lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         lru_page_max = pages / 2;
  97
  98         /* initialize ll_cache data */
  99         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 100         sbi->ll_cache.ccc_lru_max = lru_page_max;
 101         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 102         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 103         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 104
 105         cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 106         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 107
 108         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 109                                            SBI_DEFAULT_READAHEAD_MAX);
 110         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 111         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 112                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 113         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 114         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 115
 116         ll_generate_random_uuid(uuid);
 117         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 118         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 119
 120         spin_lock(&ll_sb_lock);
 121         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 122         spin_unlock(&ll_sb_lock);
 123
 124         sbi->ll_flags |= LL_SBI_VERBOSE;
 125 #ifdef ENABLE_CHECKSUM
 126         sbi->ll_flags |= LL_SBI_CHECKSUM;
 127 #endif
 128
 129 #ifdef HAVE_LRU_RESIZE_SUPPORT
 130         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 131 #endif
 132
 133         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 134                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 135                                pp_r_hist.oh_lock);
 136                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 137                                pp_w_hist.oh_lock);
 138         }
 139
 140         /* metadata statahead is enabled by default */
 141         sbi->ll_sa_max = LL_SA_RPC_DEF;
 142         cfs_atomic_set(&sbi->ll_sa_total, 0);
 143         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 144         cfs_atomic_set(&sbi->ll_agl_total, 0);
 145         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 146
 147         RETURN(sbi);
 148 }
 149
 150 void ll_free_sbi(struct super_block *sb)
 151 {
 152         struct ll_sb_info *sbi = ll_s2sbi(sb);
 153         ENTRY;
 154
 155         if (sbi != NULL) {
 156                 spin_lock(&ll_sb_lock);
 157                 cfs_list_del(&sbi->ll_list);
 158                 spin_unlock(&ll_sb_lock);
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_capa *oc = NULL;
 171         struct obd_statfs *osfs = NULL;
 172         struct ptlrpc_request *request = NULL;
 173         struct obd_connect_data *data = NULL;
 174         struct obd_uuid *uuid;
 175         struct md_op_data *op_data;
 176         struct lustre_md lmd;
 177         obd_valid valid;
 178         int size, err, checksum;
 179         ENTRY;
 180
 181         obd = class_name2obd(md);
 182         if (!obd) {
 183                 CERROR("MD %s: not setup or attached\n", md);
 184                 RETURN(-EINVAL);
 185         }
 186
 187         OBD_ALLOC_PTR(data);
 188         if (data == NULL)
 189                 RETURN(-ENOMEM);
 190
 191         OBD_ALLOC_PTR(osfs);
 192         if (osfs == NULL) {
 193                 OBD_FREE_PTR(data);
 194                 RETURN(-ENOMEM);
 195         }
 196
 197         /* indicate the features supported by this client */
 198         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 199                                   OBD_CONNECT_ATTRFID  |
 200                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 201                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 202                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 203                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 204                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 205                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 206                                   OBD_CONNECT_EINPROGRESS |
 207                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 208                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 209                                   OBD_CONNECT_MAX_EASIZE |
 210                                   OBD_CONNECT_FLOCK_DEAD |
 211                                   OBD_CONNECT_DISP_STRIPE;
 212
 213         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 214                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 215
 216 #ifdef HAVE_LRU_RESIZE_SUPPORT
 217         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 218                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 219 #endif
 220 #ifdef CONFIG_FS_POSIX_ACL
 221         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 222 #endif
 223
 224         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 225                 /* flag mdc connection as lightweight, only used for test
 226                  * purpose, use with care */
 227                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 228
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237 #ifdef HAVE_MS_FLOCK_LOCK
 238         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 239         sb->s_flags |= MS_FLOCK_LOCK;
 240 #endif
 241 #ifdef MS_HAS_NEW_AOPS
 242         sb->s_flags |= MS_HAS_NEW_AOPS;
 243 #endif
 244
 245         if (sbi->ll_flags & LL_SBI_FLOCK)
 246                 sbi->ll_fop = &ll_file_operations_flock;
 247         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 248                 sbi->ll_fop = &ll_file_operations;
 249         else
 250                 sbi->ll_fop = &ll_file_operations_noflock;
 251
 252         /* real client */
 253         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 254         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 255                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 256
 257         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 258
 259         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 260         if (err == -EBUSY) {
 261                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 262                                    "recovery, of which this client is not a "
 263                                    "part. Please wait for recovery to complete,"
 264                                    " abort, or time out.\n", md);
 265                 GOTO(out, err);
 266         } else if (err) {
 267                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 268                 GOTO(out, err);
 269         }
 270
 271         sbi->ll_md_exp->exp_connect_data = *data;
 272
 273         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 274                            LUSTRE_SEQ_METADATA);
 275         if (err) {
 276                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 277                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 278                 GOTO(out_md, err);
 279         }
 280
 281         /* For mount, we only need fs info from MDT0, and also in DNE, it
 282          * can make sure the client can be mounted as long as MDT0 is
 283          * avaible */
 284         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 285                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 286                         OBD_STATFS_FOR_MDT0);
 287         if (err)
 288                 GOTO(out_md_fid, err);
 289
 290         /* This needs to be after statfs to ensure connect has finished.
 291          * Note that "data" does NOT contain the valid connect reply.
 292          * If connecting to a 1.8 server there will be no LMV device, so
 293          * we can access the MDC export directly and exp_connect_flags will
 294          * be non-zero, but if accessing an upgraded 2.1 server it will
 295          * have the correct flags filled in.
 296          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 297         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 298         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 299             valid != CLIENT_CONNECT_MDT_REQD) {
 300                 char *buf;
 301
 302                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 303                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 304                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 305                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 306                                    "feature(s) needed for correct operation "
 307                                    "of this client (%s). Please upgrade "
 308                                    "server or downgrade client.\n",
 309                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 310                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 311                 GOTO(out_md_fid, err = -EPROTO);
 312         }
 313
 314         size = sizeof(*data);
 315         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 316                            KEY_CONN_DATA,  &size, data, NULL);
 317         if (err) {
 318                 CERROR("%s: Get connect data failed: rc = %d\n",
 319                        sbi->ll_md_exp->exp_obd->obd_name, err);
 320                 GOTO(out_md_fid, err);
 321         }
 322
 323         LASSERT(osfs->os_bsize);
 324         sb->s_blocksize = osfs->os_bsize;
 325         sb->s_blocksize_bits = log2(osfs->os_bsize);
 326         sb->s_magic = LL_SUPER_MAGIC;
 327         sb->s_maxbytes = MAX_LFS_FILESIZE;
 328         sbi->ll_namelen = osfs->os_namelen;
 329         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 330
 331         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 332             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 333                 LCONSOLE_INFO("Disabling user_xattr feature because "
 334                               "it is not supported on the server\n");
 335                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags |= MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags |= LL_SBI_ACL;
 343         } else {
 344                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags &= ~MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags &= ~LL_SBI_ACL;
 349         }
 350
 351         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 352                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 353                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client is set as remote by default.\n");
 355                 }
 356         } else {
 357                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 358                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client claims to be remote, but server "
 360                                       "rejected, forced to be local.\n");
 361                 }
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 365                 LCONSOLE_INFO("client enabled MDS capability!\n");
 366                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 370                 LCONSOLE_INFO("client enabled OSS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 375                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 378                 sbi->ll_md_brw_size = data->ocd_brw_size;
 379         else
 380                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 383                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 384
 385         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 386                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 387                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 388                                       "unknown maximum xattr size.\n", dt);
 389                 } else {
 390                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 391                         sbi->ll_xattr_cache_enabled = 1;
 392                 }
 393         }
 394
 395         obd = class_name2obd(dt);
 396         if (!obd) {
 397                 CERROR("DT %s: not setup or attached\n", dt);
 398                 GOTO(out_md_fid, err = -ENODEV);
 399         }
 400
 401         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 402                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 403                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 404                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 405                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 406                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 407                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 408                                   OBD_CONNECT_MAXBYTES |
 409                                   OBD_CONNECT_EINPROGRESS |
 410                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 411                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 412
 413         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 414                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 415
 416         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 417                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 418                  * disabled by default, because it can still be enabled on the
 419                  * fly via /proc. As a consequence, we still need to come to an
 420                  * agreement on the supported algorithms at connect time */
 421                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 422
 423                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 424                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 425                 else
 426                         data->ocd_cksum_types = cksum_types_supported_client();
 427         }
 428
 429 #ifdef HAVE_LRU_RESIZE_SUPPORT
 430         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 431 #endif
 432         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 433                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 434
 435         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 436                "ocd_grant: %d\n", data->ocd_connect_flags,
 437                data->ocd_version, data->ocd_grant);
 438
 439         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 440         obd->obd_upcall.onu_upcall = cl_ocd_update;
 441
 442         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 443
 444         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 445                           NULL);
 446         if (err == -EBUSY) {
 447                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 448                                    "recovery, of which this client is not a "
 449                                    "part.  Please wait for recovery to "
 450                                    "complete, abort, or time out.\n", dt);
 451                 GOTO(out_md, err);
 452         } else if (err) {
 453                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 454                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 455                 GOTO(out_md, err);
 456         }
 457
 458         sbi->ll_dt_exp->exp_connect_data = *data;
 459
 460         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 461                            LUSTRE_SEQ_METADATA);
 462         if (err) {
 463                 CERROR("%s: Can't init data layer FID infrastructure, "
 464                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 465                 GOTO(out_dt, err);
 466         }
 467
 468         mutex_lock(&sbi->ll_lco.lco_lock);
 469         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 470         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 471         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 472         mutex_unlock(&sbi->ll_lco.lco_lock);
 473
 474         fid_zero(&sbi->ll_root_fid);
 475         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 476         if (err) {
 477                 CERROR("cannot mds_connect: rc = %d\n", err);
 478                 GOTO(out_lock_cn_cb, err);
 479         }
 480         if (!fid_is_sane(&sbi->ll_root_fid)) {
 481                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 482                        sbi->ll_md_exp->exp_obd->obd_name,
 483                        PFID(&sbi->ll_root_fid));
 484                 GOTO(out_lock_cn_cb, err = -EINVAL);
 485         }
 486         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 487
 488         sb->s_op = &lustre_super_operations;
 489 #if THREAD_SIZE >= 8192 /*b=17630*/
 490         sb->s_export_op = &lustre_export_operations;
 491 #endif
 492
 493         /* make root inode
 494          * XXX: move this to after cbd setup? */
 495         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 496                 OBD_MD_FLMODEASIZE;
 497         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 498                 valid |= OBD_MD_FLRMTPERM;
 499         else if (sbi->ll_flags & LL_SBI_ACL)
 500                 valid |= OBD_MD_FLACL;
 501
 502         OBD_ALLOC_PTR(op_data);
 503         if (op_data == NULL)
 504                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 505
 506         op_data->op_fid1 = sbi->ll_root_fid;
 507         op_data->op_mode = 0;
 508         op_data->op_capa1 = oc;
 509         op_data->op_valid = valid;
 510
 511         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 512         if (oc)
 513                 capa_put(oc);
 514         OBD_FREE_PTR(op_data);
 515         if (err) {
 516                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 517                        sbi->ll_md_exp->exp_obd->obd_name, err);
 518                 GOTO(out_lock_cn_cb, err);
 519         }
 520
 521         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 522                                sbi->ll_md_exp, &lmd);
 523         if (err) {
 524                 CERROR("failed to understand root inode md: rc = %d\n", err);
 525                 ptlrpc_req_finished(request);
 526                 GOTO(out_lock_cn_cb, err);
 527         }
 528
 529         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 530         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 531                                             sbi->ll_flags & LL_SBI_32BIT_API),
 532                        &lmd);
 533         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 534         ptlrpc_req_finished(request);
 535
 536         if (root == NULL || IS_ERR(root)) {
 537                 if (lmd.lsm)
 538                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 539 #ifdef CONFIG_FS_POSIX_ACL
 540                 if (lmd.posix_acl) {
 541                         posix_acl_release(lmd.posix_acl);
 542                         lmd.posix_acl = NULL;
 543                 }
 544 #endif
 545                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 546                 root = NULL;
 547                 CERROR("lustre_lite: bad iget4 for root\n");
 548                 GOTO(out_root, err);
 549         }
 550
 551         err = ll_close_thread_start(&sbi->ll_lcq);
 552         if (err) {
 553                 CERROR("cannot start close thread: rc %d\n", err);
 554                 GOTO(out_root, err);
 555         }
 556
 557 #ifdef CONFIG_FS_POSIX_ACL
 558         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 559                 rct_init(&sbi->ll_rct);
 560                 et_init(&sbi->ll_et);
 561         }
 562 #endif
 563
 564         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 565         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 566                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 567                                  NULL);
 568         cl_sb_init(sb);
 569
 570         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 571                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 572                                  &sbi->ll_cache, NULL);
 573
 574         sb->s_root = d_make_root(root);
 575         if (sb->s_root == NULL) {
 576                 CERROR("%s: can't make root dentry\n",
 577                         ll_get_fsname(sb, NULL, 0));
 578                 GOTO(out_root, err = -ENOMEM);
 579         }
 580 #ifdef HAVE_DCACHE_LOCK
 581         sb->s_root->d_op = &ll_d_ops;
 582 #endif
 583
 584         sbi->ll_sdev_orig = sb->s_dev;
 585
 586         /* We set sb->s_dev equal on all lustre clients in order to support
 587          * NFS export clustering.  NFSD requires that the FSID be the same
 588          * on all clients. */
 589         /* s_dev is also used in lt_compare() to compare two fs, but that is
 590          * only a node-local comparison. */
 591         uuid = obd_get_uuid(sbi->ll_md_exp);
 592         if (uuid != NULL)
 593                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 594
 595         if (data != NULL)
 596                 OBD_FREE_PTR(data);
 597         if (osfs != NULL)
 598                 OBD_FREE_PTR(osfs);
 599         if (proc_lustre_fs_root != NULL) {
 600                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 601                                                   dt, md);
 602                 if (err < 0) {
 603                         CERROR("%s: could not register mount in lprocfs: "
 604                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 605                         err = 0;
 606                 }
 607         }
 608
 609         RETURN(err);
 610 out_root:
 611         if (root)
 612                 iput(root);
 613 out_lock_cn_cb:
 614         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 615 out_dt:
 616         obd_disconnect(sbi->ll_dt_exp);
 617         sbi->ll_dt_exp = NULL;
 618         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 619         obd_zombie_barrier();
 620 out_md_fid:
 621         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 622 out_md:
 623         obd_disconnect(sbi->ll_md_exp);
 624         sbi->ll_md_exp = NULL;
 625 out:
 626         if (data != NULL)
 627                 OBD_FREE_PTR(data);
 628         if (osfs != NULL)
 629                 OBD_FREE_PTR(osfs);
 630         return err;
 631 }
 632
 633 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 634 {
 635         int size, rc;
 636
 637         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 638         size = sizeof(int);
 639         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 640                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 641         if (rc)
 642                 CERROR("Get max mdsize error rc %d \n", rc);
 643
 644         RETURN(rc);
 645 }
 646
 647 void ll_dump_inode(struct inode *inode)
 648 {
 649         struct ll_d_hlist_node *tmp;
 650         int dentry_count = 0;
 651
 652         LASSERT(inode != NULL);
 653
 654         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 655                 dentry_count++;
 656
 657         CERROR("%s: inode %p dump: dev=%s fid="DFID
 658                " mode=%o count=%u, %d dentries\n",
 659                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 660                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 661                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 662 }
 663
 664 void lustre_dump_dentry(struct dentry *dentry, int recur)
 665 {
 666         struct list_head *tmp;
 667         int subdirs = 0;
 668
 669         LASSERT(dentry != NULL);
 670
 671         list_for_each(tmp, &dentry->d_subdirs)
 672                 subdirs++;
 673
 674         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 675                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 676                dentry->d_name.len, dentry->d_name.name,
 677                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 678                dentry->d_parent, dentry->d_inode, d_count(dentry),
 679                dentry->d_flags, dentry->d_fsdata, subdirs);
 680         if (dentry->d_inode != NULL)
 681                 ll_dump_inode(dentry->d_inode);
 682
 683         if (recur == 0)
 684                 return;
 685
 686         list_for_each(tmp, &dentry->d_subdirs) {
 687                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 688                 lustre_dump_dentry(d, recur - 1);
 689         }
 690 }
 691
 692 void client_common_put_super(struct super_block *sb)
 693 {
 694         struct ll_sb_info *sbi = ll_s2sbi(sb);
 695         ENTRY;
 696
 697 #ifdef CONFIG_FS_POSIX_ACL
 698         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 699                 et_fini(&sbi->ll_et);
 700                 rct_fini(&sbi->ll_rct);
 701         }
 702 #endif
 703
 704         ll_close_thread_shutdown(sbi->ll_lcq);
 705
 706         cl_sb_fini(sb);
 707
 708         cfs_list_del(&sbi->ll_conn_chain);
 709
 710         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 711         obd_disconnect(sbi->ll_dt_exp);
 712         sbi->ll_dt_exp = NULL;
 713         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 714          * see LU-2543. */
 715         obd_zombie_barrier();
 716
 717         lprocfs_unregister_mountpoint(sbi);
 718
 719         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 720         obd_disconnect(sbi->ll_md_exp);
 721         sbi->ll_md_exp = NULL;
 722
 723         EXIT;
 724 }
 725
 726 void ll_kill_super(struct super_block *sb)
 727 {
 728         struct ll_sb_info *sbi;
 729
 730         ENTRY;
 731
 732         /* not init sb ?*/
 733         if (!(sb->s_flags & MS_ACTIVE))
 734                 return;
 735
 736         sbi = ll_s2sbi(sb);
 737         /* we need restore s_dev from changed for clustred NFS before put_super
 738          * because new kernels have cached s_dev and change sb->s_dev in
 739          * put_super not affected real removing devices */
 740         if (sbi) {
 741                 sb->s_dev = sbi->ll_sdev_orig;
 742                 sbi->ll_umounting = 1;
 743         }
 744         EXIT;
 745 }
 746
 747 char *ll_read_opt(const char *opt, char *data)
 748 {
 749         char *value;
 750         char *retval;
 751         ENTRY;
 752
 753         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 754         if (strncmp(opt, data, strlen(opt)))
 755                 RETURN(NULL);
 756         if ((value = strchr(data, '=')) == NULL)
 757                 RETURN(NULL);
 758
 759         value++;
 760         OBD_ALLOC(retval, strlen(value) + 1);
 761         if (!retval) {
 762                 CERROR("out of memory!\n");
 763                 RETURN(NULL);
 764         }
 765
 766         memcpy(retval, value, strlen(value)+1);
 767         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 768         RETURN(retval);
 769 }
 770
 771 static inline int ll_set_opt(const char *opt, char *data, int fl)
 772 {
 773         if (strncmp(opt, data, strlen(opt)) != 0)
 774                 return(0);
 775         else
 776                 return(fl);
 777 }
 778
 779 /* non-client-specific mount options are parsed in lmd_parse */
 780 static int ll_options(char *options, int *flags)
 781 {
 782         int tmp;
 783         char *s1 = options, *s2;
 784         ENTRY;
 785
 786         if (!options)
 787                 RETURN(0);
 788
 789         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 790
 791         while (*s1) {
 792                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 793                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 794                 if (tmp) {
 795                         *flags |= tmp;
 796                         goto next;
 797                 }
 798                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 799                 if (tmp) {
 800                         *flags |= tmp;
 801                         goto next;
 802                 }
 803                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 804                 if (tmp) {
 805                         *flags |= tmp;
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 809                 if (tmp) {
 810                         *flags &= ~tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 819                 if (tmp) {
 820                         *flags &= ~tmp;
 821                         goto next;
 822                 }
 823 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 824                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 825                 if (tmp) {
 826                         /* Ignore deprecated mount option.  The client will
 827                          * always try to mount with ACL support, whether this
 828                          * is used depends on whether server supports it. */
 829                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 830                                                   "mount option 'acl'.\n");
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 834                 if (tmp) {
 835                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 836                                                   "mount option 'noacl'.\n");
 837                         goto next;
 838                 }
 839 #else
 840 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 841 #endif
 842                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857
 858                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 859                 if (tmp) {
 860                         *flags |= tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 864                 if (tmp) {
 865                         *flags &= ~tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 894                 if (tmp) {
 895                         *flags |= tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 904                 if (tmp) {
 905                         *flags &= ~tmp;
 906                         goto next;
 907                 }
 908                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 909                                    s1);
 910                 RETURN(-EINVAL);
 911
 912 next:
 913                 /* Find next opt */
 914                 s2 = strchr(s1, ',');
 915                 if (s2 == NULL)
 916                         break;
 917                 s1 = s2 + 1;
 918         }
 919         RETURN(0);
 920 }
 921
 922 void ll_lli_init(struct ll_inode_info *lli)
 923 {
 924         lli->lli_inode_magic = LLI_INODE_MAGIC;
 925         lli->lli_flags = 0;
 926         lli->lli_ioepoch = 0;
 927         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 928         spin_lock_init(&lli->lli_lock);
 929         lli->lli_posix_acl = NULL;
 930         lli->lli_remote_perms = NULL;
 931         mutex_init(&lli->lli_rmtperm_mutex);
 932         /* Do not set lli_fid, it has been initialized already. */
 933         fid_zero(&lli->lli_pfid);
 934         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 935         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 936         cfs_atomic_set(&lli->lli_open_count, 0);
 937         lli->lli_mds_capa = NULL;
 938         lli->lli_rmtperm_time = 0;
 939         lli->lli_pending_och = NULL;
 940         lli->lli_mds_read_och = NULL;
 941         lli->lli_mds_write_och = NULL;
 942         lli->lli_mds_exec_och = NULL;
 943         lli->lli_open_fd_read_count = 0;
 944         lli->lli_open_fd_write_count = 0;
 945         lli->lli_open_fd_exec_count = 0;
 946         mutex_init(&lli->lli_och_mutex);
 947         spin_lock_init(&lli->lli_agl_lock);
 948         lli->lli_has_smd = false;
 949         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 950         lli->lli_clob = NULL;
 951
 952         init_rwsem(&lli->lli_xattrs_list_rwsem);
 953         mutex_init(&lli->lli_xattrs_enq_lock);
 954
 955         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 956         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 957                 mutex_init(&lli->lli_readdir_mutex);
 958                 lli->lli_opendir_key = NULL;
 959                 lli->lli_sai = NULL;
 960                 lli->lli_def_acl = NULL;
 961                 spin_lock_init(&lli->lli_sa_lock);
 962                 lli->lli_opendir_pid = 0;
 963         } else {
 964                 sema_init(&lli->lli_size_sem, 1);
 965                 lli->lli_size_sem_owner = NULL;
 966                 lli->lli_symlink_name = NULL;
 967                 init_rwsem(&lli->lli_trunc_sem);
 968                 mutex_init(&lli->lli_write_mutex);
 969                 init_rwsem(&lli->lli_glimpse_sem);
 970                 lli->lli_glimpse_time = 0;
 971                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 972                 lli->lli_agl_index = 0;
 973                 lli->lli_async_rc = 0;
 974                 lli->lli_volatile = false;
 975         }
 976         mutex_init(&lli->lli_layout_mutex);
 977 }
 978
 979 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 980 {
 981         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 982
 983         bdi->name = "lustre";
 984         return bdi_register(bdi, NULL, "lustre-%d",
 985                             atomic_inc_return(&ll_bdi_num));
 986 }
 987
 988 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 989 {
 990         struct lustre_profile *lprof = NULL;
 991         struct lustre_sb_info *lsi = s2lsi(sb);
 992         struct ll_sb_info *sbi;
 993         char  *dt = NULL, *md = NULL;
 994         char  *profilenm = get_profile_name(sb);
 995         struct config_llog_instance *cfg;
 996         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 997         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 998         int    err;
 999         ENTRY;
1000
1001         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1002
1003         OBD_ALLOC_PTR(cfg);
1004         if (cfg == NULL)
1005                 RETURN(-ENOMEM);
1006
1007         try_module_get(THIS_MODULE);
1008
1009         /* client additional sb info */
1010         lsi->lsi_llsbi = sbi = ll_init_sbi();
1011         if (!sbi) {
1012                 module_put(THIS_MODULE);
1013                 OBD_FREE_PTR(cfg);
1014                 RETURN(-ENOMEM);
1015         }
1016
1017         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1018         if (err)
1019                 GOTO(out_free, err);
1020
1021         err = bdi_init(&lsi->lsi_bdi);
1022         if (err)
1023                 GOTO(out_free, err);
1024         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1025         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1026         err = ll_bdi_register(&lsi->lsi_bdi);
1027         if (err)
1028                 GOTO(out_free, err);
1029
1030         sb->s_bdi = &lsi->lsi_bdi;
1031 #ifndef HAVE_DCACHE_LOCK
1032         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1033         sb->s_d_op = &ll_d_ops;
1034 #endif
1035
1036         /* Generate a string unique to this super, in case some joker tries
1037            to mount the same fs at two mount points.
1038            Use the address of the super itself.*/
1039         cfg->cfg_instance = sb;
1040         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1041         cfg->cfg_callback = class_config_llog_handler;
1042         /* set up client obds */
1043         err = lustre_process_log(sb, profilenm, cfg);
1044         if (err < 0)
1045                 GOTO(out_free, err);
1046
1047         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1048         lprof = class_get_profile(profilenm);
1049         if (lprof == NULL) {
1050                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1051                                    " read from the MGS.  Does that filesystem "
1052                                    "exist?\n", profilenm);
1053                 GOTO(out_free, err = -EINVAL);
1054         }
1055         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1056                lprof->lp_md, lprof->lp_dt);
1057
1058         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1059         if (!dt)
1060                 GOTO(out_free, err = -ENOMEM);
1061         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1062
1063         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1064         if (!md)
1065                 GOTO(out_free, err = -ENOMEM);
1066         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1067
1068         /* connections, registrations, sb setup */
1069         err = client_common_fill_super(sb, md, dt, mnt);
1070
1071 out_free:
1072         if (md)
1073                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1074         if (dt)
1075                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1076         if (err)
1077                 ll_put_super(sb);
1078         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1079                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1080
1081         OBD_FREE_PTR(cfg);
1082         RETURN(err);
1083 } /* ll_fill_super */
1084
1085
1086 void lu_context_keys_dump(void);
1087
1088 void ll_put_super(struct super_block *sb)
1089 {
1090         struct config_llog_instance cfg, params_cfg;
1091         struct obd_device *obd;
1092         struct lustre_sb_info *lsi = s2lsi(sb);
1093         struct ll_sb_info *sbi = ll_s2sbi(sb);
1094         char *profilenm = get_profile_name(sb);
1095         int ccc_count, next, force = 1, rc = 0;
1096         ENTRY;
1097
1098         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1099
1100         ll_print_capa_stat(sbi);
1101
1102         cfg.cfg_instance = sb;
1103         lustre_end_log(sb, profilenm, &cfg);
1104
1105         params_cfg.cfg_instance = sb;
1106         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1107
1108         if (sbi->ll_md_exp) {
1109                 obd = class_exp2obd(sbi->ll_md_exp);
1110                 if (obd)
1111                         force = obd->obd_force;
1112         }
1113
1114         /* Wait for unstable pages to be committed to stable storage */
1115         if (force == 0) {
1116                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1117                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1118                         cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1119                         &lwi);
1120         }
1121
1122         ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1123         if (force == 0 && rc != -EINTR)
1124                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1125
1126
1127         /* We need to set force before the lov_disconnect in
1128            lustre_common_put_super, since l_d cleans up osc's as well. */
1129         if (force) {
1130                 next = 0;
1131                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1132                                                      &next)) != NULL) {
1133                         obd->obd_force = force;
1134                 }
1135         }
1136
1137         if (sbi->ll_lcq) {
1138                 /* Only if client_common_fill_super succeeded */
1139                 client_common_put_super(sb);
1140         }
1141
1142         next = 0;
1143         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1144                 class_manual_cleanup(obd);
1145         }
1146
1147         if (sbi->ll_flags & LL_SBI_VERBOSE)
1148                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1149
1150         if (profilenm)
1151                 class_del_profile(profilenm);
1152
1153         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1154                 bdi_destroy(&lsi->lsi_bdi);
1155                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1156         }
1157
1158         ll_free_sbi(sb);
1159         lsi->lsi_llsbi = NULL;
1160
1161         lustre_common_put_super(sb);
1162
1163         cl_env_cache_purge(~0);
1164
1165         module_put(THIS_MODULE);
1166
1167         EXIT;
1168 } /* client_put_super */
1169
1170 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1171 {
1172         struct inode *inode = NULL;
1173
1174         /* NOTE: we depend on atomic igrab() -bzzz */
1175         lock_res_and_lock(lock);
1176         if (lock->l_resource->lr_lvb_inode) {
1177                 struct ll_inode_info * lli;
1178                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1179                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1180                         inode = igrab(lock->l_resource->lr_lvb_inode);
1181                 } else {
1182                         inode = lock->l_resource->lr_lvb_inode;
1183                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1184                                          D_WARNING, lock, "lr_lvb_inode %p is "
1185                                          "bogus: magic %08x",
1186                                          lock->l_resource->lr_lvb_inode,
1187                                          lli->lli_inode_magic);
1188                         inode = NULL;
1189                 }
1190         }
1191         unlock_res_and_lock(lock);
1192         return inode;
1193 }
1194
1195 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1196 {
1197         struct inode *inode = NULL;
1198         /* NOTE: we depend on atomic igrab() -bzzz */
1199         lock_res_and_lock(lock);
1200         if (lock->l_ast_data) {
1201                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1202                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1203                         inode = igrab(lock->l_ast_data);
1204                 } else {
1205                         inode = lock->l_ast_data;
1206                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1207                                          D_WARNING, lock, "l_ast_data %p is "
1208                                          "bogus: magic %08x", lock->l_ast_data,
1209                                          lli->lli_inode_magic);
1210                         inode = NULL;
1211                 }
1212         }
1213         unlock_res_and_lock(lock);
1214         return inode;
1215 }
1216
1217 static void ll_dir_clear_lsm_md(struct inode *inode)
1218 {
1219         struct ll_inode_info *lli = ll_i2info(inode);
1220
1221         LASSERT(S_ISDIR(inode->i_mode));
1222
1223         if (lli->lli_lsm_md != NULL) {
1224                 lmv_free_memmd(lli->lli_lsm_md);
1225                 lli->lli_lsm_md = NULL;
1226         }
1227 }
1228
1229 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1230                                       const struct lu_fid *fid,
1231                                       struct lustre_md *md)
1232 {
1233         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1234         struct mdt_body         *body = md->body;
1235         struct inode            *inode;
1236         ino_t                   ino;
1237         ENTRY;
1238
1239         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1240         inode = iget_locked(sb, ino);
1241         if (inode == NULL) {
1242                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1243                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1244                 RETURN(ERR_PTR(-ENOENT));
1245         }
1246
1247         if (inode->i_state & I_NEW) {
1248                 struct ll_inode_info *lli = ll_i2info(inode);
1249                 struct lmv_stripe_md *lsm = md->lmv;
1250
1251                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1252                                 (body->mode & S_IFMT);
1253                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1254                          PFID(fid));
1255
1256                 LTIME_S(inode->i_mtime) = 0;
1257                 LTIME_S(inode->i_atime) = 0;
1258                 LTIME_S(inode->i_ctime) = 0;
1259                 inode->i_rdev = 0;
1260
1261                 /* initializing backing dev info. */
1262                 inode->i_mapping->backing_dev_info =
1263                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1264                 inode->i_op = &ll_dir_inode_operations;
1265                 inode->i_fop = &ll_dir_operations;
1266                 lli->lli_fid = *fid;
1267                 ll_lli_init(lli);
1268
1269                 LASSERT(lsm != NULL);
1270                 /* master stripe FID */
1271                 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1272                 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1273                        lli, PFID(fid), PFID(&lli->lli_pfid));
1274                 unlock_new_inode(inode);
1275         }
1276
1277         RETURN(inode);
1278 }
1279
1280 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1281 {
1282         struct lu_fid *fid;
1283         struct lmv_stripe_md *lsm = md->lmv;
1284         int i;
1285
1286         LASSERT(lsm != NULL);
1287         /* XXX sigh, this lsm_root initialization should be in
1288          * LMV layer, but it needs ll_iget right now, so we
1289          * put this here right now. */
1290         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1291                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1292                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1293                 if (i == 0) {
1294                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1295                 } else {
1296                         /* Unfortunately ll_iget will call ll_update_inode,
1297                          * where the initialization of slave inode is slightly
1298                          * different, so it reset lsm_md to NULL to avoid
1299                          * initializing lsm for slave inode. */
1300                         lsm->lsm_md_oinfo[i].lmo_root =
1301                                         ll_iget_anon_dir(inode->i_sb, fid, md);
1302                         if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1303                                 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1304
1305                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1306                                 return rc;
1307                         }
1308                 }
1309         }
1310
1311         /* Here is where the lsm is being initialized(fill lmo_info) after
1312          * client retrieve MD stripe information from MDT. */
1313         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1314                                 ll_md_blocking_ast);
1315 }
1316
1317 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1318                                 const struct lmv_stripe_md *lsm_md2)
1319 {
1320         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1321                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1322                lsm_md1->lsm_md_master_mdt_index ==
1323                                         lsm_md2->lsm_md_master_mdt_index &&
1324                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1325                lsm_md1->lsm_md_layout_version ==
1326                                         lsm_md2->lsm_md_layout_version &&
1327                strcmp(lsm_md1->lsm_md_pool_name,
1328                       lsm_md2->lsm_md_pool_name) == 0;
1329 }
1330
1331 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1332 {
1333         struct ll_inode_info *lli = ll_i2info(inode);
1334         struct lmv_stripe_md *lsm = md->lmv;
1335         int idx;
1336
1337         LASSERT(lsm != NULL);
1338         LASSERT(S_ISDIR(inode->i_mode));
1339         if (lli->lli_lsm_md == NULL) {
1340                 int rc;
1341
1342                 rc = ll_init_lsm_md(inode, md);
1343                 if (rc != 0) {
1344                         CERROR("%s: init "DFID" failed: rc = %d\n",
1345                                ll_get_fsname(inode->i_sb, NULL, 0),
1346                                PFID(&lli->lli_fid), rc);
1347                         return;
1348                 }
1349                 lli->lli_lsm_md = lsm;
1350                 /* set lsm_md to NULL, so the following free lustre_md
1351                  * will not free this lsm */
1352                 md->lmv = NULL;
1353                 return;
1354         }
1355
1356         /* Compare the old and new stripe information */
1357         if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1358                 CERROR("inode %p %lu mismatch\n"
1359                        "    new(%p)     vs     lli_lsm_md(%p):\n"
1360                        "    magic:      %x                   %x\n"
1361                        "    count:      %x                   %x\n"
1362                        "    master:     %x                   %x\n"
1363                        "    hash_type:  %x                   %x\n"
1364                        "    layout:     %x                   %x\n"
1365                        "    pool:       %s                   %s\n",
1366                        inode, inode->i_ino, lsm, lli->lli_lsm_md,
1367                        lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1368                        lsm->lsm_md_stripe_count,
1369                        lli->lli_lsm_md->lsm_md_stripe_count,
1370                        lsm->lsm_md_master_mdt_index,
1371                        lli->lli_lsm_md->lsm_md_master_mdt_index,
1372                        lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1373                        lsm->lsm_md_layout_version,
1374                        lli->lli_lsm_md->lsm_md_layout_version,
1375                        lsm->lsm_md_pool_name,
1376                        lli->lli_lsm_md->lsm_md_pool_name);
1377                 return;
1378         }
1379
1380         for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1381                 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1382                                &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1383                         CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1384                                "new:"DFID"\n",
1385                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1386                              PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1387                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1388                         return;
1389                 }
1390         }
1391
1392         md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1393                          md->body, ll_md_blocking_ast);
1394 }
1395
1396 void ll_clear_inode(struct inode *inode)
1397 {
1398         struct ll_inode_info *lli = ll_i2info(inode);
1399         struct ll_sb_info *sbi = ll_i2sbi(inode);
1400         ENTRY;
1401
1402         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1403                PFID(ll_inode2fid(inode)), inode);
1404
1405         if (S_ISDIR(inode->i_mode)) {
1406                 /* these should have been cleared in ll_file_release */
1407                 LASSERT(lli->lli_opendir_key == NULL);
1408                 LASSERT(lli->lli_sai == NULL);
1409                 LASSERT(lli->lli_opendir_pid == 0);
1410         }
1411
1412         spin_lock(&lli->lli_lock);
1413         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1414         spin_unlock(&lli->lli_lock);
1415         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1416
1417         LASSERT(!lli->lli_open_fd_write_count);
1418         LASSERT(!lli->lli_open_fd_read_count);
1419         LASSERT(!lli->lli_open_fd_exec_count);
1420
1421         if (lli->lli_mds_write_och)
1422                 ll_md_real_close(inode, FMODE_WRITE);
1423         if (lli->lli_mds_exec_och)
1424                 ll_md_real_close(inode, FMODE_EXEC);
1425         if (lli->lli_mds_read_och)
1426                 ll_md_real_close(inode, FMODE_READ);
1427
1428         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429                 OBD_FREE(lli->lli_symlink_name,
1430                          strlen(lli->lli_symlink_name) + 1);
1431                 lli->lli_symlink_name = NULL;
1432         }
1433
1434         ll_xattr_cache_destroy(inode);
1435
1436         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1437                 LASSERT(lli->lli_posix_acl == NULL);
1438                 if (lli->lli_remote_perms) {
1439                         free_rmtperm_hash(lli->lli_remote_perms);
1440                         lli->lli_remote_perms = NULL;
1441                 }
1442         }
1443 #ifdef CONFIG_FS_POSIX_ACL
1444         else if (lli->lli_posix_acl) {
1445                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1446                 LASSERT(lli->lli_remote_perms == NULL);
1447                 posix_acl_release(lli->lli_posix_acl);
1448                 lli->lli_posix_acl = NULL;
1449         }
1450 #endif
1451         lli->lli_inode_magic = LLI_INODE_DEAD;
1452
1453         ll_clear_inode_capas(inode);
1454         if (S_ISDIR(inode->i_mode))
1455                 ll_dir_clear_lsm_md(inode);
1456         else
1457                 LASSERT(list_empty(&lli->lli_agl_list));
1458
1459         /*
1460          * XXX This has to be done before lsm is freed below, because
1461          * cl_object still uses inode lsm.
1462          */
1463         cl_inode_fini(inode);
1464         lli->lli_has_smd = false;
1465
1466         EXIT;
1467 }
1468
1469 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1470                   struct md_open_data **mod)
1471 {
1472         struct lustre_md md;
1473         struct inode *inode = dentry->d_inode;
1474         struct ll_sb_info *sbi = ll_i2sbi(inode);
1475         struct ptlrpc_request *request = NULL;
1476         int rc, ia_valid;
1477         ENTRY;
1478
1479         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1480                                      LUSTRE_OPC_ANY, NULL);
1481         if (IS_ERR(op_data))
1482                 RETURN(PTR_ERR(op_data));
1483
1484         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1485                         &request, mod);
1486         if (rc) {
1487                 ptlrpc_req_finished(request);
1488                 if (rc == -ENOENT) {
1489                         clear_nlink(inode);
1490                         /* Unlinked special device node? Or just a race?
1491                          * Pretend we done everything. */
1492                         if (!S_ISREG(inode->i_mode) &&
1493                             !S_ISDIR(inode->i_mode)) {
1494                                 ia_valid = op_data->op_attr.ia_valid;
1495                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1496                                 rc = simple_setattr(dentry, &op_data->op_attr);
1497                                 op_data->op_attr.ia_valid = ia_valid;
1498                         }
1499                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1500                         CERROR("md_setattr fails: rc = %d\n", rc);
1501                 }
1502                 RETURN(rc);
1503         }
1504
1505         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1506                               sbi->ll_md_exp, &md);
1507         if (rc) {
1508                 ptlrpc_req_finished(request);
1509                 RETURN(rc);
1510         }
1511
1512         ia_valid = op_data->op_attr.ia_valid;
1513         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1514          * cache is not cleared yet. */
1515         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1516         rc = simple_setattr(dentry, &op_data->op_attr);
1517         op_data->op_attr.ia_valid = ia_valid;
1518
1519         /* Extract epoch data if obtained. */
1520         op_data->op_handle = md.body->handle;
1521         op_data->op_ioepoch = md.body->ioepoch;
1522
1523         ll_update_inode(inode, &md);
1524         ptlrpc_req_finished(request);
1525
1526         RETURN(rc);
1527 }
1528
1529 /* Close IO epoch and send Size-on-MDS attribute update. */
1530 static int ll_setattr_done_writing(struct inode *inode,
1531                                    struct md_op_data *op_data,
1532                                    struct md_open_data *mod)
1533 {
1534         struct ll_inode_info *lli = ll_i2info(inode);
1535         int rc = 0;
1536         ENTRY;
1537
1538         LASSERT(op_data != NULL);
1539         if (!S_ISREG(inode->i_mode))
1540                 RETURN(0);
1541
1542         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1543                op_data->op_ioepoch, PFID(&lli->lli_fid));
1544
1545         op_data->op_flags = MF_EPOCH_CLOSE;
1546         ll_done_writing_attr(inode, op_data);
1547         ll_pack_inode2opdata(inode, op_data, NULL);
1548
1549         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1550         if (rc == -EAGAIN) {
1551                 /* MDS has instructed us to obtain Size-on-MDS attribute
1552                  * from OSTs and send setattr to back to MDS. */
1553                 rc = ll_som_update(inode, op_data);
1554         } else if (rc) {
1555                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1556                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1557                        PFID(ll_inode2fid(inode)), rc);
1558         }
1559         RETURN(rc);
1560 }
1561
1562 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1563 {
1564         struct obd_capa *capa;
1565         int rc;
1566
1567         if (attr->ia_valid & ATTR_SIZE)
1568                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1569         else
1570                 capa = ll_mdscapa_get(inode);
1571
1572         rc = cl_setattr_ost(inode, attr, capa);
1573
1574         if (attr->ia_valid & ATTR_SIZE)
1575                 ll_truncate_free_capa(capa);
1576         else
1577                 capa_put(capa);
1578
1579         return rc;
1580 }
1581
1582 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1583  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1584  * keep these values until such a time that objects are allocated for it.
1585  * We do the MDS operations first, as it is checking permissions for us.
1586  * We don't to the MDS RPC if there is nothing that we want to store there,
1587  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1588  * going to do an RPC anyways.
1589  *
1590  * If we are doing a truncate, we will send the mtime and ctime updates
1591  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1592  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1593  * at the same time.
1594  *
1595  * In case of HSMimport, we only set attr on MDS.
1596  */
1597 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1598 {
1599         struct inode *inode = dentry->d_inode;
1600         struct ll_inode_info *lli = ll_i2info(inode);
1601         struct md_op_data *op_data = NULL;
1602         struct md_open_data *mod = NULL;
1603         bool file_is_released = false;
1604         int rc = 0, rc1 = 0;
1605         ENTRY;
1606
1607         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1608                "valid %x, hsm_import %d\n",
1609                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1610                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1611                hsm_import);
1612
1613         if (attr->ia_valid & ATTR_SIZE) {
1614                 /* Check new size against VFS/VM file size limit and rlimit */
1615                 rc = inode_newsize_ok(inode, attr->ia_size);
1616                 if (rc)
1617                         RETURN(rc);
1618
1619                 /* The maximum Lustre file size is variable, based on the
1620                  * OST maximum object size and number of stripes.  This
1621                  * needs another check in addition to the VFS check above. */
1622                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1623                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1624                                PFID(&lli->lli_fid), attr->ia_size,
1625                                ll_file_maxbytes(inode));
1626                         RETURN(-EFBIG);
1627                 }
1628
1629                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1630         }
1631
1632         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1633         if (attr->ia_valid & TIMES_SET_FLAGS) {
1634                 if (current_fsuid() != inode->i_uid &&
1635                     !cfs_capable(CFS_CAP_FOWNER))
1636                         RETURN(-EPERM);
1637         }
1638
1639         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1640         if (attr->ia_valid & ATTR_CTIME) {
1641                 attr->ia_ctime = CFS_CURRENT_TIME;
1642                 attr->ia_valid |= ATTR_CTIME_SET;
1643         }
1644         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1645             (attr->ia_valid & ATTR_ATIME)) {
1646                 attr->ia_atime = CFS_CURRENT_TIME;
1647                 attr->ia_valid |= ATTR_ATIME_SET;
1648         }
1649         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1650             (attr->ia_valid & ATTR_MTIME)) {
1651                 attr->ia_mtime = CFS_CURRENT_TIME;
1652                 attr->ia_valid |= ATTR_MTIME_SET;
1653         }
1654
1655         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1656                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1657                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1658                        cfs_time_current_sec());
1659
1660         /* We always do an MDS RPC, even if we're only changing the size;
1661          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1662
1663         OBD_ALLOC_PTR(op_data);
1664         if (op_data == NULL)
1665                 RETURN(-ENOMEM);
1666
1667         if (!S_ISDIR(inode->i_mode)) {
1668                 if (attr->ia_valid & ATTR_SIZE)
1669                         inode_dio_write_done(inode);
1670                 mutex_unlock(&inode->i_mutex);
1671                 down_write(&lli->lli_trunc_sem);
1672         }
1673
1674         /* truncate on a released file must failed with -ENODATA,
1675          * so size must not be set on MDS for released file
1676          * but other attributes must be set
1677          */
1678         if (S_ISREG(inode->i_mode)) {
1679                 struct lov_stripe_md *lsm;
1680                 __u32 gen;
1681
1682                 ll_layout_refresh(inode, &gen);
1683                 lsm = ccc_inode_lsm_get(inode);
1684                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1685                         file_is_released = true;
1686                 ccc_inode_lsm_put(inode, lsm);
1687
1688                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1689                         if (file_is_released) {
1690                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1691                                 if (rc < 0)
1692                                         GOTO(out, rc);
1693
1694                                 file_is_released = false;
1695                                 ll_layout_refresh(inode, &gen);
1696                         }
1697
1698                         /* If we are changing file size, file content is
1699                          * modified, flag it. */
1700                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1701                         spin_lock(&lli->lli_lock);
1702                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1703                         spin_unlock(&lli->lli_lock);
1704                         op_data->op_bias |= MDS_DATA_MODIFIED;
1705                 }
1706         }
1707
1708         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1709
1710         /* Open epoch for truncate. */
1711         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1712             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1713                 op_data->op_flags = MF_EPOCH_OPEN;
1714
1715         rc = ll_md_setattr(dentry, op_data, &mod);
1716         if (rc)
1717                 GOTO(out, rc);
1718
1719         /* RPC to MDT is sent, cancel data modification flag */
1720         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1721                 spin_lock(&lli->lli_lock);
1722                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1723                 spin_unlock(&lli->lli_lock);
1724         }
1725
1726         ll_ioepoch_open(lli, op_data->op_ioepoch);
1727         if (!S_ISREG(inode->i_mode) || file_is_released)
1728                 GOTO(out, rc = 0);
1729
1730         if (attr->ia_valid & (ATTR_SIZE |
1731                               ATTR_ATIME | ATTR_ATIME_SET |
1732                               ATTR_MTIME | ATTR_MTIME_SET)) {
1733                 /* For truncate and utimes sending attributes to OSTs, setting
1734                  * mtime/atime to the past will be performed under PW [0:EOF]
1735                  * extent lock (new_size:EOF for truncate).  It may seem
1736                  * excessive to send mtime/atime updates to OSTs when not
1737                  * setting times to past, but it is necessary due to possible
1738                  * time de-synchronization between MDT inode and OST objects */
1739                 rc = ll_setattr_ost(inode, attr);
1740         }
1741         EXIT;
1742 out:
1743         if (op_data) {
1744                 if (op_data->op_ioepoch) {
1745                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1746                         if (!rc)
1747                                 rc = rc1;
1748                 }
1749                 ll_finish_md_op_data(op_data);
1750         }
1751         if (!S_ISDIR(inode->i_mode)) {
1752                 up_write(&lli->lli_trunc_sem);
1753                 mutex_lock(&inode->i_mutex);
1754                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1755                         inode_dio_wait(inode);
1756         }
1757
1758         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1759                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1760
1761         return rc;
1762 }
1763
1764 int ll_setattr(struct dentry *de, struct iattr *attr)
1765 {
1766         int mode = de->d_inode->i_mode;
1767
1768         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1769                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1770                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1771
1772         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1773                                (ATTR_SIZE|ATTR_MODE)) &&
1774             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1775              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1776               !(attr->ia_mode & S_ISGID))))
1777                 attr->ia_valid |= ATTR_FORCE;
1778
1779         if ((mode & S_ISUID) &&
1780             !(attr->ia_mode & S_ISUID) &&
1781             !(attr->ia_valid & ATTR_KILL_SUID))
1782                 attr->ia_valid |= ATTR_KILL_SUID;
1783
1784         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1785             !(attr->ia_mode & S_ISGID) &&
1786             !(attr->ia_valid & ATTR_KILL_SGID))
1787                 attr->ia_valid |= ATTR_KILL_SGID;
1788
1789         return ll_setattr_raw(de, attr, false);
1790 }
1791
1792 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1793                        __u64 max_age, __u32 flags)
1794 {
1795         struct ll_sb_info *sbi = ll_s2sbi(sb);
1796         struct obd_statfs obd_osfs;
1797         int rc;
1798         ENTRY;
1799
1800         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1801         if (rc) {
1802                 CERROR("md_statfs fails: rc = %d\n", rc);
1803                 RETURN(rc);
1804         }
1805
1806         osfs->os_type = sb->s_magic;
1807
1808         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1809                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1810
1811         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1812                 flags |= OBD_STATFS_NODELAY;
1813
1814         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1815         if (rc) {
1816                 CERROR("obd_statfs fails: rc = %d\n", rc);
1817                 RETURN(rc);
1818         }
1819
1820         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1821                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1822                obd_osfs.os_files);
1823
1824         osfs->os_bsize = obd_osfs.os_bsize;
1825         osfs->os_blocks = obd_osfs.os_blocks;
1826         osfs->os_bfree = obd_osfs.os_bfree;
1827         osfs->os_bavail = obd_osfs.os_bavail;
1828
1829         /* If we don't have as many objects free on the OST as inodes
1830          * on the MDS, we reduce the total number of inodes to
1831          * compensate, so that the "inodes in use" number is correct.
1832          */
1833         if (obd_osfs.os_ffree < osfs->os_ffree) {
1834                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1835                         obd_osfs.os_ffree;
1836                 osfs->os_ffree = obd_osfs.os_ffree;
1837         }
1838
1839         RETURN(rc);
1840 }
1841 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1842 {
1843         struct super_block *sb = de->d_sb;
1844         struct obd_statfs osfs;
1845         __u64 fsid = huge_encode_dev(sb->s_dev);
1846         int rc;
1847
1848         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1849         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1850
1851         /* Some amount of caching on the client is allowed */
1852         rc = ll_statfs_internal(sb, &osfs,
1853                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1854                                 0);
1855         if (rc)
1856                 return rc;
1857
1858         statfs_unpack(sfs, &osfs);
1859
1860         /* We need to downshift for all 32-bit kernels, because we can't
1861          * tell if the kernel is being called via sys_statfs64() or not.
1862          * Stop before overflowing f_bsize - in which case it is better
1863          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1864         if (sizeof(long) < 8) {
1865                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1866                         sfs->f_bsize <<= 1;
1867
1868                         osfs.os_blocks >>= 1;
1869                         osfs.os_bfree >>= 1;
1870                         osfs.os_bavail >>= 1;
1871                 }
1872         }
1873
1874         sfs->f_blocks = osfs.os_blocks;
1875         sfs->f_bfree = osfs.os_bfree;
1876         sfs->f_bavail = osfs.os_bavail;
1877         sfs->f_fsid.val[0] = (__u32)fsid;
1878         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1879         return 0;
1880 }
1881
1882 void ll_inode_size_lock(struct inode *inode)
1883 {
1884         struct ll_inode_info *lli;
1885
1886         LASSERT(!S_ISDIR(inode->i_mode));
1887
1888         lli = ll_i2info(inode);
1889         LASSERT(lli->lli_size_sem_owner != current);
1890         down(&lli->lli_size_sem);
1891         LASSERT(lli->lli_size_sem_owner == NULL);
1892         lli->lli_size_sem_owner = current;
1893 }
1894
1895 void ll_inode_size_unlock(struct inode *inode)
1896 {
1897         struct ll_inode_info *lli;
1898
1899         lli = ll_i2info(inode);
1900         LASSERT(lli->lli_size_sem_owner == current);
1901         lli->lli_size_sem_owner = NULL;
1902         up(&lli->lli_size_sem);
1903 }
1904
1905 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1906 {
1907         struct ll_inode_info *lli = ll_i2info(inode);
1908         struct mdt_body *body = md->body;
1909         struct lov_stripe_md *lsm = md->lsm;
1910         struct ll_sb_info *sbi = ll_i2sbi(inode);
1911
1912         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1913         if (lsm != NULL) {
1914                 if (!lli->lli_has_smd &&
1915                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1916                         cl_file_inode_init(inode, md);
1917
1918                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1919                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1920                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1921         }
1922
1923         if (S_ISDIR(inode->i_mode) && md->lmv != NULL)
1924                 ll_update_lsm_md(inode, md);
1925
1926         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1927                 if (body->valid & OBD_MD_FLRMTPERM)
1928                         ll_update_remote_perm(inode, md->remote_perm);
1929         }
1930 #ifdef CONFIG_FS_POSIX_ACL
1931         else if (body->valid & OBD_MD_FLACL) {
1932                 spin_lock(&lli->lli_lock);
1933                 if (lli->lli_posix_acl)
1934                         posix_acl_release(lli->lli_posix_acl);
1935                 lli->lli_posix_acl = md->posix_acl;
1936                 spin_unlock(&lli->lli_lock);
1937         }
1938 #endif
1939         inode->i_ino = cl_fid_build_ino(&body->fid1,
1940                                         sbi->ll_flags & LL_SBI_32BIT_API);
1941         inode->i_generation = cl_fid_build_gen(&body->fid1);
1942
1943         if (body->valid & OBD_MD_FLATIME) {
1944                 if (body->atime > LTIME_S(inode->i_atime))
1945                         LTIME_S(inode->i_atime) = body->atime;
1946                 lli->lli_lvb.lvb_atime = body->atime;
1947         }
1948         if (body->valid & OBD_MD_FLMTIME) {
1949                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1950                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1951                                "to "LPU64"\n", inode->i_ino,
1952                                LTIME_S(inode->i_mtime), body->mtime);
1953                         LTIME_S(inode->i_mtime) = body->mtime;
1954                 }
1955                 lli->lli_lvb.lvb_mtime = body->mtime;
1956         }
1957         if (body->valid & OBD_MD_FLCTIME) {
1958                 if (body->ctime > LTIME_S(inode->i_ctime))
1959                         LTIME_S(inode->i_ctime) = body->ctime;
1960                 lli->lli_lvb.lvb_ctime = body->ctime;
1961         }
1962         if (body->valid & OBD_MD_FLMODE)
1963                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1964         if (body->valid & OBD_MD_FLTYPE)
1965                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1966         LASSERT(inode->i_mode != 0);
1967         if (S_ISREG(inode->i_mode)) {
1968                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1969         } else {
1970                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1971         }
1972         if (body->valid & OBD_MD_FLUID)
1973                 inode->i_uid = body->uid;
1974         if (body->valid & OBD_MD_FLGID)
1975                 inode->i_gid = body->gid;
1976         if (body->valid & OBD_MD_FLFLAGS)
1977                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1978         if (body->valid & OBD_MD_FLNLINK)
1979                 set_nlink(inode, body->nlink);
1980         if (body->valid & OBD_MD_FLRDEV)
1981                 inode->i_rdev = old_decode_dev(body->rdev);
1982
1983         if (body->valid & OBD_MD_FLID) {
1984                 /* FID shouldn't be changed! */
1985                 if (fid_is_sane(&lli->lli_fid)) {
1986                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1987                                  "Trying to change FID "DFID
1988                                  " to the "DFID", inode "DFID"(%p)\n",
1989                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1990                                  PFID(ll_inode2fid(inode)), inode);
1991                 } else {
1992                         lli->lli_fid = body->fid1;
1993                 }
1994         }
1995
1996         LASSERT(fid_seq(&lli->lli_fid) != 0);
1997
1998         if (body->valid & OBD_MD_FLSIZE) {
1999                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2000                     S_ISREG(inode->i_mode)) {
2001                         struct lustre_handle lockh;
2002                         ldlm_mode_t mode;
2003
2004                         /* As it is possible a blocking ast has been processed
2005                          * by this time, we need to check there is an UPDATE
2006                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2007                          * it. */
2008                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2009                                                &lockh, LDLM_FL_CBPENDING,
2010                                                LCK_CR | LCK_CW |
2011                                                LCK_PR | LCK_PW);
2012                         if (mode) {
2013                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2014                                                       LLIF_EPOCH_PENDING |
2015                                                       LLIF_SOM_DIRTY)) {
2016                                         CERROR("%s: inode "DFID" flags %u still"
2017                                                " has size authority! do not "
2018                                                "trust the size from MDS\n",
2019                                                sbi->ll_md_exp->exp_obd->obd_name,
2020                                                PFID(ll_inode2fid(inode)),
2021                                                lli->lli_flags);
2022                                 } else {
2023                                         /* Use old size assignment to avoid
2024                                          * deadlock bz14138 & bz14326 */
2025                                         i_size_write(inode, body->size);
2026                                         spin_lock(&lli->lli_lock);
2027                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2028                                         spin_unlock(&lli->lli_lock);
2029                                 }
2030                                 ldlm_lock_decref(&lockh, mode);
2031                         }
2032                 } else {
2033                         /* Use old size assignment to avoid
2034                          * deadlock bz14138 & bz14326 */
2035                         i_size_write(inode, body->size);
2036
2037                         CDEBUG(D_VFSTRACE,
2038                                "inode="DFID", updating i_size %llu\n",
2039                                PFID(ll_inode2fid(inode)),
2040                                (unsigned long long)body->size);
2041                 }
2042
2043                 if (body->valid & OBD_MD_FLBLOCKS)
2044                         inode->i_blocks = body->blocks;
2045         }
2046
2047         if (body->valid & OBD_MD_FLMDSCAPA) {
2048                 LASSERT(md->mds_capa);
2049                 ll_add_capa(inode, md->mds_capa);
2050         }
2051         if (body->valid & OBD_MD_FLOSSCAPA) {
2052                 LASSERT(md->oss_capa);
2053                 ll_add_capa(inode, md->oss_capa);
2054         }
2055
2056         if (body->valid & OBD_MD_TSTATE) {
2057                 if (body->t_state & MS_RESTORE)
2058                         lli->lli_flags |= LLIF_FILE_RESTORING;
2059         }
2060 }
2061
2062 void ll_read_inode2(struct inode *inode, void *opaque)
2063 {
2064         struct lustre_md *md = opaque;
2065         struct ll_inode_info *lli = ll_i2info(inode);
2066         ENTRY;
2067
2068         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2069                PFID(&lli->lli_fid), inode);
2070
2071         LASSERT(!lli->lli_has_smd);
2072
2073         /* Core attributes from the MDS first.  This is a new inode, and
2074          * the VFS doesn't zero times in the core inode so we have to do
2075          * it ourselves.  They will be overwritten by either MDS or OST
2076          * attributes - we just need to make sure they aren't newer. */
2077         LTIME_S(inode->i_mtime) = 0;
2078         LTIME_S(inode->i_atime) = 0;
2079         LTIME_S(inode->i_ctime) = 0;
2080         inode->i_rdev = 0;
2081         ll_update_inode(inode, md);
2082
2083         /* OIDEBUG(inode); */
2084
2085         /* initializing backing dev info. */
2086         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2087
2088
2089         if (S_ISREG(inode->i_mode)) {
2090                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2091                 inode->i_op = &ll_file_inode_operations;
2092                 inode->i_fop = sbi->ll_fop;
2093                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2094                 EXIT;
2095         } else if (S_ISDIR(inode->i_mode)) {
2096                 inode->i_op = &ll_dir_inode_operations;
2097                 inode->i_fop = &ll_dir_operations;
2098                 EXIT;
2099         } else if (S_ISLNK(inode->i_mode)) {
2100                 inode->i_op = &ll_fast_symlink_inode_operations;
2101                 EXIT;
2102         } else {
2103                 inode->i_op = &ll_special_inode_operations;
2104
2105                 init_special_inode(inode, inode->i_mode,
2106                                    inode->i_rdev);
2107
2108                 EXIT;
2109         }
2110 }
2111
2112 void ll_delete_inode(struct inode *inode)
2113 {
2114         struct cl_inode_info *lli = cl_i2info(inode);
2115         ENTRY;
2116
2117         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2118                 /* discard all dirty pages before truncating them, required by
2119                  * osc_extent implementation at LU-1030. */
2120                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2121                                    CL_FSYNC_DISCARD, 1);
2122
2123         truncate_inode_pages(&inode->i_data, 0);
2124
2125         /* Workaround for LU-118 */
2126         if (inode->i_data.nrpages) {
2127                 spin_lock_irq(&inode->i_data.tree_lock);
2128                 spin_unlock_irq(&inode->i_data.tree_lock);
2129                 LASSERTF(inode->i_data.nrpages == 0,
2130                          "inode="DFID"(%p) nrpages=%lu, see "
2131                          "http://jira.whamcloud.com/browse/LU-118\n",
2132                          PFID(ll_inode2fid(inode)), inode,
2133                          inode->i_data.nrpages);
2134         }
2135         /* Workaround end */
2136
2137 #ifdef HAVE_SBOPS_EVICT_INODE
2138         ll_clear_inode(inode);
2139 #endif
2140         clear_inode(inode);
2141
2142         EXIT;
2143 }
2144
2145 int ll_iocontrol(struct inode *inode, struct file *file,
2146                  unsigned int cmd, unsigned long arg)
2147 {
2148         struct ll_sb_info *sbi = ll_i2sbi(inode);
2149         struct ptlrpc_request *req = NULL;
2150         int rc, flags = 0;
2151         ENTRY;
2152
2153         switch(cmd) {
2154         case FSFILT_IOC_GETFLAGS: {
2155                 struct mdt_body *body;
2156                 struct md_op_data *op_data;
2157
2158                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2159                                              0, 0, LUSTRE_OPC_ANY,
2160                                              NULL);
2161                 if (IS_ERR(op_data))
2162                         RETURN(PTR_ERR(op_data));
2163
2164                 op_data->op_valid = OBD_MD_FLFLAGS;
2165                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2166                 ll_finish_md_op_data(op_data);
2167                 if (rc) {
2168                         CERROR("%s: failure inode "DFID": rc = %d\n",
2169                                sbi->ll_md_exp->exp_obd->obd_name,
2170                                PFID(ll_inode2fid(inode)), rc);
2171                         RETURN(-abs(rc));
2172                 }
2173
2174                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2175
2176                 flags = body->flags;
2177
2178                 ptlrpc_req_finished(req);
2179
2180                 RETURN(put_user(flags, (int *)arg));
2181         }
2182         case FSFILT_IOC_SETFLAGS: {
2183                 struct lov_stripe_md *lsm;
2184                 struct obd_info oinfo = { { { 0 } } };
2185                 struct md_op_data *op_data;
2186
2187                 if (get_user(flags, (int *)arg))
2188                         RETURN(-EFAULT);
2189
2190                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2191                                              LUSTRE_OPC_ANY, NULL);
2192                 if (IS_ERR(op_data))
2193                         RETURN(PTR_ERR(op_data));
2194
2195                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2196                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2197                 rc = md_setattr(sbi->ll_md_exp, op_data,
2198                                 NULL, 0, NULL, 0, &req, NULL);
2199                 ll_finish_md_op_data(op_data);
2200                 ptlrpc_req_finished(req);
2201                 if (rc)
2202                         RETURN(rc);
2203
2204                 inode->i_flags = ll_ext_to_inode_flags(flags);
2205
2206                 lsm = ccc_inode_lsm_get(inode);
2207                 if (!lsm_has_objects(lsm)) {
2208                         ccc_inode_lsm_put(inode, lsm);
2209                         RETURN(0);
2210                 }
2211
2212                 OBDO_ALLOC(oinfo.oi_oa);
2213                 if (!oinfo.oi_oa) {
2214                         ccc_inode_lsm_put(inode, lsm);
2215                         RETURN(-ENOMEM);
2216                 }
2217                 oinfo.oi_md = lsm;
2218                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2219                 oinfo.oi_oa->o_flags = flags;
2220                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2221                                        OBD_MD_FLGROUP;
2222                 oinfo.oi_capa = ll_mdscapa_get(inode);
2223                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2224                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2225                 capa_put(oinfo.oi_capa);
2226                 OBDO_FREE(oinfo.oi_oa);
2227                 ccc_inode_lsm_put(inode, lsm);
2228
2229                 if (rc && rc != -EPERM && rc != -EACCES)
2230                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2231
2232                 RETURN(rc);
2233         }
2234         default:
2235                 RETURN(-ENOSYS);
2236         }
2237
2238         RETURN(0);
2239 }
2240
2241 int ll_flush_ctx(struct inode *inode)
2242 {
2243         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2244
2245         CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2246
2247         obd_set_info_async(NULL, sbi->ll_md_exp,
2248                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2249                            0, NULL, NULL);
2250         obd_set_info_async(NULL, sbi->ll_dt_exp,
2251                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2252                            0, NULL, NULL);
2253         return 0;
2254 }
2255
2256 /* umount -f client means force down, don't save state */
2257 void ll_umount_begin(struct super_block *sb)
2258 {
2259         struct ll_sb_info *sbi = ll_s2sbi(sb);
2260         struct obd_device *obd;
2261         struct obd_ioctl_data *ioc_data;
2262         ENTRY;
2263
2264         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2265                sb->s_count, atomic_read(&sb->s_active));
2266
2267         obd = class_exp2obd(sbi->ll_md_exp);
2268         if (obd == NULL) {
2269                 CERROR("Invalid MDC connection handle "LPX64"\n",
2270                        sbi->ll_md_exp->exp_handle.h_cookie);
2271                 EXIT;
2272                 return;
2273         }
2274         obd->obd_force = 1;
2275
2276         obd = class_exp2obd(sbi->ll_dt_exp);
2277         if (obd == NULL) {
2278                 CERROR("Invalid LOV connection handle "LPX64"\n",
2279                        sbi->ll_dt_exp->exp_handle.h_cookie);
2280                 EXIT;
2281                 return;
2282         }
2283         obd->obd_force = 1;
2284
2285         OBD_ALLOC_PTR(ioc_data);
2286         if (ioc_data) {
2287                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2288                               sizeof *ioc_data, ioc_data, NULL);
2289
2290                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2291                               sizeof *ioc_data, ioc_data, NULL);
2292
2293                 OBD_FREE_PTR(ioc_data);
2294         }
2295
2296         /* Really, we'd like to wait until there are no requests outstanding,
2297          * and then continue.  For now, we just invalidate the requests,
2298          * schedule() and sleep one second if needed, and hope.
2299          */
2300         schedule();
2301         EXIT;
2302 }
2303
2304 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2305 {
2306         struct ll_sb_info *sbi = ll_s2sbi(sb);
2307         char *profilenm = get_profile_name(sb);
2308         int err;
2309         __u32 read_only;
2310
2311         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2312                 read_only = *flags & MS_RDONLY;
2313                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2314                                          sizeof(KEY_READ_ONLY),
2315                                          KEY_READ_ONLY, sizeof(read_only),
2316                                          &read_only, NULL);
2317                 if (err) {
2318                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2319                                       profilenm, read_only ?
2320                                       "read-only" : "read-write", err);
2321                         return err;
2322                 }
2323
2324                 if (read_only)
2325                         sb->s_flags |= MS_RDONLY;
2326                 else
2327                         sb->s_flags &= ~MS_RDONLY;
2328
2329                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2330                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2331                                       read_only ?  "read-only" : "read-write");
2332         }
2333         return 0;
2334 }
2335
2336 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2337                   struct super_block *sb, struct lookup_intent *it)
2338 {
2339         struct ll_sb_info *sbi = NULL;
2340         struct lustre_md md = { 0 };
2341         int rc;
2342         ENTRY;
2343
2344         LASSERT(*inode || sb);
2345         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2346         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2347                               sbi->ll_md_exp, &md);
2348         if (rc)
2349                 RETURN(rc);
2350
2351         if (*inode) {
2352                 ll_update_inode(*inode, &md);
2353         } else {
2354                 LASSERT(sb != NULL);
2355
2356                 /*
2357                  * At this point server returns to client's same fid as client
2358                  * generated for creating. So using ->fid1 is okay here.
2359                  */
2360                 LASSERT(fid_is_sane(&md.body->fid1));
2361
2362                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2363                                              sbi->ll_flags & LL_SBI_32BIT_API),
2364                                  &md);
2365                 if (*inode == NULL || IS_ERR(*inode)) {
2366 #ifdef CONFIG_FS_POSIX_ACL
2367                         if (md.posix_acl) {
2368                                 posix_acl_release(md.posix_acl);
2369                                 md.posix_acl = NULL;
2370                         }
2371 #endif
2372                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2373                         *inode = NULL;
2374                         CERROR("new_inode -fatal: rc %d\n", rc);
2375                         GOTO(out, rc);
2376                 }
2377         }
2378
2379         /* Handling piggyback layout lock.
2380          * Layout lock can be piggybacked by getattr and open request.
2381          * The lsm can be applied to inode only if it comes with a layout lock
2382          * otherwise correct layout may be overwritten, for example:
2383          * 1. proc1: mdt returns a lsm but not granting layout
2384          * 2. layout was changed by another client
2385          * 3. proc2: refresh layout and layout lock granted
2386          * 4. proc1: to apply a stale layout */
2387         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2388                 struct lustre_handle lockh;
2389                 struct ldlm_lock *lock;
2390
2391                 lockh.cookie = it->d.lustre.it_lock_handle;
2392                 lock = ldlm_handle2lock(&lockh);
2393                 LASSERT(lock != NULL);
2394                 if (ldlm_has_layout(lock)) {
2395                         struct cl_object_conf conf;
2396
2397                         memset(&conf, 0, sizeof(conf));
2398                         conf.coc_opc = OBJECT_CONF_SET;
2399                         conf.coc_inode = *inode;
2400                         conf.coc_lock = lock;
2401                         conf.u.coc_md = &md;
2402                         (void)ll_layout_conf(*inode, &conf);
2403                 }
2404                 LDLM_LOCK_PUT(lock);
2405         }
2406
2407 out:
2408         if (md.lsm != NULL)
2409                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2410         md_free_lustre_md(sbi->ll_md_exp, &md);
2411         RETURN(rc);
2412 }
2413
2414 int ll_obd_statfs(struct inode *inode, void *arg)
2415 {
2416         struct ll_sb_info *sbi = NULL;
2417         struct obd_export *exp;
2418         char *buf = NULL;
2419         struct obd_ioctl_data *data = NULL;
2420         __u32 type;
2421         __u32 flags;
2422         int len = 0, rc;
2423
2424         if (!inode || !(sbi = ll_i2sbi(inode)))
2425                 GOTO(out_statfs, rc = -EINVAL);
2426
2427         rc = obd_ioctl_getdata(&buf, &len, arg);
2428         if (rc)
2429                 GOTO(out_statfs, rc);
2430
2431         data = (void*)buf;
2432         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2433             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2434                 GOTO(out_statfs, rc = -EINVAL);
2435
2436         if (data->ioc_inllen1 != sizeof(__u32) ||
2437             data->ioc_inllen2 != sizeof(__u32) ||
2438             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2439             data->ioc_plen2 != sizeof(struct obd_uuid))
2440                 GOTO(out_statfs, rc = -EINVAL);
2441
2442         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2443         if (type & LL_STATFS_LMV)
2444                 exp = sbi->ll_md_exp;
2445         else if (type & LL_STATFS_LOV)
2446                 exp = sbi->ll_dt_exp;
2447         else
2448                 GOTO(out_statfs, rc = -ENODEV);
2449
2450         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2451         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2452         if (rc)
2453                 GOTO(out_statfs, rc);
2454 out_statfs:
2455         if (buf)
2456                 obd_ioctl_freedata(buf, len);
2457         return rc;
2458 }
2459
2460 int ll_process_config(struct lustre_cfg *lcfg)
2461 {
2462         struct super_block *sb;
2463         unsigned long x;
2464         int rc = 0;
2465         char *ptr;
2466
2467         /* The instance name contains the sb: lustre-client-aacfe000 */
2468         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2469         if (!ptr || !*(++ptr))
2470                 return -EINVAL;
2471         if (sscanf(ptr, "%lx", &x) != 1)
2472                 return -EINVAL;
2473         sb = (struct super_block *)x;
2474         /* This better be a real Lustre superblock! */
2475         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2476
2477         /* Note we have not called client_common_fill_super yet, so
2478            proc fns must be able to handle that! */
2479         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2480                                           lcfg, sb);
2481         if (rc > 0)
2482                 rc = 0;
2483         return rc;
2484 }
2485
2486 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2487 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2488                                        struct inode *i1, struct inode *i2,
2489                                        const char *name, int namelen,
2490                                        int mode, __u32 opc, void *data)
2491 {
2492         LASSERT(i1 != NULL);
2493
2494         if (namelen > ll_i2sbi(i1)->ll_namelen)
2495                 return ERR_PTR(-ENAMETOOLONG);
2496
2497         if (op_data == NULL)
2498                 OBD_ALLOC_PTR(op_data);
2499
2500         if (op_data == NULL)
2501                 return ERR_PTR(-ENOMEM);
2502
2503         ll_i2gids(op_data->op_suppgids, i1, i2);
2504         op_data->op_fid1 = *ll_inode2fid(i1);
2505         op_data->op_capa1 = ll_mdscapa_get(i1);
2506         if (S_ISDIR(i1->i_mode))
2507                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2508
2509         if (i2) {
2510                 op_data->op_fid2 = *ll_inode2fid(i2);
2511                 op_data->op_capa2 = ll_mdscapa_get(i2);
2512                 if (S_ISDIR(i2->i_mode))
2513                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2514         } else {
2515                 fid_zero(&op_data->op_fid2);
2516                 op_data->op_capa2 = NULL;
2517         }
2518
2519         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2520                 op_data->op_cli_flags |= CLI_HASH64;
2521
2522         if (ll_need_32bit_api(ll_i2sbi(i1)))
2523                 op_data->op_cli_flags |= CLI_API32;
2524
2525         op_data->op_name = name;
2526         op_data->op_namelen = namelen;
2527         op_data->op_mode = mode;
2528         op_data->op_mod_time = cfs_time_current_sec();
2529         op_data->op_fsuid = current_fsuid();
2530         op_data->op_fsgid = current_fsgid();
2531         op_data->op_cap = cfs_curproc_cap_pack();
2532         op_data->op_bias = 0;
2533         op_data->op_cli_flags = 0;
2534         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2535              filename_is_volatile(name, namelen, NULL))
2536                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2537         op_data->op_opc = opc;
2538         op_data->op_mds = 0;
2539         op_data->op_data = data;
2540
2541         /* If the file is being opened after mknod() (normally due to NFS)
2542          * try to use the default stripe data from parent directory for
2543          * allocating OST objects.  Try to pass the parent FID to MDS. */
2544         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2545             !ll_i2info(i2)->lli_has_smd) {
2546                 struct ll_inode_info *lli = ll_i2info(i2);
2547
2548                 spin_lock(&lli->lli_lock);
2549                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2550                         op_data->op_fid1 = lli->lli_pfid;
2551                 spin_unlock(&lli->lli_lock);
2552                 /** We ignore parent's capability temporary. */
2553         }
2554
2555         /* When called by ll_setattr_raw, file is i1. */
2556         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2557                 op_data->op_bias |= MDS_DATA_MODIFIED;
2558
2559         return op_data;
2560 }
2561
2562 void ll_finish_md_op_data(struct md_op_data *op_data)
2563 {
2564         capa_put(op_data->op_capa1);
2565         capa_put(op_data->op_capa2);
2566         OBD_FREE_PTR(op_data);
2567 }
2568
2569 #ifdef HAVE_SUPEROPS_USE_DENTRY
2570 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2571 #else
2572 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2573 #endif
2574 {
2575         struct ll_sb_info *sbi;
2576
2577 #ifdef HAVE_SUPEROPS_USE_DENTRY
2578         LASSERT((seq != NULL) && (dentry != NULL));
2579         sbi = ll_s2sbi(dentry->d_sb);
2580 #else
2581         LASSERT((seq != NULL) && (vfs != NULL));
2582         sbi = ll_s2sbi(vfs->mnt_sb);
2583 #endif
2584
2585         if (sbi->ll_flags & LL_SBI_NOLCK)
2586                 seq_puts(seq, ",nolock");
2587
2588         if (sbi->ll_flags & LL_SBI_FLOCK)
2589                 seq_puts(seq, ",flock");
2590
2591         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2592                 seq_puts(seq, ",localflock");
2593
2594         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2595                 seq_puts(seq, ",user_xattr");
2596
2597         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2598                 seq_puts(seq, ",lazystatfs");
2599
2600         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2601                 seq_puts(seq, ",user_fid2path");
2602
2603         RETURN(0);
2604 }
2605
2606 /**
2607  * Get obd name by cmd, and copy out to user space
2608  */
2609 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2610 {
2611         struct ll_sb_info *sbi = ll_i2sbi(inode);
2612         struct obd_device *obd;
2613         ENTRY;
2614
2615         if (cmd == OBD_IOC_GETDTNAME)
2616                 obd = class_exp2obd(sbi->ll_dt_exp);
2617         else if (cmd == OBD_IOC_GETMDNAME)
2618                 obd = class_exp2obd(sbi->ll_md_exp);
2619         else
2620                 RETURN(-EINVAL);
2621
2622         if (!obd)
2623                 RETURN(-ENOENT);
2624
2625         if (copy_to_user((void *)arg, obd->obd_name,
2626                          strlen(obd->obd_name) + 1))
2627                 RETURN(-EFAULT);
2628
2629         RETURN(0);
2630 }
2631
2632 /**
2633  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2634  * fsname will be returned in this buffer; otherwise, a static buffer will be
2635  * used to store the fsname and returned to caller.
2636  */
2637 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2638 {
2639         static char fsname_static[MTI_NAME_MAXLEN];
2640         struct lustre_sb_info *lsi = s2lsi(sb);
2641         char *ptr;
2642         int len;
2643
2644         if (buf == NULL) {
2645                 /* this means the caller wants to use static buffer
2646                  * and it doesn't care about race. Usually this is
2647                  * in error reporting path */
2648                 buf = fsname_static;
2649                 buflen = sizeof(fsname_static);
2650         }
2651
2652         len = strlen(lsi->lsi_lmd->lmd_profile);
2653         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2654         if (ptr && (strcmp(ptr, "-client") == 0))
2655                 len -= 7;
2656
2657         if (unlikely(len >= buflen))
2658                 len = buflen - 1;
2659         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2660         buf[len] = '\0';
2661
2662         return buf;
2663 }
2664
2665 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2666 {
2667         char *path = NULL;
2668
2669         struct path p;
2670
2671         p.dentry = dentry;
2672         p.mnt = current->fs->root.mnt;
2673         path_get(&p);
2674         path = d_path(&p, buf, bufsize);
2675         path_put(&p);
2676         return path;
2677 }
2678
2679 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2680 {
2681         char *buf, *path = NULL;
2682         struct dentry *dentry = NULL;
2683         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2684
2685         /* this can be called inside spin lock so use GFP_ATOMIC. */
2686         buf = (char *)__get_free_page(GFP_ATOMIC);
2687         if (buf != NULL) {
2688                 dentry = d_find_alias(page->mapping->host);
2689                 if (dentry != NULL)
2690                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2691         }
2692
2693         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2694               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2695               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2696               PFID(&obj->cob_header.coh_lu.loh_fid),
2697               (path && !IS_ERR(path)) ? path : "", ioret);
2698
2699         if (dentry != NULL)
2700                 dput(dentry);
2701
2702         if (buf != NULL)
2703                 free_page((unsigned long)buf);
2704 }