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, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  97 #ifdef HAVE_BGL_SUPPORT
  98                 lru_page_max = pages / 4;
  99 #else
 100                 lru_page_max = pages / 2;
 101 #endif
 102         } else {
 103                 lru_page_max = (pages / 4) * 3;
 104         }
 105
 106         /* initialize lru data */
 107         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 108         sbi->ll_cache.ccc_lru_max = lru_page_max;
 109         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 110         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 111         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 112
 113         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 114                                            SBI_DEFAULT_READAHEAD_MAX);
 115         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 117                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 118         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 119         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 120
 121         ll_generate_random_uuid(uuid);
 122         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 123         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 124
 125         spin_lock(&ll_sb_lock);
 126         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 127         spin_unlock(&ll_sb_lock);
 128
 129         sbi->ll_flags |= LL_SBI_VERBOSE;
 130 #ifdef ENABLE_CHECKSUM
 131         sbi->ll_flags |= LL_SBI_CHECKSUM;
 132 #endif
 133
 134 #ifdef HAVE_LRU_RESIZE_SUPPORT
 135         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 136 #endif
 137
 138         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 139                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 140                                pp_r_hist.oh_lock);
 141                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 142                                pp_w_hist.oh_lock);
 143         }
 144
 145         /* metadata statahead is enabled by default */
 146         sbi->ll_sa_max = LL_SA_RPC_DEF;
 147         cfs_atomic_set(&sbi->ll_sa_total, 0);
 148         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 149         cfs_atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151
 152         RETURN(sbi);
 153 }
 154
 155 void ll_free_sbi(struct super_block *sb)
 156 {
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         ENTRY;
 159
 160         if (sbi != NULL) {
 161                 spin_lock(&ll_sb_lock);
 162                 cfs_list_del(&sbi->ll_list);
 163                 spin_unlock(&ll_sb_lock);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static struct dentry_operations ll_d_root_ops = {
 170         .d_compare = ll_dcompare,
 171         .d_revalidate = ll_revalidate_nd,
 172 };
 173
 174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 175                                     struct vfsmount *mnt)
 176 {
 177         struct inode *root = 0;
 178         struct ll_sb_info *sbi = ll_s2sbi(sb);
 179         struct obd_device *obd;
 180         struct obd_capa *oc = NULL;
 181         struct obd_statfs *osfs = NULL;
 182         struct ptlrpc_request *request = NULL;
 183         struct obd_connect_data *data = NULL;
 184         struct obd_uuid *uuid;
 185         struct md_op_data *op_data;
 186         struct lustre_md lmd;
 187         obd_valid valid;
 188         int size, err, checksum;
 189         ENTRY;
 190
 191         obd = class_name2obd(md);
 192         if (!obd) {
 193                 CERROR("MD %s: not setup or attached\n", md);
 194                 RETURN(-EINVAL);
 195         }
 196
 197         OBD_ALLOC_PTR(data);
 198         if (data == NULL)
 199                 RETURN(-ENOMEM);
 200
 201         OBD_ALLOC_PTR(osfs);
 202         if (osfs == NULL) {
 203                 OBD_FREE_PTR(data);
 204                 RETURN(-ENOMEM);
 205         }
 206
 207         if (proc_lustre_fs_root) {
 208                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 209                                                   dt, md);
 210                 if (err < 0)
 211                         CERROR("could not register mount in /proc/fs/lustre\n");
 212         }
 213
 214         /* indicate the features supported by this client */
 215         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 216                                   OBD_CONNECT_ATTRFID  |
 217                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 218                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 219                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 220                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 221                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 222                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 223                                   OBD_CONNECT_EINPROGRESS |
 224                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 225                                   OBD_CONNECT_LAYOUTLOCK;
 226
 227         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 228                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 229
 230 #ifdef HAVE_LRU_RESIZE_SUPPORT
 231         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 232                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 233 #endif
 234 #ifdef CONFIG_FS_POSIX_ACL
 235         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 236 #endif
 237
 238         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 239                 /* flag mdc connection as lightweight, only used for test
 240                  * purpose, use with care */
 241                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 242
 243         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 244         data->ocd_version = LUSTRE_VERSION_CODE;
 245
 246         if (sb->s_flags & MS_RDONLY)
 247                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 248         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 249                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 250
 251 #ifdef HAVE_MS_FLOCK_LOCK
 252         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 253         sb->s_flags |= MS_FLOCK_LOCK;
 254 #endif
 255 #ifdef MS_HAS_NEW_AOPS
 256         sb->s_flags |= MS_HAS_NEW_AOPS;
 257 #endif
 258
 259         if (sbi->ll_flags & LL_SBI_FLOCK)
 260                 sbi->ll_fop = &ll_file_operations_flock;
 261         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 262                 sbi->ll_fop = &ll_file_operations;
 263         else
 264                 sbi->ll_fop = &ll_file_operations_noflock;
 265
 266         /* real client */
 267         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 268         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 269                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 270
 271         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 272
 273         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 274         if (err == -EBUSY) {
 275                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 276                                    "recovery, of which this client is not a "
 277                                    "part. Please wait for recovery to complete,"
 278                                    " abort, or time out.\n", md);
 279                 GOTO(out, err);
 280         } else if (err) {
 281                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 282                 GOTO(out, err);
 283         }
 284
 285         err = obd_fid_init(sbi->ll_md_exp);
 286         if (err) {
 287                 CERROR("Can't init metadata layer FID infrastructure, "
 288                        "rc %d\n", err);
 289                 GOTO(out_md, err);
 290         }
 291
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 294         if (err)
 295                 GOTO(out_md_fid, err);
 296
 297         /* This needs to be after statfs to ensure connect has finished.
 298          * Note that "data" does NOT contain the valid connect reply.
 299          * If connecting to a 1.8 server there will be no LMV device, so
 300          * we can access the MDC export directly and exp_connect_flags will
 301          * be non-zero, but if accessing an upgraded 2.1 server it will
 302          * have the correct flags filled in.
 303          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 304         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 305         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 306             valid != CLIENT_CONNECT_MDT_REQD) {
 307                 char *buf;
 308
 309                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 310                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 311                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 312                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 313                                    "feature(s) needed for correct operation "
 314                                    "of this client (%s). Please upgrade "
 315                                    "server or downgrade client.\n",
 316                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 317                 OBD_FREE(buf, CFS_PAGE_SIZE);
 318                 GOTO(out_md, err = -EPROTO);
 319         }
 320
 321         size = sizeof(*data);
 322         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 323                            KEY_CONN_DATA,  &size, data, NULL);
 324         if (err) {
 325                 CERROR("Get connect data failed: %d \n", err);
 326                 GOTO(out_md, err);
 327         }
 328
 329         LASSERT(osfs->os_bsize);
 330         sb->s_blocksize = osfs->os_bsize;
 331         sb->s_blocksize_bits = log2(osfs->os_bsize);
 332         sb->s_magic = LL_SUPER_MAGIC;
 333         sb->s_maxbytes = MAX_LFS_FILESIZE;
 334         sbi->ll_namelen = osfs->os_namelen;
 335         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 336
 337         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 338             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 339                 LCONSOLE_INFO("Disabling user_xattr feature because "
 340                               "it is not supported on the server\n");
 341                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 345 #ifdef MS_POSIXACL
 346                 sb->s_flags |= MS_POSIXACL;
 347 #endif
 348                 sbi->ll_flags |= LL_SBI_ACL;
 349         } else {
 350                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 351 #ifdef MS_POSIXACL
 352                 sb->s_flags &= ~MS_POSIXACL;
 353 #endif
 354                 sbi->ll_flags &= ~LL_SBI_ACL;
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 358                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 359                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 360                         LCONSOLE_INFO("client is set as remote by default.\n");
 361                 }
 362         } else {
 363                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 364                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 365                         LCONSOLE_INFO("client claims to be remote, but server "
 366                                       "rejected, forced to be local.\n");
 367                 }
 368         }
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 371                 LCONSOLE_INFO("client enabled MDS capability!\n");
 372                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 373         }
 374
 375         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 376                 LCONSOLE_INFO("client enabled OSS capability!\n");
 377                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 378         }
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 381                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 382
 383         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 384                 sbi->ll_md_brw_size = data->ocd_brw_size;
 385         else
 386                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 387
 388         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 389                 LCONSOLE_INFO("Layout lock feature supported.\n");
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391         }
 392
 393         obd = class_name2obd(dt);
 394         if (!obd) {
 395                 CERROR("DT %s: not setup or attached\n", dt);
 396                 GOTO(out_md_fid, err = -ENODEV);
 397         }
 398
 399         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 400                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 401                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 402                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 403                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 404                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 405                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 406                                   OBD_CONNECT_MAXBYTES |
 407                                   OBD_CONNECT_EINPROGRESS |
 408                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 409                                   OBD_CONNECT_LAYOUTLOCK;
 410
 411         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 412                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 413
 414         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 415                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 416                  * disabled by default, because it can still be enabled on the
 417                  * fly via /proc. As a consequence, we still need to come to an
 418                  * agreement on the supported algorithms at connect time */
 419                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 420
 421                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 422                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 423                 else
 424                         data->ocd_cksum_types = cksum_types_supported_client();
 425         }
 426
 427 #ifdef HAVE_LRU_RESIZE_SUPPORT
 428         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 429 #endif
 430         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 431                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 432
 433         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 434                "ocd_grant: %d\n", data->ocd_connect_flags,
 435                data->ocd_version, data->ocd_grant);
 436
 437         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 438         obd->obd_upcall.onu_upcall = cl_ocd_update;
 439
 440         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 441
 442         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 443         if (err == -EBUSY) {
 444                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 445                                    "recovery, of which this client is not a "
 446                                    "part.  Please wait for recovery to "
 447                                    "complete, abort, or time out.\n", dt);
 448                 GOTO(out_md_fid, err);
 449         } else if (err) {
 450                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 451                 GOTO(out_md_fid, err);
 452         }
 453
 454         err = obd_fid_init(sbi->ll_dt_exp);
 455         if (err) {
 456                 CERROR("Can't init data layer FID infrastructure, "
 457                        "rc %d\n", err);
 458                 GOTO(out_dt, err);
 459         }
 460
 461         mutex_lock(&sbi->ll_lco.lco_lock);
 462         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 463         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 464         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 465         mutex_unlock(&sbi->ll_lco.lco_lock);
 466
 467         fid_zero(&sbi->ll_root_fid);
 468         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 469         if (err) {
 470                 CERROR("cannot mds_connect: rc = %d\n", err);
 471                 GOTO(out_lock_cn_cb, err);
 472         }
 473         if (!fid_is_sane(&sbi->ll_root_fid)) {
 474                 CERROR("Invalid root fid during mount\n");
 475                 GOTO(out_lock_cn_cb, err = -EINVAL);
 476         }
 477         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 478
 479         sb->s_op = &lustre_super_operations;
 480 #if THREAD_SIZE >= 8192 /*b=17630*/
 481         sb->s_export_op = &lustre_export_operations;
 482 #endif
 483
 484         /* make root inode
 485          * XXX: move this to after cbd setup? */
 486         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 487         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 488                 valid |= OBD_MD_FLRMTPERM;
 489         else if (sbi->ll_flags & LL_SBI_ACL)
 490                 valid |= OBD_MD_FLACL;
 491
 492         OBD_ALLOC_PTR(op_data);
 493         if (op_data == NULL)
 494                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 495
 496         op_data->op_fid1 = sbi->ll_root_fid;
 497         op_data->op_mode = 0;
 498         op_data->op_capa1 = oc;
 499         op_data->op_valid = valid;
 500
 501         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 502         if (oc)
 503                 capa_put(oc);
 504         OBD_FREE_PTR(op_data);
 505         if (err) {
 506                 CERROR("md_getattr failed for root: rc = %d\n", err);
 507                 GOTO(out_lock_cn_cb, err);
 508         }
 509         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 510                                sbi->ll_md_exp, &lmd);
 511         if (err) {
 512                 CERROR("failed to understand root inode md: rc = %d\n", err);
 513                 ptlrpc_req_finished (request);
 514                 GOTO(out_lock_cn_cb, err);
 515         }
 516
 517         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 518         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 519         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 520         ptlrpc_req_finished(request);
 521
 522         if (root == NULL || IS_ERR(root)) {
 523                 if (lmd.lsm)
 524                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 525 #ifdef CONFIG_FS_POSIX_ACL
 526                 if (lmd.posix_acl) {
 527                         posix_acl_release(lmd.posix_acl);
 528                         lmd.posix_acl = NULL;
 529                 }
 530 #endif
 531                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 532                 root = NULL;
 533                 CERROR("lustre_lite: bad iget4 for root\n");
 534                 GOTO(out_root, err);
 535         }
 536
 537         err = ll_close_thread_start(&sbi->ll_lcq);
 538         if (err) {
 539                 CERROR("cannot start close thread: rc %d\n", err);
 540                 GOTO(out_root, err);
 541         }
 542
 543 #ifdef CONFIG_FS_POSIX_ACL
 544         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 545                 rct_init(&sbi->ll_rct);
 546                 et_init(&sbi->ll_et);
 547         }
 548 #endif
 549
 550         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 551         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 552                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 553                                  NULL);
 554         cl_sb_init(sb);
 555
 556         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 557                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 558                                  &sbi->ll_cache, NULL);
 559
 560         sb->s_root = d_make_root(root);
 561         if (sb->s_root == NULL) {
 562                 CERROR("%s: can't make root dentry\n",
 563                         ll_get_fsname(sb, NULL, 0));
 564                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 565         }
 566
 567 #ifdef HAVE_DCACHE_LOCK
 568         sb->s_root->d_op = &ll_d_root_ops;
 569 #else
 570         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 571         d_set_d_op(sb->s_root, &ll_d_root_ops);
 572         sb->s_d_op = &ll_d_ops;
 573 #endif
 574
 575         sbi->ll_sdev_orig = sb->s_dev;
 576
 577         /* We set sb->s_dev equal on all lustre clients in order to support
 578          * NFS export clustering.  NFSD requires that the FSID be the same
 579          * on all clients. */
 580         /* s_dev is also used in lt_compare() to compare two fs, but that is
 581          * only a node-local comparison. */
 582         uuid = obd_get_uuid(sbi->ll_md_exp);
 583         if (uuid != NULL)
 584                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 585
 586         if (data != NULL)
 587                 OBD_FREE_PTR(data);
 588         if (osfs != NULL)
 589                 OBD_FREE_PTR(osfs);
 590
 591         RETURN(err);
 592 out_root:
 593         if (root)
 594                 iput(root);
 595 out_lock_cn_cb:
 596         obd_fid_fini(sbi->ll_dt_exp);
 597 out_dt:
 598         obd_disconnect(sbi->ll_dt_exp);
 599         sbi->ll_dt_exp = NULL;
 600         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 601         obd_zombie_barrier();
 602 out_md_fid:
 603         obd_fid_fini(sbi->ll_md_exp);
 604 out_md:
 605         obd_disconnect(sbi->ll_md_exp);
 606         sbi->ll_md_exp = NULL;
 607 out:
 608         if (data != NULL)
 609                 OBD_FREE_PTR(data);
 610         if (osfs != NULL)
 611                 OBD_FREE_PTR(osfs);
 612         lprocfs_unregister_mountpoint(sbi);
 613         return err;
 614 }
 615
 616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 617 {
 618         int size, rc;
 619
 620         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 621         size = sizeof(int);
 622         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 623                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 624         if (rc)
 625                 CERROR("Get max mdsize error rc %d \n", rc);
 626
 627         RETURN(rc);
 628 }
 629
 630 void ll_dump_inode(struct inode *inode)
 631 {
 632         struct ll_d_hlist_node *tmp;
 633         int dentry_count = 0;
 634
 635         LASSERT(inode != NULL);
 636
 637         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 638                 dentry_count++;
 639
 640         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 641                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 642                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 643 }
 644
 645 void lustre_dump_dentry(struct dentry *dentry, int recur)
 646 {
 647         struct list_head *tmp;
 648         int subdirs = 0;
 649
 650         LASSERT(dentry != NULL);
 651
 652         list_for_each(tmp, &dentry->d_subdirs)
 653                 subdirs++;
 654
 655         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 656                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 657                dentry->d_name.len, dentry->d_name.name,
 658                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 659                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 660                dentry->d_flags, dentry->d_fsdata, subdirs);
 661         if (dentry->d_inode != NULL)
 662                 ll_dump_inode(dentry->d_inode);
 663
 664         if (recur == 0)
 665                 return;
 666
 667         list_for_each(tmp, &dentry->d_subdirs) {
 668                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 669                 lustre_dump_dentry(d, recur - 1);
 670         }
 671 }
 672
 673 void client_common_put_super(struct super_block *sb)
 674 {
 675         struct ll_sb_info *sbi = ll_s2sbi(sb);
 676         ENTRY;
 677
 678 #ifdef CONFIG_FS_POSIX_ACL
 679         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 680                 et_fini(&sbi->ll_et);
 681                 rct_fini(&sbi->ll_rct);
 682         }
 683 #endif
 684
 685         ll_close_thread_shutdown(sbi->ll_lcq);
 686
 687         cl_sb_fini(sb);
 688
 689         cfs_list_del(&sbi->ll_conn_chain);
 690
 691         obd_fid_fini(sbi->ll_dt_exp);
 692         obd_disconnect(sbi->ll_dt_exp);
 693         sbi->ll_dt_exp = NULL;
 694         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 695          * see LU-2543. */
 696         obd_zombie_barrier();
 697
 698         lprocfs_unregister_mountpoint(sbi);
 699
 700         obd_fid_fini(sbi->ll_md_exp);
 701         obd_disconnect(sbi->ll_md_exp);
 702         sbi->ll_md_exp = NULL;
 703
 704         EXIT;
 705 }
 706
 707 void ll_kill_super(struct super_block *sb)
 708 {
 709         struct ll_sb_info *sbi;
 710
 711         ENTRY;
 712
 713         /* not init sb ?*/
 714         if (!(sb->s_flags & MS_ACTIVE))
 715                 return;
 716
 717         sbi = ll_s2sbi(sb);
 718         /* we need restore s_dev from changed for clustred NFS before put_super
 719          * because new kernels have cached s_dev and change sb->s_dev in
 720          * put_super not affected real removing devices */
 721         if (sbi)
 722                 sb->s_dev = sbi->ll_sdev_orig;
 723         EXIT;
 724 }
 725
 726 char *ll_read_opt(const char *opt, char *data)
 727 {
 728         char *value;
 729         char *retval;
 730         ENTRY;
 731
 732         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 733         if (strncmp(opt, data, strlen(opt)))
 734                 RETURN(NULL);
 735         if ((value = strchr(data, '=')) == NULL)
 736                 RETURN(NULL);
 737
 738         value++;
 739         OBD_ALLOC(retval, strlen(value) + 1);
 740         if (!retval) {
 741                 CERROR("out of memory!\n");
 742                 RETURN(NULL);
 743         }
 744
 745         memcpy(retval, value, strlen(value)+1);
 746         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 747         RETURN(retval);
 748 }
 749
 750 static inline int ll_set_opt(const char *opt, char *data, int fl)
 751 {
 752         if (strncmp(opt, data, strlen(opt)) != 0)
 753                 return(0);
 754         else
 755                 return(fl);
 756 }
 757
 758 /* non-client-specific mount options are parsed in lmd_parse */
 759 static int ll_options(char *options, int *flags)
 760 {
 761         int tmp;
 762         char *s1 = options, *s2;
 763         ENTRY;
 764
 765         if (!options)
 766                 RETURN(0);
 767
 768         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 769
 770         while (*s1) {
 771                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 772                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 773                 if (tmp) {
 774                         *flags |= tmp;
 775                         goto next;
 776                 }
 777                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 778                 if (tmp) {
 779                         *flags |= tmp;
 780                         goto next;
 781                 }
 782                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 783                 if (tmp) {
 784                         *flags |= tmp;
 785                         goto next;
 786                 }
 787                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 788                 if (tmp) {
 789                         *flags &= ~tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 793                 if (tmp) {
 794                         *flags |= tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 798                 if (tmp) {
 799                         *flags &= ~tmp;
 800                         goto next;
 801                 }
 802 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 803                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 804                 if (tmp) {
 805                         /* Ignore deprecated mount option.  The client will
 806                          * always try to mount with ACL support, whether this
 807                          * is used depends on whether server supports it. */
 808                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 809                                                   "mount option 'acl'.\n");
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 813                 if (tmp) {
 814                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 815                                                   "mount option 'noacl'.\n");
 816                         goto next;
 817                 }
 818 #else
 819 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 820 #endif
 821                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836
 837                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 873                 if (tmp) {
 874                         *flags |= tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 888                                    s1);
 889                 RETURN(-EINVAL);
 890
 891 next:
 892                 /* Find next opt */
 893                 s2 = strchr(s1, ',');
 894                 if (s2 == NULL)
 895                         break;
 896                 s1 = s2 + 1;
 897         }
 898         RETURN(0);
 899 }
 900
 901 void ll_lli_init(struct ll_inode_info *lli)
 902 {
 903         lli->lli_inode_magic = LLI_INODE_MAGIC;
 904         lli->lli_flags = 0;
 905         lli->lli_ioepoch = 0;
 906         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 907         spin_lock_init(&lli->lli_lock);
 908         lli->lli_posix_acl = NULL;
 909         lli->lli_remote_perms = NULL;
 910         mutex_init(&lli->lli_rmtperm_mutex);
 911         /* Do not set lli_fid, it has been initialized already. */
 912         fid_zero(&lli->lli_pfid);
 913         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 914         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 915         cfs_atomic_set(&lli->lli_open_count, 0);
 916         lli->lli_mds_capa = NULL;
 917         lli->lli_rmtperm_time = 0;
 918         lli->lli_pending_och = NULL;
 919         lli->lli_mds_read_och = NULL;
 920         lli->lli_mds_write_och = NULL;
 921         lli->lli_mds_exec_och = NULL;
 922         lli->lli_open_fd_read_count = 0;
 923         lli->lli_open_fd_write_count = 0;
 924         lli->lli_open_fd_exec_count = 0;
 925         mutex_init(&lli->lli_och_mutex);
 926         spin_lock_init(&lli->lli_agl_lock);
 927         lli->lli_has_smd = false;
 928         lli->lli_layout_gen = LL_LAYOUT_GEN_ZERO;
 929         lli->lli_clob = NULL;
 930
 931         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 932         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 933                 mutex_init(&lli->lli_readdir_mutex);
 934                 lli->lli_opendir_key = NULL;
 935                 lli->lli_sai = NULL;
 936                 lli->lli_def_acl = NULL;
 937                 spin_lock_init(&lli->lli_sa_lock);
 938                 lli->lli_opendir_pid = 0;
 939         } else {
 940                 sema_init(&lli->lli_size_sem, 1);
 941                 lli->lli_size_sem_owner = NULL;
 942                 lli->lli_symlink_name = NULL;
 943                 init_rwsem(&lli->lli_trunc_sem);
 944                 mutex_init(&lli->lli_write_mutex);
 945                 init_rwsem(&lli->lli_glimpse_sem);
 946                 lli->lli_glimpse_time = 0;
 947                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 948                 lli->lli_agl_index = 0;
 949                 lli->lli_async_rc = 0;
 950         }
 951         mutex_init(&lli->lli_layout_mutex);
 952 }
 953
 954 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 955 {
 956 #ifdef HAVE_BDI_REGISTER
 957         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 958
 959 #ifdef HAVE_BDI_NAME
 960         bdi->name = "lustre";
 961 #endif
 962         return bdi_register(bdi, NULL, "lustre-%d",
 963                             atomic_inc_return(&ll_bdi_num));
 964 #else
 965         return 0;
 966 #endif
 967 }
 968
 969 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 970 {
 971         struct lustre_profile *lprof = NULL;
 972         struct lustre_sb_info *lsi = s2lsi(sb);
 973         struct ll_sb_info *sbi;
 974         char  *dt = NULL, *md = NULL;
 975         char  *profilenm = get_profile_name(sb);
 976         struct config_llog_instance *cfg;
 977         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 978         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 979         int    err;
 980         ENTRY;
 981
 982         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 983
 984         OBD_ALLOC_PTR(cfg);
 985         if (cfg == NULL)
 986                 RETURN(-ENOMEM);
 987
 988         cfs_module_get();
 989
 990         /* client additional sb info */
 991         lsi->lsi_llsbi = sbi = ll_init_sbi();
 992         if (!sbi) {
 993                 cfs_module_put(THIS_MODULE);
 994                 OBD_FREE_PTR(cfg);
 995                 RETURN(-ENOMEM);
 996         }
 997
 998         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 999         if (err)
1000                 GOTO(out_free, err);
1001
1002         err = ll_bdi_init(&lsi->lsi_bdi);
1003         if (err)
1004                 GOTO(out_free, err);
1005         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1006         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1007         err = ll_bdi_register(&lsi->lsi_bdi);
1008         if (err)
1009                 GOTO(out_free, err);
1010
1011 #ifdef HAVE_SB_BDI
1012         sb->s_bdi = &lsi->lsi_bdi;
1013 #endif
1014
1015         /* Generate a string unique to this super, in case some joker tries
1016            to mount the same fs at two mount points.
1017            Use the address of the super itself.*/
1018         cfg->cfg_instance = sb;
1019         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1020         cfg->cfg_callback = class_config_llog_handler;
1021         /* set up client obds */
1022         err = lustre_process_log(sb, profilenm, cfg);
1023         if (err < 0) {
1024                 CERROR("Unable to process log: %d\n", err);
1025                 GOTO(out_free, err);
1026         }
1027
1028         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1029         lprof = class_get_profile(profilenm);
1030         if (lprof == NULL) {
1031                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1032                                    " read from the MGS.  Does that filesystem "
1033                                    "exist?\n", profilenm);
1034                 GOTO(out_free, err = -EINVAL);
1035         }
1036         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1037                lprof->lp_md, lprof->lp_dt);
1038
1039         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1040         if (!dt)
1041                 GOTO(out_free, err = -ENOMEM);
1042         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1043
1044         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1045         if (!md)
1046                 GOTO(out_free, err = -ENOMEM);
1047         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1048
1049         /* connections, registrations, sb setup */
1050         err = client_common_fill_super(sb, md, dt, mnt);
1051
1052 out_free:
1053         if (md)
1054                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1055         if (dt)
1056                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1057         if (err)
1058                 ll_put_super(sb);
1059         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1060                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1061
1062         OBD_FREE_PTR(cfg);
1063         RETURN(err);
1064 } /* ll_fill_super */
1065
1066
1067 void lu_context_keys_dump(void);
1068
1069 void ll_put_super(struct super_block *sb)
1070 {
1071         struct config_llog_instance cfg;
1072         struct obd_device *obd;
1073         struct lustre_sb_info *lsi = s2lsi(sb);
1074         struct ll_sb_info *sbi = ll_s2sbi(sb);
1075         char *profilenm = get_profile_name(sb);
1076         int force = 1, next;
1077         ENTRY;
1078
1079         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1080
1081         ll_print_capa_stat(sbi);
1082
1083         cfg.cfg_instance = sb;
1084         lustre_end_log(sb, profilenm, &cfg);
1085
1086         if (sbi->ll_md_exp) {
1087                 obd = class_exp2obd(sbi->ll_md_exp);
1088                 if (obd)
1089                         force = obd->obd_force;
1090         }
1091
1092         /* We need to set force before the lov_disconnect in
1093            lustre_common_put_super, since l_d cleans up osc's as well. */
1094         if (force) {
1095                 next = 0;
1096                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1097                                                      &next)) != NULL) {
1098                         obd->obd_force = force;
1099                 }
1100         }
1101
1102         if (sbi->ll_lcq) {
1103                 /* Only if client_common_fill_super succeeded */
1104                 client_common_put_super(sb);
1105         }
1106
1107         next = 0;
1108         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1109                 class_manual_cleanup(obd);
1110         }
1111
1112         if (sbi->ll_flags & LL_SBI_VERBOSE)
1113                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1114
1115         if (profilenm)
1116                 class_del_profile(profilenm);
1117
1118         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1119                 ll_bdi_destroy(&lsi->lsi_bdi);
1120                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1121         }
1122
1123         ll_free_sbi(sb);
1124         lsi->lsi_llsbi = NULL;
1125
1126         lustre_common_put_super(sb);
1127
1128         cl_env_cache_purge(~0);
1129
1130         cfs_module_put(THIS_MODULE);
1131
1132         EXIT;
1133 } /* client_put_super */
1134
1135 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1136 {
1137         struct inode *inode = NULL;
1138         /* NOTE: we depend on atomic igrab() -bzzz */
1139         lock_res_and_lock(lock);
1140         if (lock->l_ast_data) {
1141                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1142                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1143                         inode = igrab(lock->l_ast_data);
1144                 } else {
1145                         inode = lock->l_ast_data;
1146                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1147                                          D_WARNING, lock, "l_ast_data %p is "
1148                                          "bogus: magic %08x", lock->l_ast_data,
1149                                          lli->lli_inode_magic);
1150                         inode = NULL;
1151                 }
1152         }
1153         unlock_res_and_lock(lock);
1154         return inode;
1155 }
1156
1157 static int null_if_equal(struct ldlm_lock *lock, void *data)
1158 {
1159         if (data == lock->l_ast_data) {
1160                 lock->l_ast_data = NULL;
1161
1162                 if (lock->l_req_mode != lock->l_granted_mode)
1163                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1164         }
1165
1166         return LDLM_ITER_CONTINUE;
1167 }
1168
1169 void ll_clear_inode(struct inode *inode)
1170 {
1171         struct ll_inode_info *lli = ll_i2info(inode);
1172         struct ll_sb_info *sbi = ll_i2sbi(inode);
1173         ENTRY;
1174
1175         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1176                inode->i_generation, inode);
1177
1178         if (S_ISDIR(inode->i_mode)) {
1179                 /* these should have been cleared in ll_file_release */
1180                 LASSERT(lli->lli_opendir_key == NULL);
1181                 LASSERT(lli->lli_sai == NULL);
1182                 LASSERT(lli->lli_opendir_pid == 0);
1183         }
1184
1185         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1186         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1187                          null_if_equal, inode);
1188
1189         LASSERT(!lli->lli_open_fd_write_count);
1190         LASSERT(!lli->lli_open_fd_read_count);
1191         LASSERT(!lli->lli_open_fd_exec_count);
1192
1193         if (lli->lli_mds_write_och)
1194                 ll_md_real_close(inode, FMODE_WRITE);
1195         if (lli->lli_mds_exec_och)
1196                 ll_md_real_close(inode, FMODE_EXEC);
1197         if (lli->lli_mds_read_och)
1198                 ll_md_real_close(inode, FMODE_READ);
1199
1200         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1201                 OBD_FREE(lli->lli_symlink_name,
1202                          strlen(lli->lli_symlink_name) + 1);
1203                 lli->lli_symlink_name = NULL;
1204         }
1205
1206         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1207                 LASSERT(lli->lli_posix_acl == NULL);
1208                 if (lli->lli_remote_perms) {
1209                         free_rmtperm_hash(lli->lli_remote_perms);
1210                         lli->lli_remote_perms = NULL;
1211                 }
1212         }
1213 #ifdef CONFIG_FS_POSIX_ACL
1214         else if (lli->lli_posix_acl) {
1215                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1216                 LASSERT(lli->lli_remote_perms == NULL);
1217                 posix_acl_release(lli->lli_posix_acl);
1218                 lli->lli_posix_acl = NULL;
1219         }
1220 #endif
1221         lli->lli_inode_magic = LLI_INODE_DEAD;
1222
1223         ll_clear_inode_capas(inode);
1224         if (!S_ISDIR(inode->i_mode))
1225                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1226
1227         /*
1228          * XXX This has to be done before lsm is freed below, because
1229          * cl_object still uses inode lsm.
1230          */
1231         cl_inode_fini(inode);
1232         lli->lli_has_smd = false;
1233
1234         EXIT;
1235 }
1236
1237 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1238                   struct md_open_data **mod)
1239 {
1240         struct lustre_md md;
1241         struct inode *inode = dentry->d_inode;
1242         struct ll_sb_info *sbi = ll_i2sbi(inode);
1243         struct ptlrpc_request *request = NULL;
1244         int rc, ia_valid;
1245         ENTRY;
1246
1247         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1248                                      LUSTRE_OPC_ANY, NULL);
1249         if (IS_ERR(op_data))
1250                 RETURN(PTR_ERR(op_data));
1251
1252         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1253                         &request, mod);
1254         if (rc) {
1255                 ptlrpc_req_finished(request);
1256                 if (rc == -ENOENT) {
1257                         clear_nlink(inode);
1258                         /* Unlinked special device node? Or just a race?
1259                          * Pretend we done everything. */
1260                         if (!S_ISREG(inode->i_mode) &&
1261                             !S_ISDIR(inode->i_mode)) {
1262                                 ia_valid = op_data->op_attr.ia_valid;
1263                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1264                                 rc = simple_setattr(dentry, &op_data->op_attr);
1265                                 op_data->op_attr.ia_valid = ia_valid;
1266                         }
1267                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1268                         CERROR("md_setattr fails: rc = %d\n", rc);
1269                 }
1270                 RETURN(rc);
1271         }
1272
1273         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1274                               sbi->ll_md_exp, &md);
1275         if (rc) {
1276                 ptlrpc_req_finished(request);
1277                 RETURN(rc);
1278         }
1279
1280         /* We want to adjust timestamps.
1281          * If there is at least some data in file, we cleared ATTR_SIZE
1282          * to avoid update size, otherwise it is important to do.(SOM case)
1283          * (bug 6196) */
1284         ia_valid = op_data->op_attr.ia_valid;
1285         /* Since we set ATTR_*_SET flags above, and already done permission
1286          * check, So don't let inode_change_ok() check it again.  */
1287         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1288         rc = simple_setattr(dentry, &op_data->op_attr);
1289         op_data->op_attr.ia_valid = ia_valid;
1290
1291         /* Extract epoch data if obtained. */
1292         op_data->op_handle = md.body->handle;
1293         op_data->op_ioepoch = md.body->ioepoch;
1294
1295         ll_update_inode(inode, &md);
1296         ptlrpc_req_finished(request);
1297
1298         RETURN(rc);
1299 }
1300
1301 /* Close IO epoch and send Size-on-MDS attribute update. */
1302 static int ll_setattr_done_writing(struct inode *inode,
1303                                    struct md_op_data *op_data,
1304                                    struct md_open_data *mod)
1305 {
1306         struct ll_inode_info *lli = ll_i2info(inode);
1307         int rc = 0;
1308         ENTRY;
1309
1310         LASSERT(op_data != NULL);
1311         if (!S_ISREG(inode->i_mode))
1312                 RETURN(0);
1313
1314         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1315                op_data->op_ioepoch, PFID(&lli->lli_fid));
1316
1317         op_data->op_flags = MF_EPOCH_CLOSE;
1318         ll_done_writing_attr(inode, op_data);
1319         ll_pack_inode2opdata(inode, op_data, NULL);
1320
1321         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1322         if (rc == -EAGAIN) {
1323                 /* MDS has instructed us to obtain Size-on-MDS attribute
1324                  * from OSTs and send setattr to back to MDS. */
1325                 rc = ll_som_update(inode, op_data);
1326         } else if (rc) {
1327                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1328                        inode->i_ino, rc);
1329         }
1330         RETURN(rc);
1331 }
1332
1333 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1334 {
1335         struct obd_capa *capa;
1336         int rc;
1337
1338         if (attr->ia_valid & ATTR_SIZE)
1339                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1340         else
1341                 capa = ll_mdscapa_get(inode);
1342
1343         rc = cl_setattr_ost(inode, attr, capa);
1344
1345         if (attr->ia_valid & ATTR_SIZE)
1346                 ll_truncate_free_capa(capa);
1347         else
1348                 capa_put(capa);
1349
1350         return rc;
1351 }
1352
1353 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1354 /**
1355  * inode_newsize_ok - may this inode be truncated to a given size
1356  * @inode:      the inode to be truncated
1357  * @offset:     the new size to assign to the inode
1358  * @Returns:    0 on success, -ve errno on failure
1359  *
1360  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1361  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1362  * when necessary. Caller must not proceed with inode size change if failure is
1363  * returned. @inode must be a file (not directory), with appropriate
1364  * permissions to allow truncate (inode_newsize_ok does NOT check these
1365  * conditions).
1366  *
1367  * inode_newsize_ok must be called with i_mutex held.
1368  */
1369 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1370 {
1371         if (inode->i_size < offset) {
1372                 unsigned long limit;
1373
1374                 limit = rlimit(RLIMIT_FSIZE);
1375                 if (limit != RLIM_INFINITY && offset > limit)
1376                         goto out_sig;
1377                 if (offset > inode->i_sb->s_maxbytes)
1378                         goto out_big;
1379         } else {
1380                 /*
1381                  * truncation of in-use swapfiles is disallowed - it would
1382                  * cause subsequent swapout to scribble on the now-freed
1383                  * blocks.
1384                  */
1385                 if (IS_SWAPFILE(inode))
1386                         return -ETXTBSY;
1387         }
1388
1389         return 0;
1390 out_sig:
1391         send_sig(SIGXFSZ, current, 0);
1392 out_big:
1393         return -EFBIG;
1394 }
1395 #endif
1396
1397 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1398  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1399  * keep these values until such a time that objects are allocated for it.
1400  * We do the MDS operations first, as it is checking permissions for us.
1401  * We don't to the MDS RPC if there is nothing that we want to store there,
1402  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1403  * going to do an RPC anyways.
1404  *
1405  * If we are doing a truncate, we will send the mtime and ctime updates
1406  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1407  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1408  * at the same time.
1409  */
1410 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1411 {
1412         struct lov_stripe_md *lsm;
1413         struct inode *inode = dentry->d_inode;
1414         struct ll_inode_info *lli = ll_i2info(inode);
1415         struct md_op_data *op_data = NULL;
1416         struct md_open_data *mod = NULL;
1417         int ia_valid = attr->ia_valid;
1418         int rc = 0, rc1 = 0;
1419         ENTRY;
1420
1421         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1422                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1423                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1424                 attr->ia_valid);
1425
1426         if (ia_valid & ATTR_SIZE) {
1427                 /* Check new size against VFS/VM file size limit and rlimit */
1428                 rc = inode_newsize_ok(inode, attr->ia_size);
1429                 if (rc)
1430                         RETURN(rc);
1431
1432                 /* The maximum Lustre file size is variable, based on the
1433                  * OST maximum object size and number of stripes.  This
1434                  * needs another check in addition to the VFS check above. */
1435                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1436                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1437                                PFID(&lli->lli_fid), attr->ia_size,
1438                                ll_file_maxbytes(inode));
1439                         RETURN(-EFBIG);
1440                 }
1441
1442                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1443         }
1444
1445         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1446         if (ia_valid & TIMES_SET_FLAGS) {
1447                 if (cfs_curproc_fsuid() != inode->i_uid &&
1448                     !cfs_capable(CFS_CAP_FOWNER))
1449                         RETURN(-EPERM);
1450         }
1451
1452         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1453         if (attr->ia_valid & ATTR_CTIME) {
1454                 attr->ia_ctime = CFS_CURRENT_TIME;
1455                 attr->ia_valid |= ATTR_CTIME_SET;
1456         }
1457         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1458                 attr->ia_atime = CFS_CURRENT_TIME;
1459                 attr->ia_valid |= ATTR_ATIME_SET;
1460         }
1461         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1462                 attr->ia_mtime = CFS_CURRENT_TIME;
1463                 attr->ia_valid |= ATTR_MTIME_SET;
1464         }
1465
1466         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1467                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1468                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1469                        cfs_time_current_sec());
1470
1471         /* If we are changing file size, file content is modified, flag it. */
1472         if (attr->ia_valid & ATTR_SIZE) {
1473                 spin_lock(&lli->lli_lock);
1474                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1475                 spin_unlock(&lli->lli_lock);
1476         }
1477
1478         /* We always do an MDS RPC, even if we're only changing the size;
1479          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1480
1481         OBD_ALLOC_PTR(op_data);
1482         if (op_data == NULL)
1483                 RETURN(-ENOMEM);
1484
1485         if (!S_ISDIR(inode->i_mode)) {
1486                 if (ia_valid & ATTR_SIZE)
1487                         inode_dio_write_done(inode);
1488                 mutex_unlock(&inode->i_mutex);
1489                 down_write(&lli->lli_trunc_sem);
1490                 mutex_lock(&inode->i_mutex);
1491                 if (ia_valid & ATTR_SIZE)
1492                         inode_dio_wait(inode);
1493         }
1494
1495         /* We need a steady stripe configuration for setattr to avoid
1496          * confusion. */
1497         lsm = ccc_inode_lsm_get(inode);
1498
1499         /* NB: ATTR_SIZE will only be set after this point if the size
1500          * resides on the MDS, ie, this file has no objects. */
1501         if (lsm != NULL)
1502                 attr->ia_valid &= ~ATTR_SIZE;
1503         /* can't call ll_setattr_ost() while holding a refcount of lsm */
1504         ccc_inode_lsm_put(inode, lsm);
1505
1506         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1507
1508         /* Open epoch for truncate. */
1509         if (exp_connect_som(ll_i2mdexp(inode)) &&
1510             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1511                 op_data->op_flags = MF_EPOCH_OPEN;
1512
1513         rc = ll_md_setattr(dentry, op_data, &mod);
1514         if (rc)
1515                 GOTO(out, rc);
1516
1517         /* RPC to MDT is sent, cancel data modification flag */
1518         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1519                 spin_lock(&lli->lli_lock);
1520                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1521                 spin_unlock(&lli->lli_lock);
1522         }
1523
1524         ll_ioepoch_open(lli, op_data->op_ioepoch);
1525         if (!S_ISREG(inode->i_mode))
1526                 GOTO(out, rc = 0);
1527
1528         if (ia_valid & ATTR_SIZE)
1529                 attr->ia_valid |= ATTR_SIZE;
1530         if (ia_valid & (ATTR_SIZE |
1531                         ATTR_ATIME | ATTR_ATIME_SET |
1532                         ATTR_MTIME | ATTR_MTIME_SET))
1533                 /* For truncate and utimes sending attributes to OSTs, setting
1534                  * mtime/atime to the past will be performed under PW [0:EOF]
1535                  * extent lock (new_size:EOF for truncate).  It may seem
1536                  * excessive to send mtime/atime updates to OSTs when not
1537                  * setting times to past, but it is necessary due to possible
1538                  * time de-synchronization between MDT inode and OST objects */
1539                 rc = ll_setattr_ost(inode, attr);
1540         EXIT;
1541 out:
1542         if (op_data) {
1543                 if (op_data->op_ioepoch) {
1544                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1545                         if (!rc)
1546                                 rc = rc1;
1547                 }
1548                 ll_finish_md_op_data(op_data);
1549         }
1550         if (!S_ISDIR(inode->i_mode))
1551                 up_write(&lli->lli_trunc_sem);
1552
1553         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1554                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1555
1556         return rc;
1557 }
1558
1559 int ll_setattr(struct dentry *de, struct iattr *attr)
1560 {
1561         int mode = de->d_inode->i_mode;
1562
1563         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1564                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1565                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1566
1567         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1568                                (ATTR_SIZE|ATTR_MODE)) &&
1569             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1570              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1571               !(attr->ia_mode & S_ISGID))))
1572                 attr->ia_valid |= ATTR_FORCE;
1573
1574         if ((mode & S_ISUID) &&
1575             !(attr->ia_mode & S_ISUID) &&
1576             !(attr->ia_valid & ATTR_KILL_SUID))
1577                 attr->ia_valid |= ATTR_KILL_SUID;
1578
1579         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1580             !(attr->ia_mode & S_ISGID) &&
1581             !(attr->ia_valid & ATTR_KILL_SGID))
1582                 attr->ia_valid |= ATTR_KILL_SGID;
1583
1584         return ll_setattr_raw(de, attr);
1585 }
1586
1587 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1588                        __u64 max_age, __u32 flags)
1589 {
1590         struct ll_sb_info *sbi = ll_s2sbi(sb);
1591         struct obd_statfs obd_osfs;
1592         int rc;
1593         ENTRY;
1594
1595         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1596         if (rc) {
1597                 CERROR("md_statfs fails: rc = %d\n", rc);
1598                 RETURN(rc);
1599         }
1600
1601         osfs->os_type = sb->s_magic;
1602
1603         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1604                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1605
1606         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1607                 flags |= OBD_STATFS_NODELAY;
1608
1609         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1610         if (rc) {
1611                 CERROR("obd_statfs fails: rc = %d\n", rc);
1612                 RETURN(rc);
1613         }
1614
1615         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1616                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1617                obd_osfs.os_files);
1618
1619         osfs->os_bsize = obd_osfs.os_bsize;
1620         osfs->os_blocks = obd_osfs.os_blocks;
1621         osfs->os_bfree = obd_osfs.os_bfree;
1622         osfs->os_bavail = obd_osfs.os_bavail;
1623
1624         /* If we don't have as many objects free on the OST as inodes
1625          * on the MDS, we reduce the total number of inodes to
1626          * compensate, so that the "inodes in use" number is correct.
1627          */
1628         if (obd_osfs.os_ffree < osfs->os_ffree) {
1629                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1630                         obd_osfs.os_ffree;
1631                 osfs->os_ffree = obd_osfs.os_ffree;
1632         }
1633
1634         RETURN(rc);
1635 }
1636 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1637 {
1638         struct super_block *sb = de->d_sb;
1639         struct obd_statfs osfs;
1640         int rc;
1641
1642         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1643         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1644
1645         /* Some amount of caching on the client is allowed */
1646         rc = ll_statfs_internal(sb, &osfs,
1647                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1648                                 0);
1649         if (rc)
1650                 return rc;
1651
1652         statfs_unpack(sfs, &osfs);
1653
1654         /* We need to downshift for all 32-bit kernels, because we can't
1655          * tell if the kernel is being called via sys_statfs64() or not.
1656          * Stop before overflowing f_bsize - in which case it is better
1657          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1658         if (sizeof(long) < 8) {
1659                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1660                         sfs->f_bsize <<= 1;
1661
1662                         osfs.os_blocks >>= 1;
1663                         osfs.os_bfree >>= 1;
1664                         osfs.os_bavail >>= 1;
1665                 }
1666         }
1667
1668         sfs->f_blocks = osfs.os_blocks;
1669         sfs->f_bfree = osfs.os_bfree;
1670         sfs->f_bavail = osfs.os_bavail;
1671
1672         return 0;
1673 }
1674
1675 void ll_inode_size_lock(struct inode *inode)
1676 {
1677         struct ll_inode_info *lli;
1678
1679         LASSERT(!S_ISDIR(inode->i_mode));
1680
1681         lli = ll_i2info(inode);
1682         LASSERT(lli->lli_size_sem_owner != current);
1683         down(&lli->lli_size_sem);
1684         LASSERT(lli->lli_size_sem_owner == NULL);
1685         lli->lli_size_sem_owner = current;
1686 }
1687
1688 void ll_inode_size_unlock(struct inode *inode)
1689 {
1690         struct ll_inode_info *lli;
1691
1692         lli = ll_i2info(inode);
1693         LASSERT(lli->lli_size_sem_owner == current);
1694         lli->lli_size_sem_owner = NULL;
1695         up(&lli->lli_size_sem);
1696 }
1697
1698 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1699 {
1700         struct ll_inode_info *lli = ll_i2info(inode);
1701         struct mdt_body *body = md->body;
1702         struct lov_stripe_md *lsm = md->lsm;
1703         struct ll_sb_info *sbi = ll_i2sbi(inode);
1704
1705         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1706         if (lsm != NULL) {
1707                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1708                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1709                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1710         }
1711
1712         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1713                 if (body->valid & OBD_MD_FLRMTPERM)
1714                         ll_update_remote_perm(inode, md->remote_perm);
1715         }
1716 #ifdef CONFIG_FS_POSIX_ACL
1717         else if (body->valid & OBD_MD_FLACL) {
1718                 spin_lock(&lli->lli_lock);
1719                 if (lli->lli_posix_acl)
1720                         posix_acl_release(lli->lli_posix_acl);
1721                 lli->lli_posix_acl = md->posix_acl;
1722                 spin_unlock(&lli->lli_lock);
1723         }
1724 #endif
1725         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1726         inode->i_generation = cl_fid_build_gen(&body->fid1);
1727
1728         if (body->valid & OBD_MD_FLATIME) {
1729                 if (body->atime > LTIME_S(inode->i_atime))
1730                         LTIME_S(inode->i_atime) = body->atime;
1731                 lli->lli_lvb.lvb_atime = body->atime;
1732         }
1733         if (body->valid & OBD_MD_FLMTIME) {
1734                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1735                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1736                                "to "LPU64"\n", inode->i_ino,
1737                                LTIME_S(inode->i_mtime), body->mtime);
1738                         LTIME_S(inode->i_mtime) = body->mtime;
1739                 }
1740                 lli->lli_lvb.lvb_mtime = body->mtime;
1741         }
1742         if (body->valid & OBD_MD_FLCTIME) {
1743                 if (body->ctime > LTIME_S(inode->i_ctime))
1744                         LTIME_S(inode->i_ctime) = body->ctime;
1745                 lli->lli_lvb.lvb_ctime = body->ctime;
1746         }
1747         if (body->valid & OBD_MD_FLMODE)
1748                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1749         if (body->valid & OBD_MD_FLTYPE)
1750                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1751         LASSERT(inode->i_mode != 0);
1752         if (S_ISREG(inode->i_mode)) {
1753                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1754         } else {
1755                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1756         }
1757         if (body->valid & OBD_MD_FLUID)
1758                 inode->i_uid = body->uid;
1759         if (body->valid & OBD_MD_FLGID)
1760                 inode->i_gid = body->gid;
1761         if (body->valid & OBD_MD_FLFLAGS)
1762                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1763         if (body->valid & OBD_MD_FLNLINK)
1764                 set_nlink(inode, body->nlink);
1765         if (body->valid & OBD_MD_FLRDEV)
1766                 inode->i_rdev = old_decode_dev(body->rdev);
1767
1768         if (body->valid & OBD_MD_FLID) {
1769                 /* FID shouldn't be changed! */
1770                 if (fid_is_sane(&lli->lli_fid)) {
1771                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1772                                  "Trying to change FID "DFID
1773                                  " to the "DFID", inode %lu/%u(%p)\n",
1774                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1775                                  inode->i_ino, inode->i_generation, inode);
1776                 } else
1777                         lli->lli_fid = body->fid1;
1778         }
1779
1780         LASSERT(fid_seq(&lli->lli_fid) != 0);
1781
1782         if (body->valid & OBD_MD_FLSIZE) {
1783                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1784                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1785                         struct lustre_handle lockh;
1786                         ldlm_mode_t mode;
1787
1788                         /* As it is possible a blocking ast has been processed
1789                          * by this time, we need to check there is an UPDATE
1790                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1791                          * it. */
1792                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1793                                                &lockh, LDLM_FL_CBPENDING);
1794                         if (mode) {
1795                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1796                                                       LLIF_EPOCH_PENDING |
1797                                                       LLIF_SOM_DIRTY)) {
1798                                         CERROR("ino %lu flags %u still has "
1799                                                "size authority! do not trust "
1800                                                "the size got from MDS\n",
1801                                                inode->i_ino, lli->lli_flags);
1802                                 } else {
1803                                         /* Use old size assignment to avoid
1804                                          * deadlock bz14138 & bz14326 */
1805                                         i_size_write(inode, body->size);
1806                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1807                                 }
1808                                 ldlm_lock_decref(&lockh, mode);
1809                         }
1810                 } else {
1811                         /* Use old size assignment to avoid
1812                          * deadlock bz14138 & bz14326 */
1813                         i_size_write(inode, body->size);
1814
1815                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1816                                inode->i_ino, (unsigned long long)body->size);
1817                 }
1818
1819                 if (body->valid & OBD_MD_FLBLOCKS)
1820                         inode->i_blocks = body->blocks;
1821         }
1822
1823         if (body->valid & OBD_MD_FLMDSCAPA) {
1824                 LASSERT(md->mds_capa);
1825                 ll_add_capa(inode, md->mds_capa);
1826         }
1827         if (body->valid & OBD_MD_FLOSSCAPA) {
1828                 LASSERT(md->oss_capa);
1829                 ll_add_capa(inode, md->oss_capa);
1830         }
1831 }
1832
1833 void ll_read_inode2(struct inode *inode, void *opaque)
1834 {
1835         struct lustre_md *md = opaque;
1836         struct ll_inode_info *lli = ll_i2info(inode);
1837         ENTRY;
1838
1839         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1840                PFID(&lli->lli_fid), inode);
1841
1842         LASSERT(!lli->lli_has_smd);
1843
1844         /* Core attributes from the MDS first.  This is a new inode, and
1845          * the VFS doesn't zero times in the core inode so we have to do
1846          * it ourselves.  They will be overwritten by either MDS or OST
1847          * attributes - we just need to make sure they aren't newer. */
1848         LTIME_S(inode->i_mtime) = 0;
1849         LTIME_S(inode->i_atime) = 0;
1850         LTIME_S(inode->i_ctime) = 0;
1851         inode->i_rdev = 0;
1852         ll_update_inode(inode, md);
1853
1854         /* OIDEBUG(inode); */
1855
1856         /* initializing backing dev info. */
1857         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1858
1859
1860         if (S_ISREG(inode->i_mode)) {
1861                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1862                 inode->i_op = &ll_file_inode_operations;
1863                 inode->i_fop = sbi->ll_fop;
1864                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1865                 EXIT;
1866         } else if (S_ISDIR(inode->i_mode)) {
1867                 inode->i_op = &ll_dir_inode_operations;
1868                 inode->i_fop = &ll_dir_operations;
1869                 EXIT;
1870         } else if (S_ISLNK(inode->i_mode)) {
1871                 inode->i_op = &ll_fast_symlink_inode_operations;
1872                 EXIT;
1873         } else {
1874                 inode->i_op = &ll_special_inode_operations;
1875
1876                 init_special_inode(inode, inode->i_mode,
1877                                    inode->i_rdev);
1878
1879                 EXIT;
1880         }
1881 }
1882
1883 void ll_delete_inode(struct inode *inode)
1884 {
1885         struct cl_inode_info *lli = cl_i2info(inode);
1886         ENTRY;
1887
1888         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1889                 /* discard all dirty pages before truncating them, required by
1890                  * osc_extent implementation at LU-1030. */
1891                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1892
1893         truncate_inode_pages(&inode->i_data, 0);
1894
1895         /* Workaround for LU-118 */
1896         if (inode->i_data.nrpages) {
1897                 TREE_READ_LOCK_IRQ(&inode->i_data);
1898                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1899                 LASSERTF(inode->i_data.nrpages == 0,
1900                          "inode=%lu/%u(%p) nrpages=%lu, see "
1901                          "http://jira.whamcloud.com/browse/LU-118\n",
1902                          inode->i_ino, inode->i_generation, inode,
1903                          inode->i_data.nrpages);
1904         }
1905         /* Workaround end */
1906
1907 #ifdef HAVE_SBOPS_EVICT_INODE
1908         ll_clear_inode(inode);
1909 #endif
1910         clear_inode(inode);
1911
1912         EXIT;
1913 }
1914
1915 int ll_iocontrol(struct inode *inode, struct file *file,
1916                  unsigned int cmd, unsigned long arg)
1917 {
1918         struct ll_sb_info *sbi = ll_i2sbi(inode);
1919         struct ptlrpc_request *req = NULL;
1920         int rc, flags = 0;
1921         ENTRY;
1922
1923         switch(cmd) {
1924         case FSFILT_IOC_GETFLAGS: {
1925                 struct mdt_body *body;
1926                 struct md_op_data *op_data;
1927
1928                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1929                                              0, 0, LUSTRE_OPC_ANY,
1930                                              NULL);
1931                 if (IS_ERR(op_data))
1932                         RETURN(PTR_ERR(op_data));
1933
1934                 op_data->op_valid = OBD_MD_FLFLAGS;
1935                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1936                 ll_finish_md_op_data(op_data);
1937                 if (rc) {
1938                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1939                         RETURN(-abs(rc));
1940                 }
1941
1942                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1943
1944                 flags = body->flags;
1945
1946                 ptlrpc_req_finished(req);
1947
1948                 RETURN(put_user(flags, (int *)arg));
1949         }
1950         case FSFILT_IOC_SETFLAGS: {
1951                 struct lov_stripe_md *lsm;
1952                 struct obd_info oinfo = { { { 0 } } };
1953                 struct md_op_data *op_data;
1954
1955                 if (get_user(flags, (int *)arg))
1956                         RETURN(-EFAULT);
1957
1958                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1959                                              LUSTRE_OPC_ANY, NULL);
1960                 if (IS_ERR(op_data))
1961                         RETURN(PTR_ERR(op_data));
1962
1963                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1964                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1965                 rc = md_setattr(sbi->ll_md_exp, op_data,
1966                                 NULL, 0, NULL, 0, &req, NULL);
1967                 ll_finish_md_op_data(op_data);
1968                 ptlrpc_req_finished(req);
1969                 if (rc)
1970                         RETURN(rc);
1971
1972                 inode->i_flags = ll_ext_to_inode_flags(flags);
1973
1974                 lsm = ccc_inode_lsm_get(inode);
1975                 if (lsm == NULL)
1976                         RETURN(0);
1977
1978                 OBDO_ALLOC(oinfo.oi_oa);
1979                 if (!oinfo.oi_oa) {
1980                         ccc_inode_lsm_put(inode, lsm);
1981                         RETURN(-ENOMEM);
1982                 }
1983                 oinfo.oi_md = lsm;
1984                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1985                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1986                 oinfo.oi_oa->o_flags = flags;
1987                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1988                                        OBD_MD_FLGROUP;
1989                 oinfo.oi_capa = ll_mdscapa_get(inode);
1990                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1991                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1992                 capa_put(oinfo.oi_capa);
1993                 OBDO_FREE(oinfo.oi_oa);
1994                 ccc_inode_lsm_put(inode, lsm);
1995
1996                 if (rc && rc != -EPERM && rc != -EACCES)
1997                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1998
1999                 RETURN(rc);
2000         }
2001         default:
2002                 RETURN(-ENOSYS);
2003         }
2004
2005         RETURN(0);
2006 }
2007
2008 int ll_flush_ctx(struct inode *inode)
2009 {
2010         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2011
2012         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2013
2014         obd_set_info_async(NULL, sbi->ll_md_exp,
2015                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2016                            0, NULL, NULL);
2017         obd_set_info_async(NULL, sbi->ll_dt_exp,
2018                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2019                            0, NULL, NULL);
2020         return 0;
2021 }
2022
2023 /* umount -f client means force down, don't save state */
2024 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2025 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2026 {
2027         struct super_block *sb = vfsmnt->mnt_sb;
2028 #else
2029 void ll_umount_begin(struct super_block *sb)
2030 {
2031 #endif
2032         struct ll_sb_info *sbi = ll_s2sbi(sb);
2033         struct obd_device *obd;
2034         struct obd_ioctl_data *ioc_data;
2035         ENTRY;
2036
2037 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2038         if (!(flags & MNT_FORCE)) {
2039                 EXIT;
2040                 return;
2041         }
2042 #endif
2043
2044         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2045                sb->s_count, atomic_read(&sb->s_active));
2046
2047         obd = class_exp2obd(sbi->ll_md_exp);
2048         if (obd == NULL) {
2049                 CERROR("Invalid MDC connection handle "LPX64"\n",
2050                        sbi->ll_md_exp->exp_handle.h_cookie);
2051                 EXIT;
2052                 return;
2053         }
2054         obd->obd_force = 1;
2055
2056         obd = class_exp2obd(sbi->ll_dt_exp);
2057         if (obd == NULL) {
2058                 CERROR("Invalid LOV connection handle "LPX64"\n",
2059                        sbi->ll_dt_exp->exp_handle.h_cookie);
2060                 EXIT;
2061                 return;
2062         }
2063         obd->obd_force = 1;
2064
2065         OBD_ALLOC_PTR(ioc_data);
2066         if (ioc_data) {
2067                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2068                               sizeof ioc_data, ioc_data, NULL);
2069
2070                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2071                               sizeof ioc_data, ioc_data, NULL);
2072
2073                 OBD_FREE_PTR(ioc_data);
2074         }
2075
2076
2077         /* Really, we'd like to wait until there are no requests outstanding,
2078          * and then continue.  For now, we just invalidate the requests,
2079          * schedule() and sleep one second if needed, and hope.
2080          */
2081         cfs_schedule();
2082 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2083         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2084                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2085                                                    cfs_time_seconds(1));
2086                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2087                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2088                                       "may try to umount it a bit later\n",
2089                                       atomic_read(&vfsmnt->mnt_count));
2090         }
2091 #endif
2092
2093         EXIT;
2094 }
2095
2096 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2097 {
2098         struct ll_sb_info *sbi = ll_s2sbi(sb);
2099         char *profilenm = get_profile_name(sb);
2100         int err;
2101         __u32 read_only;
2102
2103         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2104                 read_only = *flags & MS_RDONLY;
2105                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2106                                          sizeof(KEY_READ_ONLY),
2107                                          KEY_READ_ONLY, sizeof(read_only),
2108                                          &read_only, NULL);
2109                 if (err) {
2110                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2111                                       profilenm, read_only ?
2112                                       "read-only" : "read-write", err);
2113                         return err;
2114                 }
2115
2116                 if (read_only)
2117                         sb->s_flags |= MS_RDONLY;
2118                 else
2119                         sb->s_flags &= ~MS_RDONLY;
2120
2121                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2122                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2123                                       read_only ?  "read-only" : "read-write");
2124         }
2125         return 0;
2126 }
2127
2128 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2129                   struct super_block *sb, struct lookup_intent *it)
2130 {
2131         struct ll_sb_info *sbi = NULL;
2132         struct lustre_md md;
2133         int rc;
2134         ENTRY;
2135
2136         LASSERT(*inode || sb);
2137         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2138         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2139                               sbi->ll_md_exp, &md);
2140         if (rc)
2141                 RETURN(rc);
2142
2143         if (*inode) {
2144                 ll_update_inode(*inode, &md);
2145         } else {
2146                 LASSERT(sb != NULL);
2147
2148                 /*
2149                  * At this point server returns to client's same fid as client
2150                  * generated for creating. So using ->fid1 is okay here.
2151                  */
2152                 LASSERT(fid_is_sane(&md.body->fid1));
2153
2154                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2155                 if (*inode == NULL || IS_ERR(*inode)) {
2156 #ifdef CONFIG_FS_POSIX_ACL
2157                         if (md.posix_acl) {
2158                                 posix_acl_release(md.posix_acl);
2159                                 md.posix_acl = NULL;
2160                         }
2161 #endif
2162                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2163                         *inode = NULL;
2164                         CERROR("new_inode -fatal: rc %d\n", rc);
2165                         GOTO(out, rc);
2166                 }
2167         }
2168
2169         /* Handling piggyback layout lock.
2170          * Layout lock can be piggybacked by getattr and open request.
2171          * The lsm can be applied to inode only if it comes with a layout lock
2172          * otherwise correct layout may be overwritten, for example:
2173          * 1. proc1: mdt returns a lsm but not granting layout
2174          * 2. layout was changed by another client
2175          * 3. proc2: refresh layout and layout lock granted
2176          * 4. proc1: to apply a stale layout */
2177         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2178                 struct lustre_handle lockh;
2179                 struct ldlm_lock *lock;
2180
2181                 lockh.cookie = it->d.lustre.it_lock_handle;
2182                 lock = ldlm_handle2lock(&lockh);
2183                 LASSERT(lock != NULL);
2184                 if (ldlm_has_layout(lock)) {
2185                         struct cl_object_conf conf;
2186
2187                         memset(&conf, 0, sizeof(conf));
2188                         conf.coc_opc = OBJECT_CONF_SET;
2189                         conf.coc_inode = *inode;
2190                         conf.coc_lock = lock;
2191                         conf.u.coc_md = &md;
2192                         (void)ll_layout_conf(*inode, &conf);
2193                 }
2194                 LDLM_LOCK_PUT(lock);
2195         }
2196
2197 out:
2198         if (md.lsm != NULL)
2199                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2200         md_free_lustre_md(sbi->ll_md_exp, &md);
2201         RETURN(rc);
2202 }
2203
2204 int ll_obd_statfs(struct inode *inode, void *arg)
2205 {
2206         struct ll_sb_info *sbi = NULL;
2207         struct obd_export *exp;
2208         char *buf = NULL;
2209         struct obd_ioctl_data *data = NULL;
2210         __u32 type;
2211         __u32 flags;
2212         int len = 0, rc;
2213
2214         if (!inode || !(sbi = ll_i2sbi(inode)))
2215                 GOTO(out_statfs, rc = -EINVAL);
2216
2217         rc = obd_ioctl_getdata(&buf, &len, arg);
2218         if (rc)
2219                 GOTO(out_statfs, rc);
2220
2221         data = (void*)buf;
2222         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2223             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2224                 GOTO(out_statfs, rc = -EINVAL);
2225
2226         if (data->ioc_inllen1 != sizeof(__u32) ||
2227             data->ioc_inllen2 != sizeof(__u32) ||
2228             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2229             data->ioc_plen2 != sizeof(struct obd_uuid))
2230                 GOTO(out_statfs, rc = -EINVAL);
2231
2232         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2233         if (type & LL_STATFS_LMV)
2234                 exp = sbi->ll_md_exp;
2235         else if (type & LL_STATFS_LOV)
2236                 exp = sbi->ll_dt_exp;
2237         else
2238                 GOTO(out_statfs, rc = -ENODEV);
2239
2240         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2241         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2242         if (rc)
2243                 GOTO(out_statfs, rc);
2244 out_statfs:
2245         if (buf)
2246                 obd_ioctl_freedata(buf, len);
2247         return rc;
2248 }
2249
2250 int ll_process_config(struct lustre_cfg *lcfg)
2251 {
2252         char *ptr;
2253         void *sb;
2254         struct lprocfs_static_vars lvars;
2255         unsigned long x;
2256         int rc = 0;
2257
2258         lprocfs_llite_init_vars(&lvars);
2259
2260         /* The instance name contains the sb: lustre-client-aacfe000 */
2261         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2262         if (!ptr || !*(++ptr))
2263                 return -EINVAL;
2264         if (sscanf(ptr, "%lx", &x) != 1)
2265                 return -EINVAL;
2266         sb = (void *)x;
2267         /* This better be a real Lustre superblock! */
2268         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2269
2270         /* Note we have not called client_common_fill_super yet, so
2271            proc fns must be able to handle that! */
2272         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2273                                       lcfg, sb);
2274         if (rc > 0)
2275                 rc = 0;
2276         return(rc);
2277 }
2278
2279 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2280 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2281                                        struct inode *i1, struct inode *i2,
2282                                        const char *name, int namelen,
2283                                        int mode, __u32 opc, void *data)
2284 {
2285         LASSERT(i1 != NULL);
2286
2287         if (namelen > ll_i2sbi(i1)->ll_namelen)
2288                 return ERR_PTR(-ENAMETOOLONG);
2289
2290         if (op_data == NULL)
2291                 OBD_ALLOC_PTR(op_data);
2292
2293         if (op_data == NULL)
2294                 return ERR_PTR(-ENOMEM);
2295
2296         ll_i2gids(op_data->op_suppgids, i1, i2);
2297         op_data->op_fid1 = *ll_inode2fid(i1);
2298         op_data->op_capa1 = ll_mdscapa_get(i1);
2299
2300         if (i2) {
2301                 op_data->op_fid2 = *ll_inode2fid(i2);
2302                 op_data->op_capa2 = ll_mdscapa_get(i2);
2303         } else {
2304                 fid_zero(&op_data->op_fid2);
2305                 op_data->op_capa2 = NULL;
2306         }
2307
2308         op_data->op_name = name;
2309         op_data->op_namelen = namelen;
2310         op_data->op_mode = mode;
2311         op_data->op_mod_time = cfs_time_current_sec();
2312         op_data->op_fsuid = cfs_curproc_fsuid();
2313         op_data->op_fsgid = cfs_curproc_fsgid();
2314         op_data->op_cap = cfs_curproc_cap_pack();
2315         op_data->op_bias = MDS_CHECK_SPLIT;
2316         op_data->op_opc = opc;
2317         op_data->op_mds = 0;
2318         op_data->op_data = data;
2319
2320         /* If the file is being opened after mknod() (normally due to NFS)
2321          * try to use the default stripe data from parent directory for
2322          * allocating OST objects.  Try to pass the parent FID to MDS. */
2323         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2324             !ll_i2info(i2)->lli_has_smd) {
2325                 struct ll_inode_info *lli = ll_i2info(i2);
2326
2327                 spin_lock(&lli->lli_lock);
2328                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2329                         op_data->op_fid1 = lli->lli_pfid;
2330                 spin_unlock(&lli->lli_lock);
2331                 /** We ignore parent's capability temporary. */
2332         }
2333
2334         /* When called by ll_setattr_raw, file is i1. */
2335         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2336                 op_data->op_bias |= MDS_DATA_MODIFIED;
2337
2338         return op_data;
2339 }
2340
2341 void ll_finish_md_op_data(struct md_op_data *op_data)
2342 {
2343         capa_put(op_data->op_capa1);
2344         capa_put(op_data->op_capa2);
2345         OBD_FREE_PTR(op_data);
2346 }
2347
2348 #ifdef HAVE_SUPEROPS_USE_DENTRY
2349 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2350 #else
2351 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2352 #endif
2353 {
2354         struct ll_sb_info *sbi;
2355
2356 #ifdef HAVE_SUPEROPS_USE_DENTRY
2357         LASSERT((seq != NULL) && (dentry != NULL));
2358         sbi = ll_s2sbi(dentry->d_sb);
2359 #else
2360         LASSERT((seq != NULL) && (vfs != NULL));
2361         sbi = ll_s2sbi(vfs->mnt_sb);
2362 #endif
2363
2364         if (sbi->ll_flags & LL_SBI_NOLCK)
2365                 seq_puts(seq, ",nolock");
2366
2367         if (sbi->ll_flags & LL_SBI_FLOCK)
2368                 seq_puts(seq, ",flock");
2369
2370         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2371                 seq_puts(seq, ",localflock");
2372
2373         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2374                 seq_puts(seq, ",user_xattr");
2375
2376         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2377                 seq_puts(seq, ",lazystatfs");
2378
2379         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2380                 seq_puts(seq, ",user_fid2path");
2381
2382         RETURN(0);
2383 }
2384
2385 /**
2386  * Get obd name by cmd, and copy out to user space
2387  */
2388 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2389 {
2390         struct ll_sb_info *sbi = ll_i2sbi(inode);
2391         struct obd_device *obd;
2392         ENTRY;
2393
2394         if (cmd == OBD_IOC_GETDTNAME)
2395                 obd = class_exp2obd(sbi->ll_dt_exp);
2396         else if (cmd == OBD_IOC_GETMDNAME)
2397                 obd = class_exp2obd(sbi->ll_md_exp);
2398         else
2399                 RETURN(-EINVAL);
2400
2401         if (!obd)
2402                 RETURN(-ENOENT);
2403
2404         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2405                              strlen(obd->obd_name) + 1))
2406                 RETURN(-EFAULT);
2407
2408         RETURN(0);
2409 }
2410
2411 /**
2412  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2413  * fsname will be returned in this buffer; otherwise, a static buffer will be
2414  * used to store the fsname and returned to caller.
2415  */
2416 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2417 {
2418         static char fsname_static[MTI_NAME_MAXLEN];
2419         struct lustre_sb_info *lsi = s2lsi(sb);
2420         char *ptr;
2421         int len;
2422
2423         if (buf == NULL) {
2424                 /* this means the caller wants to use static buffer
2425                  * and it doesn't care about race. Usually this is
2426                  * in error reporting path */
2427                 buf = fsname_static;
2428                 buflen = sizeof(fsname_static);
2429         }
2430
2431         len = strlen(lsi->lsi_lmd->lmd_profile);
2432         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2433         if (ptr && (strcmp(ptr, "-client") == 0))
2434                 len -= 7;
2435
2436         if (unlikely(len >= buflen))
2437                 len = buflen - 1;
2438         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2439         buf[len] = '\0';
2440
2441         return buf;
2442 }
2443
2444 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2445 {
2446         char *path = NULL;
2447
2448 #ifdef HAVE_FS_STRUCT_USE_PATH
2449         struct path p;
2450
2451         p.dentry = dentry;
2452         p.mnt = current->fs->root.mnt;
2453         path_get(&p);
2454         path = d_path(&p, buf, bufsize);
2455         path_put(&p);
2456 #else
2457         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2458 #endif
2459
2460         return path;
2461 }
2462
2463 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2464 {
2465         char *buf, *path = NULL;
2466         struct dentry *dentry = NULL;
2467         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2468
2469         /* this can be called inside spin lock so use GFP_ATOMIC. */
2470         buf = (char *)__get_free_page(GFP_ATOMIC);
2471         if (buf != NULL) {
2472                 dentry = d_find_alias(page->mapping->host);
2473                 if (dentry != NULL)
2474                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2475         }
2476
2477         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2478               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2479               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2480               PFID(&obj->cob_header.coh_lu.loh_fid),
2481               (path && !IS_ERR(path)) ? path : "", ioret);
2482
2483         if (dentry != NULL)
2484                 dput(dentry);
2485
2486         if (buf != NULL)
2487                 free_page((unsigned long)buf);
2488 }