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 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/llite_lib.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/module.h>
  47 #include <linux/types.h>
  48 #include <linux/version.h>
  49 #include <linux/mm.h>
  50
  51 #include <lustre_lite.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 cfs_mem_cache_t *ll_file_data_slab;
  63
  64 CFS_LIST_HEAD(ll_super_blocks);
  65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  66
  67 #ifndef MS_HAS_NEW_AOPS
  68 extern struct address_space_operations ll_aops;
  69 extern struct address_space_operations ll_dir_aops;
  70 #else
  71 extern struct address_space_operations_ext ll_aops;
  72 extern struct address_space_operations_ext ll_dir_aops;
  73 #endif
  74
  75 #ifndef log2
  76 #define log2(n) cfs_ffz(~(n))
  77 #endif
  78
  79 static struct ll_sb_info *ll_init_sbi(void)
  80 {
  81         struct ll_sb_info *sbi = NULL;
  82         unsigned long pages;
  83         struct sysinfo si;
  84         class_uuid_t uuid;
  85         int i;
  86         ENTRY;
  87
  88         OBD_ALLOC(sbi, sizeof(*sbi));
  89         if (!sbi)
  90                 RETURN(NULL);
  91
  92         cfs_spin_lock_init(&sbi->ll_lock);
  93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  95         cfs_spin_lock_init(&sbi->ll_process_lock);
  96         sbi->ll_rw_stats_on = 0;
  97
  98         si_meminfo(&si);
  99         pages = si.totalram - si.totalhigh;
 100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 101 #ifdef HAVE_BGL_SUPPORT
 102                 sbi->ll_async_page_max = pages / 4;
 103 #else
 104                 sbi->ll_async_page_max = pages / 2;
 105 #endif
 106         } else {
 107                 sbi->ll_async_page_max = (pages / 4) * 3;
 108         }
 109
 110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 111                                            SBI_DEFAULT_READAHEAD_MAX);
 112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 117
 118         ll_generate_random_uuid(uuid);
 119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 121
 122         cfs_spin_lock(&ll_sb_lock);
 123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 124         cfs_spin_unlock(&ll_sb_lock);
 125
 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
 321         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 322          *         retval = in_file->f_op->sendfile(...);
 323          *         if (*ppos > max)
 324          *                 retval = -EOVERFLOW;
 325          *
 326          * it will check if *ppos is greater than max. However, max equals to
 327          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 328          * has been defined as a signed long long integer in linux kernel. */
 329 #if BITS_PER_LONG == 64
 330         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 331 #else
 332         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 333 #endif
 334         sbi->ll_namelen = osfs->os_namelen;
 335         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 336
 337         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 338             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 339                 LCONSOLE_INFO("Disabling user_xattr feature because "
 340                               "it is not supported on the server\n");
 341                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags |= MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags |= LL_SBI_ACL;
 349         } else {
 350                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 351 #ifdef MS_POSIXACL
 352                 sb->s_flags &= ~MS_POSIXACL;
 353 #endif
 354                 sbi->ll_flags &= ~LL_SBI_ACL;
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 358                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 359                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 360                         LCONSOLE_INFO("client is set as remote by default.\n");
 361                 }
 362         } else {
 363                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 364                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 365                         LCONSOLE_INFO("client claims to be remote, but server "
 366                                       "rejected, forced to be local.\n");
 367                 }
 368         }
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 371                 LCONSOLE_INFO("client enabled MDS capability!\n");
 372                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 373         }
 374
 375         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 376                 LCONSOLE_INFO("client enabled OSS capability!\n");
 377                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 378         }
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 381                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 382
 383         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 384                 sbi->ll_md_brw_size = data->ocd_brw_size;
 385         else
 386                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 387
 388         obd = class_name2obd(dt);
 389         if (!obd) {
 390                 CERROR("DT %s: not setup or attached\n", dt);
 391                 GOTO(out_md_fid, err = -ENODEV);
 392         }
 393
 394         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 395                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 396                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 397                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 398                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 399                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 400                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 401                                   OBD_CONNECT_MAXBYTES;
 402
 403         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 404                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 405
 406         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 407                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 408                  * disabled by default, because it can still be enabled on the
 409                  * fly via /proc. As a consequence, we still need to come to an
 410                  * agreement on the supported algorithms at connect time */
 411                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 412
 413                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 414                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 415                 else
 416                         data->ocd_cksum_types = cksum_types_supported();
 417         }
 418
 419 #ifdef HAVE_LRU_RESIZE_SUPPORT
 420         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 421 #endif
 422         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 423                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 424
 425         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 426                "ocd_grant: %d\n", data->ocd_connect_flags,
 427                data->ocd_version, data->ocd_grant);
 428
 429         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 430         obd->obd_upcall.onu_upcall = cl_ocd_update;
 431
 432         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 433
 434         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 435         if (err == -EBUSY) {
 436                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 437                                    "recovery, of which this client is not a "
 438                                    "part.  Please wait for recovery to "
 439                                    "complete, abort, or time out.\n", dt);
 440                 GOTO(out_md_fid, err);
 441         } else if (err) {
 442                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 443                 GOTO(out_md_fid, err);
 444         }
 445
 446         err = obd_fid_init(sbi->ll_dt_exp);
 447         if (err) {
 448                 CERROR("Can't init data layer FID infrastructure, "
 449                        "rc %d\n", err);
 450                 GOTO(out_dt, err);
 451         }
 452
 453         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 454         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 455         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 456         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 457         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 458
 459         fid_zero(&sbi->ll_root_fid);
 460         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 461         if (err) {
 462                 CERROR("cannot mds_connect: rc = %d\n", err);
 463                 GOTO(out_lock_cn_cb, err);
 464         }
 465         if (!fid_is_sane(&sbi->ll_root_fid)) {
 466                 CERROR("Invalid root fid during mount\n");
 467                 GOTO(out_lock_cn_cb, err = -EINVAL);
 468         }
 469         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 470
 471         sb->s_op = &lustre_super_operations;
 472 #if THREAD_SIZE >= 8192
 473         sb->s_export_op = &lustre_export_operations;
 474 #endif
 475
 476         /* make root inode
 477          * XXX: move this to after cbd setup? */
 478         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 479         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 480                 valid |= OBD_MD_FLRMTPERM;
 481         else if (sbi->ll_flags & LL_SBI_ACL)
 482                 valid |= OBD_MD_FLACL;
 483
 484         OBD_ALLOC_PTR(op_data);
 485         if (op_data == NULL)
 486                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 487
 488         op_data->op_fid1 = sbi->ll_root_fid;
 489         op_data->op_mode = 0;
 490         op_data->op_capa1 = oc;
 491         op_data->op_valid = valid;
 492
 493         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 494         if (oc)
 495                 capa_put(oc);
 496         OBD_FREE_PTR(op_data);
 497         if (err) {
 498                 CERROR("md_getattr failed for root: rc = %d\n", err);
 499                 GOTO(out_lock_cn_cb, err);
 500         }
 501         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 502                                sbi->ll_md_exp, &lmd);
 503         if (err) {
 504                 CERROR("failed to understand root inode md: rc = %d\n", err);
 505                 ptlrpc_req_finished (request);
 506                 GOTO(out_lock_cn_cb, err);
 507         }
 508
 509         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 510         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 511         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 512         ptlrpc_req_finished(request);
 513
 514         if (root == NULL || IS_ERR(root)) {
 515                 if (lmd.lsm)
 516                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 517 #ifdef CONFIG_FS_POSIX_ACL
 518                 if (lmd.posix_acl) {
 519                         posix_acl_release(lmd.posix_acl);
 520                         lmd.posix_acl = NULL;
 521                 }
 522 #endif
 523                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 524                 root = NULL;
 525                 CERROR("lustre_lite: bad iget4 for root\n");
 526                 GOTO(out_root, err);
 527         }
 528
 529         err = ll_close_thread_start(&sbi->ll_lcq);
 530         if (err) {
 531                 CERROR("cannot start close thread: rc %d\n", err);
 532                 GOTO(out_root, err);
 533         }
 534
 535 #ifdef CONFIG_FS_POSIX_ACL
 536         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 537                 rct_init(&sbi->ll_rct);
 538                 et_init(&sbi->ll_et);
 539         }
 540 #endif
 541
 542         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 543         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 544                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 545                                  NULL);
 546         cl_sb_init(sb);
 547
 548         sb->s_root = d_alloc_root(root);
 549         sb->s_root->d_op = &ll_d_root_ops;
 550
 551         sbi->ll_sdev_orig = sb->s_dev;
 552
 553         /* We set sb->s_dev equal on all lustre clients in order to support
 554          * NFS export clustering.  NFSD requires that the FSID be the same
 555          * on all clients. */
 556         /* s_dev is also used in lt_compare() to compare two fs, but that is
 557          * only a node-local comparison. */
 558         uuid = obd_get_uuid(sbi->ll_md_exp);
 559         if (uuid != NULL)
 560                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 561         sbi->ll_mnt = mnt;
 562
 563         if (data != NULL)
 564                 OBD_FREE_PTR(data);
 565         if (osfs != NULL)
 566                 OBD_FREE_PTR(osfs);
 567
 568         RETURN(err);
 569 out_root:
 570         if (root)
 571                 iput(root);
 572 out_lock_cn_cb:
 573         obd_fid_fini(sbi->ll_dt_exp);
 574 out_dt:
 575         obd_disconnect(sbi->ll_dt_exp);
 576         sbi->ll_dt_exp = NULL;
 577 out_md_fid:
 578         obd_fid_fini(sbi->ll_md_exp);
 579 out_md:
 580         obd_disconnect(sbi->ll_md_exp);
 581         sbi->ll_md_exp = NULL;
 582 out:
 583         if (data != NULL)
 584                 OBD_FREE_PTR(data);
 585         if (osfs != NULL)
 586                 OBD_FREE_PTR(osfs);
 587         lprocfs_unregister_mountpoint(sbi);
 588         return err;
 589 }
 590
 591 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 592 {
 593         int size, rc;
 594
 595         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 596         size = sizeof(int);
 597         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 598                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 599         if (rc)
 600                 CERROR("Get max mdsize error rc %d \n", rc);
 601
 602         RETURN(rc);
 603 }
 604
 605 void ll_dump_inode(struct inode *inode)
 606 {
 607         struct list_head *tmp;
 608         int dentry_count = 0;
 609
 610         LASSERT(inode != NULL);
 611
 612         list_for_each(tmp, &inode->i_dentry)
 613                 dentry_count++;
 614
 615         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 616                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 617                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 618 }
 619
 620 void lustre_dump_dentry(struct dentry *dentry, int recur)
 621 {
 622         struct list_head *tmp;
 623         int subdirs = 0;
 624
 625         LASSERT(dentry != NULL);
 626
 627         list_for_each(tmp, &dentry->d_subdirs)
 628                 subdirs++;
 629
 630         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 631                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 632                dentry->d_name.len, dentry->d_name.name,
 633                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 634                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 635                dentry->d_flags, dentry->d_fsdata, subdirs);
 636         if (dentry->d_inode != NULL)
 637                 ll_dump_inode(dentry->d_inode);
 638
 639         if (recur == 0)
 640                 return;
 641
 642        list_for_each(tmp, &dentry->d_subdirs) {
 643                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 644                 lustre_dump_dentry(d, recur - 1);
 645         }
 646 }
 647
 648 void client_common_put_super(struct super_block *sb)
 649 {
 650         struct ll_sb_info *sbi = ll_s2sbi(sb);
 651         ENTRY;
 652
 653 #ifdef CONFIG_FS_POSIX_ACL
 654         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 655                 et_fini(&sbi->ll_et);
 656                 rct_fini(&sbi->ll_rct);
 657         }
 658 #endif
 659
 660         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 661
 662         ll_close_thread_shutdown(sbi->ll_lcq);
 663
 664         cl_sb_fini(sb);
 665
 666         cfs_list_del(&sbi->ll_conn_chain);
 667
 668         obd_fid_fini(sbi->ll_dt_exp);
 669         obd_disconnect(sbi->ll_dt_exp);
 670         sbi->ll_dt_exp = NULL;
 671
 672         lprocfs_unregister_mountpoint(sbi);
 673
 674         obd_fid_fini(sbi->ll_md_exp);
 675         obd_disconnect(sbi->ll_md_exp);
 676         sbi->ll_md_exp = NULL;
 677
 678         EXIT;
 679 }
 680
 681 void ll_kill_super(struct super_block *sb)
 682 {
 683         struct ll_sb_info *sbi;
 684
 685         ENTRY;
 686
 687         /* not init sb ?*/
 688         if (!(sb->s_flags & MS_ACTIVE))
 689                 return;
 690
 691         sbi = ll_s2sbi(sb);
 692         /* we need restore s_dev from changed for clustred NFS before put_super
 693          * because new kernels have cached s_dev and change sb->s_dev in
 694          * put_super not affected real removing devices */
 695         if (sbi)
 696                 sb->s_dev = sbi->ll_sdev_orig;
 697         EXIT;
 698 }
 699
 700 char *ll_read_opt(const char *opt, char *data)
 701 {
 702         char *value;
 703         char *retval;
 704         ENTRY;
 705
 706         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 707         if (strncmp(opt, data, strlen(opt)))
 708                 RETURN(NULL);
 709         if ((value = strchr(data, '=')) == NULL)
 710                 RETURN(NULL);
 711
 712         value++;
 713         OBD_ALLOC(retval, strlen(value) + 1);
 714         if (!retval) {
 715                 CERROR("out of memory!\n");
 716                 RETURN(NULL);
 717         }
 718
 719         memcpy(retval, value, strlen(value)+1);
 720         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 721         RETURN(retval);
 722 }
 723
 724 static inline int ll_set_opt(const char *opt, char *data, int fl)
 725 {
 726         if (strncmp(opt, data, strlen(opt)) != 0)
 727                 return(0);
 728         else
 729                 return(fl);
 730 }
 731
 732 /* non-client-specific mount options are parsed in lmd_parse */
 733 static int ll_options(char *options, int *flags)
 734 {
 735         int tmp;
 736         char *s1 = options, *s2;
 737         ENTRY;
 738
 739         if (!options)
 740                 RETURN(0);
 741
 742         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 743
 744         while (*s1) {
 745                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 746                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 747                 if (tmp) {
 748                         *flags |= tmp;
 749                         goto next;
 750                 }
 751                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 752                 if (tmp) {
 753                         *flags |= tmp;
 754                         goto next;
 755                 }
 756                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 757                 if (tmp) {
 758                         *flags |= tmp;
 759                         goto next;
 760                 }
 761                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 762                 if (tmp) {
 763                         *flags &= ~tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 767                 if (tmp) {
 768                         *flags |= tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 772                 if (tmp) {
 773                         *flags &= ~tmp;
 774                         goto next;
 775                 }
 776 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 777                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 778                 if (tmp) {
 779                         /* Ignore deprecated mount option.  The client will
 780                          * always try to mount with ACL support, whether this
 781                          * is used depends on whether server supports it. */
 782                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 783                                                   "mount option 'acl'.\n");
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 787                 if (tmp) {
 788                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 789                                                   "mount option 'noacl'.\n");
 790                         goto next;
 791                 }
 792 #else
 793 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 794 #endif
 795                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 796                 if (tmp) {
 797                         *flags |= tmp;
 798                         goto next;
 799                 }
 800
 801                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 827                 if (tmp) {
 828                         *flags &= ~tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 832                 if (tmp) {
 833                         *flags |= tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841
 842                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 843                                    s1);
 844                 RETURN(-EINVAL);
 845
 846 next:
 847                 /* Find next opt */
 848                 s2 = strchr(s1, ',');
 849                 if (s2 == NULL)
 850                         break;
 851                 s1 = s2 + 1;
 852         }
 853         RETURN(0);
 854 }
 855
 856 void ll_lli_init(struct ll_inode_info *lli)
 857 {
 858         lli->lli_inode_magic = LLI_INODE_MAGIC;
 859         lli->lli_flags = 0;
 860         lli->lli_ioepoch = 0;
 861         cfs_spin_lock_init(&lli->lli_lock);
 862         lli->lli_posix_acl = NULL;
 863         lli->lli_remote_perms = NULL;
 864         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 865         /* Do not set lli_fid, it has been initialized already. */
 866         fid_zero(&lli->lli_pfid);
 867         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 868         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 869         cfs_atomic_set(&lli->lli_open_count, 0);
 870         lli->lli_mds_capa = NULL;
 871         lli->lli_rmtperm_time = 0;
 872         lli->lli_pending_och = NULL;
 873         lli->lli_mds_read_och = NULL;
 874         lli->lli_mds_write_och = NULL;
 875         lli->lli_mds_exec_och = NULL;
 876         lli->lli_open_fd_read_count = 0;
 877         lli->lli_open_fd_write_count = 0;
 878         lli->lli_open_fd_exec_count = 0;
 879         cfs_sema_init(&lli->lli_och_sem, 1);
 880         cfs_spin_lock_init(&lli->lli_agl_lock);
 881         lli->lli_smd = NULL;
 882         lli->lli_clob = NULL;
 883
 884         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 885         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 886                 cfs_sema_init(&lli->lli_readdir_sem, 1);
 887                 lli->lli_opendir_key = NULL;
 888                 lli->lli_sai = NULL;
 889                 lli->lli_sa_pos = 0;
 890                 lli->lli_def_acl = NULL;
 891                 cfs_spin_lock_init(&lli->lli_sa_lock);
 892                 lli->lli_opendir_pid = 0;
 893         } else {
 894                 cfs_sema_init(&lli->lli_size_sem, 1);
 895                 lli->lli_size_sem_owner = NULL;
 896                 lli->lli_symlink_name = NULL;
 897                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 898                 cfs_init_rwsem(&lli->lli_trunc_sem);
 899                 cfs_sema_init(&lli->lli_write_sem, 1);
 900                 lli->lli_async_rc = 0;
 901                 lli->lli_write_rc = 0;
 902                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 903                 lli->lli_glimpse_time = 0;
 904                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 905                 lli->lli_agl_index = 0;
 906         }
 907 }
 908
 909 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 910 {
 911 #ifdef HAVE_BDI_REGISTER
 912         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 913
 914 #ifdef HAVE_BDI_NAME
 915         bdi->name = "lustre";
 916 #endif
 917         return bdi_register(bdi, NULL, "lustre-%d",
 918                             atomic_inc_return(&ll_bdi_num));
 919 #else
 920         return 0;
 921 #endif
 922 }
 923
 924 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 925 {
 926         struct lustre_profile *lprof = NULL;
 927         struct lustre_sb_info *lsi = s2lsi(sb);
 928         struct ll_sb_info *sbi;
 929         char  *dt = NULL, *md = NULL;
 930         char  *profilenm = get_profile_name(sb);
 931         struct config_llog_instance *cfg;
 932         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 933         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 934         int    err;
 935         ENTRY;
 936
 937         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 938
 939         OBD_ALLOC_PTR(cfg);
 940         if (cfg == NULL)
 941                 RETURN(-ENOMEM);
 942
 943         cfs_module_get();
 944
 945         /* client additional sb info */
 946         lsi->lsi_llsbi = sbi = ll_init_sbi();
 947         if (!sbi) {
 948                 cfs_module_put(THIS_MODULE);
 949                 OBD_FREE_PTR(cfg);
 950                 RETURN(-ENOMEM);
 951         }
 952
 953         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 954         if (err)
 955                 GOTO(out_free, err);
 956
 957         err = ll_bdi_init(&lsi->lsi_bdi);
 958         if (err)
 959                 GOTO(out_free, err);
 960         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 961         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 962         err = ll_bdi_register(&lsi->lsi_bdi);
 963         if (err)
 964                 GOTO(out_free, err);
 965
 966 #ifdef HAVE_SB_BDI
 967         sb->s_bdi = &lsi->lsi_bdi;
 968 #endif
 969
 970         /* Generate a string unique to this super, in case some joker tries
 971            to mount the same fs at two mount points.
 972            Use the address of the super itself.*/
 973         cfg->cfg_instance = sb;
 974         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 975
 976         /* set up client obds */
 977         err = lustre_process_log(sb, profilenm, cfg);
 978         if (err < 0) {
 979                 CERROR("Unable to process log: %d\n", err);
 980                 GOTO(out_free, err);
 981         }
 982
 983         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 984         lprof = class_get_profile(profilenm);
 985         if (lprof == NULL) {
 986                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 987                                    " read from the MGS.  Does that filesystem "
 988                                    "exist?\n", profilenm);
 989                 GOTO(out_free, err = -EINVAL);
 990         }
 991         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 992                lprof->lp_md, lprof->lp_dt);
 993
 994         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 995         if (!dt)
 996                 GOTO(out_free, err = -ENOMEM);
 997         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 998
 999         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1000         if (!md)
1001                 GOTO(out_free, err = -ENOMEM);
1002         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1003
1004         /* connections, registrations, sb setup */
1005         err = client_common_fill_super(sb, md, dt, mnt);
1006
1007 out_free:
1008         if (md)
1009                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1010         if (dt)
1011                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1012         if (err)
1013                 ll_put_super(sb);
1014         else
1015                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1016
1017         OBD_FREE_PTR(cfg);
1018         RETURN(err);
1019 } /* ll_fill_super */
1020
1021
1022 void lu_context_keys_dump(void);
1023
1024 void ll_put_super(struct super_block *sb)
1025 {
1026         struct config_llog_instance cfg;
1027         struct obd_device *obd;
1028         struct lustre_sb_info *lsi = s2lsi(sb);
1029         struct ll_sb_info *sbi = ll_s2sbi(sb);
1030         char *profilenm = get_profile_name(sb);
1031         int force = 1, next;
1032         ENTRY;
1033
1034         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1035
1036         ll_print_capa_stat(sbi);
1037
1038         cfg.cfg_instance = sb;
1039         lustre_end_log(sb, profilenm, &cfg);
1040
1041         if (sbi->ll_md_exp) {
1042                 obd = class_exp2obd(sbi->ll_md_exp);
1043                 if (obd)
1044                         force = obd->obd_force;
1045         }
1046
1047         /* We need to set force before the lov_disconnect in
1048            lustre_common_put_super, since l_d cleans up osc's as well. */
1049         if (force) {
1050                 next = 0;
1051                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1052                                                      &next)) != NULL) {
1053                         obd->obd_force = force;
1054                 }
1055         }
1056
1057         if (sbi->ll_lcq) {
1058                 /* Only if client_common_fill_super succeeded */
1059                 client_common_put_super(sb);
1060         }
1061
1062         next = 0;
1063         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1064                 class_manual_cleanup(obd);
1065         }
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         LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1083
1084         cfs_module_put(THIS_MODULE);
1085
1086         EXIT;
1087 } /* client_put_super */
1088
1089 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1090 {
1091         struct inode *inode = NULL;
1092         /* NOTE: we depend on atomic igrab() -bzzz */
1093         lock_res_and_lock(lock);
1094         if (lock->l_ast_data) {
1095                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1096                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1097                         inode = igrab(lock->l_ast_data);
1098                 } else {
1099                         inode = lock->l_ast_data;
1100                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1101                                          D_WARNING, lock, "l_ast_data %p is "
1102                                          "bogus: magic %08x", lock->l_ast_data,
1103                                          lli->lli_inode_magic);
1104                         inode = NULL;
1105                 }
1106         }
1107         unlock_res_and_lock(lock);
1108         return inode;
1109 }
1110
1111 static int null_if_equal(struct ldlm_lock *lock, void *data)
1112 {
1113         if (data == lock->l_ast_data) {
1114                 lock->l_ast_data = NULL;
1115
1116                 if (lock->l_req_mode != lock->l_granted_mode)
1117                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1118         }
1119
1120         return LDLM_ITER_CONTINUE;
1121 }
1122
1123 void ll_clear_inode(struct inode *inode)
1124 {
1125         struct ll_inode_info *lli = ll_i2info(inode);
1126         struct ll_sb_info *sbi = ll_i2sbi(inode);
1127         ENTRY;
1128
1129         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1130                inode->i_generation, inode);
1131
1132         if (S_ISDIR(inode->i_mode)) {
1133                 /* these should have been cleared in ll_file_release */
1134                 LASSERT(lli->lli_opendir_key == NULL);
1135                 LASSERT(lli->lli_sai == NULL);
1136                 LASSERT(lli->lli_opendir_pid == 0);
1137         }
1138
1139         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1140         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1141                          null_if_equal, inode);
1142
1143         LASSERT(!lli->lli_open_fd_write_count);
1144         LASSERT(!lli->lli_open_fd_read_count);
1145         LASSERT(!lli->lli_open_fd_exec_count);
1146
1147         if (lli->lli_mds_write_och)
1148                 ll_md_real_close(inode, FMODE_WRITE);
1149         if (lli->lli_mds_exec_och)
1150                 ll_md_real_close(inode, FMODE_EXEC);
1151         if (lli->lli_mds_read_och)
1152                 ll_md_real_close(inode, FMODE_READ);
1153
1154         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1155                 OBD_FREE(lli->lli_symlink_name,
1156                          strlen(lli->lli_symlink_name) + 1);
1157                 lli->lli_symlink_name = NULL;
1158         }
1159
1160         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1161                 LASSERT(lli->lli_posix_acl == NULL);
1162                 if (lli->lli_remote_perms) {
1163                         free_rmtperm_hash(lli->lli_remote_perms);
1164                         lli->lli_remote_perms = NULL;
1165                 }
1166         }
1167 #ifdef CONFIG_FS_POSIX_ACL
1168         else if (lli->lli_posix_acl) {
1169                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1170                 LASSERT(lli->lli_remote_perms == NULL);
1171                 posix_acl_release(lli->lli_posix_acl);
1172                 lli->lli_posix_acl = NULL;
1173         }
1174 #endif
1175         lli->lli_inode_magic = LLI_INODE_DEAD;
1176
1177         ll_clear_inode_capas(inode);
1178         if (!S_ISDIR(inode->i_mode))
1179                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1180
1181         /*
1182          * XXX This has to be done before lsm is freed below, because
1183          * cl_object still uses inode lsm.
1184          */
1185         cl_inode_fini(inode);
1186
1187         if (lli->lli_smd) {
1188                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1189                 lli->lli_smd = NULL;
1190         }
1191
1192
1193         EXIT;
1194 }
1195
1196 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1197                   struct md_open_data **mod)
1198 {
1199         struct lustre_md md;
1200         struct ll_sb_info *sbi = ll_i2sbi(inode);
1201         struct ptlrpc_request *request = NULL;
1202         int rc;
1203         ENTRY;
1204
1205         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1206                                      LUSTRE_OPC_ANY, NULL);
1207         if (IS_ERR(op_data))
1208                 RETURN(PTR_ERR(op_data));
1209
1210         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1211                         &request, mod);
1212         if (rc) {
1213                 ptlrpc_req_finished(request);
1214                 if (rc == -ENOENT) {
1215                         inode->i_nlink = 0;
1216                         /* Unlinked special device node? Or just a race?
1217                          * Pretend we done everything. */
1218                         if (!S_ISREG(inode->i_mode) &&
1219                             !S_ISDIR(inode->i_mode))
1220                                 rc = inode_setattr(inode, &op_data->op_attr);
1221                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1222                         CERROR("md_setattr fails: rc = %d\n", rc);
1223                 }
1224                 RETURN(rc);
1225         }
1226
1227         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1228                               sbi->ll_md_exp, &md);
1229         if (rc) {
1230                 ptlrpc_req_finished(request);
1231                 RETURN(rc);
1232         }
1233
1234         /* We call inode_setattr to adjust timestamps.
1235          * If there is at least some data in file, we cleared ATTR_SIZE
1236          * above to avoid invoking vmtruncate, otherwise it is important
1237          * to call vmtruncate in inode_setattr to update inode->i_size
1238          * (bug 6196) */
1239         rc = inode_setattr(inode, &op_data->op_attr);
1240
1241         /* Extract epoch data if obtained. */
1242         op_data->op_handle = md.body->handle;
1243         op_data->op_ioepoch = md.body->ioepoch;
1244
1245         ll_update_inode(inode, &md);
1246         ptlrpc_req_finished(request);
1247
1248         RETURN(rc);
1249 }
1250
1251 /* Close IO epoch and send Size-on-MDS attribute update. */
1252 static int ll_setattr_done_writing(struct inode *inode,
1253                                    struct md_op_data *op_data,
1254                                    struct md_open_data *mod)
1255 {
1256         struct ll_inode_info *lli = ll_i2info(inode);
1257         int rc = 0;
1258         ENTRY;
1259
1260         LASSERT(op_data != NULL);
1261         if (!S_ISREG(inode->i_mode))
1262                 RETURN(0);
1263
1264         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1265                op_data->op_ioepoch, PFID(&lli->lli_fid));
1266
1267         op_data->op_flags = MF_EPOCH_CLOSE;
1268         ll_done_writing_attr(inode, op_data);
1269         ll_pack_inode2opdata(inode, op_data, NULL);
1270
1271         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1272         if (rc == -EAGAIN) {
1273                 /* MDS has instructed us to obtain Size-on-MDS attribute
1274                  * from OSTs and send setattr to back to MDS. */
1275                 rc = ll_som_update(inode, op_data);
1276         } else if (rc) {
1277                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1278                        inode->i_ino, rc);
1279         }
1280         RETURN(rc);
1281 }
1282
1283 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1284 {
1285         struct obd_capa *capa;
1286         int rc;
1287
1288         if (attr->ia_valid & ATTR_SIZE)
1289                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1290         else
1291                 capa = ll_mdscapa_get(inode);
1292
1293         rc = cl_setattr_ost(inode, attr, capa);
1294
1295         if (attr->ia_valid & ATTR_SIZE)
1296                 ll_truncate_free_capa(capa);
1297         else
1298                 capa_put(capa);
1299
1300         return rc;
1301 }
1302
1303 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1304  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1305  * keep these values until such a time that objects are allocated for it.
1306  * We do the MDS operations first, as it is checking permissions for us.
1307  * We don't to the MDS RPC if there is nothing that we want to store there,
1308  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1309  * going to do an RPC anyways.
1310  *
1311  * If we are doing a truncate, we will send the mtime and ctime updates
1312  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1313  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1314  * at the same time.
1315  */
1316 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1317 {
1318         struct ll_inode_info *lli = ll_i2info(inode);
1319         struct lov_stripe_md *lsm = lli->lli_smd;
1320         struct md_op_data *op_data = NULL;
1321         struct md_open_data *mod = NULL;
1322         int ia_valid = attr->ia_valid;
1323         int rc = 0, rc1 = 0;
1324         ENTRY;
1325
1326         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1327                attr->ia_valid);
1328         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1329
1330         if (ia_valid & ATTR_SIZE) {
1331                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1332                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1333                                attr->ia_size, ll_file_maxbytes(inode));
1334                         RETURN(-EFBIG);
1335                 }
1336
1337                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1338         }
1339
1340         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1341         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1342                 if (cfs_curproc_fsuid() != inode->i_uid &&
1343                     !cfs_capable(CFS_CAP_FOWNER))
1344                         RETURN(-EPERM);
1345         }
1346
1347         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1348         if (attr->ia_valid & ATTR_CTIME) {
1349                 attr->ia_ctime = CFS_CURRENT_TIME;
1350                 attr->ia_valid |= ATTR_CTIME_SET;
1351         }
1352         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1353                 attr->ia_atime = CFS_CURRENT_TIME;
1354                 attr->ia_valid |= ATTR_ATIME_SET;
1355         }
1356         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1357                 attr->ia_mtime = CFS_CURRENT_TIME;
1358                 attr->ia_valid |= ATTR_MTIME_SET;
1359         }
1360
1361         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1362                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1363                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1364                        cfs_time_current_sec());
1365
1366         /* NB: ATTR_SIZE will only be set after this point if the size
1367          * resides on the MDS, ie, this file has no objects. */
1368         if (lsm)
1369                 attr->ia_valid &= ~ATTR_SIZE;
1370
1371         /* We always do an MDS RPC, even if we're only changing the size;
1372          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1373
1374         OBD_ALLOC_PTR(op_data);
1375         if (op_data == NULL)
1376                 RETURN(-ENOMEM);
1377
1378         UNLOCK_INODE_MUTEX(inode);
1379         if (ia_valid & ATTR_SIZE)
1380                 UP_WRITE_I_ALLOC_SEM(inode);
1381         if (!S_ISDIR(inode->i_mode))
1382                 cfs_down_write(&lli->lli_trunc_sem);
1383         LOCK_INODE_MUTEX(inode);
1384         if (ia_valid & ATTR_SIZE)
1385                 DOWN_WRITE_I_ALLOC_SEM(inode);
1386
1387         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1388
1389         /* Open epoch for truncate. */
1390         if (exp_connect_som(ll_i2mdexp(inode)) &&
1391             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1392                 op_data->op_flags = MF_EPOCH_OPEN;
1393
1394         rc = ll_md_setattr(inode, op_data, &mod);
1395         if (rc)
1396                 GOTO(out, rc);
1397
1398         ll_ioepoch_open(lli, op_data->op_ioepoch);
1399         if (!lsm || !S_ISREG(inode->i_mode)) {
1400                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1401                 GOTO(out, rc = 0);
1402         }
1403
1404         if (ia_valid & ATTR_SIZE)
1405                 attr->ia_valid |= ATTR_SIZE;
1406         if (ia_valid & (ATTR_SIZE |
1407                         ATTR_ATIME | ATTR_ATIME_SET |
1408                         ATTR_MTIME | ATTR_MTIME_SET))
1409                 /* on truncate and utimes send attributes to osts, setting
1410                  * mtime/atime to past will be performed under PW 0:EOF extent
1411                  * lock (new_size:EOF for truncate)
1412                  * it may seem excessive to send mtime/atime updates to osts
1413                  * when not setting times to past, but it is necessary due to
1414                  * possible time de-synchronization */
1415                 rc = ll_setattr_ost(inode, attr);
1416         EXIT;
1417 out:
1418         if (op_data) {
1419                 if (op_data->op_ioepoch)
1420                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1421                 ll_finish_md_op_data(op_data);
1422         }
1423         if (!S_ISDIR(inode->i_mode))
1424                 cfs_up_write(&lli->lli_trunc_sem);
1425         return rc ? rc : rc1;
1426 }
1427
1428 int ll_setattr(struct dentry *de, struct iattr *attr)
1429 {
1430         int mode = de->d_inode->i_mode;
1431
1432         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1433                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1434                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1435
1436         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1437                                (ATTR_SIZE|ATTR_MODE)) &&
1438             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1439              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1440               !(attr->ia_mode & S_ISGID))))
1441                 attr->ia_valid |= ATTR_FORCE;
1442
1443         if ((mode & S_ISUID) &&
1444             !(attr->ia_mode & S_ISUID) &&
1445             !(attr->ia_valid & ATTR_KILL_SUID))
1446                 attr->ia_valid |= ATTR_KILL_SUID;
1447
1448         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1449             !(attr->ia_mode & S_ISGID) &&
1450             !(attr->ia_valid & ATTR_KILL_SGID))
1451                 attr->ia_valid |= ATTR_KILL_SGID;
1452
1453         return ll_setattr_raw(de->d_inode, attr);
1454 }
1455
1456 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1457                        __u64 max_age, __u32 flags)
1458 {
1459         struct ll_sb_info *sbi = ll_s2sbi(sb);
1460         struct obd_statfs obd_osfs;
1461         int rc;
1462         ENTRY;
1463
1464         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1465         if (rc) {
1466                 CERROR("md_statfs fails: rc = %d\n", rc);
1467                 RETURN(rc);
1468         }
1469
1470         osfs->os_type = sb->s_magic;
1471
1472         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1473                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1474
1475         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1476                 flags |= OBD_STATFS_NODELAY;
1477
1478         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1479                               &obd_osfs, max_age, flags);
1480         if (rc) {
1481                 CERROR("obd_statfs fails: rc = %d\n", rc);
1482                 RETURN(rc);
1483         }
1484
1485         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1486                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1487                obd_osfs.os_files);
1488
1489         osfs->os_bsize = obd_osfs.os_bsize;
1490         osfs->os_blocks = obd_osfs.os_blocks;
1491         osfs->os_bfree = obd_osfs.os_bfree;
1492         osfs->os_bavail = obd_osfs.os_bavail;
1493
1494         /* If we don't have as many objects free on the OST as inodes
1495          * on the MDS, we reduce the total number of inodes to
1496          * compensate, so that the "inodes in use" number is correct.
1497          */
1498         if (obd_osfs.os_ffree < osfs->os_ffree) {
1499                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1500                         obd_osfs.os_ffree;
1501                 osfs->os_ffree = obd_osfs.os_ffree;
1502         }
1503
1504         RETURN(rc);
1505 }
1506 #ifndef HAVE_STATFS_DENTRY_PARAM
1507 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1508 {
1509 #else
1510 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1511 {
1512         struct super_block *sb = de->d_sb;
1513 #endif
1514         struct obd_statfs osfs;
1515         int rc;
1516
1517         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1518         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1519
1520         /* Some amount of caching on the client is allowed */
1521         rc = ll_statfs_internal(sb, &osfs,
1522                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1523                                 0);
1524         if (rc)
1525                 return rc;
1526
1527         statfs_unpack(sfs, &osfs);
1528
1529         /* We need to downshift for all 32-bit kernels, because we can't
1530          * tell if the kernel is being called via sys_statfs64() or not.
1531          * Stop before overflowing f_bsize - in which case it is better
1532          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1533         if (sizeof(long) < 8) {
1534                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1535                         sfs->f_bsize <<= 1;
1536
1537                         osfs.os_blocks >>= 1;
1538                         osfs.os_bfree >>= 1;
1539                         osfs.os_bavail >>= 1;
1540                 }
1541         }
1542
1543         sfs->f_blocks = osfs.os_blocks;
1544         sfs->f_bfree = osfs.os_bfree;
1545         sfs->f_bavail = osfs.os_bavail;
1546
1547         return 0;
1548 }
1549
1550 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1551 {
1552         struct ll_inode_info *lli;
1553         struct lov_stripe_md *lsm;
1554
1555         LASSERT(!S_ISDIR(inode->i_mode));
1556
1557         lli = ll_i2info(inode);
1558         LASSERT(lli->lli_size_sem_owner != current);
1559         cfs_down(&lli->lli_size_sem);
1560         LASSERT(lli->lli_size_sem_owner == NULL);
1561         lli->lli_size_sem_owner = current;
1562         lsm = lli->lli_smd;
1563         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1564                  lsm, lock_lsm);
1565         if (lock_lsm)
1566                 lov_stripe_lock(lsm);
1567 }
1568
1569 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1570 {
1571         struct ll_inode_info *lli;
1572         struct lov_stripe_md *lsm;
1573
1574         lli = ll_i2info(inode);
1575         lsm = lli->lli_smd;
1576         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1577                  lsm, unlock_lsm);
1578         if (unlock_lsm)
1579                 lov_stripe_unlock(lsm);
1580         LASSERT(lli->lli_size_sem_owner == current);
1581         lli->lli_size_sem_owner = NULL;
1582         cfs_up(&lli->lli_size_sem);
1583 }
1584
1585 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1586 {
1587         struct ll_inode_info *lli = ll_i2info(inode);
1588         struct mdt_body *body = md->body;
1589         struct lov_stripe_md *lsm = md->lsm;
1590         struct ll_sb_info *sbi = ll_i2sbi(inode);
1591
1592         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1593         if (lsm != NULL) {
1594                 LASSERT(S_ISREG(inode->i_mode));
1595
1596                 cfs_down(&lli->lli_och_sem);
1597                 if (lli->lli_smd == NULL) {
1598                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1599                             lsm->lsm_magic != LOV_MAGIC_V3) {
1600                                 dump_lsm(D_ERROR, lsm);
1601                                 LBUG();
1602                         }
1603                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1604                                lsm, inode->i_ino, inode->i_generation, inode);
1605                         /* cl_file_inode_init must go before lli_smd or a race
1606                          * is possible where client thinks the file has stripes,
1607                          * but lov raid0 is not setup yet and parallel e.g.
1608                          * glimpse would try to use uninitialized lov */
1609                         cl_file_inode_init(inode, md);
1610                         cfs_spin_lock(&lli->lli_lock);
1611                         lli->lli_smd = lsm;
1612                         cfs_spin_unlock(&lli->lli_lock);
1613                         cfs_up(&lli->lli_och_sem);
1614                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1615                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1616                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1617                 } else {
1618                         cfs_up(&lli->lli_och_sem);
1619                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1620                                 lli->lli_smd->lsm_stripe_count ==
1621                                 lsm->lsm_stripe_count);
1622                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1623                                 CERROR("lsm mismatch for inode %ld\n",
1624                                        inode->i_ino);
1625                                 CERROR("lli_smd:\n");
1626                                 dump_lsm(D_ERROR, lli->lli_smd);
1627                                 CERROR("lsm:\n");
1628                                 dump_lsm(D_ERROR, lsm);
1629                                 LBUG();
1630                         }
1631                 }
1632                 if (lli->lli_smd != lsm)
1633                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1634         }
1635
1636         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1637                 if (body->valid & OBD_MD_FLRMTPERM)
1638                         ll_update_remote_perm(inode, md->remote_perm);
1639         }
1640 #ifdef CONFIG_FS_POSIX_ACL
1641         else if (body->valid & OBD_MD_FLACL) {
1642                 cfs_spin_lock(&lli->lli_lock);
1643                 if (lli->lli_posix_acl)
1644                         posix_acl_release(lli->lli_posix_acl);
1645                 lli->lli_posix_acl = md->posix_acl;
1646                 cfs_spin_unlock(&lli->lli_lock);
1647         }
1648 #endif
1649         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1650         inode->i_generation = cl_fid_build_gen(&body->fid1);
1651
1652         if (body->valid & OBD_MD_FLATIME) {
1653                 if (body->atime > LTIME_S(inode->i_atime))
1654                         LTIME_S(inode->i_atime) = body->atime;
1655                 lli->lli_lvb.lvb_atime = body->atime;
1656         }
1657         if (body->valid & OBD_MD_FLMTIME) {
1658                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1659                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1660                                "to "LPU64"\n", inode->i_ino,
1661                                LTIME_S(inode->i_mtime), body->mtime);
1662                         LTIME_S(inode->i_mtime) = body->mtime;
1663                 }
1664                 lli->lli_lvb.lvb_mtime = body->mtime;
1665         }
1666         if (body->valid & OBD_MD_FLCTIME) {
1667                 if (body->ctime > LTIME_S(inode->i_ctime))
1668                         LTIME_S(inode->i_ctime) = body->ctime;
1669                 lli->lli_lvb.lvb_ctime = body->ctime;
1670         }
1671         if (body->valid & OBD_MD_FLMODE)
1672                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1673         if (body->valid & OBD_MD_FLTYPE)
1674                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1675         LASSERT(inode->i_mode != 0);
1676         if (S_ISREG(inode->i_mode)) {
1677                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1678         } else {
1679                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1680         }
1681 #ifdef HAVE_INODE_BLKSIZE
1682         inode->i_blksize = 1<<inode->i_blkbits;
1683 #endif
1684         if (body->valid & OBD_MD_FLUID)
1685                 inode->i_uid = body->uid;
1686         if (body->valid & OBD_MD_FLGID)
1687                 inode->i_gid = body->gid;
1688         if (body->valid & OBD_MD_FLFLAGS)
1689                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1690         if (body->valid & OBD_MD_FLNLINK)
1691                 inode->i_nlink = body->nlink;
1692         if (body->valid & OBD_MD_FLRDEV)
1693                 inode->i_rdev = old_decode_dev(body->rdev);
1694
1695         if (body->valid & OBD_MD_FLID) {
1696                 /* FID shouldn't be changed! */
1697                 if (fid_is_sane(&lli->lli_fid)) {
1698                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1699                                  "Trying to change FID "DFID
1700                                  " to the "DFID", inode %lu/%u(%p)\n",
1701                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1702                                  inode->i_ino, inode->i_generation, inode);
1703                 } else
1704                         lli->lli_fid = body->fid1;
1705         }
1706
1707         LASSERT(fid_seq(&lli->lli_fid) != 0);
1708
1709         if (body->valid & OBD_MD_FLSIZE) {
1710                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1711                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1712                         struct lustre_handle lockh;
1713                         ldlm_mode_t mode;
1714
1715                         /* As it is possible a blocking ast has been processed
1716                          * by this time, we need to check there is an UPDATE
1717                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1718                          * it. */
1719                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1720                                                &lockh);
1721                         if (mode) {
1722                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1723                                                       LLIF_EPOCH_PENDING |
1724                                                       LLIF_SOM_DIRTY)) {
1725                                         CERROR("ino %lu flags %u still has "
1726                                                "size authority! do not trust "
1727                                                "the size got from MDS\n",
1728                                                inode->i_ino, lli->lli_flags);
1729                                 } else {
1730                                         /* Use old size assignment to avoid
1731                                          * deadlock bz14138 & bz14326 */
1732                                         inode->i_size = body->size;
1733                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1734                                 }
1735                                 ldlm_lock_decref(&lockh, mode);
1736                         }
1737                 } else {
1738                         /* Use old size assignment to avoid
1739                          * deadlock bz14138 & bz14326 */
1740                         inode->i_size = body->size;
1741
1742                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1743                                inode->i_ino, (unsigned long long)body->size);
1744                 }
1745
1746                 if (body->valid & OBD_MD_FLBLOCKS)
1747                         inode->i_blocks = body->blocks;
1748         }
1749
1750         if (body->valid & OBD_MD_FLMDSCAPA) {
1751                 LASSERT(md->mds_capa);
1752                 ll_add_capa(inode, md->mds_capa);
1753         }
1754         if (body->valid & OBD_MD_FLOSSCAPA) {
1755                 LASSERT(md->oss_capa);
1756                 ll_add_capa(inode, md->oss_capa);
1757         }
1758 }
1759
1760 void ll_read_inode2(struct inode *inode, void *opaque)
1761 {
1762         struct lustre_md *md = opaque;
1763         struct ll_inode_info *lli = ll_i2info(inode);
1764         ENTRY;
1765
1766         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1767                PFID(&lli->lli_fid), inode);
1768
1769         LASSERT(!lli->lli_smd);
1770
1771         /* Core attributes from the MDS first.  This is a new inode, and
1772          * the VFS doesn't zero times in the core inode so we have to do
1773          * it ourselves.  They will be overwritten by either MDS or OST
1774          * attributes - we just need to make sure they aren't newer. */
1775         LTIME_S(inode->i_mtime) = 0;
1776         LTIME_S(inode->i_atime) = 0;
1777         LTIME_S(inode->i_ctime) = 0;
1778         inode->i_rdev = 0;
1779         ll_update_inode(inode, md);
1780
1781         /* OIDEBUG(inode); */
1782
1783         /* initializing backing dev info. */
1784         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1785
1786
1787         if (S_ISREG(inode->i_mode)) {
1788                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1789                 inode->i_op = &ll_file_inode_operations;
1790                 inode->i_fop = sbi->ll_fop;
1791                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1792                 EXIT;
1793         } else if (S_ISDIR(inode->i_mode)) {
1794                 inode->i_op = &ll_dir_inode_operations;
1795                 inode->i_fop = &ll_dir_operations;
1796                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1797                 EXIT;
1798         } else if (S_ISLNK(inode->i_mode)) {
1799                 inode->i_op = &ll_fast_symlink_inode_operations;
1800                 EXIT;
1801         } else {
1802                 inode->i_op = &ll_special_inode_operations;
1803
1804                 init_special_inode(inode, inode->i_mode,
1805                                    kdev_t_to_nr(inode->i_rdev));
1806
1807                 EXIT;
1808         }
1809 }
1810
1811 void ll_delete_inode(struct inode *inode)
1812 {
1813         struct ll_sb_info *sbi = ll_i2sbi(inode);
1814         int rc;
1815         ENTRY;
1816
1817         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1818         if (rc)
1819                 CERROR("fid_delete() failed, rc %d\n", rc);
1820
1821         truncate_inode_pages(&inode->i_data, 0);
1822
1823         /* Workaround for LU-118 */
1824         if (inode->i_data.nrpages) {
1825                 TREE_READ_LOCK_IRQ(&inode->i_data);
1826                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1827                 LASSERTF(inode->i_data.nrpages == 0,
1828                          "inode=%lu/%u(%p) nrpages=%lu, see "
1829                          "http://jira.whamcloud.com/browse/LU-118\n",
1830                          inode->i_ino, inode->i_generation, inode,
1831                          inode->i_data.nrpages);
1832         }
1833         /* Workaround end */
1834
1835 #ifdef HAVE_SBOPS_EVICT_INODE
1836         ll_clear_inode(inode);
1837         end_writeback(inode);
1838 #else
1839         clear_inode(inode);
1840 #endif
1841
1842         EXIT;
1843 }
1844
1845 int ll_iocontrol(struct inode *inode, struct file *file,
1846                  unsigned int cmd, unsigned long arg)
1847 {
1848         struct ll_sb_info *sbi = ll_i2sbi(inode);
1849         struct ptlrpc_request *req = NULL;
1850         int rc, flags = 0;
1851         ENTRY;
1852
1853         switch(cmd) {
1854         case FSFILT_IOC_GETFLAGS: {
1855                 struct mdt_body *body;
1856                 struct md_op_data *op_data;
1857
1858                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1859                                              0, 0, LUSTRE_OPC_ANY,
1860                                              NULL);
1861                 if (IS_ERR(op_data))
1862                         RETURN(PTR_ERR(op_data));
1863
1864                 op_data->op_valid = OBD_MD_FLFLAGS;
1865                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1866                 ll_finish_md_op_data(op_data);
1867                 if (rc) {
1868                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1869                         RETURN(-abs(rc));
1870                 }
1871
1872                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1873
1874                 flags = body->flags;
1875
1876                 ptlrpc_req_finished(req);
1877
1878                 RETURN(put_user(flags, (int *)arg));
1879         }
1880         case FSFILT_IOC_SETFLAGS: {
1881                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1882                 struct obd_info oinfo = { { { 0 } } };
1883                 struct md_op_data *op_data;
1884
1885                 if (get_user(flags, (int *)arg))
1886                         RETURN(-EFAULT);
1887
1888                 oinfo.oi_md = lsm;
1889                 OBDO_ALLOC(oinfo.oi_oa);
1890                 if (!oinfo.oi_oa)
1891                         RETURN(-ENOMEM);
1892
1893                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1894                                              LUSTRE_OPC_ANY, NULL);
1895                 if (IS_ERR(op_data))
1896                         RETURN(PTR_ERR(op_data));
1897
1898                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1899                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1900                 rc = md_setattr(sbi->ll_md_exp, op_data,
1901                                 NULL, 0, NULL, 0, &req, NULL);
1902                 ll_finish_md_op_data(op_data);
1903                 ptlrpc_req_finished(req);
1904                 if (rc) {
1905                         OBDO_FREE(oinfo.oi_oa);
1906                         RETURN(rc);
1907                 }
1908
1909                 if (lsm == NULL) {
1910                         OBDO_FREE(oinfo.oi_oa);
1911                         GOTO(update_cache, rc);
1912                 }
1913
1914                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1915                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1916                 oinfo.oi_oa->o_flags = flags;
1917                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1918                                        OBD_MD_FLGROUP;
1919                 oinfo.oi_capa = ll_mdscapa_get(inode);
1920                 obdo_from_inode(oinfo.oi_oa, inode,
1921                                 &ll_i2info(inode)->lli_fid, 0);
1922                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1923                 capa_put(oinfo.oi_capa);
1924                 OBDO_FREE(oinfo.oi_oa);
1925                 if (rc) {
1926                         if (rc != -EPERM && rc != -EACCES)
1927                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1928                         RETURN(rc);
1929                 }
1930
1931                 EXIT;
1932 update_cache:
1933                 inode->i_flags = ll_ext_to_inode_flags(flags);
1934                 return 0;
1935         }
1936         default:
1937                 RETURN(-ENOSYS);
1938         }
1939
1940         RETURN(0);
1941 }
1942
1943 int ll_flush_ctx(struct inode *inode)
1944 {
1945         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1946
1947         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1948
1949         obd_set_info_async(sbi->ll_md_exp,
1950                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1951                            0, NULL, NULL);
1952         obd_set_info_async(sbi->ll_dt_exp,
1953                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1954                            0, NULL, NULL);
1955         return 0;
1956 }
1957
1958 /* umount -f client means force down, don't save state */
1959 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1960 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1961 {
1962         struct super_block *sb = vfsmnt->mnt_sb;
1963 #else
1964 void ll_umount_begin(struct super_block *sb)
1965 {
1966 #endif
1967         struct lustre_sb_info *lsi = s2lsi(sb);
1968         struct ll_sb_info *sbi = ll_s2sbi(sb);
1969         struct obd_device *obd;
1970         struct obd_ioctl_data *ioc_data;
1971         ENTRY;
1972
1973 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1974         if (!(flags & MNT_FORCE)) {
1975                 EXIT;
1976                 return;
1977         }
1978 #endif
1979
1980         /* Tell the MGC we got umount -f */
1981         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1982
1983         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1984                sb->s_count, atomic_read(&sb->s_active));
1985
1986         obd = class_exp2obd(sbi->ll_md_exp);
1987         if (obd == NULL) {
1988                 CERROR("Invalid MDC connection handle "LPX64"\n",
1989                        sbi->ll_md_exp->exp_handle.h_cookie);
1990                 EXIT;
1991                 return;
1992         }
1993         obd->obd_force = 1;
1994
1995         obd = class_exp2obd(sbi->ll_dt_exp);
1996         if (obd == NULL) {
1997                 CERROR("Invalid LOV connection handle "LPX64"\n",
1998                        sbi->ll_dt_exp->exp_handle.h_cookie);
1999                 EXIT;
2000                 return;
2001         }
2002         obd->obd_force = 1;
2003
2004         OBD_ALLOC_PTR(ioc_data);
2005         if (ioc_data) {
2006                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2007                               sizeof ioc_data, ioc_data, NULL);
2008
2009                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2010                               sizeof ioc_data, ioc_data, NULL);
2011
2012                 OBD_FREE_PTR(ioc_data);
2013         }
2014
2015
2016         /* Really, we'd like to wait until there are no requests outstanding,
2017          * and then continue.  For now, we just invalidate the requests,
2018          * schedule() and sleep one second if needed, and hope.
2019          */
2020         cfs_schedule();
2021 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2022         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2023                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2024                                                    cfs_time_seconds(1));
2025                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2026                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2027                                       "may try to umount it a bit later\n",
2028                                       atomic_read(&vfsmnt->mnt_count));
2029         }
2030 #endif
2031
2032         EXIT;
2033 }
2034
2035 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2036 {
2037         struct ll_sb_info *sbi = ll_s2sbi(sb);
2038         int err;
2039         __u32 read_only;
2040
2041         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2042                 read_only = *flags & MS_RDONLY;
2043                 err = obd_set_info_async(sbi->ll_md_exp,
2044                                          sizeof(KEY_READ_ONLY),
2045                                          KEY_READ_ONLY, sizeof(read_only),
2046                                          &read_only, NULL);
2047                 if (err) {
2048                         CERROR("Failed to change the read-only flag during "
2049                                "remount: %d\n", err);
2050                         return err;
2051                 }
2052
2053                 if (read_only)
2054                         sb->s_flags |= MS_RDONLY;
2055                 else
2056                         sb->s_flags &= ~MS_RDONLY;
2057         }
2058         return 0;
2059 }
2060
2061 int ll_prep_inode(struct inode **inode,
2062                   struct ptlrpc_request *req,
2063                   struct super_block *sb)
2064 {
2065         struct ll_sb_info *sbi = NULL;
2066         struct lustre_md md;
2067         int rc;
2068         ENTRY;
2069
2070         LASSERT(*inode || sb);
2071         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2072         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2073                               sbi->ll_md_exp, &md);
2074         if (rc)
2075                 RETURN(rc);
2076
2077         if (*inode) {
2078                 ll_update_inode(*inode, &md);
2079         } else {
2080                 LASSERT(sb != NULL);
2081
2082                 /*
2083                  * At this point server returns to client's same fid as client
2084                  * generated for creating. So using ->fid1 is okay here.
2085                  */
2086                 LASSERT(fid_is_sane(&md.body->fid1));
2087
2088                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2089                 if (*inode == NULL || IS_ERR(*inode)) {
2090                         if (md.lsm)
2091                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2092 #ifdef CONFIG_FS_POSIX_ACL
2093                         if (md.posix_acl) {
2094                                 posix_acl_release(md.posix_acl);
2095                                 md.posix_acl = NULL;
2096                         }
2097 #endif
2098                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2099                         *inode = NULL;
2100                         CERROR("new_inode -fatal: rc %d\n", rc);
2101                         GOTO(out, rc);
2102                 }
2103         }
2104
2105 out:
2106         md_free_lustre_md(sbi->ll_md_exp, &md);
2107         RETURN(rc);
2108 }
2109
2110 int ll_obd_statfs(struct inode *inode, void *arg)
2111 {
2112         struct ll_sb_info *sbi = NULL;
2113         struct obd_export *exp;
2114         char *buf = NULL;
2115         struct obd_ioctl_data *data = NULL;
2116         __u32 type;
2117         int len = 0, rc;
2118
2119         if (!inode || !(sbi = ll_i2sbi(inode)))
2120                 GOTO(out_statfs, rc = -EINVAL);
2121
2122         rc = obd_ioctl_getdata(&buf, &len, arg);
2123         if (rc)
2124                 GOTO(out_statfs, rc);
2125
2126         data = (void*)buf;
2127         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2128             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2129                 GOTO(out_statfs, rc = -EINVAL);
2130
2131         if (data->ioc_inllen1 != sizeof(__u32) ||
2132             data->ioc_inllen2 != sizeof(__u32) ||
2133             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2134             data->ioc_plen2 != sizeof(struct obd_uuid))
2135                 GOTO(out_statfs, rc = -EINVAL);
2136
2137         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2138         if (type == LL_STATFS_MDC)
2139                 exp = sbi->ll_md_exp;
2140         else if (type == LL_STATFS_LOV)
2141                 exp = sbi->ll_dt_exp;
2142         else
2143                 GOTO(out_statfs, rc = -ENODEV);
2144
2145         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2146         if (rc)
2147                 GOTO(out_statfs, rc);
2148 out_statfs:
2149         if (buf)
2150                 obd_ioctl_freedata(buf, len);
2151         return rc;
2152 }
2153
2154 int ll_process_config(struct lustre_cfg *lcfg)
2155 {
2156         char *ptr;
2157         void *sb;
2158         struct lprocfs_static_vars lvars;
2159         unsigned long x;
2160         int rc = 0;
2161
2162         lprocfs_llite_init_vars(&lvars);
2163
2164         /* The instance name contains the sb: lustre-client-aacfe000 */
2165         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2166         if (!ptr || !*(++ptr))
2167                 return -EINVAL;
2168         if (sscanf(ptr, "%lx", &x) != 1)
2169                 return -EINVAL;
2170         sb = (void *)x;
2171         /* This better be a real Lustre superblock! */
2172         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2173
2174         /* Note we have not called client_common_fill_super yet, so
2175            proc fns must be able to handle that! */
2176         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2177                                       lcfg, sb);
2178         if (rc > 0)
2179                 rc = 0;
2180         return(rc);
2181 }
2182
2183 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2184 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2185                                        struct inode *i1, struct inode *i2,
2186                                        const char *name, int namelen,
2187                                        int mode, __u32 opc, void *data)
2188 {
2189         LASSERT(i1 != NULL);
2190
2191         if (namelen > ll_i2sbi(i1)->ll_namelen)
2192                 return ERR_PTR(-ENAMETOOLONG);
2193
2194         if (op_data == NULL)
2195                 OBD_ALLOC_PTR(op_data);
2196
2197         if (op_data == NULL)
2198                 return ERR_PTR(-ENOMEM);
2199
2200         ll_i2gids(op_data->op_suppgids, i1, i2);
2201         op_data->op_fid1 = *ll_inode2fid(i1);
2202         op_data->op_capa1 = ll_mdscapa_get(i1);
2203
2204         if (i2) {
2205                 op_data->op_fid2 = *ll_inode2fid(i2);
2206                 op_data->op_capa2 = ll_mdscapa_get(i2);
2207         } else {
2208                 fid_zero(&op_data->op_fid2);
2209                 op_data->op_capa2 = NULL;
2210         }
2211
2212         op_data->op_name = name;
2213         op_data->op_namelen = namelen;
2214         op_data->op_mode = mode;
2215         op_data->op_mod_time = cfs_time_current_sec();
2216         op_data->op_fsuid = cfs_curproc_fsuid();
2217         op_data->op_fsgid = cfs_curproc_fsgid();
2218         op_data->op_cap = cfs_curproc_cap_pack();
2219         op_data->op_bias = MDS_CHECK_SPLIT;
2220         op_data->op_opc = opc;
2221         op_data->op_mds = 0;
2222         op_data->op_data = data;
2223
2224         /* If the file is being opened after mknod() (normally due to NFS)
2225          * try to use the default stripe data from parent directory for
2226          * allocating OST objects.  Try to pass the parent FID to MDS. */
2227         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2228             ll_i2info(i2)->lli_smd == NULL) {
2229                 struct ll_inode_info *lli = ll_i2info(i2);
2230
2231                 cfs_spin_lock(&lli->lli_lock);
2232                 if (likely(lli->lli_smd == NULL &&
2233                            !fid_is_zero(&lli->lli_pfid)))
2234                         op_data->op_fid1 = lli->lli_pfid;
2235                 cfs_spin_unlock(&lli->lli_lock);
2236                 /** We ignore parent's capability temporary. */
2237         }
2238
2239         return op_data;
2240 }
2241
2242 void ll_finish_md_op_data(struct md_op_data *op_data)
2243 {
2244         capa_put(op_data->op_capa1);
2245         capa_put(op_data->op_capa2);
2246         OBD_FREE_PTR(op_data);
2247 }
2248
2249 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2250 {
2251         struct ll_sb_info *sbi;
2252
2253         LASSERT((seq != NULL) && (vfs != NULL));
2254         sbi = ll_s2sbi(vfs->mnt_sb);
2255
2256         if (sbi->ll_flags & LL_SBI_NOLCK)
2257                 seq_puts(seq, ",nolock");
2258
2259         if (sbi->ll_flags & LL_SBI_FLOCK)
2260                 seq_puts(seq, ",flock");
2261
2262         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2263                 seq_puts(seq, ",localflock");
2264
2265         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2266                 seq_puts(seq, ",user_xattr");
2267
2268         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2269                 seq_puts(seq, ",lazystatfs");
2270
2271         RETURN(0);
2272 }