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                 lru_page_max = pages / 2;
  98         } else {
  99                 lru_page_max = (pages / 4) * 3;
 100         }
 101
 102         /* initialize lru data */
 103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 104         sbi->ll_cache.ccc_lru_max = lru_page_max;
 105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 108
 109         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 110                                            SBI_DEFAULT_READAHEAD_MAX);
 111         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 112         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 113                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 114         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 115         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 116
 117         ll_generate_random_uuid(uuid);
 118         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 119         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 120
 121         spin_lock(&ll_sb_lock);
 122         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 123         spin_unlock(&ll_sb_lock);
 124
 125         sbi->ll_flags |= LL_SBI_VERBOSE;
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 136                                pp_r_hist.oh_lock);
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145         cfs_atomic_set(&sbi->ll_agl_total, 0);
 146         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 147
 148         RETURN(sbi);
 149 }
 150
 151 void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 spin_lock(&ll_sb_lock);
 158                 cfs_list_del(&sbi->ll_list);
 159                 spin_unlock(&ll_sb_lock);
 160                 OBD_FREE(sbi, sizeof(*sbi));
 161         }
 162         EXIT;
 163 }
 164
 165 static struct dentry_operations ll_d_root_ops = {
 166         .d_compare = ll_dcompare,
 167         .d_revalidate = ll_revalidate_nd,
 168 };
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = 0;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         if (proc_lustre_fs_root) {
 204                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 205                                                   dt, md);
 206                 if (err < 0)
 207                         CERROR("could not register mount in /proc/fs/lustre\n");
 208         }
 209
 210         /* indicate the features supported by this client */
 211         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 212                                   OBD_CONNECT_ATTRFID  |
 213                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 214                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 215                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 216                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 217                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 218                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 219                                   OBD_CONNECT_EINPROGRESS |
 220                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 221                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 222
 223         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 224                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 225
 226 #ifdef HAVE_LRU_RESIZE_SUPPORT
 227         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 228                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 229 #endif
 230 #ifdef CONFIG_FS_POSIX_ACL
 231         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 232 #endif
 233
 234         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 235                 /* flag mdc connection as lightweight, only used for test
 236                  * purpose, use with care */
 237                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 238
 239         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 240         data->ocd_version = LUSTRE_VERSION_CODE;
 241
 242         if (sb->s_flags & MS_RDONLY)
 243                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 244         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 245                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 246
 247 #ifdef HAVE_MS_FLOCK_LOCK
 248         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 249         sb->s_flags |= MS_FLOCK_LOCK;
 250 #endif
 251 #ifdef MS_HAS_NEW_AOPS
 252         sb->s_flags |= MS_HAS_NEW_AOPS;
 253 #endif
 254
 255         if (sbi->ll_flags & LL_SBI_FLOCK)
 256                 sbi->ll_fop = &ll_file_operations_flock;
 257         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 258                 sbi->ll_fop = &ll_file_operations;
 259         else
 260                 sbi->ll_fop = &ll_file_operations_noflock;
 261
 262         /* real client */
 263         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 264         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 265                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 266
 267         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 268
 269         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 270         if (err == -EBUSY) {
 271                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 272                                    "recovery, of which this client is not a "
 273                                    "part. Please wait for recovery to complete,"
 274                                    " abort, or time out.\n", md);
 275                 GOTO(out, err);
 276         } else if (err) {
 277                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 278                 GOTO(out, err);
 279         }
 280
 281         sbi->ll_md_exp->exp_connect_data = *data;
 282
 283         /* For mount, we only need fs info from MDT0, and also in DNE, it
 284          * can make sure the client can be mounted as long as MDT0 is
 285          * avaible */
 286         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 287                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 288                         OBD_STATFS_FOR_MDT0);
 289         if (err)
 290                 GOTO(out_md, err);
 291
 292         /* This needs to be after statfs to ensure connect has finished.
 293          * Note that "data" does NOT contain the valid connect reply.
 294          * If connecting to a 1.8 server there will be no LMV device, so
 295          * we can access the MDC export directly and exp_connect_flags will
 296          * be non-zero, but if accessing an upgraded 2.1 server it will
 297          * have the correct flags filled in.
 298          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 299         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 300         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 301             valid != CLIENT_CONNECT_MDT_REQD) {
 302                 char *buf;
 303
 304                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 305                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 306                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 307                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 308                                    "feature(s) needed for correct operation "
 309                                    "of this client (%s). Please upgrade "
 310                                    "server or downgrade client.\n",
 311                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 312                 OBD_FREE(buf, CFS_PAGE_SIZE);
 313                 GOTO(out_md, err = -EPROTO);
 314         }
 315
 316         size = sizeof(*data);
 317         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 318                            KEY_CONN_DATA,  &size, data, NULL);
 319         if (err) {
 320                 CERROR("%s: Get connect data failed: rc = %d\n",
 321                        sbi->ll_md_exp->exp_obd->obd_name, err);
 322                 GOTO(out_md, err);
 323         }
 324
 325         LASSERT(osfs->os_bsize);
 326         sb->s_blocksize = osfs->os_bsize;
 327         sb->s_blocksize_bits = log2(osfs->os_bsize);
 328         sb->s_magic = LL_SUPER_MAGIC;
 329         sb->s_maxbytes = MAX_LFS_FILESIZE;
 330         sbi->ll_namelen = osfs->os_namelen;
 331         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 332
 333         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 334             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 335                 LCONSOLE_INFO("Disabling user_xattr feature because "
 336                               "it is not supported on the server\n");
 337                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 338         }
 339
 340         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 341 #ifdef MS_POSIXACL
 342                 sb->s_flags |= MS_POSIXACL;
 343 #endif
 344                 sbi->ll_flags |= LL_SBI_ACL;
 345         } else {
 346                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags &= ~MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags &= ~LL_SBI_ACL;
 351         }
 352
 353         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 354                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 355                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 356                         LCONSOLE_INFO("client is set as remote by default.\n");
 357                 }
 358         } else {
 359                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 360                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client claims to be remote, but server "
 362                                       "rejected, forced to be local.\n");
 363                 }
 364         }
 365
 366         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 367                 LCONSOLE_INFO("client enabled MDS capability!\n");
 368                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 369         }
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 372                 LCONSOLE_INFO("client enabled OSS capability!\n");
 373                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 377                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 380                 sbi->ll_md_brw_size = data->ocd_brw_size;
 381         else
 382                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 385                 LCONSOLE_INFO("Layout lock feature supported.\n");
 386                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 387         }
 388
 389         obd = class_name2obd(dt);
 390         if (!obd) {
 391                 CERROR("DT %s: not setup or attached\n", dt);
 392                 GOTO(out_md, err = -ENODEV);
 393         }
 394
 395         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 396                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 397                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 398                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 399                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 400                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 401                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 402                                   OBD_CONNECT_MAXBYTES |
 403                                   OBD_CONNECT_EINPROGRESS |
 404                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 405                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 406
 407         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 408                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 409
 410         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 411                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 412                  * disabled by default, because it can still be enabled on the
 413                  * fly via /proc. As a consequence, we still need to come to an
 414                  * agreement on the supported algorithms at connect time */
 415                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 416
 417                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 418                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 419                 else
 420                         data->ocd_cksum_types = cksum_types_supported_client();
 421         }
 422
 423 #ifdef HAVE_LRU_RESIZE_SUPPORT
 424         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 425 #endif
 426         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 427                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 428
 429         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 430                "ocd_grant: %d\n", data->ocd_connect_flags,
 431                data->ocd_version, data->ocd_grant);
 432
 433         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 434         obd->obd_upcall.onu_upcall = cl_ocd_update;
 435
 436         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 437
 438         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 439                           NULL);
 440         if (err == -EBUSY) {
 441                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 442                                    "recovery, of which this client is not a "
 443                                    "part.  Please wait for recovery to "
 444                                    "complete, abort, or time out.\n", dt);
 445                 GOTO(out_md, err);
 446         } else if (err) {
 447                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 448                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 449                 GOTO(out_md, err);
 450         }
 451
 452         sbi->ll_dt_exp->exp_connect_data = *data;
 453
 454         mutex_lock(&sbi->ll_lco.lco_lock);
 455         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 456         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 457         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 458         mutex_unlock(&sbi->ll_lco.lco_lock);
 459
 460         fid_zero(&sbi->ll_root_fid);
 461         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 462         if (err) {
 463                 CERROR("cannot mds_connect: rc = %d\n", err);
 464                 GOTO(out_dt, err);
 465         }
 466         if (!fid_is_sane(&sbi->ll_root_fid)) {
 467                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 468                        sbi->ll_md_exp->exp_obd->obd_name,
 469                        PFID(&sbi->ll_root_fid));
 470                 GOTO(out_dt, err = -EINVAL);
 471         }
 472         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 473
 474         sb->s_op = &lustre_super_operations;
 475 #if THREAD_SIZE >= 8192 /*b=17630*/
 476         sb->s_export_op = &lustre_export_operations;
 477 #endif
 478
 479         /* make root inode
 480          * XXX: move this to after cbd setup? */
 481         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 482         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 483                 valid |= OBD_MD_FLRMTPERM;
 484         else if (sbi->ll_flags & LL_SBI_ACL)
 485                 valid |= OBD_MD_FLACL;
 486
 487         OBD_ALLOC_PTR(op_data);
 488         if (op_data == NULL)
 489                 GOTO(out_dt, err = -ENOMEM);
 490
 491         op_data->op_fid1 = sbi->ll_root_fid;
 492         op_data->op_mode = 0;
 493         op_data->op_capa1 = oc;
 494         op_data->op_valid = valid;
 495
 496         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 497         if (oc)
 498                 capa_put(oc);
 499         OBD_FREE_PTR(op_data);
 500         if (err) {
 501                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 502                        sbi->ll_md_exp->exp_obd->obd_name, err);
 503                 GOTO(out_dt, err);
 504         }
 505
 506         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 507                                sbi->ll_md_exp, &lmd);
 508         if (err) {
 509                 CERROR("failed to understand root inode md: rc = %d\n", err);
 510                 ptlrpc_req_finished(request);
 511                 GOTO(out_dt, err);
 512         }
 513
 514         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 515         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 516         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 517         ptlrpc_req_finished(request);
 518
 519         if (root == NULL || IS_ERR(root)) {
 520                 if (lmd.lsm)
 521                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 522 #ifdef CONFIG_FS_POSIX_ACL
 523                 if (lmd.posix_acl) {
 524                         posix_acl_release(lmd.posix_acl);
 525                         lmd.posix_acl = NULL;
 526                 }
 527 #endif
 528                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 529                 root = NULL;
 530                 CERROR("lustre_lite: bad iget4 for root\n");
 531                 GOTO(out_root, err);
 532         }
 533
 534         err = ll_close_thread_start(&sbi->ll_lcq);
 535         if (err) {
 536                 CERROR("cannot start close thread: rc %d\n", err);
 537                 GOTO(out_root, err);
 538         }
 539
 540 #ifdef CONFIG_FS_POSIX_ACL
 541         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 542                 rct_init(&sbi->ll_rct);
 543                 et_init(&sbi->ll_et);
 544         }
 545 #endif
 546
 547         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 548         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 549                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 550                                  NULL);
 551         cl_sb_init(sb);
 552
 553         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 554                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 555                                  &sbi->ll_cache, NULL);
 556
 557         sb->s_root = d_make_root(root);
 558         if (sb->s_root == NULL) {
 559                 CERROR("%s: can't make root dentry\n",
 560                         ll_get_fsname(sb, NULL, 0));
 561                 GOTO(out_root, err = -ENOMEM);
 562         }
 563
 564 #ifdef HAVE_DCACHE_LOCK
 565         sb->s_root->d_op = &ll_d_root_ops;
 566 #else
 567         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 568         d_set_d_op(sb->s_root, &ll_d_root_ops);
 569         sb->s_d_op = &ll_d_ops;
 570 #endif
 571
 572         sbi->ll_sdev_orig = sb->s_dev;
 573
 574         /* We set sb->s_dev equal on all lustre clients in order to support
 575          * NFS export clustering.  NFSD requires that the FSID be the same
 576          * on all clients. */
 577         /* s_dev is also used in lt_compare() to compare two fs, but that is
 578          * only a node-local comparison. */
 579         uuid = obd_get_uuid(sbi->ll_md_exp);
 580         if (uuid != NULL)
 581                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 582
 583         if (data != NULL)
 584                 OBD_FREE_PTR(data);
 585         if (osfs != NULL)
 586                 OBD_FREE_PTR(osfs);
 587
 588         RETURN(err);
 589 out_root:
 590         if (root)
 591                 iput(root);
 592 out_dt:
 593         obd_disconnect(sbi->ll_dt_exp);
 594         sbi->ll_dt_exp = NULL;
 595         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 596         obd_zombie_barrier();
 597 out_md:
 598         obd_disconnect(sbi->ll_md_exp);
 599         sbi->ll_md_exp = NULL;
 600 out:
 601         if (data != NULL)
 602                 OBD_FREE_PTR(data);
 603         if (osfs != NULL)
 604                 OBD_FREE_PTR(osfs);
 605         lprocfs_unregister_mountpoint(sbi);
 606         return err;
 607 }
 608
 609 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 610 {
 611         int size, rc;
 612
 613         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 614         size = sizeof(int);
 615         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 616                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 617         if (rc)
 618                 CERROR("Get max mdsize error rc %d \n", rc);
 619
 620         RETURN(rc);
 621 }
 622
 623 void ll_dump_inode(struct inode *inode)
 624 {
 625         struct ll_d_hlist_node *tmp;
 626         int dentry_count = 0;
 627
 628         LASSERT(inode != NULL);
 629
 630         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 631                 dentry_count++;
 632
 633         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 634                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 635                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 636 }
 637
 638 void lustre_dump_dentry(struct dentry *dentry, int recur)
 639 {
 640         struct list_head *tmp;
 641         int subdirs = 0;
 642
 643         LASSERT(dentry != NULL);
 644
 645         list_for_each(tmp, &dentry->d_subdirs)
 646                 subdirs++;
 647
 648         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 649                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 650                dentry->d_name.len, dentry->d_name.name,
 651                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 652                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 653                dentry->d_flags, dentry->d_fsdata, subdirs);
 654         if (dentry->d_inode != NULL)
 655                 ll_dump_inode(dentry->d_inode);
 656
 657         if (recur == 0)
 658                 return;
 659
 660         list_for_each(tmp, &dentry->d_subdirs) {
 661                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 662                 lustre_dump_dentry(d, recur - 1);
 663         }
 664 }
 665
 666 void client_common_put_super(struct super_block *sb)
 667 {
 668         struct ll_sb_info *sbi = ll_s2sbi(sb);
 669         ENTRY;
 670
 671 #ifdef CONFIG_FS_POSIX_ACL
 672         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 673                 et_fini(&sbi->ll_et);
 674                 rct_fini(&sbi->ll_rct);
 675         }
 676 #endif
 677
 678         ll_close_thread_shutdown(sbi->ll_lcq);
 679
 680         cl_sb_fini(sb);
 681
 682         cfs_list_del(&sbi->ll_conn_chain);
 683
 684         obd_disconnect(sbi->ll_dt_exp);
 685         sbi->ll_dt_exp = NULL;
 686         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 687          * see LU-2543. */
 688         obd_zombie_barrier();
 689
 690         lprocfs_unregister_mountpoint(sbi);
 691
 692         obd_disconnect(sbi->ll_md_exp);
 693         sbi->ll_md_exp = NULL;
 694
 695         EXIT;
 696 }
 697
 698 void ll_kill_super(struct super_block *sb)
 699 {
 700         struct ll_sb_info *sbi;
 701
 702         ENTRY;
 703
 704         /* not init sb ?*/
 705         if (!(sb->s_flags & MS_ACTIVE))
 706                 return;
 707
 708         sbi = ll_s2sbi(sb);
 709         /* we need restore s_dev from changed for clustred NFS before put_super
 710          * because new kernels have cached s_dev and change sb->s_dev in
 711          * put_super not affected real removing devices */
 712         if (sbi)
 713                 sb->s_dev = sbi->ll_sdev_orig;
 714         EXIT;
 715 }
 716
 717 char *ll_read_opt(const char *opt, char *data)
 718 {
 719         char *value;
 720         char *retval;
 721         ENTRY;
 722
 723         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 724         if (strncmp(opt, data, strlen(opt)))
 725                 RETURN(NULL);
 726         if ((value = strchr(data, '=')) == NULL)
 727                 RETURN(NULL);
 728
 729         value++;
 730         OBD_ALLOC(retval, strlen(value) + 1);
 731         if (!retval) {
 732                 CERROR("out of memory!\n");
 733                 RETURN(NULL);
 734         }
 735
 736         memcpy(retval, value, strlen(value)+1);
 737         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 738         RETURN(retval);
 739 }
 740
 741 static inline int ll_set_opt(const char *opt, char *data, int fl)
 742 {
 743         if (strncmp(opt, data, strlen(opt)) != 0)
 744                 return(0);
 745         else
 746                 return(fl);
 747 }
 748
 749 /* non-client-specific mount options are parsed in lmd_parse */
 750 static int ll_options(char *options, int *flags)
 751 {
 752         int tmp;
 753         char *s1 = options, *s2;
 754         ENTRY;
 755
 756         if (!options)
 757                 RETURN(0);
 758
 759         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 760
 761         while (*s1) {
 762                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 763                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 764                 if (tmp) {
 765                         *flags |= tmp;
 766                         goto next;
 767                 }
 768                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 769                 if (tmp) {
 770                         *flags |= tmp;
 771                         goto next;
 772                 }
 773                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 774                 if (tmp) {
 775                         *flags |= tmp;
 776                         goto next;
 777                 }
 778                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 779                 if (tmp) {
 780                         *flags &= ~tmp;
 781                         goto next;
 782                 }
 783                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 784                 if (tmp) {
 785                         *flags |= tmp;
 786                         goto next;
 787                 }
 788                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 789                 if (tmp) {
 790                         *flags &= ~tmp;
 791                         goto next;
 792                 }
 793 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 794                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 795                 if (tmp) {
 796                         /* Ignore deprecated mount option.  The client will
 797                          * always try to mount with ACL support, whether this
 798                          * is used depends on whether server supports it. */
 799                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 800                                                   "mount option 'acl'.\n");
 801                         goto next;
 802                 }
 803                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 804                 if (tmp) {
 805                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 806                                                   "mount option 'noacl'.\n");
 807                         goto next;
 808                 }
 809 #else
 810 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 811 #endif
 812                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 823                 if (tmp) {
 824                         *flags &= ~tmp;
 825                         goto next;
 826                 }
 827
 828                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 859                 if (tmp) {
 860                         *flags |= tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 879                                    s1);
 880                 RETURN(-EINVAL);
 881
 882 next:
 883                 /* Find next opt */
 884                 s2 = strchr(s1, ',');
 885                 if (s2 == NULL)
 886                         break;
 887                 s1 = s2 + 1;
 888         }
 889         RETURN(0);
 890 }
 891
 892 void ll_lli_init(struct ll_inode_info *lli)
 893 {
 894         lli->lli_inode_magic = LLI_INODE_MAGIC;
 895         lli->lli_flags = 0;
 896         lli->lli_ioepoch = 0;
 897         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 898         spin_lock_init(&lli->lli_lock);
 899         lli->lli_posix_acl = NULL;
 900         lli->lli_remote_perms = NULL;
 901         mutex_init(&lli->lli_rmtperm_mutex);
 902         /* Do not set lli_fid, it has been initialized already. */
 903         fid_zero(&lli->lli_pfid);
 904         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 905         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 906         cfs_atomic_set(&lli->lli_open_count, 0);
 907         lli->lli_mds_capa = NULL;
 908         lli->lli_rmtperm_time = 0;
 909         lli->lli_pending_och = NULL;
 910         lli->lli_mds_read_och = NULL;
 911         lli->lli_mds_write_och = NULL;
 912         lli->lli_mds_exec_och = NULL;
 913         lli->lli_open_fd_read_count = 0;
 914         lli->lli_open_fd_write_count = 0;
 915         lli->lli_open_fd_exec_count = 0;
 916         mutex_init(&lli->lli_och_mutex);
 917         spin_lock_init(&lli->lli_agl_lock);
 918         lli->lli_has_smd = false;
 919         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 920         lli->lli_clob = NULL;
 921
 922         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 923         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 924                 mutex_init(&lli->lli_readdir_mutex);
 925                 lli->lli_opendir_key = NULL;
 926                 lli->lli_sai = NULL;
 927                 lli->lli_def_acl = NULL;
 928                 spin_lock_init(&lli->lli_sa_lock);
 929                 lli->lli_opendir_pid = 0;
 930         } else {
 931                 sema_init(&lli->lli_size_sem, 1);
 932                 lli->lli_size_sem_owner = NULL;
 933                 lli->lli_symlink_name = NULL;
 934                 init_rwsem(&lli->lli_trunc_sem);
 935                 mutex_init(&lli->lli_write_mutex);
 936                 init_rwsem(&lli->lli_glimpse_sem);
 937                 lli->lli_glimpse_time = 0;
 938                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 939                 lli->lli_agl_index = 0;
 940                 lli->lli_async_rc = 0;
 941                 lli->lli_volatile = false;
 942         }
 943         mutex_init(&lli->lli_layout_mutex);
 944 }
 945
 946 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 947 {
 948 #ifdef HAVE_BDI_REGISTER
 949         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 950
 951 #ifdef HAVE_BDI_NAME
 952         bdi->name = "lustre";
 953 #endif
 954         return bdi_register(bdi, NULL, "lustre-%d",
 955                             atomic_inc_return(&ll_bdi_num));
 956 #else
 957         return 0;
 958 #endif
 959 }
 960
 961 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 962 {
 963         struct lustre_profile *lprof = NULL;
 964         struct lustre_sb_info *lsi = s2lsi(sb);
 965         struct ll_sb_info *sbi;
 966         char  *dt = NULL, *md = NULL;
 967         char  *profilenm = get_profile_name(sb);
 968         struct config_llog_instance *cfg;
 969         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 970         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 971         int    err;
 972         ENTRY;
 973
 974         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 975
 976         OBD_ALLOC_PTR(cfg);
 977         if (cfg == NULL)
 978                 RETURN(-ENOMEM);
 979
 980         cfs_module_get();
 981
 982         /* client additional sb info */
 983         lsi->lsi_llsbi = sbi = ll_init_sbi();
 984         if (!sbi) {
 985                 cfs_module_put(THIS_MODULE);
 986                 OBD_FREE_PTR(cfg);
 987                 RETURN(-ENOMEM);
 988         }
 989
 990         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 991         if (err)
 992                 GOTO(out_free, err);
 993
 994         err = bdi_init(&lsi->lsi_bdi);
 995         if (err)
 996                 GOTO(out_free, err);
 997         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 998         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 999         err = ll_bdi_register(&lsi->lsi_bdi);
1000         if (err)
1001                 GOTO(out_free, err);
1002
1003 #ifdef HAVE_SB_BDI
1004         sb->s_bdi = &lsi->lsi_bdi;
1005 #endif
1006
1007         /* Generate a string unique to this super, in case some joker tries
1008            to mount the same fs at two mount points.
1009            Use the address of the super itself.*/
1010         cfg->cfg_instance = sb;
1011         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1012         cfg->cfg_callback = class_config_llog_handler;
1013         /* set up client obds */
1014         err = lustre_process_log(sb, profilenm, cfg);
1015         if (err < 0) {
1016                 CERROR("Unable to process log: %d\n", err);
1017                 GOTO(out_free, err);
1018         }
1019
1020         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1021         lprof = class_get_profile(profilenm);
1022         if (lprof == NULL) {
1023                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1024                                    " read from the MGS.  Does that filesystem "
1025                                    "exist?\n", profilenm);
1026                 GOTO(out_free, err = -EINVAL);
1027         }
1028         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1029                lprof->lp_md, lprof->lp_dt);
1030
1031         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1032         if (!dt)
1033                 GOTO(out_free, err = -ENOMEM);
1034         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1035
1036         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1037         if (!md)
1038                 GOTO(out_free, err = -ENOMEM);
1039         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1040
1041         /* connections, registrations, sb setup */
1042         err = client_common_fill_super(sb, md, dt, mnt);
1043
1044 out_free:
1045         if (md)
1046                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1047         if (dt)
1048                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1049         if (err)
1050                 ll_put_super(sb);
1051         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1052                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1053
1054         OBD_FREE_PTR(cfg);
1055         RETURN(err);
1056 } /* ll_fill_super */
1057
1058
1059 void lu_context_keys_dump(void);
1060
1061 void ll_put_super(struct super_block *sb)
1062 {
1063         struct config_llog_instance cfg;
1064         struct obd_device *obd;
1065         struct lustre_sb_info *lsi = s2lsi(sb);
1066         struct ll_sb_info *sbi = ll_s2sbi(sb);
1067         char *profilenm = get_profile_name(sb);
1068         int force = 1, next;
1069         ENTRY;
1070
1071         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1072
1073         ll_print_capa_stat(sbi);
1074
1075         cfg.cfg_instance = sb;
1076         lustre_end_log(sb, profilenm, &cfg);
1077
1078         if (sbi->ll_md_exp) {
1079                 obd = class_exp2obd(sbi->ll_md_exp);
1080                 if (obd)
1081                         force = obd->obd_force;
1082         }
1083
1084         /* We need to set force before the lov_disconnect in
1085            lustre_common_put_super, since l_d cleans up osc's as well. */
1086         if (force) {
1087                 next = 0;
1088                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1089                                                      &next)) != NULL) {
1090                         obd->obd_force = force;
1091                 }
1092         }
1093
1094         if (sbi->ll_lcq) {
1095                 /* Only if client_common_fill_super succeeded */
1096                 client_common_put_super(sb);
1097         }
1098
1099         next = 0;
1100         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1101                 class_manual_cleanup(obd);
1102         }
1103
1104         if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1106
1107         if (profilenm)
1108                 class_del_profile(profilenm);
1109
1110         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1111                 bdi_destroy(&lsi->lsi_bdi);
1112                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1113         }
1114
1115         ll_free_sbi(sb);
1116         lsi->lsi_llsbi = NULL;
1117
1118         lustre_common_put_super(sb);
1119
1120         cl_env_cache_purge(~0);
1121
1122         cfs_module_put(THIS_MODULE);
1123
1124         EXIT;
1125 } /* client_put_super */
1126
1127 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1128 {
1129         struct inode *inode = NULL;
1130
1131         /* NOTE: we depend on atomic igrab() -bzzz */
1132         lock_res_and_lock(lock);
1133         if (lock->l_resource->lr_lvb_inode) {
1134                 struct ll_inode_info * lli;
1135                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1136                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1137                         inode = igrab(lock->l_resource->lr_lvb_inode);
1138                 } else {
1139                         inode = lock->l_resource->lr_lvb_inode;
1140                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1141                                          D_WARNING, lock, "lr_lvb_inode %p is "
1142                                          "bogus: magic %08x",
1143                                          lock->l_resource->lr_lvb_inode,
1144                                          lli->lli_inode_magic);
1145                         inode = NULL;
1146                 }
1147         }
1148         unlock_res_and_lock(lock);
1149         return inode;
1150 }
1151
1152 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1153 {
1154         struct inode *inode = NULL;
1155         /* NOTE: we depend on atomic igrab() -bzzz */
1156         lock_res_and_lock(lock);
1157         if (lock->l_ast_data) {
1158                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1159                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1160                         inode = igrab(lock->l_ast_data);
1161                 } else {
1162                         inode = lock->l_ast_data;
1163                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1164                                          D_WARNING, lock, "l_ast_data %p is "
1165                                          "bogus: magic %08x", lock->l_ast_data,
1166                                          lli->lli_inode_magic);
1167                         inode = NULL;
1168                 }
1169         }
1170         unlock_res_and_lock(lock);
1171         return inode;
1172 }
1173
1174 void ll_clear_inode(struct inode *inode)
1175 {
1176         struct ll_inode_info *lli = ll_i2info(inode);
1177         struct ll_sb_info *sbi = ll_i2sbi(inode);
1178         ENTRY;
1179
1180         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1181                inode->i_generation, inode);
1182
1183         if (S_ISDIR(inode->i_mode)) {
1184                 /* these should have been cleared in ll_file_release */
1185                 LASSERT(lli->lli_opendir_key == NULL);
1186                 LASSERT(lli->lli_sai == NULL);
1187                 LASSERT(lli->lli_opendir_pid == 0);
1188         }
1189
1190         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1191         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1192
1193         LASSERT(!lli->lli_open_fd_write_count);
1194         LASSERT(!lli->lli_open_fd_read_count);
1195         LASSERT(!lli->lli_open_fd_exec_count);
1196
1197         if (lli->lli_mds_write_och)
1198                 ll_md_real_close(inode, FMODE_WRITE);
1199         if (lli->lli_mds_exec_och)
1200                 ll_md_real_close(inode, FMODE_EXEC);
1201         if (lli->lli_mds_read_och)
1202                 ll_md_real_close(inode, FMODE_READ);
1203
1204         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1205                 OBD_FREE(lli->lli_symlink_name,
1206                          strlen(lli->lli_symlink_name) + 1);
1207                 lli->lli_symlink_name = NULL;
1208         }
1209
1210         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1211                 LASSERT(lli->lli_posix_acl == NULL);
1212                 if (lli->lli_remote_perms) {
1213                         free_rmtperm_hash(lli->lli_remote_perms);
1214                         lli->lli_remote_perms = NULL;
1215                 }
1216         }
1217 #ifdef CONFIG_FS_POSIX_ACL
1218         else if (lli->lli_posix_acl) {
1219                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1220                 LASSERT(lli->lli_remote_perms == NULL);
1221                 posix_acl_release(lli->lli_posix_acl);
1222                 lli->lli_posix_acl = NULL;
1223         }
1224 #endif
1225         lli->lli_inode_magic = LLI_INODE_DEAD;
1226
1227         ll_clear_inode_capas(inode);
1228         if (!S_ISDIR(inode->i_mode))
1229                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1230
1231         /*
1232          * XXX This has to be done before lsm is freed below, because
1233          * cl_object still uses inode lsm.
1234          */
1235         cl_inode_fini(inode);
1236         lli->lli_has_smd = false;
1237
1238         EXIT;
1239 }
1240
1241 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1242                   struct md_open_data **mod)
1243 {
1244         struct lustre_md md;
1245         struct inode *inode = dentry->d_inode;
1246         struct ll_sb_info *sbi = ll_i2sbi(inode);
1247         struct ptlrpc_request *request = NULL;
1248         int rc, ia_valid;
1249         ENTRY;
1250
1251         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1252                                      LUSTRE_OPC_ANY, NULL);
1253         if (IS_ERR(op_data))
1254                 RETURN(PTR_ERR(op_data));
1255
1256         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1257                         &request, mod);
1258         if (rc) {
1259                 ptlrpc_req_finished(request);
1260                 if (rc == -ENOENT) {
1261                         clear_nlink(inode);
1262                         /* Unlinked special device node? Or just a race?
1263                          * Pretend we done everything. */
1264                         if (!S_ISREG(inode->i_mode) &&
1265                             !S_ISDIR(inode->i_mode)) {
1266                                 ia_valid = op_data->op_attr.ia_valid;
1267                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1268                                 rc = simple_setattr(dentry, &op_data->op_attr);
1269                                 op_data->op_attr.ia_valid = ia_valid;
1270                         }
1271                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1272                         CERROR("md_setattr fails: rc = %d\n", rc);
1273                 }
1274                 RETURN(rc);
1275         }
1276
1277         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1278                               sbi->ll_md_exp, &md);
1279         if (rc) {
1280                 ptlrpc_req_finished(request);
1281                 RETURN(rc);
1282         }
1283
1284         /* We want to adjust timestamps.
1285          * If there is at least some data in file, we cleared ATTR_SIZE
1286          * to avoid update size, otherwise it is important to do.(SOM case)
1287          * (bug 6196) */
1288         ia_valid = op_data->op_attr.ia_valid;
1289         /* Since we set ATTR_*_SET flags above, and already done permission
1290          * check, So don't let inode_change_ok() check it again.  */
1291         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1292         rc = simple_setattr(dentry, &op_data->op_attr);
1293         op_data->op_attr.ia_valid = ia_valid;
1294
1295         /* Extract epoch data if obtained. */
1296         op_data->op_handle = md.body->handle;
1297         op_data->op_ioepoch = md.body->ioepoch;
1298
1299         ll_update_inode(inode, &md);
1300         ptlrpc_req_finished(request);
1301
1302         RETURN(rc);
1303 }
1304
1305 /* Close IO epoch and send Size-on-MDS attribute update. */
1306 static int ll_setattr_done_writing(struct inode *inode,
1307                                    struct md_op_data *op_data,
1308                                    struct md_open_data *mod)
1309 {
1310         struct ll_inode_info *lli = ll_i2info(inode);
1311         int rc = 0;
1312         ENTRY;
1313
1314         LASSERT(op_data != NULL);
1315         if (!S_ISREG(inode->i_mode))
1316                 RETURN(0);
1317
1318         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1319                op_data->op_ioepoch, PFID(&lli->lli_fid));
1320
1321         op_data->op_flags = MF_EPOCH_CLOSE;
1322         ll_done_writing_attr(inode, op_data);
1323         ll_pack_inode2opdata(inode, op_data, NULL);
1324
1325         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1326         if (rc == -EAGAIN) {
1327                 /* MDS has instructed us to obtain Size-on-MDS attribute
1328                  * from OSTs and send setattr to back to MDS. */
1329                 rc = ll_som_update(inode, op_data);
1330         } else if (rc) {
1331                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1332                        inode->i_ino, rc);
1333         }
1334         RETURN(rc);
1335 }
1336
1337 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1338 {
1339         struct obd_capa *capa;
1340         int rc;
1341
1342         if (attr->ia_valid & ATTR_SIZE)
1343                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1344         else
1345                 capa = ll_mdscapa_get(inode);
1346
1347         rc = cl_setattr_ost(inode, attr, capa);
1348
1349         if (attr->ia_valid & ATTR_SIZE)
1350                 ll_truncate_free_capa(capa);
1351         else
1352                 capa_put(capa);
1353
1354         return rc;
1355 }
1356
1357 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1358 /**
1359  * inode_newsize_ok - may this inode be truncated to a given size
1360  * @inode:      the inode to be truncated
1361  * @offset:     the new size to assign to the inode
1362  * @Returns:    0 on success, -ve errno on failure
1363  *
1364  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1365  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1366  * when necessary. Caller must not proceed with inode size change if failure is
1367  * returned. @inode must be a file (not directory), with appropriate
1368  * permissions to allow truncate (inode_newsize_ok does NOT check these
1369  * conditions).
1370  *
1371  * inode_newsize_ok must be called with i_mutex held.
1372  */
1373 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1374 {
1375         if (inode->i_size < offset) {
1376                 unsigned long limit;
1377
1378                 limit = rlimit(RLIMIT_FSIZE);
1379                 if (limit != RLIM_INFINITY && offset > limit)
1380                         goto out_sig;
1381                 if (offset > inode->i_sb->s_maxbytes)
1382                         goto out_big;
1383         } else {
1384                 /*
1385                  * truncation of in-use swapfiles is disallowed - it would
1386                  * cause subsequent swapout to scribble on the now-freed
1387                  * blocks.
1388                  */
1389                 if (IS_SWAPFILE(inode))
1390                         return -ETXTBSY;
1391         }
1392
1393         return 0;
1394 out_sig:
1395         send_sig(SIGXFSZ, current, 0);
1396 out_big:
1397         return -EFBIG;
1398 }
1399 #endif
1400
1401 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1402  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1403  * keep these values until such a time that objects are allocated for it.
1404  * We do the MDS operations first, as it is checking permissions for us.
1405  * We don't to the MDS RPC if there is nothing that we want to store there,
1406  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1407  * going to do an RPC anyways.
1408  *
1409  * If we are doing a truncate, we will send the mtime and ctime updates
1410  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1411  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1412  * at the same time.
1413  */
1414 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1415 {
1416         struct lov_stripe_md *lsm;
1417         struct inode *inode = dentry->d_inode;
1418         struct ll_inode_info *lli = ll_i2info(inode);
1419         struct md_op_data *op_data = NULL;
1420         struct md_open_data *mod = NULL;
1421         int ia_valid = attr->ia_valid;
1422         int rc = 0, rc1 = 0;
1423         ENTRY;
1424
1425         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1426                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1427                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1428                 attr->ia_valid);
1429
1430         if (ia_valid & ATTR_SIZE) {
1431                 /* Check new size against VFS/VM file size limit and rlimit */
1432                 rc = inode_newsize_ok(inode, attr->ia_size);
1433                 if (rc)
1434                         RETURN(rc);
1435
1436                 /* The maximum Lustre file size is variable, based on the
1437                  * OST maximum object size and number of stripes.  This
1438                  * needs another check in addition to the VFS check above. */
1439                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1440                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1441                                PFID(&lli->lli_fid), attr->ia_size,
1442                                ll_file_maxbytes(inode));
1443                         RETURN(-EFBIG);
1444                 }
1445
1446                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1447         }
1448
1449         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1450         if (ia_valid & TIMES_SET_FLAGS) {
1451                 if (cfs_curproc_fsuid() != inode->i_uid &&
1452                     !cfs_capable(CFS_CAP_FOWNER))
1453                         RETURN(-EPERM);
1454         }
1455
1456         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1457         if (attr->ia_valid & ATTR_CTIME) {
1458                 attr->ia_ctime = CFS_CURRENT_TIME;
1459                 attr->ia_valid |= ATTR_CTIME_SET;
1460         }
1461         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1462                 attr->ia_atime = CFS_CURRENT_TIME;
1463                 attr->ia_valid |= ATTR_ATIME_SET;
1464         }
1465         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1466                 attr->ia_mtime = CFS_CURRENT_TIME;
1467                 attr->ia_valid |= ATTR_MTIME_SET;
1468         }
1469
1470         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1471                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1472                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1473                        cfs_time_current_sec());
1474
1475         /* If we are changing file size, file content is modified, flag it. */
1476         if (attr->ia_valid & ATTR_SIZE) {
1477                 spin_lock(&lli->lli_lock);
1478                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1479                 spin_unlock(&lli->lli_lock);
1480         }
1481
1482         /* We always do an MDS RPC, even if we're only changing the size;
1483          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1484
1485         OBD_ALLOC_PTR(op_data);
1486         if (op_data == NULL)
1487                 RETURN(-ENOMEM);
1488
1489         if (!S_ISDIR(inode->i_mode)) {
1490                 if (ia_valid & ATTR_SIZE)
1491                         inode_dio_write_done(inode);
1492                 mutex_unlock(&inode->i_mutex);
1493                 down_write(&lli->lli_trunc_sem);
1494         }
1495
1496         /* We need a steady stripe configuration for setattr to avoid
1497          * confusion. */
1498         lsm = ccc_inode_lsm_get(inode);
1499
1500         /* NB: ATTR_SIZE will only be set after this point if the size
1501          * resides on the MDS, ie, this file has no objects. */
1502         if (lsm != NULL)
1503                 attr->ia_valid &= ~ATTR_SIZE;
1504         /* can't call ll_setattr_ost() while holding a refcount of lsm */
1505         ccc_inode_lsm_put(inode, lsm);
1506
1507         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1508
1509         /* Open epoch for truncate. */
1510         if (exp_connect_som(ll_i2mdexp(inode)) &&
1511             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1512                 op_data->op_flags = MF_EPOCH_OPEN;
1513
1514         rc = ll_md_setattr(dentry, op_data, &mod);
1515         if (rc)
1516                 GOTO(out, rc);
1517
1518         /* RPC to MDT is sent, cancel data modification flag */
1519         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1520                 spin_lock(&lli->lli_lock);
1521                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1522                 spin_unlock(&lli->lli_lock);
1523         }
1524
1525         ll_ioepoch_open(lli, op_data->op_ioepoch);
1526         if (!S_ISREG(inode->i_mode))
1527                 GOTO(out, rc = 0);
1528
1529         if (ia_valid & ATTR_SIZE)
1530                 attr->ia_valid |= ATTR_SIZE;
1531         if (ia_valid & (ATTR_SIZE |
1532                         ATTR_ATIME | ATTR_ATIME_SET |
1533                         ATTR_MTIME | ATTR_MTIME_SET))
1534                 /* For truncate and utimes sending attributes to OSTs, setting
1535                  * mtime/atime to the past will be performed under PW [0:EOF]
1536                  * extent lock (new_size:EOF for truncate).  It may seem
1537                  * excessive to send mtime/atime updates to OSTs when not
1538                  * setting times to past, but it is necessary due to possible
1539                  * time de-synchronization between MDT inode and OST objects */
1540                 rc = ll_setattr_ost(inode, attr);
1541         EXIT;
1542 out:
1543         if (op_data) {
1544                 if (op_data->op_ioepoch) {
1545                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1546                         if (!rc)
1547                                 rc = rc1;
1548                 }
1549                 ll_finish_md_op_data(op_data);
1550         }
1551         if (!S_ISDIR(inode->i_mode)) {
1552                 up_write(&lli->lli_trunc_sem);
1553                 mutex_lock(&inode->i_mutex);
1554                 if (ia_valid & ATTR_SIZE)
1555                         inode_dio_wait(inode);
1556         }
1557
1558         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1559                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1560
1561         return rc;
1562 }
1563
1564 int ll_setattr(struct dentry *de, struct iattr *attr)
1565 {
1566         int mode = de->d_inode->i_mode;
1567
1568         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1569                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1570                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1571
1572         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1573                                (ATTR_SIZE|ATTR_MODE)) &&
1574             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1575              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1576               !(attr->ia_mode & S_ISGID))))
1577                 attr->ia_valid |= ATTR_FORCE;
1578
1579         if ((mode & S_ISUID) &&
1580             !(attr->ia_mode & S_ISUID) &&
1581             !(attr->ia_valid & ATTR_KILL_SUID))
1582                 attr->ia_valid |= ATTR_KILL_SUID;
1583
1584         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1585             !(attr->ia_mode & S_ISGID) &&
1586             !(attr->ia_valid & ATTR_KILL_SGID))
1587                 attr->ia_valid |= ATTR_KILL_SGID;
1588
1589         return ll_setattr_raw(de, attr);
1590 }
1591
1592 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1593                        __u64 max_age, __u32 flags)
1594 {
1595         struct ll_sb_info *sbi = ll_s2sbi(sb);
1596         struct obd_statfs obd_osfs;
1597         int rc;
1598         ENTRY;
1599
1600         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1601         if (rc) {
1602                 CERROR("md_statfs fails: rc = %d\n", rc);
1603                 RETURN(rc);
1604         }
1605
1606         osfs->os_type = sb->s_magic;
1607
1608         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1609                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1610
1611         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1612                 flags |= OBD_STATFS_NODELAY;
1613
1614         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1615         if (rc) {
1616                 CERROR("obd_statfs fails: rc = %d\n", rc);
1617                 RETURN(rc);
1618         }
1619
1620         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1621                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1622                obd_osfs.os_files);
1623
1624         osfs->os_bsize = obd_osfs.os_bsize;
1625         osfs->os_blocks = obd_osfs.os_blocks;
1626         osfs->os_bfree = obd_osfs.os_bfree;
1627         osfs->os_bavail = obd_osfs.os_bavail;
1628
1629         /* If we don't have as many objects free on the OST as inodes
1630          * on the MDS, we reduce the total number of inodes to
1631          * compensate, so that the "inodes in use" number is correct.
1632          */
1633         if (obd_osfs.os_ffree < osfs->os_ffree) {
1634                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1635                         obd_osfs.os_ffree;
1636                 osfs->os_ffree = obd_osfs.os_ffree;
1637         }
1638
1639         RETURN(rc);
1640 }
1641 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1642 {
1643         struct super_block *sb = de->d_sb;
1644         struct obd_statfs osfs;
1645         int rc;
1646
1647         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1648         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1649
1650         /* Some amount of caching on the client is allowed */
1651         rc = ll_statfs_internal(sb, &osfs,
1652                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1653                                 0);
1654         if (rc)
1655                 return rc;
1656
1657         statfs_unpack(sfs, &osfs);
1658
1659         /* We need to downshift for all 32-bit kernels, because we can't
1660          * tell if the kernel is being called via sys_statfs64() or not.
1661          * Stop before overflowing f_bsize - in which case it is better
1662          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1663         if (sizeof(long) < 8) {
1664                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1665                         sfs->f_bsize <<= 1;
1666
1667                         osfs.os_blocks >>= 1;
1668                         osfs.os_bfree >>= 1;
1669                         osfs.os_bavail >>= 1;
1670                 }
1671         }
1672
1673         sfs->f_blocks = osfs.os_blocks;
1674         sfs->f_bfree = osfs.os_bfree;
1675         sfs->f_bavail = osfs.os_bavail;
1676
1677         return 0;
1678 }
1679
1680 void ll_inode_size_lock(struct inode *inode)
1681 {
1682         struct ll_inode_info *lli;
1683
1684         LASSERT(!S_ISDIR(inode->i_mode));
1685
1686         lli = ll_i2info(inode);
1687         LASSERT(lli->lli_size_sem_owner != current);
1688         down(&lli->lli_size_sem);
1689         LASSERT(lli->lli_size_sem_owner == NULL);
1690         lli->lli_size_sem_owner = current;
1691 }
1692
1693 void ll_inode_size_unlock(struct inode *inode)
1694 {
1695         struct ll_inode_info *lli;
1696
1697         lli = ll_i2info(inode);
1698         LASSERT(lli->lli_size_sem_owner == current);
1699         lli->lli_size_sem_owner = NULL;
1700         up(&lli->lli_size_sem);
1701 }
1702
1703 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1704 {
1705         struct ll_inode_info *lli = ll_i2info(inode);
1706         struct mdt_body *body = md->body;
1707         struct lov_stripe_md *lsm = md->lsm;
1708         struct ll_sb_info *sbi = ll_i2sbi(inode);
1709
1710         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1711         if (lsm != NULL) {
1712                 if (!lli->lli_has_smd &&
1713                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1714                         cl_file_inode_init(inode, md);
1715
1716                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1717                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1718                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1719         }
1720
1721         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1722                 if (body->valid & OBD_MD_FLRMTPERM)
1723                         ll_update_remote_perm(inode, md->remote_perm);
1724         }
1725 #ifdef CONFIG_FS_POSIX_ACL
1726         else if (body->valid & OBD_MD_FLACL) {
1727                 spin_lock(&lli->lli_lock);
1728                 if (lli->lli_posix_acl)
1729                         posix_acl_release(lli->lli_posix_acl);
1730                 lli->lli_posix_acl = md->posix_acl;
1731                 spin_unlock(&lli->lli_lock);
1732         }
1733 #endif
1734         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1735         inode->i_generation = cl_fid_build_gen(&body->fid1);
1736
1737         if (body->valid & OBD_MD_FLATIME) {
1738                 if (body->atime > LTIME_S(inode->i_atime))
1739                         LTIME_S(inode->i_atime) = body->atime;
1740                 lli->lli_lvb.lvb_atime = body->atime;
1741         }
1742         if (body->valid & OBD_MD_FLMTIME) {
1743                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1744                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1745                                "to "LPU64"\n", inode->i_ino,
1746                                LTIME_S(inode->i_mtime), body->mtime);
1747                         LTIME_S(inode->i_mtime) = body->mtime;
1748                 }
1749                 lli->lli_lvb.lvb_mtime = body->mtime;
1750         }
1751         if (body->valid & OBD_MD_FLCTIME) {
1752                 if (body->ctime > LTIME_S(inode->i_ctime))
1753                         LTIME_S(inode->i_ctime) = body->ctime;
1754                 lli->lli_lvb.lvb_ctime = body->ctime;
1755         }
1756         if (body->valid & OBD_MD_FLMODE)
1757                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1758         if (body->valid & OBD_MD_FLTYPE)
1759                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1760         LASSERT(inode->i_mode != 0);
1761         if (S_ISREG(inode->i_mode)) {
1762                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1763         } else {
1764                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1765         }
1766         if (body->valid & OBD_MD_FLUID)
1767                 inode->i_uid = body->uid;
1768         if (body->valid & OBD_MD_FLGID)
1769                 inode->i_gid = body->gid;
1770         if (body->valid & OBD_MD_FLFLAGS)
1771                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1772         if (body->valid & OBD_MD_FLNLINK)
1773                 set_nlink(inode, body->nlink);
1774         if (body->valid & OBD_MD_FLRDEV)
1775                 inode->i_rdev = old_decode_dev(body->rdev);
1776
1777         if (body->valid & OBD_MD_FLID) {
1778                 /* FID shouldn't be changed! */
1779                 if (fid_is_sane(&lli->lli_fid)) {
1780                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1781                                  "Trying to change FID "DFID
1782                                  " to the "DFID", inode %lu/%u(%p)\n",
1783                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1784                                  inode->i_ino, inode->i_generation, inode);
1785                 } else
1786                         lli->lli_fid = body->fid1;
1787         }
1788
1789         LASSERT(fid_seq(&lli->lli_fid) != 0);
1790
1791         if (body->valid & OBD_MD_FLSIZE) {
1792                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1793                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1794                         struct lustre_handle lockh;
1795                         ldlm_mode_t mode;
1796
1797                         /* As it is possible a blocking ast has been processed
1798                          * by this time, we need to check there is an UPDATE
1799                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1800                          * it. */
1801                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1802                                                &lockh, LDLM_FL_CBPENDING);
1803                         if (mode) {
1804                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1805                                                       LLIF_EPOCH_PENDING |
1806                                                       LLIF_SOM_DIRTY)) {
1807                                         CERROR("ino %lu flags %u still has "
1808                                                "size authority! do not trust "
1809                                                "the size got from MDS\n",
1810                                                inode->i_ino, lli->lli_flags);
1811                                 } else {
1812                                         /* Use old size assignment to avoid
1813                                          * deadlock bz14138 & bz14326 */
1814                                         i_size_write(inode, body->size);
1815                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1816                                 }
1817                                 ldlm_lock_decref(&lockh, mode);
1818                         }
1819                 } else {
1820                         /* Use old size assignment to avoid
1821                          * deadlock bz14138 & bz14326 */
1822                         i_size_write(inode, body->size);
1823
1824                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1825                                inode->i_ino, (unsigned long long)body->size);
1826                 }
1827
1828                 if (body->valid & OBD_MD_FLBLOCKS)
1829                         inode->i_blocks = body->blocks;
1830         }
1831
1832         if (body->valid & OBD_MD_FLMDSCAPA) {
1833                 LASSERT(md->mds_capa);
1834                 ll_add_capa(inode, md->mds_capa);
1835         }
1836         if (body->valid & OBD_MD_FLOSSCAPA) {
1837                 LASSERT(md->oss_capa);
1838                 ll_add_capa(inode, md->oss_capa);
1839         }
1840 }
1841
1842 void ll_read_inode2(struct inode *inode, void *opaque)
1843 {
1844         struct lustre_md *md = opaque;
1845         struct ll_inode_info *lli = ll_i2info(inode);
1846         ENTRY;
1847
1848         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1849                PFID(&lli->lli_fid), inode);
1850
1851         LASSERT(!lli->lli_has_smd);
1852
1853         /* Core attributes from the MDS first.  This is a new inode, and
1854          * the VFS doesn't zero times in the core inode so we have to do
1855          * it ourselves.  They will be overwritten by either MDS or OST
1856          * attributes - we just need to make sure they aren't newer. */
1857         LTIME_S(inode->i_mtime) = 0;
1858         LTIME_S(inode->i_atime) = 0;
1859         LTIME_S(inode->i_ctime) = 0;
1860         inode->i_rdev = 0;
1861         ll_update_inode(inode, md);
1862
1863         /* OIDEBUG(inode); */
1864
1865         /* initializing backing dev info. */
1866         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1867
1868
1869         if (S_ISREG(inode->i_mode)) {
1870                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1871                 inode->i_op = &ll_file_inode_operations;
1872                 inode->i_fop = sbi->ll_fop;
1873                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1874                 EXIT;
1875         } else if (S_ISDIR(inode->i_mode)) {
1876                 inode->i_op = &ll_dir_inode_operations;
1877                 inode->i_fop = &ll_dir_operations;
1878                 EXIT;
1879         } else if (S_ISLNK(inode->i_mode)) {
1880                 inode->i_op = &ll_fast_symlink_inode_operations;
1881                 EXIT;
1882         } else {
1883                 inode->i_op = &ll_special_inode_operations;
1884
1885                 init_special_inode(inode, inode->i_mode,
1886                                    inode->i_rdev);
1887
1888                 EXIT;
1889         }
1890 }
1891
1892 void ll_delete_inode(struct inode *inode)
1893 {
1894         struct cl_inode_info *lli = cl_i2info(inode);
1895         ENTRY;
1896
1897         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1898                 /* discard all dirty pages before truncating them, required by
1899                  * osc_extent implementation at LU-1030. */
1900                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1901
1902         truncate_inode_pages(&inode->i_data, 0);
1903
1904         /* Workaround for LU-118 */
1905         if (inode->i_data.nrpages) {
1906                 TREE_READ_LOCK_IRQ(&inode->i_data);
1907                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1908                 LASSERTF(inode->i_data.nrpages == 0,
1909                          "inode=%lu/%u(%p) nrpages=%lu, see "
1910                          "http://jira.whamcloud.com/browse/LU-118\n",
1911                          inode->i_ino, inode->i_generation, inode,
1912                          inode->i_data.nrpages);
1913         }
1914         /* Workaround end */
1915
1916 #ifdef HAVE_SBOPS_EVICT_INODE
1917         ll_clear_inode(inode);
1918 #endif
1919         clear_inode(inode);
1920
1921         EXIT;
1922 }
1923
1924 int ll_iocontrol(struct inode *inode, struct file *file,
1925                  unsigned int cmd, unsigned long arg)
1926 {
1927         struct ll_sb_info *sbi = ll_i2sbi(inode);
1928         struct ptlrpc_request *req = NULL;
1929         int rc, flags = 0;
1930         ENTRY;
1931
1932         switch(cmd) {
1933         case FSFILT_IOC_GETFLAGS: {
1934                 struct mdt_body *body;
1935                 struct md_op_data *op_data;
1936
1937                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1938                                              0, 0, LUSTRE_OPC_ANY,
1939                                              NULL);
1940                 if (IS_ERR(op_data))
1941                         RETURN(PTR_ERR(op_data));
1942
1943                 op_data->op_valid = OBD_MD_FLFLAGS;
1944                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1945                 ll_finish_md_op_data(op_data);
1946                 if (rc) {
1947                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1948                         RETURN(-abs(rc));
1949                 }
1950
1951                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1952
1953                 flags = body->flags;
1954
1955                 ptlrpc_req_finished(req);
1956
1957                 RETURN(put_user(flags, (int *)arg));
1958         }
1959         case FSFILT_IOC_SETFLAGS: {
1960                 struct lov_stripe_md *lsm;
1961                 struct obd_info oinfo = { { { 0 } } };
1962                 struct md_op_data *op_data;
1963
1964                 if (get_user(flags, (int *)arg))
1965                         RETURN(-EFAULT);
1966
1967                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1968                                              LUSTRE_OPC_ANY, NULL);
1969                 if (IS_ERR(op_data))
1970                         RETURN(PTR_ERR(op_data));
1971
1972                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1973                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1974                 rc = md_setattr(sbi->ll_md_exp, op_data,
1975                                 NULL, 0, NULL, 0, &req, NULL);
1976                 ll_finish_md_op_data(op_data);
1977                 ptlrpc_req_finished(req);
1978                 if (rc)
1979                         RETURN(rc);
1980
1981                 inode->i_flags = ll_ext_to_inode_flags(flags);
1982
1983                 lsm = ccc_inode_lsm_get(inode);
1984                 if (lsm == NULL)
1985                         RETURN(0);
1986
1987                 OBDO_ALLOC(oinfo.oi_oa);
1988                 if (!oinfo.oi_oa) {
1989                         ccc_inode_lsm_put(inode, lsm);
1990                         RETURN(-ENOMEM);
1991                 }
1992                 oinfo.oi_md = lsm;
1993                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1994                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1995                 oinfo.oi_oa->o_flags = flags;
1996                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1997                                        OBD_MD_FLGROUP;
1998                 oinfo.oi_capa = ll_mdscapa_get(inode);
1999                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2000                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2001                 capa_put(oinfo.oi_capa);
2002                 OBDO_FREE(oinfo.oi_oa);
2003                 ccc_inode_lsm_put(inode, lsm);
2004
2005                 if (rc && rc != -EPERM && rc != -EACCES)
2006                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2007
2008                 RETURN(rc);
2009         }
2010         default:
2011                 RETURN(-ENOSYS);
2012         }
2013
2014         RETURN(0);
2015 }
2016
2017 int ll_flush_ctx(struct inode *inode)
2018 {
2019         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2020
2021         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2022
2023         obd_set_info_async(NULL, sbi->ll_md_exp,
2024                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2025                            0, NULL, NULL);
2026         obd_set_info_async(NULL, sbi->ll_dt_exp,
2027                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2028                            0, NULL, NULL);
2029         return 0;
2030 }
2031
2032 /* umount -f client means force down, don't save state */
2033 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2034 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2035 {
2036         struct super_block *sb = vfsmnt->mnt_sb;
2037 #else
2038 void ll_umount_begin(struct super_block *sb)
2039 {
2040 #endif
2041         struct ll_sb_info *sbi = ll_s2sbi(sb);
2042         struct obd_device *obd;
2043         struct obd_ioctl_data *ioc_data;
2044         ENTRY;
2045
2046 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2047         if (!(flags & MNT_FORCE)) {
2048                 EXIT;
2049                 return;
2050         }
2051 #endif
2052
2053         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2054                sb->s_count, atomic_read(&sb->s_active));
2055
2056         obd = class_exp2obd(sbi->ll_md_exp);
2057         if (obd == NULL) {
2058                 CERROR("Invalid MDC connection handle "LPX64"\n",
2059                        sbi->ll_md_exp->exp_handle.h_cookie);
2060                 EXIT;
2061                 return;
2062         }
2063         obd->obd_force = 1;
2064
2065         obd = class_exp2obd(sbi->ll_dt_exp);
2066         if (obd == NULL) {
2067                 CERROR("Invalid LOV connection handle "LPX64"\n",
2068                        sbi->ll_dt_exp->exp_handle.h_cookie);
2069                 EXIT;
2070                 return;
2071         }
2072         obd->obd_force = 1;
2073
2074         OBD_ALLOC_PTR(ioc_data);
2075         if (ioc_data) {
2076                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2077                               sizeof ioc_data, ioc_data, NULL);
2078
2079                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2080                               sizeof ioc_data, ioc_data, NULL);
2081
2082                 OBD_FREE_PTR(ioc_data);
2083         }
2084
2085
2086         /* Really, we'd like to wait until there are no requests outstanding,
2087          * and then continue.  For now, we just invalidate the requests,
2088          * schedule() and sleep one second if needed, and hope.
2089          */
2090         cfs_schedule();
2091 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2092         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2093                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2094                                                    cfs_time_seconds(1));
2095                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2096                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2097                                       "may try to umount it a bit later\n",
2098                                       atomic_read(&vfsmnt->mnt_count));
2099         }
2100 #endif
2101
2102         EXIT;
2103 }
2104
2105 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2106 {
2107         struct ll_sb_info *sbi = ll_s2sbi(sb);
2108         char *profilenm = get_profile_name(sb);
2109         int err;
2110         __u32 read_only;
2111
2112         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2113                 read_only = *flags & MS_RDONLY;
2114                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2115                                          sizeof(KEY_READ_ONLY),
2116                                          KEY_READ_ONLY, sizeof(read_only),
2117                                          &read_only, NULL);
2118                 if (err) {
2119                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2120                                       profilenm, read_only ?
2121                                       "read-only" : "read-write", err);
2122                         return err;
2123                 }
2124
2125                 if (read_only)
2126                         sb->s_flags |= MS_RDONLY;
2127                 else
2128                         sb->s_flags &= ~MS_RDONLY;
2129
2130                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2131                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2132                                       read_only ?  "read-only" : "read-write");
2133         }
2134         return 0;
2135 }
2136
2137 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2138                   struct super_block *sb, struct lookup_intent *it)
2139 {
2140         struct ll_sb_info *sbi = NULL;
2141         struct lustre_md md;
2142         int rc;
2143         ENTRY;
2144
2145         LASSERT(*inode || sb);
2146         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2147         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2148                               sbi->ll_md_exp, &md);
2149         if (rc)
2150                 RETURN(rc);
2151
2152         if (*inode) {
2153                 ll_update_inode(*inode, &md);
2154         } else {
2155                 LASSERT(sb != NULL);
2156
2157                 /*
2158                  * At this point server returns to client's same fid as client
2159                  * generated for creating. So using ->fid1 is okay here.
2160                  */
2161                 LASSERT(fid_is_sane(&md.body->fid1));
2162
2163                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2164                 if (*inode == NULL || IS_ERR(*inode)) {
2165 #ifdef CONFIG_FS_POSIX_ACL
2166                         if (md.posix_acl) {
2167                                 posix_acl_release(md.posix_acl);
2168                                 md.posix_acl = NULL;
2169                         }
2170 #endif
2171                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2172                         *inode = NULL;
2173                         CERROR("new_inode -fatal: rc %d\n", rc);
2174                         GOTO(out, rc);
2175                 }
2176         }
2177
2178         /* Handling piggyback layout lock.
2179          * Layout lock can be piggybacked by getattr and open request.
2180          * The lsm can be applied to inode only if it comes with a layout lock
2181          * otherwise correct layout may be overwritten, for example:
2182          * 1. proc1: mdt returns a lsm but not granting layout
2183          * 2. layout was changed by another client
2184          * 3. proc2: refresh layout and layout lock granted
2185          * 4. proc1: to apply a stale layout */
2186         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2187                 struct lustre_handle lockh;
2188                 struct ldlm_lock *lock;
2189
2190                 lockh.cookie = it->d.lustre.it_lock_handle;
2191                 lock = ldlm_handle2lock(&lockh);
2192                 LASSERT(lock != NULL);
2193                 if (ldlm_has_layout(lock)) {
2194                         struct cl_object_conf conf;
2195
2196                         memset(&conf, 0, sizeof(conf));
2197                         conf.coc_opc = OBJECT_CONF_SET;
2198                         conf.coc_inode = *inode;
2199                         conf.coc_lock = lock;
2200                         conf.u.coc_md = &md;
2201                         (void)ll_layout_conf(*inode, &conf);
2202                 }
2203                 LDLM_LOCK_PUT(lock);
2204         }
2205
2206 out:
2207         if (md.lsm != NULL)
2208                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2209         md_free_lustre_md(sbi->ll_md_exp, &md);
2210         RETURN(rc);
2211 }
2212
2213 int ll_obd_statfs(struct inode *inode, void *arg)
2214 {
2215         struct ll_sb_info *sbi = NULL;
2216         struct obd_export *exp;
2217         char *buf = NULL;
2218         struct obd_ioctl_data *data = NULL;
2219         __u32 type;
2220         __u32 flags;
2221         int len = 0, rc;
2222
2223         if (!inode || !(sbi = ll_i2sbi(inode)))
2224                 GOTO(out_statfs, rc = -EINVAL);
2225
2226         rc = obd_ioctl_getdata(&buf, &len, arg);
2227         if (rc)
2228                 GOTO(out_statfs, rc);
2229
2230         data = (void*)buf;
2231         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2232             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2233                 GOTO(out_statfs, rc = -EINVAL);
2234
2235         if (data->ioc_inllen1 != sizeof(__u32) ||
2236             data->ioc_inllen2 != sizeof(__u32) ||
2237             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2238             data->ioc_plen2 != sizeof(struct obd_uuid))
2239                 GOTO(out_statfs, rc = -EINVAL);
2240
2241         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2242         if (type & LL_STATFS_LMV)
2243                 exp = sbi->ll_md_exp;
2244         else if (type & LL_STATFS_LOV)
2245                 exp = sbi->ll_dt_exp;
2246         else
2247                 GOTO(out_statfs, rc = -ENODEV);
2248
2249         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2250         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2251         if (rc)
2252                 GOTO(out_statfs, rc);
2253 out_statfs:
2254         if (buf)
2255                 obd_ioctl_freedata(buf, len);
2256         return rc;
2257 }
2258
2259 int ll_process_config(struct lustre_cfg *lcfg)
2260 {
2261         char *ptr;
2262         void *sb;
2263         struct lprocfs_static_vars lvars;
2264         unsigned long x;
2265         int rc = 0;
2266
2267         lprocfs_llite_init_vars(&lvars);
2268
2269         /* The instance name contains the sb: lustre-client-aacfe000 */
2270         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2271         if (!ptr || !*(++ptr))
2272                 return -EINVAL;
2273         if (sscanf(ptr, "%lx", &x) != 1)
2274                 return -EINVAL;
2275         sb = (void *)x;
2276         /* This better be a real Lustre superblock! */
2277         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2278
2279         /* Note we have not called client_common_fill_super yet, so
2280            proc fns must be able to handle that! */
2281         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2282                                       lcfg, sb);
2283         if (rc > 0)
2284                 rc = 0;
2285         return(rc);
2286 }
2287
2288 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2289 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2290                                        struct inode *i1, struct inode *i2,
2291                                        const char *name, int namelen,
2292                                        int mode, __u32 opc, void *data)
2293 {
2294         LASSERT(i1 != NULL);
2295
2296         if (namelen > ll_i2sbi(i1)->ll_namelen)
2297                 return ERR_PTR(-ENAMETOOLONG);
2298
2299         if (op_data == NULL)
2300                 OBD_ALLOC_PTR(op_data);
2301
2302         if (op_data == NULL)
2303                 return ERR_PTR(-ENOMEM);
2304
2305         ll_i2gids(op_data->op_suppgids, i1, i2);
2306         op_data->op_fid1 = *ll_inode2fid(i1);
2307         op_data->op_capa1 = ll_mdscapa_get(i1);
2308
2309         if (i2) {
2310                 op_data->op_fid2 = *ll_inode2fid(i2);
2311                 op_data->op_capa2 = ll_mdscapa_get(i2);
2312         } else {
2313                 fid_zero(&op_data->op_fid2);
2314                 op_data->op_capa2 = NULL;
2315         }
2316
2317         op_data->op_name = name;
2318         op_data->op_namelen = namelen;
2319         op_data->op_mode = mode;
2320         op_data->op_mod_time = cfs_time_current_sec();
2321         op_data->op_fsuid = cfs_curproc_fsuid();
2322         op_data->op_fsgid = cfs_curproc_fsgid();
2323         op_data->op_cap = cfs_curproc_cap_pack();
2324         op_data->op_bias = 0;
2325         op_data->op_cli_flags = 0;
2326         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2327              filename_is_volatile(name, namelen, NULL))
2328                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2329         op_data->op_opc = opc;
2330         op_data->op_mds = 0;
2331         op_data->op_data = data;
2332
2333         /* If the file is being opened after mknod() (normally due to NFS)
2334          * try to use the default stripe data from parent directory for
2335          * allocating OST objects.  Try to pass the parent FID to MDS. */
2336         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2337             !ll_i2info(i2)->lli_has_smd) {
2338                 struct ll_inode_info *lli = ll_i2info(i2);
2339
2340                 spin_lock(&lli->lli_lock);
2341                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2342                         op_data->op_fid1 = lli->lli_pfid;
2343                 spin_unlock(&lli->lli_lock);
2344                 /** We ignore parent's capability temporary. */
2345         }
2346
2347         /* When called by ll_setattr_raw, file is i1. */
2348         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2349                 op_data->op_bias |= MDS_DATA_MODIFIED;
2350
2351         return op_data;
2352 }
2353
2354 void ll_finish_md_op_data(struct md_op_data *op_data)
2355 {
2356         capa_put(op_data->op_capa1);
2357         capa_put(op_data->op_capa2);
2358         OBD_FREE_PTR(op_data);
2359 }
2360
2361 #ifdef HAVE_SUPEROPS_USE_DENTRY
2362 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2363 #else
2364 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2365 #endif
2366 {
2367         struct ll_sb_info *sbi;
2368
2369 #ifdef HAVE_SUPEROPS_USE_DENTRY
2370         LASSERT((seq != NULL) && (dentry != NULL));
2371         sbi = ll_s2sbi(dentry->d_sb);
2372 #else
2373         LASSERT((seq != NULL) && (vfs != NULL));
2374         sbi = ll_s2sbi(vfs->mnt_sb);
2375 #endif
2376
2377         if (sbi->ll_flags & LL_SBI_NOLCK)
2378                 seq_puts(seq, ",nolock");
2379
2380         if (sbi->ll_flags & LL_SBI_FLOCK)
2381                 seq_puts(seq, ",flock");
2382
2383         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2384                 seq_puts(seq, ",localflock");
2385
2386         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2387                 seq_puts(seq, ",user_xattr");
2388
2389         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2390                 seq_puts(seq, ",lazystatfs");
2391
2392         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2393                 seq_puts(seq, ",user_fid2path");
2394
2395         RETURN(0);
2396 }
2397
2398 /**
2399  * Get obd name by cmd, and copy out to user space
2400  */
2401 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2402 {
2403         struct ll_sb_info *sbi = ll_i2sbi(inode);
2404         struct obd_device *obd;
2405         ENTRY;
2406
2407         if (cmd == OBD_IOC_GETDTNAME)
2408                 obd = class_exp2obd(sbi->ll_dt_exp);
2409         else if (cmd == OBD_IOC_GETMDNAME)
2410                 obd = class_exp2obd(sbi->ll_md_exp);
2411         else
2412                 RETURN(-EINVAL);
2413
2414         if (!obd)
2415                 RETURN(-ENOENT);
2416
2417         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2418                              strlen(obd->obd_name) + 1))
2419                 RETURN(-EFAULT);
2420
2421         RETURN(0);
2422 }
2423
2424 /**
2425  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2426  * fsname will be returned in this buffer; otherwise, a static buffer will be
2427  * used to store the fsname and returned to caller.
2428  */
2429 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2430 {
2431         static char fsname_static[MTI_NAME_MAXLEN];
2432         struct lustre_sb_info *lsi = s2lsi(sb);
2433         char *ptr;
2434         int len;
2435
2436         if (buf == NULL) {
2437                 /* this means the caller wants to use static buffer
2438                  * and it doesn't care about race. Usually this is
2439                  * in error reporting path */
2440                 buf = fsname_static;
2441                 buflen = sizeof(fsname_static);
2442         }
2443
2444         len = strlen(lsi->lsi_lmd->lmd_profile);
2445         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2446         if (ptr && (strcmp(ptr, "-client") == 0))
2447                 len -= 7;
2448
2449         if (unlikely(len >= buflen))
2450                 len = buflen - 1;
2451         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2452         buf[len] = '\0';
2453
2454         return buf;
2455 }
2456
2457 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2458 {
2459         char *path = NULL;
2460
2461 #ifdef HAVE_FS_STRUCT_USE_PATH
2462         struct path p;
2463
2464         p.dentry = dentry;
2465         p.mnt = current->fs->root.mnt;
2466         path_get(&p);
2467         path = d_path(&p, buf, bufsize);
2468         path_put(&p);
2469 #else
2470         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2471 #endif
2472
2473         return path;
2474 }
2475
2476 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2477 {
2478         char *buf, *path = NULL;
2479         struct dentry *dentry = NULL;
2480         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2481
2482         /* this can be called inside spin lock so use GFP_ATOMIC. */
2483         buf = (char *)__get_free_page(GFP_ATOMIC);
2484         if (buf != NULL) {
2485                 dentry = d_find_alias(page->mapping->host);
2486                 if (dentry != NULL)
2487                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2488         }
2489
2490         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2491               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2492               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2493               PFID(&obj->cob_header.coh_lu.loh_fid),
2494               (path && !IS_ERR(path)) ? path : "", ioret);
2495
2496         if (dentry != NULL)
2497                 dput(dentry);
2498
2499         if (buf != NULL)
2500                 free_page((unsigned long)buf);
2501 }