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 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  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         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 231         data->ocd_version = LUSTRE_VERSION_CODE;
 232
 233         if (sb->s_flags & MS_RDONLY)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 235         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 236                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 237
 238 #ifdef HAVE_MS_FLOCK_LOCK
 239         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 240         sb->s_flags |= MS_FLOCK_LOCK;
 241 #endif
 242 #ifdef MS_HAS_NEW_AOPS
 243         sb->s_flags |= MS_HAS_NEW_AOPS;
 244 #endif
 245
 246         if (sbi->ll_flags & LL_SBI_FLOCK)
 247                 sbi->ll_fop = &ll_file_operations_flock;
 248         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 249                 sbi->ll_fop = &ll_file_operations;
 250         else
 251                 sbi->ll_fop = &ll_file_operations_noflock;
 252
 253         /* real client */
 254         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 255         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 256                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 257
 258         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 259
 260         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 261         if (err == -EBUSY) {
 262                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 263                                    "recovery, of which this client is not a "
 264                                    "part. Please wait for recovery to complete,"
 265                                    " abort, or time out.\n", md);
 266                 GOTO(out, err);
 267         } else if (err) {
 268                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 269                 GOTO(out, err);
 270         }
 271
 272         err = obd_fid_init(sbi->ll_md_exp);
 273         if (err) {
 274                 CERROR("Can't init metadata layer FID infrastructure, "
 275                        "rc %d\n", err);
 276                 GOTO(out_md, err);
 277         }
 278
 279         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 280                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 281         if (err)
 282                 GOTO(out_md_fid, err);
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 292         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 297                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, CFS_PAGE_SIZE);
 305                 GOTO(out_md, err = -EPROTO);
 306         }
 307
 308         size = sizeof(*data);
 309         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 310                            KEY_CONN_DATA,  &size, data, NULL);
 311         if (err) {
 312                 CERROR("Get connect data failed: %d \n", err);
 313                 GOTO(out_md, err);
 314         }
 315
 316         LASSERT(osfs->os_bsize);
 317         sb->s_blocksize = osfs->os_bsize;
 318         sb->s_blocksize_bits = log2(osfs->os_bsize);
 319         sb->s_magic = LL_SUPER_MAGIC;
 320         sb->s_maxbytes = MAX_LFS_FILESIZE;
 321         sbi->ll_namelen = osfs->os_namelen;
 322         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 323
 324         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 325             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 326                 LCONSOLE_INFO("Disabling user_xattr feature because "
 327                               "it is not supported on the server\n");
 328                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 332 #ifdef MS_POSIXACL
 333                 sb->s_flags |= MS_POSIXACL;
 334 #endif
 335                 sbi->ll_flags |= LL_SBI_ACL;
 336         } else {
 337                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 338 #ifdef MS_POSIXACL
 339                 sb->s_flags &= ~MS_POSIXACL;
 340 #endif
 341                 sbi->ll_flags &= ~LL_SBI_ACL;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 345                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 346                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 347                         LCONSOLE_INFO("client is set as remote by default.\n");
 348                 }
 349         } else {
 350                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 351                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 352                         LCONSOLE_INFO("client claims to be remote, but server "
 353                                       "rejected, forced to be local.\n");
 354                 }
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 358                 LCONSOLE_INFO("client enabled MDS capability!\n");
 359                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 360         }
 361
 362         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 363                 LCONSOLE_INFO("client enabled OSS capability!\n");
 364                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 368                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 371                 sbi->ll_md_brw_size = data->ocd_brw_size;
 372         else
 373                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 374
 375         obd = class_name2obd(dt);
 376         if (!obd) {
 377                 CERROR("DT %s: not setup or attached\n", dt);
 378                 GOTO(out_md_fid, err = -ENODEV);
 379         }
 380
 381         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 382                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 383                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 384                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 385                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 386                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 387                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 388                                   OBD_CONNECT_MAXBYTES |
 389                                   OBD_CONNECT_EINPROGRESS |
 390                                   OBD_CONNECT_JOBSTATS;
 391
 392         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 393                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 394
 395         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 396                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 397                  * disabled by default, because it can still be enabled on the
 398                  * fly via /proc. As a consequence, we still need to come to an
 399                  * agreement on the supported algorithms at connect time */
 400                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 401
 402                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 403                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 404                 else
 405                         data->ocd_cksum_types = cksum_types_supported();
 406         }
 407
 408 #ifdef HAVE_LRU_RESIZE_SUPPORT
 409         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 410 #endif
 411         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 412                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 413
 414         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 415                "ocd_grant: %d\n", data->ocd_connect_flags,
 416                data->ocd_version, data->ocd_grant);
 417
 418         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 419         obd->obd_upcall.onu_upcall = cl_ocd_update;
 420
 421         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 422
 423         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 424         if (err == -EBUSY) {
 425                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 426                                    "recovery, of which this client is not a "
 427                                    "part.  Please wait for recovery to "
 428                                    "complete, abort, or time out.\n", dt);
 429                 GOTO(out_md_fid, err);
 430         } else if (err) {
 431                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 432                 GOTO(out_md_fid, err);
 433         }
 434
 435         err = obd_fid_init(sbi->ll_dt_exp);
 436         if (err) {
 437                 CERROR("Can't init data layer FID infrastructure, "
 438                        "rc %d\n", err);
 439                 GOTO(out_dt, err);
 440         }
 441
 442         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 443         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 444         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 445         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 446         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 447
 448         fid_zero(&sbi->ll_root_fid);
 449         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 450         if (err) {
 451                 CERROR("cannot mds_connect: rc = %d\n", err);
 452                 GOTO(out_lock_cn_cb, err);
 453         }
 454         if (!fid_is_sane(&sbi->ll_root_fid)) {
 455                 CERROR("Invalid root fid during mount\n");
 456                 GOTO(out_lock_cn_cb, err = -EINVAL);
 457         }
 458         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 459
 460         sb->s_op = &lustre_super_operations;
 461 #if THREAD_SIZE >= 8192
 462         sb->s_export_op = &lustre_export_operations;
 463 #endif
 464
 465         /* make root inode
 466          * XXX: move this to after cbd setup? */
 467         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 468         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 469                 valid |= OBD_MD_FLRMTPERM;
 470         else if (sbi->ll_flags & LL_SBI_ACL)
 471                 valid |= OBD_MD_FLACL;
 472
 473         OBD_ALLOC_PTR(op_data);
 474         if (op_data == NULL)
 475                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 476
 477         op_data->op_fid1 = sbi->ll_root_fid;
 478         op_data->op_mode = 0;
 479         op_data->op_capa1 = oc;
 480         op_data->op_valid = valid;
 481
 482         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 483         if (oc)
 484                 capa_put(oc);
 485         OBD_FREE_PTR(op_data);
 486         if (err) {
 487                 CERROR("md_getattr failed for root: rc = %d\n", err);
 488                 GOTO(out_lock_cn_cb, err);
 489         }
 490         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 491                                sbi->ll_md_exp, &lmd);
 492         if (err) {
 493                 CERROR("failed to understand root inode md: rc = %d\n", err);
 494                 ptlrpc_req_finished (request);
 495                 GOTO(out_lock_cn_cb, err);
 496         }
 497
 498         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 499         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 500         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 501         ptlrpc_req_finished(request);
 502
 503         if (root == NULL || IS_ERR(root)) {
 504                 if (lmd.lsm)
 505                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 506 #ifdef CONFIG_FS_POSIX_ACL
 507                 if (lmd.posix_acl) {
 508                         posix_acl_release(lmd.posix_acl);
 509                         lmd.posix_acl = NULL;
 510                 }
 511 #endif
 512                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 513                 root = NULL;
 514                 CERROR("lustre_lite: bad iget4 for root\n");
 515                 GOTO(out_root, err);
 516         }
 517
 518         err = ll_close_thread_start(&sbi->ll_lcq);
 519         if (err) {
 520                 CERROR("cannot start close thread: rc %d\n", err);
 521                 GOTO(out_root, err);
 522         }
 523
 524 #ifdef CONFIG_FS_POSIX_ACL
 525         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 526                 rct_init(&sbi->ll_rct);
 527                 et_init(&sbi->ll_et);
 528         }
 529 #endif
 530
 531         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 532         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 533                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 534                                  NULL);
 535         cl_sb_init(sb);
 536
 537         sb->s_root = d_alloc_root(root);
 538 #ifdef DCACHE_OP_HASH
 539         d_set_d_op(sb->s_root, &ll_d_root_ops);
 540         sb->s_d_op = &ll_d_ops;
 541 #else
 542         sb->s_root->d_op = &ll_d_root_ops;
 543 #endif
 544
 545         sbi->ll_sdev_orig = sb->s_dev;
 546
 547         /* We set sb->s_dev equal on all lustre clients in order to support
 548          * NFS export clustering.  NFSD requires that the FSID be the same
 549          * on all clients. */
 550         /* s_dev is also used in lt_compare() to compare two fs, but that is
 551          * only a node-local comparison. */
 552         uuid = obd_get_uuid(sbi->ll_md_exp);
 553         if (uuid != NULL)
 554                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 555         sbi->ll_mnt = mnt;
 556
 557         if (data != NULL)
 558                 OBD_FREE_PTR(data);
 559         if (osfs != NULL)
 560                 OBD_FREE_PTR(osfs);
 561
 562         RETURN(err);
 563 out_root:
 564         if (root)
 565                 iput(root);
 566 out_lock_cn_cb:
 567         obd_fid_fini(sbi->ll_dt_exp);
 568 out_dt:
 569         obd_disconnect(sbi->ll_dt_exp);
 570         sbi->ll_dt_exp = NULL;
 571 out_md_fid:
 572         obd_fid_fini(sbi->ll_md_exp);
 573 out_md:
 574         obd_disconnect(sbi->ll_md_exp);
 575         sbi->ll_md_exp = NULL;
 576 out:
 577         if (data != NULL)
 578                 OBD_FREE_PTR(data);
 579         if (osfs != NULL)
 580                 OBD_FREE_PTR(osfs);
 581         lprocfs_unregister_mountpoint(sbi);
 582         return err;
 583 }
 584
 585 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 586 {
 587         int size, rc;
 588
 589         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 590         size = sizeof(int);
 591         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 592                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 593         if (rc)
 594                 CERROR("Get max mdsize error rc %d \n", rc);
 595
 596         RETURN(rc);
 597 }
 598
 599 void ll_dump_inode(struct inode *inode)
 600 {
 601         struct list_head *tmp;
 602         int dentry_count = 0;
 603
 604         LASSERT(inode != NULL);
 605
 606         list_for_each(tmp, &inode->i_dentry)
 607                 dentry_count++;
 608
 609         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 610                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 611                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 612 }
 613
 614 void lustre_dump_dentry(struct dentry *dentry, int recur)
 615 {
 616         struct list_head *tmp;
 617         int subdirs = 0;
 618
 619         LASSERT(dentry != NULL);
 620
 621         list_for_each(tmp, &dentry->d_subdirs)
 622                 subdirs++;
 623
 624         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 625                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 626                dentry->d_name.len, dentry->d_name.name,
 627                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 628                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 629                dentry->d_flags, dentry->d_fsdata, subdirs);
 630         if (dentry->d_inode != NULL)
 631                 ll_dump_inode(dentry->d_inode);
 632
 633         if (recur == 0)
 634                 return;
 635
 636        list_for_each(tmp, &dentry->d_subdirs) {
 637                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 638                 lustre_dump_dentry(d, recur - 1);
 639         }
 640 }
 641
 642 void client_common_put_super(struct super_block *sb)
 643 {
 644         struct ll_sb_info *sbi = ll_s2sbi(sb);
 645         ENTRY;
 646
 647 #ifdef CONFIG_FS_POSIX_ACL
 648         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 649                 et_fini(&sbi->ll_et);
 650                 rct_fini(&sbi->ll_rct);
 651         }
 652 #endif
 653
 654         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 655
 656         ll_close_thread_shutdown(sbi->ll_lcq);
 657
 658         cl_sb_fini(sb);
 659
 660         cfs_list_del(&sbi->ll_conn_chain);
 661
 662         obd_fid_fini(sbi->ll_dt_exp);
 663         obd_disconnect(sbi->ll_dt_exp);
 664         sbi->ll_dt_exp = NULL;
 665
 666         lprocfs_unregister_mountpoint(sbi);
 667
 668         obd_fid_fini(sbi->ll_md_exp);
 669         obd_disconnect(sbi->ll_md_exp);
 670         sbi->ll_md_exp = NULL;
 671
 672         EXIT;
 673 }
 674
 675 void ll_kill_super(struct super_block *sb)
 676 {
 677         struct ll_sb_info *sbi;
 678
 679         ENTRY;
 680
 681         /* not init sb ?*/
 682         if (!(sb->s_flags & MS_ACTIVE))
 683                 return;
 684
 685         sbi = ll_s2sbi(sb);
 686         /* we need restore s_dev from changed for clustred NFS before put_super
 687          * because new kernels have cached s_dev and change sb->s_dev in
 688          * put_super not affected real removing devices */
 689         if (sbi)
 690                 sb->s_dev = sbi->ll_sdev_orig;
 691         EXIT;
 692 }
 693
 694 char *ll_read_opt(const char *opt, char *data)
 695 {
 696         char *value;
 697         char *retval;
 698         ENTRY;
 699
 700         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 701         if (strncmp(opt, data, strlen(opt)))
 702                 RETURN(NULL);
 703         if ((value = strchr(data, '=')) == NULL)
 704                 RETURN(NULL);
 705
 706         value++;
 707         OBD_ALLOC(retval, strlen(value) + 1);
 708         if (!retval) {
 709                 CERROR("out of memory!\n");
 710                 RETURN(NULL);
 711         }
 712
 713         memcpy(retval, value, strlen(value)+1);
 714         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 715         RETURN(retval);
 716 }
 717
 718 static inline int ll_set_opt(const char *opt, char *data, int fl)
 719 {
 720         if (strncmp(opt, data, strlen(opt)) != 0)
 721                 return(0);
 722         else
 723                 return(fl);
 724 }
 725
 726 /* non-client-specific mount options are parsed in lmd_parse */
 727 static int ll_options(char *options, int *flags)
 728 {
 729         int tmp;
 730         char *s1 = options, *s2;
 731         ENTRY;
 732
 733         if (!options)
 734                 RETURN(0);
 735
 736         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 737
 738         while (*s1) {
 739                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 740                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 741                 if (tmp) {
 742                         *flags |= tmp;
 743                         goto next;
 744                 }
 745                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 746                 if (tmp) {
 747                         *flags |= tmp;
 748                         goto next;
 749                 }
 750                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 751                 if (tmp) {
 752                         *flags |= tmp;
 753                         goto next;
 754                 }
 755                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 756                 if (tmp) {
 757                         *flags &= ~tmp;
 758                         goto next;
 759                 }
 760                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 761                 if (tmp) {
 762                         *flags |= tmp;
 763                         goto next;
 764                 }
 765                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 766                 if (tmp) {
 767                         *flags &= ~tmp;
 768                         goto next;
 769                 }
 770 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 771                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 772                 if (tmp) {
 773                         /* Ignore deprecated mount option.  The client will
 774                          * always try to mount with ACL support, whether this
 775                          * is used depends on whether server supports it. */
 776                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 777                                                   "mount option 'acl'.\n");
 778                         goto next;
 779                 }
 780                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 781                 if (tmp) {
 782                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 783                                                   "mount option 'noacl'.\n");
 784                         goto next;
 785                 }
 786 #else
 787 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 788 #endif
 789                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794
 795                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 796                 if (tmp) {
 797                         *flags |= tmp;
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 801                 if (tmp) {
 802                         *flags &= ~tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 806                 if (tmp) {
 807                         *flags |= tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 811                 if (tmp) {
 812                         *flags &= ~tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 821                 if (tmp) {
 822                         *flags &= ~tmp;
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 826                 if (tmp) {
 827                         *flags |= tmp;
 828                         goto next;
 829                 }
 830                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 841                 if (tmp) {
 842                         *flags &= ~tmp;
 843                         goto next;
 844                 }
 845                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 846                                    s1);
 847                 RETURN(-EINVAL);
 848
 849 next:
 850                 /* Find next opt */
 851                 s2 = strchr(s1, ',');
 852                 if (s2 == NULL)
 853                         break;
 854                 s1 = s2 + 1;
 855         }
 856         RETURN(0);
 857 }
 858
 859 void ll_lli_init(struct ll_inode_info *lli)
 860 {
 861         lli->lli_inode_magic = LLI_INODE_MAGIC;
 862         lli->lli_flags = 0;
 863         lli->lli_ioepoch = 0;
 864         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 865         cfs_spin_lock_init(&lli->lli_lock);
 866         lli->lli_posix_acl = NULL;
 867         lli->lli_remote_perms = NULL;
 868         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 869         /* Do not set lli_fid, it has been initialized already. */
 870         fid_zero(&lli->lli_pfid);
 871         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 872         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 873         cfs_atomic_set(&lli->lli_open_count, 0);
 874         lli->lli_mds_capa = NULL;
 875         lli->lli_rmtperm_time = 0;
 876         lli->lli_pending_och = NULL;
 877         lli->lli_mds_read_och = NULL;
 878         lli->lli_mds_write_och = NULL;
 879         lli->lli_mds_exec_och = NULL;
 880         lli->lli_open_fd_read_count = 0;
 881         lli->lli_open_fd_write_count = 0;
 882         lli->lli_open_fd_exec_count = 0;
 883         cfs_mutex_init(&lli->lli_och_mutex);
 884         cfs_spin_lock_init(&lli->lli_agl_lock);
 885         lli->lli_smd = NULL;
 886         lli->lli_clob = NULL;
 887
 888         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 889         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 890                 cfs_mutex_init(&lli->lli_readdir_mutex);
 891                 lli->lli_opendir_key = NULL;
 892                 lli->lli_sai = NULL;
 893                 lli->lli_sa_pos = 0;
 894                 lli->lli_def_acl = NULL;
 895                 cfs_spin_lock_init(&lli->lli_sa_lock);
 896                 lli->lli_opendir_pid = 0;
 897         } else {
 898                 cfs_sema_init(&lli->lli_size_sem, 1);
 899                 lli->lli_size_sem_owner = NULL;
 900                 lli->lli_symlink_name = NULL;
 901                 cfs_init_rwsem(&lli->lli_trunc_sem);
 902                 cfs_mutex_init(&lli->lli_write_mutex);
 903                 lli->lli_async_rc = 0;
 904                 lli->lli_write_rc = 0;
 905                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 906                 lli->lli_glimpse_time = 0;
 907                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 908                 lli->lli_agl_index = 0;
 909         }
 910 }
 911
 912 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 913 {
 914 #ifdef HAVE_BDI_REGISTER
 915         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 916
 917 #ifdef HAVE_BDI_NAME
 918         bdi->name = "lustre";
 919 #endif
 920         return bdi_register(bdi, NULL, "lustre-%d",
 921                             atomic_inc_return(&ll_bdi_num));
 922 #else
 923         return 0;
 924 #endif
 925 }
 926
 927 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 928 {
 929         struct lustre_profile *lprof = NULL;
 930         struct lustre_sb_info *lsi = s2lsi(sb);
 931         struct ll_sb_info *sbi;
 932         char  *dt = NULL, *md = NULL;
 933         char  *profilenm = get_profile_name(sb);
 934         struct config_llog_instance *cfg;
 935         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 936         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 937         int    err;
 938         ENTRY;
 939
 940         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 941
 942         OBD_ALLOC_PTR(cfg);
 943         if (cfg == NULL)
 944                 RETURN(-ENOMEM);
 945
 946         cfs_module_get();
 947
 948         /* client additional sb info */
 949         lsi->lsi_llsbi = sbi = ll_init_sbi();
 950         if (!sbi) {
 951                 cfs_module_put(THIS_MODULE);
 952                 OBD_FREE_PTR(cfg);
 953                 RETURN(-ENOMEM);
 954         }
 955
 956         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 957         if (err)
 958                 GOTO(out_free, err);
 959
 960         err = ll_bdi_init(&lsi->lsi_bdi);
 961         if (err)
 962                 GOTO(out_free, err);
 963         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 964         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 965         err = ll_bdi_register(&lsi->lsi_bdi);
 966         if (err)
 967                 GOTO(out_free, err);
 968
 969 #ifdef HAVE_SB_BDI
 970         sb->s_bdi = &lsi->lsi_bdi;
 971 #endif
 972
 973         /* Generate a string unique to this super, in case some joker tries
 974            to mount the same fs at two mount points.
 975            Use the address of the super itself.*/
 976         cfg->cfg_instance = sb;
 977         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 978
 979         /* set up client obds */
 980         err = lustre_process_log(sb, profilenm, cfg);
 981         if (err < 0) {
 982                 CERROR("Unable to process log: %d\n", err);
 983                 GOTO(out_free, err);
 984         }
 985
 986         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 987         lprof = class_get_profile(profilenm);
 988         if (lprof == NULL) {
 989                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 990                                    " read from the MGS.  Does that filesystem "
 991                                    "exist?\n", profilenm);
 992                 GOTO(out_free, err = -EINVAL);
 993         }
 994         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 995                lprof->lp_md, lprof->lp_dt);
 996
 997         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 998         if (!dt)
 999                 GOTO(out_free, err = -ENOMEM);
1000         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1001
1002         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1003         if (!md)
1004                 GOTO(out_free, err = -ENOMEM);
1005         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1006
1007         /* connections, registrations, sb setup */
1008         err = client_common_fill_super(sb, md, dt, mnt);
1009
1010 out_free:
1011         if (md)
1012                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1013         if (dt)
1014                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1015         if (err)
1016                 ll_put_super(sb);
1017         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1018                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1019
1020         OBD_FREE_PTR(cfg);
1021         RETURN(err);
1022 } /* ll_fill_super */
1023
1024
1025 void lu_context_keys_dump(void);
1026
1027 void ll_put_super(struct super_block *sb)
1028 {
1029         struct config_llog_instance cfg;
1030         struct obd_device *obd;
1031         struct lustre_sb_info *lsi = s2lsi(sb);
1032         struct ll_sb_info *sbi = ll_s2sbi(sb);
1033         char *profilenm = get_profile_name(sb);
1034         int force = 1, next;
1035         ENTRY;
1036
1037         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1038
1039         ll_print_capa_stat(sbi);
1040
1041         cfg.cfg_instance = sb;
1042         lustre_end_log(sb, profilenm, &cfg);
1043
1044         if (sbi->ll_md_exp) {
1045                 obd = class_exp2obd(sbi->ll_md_exp);
1046                 if (obd)
1047                         force = obd->obd_force;
1048         }
1049
1050         /* We need to set force before the lov_disconnect in
1051            lustre_common_put_super, since l_d cleans up osc's as well. */
1052         if (force) {
1053                 next = 0;
1054                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1055                                                      &next)) != NULL) {
1056                         obd->obd_force = force;
1057                 }
1058         }
1059
1060         if (sbi->ll_lcq) {
1061                 /* Only if client_common_fill_super succeeded */
1062                 client_common_put_super(sb);
1063         }
1064
1065         next = 0;
1066         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1067                 class_manual_cleanup(obd);
1068         }
1069
1070         if (sbi->ll_flags & LL_SBI_VERBOSE)
1071                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1072
1073         if (profilenm)
1074                 class_del_profile(profilenm);
1075
1076         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1077                 ll_bdi_destroy(&lsi->lsi_bdi);
1078                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1079         }
1080
1081         ll_free_sbi(sb);
1082         lsi->lsi_llsbi = NULL;
1083
1084         lustre_common_put_super(sb);
1085
1086         cl_env_cache_purge(~0);
1087
1088         cfs_module_put(THIS_MODULE);
1089
1090         EXIT;
1091 } /* client_put_super */
1092
1093 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1094 {
1095         struct inode *inode = NULL;
1096         /* NOTE: we depend on atomic igrab() -bzzz */
1097         lock_res_and_lock(lock);
1098         if (lock->l_ast_data) {
1099                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1100                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1101                         inode = igrab(lock->l_ast_data);
1102                 } else {
1103                         inode = lock->l_ast_data;
1104                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1105                                          D_WARNING, lock, "l_ast_data %p is "
1106                                          "bogus: magic %08x", lock->l_ast_data,
1107                                          lli->lli_inode_magic);
1108                         inode = NULL;
1109                 }
1110         }
1111         unlock_res_and_lock(lock);
1112         return inode;
1113 }
1114
1115 static int null_if_equal(struct ldlm_lock *lock, void *data)
1116 {
1117         if (data == lock->l_ast_data) {
1118                 lock->l_ast_data = NULL;
1119
1120                 if (lock->l_req_mode != lock->l_granted_mode)
1121                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1122         }
1123
1124         return LDLM_ITER_CONTINUE;
1125 }
1126
1127 void ll_clear_inode(struct inode *inode)
1128 {
1129         struct ll_inode_info *lli = ll_i2info(inode);
1130         struct ll_sb_info *sbi = ll_i2sbi(inode);
1131         ENTRY;
1132
1133         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1134                inode->i_generation, inode);
1135
1136         if (S_ISDIR(inode->i_mode)) {
1137                 /* these should have been cleared in ll_file_release */
1138                 LASSERT(lli->lli_opendir_key == NULL);
1139                 LASSERT(lli->lli_sai == NULL);
1140                 LASSERT(lli->lli_opendir_pid == 0);
1141         }
1142
1143         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1144         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1145                          null_if_equal, inode);
1146
1147         LASSERT(!lli->lli_open_fd_write_count);
1148         LASSERT(!lli->lli_open_fd_read_count);
1149         LASSERT(!lli->lli_open_fd_exec_count);
1150
1151         if (lli->lli_mds_write_och)
1152                 ll_md_real_close(inode, FMODE_WRITE);
1153         if (lli->lli_mds_exec_och)
1154                 ll_md_real_close(inode, FMODE_EXEC);
1155         if (lli->lli_mds_read_och)
1156                 ll_md_real_close(inode, FMODE_READ);
1157
1158         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1159                 OBD_FREE(lli->lli_symlink_name,
1160                          strlen(lli->lli_symlink_name) + 1);
1161                 lli->lli_symlink_name = NULL;
1162         }
1163
1164         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1165                 LASSERT(lli->lli_posix_acl == NULL);
1166                 if (lli->lli_remote_perms) {
1167                         free_rmtperm_hash(lli->lli_remote_perms);
1168                         lli->lli_remote_perms = NULL;
1169                 }
1170         }
1171 #ifdef CONFIG_FS_POSIX_ACL
1172         else if (lli->lli_posix_acl) {
1173                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1174                 LASSERT(lli->lli_remote_perms == NULL);
1175                 posix_acl_release(lli->lli_posix_acl);
1176                 lli->lli_posix_acl = NULL;
1177         }
1178 #endif
1179         lli->lli_inode_magic = LLI_INODE_DEAD;
1180
1181         ll_clear_inode_capas(inode);
1182         if (!S_ISDIR(inode->i_mode))
1183                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1184
1185         /*
1186          * XXX This has to be done before lsm is freed below, because
1187          * cl_object still uses inode lsm.
1188          */
1189         cl_inode_fini(inode);
1190
1191         if (lli->lli_smd) {
1192                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1193                 lli->lli_smd = NULL;
1194         }
1195
1196
1197         EXIT;
1198 }
1199
1200 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1201                   struct md_open_data **mod)
1202 {
1203         struct lustre_md md;
1204         struct inode *inode = dentry->d_inode;
1205         struct ll_sb_info *sbi = ll_i2sbi(inode);
1206         struct ptlrpc_request *request = NULL;
1207         int rc, ia_valid;
1208         ENTRY;
1209
1210         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1211                                      LUSTRE_OPC_ANY, NULL);
1212         if (IS_ERR(op_data))
1213                 RETURN(PTR_ERR(op_data));
1214
1215         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1216                         &request, mod);
1217         if (rc) {
1218                 ptlrpc_req_finished(request);
1219                 if (rc == -ENOENT) {
1220                         inode->i_nlink = 0;
1221                         /* Unlinked special device node? Or just a race?
1222                          * Pretend we done everything. */
1223                         if (!S_ISREG(inode->i_mode) &&
1224                             !S_ISDIR(inode->i_mode)) {
1225                                 ia_valid = op_data->op_attr.ia_valid;
1226                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1227                                 rc = simple_setattr(dentry, &op_data->op_attr);
1228                                 op_data->op_attr.ia_valid = ia_valid;
1229                         }
1230                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1231                         CERROR("md_setattr fails: rc = %d\n", rc);
1232                 }
1233                 RETURN(rc);
1234         }
1235
1236         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1237                               sbi->ll_md_exp, &md);
1238         if (rc) {
1239                 ptlrpc_req_finished(request);
1240                 RETURN(rc);
1241         }
1242
1243         /* We want to adjust timestamps.
1244          * If there is at least some data in file, we cleared ATTR_SIZE
1245          * to avoid update size, otherwise it is important to do.(SOM case)
1246          * (bug 6196) */
1247         ia_valid = op_data->op_attr.ia_valid;
1248         /* Since we set ATTR_*_SET flags above, and already done permission
1249          * check, So don't let inode_change_ok() check it again.  */
1250         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1251         rc = simple_setattr(dentry, &op_data->op_attr);
1252         op_data->op_attr.ia_valid = ia_valid;
1253
1254         /* Extract epoch data if obtained. */
1255         op_data->op_handle = md.body->handle;
1256         op_data->op_ioepoch = md.body->ioepoch;
1257
1258         ll_update_inode(inode, &md);
1259         ptlrpc_req_finished(request);
1260
1261         RETURN(rc);
1262 }
1263
1264 /* Close IO epoch and send Size-on-MDS attribute update. */
1265 static int ll_setattr_done_writing(struct inode *inode,
1266                                    struct md_op_data *op_data,
1267                                    struct md_open_data *mod)
1268 {
1269         struct ll_inode_info *lli = ll_i2info(inode);
1270         int rc = 0;
1271         ENTRY;
1272
1273         LASSERT(op_data != NULL);
1274         if (!S_ISREG(inode->i_mode))
1275                 RETURN(0);
1276
1277         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1278                op_data->op_ioepoch, PFID(&lli->lli_fid));
1279
1280         op_data->op_flags = MF_EPOCH_CLOSE;
1281         ll_done_writing_attr(inode, op_data);
1282         ll_pack_inode2opdata(inode, op_data, NULL);
1283
1284         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1285         if (rc == -EAGAIN) {
1286                 /* MDS has instructed us to obtain Size-on-MDS attribute
1287                  * from OSTs and send setattr to back to MDS. */
1288                 rc = ll_som_update(inode, op_data);
1289         } else if (rc) {
1290                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1291                        inode->i_ino, rc);
1292         }
1293         RETURN(rc);
1294 }
1295
1296 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1297 {
1298         struct obd_capa *capa;
1299         int rc;
1300
1301         if (attr->ia_valid & ATTR_SIZE)
1302                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1303         else
1304                 capa = ll_mdscapa_get(inode);
1305
1306         rc = cl_setattr_ost(inode, attr, capa);
1307
1308         if (attr->ia_valid & ATTR_SIZE)
1309                 ll_truncate_free_capa(capa);
1310         else
1311                 capa_put(capa);
1312
1313         return rc;
1314 }
1315
1316 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1317  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1318  * keep these values until such a time that objects are allocated for it.
1319  * We do the MDS operations first, as it is checking permissions for us.
1320  * We don't to the MDS RPC if there is nothing that we want to store there,
1321  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1322  * going to do an RPC anyways.
1323  *
1324  * If we are doing a truncate, we will send the mtime and ctime updates
1325  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1326  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1327  * at the same time.
1328  */
1329 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1330 {
1331         struct inode *inode = dentry->d_inode;
1332         struct ll_inode_info *lli = ll_i2info(inode);
1333         struct lov_stripe_md *lsm = lli->lli_smd;
1334         struct md_op_data *op_data = NULL;
1335         struct md_open_data *mod = NULL;
1336         int ia_valid = attr->ia_valid;
1337         int rc = 0, rc1 = 0;
1338         ENTRY;
1339
1340         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1341                attr->ia_valid);
1342
1343         if (ia_valid & ATTR_SIZE) {
1344                 /* Check new size against VFS/VM file size limit and rlimit */
1345                 rc = inode_newsize_ok(inode, attr->ia_size);
1346                 if (rc)
1347                         RETURN(rc);
1348
1349                 /* The maximum Lustre file size is variable, based on the
1350                  * OST maximum object size and number of stripes.  This
1351                  * needs another check in addition to the VFS check above. */
1352                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1353                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1354                                PFID(&lli->lli_fid), attr->ia_size,
1355                                ll_file_maxbytes(inode));
1356                         RETURN(-EFBIG);
1357                 }
1358
1359                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1360         }
1361
1362         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1363         if (ia_valid & TIMES_SET_FLAGS) {
1364                 if (cfs_curproc_fsuid() != inode->i_uid &&
1365                     !cfs_capable(CFS_CAP_FOWNER))
1366                         RETURN(-EPERM);
1367         }
1368
1369         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1370         if (attr->ia_valid & ATTR_CTIME) {
1371                 attr->ia_ctime = CFS_CURRENT_TIME;
1372                 attr->ia_valid |= ATTR_CTIME_SET;
1373         }
1374         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1375                 attr->ia_atime = CFS_CURRENT_TIME;
1376                 attr->ia_valid |= ATTR_ATIME_SET;
1377         }
1378         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1379                 attr->ia_mtime = CFS_CURRENT_TIME;
1380                 attr->ia_valid |= ATTR_MTIME_SET;
1381         }
1382
1383         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1384                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1385                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1386                        cfs_time_current_sec());
1387
1388         /* NB: ATTR_SIZE will only be set after this point if the size
1389          * resides on the MDS, ie, this file has no objects. */
1390         if (lsm)
1391                 attr->ia_valid &= ~ATTR_SIZE;
1392
1393         /* We always do an MDS RPC, even if we're only changing the size;
1394          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1395
1396         OBD_ALLOC_PTR(op_data);
1397         if (op_data == NULL)
1398                 RETURN(-ENOMEM);
1399
1400         if (!S_ISDIR(inode->i_mode)) {
1401                 if (ia_valid & ATTR_SIZE)
1402                         UP_WRITE_I_ALLOC_SEM(inode);
1403                 UNLOCK_INODE_MUTEX(inode);
1404                 cfs_down_write(&lli->lli_trunc_sem);
1405                 LOCK_INODE_MUTEX(inode);
1406                 if (ia_valid & ATTR_SIZE)
1407                         DOWN_WRITE_I_ALLOC_SEM(inode);
1408         }
1409
1410         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1411
1412         /* Open epoch for truncate. */
1413         if (exp_connect_som(ll_i2mdexp(inode)) &&
1414             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1415                 op_data->op_flags = MF_EPOCH_OPEN;
1416
1417         rc = ll_md_setattr(dentry, op_data, &mod);
1418         if (rc)
1419                 GOTO(out, rc);
1420
1421         ll_ioepoch_open(lli, op_data->op_ioepoch);
1422         if (!lsm || !S_ISREG(inode->i_mode)) {
1423                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1424                 GOTO(out, rc = 0);
1425         }
1426
1427         if (ia_valid & ATTR_SIZE)
1428                 attr->ia_valid |= ATTR_SIZE;
1429         if (ia_valid & (ATTR_SIZE |
1430                         ATTR_ATIME | ATTR_ATIME_SET |
1431                         ATTR_MTIME | ATTR_MTIME_SET))
1432                 /* For truncate and utimes sending attributes to OSTs, setting
1433                  * mtime/atime to the past will be performed under PW [0:EOF]
1434                  * extent lock (new_size:EOF for truncate).  It may seem
1435                  * excessive to send mtime/atime updates to OSTs when not
1436                  * setting times to past, but it is necessary due to possible
1437                  * time de-synchronization between MDT inode and OST objects */
1438                 rc = ll_setattr_ost(inode, attr);
1439         EXIT;
1440 out:
1441         if (op_data) {
1442                 if (op_data->op_ioepoch) {
1443                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1444                         if (!rc)
1445                                 rc = rc1;
1446                 }
1447                 ll_finish_md_op_data(op_data);
1448         }
1449         if (!S_ISDIR(inode->i_mode))
1450                 cfs_up_write(&lli->lli_trunc_sem);
1451
1452         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1453                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1454
1455         return rc;
1456 }
1457
1458 int ll_setattr(struct dentry *de, struct iattr *attr)
1459 {
1460         int mode = de->d_inode->i_mode;
1461
1462         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1463                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1464                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1465
1466         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1467                                (ATTR_SIZE|ATTR_MODE)) &&
1468             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1469              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1470               !(attr->ia_mode & S_ISGID))))
1471                 attr->ia_valid |= ATTR_FORCE;
1472
1473         if ((mode & S_ISUID) &&
1474             !(attr->ia_mode & S_ISUID) &&
1475             !(attr->ia_valid & ATTR_KILL_SUID))
1476                 attr->ia_valid |= ATTR_KILL_SUID;
1477
1478         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1479             !(attr->ia_mode & S_ISGID) &&
1480             !(attr->ia_valid & ATTR_KILL_SGID))
1481                 attr->ia_valid |= ATTR_KILL_SGID;
1482
1483         return ll_setattr_raw(de, attr);
1484 }
1485
1486 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1487                        __u64 max_age, __u32 flags)
1488 {
1489         struct ll_sb_info *sbi = ll_s2sbi(sb);
1490         struct obd_statfs obd_osfs;
1491         int rc;
1492         ENTRY;
1493
1494         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1495         if (rc) {
1496                 CERROR("md_statfs fails: rc = %d\n", rc);
1497                 RETURN(rc);
1498         }
1499
1500         osfs->os_type = sb->s_magic;
1501
1502         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1503                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1504
1505         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1506                 flags |= OBD_STATFS_NODELAY;
1507
1508         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1509         if (rc) {
1510                 CERROR("obd_statfs fails: rc = %d\n", rc);
1511                 RETURN(rc);
1512         }
1513
1514         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1515                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1516                obd_osfs.os_files);
1517
1518         osfs->os_bsize = obd_osfs.os_bsize;
1519         osfs->os_blocks = obd_osfs.os_blocks;
1520         osfs->os_bfree = obd_osfs.os_bfree;
1521         osfs->os_bavail = obd_osfs.os_bavail;
1522
1523         /* If we don't have as many objects free on the OST as inodes
1524          * on the MDS, we reduce the total number of inodes to
1525          * compensate, so that the "inodes in use" number is correct.
1526          */
1527         if (obd_osfs.os_ffree < osfs->os_ffree) {
1528                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1529                         obd_osfs.os_ffree;
1530                 osfs->os_ffree = obd_osfs.os_ffree;
1531         }
1532
1533         RETURN(rc);
1534 }
1535 #ifndef HAVE_STATFS_DENTRY_PARAM
1536 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1537 {
1538 #else
1539 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1540 {
1541         struct super_block *sb = de->d_sb;
1542 #endif
1543         struct obd_statfs osfs;
1544         int rc;
1545
1546         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1547         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1548
1549         /* Some amount of caching on the client is allowed */
1550         rc = ll_statfs_internal(sb, &osfs,
1551                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1552                                 0);
1553         if (rc)
1554                 return rc;
1555
1556         statfs_unpack(sfs, &osfs);
1557
1558         /* We need to downshift for all 32-bit kernels, because we can't
1559          * tell if the kernel is being called via sys_statfs64() or not.
1560          * Stop before overflowing f_bsize - in which case it is better
1561          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1562         if (sizeof(long) < 8) {
1563                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1564                         sfs->f_bsize <<= 1;
1565
1566                         osfs.os_blocks >>= 1;
1567                         osfs.os_bfree >>= 1;
1568                         osfs.os_bavail >>= 1;
1569                 }
1570         }
1571
1572         sfs->f_blocks = osfs.os_blocks;
1573         sfs->f_bfree = osfs.os_bfree;
1574         sfs->f_bavail = osfs.os_bavail;
1575
1576         return 0;
1577 }
1578
1579 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1580 {
1581         struct ll_inode_info *lli;
1582         struct lov_stripe_md *lsm;
1583
1584         LASSERT(!S_ISDIR(inode->i_mode));
1585
1586         lli = ll_i2info(inode);
1587         LASSERT(lli->lli_size_sem_owner != current);
1588         cfs_down(&lli->lli_size_sem);
1589         LASSERT(lli->lli_size_sem_owner == NULL);
1590         lli->lli_size_sem_owner = current;
1591         lsm = lli->lli_smd;
1592         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1593                  lsm, lock_lsm);
1594         if (lock_lsm)
1595                 lov_stripe_lock(lsm);
1596 }
1597
1598 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1599 {
1600         struct ll_inode_info *lli;
1601         struct lov_stripe_md *lsm;
1602
1603         lli = ll_i2info(inode);
1604         lsm = lli->lli_smd;
1605         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1606                  lsm, unlock_lsm);
1607         if (unlock_lsm)
1608                 lov_stripe_unlock(lsm);
1609         LASSERT(lli->lli_size_sem_owner == current);
1610         lli->lli_size_sem_owner = NULL;
1611         cfs_up(&lli->lli_size_sem);
1612 }
1613
1614 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1615 {
1616         struct ll_inode_info *lli = ll_i2info(inode);
1617         struct mdt_body *body = md->body;
1618         struct lov_stripe_md *lsm = md->lsm;
1619         struct ll_sb_info *sbi = ll_i2sbi(inode);
1620
1621         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1622         if (lsm != NULL) {
1623                 LASSERT(S_ISREG(inode->i_mode));
1624
1625                 cfs_mutex_lock(&lli->lli_och_mutex);
1626                 if (lli->lli_smd == NULL) {
1627                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1628                             lsm->lsm_magic != LOV_MAGIC_V3) {
1629                                 dump_lsm(D_ERROR, lsm);
1630                                 LBUG();
1631                         }
1632                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1633                                lsm, inode->i_ino, inode->i_generation, inode);
1634                         /* cl_file_inode_init must go before lli_smd or a race
1635                          * is possible where client thinks the file has stripes,
1636                          * but lov raid0 is not setup yet and parallel e.g.
1637                          * glimpse would try to use uninitialized lov */
1638                         cl_file_inode_init(inode, md);
1639                         cfs_spin_lock(&lli->lli_lock);
1640                         lli->lli_smd = lsm;
1641                         cfs_spin_unlock(&lli->lli_lock);
1642                         cfs_mutex_unlock(&lli->lli_och_mutex);
1643                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1644                         if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1645                                 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1646                 } else {
1647                         cfs_mutex_unlock(&lli->lli_och_mutex);
1648                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1649                                 lli->lli_smd->lsm_stripe_count ==
1650                                 lsm->lsm_stripe_count);
1651                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1652                                 CERROR("lsm mismatch for inode %ld\n",
1653                                        inode->i_ino);
1654                                 CERROR("lli_smd:\n");
1655                                 dump_lsm(D_ERROR, lli->lli_smd);
1656                                 CERROR("lsm:\n");
1657                                 dump_lsm(D_ERROR, lsm);
1658                                 LBUG();
1659                         }
1660                 }
1661                 if (lli->lli_smd != lsm)
1662                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1663         }
1664
1665         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1666                 if (body->valid & OBD_MD_FLRMTPERM)
1667                         ll_update_remote_perm(inode, md->remote_perm);
1668         }
1669 #ifdef CONFIG_FS_POSIX_ACL
1670         else if (body->valid & OBD_MD_FLACL) {
1671                 cfs_spin_lock(&lli->lli_lock);
1672                 if (lli->lli_posix_acl)
1673                         posix_acl_release(lli->lli_posix_acl);
1674                 lli->lli_posix_acl = md->posix_acl;
1675                 cfs_spin_unlock(&lli->lli_lock);
1676         }
1677 #endif
1678         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1679         inode->i_generation = cl_fid_build_gen(&body->fid1);
1680
1681         if (body->valid & OBD_MD_FLATIME) {
1682                 if (body->atime > LTIME_S(inode->i_atime))
1683                         LTIME_S(inode->i_atime) = body->atime;
1684                 lli->lli_lvb.lvb_atime = body->atime;
1685         }
1686         if (body->valid & OBD_MD_FLMTIME) {
1687                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1688                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1689                                "to "LPU64"\n", inode->i_ino,
1690                                LTIME_S(inode->i_mtime), body->mtime);
1691                         LTIME_S(inode->i_mtime) = body->mtime;
1692                 }
1693                 lli->lli_lvb.lvb_mtime = body->mtime;
1694         }
1695         if (body->valid & OBD_MD_FLCTIME) {
1696                 if (body->ctime > LTIME_S(inode->i_ctime))
1697                         LTIME_S(inode->i_ctime) = body->ctime;
1698                 lli->lli_lvb.lvb_ctime = body->ctime;
1699         }
1700         if (body->valid & OBD_MD_FLMODE)
1701                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1702         if (body->valid & OBD_MD_FLTYPE)
1703                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1704         LASSERT(inode->i_mode != 0);
1705         if (S_ISREG(inode->i_mode)) {
1706                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1707         } else {
1708                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1709         }
1710 #ifdef HAVE_INODE_BLKSIZE
1711         inode->i_blksize = 1<<inode->i_blkbits;
1712 #endif
1713         if (body->valid & OBD_MD_FLUID)
1714                 inode->i_uid = body->uid;
1715         if (body->valid & OBD_MD_FLGID)
1716                 inode->i_gid = body->gid;
1717         if (body->valid & OBD_MD_FLFLAGS)
1718                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1719         if (body->valid & OBD_MD_FLNLINK)
1720                 inode->i_nlink = body->nlink;
1721         if (body->valid & OBD_MD_FLRDEV)
1722                 inode->i_rdev = old_decode_dev(body->rdev);
1723
1724         if (body->valid & OBD_MD_FLID) {
1725                 /* FID shouldn't be changed! */
1726                 if (fid_is_sane(&lli->lli_fid)) {
1727                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1728                                  "Trying to change FID "DFID
1729                                  " to the "DFID", inode %lu/%u(%p)\n",
1730                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1731                                  inode->i_ino, inode->i_generation, inode);
1732                 } else
1733                         lli->lli_fid = body->fid1;
1734         }
1735
1736         LASSERT(fid_seq(&lli->lli_fid) != 0);
1737
1738         if (body->valid & OBD_MD_FLSIZE) {
1739                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1740                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1741                         struct lustre_handle lockh;
1742                         ldlm_mode_t mode;
1743
1744                         /* As it is possible a blocking ast has been processed
1745                          * by this time, we need to check there is an UPDATE
1746                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1747                          * it. */
1748                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1749                                                &lockh);
1750                         if (mode) {
1751                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1752                                                       LLIF_EPOCH_PENDING |
1753                                                       LLIF_SOM_DIRTY)) {
1754                                         CERROR("ino %lu flags %u still has "
1755                                                "size authority! do not trust "
1756                                                "the size got from MDS\n",
1757                                                inode->i_ino, lli->lli_flags);
1758                                 } else {
1759                                         /* Use old size assignment to avoid
1760                                          * deadlock bz14138 & bz14326 */
1761                                         i_size_write(inode, body->size);
1762                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1763                                 }
1764                                 ldlm_lock_decref(&lockh, mode);
1765                         }
1766                 } else {
1767                         /* Use old size assignment to avoid
1768                          * deadlock bz14138 & bz14326 */
1769                         i_size_write(inode, body->size);
1770
1771                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1772                                inode->i_ino, (unsigned long long)body->size);
1773                 }
1774
1775                 if (body->valid & OBD_MD_FLBLOCKS)
1776                         inode->i_blocks = body->blocks;
1777         }
1778
1779         if (body->valid & OBD_MD_FLMDSCAPA) {
1780                 LASSERT(md->mds_capa);
1781                 ll_add_capa(inode, md->mds_capa);
1782         }
1783         if (body->valid & OBD_MD_FLOSSCAPA) {
1784                 LASSERT(md->oss_capa);
1785                 ll_add_capa(inode, md->oss_capa);
1786         }
1787 }
1788
1789 void ll_read_inode2(struct inode *inode, void *opaque)
1790 {
1791         struct lustre_md *md = opaque;
1792         struct ll_inode_info *lli = ll_i2info(inode);
1793         ENTRY;
1794
1795         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1796                PFID(&lli->lli_fid), inode);
1797
1798         LASSERT(!lli->lli_smd);
1799
1800         /* Core attributes from the MDS first.  This is a new inode, and
1801          * the VFS doesn't zero times in the core inode so we have to do
1802          * it ourselves.  They will be overwritten by either MDS or OST
1803          * attributes - we just need to make sure they aren't newer. */
1804         LTIME_S(inode->i_mtime) = 0;
1805         LTIME_S(inode->i_atime) = 0;
1806         LTIME_S(inode->i_ctime) = 0;
1807         inode->i_rdev = 0;
1808         ll_update_inode(inode, md);
1809
1810         /* OIDEBUG(inode); */
1811
1812         /* initializing backing dev info. */
1813         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1814
1815
1816         if (S_ISREG(inode->i_mode)) {
1817                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1818                 inode->i_op = &ll_file_inode_operations;
1819                 inode->i_fop = sbi->ll_fop;
1820                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1821                 EXIT;
1822         } else if (S_ISDIR(inode->i_mode)) {
1823                 inode->i_op = &ll_dir_inode_operations;
1824                 inode->i_fop = &ll_dir_operations;
1825                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1826                 EXIT;
1827         } else if (S_ISLNK(inode->i_mode)) {
1828                 inode->i_op = &ll_fast_symlink_inode_operations;
1829                 EXIT;
1830         } else {
1831                 inode->i_op = &ll_special_inode_operations;
1832
1833                 init_special_inode(inode, inode->i_mode,
1834                                    kdev_t_to_nr(inode->i_rdev));
1835
1836                 EXIT;
1837         }
1838 }
1839
1840 void ll_delete_inode(struct inode *inode)
1841 {
1842         struct ll_sb_info *sbi = ll_i2sbi(inode);
1843         int rc;
1844         ENTRY;
1845
1846         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1847         if (rc)
1848                 CERROR("fid_delete() failed, rc %d\n", rc);
1849
1850         truncate_inode_pages(&inode->i_data, 0);
1851
1852         /* Workaround for LU-118 */
1853         if (inode->i_data.nrpages) {
1854                 TREE_READ_LOCK_IRQ(&inode->i_data);
1855                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1856                 LASSERTF(inode->i_data.nrpages == 0,
1857                          "inode=%lu/%u(%p) nrpages=%lu, see "
1858                          "http://jira.whamcloud.com/browse/LU-118\n",
1859                          inode->i_ino, inode->i_generation, inode,
1860                          inode->i_data.nrpages);
1861         }
1862         /* Workaround end */
1863
1864 #ifdef HAVE_SBOPS_EVICT_INODE
1865         ll_clear_inode(inode);
1866         end_writeback(inode);
1867 #else
1868         clear_inode(inode);
1869 #endif
1870
1871         EXIT;
1872 }
1873
1874 int ll_iocontrol(struct inode *inode, struct file *file,
1875                  unsigned int cmd, unsigned long arg)
1876 {
1877         struct ll_sb_info *sbi = ll_i2sbi(inode);
1878         struct ptlrpc_request *req = NULL;
1879         int rc, flags = 0;
1880         ENTRY;
1881
1882         switch(cmd) {
1883         case FSFILT_IOC_GETFLAGS: {
1884                 struct mdt_body *body;
1885                 struct md_op_data *op_data;
1886
1887                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1888                                              0, 0, LUSTRE_OPC_ANY,
1889                                              NULL);
1890                 if (IS_ERR(op_data))
1891                         RETURN(PTR_ERR(op_data));
1892
1893                 op_data->op_valid = OBD_MD_FLFLAGS;
1894                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1895                 ll_finish_md_op_data(op_data);
1896                 if (rc) {
1897                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1898                         RETURN(-abs(rc));
1899                 }
1900
1901                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1902
1903                 flags = body->flags;
1904
1905                 ptlrpc_req_finished(req);
1906
1907                 RETURN(put_user(flags, (int *)arg));
1908         }
1909         case FSFILT_IOC_SETFLAGS: {
1910                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1911                 struct obd_info oinfo = { { { 0 } } };
1912                 struct md_op_data *op_data;
1913
1914                 if (get_user(flags, (int *)arg))
1915                         RETURN(-EFAULT);
1916
1917                 oinfo.oi_md = lsm;
1918                 OBDO_ALLOC(oinfo.oi_oa);
1919                 if (!oinfo.oi_oa)
1920                         RETURN(-ENOMEM);
1921
1922                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1923                                              LUSTRE_OPC_ANY, NULL);
1924                 if (IS_ERR(op_data))
1925                         RETURN(PTR_ERR(op_data));
1926
1927                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1928                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1929                 rc = md_setattr(sbi->ll_md_exp, op_data,
1930                                 NULL, 0, NULL, 0, &req, NULL);
1931                 ll_finish_md_op_data(op_data);
1932                 ptlrpc_req_finished(req);
1933                 if (rc) {
1934                         OBDO_FREE(oinfo.oi_oa);
1935                         RETURN(rc);
1936                 }
1937
1938                 if (lsm == NULL) {
1939                         OBDO_FREE(oinfo.oi_oa);
1940                         GOTO(update_cache, rc);
1941                 }
1942
1943                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1944                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1945                 oinfo.oi_oa->o_flags = flags;
1946                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1947                                        OBD_MD_FLGROUP;
1948                 oinfo.oi_capa = ll_mdscapa_get(inode);
1949                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1950                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1951                 capa_put(oinfo.oi_capa);
1952                 OBDO_FREE(oinfo.oi_oa);
1953                 if (rc) {
1954                         if (rc != -EPERM && rc != -EACCES)
1955                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1956                         RETURN(rc);
1957                 }
1958
1959                 EXIT;
1960 update_cache:
1961                 inode->i_flags = ll_ext_to_inode_flags(flags);
1962                 return 0;
1963         }
1964         default:
1965                 RETURN(-ENOSYS);
1966         }
1967
1968         RETURN(0);
1969 }
1970
1971 int ll_flush_ctx(struct inode *inode)
1972 {
1973         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1974
1975         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1976
1977         obd_set_info_async(NULL, sbi->ll_md_exp,
1978                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1979                            0, NULL, NULL);
1980         obd_set_info_async(NULL, sbi->ll_dt_exp,
1981                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1982                            0, NULL, NULL);
1983         return 0;
1984 }
1985
1986 /* umount -f client means force down, don't save state */
1987 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1988 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1989 {
1990         struct super_block *sb = vfsmnt->mnt_sb;
1991 #else
1992 void ll_umount_begin(struct super_block *sb)
1993 {
1994 #endif
1995         struct lustre_sb_info *lsi = s2lsi(sb);
1996         struct ll_sb_info *sbi = ll_s2sbi(sb);
1997         struct obd_device *obd;
1998         struct obd_ioctl_data *ioc_data;
1999         ENTRY;
2000
2001 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2002         if (!(flags & MNT_FORCE)) {
2003                 EXIT;
2004                 return;
2005         }
2006 #endif
2007
2008         /* Tell the MGC we got umount -f */
2009         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2010
2011         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2012                sb->s_count, atomic_read(&sb->s_active));
2013
2014         obd = class_exp2obd(sbi->ll_md_exp);
2015         if (obd == NULL) {
2016                 CERROR("Invalid MDC connection handle "LPX64"\n",
2017                        sbi->ll_md_exp->exp_handle.h_cookie);
2018                 EXIT;
2019                 return;
2020         }
2021         obd->obd_force = 1;
2022
2023         obd = class_exp2obd(sbi->ll_dt_exp);
2024         if (obd == NULL) {
2025                 CERROR("Invalid LOV connection handle "LPX64"\n",
2026                        sbi->ll_dt_exp->exp_handle.h_cookie);
2027                 EXIT;
2028                 return;
2029         }
2030         obd->obd_force = 1;
2031
2032         OBD_ALLOC_PTR(ioc_data);
2033         if (ioc_data) {
2034                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2035                               sizeof ioc_data, ioc_data, NULL);
2036
2037                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2038                               sizeof ioc_data, ioc_data, NULL);
2039
2040                 OBD_FREE_PTR(ioc_data);
2041         }
2042
2043
2044         /* Really, we'd like to wait until there are no requests outstanding,
2045          * and then continue.  For now, we just invalidate the requests,
2046          * schedule() and sleep one second if needed, and hope.
2047          */
2048         cfs_schedule();
2049 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2050         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2051                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2052                                                    cfs_time_seconds(1));
2053                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2054                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2055                                       "may try to umount it a bit later\n",
2056                                       atomic_read(&vfsmnt->mnt_count));
2057         }
2058 #endif
2059
2060         EXIT;
2061 }
2062
2063 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2064 {
2065         struct ll_sb_info *sbi = ll_s2sbi(sb);
2066         char *profilenm = get_profile_name(sb);
2067         int err;
2068         __u32 read_only;
2069
2070         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2071                 read_only = *flags & MS_RDONLY;
2072                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2073                                          sizeof(KEY_READ_ONLY),
2074                                          KEY_READ_ONLY, sizeof(read_only),
2075                                          &read_only, NULL);
2076                 if (err) {
2077                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2078                                       profilenm, read_only ?
2079                                       "read-only" : "read-write", err);
2080                         return err;
2081                 }
2082
2083                 if (read_only)
2084                         sb->s_flags |= MS_RDONLY;
2085                 else
2086                         sb->s_flags &= ~MS_RDONLY;
2087
2088                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2089                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2090                                       read_only ?  "read-only" : "read-write");
2091         }
2092         return 0;
2093 }
2094
2095 int ll_prep_inode(struct inode **inode,
2096                   struct ptlrpc_request *req,
2097                   struct super_block *sb)
2098 {
2099         struct ll_sb_info *sbi = NULL;
2100         struct lustre_md md;
2101         int rc;
2102         ENTRY;
2103
2104         LASSERT(*inode || sb);
2105         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2106         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2107                               sbi->ll_md_exp, &md);
2108         if (rc)
2109                 RETURN(rc);
2110
2111         if (*inode) {
2112                 ll_update_inode(*inode, &md);
2113         } else {
2114                 LASSERT(sb != NULL);
2115
2116                 /*
2117                  * At this point server returns to client's same fid as client
2118                  * generated for creating. So using ->fid1 is okay here.
2119                  */
2120                 LASSERT(fid_is_sane(&md.body->fid1));
2121
2122                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2123                 if (*inode == NULL || IS_ERR(*inode)) {
2124                         if (md.lsm)
2125                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2126 #ifdef CONFIG_FS_POSIX_ACL
2127                         if (md.posix_acl) {
2128                                 posix_acl_release(md.posix_acl);
2129                                 md.posix_acl = NULL;
2130                         }
2131 #endif
2132                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2133                         *inode = NULL;
2134                         CERROR("new_inode -fatal: rc %d\n", rc);
2135                         GOTO(out, rc);
2136                 }
2137         }
2138
2139 out:
2140         md_free_lustre_md(sbi->ll_md_exp, &md);
2141         RETURN(rc);
2142 }
2143
2144 int ll_obd_statfs(struct inode *inode, void *arg)
2145 {
2146         struct ll_sb_info *sbi = NULL;
2147         struct obd_export *exp;
2148         char *buf = NULL;
2149         struct obd_ioctl_data *data = NULL;
2150         __u32 type;
2151         int len = 0, rc;
2152
2153         if (!inode || !(sbi = ll_i2sbi(inode)))
2154                 GOTO(out_statfs, rc = -EINVAL);
2155
2156         rc = obd_ioctl_getdata(&buf, &len, arg);
2157         if (rc)
2158                 GOTO(out_statfs, rc);
2159
2160         data = (void*)buf;
2161         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2162             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2163                 GOTO(out_statfs, rc = -EINVAL);
2164
2165         if (data->ioc_inllen1 != sizeof(__u32) ||
2166             data->ioc_inllen2 != sizeof(__u32) ||
2167             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2168             data->ioc_plen2 != sizeof(struct obd_uuid))
2169                 GOTO(out_statfs, rc = -EINVAL);
2170
2171         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2172         if (type == LL_STATFS_LMV)
2173                 exp = sbi->ll_md_exp;
2174         else if (type == LL_STATFS_LOV)
2175                 exp = sbi->ll_dt_exp;
2176         else
2177                 GOTO(out_statfs, rc = -ENODEV);
2178
2179         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2180         if (rc)
2181                 GOTO(out_statfs, rc);
2182 out_statfs:
2183         if (buf)
2184                 obd_ioctl_freedata(buf, len);
2185         return rc;
2186 }
2187
2188 int ll_process_config(struct lustre_cfg *lcfg)
2189 {
2190         char *ptr;
2191         void *sb;
2192         struct lprocfs_static_vars lvars;
2193         unsigned long x;
2194         int rc = 0;
2195
2196         lprocfs_llite_init_vars(&lvars);
2197
2198         /* The instance name contains the sb: lustre-client-aacfe000 */
2199         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2200         if (!ptr || !*(++ptr))
2201                 return -EINVAL;
2202         if (sscanf(ptr, "%lx", &x) != 1)
2203                 return -EINVAL;
2204         sb = (void *)x;
2205         /* This better be a real Lustre superblock! */
2206         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2207
2208         /* Note we have not called client_common_fill_super yet, so
2209            proc fns must be able to handle that! */
2210         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2211                                       lcfg, sb);
2212         if (rc > 0)
2213                 rc = 0;
2214         return(rc);
2215 }
2216
2217 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2218 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2219                                        struct inode *i1, struct inode *i2,
2220                                        const char *name, int namelen,
2221                                        int mode, __u32 opc, void *data)
2222 {
2223         LASSERT(i1 != NULL);
2224
2225         if (namelen > ll_i2sbi(i1)->ll_namelen)
2226                 return ERR_PTR(-ENAMETOOLONG);
2227
2228         if (op_data == NULL)
2229                 OBD_ALLOC_PTR(op_data);
2230
2231         if (op_data == NULL)
2232                 return ERR_PTR(-ENOMEM);
2233
2234         ll_i2gids(op_data->op_suppgids, i1, i2);
2235         op_data->op_fid1 = *ll_inode2fid(i1);
2236         op_data->op_capa1 = ll_mdscapa_get(i1);
2237
2238         if (i2) {
2239                 op_data->op_fid2 = *ll_inode2fid(i2);
2240                 op_data->op_capa2 = ll_mdscapa_get(i2);
2241         } else {
2242                 fid_zero(&op_data->op_fid2);
2243                 op_data->op_capa2 = NULL;
2244         }
2245
2246         op_data->op_name = name;
2247         op_data->op_namelen = namelen;
2248         op_data->op_mode = mode;
2249         op_data->op_mod_time = cfs_time_current_sec();
2250         op_data->op_fsuid = cfs_curproc_fsuid();
2251         op_data->op_fsgid = cfs_curproc_fsgid();
2252         op_data->op_cap = cfs_curproc_cap_pack();
2253         op_data->op_bias = MDS_CHECK_SPLIT;
2254         op_data->op_opc = opc;
2255         op_data->op_mds = 0;
2256         op_data->op_data = data;
2257
2258         /* If the file is being opened after mknod() (normally due to NFS)
2259          * try to use the default stripe data from parent directory for
2260          * allocating OST objects.  Try to pass the parent FID to MDS. */
2261         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2262             ll_i2info(i2)->lli_smd == NULL) {
2263                 struct ll_inode_info *lli = ll_i2info(i2);
2264
2265                 cfs_spin_lock(&lli->lli_lock);
2266                 if (likely(lli->lli_smd == NULL &&
2267                            !fid_is_zero(&lli->lli_pfid)))
2268                         op_data->op_fid1 = lli->lli_pfid;
2269                 cfs_spin_unlock(&lli->lli_lock);
2270                 /** We ignore parent's capability temporary. */
2271         }
2272
2273         return op_data;
2274 }
2275
2276 void ll_finish_md_op_data(struct md_op_data *op_data)
2277 {
2278         capa_put(op_data->op_capa1);
2279         capa_put(op_data->op_capa2);
2280         OBD_FREE_PTR(op_data);
2281 }
2282
2283 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2284 {
2285         struct ll_sb_info *sbi;
2286
2287         LASSERT((seq != NULL) && (vfs != NULL));
2288         sbi = ll_s2sbi(vfs->mnt_sb);
2289
2290         if (sbi->ll_flags & LL_SBI_NOLCK)
2291                 seq_puts(seq, ",nolock");
2292
2293         if (sbi->ll_flags & LL_SBI_FLOCK)
2294                 seq_puts(seq, ",flock");
2295
2296         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2297                 seq_puts(seq, ",localflock");
2298
2299         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2300                 seq_puts(seq, ",user_xattr");
2301
2302         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2303                 seq_puts(seq, ",lazystatfs");
2304
2305         RETURN(0);
2306 }
2307
2308 /**
2309  * Get obd name by cmd, and copy out to user space
2310  */
2311 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2312 {
2313         struct ll_sb_info *sbi = ll_i2sbi(inode);
2314         struct obd_device *obd;
2315         ENTRY;
2316
2317         if (cmd == OBD_IOC_GETDTNAME)
2318                 obd = class_exp2obd(sbi->ll_dt_exp);
2319         else if (cmd == OBD_IOC_GETMDNAME)
2320                 obd = class_exp2obd(sbi->ll_md_exp);
2321         else
2322                 RETURN(-EINVAL);
2323
2324         if (!obd)
2325                 RETURN(-ENOENT);
2326
2327         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2328                              strlen(obd->obd_name) + 1))
2329                 RETURN(-EFAULT);
2330
2331         RETURN(0);
2332 }