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