lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  97                 lru_page_max = pages / 2;
  98         } else {
  99                 lru_page_max = (pages / 4) * 3;
 100         }
 101
 102         /* initialize lru data */
 103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 104         sbi->ll_cache.ccc_lru_max = lru_page_max;
 105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 108
 109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 110                                            SBI_DEFAULT_READAHEAD_MAX);
 111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 116
 117         ll_generate_random_uuid(uuid);
 118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 120
 121         spin_lock(&ll_sb_lock);
 122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 123         spin_unlock(&ll_sb_lock);
 124
 125         sbi->ll_flags |= LL_SBI_VERBOSE;
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 136                                pp_r_hist.oh_lock);
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145         cfs_atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147
 148         RETURN(sbi);
 149 }
 150
 151 void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 spin_lock(&ll_sb_lock);
 158                 cfs_list_del(&sbi->ll_list);
 159                 spin_unlock(&ll_sb_lock);
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static struct dentry_operations ll_d_root_ops = {
 166         .d_compare = ll_dcompare,
 167         .d_revalidate = ll_revalidate_nd,
 168 };
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = 0;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         if (proc_lustre_fs_root) {
 204                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 205                                                   dt, md);
 206                 if (err < 0)
 207                         CERROR("could not register mount in /proc/fs/lustre\n");
 208         }
 209
 210         /* indicate the features supported by this client */
 211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 212                                   OBD_CONNECT_ATTRFID  |
 213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 214                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 215                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 216                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 217                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 218                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 219                                   OBD_CONNECT_EINPROGRESS |
 220                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 221                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 222
 223         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 224                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 225
 226 #ifdef HAVE_LRU_RESIZE_SUPPORT
 227         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 228                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 229 #endif
 230 #ifdef CONFIG_FS_POSIX_ACL
 231         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 232 #endif
 233
 234         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 235                 /* flag mdc connection as lightweight, only used for test
 236                  * purpose, use with care */
 237                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 238
 239         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 240         data->ocd_version = LUSTRE_VERSION_CODE;
 241
 242         if (sb->s_flags & MS_RDONLY)
 243                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 244         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 245                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 246
 247 #ifdef HAVE_MS_FLOCK_LOCK
 248         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 249         sb->s_flags |= MS_FLOCK_LOCK;
 250 #endif
 251 #ifdef MS_HAS_NEW_AOPS
 252         sb->s_flags |= MS_HAS_NEW_AOPS;
 253 #endif
 254
 255         if (sbi->ll_flags & LL_SBI_FLOCK)
 256                 sbi->ll_fop = &ll_file_operations_flock;
 257         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 258                 sbi->ll_fop = &ll_file_operations;
 259         else
 260                 sbi->ll_fop = &ll_file_operations_noflock;
 261
 262         /* real client */
 263         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 264         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 265                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 266
 267         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 268
 269         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 270         if (err == -EBUSY) {
 271                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 272                                    "recovery, of which this client is not a "
 273                                    "part. Please wait for recovery to complete,"
 274                                    " abort, or time out.\n", md);
 275                 GOTO(out, err);
 276         } else if (err) {
 277                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 278                 GOTO(out, err);
 279         }
 280
 281         sbi->ll_md_exp->exp_connect_data = *data;
 282
 283         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 284                            LUSTRE_SEQ_METADATA);
 285         if (err) {
 286                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 287                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 288                 GOTO(out_md, err);
 289         }
 290
 291         /* For mount, we only need fs info from MDT0, and also in DNE, it
 292          * can make sure the client can be mounted as long as MDT0 is
 293          * avaible */
 294         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 295                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 296                         OBD_STATFS_FOR_MDT0);
 297         if (err)
 298                 GOTO(out_md_fid, err);
 299
 300         /* This needs to be after statfs to ensure connect has finished.
 301          * Note that "data" does NOT contain the valid connect reply.
 302          * If connecting to a 1.8 server there will be no LMV device, so
 303          * we can access the MDC export directly and exp_connect_flags will
 304          * be non-zero, but if accessing an upgraded 2.1 server it will
 305          * have the correct flags filled in.
 306          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 307         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 308         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 309             valid != CLIENT_CONNECT_MDT_REQD) {
 310                 char *buf;
 311
 312                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 313                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 314                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 315                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 316                                    "feature(s) needed for correct operation "
 317                                    "of this client (%s). Please upgrade "
 318                                    "server or downgrade client.\n",
 319                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 320                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 321                 GOTO(out_md_fid, err = -EPROTO);
 322         }
 323
 324         size = sizeof(*data);
 325         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 326                            KEY_CONN_DATA,  &size, data, NULL);
 327         if (err) {
 328                 CERROR("%s: Get connect data failed: rc = %d\n",
 329                        sbi->ll_md_exp->exp_obd->obd_name, err);
 330                 GOTO(out_md_fid, err);
 331         }
 332
 333         LASSERT(osfs->os_bsize);
 334         sb->s_blocksize = osfs->os_bsize;
 335         sb->s_blocksize_bits = log2(osfs->os_bsize);
 336         sb->s_magic = LL_SUPER_MAGIC;
 337         sb->s_maxbytes = MAX_LFS_FILESIZE;
 338         sbi->ll_namelen = osfs->os_namelen;
 339         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 340
 341         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 342             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 343                 LCONSOLE_INFO("Disabling user_xattr feature because "
 344                               "it is not supported on the server\n");
 345                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 349 #ifdef MS_POSIXACL
 350                 sb->s_flags |= MS_POSIXACL;
 351 #endif
 352                 sbi->ll_flags |= LL_SBI_ACL;
 353         } else {
 354                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 355 #ifdef MS_POSIXACL
 356                 sb->s_flags &= ~MS_POSIXACL;
 357 #endif
 358                 sbi->ll_flags &= ~LL_SBI_ACL;
 359         }
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 362                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 363                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client is set as remote by default.\n");
 365                 }
 366         } else {
 367                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 368                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 369                         LCONSOLE_INFO("client claims to be remote, but server "
 370                                       "rejected, forced to be local.\n");
 371                 }
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 375                 LCONSOLE_INFO("client enabled MDS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 380                 LCONSOLE_INFO("client enabled OSS capability!\n");
 381                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 382         }
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 385                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 388                 sbi->ll_md_brw_size = data->ocd_brw_size;
 389         else
 390                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 391
 392         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 393                 LCONSOLE_INFO("Layout lock feature supported.\n");
 394                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 395         }
 396
 397         obd = class_name2obd(dt);
 398         if (!obd) {
 399                 CERROR("DT %s: not setup or attached\n", dt);
 400                 GOTO(out_md_fid, err = -ENODEV);
 401         }
 402
 403         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 404                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 405                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 406                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 407                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 408                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 409                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 410                                   OBD_CONNECT_MAXBYTES |
 411                                   OBD_CONNECT_EINPROGRESS |
 412                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 413                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 414
 415         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 416                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 417
 418         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 419                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 420                  * disabled by default, because it can still be enabled on the
 421                  * fly via /proc. As a consequence, we still need to come to an
 422                  * agreement on the supported algorithms at connect time */
 423                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 424
 425                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 426                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 427                 else
 428                         data->ocd_cksum_types = cksum_types_supported_client();
 429         }
 430
 431 #ifdef HAVE_LRU_RESIZE_SUPPORT
 432         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 433 #endif
 434         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 435                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 436
 437         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 438                "ocd_grant: %d\n", data->ocd_connect_flags,
 439                data->ocd_version, data->ocd_grant);
 440
 441         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 442         obd->obd_upcall.onu_upcall = cl_ocd_update;
 443
 444         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 445
 446         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 447                           NULL);
 448         if (err == -EBUSY) {
 449                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 450                                    "recovery, of which this client is not a "
 451                                    "part.  Please wait for recovery to "
 452                                    "complete, abort, or time out.\n", dt);
 453                 GOTO(out_md, err);
 454         } else if (err) {
 455                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 456                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 457                 GOTO(out_md, err);
 458         }
 459
 460         sbi->ll_dt_exp->exp_connect_data = *data;
 461
 462         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 463                            LUSTRE_SEQ_METADATA);
 464         if (err) {
 465                 CERROR("%s: Can't init data layer FID infrastructure, "
 466                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 467                 GOTO(out_dt, err);
 468         }
 469
 470         mutex_lock(&sbi->ll_lco.lco_lock);
 471         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 472         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 473         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 474         mutex_unlock(&sbi->ll_lco.lco_lock);
 475
 476         fid_zero(&sbi->ll_root_fid);
 477         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 478         if (err) {
 479                 CERROR("cannot mds_connect: rc = %d\n", err);
 480                 GOTO(out_lock_cn_cb, err);
 481         }
 482         if (!fid_is_sane(&sbi->ll_root_fid)) {
 483                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 484                        sbi->ll_md_exp->exp_obd->obd_name,
 485                        PFID(&sbi->ll_root_fid));
 486                 GOTO(out_lock_cn_cb, err = -EINVAL);
 487         }
 488         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 489
 490         sb->s_op = &lustre_super_operations;
 491 #if THREAD_SIZE >= 8192 /*b=17630*/
 492         sb->s_export_op = &lustre_export_operations;
 493 #endif
 494
 495         /* make root inode
 496          * XXX: move this to after cbd setup? */
 497         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 498         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 499                 valid |= OBD_MD_FLRMTPERM;
 500         else if (sbi->ll_flags & LL_SBI_ACL)
 501                 valid |= OBD_MD_FLACL;
 502
 503         OBD_ALLOC_PTR(op_data);
 504         if (op_data == NULL)
 505                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 506
 507         op_data->op_fid1 = sbi->ll_root_fid;
 508         op_data->op_mode = 0;
 509         op_data->op_capa1 = oc;
 510         op_data->op_valid = valid;
 511
 512         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 513         if (oc)
 514                 capa_put(oc);
 515         OBD_FREE_PTR(op_data);
 516         if (err) {
 517                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 518                        sbi->ll_md_exp->exp_obd->obd_name, err);
 519                 GOTO(out_lock_cn_cb, err);
 520         }
 521
 522         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 523                                sbi->ll_md_exp, &lmd);
 524         if (err) {
 525                 CERROR("failed to understand root inode md: rc = %d\n", err);
 526                 ptlrpc_req_finished(request);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 531         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 532                                             sbi->ll_flags & LL_SBI_32BIT_API),
 533                        &lmd);
 534         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 535         ptlrpc_req_finished(request);
 536
 537         if (root == NULL || IS_ERR(root)) {
 538                 if (lmd.lsm)
 539                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 540 #ifdef CONFIG_FS_POSIX_ACL
 541                 if (lmd.posix_acl) {
 542                         posix_acl_release(lmd.posix_acl);
 543                         lmd.posix_acl = NULL;
 544                 }
 545 #endif
 546                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 547                 root = NULL;
 548                 CERROR("lustre_lite: bad iget4 for root\n");
 549                 GOTO(out_root, err);
 550         }
 551
 552         err = ll_close_thread_start(&sbi->ll_lcq);
 553         if (err) {
 554                 CERROR("cannot start close thread: rc %d\n", err);
 555                 GOTO(out_root, err);
 556         }
 557
 558 #ifdef CONFIG_FS_POSIX_ACL
 559         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 560                 rct_init(&sbi->ll_rct);
 561                 et_init(&sbi->ll_et);
 562         }
 563 #endif
 564
 565         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 566         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 567                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 568                                  NULL);
 569         cl_sb_init(sb);
 570
 571         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 572                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 573                                  &sbi->ll_cache, NULL);
 574
 575         sb->s_root = d_make_root(root);
 576         if (sb->s_root == NULL) {
 577                 CERROR("%s: can't make root dentry\n",
 578                         ll_get_fsname(sb, NULL, 0));
 579                 GOTO(out_root, err = -ENOMEM);
 580         }
 581
 582 #ifdef HAVE_DCACHE_LOCK
 583         sb->s_root->d_op = &ll_d_root_ops;
 584 #else
 585         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 586         d_set_d_op(sb->s_root, &ll_d_root_ops);
 587         sb->s_d_op = &ll_d_ops;
 588 #endif
 589
 590         sbi->ll_sdev_orig = sb->s_dev;
 591
 592         /* We set sb->s_dev equal on all lustre clients in order to support
 593          * NFS export clustering.  NFSD requires that the FSID be the same
 594          * on all clients. */
 595         /* s_dev is also used in lt_compare() to compare two fs, but that is
 596          * only a node-local comparison. */
 597         uuid = obd_get_uuid(sbi->ll_md_exp);
 598         if (uuid != NULL) {
 599                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 600                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 601         }
 602
 603         if (data != NULL)
 604                 OBD_FREE_PTR(data);
 605         if (osfs != NULL)
 606                 OBD_FREE_PTR(osfs);
 607
 608         RETURN(err);
 609 out_root:
 610         if (root)
 611                 iput(root);
 612 out_lock_cn_cb:
 613         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 614 out_dt:
 615         obd_disconnect(sbi->ll_dt_exp);
 616         sbi->ll_dt_exp = NULL;
 617         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 618         obd_zombie_barrier();
 619 out_md_fid:
 620         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 621 out_md:
 622         obd_disconnect(sbi->ll_md_exp);
 623         sbi->ll_md_exp = NULL;
 624 out:
 625         if (data != NULL)
 626                 OBD_FREE_PTR(data);
 627         if (osfs != NULL)
 628                 OBD_FREE_PTR(osfs);
 629         lprocfs_unregister_mountpoint(sbi);
 630         return err;
 631 }
 632
 633 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 634 {
 635         int size, rc;
 636
 637         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 638         size = sizeof(int);
 639         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 640                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 641         if (rc)
 642                 CERROR("Get max mdsize error rc %d \n", rc);
 643
 644         RETURN(rc);
 645 }
 646
 647 void ll_dump_inode(struct inode *inode)
 648 {
 649         struct ll_d_hlist_node *tmp;
 650         int dentry_count = 0;
 651
 652         LASSERT(inode != NULL);
 653
 654         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 655                 dentry_count++;
 656
 657         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 658                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 659                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 660 }
 661
 662 void lustre_dump_dentry(struct dentry *dentry, int recur)
 663 {
 664         struct list_head *tmp;
 665         int subdirs = 0;
 666
 667         LASSERT(dentry != NULL);
 668
 669         list_for_each(tmp, &dentry->d_subdirs)
 670                 subdirs++;
 671
 672         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 673                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 674                dentry->d_name.len, dentry->d_name.name,
 675                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 676                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 677                dentry->d_flags, dentry->d_fsdata, subdirs);
 678         if (dentry->d_inode != NULL)
 679                 ll_dump_inode(dentry->d_inode);
 680
 681         if (recur == 0)
 682                 return;
 683
 684         list_for_each(tmp, &dentry->d_subdirs) {
 685                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 686                 lustre_dump_dentry(d, recur - 1);
 687         }
 688 }
 689
 690 void client_common_put_super(struct super_block *sb)
 691 {
 692         struct ll_sb_info *sbi = ll_s2sbi(sb);
 693         ENTRY;
 694
 695 #ifdef CONFIG_FS_POSIX_ACL
 696         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 697                 et_fini(&sbi->ll_et);
 698                 rct_fini(&sbi->ll_rct);
 699         }
 700 #endif
 701
 702         ll_close_thread_shutdown(sbi->ll_lcq);
 703
 704         cl_sb_fini(sb);
 705
 706         cfs_list_del(&sbi->ll_conn_chain);
 707
 708         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 709         obd_disconnect(sbi->ll_dt_exp);
 710         sbi->ll_dt_exp = NULL;
 711         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 712          * see LU-2543. */
 713         obd_zombie_barrier();
 714
 715         lprocfs_unregister_mountpoint(sbi);
 716
 717         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 718         obd_disconnect(sbi->ll_md_exp);
 719         sbi->ll_md_exp = NULL;
 720
 721         EXIT;
 722 }
 723
 724 void ll_kill_super(struct super_block *sb)
 725 {
 726         struct ll_sb_info *sbi;
 727
 728         ENTRY;
 729
 730         /* not init sb ?*/
 731         if (!(sb->s_flags & MS_ACTIVE))
 732                 return;
 733
 734         sbi = ll_s2sbi(sb);
 735         /* we need restore s_dev from changed for clustred NFS before put_super
 736          * because new kernels have cached s_dev and change sb->s_dev in
 737          * put_super not affected real removing devices */
 738         if (sbi) {
 739                 sb->s_dev = sbi->ll_sdev_orig;
 740                 sbi->ll_umounting = 1;
 741         }
 742         EXIT;
 743 }
 744
 745 char *ll_read_opt(const char *opt, char *data)
 746 {
 747         char *value;
 748         char *retval;
 749         ENTRY;
 750
 751         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 752         if (strncmp(opt, data, strlen(opt)))
 753                 RETURN(NULL);
 754         if ((value = strchr(data, '=')) == NULL)
 755                 RETURN(NULL);
 756
 757         value++;
 758         OBD_ALLOC(retval, strlen(value) + 1);
 759         if (!retval) {
 760                 CERROR("out of memory!\n");
 761                 RETURN(NULL);
 762         }
 763
 764         memcpy(retval, value, strlen(value)+1);
 765         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 766         RETURN(retval);
 767 }
 768
 769 static inline int ll_set_opt(const char *opt, char *data, int fl)
 770 {
 771         if (strncmp(opt, data, strlen(opt)) != 0)
 772                 return(0);
 773         else
 774                 return(fl);
 775 }
 776
 777 /* non-client-specific mount options are parsed in lmd_parse */
 778 static int ll_options(char *options, int *flags)
 779 {
 780         int tmp;
 781         char *s1 = options, *s2;
 782         ENTRY;
 783
 784         if (!options)
 785                 RETURN(0);
 786
 787         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 788
 789         while (*s1) {
 790                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 791                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 822                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 823                 if (tmp) {
 824                         /* Ignore deprecated mount option.  The client will
 825                          * always try to mount with ACL support, whether this
 826                          * is used depends on whether server supports it. */
 827                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 828                                                   "mount option 'acl'.\n");
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 832                 if (tmp) {
 833                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 834                                                   "mount option 'noacl'.\n");
 835                         goto next;
 836                 }
 837 #else
 838 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 839 #endif
 840                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 846                 if (tmp) {
 847                         *flags |= tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 851                 if (tmp) {
 852                         *flags &= ~tmp;
 853                         goto next;
 854                 }
 855
 856                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 862                 if (tmp) {
 863                         *flags &= ~tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 902                 if (tmp) {
 903                         *flags &= ~tmp;
 904                         goto next;
 905                 }
 906                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 907                                    s1);
 908                 RETURN(-EINVAL);
 909
 910 next:
 911                 /* Find next opt */
 912                 s2 = strchr(s1, ',');
 913                 if (s2 == NULL)
 914                         break;
 915                 s1 = s2 + 1;
 916         }
 917         RETURN(0);
 918 }
 919
 920 void ll_lli_init(struct ll_inode_info *lli)
 921 {
 922         lli->lli_inode_magic = LLI_INODE_MAGIC;
 923         lli->lli_flags = 0;
 924         lli->lli_ioepoch = 0;
 925         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 926         spin_lock_init(&lli->lli_lock);
 927         lli->lli_posix_acl = NULL;
 928         lli->lli_remote_perms = NULL;
 929         mutex_init(&lli->lli_rmtperm_mutex);
 930         /* Do not set lli_fid, it has been initialized already. */
 931         fid_zero(&lli->lli_pfid);
 932         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 933         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 934         cfs_atomic_set(&lli->lli_open_count, 0);
 935         lli->lli_mds_capa = NULL;
 936         lli->lli_rmtperm_time = 0;
 937         lli->lli_pending_och = NULL;
 938         lli->lli_mds_read_och = NULL;
 939         lli->lli_mds_write_och = NULL;
 940         lli->lli_mds_exec_och = NULL;
 941         lli->lli_open_fd_read_count = 0;
 942         lli->lli_open_fd_write_count = 0;
 943         lli->lli_open_fd_exec_count = 0;
 944         mutex_init(&lli->lli_och_mutex);
 945         spin_lock_init(&lli->lli_agl_lock);
 946         lli->lli_has_smd = false;
 947         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 948         lli->lli_clob = NULL;
 949
 950         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 951         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 952                 mutex_init(&lli->lli_readdir_mutex);
 953                 lli->lli_opendir_key = NULL;
 954                 lli->lli_sai = NULL;
 955                 lli->lli_def_acl = NULL;
 956                 spin_lock_init(&lli->lli_sa_lock);
 957                 lli->lli_opendir_pid = 0;
 958         } else {
 959                 sema_init(&lli->lli_size_sem, 1);
 960                 lli->lli_size_sem_owner = NULL;
 961                 lli->lli_symlink_name = NULL;
 962                 init_rwsem(&lli->lli_trunc_sem);
 963                 mutex_init(&lli->lli_write_mutex);
 964                 init_rwsem(&lli->lli_glimpse_sem);
 965                 lli->lli_glimpse_time = 0;
 966                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 967                 lli->lli_agl_index = 0;
 968                 lli->lli_async_rc = 0;
 969                 lli->lli_volatile = false;
 970         }
 971         mutex_init(&lli->lli_layout_mutex);
 972 }
 973
 974 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 975 {
 976         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 977
 978         bdi->name = "lustre";
 979         return bdi_register(bdi, NULL, "lustre-%d",
 980                             atomic_inc_return(&ll_bdi_num));
 981 }
 982
 983 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 984 {
 985         struct lustre_profile *lprof = NULL;
 986         struct lustre_sb_info *lsi = s2lsi(sb);
 987         struct ll_sb_info *sbi;
 988         char  *dt = NULL, *md = NULL;
 989         char  *profilenm = get_profile_name(sb);
 990         struct config_llog_instance *cfg;
 991         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 992         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 993         int    err;
 994         ENTRY;
 995
 996         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 997
 998         OBD_ALLOC_PTR(cfg);
 999         if (cfg == NULL)
1000                 RETURN(-ENOMEM);
1001
1002         cfs_module_get();
1003
1004         /* client additional sb info */
1005         lsi->lsi_llsbi = sbi = ll_init_sbi();
1006         if (!sbi) {
1007                 cfs_module_put(THIS_MODULE);
1008                 OBD_FREE_PTR(cfg);
1009                 RETURN(-ENOMEM);
1010         }
1011
1012         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1013         if (err)
1014                 GOTO(out_free, err);
1015
1016         err = bdi_init(&lsi->lsi_bdi);
1017         if (err)
1018                 GOTO(out_free, err);
1019         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1020         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1021         err = ll_bdi_register(&lsi->lsi_bdi);
1022         if (err)
1023                 GOTO(out_free, err);
1024
1025         sb->s_bdi = &lsi->lsi_bdi;
1026
1027         /* Generate a string unique to this super, in case some joker tries
1028            to mount the same fs at two mount points.
1029            Use the address of the super itself.*/
1030         cfg->cfg_instance = sb;
1031         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1032         cfg->cfg_callback = class_config_llog_handler;
1033         /* set up client obds */
1034         err = lustre_process_log(sb, profilenm, cfg);
1035         if (err < 0) {
1036                 CERROR("Unable to process log: %d\n", err);
1037                 GOTO(out_free, err);
1038         }
1039
1040         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1041         lprof = class_get_profile(profilenm);
1042         if (lprof == NULL) {
1043                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1044                                    " read from the MGS.  Does that filesystem "
1045                                    "exist?\n", profilenm);
1046                 GOTO(out_free, err = -EINVAL);
1047         }
1048         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1049                lprof->lp_md, lprof->lp_dt);
1050
1051         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1052         if (!dt)
1053                 GOTO(out_free, err = -ENOMEM);
1054         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1055
1056         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1057         if (!md)
1058                 GOTO(out_free, err = -ENOMEM);
1059         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1060
1061         /* connections, registrations, sb setup */
1062         err = client_common_fill_super(sb, md, dt, mnt);
1063
1064 out_free:
1065         if (md)
1066                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1067         if (dt)
1068                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1069         if (err)
1070                 ll_put_super(sb);
1071         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1072                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1073
1074         OBD_FREE_PTR(cfg);
1075         RETURN(err);
1076 } /* ll_fill_super */
1077
1078
1079 void lu_context_keys_dump(void);
1080
1081 void ll_put_super(struct super_block *sb)
1082 {
1083         struct config_llog_instance cfg;
1084         struct obd_device *obd;
1085         struct lustre_sb_info *lsi = s2lsi(sb);
1086         struct ll_sb_info *sbi = ll_s2sbi(sb);
1087         char *profilenm = get_profile_name(sb);
1088         int force = 1, next;
1089         ENTRY;
1090
1091         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1092
1093         ll_print_capa_stat(sbi);
1094
1095         cfg.cfg_instance = sb;
1096         lustre_end_log(sb, profilenm, &cfg);
1097
1098         if (sbi->ll_md_exp) {
1099                 obd = class_exp2obd(sbi->ll_md_exp);
1100                 if (obd)
1101                         force = obd->obd_force;
1102         }
1103
1104         /* We need to set force before the lov_disconnect in
1105            lustre_common_put_super, since l_d cleans up osc's as well. */
1106         if (force) {
1107                 next = 0;
1108                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1109                                                      &next)) != NULL) {
1110                         obd->obd_force = force;
1111                 }
1112         }
1113
1114         if (sbi->ll_lcq) {
1115                 /* Only if client_common_fill_super succeeded */
1116                 client_common_put_super(sb);
1117         }
1118
1119         next = 0;
1120         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1121                 class_manual_cleanup(obd);
1122         }
1123
1124         if (sbi->ll_flags & LL_SBI_VERBOSE)
1125                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1126
1127         if (profilenm)
1128                 class_del_profile(profilenm);
1129
1130         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1131                 bdi_destroy(&lsi->lsi_bdi);
1132                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1133         }
1134
1135         ll_free_sbi(sb);
1136         lsi->lsi_llsbi = NULL;
1137
1138         lustre_common_put_super(sb);
1139
1140         cfs_module_put(THIS_MODULE);
1141
1142         EXIT;
1143 } /* client_put_super */
1144
1145 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1146 {
1147         struct inode *inode = NULL;
1148
1149         /* NOTE: we depend on atomic igrab() -bzzz */
1150         lock_res_and_lock(lock);
1151         if (lock->l_resource->lr_lvb_inode) {
1152                 struct ll_inode_info * lli;
1153                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1154                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1155                         inode = igrab(lock->l_resource->lr_lvb_inode);
1156                 } else {
1157                         inode = lock->l_resource->lr_lvb_inode;
1158                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1159                                          D_WARNING, lock, "lr_lvb_inode %p is "
1160                                          "bogus: magic %08x",
1161                                          lock->l_resource->lr_lvb_inode,
1162                                          lli->lli_inode_magic);
1163                         inode = NULL;
1164                 }
1165         }
1166         unlock_res_and_lock(lock);
1167         return inode;
1168 }
1169
1170 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1171 {
1172         struct inode *inode = NULL;
1173         /* NOTE: we depend on atomic igrab() -bzzz */
1174         lock_res_and_lock(lock);
1175         if (lock->l_ast_data) {
1176                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1177                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1178                         inode = igrab(lock->l_ast_data);
1179                 } else {
1180                         inode = lock->l_ast_data;
1181                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1182                                          D_WARNING, lock, "l_ast_data %p is "
1183                                          "bogus: magic %08x", lock->l_ast_data,
1184                                          lli->lli_inode_magic);
1185                         inode = NULL;
1186                 }
1187         }
1188         unlock_res_and_lock(lock);
1189         return inode;
1190 }
1191
1192 void ll_clear_inode(struct inode *inode)
1193 {
1194         struct ll_inode_info *lli = ll_i2info(inode);
1195         struct ll_sb_info *sbi = ll_i2sbi(inode);
1196         ENTRY;
1197
1198         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1199                inode->i_generation, inode);
1200
1201         if (S_ISDIR(inode->i_mode)) {
1202                 /* these should have been cleared in ll_file_release */
1203                 LASSERT(lli->lli_opendir_key == NULL);
1204                 LASSERT(lli->lli_sai == NULL);
1205                 LASSERT(lli->lli_opendir_pid == 0);
1206         }
1207
1208         spin_lock(&lli->lli_lock);
1209         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1210         spin_unlock(&lli->lli_lock);
1211         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1212
1213         LASSERT(!lli->lli_open_fd_write_count);
1214         LASSERT(!lli->lli_open_fd_read_count);
1215         LASSERT(!lli->lli_open_fd_exec_count);
1216
1217         if (lli->lli_mds_write_och)
1218                 ll_md_real_close(inode, FMODE_WRITE);
1219         if (lli->lli_mds_exec_och)
1220                 ll_md_real_close(inode, FMODE_EXEC);
1221         if (lli->lli_mds_read_och)
1222                 ll_md_real_close(inode, FMODE_READ);
1223
1224         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1225                 OBD_FREE(lli->lli_symlink_name,
1226                          strlen(lli->lli_symlink_name) + 1);
1227                 lli->lli_symlink_name = NULL;
1228         }
1229
1230         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1231                 LASSERT(lli->lli_posix_acl == NULL);
1232                 if (lli->lli_remote_perms) {
1233                         free_rmtperm_hash(lli->lli_remote_perms);
1234                         lli->lli_remote_perms = NULL;
1235                 }
1236         }
1237 #ifdef CONFIG_FS_POSIX_ACL
1238         else if (lli->lli_posix_acl) {
1239                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1240                 LASSERT(lli->lli_remote_perms == NULL);
1241                 posix_acl_release(lli->lli_posix_acl);
1242                 lli->lli_posix_acl = NULL;
1243         }
1244 #endif
1245         lli->lli_inode_magic = LLI_INODE_DEAD;
1246
1247         ll_clear_inode_capas(inode);
1248         if (!S_ISDIR(inode->i_mode))
1249                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1250
1251         /*
1252          * XXX This has to be done before lsm is freed below, because
1253          * cl_object still uses inode lsm.
1254          */
1255         cl_inode_fini(inode);
1256         lli->lli_has_smd = false;
1257
1258         EXIT;
1259 }
1260
1261 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1262                   struct md_open_data **mod)
1263 {
1264         struct lustre_md md;
1265         struct inode *inode = dentry->d_inode;
1266         struct ll_sb_info *sbi = ll_i2sbi(inode);
1267         struct ptlrpc_request *request = NULL;
1268         int rc, ia_valid;
1269         ENTRY;
1270
1271         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1272                                      LUSTRE_OPC_ANY, NULL);
1273         if (IS_ERR(op_data))
1274                 RETURN(PTR_ERR(op_data));
1275
1276         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1277                         &request, mod);
1278         if (rc) {
1279                 ptlrpc_req_finished(request);
1280                 if (rc == -ENOENT) {
1281                         clear_nlink(inode);
1282                         /* Unlinked special device node? Or just a race?
1283                          * Pretend we done everything. */
1284                         if (!S_ISREG(inode->i_mode) &&
1285                             !S_ISDIR(inode->i_mode)) {
1286                                 ia_valid = op_data->op_attr.ia_valid;
1287                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1288                                 rc = simple_setattr(dentry, &op_data->op_attr);
1289                                 op_data->op_attr.ia_valid = ia_valid;
1290                         }
1291                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1292                         CERROR("md_setattr fails: rc = %d\n", rc);
1293                 }
1294                 RETURN(rc);
1295         }
1296
1297         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1298                               sbi->ll_md_exp, &md);
1299         if (rc) {
1300                 ptlrpc_req_finished(request);
1301                 RETURN(rc);
1302         }
1303
1304         ia_valid = op_data->op_attr.ia_valid;
1305         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1306          * cache is not cleared yet. */
1307         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1308         rc = simple_setattr(dentry, &op_data->op_attr);
1309         op_data->op_attr.ia_valid = ia_valid;
1310
1311         /* Extract epoch data if obtained. */
1312         op_data->op_handle = md.body->handle;
1313         op_data->op_ioepoch = md.body->ioepoch;
1314
1315         ll_update_inode(inode, &md);
1316         ptlrpc_req_finished(request);
1317
1318         RETURN(rc);
1319 }
1320
1321 /* Close IO epoch and send Size-on-MDS attribute update. */
1322 static int ll_setattr_done_writing(struct inode *inode,
1323                                    struct md_op_data *op_data,
1324                                    struct md_open_data *mod)
1325 {
1326         struct ll_inode_info *lli = ll_i2info(inode);
1327         int rc = 0;
1328         ENTRY;
1329
1330         LASSERT(op_data != NULL);
1331         if (!S_ISREG(inode->i_mode))
1332                 RETURN(0);
1333
1334         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1335                op_data->op_ioepoch, PFID(&lli->lli_fid));
1336
1337         op_data->op_flags = MF_EPOCH_CLOSE;
1338         ll_done_writing_attr(inode, op_data);
1339         ll_pack_inode2opdata(inode, op_data, NULL);
1340
1341         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1342         if (rc == -EAGAIN) {
1343                 /* MDS has instructed us to obtain Size-on-MDS attribute
1344                  * from OSTs and send setattr to back to MDS. */
1345                 rc = ll_som_update(inode, op_data);
1346         } else if (rc) {
1347                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1348                        inode->i_ino, rc);
1349         }
1350         RETURN(rc);
1351 }
1352
1353 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1354 {
1355         struct obd_capa *capa;
1356         int rc;
1357
1358         if (attr->ia_valid & ATTR_SIZE)
1359                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1360         else
1361                 capa = ll_mdscapa_get(inode);
1362
1363         rc = cl_setattr_ost(inode, attr, capa);
1364
1365         if (attr->ia_valid & ATTR_SIZE)
1366                 ll_truncate_free_capa(capa);
1367         else
1368                 capa_put(capa);
1369
1370         return rc;
1371 }
1372
1373 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1374  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1375  * keep these values until such a time that objects are allocated for it.
1376  * We do the MDS operations first, as it is checking permissions for us.
1377  * We don't to the MDS RPC if there is nothing that we want to store there,
1378  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1379  * going to do an RPC anyways.
1380  *
1381  * If we are doing a truncate, we will send the mtime and ctime updates
1382  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1383  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1384  * at the same time.
1385  */
1386 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1387 {
1388         struct inode *inode = dentry->d_inode;
1389         struct ll_inode_info *lli = ll_i2info(inode);
1390         struct md_op_data *op_data = NULL;
1391         struct md_open_data *mod = NULL;
1392         bool file_is_released = false;
1393         int rc = 0, rc1 = 0;
1394         ENTRY;
1395
1396         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1397                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1398                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1399                 attr->ia_valid);
1400
1401         if (attr->ia_valid & ATTR_SIZE) {
1402                 /* Check new size against VFS/VM file size limit and rlimit */
1403                 rc = inode_newsize_ok(inode, attr->ia_size);
1404                 if (rc)
1405                         RETURN(rc);
1406
1407                 /* The maximum Lustre file size is variable, based on the
1408                  * OST maximum object size and number of stripes.  This
1409                  * needs another check in addition to the VFS check above. */
1410                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1411                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1412                                PFID(&lli->lli_fid), attr->ia_size,
1413                                ll_file_maxbytes(inode));
1414                         RETURN(-EFBIG);
1415                 }
1416
1417                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1418         }
1419
1420         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1421         if (attr->ia_valid & TIMES_SET_FLAGS) {
1422                 if (cfs_curproc_fsuid() != inode->i_uid &&
1423                     !cfs_capable(CFS_CAP_FOWNER))
1424                         RETURN(-EPERM);
1425         }
1426
1427         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1428         if (attr->ia_valid & ATTR_CTIME) {
1429                 attr->ia_ctime = CFS_CURRENT_TIME;
1430                 attr->ia_valid |= ATTR_CTIME_SET;
1431         }
1432         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1433             (attr->ia_valid & ATTR_ATIME)) {
1434                 attr->ia_atime = CFS_CURRENT_TIME;
1435                 attr->ia_valid |= ATTR_ATIME_SET;
1436         }
1437         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1438             (attr->ia_valid & ATTR_MTIME)) {
1439                 attr->ia_mtime = CFS_CURRENT_TIME;
1440                 attr->ia_valid |= ATTR_MTIME_SET;
1441         }
1442
1443         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1444                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1445                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1446                        cfs_time_current_sec());
1447
1448         /* If we are changing file size, file content is modified, flag it. */
1449         if (attr->ia_valid & ATTR_SIZE) {
1450                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1451                 spin_lock(&lli->lli_lock);
1452                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1453                 spin_unlock(&lli->lli_lock);
1454         }
1455
1456         /* We always do an MDS RPC, even if we're only changing the size;
1457          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1458
1459         OBD_ALLOC_PTR(op_data);
1460         if (op_data == NULL)
1461                 RETURN(-ENOMEM);
1462
1463         if (!S_ISDIR(inode->i_mode)) {
1464                 if (attr->ia_valid & ATTR_SIZE)
1465                         inode_dio_write_done(inode);
1466                 mutex_unlock(&inode->i_mutex);
1467                 down_write(&lli->lli_trunc_sem);
1468         }
1469
1470         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1471
1472         /* Open epoch for truncate. */
1473         if (exp_connect_som(ll_i2mdexp(inode)) &&
1474             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1475                 op_data->op_flags = MF_EPOCH_OPEN;
1476
1477         /* truncate on a released file must failed with -ENODATA,
1478          * so size must not be set on MDS for released file
1479          * but other attributes must be set
1480          */
1481         if (S_ISREG(inode->i_mode)) {
1482                 struct lov_stripe_md *lsm;
1483                 __u32 gen;
1484
1485                 ll_layout_refresh(inode, &gen);
1486                 lsm = ccc_inode_lsm_get(inode);
1487                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1488                         file_is_released = true;
1489                 ccc_inode_lsm_put(inode, lsm);
1490         }
1491
1492         /* clear size attr for released file
1493          * we clear the attribute send to MDT in op_data, not the original
1494          * received from caller in attr which is used later to
1495          * decide return code */
1496         if (file_is_released && (attr->ia_valid & ATTR_SIZE))
1497                 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1498
1499         rc = ll_md_setattr(dentry, op_data, &mod);
1500         if (rc)
1501                 GOTO(out, rc);
1502
1503         /* truncate failed, others succeed */
1504         if (file_is_released) {
1505                 if (attr->ia_valid & ATTR_SIZE)
1506                         GOTO(out, rc = -ENODATA);
1507                 else
1508                         GOTO(out, rc = 0);
1509         }
1510
1511         /* RPC to MDT is sent, cancel data modification flag */
1512         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1513                 spin_lock(&lli->lli_lock);
1514                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1515                 spin_unlock(&lli->lli_lock);
1516         }
1517
1518         ll_ioepoch_open(lli, op_data->op_ioepoch);
1519         if (!S_ISREG(inode->i_mode))
1520                 GOTO(out, rc = 0);
1521
1522         if (attr->ia_valid & (ATTR_SIZE |
1523                               ATTR_ATIME | ATTR_ATIME_SET |
1524                               ATTR_MTIME | ATTR_MTIME_SET)) {
1525                 /* For truncate and utimes sending attributes to OSTs, setting
1526                  * mtime/atime to the past will be performed under PW [0:EOF]
1527                  * extent lock (new_size:EOF for truncate).  It may seem
1528                  * excessive to send mtime/atime updates to OSTs when not
1529                  * setting times to past, but it is necessary due to possible
1530                  * time de-synchronization between MDT inode and OST objects */
1531                 rc = ll_setattr_ost(inode, attr);
1532         }
1533         EXIT;
1534 out:
1535         if (op_data) {
1536                 if (op_data->op_ioepoch) {
1537                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1538                         if (!rc)
1539                                 rc = rc1;
1540                 }
1541                 ll_finish_md_op_data(op_data);
1542         }
1543         if (!S_ISDIR(inode->i_mode)) {
1544                 up_write(&lli->lli_trunc_sem);
1545                 mutex_lock(&inode->i_mutex);
1546                 if (attr->ia_valid & ATTR_SIZE)
1547                         inode_dio_wait(inode);
1548         }
1549
1550         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1551                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1552
1553         return rc;
1554 }
1555
1556 int ll_setattr(struct dentry *de, struct iattr *attr)
1557 {
1558         int mode = de->d_inode->i_mode;
1559
1560         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1561                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1562                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1563
1564         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1565                                (ATTR_SIZE|ATTR_MODE)) &&
1566             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1567              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1568               !(attr->ia_mode & S_ISGID))))
1569                 attr->ia_valid |= ATTR_FORCE;
1570
1571         if ((mode & S_ISUID) &&
1572             !(attr->ia_mode & S_ISUID) &&
1573             !(attr->ia_valid & ATTR_KILL_SUID))
1574                 attr->ia_valid |= ATTR_KILL_SUID;
1575
1576         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1577             !(attr->ia_mode & S_ISGID) &&
1578             !(attr->ia_valid & ATTR_KILL_SGID))
1579                 attr->ia_valid |= ATTR_KILL_SGID;
1580
1581         return ll_setattr_raw(de, attr);
1582 }
1583
1584 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1585                        __u64 max_age, __u32 flags)
1586 {
1587         struct ll_sb_info *sbi = ll_s2sbi(sb);
1588         struct obd_statfs obd_osfs;
1589         int rc;
1590         ENTRY;
1591
1592         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1593         if (rc) {
1594                 CERROR("md_statfs fails: rc = %d\n", rc);
1595                 RETURN(rc);
1596         }
1597
1598         osfs->os_type = sb->s_magic;
1599
1600         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1601                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1602
1603         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1604                 flags |= OBD_STATFS_NODELAY;
1605
1606         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1607         if (rc) {
1608                 CERROR("obd_statfs fails: rc = %d\n", rc);
1609                 RETURN(rc);
1610         }
1611
1612         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1613                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1614                obd_osfs.os_files);
1615
1616         osfs->os_bsize = obd_osfs.os_bsize;
1617         osfs->os_blocks = obd_osfs.os_blocks;
1618         osfs->os_bfree = obd_osfs.os_bfree;
1619         osfs->os_bavail = obd_osfs.os_bavail;
1620
1621         /* If we don't have as many objects free on the OST as inodes
1622          * on the MDS, we reduce the total number of inodes to
1623          * compensate, so that the "inodes in use" number is correct.
1624          */
1625         if (obd_osfs.os_ffree < osfs->os_ffree) {
1626                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1627                         obd_osfs.os_ffree;
1628                 osfs->os_ffree = obd_osfs.os_ffree;
1629         }
1630
1631         RETURN(rc);
1632 }
1633 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1634 {
1635         struct super_block *sb = de->d_sb;
1636         struct obd_statfs osfs;
1637         int rc;
1638
1639         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1640         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1641
1642         /* Some amount of caching on the client is allowed */
1643         rc = ll_statfs_internal(sb, &osfs,
1644                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1645                                 0);
1646         if (rc)
1647                 return rc;
1648
1649         statfs_unpack(sfs, &osfs);
1650
1651         /* We need to downshift for all 32-bit kernels, because we can't
1652          * tell if the kernel is being called via sys_statfs64() or not.
1653          * Stop before overflowing f_bsize - in which case it is better
1654          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1655         if (sizeof(long) < 8) {
1656                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1657                         sfs->f_bsize <<= 1;
1658
1659                         osfs.os_blocks >>= 1;
1660                         osfs.os_bfree >>= 1;
1661                         osfs.os_bavail >>= 1;
1662                 }
1663         }
1664
1665         sfs->f_blocks = osfs.os_blocks;
1666         sfs->f_bfree = osfs.os_bfree;
1667         sfs->f_bavail = osfs.os_bavail;
1668         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1669         return 0;
1670 }
1671
1672 void ll_inode_size_lock(struct inode *inode)
1673 {
1674         struct ll_inode_info *lli;
1675
1676         LASSERT(!S_ISDIR(inode->i_mode));
1677
1678         lli = ll_i2info(inode);
1679         LASSERT(lli->lli_size_sem_owner != current);
1680         down(&lli->lli_size_sem);
1681         LASSERT(lli->lli_size_sem_owner == NULL);
1682         lli->lli_size_sem_owner = current;
1683 }
1684
1685 void ll_inode_size_unlock(struct inode *inode)
1686 {
1687         struct ll_inode_info *lli;
1688
1689         lli = ll_i2info(inode);
1690         LASSERT(lli->lli_size_sem_owner == current);
1691         lli->lli_size_sem_owner = NULL;
1692         up(&lli->lli_size_sem);
1693 }
1694
1695 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1696 {
1697         struct ll_inode_info *lli = ll_i2info(inode);
1698         struct mdt_body *body = md->body;
1699         struct lov_stripe_md *lsm = md->lsm;
1700         struct ll_sb_info *sbi = ll_i2sbi(inode);
1701
1702         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1703         if (lsm != NULL) {
1704                 if (!lli->lli_has_smd &&
1705                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1706                         cl_file_inode_init(inode, md);
1707
1708                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1709                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1710                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1711         }
1712
1713         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1714                 if (body->valid & OBD_MD_FLRMTPERM)
1715                         ll_update_remote_perm(inode, md->remote_perm);
1716         }
1717 #ifdef CONFIG_FS_POSIX_ACL
1718         else if (body->valid & OBD_MD_FLACL) {
1719                 spin_lock(&lli->lli_lock);
1720                 if (lli->lli_posix_acl)
1721                         posix_acl_release(lli->lli_posix_acl);
1722                 lli->lli_posix_acl = md->posix_acl;
1723                 spin_unlock(&lli->lli_lock);
1724         }
1725 #endif
1726         inode->i_ino = cl_fid_build_ino(&body->fid1,
1727                                         sbi->ll_flags & LL_SBI_32BIT_API);
1728         inode->i_generation = cl_fid_build_gen(&body->fid1);
1729
1730         if (body->valid & OBD_MD_FLATIME) {
1731                 if (body->atime > LTIME_S(inode->i_atime))
1732                         LTIME_S(inode->i_atime) = body->atime;
1733                 lli->lli_lvb.lvb_atime = body->atime;
1734         }
1735         if (body->valid & OBD_MD_FLMTIME) {
1736                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1737                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1738                                "to "LPU64"\n", inode->i_ino,
1739                                LTIME_S(inode->i_mtime), body->mtime);
1740                         LTIME_S(inode->i_mtime) = body->mtime;
1741                 }
1742                 lli->lli_lvb.lvb_mtime = body->mtime;
1743         }
1744         if (body->valid & OBD_MD_FLCTIME) {
1745                 if (body->ctime > LTIME_S(inode->i_ctime))
1746                         LTIME_S(inode->i_ctime) = body->ctime;
1747                 lli->lli_lvb.lvb_ctime = body->ctime;
1748         }
1749         if (body->valid & OBD_MD_FLMODE)
1750                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1751         if (body->valid & OBD_MD_FLTYPE)
1752                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1753         LASSERT(inode->i_mode != 0);
1754         if (S_ISREG(inode->i_mode)) {
1755                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1756         } else {
1757                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1758         }
1759         if (body->valid & OBD_MD_FLUID)
1760                 inode->i_uid = body->uid;
1761         if (body->valid & OBD_MD_FLGID)
1762                 inode->i_gid = body->gid;
1763         if (body->valid & OBD_MD_FLFLAGS)
1764                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1765         if (body->valid & OBD_MD_FLNLINK)
1766                 set_nlink(inode, body->nlink);
1767         if (body->valid & OBD_MD_FLRDEV)
1768                 inode->i_rdev = old_decode_dev(body->rdev);
1769
1770         if (body->valid & OBD_MD_FLID) {
1771                 /* FID shouldn't be changed! */
1772                 if (fid_is_sane(&lli->lli_fid)) {
1773                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1774                                  "Trying to change FID "DFID
1775                                  " to the "DFID", inode %lu/%u(%p)\n",
1776                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1777                                  inode->i_ino, inode->i_generation, inode);
1778                 } else
1779                         lli->lli_fid = body->fid1;
1780         }
1781
1782         LASSERT(fid_seq(&lli->lli_fid) != 0);
1783
1784         if (body->valid & OBD_MD_FLSIZE) {
1785                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1786                     S_ISREG(inode->i_mode)) {
1787                         struct lustre_handle lockh;
1788                         ldlm_mode_t mode;
1789
1790                         /* As it is possible a blocking ast has been processed
1791                          * by this time, we need to check there is an UPDATE
1792                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1793                          * it. */
1794                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1795                                                &lockh, LDLM_FL_CBPENDING);
1796                         if (mode) {
1797                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1798                                                       LLIF_EPOCH_PENDING |
1799                                                       LLIF_SOM_DIRTY)) {
1800                                         CERROR("ino %lu flags %u still has "
1801                                                "size authority! do not trust "
1802                                                "the size got from MDS\n",
1803                                                inode->i_ino, lli->lli_flags);
1804                                 } else {
1805                                         /* Use old size assignment to avoid
1806                                          * deadlock bz14138 & bz14326 */
1807                                         i_size_write(inode, body->size);
1808                                         spin_lock(&lli->lli_lock);
1809                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1810                                         spin_unlock(&lli->lli_lock);
1811                                 }
1812                                 ldlm_lock_decref(&lockh, mode);
1813                         }
1814                 } else {
1815                         /* Use old size assignment to avoid
1816                          * deadlock bz14138 & bz14326 */
1817                         i_size_write(inode, body->size);
1818
1819                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1820                                inode->i_ino, (unsigned long long)body->size);
1821                 }
1822
1823                 if (body->valid & OBD_MD_FLBLOCKS)
1824                         inode->i_blocks = body->blocks;
1825         }
1826
1827         if (body->valid & OBD_MD_FLMDSCAPA) {
1828                 LASSERT(md->mds_capa);
1829                 ll_add_capa(inode, md->mds_capa);
1830         }
1831         if (body->valid & OBD_MD_FLOSSCAPA) {
1832                 LASSERT(md->oss_capa);
1833                 ll_add_capa(inode, md->oss_capa);
1834         }
1835
1836         if (body->valid & OBD_MD_TSTATE) {
1837                 if (body->t_state & MS_RESTORE)
1838                         lli->lli_flags |= LLIF_FILE_RESTORING;
1839         }
1840 }
1841
1842 void ll_read_inode2(struct inode *inode, void *opaque)
1843 {
1844         struct lustre_md *md = opaque;
1845         struct ll_inode_info *lli = ll_i2info(inode);
1846         ENTRY;
1847
1848         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1849                PFID(&lli->lli_fid), inode);
1850
1851         LASSERT(!lli->lli_has_smd);
1852
1853         /* Core attributes from the MDS first.  This is a new inode, and
1854          * the VFS doesn't zero times in the core inode so we have to do
1855          * it ourselves.  They will be overwritten by either MDS or OST
1856          * attributes - we just need to make sure they aren't newer. */
1857         LTIME_S(inode->i_mtime) = 0;
1858         LTIME_S(inode->i_atime) = 0;
1859         LTIME_S(inode->i_ctime) = 0;
1860         inode->i_rdev = 0;
1861         ll_update_inode(inode, md);
1862
1863         /* OIDEBUG(inode); */
1864
1865         /* initializing backing dev info. */
1866         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1867
1868
1869         if (S_ISREG(inode->i_mode)) {
1870                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1871                 inode->i_op = &ll_file_inode_operations;
1872                 inode->i_fop = sbi->ll_fop;
1873                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1874                 EXIT;
1875         } else if (S_ISDIR(inode->i_mode)) {
1876                 inode->i_op = &ll_dir_inode_operations;
1877                 inode->i_fop = &ll_dir_operations;
1878                 EXIT;
1879         } else if (S_ISLNK(inode->i_mode)) {
1880                 inode->i_op = &ll_fast_symlink_inode_operations;
1881                 EXIT;
1882         } else {
1883                 inode->i_op = &ll_special_inode_operations;
1884
1885                 init_special_inode(inode, inode->i_mode,
1886                                    inode->i_rdev);
1887
1888                 EXIT;
1889         }
1890 }
1891
1892 void ll_delete_inode(struct inode *inode)
1893 {
1894         struct cl_inode_info *lli = cl_i2info(inode);
1895         ENTRY;
1896
1897         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1898                 /* discard all dirty pages before truncating them, required by
1899                  * osc_extent implementation at LU-1030. */
1900                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1901                                    CL_FSYNC_DISCARD, 1);
1902
1903         truncate_inode_pages(&inode->i_data, 0);
1904
1905         /* Workaround for LU-118 */
1906         if (inode->i_data.nrpages) {
1907                 TREE_READ_LOCK_IRQ(&inode->i_data);
1908                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1909                 LASSERTF(inode->i_data.nrpages == 0,
1910                          "inode=%lu/%u(%p) nrpages=%lu, see "
1911                          "http://jira.whamcloud.com/browse/LU-118\n",
1912                          inode->i_ino, inode->i_generation, inode,
1913                          inode->i_data.nrpages);
1914         }
1915         /* Workaround end */
1916
1917 #ifdef HAVE_SBOPS_EVICT_INODE
1918         ll_clear_inode(inode);
1919 #endif
1920         clear_inode(inode);
1921
1922         EXIT;
1923 }
1924
1925 int ll_iocontrol(struct inode *inode, struct file *file,
1926                  unsigned int cmd, unsigned long arg)
1927 {
1928         struct ll_sb_info *sbi = ll_i2sbi(inode);
1929         struct ptlrpc_request *req = NULL;
1930         int rc, flags = 0;
1931         ENTRY;
1932
1933         switch(cmd) {
1934         case FSFILT_IOC_GETFLAGS: {
1935                 struct mdt_body *body;
1936                 struct md_op_data *op_data;
1937
1938                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1939                                              0, 0, LUSTRE_OPC_ANY,
1940                                              NULL);
1941                 if (IS_ERR(op_data))
1942                         RETURN(PTR_ERR(op_data));
1943
1944                 op_data->op_valid = OBD_MD_FLFLAGS;
1945                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1946                 ll_finish_md_op_data(op_data);
1947                 if (rc) {
1948                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1949                         RETURN(-abs(rc));
1950                 }
1951
1952                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1953
1954                 flags = body->flags;
1955
1956                 ptlrpc_req_finished(req);
1957
1958                 RETURN(put_user(flags, (int *)arg));
1959         }
1960         case FSFILT_IOC_SETFLAGS: {
1961                 struct lov_stripe_md *lsm;
1962                 struct obd_info oinfo = { { { 0 } } };
1963                 struct md_op_data *op_data;
1964
1965                 if (get_user(flags, (int *)arg))
1966                         RETURN(-EFAULT);
1967
1968                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1969                                              LUSTRE_OPC_ANY, NULL);
1970                 if (IS_ERR(op_data))
1971                         RETURN(PTR_ERR(op_data));
1972
1973                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1974                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1975                 rc = md_setattr(sbi->ll_md_exp, op_data,
1976                                 NULL, 0, NULL, 0, &req, NULL);
1977                 ll_finish_md_op_data(op_data);
1978                 ptlrpc_req_finished(req);
1979                 if (rc)
1980                         RETURN(rc);
1981
1982                 inode->i_flags = ll_ext_to_inode_flags(flags);
1983
1984                 lsm = ccc_inode_lsm_get(inode);
1985                 if (!lsm_has_objects(lsm)) {
1986                         ccc_inode_lsm_put(inode, lsm);
1987                         RETURN(0);
1988                 }
1989
1990                 OBDO_ALLOC(oinfo.oi_oa);
1991                 if (!oinfo.oi_oa) {
1992                         ccc_inode_lsm_put(inode, lsm);
1993                         RETURN(-ENOMEM);
1994                 }
1995                 oinfo.oi_md = lsm;
1996                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1997                 oinfo.oi_oa->o_flags = flags;
1998                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1999                                        OBD_MD_FLGROUP;
2000                 oinfo.oi_capa = ll_mdscapa_get(inode);
2001                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2002                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2003                 capa_put(oinfo.oi_capa);
2004                 OBDO_FREE(oinfo.oi_oa);
2005                 ccc_inode_lsm_put(inode, lsm);
2006
2007                 if (rc && rc != -EPERM && rc != -EACCES)
2008                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2009
2010                 RETURN(rc);
2011         }
2012         default:
2013                 RETURN(-ENOSYS);
2014         }
2015
2016         RETURN(0);
2017 }
2018
2019 int ll_flush_ctx(struct inode *inode)
2020 {
2021         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2022
2023         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2024
2025         obd_set_info_async(NULL, sbi->ll_md_exp,
2026                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2027                            0, NULL, NULL);
2028         obd_set_info_async(NULL, sbi->ll_dt_exp,
2029                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2030                            0, NULL, NULL);
2031         return 0;
2032 }
2033
2034 /* umount -f client means force down, don't save state */
2035 void ll_umount_begin(struct super_block *sb)
2036 {
2037         struct ll_sb_info *sbi = ll_s2sbi(sb);
2038         struct obd_device *obd;
2039         struct obd_ioctl_data *ioc_data;
2040         ENTRY;
2041
2042         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2043                sb->s_count, atomic_read(&sb->s_active));
2044
2045         obd = class_exp2obd(sbi->ll_md_exp);
2046         if (obd == NULL) {
2047                 CERROR("Invalid MDC connection handle "LPX64"\n",
2048                        sbi->ll_md_exp->exp_handle.h_cookie);
2049                 EXIT;
2050                 return;
2051         }
2052         obd->obd_force = 1;
2053
2054         obd = class_exp2obd(sbi->ll_dt_exp);
2055         if (obd == NULL) {
2056                 CERROR("Invalid LOV connection handle "LPX64"\n",
2057                        sbi->ll_dt_exp->exp_handle.h_cookie);
2058                 EXIT;
2059                 return;
2060         }
2061         obd->obd_force = 1;
2062
2063         OBD_ALLOC_PTR(ioc_data);
2064         if (ioc_data) {
2065                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2066                               sizeof *ioc_data, ioc_data, NULL);
2067
2068                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2069                               sizeof *ioc_data, ioc_data, NULL);
2070
2071                 OBD_FREE_PTR(ioc_data);
2072         }
2073
2074         /* Really, we'd like to wait until there are no requests outstanding,
2075          * and then continue.  For now, we just invalidate the requests,
2076          * schedule() and sleep one second if needed, and hope.
2077          */
2078         cfs_schedule();
2079         EXIT;
2080 }
2081
2082 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2083 {
2084         struct ll_sb_info *sbi = ll_s2sbi(sb);
2085         char *profilenm = get_profile_name(sb);
2086         int err;
2087         __u32 read_only;
2088
2089         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2090                 read_only = *flags & MS_RDONLY;
2091                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2092                                          sizeof(KEY_READ_ONLY),
2093                                          KEY_READ_ONLY, sizeof(read_only),
2094                                          &read_only, NULL);
2095                 if (err) {
2096                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2097                                       profilenm, read_only ?
2098                                       "read-only" : "read-write", err);
2099                         return err;
2100                 }
2101
2102                 if (read_only)
2103                         sb->s_flags |= MS_RDONLY;
2104                 else
2105                         sb->s_flags &= ~MS_RDONLY;
2106
2107                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2108                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2109                                       read_only ?  "read-only" : "read-write");
2110         }
2111         return 0;
2112 }
2113
2114 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2115                   struct super_block *sb, struct lookup_intent *it)
2116 {
2117         struct ll_sb_info *sbi = NULL;
2118         struct lustre_md md;
2119         int rc;
2120         ENTRY;
2121
2122         LASSERT(*inode || sb);
2123         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2124         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2125                               sbi->ll_md_exp, &md);
2126         if (rc)
2127                 RETURN(rc);
2128
2129         if (*inode) {
2130                 ll_update_inode(*inode, &md);
2131         } else {
2132                 LASSERT(sb != NULL);
2133
2134                 /*
2135                  * At this point server returns to client's same fid as client
2136                  * generated for creating. So using ->fid1 is okay here.
2137                  */
2138                 LASSERT(fid_is_sane(&md.body->fid1));
2139
2140                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2141                                              sbi->ll_flags & LL_SBI_32BIT_API),
2142                                  &md);
2143                 if (*inode == NULL || IS_ERR(*inode)) {
2144 #ifdef CONFIG_FS_POSIX_ACL
2145                         if (md.posix_acl) {
2146                                 posix_acl_release(md.posix_acl);
2147                                 md.posix_acl = NULL;
2148                         }
2149 #endif
2150                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2151                         *inode = NULL;
2152                         CERROR("new_inode -fatal: rc %d\n", rc);
2153                         GOTO(out, rc);
2154                 }
2155         }
2156
2157         /* Handling piggyback layout lock.
2158          * Layout lock can be piggybacked by getattr and open request.
2159          * The lsm can be applied to inode only if it comes with a layout lock
2160          * otherwise correct layout may be overwritten, for example:
2161          * 1. proc1: mdt returns a lsm but not granting layout
2162          * 2. layout was changed by another client
2163          * 3. proc2: refresh layout and layout lock granted
2164          * 4. proc1: to apply a stale layout */
2165         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2166                 struct lustre_handle lockh;
2167                 struct ldlm_lock *lock;
2168
2169                 lockh.cookie = it->d.lustre.it_lock_handle;
2170                 lock = ldlm_handle2lock(&lockh);
2171                 LASSERT(lock != NULL);
2172                 if (ldlm_has_layout(lock)) {
2173                         struct cl_object_conf conf;
2174
2175                         memset(&conf, 0, sizeof(conf));
2176                         conf.coc_opc = OBJECT_CONF_SET;
2177                         conf.coc_inode = *inode;
2178                         conf.coc_lock = lock;
2179                         conf.u.coc_md = &md;
2180                         (void)ll_layout_conf(*inode, &conf);
2181                 }
2182                 LDLM_LOCK_PUT(lock);
2183         }
2184
2185 out:
2186         if (md.lsm != NULL)
2187                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2188         md_free_lustre_md(sbi->ll_md_exp, &md);
2189         RETURN(rc);
2190 }
2191
2192 int ll_obd_statfs(struct inode *inode, void *arg)
2193 {
2194         struct ll_sb_info *sbi = NULL;
2195         struct obd_export *exp;
2196         char *buf = NULL;
2197         struct obd_ioctl_data *data = NULL;
2198         __u32 type;
2199         __u32 flags;
2200         int len = 0, rc;
2201
2202         if (!inode || !(sbi = ll_i2sbi(inode)))
2203                 GOTO(out_statfs, rc = -EINVAL);
2204
2205         rc = obd_ioctl_getdata(&buf, &len, arg);
2206         if (rc)
2207                 GOTO(out_statfs, rc);
2208
2209         data = (void*)buf;
2210         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2211             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2212                 GOTO(out_statfs, rc = -EINVAL);
2213
2214         if (data->ioc_inllen1 != sizeof(__u32) ||
2215             data->ioc_inllen2 != sizeof(__u32) ||
2216             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2217             data->ioc_plen2 != sizeof(struct obd_uuid))
2218                 GOTO(out_statfs, rc = -EINVAL);
2219
2220         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2221         if (type & LL_STATFS_LMV)
2222                 exp = sbi->ll_md_exp;
2223         else if (type & LL_STATFS_LOV)
2224                 exp = sbi->ll_dt_exp;
2225         else
2226                 GOTO(out_statfs, rc = -ENODEV);
2227
2228         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2229         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2230         if (rc)
2231                 GOTO(out_statfs, rc);
2232 out_statfs:
2233         if (buf)
2234                 obd_ioctl_freedata(buf, len);
2235         return rc;
2236 }
2237
2238 int ll_process_config(struct lustre_cfg *lcfg)
2239 {
2240         char *ptr;
2241         void *sb;
2242         struct lprocfs_static_vars lvars;
2243         unsigned long x;
2244         int rc = 0;
2245
2246         lprocfs_llite_init_vars(&lvars);
2247
2248         /* The instance name contains the sb: lustre-client-aacfe000 */
2249         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2250         if (!ptr || !*(++ptr))
2251                 return -EINVAL;
2252         if (sscanf(ptr, "%lx", &x) != 1)
2253                 return -EINVAL;
2254         sb = (void *)x;
2255         /* This better be a real Lustre superblock! */
2256         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2257
2258         /* Note we have not called client_common_fill_super yet, so
2259            proc fns must be able to handle that! */
2260         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2261                                       lcfg, sb);
2262         if (rc > 0)
2263                 rc = 0;
2264         return(rc);
2265 }
2266
2267 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2268 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2269                                        struct inode *i1, struct inode *i2,
2270                                        const char *name, int namelen,
2271                                        int mode, __u32 opc, void *data)
2272 {
2273         LASSERT(i1 != NULL);
2274
2275         if (namelen > ll_i2sbi(i1)->ll_namelen)
2276                 return ERR_PTR(-ENAMETOOLONG);
2277
2278         if (op_data == NULL)
2279                 OBD_ALLOC_PTR(op_data);
2280
2281         if (op_data == NULL)
2282                 return ERR_PTR(-ENOMEM);
2283
2284         ll_i2gids(op_data->op_suppgids, i1, i2);
2285         op_data->op_fid1 = *ll_inode2fid(i1);
2286         op_data->op_capa1 = ll_mdscapa_get(i1);
2287
2288         if (i2) {
2289                 op_data->op_fid2 = *ll_inode2fid(i2);
2290                 op_data->op_capa2 = ll_mdscapa_get(i2);
2291         } else {
2292                 fid_zero(&op_data->op_fid2);
2293                 op_data->op_capa2 = NULL;
2294         }
2295
2296         op_data->op_name = name;
2297         op_data->op_namelen = namelen;
2298         op_data->op_mode = mode;
2299         op_data->op_mod_time = cfs_time_current_sec();
2300         op_data->op_fsuid = cfs_curproc_fsuid();
2301         op_data->op_fsgid = cfs_curproc_fsgid();
2302         op_data->op_cap = cfs_curproc_cap_pack();
2303         op_data->op_bias = 0;
2304         op_data->op_cli_flags = 0;
2305         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2306              filename_is_volatile(name, namelen, NULL))
2307                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2308         op_data->op_opc = opc;
2309         op_data->op_mds = 0;
2310         op_data->op_data = data;
2311
2312         /* If the file is being opened after mknod() (normally due to NFS)
2313          * try to use the default stripe data from parent directory for
2314          * allocating OST objects.  Try to pass the parent FID to MDS. */
2315         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2316             !ll_i2info(i2)->lli_has_smd) {
2317                 struct ll_inode_info *lli = ll_i2info(i2);
2318
2319                 spin_lock(&lli->lli_lock);
2320                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2321                         op_data->op_fid1 = lli->lli_pfid;
2322                 spin_unlock(&lli->lli_lock);
2323                 /** We ignore parent's capability temporary. */
2324         }
2325
2326         /* When called by ll_setattr_raw, file is i1. */
2327         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2328                 op_data->op_bias |= MDS_DATA_MODIFIED;
2329
2330         return op_data;
2331 }
2332
2333 void ll_finish_md_op_data(struct md_op_data *op_data)
2334 {
2335         capa_put(op_data->op_capa1);
2336         capa_put(op_data->op_capa2);
2337         OBD_FREE_PTR(op_data);
2338 }
2339
2340 #ifdef HAVE_SUPEROPS_USE_DENTRY
2341 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2342 #else
2343 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2344 #endif
2345 {
2346         struct ll_sb_info *sbi;
2347
2348 #ifdef HAVE_SUPEROPS_USE_DENTRY
2349         LASSERT((seq != NULL) && (dentry != NULL));
2350         sbi = ll_s2sbi(dentry->d_sb);
2351 #else
2352         LASSERT((seq != NULL) && (vfs != NULL));
2353         sbi = ll_s2sbi(vfs->mnt_sb);
2354 #endif
2355
2356         if (sbi->ll_flags & LL_SBI_NOLCK)
2357                 seq_puts(seq, ",nolock");
2358
2359         if (sbi->ll_flags & LL_SBI_FLOCK)
2360                 seq_puts(seq, ",flock");
2361
2362         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2363                 seq_puts(seq, ",localflock");
2364
2365         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2366                 seq_puts(seq, ",user_xattr");
2367
2368         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2369                 seq_puts(seq, ",lazystatfs");
2370
2371         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2372                 seq_puts(seq, ",user_fid2path");
2373
2374         RETURN(0);
2375 }
2376
2377 /**
2378  * Get obd name by cmd, and copy out to user space
2379  */
2380 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2381 {
2382         struct ll_sb_info *sbi = ll_i2sbi(inode);
2383         struct obd_device *obd;
2384         ENTRY;
2385
2386         if (cmd == OBD_IOC_GETDTNAME)
2387                 obd = class_exp2obd(sbi->ll_dt_exp);
2388         else if (cmd == OBD_IOC_GETMDNAME)
2389                 obd = class_exp2obd(sbi->ll_md_exp);
2390         else
2391                 RETURN(-EINVAL);
2392
2393         if (!obd)
2394                 RETURN(-ENOENT);
2395
2396         if (copy_to_user((void *)arg, obd->obd_name,
2397                          strlen(obd->obd_name) + 1))
2398                 RETURN(-EFAULT);
2399
2400         RETURN(0);
2401 }
2402
2403 /**
2404  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2405  * fsname will be returned in this buffer; otherwise, a static buffer will be
2406  * used to store the fsname and returned to caller.
2407  */
2408 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2409 {
2410         static char fsname_static[MTI_NAME_MAXLEN];
2411         struct lustre_sb_info *lsi = s2lsi(sb);
2412         char *ptr;
2413         int len;
2414
2415         if (buf == NULL) {
2416                 /* this means the caller wants to use static buffer
2417                  * and it doesn't care about race. Usually this is
2418                  * in error reporting path */
2419                 buf = fsname_static;
2420                 buflen = sizeof(fsname_static);
2421         }
2422
2423         len = strlen(lsi->lsi_lmd->lmd_profile);
2424         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2425         if (ptr && (strcmp(ptr, "-client") == 0))
2426                 len -= 7;
2427
2428         if (unlikely(len >= buflen))
2429                 len = buflen - 1;
2430         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2431         buf[len] = '\0';
2432
2433         return buf;
2434 }
2435
2436 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2437 {
2438         char *path = NULL;
2439
2440 #ifdef HAVE_FS_STRUCT_USE_PATH
2441         struct path p;
2442
2443         p.dentry = dentry;
2444         p.mnt = current->fs->root.mnt;
2445         path_get(&p);
2446         path = d_path(&p, buf, bufsize);
2447         path_put(&p);
2448 #else
2449         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2450 #endif
2451
2452         return path;
2453 }
2454
2455 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2456 {
2457         char *buf, *path = NULL;
2458         struct dentry *dentry = NULL;
2459         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2460
2461         /* this can be called inside spin lock so use GFP_ATOMIC. */
2462         buf = (char *)__get_free_page(GFP_ATOMIC);
2463         if (buf != NULL) {
2464                 dentry = d_find_alias(page->mapping->host);
2465                 if (dentry != NULL)
2466                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2467         }
2468
2469         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2470               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2471               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2472               PFID(&obj->cob_header.coh_lu.loh_fid),
2473               (path && !IS_ERR(path)) ? path : "", ioret);
2474
2475         if (dentry != NULL)
2476                 dput(dentry);
2477
2478         if (buf != NULL)
2479                 free_page((unsigned long)buf);
2480 }