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, 2013, Intel Corporation.
  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 struct kmem_cache *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  97                 lru_page_max = pages / 2;
  98         } else {
  99                 lru_page_max = (pages / 4) * 3;
 100         }
 101
 102         /* initialize lru data */
 103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 104         sbi->ll_cache.ccc_lru_max = lru_page_max;
 105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 108
 109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 110                                            SBI_DEFAULT_READAHEAD_MAX);
 111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 116
 117         ll_generate_random_uuid(uuid);
 118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 120
 121         spin_lock(&ll_sb_lock);
 122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 123         spin_unlock(&ll_sb_lock);
 124
 125         sbi->ll_flags |= LL_SBI_VERBOSE;
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 136                                pp_r_hist.oh_lock);
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145         cfs_atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147
 148         RETURN(sbi);
 149 }
 150
 151 void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 spin_lock(&ll_sb_lock);
 158                 cfs_list_del(&sbi->ll_list);
 159                 spin_unlock(&ll_sb_lock);
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 166                                     struct vfsmount *mnt)
 167 {
 168         struct inode *root = 0;
 169         struct ll_sb_info *sbi = ll_s2sbi(sb);
 170         struct obd_device *obd;
 171         struct obd_capa *oc = NULL;
 172         struct obd_statfs *osfs = NULL;
 173         struct ptlrpc_request *request = NULL;
 174         struct obd_connect_data *data = NULL;
 175         struct obd_uuid *uuid;
 176         struct md_op_data *op_data;
 177         struct lustre_md lmd;
 178         obd_valid valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         OBD_ALLOC_PTR(osfs);
 193         if (osfs == NULL) {
 194                 OBD_FREE_PTR(data);
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         /* indicate the features supported by this client */
 199         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 200                                   OBD_CONNECT_ATTRFID  |
 201                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 202                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 203                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 204                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 205                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 206                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 207                                   OBD_CONNECT_EINPROGRESS |
 208                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 209                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 210                                   OBD_CONNECT_MAX_EASIZE |
 211                                   OBD_CONNECT_FLOCK_DEAD;
 212
 213         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 214                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 215
 216 #ifdef HAVE_LRU_RESIZE_SUPPORT
 217         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 218                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 219 #endif
 220 #ifdef CONFIG_FS_POSIX_ACL
 221         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 222 #endif
 223
 224         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 225                 /* flag mdc connection as lightweight, only used for test
 226                  * purpose, use with care */
 227                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 228
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237 #ifdef HAVE_MS_FLOCK_LOCK
 238         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 239         sb->s_flags |= MS_FLOCK_LOCK;
 240 #endif
 241 #ifdef MS_HAS_NEW_AOPS
 242         sb->s_flags |= MS_HAS_NEW_AOPS;
 243 #endif
 244
 245         if (sbi->ll_flags & LL_SBI_FLOCK)
 246                 sbi->ll_fop = &ll_file_operations_flock;
 247         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 248                 sbi->ll_fop = &ll_file_operations;
 249         else
 250                 sbi->ll_fop = &ll_file_operations_noflock;
 251
 252         /* real client */
 253         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 254         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 255                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 256
 257         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 258
 259         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 260         if (err == -EBUSY) {
 261                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 262                                    "recovery, of which this client is not a "
 263                                    "part. Please wait for recovery to complete,"
 264                                    " abort, or time out.\n", md);
 265                 GOTO(out, err);
 266         } else if (err) {
 267                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 268                 GOTO(out, err);
 269         }
 270
 271         sbi->ll_md_exp->exp_connect_data = *data;
 272
 273         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 274                            LUSTRE_SEQ_METADATA);
 275         if (err) {
 276                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 277                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 278                 GOTO(out_md, err);
 279         }
 280
 281         /* For mount, we only need fs info from MDT0, and also in DNE, it
 282          * can make sure the client can be mounted as long as MDT0 is
 283          * avaible */
 284         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 285                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 286                         OBD_STATFS_FOR_MDT0);
 287         if (err)
 288                 GOTO(out_md_fid, err);
 289
 290         /* This needs to be after statfs to ensure connect has finished.
 291          * Note that "data" does NOT contain the valid connect reply.
 292          * If connecting to a 1.8 server there will be no LMV device, so
 293          * we can access the MDC export directly and exp_connect_flags will
 294          * be non-zero, but if accessing an upgraded 2.1 server it will
 295          * have the correct flags filled in.
 296          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 297         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 298         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 299             valid != CLIENT_CONNECT_MDT_REQD) {
 300                 char *buf;
 301
 302                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 303                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 304                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 305                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 306                                    "feature(s) needed for correct operation "
 307                                    "of this client (%s). Please upgrade "
 308                                    "server or downgrade client.\n",
 309                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 310                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 311                 GOTO(out_md_fid, err = -EPROTO);
 312         }
 313
 314         size = sizeof(*data);
 315         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 316                            KEY_CONN_DATA,  &size, data, NULL);
 317         if (err) {
 318                 CERROR("%s: Get connect data failed: rc = %d\n",
 319                        sbi->ll_md_exp->exp_obd->obd_name, err);
 320                 GOTO(out_md_fid, err);
 321         }
 322
 323         LASSERT(osfs->os_bsize);
 324         sb->s_blocksize = osfs->os_bsize;
 325         sb->s_blocksize_bits = log2(osfs->os_bsize);
 326         sb->s_magic = LL_SUPER_MAGIC;
 327         sb->s_maxbytes = MAX_LFS_FILESIZE;
 328         sbi->ll_namelen = osfs->os_namelen;
 329         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 330
 331         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 332             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 333                 LCONSOLE_INFO("Disabling user_xattr feature because "
 334                               "it is not supported on the server\n");
 335                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags |= MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags |= LL_SBI_ACL;
 343         } else {
 344                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags &= ~MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags &= ~LL_SBI_ACL;
 349         }
 350
 351         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 352                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 353                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client is set as remote by default.\n");
 355                 }
 356         } else {
 357                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 358                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client claims to be remote, but server "
 360                                       "rejected, forced to be local.\n");
 361                 }
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 365                 LCONSOLE_INFO("client enabled MDS capability!\n");
 366                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 370                 LCONSOLE_INFO("client enabled OSS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 375                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 378                 sbi->ll_md_brw_size = data->ocd_brw_size;
 379         else
 380                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 383                 LCONSOLE_INFO("Layout lock feature supported.\n");
 384                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 385         }
 386
 387         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 388                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 389                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 390                                       "unknown maximum xattr size.\n", dt);
 391                 } else {
 392                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 393                         sbi->ll_xattr_cache_enabled = 1;
 394                 }
 395         }
 396
 397         obd = class_name2obd(dt);
 398         if (!obd) {
 399                 CERROR("DT %s: not setup or attached\n", dt);
 400                 GOTO(out_md_fid, err = -ENODEV);
 401         }
 402
 403         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 404                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 405                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 406                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 407                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 408                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 409                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 410                                   OBD_CONNECT_MAXBYTES |
 411                                   OBD_CONNECT_EINPROGRESS |
 412                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 413                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 414
 415         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 416                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 417
 418         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 419                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 420                  * disabled by default, because it can still be enabled on the
 421                  * fly via /proc. As a consequence, we still need to come to an
 422                  * agreement on the supported algorithms at connect time */
 423                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 424
 425                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 426                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 427                 else
 428                         data->ocd_cksum_types = cksum_types_supported_client();
 429         }
 430
 431 #ifdef HAVE_LRU_RESIZE_SUPPORT
 432         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 433 #endif
 434         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 435                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 436
 437         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 438                "ocd_grant: %d\n", data->ocd_connect_flags,
 439                data->ocd_version, data->ocd_grant);
 440
 441         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 442         obd->obd_upcall.onu_upcall = cl_ocd_update;
 443
 444         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 445
 446         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 447                           NULL);
 448         if (err == -EBUSY) {
 449                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 450                                    "recovery, of which this client is not a "
 451                                    "part.  Please wait for recovery to "
 452                                    "complete, abort, or time out.\n", dt);
 453                 GOTO(out_md, err);
 454         } else if (err) {
 455                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 456                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 457                 GOTO(out_md, err);
 458         }
 459
 460         sbi->ll_dt_exp->exp_connect_data = *data;
 461
 462         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 463                            LUSTRE_SEQ_METADATA);
 464         if (err) {
 465                 CERROR("%s: Can't init data layer FID infrastructure, "
 466                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 467                 GOTO(out_dt, err);
 468         }
 469
 470         mutex_lock(&sbi->ll_lco.lco_lock);
 471         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 472         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 473         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 474         mutex_unlock(&sbi->ll_lco.lco_lock);
 475
 476         fid_zero(&sbi->ll_root_fid);
 477         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 478         if (err) {
 479                 CERROR("cannot mds_connect: rc = %d\n", err);
 480                 GOTO(out_lock_cn_cb, err);
 481         }
 482         if (!fid_is_sane(&sbi->ll_root_fid)) {
 483                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 484                        sbi->ll_md_exp->exp_obd->obd_name,
 485                        PFID(&sbi->ll_root_fid));
 486                 GOTO(out_lock_cn_cb, err = -EINVAL);
 487         }
 488         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 489
 490         sb->s_op = &lustre_super_operations;
 491 #if THREAD_SIZE >= 8192 /*b=17630*/
 492         sb->s_export_op = &lustre_export_operations;
 493 #endif
 494
 495         /* make root inode
 496          * XXX: move this to after cbd setup? */
 497         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 498         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 499                 valid |= OBD_MD_FLRMTPERM;
 500         else if (sbi->ll_flags & LL_SBI_ACL)
 501                 valid |= OBD_MD_FLACL;
 502
 503         OBD_ALLOC_PTR(op_data);
 504         if (op_data == NULL)
 505                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 506
 507         op_data->op_fid1 = sbi->ll_root_fid;
 508         op_data->op_mode = 0;
 509         op_data->op_capa1 = oc;
 510         op_data->op_valid = valid;
 511
 512         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 513         if (oc)
 514                 capa_put(oc);
 515         OBD_FREE_PTR(op_data);
 516         if (err) {
 517                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 518                        sbi->ll_md_exp->exp_obd->obd_name, err);
 519                 GOTO(out_lock_cn_cb, err);
 520         }
 521
 522         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 523                                sbi->ll_md_exp, &lmd);
 524         if (err) {
 525                 CERROR("failed to understand root inode md: rc = %d\n", err);
 526                 ptlrpc_req_finished(request);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 531         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 532                                             sbi->ll_flags & LL_SBI_32BIT_API),
 533                        &lmd);
 534         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 535         ptlrpc_req_finished(request);
 536
 537         if (root == NULL || IS_ERR(root)) {
 538                 if (lmd.lsm)
 539                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 540 #ifdef CONFIG_FS_POSIX_ACL
 541                 if (lmd.posix_acl) {
 542                         posix_acl_release(lmd.posix_acl);
 543                         lmd.posix_acl = NULL;
 544                 }
 545 #endif
 546                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 547                 root = NULL;
 548                 CERROR("lustre_lite: bad iget4 for root\n");
 549                 GOTO(out_root, err);
 550         }
 551
 552         err = ll_close_thread_start(&sbi->ll_lcq);
 553         if (err) {
 554                 CERROR("cannot start close thread: rc %d\n", err);
 555                 GOTO(out_root, err);
 556         }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 560                 rct_init(&sbi->ll_rct);
 561                 et_init(&sbi->ll_et);
 562         }
 563 #endif
 564
 565         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 566         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 567                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 568                                  NULL);
 569         cl_sb_init(sb);
 570
 571         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 572                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 573                                  &sbi->ll_cache, NULL);
 574
 575         sb->s_root = d_make_root(root);
 576         if (sb->s_root == NULL) {
 577                 CERROR("%s: can't make root dentry\n",
 578                         ll_get_fsname(sb, NULL, 0));
 579                 GOTO(out_root, err = -ENOMEM);
 580         }
 581 #ifdef HAVE_DCACHE_LOCK
 582         sb->s_root->d_op = &ll_d_ops;
 583 #endif
 584
 585         sbi->ll_sdev_orig = sb->s_dev;
 586
 587         /* We set sb->s_dev equal on all lustre clients in order to support
 588          * NFS export clustering.  NFSD requires that the FSID be the same
 589          * on all clients. */
 590         /* s_dev is also used in lt_compare() to compare two fs, but that is
 591          * only a node-local comparison. */
 592         uuid = obd_get_uuid(sbi->ll_md_exp);
 593         if (uuid != NULL)
 594                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 595
 596         if (data != NULL)
 597                 OBD_FREE_PTR(data);
 598         if (osfs != NULL)
 599                 OBD_FREE_PTR(osfs);
 600         if (proc_lustre_fs_root != NULL) {
 601                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 602                                                   dt, md);
 603                 if (err < 0) {
 604                         CERROR("%s: could not register mount in lprocfs: "
 605                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 606                         err = 0;
 607                 }
 608         }
 609
 610         RETURN(err);
 611 out_root:
 612         if (root)
 613                 iput(root);
 614 out_lock_cn_cb:
 615         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 616 out_dt:
 617         obd_disconnect(sbi->ll_dt_exp);
 618         sbi->ll_dt_exp = NULL;
 619         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 620         obd_zombie_barrier();
 621 out_md_fid:
 622         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 623 out_md:
 624         obd_disconnect(sbi->ll_md_exp);
 625         sbi->ll_md_exp = NULL;
 626 out:
 627         if (data != NULL)
 628                 OBD_FREE_PTR(data);
 629         if (osfs != NULL)
 630                 OBD_FREE_PTR(osfs);
 631         return err;
 632 }
 633
 634 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 635 {
 636         int size, rc;
 637
 638         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 639         size = sizeof(int);
 640         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 641                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 642         if (rc)
 643                 CERROR("Get max mdsize error rc %d \n", rc);
 644
 645         RETURN(rc);
 646 }
 647
 648 void ll_dump_inode(struct inode *inode)
 649 {
 650         struct ll_d_hlist_node *tmp;
 651         int dentry_count = 0;
 652
 653         LASSERT(inode != NULL);
 654
 655         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 656                 dentry_count++;
 657
 658         CERROR("%s: inode %p dump: dev=%s fid="DFID
 659                " mode=%o count=%u, %d dentries\n",
 660                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 661                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 662                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 663 }
 664
 665 void lustre_dump_dentry(struct dentry *dentry, int recur)
 666 {
 667         struct list_head *tmp;
 668         int subdirs = 0;
 669
 670         LASSERT(dentry != NULL);
 671
 672         list_for_each(tmp, &dentry->d_subdirs)
 673                 subdirs++;
 674
 675         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 676                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 677                dentry->d_name.len, dentry->d_name.name,
 678                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 679                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 680                dentry->d_flags, dentry->d_fsdata, subdirs);
 681         if (dentry->d_inode != NULL)
 682                 ll_dump_inode(dentry->d_inode);
 683
 684         if (recur == 0)
 685                 return;
 686
 687         list_for_each(tmp, &dentry->d_subdirs) {
 688                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 689                 lustre_dump_dentry(d, recur - 1);
 690         }
 691 }
 692
 693 void client_common_put_super(struct super_block *sb)
 694 {
 695         struct ll_sb_info *sbi = ll_s2sbi(sb);
 696         ENTRY;
 697
 698 #ifdef CONFIG_FS_POSIX_ACL
 699         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 700                 et_fini(&sbi->ll_et);
 701                 rct_fini(&sbi->ll_rct);
 702         }
 703 #endif
 704
 705         ll_close_thread_shutdown(sbi->ll_lcq);
 706
 707         cl_sb_fini(sb);
 708
 709         cfs_list_del(&sbi->ll_conn_chain);
 710
 711         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 712         obd_disconnect(sbi->ll_dt_exp);
 713         sbi->ll_dt_exp = NULL;
 714         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 715          * see LU-2543. */
 716         obd_zombie_barrier();
 717
 718         lprocfs_unregister_mountpoint(sbi);
 719
 720         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 721         obd_disconnect(sbi->ll_md_exp);
 722         sbi->ll_md_exp = NULL;
 723
 724         EXIT;
 725 }
 726
 727 void ll_kill_super(struct super_block *sb)
 728 {
 729         struct ll_sb_info *sbi;
 730
 731         ENTRY;
 732
 733         /* not init sb ?*/
 734         if (!(sb->s_flags & MS_ACTIVE))
 735                 return;
 736
 737         sbi = ll_s2sbi(sb);
 738         /* we need restore s_dev from changed for clustred NFS before put_super
 739          * because new kernels have cached s_dev and change sb->s_dev in
 740          * put_super not affected real removing devices */
 741         if (sbi) {
 742                 sb->s_dev = sbi->ll_sdev_orig;
 743                 sbi->ll_umounting = 1;
 744         }
 745         EXIT;
 746 }
 747
 748 char *ll_read_opt(const char *opt, char *data)
 749 {
 750         char *value;
 751         char *retval;
 752         ENTRY;
 753
 754         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 755         if (strncmp(opt, data, strlen(opt)))
 756                 RETURN(NULL);
 757         if ((value = strchr(data, '=')) == NULL)
 758                 RETURN(NULL);
 759
 760         value++;
 761         OBD_ALLOC(retval, strlen(value) + 1);
 762         if (!retval) {
 763                 CERROR("out of memory!\n");
 764                 RETURN(NULL);
 765         }
 766
 767         memcpy(retval, value, strlen(value)+1);
 768         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 769         RETURN(retval);
 770 }
 771
 772 static inline int ll_set_opt(const char *opt, char *data, int fl)
 773 {
 774         if (strncmp(opt, data, strlen(opt)) != 0)
 775                 return(0);
 776         else
 777                 return(fl);
 778 }
 779
 780 /* non-client-specific mount options are parsed in lmd_parse */
 781 static int ll_options(char *options, int *flags)
 782 {
 783         int tmp;
 784         char *s1 = options, *s2;
 785         ENTRY;
 786
 787         if (!options)
 788                 RETURN(0);
 789
 790         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 791
 792         while (*s1) {
 793                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 794                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 795                 if (tmp) {
 796                         *flags |= tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 805                 if (tmp) {
 806                         *flags |= tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 810                 if (tmp) {
 811                         *flags &= ~tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 820                 if (tmp) {
 821                         *flags &= ~tmp;
 822                         goto next;
 823                 }
 824 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 51, 0)
 825                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 826                 if (tmp) {
 827                         /* Ignore deprecated mount option.  The client will
 828                          * always try to mount with ACL support, whether this
 829                          * is used depends on whether server supports it. */
 830                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 831                                                   "mount option 'acl'.\n");
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 835                 if (tmp) {
 836                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 837                                                   "mount option 'noacl'.\n");
 838                         goto next;
 839                 }
 840 #else
 841 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 842 #endif
 843                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 844                 if (tmp) {
 845                         *flags |= tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858
 859                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 860                 if (tmp) {
 861                         *flags |= tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 865                 if (tmp) {
 866                         *flags &= ~tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 875                 if (tmp) {
 876                         *flags &= ~tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 895                 if (tmp) {
 896                         *flags |= tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 900                 if (tmp) {
 901                         *flags |= tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 905                 if (tmp) {
 906                         *flags &= ~tmp;
 907                         goto next;
 908                 }
 909                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 910                                    s1);
 911                 RETURN(-EINVAL);
 912
 913 next:
 914                 /* Find next opt */
 915                 s2 = strchr(s1, ',');
 916                 if (s2 == NULL)
 917                         break;
 918                 s1 = s2 + 1;
 919         }
 920         RETURN(0);
 921 }
 922
 923 void ll_lli_init(struct ll_inode_info *lli)
 924 {
 925         lli->lli_inode_magic = LLI_INODE_MAGIC;
 926         lli->lli_flags = 0;
 927         lli->lli_ioepoch = 0;
 928         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 929         spin_lock_init(&lli->lli_lock);
 930         lli->lli_posix_acl = NULL;
 931         lli->lli_remote_perms = NULL;
 932         mutex_init(&lli->lli_rmtperm_mutex);
 933         /* Do not set lli_fid, it has been initialized already. */
 934         fid_zero(&lli->lli_pfid);
 935         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 936         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 937         cfs_atomic_set(&lli->lli_open_count, 0);
 938         lli->lli_mds_capa = NULL;
 939         lli->lli_rmtperm_time = 0;
 940         lli->lli_pending_och = NULL;
 941         lli->lli_mds_read_och = NULL;
 942         lli->lli_mds_write_och = NULL;
 943         lli->lli_mds_exec_och = NULL;
 944         lli->lli_open_fd_read_count = 0;
 945         lli->lli_open_fd_write_count = 0;
 946         lli->lli_open_fd_exec_count = 0;
 947         mutex_init(&lli->lli_och_mutex);
 948         spin_lock_init(&lli->lli_agl_lock);
 949         lli->lli_has_smd = false;
 950         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 951         lli->lli_clob = NULL;
 952
 953         init_rwsem(&lli->lli_xattrs_list_rwsem);
 954         mutex_init(&lli->lli_xattrs_enq_lock);
 955
 956         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 957         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 958                 mutex_init(&lli->lli_readdir_mutex);
 959                 lli->lli_opendir_key = NULL;
 960                 lli->lli_sai = NULL;
 961                 lli->lli_def_acl = NULL;
 962                 spin_lock_init(&lli->lli_sa_lock);
 963                 lli->lli_opendir_pid = 0;
 964         } else {
 965                 sema_init(&lli->lli_size_sem, 1);
 966                 lli->lli_size_sem_owner = NULL;
 967                 lli->lli_symlink_name = NULL;
 968                 init_rwsem(&lli->lli_trunc_sem);
 969                 mutex_init(&lli->lli_write_mutex);
 970                 init_rwsem(&lli->lli_glimpse_sem);
 971                 lli->lli_glimpse_time = 0;
 972                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 973                 lli->lli_agl_index = 0;
 974                 lli->lli_async_rc = 0;
 975                 lli->lli_volatile = false;
 976         }
 977         mutex_init(&lli->lli_layout_mutex);
 978 }
 979
 980 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 981 {
 982         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 983
 984         bdi->name = "lustre";
 985         return bdi_register(bdi, NULL, "lustre-%d",
 986                             atomic_inc_return(&ll_bdi_num));
 987 }
 988
 989 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 990 {
 991         struct lustre_profile *lprof = NULL;
 992         struct lustre_sb_info *lsi = s2lsi(sb);
 993         struct ll_sb_info *sbi;
 994         char  *dt = NULL, *md = NULL;
 995         char  *profilenm = get_profile_name(sb);
 996         struct config_llog_instance *cfg;
 997         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 998         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 999         int    err;
1000         ENTRY;
1001
1002         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1003
1004         OBD_ALLOC_PTR(cfg);
1005         if (cfg == NULL)
1006                 RETURN(-ENOMEM);
1007
1008         try_module_get(THIS_MODULE);
1009
1010         /* client additional sb info */
1011         lsi->lsi_llsbi = sbi = ll_init_sbi();
1012         if (!sbi) {
1013                 module_put(THIS_MODULE);
1014                 OBD_FREE_PTR(cfg);
1015                 RETURN(-ENOMEM);
1016         }
1017
1018         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1019         if (err)
1020                 GOTO(out_free, err);
1021
1022         err = bdi_init(&lsi->lsi_bdi);
1023         if (err)
1024                 GOTO(out_free, err);
1025         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1026         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1027         err = ll_bdi_register(&lsi->lsi_bdi);
1028         if (err)
1029                 GOTO(out_free, err);
1030
1031         sb->s_bdi = &lsi->lsi_bdi;
1032 #ifndef HAVE_DCACHE_LOCK
1033         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1034         sb->s_d_op = &ll_d_ops;
1035 #endif
1036
1037         /* Generate a string unique to this super, in case some joker tries
1038            to mount the same fs at two mount points.
1039            Use the address of the super itself.*/
1040         cfg->cfg_instance = sb;
1041         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1042         cfg->cfg_callback = class_config_llog_handler;
1043         /* set up client obds */
1044         err = lustre_process_log(sb, profilenm, cfg);
1045         if (err < 0) {
1046                 CERROR("Unable to process log: %d\n", err);
1047                 GOTO(out_free, err);
1048         }
1049
1050         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1051         lprof = class_get_profile(profilenm);
1052         if (lprof == NULL) {
1053                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1054                                    " read from the MGS.  Does that filesystem "
1055                                    "exist?\n", profilenm);
1056                 GOTO(out_free, err = -EINVAL);
1057         }
1058         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1059                lprof->lp_md, lprof->lp_dt);
1060
1061         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1062         if (!dt)
1063                 GOTO(out_free, err = -ENOMEM);
1064         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1065
1066         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1067         if (!md)
1068                 GOTO(out_free, err = -ENOMEM);
1069         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1070
1071         /* connections, registrations, sb setup */
1072         err = client_common_fill_super(sb, md, dt, mnt);
1073
1074 out_free:
1075         if (md)
1076                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1077         if (dt)
1078                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1079         if (err)
1080                 ll_put_super(sb);
1081         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1082                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1083
1084         OBD_FREE_PTR(cfg);
1085         RETURN(err);
1086 } /* ll_fill_super */
1087
1088
1089 void lu_context_keys_dump(void);
1090
1091 void ll_put_super(struct super_block *sb)
1092 {
1093         struct config_llog_instance cfg, params_cfg;
1094         struct obd_device *obd;
1095         struct lustre_sb_info *lsi = s2lsi(sb);
1096         struct ll_sb_info *sbi = ll_s2sbi(sb);
1097         char *profilenm = get_profile_name(sb);
1098         int force = 1, next;
1099         ENTRY;
1100
1101         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1102
1103         ll_print_capa_stat(sbi);
1104
1105         cfg.cfg_instance = sb;
1106         lustre_end_log(sb, profilenm, &cfg);
1107
1108         params_cfg.cfg_instance = sb;
1109         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1110
1111         if (sbi->ll_md_exp) {
1112                 obd = class_exp2obd(sbi->ll_md_exp);
1113                 if (obd)
1114                         force = obd->obd_force;
1115         }
1116
1117         /* We need to set force before the lov_disconnect in
1118            lustre_common_put_super, since l_d cleans up osc's as well. */
1119         if (force) {
1120                 next = 0;
1121                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1122                                                      &next)) != NULL) {
1123                         obd->obd_force = force;
1124                 }
1125         }
1126
1127         if (sbi->ll_lcq) {
1128                 /* Only if client_common_fill_super succeeded */
1129                 client_common_put_super(sb);
1130         }
1131
1132         next = 0;
1133         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1134                 class_manual_cleanup(obd);
1135         }
1136
1137         if (sbi->ll_flags & LL_SBI_VERBOSE)
1138                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1139
1140         if (profilenm)
1141                 class_del_profile(profilenm);
1142
1143         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1144                 bdi_destroy(&lsi->lsi_bdi);
1145                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1146         }
1147
1148         ll_free_sbi(sb);
1149         lsi->lsi_llsbi = NULL;
1150
1151         lustre_common_put_super(sb);
1152
1153         module_put(THIS_MODULE);
1154
1155         EXIT;
1156 } /* client_put_super */
1157
1158 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1159 {
1160         struct inode *inode = NULL;
1161
1162         /* NOTE: we depend on atomic igrab() -bzzz */
1163         lock_res_and_lock(lock);
1164         if (lock->l_resource->lr_lvb_inode) {
1165                 struct ll_inode_info * lli;
1166                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1167                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1168                         inode = igrab(lock->l_resource->lr_lvb_inode);
1169                 } else {
1170                         inode = lock->l_resource->lr_lvb_inode;
1171                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1172                                          D_WARNING, lock, "lr_lvb_inode %p is "
1173                                          "bogus: magic %08x",
1174                                          lock->l_resource->lr_lvb_inode,
1175                                          lli->lli_inode_magic);
1176                         inode = NULL;
1177                 }
1178         }
1179         unlock_res_and_lock(lock);
1180         return inode;
1181 }
1182
1183 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1184 {
1185         struct inode *inode = NULL;
1186         /* NOTE: we depend on atomic igrab() -bzzz */
1187         lock_res_and_lock(lock);
1188         if (lock->l_ast_data) {
1189                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1190                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1191                         inode = igrab(lock->l_ast_data);
1192                 } else {
1193                         inode = lock->l_ast_data;
1194                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1195                                          D_WARNING, lock, "l_ast_data %p is "
1196                                          "bogus: magic %08x", lock->l_ast_data,
1197                                          lli->lli_inode_magic);
1198                         inode = NULL;
1199                 }
1200         }
1201         unlock_res_and_lock(lock);
1202         return inode;
1203 }
1204
1205 void ll_clear_inode(struct inode *inode)
1206 {
1207         struct ll_inode_info *lli = ll_i2info(inode);
1208         struct ll_sb_info *sbi = ll_i2sbi(inode);
1209         ENTRY;
1210
1211         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1212                PFID(ll_inode2fid(inode)), inode);
1213
1214         if (S_ISDIR(inode->i_mode)) {
1215                 /* these should have been cleared in ll_file_release */
1216                 LASSERT(lli->lli_opendir_key == NULL);
1217                 LASSERT(lli->lli_sai == NULL);
1218                 LASSERT(lli->lli_opendir_pid == 0);
1219         }
1220
1221         spin_lock(&lli->lli_lock);
1222         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1223         spin_unlock(&lli->lli_lock);
1224         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1225
1226         LASSERT(!lli->lli_open_fd_write_count);
1227         LASSERT(!lli->lli_open_fd_read_count);
1228         LASSERT(!lli->lli_open_fd_exec_count);
1229
1230         if (lli->lli_mds_write_och)
1231                 ll_md_real_close(inode, FMODE_WRITE);
1232         if (lli->lli_mds_exec_och)
1233                 ll_md_real_close(inode, FMODE_EXEC);
1234         if (lli->lli_mds_read_och)
1235                 ll_md_real_close(inode, FMODE_READ);
1236
1237         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1238                 OBD_FREE(lli->lli_symlink_name,
1239                          strlen(lli->lli_symlink_name) + 1);
1240                 lli->lli_symlink_name = NULL;
1241         }
1242
1243         ll_xattr_cache_destroy(inode);
1244
1245         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1246                 LASSERT(lli->lli_posix_acl == NULL);
1247                 if (lli->lli_remote_perms) {
1248                         free_rmtperm_hash(lli->lli_remote_perms);
1249                         lli->lli_remote_perms = NULL;
1250                 }
1251         }
1252 #ifdef CONFIG_FS_POSIX_ACL
1253         else if (lli->lli_posix_acl) {
1254                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1255                 LASSERT(lli->lli_remote_perms == NULL);
1256                 posix_acl_release(lli->lli_posix_acl);
1257                 lli->lli_posix_acl = NULL;
1258         }
1259 #endif
1260         lli->lli_inode_magic = LLI_INODE_DEAD;
1261
1262         ll_clear_inode_capas(inode);
1263         if (!S_ISDIR(inode->i_mode))
1264                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1265
1266         /*
1267          * XXX This has to be done before lsm is freed below, because
1268          * cl_object still uses inode lsm.
1269          */
1270         cl_inode_fini(inode);
1271         lli->lli_has_smd = false;
1272
1273         EXIT;
1274 }
1275
1276 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1277                   struct md_open_data **mod)
1278 {
1279         struct lustre_md md;
1280         struct inode *inode = dentry->d_inode;
1281         struct ll_sb_info *sbi = ll_i2sbi(inode);
1282         struct ptlrpc_request *request = NULL;
1283         int rc, ia_valid;
1284         ENTRY;
1285
1286         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1287                                      LUSTRE_OPC_ANY, NULL);
1288         if (IS_ERR(op_data))
1289                 RETURN(PTR_ERR(op_data));
1290
1291         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1292                         &request, mod);
1293         if (rc) {
1294                 ptlrpc_req_finished(request);
1295                 if (rc == -ENOENT) {
1296                         clear_nlink(inode);
1297                         /* Unlinked special device node? Or just a race?
1298                          * Pretend we done everything. */
1299                         if (!S_ISREG(inode->i_mode) &&
1300                             !S_ISDIR(inode->i_mode)) {
1301                                 ia_valid = op_data->op_attr.ia_valid;
1302                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1303                                 rc = simple_setattr(dentry, &op_data->op_attr);
1304                                 op_data->op_attr.ia_valid = ia_valid;
1305                         }
1306                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1307                         CERROR("md_setattr fails: rc = %d\n", rc);
1308                 }
1309                 RETURN(rc);
1310         }
1311
1312         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1313                               sbi->ll_md_exp, &md);
1314         if (rc) {
1315                 ptlrpc_req_finished(request);
1316                 RETURN(rc);
1317         }
1318
1319         ia_valid = op_data->op_attr.ia_valid;
1320         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1321          * cache is not cleared yet. */
1322         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1323         rc = simple_setattr(dentry, &op_data->op_attr);
1324         op_data->op_attr.ia_valid = ia_valid;
1325
1326         /* Extract epoch data if obtained. */
1327         op_data->op_handle = md.body->handle;
1328         op_data->op_ioepoch = md.body->ioepoch;
1329
1330         ll_update_inode(inode, &md);
1331         ptlrpc_req_finished(request);
1332
1333         RETURN(rc);
1334 }
1335
1336 /* Close IO epoch and send Size-on-MDS attribute update. */
1337 static int ll_setattr_done_writing(struct inode *inode,
1338                                    struct md_op_data *op_data,
1339                                    struct md_open_data *mod)
1340 {
1341         struct ll_inode_info *lli = ll_i2info(inode);
1342         int rc = 0;
1343         ENTRY;
1344
1345         LASSERT(op_data != NULL);
1346         if (!S_ISREG(inode->i_mode))
1347                 RETURN(0);
1348
1349         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1350                op_data->op_ioepoch, PFID(&lli->lli_fid));
1351
1352         op_data->op_flags = MF_EPOCH_CLOSE;
1353         ll_done_writing_attr(inode, op_data);
1354         ll_pack_inode2opdata(inode, op_data, NULL);
1355
1356         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1357         if (rc == -EAGAIN) {
1358                 /* MDS has instructed us to obtain Size-on-MDS attribute
1359                  * from OSTs and send setattr to back to MDS. */
1360                 rc = ll_som_update(inode, op_data);
1361         } else if (rc) {
1362                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1363                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1364                        PFID(ll_inode2fid(inode)), rc);
1365         }
1366         RETURN(rc);
1367 }
1368
1369 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1370 {
1371         struct obd_capa *capa;
1372         int rc;
1373
1374         if (attr->ia_valid & ATTR_SIZE)
1375                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1376         else
1377                 capa = ll_mdscapa_get(inode);
1378
1379         rc = cl_setattr_ost(inode, attr, capa);
1380
1381         if (attr->ia_valid & ATTR_SIZE)
1382                 ll_truncate_free_capa(capa);
1383         else
1384                 capa_put(capa);
1385
1386         return rc;
1387 }
1388
1389 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1390  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1391  * keep these values until such a time that objects are allocated for it.
1392  * We do the MDS operations first, as it is checking permissions for us.
1393  * We don't to the MDS RPC if there is nothing that we want to store there,
1394  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1395  * going to do an RPC anyways.
1396  *
1397  * If we are doing a truncate, we will send the mtime and ctime updates
1398  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1399  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1400  * at the same time.
1401  *
1402  * In case of HSMimport, we only set attr on MDS.
1403  */
1404 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1405 {
1406         struct inode *inode = dentry->d_inode;
1407         struct ll_inode_info *lli = ll_i2info(inode);
1408         struct md_op_data *op_data = NULL;
1409         struct md_open_data *mod = NULL;
1410         bool file_is_released = false;
1411         int rc = 0, rc1 = 0;
1412         ENTRY;
1413
1414         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1415                "valid %x, hsm_import %d\n",
1416                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1417                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1418                hsm_import);
1419
1420         if (attr->ia_valid & ATTR_SIZE) {
1421                 /* Check new size against VFS/VM file size limit and rlimit */
1422                 rc = inode_newsize_ok(inode, attr->ia_size);
1423                 if (rc)
1424                         RETURN(rc);
1425
1426                 /* The maximum Lustre file size is variable, based on the
1427                  * OST maximum object size and number of stripes.  This
1428                  * needs another check in addition to the VFS check above. */
1429                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1430                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1431                                PFID(&lli->lli_fid), attr->ia_size,
1432                                ll_file_maxbytes(inode));
1433                         RETURN(-EFBIG);
1434                 }
1435
1436                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1437         }
1438
1439         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1440         if (attr->ia_valid & TIMES_SET_FLAGS) {
1441                 if (current_fsuid() != inode->i_uid &&
1442                     !cfs_capable(CFS_CAP_FOWNER))
1443                         RETURN(-EPERM);
1444         }
1445
1446         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1447         if (attr->ia_valid & ATTR_CTIME) {
1448                 attr->ia_ctime = CFS_CURRENT_TIME;
1449                 attr->ia_valid |= ATTR_CTIME_SET;
1450         }
1451         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1452             (attr->ia_valid & ATTR_ATIME)) {
1453                 attr->ia_atime = CFS_CURRENT_TIME;
1454                 attr->ia_valid |= ATTR_ATIME_SET;
1455         }
1456         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1457             (attr->ia_valid & ATTR_MTIME)) {
1458                 attr->ia_mtime = CFS_CURRENT_TIME;
1459                 attr->ia_valid |= ATTR_MTIME_SET;
1460         }
1461
1462         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1463                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1464                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1465                        cfs_time_current_sec());
1466
1467         /* We always do an MDS RPC, even if we're only changing the size;
1468          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1469
1470         OBD_ALLOC_PTR(op_data);
1471         if (op_data == NULL)
1472                 RETURN(-ENOMEM);
1473
1474         if (!S_ISDIR(inode->i_mode)) {
1475                 if (attr->ia_valid & ATTR_SIZE)
1476                         inode_dio_write_done(inode);
1477                 mutex_unlock(&inode->i_mutex);
1478                 down_write(&lli->lli_trunc_sem);
1479         }
1480
1481         /* truncate on a released file must failed with -ENODATA,
1482          * so size must not be set on MDS for released file
1483          * but other attributes must be set
1484          */
1485         if (S_ISREG(inode->i_mode)) {
1486                 struct lov_stripe_md *lsm;
1487                 __u32 gen;
1488
1489                 ll_layout_refresh(inode, &gen);
1490                 lsm = ccc_inode_lsm_get(inode);
1491                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1492                         file_is_released = true;
1493                 ccc_inode_lsm_put(inode, lsm);
1494
1495                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1496                         if (file_is_released) {
1497                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1498                                 if (rc < 0)
1499                                         GOTO(out, rc);
1500
1501                                 file_is_released = false;
1502                                 ll_layout_refresh(inode, &gen);
1503                         }
1504
1505                         /* If we are changing file size, file content is
1506                          * modified, flag it. */
1507                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1508                         spin_lock(&lli->lli_lock);
1509                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1510                         spin_unlock(&lli->lli_lock);
1511                         op_data->op_bias |= MDS_DATA_MODIFIED;
1512                 }
1513         }
1514
1515         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1516
1517         /* Open epoch for truncate. */
1518         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1519             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1520                 op_data->op_flags = MF_EPOCH_OPEN;
1521
1522         rc = ll_md_setattr(dentry, op_data, &mod);
1523         if (rc)
1524                 GOTO(out, rc);
1525
1526         /* RPC to MDT is sent, cancel data modification flag */
1527         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1528                 spin_lock(&lli->lli_lock);
1529                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1530                 spin_unlock(&lli->lli_lock);
1531         }
1532
1533         ll_ioepoch_open(lli, op_data->op_ioepoch);
1534         if (!S_ISREG(inode->i_mode) || file_is_released)
1535                 GOTO(out, rc = 0);
1536
1537         if (attr->ia_valid & (ATTR_SIZE |
1538                               ATTR_ATIME | ATTR_ATIME_SET |
1539                               ATTR_MTIME | ATTR_MTIME_SET)) {
1540                 /* For truncate and utimes sending attributes to OSTs, setting
1541                  * mtime/atime to the past will be performed under PW [0:EOF]
1542                  * extent lock (new_size:EOF for truncate).  It may seem
1543                  * excessive to send mtime/atime updates to OSTs when not
1544                  * setting times to past, but it is necessary due to possible
1545                  * time de-synchronization between MDT inode and OST objects */
1546                 rc = ll_setattr_ost(inode, attr);
1547         }
1548         EXIT;
1549 out:
1550         if (op_data) {
1551                 if (op_data->op_ioepoch) {
1552                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1553                         if (!rc)
1554                                 rc = rc1;
1555                 }
1556                 ll_finish_md_op_data(op_data);
1557         }
1558         if (!S_ISDIR(inode->i_mode)) {
1559                 up_write(&lli->lli_trunc_sem);
1560                 mutex_lock(&inode->i_mutex);
1561                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1562                         inode_dio_wait(inode);
1563         }
1564
1565         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1566                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1567
1568         return rc;
1569 }
1570
1571 int ll_setattr(struct dentry *de, struct iattr *attr)
1572 {
1573         int mode = de->d_inode->i_mode;
1574
1575         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1576                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1577                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1578
1579         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1580                                (ATTR_SIZE|ATTR_MODE)) &&
1581             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1582              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1583               !(attr->ia_mode & S_ISGID))))
1584                 attr->ia_valid |= ATTR_FORCE;
1585
1586         if ((mode & S_ISUID) &&
1587             !(attr->ia_mode & S_ISUID) &&
1588             !(attr->ia_valid & ATTR_KILL_SUID))
1589                 attr->ia_valid |= ATTR_KILL_SUID;
1590
1591         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1592             !(attr->ia_mode & S_ISGID) &&
1593             !(attr->ia_valid & ATTR_KILL_SGID))
1594                 attr->ia_valid |= ATTR_KILL_SGID;
1595
1596         return ll_setattr_raw(de, attr, false);
1597 }
1598
1599 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1600                        __u64 max_age, __u32 flags)
1601 {
1602         struct ll_sb_info *sbi = ll_s2sbi(sb);
1603         struct obd_statfs obd_osfs;
1604         int rc;
1605         ENTRY;
1606
1607         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1608         if (rc) {
1609                 CERROR("md_statfs fails: rc = %d\n", rc);
1610                 RETURN(rc);
1611         }
1612
1613         osfs->os_type = sb->s_magic;
1614
1615         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1616                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1617
1618         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1619                 flags |= OBD_STATFS_NODELAY;
1620
1621         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1622         if (rc) {
1623                 CERROR("obd_statfs fails: rc = %d\n", rc);
1624                 RETURN(rc);
1625         }
1626
1627         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1628                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1629                obd_osfs.os_files);
1630
1631         osfs->os_bsize = obd_osfs.os_bsize;
1632         osfs->os_blocks = obd_osfs.os_blocks;
1633         osfs->os_bfree = obd_osfs.os_bfree;
1634         osfs->os_bavail = obd_osfs.os_bavail;
1635
1636         /* If we don't have as many objects free on the OST as inodes
1637          * on the MDS, we reduce the total number of inodes to
1638          * compensate, so that the "inodes in use" number is correct.
1639          */
1640         if (obd_osfs.os_ffree < osfs->os_ffree) {
1641                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1642                         obd_osfs.os_ffree;
1643                 osfs->os_ffree = obd_osfs.os_ffree;
1644         }
1645
1646         RETURN(rc);
1647 }
1648 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1649 {
1650         struct super_block *sb = de->d_sb;
1651         struct obd_statfs osfs;
1652         __u64 fsid = huge_encode_dev(sb->s_dev);
1653         int rc;
1654
1655         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1656         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1657
1658         /* Some amount of caching on the client is allowed */
1659         rc = ll_statfs_internal(sb, &osfs,
1660                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1661                                 0);
1662         if (rc)
1663                 return rc;
1664
1665         statfs_unpack(sfs, &osfs);
1666
1667         /* We need to downshift for all 32-bit kernels, because we can't
1668          * tell if the kernel is being called via sys_statfs64() or not.
1669          * Stop before overflowing f_bsize - in which case it is better
1670          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1671         if (sizeof(long) < 8) {
1672                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1673                         sfs->f_bsize <<= 1;
1674
1675                         osfs.os_blocks >>= 1;
1676                         osfs.os_bfree >>= 1;
1677                         osfs.os_bavail >>= 1;
1678                 }
1679         }
1680
1681         sfs->f_blocks = osfs.os_blocks;
1682         sfs->f_bfree = osfs.os_bfree;
1683         sfs->f_bavail = osfs.os_bavail;
1684         sfs->f_fsid.val[0] = (__u32)fsid;
1685         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1686         return 0;
1687 }
1688
1689 void ll_inode_size_lock(struct inode *inode)
1690 {
1691         struct ll_inode_info *lli;
1692
1693         LASSERT(!S_ISDIR(inode->i_mode));
1694
1695         lli = ll_i2info(inode);
1696         LASSERT(lli->lli_size_sem_owner != current);
1697         down(&lli->lli_size_sem);
1698         LASSERT(lli->lli_size_sem_owner == NULL);
1699         lli->lli_size_sem_owner = current;
1700 }
1701
1702 void ll_inode_size_unlock(struct inode *inode)
1703 {
1704         struct ll_inode_info *lli;
1705
1706         lli = ll_i2info(inode);
1707         LASSERT(lli->lli_size_sem_owner == current);
1708         lli->lli_size_sem_owner = NULL;
1709         up(&lli->lli_size_sem);
1710 }
1711
1712 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1713 {
1714         struct ll_inode_info *lli = ll_i2info(inode);
1715         struct mdt_body *body = md->body;
1716         struct lov_stripe_md *lsm = md->lsm;
1717         struct ll_sb_info *sbi = ll_i2sbi(inode);
1718
1719         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1720         if (lsm != NULL) {
1721                 if (!lli->lli_has_smd &&
1722                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1723                         cl_file_inode_init(inode, md);
1724
1725                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1726                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1727                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1728         }
1729
1730         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1731                 if (body->valid & OBD_MD_FLRMTPERM)
1732                         ll_update_remote_perm(inode, md->remote_perm);
1733         }
1734 #ifdef CONFIG_FS_POSIX_ACL
1735         else if (body->valid & OBD_MD_FLACL) {
1736                 spin_lock(&lli->lli_lock);
1737                 if (lli->lli_posix_acl)
1738                         posix_acl_release(lli->lli_posix_acl);
1739                 lli->lli_posix_acl = md->posix_acl;
1740                 spin_unlock(&lli->lli_lock);
1741         }
1742 #endif
1743         inode->i_ino = cl_fid_build_ino(&body->fid1,
1744                                         sbi->ll_flags & LL_SBI_32BIT_API);
1745         inode->i_generation = cl_fid_build_gen(&body->fid1);
1746
1747         if (body->valid & OBD_MD_FLATIME) {
1748                 if (body->atime > LTIME_S(inode->i_atime))
1749                         LTIME_S(inode->i_atime) = body->atime;
1750                 lli->lli_lvb.lvb_atime = body->atime;
1751         }
1752         if (body->valid & OBD_MD_FLMTIME) {
1753                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1754                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1755                                "to "LPU64"\n", inode->i_ino,
1756                                LTIME_S(inode->i_mtime), body->mtime);
1757                         LTIME_S(inode->i_mtime) = body->mtime;
1758                 }
1759                 lli->lli_lvb.lvb_mtime = body->mtime;
1760         }
1761         if (body->valid & OBD_MD_FLCTIME) {
1762                 if (body->ctime > LTIME_S(inode->i_ctime))
1763                         LTIME_S(inode->i_ctime) = body->ctime;
1764                 lli->lli_lvb.lvb_ctime = body->ctime;
1765         }
1766         if (body->valid & OBD_MD_FLMODE)
1767                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1768         if (body->valid & OBD_MD_FLTYPE)
1769                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1770         LASSERT(inode->i_mode != 0);
1771         if (S_ISREG(inode->i_mode)) {
1772                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1773         } else {
1774                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1775         }
1776         if (body->valid & OBD_MD_FLUID)
1777                 inode->i_uid = body->uid;
1778         if (body->valid & OBD_MD_FLGID)
1779                 inode->i_gid = body->gid;
1780         if (body->valid & OBD_MD_FLFLAGS)
1781                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1782         if (body->valid & OBD_MD_FLNLINK)
1783                 set_nlink(inode, body->nlink);
1784         if (body->valid & OBD_MD_FLRDEV)
1785                 inode->i_rdev = old_decode_dev(body->rdev);
1786
1787         if (body->valid & OBD_MD_FLID) {
1788                 /* FID shouldn't be changed! */
1789                 if (fid_is_sane(&lli->lli_fid)) {
1790                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1791                                  "Trying to change FID "DFID
1792                                  " to the "DFID", inode "DFID"(%p)\n",
1793                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1794                                  PFID(ll_inode2fid(inode)), inode);
1795                 } else
1796                         lli->lli_fid = body->fid1;
1797         }
1798
1799         LASSERT(fid_seq(&lli->lli_fid) != 0);
1800
1801         if (body->valid & OBD_MD_FLSIZE) {
1802                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1803                     S_ISREG(inode->i_mode)) {
1804                         struct lustre_handle lockh;
1805                         ldlm_mode_t mode;
1806
1807                         /* As it is possible a blocking ast has been processed
1808                          * by this time, we need to check there is an UPDATE
1809                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1810                          * it. */
1811                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1812                                                &lockh, LDLM_FL_CBPENDING,
1813                                                LCK_CR | LCK_CW |
1814                                                LCK_PR | LCK_PW);
1815                         if (mode) {
1816                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1817                                                       LLIF_EPOCH_PENDING |
1818                                                       LLIF_SOM_DIRTY)) {
1819                                         CERROR("%s: inode "DFID" flags %u still"
1820                                                " has size authority! do not "
1821                                                "trust the size from MDS\n",
1822                                                sbi->ll_md_exp->exp_obd->obd_name,
1823                                                PFID(ll_inode2fid(inode)),
1824                                                lli->lli_flags);
1825                                 } else {
1826                                         /* Use old size assignment to avoid
1827                                          * deadlock bz14138 & bz14326 */
1828                                         i_size_write(inode, body->size);
1829                                         spin_lock(&lli->lli_lock);
1830                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1831                                         spin_unlock(&lli->lli_lock);
1832                                 }
1833                                 ldlm_lock_decref(&lockh, mode);
1834                         }
1835                 } else {
1836                         /* Use old size assignment to avoid
1837                          * deadlock bz14138 & bz14326 */
1838                         i_size_write(inode, body->size);
1839
1840                         CDEBUG(D_VFSTRACE,
1841                                "inode="DFID", updating i_size %llu\n",
1842                                PFID(ll_inode2fid(inode)),
1843                                (unsigned long long)body->size);
1844                 }
1845
1846                 if (body->valid & OBD_MD_FLBLOCKS)
1847                         inode->i_blocks = body->blocks;
1848         }
1849
1850         if (body->valid & OBD_MD_FLMDSCAPA) {
1851                 LASSERT(md->mds_capa);
1852                 ll_add_capa(inode, md->mds_capa);
1853         }
1854         if (body->valid & OBD_MD_FLOSSCAPA) {
1855                 LASSERT(md->oss_capa);
1856                 ll_add_capa(inode, md->oss_capa);
1857         }
1858
1859         if (body->valid & OBD_MD_TSTATE) {
1860                 if (body->t_state & MS_RESTORE)
1861                         lli->lli_flags |= LLIF_FILE_RESTORING;
1862         }
1863 }
1864
1865 void ll_read_inode2(struct inode *inode, void *opaque)
1866 {
1867         struct lustre_md *md = opaque;
1868         struct ll_inode_info *lli = ll_i2info(inode);
1869         ENTRY;
1870
1871         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1872                PFID(&lli->lli_fid), inode);
1873
1874         LASSERT(!lli->lli_has_smd);
1875
1876         /* Core attributes from the MDS first.  This is a new inode, and
1877          * the VFS doesn't zero times in the core inode so we have to do
1878          * it ourselves.  They will be overwritten by either MDS or OST
1879          * attributes - we just need to make sure they aren't newer. */
1880         LTIME_S(inode->i_mtime) = 0;
1881         LTIME_S(inode->i_atime) = 0;
1882         LTIME_S(inode->i_ctime) = 0;
1883         inode->i_rdev = 0;
1884         ll_update_inode(inode, md);
1885
1886         /* OIDEBUG(inode); */
1887
1888         /* initializing backing dev info. */
1889         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1890
1891
1892         if (S_ISREG(inode->i_mode)) {
1893                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1894                 inode->i_op = &ll_file_inode_operations;
1895                 inode->i_fop = sbi->ll_fop;
1896                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1897                 EXIT;
1898         } else if (S_ISDIR(inode->i_mode)) {
1899                 inode->i_op = &ll_dir_inode_operations;
1900                 inode->i_fop = &ll_dir_operations;
1901                 EXIT;
1902         } else if (S_ISLNK(inode->i_mode)) {
1903                 inode->i_op = &ll_fast_symlink_inode_operations;
1904                 EXIT;
1905         } else {
1906                 inode->i_op = &ll_special_inode_operations;
1907
1908                 init_special_inode(inode, inode->i_mode,
1909                                    inode->i_rdev);
1910
1911                 EXIT;
1912         }
1913 }
1914
1915 void ll_delete_inode(struct inode *inode)
1916 {
1917         struct cl_inode_info *lli = cl_i2info(inode);
1918         ENTRY;
1919
1920         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1921                 /* discard all dirty pages before truncating them, required by
1922                  * osc_extent implementation at LU-1030. */
1923                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1924                                    CL_FSYNC_DISCARD, 1);
1925
1926         truncate_inode_pages(&inode->i_data, 0);
1927
1928         /* Workaround for LU-118 */
1929         if (inode->i_data.nrpages) {
1930                 spin_lock_irq(&inode->i_data.tree_lock);
1931                 spin_unlock_irq(&inode->i_data.tree_lock);
1932                 LASSERTF(inode->i_data.nrpages == 0,
1933                          "inode="DFID"(%p) nrpages=%lu, see "
1934                          "http://jira.whamcloud.com/browse/LU-118\n",
1935                          PFID(ll_inode2fid(inode)), inode,
1936                          inode->i_data.nrpages);
1937         }
1938         /* Workaround end */
1939
1940 #ifdef HAVE_SBOPS_EVICT_INODE
1941         ll_clear_inode(inode);
1942 #endif
1943         clear_inode(inode);
1944
1945         EXIT;
1946 }
1947
1948 int ll_iocontrol(struct inode *inode, struct file *file,
1949                  unsigned int cmd, unsigned long arg)
1950 {
1951         struct ll_sb_info *sbi = ll_i2sbi(inode);
1952         struct ptlrpc_request *req = NULL;
1953         int rc, flags = 0;
1954         ENTRY;
1955
1956         switch(cmd) {
1957         case FSFILT_IOC_GETFLAGS: {
1958                 struct mdt_body *body;
1959                 struct md_op_data *op_data;
1960
1961                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1962                                              0, 0, LUSTRE_OPC_ANY,
1963                                              NULL);
1964                 if (IS_ERR(op_data))
1965                         RETURN(PTR_ERR(op_data));
1966
1967                 op_data->op_valid = OBD_MD_FLFLAGS;
1968                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1969                 ll_finish_md_op_data(op_data);
1970                 if (rc) {
1971                         CERROR("%s: failure inode "DFID": rc = %d\n",
1972                                sbi->ll_md_exp->exp_obd->obd_name,
1973                                PFID(ll_inode2fid(inode)), rc);
1974                         RETURN(-abs(rc));
1975                 }
1976
1977                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1978
1979                 flags = body->flags;
1980
1981                 ptlrpc_req_finished(req);
1982
1983                 RETURN(put_user(flags, (int *)arg));
1984         }
1985         case FSFILT_IOC_SETFLAGS: {
1986                 struct lov_stripe_md *lsm;
1987                 struct obd_info oinfo = { { { 0 } } };
1988                 struct md_op_data *op_data;
1989
1990                 if (get_user(flags, (int *)arg))
1991                         RETURN(-EFAULT);
1992
1993                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1994                                              LUSTRE_OPC_ANY, NULL);
1995                 if (IS_ERR(op_data))
1996                         RETURN(PTR_ERR(op_data));
1997
1998                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1999                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2000                 rc = md_setattr(sbi->ll_md_exp, op_data,
2001                                 NULL, 0, NULL, 0, &req, NULL);
2002                 ll_finish_md_op_data(op_data);
2003                 ptlrpc_req_finished(req);
2004                 if (rc)
2005                         RETURN(rc);
2006
2007                 inode->i_flags = ll_ext_to_inode_flags(flags);
2008
2009                 lsm = ccc_inode_lsm_get(inode);
2010                 if (!lsm_has_objects(lsm)) {
2011                         ccc_inode_lsm_put(inode, lsm);
2012                         RETURN(0);
2013                 }
2014
2015                 OBDO_ALLOC(oinfo.oi_oa);
2016                 if (!oinfo.oi_oa) {
2017                         ccc_inode_lsm_put(inode, lsm);
2018                         RETURN(-ENOMEM);
2019                 }
2020                 oinfo.oi_md = lsm;
2021                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2022                 oinfo.oi_oa->o_flags = flags;
2023                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2024                                        OBD_MD_FLGROUP;
2025                 oinfo.oi_capa = ll_mdscapa_get(inode);
2026                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2027                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2028                 capa_put(oinfo.oi_capa);
2029                 OBDO_FREE(oinfo.oi_oa);
2030                 ccc_inode_lsm_put(inode, lsm);
2031
2032                 if (rc && rc != -EPERM && rc != -EACCES)
2033                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2034
2035                 RETURN(rc);
2036         }
2037         default:
2038                 RETURN(-ENOSYS);
2039         }
2040
2041         RETURN(0);
2042 }
2043
2044 int ll_flush_ctx(struct inode *inode)
2045 {
2046         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2047
2048         CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2049
2050         obd_set_info_async(NULL, sbi->ll_md_exp,
2051                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2052                            0, NULL, NULL);
2053         obd_set_info_async(NULL, sbi->ll_dt_exp,
2054                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2055                            0, NULL, NULL);
2056         return 0;
2057 }
2058
2059 /* umount -f client means force down, don't save state */
2060 void ll_umount_begin(struct super_block *sb)
2061 {
2062         struct ll_sb_info *sbi = ll_s2sbi(sb);
2063         struct obd_device *obd;
2064         struct obd_ioctl_data *ioc_data;
2065         ENTRY;
2066
2067         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2068                sb->s_count, atomic_read(&sb->s_active));
2069
2070         obd = class_exp2obd(sbi->ll_md_exp);
2071         if (obd == NULL) {
2072                 CERROR("Invalid MDC connection handle "LPX64"\n",
2073                        sbi->ll_md_exp->exp_handle.h_cookie);
2074                 EXIT;
2075                 return;
2076         }
2077         obd->obd_force = 1;
2078
2079         obd = class_exp2obd(sbi->ll_dt_exp);
2080         if (obd == NULL) {
2081                 CERROR("Invalid LOV connection handle "LPX64"\n",
2082                        sbi->ll_dt_exp->exp_handle.h_cookie);
2083                 EXIT;
2084                 return;
2085         }
2086         obd->obd_force = 1;
2087
2088         OBD_ALLOC_PTR(ioc_data);
2089         if (ioc_data) {
2090                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2091                               sizeof *ioc_data, ioc_data, NULL);
2092
2093                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2094                               sizeof *ioc_data, ioc_data, NULL);
2095
2096                 OBD_FREE_PTR(ioc_data);
2097         }
2098
2099         /* Really, we'd like to wait until there are no requests outstanding,
2100          * and then continue.  For now, we just invalidate the requests,
2101          * schedule() and sleep one second if needed, and hope.
2102          */
2103         schedule();
2104         EXIT;
2105 }
2106
2107 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2108 {
2109         struct ll_sb_info *sbi = ll_s2sbi(sb);
2110         char *profilenm = get_profile_name(sb);
2111         int err;
2112         __u32 read_only;
2113
2114         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2115                 read_only = *flags & MS_RDONLY;
2116                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2117                                          sizeof(KEY_READ_ONLY),
2118                                          KEY_READ_ONLY, sizeof(read_only),
2119                                          &read_only, NULL);
2120                 if (err) {
2121                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2122                                       profilenm, read_only ?
2123                                       "read-only" : "read-write", err);
2124                         return err;
2125                 }
2126
2127                 if (read_only)
2128                         sb->s_flags |= MS_RDONLY;
2129                 else
2130                         sb->s_flags &= ~MS_RDONLY;
2131
2132                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2133                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2134                                       read_only ?  "read-only" : "read-write");
2135         }
2136         return 0;
2137 }
2138
2139 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2140                   struct super_block *sb, struct lookup_intent *it)
2141 {
2142         struct ll_sb_info *sbi = NULL;
2143         struct lustre_md md;
2144         int rc;
2145         ENTRY;
2146
2147         LASSERT(*inode || sb);
2148         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2149         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2150                               sbi->ll_md_exp, &md);
2151         if (rc)
2152                 RETURN(rc);
2153
2154         if (*inode) {
2155                 ll_update_inode(*inode, &md);
2156         } else {
2157                 LASSERT(sb != NULL);
2158
2159                 /*
2160                  * At this point server returns to client's same fid as client
2161                  * generated for creating. So using ->fid1 is okay here.
2162                  */
2163                 LASSERT(fid_is_sane(&md.body->fid1));
2164
2165                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2166                                              sbi->ll_flags & LL_SBI_32BIT_API),
2167                                  &md);
2168                 if (*inode == NULL || IS_ERR(*inode)) {
2169 #ifdef CONFIG_FS_POSIX_ACL
2170                         if (md.posix_acl) {
2171                                 posix_acl_release(md.posix_acl);
2172                                 md.posix_acl = NULL;
2173                         }
2174 #endif
2175                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2176                         *inode = NULL;
2177                         CERROR("new_inode -fatal: rc %d\n", rc);
2178                         GOTO(out, rc);
2179                 }
2180         }
2181
2182         /* Handling piggyback layout lock.
2183          * Layout lock can be piggybacked by getattr and open request.
2184          * The lsm can be applied to inode only if it comes with a layout lock
2185          * otherwise correct layout may be overwritten, for example:
2186          * 1. proc1: mdt returns a lsm but not granting layout
2187          * 2. layout was changed by another client
2188          * 3. proc2: refresh layout and layout lock granted
2189          * 4. proc1: to apply a stale layout */
2190         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2191                 struct lustre_handle lockh;
2192                 struct ldlm_lock *lock;
2193
2194                 lockh.cookie = it->d.lustre.it_lock_handle;
2195                 lock = ldlm_handle2lock(&lockh);
2196                 LASSERT(lock != NULL);
2197                 if (ldlm_has_layout(lock)) {
2198                         struct cl_object_conf conf;
2199
2200                         memset(&conf, 0, sizeof(conf));
2201                         conf.coc_opc = OBJECT_CONF_SET;
2202                         conf.coc_inode = *inode;
2203                         conf.coc_lock = lock;
2204                         conf.u.coc_md = &md;
2205                         (void)ll_layout_conf(*inode, &conf);
2206                 }
2207                 LDLM_LOCK_PUT(lock);
2208         }
2209
2210 out:
2211         if (md.lsm != NULL)
2212                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2213         md_free_lustre_md(sbi->ll_md_exp, &md);
2214         RETURN(rc);
2215 }
2216
2217 int ll_obd_statfs(struct inode *inode, void *arg)
2218 {
2219         struct ll_sb_info *sbi = NULL;
2220         struct obd_export *exp;
2221         char *buf = NULL;
2222         struct obd_ioctl_data *data = NULL;
2223         __u32 type;
2224         __u32 flags;
2225         int len = 0, rc;
2226
2227         if (!inode || !(sbi = ll_i2sbi(inode)))
2228                 GOTO(out_statfs, rc = -EINVAL);
2229
2230         rc = obd_ioctl_getdata(&buf, &len, arg);
2231         if (rc)
2232                 GOTO(out_statfs, rc);
2233
2234         data = (void*)buf;
2235         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2236             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2237                 GOTO(out_statfs, rc = -EINVAL);
2238
2239         if (data->ioc_inllen1 != sizeof(__u32) ||
2240             data->ioc_inllen2 != sizeof(__u32) ||
2241             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2242             data->ioc_plen2 != sizeof(struct obd_uuid))
2243                 GOTO(out_statfs, rc = -EINVAL);
2244
2245         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2246         if (type & LL_STATFS_LMV)
2247                 exp = sbi->ll_md_exp;
2248         else if (type & LL_STATFS_LOV)
2249                 exp = sbi->ll_dt_exp;
2250         else
2251                 GOTO(out_statfs, rc = -ENODEV);
2252
2253         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2254         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2255         if (rc)
2256                 GOTO(out_statfs, rc);
2257 out_statfs:
2258         if (buf)
2259                 obd_ioctl_freedata(buf, len);
2260         return rc;
2261 }
2262
2263 int ll_process_config(struct lustre_cfg *lcfg)
2264 {
2265         char *ptr;
2266         void *sb;
2267         struct lprocfs_static_vars lvars;
2268         unsigned long x;
2269         int rc = 0;
2270
2271         lprocfs_llite_init_vars(&lvars);
2272
2273         /* The instance name contains the sb: lustre-client-aacfe000 */
2274         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2275         if (!ptr || !*(++ptr))
2276                 return -EINVAL;
2277         if (sscanf(ptr, "%lx", &x) != 1)
2278                 return -EINVAL;
2279         sb = (void *)x;
2280         /* This better be a real Lustre superblock! */
2281         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2282
2283         /* Note we have not called client_common_fill_super yet, so
2284            proc fns must be able to handle that! */
2285         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2286                                       lcfg, sb);
2287         if (rc > 0)
2288                 rc = 0;
2289         return(rc);
2290 }
2291
2292 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2293 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2294                                        struct inode *i1, struct inode *i2,
2295                                        const char *name, int namelen,
2296                                        int mode, __u32 opc, void *data)
2297 {
2298         LASSERT(i1 != NULL);
2299
2300         if (namelen > ll_i2sbi(i1)->ll_namelen)
2301                 return ERR_PTR(-ENAMETOOLONG);
2302
2303         if (op_data == NULL)
2304                 OBD_ALLOC_PTR(op_data);
2305
2306         if (op_data == NULL)
2307                 return ERR_PTR(-ENOMEM);
2308
2309         ll_i2gids(op_data->op_suppgids, i1, i2);
2310         op_data->op_fid1 = *ll_inode2fid(i1);
2311         op_data->op_capa1 = ll_mdscapa_get(i1);
2312
2313         if (i2) {
2314                 op_data->op_fid2 = *ll_inode2fid(i2);
2315                 op_data->op_capa2 = ll_mdscapa_get(i2);
2316         } else {
2317                 fid_zero(&op_data->op_fid2);
2318                 op_data->op_capa2 = NULL;
2319         }
2320
2321         op_data->op_name = name;
2322         op_data->op_namelen = namelen;
2323         op_data->op_mode = mode;
2324         op_data->op_mod_time = cfs_time_current_sec();
2325         op_data->op_fsuid = current_fsuid();
2326         op_data->op_fsgid = current_fsgid();
2327         op_data->op_cap = cfs_curproc_cap_pack();
2328         op_data->op_bias = 0;
2329         op_data->op_cli_flags = 0;
2330         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2331              filename_is_volatile(name, namelen, NULL))
2332                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2333         op_data->op_opc = opc;
2334         op_data->op_mds = 0;
2335         op_data->op_data = data;
2336
2337         /* If the file is being opened after mknod() (normally due to NFS)
2338          * try to use the default stripe data from parent directory for
2339          * allocating OST objects.  Try to pass the parent FID to MDS. */
2340         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2341             !ll_i2info(i2)->lli_has_smd) {
2342                 struct ll_inode_info *lli = ll_i2info(i2);
2343
2344                 spin_lock(&lli->lli_lock);
2345                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2346                         op_data->op_fid1 = lli->lli_pfid;
2347                 spin_unlock(&lli->lli_lock);
2348                 /** We ignore parent's capability temporary. */
2349         }
2350
2351         /* When called by ll_setattr_raw, file is i1. */
2352         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2353                 op_data->op_bias |= MDS_DATA_MODIFIED;
2354
2355         return op_data;
2356 }
2357
2358 void ll_finish_md_op_data(struct md_op_data *op_data)
2359 {
2360         capa_put(op_data->op_capa1);
2361         capa_put(op_data->op_capa2);
2362         OBD_FREE_PTR(op_data);
2363 }
2364
2365 #ifdef HAVE_SUPEROPS_USE_DENTRY
2366 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2367 #else
2368 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2369 #endif
2370 {
2371         struct ll_sb_info *sbi;
2372
2373 #ifdef HAVE_SUPEROPS_USE_DENTRY
2374         LASSERT((seq != NULL) && (dentry != NULL));
2375         sbi = ll_s2sbi(dentry->d_sb);
2376 #else
2377         LASSERT((seq != NULL) && (vfs != NULL));
2378         sbi = ll_s2sbi(vfs->mnt_sb);
2379 #endif
2380
2381         if (sbi->ll_flags & LL_SBI_NOLCK)
2382                 seq_puts(seq, ",nolock");
2383
2384         if (sbi->ll_flags & LL_SBI_FLOCK)
2385                 seq_puts(seq, ",flock");
2386
2387         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2388                 seq_puts(seq, ",localflock");
2389
2390         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2391                 seq_puts(seq, ",user_xattr");
2392
2393         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2394                 seq_puts(seq, ",lazystatfs");
2395
2396         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2397                 seq_puts(seq, ",user_fid2path");
2398
2399         RETURN(0);
2400 }
2401
2402 /**
2403  * Get obd name by cmd, and copy out to user space
2404  */
2405 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2406 {
2407         struct ll_sb_info *sbi = ll_i2sbi(inode);
2408         struct obd_device *obd;
2409         ENTRY;
2410
2411         if (cmd == OBD_IOC_GETDTNAME)
2412                 obd = class_exp2obd(sbi->ll_dt_exp);
2413         else if (cmd == OBD_IOC_GETMDNAME)
2414                 obd = class_exp2obd(sbi->ll_md_exp);
2415         else
2416                 RETURN(-EINVAL);
2417
2418         if (!obd)
2419                 RETURN(-ENOENT);
2420
2421         if (copy_to_user((void *)arg, obd->obd_name,
2422                          strlen(obd->obd_name) + 1))
2423                 RETURN(-EFAULT);
2424
2425         RETURN(0);
2426 }
2427
2428 /**
2429  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2430  * fsname will be returned in this buffer; otherwise, a static buffer will be
2431  * used to store the fsname and returned to caller.
2432  */
2433 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2434 {
2435         static char fsname_static[MTI_NAME_MAXLEN];
2436         struct lustre_sb_info *lsi = s2lsi(sb);
2437         char *ptr;
2438         int len;
2439
2440         if (buf == NULL) {
2441                 /* this means the caller wants to use static buffer
2442                  * and it doesn't care about race. Usually this is
2443                  * in error reporting path */
2444                 buf = fsname_static;
2445                 buflen = sizeof(fsname_static);
2446         }
2447
2448         len = strlen(lsi->lsi_lmd->lmd_profile);
2449         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2450         if (ptr && (strcmp(ptr, "-client") == 0))
2451                 len -= 7;
2452
2453         if (unlikely(len >= buflen))
2454                 len = buflen - 1;
2455         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2456         buf[len] = '\0';
2457
2458         return buf;
2459 }
2460
2461 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2462 {
2463         char *path = NULL;
2464
2465         struct path p;
2466
2467         p.dentry = dentry;
2468         p.mnt = current->fs->root.mnt;
2469         path_get(&p);
2470         path = d_path(&p, buf, bufsize);
2471         path_put(&p);
2472         return path;
2473 }
2474
2475 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2476 {
2477         char *buf, *path = NULL;
2478         struct dentry *dentry = NULL;
2479         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2480
2481         /* this can be called inside spin lock so use GFP_ATOMIC. */
2482         buf = (char *)__get_free_page(GFP_ATOMIC);
2483         if (buf != NULL) {
2484                 dentry = d_find_alias(page->mapping->host);
2485                 if (dentry != NULL)
2486                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2487         }
2488
2489         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2490               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2491               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2492               PFID(&obj->cob_header.coh_lu.loh_fid),
2493               (path && !IS_ERR(path)) ? path : "", ioret);
2494
2495         if (dentry != NULL)
2496                 dput(dentry);
2497
2498         if (buf != NULL)
2499                 free_page((unsigned long)buf);
2500 }