lustre/llite/llite_lib.c

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