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         lru_page_max = pages / 2;
  97
  98         /* initialize ll_cache data */
  99         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 100         sbi->ll_cache.ccc_lru_max = lru_page_max;
 101         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 102         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 103         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 104
 105         cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 106         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 107
 108         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 109                                            SBI_DEFAULT_READAHEAD_MAX);
 110         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 111         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 112                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 113         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 114         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 115
 116         ll_generate_random_uuid(uuid);
 117         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 118         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 119
 120         spin_lock(&ll_sb_lock);
 121         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 122         spin_unlock(&ll_sb_lock);
 123
 124         sbi->ll_flags |= LL_SBI_VERBOSE;
 125 #ifdef ENABLE_CHECKSUM
 126         sbi->ll_flags |= LL_SBI_CHECKSUM;
 127 #endif
 128
 129 #ifdef HAVE_LRU_RESIZE_SUPPORT
 130         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 131 #endif
 132
 133         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 134                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 135                                pp_r_hist.oh_lock);
 136                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 137                                pp_w_hist.oh_lock);
 138         }
 139
 140         /* metadata statahead is enabled by default */
 141         sbi->ll_sa_max = LL_SA_RPC_DEF;
 142         cfs_atomic_set(&sbi->ll_sa_total, 0);
 143         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 144         cfs_atomic_set(&sbi->ll_agl_total, 0);
 145         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 146
 147         RETURN(sbi);
 148 }
 149
 150 void ll_free_sbi(struct super_block *sb)
 151 {
 152         struct ll_sb_info *sbi = ll_s2sbi(sb);
 153         ENTRY;
 154
 155         if (sbi != NULL) {
 156                 spin_lock(&ll_sb_lock);
 157                 cfs_list_del(&sbi->ll_list);
 158                 spin_unlock(&ll_sb_lock);
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_capa *oc = NULL;
 171         struct obd_statfs *osfs = NULL;
 172         struct ptlrpc_request *request = NULL;
 173         struct obd_connect_data *data = NULL;
 174         struct obd_uuid *uuid;
 175         struct md_op_data *op_data;
 176         struct lustre_md lmd;
 177         obd_valid valid;
 178         int size, err, checksum;
 179         ENTRY;
 180
 181         obd = class_name2obd(md);
 182         if (!obd) {
 183                 CERROR("MD %s: not setup or attached\n", md);
 184                 RETURN(-EINVAL);
 185         }
 186
 187         OBD_ALLOC_PTR(data);
 188         if (data == NULL)
 189                 RETURN(-ENOMEM);
 190
 191         OBD_ALLOC_PTR(osfs);
 192         if (osfs == NULL) {
 193                 OBD_FREE_PTR(data);
 194                 RETURN(-ENOMEM);
 195         }
 196
 197         /* indicate the features supported by this client */
 198         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 199                                   OBD_CONNECT_ATTRFID  |
 200                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 201                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 202                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 203                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 204                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 205                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 206                                   OBD_CONNECT_EINPROGRESS |
 207                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 208                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 209                                   OBD_CONNECT_MAX_EASIZE |
 210                                   OBD_CONNECT_FLOCK_DEAD |
 211                                   OBD_CONNECT_DISP_STRIPE;
 212
 213         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 214                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 215
 216 #ifdef HAVE_LRU_RESIZE_SUPPORT
 217         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 218                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 219 #endif
 220 #ifdef CONFIG_FS_POSIX_ACL
 221         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 222 #endif
 223
 224         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 225                 /* flag mdc connection as lightweight, only used for test
 226                  * purpose, use with care */
 227                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 228
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237 #ifdef HAVE_MS_FLOCK_LOCK
 238         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 239         sb->s_flags |= MS_FLOCK_LOCK;
 240 #endif
 241 #ifdef MS_HAS_NEW_AOPS
 242         sb->s_flags |= MS_HAS_NEW_AOPS;
 243 #endif
 244
 245         if (sbi->ll_flags & LL_SBI_FLOCK)
 246                 sbi->ll_fop = &ll_file_operations_flock;
 247         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 248                 sbi->ll_fop = &ll_file_operations;
 249         else
 250                 sbi->ll_fop = &ll_file_operations_noflock;
 251
 252         /* real client */
 253         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 254         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 255                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 256
 257         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 258
 259         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 260         if (err == -EBUSY) {
 261                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 262                                    "recovery, of which this client is not a "
 263                                    "part. Please wait for recovery to complete,"
 264                                    " abort, or time out.\n", md);
 265                 GOTO(out, err);
 266         } else if (err) {
 267                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 268                 GOTO(out, err);
 269         }
 270
 271         sbi->ll_md_exp->exp_connect_data = *data;
 272
 273         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 274                            LUSTRE_SEQ_METADATA);
 275         if (err) {
 276                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 277                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 278                 GOTO(out_md, err);
 279         }
 280
 281         /* For mount, we only need fs info from MDT0, and also in DNE, it
 282          * can make sure the client can be mounted as long as MDT0 is
 283          * avaible */
 284         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 285                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 286                         OBD_STATFS_FOR_MDT0);
 287         if (err)
 288                 GOTO(out_md_fid, err);
 289
 290         /* This needs to be after statfs to ensure connect has finished.
 291          * Note that "data" does NOT contain the valid connect reply.
 292          * If connecting to a 1.8 server there will be no LMV device, so
 293          * we can access the MDC export directly and exp_connect_flags will
 294          * be non-zero, but if accessing an upgraded 2.1 server it will
 295          * have the correct flags filled in.
 296          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 297         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 298         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 299             valid != CLIENT_CONNECT_MDT_REQD) {
 300                 char *buf;
 301
 302                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 303                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 304                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 305                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 306                                    "feature(s) needed for correct operation "
 307                                    "of this client (%s). Please upgrade "
 308                                    "server or downgrade client.\n",
 309                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 310                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 311                 GOTO(out_md_fid, err = -EPROTO);
 312         }
 313
 314         size = sizeof(*data);
 315         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 316                            KEY_CONN_DATA,  &size, data, NULL);
 317         if (err) {
 318                 CERROR("%s: Get connect data failed: rc = %d\n",
 319                        sbi->ll_md_exp->exp_obd->obd_name, err);
 320                 GOTO(out_md_fid, err);
 321         }
 322
 323         LASSERT(osfs->os_bsize);
 324         sb->s_blocksize = osfs->os_bsize;
 325         sb->s_blocksize_bits = log2(osfs->os_bsize);
 326         sb->s_magic = LL_SUPER_MAGIC;
 327         sb->s_maxbytes = MAX_LFS_FILESIZE;
 328         sbi->ll_namelen = osfs->os_namelen;
 329         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 330
 331         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 332             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 333                 LCONSOLE_INFO("Disabling user_xattr feature because "
 334                               "it is not supported on the server\n");
 335                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags |= MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags |= LL_SBI_ACL;
 343         } else {
 344                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags &= ~MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags &= ~LL_SBI_ACL;
 349         }
 350
 351         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 352                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 353                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client is set as remote by default.\n");
 355                 }
 356         } else {
 357                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 358                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client claims to be remote, but server "
 360                                       "rejected, forced to be local.\n");
 361                 }
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 365                 LCONSOLE_INFO("client enabled MDS capability!\n");
 366                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 370                 LCONSOLE_INFO("client enabled OSS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 375                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 378                 sbi->ll_md_brw_size = data->ocd_brw_size;
 379         else
 380                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 383                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 384
 385         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 386                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 387                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 388                                       "unknown maximum xattr size.\n", dt);
 389                 } else {
 390                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 391                         sbi->ll_xattr_cache_enabled = 1;
 392                 }
 393         }
 394
 395         obd = class_name2obd(dt);
 396         if (!obd) {
 397                 CERROR("DT %s: not setup or attached\n", dt);
 398                 GOTO(out_md_fid, err = -ENODEV);
 399         }
 400
 401         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 402                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 403                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 404                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 405                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 406                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 407                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 408                                   OBD_CONNECT_MAXBYTES |
 409                                   OBD_CONNECT_EINPROGRESS |
 410                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 411                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 412
 413         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 414                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 415
 416         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 417                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 418                  * disabled by default, because it can still be enabled on the
 419                  * fly via /proc. As a consequence, we still need to come to an
 420                  * agreement on the supported algorithms at connect time */
 421                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 422
 423                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 424                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 425                 else
 426                         data->ocd_cksum_types = cksum_types_supported_client();
 427         }
 428
 429 #ifdef HAVE_LRU_RESIZE_SUPPORT
 430         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 431 #endif
 432         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 433                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 434
 435         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 436                "ocd_grant: %d\n", data->ocd_connect_flags,
 437                data->ocd_version, data->ocd_grant);
 438
 439         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 440         obd->obd_upcall.onu_upcall = cl_ocd_update;
 441
 442         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 443
 444         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 445                           NULL);
 446         if (err == -EBUSY) {
 447                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 448                                    "recovery, of which this client is not a "
 449                                    "part.  Please wait for recovery to "
 450                                    "complete, abort, or time out.\n", dt);
 451                 GOTO(out_md, err);
 452         } else if (err) {
 453                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 454                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 455                 GOTO(out_md, err);
 456         }
 457
 458         sbi->ll_dt_exp->exp_connect_data = *data;
 459
 460         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 461                            LUSTRE_SEQ_METADATA);
 462         if (err) {
 463                 CERROR("%s: Can't init data layer FID infrastructure, "
 464                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 465                 GOTO(out_dt, err);
 466         }
 467
 468         mutex_lock(&sbi->ll_lco.lco_lock);
 469         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 470         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 471         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 472         mutex_unlock(&sbi->ll_lco.lco_lock);
 473
 474         fid_zero(&sbi->ll_root_fid);
 475         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 476         if (err) {
 477                 CERROR("cannot mds_connect: rc = %d\n", err);
 478                 GOTO(out_lock_cn_cb, err);
 479         }
 480         if (!fid_is_sane(&sbi->ll_root_fid)) {
 481                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 482                        sbi->ll_md_exp->exp_obd->obd_name,
 483                        PFID(&sbi->ll_root_fid));
 484                 GOTO(out_lock_cn_cb, err = -EINVAL);
 485         }
 486         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 487
 488         sb->s_op = &lustre_super_operations;
 489 #if THREAD_SIZE >= 8192 /*b=17630*/
 490         sb->s_export_op = &lustre_export_operations;
 491 #endif
 492
 493         /* make root inode
 494          * XXX: move this to after cbd setup? */
 495         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 496                 OBD_MD_FLMODEASIZE;
 497         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 498                 valid |= OBD_MD_FLRMTPERM;
 499         else if (sbi->ll_flags & LL_SBI_ACL)
 500                 valid |= OBD_MD_FLACL;
 501
 502         OBD_ALLOC_PTR(op_data);
 503         if (op_data == NULL)
 504                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 505
 506         op_data->op_fid1 = sbi->ll_root_fid;
 507         op_data->op_mode = 0;
 508         op_data->op_capa1 = oc;
 509         op_data->op_valid = valid;
 510
 511         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 512         if (oc)
 513                 capa_put(oc);
 514         OBD_FREE_PTR(op_data);
 515         if (err) {
 516                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 517                        sbi->ll_md_exp->exp_obd->obd_name, err);
 518                 GOTO(out_lock_cn_cb, err);
 519         }
 520
 521         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 522                                sbi->ll_md_exp, &lmd);
 523         if (err) {
 524                 CERROR("failed to understand root inode md: rc = %d\n", err);
 525                 ptlrpc_req_finished(request);
 526                 GOTO(out_lock_cn_cb, err);
 527         }
 528
 529         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 530         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 531                                             sbi->ll_flags & LL_SBI_32BIT_API),
 532                        &lmd);
 533         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 534         ptlrpc_req_finished(request);
 535
 536         if (root == NULL || IS_ERR(root)) {
 537                 if (lmd.lsm)
 538                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 539 #ifdef CONFIG_FS_POSIX_ACL
 540                 if (lmd.posix_acl) {
 541                         posix_acl_release(lmd.posix_acl);
 542                         lmd.posix_acl = NULL;
 543                 }
 544 #endif
 545                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 546                 root = NULL;
 547                 CERROR("lustre_lite: bad iget4 for root\n");
 548                 GOTO(out_root, err);
 549         }
 550
 551         err = ll_close_thread_start(&sbi->ll_lcq);
 552         if (err) {
 553                 CERROR("cannot start close thread: rc %d\n", err);
 554                 GOTO(out_root, err);
 555         }
 556
 557 #ifdef CONFIG_FS_POSIX_ACL
 558         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 559                 rct_init(&sbi->ll_rct);
 560                 et_init(&sbi->ll_et);
 561         }
 562 #endif
 563
 564         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 565         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 566                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 567                                  NULL);
 568         cl_sb_init(sb);
 569
 570         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 571                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 572                                  &sbi->ll_cache, NULL);
 573
 574         sb->s_root = d_make_root(root);
 575         if (sb->s_root == NULL) {
 576                 CERROR("%s: can't make root dentry\n",
 577                         ll_get_fsname(sb, NULL, 0));
 578                 GOTO(out_root, err = -ENOMEM);
 579         }
 580 #ifdef HAVE_DCACHE_LOCK
 581         sb->s_root->d_op = &ll_d_ops;
 582 #endif
 583
 584         sbi->ll_sdev_orig = sb->s_dev;
 585
 586         /* We set sb->s_dev equal on all lustre clients in order to support
 587          * NFS export clustering.  NFSD requires that the FSID be the same
 588          * on all clients. */
 589         /* s_dev is also used in lt_compare() to compare two fs, but that is
 590          * only a node-local comparison. */
 591         uuid = obd_get_uuid(sbi->ll_md_exp);
 592         if (uuid != NULL)
 593                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 594
 595         if (data != NULL)
 596                 OBD_FREE_PTR(data);
 597         if (osfs != NULL)
 598                 OBD_FREE_PTR(osfs);
 599         if (proc_lustre_fs_root != NULL) {
 600                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 601                                                   dt, md);
 602                 if (err < 0) {
 603                         CERROR("%s: could not register mount in lprocfs: "
 604                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 605                         err = 0;
 606                 }
 607         }
 608
 609         RETURN(err);
 610 out_root:
 611         if (root)
 612                 iput(root);
 613 out_lock_cn_cb:
 614         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 615 out_dt:
 616         obd_disconnect(sbi->ll_dt_exp);
 617         sbi->ll_dt_exp = NULL;
 618         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 619         obd_zombie_barrier();
 620 out_md_fid:
 621         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 622 out_md:
 623         obd_disconnect(sbi->ll_md_exp);
 624         sbi->ll_md_exp = NULL;
 625 out:
 626         if (data != NULL)
 627                 OBD_FREE_PTR(data);
 628         if (osfs != NULL)
 629                 OBD_FREE_PTR(osfs);
 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 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 648 {
 649         int size, rc;
 650
 651         size = sizeof(int);
 652         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 653                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 654         if (rc)
 655                 CERROR("Get default mdsize error rc %d\n", rc);
 656
 657         RETURN(rc);
 658 }
 659
 660 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 661 {
 662         int size, rc;
 663
 664         size = sizeof(int);
 665         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 666                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 667         if (rc)
 668                 CERROR("Get max cookiesize error rc %d\n", rc);
 669
 670         RETURN(rc);
 671 }
 672
 673 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 674 {
 675         int size, rc;
 676
 677         size = sizeof(int);
 678         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 679                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 680         if (rc)
 681                 CERROR("Get default cookiesize error rc %d\n", rc);
 682
 683         RETURN(rc);
 684 }
 685
 686 void ll_dump_inode(struct inode *inode)
 687 {
 688         struct ll_d_hlist_node *tmp;
 689         int dentry_count = 0;
 690
 691         LASSERT(inode != NULL);
 692
 693         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 694                 dentry_count++;
 695
 696         CERROR("%s: inode %p dump: dev=%s fid="DFID
 697                " mode=%o count=%u, %d dentries\n",
 698                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 699                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 700                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 701 }
 702
 703 void lustre_dump_dentry(struct dentry *dentry, int recur)
 704 {
 705         struct list_head *tmp;
 706         int subdirs = 0;
 707
 708         LASSERT(dentry != NULL);
 709
 710         list_for_each(tmp, &dentry->d_subdirs)
 711                 subdirs++;
 712
 713         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 714                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 715                dentry->d_name.len, dentry->d_name.name,
 716                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 717                dentry->d_parent, dentry->d_inode, d_count(dentry),
 718                dentry->d_flags, dentry->d_fsdata, subdirs);
 719         if (dentry->d_inode != NULL)
 720                 ll_dump_inode(dentry->d_inode);
 721
 722         if (recur == 0)
 723                 return;
 724
 725         list_for_each(tmp, &dentry->d_subdirs) {
 726                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 727                 lustre_dump_dentry(d, recur - 1);
 728         }
 729 }
 730
 731 void client_common_put_super(struct super_block *sb)
 732 {
 733         struct ll_sb_info *sbi = ll_s2sbi(sb);
 734         ENTRY;
 735
 736 #ifdef CONFIG_FS_POSIX_ACL
 737         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 738                 et_fini(&sbi->ll_et);
 739                 rct_fini(&sbi->ll_rct);
 740         }
 741 #endif
 742
 743         ll_close_thread_shutdown(sbi->ll_lcq);
 744
 745         cl_sb_fini(sb);
 746
 747         cfs_list_del(&sbi->ll_conn_chain);
 748
 749         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 750         obd_disconnect(sbi->ll_dt_exp);
 751         sbi->ll_dt_exp = NULL;
 752         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 753          * see LU-2543. */
 754         obd_zombie_barrier();
 755
 756         lprocfs_unregister_mountpoint(sbi);
 757
 758         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 759         obd_disconnect(sbi->ll_md_exp);
 760         sbi->ll_md_exp = NULL;
 761
 762         EXIT;
 763 }
 764
 765 void ll_kill_super(struct super_block *sb)
 766 {
 767         struct ll_sb_info *sbi;
 768
 769         ENTRY;
 770
 771         /* not init sb ?*/
 772         if (!(sb->s_flags & MS_ACTIVE))
 773                 return;
 774
 775         sbi = ll_s2sbi(sb);
 776         /* we need restore s_dev from changed for clustred NFS before put_super
 777          * because new kernels have cached s_dev and change sb->s_dev in
 778          * put_super not affected real removing devices */
 779         if (sbi) {
 780                 sb->s_dev = sbi->ll_sdev_orig;
 781                 sbi->ll_umounting = 1;
 782         }
 783         EXIT;
 784 }
 785
 786 char *ll_read_opt(const char *opt, char *data)
 787 {
 788         char *value;
 789         char *retval;
 790         ENTRY;
 791
 792         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 793         if (strncmp(opt, data, strlen(opt)))
 794                 RETURN(NULL);
 795         if ((value = strchr(data, '=')) == NULL)
 796                 RETURN(NULL);
 797
 798         value++;
 799         OBD_ALLOC(retval, strlen(value) + 1);
 800         if (!retval) {
 801                 CERROR("out of memory!\n");
 802                 RETURN(NULL);
 803         }
 804
 805         memcpy(retval, value, strlen(value)+1);
 806         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 807         RETURN(retval);
 808 }
 809
 810 static inline int ll_set_opt(const char *opt, char *data, int fl)
 811 {
 812         if (strncmp(opt, data, strlen(opt)) != 0)
 813                 return(0);
 814         else
 815                 return(fl);
 816 }
 817
 818 /* non-client-specific mount options are parsed in lmd_parse */
 819 static int ll_options(char *options, int *flags)
 820 {
 821         int tmp;
 822         char *s1 = options, *s2;
 823         ENTRY;
 824
 825         if (!options)
 826                 RETURN(0);
 827
 828         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 829
 830         while (*s1) {
 831                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 832                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 863                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 864                 if (tmp) {
 865                         /* Ignore deprecated mount option.  The client will
 866                          * always try to mount with ACL support, whether this
 867                          * is used depends on whether server supports it. */
 868                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 869                                                   "mount option 'acl'.\n");
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 873                 if (tmp) {
 874                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 875                                                   "mount option 'noacl'.\n");
 876                         goto next;
 877                 }
 878 #else
 879 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 880 #endif
 881                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 882                 if (tmp) {
 883                         *flags |= tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 892                 if (tmp) {
 893                         *flags &= ~tmp;
 894                         goto next;
 895                 }
 896
 897                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 908                 if (tmp) {
 909                         *flags |= tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 913                 if (tmp) {
 914                         *flags &= ~tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 918                 if (tmp) {
 919                         *flags |= tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 923                 if (tmp) {
 924                         *flags &= ~tmp;
 925                         goto next;
 926                 }
 927                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 928                 if (tmp) {
 929                         *flags |= tmp;
 930                         goto next;
 931                 }
 932                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 938                 if (tmp) {
 939                         *flags |= tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 943                 if (tmp) {
 944                         *flags &= ~tmp;
 945                         goto next;
 946                 }
 947                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 948                                    s1);
 949                 RETURN(-EINVAL);
 950
 951 next:
 952                 /* Find next opt */
 953                 s2 = strchr(s1, ',');
 954                 if (s2 == NULL)
 955                         break;
 956                 s1 = s2 + 1;
 957         }
 958         RETURN(0);
 959 }
 960
 961 void ll_lli_init(struct ll_inode_info *lli)
 962 {
 963         lli->lli_inode_magic = LLI_INODE_MAGIC;
 964         lli->lli_flags = 0;
 965         lli->lli_ioepoch = 0;
 966         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 967         spin_lock_init(&lli->lli_lock);
 968         lli->lli_posix_acl = NULL;
 969         lli->lli_remote_perms = NULL;
 970         mutex_init(&lli->lli_rmtperm_mutex);
 971         /* Do not set lli_fid, it has been initialized already. */
 972         fid_zero(&lli->lli_pfid);
 973         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 974         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 975         cfs_atomic_set(&lli->lli_open_count, 0);
 976         lli->lli_mds_capa = NULL;
 977         lli->lli_rmtperm_time = 0;
 978         lli->lli_pending_och = NULL;
 979         lli->lli_mds_read_och = NULL;
 980         lli->lli_mds_write_och = NULL;
 981         lli->lli_mds_exec_och = NULL;
 982         lli->lli_open_fd_read_count = 0;
 983         lli->lli_open_fd_write_count = 0;
 984         lli->lli_open_fd_exec_count = 0;
 985         mutex_init(&lli->lli_och_mutex);
 986         spin_lock_init(&lli->lli_agl_lock);
 987         lli->lli_has_smd = false;
 988         spin_lock_init(&lli->lli_layout_lock);
 989         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 990         lli->lli_clob = NULL;
 991
 992         init_rwsem(&lli->lli_xattrs_list_rwsem);
 993         mutex_init(&lli->lli_xattrs_enq_lock);
 994
 995         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 996         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 997                 mutex_init(&lli->lli_readdir_mutex);
 998                 lli->lli_opendir_key = NULL;
 999                 lli->lli_sai = NULL;
1000                 lli->lli_def_acl = NULL;
1001                 spin_lock_init(&lli->lli_sa_lock);
1002                 lli->lli_opendir_pid = 0;
1003         } else {
1004                 mutex_init(&lli->lli_size_mutex);
1005                 lli->lli_symlink_name = NULL;
1006                 init_rwsem(&lli->lli_trunc_sem);
1007                 mutex_init(&lli->lli_write_mutex);
1008                 init_rwsem(&lli->lli_glimpse_sem);
1009                 lli->lli_glimpse_time = 0;
1010                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
1011                 lli->lli_agl_index = 0;
1012                 lli->lli_async_rc = 0;
1013                 lli->lli_volatile = false;
1014         }
1015         mutex_init(&lli->lli_layout_mutex);
1016 }
1017
1018 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1019 {
1020         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1021
1022         bdi->name = "lustre";
1023         return bdi_register(bdi, NULL, "lustre-%d",
1024                             atomic_inc_return(&ll_bdi_num));
1025 }
1026
1027 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1028 {
1029         struct lustre_profile *lprof = NULL;
1030         struct lustre_sb_info *lsi = s2lsi(sb);
1031         struct ll_sb_info *sbi;
1032         char  *dt = NULL, *md = NULL;
1033         char  *profilenm = get_profile_name(sb);
1034         struct config_llog_instance *cfg;
1035         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1036         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1037         int    err;
1038         ENTRY;
1039
1040         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1041
1042         OBD_ALLOC_PTR(cfg);
1043         if (cfg == NULL)
1044                 RETURN(-ENOMEM);
1045
1046         try_module_get(THIS_MODULE);
1047
1048         /* client additional sb info */
1049         lsi->lsi_llsbi = sbi = ll_init_sbi();
1050         if (!sbi) {
1051                 module_put(THIS_MODULE);
1052                 OBD_FREE_PTR(cfg);
1053                 RETURN(-ENOMEM);
1054         }
1055
1056         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1057         if (err)
1058                 GOTO(out_free, err);
1059
1060         err = bdi_init(&lsi->lsi_bdi);
1061         if (err)
1062                 GOTO(out_free, err);
1063         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1064         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1065         err = ll_bdi_register(&lsi->lsi_bdi);
1066         if (err)
1067                 GOTO(out_free, err);
1068
1069         sb->s_bdi = &lsi->lsi_bdi;
1070 #ifndef HAVE_DCACHE_LOCK
1071         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1072         sb->s_d_op = &ll_d_ops;
1073 #endif
1074
1075         /* Generate a string unique to this super, in case some joker tries
1076            to mount the same fs at two mount points.
1077            Use the address of the super itself.*/
1078         cfg->cfg_instance = sb;
1079         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1080         cfg->cfg_callback = class_config_llog_handler;
1081         /* set up client obds */
1082         err = lustre_process_log(sb, profilenm, cfg);
1083         if (err < 0)
1084                 GOTO(out_free, err);
1085
1086         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1087         lprof = class_get_profile(profilenm);
1088         if (lprof == NULL) {
1089                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1090                                    " read from the MGS.  Does that filesystem "
1091                                    "exist?\n", profilenm);
1092                 GOTO(out_free, err = -EINVAL);
1093         }
1094         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1095                lprof->lp_md, lprof->lp_dt);
1096
1097         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1098         if (!dt)
1099                 GOTO(out_free, err = -ENOMEM);
1100         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1101
1102         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1103         if (!md)
1104                 GOTO(out_free, err = -ENOMEM);
1105         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1106
1107         /* connections, registrations, sb setup */
1108         err = client_common_fill_super(sb, md, dt, mnt);
1109
1110 out_free:
1111         if (md)
1112                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1113         if (dt)
1114                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1115         if (err)
1116                 ll_put_super(sb);
1117         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1118                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1119
1120         OBD_FREE_PTR(cfg);
1121         RETURN(err);
1122 } /* ll_fill_super */
1123
1124
1125 void lu_context_keys_dump(void);
1126
1127 void ll_put_super(struct super_block *sb)
1128 {
1129         struct config_llog_instance cfg, params_cfg;
1130         struct obd_device *obd;
1131         struct lustre_sb_info *lsi = s2lsi(sb);
1132         struct ll_sb_info *sbi = ll_s2sbi(sb);
1133         char *profilenm = get_profile_name(sb);
1134         int ccc_count, next, force = 1, rc = 0;
1135         ENTRY;
1136
1137         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1138
1139         ll_print_capa_stat(sbi);
1140
1141         cfg.cfg_instance = sb;
1142         lustre_end_log(sb, profilenm, &cfg);
1143
1144         params_cfg.cfg_instance = sb;
1145         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1146
1147         if (sbi->ll_md_exp) {
1148                 obd = class_exp2obd(sbi->ll_md_exp);
1149                 if (obd)
1150                         force = obd->obd_force;
1151         }
1152
1153         /* Wait for unstable pages to be committed to stable storage */
1154         if (force == 0) {
1155                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1156                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1157                         cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1158                         &lwi);
1159         }
1160
1161         ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1162         if (force == 0 && rc != -EINTR)
1163                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1164
1165
1166         /* We need to set force before the lov_disconnect in
1167            lustre_common_put_super, since l_d cleans up osc's as well. */
1168         if (force) {
1169                 next = 0;
1170                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1171                                                      &next)) != NULL) {
1172                         obd->obd_force = force;
1173                 }
1174         }
1175
1176         if (sbi->ll_lcq) {
1177                 /* Only if client_common_fill_super succeeded */
1178                 client_common_put_super(sb);
1179         }
1180
1181         next = 0;
1182         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1183                 class_manual_cleanup(obd);
1184         }
1185
1186         if (sbi->ll_flags & LL_SBI_VERBOSE)
1187                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1188
1189         if (profilenm)
1190                 class_del_profile(profilenm);
1191
1192         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1193                 bdi_destroy(&lsi->lsi_bdi);
1194                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1195         }
1196
1197         ll_free_sbi(sb);
1198         lsi->lsi_llsbi = NULL;
1199
1200         lustre_common_put_super(sb);
1201
1202         cl_env_cache_purge(~0);
1203
1204         module_put(THIS_MODULE);
1205
1206         EXIT;
1207 } /* client_put_super */
1208
1209 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1210 {
1211         struct inode *inode = NULL;
1212
1213         /* NOTE: we depend on atomic igrab() -bzzz */
1214         lock_res_and_lock(lock);
1215         if (lock->l_resource->lr_lvb_inode) {
1216                 struct ll_inode_info * lli;
1217                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1218                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1219                         inode = igrab(lock->l_resource->lr_lvb_inode);
1220                 } else {
1221                         inode = lock->l_resource->lr_lvb_inode;
1222                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1223                                          D_WARNING, lock, "lr_lvb_inode %p is "
1224                                          "bogus: magic %08x",
1225                                          lock->l_resource->lr_lvb_inode,
1226                                          lli->lli_inode_magic);
1227                         inode = NULL;
1228                 }
1229         }
1230         unlock_res_and_lock(lock);
1231         return inode;
1232 }
1233
1234 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1235 {
1236         struct inode *inode = NULL;
1237         /* NOTE: we depend on atomic igrab() -bzzz */
1238         lock_res_and_lock(lock);
1239         if (lock->l_ast_data) {
1240                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1241                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1242                         inode = igrab(lock->l_ast_data);
1243                 } else {
1244                         inode = lock->l_ast_data;
1245                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1246                                          D_WARNING, lock, "l_ast_data %p is "
1247                                          "bogus: magic %08x", lock->l_ast_data,
1248                                          lli->lli_inode_magic);
1249                         inode = NULL;
1250                 }
1251         }
1252         unlock_res_and_lock(lock);
1253         return inode;
1254 }
1255
1256 static void ll_dir_clear_lsm_md(struct inode *inode)
1257 {
1258         struct ll_inode_info *lli = ll_i2info(inode);
1259
1260         LASSERT(S_ISDIR(inode->i_mode));
1261
1262         if (lli->lli_lsm_md != NULL) {
1263                 lmv_free_memmd(lli->lli_lsm_md);
1264                 lli->lli_lsm_md = NULL;
1265         }
1266 }
1267
1268 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1269                                       const struct lu_fid *fid,
1270                                       struct lustre_md *md)
1271 {
1272         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1273         struct mdt_body         *body = md->body;
1274         struct inode            *inode;
1275         ino_t                   ino;
1276         ENTRY;
1277
1278         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1279         inode = iget_locked(sb, ino);
1280         if (inode == NULL) {
1281                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1282                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1283                 RETURN(ERR_PTR(-ENOENT));
1284         }
1285
1286         if (inode->i_state & I_NEW) {
1287                 struct ll_inode_info *lli = ll_i2info(inode);
1288                 struct lmv_stripe_md *lsm = md->lmv;
1289
1290                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1291                                 (body->mode & S_IFMT);
1292                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1293                          PFID(fid));
1294
1295                 LTIME_S(inode->i_mtime) = 0;
1296                 LTIME_S(inode->i_atime) = 0;
1297                 LTIME_S(inode->i_ctime) = 0;
1298                 inode->i_rdev = 0;
1299
1300                 /* initializing backing dev info. */
1301                 inode->i_mapping->backing_dev_info =
1302                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1303                 inode->i_op = &ll_dir_inode_operations;
1304                 inode->i_fop = &ll_dir_operations;
1305                 lli->lli_fid = *fid;
1306                 ll_lli_init(lli);
1307
1308                 LASSERT(lsm != NULL);
1309                 /* master stripe FID */
1310                 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1311                 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1312                        lli, PFID(fid), PFID(&lli->lli_pfid));
1313                 unlock_new_inode(inode);
1314         }
1315
1316         RETURN(inode);
1317 }
1318
1319 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1320 {
1321         struct lu_fid *fid;
1322         struct lmv_stripe_md *lsm = md->lmv;
1323         int i;
1324
1325         LASSERT(lsm != NULL);
1326         /* XXX sigh, this lsm_root initialization should be in
1327          * LMV layer, but it needs ll_iget right now, so we
1328          * put this here right now. */
1329         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1330                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1331                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1332                 if (i == 0) {
1333                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1334                 } else {
1335                         /* Unfortunately ll_iget will call ll_update_inode,
1336                          * where the initialization of slave inode is slightly
1337                          * different, so it reset lsm_md to NULL to avoid
1338                          * initializing lsm for slave inode. */
1339                         lsm->lsm_md_oinfo[i].lmo_root =
1340                                         ll_iget_anon_dir(inode->i_sb, fid, md);
1341                         if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1342                                 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1343
1344                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1345                                 return rc;
1346                         }
1347                 }
1348         }
1349
1350         /* Here is where the lsm is being initialized(fill lmo_info) after
1351          * client retrieve MD stripe information from MDT. */
1352         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1353                                 ll_md_blocking_ast);
1354 }
1355
1356 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1357                                 const struct lmv_stripe_md *lsm_md2)
1358 {
1359         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1360                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1361                lsm_md1->lsm_md_master_mdt_index ==
1362                                         lsm_md2->lsm_md_master_mdt_index &&
1363                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1364                lsm_md1->lsm_md_layout_version ==
1365                                         lsm_md2->lsm_md_layout_version &&
1366                strcmp(lsm_md1->lsm_md_pool_name,
1367                       lsm_md2->lsm_md_pool_name) == 0;
1368 }
1369
1370 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1371 {
1372         struct ll_inode_info *lli = ll_i2info(inode);
1373         struct lmv_stripe_md *lsm = md->lmv;
1374         int idx;
1375         ENTRY;
1376
1377         LASSERT(S_ISDIR(inode->i_mode));
1378         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1379                PFID(ll_inode2fid(inode)));
1380
1381         /* no striped information from request. */
1382         if (lsm == NULL) {
1383                 if (lli->lli_lsm_md == NULL) {
1384                         RETURN_EXIT;
1385                 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1386                         /* migration is done, the temporay MIGRATE layout has
1387                          * been removed */
1388                         CDEBUG(D_INODE, DFID" finish migration.\n",
1389                                PFID(ll_inode2fid(inode)));
1390                         lmv_free_memmd(lli->lli_lsm_md);
1391                         lli->lli_lsm_md = NULL;
1392                         RETURN_EXIT;
1393                 } else {
1394                         /* The lustre_md from req does not include stripeEA,
1395                          * see ll_md_setattr */
1396                         RETURN_EXIT;
1397                 }
1398         }
1399
1400         /* set the directory layout */
1401         if (lli->lli_lsm_md == NULL) {
1402                 int rc;
1403
1404                 rc = ll_init_lsm_md(inode, md);
1405                 if (rc != 0) {
1406                         CERROR("%s: init "DFID" failed: rc = %d\n",
1407                                ll_get_fsname(inode->i_sb, NULL, 0),
1408                                PFID(&lli->lli_fid), rc);
1409                         return;
1410                 }
1411                 lli->lli_lsm_md = lsm;
1412                 /* set lsm_md to NULL, so the following free lustre_md
1413                  * will not free this lsm */
1414                 md->lmv = NULL;
1415                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1416                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1417                 RETURN_EXIT;
1418         }
1419
1420         /* Compare the old and new stripe information */
1421         if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1422                 CERROR("inode %p %lu mismatch\n"
1423                        "    new(%p)     vs     lli_lsm_md(%p):\n"
1424                        "    magic:      %x                   %x\n"
1425                        "    count:      %x                   %x\n"
1426                        "    master:     %x                   %x\n"
1427                        "    hash_type:  %x                   %x\n"
1428                        "    layout:     %x                   %x\n"
1429                        "    pool:       %s                   %s\n",
1430                        inode, inode->i_ino, lsm, lli->lli_lsm_md,
1431                        lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1432                        lsm->lsm_md_stripe_count,
1433                        lli->lli_lsm_md->lsm_md_stripe_count,
1434                        lsm->lsm_md_master_mdt_index,
1435                        lli->lli_lsm_md->lsm_md_master_mdt_index,
1436                        lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1437                        lsm->lsm_md_layout_version,
1438                        lli->lli_lsm_md->lsm_md_layout_version,
1439                        lsm->lsm_md_pool_name,
1440                        lli->lli_lsm_md->lsm_md_pool_name);
1441                 return;
1442         }
1443
1444         for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1445                 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1446                                &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1447                         CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1448                                "new:"DFID"\n",
1449                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1450                              PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1451                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1452                         return;
1453                 }
1454         }
1455
1456         md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1457                          md->body, ll_md_blocking_ast);
1458
1459         RETURN_EXIT;
1460 }
1461
1462 void ll_clear_inode(struct inode *inode)
1463 {
1464         struct ll_inode_info *lli = ll_i2info(inode);
1465         struct ll_sb_info *sbi = ll_i2sbi(inode);
1466         ENTRY;
1467
1468         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1469                PFID(ll_inode2fid(inode)), inode);
1470
1471         if (S_ISDIR(inode->i_mode)) {
1472                 /* these should have been cleared in ll_file_release */
1473                 LASSERT(lli->lli_opendir_key == NULL);
1474                 LASSERT(lli->lli_sai == NULL);
1475                 LASSERT(lli->lli_opendir_pid == 0);
1476         }
1477
1478         spin_lock(&lli->lli_lock);
1479         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1480         spin_unlock(&lli->lli_lock);
1481         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1482
1483         LASSERT(!lli->lli_open_fd_write_count);
1484         LASSERT(!lli->lli_open_fd_read_count);
1485         LASSERT(!lli->lli_open_fd_exec_count);
1486
1487         if (lli->lli_mds_write_och)
1488                 ll_md_real_close(inode, FMODE_WRITE);
1489         if (lli->lli_mds_exec_och)
1490                 ll_md_real_close(inode, FMODE_EXEC);
1491         if (lli->lli_mds_read_och)
1492                 ll_md_real_close(inode, FMODE_READ);
1493
1494         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1495                 OBD_FREE(lli->lli_symlink_name,
1496                          strlen(lli->lli_symlink_name) + 1);
1497                 lli->lli_symlink_name = NULL;
1498         }
1499
1500         ll_xattr_cache_destroy(inode);
1501
1502         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1503                 LASSERT(lli->lli_posix_acl == NULL);
1504                 if (lli->lli_remote_perms) {
1505                         free_rmtperm_hash(lli->lli_remote_perms);
1506                         lli->lli_remote_perms = NULL;
1507                 }
1508         }
1509 #ifdef CONFIG_FS_POSIX_ACL
1510         else if (lli->lli_posix_acl) {
1511                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1512                 LASSERT(lli->lli_remote_perms == NULL);
1513                 posix_acl_release(lli->lli_posix_acl);
1514                 lli->lli_posix_acl = NULL;
1515         }
1516 #endif
1517         lli->lli_inode_magic = LLI_INODE_DEAD;
1518
1519         ll_clear_inode_capas(inode);
1520         if (S_ISDIR(inode->i_mode))
1521                 ll_dir_clear_lsm_md(inode);
1522         else
1523                 LASSERT(list_empty(&lli->lli_agl_list));
1524
1525         /*
1526          * XXX This has to be done before lsm is freed below, because
1527          * cl_object still uses inode lsm.
1528          */
1529         cl_inode_fini(inode);
1530         lli->lli_has_smd = false;
1531
1532         EXIT;
1533 }
1534
1535 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1536                   struct md_open_data **mod)
1537 {
1538         struct lustre_md md;
1539         struct inode *inode = dentry->d_inode;
1540         struct ll_sb_info *sbi = ll_i2sbi(inode);
1541         struct ptlrpc_request *request = NULL;
1542         int rc, ia_valid;
1543         ENTRY;
1544
1545         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1546                                      LUSTRE_OPC_ANY, NULL);
1547         if (IS_ERR(op_data))
1548                 RETURN(PTR_ERR(op_data));
1549
1550         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1551                         &request, mod);
1552         if (rc) {
1553                 ptlrpc_req_finished(request);
1554                 if (rc == -ENOENT) {
1555                         clear_nlink(inode);
1556                         /* Unlinked special device node? Or just a race?
1557                          * Pretend we done everything. */
1558                         if (!S_ISREG(inode->i_mode) &&
1559                             !S_ISDIR(inode->i_mode)) {
1560                                 ia_valid = op_data->op_attr.ia_valid;
1561                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1562                                 rc = simple_setattr(dentry, &op_data->op_attr);
1563                                 op_data->op_attr.ia_valid = ia_valid;
1564                         }
1565                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1566                         CERROR("md_setattr fails: rc = %d\n", rc);
1567                 }
1568                 RETURN(rc);
1569         }
1570
1571         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1572                               sbi->ll_md_exp, &md);
1573         if (rc) {
1574                 ptlrpc_req_finished(request);
1575                 RETURN(rc);
1576         }
1577
1578         ia_valid = op_data->op_attr.ia_valid;
1579         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1580          * cache is not cleared yet. */
1581         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1582         rc = simple_setattr(dentry, &op_data->op_attr);
1583         op_data->op_attr.ia_valid = ia_valid;
1584
1585         /* Extract epoch data if obtained. */
1586         op_data->op_handle = md.body->handle;
1587         op_data->op_ioepoch = md.body->ioepoch;
1588
1589         ll_update_inode(inode, &md);
1590         ptlrpc_req_finished(request);
1591
1592         RETURN(rc);
1593 }
1594
1595 /* Close IO epoch and send Size-on-MDS attribute update. */
1596 static int ll_setattr_done_writing(struct inode *inode,
1597                                    struct md_op_data *op_data,
1598                                    struct md_open_data *mod)
1599 {
1600         struct ll_inode_info *lli = ll_i2info(inode);
1601         int rc = 0;
1602         ENTRY;
1603
1604         LASSERT(op_data != NULL);
1605         if (!S_ISREG(inode->i_mode))
1606                 RETURN(0);
1607
1608         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1609                op_data->op_ioepoch, PFID(&lli->lli_fid));
1610
1611         op_data->op_flags = MF_EPOCH_CLOSE;
1612         ll_done_writing_attr(inode, op_data);
1613         ll_pack_inode2opdata(inode, op_data, NULL);
1614
1615         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1616         if (rc == -EAGAIN) {
1617                 /* MDS has instructed us to obtain Size-on-MDS attribute
1618                  * from OSTs and send setattr to back to MDS. */
1619                 rc = ll_som_update(inode, op_data);
1620         } else if (rc) {
1621                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1622                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1623                        PFID(ll_inode2fid(inode)), rc);
1624         }
1625         RETURN(rc);
1626 }
1627
1628 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1629 {
1630         struct obd_capa *capa;
1631         int rc;
1632
1633         if (attr->ia_valid & ATTR_SIZE)
1634                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1635         else
1636                 capa = ll_mdscapa_get(inode);
1637
1638         rc = cl_setattr_ost(inode, attr, capa);
1639
1640         if (attr->ia_valid & ATTR_SIZE)
1641                 ll_truncate_free_capa(capa);
1642         else
1643                 capa_put(capa);
1644
1645         return rc;
1646 }
1647
1648 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1649  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1650  * keep these values until such a time that objects are allocated for it.
1651  * We do the MDS operations first, as it is checking permissions for us.
1652  * We don't to the MDS RPC if there is nothing that we want to store there,
1653  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1654  * going to do an RPC anyways.
1655  *
1656  * If we are doing a truncate, we will send the mtime and ctime updates
1657  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1658  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1659  * at the same time.
1660  *
1661  * In case of HSMimport, we only set attr on MDS.
1662  */
1663 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1664 {
1665         struct inode *inode = dentry->d_inode;
1666         struct ll_inode_info *lli = ll_i2info(inode);
1667         struct md_op_data *op_data = NULL;
1668         struct md_open_data *mod = NULL;
1669         bool file_is_released = false;
1670         int rc = 0, rc1 = 0;
1671         ENTRY;
1672
1673         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1674                "valid %x, hsm_import %d\n",
1675                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1676                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1677                hsm_import);
1678
1679         if (attr->ia_valid & ATTR_SIZE) {
1680                 /* Check new size against VFS/VM file size limit and rlimit */
1681                 rc = inode_newsize_ok(inode, attr->ia_size);
1682                 if (rc)
1683                         RETURN(rc);
1684
1685                 /* The maximum Lustre file size is variable, based on the
1686                  * OST maximum object size and number of stripes.  This
1687                  * needs another check in addition to the VFS check above. */
1688                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1689                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1690                                PFID(&lli->lli_fid), attr->ia_size,
1691                                ll_file_maxbytes(inode));
1692                         RETURN(-EFBIG);
1693                 }
1694
1695                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1696         }
1697
1698         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1699         if (attr->ia_valid & TIMES_SET_FLAGS) {
1700                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1701                     !cfs_capable(CFS_CAP_FOWNER))
1702                         RETURN(-EPERM);
1703         }
1704
1705         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1706         if (attr->ia_valid & ATTR_CTIME) {
1707                 attr->ia_ctime = CFS_CURRENT_TIME;
1708                 attr->ia_valid |= ATTR_CTIME_SET;
1709         }
1710         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1711             (attr->ia_valid & ATTR_ATIME)) {
1712                 attr->ia_atime = CFS_CURRENT_TIME;
1713                 attr->ia_valid |= ATTR_ATIME_SET;
1714         }
1715         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1716             (attr->ia_valid & ATTR_MTIME)) {
1717                 attr->ia_mtime = CFS_CURRENT_TIME;
1718                 attr->ia_valid |= ATTR_MTIME_SET;
1719         }
1720
1721         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1722                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1723                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1724                        cfs_time_current_sec());
1725
1726         /* We always do an MDS RPC, even if we're only changing the size;
1727          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1728
1729         OBD_ALLOC_PTR(op_data);
1730         if (op_data == NULL)
1731                 RETURN(-ENOMEM);
1732
1733         if (!S_ISDIR(inode->i_mode)) {
1734                 if (attr->ia_valid & ATTR_SIZE)
1735                         inode_dio_write_done(inode);
1736                 mutex_unlock(&inode->i_mutex);
1737         }
1738
1739         /* truncate on a released file must failed with -ENODATA,
1740          * so size must not be set on MDS for released file
1741          * but other attributes must be set
1742          */
1743         if (S_ISREG(inode->i_mode)) {
1744                 struct lov_stripe_md *lsm;
1745                 __u32 gen;
1746
1747                 ll_layout_refresh(inode, &gen);
1748                 lsm = ccc_inode_lsm_get(inode);
1749                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1750                         file_is_released = true;
1751                 ccc_inode_lsm_put(inode, lsm);
1752
1753                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1754                         if (file_is_released) {
1755                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1756                                 if (rc < 0)
1757                                         GOTO(out, rc);
1758
1759                                 file_is_released = false;
1760                                 ll_layout_refresh(inode, &gen);
1761                         }
1762
1763                         /* If we are changing file size, file content is
1764                          * modified, flag it. */
1765                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1766                         spin_lock(&lli->lli_lock);
1767                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1768                         spin_unlock(&lli->lli_lock);
1769                         op_data->op_bias |= MDS_DATA_MODIFIED;
1770                 }
1771         }
1772
1773         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1774
1775         /* Open epoch for truncate. */
1776         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1777             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1778                 op_data->op_flags = MF_EPOCH_OPEN;
1779
1780         rc = ll_md_setattr(dentry, op_data, &mod);
1781         if (rc)
1782                 GOTO(out, rc);
1783
1784         /* RPC to MDT is sent, cancel data modification flag */
1785         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1786                 spin_lock(&lli->lli_lock);
1787                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1788                 spin_unlock(&lli->lli_lock);
1789         }
1790
1791         ll_ioepoch_open(lli, op_data->op_ioepoch);
1792         if (!S_ISREG(inode->i_mode) || file_is_released)
1793                 GOTO(out, rc = 0);
1794
1795         if (attr->ia_valid & (ATTR_SIZE |
1796                               ATTR_ATIME | ATTR_ATIME_SET |
1797                               ATTR_MTIME | ATTR_MTIME_SET)) {
1798                 /* For truncate and utimes sending attributes to OSTs, setting
1799                  * mtime/atime to the past will be performed under PW [0:EOF]
1800                  * extent lock (new_size:EOF for truncate).  It may seem
1801                  * excessive to send mtime/atime updates to OSTs when not
1802                  * setting times to past, but it is necessary due to possible
1803                  * time de-synchronization between MDT inode and OST objects */
1804                 if (attr->ia_valid & ATTR_SIZE)
1805                         down_write(&lli->lli_trunc_sem);
1806                 rc = ll_setattr_ost(inode, attr);
1807                 if (attr->ia_valid & ATTR_SIZE)
1808                         up_write(&lli->lli_trunc_sem);
1809         }
1810         EXIT;
1811 out:
1812         if (op_data) {
1813                 if (op_data->op_ioepoch) {
1814                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1815                         if (!rc)
1816                                 rc = rc1;
1817                 }
1818                 ll_finish_md_op_data(op_data);
1819         }
1820         if (!S_ISDIR(inode->i_mode)) {
1821                 mutex_lock(&inode->i_mutex);
1822                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1823                         inode_dio_wait(inode);
1824         }
1825
1826         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1827                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1828
1829         return rc;
1830 }
1831
1832 int ll_setattr(struct dentry *de, struct iattr *attr)
1833 {
1834         int mode = de->d_inode->i_mode;
1835
1836         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1837                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1838                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1839
1840         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1841                                (ATTR_SIZE|ATTR_MODE)) &&
1842             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1843              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1844               !(attr->ia_mode & S_ISGID))))
1845                 attr->ia_valid |= ATTR_FORCE;
1846
1847         if ((mode & S_ISUID) &&
1848             !(attr->ia_mode & S_ISUID) &&
1849             !(attr->ia_valid & ATTR_KILL_SUID))
1850                 attr->ia_valid |= ATTR_KILL_SUID;
1851
1852         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1853             !(attr->ia_mode & S_ISGID) &&
1854             !(attr->ia_valid & ATTR_KILL_SGID))
1855                 attr->ia_valid |= ATTR_KILL_SGID;
1856
1857         return ll_setattr_raw(de, attr, false);
1858 }
1859
1860 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1861                        __u64 max_age, __u32 flags)
1862 {
1863         struct ll_sb_info *sbi = ll_s2sbi(sb);
1864         struct obd_statfs obd_osfs;
1865         int rc;
1866         ENTRY;
1867
1868         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1869         if (rc) {
1870                 CERROR("md_statfs fails: rc = %d\n", rc);
1871                 RETURN(rc);
1872         }
1873
1874         osfs->os_type = sb->s_magic;
1875
1876         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1877                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1878
1879         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1880                 flags |= OBD_STATFS_NODELAY;
1881
1882         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1883         if (rc) {
1884                 CERROR("obd_statfs fails: rc = %d\n", rc);
1885                 RETURN(rc);
1886         }
1887
1888         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1889                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1890                obd_osfs.os_files);
1891
1892         osfs->os_bsize = obd_osfs.os_bsize;
1893         osfs->os_blocks = obd_osfs.os_blocks;
1894         osfs->os_bfree = obd_osfs.os_bfree;
1895         osfs->os_bavail = obd_osfs.os_bavail;
1896
1897         /* If we don't have as many objects free on the OST as inodes
1898          * on the MDS, we reduce the total number of inodes to
1899          * compensate, so that the "inodes in use" number is correct.
1900          */
1901         if (obd_osfs.os_ffree < osfs->os_ffree) {
1902                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1903                         obd_osfs.os_ffree;
1904                 osfs->os_ffree = obd_osfs.os_ffree;
1905         }
1906
1907         RETURN(rc);
1908 }
1909 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1910 {
1911         struct super_block *sb = de->d_sb;
1912         struct obd_statfs osfs;
1913         __u64 fsid = huge_encode_dev(sb->s_dev);
1914         int rc;
1915
1916         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1917         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1918
1919         /* Some amount of caching on the client is allowed */
1920         rc = ll_statfs_internal(sb, &osfs,
1921                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1922                                 0);
1923         if (rc)
1924                 return rc;
1925
1926         statfs_unpack(sfs, &osfs);
1927
1928         /* We need to downshift for all 32-bit kernels, because we can't
1929          * tell if the kernel is being called via sys_statfs64() or not.
1930          * Stop before overflowing f_bsize - in which case it is better
1931          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1932         if (sizeof(long) < 8) {
1933                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1934                         sfs->f_bsize <<= 1;
1935
1936                         osfs.os_blocks >>= 1;
1937                         osfs.os_bfree >>= 1;
1938                         osfs.os_bavail >>= 1;
1939                 }
1940         }
1941
1942         sfs->f_blocks = osfs.os_blocks;
1943         sfs->f_bfree = osfs.os_bfree;
1944         sfs->f_bavail = osfs.os_bavail;
1945         sfs->f_fsid.val[0] = (__u32)fsid;
1946         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1947         return 0;
1948 }
1949
1950 void ll_inode_size_lock(struct inode *inode)
1951 {
1952         struct ll_inode_info *lli;
1953
1954         LASSERT(!S_ISDIR(inode->i_mode));
1955
1956         lli = ll_i2info(inode);
1957         mutex_lock(&lli->lli_size_mutex);
1958 }
1959
1960 void ll_inode_size_unlock(struct inode *inode)
1961 {
1962         struct ll_inode_info *lli;
1963
1964         lli = ll_i2info(inode);
1965         mutex_unlock(&lli->lli_size_mutex);
1966 }
1967
1968 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1969 {
1970         struct ll_inode_info *lli = ll_i2info(inode);
1971         struct mdt_body *body = md->body;
1972         struct lov_stripe_md *lsm = md->lsm;
1973         struct ll_sb_info *sbi = ll_i2sbi(inode);
1974
1975         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1976         if (lsm != NULL) {
1977                 if (!lli->lli_has_smd &&
1978                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1979                         cl_file_inode_init(inode, md);
1980
1981                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1982                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1983                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1984         }
1985
1986         if (S_ISDIR(inode->i_mode))
1987                 ll_update_lsm_md(inode, md);
1988
1989         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1990                 if (body->valid & OBD_MD_FLRMTPERM)
1991                         ll_update_remote_perm(inode, md->remote_perm);
1992         }
1993 #ifdef CONFIG_FS_POSIX_ACL
1994         else if (body->valid & OBD_MD_FLACL) {
1995                 spin_lock(&lli->lli_lock);
1996                 if (lli->lli_posix_acl)
1997                         posix_acl_release(lli->lli_posix_acl);
1998                 lli->lli_posix_acl = md->posix_acl;
1999                 spin_unlock(&lli->lli_lock);
2000         }
2001 #endif
2002         inode->i_ino = cl_fid_build_ino(&body->fid1,
2003                                         sbi->ll_flags & LL_SBI_32BIT_API);
2004         inode->i_generation = cl_fid_build_gen(&body->fid1);
2005
2006         if (body->valid & OBD_MD_FLATIME) {
2007                 if (body->atime > LTIME_S(inode->i_atime))
2008                         LTIME_S(inode->i_atime) = body->atime;
2009                 lli->lli_lvb.lvb_atime = body->atime;
2010         }
2011         if (body->valid & OBD_MD_FLMTIME) {
2012                 if (body->mtime > LTIME_S(inode->i_mtime)) {
2013                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
2014                                "to "LPU64"\n", inode->i_ino,
2015                                LTIME_S(inode->i_mtime), body->mtime);
2016                         LTIME_S(inode->i_mtime) = body->mtime;
2017                 }
2018                 lli->lli_lvb.lvb_mtime = body->mtime;
2019         }
2020         if (body->valid & OBD_MD_FLCTIME) {
2021                 if (body->ctime > LTIME_S(inode->i_ctime))
2022                         LTIME_S(inode->i_ctime) = body->ctime;
2023                 lli->lli_lvb.lvb_ctime = body->ctime;
2024         }
2025         if (body->valid & OBD_MD_FLMODE)
2026                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
2027         if (body->valid & OBD_MD_FLTYPE)
2028                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
2029         LASSERT(inode->i_mode != 0);
2030         if (S_ISREG(inode->i_mode)) {
2031                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
2032         } else {
2033                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2034         }
2035         if (body->valid & OBD_MD_FLUID)
2036                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
2037         if (body->valid & OBD_MD_FLGID)
2038                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
2039         if (body->valid & OBD_MD_FLFLAGS)
2040                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
2041         if (body->valid & OBD_MD_FLNLINK)
2042                 set_nlink(inode, body->nlink);
2043         if (body->valid & OBD_MD_FLRDEV)
2044                 inode->i_rdev = old_decode_dev(body->rdev);
2045
2046         if (body->valid & OBD_MD_FLID) {
2047                 /* FID shouldn't be changed! */
2048                 if (fid_is_sane(&lli->lli_fid)) {
2049                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
2050                                  "Trying to change FID "DFID
2051                                  " to the "DFID", inode "DFID"(%p)\n",
2052                                  PFID(&lli->lli_fid), PFID(&body->fid1),
2053                                  PFID(ll_inode2fid(inode)), inode);
2054                 } else {
2055                         lli->lli_fid = body->fid1;
2056                 }
2057         }
2058
2059         LASSERT(fid_seq(&lli->lli_fid) != 0);
2060
2061         if (body->valid & OBD_MD_FLSIZE) {
2062                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2063                     S_ISREG(inode->i_mode)) {
2064                         struct lustre_handle lockh;
2065                         ldlm_mode_t mode;
2066
2067                         /* As it is possible a blocking ast has been processed
2068                          * by this time, we need to check there is an UPDATE
2069                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2070                          * it. */
2071                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2072                                                &lockh, LDLM_FL_CBPENDING,
2073                                                LCK_CR | LCK_CW |
2074                                                LCK_PR | LCK_PW);
2075                         if (mode) {
2076                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2077                                                       LLIF_EPOCH_PENDING |
2078                                                       LLIF_SOM_DIRTY)) {
2079                                         CERROR("%s: inode "DFID" flags %u still"
2080                                                " has size authority! do not "
2081                                                "trust the size from MDS\n",
2082                                                sbi->ll_md_exp->exp_obd->obd_name,
2083                                                PFID(ll_inode2fid(inode)),
2084                                                lli->lli_flags);
2085                                 } else {
2086                                         /* Use old size assignment to avoid
2087                                          * deadlock bz14138 & bz14326 */
2088                                         i_size_write(inode, body->size);
2089                                         spin_lock(&lli->lli_lock);
2090                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2091                                         spin_unlock(&lli->lli_lock);
2092                                 }
2093                                 ldlm_lock_decref(&lockh, mode);
2094                         }
2095                 } else {
2096                         /* Use old size assignment to avoid
2097                          * deadlock bz14138 & bz14326 */
2098                         i_size_write(inode, body->size);
2099
2100                         CDEBUG(D_VFSTRACE,
2101                                "inode="DFID", updating i_size %llu\n",
2102                                PFID(ll_inode2fid(inode)),
2103                                (unsigned long long)body->size);
2104                 }
2105
2106                 if (body->valid & OBD_MD_FLBLOCKS)
2107                         inode->i_blocks = body->blocks;
2108         }
2109
2110         if (body->valid & OBD_MD_FLMDSCAPA) {
2111                 LASSERT(md->mds_capa);
2112                 ll_add_capa(inode, md->mds_capa);
2113         }
2114         if (body->valid & OBD_MD_FLOSSCAPA) {
2115                 LASSERT(md->oss_capa);
2116                 ll_add_capa(inode, md->oss_capa);
2117         }
2118
2119         if (body->valid & OBD_MD_TSTATE) {
2120                 if (body->t_state & MS_RESTORE)
2121                         lli->lli_flags |= LLIF_FILE_RESTORING;
2122         }
2123 }
2124
2125 void ll_read_inode2(struct inode *inode, void *opaque)
2126 {
2127         struct lustre_md *md = opaque;
2128         struct ll_inode_info *lli = ll_i2info(inode);
2129         ENTRY;
2130
2131         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2132                PFID(&lli->lli_fid), inode);
2133
2134         LASSERT(!lli->lli_has_smd);
2135
2136         /* Core attributes from the MDS first.  This is a new inode, and
2137          * the VFS doesn't zero times in the core inode so we have to do
2138          * it ourselves.  They will be overwritten by either MDS or OST
2139          * attributes - we just need to make sure they aren't newer. */
2140         LTIME_S(inode->i_mtime) = 0;
2141         LTIME_S(inode->i_atime) = 0;
2142         LTIME_S(inode->i_ctime) = 0;
2143         inode->i_rdev = 0;
2144         ll_update_inode(inode, md);
2145
2146         /* OIDEBUG(inode); */
2147
2148         /* initializing backing dev info. */
2149         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2150
2151
2152         if (S_ISREG(inode->i_mode)) {
2153                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2154                 inode->i_op = &ll_file_inode_operations;
2155                 inode->i_fop = sbi->ll_fop;
2156                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2157                 EXIT;
2158         } else if (S_ISDIR(inode->i_mode)) {
2159                 inode->i_op = &ll_dir_inode_operations;
2160                 inode->i_fop = &ll_dir_operations;
2161                 EXIT;
2162         } else if (S_ISLNK(inode->i_mode)) {
2163                 inode->i_op = &ll_fast_symlink_inode_operations;
2164                 EXIT;
2165         } else {
2166                 inode->i_op = &ll_special_inode_operations;
2167
2168                 init_special_inode(inode, inode->i_mode,
2169                                    inode->i_rdev);
2170
2171                 EXIT;
2172         }
2173 }
2174
2175 void ll_delete_inode(struct inode *inode)
2176 {
2177         struct cl_inode_info *lli = cl_i2info(inode);
2178         ENTRY;
2179
2180         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2181                 /* discard all dirty pages before truncating them, required by
2182                  * osc_extent implementation at LU-1030. */
2183                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2184                                    CL_FSYNC_DISCARD, 1);
2185
2186         truncate_inode_pages(&inode->i_data, 0);
2187
2188         /* Workaround for LU-118 */
2189         if (inode->i_data.nrpages) {
2190                 spin_lock_irq(&inode->i_data.tree_lock);
2191                 spin_unlock_irq(&inode->i_data.tree_lock);
2192                 LASSERTF(inode->i_data.nrpages == 0,
2193                          "inode="DFID"(%p) nrpages=%lu, see "
2194                          "http://jira.whamcloud.com/browse/LU-118\n",
2195                          PFID(ll_inode2fid(inode)), inode,
2196                          inode->i_data.nrpages);
2197         }
2198         /* Workaround end */
2199
2200 #ifdef HAVE_SBOPS_EVICT_INODE
2201         ll_clear_inode(inode);
2202 #endif
2203         clear_inode(inode);
2204
2205         EXIT;
2206 }
2207
2208 int ll_iocontrol(struct inode *inode, struct file *file,
2209                  unsigned int cmd, unsigned long arg)
2210 {
2211         struct ll_sb_info *sbi = ll_i2sbi(inode);
2212         struct ptlrpc_request *req = NULL;
2213         int rc, flags = 0;
2214         ENTRY;
2215
2216         switch(cmd) {
2217         case FSFILT_IOC_GETFLAGS: {
2218                 struct mdt_body *body;
2219                 struct md_op_data *op_data;
2220
2221                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2222                                              0, 0, LUSTRE_OPC_ANY,
2223                                              NULL);
2224                 if (IS_ERR(op_data))
2225                         RETURN(PTR_ERR(op_data));
2226
2227                 op_data->op_valid = OBD_MD_FLFLAGS;
2228                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2229                 ll_finish_md_op_data(op_data);
2230                 if (rc) {
2231                         CERROR("%s: failure inode "DFID": rc = %d\n",
2232                                sbi->ll_md_exp->exp_obd->obd_name,
2233                                PFID(ll_inode2fid(inode)), rc);
2234                         RETURN(-abs(rc));
2235                 }
2236
2237                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2238
2239                 flags = body->flags;
2240
2241                 ptlrpc_req_finished(req);
2242
2243                 RETURN(put_user(flags, (int *)arg));
2244         }
2245         case FSFILT_IOC_SETFLAGS: {
2246                 struct lov_stripe_md *lsm;
2247                 struct obd_info oinfo = { { { 0 } } };
2248                 struct md_op_data *op_data;
2249
2250                 if (get_user(flags, (int *)arg))
2251                         RETURN(-EFAULT);
2252
2253                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2254                                              LUSTRE_OPC_ANY, NULL);
2255                 if (IS_ERR(op_data))
2256                         RETURN(PTR_ERR(op_data));
2257
2258                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2259                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2260                 rc = md_setattr(sbi->ll_md_exp, op_data,
2261                                 NULL, 0, NULL, 0, &req, NULL);
2262                 ll_finish_md_op_data(op_data);
2263                 ptlrpc_req_finished(req);
2264                 if (rc)
2265                         RETURN(rc);
2266
2267                 inode->i_flags = ll_ext_to_inode_flags(flags);
2268
2269                 lsm = ccc_inode_lsm_get(inode);
2270                 if (!lsm_has_objects(lsm)) {
2271                         ccc_inode_lsm_put(inode, lsm);
2272                         RETURN(0);
2273                 }
2274
2275                 OBDO_ALLOC(oinfo.oi_oa);
2276                 if (!oinfo.oi_oa) {
2277                         ccc_inode_lsm_put(inode, lsm);
2278                         RETURN(-ENOMEM);
2279                 }
2280                 oinfo.oi_md = lsm;
2281                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2282                 oinfo.oi_oa->o_flags = flags;
2283                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2284                                        OBD_MD_FLGROUP;
2285                 oinfo.oi_capa = ll_mdscapa_get(inode);
2286                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2287                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2288                 capa_put(oinfo.oi_capa);
2289                 OBDO_FREE(oinfo.oi_oa);
2290                 ccc_inode_lsm_put(inode, lsm);
2291
2292                 if (rc && rc != -EPERM && rc != -EACCES)
2293                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2294
2295                 RETURN(rc);
2296         }
2297         default:
2298                 RETURN(-ENOSYS);
2299         }
2300
2301         RETURN(0);
2302 }
2303
2304 int ll_flush_ctx(struct inode *inode)
2305 {
2306         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2307
2308         CDEBUG(D_SEC, "flush context for user %d\n",
2309                from_kuid(&init_user_ns, current_uid()));
2310
2311         obd_set_info_async(NULL, sbi->ll_md_exp,
2312                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2313                            0, NULL, NULL);
2314         obd_set_info_async(NULL, sbi->ll_dt_exp,
2315                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2316                            0, NULL, NULL);
2317         return 0;
2318 }
2319
2320 /* umount -f client means force down, don't save state */
2321 void ll_umount_begin(struct super_block *sb)
2322 {
2323         struct ll_sb_info *sbi = ll_s2sbi(sb);
2324         struct obd_device *obd;
2325         struct obd_ioctl_data *ioc_data;
2326         ENTRY;
2327
2328         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2329                sb->s_count, atomic_read(&sb->s_active));
2330
2331         obd = class_exp2obd(sbi->ll_md_exp);
2332         if (obd == NULL) {
2333                 CERROR("Invalid MDC connection handle "LPX64"\n",
2334                        sbi->ll_md_exp->exp_handle.h_cookie);
2335                 EXIT;
2336                 return;
2337         }
2338         obd->obd_force = 1;
2339
2340         obd = class_exp2obd(sbi->ll_dt_exp);
2341         if (obd == NULL) {
2342                 CERROR("Invalid LOV connection handle "LPX64"\n",
2343                        sbi->ll_dt_exp->exp_handle.h_cookie);
2344                 EXIT;
2345                 return;
2346         }
2347         obd->obd_force = 1;
2348
2349         OBD_ALLOC_PTR(ioc_data);
2350         if (ioc_data) {
2351                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2352                               sizeof *ioc_data, ioc_data, NULL);
2353
2354                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2355                               sizeof *ioc_data, ioc_data, NULL);
2356
2357                 OBD_FREE_PTR(ioc_data);
2358         }
2359
2360         /* Really, we'd like to wait until there are no requests outstanding,
2361          * and then continue.  For now, we just invalidate the requests,
2362          * schedule() and sleep one second if needed, and hope.
2363          */
2364         schedule();
2365         EXIT;
2366 }
2367
2368 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2369 {
2370         struct ll_sb_info *sbi = ll_s2sbi(sb);
2371         char *profilenm = get_profile_name(sb);
2372         int err;
2373         __u32 read_only;
2374
2375         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2376                 read_only = *flags & MS_RDONLY;
2377                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2378                                          sizeof(KEY_READ_ONLY),
2379                                          KEY_READ_ONLY, sizeof(read_only),
2380                                          &read_only, NULL);
2381                 if (err) {
2382                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2383                                       profilenm, read_only ?
2384                                       "read-only" : "read-write", err);
2385                         return err;
2386                 }
2387
2388                 if (read_only)
2389                         sb->s_flags |= MS_RDONLY;
2390                 else
2391                         sb->s_flags &= ~MS_RDONLY;
2392
2393                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2394                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2395                                       read_only ?  "read-only" : "read-write");
2396         }
2397         return 0;
2398 }
2399
2400 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2401                   struct super_block *sb, struct lookup_intent *it)
2402 {
2403         struct ll_sb_info *sbi = NULL;
2404         struct lustre_md md = { 0 };
2405         int rc;
2406         ENTRY;
2407
2408         LASSERT(*inode || sb);
2409         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2410         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2411                               sbi->ll_md_exp, &md);
2412         if (rc)
2413                 RETURN(rc);
2414
2415         if (*inode) {
2416                 ll_update_inode(*inode, &md);
2417         } else {
2418                 LASSERT(sb != NULL);
2419
2420                 /*
2421                  * At this point server returns to client's same fid as client
2422                  * generated for creating. So using ->fid1 is okay here.
2423                  */
2424                 LASSERT(fid_is_sane(&md.body->fid1));
2425
2426                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2427                                              sbi->ll_flags & LL_SBI_32BIT_API),
2428                                  &md);
2429                 if (*inode == NULL || IS_ERR(*inode)) {
2430 #ifdef CONFIG_FS_POSIX_ACL
2431                         if (md.posix_acl) {
2432                                 posix_acl_release(md.posix_acl);
2433                                 md.posix_acl = NULL;
2434                         }
2435 #endif
2436                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2437                         *inode = NULL;
2438                         CERROR("new_inode -fatal: rc %d\n", rc);
2439                         GOTO(out, rc);
2440                 }
2441         }
2442
2443         /* Handling piggyback layout lock.
2444          * Layout lock can be piggybacked by getattr and open request.
2445          * The lsm can be applied to inode only if it comes with a layout lock
2446          * otherwise correct layout may be overwritten, for example:
2447          * 1. proc1: mdt returns a lsm but not granting layout
2448          * 2. layout was changed by another client
2449          * 3. proc2: refresh layout and layout lock granted
2450          * 4. proc1: to apply a stale layout */
2451         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2452                 struct lustre_handle lockh;
2453                 struct ldlm_lock *lock;
2454
2455                 lockh.cookie = it->d.lustre.it_lock_handle;
2456                 lock = ldlm_handle2lock(&lockh);
2457                 LASSERT(lock != NULL);
2458                 if (ldlm_has_layout(lock)) {
2459                         struct cl_object_conf conf;
2460
2461                         memset(&conf, 0, sizeof(conf));
2462                         conf.coc_opc = OBJECT_CONF_SET;
2463                         conf.coc_inode = *inode;
2464                         conf.coc_lock = lock;
2465                         conf.u.coc_md = &md;
2466                         (void)ll_layout_conf(*inode, &conf);
2467                 }
2468                 LDLM_LOCK_PUT(lock);
2469         }
2470
2471 out:
2472         if (md.lsm != NULL)
2473                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2474         md_free_lustre_md(sbi->ll_md_exp, &md);
2475         RETURN(rc);
2476 }
2477
2478 int ll_obd_statfs(struct inode *inode, void *arg)
2479 {
2480         struct ll_sb_info *sbi = NULL;
2481         struct obd_export *exp;
2482         char *buf = NULL;
2483         struct obd_ioctl_data *data = NULL;
2484         __u32 type;
2485         __u32 flags;
2486         int len = 0, rc;
2487
2488         if (!inode || !(sbi = ll_i2sbi(inode)))
2489                 GOTO(out_statfs, rc = -EINVAL);
2490
2491         rc = obd_ioctl_getdata(&buf, &len, arg);
2492         if (rc)
2493                 GOTO(out_statfs, rc);
2494
2495         data = (void*)buf;
2496         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2497             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2498                 GOTO(out_statfs, rc = -EINVAL);
2499
2500         if (data->ioc_inllen1 != sizeof(__u32) ||
2501             data->ioc_inllen2 != sizeof(__u32) ||
2502             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2503             data->ioc_plen2 != sizeof(struct obd_uuid))
2504                 GOTO(out_statfs, rc = -EINVAL);
2505
2506         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2507         if (type & LL_STATFS_LMV)
2508                 exp = sbi->ll_md_exp;
2509         else if (type & LL_STATFS_LOV)
2510                 exp = sbi->ll_dt_exp;
2511         else
2512                 GOTO(out_statfs, rc = -ENODEV);
2513
2514         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2515         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2516         if (rc)
2517                 GOTO(out_statfs, rc);
2518 out_statfs:
2519         if (buf)
2520                 obd_ioctl_freedata(buf, len);
2521         return rc;
2522 }
2523
2524 int ll_process_config(struct lustre_cfg *lcfg)
2525 {
2526         struct super_block *sb;
2527         unsigned long x;
2528         int rc = 0;
2529         char *ptr;
2530
2531         /* The instance name contains the sb: lustre-client-aacfe000 */
2532         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2533         if (!ptr || !*(++ptr))
2534                 return -EINVAL;
2535         if (sscanf(ptr, "%lx", &x) != 1)
2536                 return -EINVAL;
2537         sb = (struct super_block *)x;
2538         /* This better be a real Lustre superblock! */
2539         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2540
2541         /* Note we have not called client_common_fill_super yet, so
2542            proc fns must be able to handle that! */
2543         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2544                                           lcfg, sb);
2545         if (rc > 0)
2546                 rc = 0;
2547         return rc;
2548 }
2549
2550 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2551 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2552                                        struct inode *i1, struct inode *i2,
2553                                        const char *name, int namelen,
2554                                        int mode, __u32 opc, void *data)
2555 {
2556         LASSERT(i1 != NULL);
2557
2558         if (namelen > ll_i2sbi(i1)->ll_namelen)
2559                 return ERR_PTR(-ENAMETOOLONG);
2560
2561         if (op_data == NULL)
2562                 OBD_ALLOC_PTR(op_data);
2563
2564         if (op_data == NULL)
2565                 return ERR_PTR(-ENOMEM);
2566
2567         ll_i2gids(op_data->op_suppgids, i1, i2);
2568         op_data->op_fid1 = *ll_inode2fid(i1);
2569         op_data->op_capa1 = ll_mdscapa_get(i1);
2570         if (S_ISDIR(i1->i_mode))
2571                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2572
2573         if (i2) {
2574                 op_data->op_fid2 = *ll_inode2fid(i2);
2575                 op_data->op_capa2 = ll_mdscapa_get(i2);
2576                 if (S_ISDIR(i2->i_mode))
2577                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2578         } else {
2579                 fid_zero(&op_data->op_fid2);
2580                 op_data->op_capa2 = NULL;
2581         }
2582
2583         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2584                 op_data->op_cli_flags |= CLI_HASH64;
2585
2586         if (ll_need_32bit_api(ll_i2sbi(i1)))
2587                 op_data->op_cli_flags |= CLI_API32;
2588
2589         op_data->op_name = name;
2590         op_data->op_namelen = namelen;
2591         op_data->op_mode = mode;
2592         op_data->op_mod_time = cfs_time_current_sec();
2593         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2594         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2595         op_data->op_cap = cfs_curproc_cap_pack();
2596         op_data->op_bias = 0;
2597         op_data->op_cli_flags = 0;
2598         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2599              filename_is_volatile(name, namelen, NULL))
2600                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2601         op_data->op_mds = 0;
2602         op_data->op_data = data;
2603
2604         /* If the file is being opened after mknod() (normally due to NFS)
2605          * try to use the default stripe data from parent directory for
2606          * allocating OST objects.  Try to pass the parent FID to MDS. */
2607         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2608             !ll_i2info(i2)->lli_has_smd) {
2609                 struct ll_inode_info *lli = ll_i2info(i2);
2610
2611                 spin_lock(&lli->lli_lock);
2612                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2613                         op_data->op_fid1 = lli->lli_pfid;
2614                 spin_unlock(&lli->lli_lock);
2615                 /** We ignore parent's capability temporary. */
2616         }
2617
2618         /* When called by ll_setattr_raw, file is i1. */
2619         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2620                 op_data->op_bias |= MDS_DATA_MODIFIED;
2621
2622         return op_data;
2623 }
2624
2625 void ll_finish_md_op_data(struct md_op_data *op_data)
2626 {
2627         capa_put(op_data->op_capa1);
2628         capa_put(op_data->op_capa2);
2629         OBD_FREE_PTR(op_data);
2630 }
2631
2632 #ifdef HAVE_SUPEROPS_USE_DENTRY
2633 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2634 #else
2635 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2636 #endif
2637 {
2638         struct ll_sb_info *sbi;
2639
2640 #ifdef HAVE_SUPEROPS_USE_DENTRY
2641         LASSERT((seq != NULL) && (dentry != NULL));
2642         sbi = ll_s2sbi(dentry->d_sb);
2643 #else
2644         LASSERT((seq != NULL) && (vfs != NULL));
2645         sbi = ll_s2sbi(vfs->mnt_sb);
2646 #endif
2647
2648         if (sbi->ll_flags & LL_SBI_NOLCK)
2649                 seq_puts(seq, ",nolock");
2650
2651         if (sbi->ll_flags & LL_SBI_FLOCK)
2652                 seq_puts(seq, ",flock");
2653
2654         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2655                 seq_puts(seq, ",localflock");
2656
2657         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2658                 seq_puts(seq, ",user_xattr");
2659
2660         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2661                 seq_puts(seq, ",lazystatfs");
2662
2663         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2664                 seq_puts(seq, ",user_fid2path");
2665
2666         RETURN(0);
2667 }
2668
2669 /**
2670  * Get obd name by cmd, and copy out to user space
2671  */
2672 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2673 {
2674         struct ll_sb_info *sbi = ll_i2sbi(inode);
2675         struct obd_device *obd;
2676         ENTRY;
2677
2678         if (cmd == OBD_IOC_GETDTNAME)
2679                 obd = class_exp2obd(sbi->ll_dt_exp);
2680         else if (cmd == OBD_IOC_GETMDNAME)
2681                 obd = class_exp2obd(sbi->ll_md_exp);
2682         else
2683                 RETURN(-EINVAL);
2684
2685         if (!obd)
2686                 RETURN(-ENOENT);
2687
2688         if (copy_to_user((void *)arg, obd->obd_name,
2689                          strlen(obd->obd_name) + 1))
2690                 RETURN(-EFAULT);
2691
2692         RETURN(0);
2693 }
2694
2695 /**
2696  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2697  * fsname will be returned in this buffer; otherwise, a static buffer will be
2698  * used to store the fsname and returned to caller.
2699  */
2700 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2701 {
2702         static char fsname_static[MTI_NAME_MAXLEN];
2703         struct lustre_sb_info *lsi = s2lsi(sb);
2704         char *ptr;
2705         int len;
2706
2707         if (buf == NULL) {
2708                 /* this means the caller wants to use static buffer
2709                  * and it doesn't care about race. Usually this is
2710                  * in error reporting path */
2711                 buf = fsname_static;
2712                 buflen = sizeof(fsname_static);
2713         }
2714
2715         len = strlen(lsi->lsi_lmd->lmd_profile);
2716         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2717         if (ptr && (strcmp(ptr, "-client") == 0))
2718                 len -= 7;
2719
2720         if (unlikely(len >= buflen))
2721                 len = buflen - 1;
2722         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2723         buf[len] = '\0';
2724
2725         return buf;
2726 }
2727
2728 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2729 {
2730         char *path = NULL;
2731
2732         struct path p;
2733
2734         p.dentry = dentry;
2735         p.mnt = current->fs->root.mnt;
2736         path_get(&p);
2737         path = d_path(&p, buf, bufsize);
2738         path_put(&p);
2739         return path;
2740 }
2741
2742 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2743 {
2744         char *buf, *path = NULL;
2745         struct dentry *dentry = NULL;
2746         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2747
2748         /* this can be called inside spin lock so use GFP_ATOMIC. */
2749         buf = (char *)__get_free_page(GFP_ATOMIC);
2750         if (buf != NULL) {
2751                 dentry = d_find_alias(page->mapping->host);
2752                 if (dentry != NULL)
2753                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2754         }
2755
2756         CDEBUG(D_WARNING,
2757                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2758                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2759                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2760                PFID(&obj->cob_header.coh_lu.loh_fid),
2761                (path && !IS_ERR(path)) ? path : "", ioret);
2762
2763         if (dentry != NULL)
2764                 dput(dentry);
2765
2766         if (buf != NULL)
2767                 free_page((unsigned long)buf);
2768 }