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                 lli->lli_lmv_md = NULL;
 964         } else {
 965                 sema_init(&lli->lli_size_sem, 1);
 966                 lli->lli_size_sem_owner = NULL;
 967                 lli->lli_symlink_name = NULL;
 968                 init_rwsem(&lli->lli_trunc_sem);
 969                 mutex_init(&lli->lli_write_mutex);
 970                 init_rwsem(&lli->lli_glimpse_sem);
 971                 lli->lli_glimpse_time = 0;
 972                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 973                 lli->lli_agl_index = 0;
 974                 lli->lli_async_rc = 0;
 975                 lli->lli_volatile = false;
 976         }
 977         mutex_init(&lli->lli_layout_mutex);
 978 }
 979
 980 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 981 {
 982         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 983
 984         bdi->name = "lustre";
 985         return bdi_register(bdi, NULL, "lustre-%d",
 986                             atomic_inc_return(&ll_bdi_num));
 987 }
 988
 989 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 990 {
 991         struct lustre_profile *lprof = NULL;
 992         struct lustre_sb_info *lsi = s2lsi(sb);
 993         struct ll_sb_info *sbi;
 994         char  *dt = NULL, *md = NULL;
 995         char  *profilenm = get_profile_name(sb);
 996         struct config_llog_instance *cfg;
 997         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 998         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 999         int    err;
1000         ENTRY;
1001
1002         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1003
1004         OBD_ALLOC_PTR(cfg);
1005         if (cfg == NULL)
1006                 RETURN(-ENOMEM);
1007
1008         try_module_get(THIS_MODULE);
1009
1010         /* client additional sb info */
1011         lsi->lsi_llsbi = sbi = ll_init_sbi();
1012         if (!sbi) {
1013                 module_put(THIS_MODULE);
1014                 OBD_FREE_PTR(cfg);
1015                 RETURN(-ENOMEM);
1016         }
1017
1018         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1019         if (err)
1020                 GOTO(out_free, err);
1021
1022         err = bdi_init(&lsi->lsi_bdi);
1023         if (err)
1024                 GOTO(out_free, err);
1025         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1026         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1027         err = ll_bdi_register(&lsi->lsi_bdi);
1028         if (err)
1029                 GOTO(out_free, err);
1030
1031         sb->s_bdi = &lsi->lsi_bdi;
1032 #ifndef HAVE_DCACHE_LOCK
1033         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1034         sb->s_d_op = &ll_d_ops;
1035 #endif
1036
1037         /* Generate a string unique to this super, in case some joker tries
1038            to mount the same fs at two mount points.
1039            Use the address of the super itself.*/
1040         cfg->cfg_instance = sb;
1041         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1042         cfg->cfg_callback = class_config_llog_handler;
1043         /* set up client obds */
1044         err = lustre_process_log(sb, profilenm, cfg);
1045         if (err < 0)
1046                 GOTO(out_free, err);
1047
1048         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1049         lprof = class_get_profile(profilenm);
1050         if (lprof == NULL) {
1051                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1052                                    " read from the MGS.  Does that filesystem "
1053                                    "exist?\n", profilenm);
1054                 GOTO(out_free, err = -EINVAL);
1055         }
1056         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1057                lprof->lp_md, lprof->lp_dt);
1058
1059         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1060         if (!dt)
1061                 GOTO(out_free, err = -ENOMEM);
1062         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1063
1064         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1065         if (!md)
1066                 GOTO(out_free, err = -ENOMEM);
1067         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1068
1069         /* connections, registrations, sb setup */
1070         err = client_common_fill_super(sb, md, dt, mnt);
1071
1072 out_free:
1073         if (md)
1074                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1075         if (dt)
1076                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1077         if (err)
1078                 ll_put_super(sb);
1079         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1080                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1081
1082         OBD_FREE_PTR(cfg);
1083         RETURN(err);
1084 } /* ll_fill_super */
1085
1086
1087 void lu_context_keys_dump(void);
1088
1089 void ll_put_super(struct super_block *sb)
1090 {
1091         struct config_llog_instance cfg, params_cfg;
1092         struct obd_device *obd;
1093         struct lustre_sb_info *lsi = s2lsi(sb);
1094         struct ll_sb_info *sbi = ll_s2sbi(sb);
1095         char *profilenm = get_profile_name(sb);
1096         int ccc_count, next, force = 1, rc = 0;
1097         ENTRY;
1098
1099         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1100
1101         ll_print_capa_stat(sbi);
1102
1103         cfg.cfg_instance = sb;
1104         lustre_end_log(sb, profilenm, &cfg);
1105
1106         params_cfg.cfg_instance = sb;
1107         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1108
1109         if (sbi->ll_md_exp) {
1110                 obd = class_exp2obd(sbi->ll_md_exp);
1111                 if (obd)
1112                         force = obd->obd_force;
1113         }
1114
1115         /* Wait for unstable pages to be committed to stable storage */
1116         if (force == 0) {
1117                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1118                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1119                         cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1120                         &lwi);
1121         }
1122
1123         ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1124         if (force == 0 && rc != -EINTR)
1125                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1126
1127
1128         /* We need to set force before the lov_disconnect in
1129            lustre_common_put_super, since l_d cleans up osc's as well. */
1130         if (force) {
1131                 next = 0;
1132                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1133                                                      &next)) != NULL) {
1134                         obd->obd_force = force;
1135                 }
1136         }
1137
1138         if (sbi->ll_lcq) {
1139                 /* Only if client_common_fill_super succeeded */
1140                 client_common_put_super(sb);
1141         }
1142
1143         next = 0;
1144         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1145                 class_manual_cleanup(obd);
1146         }
1147
1148         if (sbi->ll_flags & LL_SBI_VERBOSE)
1149                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1150
1151         if (profilenm)
1152                 class_del_profile(profilenm);
1153
1154         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1155                 bdi_destroy(&lsi->lsi_bdi);
1156                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1157         }
1158
1159         ll_free_sbi(sb);
1160         lsi->lsi_llsbi = NULL;
1161
1162         lustre_common_put_super(sb);
1163
1164         cl_env_cache_purge(~0);
1165
1166         module_put(THIS_MODULE);
1167
1168         EXIT;
1169 } /* client_put_super */
1170
1171 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1172 {
1173         struct inode *inode = NULL;
1174
1175         /* NOTE: we depend on atomic igrab() -bzzz */
1176         lock_res_and_lock(lock);
1177         if (lock->l_resource->lr_lvb_inode) {
1178                 struct ll_inode_info * lli;
1179                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1180                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1181                         inode = igrab(lock->l_resource->lr_lvb_inode);
1182                 } else {
1183                         inode = lock->l_resource->lr_lvb_inode;
1184                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1185                                          D_WARNING, lock, "lr_lvb_inode %p is "
1186                                          "bogus: magic %08x",
1187                                          lock->l_resource->lr_lvb_inode,
1188                                          lli->lli_inode_magic);
1189                         inode = NULL;
1190                 }
1191         }
1192         unlock_res_and_lock(lock);
1193         return inode;
1194 }
1195
1196 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1197 {
1198         struct inode *inode = NULL;
1199         /* NOTE: we depend on atomic igrab() -bzzz */
1200         lock_res_and_lock(lock);
1201         if (lock->l_ast_data) {
1202                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1203                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1204                         inode = igrab(lock->l_ast_data);
1205                 } else {
1206                         inode = lock->l_ast_data;
1207                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1208                                          D_WARNING, lock, "l_ast_data %p is "
1209                                          "bogus: magic %08x", lock->l_ast_data,
1210                                          lli->lli_inode_magic);
1211                         inode = NULL;
1212                 }
1213         }
1214         unlock_res_and_lock(lock);
1215         return inode;
1216 }
1217
1218 static void ll_dir_clear_lsm_md(struct inode *inode)
1219 {
1220         struct ll_inode_info *lli = ll_i2info(inode);
1221
1222         LASSERT(S_ISDIR(inode->i_mode));
1223
1224         if (lli->lli_lsm_md != NULL) {
1225                 lmv_free_memmd(lli->lli_lsm_md);
1226                 lli->lli_lsm_md = NULL;
1227         }
1228 }
1229
1230 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1231                                       const struct lu_fid *fid,
1232                                       struct lustre_md *md)
1233 {
1234         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1235         struct mdt_body         *body = md->body;
1236         struct inode            *inode;
1237         ino_t                   ino;
1238         ENTRY;
1239
1240         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1241         inode = iget_locked(sb, ino);
1242         if (inode == NULL) {
1243                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1244                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1245                 RETURN(ERR_PTR(-ENOENT));
1246         }
1247
1248         if (inode->i_state & I_NEW) {
1249                 struct ll_inode_info *lli = ll_i2info(inode);
1250                 struct lmv_stripe_md *lsm = md->lmv;
1251
1252                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1253                                 (body->mode & S_IFMT);
1254                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1255                          PFID(fid));
1256
1257                 LTIME_S(inode->i_mtime) = 0;
1258                 LTIME_S(inode->i_atime) = 0;
1259                 LTIME_S(inode->i_ctime) = 0;
1260                 inode->i_rdev = 0;
1261
1262                 /* initializing backing dev info. */
1263                 inode->i_mapping->backing_dev_info =
1264                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1265                 inode->i_op = &ll_dir_inode_operations;
1266                 inode->i_fop = &ll_dir_operations;
1267                 lli->lli_fid = *fid;
1268                 ll_lli_init(lli);
1269
1270                 LASSERT(lsm != NULL);
1271                 /* master stripe FID */
1272                 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1273                 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1274                        lli, PFID(fid), PFID(&lli->lli_pfid));
1275                 unlock_new_inode(inode);
1276         }
1277
1278         RETURN(inode);
1279 }
1280
1281 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1282 {
1283         struct lu_fid *fid;
1284         struct lmv_stripe_md *lsm = md->lmv;
1285         int i;
1286
1287         LASSERT(lsm != NULL);
1288         /* XXX sigh, this lsm_root initialization should be in
1289          * LMV layer, but it needs ll_iget right now, so we
1290          * put this here right now. */
1291         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1292                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1293                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1294                 if (i == 0) {
1295                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1296                 } else {
1297                         /* Unfortunately ll_iget will call ll_update_inode,
1298                          * where the initialization of slave inode is slightly
1299                          * different, so it reset lsm_md to NULL to avoid
1300                          * initializing lsm for slave inode. */
1301                         lsm->lsm_md_oinfo[i].lmo_root =
1302                                         ll_iget_anon_dir(inode->i_sb, fid, md);
1303                         if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1304                                 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1305
1306                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1307                                 return rc;
1308                         }
1309                 }
1310         }
1311
1312         /* Here is where the lsm is being initialized(fill lmo_info) after
1313          * client retrieve MD stripe information from MDT. */
1314         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1315                                 ll_md_blocking_ast);
1316 }
1317
1318 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1319                                 const struct lmv_stripe_md *lsm_md2)
1320 {
1321         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1322                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1323                lsm_md1->lsm_md_master_mdt_index ==
1324                                         lsm_md2->lsm_md_master_mdt_index &&
1325                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1326                lsm_md1->lsm_md_layout_version ==
1327                                         lsm_md2->lsm_md_layout_version &&
1328                strcmp(lsm_md1->lsm_md_pool_name,
1329                       lsm_md2->lsm_md_pool_name) == 0;
1330 }
1331
1332 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1333 {
1334         struct ll_inode_info *lli = ll_i2info(inode);
1335         struct lmv_stripe_md *lsm = md->lmv;
1336         int idx;
1337         ENTRY;
1338
1339         LASSERT(S_ISDIR(inode->i_mode));
1340         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1341                PFID(ll_inode2fid(inode)));
1342
1343         /* no striped information from request. */
1344         if (lsm == NULL) {
1345                 if (lli->lli_lsm_md == NULL) {
1346                         RETURN_EXIT;
1347                 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1348                         /* migration is done, the temporay MIGRATE layout has
1349                          * been removed */
1350                         CDEBUG(D_INODE, DFID" finish migration.\n",
1351                                PFID(ll_inode2fid(inode)));
1352                         lmv_free_memmd(lli->lli_lsm_md);
1353                         lli->lli_lsm_md = NULL;
1354                         RETURN_EXIT;
1355                 } else {
1356                         /* The lustre_md from req does not include stripeEA,
1357                          * see ll_md_setattr */
1358                         RETURN_EXIT;
1359                 }
1360         }
1361
1362         /* set the directory layout */
1363         if (lli->lli_lsm_md == NULL) {
1364                 int rc;
1365
1366                 rc = ll_init_lsm_md(inode, md);
1367                 if (rc != 0) {
1368                         CERROR("%s: init "DFID" failed: rc = %d\n",
1369                                ll_get_fsname(inode->i_sb, NULL, 0),
1370                                PFID(&lli->lli_fid), rc);
1371                         return;
1372                 }
1373                 lli->lli_lsm_md = lsm;
1374                 /* set lsm_md to NULL, so the following free lustre_md
1375                  * will not free this lsm */
1376                 md->lmv = NULL;
1377                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1378                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1379                 RETURN_EXIT;
1380         }
1381
1382         /* Compare the old and new stripe information */
1383         if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1384                 CERROR("inode %p %lu mismatch\n"
1385                        "    new(%p)     vs     lli_lsm_md(%p):\n"
1386                        "    magic:      %x                   %x\n"
1387                        "    count:      %x                   %x\n"
1388                        "    master:     %x                   %x\n"
1389                        "    hash_type:  %x                   %x\n"
1390                        "    layout:     %x                   %x\n"
1391                        "    pool:       %s                   %s\n",
1392                        inode, inode->i_ino, lsm, lli->lli_lsm_md,
1393                        lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1394                        lsm->lsm_md_stripe_count,
1395                        lli->lli_lsm_md->lsm_md_stripe_count,
1396                        lsm->lsm_md_master_mdt_index,
1397                        lli->lli_lsm_md->lsm_md_master_mdt_index,
1398                        lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1399                        lsm->lsm_md_layout_version,
1400                        lli->lli_lsm_md->lsm_md_layout_version,
1401                        lsm->lsm_md_pool_name,
1402                        lli->lli_lsm_md->lsm_md_pool_name);
1403                 return;
1404         }
1405
1406         for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1407                 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1408                                &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1409                         CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1410                                "new:"DFID"\n",
1411                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1412                              PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1413                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1414                         return;
1415                 }
1416         }
1417
1418         md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1419                          md->body, ll_md_blocking_ast);
1420
1421         RETURN_EXIT;
1422 }
1423
1424 void ll_clear_inode(struct inode *inode)
1425 {
1426         struct ll_inode_info *lli = ll_i2info(inode);
1427         struct ll_sb_info *sbi = ll_i2sbi(inode);
1428         ENTRY;
1429
1430         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1431                PFID(ll_inode2fid(inode)), inode);
1432
1433         if (S_ISDIR(inode->i_mode)) {
1434                 /* these should have been cleared in ll_file_release */
1435                 LASSERT(lli->lli_opendir_key == NULL);
1436                 LASSERT(lli->lli_sai == NULL);
1437                 LASSERT(lli->lli_opendir_pid == 0);
1438         }
1439
1440         spin_lock(&lli->lli_lock);
1441         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1442         spin_unlock(&lli->lli_lock);
1443         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1444
1445         LASSERT(!lli->lli_open_fd_write_count);
1446         LASSERT(!lli->lli_open_fd_read_count);
1447         LASSERT(!lli->lli_open_fd_exec_count);
1448
1449         if (lli->lli_mds_write_och)
1450                 ll_md_real_close(inode, FMODE_WRITE);
1451         if (lli->lli_mds_exec_och)
1452                 ll_md_real_close(inode, FMODE_EXEC);
1453         if (lli->lli_mds_read_och)
1454                 ll_md_real_close(inode, FMODE_READ);
1455
1456         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1457                 OBD_FREE(lli->lli_symlink_name,
1458                          strlen(lli->lli_symlink_name) + 1);
1459                 lli->lli_symlink_name = NULL;
1460         }
1461
1462         ll_xattr_cache_destroy(inode);
1463
1464         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1465                 LASSERT(lli->lli_posix_acl == NULL);
1466                 if (lli->lli_remote_perms) {
1467                         free_rmtperm_hash(lli->lli_remote_perms);
1468                         lli->lli_remote_perms = NULL;
1469                 }
1470         }
1471 #ifdef CONFIG_FS_POSIX_ACL
1472         else if (lli->lli_posix_acl) {
1473                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1474                 LASSERT(lli->lli_remote_perms == NULL);
1475                 posix_acl_release(lli->lli_posix_acl);
1476                 lli->lli_posix_acl = NULL;
1477         }
1478 #endif
1479         lli->lli_inode_magic = LLI_INODE_DEAD;
1480
1481         ll_clear_inode_capas(inode);
1482         if (S_ISDIR(inode->i_mode))
1483                 ll_dir_clear_lsm_md(inode);
1484         else
1485                 LASSERT(list_empty(&lli->lli_agl_list));
1486
1487         /*
1488          * XXX This has to be done before lsm is freed below, because
1489          * cl_object still uses inode lsm.
1490          */
1491         cl_inode_fini(inode);
1492         lli->lli_has_smd = false;
1493
1494         EXIT;
1495 }
1496
1497 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1498                   struct md_open_data **mod)
1499 {
1500         struct lustre_md md;
1501         struct inode *inode = dentry->d_inode;
1502         struct ll_sb_info *sbi = ll_i2sbi(inode);
1503         struct ptlrpc_request *request = NULL;
1504         int rc, ia_valid;
1505         ENTRY;
1506
1507         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1508                                      LUSTRE_OPC_ANY, NULL);
1509         if (IS_ERR(op_data))
1510                 RETURN(PTR_ERR(op_data));
1511
1512         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1513                         &request, mod);
1514         if (rc) {
1515                 ptlrpc_req_finished(request);
1516                 if (rc == -ENOENT) {
1517                         clear_nlink(inode);
1518                         /* Unlinked special device node? Or just a race?
1519                          * Pretend we done everything. */
1520                         if (!S_ISREG(inode->i_mode) &&
1521                             !S_ISDIR(inode->i_mode)) {
1522                                 ia_valid = op_data->op_attr.ia_valid;
1523                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1524                                 rc = simple_setattr(dentry, &op_data->op_attr);
1525                                 op_data->op_attr.ia_valid = ia_valid;
1526                         }
1527                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1528                         CERROR("md_setattr fails: rc = %d\n", rc);
1529                 }
1530                 RETURN(rc);
1531         }
1532
1533         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1534                               sbi->ll_md_exp, &md);
1535         if (rc) {
1536                 ptlrpc_req_finished(request);
1537                 RETURN(rc);
1538         }
1539
1540         ia_valid = op_data->op_attr.ia_valid;
1541         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1542          * cache is not cleared yet. */
1543         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1544         rc = simple_setattr(dentry, &op_data->op_attr);
1545         op_data->op_attr.ia_valid = ia_valid;
1546
1547         /* Extract epoch data if obtained. */
1548         op_data->op_handle = md.body->handle;
1549         op_data->op_ioepoch = md.body->ioepoch;
1550
1551         ll_update_inode(inode, &md);
1552         ptlrpc_req_finished(request);
1553
1554         RETURN(rc);
1555 }
1556
1557 /* Close IO epoch and send Size-on-MDS attribute update. */
1558 static int ll_setattr_done_writing(struct inode *inode,
1559                                    struct md_op_data *op_data,
1560                                    struct md_open_data *mod)
1561 {
1562         struct ll_inode_info *lli = ll_i2info(inode);
1563         int rc = 0;
1564         ENTRY;
1565
1566         LASSERT(op_data != NULL);
1567         if (!S_ISREG(inode->i_mode))
1568                 RETURN(0);
1569
1570         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1571                op_data->op_ioepoch, PFID(&lli->lli_fid));
1572
1573         op_data->op_flags = MF_EPOCH_CLOSE;
1574         ll_done_writing_attr(inode, op_data);
1575         ll_pack_inode2opdata(inode, op_data, NULL);
1576
1577         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1578         if (rc == -EAGAIN) {
1579                 /* MDS has instructed us to obtain Size-on-MDS attribute
1580                  * from OSTs and send setattr to back to MDS. */
1581                 rc = ll_som_update(inode, op_data);
1582         } else if (rc) {
1583                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1584                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1585                        PFID(ll_inode2fid(inode)), rc);
1586         }
1587         RETURN(rc);
1588 }
1589
1590 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1591 {
1592         struct obd_capa *capa;
1593         int rc;
1594
1595         if (attr->ia_valid & ATTR_SIZE)
1596                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1597         else
1598                 capa = ll_mdscapa_get(inode);
1599
1600         rc = cl_setattr_ost(inode, attr, capa);
1601
1602         if (attr->ia_valid & ATTR_SIZE)
1603                 ll_truncate_free_capa(capa);
1604         else
1605                 capa_put(capa);
1606
1607         return rc;
1608 }
1609
1610 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1611  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1612  * keep these values until such a time that objects are allocated for it.
1613  * We do the MDS operations first, as it is checking permissions for us.
1614  * We don't to the MDS RPC if there is nothing that we want to store there,
1615  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1616  * going to do an RPC anyways.
1617  *
1618  * If we are doing a truncate, we will send the mtime and ctime updates
1619  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1620  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1621  * at the same time.
1622  *
1623  * In case of HSMimport, we only set attr on MDS.
1624  */
1625 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1626 {
1627         struct inode *inode = dentry->d_inode;
1628         struct ll_inode_info *lli = ll_i2info(inode);
1629         struct md_op_data *op_data = NULL;
1630         struct md_open_data *mod = NULL;
1631         bool file_is_released = false;
1632         int rc = 0, rc1 = 0;
1633         ENTRY;
1634
1635         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1636                "valid %x, hsm_import %d\n",
1637                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1638                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1639                hsm_import);
1640
1641         if (attr->ia_valid & ATTR_SIZE) {
1642                 /* Check new size against VFS/VM file size limit and rlimit */
1643                 rc = inode_newsize_ok(inode, attr->ia_size);
1644                 if (rc)
1645                         RETURN(rc);
1646
1647                 /* The maximum Lustre file size is variable, based on the
1648                  * OST maximum object size and number of stripes.  This
1649                  * needs another check in addition to the VFS check above. */
1650                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1651                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1652                                PFID(&lli->lli_fid), attr->ia_size,
1653                                ll_file_maxbytes(inode));
1654                         RETURN(-EFBIG);
1655                 }
1656
1657                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1658         }
1659
1660         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1661         if (attr->ia_valid & TIMES_SET_FLAGS) {
1662                 if (current_fsuid() != inode->i_uid &&
1663                     !cfs_capable(CFS_CAP_FOWNER))
1664                         RETURN(-EPERM);
1665         }
1666
1667         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1668         if (attr->ia_valid & ATTR_CTIME) {
1669                 attr->ia_ctime = CFS_CURRENT_TIME;
1670                 attr->ia_valid |= ATTR_CTIME_SET;
1671         }
1672         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1673             (attr->ia_valid & ATTR_ATIME)) {
1674                 attr->ia_atime = CFS_CURRENT_TIME;
1675                 attr->ia_valid |= ATTR_ATIME_SET;
1676         }
1677         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1678             (attr->ia_valid & ATTR_MTIME)) {
1679                 attr->ia_mtime = CFS_CURRENT_TIME;
1680                 attr->ia_valid |= ATTR_MTIME_SET;
1681         }
1682
1683         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1684                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1685                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1686                        cfs_time_current_sec());
1687
1688         /* We always do an MDS RPC, even if we're only changing the size;
1689          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1690
1691         OBD_ALLOC_PTR(op_data);
1692         if (op_data == NULL)
1693                 RETURN(-ENOMEM);
1694
1695         if (!S_ISDIR(inode->i_mode)) {
1696                 if (attr->ia_valid & ATTR_SIZE)
1697                         inode_dio_write_done(inode);
1698                 mutex_unlock(&inode->i_mutex);
1699                 down_write(&lli->lli_trunc_sem);
1700         }
1701
1702         /* truncate on a released file must failed with -ENODATA,
1703          * so size must not be set on MDS for released file
1704          * but other attributes must be set
1705          */
1706         if (S_ISREG(inode->i_mode)) {
1707                 struct lov_stripe_md *lsm;
1708                 __u32 gen;
1709
1710                 ll_layout_refresh(inode, &gen);
1711                 lsm = ccc_inode_lsm_get(inode);
1712                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1713                         file_is_released = true;
1714                 ccc_inode_lsm_put(inode, lsm);
1715
1716                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1717                         if (file_is_released) {
1718                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1719                                 if (rc < 0)
1720                                         GOTO(out, rc);
1721
1722                                 file_is_released = false;
1723                                 ll_layout_refresh(inode, &gen);
1724                         }
1725
1726                         /* If we are changing file size, file content is
1727                          * modified, flag it. */
1728                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1729                         spin_lock(&lli->lli_lock);
1730                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1731                         spin_unlock(&lli->lli_lock);
1732                         op_data->op_bias |= MDS_DATA_MODIFIED;
1733                 }
1734         }
1735
1736         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1737
1738         /* Open epoch for truncate. */
1739         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1740             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1741                 op_data->op_flags = MF_EPOCH_OPEN;
1742
1743         rc = ll_md_setattr(dentry, op_data, &mod);
1744         if (rc)
1745                 GOTO(out, rc);
1746
1747         /* RPC to MDT is sent, cancel data modification flag */
1748         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1749                 spin_lock(&lli->lli_lock);
1750                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1751                 spin_unlock(&lli->lli_lock);
1752         }
1753
1754         ll_ioepoch_open(lli, op_data->op_ioepoch);
1755         if (!S_ISREG(inode->i_mode) || file_is_released)
1756                 GOTO(out, rc = 0);
1757
1758         if (attr->ia_valid & (ATTR_SIZE |
1759                               ATTR_ATIME | ATTR_ATIME_SET |
1760                               ATTR_MTIME | ATTR_MTIME_SET)) {
1761                 /* For truncate and utimes sending attributes to OSTs, setting
1762                  * mtime/atime to the past will be performed under PW [0:EOF]
1763                  * extent lock (new_size:EOF for truncate).  It may seem
1764                  * excessive to send mtime/atime updates to OSTs when not
1765                  * setting times to past, but it is necessary due to possible
1766                  * time de-synchronization between MDT inode and OST objects */
1767                 rc = ll_setattr_ost(inode, attr);
1768         }
1769         EXIT;
1770 out:
1771         if (op_data) {
1772                 if (op_data->op_ioepoch) {
1773                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1774                         if (!rc)
1775                                 rc = rc1;
1776                 }
1777                 ll_finish_md_op_data(op_data);
1778         }
1779         if (!S_ISDIR(inode->i_mode)) {
1780                 up_write(&lli->lli_trunc_sem);
1781                 mutex_lock(&inode->i_mutex);
1782                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1783                         inode_dio_wait(inode);
1784         }
1785
1786         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1787                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1788
1789         return rc;
1790 }
1791
1792 int ll_setattr(struct dentry *de, struct iattr *attr)
1793 {
1794         int mode = de->d_inode->i_mode;
1795
1796         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1797                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1798                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1799
1800         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1801                                (ATTR_SIZE|ATTR_MODE)) &&
1802             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1803              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1804               !(attr->ia_mode & S_ISGID))))
1805                 attr->ia_valid |= ATTR_FORCE;
1806
1807         if ((mode & S_ISUID) &&
1808             !(attr->ia_mode & S_ISUID) &&
1809             !(attr->ia_valid & ATTR_KILL_SUID))
1810                 attr->ia_valid |= ATTR_KILL_SUID;
1811
1812         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1813             !(attr->ia_mode & S_ISGID) &&
1814             !(attr->ia_valid & ATTR_KILL_SGID))
1815                 attr->ia_valid |= ATTR_KILL_SGID;
1816
1817         return ll_setattr_raw(de, attr, false);
1818 }
1819
1820 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1821                        __u64 max_age, __u32 flags)
1822 {
1823         struct ll_sb_info *sbi = ll_s2sbi(sb);
1824         struct obd_statfs obd_osfs;
1825         int rc;
1826         ENTRY;
1827
1828         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1829         if (rc) {
1830                 CERROR("md_statfs fails: rc = %d\n", rc);
1831                 RETURN(rc);
1832         }
1833
1834         osfs->os_type = sb->s_magic;
1835
1836         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1837                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1838
1839         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1840                 flags |= OBD_STATFS_NODELAY;
1841
1842         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1843         if (rc) {
1844                 CERROR("obd_statfs fails: rc = %d\n", rc);
1845                 RETURN(rc);
1846         }
1847
1848         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1849                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1850                obd_osfs.os_files);
1851
1852         osfs->os_bsize = obd_osfs.os_bsize;
1853         osfs->os_blocks = obd_osfs.os_blocks;
1854         osfs->os_bfree = obd_osfs.os_bfree;
1855         osfs->os_bavail = obd_osfs.os_bavail;
1856
1857         /* If we don't have as many objects free on the OST as inodes
1858          * on the MDS, we reduce the total number of inodes to
1859          * compensate, so that the "inodes in use" number is correct.
1860          */
1861         if (obd_osfs.os_ffree < osfs->os_ffree) {
1862                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1863                         obd_osfs.os_ffree;
1864                 osfs->os_ffree = obd_osfs.os_ffree;
1865         }
1866
1867         RETURN(rc);
1868 }
1869 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1870 {
1871         struct super_block *sb = de->d_sb;
1872         struct obd_statfs osfs;
1873         __u64 fsid = huge_encode_dev(sb->s_dev);
1874         int rc;
1875
1876         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1877         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1878
1879         /* Some amount of caching on the client is allowed */
1880         rc = ll_statfs_internal(sb, &osfs,
1881                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1882                                 0);
1883         if (rc)
1884                 return rc;
1885
1886         statfs_unpack(sfs, &osfs);
1887
1888         /* We need to downshift for all 32-bit kernels, because we can't
1889          * tell if the kernel is being called via sys_statfs64() or not.
1890          * Stop before overflowing f_bsize - in which case it is better
1891          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1892         if (sizeof(long) < 8) {
1893                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1894                         sfs->f_bsize <<= 1;
1895
1896                         osfs.os_blocks >>= 1;
1897                         osfs.os_bfree >>= 1;
1898                         osfs.os_bavail >>= 1;
1899                 }
1900         }
1901
1902         sfs->f_blocks = osfs.os_blocks;
1903         sfs->f_bfree = osfs.os_bfree;
1904         sfs->f_bavail = osfs.os_bavail;
1905         sfs->f_fsid.val[0] = (__u32)fsid;
1906         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1907         return 0;
1908 }
1909
1910 void ll_inode_size_lock(struct inode *inode)
1911 {
1912         struct ll_inode_info *lli;
1913
1914         LASSERT(!S_ISDIR(inode->i_mode));
1915
1916         lli = ll_i2info(inode);
1917         LASSERT(lli->lli_size_sem_owner != current);
1918         down(&lli->lli_size_sem);
1919         LASSERT(lli->lli_size_sem_owner == NULL);
1920         lli->lli_size_sem_owner = current;
1921 }
1922
1923 void ll_inode_size_unlock(struct inode *inode)
1924 {
1925         struct ll_inode_info *lli;
1926
1927         lli = ll_i2info(inode);
1928         LASSERT(lli->lli_size_sem_owner == current);
1929         lli->lli_size_sem_owner = NULL;
1930         up(&lli->lli_size_sem);
1931 }
1932
1933 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1934 {
1935         struct ll_inode_info *lli = ll_i2info(inode);
1936         struct mdt_body *body = md->body;
1937         struct lov_stripe_md *lsm = md->lsm;
1938         struct ll_sb_info *sbi = ll_i2sbi(inode);
1939
1940         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1941         if (lsm != NULL) {
1942                 if (!lli->lli_has_smd &&
1943                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1944                         cl_file_inode_init(inode, md);
1945
1946                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1947                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1948                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1949         }
1950
1951         if (S_ISDIR(inode->i_mode))
1952                 ll_update_lsm_md(inode, md);
1953
1954         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1955                 if (body->valid & OBD_MD_FLRMTPERM)
1956                         ll_update_remote_perm(inode, md->remote_perm);
1957         }
1958 #ifdef CONFIG_FS_POSIX_ACL
1959         else if (body->valid & OBD_MD_FLACL) {
1960                 spin_lock(&lli->lli_lock);
1961                 if (lli->lli_posix_acl)
1962                         posix_acl_release(lli->lli_posix_acl);
1963                 lli->lli_posix_acl = md->posix_acl;
1964                 spin_unlock(&lli->lli_lock);
1965         }
1966 #endif
1967         inode->i_ino = cl_fid_build_ino(&body->fid1,
1968                                         sbi->ll_flags & LL_SBI_32BIT_API);
1969         inode->i_generation = cl_fid_build_gen(&body->fid1);
1970
1971         if (body->valid & OBD_MD_FLATIME) {
1972                 if (body->atime > LTIME_S(inode->i_atime))
1973                         LTIME_S(inode->i_atime) = body->atime;
1974                 lli->lli_lvb.lvb_atime = body->atime;
1975         }
1976         if (body->valid & OBD_MD_FLMTIME) {
1977                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1978                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1979                                "to "LPU64"\n", inode->i_ino,
1980                                LTIME_S(inode->i_mtime), body->mtime);
1981                         LTIME_S(inode->i_mtime) = body->mtime;
1982                 }
1983                 lli->lli_lvb.lvb_mtime = body->mtime;
1984         }
1985         if (body->valid & OBD_MD_FLCTIME) {
1986                 if (body->ctime > LTIME_S(inode->i_ctime))
1987                         LTIME_S(inode->i_ctime) = body->ctime;
1988                 lli->lli_lvb.lvb_ctime = body->ctime;
1989         }
1990         if (body->valid & OBD_MD_FLMODE)
1991                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1992         if (body->valid & OBD_MD_FLTYPE)
1993                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1994         LASSERT(inode->i_mode != 0);
1995         if (S_ISREG(inode->i_mode)) {
1996                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1997         } else {
1998                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1999         }
2000         if (body->valid & OBD_MD_FLUID)
2001                 inode->i_uid = body->uid;
2002         if (body->valid & OBD_MD_FLGID)
2003                 inode->i_gid = body->gid;
2004         if (body->valid & OBD_MD_FLFLAGS)
2005                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
2006         if (body->valid & OBD_MD_FLNLINK)
2007                 set_nlink(inode, body->nlink);
2008         if (body->valid & OBD_MD_FLRDEV)
2009                 inode->i_rdev = old_decode_dev(body->rdev);
2010
2011         if (body->valid & OBD_MD_FLID) {
2012                 /* FID shouldn't be changed! */
2013                 if (fid_is_sane(&lli->lli_fid)) {
2014                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
2015                                  "Trying to change FID "DFID
2016                                  " to the "DFID", inode "DFID"(%p)\n",
2017                                  PFID(&lli->lli_fid), PFID(&body->fid1),
2018                                  PFID(ll_inode2fid(inode)), inode);
2019                 } else {
2020                         lli->lli_fid = body->fid1;
2021                 }
2022         }
2023
2024         LASSERT(fid_seq(&lli->lli_fid) != 0);
2025
2026         if (body->valid & OBD_MD_FLSIZE) {
2027                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2028                     S_ISREG(inode->i_mode)) {
2029                         struct lustre_handle lockh;
2030                         ldlm_mode_t mode;
2031
2032                         /* As it is possible a blocking ast has been processed
2033                          * by this time, we need to check there is an UPDATE
2034                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2035                          * it. */
2036                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2037                                                &lockh, LDLM_FL_CBPENDING,
2038                                                LCK_CR | LCK_CW |
2039                                                LCK_PR | LCK_PW);
2040                         if (mode) {
2041                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2042                                                       LLIF_EPOCH_PENDING |
2043                                                       LLIF_SOM_DIRTY)) {
2044                                         CERROR("%s: inode "DFID" flags %u still"
2045                                                " has size authority! do not "
2046                                                "trust the size from MDS\n",
2047                                                sbi->ll_md_exp->exp_obd->obd_name,
2048                                                PFID(ll_inode2fid(inode)),
2049                                                lli->lli_flags);
2050                                 } else {
2051                                         /* Use old size assignment to avoid
2052                                          * deadlock bz14138 & bz14326 */
2053                                         i_size_write(inode, body->size);
2054                                         spin_lock(&lli->lli_lock);
2055                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2056                                         spin_unlock(&lli->lli_lock);
2057                                 }
2058                                 ldlm_lock_decref(&lockh, mode);
2059                         }
2060                 } else {
2061                         /* Use old size assignment to avoid
2062                          * deadlock bz14138 & bz14326 */
2063                         i_size_write(inode, body->size);
2064
2065                         CDEBUG(D_VFSTRACE,
2066                                "inode="DFID", updating i_size %llu\n",
2067                                PFID(ll_inode2fid(inode)),
2068                                (unsigned long long)body->size);
2069                 }
2070
2071                 if (body->valid & OBD_MD_FLBLOCKS)
2072                         inode->i_blocks = body->blocks;
2073         }
2074
2075         if (body->valid & OBD_MD_FLMDSCAPA) {
2076                 LASSERT(md->mds_capa);
2077                 ll_add_capa(inode, md->mds_capa);
2078         }
2079         if (body->valid & OBD_MD_FLOSSCAPA) {
2080                 LASSERT(md->oss_capa);
2081                 ll_add_capa(inode, md->oss_capa);
2082         }
2083
2084         if (body->valid & OBD_MD_TSTATE) {
2085                 if (body->t_state & MS_RESTORE)
2086                         lli->lli_flags |= LLIF_FILE_RESTORING;
2087         }
2088 }
2089
2090 void ll_read_inode2(struct inode *inode, void *opaque)
2091 {
2092         struct lustre_md *md = opaque;
2093         struct ll_inode_info *lli = ll_i2info(inode);
2094         ENTRY;
2095
2096         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2097                PFID(&lli->lli_fid), inode);
2098
2099         LASSERT(!lli->lli_has_smd);
2100
2101         /* Core attributes from the MDS first.  This is a new inode, and
2102          * the VFS doesn't zero times in the core inode so we have to do
2103          * it ourselves.  They will be overwritten by either MDS or OST
2104          * attributes - we just need to make sure they aren't newer. */
2105         LTIME_S(inode->i_mtime) = 0;
2106         LTIME_S(inode->i_atime) = 0;
2107         LTIME_S(inode->i_ctime) = 0;
2108         inode->i_rdev = 0;
2109         ll_update_inode(inode, md);
2110
2111         /* OIDEBUG(inode); */
2112
2113         /* initializing backing dev info. */
2114         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2115
2116
2117         if (S_ISREG(inode->i_mode)) {
2118                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2119                 inode->i_op = &ll_file_inode_operations;
2120                 inode->i_fop = sbi->ll_fop;
2121                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2122                 EXIT;
2123         } else if (S_ISDIR(inode->i_mode)) {
2124                 inode->i_op = &ll_dir_inode_operations;
2125                 inode->i_fop = &ll_dir_operations;
2126                 EXIT;
2127         } else if (S_ISLNK(inode->i_mode)) {
2128                 inode->i_op = &ll_fast_symlink_inode_operations;
2129                 EXIT;
2130         } else {
2131                 inode->i_op = &ll_special_inode_operations;
2132
2133                 init_special_inode(inode, inode->i_mode,
2134                                    inode->i_rdev);
2135
2136                 EXIT;
2137         }
2138 }
2139
2140 void ll_delete_inode(struct inode *inode)
2141 {
2142         struct cl_inode_info *lli = cl_i2info(inode);
2143         ENTRY;
2144
2145         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2146                 /* discard all dirty pages before truncating them, required by
2147                  * osc_extent implementation at LU-1030. */
2148                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2149                                    CL_FSYNC_DISCARD, 1);
2150
2151         truncate_inode_pages(&inode->i_data, 0);
2152
2153         /* Workaround for LU-118 */
2154         if (inode->i_data.nrpages) {
2155                 spin_lock_irq(&inode->i_data.tree_lock);
2156                 spin_unlock_irq(&inode->i_data.tree_lock);
2157                 LASSERTF(inode->i_data.nrpages == 0,
2158                          "inode="DFID"(%p) nrpages=%lu, see "
2159                          "http://jira.whamcloud.com/browse/LU-118\n",
2160                          PFID(ll_inode2fid(inode)), inode,
2161                          inode->i_data.nrpages);
2162         }
2163         /* Workaround end */
2164
2165 #ifdef HAVE_SBOPS_EVICT_INODE
2166         ll_clear_inode(inode);
2167 #endif
2168         clear_inode(inode);
2169
2170         EXIT;
2171 }
2172
2173 int ll_iocontrol(struct inode *inode, struct file *file,
2174                  unsigned int cmd, unsigned long arg)
2175 {
2176         struct ll_sb_info *sbi = ll_i2sbi(inode);
2177         struct ptlrpc_request *req = NULL;
2178         int rc, flags = 0;
2179         ENTRY;
2180
2181         switch(cmd) {
2182         case FSFILT_IOC_GETFLAGS: {
2183                 struct mdt_body *body;
2184                 struct md_op_data *op_data;
2185
2186                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2187                                              0, 0, LUSTRE_OPC_ANY,
2188                                              NULL);
2189                 if (IS_ERR(op_data))
2190                         RETURN(PTR_ERR(op_data));
2191
2192                 op_data->op_valid = OBD_MD_FLFLAGS;
2193                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2194                 ll_finish_md_op_data(op_data);
2195                 if (rc) {
2196                         CERROR("%s: failure inode "DFID": rc = %d\n",
2197                                sbi->ll_md_exp->exp_obd->obd_name,
2198                                PFID(ll_inode2fid(inode)), rc);
2199                         RETURN(-abs(rc));
2200                 }
2201
2202                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2203
2204                 flags = body->flags;
2205
2206                 ptlrpc_req_finished(req);
2207
2208                 RETURN(put_user(flags, (int *)arg));
2209         }
2210         case FSFILT_IOC_SETFLAGS: {
2211                 struct lov_stripe_md *lsm;
2212                 struct obd_info oinfo = { { { 0 } } };
2213                 struct md_op_data *op_data;
2214
2215                 if (get_user(flags, (int *)arg))
2216                         RETURN(-EFAULT);
2217
2218                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2219                                              LUSTRE_OPC_ANY, NULL);
2220                 if (IS_ERR(op_data))
2221                         RETURN(PTR_ERR(op_data));
2222
2223                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2224                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2225                 rc = md_setattr(sbi->ll_md_exp, op_data,
2226                                 NULL, 0, NULL, 0, &req, NULL);
2227                 ll_finish_md_op_data(op_data);
2228                 ptlrpc_req_finished(req);
2229                 if (rc)
2230                         RETURN(rc);
2231
2232                 inode->i_flags = ll_ext_to_inode_flags(flags);
2233
2234                 lsm = ccc_inode_lsm_get(inode);
2235                 if (!lsm_has_objects(lsm)) {
2236                         ccc_inode_lsm_put(inode, lsm);
2237                         RETURN(0);
2238                 }
2239
2240                 OBDO_ALLOC(oinfo.oi_oa);
2241                 if (!oinfo.oi_oa) {
2242                         ccc_inode_lsm_put(inode, lsm);
2243                         RETURN(-ENOMEM);
2244                 }
2245                 oinfo.oi_md = lsm;
2246                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2247                 oinfo.oi_oa->o_flags = flags;
2248                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2249                                        OBD_MD_FLGROUP;
2250                 oinfo.oi_capa = ll_mdscapa_get(inode);
2251                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2252                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2253                 capa_put(oinfo.oi_capa);
2254                 OBDO_FREE(oinfo.oi_oa);
2255                 ccc_inode_lsm_put(inode, lsm);
2256
2257                 if (rc && rc != -EPERM && rc != -EACCES)
2258                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2259
2260                 RETURN(rc);
2261         }
2262         default:
2263                 RETURN(-ENOSYS);
2264         }
2265
2266         RETURN(0);
2267 }
2268
2269 int ll_flush_ctx(struct inode *inode)
2270 {
2271         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2272
2273         CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2274
2275         obd_set_info_async(NULL, sbi->ll_md_exp,
2276                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2277                            0, NULL, NULL);
2278         obd_set_info_async(NULL, sbi->ll_dt_exp,
2279                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2280                            0, NULL, NULL);
2281         return 0;
2282 }
2283
2284 /* umount -f client means force down, don't save state */
2285 void ll_umount_begin(struct super_block *sb)
2286 {
2287         struct ll_sb_info *sbi = ll_s2sbi(sb);
2288         struct obd_device *obd;
2289         struct obd_ioctl_data *ioc_data;
2290         ENTRY;
2291
2292         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2293                sb->s_count, atomic_read(&sb->s_active));
2294
2295         obd = class_exp2obd(sbi->ll_md_exp);
2296         if (obd == NULL) {
2297                 CERROR("Invalid MDC connection handle "LPX64"\n",
2298                        sbi->ll_md_exp->exp_handle.h_cookie);
2299                 EXIT;
2300                 return;
2301         }
2302         obd->obd_force = 1;
2303
2304         obd = class_exp2obd(sbi->ll_dt_exp);
2305         if (obd == NULL) {
2306                 CERROR("Invalid LOV connection handle "LPX64"\n",
2307                        sbi->ll_dt_exp->exp_handle.h_cookie);
2308                 EXIT;
2309                 return;
2310         }
2311         obd->obd_force = 1;
2312
2313         OBD_ALLOC_PTR(ioc_data);
2314         if (ioc_data) {
2315                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2316                               sizeof *ioc_data, ioc_data, NULL);
2317
2318                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2319                               sizeof *ioc_data, ioc_data, NULL);
2320
2321                 OBD_FREE_PTR(ioc_data);
2322         }
2323
2324         /* Really, we'd like to wait until there are no requests outstanding,
2325          * and then continue.  For now, we just invalidate the requests,
2326          * schedule() and sleep one second if needed, and hope.
2327          */
2328         schedule();
2329         EXIT;
2330 }
2331
2332 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2333 {
2334         struct ll_sb_info *sbi = ll_s2sbi(sb);
2335         char *profilenm = get_profile_name(sb);
2336         int err;
2337         __u32 read_only;
2338
2339         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2340                 read_only = *flags & MS_RDONLY;
2341                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2342                                          sizeof(KEY_READ_ONLY),
2343                                          KEY_READ_ONLY, sizeof(read_only),
2344                                          &read_only, NULL);
2345                 if (err) {
2346                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2347                                       profilenm, read_only ?
2348                                       "read-only" : "read-write", err);
2349                         return err;
2350                 }
2351
2352                 if (read_only)
2353                         sb->s_flags |= MS_RDONLY;
2354                 else
2355                         sb->s_flags &= ~MS_RDONLY;
2356
2357                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2358                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2359                                       read_only ?  "read-only" : "read-write");
2360         }
2361         return 0;
2362 }
2363
2364 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2365                   struct super_block *sb, struct lookup_intent *it)
2366 {
2367         struct ll_sb_info *sbi = NULL;
2368         struct lustre_md md = { 0 };
2369         int rc;
2370         ENTRY;
2371
2372         LASSERT(*inode || sb);
2373         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2374         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2375                               sbi->ll_md_exp, &md);
2376         if (rc)
2377                 RETURN(rc);
2378
2379         if (*inode) {
2380                 ll_update_inode(*inode, &md);
2381         } else {
2382                 LASSERT(sb != NULL);
2383
2384                 /*
2385                  * At this point server returns to client's same fid as client
2386                  * generated for creating. So using ->fid1 is okay here.
2387                  */
2388                 LASSERT(fid_is_sane(&md.body->fid1));
2389
2390                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2391                                              sbi->ll_flags & LL_SBI_32BIT_API),
2392                                  &md);
2393                 if (*inode == NULL || IS_ERR(*inode)) {
2394 #ifdef CONFIG_FS_POSIX_ACL
2395                         if (md.posix_acl) {
2396                                 posix_acl_release(md.posix_acl);
2397                                 md.posix_acl = NULL;
2398                         }
2399 #endif
2400                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2401                         *inode = NULL;
2402                         CERROR("new_inode -fatal: rc %d\n", rc);
2403                         GOTO(out, rc);
2404                 }
2405         }
2406
2407         /* Handling piggyback layout lock.
2408          * Layout lock can be piggybacked by getattr and open request.
2409          * The lsm can be applied to inode only if it comes with a layout lock
2410          * otherwise correct layout may be overwritten, for example:
2411          * 1. proc1: mdt returns a lsm but not granting layout
2412          * 2. layout was changed by another client
2413          * 3. proc2: refresh layout and layout lock granted
2414          * 4. proc1: to apply a stale layout */
2415         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2416                 struct lustre_handle lockh;
2417                 struct ldlm_lock *lock;
2418
2419                 lockh.cookie = it->d.lustre.it_lock_handle;
2420                 lock = ldlm_handle2lock(&lockh);
2421                 LASSERT(lock != NULL);
2422                 if (ldlm_has_layout(lock)) {
2423                         struct cl_object_conf conf;
2424
2425                         memset(&conf, 0, sizeof(conf));
2426                         conf.coc_opc = OBJECT_CONF_SET;
2427                         conf.coc_inode = *inode;
2428                         conf.coc_lock = lock;
2429                         conf.u.coc_md = &md;
2430                         (void)ll_layout_conf(*inode, &conf);
2431                 }
2432                 LDLM_LOCK_PUT(lock);
2433         }
2434
2435 out:
2436         if (md.lsm != NULL)
2437                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2438         md_free_lustre_md(sbi->ll_md_exp, &md);
2439         RETURN(rc);
2440 }
2441
2442 int ll_obd_statfs(struct inode *inode, void *arg)
2443 {
2444         struct ll_sb_info *sbi = NULL;
2445         struct obd_export *exp;
2446         char *buf = NULL;
2447         struct obd_ioctl_data *data = NULL;
2448         __u32 type;
2449         __u32 flags;
2450         int len = 0, rc;
2451
2452         if (!inode || !(sbi = ll_i2sbi(inode)))
2453                 GOTO(out_statfs, rc = -EINVAL);
2454
2455         rc = obd_ioctl_getdata(&buf, &len, arg);
2456         if (rc)
2457                 GOTO(out_statfs, rc);
2458
2459         data = (void*)buf;
2460         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2461             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2462                 GOTO(out_statfs, rc = -EINVAL);
2463
2464         if (data->ioc_inllen1 != sizeof(__u32) ||
2465             data->ioc_inllen2 != sizeof(__u32) ||
2466             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2467             data->ioc_plen2 != sizeof(struct obd_uuid))
2468                 GOTO(out_statfs, rc = -EINVAL);
2469
2470         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2471         if (type & LL_STATFS_LMV)
2472                 exp = sbi->ll_md_exp;
2473         else if (type & LL_STATFS_LOV)
2474                 exp = sbi->ll_dt_exp;
2475         else
2476                 GOTO(out_statfs, rc = -ENODEV);
2477
2478         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2479         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2480         if (rc)
2481                 GOTO(out_statfs, rc);
2482 out_statfs:
2483         if (buf)
2484                 obd_ioctl_freedata(buf, len);
2485         return rc;
2486 }
2487
2488 int ll_process_config(struct lustre_cfg *lcfg)
2489 {
2490         struct super_block *sb;
2491         unsigned long x;
2492         int rc = 0;
2493         char *ptr;
2494
2495         /* The instance name contains the sb: lustre-client-aacfe000 */
2496         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2497         if (!ptr || !*(++ptr))
2498                 return -EINVAL;
2499         if (sscanf(ptr, "%lx", &x) != 1)
2500                 return -EINVAL;
2501         sb = (struct super_block *)x;
2502         /* This better be a real Lustre superblock! */
2503         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2504
2505         /* Note we have not called client_common_fill_super yet, so
2506            proc fns must be able to handle that! */
2507         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2508                                           lcfg, sb);
2509         if (rc > 0)
2510                 rc = 0;
2511         return rc;
2512 }
2513
2514 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2515 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2516                                        struct inode *i1, struct inode *i2,
2517                                        const char *name, int namelen,
2518                                        int mode, __u32 opc, void *data)
2519 {
2520         LASSERT(i1 != NULL);
2521
2522         if (namelen > ll_i2sbi(i1)->ll_namelen)
2523                 return ERR_PTR(-ENAMETOOLONG);
2524
2525         if (op_data == NULL)
2526                 OBD_ALLOC_PTR(op_data);
2527
2528         if (op_data == NULL)
2529                 return ERR_PTR(-ENOMEM);
2530
2531         ll_i2gids(op_data->op_suppgids, i1, i2);
2532         op_data->op_fid1 = *ll_inode2fid(i1);
2533         op_data->op_capa1 = ll_mdscapa_get(i1);
2534         if (S_ISDIR(i1->i_mode))
2535                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2536
2537         if (i2) {
2538                 op_data->op_fid2 = *ll_inode2fid(i2);
2539                 op_data->op_capa2 = ll_mdscapa_get(i2);
2540                 if (S_ISDIR(i2->i_mode))
2541                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2542         } else {
2543                 fid_zero(&op_data->op_fid2);
2544                 op_data->op_capa2 = NULL;
2545         }
2546
2547         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2548                 op_data->op_cli_flags |= CLI_HASH64;
2549
2550         if (ll_need_32bit_api(ll_i2sbi(i1)))
2551                 op_data->op_cli_flags |= CLI_API32;
2552
2553         op_data->op_name = name;
2554         op_data->op_namelen = namelen;
2555         op_data->op_mode = mode;
2556         op_data->op_mod_time = cfs_time_current_sec();
2557         op_data->op_fsuid = current_fsuid();
2558         op_data->op_fsgid = current_fsgid();
2559         op_data->op_cap = cfs_curproc_cap_pack();
2560         op_data->op_bias = 0;
2561         op_data->op_cli_flags = 0;
2562         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2563              filename_is_volatile(name, namelen, NULL))
2564                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2565         op_data->op_mds = 0;
2566         op_data->op_data = data;
2567
2568         /* If the file is being opened after mknod() (normally due to NFS)
2569          * try to use the default stripe data from parent directory for
2570          * allocating OST objects.  Try to pass the parent FID to MDS. */
2571         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2572             !ll_i2info(i2)->lli_has_smd) {
2573                 struct ll_inode_info *lli = ll_i2info(i2);
2574
2575                 spin_lock(&lli->lli_lock);
2576                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2577                         op_data->op_fid1 = lli->lli_pfid;
2578                 spin_unlock(&lli->lli_lock);
2579                 /** We ignore parent's capability temporary. */
2580         }
2581
2582         /* When called by ll_setattr_raw, file is i1. */
2583         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2584                 op_data->op_bias |= MDS_DATA_MODIFIED;
2585
2586         return op_data;
2587 }
2588
2589 void ll_finish_md_op_data(struct md_op_data *op_data)
2590 {
2591         capa_put(op_data->op_capa1);
2592         capa_put(op_data->op_capa2);
2593         OBD_FREE_PTR(op_data);
2594 }
2595
2596 #ifdef HAVE_SUPEROPS_USE_DENTRY
2597 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2598 #else
2599 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2600 #endif
2601 {
2602         struct ll_sb_info *sbi;
2603
2604 #ifdef HAVE_SUPEROPS_USE_DENTRY
2605         LASSERT((seq != NULL) && (dentry != NULL));
2606         sbi = ll_s2sbi(dentry->d_sb);
2607 #else
2608         LASSERT((seq != NULL) && (vfs != NULL));
2609         sbi = ll_s2sbi(vfs->mnt_sb);
2610 #endif
2611
2612         if (sbi->ll_flags & LL_SBI_NOLCK)
2613                 seq_puts(seq, ",nolock");
2614
2615         if (sbi->ll_flags & LL_SBI_FLOCK)
2616                 seq_puts(seq, ",flock");
2617
2618         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2619                 seq_puts(seq, ",localflock");
2620
2621         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2622                 seq_puts(seq, ",user_xattr");
2623
2624         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2625                 seq_puts(seq, ",lazystatfs");
2626
2627         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2628                 seq_puts(seq, ",user_fid2path");
2629
2630         RETURN(0);
2631 }
2632
2633 /**
2634  * Get obd name by cmd, and copy out to user space
2635  */
2636 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2637 {
2638         struct ll_sb_info *sbi = ll_i2sbi(inode);
2639         struct obd_device *obd;
2640         ENTRY;
2641
2642         if (cmd == OBD_IOC_GETDTNAME)
2643                 obd = class_exp2obd(sbi->ll_dt_exp);
2644         else if (cmd == OBD_IOC_GETMDNAME)
2645                 obd = class_exp2obd(sbi->ll_md_exp);
2646         else
2647                 RETURN(-EINVAL);
2648
2649         if (!obd)
2650                 RETURN(-ENOENT);
2651
2652         if (copy_to_user((void *)arg, obd->obd_name,
2653                          strlen(obd->obd_name) + 1))
2654                 RETURN(-EFAULT);
2655
2656         RETURN(0);
2657 }
2658
2659 /**
2660  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2661  * fsname will be returned in this buffer; otherwise, a static buffer will be
2662  * used to store the fsname and returned to caller.
2663  */
2664 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2665 {
2666         static char fsname_static[MTI_NAME_MAXLEN];
2667         struct lustre_sb_info *lsi = s2lsi(sb);
2668         char *ptr;
2669         int len;
2670
2671         if (buf == NULL) {
2672                 /* this means the caller wants to use static buffer
2673                  * and it doesn't care about race. Usually this is
2674                  * in error reporting path */
2675                 buf = fsname_static;
2676                 buflen = sizeof(fsname_static);
2677         }
2678
2679         len = strlen(lsi->lsi_lmd->lmd_profile);
2680         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2681         if (ptr && (strcmp(ptr, "-client") == 0))
2682                 len -= 7;
2683
2684         if (unlikely(len >= buflen))
2685                 len = buflen - 1;
2686         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2687         buf[len] = '\0';
2688
2689         return buf;
2690 }
2691
2692 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2693 {
2694         char *path = NULL;
2695
2696         struct path p;
2697
2698         p.dentry = dentry;
2699         p.mnt = current->fs->root.mnt;
2700         path_get(&p);
2701         path = d_path(&p, buf, bufsize);
2702         path_put(&p);
2703         return path;
2704 }
2705
2706 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2707 {
2708         char *buf, *path = NULL;
2709         struct dentry *dentry = NULL;
2710         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2711
2712         /* this can be called inside spin lock so use GFP_ATOMIC. */
2713         buf = (char *)__get_free_page(GFP_ATOMIC);
2714         if (buf != NULL) {
2715                 dentry = d_find_alias(page->mapping->host);
2716                 if (dentry != NULL)
2717                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2718         }
2719
2720         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2721               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2722               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2723               PFID(&obj->cob_header.coh_lu.loh_fid),
2724               (path && !IS_ERR(path)) ? path : "", ioret);
2725
2726         if (dentry != NULL)
2727                 dput(dentry);
2728
2729         if (buf != NULL)
2730                 free_page((unsigned long)buf);
2731 }