lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67 #else
  68 extern struct address_space_operations_ext ll_aops;
  69 extern struct address_space_operations_ext ll_dir_aops;
  70 #endif
  71
  72 #ifndef log2
  73 #define log2(n) cfs_ffz(~(n))
  74 #endif
  75
  76 static struct ll_sb_info *ll_init_sbi(void)
  77 {
  78         struct ll_sb_info *sbi = NULL;
  79         unsigned long pages;
  80         struct sysinfo si;
  81         class_uuid_t uuid;
  82         int i;
  83         ENTRY;
  84
  85         OBD_ALLOC(sbi, sizeof(*sbi));
  86         if (!sbi)
  87                 RETURN(NULL);
  88
  89         cfs_spin_lock_init(&sbi->ll_lock);
  90         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  91         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  92         cfs_spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  98 #ifdef HAVE_BGL_SUPPORT
  99                 sbi->ll_async_page_max = pages / 4;
 100 #else
 101                 sbi->ll_async_page_max = pages / 2;
 102 #endif
 103         } else {
 104                 sbi->ll_async_page_max = (pages / 4) * 3;
 105         }
 106
 107         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 108                                            SBI_DEFAULT_READAHEAD_MAX);
 109         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 110         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 111                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 112         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 113         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 114
 115         ll_generate_random_uuid(uuid);
 116         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 117         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 118
 119         cfs_spin_lock(&ll_sb_lock);
 120         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 121         cfs_spin_unlock(&ll_sb_lock);
 122
 123         sbi->ll_flags |= LL_SBI_VERBOSE;
 124 #ifdef ENABLE_CHECKSUM
 125         sbi->ll_flags |= LL_SBI_CHECKSUM;
 126 #endif
 127
 128 #ifdef HAVE_LRU_RESIZE_SUPPORT
 129         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 130 #endif
 131
 132         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 133                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 134                                    pp_r_hist.oh_lock);
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_w_hist.oh_lock);
 137         }
 138
 139         /* metadata statahead is enabled by default */
 140         sbi->ll_sa_max = LL_SA_RPC_DEF;
 141         cfs_atomic_set(&sbi->ll_sa_total, 0);
 142         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 143         cfs_atomic_set(&sbi->ll_agl_total, 0);
 144         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         if (proc_lustre_fs_root) {
 202                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 203                                                   dt, md);
 204                 if (err < 0)
 205                         CERROR("could not register mount in /proc/fs/lustre\n");
 206         }
 207
 208         /* indicate the features supported by this client */
 209         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 210                                   OBD_CONNECT_ATTRFID  |
 211                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 212                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 213                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 214                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 215                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 216                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 217                                   OBD_CONNECT_EINPROGRESS |
 218                                   OBD_CONNECT_JOBSTATS;
 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
 231         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 232                 /* flag mdc connection as lightweight, only used for test
 233                  * purpose, use with care */
 234                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 235
 236         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 237         data->ocd_version = LUSTRE_VERSION_CODE;
 238
 239         if (sb->s_flags & MS_RDONLY)
 240                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 241         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 242                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 243
 244 #ifdef HAVE_MS_FLOCK_LOCK
 245         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 246         sb->s_flags |= MS_FLOCK_LOCK;
 247 #endif
 248 #ifdef MS_HAS_NEW_AOPS
 249         sb->s_flags |= MS_HAS_NEW_AOPS;
 250 #endif
 251
 252         if (sbi->ll_flags & LL_SBI_FLOCK)
 253                 sbi->ll_fop = &ll_file_operations_flock;
 254         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 255                 sbi->ll_fop = &ll_file_operations;
 256         else
 257                 sbi->ll_fop = &ll_file_operations_noflock;
 258
 259         /* real client */
 260         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 261         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 262                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 263
 264         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 265
 266         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 267         if (err == -EBUSY) {
 268                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 269                                    "recovery, of which this client is not a "
 270                                    "part. Please wait for recovery to complete,"
 271                                    " abort, or time out.\n", md);
 272                 GOTO(out, err);
 273         } else if (err) {
 274                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 275                 GOTO(out, err);
 276         }
 277
 278         err = obd_fid_init(sbi->ll_md_exp);
 279         if (err) {
 280                 CERROR("Can't init metadata layer FID infrastructure, "
 281                        "rc %d\n", err);
 282                 GOTO(out_md, err);
 283         }
 284
 285         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 286                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 287         if (err)
 288                 GOTO(out_md_fid, err);
 289
 290         /* This needs to be after statfs to ensure connect has finished.
 291          * Note that "data" does NOT contain the valid connect reply.
 292          * If connecting to a 1.8 server there will be no LMV device, so
 293          * we can access the MDC export directly and exp_connect_flags will
 294          * be non-zero, but if accessing an upgraded 2.1 server it will
 295          * have the correct flags filled in.
 296          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 297         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 298         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 299             valid != CLIENT_CONNECT_MDT_REQD) {
 300                 char *buf;
 301
 302                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 303                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 304                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 305                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 306                                    "feature(s) needed for correct operation "
 307                                    "of this client (%s). Please upgrade "
 308                                    "server or downgrade client.\n",
 309                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 310                 OBD_FREE(buf, CFS_PAGE_SIZE);
 311                 GOTO(out_md, err = -EPROTO);
 312         }
 313
 314         size = sizeof(*data);
 315         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 316                            KEY_CONN_DATA,  &size, data, NULL);
 317         if (err) {
 318                 CERROR("Get connect data failed: %d \n", err);
 319                 GOTO(out_md, err);
 320         }
 321
 322         LASSERT(osfs->os_bsize);
 323         sb->s_blocksize = osfs->os_bsize;
 324         sb->s_blocksize_bits = log2(osfs->os_bsize);
 325         sb->s_magic = LL_SUPER_MAGIC;
 326         sb->s_maxbytes = MAX_LFS_FILESIZE;
 327         sbi->ll_namelen = osfs->os_namelen;
 328         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 329
 330         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 331             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 332                 LCONSOLE_INFO("Disabling user_xattr feature because "
 333                               "it is not supported on the server\n");
 334                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 335         }
 336
 337         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 338 #ifdef MS_POSIXACL
 339                 sb->s_flags |= MS_POSIXACL;
 340 #endif
 341                 sbi->ll_flags |= LL_SBI_ACL;
 342         } else {
 343                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 344 #ifdef MS_POSIXACL
 345                 sb->s_flags &= ~MS_POSIXACL;
 346 #endif
 347                 sbi->ll_flags &= ~LL_SBI_ACL;
 348         }
 349
 350         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 351                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 352                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 353                         LCONSOLE_INFO("client is set as remote by default.\n");
 354                 }
 355         } else {
 356                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 357                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 358                         LCONSOLE_INFO("client claims to be remote, but server "
 359                                       "rejected, forced to be local.\n");
 360                 }
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 364                 LCONSOLE_INFO("client enabled MDS capability!\n");
 365                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 369                 LCONSOLE_INFO("client enabled OSS capability!\n");
 370                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 374                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 377                 sbi->ll_md_brw_size = data->ocd_brw_size;
 378         else
 379                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 382                 LCONSOLE_INFO("Layout lock feature supported.\n");
 383                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 384         }
 385
 386         obd = class_name2obd(dt);
 387         if (!obd) {
 388                 CERROR("DT %s: not setup or attached\n", dt);
 389                 GOTO(out_md_fid, err = -ENODEV);
 390         }
 391
 392         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 393                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 394                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 395                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 396                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 397                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 398                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 399                                   OBD_CONNECT_MAXBYTES |
 400                                   OBD_CONNECT_EINPROGRESS |
 401                                   OBD_CONNECT_JOBSTATS;
 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_client();
 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_lock(&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_unlock(&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 /*b=17630*/ && !defined(HAVE_FSTYPE_MOUNT) /*LU-812*/
 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(NULL, 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 #ifdef HAVE_DCACHE_LOCK
 550         sb->s_root->d_op = &ll_d_root_ops;
 551 #else
 552         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 553         d_set_d_op(sb->s_root, &ll_d_root_ops);
 554         sb->s_d_op = &ll_d_ops;
 555 #endif
 556
 557         sbi->ll_sdev_orig = sb->s_dev;
 558
 559         /* We set sb->s_dev equal on all lustre clients in order to support
 560          * NFS export clustering.  NFSD requires that the FSID be the same
 561          * on all clients. */
 562         /* s_dev is also used in lt_compare() to compare two fs, but that is
 563          * only a node-local comparison. */
 564         uuid = obd_get_uuid(sbi->ll_md_exp);
 565         if (uuid != NULL)
 566                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 567         sbi->ll_mnt = mnt;
 568
 569         if (data != NULL)
 570                 OBD_FREE_PTR(data);
 571         if (osfs != NULL)
 572                 OBD_FREE_PTR(osfs);
 573
 574         RETURN(err);
 575 out_root:
 576         if (root)
 577                 iput(root);
 578 out_lock_cn_cb:
 579         obd_fid_fini(sbi->ll_dt_exp);
 580 out_dt:
 581         obd_disconnect(sbi->ll_dt_exp);
 582         sbi->ll_dt_exp = NULL;
 583 out_md_fid:
 584         obd_fid_fini(sbi->ll_md_exp);
 585 out_md:
 586         obd_disconnect(sbi->ll_md_exp);
 587         sbi->ll_md_exp = NULL;
 588 out:
 589         if (data != NULL)
 590                 OBD_FREE_PTR(data);
 591         if (osfs != NULL)
 592                 OBD_FREE_PTR(osfs);
 593         lprocfs_unregister_mountpoint(sbi);
 594         return err;
 595 }
 596
 597 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 598 {
 599         int size, rc;
 600
 601         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 602         size = sizeof(int);
 603         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 604                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 605         if (rc)
 606                 CERROR("Get max mdsize error rc %d \n", rc);
 607
 608         RETURN(rc);
 609 }
 610
 611 void ll_dump_inode(struct inode *inode)
 612 {
 613         struct list_head *tmp;
 614         int dentry_count = 0;
 615
 616         LASSERT(inode != NULL);
 617
 618         list_for_each(tmp, &inode->i_dentry)
 619                 dentry_count++;
 620
 621         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 622                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 623                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 624 }
 625
 626 void lustre_dump_dentry(struct dentry *dentry, int recur)
 627 {
 628         struct list_head *tmp;
 629         int subdirs = 0;
 630
 631         LASSERT(dentry != NULL);
 632
 633         list_for_each(tmp, &dentry->d_subdirs)
 634                 subdirs++;
 635
 636         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 637                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 638                dentry->d_name.len, dentry->d_name.name,
 639                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 640                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 641                dentry->d_flags, dentry->d_fsdata, subdirs);
 642         if (dentry->d_inode != NULL)
 643                 ll_dump_inode(dentry->d_inode);
 644
 645         if (recur == 0)
 646                 return;
 647
 648         list_for_each(tmp, &dentry->d_subdirs) {
 649                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 650                 lustre_dump_dentry(d, recur - 1);
 651         }
 652 }
 653
 654 void client_common_put_super(struct super_block *sb)
 655 {
 656         struct ll_sb_info *sbi = ll_s2sbi(sb);
 657         ENTRY;
 658
 659 #ifdef CONFIG_FS_POSIX_ACL
 660         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 661                 et_fini(&sbi->ll_et);
 662                 rct_fini(&sbi->ll_rct);
 663         }
 664 #endif
 665
 666         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 667
 668         ll_close_thread_shutdown(sbi->ll_lcq);
 669
 670         cl_sb_fini(sb);
 671
 672         cfs_list_del(&sbi->ll_conn_chain);
 673
 674         obd_fid_fini(sbi->ll_dt_exp);
 675         obd_disconnect(sbi->ll_dt_exp);
 676         sbi->ll_dt_exp = NULL;
 677
 678         lprocfs_unregister_mountpoint(sbi);
 679
 680         obd_fid_fini(sbi->ll_md_exp);
 681         obd_disconnect(sbi->ll_md_exp);
 682         sbi->ll_md_exp = NULL;
 683
 684         EXIT;
 685 }
 686
 687 void ll_kill_super(struct super_block *sb)
 688 {
 689         struct ll_sb_info *sbi;
 690
 691         ENTRY;
 692
 693         /* not init sb ?*/
 694         if (!(sb->s_flags & MS_ACTIVE))
 695                 return;
 696
 697         sbi = ll_s2sbi(sb);
 698         /* we need restore s_dev from changed for clustred NFS before put_super
 699          * because new kernels have cached s_dev and change sb->s_dev in
 700          * put_super not affected real removing devices */
 701         if (sbi)
 702                 sb->s_dev = sbi->ll_sdev_orig;
 703         EXIT;
 704 }
 705
 706 char *ll_read_opt(const char *opt, char *data)
 707 {
 708         char *value;
 709         char *retval;
 710         ENTRY;
 711
 712         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 713         if (strncmp(opt, data, strlen(opt)))
 714                 RETURN(NULL);
 715         if ((value = strchr(data, '=')) == NULL)
 716                 RETURN(NULL);
 717
 718         value++;
 719         OBD_ALLOC(retval, strlen(value) + 1);
 720         if (!retval) {
 721                 CERROR("out of memory!\n");
 722                 RETURN(NULL);
 723         }
 724
 725         memcpy(retval, value, strlen(value)+1);
 726         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 727         RETURN(retval);
 728 }
 729
 730 static inline int ll_set_opt(const char *opt, char *data, int fl)
 731 {
 732         if (strncmp(opt, data, strlen(opt)) != 0)
 733                 return(0);
 734         else
 735                 return(fl);
 736 }
 737
 738 /* non-client-specific mount options are parsed in lmd_parse */
 739 static int ll_options(char *options, int *flags)
 740 {
 741         int tmp;
 742         char *s1 = options, *s2;
 743         ENTRY;
 744
 745         if (!options)
 746                 RETURN(0);
 747
 748         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 749
 750         while (*s1) {
 751                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 752                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 753                 if (tmp) {
 754                         *flags |= tmp;
 755                         goto next;
 756                 }
 757                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 758                 if (tmp) {
 759                         *flags |= tmp;
 760                         goto next;
 761                 }
 762                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 763                 if (tmp) {
 764                         *flags |= tmp;
 765                         goto next;
 766                 }
 767                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 768                 if (tmp) {
 769                         *flags &= ~tmp;
 770                         goto next;
 771                 }
 772                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 773                 if (tmp) {
 774                         *flags |= tmp;
 775                         goto next;
 776                 }
 777                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 778                 if (tmp) {
 779                         *flags &= ~tmp;
 780                         goto next;
 781                 }
 782 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 783                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 784                 if (tmp) {
 785                         /* Ignore deprecated mount option.  The client will
 786                          * always try to mount with ACL support, whether this
 787                          * is used depends on whether server supports it. */
 788                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 789                                                   "mount option 'acl'.\n");
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 793                 if (tmp) {
 794                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 795                                                   "mount option 'noacl'.\n");
 796                         goto next;
 797                 }
 798 #else
 799 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 800 #endif
 801                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806
 807                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 813                 if (tmp) {
 814                         *flags &= ~tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 823                 if (tmp) {
 824                         *flags &= ~tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 828                 if (tmp) {
 829                         *flags |= tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 833                 if (tmp) {
 834                         *flags &= ~tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 858                                    s1);
 859                 RETURN(-EINVAL);
 860
 861 next:
 862                 /* Find next opt */
 863                 s2 = strchr(s1, ',');
 864                 if (s2 == NULL)
 865                         break;
 866                 s1 = s2 + 1;
 867         }
 868         RETURN(0);
 869 }
 870
 871 void ll_lli_init(struct ll_inode_info *lli)
 872 {
 873         lli->lli_inode_magic = LLI_INODE_MAGIC;
 874         lli->lli_flags = 0;
 875         lli->lli_ioepoch = 0;
 876         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 877         cfs_spin_lock_init(&lli->lli_lock);
 878         lli->lli_posix_acl = NULL;
 879         lli->lli_remote_perms = NULL;
 880         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 881         /* Do not set lli_fid, it has been initialized already. */
 882         fid_zero(&lli->lli_pfid);
 883         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 884         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 885         cfs_atomic_set(&lli->lli_open_count, 0);
 886         lli->lli_mds_capa = NULL;
 887         lli->lli_rmtperm_time = 0;
 888         lli->lli_pending_och = NULL;
 889         lli->lli_mds_read_och = NULL;
 890         lli->lli_mds_write_och = NULL;
 891         lli->lli_mds_exec_och = NULL;
 892         lli->lli_open_fd_read_count = 0;
 893         lli->lli_open_fd_write_count = 0;
 894         lli->lli_open_fd_exec_count = 0;
 895         cfs_mutex_init(&lli->lli_och_mutex);
 896         cfs_spin_lock_init(&lli->lli_agl_lock);
 897         lli->lli_has_smd = false;
 898         lli->lli_clob = NULL;
 899
 900         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 901         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 902                 cfs_mutex_init(&lli->lli_readdir_mutex);
 903                 lli->lli_opendir_key = NULL;
 904                 lli->lli_sai = NULL;
 905                 lli->lli_sa_pos = 0;
 906                 lli->lli_def_acl = NULL;
 907                 cfs_spin_lock_init(&lli->lli_sa_lock);
 908                 lli->lli_opendir_pid = 0;
 909         } else {
 910                 cfs_sema_init(&lli->lli_size_sem, 1);
 911                 lli->lli_size_sem_owner = NULL;
 912                 lli->lli_symlink_name = NULL;
 913                 cfs_init_rwsem(&lli->lli_trunc_sem);
 914                 cfs_mutex_init(&lli->lli_write_mutex);
 915                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 916                 lli->lli_glimpse_time = 0;
 917                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 918                 lli->lli_agl_index = 0;
 919                 lli->lli_async_rc = 0;
 920         }
 921         cfs_mutex_init(&lli->lli_layout_mutex);
 922 }
 923
 924 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 925 {
 926 #ifdef HAVE_BDI_REGISTER
 927         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 928
 929 #ifdef HAVE_BDI_NAME
 930         bdi->name = "lustre";
 931 #endif
 932         return bdi_register(bdi, NULL, "lustre-%d",
 933                             atomic_inc_return(&ll_bdi_num));
 934 #else
 935         return 0;
 936 #endif
 937 }
 938
 939 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 940 {
 941         struct lustre_profile *lprof = NULL;
 942         struct lustre_sb_info *lsi = s2lsi(sb);
 943         struct ll_sb_info *sbi;
 944         char  *dt = NULL, *md = NULL;
 945         char  *profilenm = get_profile_name(sb);
 946         struct config_llog_instance *cfg;
 947         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 948         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 949         int    err;
 950         ENTRY;
 951
 952         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 953
 954         OBD_ALLOC_PTR(cfg);
 955         if (cfg == NULL)
 956                 RETURN(-ENOMEM);
 957
 958         cfs_module_get();
 959
 960         /* client additional sb info */
 961         lsi->lsi_llsbi = sbi = ll_init_sbi();
 962         if (!sbi) {
 963                 cfs_module_put(THIS_MODULE);
 964                 OBD_FREE_PTR(cfg);
 965                 RETURN(-ENOMEM);
 966         }
 967
 968         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 969         if (err)
 970                 GOTO(out_free, err);
 971
 972         err = ll_bdi_init(&lsi->lsi_bdi);
 973         if (err)
 974                 GOTO(out_free, err);
 975         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 976         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 977         err = ll_bdi_register(&lsi->lsi_bdi);
 978         if (err)
 979                 GOTO(out_free, err);
 980
 981 #ifdef HAVE_SB_BDI
 982         sb->s_bdi = &lsi->lsi_bdi;
 983 #endif
 984
 985         /* Generate a string unique to this super, in case some joker tries
 986            to mount the same fs at two mount points.
 987            Use the address of the super itself.*/
 988         cfg->cfg_instance = sb;
 989         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 990
 991         /* set up client obds */
 992         err = lustre_process_log(sb, profilenm, cfg);
 993         if (err < 0) {
 994                 CERROR("Unable to process log: %d\n", err);
 995                 GOTO(out_free, err);
 996         }
 997
 998         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 999         lprof = class_get_profile(profilenm);
1000         if (lprof == NULL) {
1001                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1002                                    " read from the MGS.  Does that filesystem "
1003                                    "exist?\n", profilenm);
1004                 GOTO(out_free, err = -EINVAL);
1005         }
1006         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1007                lprof->lp_md, lprof->lp_dt);
1008
1009         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1010         if (!dt)
1011                 GOTO(out_free, err = -ENOMEM);
1012         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1013
1014         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1015         if (!md)
1016                 GOTO(out_free, err = -ENOMEM);
1017         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1018
1019         /* connections, registrations, sb setup */
1020         err = client_common_fill_super(sb, md, dt, mnt);
1021
1022 out_free:
1023         if (md)
1024                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1025         if (dt)
1026                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1027         if (err)
1028                 ll_put_super(sb);
1029         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1030                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1031
1032         OBD_FREE_PTR(cfg);
1033         RETURN(err);
1034 } /* ll_fill_super */
1035
1036
1037 void lu_context_keys_dump(void);
1038
1039 void ll_put_super(struct super_block *sb)
1040 {
1041         struct config_llog_instance cfg;
1042         struct obd_device *obd;
1043         struct lustre_sb_info *lsi = s2lsi(sb);
1044         struct ll_sb_info *sbi = ll_s2sbi(sb);
1045         char *profilenm = get_profile_name(sb);
1046         int force = 1, next;
1047         ENTRY;
1048
1049         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1050
1051         ll_print_capa_stat(sbi);
1052
1053         cfg.cfg_instance = sb;
1054         lustre_end_log(sb, profilenm, &cfg);
1055
1056         if (sbi->ll_md_exp) {
1057                 obd = class_exp2obd(sbi->ll_md_exp);
1058                 if (obd)
1059                         force = obd->obd_force;
1060         }
1061
1062         /* We need to set force before the lov_disconnect in
1063            lustre_common_put_super, since l_d cleans up osc's as well. */
1064         if (force) {
1065                 next = 0;
1066                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1067                                                      &next)) != NULL) {
1068                         obd->obd_force = force;
1069                 }
1070         }
1071
1072         if (sbi->ll_lcq) {
1073                 /* Only if client_common_fill_super succeeded */
1074                 client_common_put_super(sb);
1075         }
1076
1077         next = 0;
1078         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1079                 class_manual_cleanup(obd);
1080         }
1081
1082         if (sbi->ll_flags & LL_SBI_VERBOSE)
1083                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1084
1085         if (profilenm)
1086                 class_del_profile(profilenm);
1087
1088         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1089                 ll_bdi_destroy(&lsi->lsi_bdi);
1090                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1091         }
1092
1093         ll_free_sbi(sb);
1094         lsi->lsi_llsbi = NULL;
1095
1096         lustre_common_put_super(sb);
1097
1098         cl_env_cache_purge(~0);
1099
1100         cfs_module_put(THIS_MODULE);
1101
1102         EXIT;
1103 } /* client_put_super */
1104
1105 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1106 {
1107         struct inode *inode = NULL;
1108         /* NOTE: we depend on atomic igrab() -bzzz */
1109         lock_res_and_lock(lock);
1110         if (lock->l_ast_data) {
1111                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1112                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1113                         inode = igrab(lock->l_ast_data);
1114                 } else {
1115                         inode = lock->l_ast_data;
1116                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1117                                          D_WARNING, lock, "l_ast_data %p is "
1118                                          "bogus: magic %08x", lock->l_ast_data,
1119                                          lli->lli_inode_magic);
1120                         inode = NULL;
1121                 }
1122         }
1123         unlock_res_and_lock(lock);
1124         return inode;
1125 }
1126
1127 static int null_if_equal(struct ldlm_lock *lock, void *data)
1128 {
1129         if (data == lock->l_ast_data) {
1130                 lock->l_ast_data = NULL;
1131
1132                 if (lock->l_req_mode != lock->l_granted_mode)
1133                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1134         }
1135
1136         return LDLM_ITER_CONTINUE;
1137 }
1138
1139 void ll_clear_inode(struct inode *inode)
1140 {
1141         struct ll_inode_info *lli = ll_i2info(inode);
1142         struct ll_sb_info *sbi = ll_i2sbi(inode);
1143         ENTRY;
1144
1145         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1146                inode->i_generation, inode);
1147
1148         if (S_ISDIR(inode->i_mode)) {
1149                 /* these should have been cleared in ll_file_release */
1150                 LASSERT(lli->lli_opendir_key == NULL);
1151                 LASSERT(lli->lli_sai == NULL);
1152                 LASSERT(lli->lli_opendir_pid == 0);
1153         }
1154
1155         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1156         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1157                          null_if_equal, inode);
1158
1159         LASSERT(!lli->lli_open_fd_write_count);
1160         LASSERT(!lli->lli_open_fd_read_count);
1161         LASSERT(!lli->lli_open_fd_exec_count);
1162
1163         if (lli->lli_mds_write_och)
1164                 ll_md_real_close(inode, FMODE_WRITE);
1165         if (lli->lli_mds_exec_och)
1166                 ll_md_real_close(inode, FMODE_EXEC);
1167         if (lli->lli_mds_read_och)
1168                 ll_md_real_close(inode, FMODE_READ);
1169
1170         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1171                 OBD_FREE(lli->lli_symlink_name,
1172                          strlen(lli->lli_symlink_name) + 1);
1173                 lli->lli_symlink_name = NULL;
1174         }
1175
1176         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1177                 LASSERT(lli->lli_posix_acl == NULL);
1178                 if (lli->lli_remote_perms) {
1179                         free_rmtperm_hash(lli->lli_remote_perms);
1180                         lli->lli_remote_perms = NULL;
1181                 }
1182         }
1183 #ifdef CONFIG_FS_POSIX_ACL
1184         else if (lli->lli_posix_acl) {
1185                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1186                 LASSERT(lli->lli_remote_perms == NULL);
1187                 posix_acl_release(lli->lli_posix_acl);
1188                 lli->lli_posix_acl = NULL;
1189         }
1190 #endif
1191         lli->lli_inode_magic = LLI_INODE_DEAD;
1192
1193         ll_clear_inode_capas(inode);
1194         if (!S_ISDIR(inode->i_mode))
1195                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1196
1197         /*
1198          * XXX This has to be done before lsm is freed below, because
1199          * cl_object still uses inode lsm.
1200          */
1201         cl_inode_fini(inode);
1202         lli->lli_has_smd = false;
1203
1204         EXIT;
1205 }
1206
1207 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1208                   struct md_open_data **mod)
1209 {
1210         struct lustre_md md;
1211         struct inode *inode = dentry->d_inode;
1212         struct ll_sb_info *sbi = ll_i2sbi(inode);
1213         struct ptlrpc_request *request = NULL;
1214         int rc, ia_valid;
1215         ENTRY;
1216
1217         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1218                                      LUSTRE_OPC_ANY, NULL);
1219         if (IS_ERR(op_data))
1220                 RETURN(PTR_ERR(op_data));
1221
1222         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1223                         &request, mod);
1224         if (rc) {
1225                 ptlrpc_req_finished(request);
1226                 if (rc == -ENOENT) {
1227                         inode->i_nlink = 0;
1228                         /* Unlinked special device node? Or just a race?
1229                          * Pretend we done everything. */
1230                         if (!S_ISREG(inode->i_mode) &&
1231                             !S_ISDIR(inode->i_mode)) {
1232                                 ia_valid = op_data->op_attr.ia_valid;
1233                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1234                                 rc = simple_setattr(dentry, &op_data->op_attr);
1235                                 op_data->op_attr.ia_valid = ia_valid;
1236                         }
1237                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1238                         CERROR("md_setattr fails: rc = %d\n", rc);
1239                 }
1240                 RETURN(rc);
1241         }
1242
1243         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1244                               sbi->ll_md_exp, &md);
1245         if (rc) {
1246                 ptlrpc_req_finished(request);
1247                 RETURN(rc);
1248         }
1249
1250         /* We want to adjust timestamps.
1251          * If there is at least some data in file, we cleared ATTR_SIZE
1252          * to avoid update size, otherwise it is important to do.(SOM case)
1253          * (bug 6196) */
1254         ia_valid = op_data->op_attr.ia_valid;
1255         /* Since we set ATTR_*_SET flags above, and already done permission
1256          * check, So don't let inode_change_ok() check it again.  */
1257         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1258         rc = simple_setattr(dentry, &op_data->op_attr);
1259         op_data->op_attr.ia_valid = ia_valid;
1260
1261         /* Extract epoch data if obtained. */
1262         op_data->op_handle = md.body->handle;
1263         op_data->op_ioepoch = md.body->ioepoch;
1264
1265         ll_update_inode(inode, &md);
1266         ptlrpc_req_finished(request);
1267
1268         RETURN(rc);
1269 }
1270
1271 /* Close IO epoch and send Size-on-MDS attribute update. */
1272 static int ll_setattr_done_writing(struct inode *inode,
1273                                    struct md_op_data *op_data,
1274                                    struct md_open_data *mod)
1275 {
1276         struct ll_inode_info *lli = ll_i2info(inode);
1277         int rc = 0;
1278         ENTRY;
1279
1280         LASSERT(op_data != NULL);
1281         if (!S_ISREG(inode->i_mode))
1282                 RETURN(0);
1283
1284         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1285                op_data->op_ioepoch, PFID(&lli->lli_fid));
1286
1287         op_data->op_flags = MF_EPOCH_CLOSE;
1288         ll_done_writing_attr(inode, op_data);
1289         ll_pack_inode2opdata(inode, op_data, NULL);
1290
1291         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1292         if (rc == -EAGAIN) {
1293                 /* MDS has instructed us to obtain Size-on-MDS attribute
1294                  * from OSTs and send setattr to back to MDS. */
1295                 rc = ll_som_update(inode, op_data);
1296         } else if (rc) {
1297                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1298                        inode->i_ino, rc);
1299         }
1300         RETURN(rc);
1301 }
1302
1303 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1304 {
1305         struct obd_capa *capa;
1306         int rc;
1307
1308         if (attr->ia_valid & ATTR_SIZE)
1309                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1310         else
1311                 capa = ll_mdscapa_get(inode);
1312
1313         rc = cl_setattr_ost(inode, attr, capa);
1314
1315         if (attr->ia_valid & ATTR_SIZE)
1316                 ll_truncate_free_capa(capa);
1317         else
1318                 capa_put(capa);
1319
1320         return rc;
1321 }
1322
1323 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1324 /**
1325  * inode_newsize_ok - may this inode be truncated to a given size
1326  * @inode:      the inode to be truncated
1327  * @offset:     the new size to assign to the inode
1328  * @Returns:    0 on success, -ve errno on failure
1329  *
1330  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1331  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1332  * when necessary. Caller must not proceed with inode size change if failure is
1333  * returned. @inode must be a file (not directory), with appropriate
1334  * permissions to allow truncate (inode_newsize_ok does NOT check these
1335  * conditions).
1336  *
1337  * inode_newsize_ok must be called with i_mutex held.
1338  */
1339 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1340 {
1341         if (inode->i_size < offset) {
1342                 unsigned long limit;
1343
1344                 limit = rlimit(RLIMIT_FSIZE);
1345                 if (limit != RLIM_INFINITY && offset > limit)
1346                         goto out_sig;
1347                 if (offset > inode->i_sb->s_maxbytes)
1348                         goto out_big;
1349         } else {
1350                 /*
1351                  * truncation of in-use swapfiles is disallowed - it would
1352                  * cause subsequent swapout to scribble on the now-freed
1353                  * blocks.
1354                  */
1355                 if (IS_SWAPFILE(inode))
1356                         return -ETXTBSY;
1357         }
1358
1359         return 0;
1360 out_sig:
1361         send_sig(SIGXFSZ, current, 0);
1362 out_big:
1363         return -EFBIG;
1364 }
1365 #endif
1366
1367 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1368  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1369  * keep these values until such a time that objects are allocated for it.
1370  * We do the MDS operations first, as it is checking permissions for us.
1371  * We don't to the MDS RPC if there is nothing that we want to store there,
1372  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1373  * going to do an RPC anyways.
1374  *
1375  * If we are doing a truncate, we will send the mtime and ctime updates
1376  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1377  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1378  * at the same time.
1379  */
1380 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1381 {
1382         struct lov_stripe_md *lsm;
1383         struct inode *inode = dentry->d_inode;
1384         struct ll_inode_info *lli = ll_i2info(inode);
1385         struct md_op_data *op_data = NULL;
1386         struct md_open_data *mod = NULL;
1387         int ia_valid = attr->ia_valid;
1388         int rc = 0, rc1 = 0;
1389         ENTRY;
1390
1391         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1392                attr->ia_valid);
1393
1394         if (ia_valid & ATTR_SIZE) {
1395                 /* Check new size against VFS/VM file size limit and rlimit */
1396                 rc = inode_newsize_ok(inode, attr->ia_size);
1397                 if (rc)
1398                         RETURN(rc);
1399
1400                 /* The maximum Lustre file size is variable, based on the
1401                  * OST maximum object size and number of stripes.  This
1402                  * needs another check in addition to the VFS check above. */
1403                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1404                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1405                                PFID(&lli->lli_fid), attr->ia_size,
1406                                ll_file_maxbytes(inode));
1407                         RETURN(-EFBIG);
1408                 }
1409
1410                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1411         }
1412
1413         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1414         if (ia_valid & TIMES_SET_FLAGS) {
1415                 if (cfs_curproc_fsuid() != inode->i_uid &&
1416                     !cfs_capable(CFS_CAP_FOWNER))
1417                         RETURN(-EPERM);
1418         }
1419
1420         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1421         if (attr->ia_valid & ATTR_CTIME) {
1422                 attr->ia_ctime = CFS_CURRENT_TIME;
1423                 attr->ia_valid |= ATTR_CTIME_SET;
1424         }
1425         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1426                 attr->ia_atime = CFS_CURRENT_TIME;
1427                 attr->ia_valid |= ATTR_ATIME_SET;
1428         }
1429         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1430                 attr->ia_mtime = CFS_CURRENT_TIME;
1431                 attr->ia_valid |= ATTR_MTIME_SET;
1432         }
1433
1434         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1435                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1436                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1437                        cfs_time_current_sec());
1438
1439         /* We always do an MDS RPC, even if we're only changing the size;
1440          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1441
1442         OBD_ALLOC_PTR(op_data);
1443         if (op_data == NULL)
1444                 RETURN(-ENOMEM);
1445
1446         if (!S_ISDIR(inode->i_mode)) {
1447                 if (ia_valid & ATTR_SIZE)
1448                         UP_WRITE_I_ALLOC_SEM(inode);
1449                 mutex_unlock(&inode->i_mutex);
1450                 cfs_down_write(&lli->lli_trunc_sem);
1451                 mutex_lock(&inode->i_mutex);
1452                 if (ia_valid & ATTR_SIZE)
1453                         DOWN_WRITE_I_ALLOC_SEM(inode);
1454         }
1455
1456         /* We need a steady stripe configuration for setattr to avoid
1457          * confusion. */
1458         lsm = ccc_inode_lsm_get(inode);
1459
1460         /* NB: ATTR_SIZE will only be set after this point if the size
1461          * resides on the MDS, ie, this file has no objects. */
1462         if (lsm != NULL)
1463                 attr->ia_valid &= ~ATTR_SIZE;
1464
1465         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1466
1467         /* Open epoch for truncate. */
1468         if (exp_connect_som(ll_i2mdexp(inode)) &&
1469             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1470                 op_data->op_flags = MF_EPOCH_OPEN;
1471
1472         rc = ll_md_setattr(dentry, op_data, &mod);
1473         if (rc)
1474                 GOTO(out, rc);
1475
1476         ll_ioepoch_open(lli, op_data->op_ioepoch);
1477         if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1478                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1479                 GOTO(out, rc = 0);
1480         }
1481
1482         if (ia_valid & ATTR_SIZE)
1483                 attr->ia_valid |= ATTR_SIZE;
1484         if (ia_valid & (ATTR_SIZE |
1485                         ATTR_ATIME | ATTR_ATIME_SET |
1486                         ATTR_MTIME | ATTR_MTIME_SET))
1487                 /* For truncate and utimes sending attributes to OSTs, setting
1488                  * mtime/atime to the past will be performed under PW [0:EOF]
1489                  * extent lock (new_size:EOF for truncate).  It may seem
1490                  * excessive to send mtime/atime updates to OSTs when not
1491                  * setting times to past, but it is necessary due to possible
1492                  * time de-synchronization between MDT inode and OST objects */
1493                 rc = ll_setattr_ost(inode, attr);
1494         EXIT;
1495 out:
1496         ccc_inode_lsm_put(inode, lsm);
1497         if (op_data) {
1498                 if (op_data->op_ioepoch) {
1499                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1500                         if (!rc)
1501                                 rc = rc1;
1502                 }
1503                 ll_finish_md_op_data(op_data);
1504         }
1505         if (!S_ISDIR(inode->i_mode))
1506                 cfs_up_write(&lli->lli_trunc_sem);
1507
1508         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1509                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1510
1511         return rc;
1512 }
1513
1514 int ll_setattr(struct dentry *de, struct iattr *attr)
1515 {
1516         int mode = de->d_inode->i_mode;
1517
1518         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1519                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1520                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1521
1522         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1523                                (ATTR_SIZE|ATTR_MODE)) &&
1524             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1525              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1526               !(attr->ia_mode & S_ISGID))))
1527                 attr->ia_valid |= ATTR_FORCE;
1528
1529         if ((mode & S_ISUID) &&
1530             !(attr->ia_mode & S_ISUID) &&
1531             !(attr->ia_valid & ATTR_KILL_SUID))
1532                 attr->ia_valid |= ATTR_KILL_SUID;
1533
1534         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1535             !(attr->ia_mode & S_ISGID) &&
1536             !(attr->ia_valid & ATTR_KILL_SGID))
1537                 attr->ia_valid |= ATTR_KILL_SGID;
1538
1539         return ll_setattr_raw(de, attr);
1540 }
1541
1542 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1543                        __u64 max_age, __u32 flags)
1544 {
1545         struct ll_sb_info *sbi = ll_s2sbi(sb);
1546         struct obd_statfs obd_osfs;
1547         int rc;
1548         ENTRY;
1549
1550         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1551         if (rc) {
1552                 CERROR("md_statfs fails: rc = %d\n", rc);
1553                 RETURN(rc);
1554         }
1555
1556         osfs->os_type = sb->s_magic;
1557
1558         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1559                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1560
1561         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1562                 flags |= OBD_STATFS_NODELAY;
1563
1564         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1565         if (rc) {
1566                 CERROR("obd_statfs fails: rc = %d\n", rc);
1567                 RETURN(rc);
1568         }
1569
1570         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1571                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1572                obd_osfs.os_files);
1573
1574         osfs->os_bsize = obd_osfs.os_bsize;
1575         osfs->os_blocks = obd_osfs.os_blocks;
1576         osfs->os_bfree = obd_osfs.os_bfree;
1577         osfs->os_bavail = obd_osfs.os_bavail;
1578
1579         /* If we don't have as many objects free on the OST as inodes
1580          * on the MDS, we reduce the total number of inodes to
1581          * compensate, so that the "inodes in use" number is correct.
1582          */
1583         if (obd_osfs.os_ffree < osfs->os_ffree) {
1584                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1585                         obd_osfs.os_ffree;
1586                 osfs->os_ffree = obd_osfs.os_ffree;
1587         }
1588
1589         RETURN(rc);
1590 }
1591 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1592 {
1593         struct super_block *sb = de->d_sb;
1594         struct obd_statfs osfs;
1595         int rc;
1596
1597         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1598         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1599
1600         /* Some amount of caching on the client is allowed */
1601         rc = ll_statfs_internal(sb, &osfs,
1602                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1603                                 0);
1604         if (rc)
1605                 return rc;
1606
1607         statfs_unpack(sfs, &osfs);
1608
1609         /* We need to downshift for all 32-bit kernels, because we can't
1610          * tell if the kernel is being called via sys_statfs64() or not.
1611          * Stop before overflowing f_bsize - in which case it is better
1612          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1613         if (sizeof(long) < 8) {
1614                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1615                         sfs->f_bsize <<= 1;
1616
1617                         osfs.os_blocks >>= 1;
1618                         osfs.os_bfree >>= 1;
1619                         osfs.os_bavail >>= 1;
1620                 }
1621         }
1622
1623         sfs->f_blocks = osfs.os_blocks;
1624         sfs->f_bfree = osfs.os_bfree;
1625         sfs->f_bavail = osfs.os_bavail;
1626
1627         return 0;
1628 }
1629
1630 void ll_inode_size_lock(struct inode *inode)
1631 {
1632         struct ll_inode_info *lli;
1633
1634         LASSERT(!S_ISDIR(inode->i_mode));
1635
1636         lli = ll_i2info(inode);
1637         LASSERT(lli->lli_size_sem_owner != current);
1638         cfs_down(&lli->lli_size_sem);
1639         LASSERT(lli->lli_size_sem_owner == NULL);
1640         lli->lli_size_sem_owner = current;
1641 }
1642
1643 void ll_inode_size_unlock(struct inode *inode)
1644 {
1645         struct ll_inode_info *lli;
1646
1647         lli = ll_i2info(inode);
1648         LASSERT(lli->lli_size_sem_owner == current);
1649         lli->lli_size_sem_owner = NULL;
1650         cfs_up(&lli->lli_size_sem);
1651 }
1652
1653 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1654 {
1655         struct ll_inode_info *lli = ll_i2info(inode);
1656         struct mdt_body *body = md->body;
1657         struct lov_stripe_md *lsm = md->lsm;
1658         struct ll_sb_info *sbi = ll_i2sbi(inode);
1659
1660         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1661         if (lsm != NULL) {
1662                 LASSERT(S_ISREG(inode->i_mode));
1663                 cfs_mutex_lock(&lli->lli_och_mutex);
1664                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1665                                 lsm, inode->i_ino, inode->i_generation, inode);
1666                 /* cl_file_inode_init must go before lli_has_smd or a race
1667                  * is possible where client thinks the file has stripes,
1668                  * but lov raid0 is not setup yet and parallel e.g.
1669                  * glimpse would try to use uninitialized lov */
1670                 if (cl_file_inode_init(inode, md) == 0)
1671                         lli->lli_has_smd = true;
1672                 cfs_mutex_unlock(&lli->lli_och_mutex);
1673
1674                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1675                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1676                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1677                 if (md->lsm != NULL)
1678                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1679         }
1680
1681         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1682                 if (body->valid & OBD_MD_FLRMTPERM)
1683                         ll_update_remote_perm(inode, md->remote_perm);
1684         }
1685 #ifdef CONFIG_FS_POSIX_ACL
1686         else if (body->valid & OBD_MD_FLACL) {
1687                 cfs_spin_lock(&lli->lli_lock);
1688                 if (lli->lli_posix_acl)
1689                         posix_acl_release(lli->lli_posix_acl);
1690                 lli->lli_posix_acl = md->posix_acl;
1691                 cfs_spin_unlock(&lli->lli_lock);
1692         }
1693 #endif
1694         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1695         inode->i_generation = cl_fid_build_gen(&body->fid1);
1696
1697         if (body->valid & OBD_MD_FLATIME) {
1698                 if (body->atime > LTIME_S(inode->i_atime))
1699                         LTIME_S(inode->i_atime) = body->atime;
1700                 lli->lli_lvb.lvb_atime = body->atime;
1701         }
1702         if (body->valid & OBD_MD_FLMTIME) {
1703                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1704                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1705                                "to "LPU64"\n", inode->i_ino,
1706                                LTIME_S(inode->i_mtime), body->mtime);
1707                         LTIME_S(inode->i_mtime) = body->mtime;
1708                 }
1709                 lli->lli_lvb.lvb_mtime = body->mtime;
1710         }
1711         if (body->valid & OBD_MD_FLCTIME) {
1712                 if (body->ctime > LTIME_S(inode->i_ctime))
1713                         LTIME_S(inode->i_ctime) = body->ctime;
1714                 lli->lli_lvb.lvb_ctime = body->ctime;
1715         }
1716         if (body->valid & OBD_MD_FLMODE)
1717                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1718         if (body->valid & OBD_MD_FLTYPE)
1719                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1720         LASSERT(inode->i_mode != 0);
1721         if (S_ISREG(inode->i_mode)) {
1722                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1723         } else {
1724                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1725         }
1726         if (body->valid & OBD_MD_FLUID)
1727                 inode->i_uid = body->uid;
1728         if (body->valid & OBD_MD_FLGID)
1729                 inode->i_gid = body->gid;
1730         if (body->valid & OBD_MD_FLFLAGS)
1731                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1732         if (body->valid & OBD_MD_FLNLINK)
1733                 inode->i_nlink = body->nlink;
1734         if (body->valid & OBD_MD_FLRDEV)
1735                 inode->i_rdev = old_decode_dev(body->rdev);
1736
1737         if (body->valid & OBD_MD_FLID) {
1738                 /* FID shouldn't be changed! */
1739                 if (fid_is_sane(&lli->lli_fid)) {
1740                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1741                                  "Trying to change FID "DFID
1742                                  " to the "DFID", inode %lu/%u(%p)\n",
1743                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1744                                  inode->i_ino, inode->i_generation, inode);
1745                 } else
1746                         lli->lli_fid = body->fid1;
1747         }
1748
1749         LASSERT(fid_seq(&lli->lli_fid) != 0);
1750
1751         if (body->valid & OBD_MD_FLSIZE) {
1752                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1753                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1754                         struct lustre_handle lockh;
1755                         ldlm_mode_t mode;
1756
1757                         /* As it is possible a blocking ast has been processed
1758                          * by this time, we need to check there is an UPDATE
1759                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1760                          * it. */
1761                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1762                                                &lockh);
1763                         if (mode) {
1764                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1765                                                       LLIF_EPOCH_PENDING |
1766                                                       LLIF_SOM_DIRTY)) {
1767                                         CERROR("ino %lu flags %u still has "
1768                                                "size authority! do not trust "
1769                                                "the size got from MDS\n",
1770                                                inode->i_ino, lli->lli_flags);
1771                                 } else {
1772                                         /* Use old size assignment to avoid
1773                                          * deadlock bz14138 & bz14326 */
1774                                         i_size_write(inode, body->size);
1775                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1776                                 }
1777                                 ldlm_lock_decref(&lockh, mode);
1778                         }
1779                 } else {
1780                         /* Use old size assignment to avoid
1781                          * deadlock bz14138 & bz14326 */
1782                         i_size_write(inode, body->size);
1783
1784                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1785                                inode->i_ino, (unsigned long long)body->size);
1786                 }
1787
1788                 if (body->valid & OBD_MD_FLBLOCKS)
1789                         inode->i_blocks = body->blocks;
1790         }
1791
1792         if (body->valid & OBD_MD_FLMDSCAPA) {
1793                 LASSERT(md->mds_capa);
1794                 ll_add_capa(inode, md->mds_capa);
1795         }
1796         if (body->valid & OBD_MD_FLOSSCAPA) {
1797                 LASSERT(md->oss_capa);
1798                 ll_add_capa(inode, md->oss_capa);
1799         }
1800 }
1801
1802 void ll_read_inode2(struct inode *inode, void *opaque)
1803 {
1804         struct lustre_md *md = opaque;
1805         struct ll_inode_info *lli = ll_i2info(inode);
1806         ENTRY;
1807
1808         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1809                PFID(&lli->lli_fid), inode);
1810
1811         LASSERT(!lli->lli_has_smd);
1812
1813         /* Core attributes from the MDS first.  This is a new inode, and
1814          * the VFS doesn't zero times in the core inode so we have to do
1815          * it ourselves.  They will be overwritten by either MDS or OST
1816          * attributes - we just need to make sure they aren't newer. */
1817         LTIME_S(inode->i_mtime) = 0;
1818         LTIME_S(inode->i_atime) = 0;
1819         LTIME_S(inode->i_ctime) = 0;
1820         inode->i_rdev = 0;
1821         ll_update_inode(inode, md);
1822
1823         /* OIDEBUG(inode); */
1824
1825         /* initializing backing dev info. */
1826         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1827
1828
1829         if (S_ISREG(inode->i_mode)) {
1830                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1831                 inode->i_op = &ll_file_inode_operations;
1832                 inode->i_fop = sbi->ll_fop;
1833                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1834                 EXIT;
1835         } else if (S_ISDIR(inode->i_mode)) {
1836                 inode->i_op = &ll_dir_inode_operations;
1837                 inode->i_fop = &ll_dir_operations;
1838                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1839                 EXIT;
1840         } else if (S_ISLNK(inode->i_mode)) {
1841                 inode->i_op = &ll_fast_symlink_inode_operations;
1842                 EXIT;
1843         } else {
1844                 inode->i_op = &ll_special_inode_operations;
1845
1846                 init_special_inode(inode, inode->i_mode,
1847                                    kdev_t_to_nr(inode->i_rdev));
1848
1849                 EXIT;
1850         }
1851 }
1852
1853 void ll_delete_inode(struct inode *inode)
1854 {
1855         struct cl_inode_info *lli = cl_i2info(inode);
1856         struct ll_sb_info    *sbi = ll_i2sbi(inode);
1857         int rc;
1858         ENTRY;
1859
1860         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1861         if (rc)
1862                 CERROR("fid_delete() failed, rc %d\n", rc);
1863
1864         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1865                 /* discard all dirty pages before truncating them, required by
1866                  * osc_extent implementation at LU-1030. */
1867                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1868
1869         truncate_inode_pages(&inode->i_data, 0);
1870
1871         /* Workaround for LU-118 */
1872         if (inode->i_data.nrpages) {
1873                 TREE_READ_LOCK_IRQ(&inode->i_data);
1874                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1875                 LASSERTF(inode->i_data.nrpages == 0,
1876                          "inode=%lu/%u(%p) nrpages=%lu, see "
1877                          "http://jira.whamcloud.com/browse/LU-118\n",
1878                          inode->i_ino, inode->i_generation, inode,
1879                          inode->i_data.nrpages);
1880         }
1881         /* Workaround end */
1882
1883 #ifdef HAVE_SBOPS_EVICT_INODE
1884         ll_clear_inode(inode);
1885         end_writeback(inode);
1886 #else
1887         clear_inode(inode);
1888 #endif
1889
1890         EXIT;
1891 }
1892
1893 int ll_iocontrol(struct inode *inode, struct file *file,
1894                  unsigned int cmd, unsigned long arg)
1895 {
1896         struct ll_sb_info *sbi = ll_i2sbi(inode);
1897         struct ptlrpc_request *req = NULL;
1898         int rc, flags = 0;
1899         ENTRY;
1900
1901         switch(cmd) {
1902         case FSFILT_IOC_GETFLAGS: {
1903                 struct mdt_body *body;
1904                 struct md_op_data *op_data;
1905
1906                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1907                                              0, 0, LUSTRE_OPC_ANY,
1908                                              NULL);
1909                 if (IS_ERR(op_data))
1910                         RETURN(PTR_ERR(op_data));
1911
1912                 op_data->op_valid = OBD_MD_FLFLAGS;
1913                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1914                 ll_finish_md_op_data(op_data);
1915                 if (rc) {
1916                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1917                         RETURN(-abs(rc));
1918                 }
1919
1920                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1921
1922                 flags = body->flags;
1923
1924                 ptlrpc_req_finished(req);
1925
1926                 RETURN(put_user(flags, (int *)arg));
1927         }
1928         case FSFILT_IOC_SETFLAGS: {
1929                 struct lov_stripe_md *lsm;
1930                 struct obd_info oinfo = { { { 0 } } };
1931                 struct md_op_data *op_data;
1932
1933                 if (get_user(flags, (int *)arg))
1934                         RETURN(-EFAULT);
1935
1936                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1937                                              LUSTRE_OPC_ANY, NULL);
1938                 if (IS_ERR(op_data))
1939                         RETURN(PTR_ERR(op_data));
1940
1941                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1942                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1943                 rc = md_setattr(sbi->ll_md_exp, op_data,
1944                                 NULL, 0, NULL, 0, &req, NULL);
1945                 ll_finish_md_op_data(op_data);
1946                 ptlrpc_req_finished(req);
1947                 if (rc)
1948                         RETURN(rc);
1949
1950                 OBDO_ALLOC(oinfo.oi_oa);
1951                 if (!oinfo.oi_oa)
1952                         RETURN(-ENOMEM);
1953
1954                 lsm = ccc_inode_lsm_get(inode);
1955                 if (lsm == NULL) {
1956                         inode->i_flags = ll_ext_to_inode_flags(flags);
1957                         OBDO_FREE(oinfo.oi_oa);
1958                         RETURN(0);
1959                 }
1960
1961                 oinfo.oi_md = lsm;
1962                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1963                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1964                 oinfo.oi_oa->o_flags = flags;
1965                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1966                                        OBD_MD_FLGROUP;
1967                 oinfo.oi_capa = ll_mdscapa_get(inode);
1968                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1969                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1970                 capa_put(oinfo.oi_capa);
1971                 OBDO_FREE(oinfo.oi_oa);
1972                 ccc_inode_lsm_put(inode, lsm);
1973
1974                 if (rc && rc != -EPERM && rc != -EACCES)
1975                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1976
1977                 RETURN(rc);
1978         }
1979         default:
1980                 RETURN(-ENOSYS);
1981         }
1982
1983         RETURN(0);
1984 }
1985
1986 int ll_flush_ctx(struct inode *inode)
1987 {
1988         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1989
1990         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1991
1992         obd_set_info_async(NULL, sbi->ll_md_exp,
1993                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1994                            0, NULL, NULL);
1995         obd_set_info_async(NULL, sbi->ll_dt_exp,
1996                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1997                            0, NULL, NULL);
1998         return 0;
1999 }
2000
2001 /* umount -f client means force down, don't save state */
2002 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2003 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2004 {
2005         struct super_block *sb = vfsmnt->mnt_sb;
2006 #else
2007 void ll_umount_begin(struct super_block *sb)
2008 {
2009 #endif
2010         struct ll_sb_info *sbi = ll_s2sbi(sb);
2011         struct obd_device *obd;
2012         struct obd_ioctl_data *ioc_data;
2013         ENTRY;
2014
2015 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2016         if (!(flags & MNT_FORCE)) {
2017                 EXIT;
2018                 return;
2019         }
2020 #endif
2021
2022         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2023                sb->s_count, atomic_read(&sb->s_active));
2024
2025         obd = class_exp2obd(sbi->ll_md_exp);
2026         if (obd == NULL) {
2027                 CERROR("Invalid MDC connection handle "LPX64"\n",
2028                        sbi->ll_md_exp->exp_handle.h_cookie);
2029                 EXIT;
2030                 return;
2031         }
2032         obd->obd_force = 1;
2033
2034         obd = class_exp2obd(sbi->ll_dt_exp);
2035         if (obd == NULL) {
2036                 CERROR("Invalid LOV connection handle "LPX64"\n",
2037                        sbi->ll_dt_exp->exp_handle.h_cookie);
2038                 EXIT;
2039                 return;
2040         }
2041         obd->obd_force = 1;
2042
2043         OBD_ALLOC_PTR(ioc_data);
2044         if (ioc_data) {
2045                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2046                               sizeof ioc_data, ioc_data, NULL);
2047
2048                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2049                               sizeof ioc_data, ioc_data, NULL);
2050
2051                 OBD_FREE_PTR(ioc_data);
2052         }
2053
2054
2055         /* Really, we'd like to wait until there are no requests outstanding,
2056          * and then continue.  For now, we just invalidate the requests,
2057          * schedule() and sleep one second if needed, and hope.
2058          */
2059         cfs_schedule();
2060 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2061         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2062                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2063                                                    cfs_time_seconds(1));
2064                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2065                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2066                                       "may try to umount it a bit later\n",
2067                                       atomic_read(&vfsmnt->mnt_count));
2068         }
2069 #endif
2070
2071         EXIT;
2072 }
2073
2074 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2075 {
2076         struct ll_sb_info *sbi = ll_s2sbi(sb);
2077         char *profilenm = get_profile_name(sb);
2078         int err;
2079         __u32 read_only;
2080
2081         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2082                 read_only = *flags & MS_RDONLY;
2083                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2084                                          sizeof(KEY_READ_ONLY),
2085                                          KEY_READ_ONLY, sizeof(read_only),
2086                                          &read_only, NULL);
2087                 if (err) {
2088                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2089                                       profilenm, read_only ?
2090                                       "read-only" : "read-write", err);
2091                         return err;
2092                 }
2093
2094                 if (read_only)
2095                         sb->s_flags |= MS_RDONLY;
2096                 else
2097                         sb->s_flags &= ~MS_RDONLY;
2098
2099                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2100                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2101                                       read_only ?  "read-only" : "read-write");
2102         }
2103         return 0;
2104 }
2105
2106 int ll_prep_inode(struct inode **inode,
2107                   struct ptlrpc_request *req,
2108                   struct super_block *sb)
2109 {
2110         struct ll_sb_info *sbi = NULL;
2111         struct lustre_md md;
2112         __u64 ibits;
2113         int rc;
2114         ENTRY;
2115
2116         LASSERT(*inode || sb);
2117         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2118         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2119                               sbi->ll_md_exp, &md);
2120         if (rc)
2121                 RETURN(rc);
2122
2123         if (*inode) {
2124                 ll_update_inode(*inode, &md);
2125         } else {
2126                 LASSERT(sb != NULL);
2127
2128                 /*
2129                  * At this point server returns to client's same fid as client
2130                  * generated for creating. So using ->fid1 is okay here.
2131                  */
2132                 LASSERT(fid_is_sane(&md.body->fid1));
2133
2134                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2135                 if (*inode == NULL || IS_ERR(*inode)) {
2136                         if (md.lsm)
2137                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2138 #ifdef CONFIG_FS_POSIX_ACL
2139                         if (md.posix_acl) {
2140                                 posix_acl_release(md.posix_acl);
2141                                 md.posix_acl = NULL;
2142                         }
2143 #endif
2144                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2145                         *inode = NULL;
2146                         CERROR("new_inode -fatal: rc %d\n", rc);
2147                         GOTO(out, rc);
2148                 }
2149         }
2150
2151         /* sanity check for LAYOUT lock. */
2152         ibits = MDS_INODELOCK_LAYOUT;
2153         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2154             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2155                 char *fsname = ll_get_fsname(*inode);
2156                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2157                         fsname, PFID(ll_inode2fid(*inode)),
2158                         *inode);
2159                 if (fsname)
2160                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
2161         }
2162
2163 out:
2164         md_free_lustre_md(sbi->ll_md_exp, &md);
2165         RETURN(rc);
2166 }
2167
2168 int ll_obd_statfs(struct inode *inode, void *arg)
2169 {
2170         struct ll_sb_info *sbi = NULL;
2171         struct obd_export *exp;
2172         char *buf = NULL;
2173         struct obd_ioctl_data *data = NULL;
2174         __u32 type;
2175         int len = 0, rc;
2176
2177         if (!inode || !(sbi = ll_i2sbi(inode)))
2178                 GOTO(out_statfs, rc = -EINVAL);
2179
2180         rc = obd_ioctl_getdata(&buf, &len, arg);
2181         if (rc)
2182                 GOTO(out_statfs, rc);
2183
2184         data = (void*)buf;
2185         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2186             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2187                 GOTO(out_statfs, rc = -EINVAL);
2188
2189         if (data->ioc_inllen1 != sizeof(__u32) ||
2190             data->ioc_inllen2 != sizeof(__u32) ||
2191             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2192             data->ioc_plen2 != sizeof(struct obd_uuid))
2193                 GOTO(out_statfs, rc = -EINVAL);
2194
2195         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2196         if (type == LL_STATFS_LMV)
2197                 exp = sbi->ll_md_exp;
2198         else if (type == LL_STATFS_LOV)
2199                 exp = sbi->ll_dt_exp;
2200         else
2201                 GOTO(out_statfs, rc = -ENODEV);
2202
2203         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2204         if (rc)
2205                 GOTO(out_statfs, rc);
2206 out_statfs:
2207         if (buf)
2208                 obd_ioctl_freedata(buf, len);
2209         return rc;
2210 }
2211
2212 int ll_process_config(struct lustre_cfg *lcfg)
2213 {
2214         char *ptr;
2215         void *sb;
2216         struct lprocfs_static_vars lvars;
2217         unsigned long x;
2218         int rc = 0;
2219
2220         lprocfs_llite_init_vars(&lvars);
2221
2222         /* The instance name contains the sb: lustre-client-aacfe000 */
2223         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2224         if (!ptr || !*(++ptr))
2225                 return -EINVAL;
2226         if (sscanf(ptr, "%lx", &x) != 1)
2227                 return -EINVAL;
2228         sb = (void *)x;
2229         /* This better be a real Lustre superblock! */
2230         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2231
2232         /* Note we have not called client_common_fill_super yet, so
2233            proc fns must be able to handle that! */
2234         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2235                                       lcfg, sb);
2236         if (rc > 0)
2237                 rc = 0;
2238         return(rc);
2239 }
2240
2241 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2242 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2243                                        struct inode *i1, struct inode *i2,
2244                                        const char *name, int namelen,
2245                                        int mode, __u32 opc, void *data)
2246 {
2247         LASSERT(i1 != NULL);
2248
2249         if (namelen > ll_i2sbi(i1)->ll_namelen)
2250                 return ERR_PTR(-ENAMETOOLONG);
2251
2252         if (op_data == NULL)
2253                 OBD_ALLOC_PTR(op_data);
2254
2255         if (op_data == NULL)
2256                 return ERR_PTR(-ENOMEM);
2257
2258         ll_i2gids(op_data->op_suppgids, i1, i2);
2259         op_data->op_fid1 = *ll_inode2fid(i1);
2260         op_data->op_capa1 = ll_mdscapa_get(i1);
2261
2262         if (i2) {
2263                 op_data->op_fid2 = *ll_inode2fid(i2);
2264                 op_data->op_capa2 = ll_mdscapa_get(i2);
2265         } else {
2266                 fid_zero(&op_data->op_fid2);
2267                 op_data->op_capa2 = NULL;
2268         }
2269
2270         op_data->op_name = name;
2271         op_data->op_namelen = namelen;
2272         op_data->op_mode = mode;
2273         op_data->op_mod_time = cfs_time_current_sec();
2274         op_data->op_fsuid = cfs_curproc_fsuid();
2275         op_data->op_fsgid = cfs_curproc_fsgid();
2276         op_data->op_cap = cfs_curproc_cap_pack();
2277         op_data->op_bias = MDS_CHECK_SPLIT;
2278         op_data->op_opc = opc;
2279         op_data->op_mds = 0;
2280         op_data->op_data = data;
2281
2282         /* If the file is being opened after mknod() (normally due to NFS)
2283          * try to use the default stripe data from parent directory for
2284          * allocating OST objects.  Try to pass the parent FID to MDS. */
2285         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2286             !ll_i2info(i2)->lli_has_smd) {
2287                 struct ll_inode_info *lli = ll_i2info(i2);
2288
2289                 cfs_spin_lock(&lli->lli_lock);
2290                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2291                         op_data->op_fid1 = lli->lli_pfid;
2292                 cfs_spin_unlock(&lli->lli_lock);
2293                 /** We ignore parent's capability temporary. */
2294         }
2295
2296         return op_data;
2297 }
2298
2299 void ll_finish_md_op_data(struct md_op_data *op_data)
2300 {
2301         capa_put(op_data->op_capa1);
2302         capa_put(op_data->op_capa2);
2303         OBD_FREE_PTR(op_data);
2304 }
2305
2306 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2307 {
2308         struct ll_sb_info *sbi;
2309
2310         LASSERT((seq != NULL) && (vfs != NULL));
2311         sbi = ll_s2sbi(vfs->mnt_sb);
2312
2313         if (sbi->ll_flags & LL_SBI_NOLCK)
2314                 seq_puts(seq, ",nolock");
2315
2316         if (sbi->ll_flags & LL_SBI_FLOCK)
2317                 seq_puts(seq, ",flock");
2318
2319         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2320                 seq_puts(seq, ",localflock");
2321
2322         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2323                 seq_puts(seq, ",user_xattr");
2324
2325         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2326                 seq_puts(seq, ",lazystatfs");
2327
2328         RETURN(0);
2329 }
2330
2331 /**
2332  * Get obd name by cmd, and copy out to user space
2333  */
2334 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2335 {
2336         struct ll_sb_info *sbi = ll_i2sbi(inode);
2337         struct obd_device *obd;
2338         ENTRY;
2339
2340         if (cmd == OBD_IOC_GETDTNAME)
2341                 obd = class_exp2obd(sbi->ll_dt_exp);
2342         else if (cmd == OBD_IOC_GETMDNAME)
2343                 obd = class_exp2obd(sbi->ll_md_exp);
2344         else
2345                 RETURN(-EINVAL);
2346
2347         if (!obd)
2348                 RETURN(-ENOENT);
2349
2350         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2351                              strlen(obd->obd_name) + 1))
2352                 RETURN(-EFAULT);
2353
2354         RETURN(0);
2355 }