lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  97                 lru_page_max = pages / 2;
  98         } else {
  99                 lru_page_max = (pages / 4) * 3;
 100         }
 101
 102         /* initialize lru data */
 103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 104         sbi->ll_cache.ccc_lru_max = lru_page_max;
 105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 108
 109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 110                                            SBI_DEFAULT_READAHEAD_MAX);
 111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 116
 117         ll_generate_random_uuid(uuid);
 118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 120
 121         spin_lock(&ll_sb_lock);
 122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 123         spin_unlock(&ll_sb_lock);
 124
 125         sbi->ll_flags |= LL_SBI_VERBOSE;
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 136                                pp_r_hist.oh_lock);
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145         cfs_atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147
 148         RETURN(sbi);
 149 }
 150
 151 void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 spin_lock(&ll_sb_lock);
 158                 cfs_list_del(&sbi->ll_list);
 159                 spin_unlock(&ll_sb_lock);
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 166                                     struct vfsmount *mnt)
 167 {
 168         struct inode *root = 0;
 169         struct ll_sb_info *sbi = ll_s2sbi(sb);
 170         struct obd_device *obd;
 171         struct obd_capa *oc = NULL;
 172         struct obd_statfs *osfs = NULL;
 173         struct ptlrpc_request *request = NULL;
 174         struct obd_connect_data *data = NULL;
 175         struct obd_uuid *uuid;
 176         struct md_op_data *op_data;
 177         struct lustre_md lmd;
 178         obd_valid valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         OBD_ALLOC_PTR(osfs);
 193         if (osfs == NULL) {
 194                 OBD_FREE_PTR(data);
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         if (proc_lustre_fs_root) {
 199                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 200                                                   dt, md);
 201                 if (err < 0)
 202                         CERROR("could not register mount in /proc/fs/lustre\n");
 203         }
 204
 205         /* indicate the features supported by this client */
 206         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 207                                   OBD_CONNECT_ATTRFID  |
 208                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 209                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 210                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 211                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 212                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 213                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 214                                   OBD_CONNECT_EINPROGRESS |
 215                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 216                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 217                                   OBD_CONNECT_MAX_EASIZE;
 218
 219         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 220                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 221
 222 #ifdef HAVE_LRU_RESIZE_SUPPORT
 223         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 224                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 225 #endif
 226 #ifdef CONFIG_FS_POSIX_ACL
 227         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 228 #endif
 229
 230         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 231                 /* flag mdc connection as lightweight, only used for test
 232                  * purpose, use with care */
 233                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 234
 235         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 236         data->ocd_version = LUSTRE_VERSION_CODE;
 237
 238         if (sb->s_flags & MS_RDONLY)
 239                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 240         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 241                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 242
 243 #ifdef HAVE_MS_FLOCK_LOCK
 244         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 245         sb->s_flags |= MS_FLOCK_LOCK;
 246 #endif
 247 #ifdef MS_HAS_NEW_AOPS
 248         sb->s_flags |= MS_HAS_NEW_AOPS;
 249 #endif
 250
 251         if (sbi->ll_flags & LL_SBI_FLOCK)
 252                 sbi->ll_fop = &ll_file_operations_flock;
 253         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 254                 sbi->ll_fop = &ll_file_operations;
 255         else
 256                 sbi->ll_fop = &ll_file_operations_noflock;
 257
 258         /* real client */
 259         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 260         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 261                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 262
 263         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 264
 265         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 266         if (err == -EBUSY) {
 267                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 268                                    "recovery, of which this client is not a "
 269                                    "part. Please wait for recovery to complete,"
 270                                    " abort, or time out.\n", md);
 271                 GOTO(out, err);
 272         } else if (err) {
 273                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 274                 GOTO(out, err);
 275         }
 276
 277         sbi->ll_md_exp->exp_connect_data = *data;
 278
 279         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 280                            LUSTRE_SEQ_METADATA);
 281         if (err) {
 282                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 283                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 284                 GOTO(out_md, err);
 285         }
 286
 287         /* For mount, we only need fs info from MDT0, and also in DNE, it
 288          * can make sure the client can be mounted as long as MDT0 is
 289          * avaible */
 290         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 291                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 292                         OBD_STATFS_FOR_MDT0);
 293         if (err)
 294                 GOTO(out_md_fid, err);
 295
 296         /* This needs to be after statfs to ensure connect has finished.
 297          * Note that "data" does NOT contain the valid connect reply.
 298          * If connecting to a 1.8 server there will be no LMV device, so
 299          * we can access the MDC export directly and exp_connect_flags will
 300          * be non-zero, but if accessing an upgraded 2.1 server it will
 301          * have the correct flags filled in.
 302          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 303         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 304         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 305             valid != CLIENT_CONNECT_MDT_REQD) {
 306                 char *buf;
 307
 308                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 309                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 310                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 311                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 312                                    "feature(s) needed for correct operation "
 313                                    "of this client (%s). Please upgrade "
 314                                    "server or downgrade client.\n",
 315                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 316                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 317                 GOTO(out_md_fid, err = -EPROTO);
 318         }
 319
 320         size = sizeof(*data);
 321         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 322                            KEY_CONN_DATA,  &size, data, NULL);
 323         if (err) {
 324                 CERROR("%s: Get connect data failed: rc = %d\n",
 325                        sbi->ll_md_exp->exp_obd->obd_name, err);
 326                 GOTO(out_md_fid, err);
 327         }
 328
 329         LASSERT(osfs->os_bsize);
 330         sb->s_blocksize = osfs->os_bsize;
 331         sb->s_blocksize_bits = log2(osfs->os_bsize);
 332         sb->s_magic = LL_SUPER_MAGIC;
 333         sb->s_maxbytes = MAX_LFS_FILESIZE;
 334         sbi->ll_namelen = osfs->os_namelen;
 335         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 336
 337         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 338             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 339                 LCONSOLE_INFO("Disabling user_xattr feature because "
 340                               "it is not supported on the server\n");
 341                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags |= MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags |= LL_SBI_ACL;
 349         } else {
 350                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 351 #ifdef MS_POSIXACL
 352                 sb->s_flags &= ~MS_POSIXACL;
 353 #endif
 354                 sbi->ll_flags &= ~LL_SBI_ACL;
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 358                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 359                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 360                         LCONSOLE_INFO("client is set as remote by default.\n");
 361                 }
 362         } else {
 363                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 364                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 365                         LCONSOLE_INFO("client claims to be remote, but server "
 366                                       "rejected, forced to be local.\n");
 367                 }
 368         }
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 371                 LCONSOLE_INFO("client enabled MDS capability!\n");
 372                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 373         }
 374
 375         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 376                 LCONSOLE_INFO("client enabled OSS capability!\n");
 377                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 378         }
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 381                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 382
 383         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 384                 sbi->ll_md_brw_size = data->ocd_brw_size;
 385         else
 386                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 387
 388         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 389                 LCONSOLE_INFO("Layout lock feature supported.\n");
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391         }
 392
 393         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 394                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 395                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 396                                       "unknown maximum xattr size.\n", dt);
 397                 } else {
 398                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 399                         sbi->ll_xattr_cache_enabled = 1;
 400                 }
 401         }
 402
 403         obd = class_name2obd(dt);
 404         if (!obd) {
 405                 CERROR("DT %s: not setup or attached\n", dt);
 406                 GOTO(out_md_fid, err = -ENODEV);
 407         }
 408
 409         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 410                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 411                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 412                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 413                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 414                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 415                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 416                                   OBD_CONNECT_MAXBYTES |
 417                                   OBD_CONNECT_EINPROGRESS |
 418                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 419                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 420
 421         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 422                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 423
 424         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 425                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 426                  * disabled by default, because it can still be enabled on the
 427                  * fly via /proc. As a consequence, we still need to come to an
 428                  * agreement on the supported algorithms at connect time */
 429                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 430
 431                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 432                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 433                 else
 434                         data->ocd_cksum_types = cksum_types_supported_client();
 435         }
 436
 437 #ifdef HAVE_LRU_RESIZE_SUPPORT
 438         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 439 #endif
 440         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 441                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 442
 443         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 444                "ocd_grant: %d\n", data->ocd_connect_flags,
 445                data->ocd_version, data->ocd_grant);
 446
 447         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 448         obd->obd_upcall.onu_upcall = cl_ocd_update;
 449
 450         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 451
 452         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 453                           NULL);
 454         if (err == -EBUSY) {
 455                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 456                                    "recovery, of which this client is not a "
 457                                    "part.  Please wait for recovery to "
 458                                    "complete, abort, or time out.\n", dt);
 459                 GOTO(out_md, err);
 460         } else if (err) {
 461                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 462                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 463                 GOTO(out_md, err);
 464         }
 465
 466         sbi->ll_dt_exp->exp_connect_data = *data;
 467
 468         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 469                            LUSTRE_SEQ_METADATA);
 470         if (err) {
 471                 CERROR("%s: Can't init data layer FID infrastructure, "
 472                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 473                 GOTO(out_dt, err);
 474         }
 475
 476         mutex_lock(&sbi->ll_lco.lco_lock);
 477         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 478         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 479         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 480         mutex_unlock(&sbi->ll_lco.lco_lock);
 481
 482         fid_zero(&sbi->ll_root_fid);
 483         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 484         if (err) {
 485                 CERROR("cannot mds_connect: rc = %d\n", err);
 486                 GOTO(out_lock_cn_cb, err);
 487         }
 488         if (!fid_is_sane(&sbi->ll_root_fid)) {
 489                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 490                        sbi->ll_md_exp->exp_obd->obd_name,
 491                        PFID(&sbi->ll_root_fid));
 492                 GOTO(out_lock_cn_cb, err = -EINVAL);
 493         }
 494         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 495
 496         sb->s_op = &lustre_super_operations;
 497 #if THREAD_SIZE >= 8192 /*b=17630*/
 498         sb->s_export_op = &lustre_export_operations;
 499 #endif
 500
 501         /* make root inode
 502          * XXX: move this to after cbd setup? */
 503         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 504         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 505                 valid |= OBD_MD_FLRMTPERM;
 506         else if (sbi->ll_flags & LL_SBI_ACL)
 507                 valid |= OBD_MD_FLACL;
 508
 509         OBD_ALLOC_PTR(op_data);
 510         if (op_data == NULL)
 511                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 512
 513         op_data->op_fid1 = sbi->ll_root_fid;
 514         op_data->op_mode = 0;
 515         op_data->op_capa1 = oc;
 516         op_data->op_valid = valid;
 517
 518         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 519         if (oc)
 520                 capa_put(oc);
 521         OBD_FREE_PTR(op_data);
 522         if (err) {
 523                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 524                        sbi->ll_md_exp->exp_obd->obd_name, err);
 525                 GOTO(out_lock_cn_cb, err);
 526         }
 527
 528         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 529                                sbi->ll_md_exp, &lmd);
 530         if (err) {
 531                 CERROR("failed to understand root inode md: rc = %d\n", err);
 532                 ptlrpc_req_finished(request);
 533                 GOTO(out_lock_cn_cb, err);
 534         }
 535
 536         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 537         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 538                                             sbi->ll_flags & LL_SBI_32BIT_API),
 539                        &lmd);
 540         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 541         ptlrpc_req_finished(request);
 542
 543         if (root == NULL || IS_ERR(root)) {
 544                 if (lmd.lsm)
 545                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 546 #ifdef CONFIG_FS_POSIX_ACL
 547                 if (lmd.posix_acl) {
 548                         posix_acl_release(lmd.posix_acl);
 549                         lmd.posix_acl = NULL;
 550                 }
 551 #endif
 552                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 553                 root = NULL;
 554                 CERROR("lustre_lite: bad iget4 for root\n");
 555                 GOTO(out_root, err);
 556         }
 557
 558         err = ll_close_thread_start(&sbi->ll_lcq);
 559         if (err) {
 560                 CERROR("cannot start close thread: rc %d\n", err);
 561                 GOTO(out_root, err);
 562         }
 563
 564 #ifdef CONFIG_FS_POSIX_ACL
 565         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 566                 rct_init(&sbi->ll_rct);
 567                 et_init(&sbi->ll_et);
 568         }
 569 #endif
 570
 571         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 572         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 573                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 574                                  NULL);
 575         cl_sb_init(sb);
 576
 577         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 578                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 579                                  &sbi->ll_cache, NULL);
 580
 581         sb->s_root = d_make_root(root);
 582         if (sb->s_root == NULL) {
 583                 CERROR("%s: can't make root dentry\n",
 584                         ll_get_fsname(sb, NULL, 0));
 585                 GOTO(out_root, err = -ENOMEM);
 586         }
 587 #ifdef HAVE_DCACHE_LOCK
 588         sb->s_root->d_op = &ll_d_ops;
 589 #endif
 590
 591         sbi->ll_sdev_orig = sb->s_dev;
 592
 593         /* We set sb->s_dev equal on all lustre clients in order to support
 594          * NFS export clustering.  NFSD requires that the FSID be the same
 595          * on all clients. */
 596         /* s_dev is also used in lt_compare() to compare two fs, but that is
 597          * only a node-local comparison. */
 598         uuid = obd_get_uuid(sbi->ll_md_exp);
 599         if (uuid != NULL) {
 600                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 601                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 602         }
 603
 604         if (data != NULL)
 605                 OBD_FREE_PTR(data);
 606         if (osfs != NULL)
 607                 OBD_FREE_PTR(osfs);
 608
 609         RETURN(err);
 610 out_root:
 611         if (root)
 612                 iput(root);
 613 out_lock_cn_cb:
 614         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 615 out_dt:
 616         obd_disconnect(sbi->ll_dt_exp);
 617         sbi->ll_dt_exp = NULL;
 618         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 619         obd_zombie_barrier();
 620 out_md_fid:
 621         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 622 out_md:
 623         obd_disconnect(sbi->ll_md_exp);
 624         sbi->ll_md_exp = NULL;
 625 out:
 626         if (data != NULL)
 627                 OBD_FREE_PTR(data);
 628         if (osfs != NULL)
 629                 OBD_FREE_PTR(osfs);
 630         lprocfs_unregister_mountpoint(sbi);
 631         return err;
 632 }
 633
 634 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 635 {
 636         int size, rc;
 637
 638         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 639         size = sizeof(int);
 640         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 641                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 642         if (rc)
 643                 CERROR("Get max mdsize error rc %d \n", rc);
 644
 645         RETURN(rc);
 646 }
 647
 648 void ll_dump_inode(struct inode *inode)
 649 {
 650         struct ll_d_hlist_node *tmp;
 651         int dentry_count = 0;
 652
 653         LASSERT(inode != NULL);
 654
 655         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 656                 dentry_count++;
 657
 658         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 659                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 660                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 661 }
 662
 663 void lustre_dump_dentry(struct dentry *dentry, int recur)
 664 {
 665         struct list_head *tmp;
 666         int subdirs = 0;
 667
 668         LASSERT(dentry != NULL);
 669
 670         list_for_each(tmp, &dentry->d_subdirs)
 671                 subdirs++;
 672
 673         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 674                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 675                dentry->d_name.len, dentry->d_name.name,
 676                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 677                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 678                dentry->d_flags, dentry->d_fsdata, subdirs);
 679         if (dentry->d_inode != NULL)
 680                 ll_dump_inode(dentry->d_inode);
 681
 682         if (recur == 0)
 683                 return;
 684
 685         list_for_each(tmp, &dentry->d_subdirs) {
 686                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 687                 lustre_dump_dentry(d, recur - 1);
 688         }
 689 }
 690
 691 void client_common_put_super(struct super_block *sb)
 692 {
 693         struct ll_sb_info *sbi = ll_s2sbi(sb);
 694         ENTRY;
 695
 696 #ifdef CONFIG_FS_POSIX_ACL
 697         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 698                 et_fini(&sbi->ll_et);
 699                 rct_fini(&sbi->ll_rct);
 700         }
 701 #endif
 702
 703         ll_close_thread_shutdown(sbi->ll_lcq);
 704
 705         cl_sb_fini(sb);
 706
 707         cfs_list_del(&sbi->ll_conn_chain);
 708
 709         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 710         obd_disconnect(sbi->ll_dt_exp);
 711         sbi->ll_dt_exp = NULL;
 712         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 713          * see LU-2543. */
 714         obd_zombie_barrier();
 715
 716         lprocfs_unregister_mountpoint(sbi);
 717
 718         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 719         obd_disconnect(sbi->ll_md_exp);
 720         sbi->ll_md_exp = NULL;
 721
 722         EXIT;
 723 }
 724
 725 void ll_kill_super(struct super_block *sb)
 726 {
 727         struct ll_sb_info *sbi;
 728
 729         ENTRY;
 730
 731         /* not init sb ?*/
 732         if (!(sb->s_flags & MS_ACTIVE))
 733                 return;
 734
 735         sbi = ll_s2sbi(sb);
 736         /* we need restore s_dev from changed for clustred NFS before put_super
 737          * because new kernels have cached s_dev and change sb->s_dev in
 738          * put_super not affected real removing devices */
 739         if (sbi) {
 740                 sb->s_dev = sbi->ll_sdev_orig;
 741                 sbi->ll_umounting = 1;
 742         }
 743         EXIT;
 744 }
 745
 746 char *ll_read_opt(const char *opt, char *data)
 747 {
 748         char *value;
 749         char *retval;
 750         ENTRY;
 751
 752         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 753         if (strncmp(opt, data, strlen(opt)))
 754                 RETURN(NULL);
 755         if ((value = strchr(data, '=')) == NULL)
 756                 RETURN(NULL);
 757
 758         value++;
 759         OBD_ALLOC(retval, strlen(value) + 1);
 760         if (!retval) {
 761                 CERROR("out of memory!\n");
 762                 RETURN(NULL);
 763         }
 764
 765         memcpy(retval, value, strlen(value)+1);
 766         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 767         RETURN(retval);
 768 }
 769
 770 static inline int ll_set_opt(const char *opt, char *data, int fl)
 771 {
 772         if (strncmp(opt, data, strlen(opt)) != 0)
 773                 return(0);
 774         else
 775                 return(fl);
 776 }
 777
 778 /* non-client-specific mount options are parsed in lmd_parse */
 779 static int ll_options(char *options, int *flags)
 780 {
 781         int tmp;
 782         char *s1 = options, *s2;
 783         ENTRY;
 784
 785         if (!options)
 786                 RETURN(0);
 787
 788         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 789
 790         while (*s1) {
 791                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 792                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 793                 if (tmp) {
 794                         *flags |= tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 808                 if (tmp) {
 809                         *flags &= ~tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 818                 if (tmp) {
 819                         *flags &= ~tmp;
 820                         goto next;
 821                 }
 822 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 823                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 824                 if (tmp) {
 825                         /* Ignore deprecated mount option.  The client will
 826                          * always try to mount with ACL support, whether this
 827                          * is used depends on whether server supports it. */
 828                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 829                                                   "mount option 'acl'.\n");
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 833                 if (tmp) {
 834                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 835                                                   "mount option 'noacl'.\n");
 836                         goto next;
 837                 }
 838 #else
 839 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 840 #endif
 841                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 842                 if (tmp) {
 843                         *flags |= tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856
 857                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 893                 if (tmp) {
 894                         *flags |= tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 903                 if (tmp) {
 904                         *flags &= ~tmp;
 905                         goto next;
 906                 }
 907                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 908                                    s1);
 909                 RETURN(-EINVAL);
 910
 911 next:
 912                 /* Find next opt */
 913                 s2 = strchr(s1, ',');
 914                 if (s2 == NULL)
 915                         break;
 916                 s1 = s2 + 1;
 917         }
 918         RETURN(0);
 919 }
 920
 921 void ll_lli_init(struct ll_inode_info *lli)
 922 {
 923         lli->lli_inode_magic = LLI_INODE_MAGIC;
 924         lli->lli_flags = 0;
 925         lli->lli_ioepoch = 0;
 926         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 927         spin_lock_init(&lli->lli_lock);
 928         lli->lli_posix_acl = NULL;
 929         lli->lli_remote_perms = NULL;
 930         mutex_init(&lli->lli_rmtperm_mutex);
 931         /* Do not set lli_fid, it has been initialized already. */
 932         fid_zero(&lli->lli_pfid);
 933         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 934         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 935         cfs_atomic_set(&lli->lli_open_count, 0);
 936         lli->lli_mds_capa = NULL;
 937         lli->lli_rmtperm_time = 0;
 938         lli->lli_pending_och = NULL;
 939         lli->lli_mds_read_och = NULL;
 940         lli->lli_mds_write_och = NULL;
 941         lli->lli_mds_exec_och = NULL;
 942         lli->lli_open_fd_read_count = 0;
 943         lli->lli_open_fd_write_count = 0;
 944         lli->lli_open_fd_exec_count = 0;
 945         mutex_init(&lli->lli_och_mutex);
 946         spin_lock_init(&lli->lli_agl_lock);
 947         lli->lli_has_smd = false;
 948         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 949         lli->lli_clob = NULL;
 950
 951         init_rwsem(&lli->lli_xattrs_list_rwsem);
 952         mutex_init(&lli->lli_xattrs_enq_lock);
 953
 954         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 955         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 956                 mutex_init(&lli->lli_readdir_mutex);
 957                 lli->lli_opendir_key = NULL;
 958                 lli->lli_sai = NULL;
 959                 lli->lli_def_acl = NULL;
 960                 spin_lock_init(&lli->lli_sa_lock);
 961                 lli->lli_opendir_pid = 0;
 962         } else {
 963                 sema_init(&lli->lli_size_sem, 1);
 964                 lli->lli_size_sem_owner = NULL;
 965                 lli->lli_symlink_name = NULL;
 966                 init_rwsem(&lli->lli_trunc_sem);
 967                 mutex_init(&lli->lli_write_mutex);
 968                 init_rwsem(&lli->lli_glimpse_sem);
 969                 lli->lli_glimpse_time = 0;
 970                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 971                 lli->lli_agl_index = 0;
 972                 lli->lli_async_rc = 0;
 973                 lli->lli_volatile = false;
 974         }
 975         mutex_init(&lli->lli_layout_mutex);
 976 }
 977
 978 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 979 {
 980         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 981
 982         bdi->name = "lustre";
 983         return bdi_register(bdi, NULL, "lustre-%d",
 984                             atomic_inc_return(&ll_bdi_num));
 985 }
 986
 987 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 988 {
 989         struct lustre_profile *lprof = NULL;
 990         struct lustre_sb_info *lsi = s2lsi(sb);
 991         struct ll_sb_info *sbi;
 992         char  *dt = NULL, *md = NULL;
 993         char  *profilenm = get_profile_name(sb);
 994         struct config_llog_instance *cfg;
 995         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 996         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 997         int    err;
 998         ENTRY;
 999
1000         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1001
1002         OBD_ALLOC_PTR(cfg);
1003         if (cfg == NULL)
1004                 RETURN(-ENOMEM);
1005
1006         cfs_module_get();
1007
1008         /* client additional sb info */
1009         lsi->lsi_llsbi = sbi = ll_init_sbi();
1010         if (!sbi) {
1011                 cfs_module_put(THIS_MODULE);
1012                 OBD_FREE_PTR(cfg);
1013                 RETURN(-ENOMEM);
1014         }
1015
1016         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1017         if (err)
1018                 GOTO(out_free, err);
1019
1020         err = bdi_init(&lsi->lsi_bdi);
1021         if (err)
1022                 GOTO(out_free, err);
1023         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1024         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1025         err = ll_bdi_register(&lsi->lsi_bdi);
1026         if (err)
1027                 GOTO(out_free, err);
1028
1029         sb->s_bdi = &lsi->lsi_bdi;
1030 #ifndef HAVE_DCACHE_LOCK
1031         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1032         sb->s_d_op = &ll_d_ops;
1033 #endif
1034
1035         /* Generate a string unique to this super, in case some joker tries
1036            to mount the same fs at two mount points.
1037            Use the address of the super itself.*/
1038         cfg->cfg_instance = sb;
1039         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1040         cfg->cfg_callback = class_config_llog_handler;
1041         /* set up client obds */
1042         err = lustre_process_log(sb, profilenm, cfg);
1043         if (err < 0) {
1044                 CERROR("Unable to process log: %d\n", err);
1045                 GOTO(out_free, err);
1046         }
1047
1048         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1049         lprof = class_get_profile(profilenm);
1050         if (lprof == NULL) {
1051                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1052                                    " read from the MGS.  Does that filesystem "
1053                                    "exist?\n", profilenm);
1054                 GOTO(out_free, err = -EINVAL);
1055         }
1056         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1057                lprof->lp_md, lprof->lp_dt);
1058
1059         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1060         if (!dt)
1061                 GOTO(out_free, err = -ENOMEM);
1062         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1063
1064         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1065         if (!md)
1066                 GOTO(out_free, err = -ENOMEM);
1067         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1068
1069         /* connections, registrations, sb setup */
1070         err = client_common_fill_super(sb, md, dt, mnt);
1071
1072 out_free:
1073         if (md)
1074                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1075         if (dt)
1076                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1077         if (err)
1078                 ll_put_super(sb);
1079         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1080                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1081
1082         OBD_FREE_PTR(cfg);
1083         RETURN(err);
1084 } /* ll_fill_super */
1085
1086
1087 void lu_context_keys_dump(void);
1088
1089 void ll_put_super(struct super_block *sb)
1090 {
1091         struct config_llog_instance cfg, params_cfg;
1092         struct obd_device *obd;
1093         struct lustre_sb_info *lsi = s2lsi(sb);
1094         struct ll_sb_info *sbi = ll_s2sbi(sb);
1095         char *profilenm = get_profile_name(sb);
1096         int force = 1, next;
1097         ENTRY;
1098
1099         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1100
1101         ll_print_capa_stat(sbi);
1102
1103         cfg.cfg_instance = sb;
1104         lustre_end_log(sb, profilenm, &cfg);
1105
1106         params_cfg.cfg_instance = sb;
1107         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1108
1109         if (sbi->ll_md_exp) {
1110                 obd = class_exp2obd(sbi->ll_md_exp);
1111                 if (obd)
1112                         force = obd->obd_force;
1113         }
1114
1115         /* We need to set force before the lov_disconnect in
1116            lustre_common_put_super, since l_d cleans up osc's as well. */
1117         if (force) {
1118                 next = 0;
1119                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1120                                                      &next)) != NULL) {
1121                         obd->obd_force = force;
1122                 }
1123         }
1124
1125         if (sbi->ll_lcq) {
1126                 /* Only if client_common_fill_super succeeded */
1127                 client_common_put_super(sb);
1128         }
1129
1130         next = 0;
1131         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1132                 class_manual_cleanup(obd);
1133         }
1134
1135         if (sbi->ll_flags & LL_SBI_VERBOSE)
1136                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1137
1138         if (profilenm)
1139                 class_del_profile(profilenm);
1140
1141         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1142                 bdi_destroy(&lsi->lsi_bdi);
1143                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1144         }
1145
1146         ll_free_sbi(sb);
1147         lsi->lsi_llsbi = NULL;
1148
1149         lustre_common_put_super(sb);
1150
1151         cfs_module_put(THIS_MODULE);
1152
1153         EXIT;
1154 } /* client_put_super */
1155
1156 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1157 {
1158         struct inode *inode = NULL;
1159
1160         /* NOTE: we depend on atomic igrab() -bzzz */
1161         lock_res_and_lock(lock);
1162         if (lock->l_resource->lr_lvb_inode) {
1163                 struct ll_inode_info * lli;
1164                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1165                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1166                         inode = igrab(lock->l_resource->lr_lvb_inode);
1167                 } else {
1168                         inode = lock->l_resource->lr_lvb_inode;
1169                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1170                                          D_WARNING, lock, "lr_lvb_inode %p is "
1171                                          "bogus: magic %08x",
1172                                          lock->l_resource->lr_lvb_inode,
1173                                          lli->lli_inode_magic);
1174                         inode = NULL;
1175                 }
1176         }
1177         unlock_res_and_lock(lock);
1178         return inode;
1179 }
1180
1181 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1182 {
1183         struct inode *inode = NULL;
1184         /* NOTE: we depend on atomic igrab() -bzzz */
1185         lock_res_and_lock(lock);
1186         if (lock->l_ast_data) {
1187                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1188                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1189                         inode = igrab(lock->l_ast_data);
1190                 } else {
1191                         inode = lock->l_ast_data;
1192                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1193                                          D_WARNING, lock, "l_ast_data %p is "
1194                                          "bogus: magic %08x", lock->l_ast_data,
1195                                          lli->lli_inode_magic);
1196                         inode = NULL;
1197                 }
1198         }
1199         unlock_res_and_lock(lock);
1200         return inode;
1201 }
1202
1203 void ll_clear_inode(struct inode *inode)
1204 {
1205         struct ll_inode_info *lli = ll_i2info(inode);
1206         struct ll_sb_info *sbi = ll_i2sbi(inode);
1207         ENTRY;
1208
1209         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1210                inode->i_generation, inode);
1211
1212         if (S_ISDIR(inode->i_mode)) {
1213                 /* these should have been cleared in ll_file_release */
1214                 LASSERT(lli->lli_opendir_key == NULL);
1215                 LASSERT(lli->lli_sai == NULL);
1216                 LASSERT(lli->lli_opendir_pid == 0);
1217         }
1218
1219         spin_lock(&lli->lli_lock);
1220         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1221         spin_unlock(&lli->lli_lock);
1222         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1223
1224         LASSERT(!lli->lli_open_fd_write_count);
1225         LASSERT(!lli->lli_open_fd_read_count);
1226         LASSERT(!lli->lli_open_fd_exec_count);
1227
1228         if (lli->lli_mds_write_och)
1229                 ll_md_real_close(inode, FMODE_WRITE);
1230         if (lli->lli_mds_exec_och)
1231                 ll_md_real_close(inode, FMODE_EXEC);
1232         if (lli->lli_mds_read_och)
1233                 ll_md_real_close(inode, FMODE_READ);
1234
1235         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1236                 OBD_FREE(lli->lli_symlink_name,
1237                          strlen(lli->lli_symlink_name) + 1);
1238                 lli->lli_symlink_name = NULL;
1239         }
1240
1241         ll_xattr_cache_destroy(inode);
1242
1243         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1244                 LASSERT(lli->lli_posix_acl == NULL);
1245                 if (lli->lli_remote_perms) {
1246                         free_rmtperm_hash(lli->lli_remote_perms);
1247                         lli->lli_remote_perms = NULL;
1248                 }
1249         }
1250 #ifdef CONFIG_FS_POSIX_ACL
1251         else if (lli->lli_posix_acl) {
1252                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1253                 LASSERT(lli->lli_remote_perms == NULL);
1254                 posix_acl_release(lli->lli_posix_acl);
1255                 lli->lli_posix_acl = NULL;
1256         }
1257 #endif
1258         lli->lli_inode_magic = LLI_INODE_DEAD;
1259
1260         ll_clear_inode_capas(inode);
1261         if (!S_ISDIR(inode->i_mode))
1262                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1263
1264         /*
1265          * XXX This has to be done before lsm is freed below, because
1266          * cl_object still uses inode lsm.
1267          */
1268         cl_inode_fini(inode);
1269         lli->lli_has_smd = false;
1270
1271         EXIT;
1272 }
1273
1274 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1275                   struct md_open_data **mod)
1276 {
1277         struct lustre_md md;
1278         struct inode *inode = dentry->d_inode;
1279         struct ll_sb_info *sbi = ll_i2sbi(inode);
1280         struct ptlrpc_request *request = NULL;
1281         int rc, ia_valid;
1282         ENTRY;
1283
1284         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1285                                      LUSTRE_OPC_ANY, NULL);
1286         if (IS_ERR(op_data))
1287                 RETURN(PTR_ERR(op_data));
1288
1289         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1290                         &request, mod);
1291         if (rc) {
1292                 ptlrpc_req_finished(request);
1293                 if (rc == -ENOENT) {
1294                         clear_nlink(inode);
1295                         /* Unlinked special device node? Or just a race?
1296                          * Pretend we done everything. */
1297                         if (!S_ISREG(inode->i_mode) &&
1298                             !S_ISDIR(inode->i_mode)) {
1299                                 ia_valid = op_data->op_attr.ia_valid;
1300                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1301                                 rc = simple_setattr(dentry, &op_data->op_attr);
1302                                 op_data->op_attr.ia_valid = ia_valid;
1303                         }
1304                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1305                         CERROR("md_setattr fails: rc = %d\n", rc);
1306                 }
1307                 RETURN(rc);
1308         }
1309
1310         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1311                               sbi->ll_md_exp, &md);
1312         if (rc) {
1313                 ptlrpc_req_finished(request);
1314                 RETURN(rc);
1315         }
1316
1317         ia_valid = op_data->op_attr.ia_valid;
1318         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1319          * cache is not cleared yet. */
1320         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1321         rc = simple_setattr(dentry, &op_data->op_attr);
1322         op_data->op_attr.ia_valid = ia_valid;
1323
1324         /* Extract epoch data if obtained. */
1325         op_data->op_handle = md.body->handle;
1326         op_data->op_ioepoch = md.body->ioepoch;
1327
1328         ll_update_inode(inode, &md);
1329         ptlrpc_req_finished(request);
1330
1331         RETURN(rc);
1332 }
1333
1334 /* Close IO epoch and send Size-on-MDS attribute update. */
1335 static int ll_setattr_done_writing(struct inode *inode,
1336                                    struct md_op_data *op_data,
1337                                    struct md_open_data *mod)
1338 {
1339         struct ll_inode_info *lli = ll_i2info(inode);
1340         int rc = 0;
1341         ENTRY;
1342
1343         LASSERT(op_data != NULL);
1344         if (!S_ISREG(inode->i_mode))
1345                 RETURN(0);
1346
1347         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1348                op_data->op_ioepoch, PFID(&lli->lli_fid));
1349
1350         op_data->op_flags = MF_EPOCH_CLOSE;
1351         ll_done_writing_attr(inode, op_data);
1352         ll_pack_inode2opdata(inode, op_data, NULL);
1353
1354         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1355         if (rc == -EAGAIN) {
1356                 /* MDS has instructed us to obtain Size-on-MDS attribute
1357                  * from OSTs and send setattr to back to MDS. */
1358                 rc = ll_som_update(inode, op_data);
1359         } else if (rc) {
1360                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1361                        inode->i_ino, rc);
1362         }
1363         RETURN(rc);
1364 }
1365
1366 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1367 {
1368         struct obd_capa *capa;
1369         int rc;
1370
1371         if (attr->ia_valid & ATTR_SIZE)
1372                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1373         else
1374                 capa = ll_mdscapa_get(inode);
1375
1376         rc = cl_setattr_ost(inode, attr, capa);
1377
1378         if (attr->ia_valid & ATTR_SIZE)
1379                 ll_truncate_free_capa(capa);
1380         else
1381                 capa_put(capa);
1382
1383         return rc;
1384 }
1385
1386 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1387  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1388  * keep these values until such a time that objects are allocated for it.
1389  * We do the MDS operations first, as it is checking permissions for us.
1390  * We don't to the MDS RPC if there is nothing that we want to store there,
1391  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1392  * going to do an RPC anyways.
1393  *
1394  * If we are doing a truncate, we will send the mtime and ctime updates
1395  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1396  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1397  * at the same time.
1398  *
1399  * In case of HSMimport, we only set attr on MDS.
1400  */
1401 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1402 {
1403         struct inode *inode = dentry->d_inode;
1404         struct ll_inode_info *lli = ll_i2info(inode);
1405         struct md_op_data *op_data = NULL;
1406         struct md_open_data *mod = NULL;
1407         bool file_is_released = false;
1408         int rc = 0, rc1 = 0;
1409         ENTRY;
1410
1411         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1412                 "valid %x, hsm_import %d\n",
1413                 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1414                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1415                 attr->ia_valid, hsm_import);
1416
1417         if (attr->ia_valid & ATTR_SIZE) {
1418                 /* Check new size against VFS/VM file size limit and rlimit */
1419                 rc = inode_newsize_ok(inode, attr->ia_size);
1420                 if (rc)
1421                         RETURN(rc);
1422
1423                 /* The maximum Lustre file size is variable, based on the
1424                  * OST maximum object size and number of stripes.  This
1425                  * needs another check in addition to the VFS check above. */
1426                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1427                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1428                                PFID(&lli->lli_fid), attr->ia_size,
1429                                ll_file_maxbytes(inode));
1430                         RETURN(-EFBIG);
1431                 }
1432
1433                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1434         }
1435
1436         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1437         if (attr->ia_valid & TIMES_SET_FLAGS) {
1438                 if (cfs_curproc_fsuid() != inode->i_uid &&
1439                     !cfs_capable(CFS_CAP_FOWNER))
1440                         RETURN(-EPERM);
1441         }
1442
1443         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1444         if (attr->ia_valid & ATTR_CTIME) {
1445                 attr->ia_ctime = CFS_CURRENT_TIME;
1446                 attr->ia_valid |= ATTR_CTIME_SET;
1447         }
1448         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1449             (attr->ia_valid & ATTR_ATIME)) {
1450                 attr->ia_atime = CFS_CURRENT_TIME;
1451                 attr->ia_valid |= ATTR_ATIME_SET;
1452         }
1453         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1454             (attr->ia_valid & ATTR_MTIME)) {
1455                 attr->ia_mtime = CFS_CURRENT_TIME;
1456                 attr->ia_valid |= ATTR_MTIME_SET;
1457         }
1458
1459         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1460                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1461                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1462                        cfs_time_current_sec());
1463
1464         /* If we are changing file size, file content is modified, flag it. */
1465         if (attr->ia_valid & ATTR_SIZE) {
1466                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1467                 spin_lock(&lli->lli_lock);
1468                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1469                 spin_unlock(&lli->lli_lock);
1470         }
1471
1472         /* We always do an MDS RPC, even if we're only changing the size;
1473          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1474
1475         OBD_ALLOC_PTR(op_data);
1476         if (op_data == NULL)
1477                 RETURN(-ENOMEM);
1478
1479         if (!S_ISDIR(inode->i_mode)) {
1480                 if (attr->ia_valid & ATTR_SIZE)
1481                         inode_dio_write_done(inode);
1482                 mutex_unlock(&inode->i_mutex);
1483                 down_write(&lli->lli_trunc_sem);
1484         }
1485
1486         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1487
1488         /* Open epoch for truncate. */
1489         if (exp_connect_som(ll_i2mdexp(inode)) &&
1490             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1491                 op_data->op_flags = MF_EPOCH_OPEN;
1492
1493         /* truncate on a released file must failed with -ENODATA,
1494          * so size must not be set on MDS for released file
1495          * but other attributes must be set
1496          */
1497         if (S_ISREG(inode->i_mode)) {
1498                 struct lov_stripe_md *lsm;
1499                 __u32 gen;
1500
1501                 ll_layout_refresh(inode, &gen);
1502                 lsm = ccc_inode_lsm_get(inode);
1503                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1504                         file_is_released = true;
1505                 ccc_inode_lsm_put(inode, lsm);
1506         }
1507
1508         /* if not in HSM import mode, clear size attr for released file
1509          * we clear the attribute send to MDT in op_data, not the original
1510          * received from caller in attr which is used later to
1511          * decide return code */
1512         if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1513                 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1514
1515         rc = ll_md_setattr(dentry, op_data, &mod);
1516         if (rc)
1517                 GOTO(out, rc);
1518
1519         /* truncate failed (only when non HSM import), others succeed */
1520         if (file_is_released) {
1521                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1522                         GOTO(out, rc = -ENODATA);
1523                 else
1524                         GOTO(out, rc = 0);
1525         }
1526
1527         /* RPC to MDT is sent, cancel data modification flag */
1528         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1529                 spin_lock(&lli->lli_lock);
1530                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1531                 spin_unlock(&lli->lli_lock);
1532         }
1533
1534         ll_ioepoch_open(lli, op_data->op_ioepoch);
1535         if (!S_ISREG(inode->i_mode))
1536                 GOTO(out, rc = 0);
1537
1538         if (attr->ia_valid & (ATTR_SIZE |
1539                               ATTR_ATIME | ATTR_ATIME_SET |
1540                               ATTR_MTIME | ATTR_MTIME_SET)) {
1541                 /* For truncate and utimes sending attributes to OSTs, setting
1542                  * mtime/atime to the past will be performed under PW [0:EOF]
1543                  * extent lock (new_size:EOF for truncate).  It may seem
1544                  * excessive to send mtime/atime updates to OSTs when not
1545                  * setting times to past, but it is necessary due to possible
1546                  * time de-synchronization between MDT inode and OST objects */
1547                 rc = ll_setattr_ost(inode, attr);
1548         }
1549         EXIT;
1550 out:
1551         if (op_data) {
1552                 if (op_data->op_ioepoch) {
1553                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1554                         if (!rc)
1555                                 rc = rc1;
1556                 }
1557                 ll_finish_md_op_data(op_data);
1558         }
1559         if (!S_ISDIR(inode->i_mode)) {
1560                 up_write(&lli->lli_trunc_sem);
1561                 mutex_lock(&inode->i_mutex);
1562                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1563                         inode_dio_wait(inode);
1564         }
1565
1566         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1567                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1568
1569         return rc;
1570 }
1571
1572 int ll_setattr(struct dentry *de, struct iattr *attr)
1573 {
1574         int mode = de->d_inode->i_mode;
1575
1576         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1577                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1578                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1579
1580         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1581                                (ATTR_SIZE|ATTR_MODE)) &&
1582             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1583              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1584               !(attr->ia_mode & S_ISGID))))
1585                 attr->ia_valid |= ATTR_FORCE;
1586
1587         if ((mode & S_ISUID) &&
1588             !(attr->ia_mode & S_ISUID) &&
1589             !(attr->ia_valid & ATTR_KILL_SUID))
1590                 attr->ia_valid |= ATTR_KILL_SUID;
1591
1592         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1593             !(attr->ia_mode & S_ISGID) &&
1594             !(attr->ia_valid & ATTR_KILL_SGID))
1595                 attr->ia_valid |= ATTR_KILL_SGID;
1596
1597         return ll_setattr_raw(de, attr, false);
1598 }
1599
1600 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1601                        __u64 max_age, __u32 flags)
1602 {
1603         struct ll_sb_info *sbi = ll_s2sbi(sb);
1604         struct obd_statfs obd_osfs;
1605         int rc;
1606         ENTRY;
1607
1608         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1609         if (rc) {
1610                 CERROR("md_statfs fails: rc = %d\n", rc);
1611                 RETURN(rc);
1612         }
1613
1614         osfs->os_type = sb->s_magic;
1615
1616         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1617                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1618
1619         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1620                 flags |= OBD_STATFS_NODELAY;
1621
1622         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1623         if (rc) {
1624                 CERROR("obd_statfs fails: rc = %d\n", rc);
1625                 RETURN(rc);
1626         }
1627
1628         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1629                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1630                obd_osfs.os_files);
1631
1632         osfs->os_bsize = obd_osfs.os_bsize;
1633         osfs->os_blocks = obd_osfs.os_blocks;
1634         osfs->os_bfree = obd_osfs.os_bfree;
1635         osfs->os_bavail = obd_osfs.os_bavail;
1636
1637         /* If we don't have as many objects free on the OST as inodes
1638          * on the MDS, we reduce the total number of inodes to
1639          * compensate, so that the "inodes in use" number is correct.
1640          */
1641         if (obd_osfs.os_ffree < osfs->os_ffree) {
1642                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1643                         obd_osfs.os_ffree;
1644                 osfs->os_ffree = obd_osfs.os_ffree;
1645         }
1646
1647         RETURN(rc);
1648 }
1649 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1650 {
1651         struct super_block *sb = de->d_sb;
1652         struct obd_statfs osfs;
1653         int rc;
1654
1655         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1656         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1657
1658         /* Some amount of caching on the client is allowed */
1659         rc = ll_statfs_internal(sb, &osfs,
1660                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1661                                 0);
1662         if (rc)
1663                 return rc;
1664
1665         statfs_unpack(sfs, &osfs);
1666
1667         /* We need to downshift for all 32-bit kernels, because we can't
1668          * tell if the kernel is being called via sys_statfs64() or not.
1669          * Stop before overflowing f_bsize - in which case it is better
1670          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1671         if (sizeof(long) < 8) {
1672                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1673                         sfs->f_bsize <<= 1;
1674
1675                         osfs.os_blocks >>= 1;
1676                         osfs.os_bfree >>= 1;
1677                         osfs.os_bavail >>= 1;
1678                 }
1679         }
1680
1681         sfs->f_blocks = osfs.os_blocks;
1682         sfs->f_bfree = osfs.os_bfree;
1683         sfs->f_bavail = osfs.os_bavail;
1684         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1685         return 0;
1686 }
1687
1688 void ll_inode_size_lock(struct inode *inode)
1689 {
1690         struct ll_inode_info *lli;
1691
1692         LASSERT(!S_ISDIR(inode->i_mode));
1693
1694         lli = ll_i2info(inode);
1695         LASSERT(lli->lli_size_sem_owner != current);
1696         down(&lli->lli_size_sem);
1697         LASSERT(lli->lli_size_sem_owner == NULL);
1698         lli->lli_size_sem_owner = current;
1699 }
1700
1701 void ll_inode_size_unlock(struct inode *inode)
1702 {
1703         struct ll_inode_info *lli;
1704
1705         lli = ll_i2info(inode);
1706         LASSERT(lli->lli_size_sem_owner == current);
1707         lli->lli_size_sem_owner = NULL;
1708         up(&lli->lli_size_sem);
1709 }
1710
1711 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1712 {
1713         struct ll_inode_info *lli = ll_i2info(inode);
1714         struct mdt_body *body = md->body;
1715         struct lov_stripe_md *lsm = md->lsm;
1716         struct ll_sb_info *sbi = ll_i2sbi(inode);
1717
1718         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1719         if (lsm != NULL) {
1720                 if (!lli->lli_has_smd &&
1721                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1722                         cl_file_inode_init(inode, md);
1723
1724                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1725                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1726                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1727         }
1728
1729         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1730                 if (body->valid & OBD_MD_FLRMTPERM)
1731                         ll_update_remote_perm(inode, md->remote_perm);
1732         }
1733 #ifdef CONFIG_FS_POSIX_ACL
1734         else if (body->valid & OBD_MD_FLACL) {
1735                 spin_lock(&lli->lli_lock);
1736                 if (lli->lli_posix_acl)
1737                         posix_acl_release(lli->lli_posix_acl);
1738                 lli->lli_posix_acl = md->posix_acl;
1739                 spin_unlock(&lli->lli_lock);
1740         }
1741 #endif
1742         inode->i_ino = cl_fid_build_ino(&body->fid1,
1743                                         sbi->ll_flags & LL_SBI_32BIT_API);
1744         inode->i_generation = cl_fid_build_gen(&body->fid1);
1745
1746         if (body->valid & OBD_MD_FLATIME) {
1747                 if (body->atime > LTIME_S(inode->i_atime))
1748                         LTIME_S(inode->i_atime) = body->atime;
1749                 lli->lli_lvb.lvb_atime = body->atime;
1750         }
1751         if (body->valid & OBD_MD_FLMTIME) {
1752                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1753                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1754                                "to "LPU64"\n", inode->i_ino,
1755                                LTIME_S(inode->i_mtime), body->mtime);
1756                         LTIME_S(inode->i_mtime) = body->mtime;
1757                 }
1758                 lli->lli_lvb.lvb_mtime = body->mtime;
1759         }
1760         if (body->valid & OBD_MD_FLCTIME) {
1761                 if (body->ctime > LTIME_S(inode->i_ctime))
1762                         LTIME_S(inode->i_ctime) = body->ctime;
1763                 lli->lli_lvb.lvb_ctime = body->ctime;
1764         }
1765         if (body->valid & OBD_MD_FLMODE)
1766                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1767         if (body->valid & OBD_MD_FLTYPE)
1768                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1769         LASSERT(inode->i_mode != 0);
1770         if (S_ISREG(inode->i_mode)) {
1771                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1772         } else {
1773                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1774         }
1775         if (body->valid & OBD_MD_FLUID)
1776                 inode->i_uid = body->uid;
1777         if (body->valid & OBD_MD_FLGID)
1778                 inode->i_gid = body->gid;
1779         if (body->valid & OBD_MD_FLFLAGS)
1780                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1781         if (body->valid & OBD_MD_FLNLINK)
1782                 set_nlink(inode, body->nlink);
1783         if (body->valid & OBD_MD_FLRDEV)
1784                 inode->i_rdev = old_decode_dev(body->rdev);
1785
1786         if (body->valid & OBD_MD_FLID) {
1787                 /* FID shouldn't be changed! */
1788                 if (fid_is_sane(&lli->lli_fid)) {
1789                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1790                                  "Trying to change FID "DFID
1791                                  " to the "DFID", inode %lu/%u(%p)\n",
1792                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1793                                  inode->i_ino, inode->i_generation, inode);
1794                 } else
1795                         lli->lli_fid = body->fid1;
1796         }
1797
1798         LASSERT(fid_seq(&lli->lli_fid) != 0);
1799
1800         if (body->valid & OBD_MD_FLSIZE) {
1801                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1802                     S_ISREG(inode->i_mode)) {
1803                         struct lustre_handle lockh;
1804                         ldlm_mode_t mode;
1805
1806                         /* As it is possible a blocking ast has been processed
1807                          * by this time, we need to check there is an UPDATE
1808                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1809                          * it. */
1810                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1811                                                &lockh, LDLM_FL_CBPENDING,
1812                                                LCK_CR | LCK_CW |
1813                                                LCK_PR | LCK_PW);
1814                         if (mode) {
1815                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1816                                                       LLIF_EPOCH_PENDING |
1817                                                       LLIF_SOM_DIRTY)) {
1818                                         CERROR("ino %lu flags %u still has "
1819                                                "size authority! do not trust "
1820                                                "the size got from MDS\n",
1821                                                inode->i_ino, lli->lli_flags);
1822                                 } else {
1823                                         /* Use old size assignment to avoid
1824                                          * deadlock bz14138 & bz14326 */
1825                                         i_size_write(inode, body->size);
1826                                         spin_lock(&lli->lli_lock);
1827                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1828                                         spin_unlock(&lli->lli_lock);
1829                                 }
1830                                 ldlm_lock_decref(&lockh, mode);
1831                         }
1832                 } else {
1833                         /* Use old size assignment to avoid
1834                          * deadlock bz14138 & bz14326 */
1835                         i_size_write(inode, body->size);
1836
1837                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1838                                inode->i_ino, (unsigned long long)body->size);
1839                 }
1840
1841                 if (body->valid & OBD_MD_FLBLOCKS)
1842                         inode->i_blocks = body->blocks;
1843         }
1844
1845         if (body->valid & OBD_MD_FLMDSCAPA) {
1846                 LASSERT(md->mds_capa);
1847                 ll_add_capa(inode, md->mds_capa);
1848         }
1849         if (body->valid & OBD_MD_FLOSSCAPA) {
1850                 LASSERT(md->oss_capa);
1851                 ll_add_capa(inode, md->oss_capa);
1852         }
1853
1854         if (body->valid & OBD_MD_TSTATE) {
1855                 if (body->t_state & MS_RESTORE)
1856                         lli->lli_flags |= LLIF_FILE_RESTORING;
1857         }
1858 }
1859
1860 void ll_read_inode2(struct inode *inode, void *opaque)
1861 {
1862         struct lustre_md *md = opaque;
1863         struct ll_inode_info *lli = ll_i2info(inode);
1864         ENTRY;
1865
1866         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1867                PFID(&lli->lli_fid), inode);
1868
1869         LASSERT(!lli->lli_has_smd);
1870
1871         /* Core attributes from the MDS first.  This is a new inode, and
1872          * the VFS doesn't zero times in the core inode so we have to do
1873          * it ourselves.  They will be overwritten by either MDS or OST
1874          * attributes - we just need to make sure they aren't newer. */
1875         LTIME_S(inode->i_mtime) = 0;
1876         LTIME_S(inode->i_atime) = 0;
1877         LTIME_S(inode->i_ctime) = 0;
1878         inode->i_rdev = 0;
1879         ll_update_inode(inode, md);
1880
1881         /* OIDEBUG(inode); */
1882
1883         /* initializing backing dev info. */
1884         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1885
1886
1887         if (S_ISREG(inode->i_mode)) {
1888                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1889                 inode->i_op = &ll_file_inode_operations;
1890                 inode->i_fop = sbi->ll_fop;
1891                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1892                 EXIT;
1893         } else if (S_ISDIR(inode->i_mode)) {
1894                 inode->i_op = &ll_dir_inode_operations;
1895                 inode->i_fop = &ll_dir_operations;
1896                 EXIT;
1897         } else if (S_ISLNK(inode->i_mode)) {
1898                 inode->i_op = &ll_fast_symlink_inode_operations;
1899                 EXIT;
1900         } else {
1901                 inode->i_op = &ll_special_inode_operations;
1902
1903                 init_special_inode(inode, inode->i_mode,
1904                                    inode->i_rdev);
1905
1906                 EXIT;
1907         }
1908 }
1909
1910 void ll_delete_inode(struct inode *inode)
1911 {
1912         struct cl_inode_info *lli = cl_i2info(inode);
1913         ENTRY;
1914
1915         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1916                 /* discard all dirty pages before truncating them, required by
1917                  * osc_extent implementation at LU-1030. */
1918                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1919                                    CL_FSYNC_DISCARD, 1);
1920
1921         truncate_inode_pages(&inode->i_data, 0);
1922
1923         /* Workaround for LU-118 */
1924         if (inode->i_data.nrpages) {
1925                 TREE_READ_LOCK_IRQ(&inode->i_data);
1926                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1927                 LASSERTF(inode->i_data.nrpages == 0,
1928                          "inode=%lu/%u(%p) nrpages=%lu, see "
1929                          "http://jira.whamcloud.com/browse/LU-118\n",
1930                          inode->i_ino, inode->i_generation, inode,
1931                          inode->i_data.nrpages);
1932         }
1933         /* Workaround end */
1934
1935 #ifdef HAVE_SBOPS_EVICT_INODE
1936         ll_clear_inode(inode);
1937 #endif
1938         clear_inode(inode);
1939
1940         EXIT;
1941 }
1942
1943 int ll_iocontrol(struct inode *inode, struct file *file,
1944                  unsigned int cmd, unsigned long arg)
1945 {
1946         struct ll_sb_info *sbi = ll_i2sbi(inode);
1947         struct ptlrpc_request *req = NULL;
1948         int rc, flags = 0;
1949         ENTRY;
1950
1951         switch(cmd) {
1952         case FSFILT_IOC_GETFLAGS: {
1953                 struct mdt_body *body;
1954                 struct md_op_data *op_data;
1955
1956                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1957                                              0, 0, LUSTRE_OPC_ANY,
1958                                              NULL);
1959                 if (IS_ERR(op_data))
1960                         RETURN(PTR_ERR(op_data));
1961
1962                 op_data->op_valid = OBD_MD_FLFLAGS;
1963                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1964                 ll_finish_md_op_data(op_data);
1965                 if (rc) {
1966                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1967                         RETURN(-abs(rc));
1968                 }
1969
1970                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1971
1972                 flags = body->flags;
1973
1974                 ptlrpc_req_finished(req);
1975
1976                 RETURN(put_user(flags, (int *)arg));
1977         }
1978         case FSFILT_IOC_SETFLAGS: {
1979                 struct lov_stripe_md *lsm;
1980                 struct obd_info oinfo = { { { 0 } } };
1981                 struct md_op_data *op_data;
1982
1983                 if (get_user(flags, (int *)arg))
1984                         RETURN(-EFAULT);
1985
1986                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1987                                              LUSTRE_OPC_ANY, NULL);
1988                 if (IS_ERR(op_data))
1989                         RETURN(PTR_ERR(op_data));
1990
1991                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1992                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1993                 rc = md_setattr(sbi->ll_md_exp, op_data,
1994                                 NULL, 0, NULL, 0, &req, NULL);
1995                 ll_finish_md_op_data(op_data);
1996                 ptlrpc_req_finished(req);
1997                 if (rc)
1998                         RETURN(rc);
1999
2000                 inode->i_flags = ll_ext_to_inode_flags(flags);
2001
2002                 lsm = ccc_inode_lsm_get(inode);
2003                 if (!lsm_has_objects(lsm)) {
2004                         ccc_inode_lsm_put(inode, lsm);
2005                         RETURN(0);
2006                 }
2007
2008                 OBDO_ALLOC(oinfo.oi_oa);
2009                 if (!oinfo.oi_oa) {
2010                         ccc_inode_lsm_put(inode, lsm);
2011                         RETURN(-ENOMEM);
2012                 }
2013                 oinfo.oi_md = lsm;
2014                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2015                 oinfo.oi_oa->o_flags = flags;
2016                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2017                                        OBD_MD_FLGROUP;
2018                 oinfo.oi_capa = ll_mdscapa_get(inode);
2019                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2020                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2021                 capa_put(oinfo.oi_capa);
2022                 OBDO_FREE(oinfo.oi_oa);
2023                 ccc_inode_lsm_put(inode, lsm);
2024
2025                 if (rc && rc != -EPERM && rc != -EACCES)
2026                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2027
2028                 RETURN(rc);
2029         }
2030         default:
2031                 RETURN(-ENOSYS);
2032         }
2033
2034         RETURN(0);
2035 }
2036
2037 int ll_flush_ctx(struct inode *inode)
2038 {
2039         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2040
2041         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2042
2043         obd_set_info_async(NULL, sbi->ll_md_exp,
2044                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2045                            0, NULL, NULL);
2046         obd_set_info_async(NULL, sbi->ll_dt_exp,
2047                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2048                            0, NULL, NULL);
2049         return 0;
2050 }
2051
2052 /* umount -f client means force down, don't save state */
2053 void ll_umount_begin(struct super_block *sb)
2054 {
2055         struct ll_sb_info *sbi = ll_s2sbi(sb);
2056         struct obd_device *obd;
2057         struct obd_ioctl_data *ioc_data;
2058         ENTRY;
2059
2060         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2061                sb->s_count, atomic_read(&sb->s_active));
2062
2063         obd = class_exp2obd(sbi->ll_md_exp);
2064         if (obd == NULL) {
2065                 CERROR("Invalid MDC connection handle "LPX64"\n",
2066                        sbi->ll_md_exp->exp_handle.h_cookie);
2067                 EXIT;
2068                 return;
2069         }
2070         obd->obd_force = 1;
2071
2072         obd = class_exp2obd(sbi->ll_dt_exp);
2073         if (obd == NULL) {
2074                 CERROR("Invalid LOV connection handle "LPX64"\n",
2075                        sbi->ll_dt_exp->exp_handle.h_cookie);
2076                 EXIT;
2077                 return;
2078         }
2079         obd->obd_force = 1;
2080
2081         OBD_ALLOC_PTR(ioc_data);
2082         if (ioc_data) {
2083                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2084                               sizeof *ioc_data, ioc_data, NULL);
2085
2086                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2087                               sizeof *ioc_data, ioc_data, NULL);
2088
2089                 OBD_FREE_PTR(ioc_data);
2090         }
2091
2092         /* Really, we'd like to wait until there are no requests outstanding,
2093          * and then continue.  For now, we just invalidate the requests,
2094          * schedule() and sleep one second if needed, and hope.
2095          */
2096         cfs_schedule();
2097         EXIT;
2098 }
2099
2100 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2101 {
2102         struct ll_sb_info *sbi = ll_s2sbi(sb);
2103         char *profilenm = get_profile_name(sb);
2104         int err;
2105         __u32 read_only;
2106
2107         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2108                 read_only = *flags & MS_RDONLY;
2109                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2110                                          sizeof(KEY_READ_ONLY),
2111                                          KEY_READ_ONLY, sizeof(read_only),
2112                                          &read_only, NULL);
2113                 if (err) {
2114                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2115                                       profilenm, read_only ?
2116                                       "read-only" : "read-write", err);
2117                         return err;
2118                 }
2119
2120                 if (read_only)
2121                         sb->s_flags |= MS_RDONLY;
2122                 else
2123                         sb->s_flags &= ~MS_RDONLY;
2124
2125                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2126                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2127                                       read_only ?  "read-only" : "read-write");
2128         }
2129         return 0;
2130 }
2131
2132 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2133                   struct super_block *sb, struct lookup_intent *it)
2134 {
2135         struct ll_sb_info *sbi = NULL;
2136         struct lustre_md md;
2137         int rc;
2138         ENTRY;
2139
2140         LASSERT(*inode || sb);
2141         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2142         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2143                               sbi->ll_md_exp, &md);
2144         if (rc)
2145                 RETURN(rc);
2146
2147         if (*inode) {
2148                 ll_update_inode(*inode, &md);
2149         } else {
2150                 LASSERT(sb != NULL);
2151
2152                 /*
2153                  * At this point server returns to client's same fid as client
2154                  * generated for creating. So using ->fid1 is okay here.
2155                  */
2156                 LASSERT(fid_is_sane(&md.body->fid1));
2157
2158                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2159                                              sbi->ll_flags & LL_SBI_32BIT_API),
2160                                  &md);
2161                 if (*inode == NULL || IS_ERR(*inode)) {
2162 #ifdef CONFIG_FS_POSIX_ACL
2163                         if (md.posix_acl) {
2164                                 posix_acl_release(md.posix_acl);
2165                                 md.posix_acl = NULL;
2166                         }
2167 #endif
2168                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2169                         *inode = NULL;
2170                         CERROR("new_inode -fatal: rc %d\n", rc);
2171                         GOTO(out, rc);
2172                 }
2173         }
2174
2175         /* Handling piggyback layout lock.
2176          * Layout lock can be piggybacked by getattr and open request.
2177          * The lsm can be applied to inode only if it comes with a layout lock
2178          * otherwise correct layout may be overwritten, for example:
2179          * 1. proc1: mdt returns a lsm but not granting layout
2180          * 2. layout was changed by another client
2181          * 3. proc2: refresh layout and layout lock granted
2182          * 4. proc1: to apply a stale layout */
2183         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2184                 struct lustre_handle lockh;
2185                 struct ldlm_lock *lock;
2186
2187                 lockh.cookie = it->d.lustre.it_lock_handle;
2188                 lock = ldlm_handle2lock(&lockh);
2189                 LASSERT(lock != NULL);
2190                 if (ldlm_has_layout(lock)) {
2191                         struct cl_object_conf conf;
2192
2193                         memset(&conf, 0, sizeof(conf));
2194                         conf.coc_opc = OBJECT_CONF_SET;
2195                         conf.coc_inode = *inode;
2196                         conf.coc_lock = lock;
2197                         conf.u.coc_md = &md;
2198                         (void)ll_layout_conf(*inode, &conf);
2199                 }
2200                 LDLM_LOCK_PUT(lock);
2201         }
2202
2203 out:
2204         if (md.lsm != NULL)
2205                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2206         md_free_lustre_md(sbi->ll_md_exp, &md);
2207         RETURN(rc);
2208 }
2209
2210 int ll_obd_statfs(struct inode *inode, void *arg)
2211 {
2212         struct ll_sb_info *sbi = NULL;
2213         struct obd_export *exp;
2214         char *buf = NULL;
2215         struct obd_ioctl_data *data = NULL;
2216         __u32 type;
2217         __u32 flags;
2218         int len = 0, rc;
2219
2220         if (!inode || !(sbi = ll_i2sbi(inode)))
2221                 GOTO(out_statfs, rc = -EINVAL);
2222
2223         rc = obd_ioctl_getdata(&buf, &len, arg);
2224         if (rc)
2225                 GOTO(out_statfs, rc);
2226
2227         data = (void*)buf;
2228         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2229             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2230                 GOTO(out_statfs, rc = -EINVAL);
2231
2232         if (data->ioc_inllen1 != sizeof(__u32) ||
2233             data->ioc_inllen2 != sizeof(__u32) ||
2234             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2235             data->ioc_plen2 != sizeof(struct obd_uuid))
2236                 GOTO(out_statfs, rc = -EINVAL);
2237
2238         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2239         if (type & LL_STATFS_LMV)
2240                 exp = sbi->ll_md_exp;
2241         else if (type & LL_STATFS_LOV)
2242                 exp = sbi->ll_dt_exp;
2243         else
2244                 GOTO(out_statfs, rc = -ENODEV);
2245
2246         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2247         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2248         if (rc)
2249                 GOTO(out_statfs, rc);
2250 out_statfs:
2251         if (buf)
2252                 obd_ioctl_freedata(buf, len);
2253         return rc;
2254 }
2255
2256 int ll_process_config(struct lustre_cfg *lcfg)
2257 {
2258         char *ptr;
2259         void *sb;
2260         struct lprocfs_static_vars lvars;
2261         unsigned long x;
2262         int rc = 0;
2263
2264         lprocfs_llite_init_vars(&lvars);
2265
2266         /* The instance name contains the sb: lustre-client-aacfe000 */
2267         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2268         if (!ptr || !*(++ptr))
2269                 return -EINVAL;
2270         if (sscanf(ptr, "%lx", &x) != 1)
2271                 return -EINVAL;
2272         sb = (void *)x;
2273         /* This better be a real Lustre superblock! */
2274         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2275
2276         /* Note we have not called client_common_fill_super yet, so
2277            proc fns must be able to handle that! */
2278         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2279                                       lcfg, sb);
2280         if (rc > 0)
2281                 rc = 0;
2282         return(rc);
2283 }
2284
2285 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2286 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2287                                        struct inode *i1, struct inode *i2,
2288                                        const char *name, int namelen,
2289                                        int mode, __u32 opc, void *data)
2290 {
2291         LASSERT(i1 != NULL);
2292
2293         if (namelen > ll_i2sbi(i1)->ll_namelen)
2294                 return ERR_PTR(-ENAMETOOLONG);
2295
2296         if (op_data == NULL)
2297                 OBD_ALLOC_PTR(op_data);
2298
2299         if (op_data == NULL)
2300                 return ERR_PTR(-ENOMEM);
2301
2302         ll_i2gids(op_data->op_suppgids, i1, i2);
2303         op_data->op_fid1 = *ll_inode2fid(i1);
2304         op_data->op_capa1 = ll_mdscapa_get(i1);
2305
2306         if (i2) {
2307                 op_data->op_fid2 = *ll_inode2fid(i2);
2308                 op_data->op_capa2 = ll_mdscapa_get(i2);
2309         } else {
2310                 fid_zero(&op_data->op_fid2);
2311                 op_data->op_capa2 = NULL;
2312         }
2313
2314         op_data->op_name = name;
2315         op_data->op_namelen = namelen;
2316         op_data->op_mode = mode;
2317         op_data->op_mod_time = cfs_time_current_sec();
2318         op_data->op_fsuid = cfs_curproc_fsuid();
2319         op_data->op_fsgid = cfs_curproc_fsgid();
2320         op_data->op_cap = cfs_curproc_cap_pack();
2321         op_data->op_bias = 0;
2322         op_data->op_cli_flags = 0;
2323         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2324              filename_is_volatile(name, namelen, NULL))
2325                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2326         op_data->op_opc = opc;
2327         op_data->op_mds = 0;
2328         op_data->op_data = data;
2329
2330         /* If the file is being opened after mknod() (normally due to NFS)
2331          * try to use the default stripe data from parent directory for
2332          * allocating OST objects.  Try to pass the parent FID to MDS. */
2333         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2334             !ll_i2info(i2)->lli_has_smd) {
2335                 struct ll_inode_info *lli = ll_i2info(i2);
2336
2337                 spin_lock(&lli->lli_lock);
2338                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2339                         op_data->op_fid1 = lli->lli_pfid;
2340                 spin_unlock(&lli->lli_lock);
2341                 /** We ignore parent's capability temporary. */
2342         }
2343
2344         /* When called by ll_setattr_raw, file is i1. */
2345         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2346                 op_data->op_bias |= MDS_DATA_MODIFIED;
2347
2348         return op_data;
2349 }
2350
2351 void ll_finish_md_op_data(struct md_op_data *op_data)
2352 {
2353         capa_put(op_data->op_capa1);
2354         capa_put(op_data->op_capa2);
2355         OBD_FREE_PTR(op_data);
2356 }
2357
2358 #ifdef HAVE_SUPEROPS_USE_DENTRY
2359 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2360 #else
2361 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2362 #endif
2363 {
2364         struct ll_sb_info *sbi;
2365
2366 #ifdef HAVE_SUPEROPS_USE_DENTRY
2367         LASSERT((seq != NULL) && (dentry != NULL));
2368         sbi = ll_s2sbi(dentry->d_sb);
2369 #else
2370         LASSERT((seq != NULL) && (vfs != NULL));
2371         sbi = ll_s2sbi(vfs->mnt_sb);
2372 #endif
2373
2374         if (sbi->ll_flags & LL_SBI_NOLCK)
2375                 seq_puts(seq, ",nolock");
2376
2377         if (sbi->ll_flags & LL_SBI_FLOCK)
2378                 seq_puts(seq, ",flock");
2379
2380         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2381                 seq_puts(seq, ",localflock");
2382
2383         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2384                 seq_puts(seq, ",user_xattr");
2385
2386         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2387                 seq_puts(seq, ",lazystatfs");
2388
2389         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2390                 seq_puts(seq, ",user_fid2path");
2391
2392         RETURN(0);
2393 }
2394
2395 /**
2396  * Get obd name by cmd, and copy out to user space
2397  */
2398 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2399 {
2400         struct ll_sb_info *sbi = ll_i2sbi(inode);
2401         struct obd_device *obd;
2402         ENTRY;
2403
2404         if (cmd == OBD_IOC_GETDTNAME)
2405                 obd = class_exp2obd(sbi->ll_dt_exp);
2406         else if (cmd == OBD_IOC_GETMDNAME)
2407                 obd = class_exp2obd(sbi->ll_md_exp);
2408         else
2409                 RETURN(-EINVAL);
2410
2411         if (!obd)
2412                 RETURN(-ENOENT);
2413
2414         if (copy_to_user((void *)arg, obd->obd_name,
2415                          strlen(obd->obd_name) + 1))
2416                 RETURN(-EFAULT);
2417
2418         RETURN(0);
2419 }
2420
2421 /**
2422  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2423  * fsname will be returned in this buffer; otherwise, a static buffer will be
2424  * used to store the fsname and returned to caller.
2425  */
2426 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2427 {
2428         static char fsname_static[MTI_NAME_MAXLEN];
2429         struct lustre_sb_info *lsi = s2lsi(sb);
2430         char *ptr;
2431         int len;
2432
2433         if (buf == NULL) {
2434                 /* this means the caller wants to use static buffer
2435                  * and it doesn't care about race. Usually this is
2436                  * in error reporting path */
2437                 buf = fsname_static;
2438                 buflen = sizeof(fsname_static);
2439         }
2440
2441         len = strlen(lsi->lsi_lmd->lmd_profile);
2442         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2443         if (ptr && (strcmp(ptr, "-client") == 0))
2444                 len -= 7;
2445
2446         if (unlikely(len >= buflen))
2447                 len = buflen - 1;
2448         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2449         buf[len] = '\0';
2450
2451         return buf;
2452 }
2453
2454 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2455 {
2456         char *path = NULL;
2457
2458 #ifdef HAVE_FS_STRUCT_USE_PATH
2459         struct path p;
2460
2461         p.dentry = dentry;
2462         p.mnt = current->fs->root.mnt;
2463         path_get(&p);
2464         path = d_path(&p, buf, bufsize);
2465         path_put(&p);
2466 #else
2467         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2468 #endif
2469
2470         return path;
2471 }
2472
2473 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2474 {
2475         char *buf, *path = NULL;
2476         struct dentry *dentry = NULL;
2477         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2478
2479         /* this can be called inside spin lock so use GFP_ATOMIC. */
2480         buf = (char *)__get_free_page(GFP_ATOMIC);
2481         if (buf != NULL) {
2482                 dentry = d_find_alias(page->mapping->host);
2483                 if (dentry != NULL)
2484                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2485         }
2486
2487         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2488               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2489               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2490               PFID(&obj->cob_header.coh_lu.loh_fid),
2491               (path && !IS_ERR(path)) ? path : "", ioret);
2492
2493         if (dentry != NULL)
2494                 dput(dentry);
2495
2496         if (buf != NULL)
2497                 free_page((unsigned long)buf);
2498 }