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