2136         md_free_lustre_md(sbi->ll_md_exp, &md);
2137         RETURN(rc);
2138 }
2139
2140 int ll_obd_statfs(struct inode *inode, void *arg)
2141 {
2142         struct ll_sb_info *sbi = NULL;
2143         struct obd_export *exp;
2144         char *buf = NULL;
2145         struct obd_ioctl_data *data = NULL;
2146         __u32 type;
2147         int len = 0, rc;
2148
2149         if (!inode || !(sbi = ll_i2sbi(inode)))
2150                 GOTO(out_statfs, rc = -EINVAL);
2151
2152         rc = obd_ioctl_getdata(&buf, &len, arg);
2153         if (rc)
2154                 GOTO(out_statfs, rc);
2155
2156         data = (void*)buf;
2157         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2158             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2159                 GOTO(out_statfs, rc = -EINVAL);
2160
2161         if (data->ioc_inllen1 != sizeof(__u32) ||
2162             data->ioc_inllen2 != sizeof(__u32) ||
2163             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2164             data->ioc_plen2 != sizeof(struct obd_uuid))
2165                 GOTO(out_statfs, rc = -EINVAL);
2166
2167         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2168         if (type == LL_STATFS_LMV)
2169                 exp = sbi->ll_md_exp;
2170         else if (type == LL_STATFS_LOV)
2171                 exp = sbi->ll_dt_exp;
2172         else
2173                 GOTO(out_statfs, rc = -ENODEV);
2174
2175         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2176         if (rc)
2177                 GOTO(out_statfs, rc);
2178 out_statfs:
2179         if (buf)
2180                 obd_ioctl_freedata(buf, len);
2181         return rc;
2182 }
2183
2184 int ll_process_config(struct lustre_cfg *lcfg)
2185 {
2186         char *ptr;
2187         void *sb;
2188         struct lprocfs_static_vars lvars;
2189         unsigned long x;
2190         int rc = 0;
2191
2192         lprocfs_llite_init_vars(&lvars);
2193
2194         /* The instance name contains the sb: lustre-client-aacfe000 */
2195         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2196         if (!ptr || !*(++ptr))
2197                 return -EINVAL;
2198         if (sscanf(ptr, "%lx", &x) != 1)
2199                 return -EINVAL;
2200         sb = (void *)x;
2201         /* This better be a real Lustre superblock! */
2202         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2203
2204         /* Note we have not called client_common_fill_super yet, so
2205            proc fns must be able to handle that! */
2206         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2207                                       lcfg, sb);
2208         if (rc > 0)
2209                 rc = 0;
2210         return(rc);
2211 }
2212
2213 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2214 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2215                                        struct inode *i1, struct inode *i2,
2216                                        const char *name, int namelen,
2217                                        int mode, __u32 opc, void *data)
2218 {
2219         LASSERT(i1 != NULL);
2220
2221         if (namelen > ll_i2sbi(i1)->ll_namelen)
2222                 return ERR_PTR(-ENAMETOOLONG);
2223
2224         if (op_data == NULL)
2225                 OBD_ALLOC_PTR(op_data);
2226
2227         if (op_data == NULL)
2228                 return ERR_PTR(-ENOMEM);
2229
2230         ll_i2gids(op_data->op_suppgids, i1, i2);
2231         op_data->op_fid1 = *ll_inode2fid(i1);
2232         op_data->op_capa1 = ll_mdscapa_get(i1);
2233
2234         if (i2) {
2235                 op_data->op_fid2 = *ll_inode2fid(i2);
2236                 op_data->op_capa2 = ll_mdscapa_get(i2);
2237         } else {
2238                 fid_zero(&op_data->op_fid2);
2239                 op_data->op_capa2 = NULL;
2240         }
2241
2242         op_data->op_name = name;
2243         op_data->op_namelen = namelen;
2244         op_data->op_mode = mode;
2245         op_data->op_mod_time = cfs_time_current_sec();
2246         op_data->op_fsuid = cfs_curproc_fsuid();
2247         op_data->op_fsgid = cfs_curproc_fsgid();
2248         op_data->op_cap = cfs_curproc_cap_pack();
2249         op_data->op_bias = MDS_CHECK_SPLIT;
2250         op_data->op_opc = opc;
2251         op_data->op_mds = 0;
2252         op_data->op_data = data;
2253
2254         /* If the file is being opened after mknod() (normally due to NFS)
2255          * try to use the default stripe data from parent directory for
2256          * allocating OST objects.  Try to pass the parent FID to MDS. */
2257         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2258             ll_i2info(i2)->lli_smd == NULL) {
2259                 struct ll_inode_info *lli = ll_i2info(i2);
2260
2261                 cfs_spin_lock(&lli->lli_lock);
2262                 if (likely(lli->lli_smd == NULL &&
2263                            !fid_is_zero(&lli->lli_pfid)))
2264                         op_data->op_fid1 = lli->lli_pfid;
2265                 cfs_spin_unlock(&lli->lli_lock);
2266                 /** We ignore parent's capability temporary. */
2267         }
2268
2269         return op_data;
2270 }
2271
2272 void ll_finish_md_op_data(struct md_op_data *op_data)
2273 {
2274         capa_put(op_data->op_capa1);
2275         capa_put(op_data->op_capa2);
2276         OBD_FREE_PTR(op_data);
2277 }
2278
2279 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2280 {
2281         struct ll_sb_info *sbi;
2282
2283         LASSERT((seq != NULL) && (vfs != NULL));
2284         sbi = ll_s2sbi(vfs->mnt_sb);
2285
2286         if (sbi->ll_flags & LL_SBI_NOLCK)
2287                 seq_puts(seq, ",nolock");
2288
2289         if (sbi->ll_flags & LL_SBI_FLOCK)
2290                 seq_puts(seq, ",flock");
2291
2292         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2293                 seq_puts(seq, ",localflock");
2294
2295         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2296                 seq_puts(seq, ",user_xattr");
2297
2298         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2299                 seq_puts(seq, ",lazystatfs");
2300
2301         RETURN(0);
2302 }
2303
2304 /**
2305  * Get obd name by cmd, and copy out to user space
2306  */
2307 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2308 {
2309         struct ll_sb_info *sbi = ll_i2sbi(inode);
2310         struct obd_device *obd;
2311         ENTRY;
2312
2313         if (cmd == OBD_IOC_GETDTNAME)
2314                 obd = class_exp2obd(sbi->ll_dt_exp);
2315         else if (cmd == OBD_IOC_GETMDNAME)
2316                 obd = class_exp2obd(sbi->ll_md_exp);
2317         else
2318                 RETURN(-EINVAL);
2319
2320         if (!obd)
2321                 RETURN(-ENOENT);
2322
2323         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2324                              strlen(obd->obd_name) + 1))
2325                 RETURN(-EFAULT);
2326
2327         RETURN(0);
2328 }