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 struct dentry_operations ll_d_root_ops = {
 166         .d_compare = ll_dcompare,
 167         .d_revalidate = ll_revalidate_nd,
 168 };
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = 0;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         if (proc_lustre_fs_root) {
 204                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 205                                                   dt, md);
 206                 if (err < 0)
 207                         CERROR("could not register mount in /proc/fs/lustre\n");
 208         }
 209
 210         /* indicate the features supported by this client */
 211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 212                                   OBD_CONNECT_ATTRFID  |
 213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 214                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 215                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 216                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 217                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 218                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 219                                   OBD_CONNECT_EINPROGRESS |
 220                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 221                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 222
 223         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 224                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 225
 226 #ifdef HAVE_LRU_RESIZE_SUPPORT
 227         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 228                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 229 #endif
 230 #ifdef CONFIG_FS_POSIX_ACL
 231         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 232 #endif
 233
 234         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 235                 /* flag mdc connection as lightweight, only used for test
 236                  * purpose, use with care */
 237                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 238
 239         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 240         data->ocd_version = LUSTRE_VERSION_CODE;
 241
 242         if (sb->s_flags & MS_RDONLY)
 243                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 244         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 245                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 246
 247 #ifdef HAVE_MS_FLOCK_LOCK
 248         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 249         sb->s_flags |= MS_FLOCK_LOCK;
 250 #endif
 251 #ifdef MS_HAS_NEW_AOPS
 252         sb->s_flags |= MS_HAS_NEW_AOPS;
 253 #endif
 254
 255         if (sbi->ll_flags & LL_SBI_FLOCK)
 256                 sbi->ll_fop = &ll_file_operations_flock;
 257         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 258                 sbi->ll_fop = &ll_file_operations;
 259         else
 260                 sbi->ll_fop = &ll_file_operations_noflock;
 261
 262         /* real client */
 263         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 264         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 265                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 266
 267         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 268
 269         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 270         if (err == -EBUSY) {
 271                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 272                                    "recovery, of which this client is not a "
 273                                    "part. Please wait for recovery to complete,"
 274                                    " abort, or time out.\n", md);
 275                 GOTO(out, err);
 276         } else if (err) {
 277                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 278                 GOTO(out, err);
 279         }
 280
 281         sbi->ll_md_exp->exp_connect_data = *data;
 282
 283         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 284                            LUSTRE_SEQ_METADATA);
 285         if (err) {
 286                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 287                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 288                 GOTO(out_md, err);
 289         }
 290
 291         /* For mount, we only need fs info from MDT0, and also in DNE, it
 292          * can make sure the client can be mounted as long as MDT0 is
 293          * avaible */
 294         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 295                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 296                         OBD_STATFS_FOR_MDT0);
 297         if (err)
 298                 GOTO(out_md_fid, err);
 299
 300         /* This needs to be after statfs to ensure connect has finished.
 301          * Note that "data" does NOT contain the valid connect reply.
 302          * If connecting to a 1.8 server there will be no LMV device, so
 303          * we can access the MDC export directly and exp_connect_flags will
 304          * be non-zero, but if accessing an upgraded 2.1 server it will
 305          * have the correct flags filled in.
 306          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 307         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 308         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 309             valid != CLIENT_CONNECT_MDT_REQD) {
 310                 char *buf;
 311
 312                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 313                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 314                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 315                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 316                                    "feature(s) needed for correct operation "
 317                                    "of this client (%s). Please upgrade "
 318                                    "server or downgrade client.\n",
 319                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 320                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 321                 GOTO(out_md_fid, err = -EPROTO);
 322         }
 323
 324         size = sizeof(*data);
 325         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 326                            KEY_CONN_DATA,  &size, data, NULL);
 327         if (err) {
 328                 CERROR("%s: Get connect data failed: rc = %d\n",
 329                        sbi->ll_md_exp->exp_obd->obd_name, err);
 330                 GOTO(out_md_fid, err);
 331         }
 332
 333         LASSERT(osfs->os_bsize);
 334         sb->s_blocksize = osfs->os_bsize;
 335         sb->s_blocksize_bits = log2(osfs->os_bsize);
 336         sb->s_magic = LL_SUPER_MAGIC;
 337         sb->s_maxbytes = MAX_LFS_FILESIZE;
 338         sbi->ll_namelen = osfs->os_namelen;
 339         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 340
 341         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 342             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 343                 LCONSOLE_INFO("Disabling user_xattr feature because "
 344                               "it is not supported on the server\n");
 345                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 349 #ifdef MS_POSIXACL
 350                 sb->s_flags |= MS_POSIXACL;
 351 #endif
 352                 sbi->ll_flags |= LL_SBI_ACL;
 353         } else {
 354                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 355 #ifdef MS_POSIXACL
 356                 sb->s_flags &= ~MS_POSIXACL;
 357 #endif
 358                 sbi->ll_flags &= ~LL_SBI_ACL;
 359         }
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 362                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 363                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client is set as remote by default.\n");
 365                 }
 366         } else {
 367                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 368                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 369                         LCONSOLE_INFO("client claims to be remote, but server "
 370                                       "rejected, forced to be local.\n");
 371                 }
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 375                 LCONSOLE_INFO("client enabled MDS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 380                 LCONSOLE_INFO("client enabled OSS capability!\n");
 381                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 382         }
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 385                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 388                 sbi->ll_md_brw_size = data->ocd_brw_size;
 389         else
 390                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 391
 392         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 393                 LCONSOLE_INFO("Layout lock feature supported.\n");
 394                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 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
 582 #ifdef HAVE_DCACHE_LOCK
 583         sb->s_root->d_op = &ll_d_root_ops;
 584 #else
 585         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 586         d_set_d_op(sb->s_root, &ll_d_root_ops);
 587         sb->s_d_op = &ll_d_ops;
 588 #endif
 589
 590         sbi->ll_sdev_orig = sb->s_dev;
 591
 592         /* We set sb->s_dev equal on all lustre clients in order to support
 593          * NFS export clustering.  NFSD requires that the FSID be the same
 594          * on all clients. */
 595         /* s_dev is also used in lt_compare() to compare two fs, but that is
 596          * only a node-local comparison. */
 597         uuid = obd_get_uuid(sbi->ll_md_exp);
 598         if (uuid != NULL) {
 599                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 600                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 601         }
 602
 603         if (data != NULL)
 604                 OBD_FREE_PTR(data);
 605         if (osfs != NULL)
 606                 OBD_FREE_PTR(osfs);
 607
 608         RETURN(err);
 609 out_root:
 610         if (root)
 611                 iput(root);
 612 out_lock_cn_cb:
 613         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 614 out_dt:
 615         obd_disconnect(sbi->ll_dt_exp);
 616         sbi->ll_dt_exp = NULL;
 617         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 618         obd_zombie_barrier();
 619 out_md_fid:
 620         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 621 out_md:
 622         obd_disconnect(sbi->ll_md_exp);
 623         sbi->ll_md_exp = NULL;
 624 out:
 625         if (data != NULL)
 626                 OBD_FREE_PTR(data);
 627         if (osfs != NULL)
 628                 OBD_FREE_PTR(osfs);
 629         lprocfs_unregister_mountpoint(sbi);
 630         return err;
 631 }
 632
 633 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 634 {
 635         int size, rc;
 636
 637         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 638         size = sizeof(int);
 639         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 640                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 641         if (rc)
 642                 CERROR("Get max mdsize error rc %d \n", rc);
 643
 644         RETURN(rc);
 645 }
 646
 647 void ll_dump_inode(struct inode *inode)
 648 {
 649         struct ll_d_hlist_node *tmp;
 650         int dentry_count = 0;
 651
 652         LASSERT(inode != NULL);
 653
 654         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 655                 dentry_count++;
 656
 657         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 658                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 659                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 660 }
 661
 662 void lustre_dump_dentry(struct dentry *dentry, int recur)
 663 {
 664         struct list_head *tmp;
 665         int subdirs = 0;
 666
 667         LASSERT(dentry != NULL);
 668
 669         list_for_each(tmp, &dentry->d_subdirs)
 670                 subdirs++;
 671
 672         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 673                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 674                dentry->d_name.len, dentry->d_name.name,
 675                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 676                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 677                dentry->d_flags, dentry->d_fsdata, subdirs);
 678         if (dentry->d_inode != NULL)
 679                 ll_dump_inode(dentry->d_inode);
 680
 681         if (recur == 0)
 682                 return;
 683
 684         list_for_each(tmp, &dentry->d_subdirs) {
 685                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 686                 lustre_dump_dentry(d, recur - 1);
 687         }
 688 }
 689
 690 void client_common_put_super(struct super_block *sb)
 691 {
 692         struct ll_sb_info *sbi = ll_s2sbi(sb);
 693         ENTRY;
 694
 695 #ifdef CONFIG_FS_POSIX_ACL
 696         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 697                 et_fini(&sbi->ll_et);
 698                 rct_fini(&sbi->ll_rct);
 699         }
 700 #endif
 701
 702         ll_close_thread_shutdown(sbi->ll_lcq);
 703
 704         cl_sb_fini(sb);
 705
 706         cfs_list_del(&sbi->ll_conn_chain);
 707
 708         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 709         obd_disconnect(sbi->ll_dt_exp);
 710         sbi->ll_dt_exp = NULL;
 711         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 712          * see LU-2543. */
 713         obd_zombie_barrier();
 714
 715         lprocfs_unregister_mountpoint(sbi);
 716
 717         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 718         obd_disconnect(sbi->ll_md_exp);
 719         sbi->ll_md_exp = NULL;
 720
 721         EXIT;
 722 }
 723
 724 void ll_kill_super(struct super_block *sb)
 725 {
 726         struct ll_sb_info *sbi;
 727
 728         ENTRY;
 729
 730         /* not init sb ?*/
 731         if (!(sb->s_flags & MS_ACTIVE))
 732                 return;
 733
 734         sbi = ll_s2sbi(sb);
 735         /* we need restore s_dev from changed for clustred NFS before put_super
 736          * because new kernels have cached s_dev and change sb->s_dev in
 737          * put_super not affected real removing devices */
 738         if (sbi) {
 739                 sb->s_dev = sbi->ll_sdev_orig;
 740                 sbi->ll_umounting = 1;
 741         }
 742         EXIT;
 743 }
 744
 745 char *ll_read_opt(const char *opt, char *data)
 746 {
 747         char *value;
 748         char *retval;
 749         ENTRY;
 750
 751         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 752         if (strncmp(opt, data, strlen(opt)))
 753                 RETURN(NULL);
 754         if ((value = strchr(data, '=')) == NULL)
 755                 RETURN(NULL);
 756
 757         value++;
 758         OBD_ALLOC(retval, strlen(value) + 1);
 759         if (!retval) {
 760                 CERROR("out of memory!\n");
 761                 RETURN(NULL);
 762         }
 763
 764         memcpy(retval, value, strlen(value)+1);
 765         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 766         RETURN(retval);
 767 }
 768
 769 static inline int ll_set_opt(const char *opt, char *data, int fl)
 770 {
 771         if (strncmp(opt, data, strlen(opt)) != 0)
 772                 return(0);
 773         else
 774                 return(fl);
 775 }
 776
 777 /* non-client-specific mount options are parsed in lmd_parse */
 778 static int ll_options(char *options, int *flags)
 779 {
 780         int tmp;
 781         char *s1 = options, *s2;
 782         ENTRY;
 783
 784         if (!options)
 785                 RETURN(0);
 786
 787         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 788
 789         while (*s1) {
 790                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 791                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 822                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 823                 if (tmp) {
 824                         /* Ignore deprecated mount option.  The client will
 825                          * always try to mount with ACL support, whether this
 826                          * is used depends on whether server supports it. */
 827                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 828                                                   "mount option 'acl'.\n");
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 832                 if (tmp) {
 833                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 834                                                   "mount option 'noacl'.\n");
 835                         goto next;
 836                 }
 837 #else
 838 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 839 #endif
 840                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 846                 if (tmp) {
 847                         *flags |= tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 851                 if (tmp) {
 852                         *flags &= ~tmp;
 853                         goto next;
 854                 }
 855
 856                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 862                 if (tmp) {
 863                         *flags &= ~tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 902                 if (tmp) {
 903                         *flags &= ~tmp;
 904                         goto next;
 905                 }
 906                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 907                                    s1);
 908                 RETURN(-EINVAL);
 909
 910 next:
 911                 /* Find next opt */
 912                 s2 = strchr(s1, ',');
 913                 if (s2 == NULL)
 914                         break;
 915                 s1 = s2 + 1;
 916         }
 917         RETURN(0);
 918 }
 919
 920 void ll_lli_init(struct ll_inode_info *lli)
 921 {
 922         lli->lli_inode_magic = LLI_INODE_MAGIC;
 923         lli->lli_flags = 0;
 924         lli->lli_ioepoch = 0;
 925         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 926         spin_lock_init(&lli->lli_lock);
 927         lli->lli_posix_acl = NULL;
 928         lli->lli_remote_perms = NULL;
 929         mutex_init(&lli->lli_rmtperm_mutex);
 930         /* Do not set lli_fid, it has been initialized already. */
 931         fid_zero(&lli->lli_pfid);
 932         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 933         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 934         cfs_atomic_set(&lli->lli_open_count, 0);
 935         lli->lli_mds_capa = NULL;
 936         lli->lli_rmtperm_time = 0;
 937         lli->lli_pending_och = NULL;
 938         lli->lli_mds_read_och = NULL;
 939         lli->lli_mds_write_och = NULL;
 940         lli->lli_mds_exec_och = NULL;
 941         lli->lli_open_fd_read_count = 0;
 942         lli->lli_open_fd_write_count = 0;
 943         lli->lli_open_fd_exec_count = 0;
 944         mutex_init(&lli->lli_och_mutex);
 945         spin_lock_init(&lli->lli_agl_lock);
 946         lli->lli_has_smd = false;
 947         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 948         lli->lli_clob = NULL;
 949
 950         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 951         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 952                 mutex_init(&lli->lli_readdir_mutex);
 953                 lli->lli_opendir_key = NULL;
 954                 lli->lli_sai = NULL;
 955                 lli->lli_def_acl = NULL;
 956                 spin_lock_init(&lli->lli_sa_lock);
 957                 lli->lli_opendir_pid = 0;
 958         } else {
 959                 sema_init(&lli->lli_size_sem, 1);
 960                 lli->lli_size_sem_owner = NULL;
 961                 lli->lli_symlink_name = NULL;
 962                 init_rwsem(&lli->lli_trunc_sem);
 963                 mutex_init(&lli->lli_write_mutex);
 964                 init_rwsem(&lli->lli_glimpse_sem);
 965                 lli->lli_glimpse_time = 0;
 966                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 967                 lli->lli_agl_index = 0;
 968                 lli->lli_async_rc = 0;
 969                 lli->lli_volatile = false;
 970         }
 971         mutex_init(&lli->lli_layout_mutex);
 972 }
 973
 974 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 975 {
 976 #ifdef HAVE_BDI_REGISTER
 977         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 978
 979 #ifdef HAVE_BDI_NAME
 980         bdi->name = "lustre";
 981 #endif
 982         return bdi_register(bdi, NULL, "lustre-%d",
 983                             atomic_inc_return(&ll_bdi_num));
 984 #else
 985         return 0;
 986 #endif
 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         cfs_module_get();
1009
1010         /* client additional sb info */
1011         lsi->lsi_llsbi = sbi = ll_init_sbi();
1012         if (!sbi) {
1013                 cfs_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 #ifdef HAVE_SB_BDI
1032         sb->s_bdi = &lsi->lsi_bdi;
1033 #endif
1034
1035         /* Generate a string unique to this super, in case some joker tries
1036            to mount the same fs at two mount points.
1037            Use the address of the super itself.*/
1038         cfg->cfg_instance = sb;
1039         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1040         cfg->cfg_callback = class_config_llog_handler;
1041         /* set up client obds */
1042         err = lustre_process_log(sb, profilenm, cfg);
1043         if (err < 0) {
1044                 CERROR("Unable to process log: %d\n", err);
1045                 GOTO(out_free, err);
1046         }
1047
1048         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1049         lprof = class_get_profile(profilenm);
1050         if (lprof == NULL) {
1051                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1052                                    " read from the MGS.  Does that filesystem "
1053                                    "exist?\n", profilenm);
1054                 GOTO(out_free, err = -EINVAL);
1055         }
1056         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1057                lprof->lp_md, lprof->lp_dt);
1058
1059         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1060         if (!dt)
1061                 GOTO(out_free, err = -ENOMEM);
1062         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1063
1064         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1065         if (!md)
1066                 GOTO(out_free, err = -ENOMEM);
1067         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1068
1069         /* connections, registrations, sb setup */
1070         err = client_common_fill_super(sb, md, dt, mnt);
1071
1072 out_free:
1073         if (md)
1074                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1075         if (dt)
1076                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1077         if (err)
1078                 ll_put_super(sb);
1079         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1080                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1081
1082         OBD_FREE_PTR(cfg);
1083         RETURN(err);
1084 } /* ll_fill_super */
1085
1086
1087 void lu_context_keys_dump(void);
1088
1089 void ll_put_super(struct super_block *sb)
1090 {
1091         struct config_llog_instance cfg;
1092         struct obd_device *obd;
1093         struct lustre_sb_info *lsi = s2lsi(sb);
1094         struct ll_sb_info *sbi = ll_s2sbi(sb);
1095         char *profilenm = get_profile_name(sb);
1096         int force = 1, next;
1097         ENTRY;
1098
1099         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1100
1101         ll_print_capa_stat(sbi);
1102
1103         cfg.cfg_instance = sb;
1104         lustre_end_log(sb, profilenm, &cfg);
1105
1106         if (sbi->ll_md_exp) {
1107                 obd = class_exp2obd(sbi->ll_md_exp);
1108                 if (obd)
1109                         force = obd->obd_force;
1110         }
1111
1112         /* We need to set force before the lov_disconnect in
1113            lustre_common_put_super, since l_d cleans up osc's as well. */
1114         if (force) {
1115                 next = 0;
1116                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1117                                                      &next)) != NULL) {
1118                         obd->obd_force = force;
1119                 }
1120         }
1121
1122         if (sbi->ll_lcq) {
1123                 /* Only if client_common_fill_super succeeded */
1124                 client_common_put_super(sb);
1125         }
1126
1127         next = 0;
1128         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1129                 class_manual_cleanup(obd);
1130         }
1131
1132         if (sbi->ll_flags & LL_SBI_VERBOSE)
1133                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1134
1135         if (profilenm)
1136                 class_del_profile(profilenm);
1137
1138         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1139                 bdi_destroy(&lsi->lsi_bdi);
1140                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1141         }
1142
1143         ll_free_sbi(sb);
1144         lsi->lsi_llsbi = NULL;
1145
1146         lustre_common_put_super(sb);
1147
1148         cfs_module_put(THIS_MODULE);
1149
1150         EXIT;
1151 } /* client_put_super */
1152
1153 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1154 {
1155         struct inode *inode = NULL;
1156
1157         /* NOTE: we depend on atomic igrab() -bzzz */
1158         lock_res_and_lock(lock);
1159         if (lock->l_resource->lr_lvb_inode) {
1160                 struct ll_inode_info * lli;
1161                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1162                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1163                         inode = igrab(lock->l_resource->lr_lvb_inode);
1164                 } else {
1165                         inode = lock->l_resource->lr_lvb_inode;
1166                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1167                                          D_WARNING, lock, "lr_lvb_inode %p is "
1168                                          "bogus: magic %08x",
1169                                          lock->l_resource->lr_lvb_inode,
1170                                          lli->lli_inode_magic);
1171                         inode = NULL;
1172                 }
1173         }
1174         unlock_res_and_lock(lock);
1175         return inode;
1176 }
1177
1178 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1179 {
1180         struct inode *inode = NULL;
1181         /* NOTE: we depend on atomic igrab() -bzzz */
1182         lock_res_and_lock(lock);
1183         if (lock->l_ast_data) {
1184                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1185                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1186                         inode = igrab(lock->l_ast_data);
1187                 } else {
1188                         inode = lock->l_ast_data;
1189                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1190                                          D_WARNING, lock, "l_ast_data %p is "
1191                                          "bogus: magic %08x", lock->l_ast_data,
1192                                          lli->lli_inode_magic);
1193                         inode = NULL;
1194                 }
1195         }
1196         unlock_res_and_lock(lock);
1197         return inode;
1198 }
1199
1200 void ll_clear_inode(struct inode *inode)
1201 {
1202         struct ll_inode_info *lli = ll_i2info(inode);
1203         struct ll_sb_info *sbi = ll_i2sbi(inode);
1204         ENTRY;
1205
1206         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1207                inode->i_generation, inode);
1208
1209         if (S_ISDIR(inode->i_mode)) {
1210                 /* these should have been cleared in ll_file_release */
1211                 LASSERT(lli->lli_opendir_key == NULL);
1212                 LASSERT(lli->lli_sai == NULL);
1213                 LASSERT(lli->lli_opendir_pid == 0);
1214         }
1215
1216         spin_lock(&lli->lli_lock);
1217         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1218         spin_unlock(&lli->lli_lock);
1219         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1220
1221         LASSERT(!lli->lli_open_fd_write_count);
1222         LASSERT(!lli->lli_open_fd_read_count);
1223         LASSERT(!lli->lli_open_fd_exec_count);
1224
1225         if (lli->lli_mds_write_och)
1226                 ll_md_real_close(inode, FMODE_WRITE);
1227         if (lli->lli_mds_exec_och)
1228                 ll_md_real_close(inode, FMODE_EXEC);
1229         if (lli->lli_mds_read_och)
1230                 ll_md_real_close(inode, FMODE_READ);
1231
1232         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1233                 OBD_FREE(lli->lli_symlink_name,
1234                          strlen(lli->lli_symlink_name) + 1);
1235                 lli->lli_symlink_name = NULL;
1236         }
1237
1238         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1239                 LASSERT(lli->lli_posix_acl == NULL);
1240                 if (lli->lli_remote_perms) {
1241                         free_rmtperm_hash(lli->lli_remote_perms);
1242                         lli->lli_remote_perms = NULL;
1243                 }
1244         }
1245 #ifdef CONFIG_FS_POSIX_ACL
1246         else if (lli->lli_posix_acl) {
1247                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1248                 LASSERT(lli->lli_remote_perms == NULL);
1249                 posix_acl_release(lli->lli_posix_acl);
1250                 lli->lli_posix_acl = NULL;
1251         }
1252 #endif
1253         lli->lli_inode_magic = LLI_INODE_DEAD;
1254
1255         ll_clear_inode_capas(inode);
1256         if (!S_ISDIR(inode->i_mode))
1257                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1258
1259         /*
1260          * XXX This has to be done before lsm is freed below, because
1261          * cl_object still uses inode lsm.
1262          */
1263         cl_inode_fini(inode);
1264         lli->lli_has_smd = false;
1265
1266         EXIT;
1267 }
1268
1269 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1270                   struct md_open_data **mod)
1271 {
1272         struct lustre_md md;
1273         struct inode *inode = dentry->d_inode;
1274         struct ll_sb_info *sbi = ll_i2sbi(inode);
1275         struct ptlrpc_request *request = NULL;
1276         int rc, ia_valid;
1277         ENTRY;
1278
1279         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1280                                      LUSTRE_OPC_ANY, NULL);
1281         if (IS_ERR(op_data))
1282                 RETURN(PTR_ERR(op_data));
1283
1284         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1285                         &request, mod);
1286         if (rc) {
1287                 ptlrpc_req_finished(request);
1288                 if (rc == -ENOENT) {
1289                         clear_nlink(inode);
1290                         /* Unlinked special device node? Or just a race?
1291                          * Pretend we done everything. */
1292                         if (!S_ISREG(inode->i_mode) &&
1293                             !S_ISDIR(inode->i_mode)) {
1294                                 ia_valid = op_data->op_attr.ia_valid;
1295                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1296                                 rc = simple_setattr(dentry, &op_data->op_attr);
1297                                 op_data->op_attr.ia_valid = ia_valid;
1298                         }
1299                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1300                         CERROR("md_setattr fails: rc = %d\n", rc);
1301                 }
1302                 RETURN(rc);
1303         }
1304
1305         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1306                               sbi->ll_md_exp, &md);
1307         if (rc) {
1308                 ptlrpc_req_finished(request);
1309                 RETURN(rc);
1310         }
1311
1312         ia_valid = op_data->op_attr.ia_valid;
1313         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1314          * cache is not cleared yet. */
1315         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1316         rc = simple_setattr(dentry, &op_data->op_attr);
1317         op_data->op_attr.ia_valid = ia_valid;
1318
1319         /* Extract epoch data if obtained. */
1320         op_data->op_handle = md.body->handle;
1321         op_data->op_ioepoch = md.body->ioepoch;
1322
1323         ll_update_inode(inode, &md);
1324         ptlrpc_req_finished(request);
1325
1326         RETURN(rc);
1327 }
1328
1329 /* Close IO epoch and send Size-on-MDS attribute update. */
1330 static int ll_setattr_done_writing(struct inode *inode,
1331                                    struct md_op_data *op_data,
1332                                    struct md_open_data *mod)
1333 {
1334         struct ll_inode_info *lli = ll_i2info(inode);
1335         int rc = 0;
1336         ENTRY;
1337
1338         LASSERT(op_data != NULL);
1339         if (!S_ISREG(inode->i_mode))
1340                 RETURN(0);
1341
1342         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1343                op_data->op_ioepoch, PFID(&lli->lli_fid));
1344
1345         op_data->op_flags = MF_EPOCH_CLOSE;
1346         ll_done_writing_attr(inode, op_data);
1347         ll_pack_inode2opdata(inode, op_data, NULL);
1348
1349         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1350         if (rc == -EAGAIN) {
1351                 /* MDS has instructed us to obtain Size-on-MDS attribute
1352                  * from OSTs and send setattr to back to MDS. */
1353                 rc = ll_som_update(inode, op_data);
1354         } else if (rc) {
1355                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1356                        inode->i_ino, rc);
1357         }
1358         RETURN(rc);
1359 }
1360
1361 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1362 {
1363         struct obd_capa *capa;
1364         int rc;
1365
1366         if (attr->ia_valid & ATTR_SIZE)
1367                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1368         else
1369                 capa = ll_mdscapa_get(inode);
1370
1371         rc = cl_setattr_ost(inode, attr, capa);
1372
1373         if (attr->ia_valid & ATTR_SIZE)
1374                 ll_truncate_free_capa(capa);
1375         else
1376                 capa_put(capa);
1377
1378         return rc;
1379 }
1380
1381 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1382 /**
1383  * inode_newsize_ok - may this inode be truncated to a given size
1384  * @inode:      the inode to be truncated
1385  * @offset:     the new size to assign to the inode
1386  * @Returns:    0 on success, -ve errno on failure
1387  *
1388  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1389  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1390  * when necessary. Caller must not proceed with inode size change if failure is
1391  * returned. @inode must be a file (not directory), with appropriate
1392  * permissions to allow truncate (inode_newsize_ok does NOT check these
1393  * conditions).
1394  *
1395  * inode_newsize_ok must be called with i_mutex held.
1396  */
1397 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1398 {
1399         if (inode->i_size < offset) {
1400                 unsigned long limit;
1401
1402                 limit = rlimit(RLIMIT_FSIZE);
1403                 if (limit != RLIM_INFINITY && offset > limit)
1404                         goto out_sig;
1405                 if (offset > inode->i_sb->s_maxbytes)
1406                         goto out_big;
1407         } else {
1408                 /*
1409                  * truncation of in-use swapfiles is disallowed - it would
1410                  * cause subsequent swapout to scribble on the now-freed
1411                  * blocks.
1412                  */
1413                 if (IS_SWAPFILE(inode))
1414                         return -ETXTBSY;
1415         }
1416
1417         return 0;
1418 out_sig:
1419         send_sig(SIGXFSZ, current, 0);
1420 out_big:
1421         return -EFBIG;
1422 }
1423 #endif
1424
1425 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1426  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1427  * keep these values until such a time that objects are allocated for it.
1428  * We do the MDS operations first, as it is checking permissions for us.
1429  * We don't to the MDS RPC if there is nothing that we want to store there,
1430  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1431  * going to do an RPC anyways.
1432  *
1433  * If we are doing a truncate, we will send the mtime and ctime updates
1434  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1435  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1436  * at the same time.
1437  */
1438 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1439 {
1440         struct inode *inode = dentry->d_inode;
1441         struct ll_inode_info *lli = ll_i2info(inode);
1442         struct md_op_data *op_data = NULL;
1443         struct md_open_data *mod = NULL;
1444         int rc = 0, rc1 = 0;
1445         ENTRY;
1446
1447         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1448                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1449                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1450                 attr->ia_valid);
1451
1452         if (attr->ia_valid & ATTR_SIZE) {
1453                 /* Check new size against VFS/VM file size limit and rlimit */
1454                 rc = inode_newsize_ok(inode, attr->ia_size);
1455                 if (rc)
1456                         RETURN(rc);
1457
1458                 /* The maximum Lustre file size is variable, based on the
1459                  * OST maximum object size and number of stripes.  This
1460                  * needs another check in addition to the VFS check above. */
1461                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1462                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1463                                PFID(&lli->lli_fid), attr->ia_size,
1464                                ll_file_maxbytes(inode));
1465                         RETURN(-EFBIG);
1466                 }
1467
1468                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1469         }
1470
1471         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1472         if (attr->ia_valid & TIMES_SET_FLAGS) {
1473                 if (cfs_curproc_fsuid() != inode->i_uid &&
1474                     !cfs_capable(CFS_CAP_FOWNER))
1475                         RETURN(-EPERM);
1476         }
1477
1478         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1479         if (attr->ia_valid & ATTR_CTIME) {
1480                 attr->ia_ctime = CFS_CURRENT_TIME;
1481                 attr->ia_valid |= ATTR_CTIME_SET;
1482         }
1483         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1484             (attr->ia_valid & ATTR_ATIME)) {
1485                 attr->ia_atime = CFS_CURRENT_TIME;
1486                 attr->ia_valid |= ATTR_ATIME_SET;
1487         }
1488         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1489             (attr->ia_valid & ATTR_MTIME)) {
1490                 attr->ia_mtime = CFS_CURRENT_TIME;
1491                 attr->ia_valid |= ATTR_MTIME_SET;
1492         }
1493
1494         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1495                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1496                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1497                        cfs_time_current_sec());
1498
1499         /* If we are changing file size, file content is modified, flag it. */
1500         if (attr->ia_valid & ATTR_SIZE) {
1501                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1502                 spin_lock(&lli->lli_lock);
1503                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1504                 spin_unlock(&lli->lli_lock);
1505         }
1506
1507         /* We always do an MDS RPC, even if we're only changing the size;
1508          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1509
1510         OBD_ALLOC_PTR(op_data);
1511         if (op_data == NULL)
1512                 RETURN(-ENOMEM);
1513
1514         if (!S_ISDIR(inode->i_mode)) {
1515                 if (attr->ia_valid & ATTR_SIZE)
1516                         inode_dio_write_done(inode);
1517                 mutex_unlock(&inode->i_mutex);
1518                 down_write(&lli->lli_trunc_sem);
1519         }
1520
1521         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1522
1523         /* Open epoch for truncate. */
1524         if (exp_connect_som(ll_i2mdexp(inode)) &&
1525             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1526                 op_data->op_flags = MF_EPOCH_OPEN;
1527
1528         rc = ll_md_setattr(dentry, op_data, &mod);
1529         if (rc)
1530                 GOTO(out, rc);
1531
1532         /* RPC to MDT is sent, cancel data modification flag */
1533         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1534                 spin_lock(&lli->lli_lock);
1535                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1536                 spin_unlock(&lli->lli_lock);
1537         }
1538
1539         ll_ioepoch_open(lli, op_data->op_ioepoch);
1540         if (!S_ISREG(inode->i_mode))
1541                 GOTO(out, rc = 0);
1542
1543         if (attr->ia_valid & (ATTR_SIZE |
1544                               ATTR_ATIME | ATTR_ATIME_SET |
1545                               ATTR_MTIME | ATTR_MTIME_SET))
1546                 /* For truncate and utimes sending attributes to OSTs, setting
1547                  * mtime/atime to the past will be performed under PW [0:EOF]
1548                  * extent lock (new_size:EOF for truncate).  It may seem
1549                  * excessive to send mtime/atime updates to OSTs when not
1550                  * setting times to past, but it is necessary due to possible
1551                  * time de-synchronization between MDT inode and OST objects */
1552                 rc = ll_setattr_ost(inode, attr);
1553         EXIT;
1554 out:
1555         if (op_data) {
1556                 if (op_data->op_ioepoch) {
1557                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1558                         if (!rc)
1559                                 rc = rc1;
1560                 }
1561                 ll_finish_md_op_data(op_data);
1562         }
1563         if (!S_ISDIR(inode->i_mode)) {
1564                 up_write(&lli->lli_trunc_sem);
1565                 mutex_lock(&inode->i_mutex);
1566                 if (attr->ia_valid & ATTR_SIZE)
1567                         inode_dio_wait(inode);
1568         }
1569
1570         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1571                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1572
1573         return rc;
1574 }
1575
1576 int ll_setattr(struct dentry *de, struct iattr *attr)
1577 {
1578         int mode = de->d_inode->i_mode;
1579
1580         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1581                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1582                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1583
1584         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1585                                (ATTR_SIZE|ATTR_MODE)) &&
1586             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1587              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1588               !(attr->ia_mode & S_ISGID))))
1589                 attr->ia_valid |= ATTR_FORCE;
1590
1591         if ((mode & S_ISUID) &&
1592             !(attr->ia_mode & S_ISUID) &&
1593             !(attr->ia_valid & ATTR_KILL_SUID))
1594                 attr->ia_valid |= ATTR_KILL_SUID;
1595
1596         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1597             !(attr->ia_mode & S_ISGID) &&
1598             !(attr->ia_valid & ATTR_KILL_SGID))
1599                 attr->ia_valid |= ATTR_KILL_SGID;
1600
1601         return ll_setattr_raw(de, attr);
1602 }
1603
1604 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1605                        __u64 max_age, __u32 flags)
1606 {
1607         struct ll_sb_info *sbi = ll_s2sbi(sb);
1608         struct obd_statfs obd_osfs;
1609         int rc;
1610         ENTRY;
1611
1612         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1613         if (rc) {
1614                 CERROR("md_statfs fails: rc = %d\n", rc);
1615                 RETURN(rc);
1616         }
1617
1618         osfs->os_type = sb->s_magic;
1619
1620         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1621                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1622
1623         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1624                 flags |= OBD_STATFS_NODELAY;
1625
1626         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1627         if (rc) {
1628                 CERROR("obd_statfs fails: rc = %d\n", rc);
1629                 RETURN(rc);
1630         }
1631
1632         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1633                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1634                obd_osfs.os_files);
1635
1636         osfs->os_bsize = obd_osfs.os_bsize;
1637         osfs->os_blocks = obd_osfs.os_blocks;
1638         osfs->os_bfree = obd_osfs.os_bfree;
1639         osfs->os_bavail = obd_osfs.os_bavail;
1640
1641         /* If we don't have as many objects free on the OST as inodes
1642          * on the MDS, we reduce the total number of inodes to
1643          * compensate, so that the "inodes in use" number is correct.
1644          */
1645         if (obd_osfs.os_ffree < osfs->os_ffree) {
1646                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1647                         obd_osfs.os_ffree;
1648                 osfs->os_ffree = obd_osfs.os_ffree;
1649         }
1650
1651         RETURN(rc);
1652 }
1653 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1654 {
1655         struct super_block *sb = de->d_sb;
1656         struct obd_statfs osfs;
1657         int rc;
1658
1659         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1660         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1661
1662         /* Some amount of caching on the client is allowed */
1663         rc = ll_statfs_internal(sb, &osfs,
1664                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1665                                 0);
1666         if (rc)
1667                 return rc;
1668
1669         statfs_unpack(sfs, &osfs);
1670
1671         /* We need to downshift for all 32-bit kernels, because we can't
1672          * tell if the kernel is being called via sys_statfs64() or not.
1673          * Stop before overflowing f_bsize - in which case it is better
1674          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1675         if (sizeof(long) < 8) {
1676                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1677                         sfs->f_bsize <<= 1;
1678
1679                         osfs.os_blocks >>= 1;
1680                         osfs.os_bfree >>= 1;
1681                         osfs.os_bavail >>= 1;
1682                 }
1683         }
1684
1685         sfs->f_blocks = osfs.os_blocks;
1686         sfs->f_bfree = osfs.os_bfree;
1687         sfs->f_bavail = osfs.os_bavail;
1688         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1689         return 0;
1690 }
1691
1692 void ll_inode_size_lock(struct inode *inode)
1693 {
1694         struct ll_inode_info *lli;
1695
1696         LASSERT(!S_ISDIR(inode->i_mode));
1697
1698         lli = ll_i2info(inode);
1699         LASSERT(lli->lli_size_sem_owner != current);
1700         down(&lli->lli_size_sem);
1701         LASSERT(lli->lli_size_sem_owner == NULL);
1702         lli->lli_size_sem_owner = current;
1703 }
1704
1705 void ll_inode_size_unlock(struct inode *inode)
1706 {
1707         struct ll_inode_info *lli;
1708
1709         lli = ll_i2info(inode);
1710         LASSERT(lli->lli_size_sem_owner == current);
1711         lli->lli_size_sem_owner = NULL;
1712         up(&lli->lli_size_sem);
1713 }
1714
1715 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1716 {
1717         struct ll_inode_info *lli = ll_i2info(inode);
1718         struct mdt_body *body = md->body;
1719         struct lov_stripe_md *lsm = md->lsm;
1720         struct ll_sb_info *sbi = ll_i2sbi(inode);
1721
1722         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1723         if (lsm != NULL) {
1724                 if (!lli->lli_has_smd &&
1725                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1726                         cl_file_inode_init(inode, md);
1727
1728                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1729                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1730                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1731         }
1732
1733         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1734                 if (body->valid & OBD_MD_FLRMTPERM)
1735                         ll_update_remote_perm(inode, md->remote_perm);
1736         }
1737 #ifdef CONFIG_FS_POSIX_ACL
1738         else if (body->valid & OBD_MD_FLACL) {
1739                 spin_lock(&lli->lli_lock);
1740                 if (lli->lli_posix_acl)
1741                         posix_acl_release(lli->lli_posix_acl);
1742                 lli->lli_posix_acl = md->posix_acl;
1743                 spin_unlock(&lli->lli_lock);
1744         }
1745 #endif
1746         inode->i_ino = cl_fid_build_ino(&body->fid1,
1747                                         sbi->ll_flags & LL_SBI_32BIT_API);
1748         inode->i_generation = cl_fid_build_gen(&body->fid1);
1749
1750         if (body->valid & OBD_MD_FLATIME) {
1751                 if (body->atime > LTIME_S(inode->i_atime))
1752                         LTIME_S(inode->i_atime) = body->atime;
1753                 lli->lli_lvb.lvb_atime = body->atime;
1754         }
1755         if (body->valid & OBD_MD_FLMTIME) {
1756                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1757                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1758                                "to "LPU64"\n", inode->i_ino,
1759                                LTIME_S(inode->i_mtime), body->mtime);
1760                         LTIME_S(inode->i_mtime) = body->mtime;
1761                 }
1762                 lli->lli_lvb.lvb_mtime = body->mtime;
1763         }
1764         if (body->valid & OBD_MD_FLCTIME) {
1765                 if (body->ctime > LTIME_S(inode->i_ctime))
1766                         LTIME_S(inode->i_ctime) = body->ctime;
1767                 lli->lli_lvb.lvb_ctime = body->ctime;
1768         }
1769         if (body->valid & OBD_MD_FLMODE)
1770                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1771         if (body->valid & OBD_MD_FLTYPE)
1772                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1773         LASSERT(inode->i_mode != 0);
1774         if (S_ISREG(inode->i_mode)) {
1775                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1776         } else {
1777                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1778         }
1779         if (body->valid & OBD_MD_FLUID)
1780                 inode->i_uid = body->uid;
1781         if (body->valid & OBD_MD_FLGID)
1782                 inode->i_gid = body->gid;
1783         if (body->valid & OBD_MD_FLFLAGS)
1784                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1785         if (body->valid & OBD_MD_FLNLINK)
1786                 set_nlink(inode, body->nlink);
1787         if (body->valid & OBD_MD_FLRDEV)
1788                 inode->i_rdev = old_decode_dev(body->rdev);
1789
1790         if (body->valid & OBD_MD_FLID) {
1791                 /* FID shouldn't be changed! */
1792                 if (fid_is_sane(&lli->lli_fid)) {
1793                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1794                                  "Trying to change FID "DFID
1795                                  " to the "DFID", inode %lu/%u(%p)\n",
1796                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1797                                  inode->i_ino, inode->i_generation, inode);
1798                 } else
1799                         lli->lli_fid = body->fid1;
1800         }
1801
1802         LASSERT(fid_seq(&lli->lli_fid) != 0);
1803
1804         if (body->valid & OBD_MD_FLSIZE) {
1805                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1806                     S_ISREG(inode->i_mode)) {
1807                         struct lustre_handle lockh;
1808                         ldlm_mode_t mode;
1809
1810                         /* As it is possible a blocking ast has been processed
1811                          * by this time, we need to check there is an UPDATE
1812                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1813                          * it. */
1814                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1815                                                &lockh, LDLM_FL_CBPENDING);
1816                         if (mode) {
1817                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1818                                                       LLIF_EPOCH_PENDING |
1819                                                       LLIF_SOM_DIRTY)) {
1820                                         CERROR("ino %lu flags %u still has "
1821                                                "size authority! do not trust "
1822                                                "the size got from MDS\n",
1823                                                inode->i_ino, lli->lli_flags);
1824                                 } else {
1825                                         /* Use old size assignment to avoid
1826                                          * deadlock bz14138 & bz14326 */
1827                                         i_size_write(inode, body->size);
1828                                         spin_lock(&lli->lli_lock);
1829                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1830                                         spin_unlock(&lli->lli_lock);
1831                                 }
1832                                 ldlm_lock_decref(&lockh, mode);
1833                         }
1834                 } else {
1835                         /* Use old size assignment to avoid
1836                          * deadlock bz14138 & bz14326 */
1837                         i_size_write(inode, body->size);
1838
1839                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1840                                inode->i_ino, (unsigned long long)body->size);
1841                 }
1842
1843                 if (body->valid & OBD_MD_FLBLOCKS)
1844                         inode->i_blocks = body->blocks;
1845         }
1846
1847         if (body->valid & OBD_MD_FLMDSCAPA) {
1848                 LASSERT(md->mds_capa);
1849                 ll_add_capa(inode, md->mds_capa);
1850         }
1851         if (body->valid & OBD_MD_FLOSSCAPA) {
1852                 LASSERT(md->oss_capa);
1853                 ll_add_capa(inode, md->oss_capa);
1854         }
1855 }
1856
1857 void ll_read_inode2(struct inode *inode, void *opaque)
1858 {
1859         struct lustre_md *md = opaque;
1860         struct ll_inode_info *lli = ll_i2info(inode);
1861         ENTRY;
1862
1863         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1864                PFID(&lli->lli_fid), inode);
1865
1866         LASSERT(!lli->lli_has_smd);
1867
1868         /* Core attributes from the MDS first.  This is a new inode, and
1869          * the VFS doesn't zero times in the core inode so we have to do
1870          * it ourselves.  They will be overwritten by either MDS or OST
1871          * attributes - we just need to make sure they aren't newer. */
1872         LTIME_S(inode->i_mtime) = 0;
1873         LTIME_S(inode->i_atime) = 0;
1874         LTIME_S(inode->i_ctime) = 0;
1875         inode->i_rdev = 0;
1876         ll_update_inode(inode, md);
1877
1878         /* OIDEBUG(inode); */
1879
1880         /* initializing backing dev info. */
1881         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1882
1883
1884         if (S_ISREG(inode->i_mode)) {
1885                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1886                 inode->i_op = &ll_file_inode_operations;
1887                 inode->i_fop = sbi->ll_fop;
1888                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1889                 EXIT;
1890         } else if (S_ISDIR(inode->i_mode)) {
1891                 inode->i_op = &ll_dir_inode_operations;
1892                 inode->i_fop = &ll_dir_operations;
1893                 EXIT;
1894         } else if (S_ISLNK(inode->i_mode)) {
1895                 inode->i_op = &ll_fast_symlink_inode_operations;
1896                 EXIT;
1897         } else {
1898                 inode->i_op = &ll_special_inode_operations;
1899
1900                 init_special_inode(inode, inode->i_mode,
1901                                    inode->i_rdev);
1902
1903                 EXIT;
1904         }
1905 }
1906
1907 void ll_delete_inode(struct inode *inode)
1908 {
1909         struct cl_inode_info *lli = cl_i2info(inode);
1910         ENTRY;
1911
1912         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1913                 /* discard all dirty pages before truncating them, required by
1914                  * osc_extent implementation at LU-1030. */
1915                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1916                                    CL_FSYNC_DISCARD, 1);
1917
1918         truncate_inode_pages(&inode->i_data, 0);
1919
1920         /* Workaround for LU-118 */
1921         if (inode->i_data.nrpages) {
1922                 TREE_READ_LOCK_IRQ(&inode->i_data);
1923                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1924                 LASSERTF(inode->i_data.nrpages == 0,
1925                          "inode=%lu/%u(%p) nrpages=%lu, see "
1926                          "http://jira.whamcloud.com/browse/LU-118\n",
1927                          inode->i_ino, inode->i_generation, inode,
1928                          inode->i_data.nrpages);
1929         }
1930         /* Workaround end */
1931
1932 #ifdef HAVE_SBOPS_EVICT_INODE
1933         ll_clear_inode(inode);
1934 #endif
1935         clear_inode(inode);
1936
1937         EXIT;
1938 }
1939
1940 int ll_iocontrol(struct inode *inode, struct file *file,
1941                  unsigned int cmd, unsigned long arg)
1942 {
1943         struct ll_sb_info *sbi = ll_i2sbi(inode);
1944         struct ptlrpc_request *req = NULL;
1945         int rc, flags = 0;
1946         ENTRY;
1947
1948         switch(cmd) {
1949         case FSFILT_IOC_GETFLAGS: {
1950                 struct mdt_body *body;
1951                 struct md_op_data *op_data;
1952
1953                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1954                                              0, 0, LUSTRE_OPC_ANY,
1955                                              NULL);
1956                 if (IS_ERR(op_data))
1957                         RETURN(PTR_ERR(op_data));
1958
1959                 op_data->op_valid = OBD_MD_FLFLAGS;
1960                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1961                 ll_finish_md_op_data(op_data);
1962                 if (rc) {
1963                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1964                         RETURN(-abs(rc));
1965                 }
1966
1967                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1968
1969                 flags = body->flags;
1970
1971                 ptlrpc_req_finished(req);
1972
1973                 RETURN(put_user(flags, (int *)arg));
1974         }
1975         case FSFILT_IOC_SETFLAGS: {
1976                 struct lov_stripe_md *lsm;
1977                 struct obd_info oinfo = { { { 0 } } };
1978                 struct md_op_data *op_data;
1979
1980                 if (get_user(flags, (int *)arg))
1981                         RETURN(-EFAULT);
1982
1983                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1984                                              LUSTRE_OPC_ANY, NULL);
1985                 if (IS_ERR(op_data))
1986                         RETURN(PTR_ERR(op_data));
1987
1988                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1989                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1990                 rc = md_setattr(sbi->ll_md_exp, op_data,
1991                                 NULL, 0, NULL, 0, &req, NULL);
1992                 ll_finish_md_op_data(op_data);
1993                 ptlrpc_req_finished(req);
1994                 if (rc)
1995                         RETURN(rc);
1996
1997                 inode->i_flags = ll_ext_to_inode_flags(flags);
1998
1999                 lsm = ccc_inode_lsm_get(inode);
2000                 if (!lsm_has_objects(lsm)) {
2001                         ccc_inode_lsm_put(inode, lsm);
2002                         RETURN(0);
2003                 }
2004
2005                 OBDO_ALLOC(oinfo.oi_oa);
2006                 if (!oinfo.oi_oa) {
2007                         ccc_inode_lsm_put(inode, lsm);
2008                         RETURN(-ENOMEM);
2009                 }
2010                 oinfo.oi_md = lsm;
2011                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2012                 oinfo.oi_oa->o_flags = flags;
2013                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2014                                        OBD_MD_FLGROUP;
2015                 oinfo.oi_capa = ll_mdscapa_get(inode);
2016                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2017                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2018                 capa_put(oinfo.oi_capa);
2019                 OBDO_FREE(oinfo.oi_oa);
2020                 ccc_inode_lsm_put(inode, lsm);
2021
2022                 if (rc && rc != -EPERM && rc != -EACCES)
2023                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2024
2025                 RETURN(rc);
2026         }
2027         default:
2028                 RETURN(-ENOSYS);
2029         }
2030
2031         RETURN(0);
2032 }
2033
2034 int ll_flush_ctx(struct inode *inode)
2035 {
2036         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2037
2038         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2039
2040         obd_set_info_async(NULL, sbi->ll_md_exp,
2041                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2042                            0, NULL, NULL);
2043         obd_set_info_async(NULL, sbi->ll_dt_exp,
2044                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2045                            0, NULL, NULL);
2046         return 0;
2047 }
2048
2049 /* umount -f client means force down, don't save state */
2050 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2051 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2052 {
2053         struct super_block *sb = vfsmnt->mnt_sb;
2054 #else
2055 void ll_umount_begin(struct super_block *sb)
2056 {
2057 #endif
2058         struct ll_sb_info *sbi = ll_s2sbi(sb);
2059         struct obd_device *obd;
2060         struct obd_ioctl_data *ioc_data;
2061         ENTRY;
2062
2063 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2064         if (!(flags & MNT_FORCE)) {
2065                 EXIT;
2066                 return;
2067         }
2068 #endif
2069
2070         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2071                sb->s_count, atomic_read(&sb->s_active));
2072
2073         obd = class_exp2obd(sbi->ll_md_exp);
2074         if (obd == NULL) {
2075                 CERROR("Invalid MDC connection handle "LPX64"\n",
2076                        sbi->ll_md_exp->exp_handle.h_cookie);
2077                 EXIT;
2078                 return;
2079         }
2080         obd->obd_force = 1;
2081
2082         obd = class_exp2obd(sbi->ll_dt_exp);
2083         if (obd == NULL) {
2084                 CERROR("Invalid LOV connection handle "LPX64"\n",
2085                        sbi->ll_dt_exp->exp_handle.h_cookie);
2086                 EXIT;
2087                 return;
2088         }
2089         obd->obd_force = 1;
2090
2091         OBD_ALLOC_PTR(ioc_data);
2092         if (ioc_data) {
2093                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2094                               sizeof *ioc_data, ioc_data, NULL);
2095
2096                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2097                               sizeof *ioc_data, ioc_data, NULL);
2098
2099                 OBD_FREE_PTR(ioc_data);
2100         }
2101
2102         /* Really, we'd like to wait until there are no requests outstanding,
2103          * and then continue.  For now, we just invalidate the requests,
2104          * schedule() and sleep one second if needed, and hope.
2105          */
2106         cfs_schedule();
2107 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2108         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2109                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2110                                                    cfs_time_seconds(1));
2111                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2112                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2113                                       "may try to umount it a bit later\n",
2114                                       atomic_read(&vfsmnt->mnt_count));
2115         }
2116 #endif
2117
2118         EXIT;
2119 }
2120
2121 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2122 {
2123         struct ll_sb_info *sbi = ll_s2sbi(sb);
2124         char *profilenm = get_profile_name(sb);
2125         int err;
2126         __u32 read_only;
2127
2128         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2129                 read_only = *flags & MS_RDONLY;
2130                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2131                                          sizeof(KEY_READ_ONLY),
2132                                          KEY_READ_ONLY, sizeof(read_only),
2133                                          &read_only, NULL);
2134                 if (err) {
2135                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2136                                       profilenm, read_only ?
2137                                       "read-only" : "read-write", err);
2138                         return err;
2139                 }
2140
2141                 if (read_only)
2142                         sb->s_flags |= MS_RDONLY;
2143                 else
2144                         sb->s_flags &= ~MS_RDONLY;
2145
2146                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2147                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2148                                       read_only ?  "read-only" : "read-write");
2149         }
2150         return 0;
2151 }
2152
2153 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2154                   struct super_block *sb, struct lookup_intent *it)
2155 {
2156         struct ll_sb_info *sbi = NULL;
2157         struct lustre_md md;
2158         int rc;
2159         ENTRY;
2160
2161         LASSERT(*inode || sb);
2162         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2163         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2164                               sbi->ll_md_exp, &md);
2165         if (rc)
2166                 RETURN(rc);
2167
2168         if (*inode) {
2169                 ll_update_inode(*inode, &md);
2170         } else {
2171                 LASSERT(sb != NULL);
2172
2173                 /*
2174                  * At this point server returns to client's same fid as client
2175                  * generated for creating. So using ->fid1 is okay here.
2176                  */
2177                 LASSERT(fid_is_sane(&md.body->fid1));
2178
2179                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2180                                              sbi->ll_flags & LL_SBI_32BIT_API),
2181                                  &md);
2182                 if (*inode == NULL || IS_ERR(*inode)) {
2183 #ifdef CONFIG_FS_POSIX_ACL
2184                         if (md.posix_acl) {
2185                                 posix_acl_release(md.posix_acl);
2186                                 md.posix_acl = NULL;
2187                         }
2188 #endif
2189                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2190                         *inode = NULL;
2191                         CERROR("new_inode -fatal: rc %d\n", rc);
2192                         GOTO(out, rc);
2193                 }
2194         }
2195
2196         /* Handling piggyback layout lock.
2197          * Layout lock can be piggybacked by getattr and open request.
2198          * The lsm can be applied to inode only if it comes with a layout lock
2199          * otherwise correct layout may be overwritten, for example:
2200          * 1. proc1: mdt returns a lsm but not granting layout
2201          * 2. layout was changed by another client
2202          * 3. proc2: refresh layout and layout lock granted
2203          * 4. proc1: to apply a stale layout */
2204         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2205                 struct lustre_handle lockh;
2206                 struct ldlm_lock *lock;
2207
2208                 lockh.cookie = it->d.lustre.it_lock_handle;
2209                 lock = ldlm_handle2lock(&lockh);
2210                 LASSERT(lock != NULL);
2211                 if (ldlm_has_layout(lock)) {
2212                         struct cl_object_conf conf;
2213
2214                         memset(&conf, 0, sizeof(conf));
2215                         conf.coc_opc = OBJECT_CONF_SET;
2216                         conf.coc_inode = *inode;
2217                         conf.coc_lock = lock;
2218                         conf.u.coc_md = &md;
2219                         (void)ll_layout_conf(*inode, &conf);
2220                 }
2221                 LDLM_LOCK_PUT(lock);
2222         }
2223
2224 out:
2225         if (md.lsm != NULL)
2226                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2227         md_free_lustre_md(sbi->ll_md_exp, &md);
2228         RETURN(rc);
2229 }
2230
2231 int ll_obd_statfs(struct inode *inode, void *arg)
2232 {
2233         struct ll_sb_info *sbi = NULL;
2234         struct obd_export *exp;
2235         char *buf = NULL;
2236         struct obd_ioctl_data *data = NULL;
2237         __u32 type;
2238         __u32 flags;
2239         int len = 0, rc;
2240
2241         if (!inode || !(sbi = ll_i2sbi(inode)))
2242                 GOTO(out_statfs, rc = -EINVAL);
2243
2244         rc = obd_ioctl_getdata(&buf, &len, arg);
2245         if (rc)
2246                 GOTO(out_statfs, rc);
2247
2248         data = (void*)buf;
2249         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2250             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2251                 GOTO(out_statfs, rc = -EINVAL);
2252
2253         if (data->ioc_inllen1 != sizeof(__u32) ||
2254             data->ioc_inllen2 != sizeof(__u32) ||
2255             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2256             data->ioc_plen2 != sizeof(struct obd_uuid))
2257                 GOTO(out_statfs, rc = -EINVAL);
2258
2259         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2260         if (type & LL_STATFS_LMV)
2261                 exp = sbi->ll_md_exp;
2262         else if (type & LL_STATFS_LOV)
2263                 exp = sbi->ll_dt_exp;
2264         else
2265                 GOTO(out_statfs, rc = -ENODEV);
2266
2267         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2268         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2269         if (rc)
2270                 GOTO(out_statfs, rc);
2271 out_statfs:
2272         if (buf)
2273                 obd_ioctl_freedata(buf, len);
2274         return rc;
2275 }
2276
2277 int ll_process_config(struct lustre_cfg *lcfg)
2278 {
2279         char *ptr;
2280         void *sb;
2281         struct lprocfs_static_vars lvars;
2282         unsigned long x;
2283         int rc = 0;
2284
2285         lprocfs_llite_init_vars(&lvars);
2286
2287         /* The instance name contains the sb: lustre-client-aacfe000 */
2288         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2289         if (!ptr || !*(++ptr))
2290                 return -EINVAL;
2291         if (sscanf(ptr, "%lx", &x) != 1)
2292                 return -EINVAL;
2293         sb = (void *)x;
2294         /* This better be a real Lustre superblock! */
2295         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2296
2297         /* Note we have not called client_common_fill_super yet, so
2298            proc fns must be able to handle that! */
2299         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2300                                       lcfg, sb);
2301         if (rc > 0)
2302                 rc = 0;
2303         return(rc);
2304 }
2305
2306 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2307 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2308                                        struct inode *i1, struct inode *i2,
2309                                        const char *name, int namelen,
2310                                        int mode, __u32 opc, void *data)
2311 {
2312         LASSERT(i1 != NULL);
2313
2314         if (namelen > ll_i2sbi(i1)->ll_namelen)
2315                 return ERR_PTR(-ENAMETOOLONG);
2316
2317         if (op_data == NULL)
2318                 OBD_ALLOC_PTR(op_data);
2319
2320         if (op_data == NULL)
2321                 return ERR_PTR(-ENOMEM);
2322
2323         ll_i2gids(op_data->op_suppgids, i1, i2);
2324         op_data->op_fid1 = *ll_inode2fid(i1);
2325         op_data->op_capa1 = ll_mdscapa_get(i1);
2326
2327         if (i2) {
2328                 op_data->op_fid2 = *ll_inode2fid(i2);
2329                 op_data->op_capa2 = ll_mdscapa_get(i2);
2330         } else {
2331                 fid_zero(&op_data->op_fid2);
2332                 op_data->op_capa2 = NULL;
2333         }
2334
2335         op_data->op_name = name;
2336         op_data->op_namelen = namelen;
2337         op_data->op_mode = mode;
2338         op_data->op_mod_time = cfs_time_current_sec();
2339         op_data->op_fsuid = cfs_curproc_fsuid();
2340         op_data->op_fsgid = cfs_curproc_fsgid();
2341         op_data->op_cap = cfs_curproc_cap_pack();
2342         op_data->op_bias = 0;
2343         op_data->op_cli_flags = 0;
2344         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2345              filename_is_volatile(name, namelen, NULL))
2346                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2347         op_data->op_opc = opc;
2348         op_data->op_mds = 0;
2349         op_data->op_data = data;
2350
2351         /* If the file is being opened after mknod() (normally due to NFS)
2352          * try to use the default stripe data from parent directory for
2353          * allocating OST objects.  Try to pass the parent FID to MDS. */
2354         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2355             !ll_i2info(i2)->lli_has_smd) {
2356                 struct ll_inode_info *lli = ll_i2info(i2);
2357
2358                 spin_lock(&lli->lli_lock);
2359                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2360                         op_data->op_fid1 = lli->lli_pfid;
2361                 spin_unlock(&lli->lli_lock);
2362                 /** We ignore parent's capability temporary. */
2363         }
2364
2365         /* When called by ll_setattr_raw, file is i1. */
2366         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2367                 op_data->op_bias |= MDS_DATA_MODIFIED;
2368
2369         return op_data;
2370 }
2371
2372 void ll_finish_md_op_data(struct md_op_data *op_data)
2373 {
2374         capa_put(op_data->op_capa1);
2375         capa_put(op_data->op_capa2);
2376         OBD_FREE_PTR(op_data);
2377 }
2378
2379 #ifdef HAVE_SUPEROPS_USE_DENTRY
2380 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2381 #else
2382 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2383 #endif
2384 {
2385         struct ll_sb_info *sbi;
2386
2387 #ifdef HAVE_SUPEROPS_USE_DENTRY
2388         LASSERT((seq != NULL) && (dentry != NULL));
2389         sbi = ll_s2sbi(dentry->d_sb);
2390 #else
2391         LASSERT((seq != NULL) && (vfs != NULL));
2392         sbi = ll_s2sbi(vfs->mnt_sb);
2393 #endif
2394
2395         if (sbi->ll_flags & LL_SBI_NOLCK)
2396                 seq_puts(seq, ",nolock");
2397
2398         if (sbi->ll_flags & LL_SBI_FLOCK)
2399                 seq_puts(seq, ",flock");
2400
2401         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2402                 seq_puts(seq, ",localflock");
2403
2404         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2405                 seq_puts(seq, ",user_xattr");
2406
2407         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2408                 seq_puts(seq, ",lazystatfs");
2409
2410         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2411                 seq_puts(seq, ",user_fid2path");
2412
2413         RETURN(0);
2414 }
2415
2416 /**
2417  * Get obd name by cmd, and copy out to user space
2418  */
2419 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2420 {
2421         struct ll_sb_info *sbi = ll_i2sbi(inode);
2422         struct obd_device *obd;
2423         ENTRY;
2424
2425         if (cmd == OBD_IOC_GETDTNAME)
2426                 obd = class_exp2obd(sbi->ll_dt_exp);
2427         else if (cmd == OBD_IOC_GETMDNAME)
2428                 obd = class_exp2obd(sbi->ll_md_exp);
2429         else
2430                 RETURN(-EINVAL);
2431
2432         if (!obd)
2433                 RETURN(-ENOENT);
2434
2435         if (copy_to_user((void *)arg, obd->obd_name,
2436                          strlen(obd->obd_name) + 1))
2437                 RETURN(-EFAULT);
2438
2439         RETURN(0);
2440 }
2441
2442 /**
2443  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2444  * fsname will be returned in this buffer; otherwise, a static buffer will be
2445  * used to store the fsname and returned to caller.
2446  */
2447 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2448 {
2449         static char fsname_static[MTI_NAME_MAXLEN];
2450         struct lustre_sb_info *lsi = s2lsi(sb);
2451         char *ptr;
2452         int len;
2453
2454         if (buf == NULL) {
2455                 /* this means the caller wants to use static buffer
2456                  * and it doesn't care about race. Usually this is
2457                  * in error reporting path */
2458                 buf = fsname_static;
2459                 buflen = sizeof(fsname_static);
2460         }
2461
2462         len = strlen(lsi->lsi_lmd->lmd_profile);
2463         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2464         if (ptr && (strcmp(ptr, "-client") == 0))
2465                 len -= 7;
2466
2467         if (unlikely(len >= buflen))
2468                 len = buflen - 1;
2469         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2470         buf[len] = '\0';
2471
2472         return buf;
2473 }
2474
2475 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2476 {
2477         char *path = NULL;
2478
2479 #ifdef HAVE_FS_STRUCT_USE_PATH
2480         struct path p;
2481
2482         p.dentry = dentry;
2483         p.mnt = current->fs->root.mnt;
2484         path_get(&p);
2485         path = d_path(&p, buf, bufsize);
2486         path_put(&p);
2487 #else
2488         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2489 #endif
2490
2491         return path;
2492 }
2493
2494 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2495 {
2496         char *buf, *path = NULL;
2497         struct dentry *dentry = NULL;
2498         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2499
2500         /* this can be called inside spin lock so use GFP_ATOMIC. */
2501         buf = (char *)__get_free_page(GFP_ATOMIC);
2502         if (buf != NULL) {
2503                 dentry = d_find_alias(page->mapping->host);
2504                 if (dentry != NULL)
2505                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2506         }
2507
2508         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2509               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2510               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2511               PFID(&obj->cob_header.coh_lu.loh_fid),
2512               (path && !IS_ERR(path)) ? path : "", ioret);
2513
2514         if (dentry != NULL)
2515                 dput(dentry);
2516
2517         if (buf != NULL)
2518                 free_page((unsigned long)buf);
2519 }