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, Whamcloud, Inc.
  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 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67 #else
  68 extern struct address_space_operations_ext ll_aops;
  69 extern struct address_space_operations_ext ll_dir_aops;
  70 #endif
  71
  72 #ifndef log2
  73 #define log2(n) cfs_ffz(~(n))
  74 #endif
  75
  76 static struct ll_sb_info *ll_init_sbi(void)
  77 {
  78         struct ll_sb_info *sbi = NULL;
  79         unsigned long pages;
  80         struct sysinfo si;
  81         class_uuid_t uuid;
  82         int i;
  83         ENTRY;
  84
  85         OBD_ALLOC(sbi, sizeof(*sbi));
  86         if (!sbi)
  87                 RETURN(NULL);
  88
  89         cfs_spin_lock_init(&sbi->ll_lock);
  90         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  91         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  92         cfs_spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  98 #ifdef HAVE_BGL_SUPPORT
  99                 sbi->ll_async_page_max = pages / 4;
 100 #else
 101                 sbi->ll_async_page_max = pages / 2;
 102 #endif
 103         } else {
 104                 sbi->ll_async_page_max = (pages / 4) * 3;
 105         }
 106
 107         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 108                                            SBI_DEFAULT_READAHEAD_MAX);
 109         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 110         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 111                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 112         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 113         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 114
 115         ll_generate_random_uuid(uuid);
 116         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 117         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 118
 119         cfs_spin_lock(&ll_sb_lock);
 120         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 121         cfs_spin_unlock(&ll_sb_lock);
 122
 123         sbi->ll_flags |= LL_SBI_VERBOSE;
 124 #ifdef ENABLE_CHECKSUM
 125         sbi->ll_flags |= LL_SBI_CHECKSUM;
 126 #endif
 127
 128 #ifdef HAVE_LRU_RESIZE_SUPPORT
 129         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 130 #endif
 131
 132         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 133                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 134                                    pp_r_hist.oh_lock);
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_w_hist.oh_lock);
 137         }
 138
 139         /* metadata statahead is enabled by default */
 140         sbi->ll_sa_max = LL_SA_RPC_DEF;
 141         cfs_atomic_set(&sbi->ll_sa_total, 0);
 142         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 143         cfs_atomic_set(&sbi->ll_agl_total, 0);
 144         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         if (proc_lustre_fs_root) {
 202                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 203                                                   dt, md);
 204                 if (err < 0)
 205                         CERROR("could not register mount in /proc/fs/lustre\n");
 206         }
 207
 208         /* indicate the features supported by this client */
 209         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 210                                   OBD_CONNECT_ATTRFID  |
 211                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 212                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 213                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 214                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 215                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 216                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 217                                   OBD_CONNECT_EINPROGRESS |
 218                                   OBD_CONNECT_JOBSTATS;
 219
 220         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 221                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 222
 223 #ifdef HAVE_LRU_RESIZE_SUPPORT
 224         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 225                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 226 #endif
 227 #ifdef CONFIG_FS_POSIX_ACL
 228         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 229 #endif
 230         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 231         data->ocd_version = LUSTRE_VERSION_CODE;
 232
 233         if (sb->s_flags & MS_RDONLY)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 235         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 236                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 237
 238 #ifdef HAVE_MS_FLOCK_LOCK
 239         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 240         sb->s_flags |= MS_FLOCK_LOCK;
 241 #endif
 242 #ifdef MS_HAS_NEW_AOPS
 243         sb->s_flags |= MS_HAS_NEW_AOPS;
 244 #endif
 245
 246         if (sbi->ll_flags & LL_SBI_FLOCK)
 247                 sbi->ll_fop = &ll_file_operations_flock;
 248         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 249                 sbi->ll_fop = &ll_file_operations;
 250         else
 251                 sbi->ll_fop = &ll_file_operations_noflock;
 252
 253         /* real client */
 254         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 255         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 256                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 257
 258         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 259
 260         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 261         if (err == -EBUSY) {
 262                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 263                                    "recovery, of which this client is not a "
 264                                    "part. Please wait for recovery to complete,"
 265                                    " abort, or time out.\n", md);
 266                 GOTO(out, err);
 267         } else if (err) {
 268                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 269                 GOTO(out, err);
 270         }
 271
 272         err = obd_fid_init(sbi->ll_md_exp);
 273         if (err) {
 274                 CERROR("Can't init metadata layer FID infrastructure, "
 275                        "rc %d\n", err);
 276                 GOTO(out_md, err);
 277         }
 278
 279         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 280                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 281         if (err)
 282                 GOTO(out_md_fid, err);
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 292         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 297                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, CFS_PAGE_SIZE);
 305                 GOTO(out_md, err = -EPROTO);
 306         }
 307
 308         size = sizeof(*data);
 309         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 310                            KEY_CONN_DATA,  &size, data, NULL);
 311         if (err) {
 312                 CERROR("Get connect data failed: %d \n", err);
 313                 GOTO(out_md, err);
 314         }
 315
 316         LASSERT(osfs->os_bsize);
 317         sb->s_blocksize = osfs->os_bsize;
 318         sb->s_blocksize_bits = log2(osfs->os_bsize);
 319         sb->s_magic = LL_SUPER_MAGIC;
 320         sb->s_maxbytes = MAX_LFS_FILESIZE;
 321         sbi->ll_namelen = osfs->os_namelen;
 322         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 323
 324         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 325             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 326                 LCONSOLE_INFO("Disabling user_xattr feature because "
 327                               "it is not supported on the server\n");
 328                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 332 #ifdef MS_POSIXACL
 333                 sb->s_flags |= MS_POSIXACL;
 334 #endif
 335                 sbi->ll_flags |= LL_SBI_ACL;
 336         } else {
 337                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 338 #ifdef MS_POSIXACL
 339                 sb->s_flags &= ~MS_POSIXACL;
 340 #endif
 341                 sbi->ll_flags &= ~LL_SBI_ACL;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 345                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 346                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 347                         LCONSOLE_INFO("client is set as remote by default.\n");
 348                 }
 349         } else {
 350                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 351                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 352                         LCONSOLE_INFO("client claims to be remote, but server "
 353                                       "rejected, forced to be local.\n");
 354                 }
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 358                 LCONSOLE_INFO("client enabled MDS capability!\n");
 359                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 360         }
 361
 362         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 363                 LCONSOLE_INFO("client enabled OSS capability!\n");
 364                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 368                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 371                 sbi->ll_md_brw_size = data->ocd_brw_size;
 372         else
 373                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 374
 375         obd = class_name2obd(dt);
 376         if (!obd) {
 377                 CERROR("DT %s: not setup or attached\n", dt);
 378                 GOTO(out_md_fid, err = -ENODEV);
 379         }
 380
 381         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 382                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 383                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 384                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 385                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 386                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 387                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 388                                   OBD_CONNECT_MAXBYTES |
 389                                   OBD_CONNECT_EINPROGRESS |
 390                                   OBD_CONNECT_JOBSTATS;
 391
 392         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 393                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 394
 395         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 396                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 397                  * disabled by default, because it can still be enabled on the
 398                  * fly via /proc. As a consequence, we still need to come to an
 399                  * agreement on the supported algorithms at connect time */
 400                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 401
 402                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 403                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 404                 else
 405                         data->ocd_cksum_types = cksum_types_supported_client();
 406         }
 407
 408 #ifdef HAVE_LRU_RESIZE_SUPPORT
 409         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 410 #endif
 411         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 412                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 413
 414         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 415                "ocd_grant: %d\n", data->ocd_connect_flags,
 416                data->ocd_version, data->ocd_grant);
 417
 418         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 419         obd->obd_upcall.onu_upcall = cl_ocd_update;
 420
 421         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 422
 423         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 424         if (err == -EBUSY) {
 425                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 426                                    "recovery, of which this client is not a "
 427                                    "part.  Please wait for recovery to "
 428                                    "complete, abort, or time out.\n", dt);
 429                 GOTO(out_md_fid, err);
 430         } else if (err) {
 431                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 432                 GOTO(out_md_fid, err);
 433         }
 434
 435         err = obd_fid_init(sbi->ll_dt_exp);
 436         if (err) {
 437                 CERROR("Can't init data layer FID infrastructure, "
 438                        "rc %d\n", err);
 439                 GOTO(out_dt, err);
 440         }
 441
 442         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 443         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 444         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 445         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 446         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 447
 448         fid_zero(&sbi->ll_root_fid);
 449         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 450         if (err) {
 451                 CERROR("cannot mds_connect: rc = %d\n", err);
 452                 GOTO(out_lock_cn_cb, err);
 453         }
 454         if (!fid_is_sane(&sbi->ll_root_fid)) {
 455                 CERROR("Invalid root fid during mount\n");
 456                 GOTO(out_lock_cn_cb, err = -EINVAL);
 457         }
 458         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 459
 460         sb->s_op = &lustre_super_operations;
 461 #if THREAD_SIZE >= 8192
 462         sb->s_export_op = &lustre_export_operations;
 463 #endif
 464
 465         /* make root inode
 466          * XXX: move this to after cbd setup? */
 467         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 468         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 469                 valid |= OBD_MD_FLRMTPERM;
 470         else if (sbi->ll_flags & LL_SBI_ACL)
 471                 valid |= OBD_MD_FLACL;
 472
 473         OBD_ALLOC_PTR(op_data);
 474         if (op_data == NULL)
 475                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 476
 477         op_data->op_fid1 = sbi->ll_root_fid;
 478         op_data->op_mode = 0;
 479         op_data->op_capa1 = oc;
 480         op_data->op_valid = valid;
 481
 482         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 483         if (oc)
 484                 capa_put(oc);
 485         OBD_FREE_PTR(op_data);
 486         if (err) {
 487                 CERROR("md_getattr failed for root: rc = %d\n", err);
 488                 GOTO(out_lock_cn_cb, err);
 489         }
 490         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 491                                sbi->ll_md_exp, &lmd);
 492         if (err) {
 493                 CERROR("failed to understand root inode md: rc = %d\n", err);
 494                 ptlrpc_req_finished (request);
 495                 GOTO(out_lock_cn_cb, err);
 496         }
 497
 498         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 499         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 500         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 501         ptlrpc_req_finished(request);
 502
 503         if (root == NULL || IS_ERR(root)) {
 504                 if (lmd.lsm)
 505                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 506 #ifdef CONFIG_FS_POSIX_ACL
 507                 if (lmd.posix_acl) {
 508                         posix_acl_release(lmd.posix_acl);
 509                         lmd.posix_acl = NULL;
 510                 }
 511 #endif
 512                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 513                 root = NULL;
 514                 CERROR("lustre_lite: bad iget4 for root\n");
 515                 GOTO(out_root, err);
 516         }
 517
 518         err = ll_close_thread_start(&sbi->ll_lcq);
 519         if (err) {
 520                 CERROR("cannot start close thread: rc %d\n", err);
 521                 GOTO(out_root, err);
 522         }
 523
 524 #ifdef CONFIG_FS_POSIX_ACL
 525         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 526                 rct_init(&sbi->ll_rct);
 527                 et_init(&sbi->ll_et);
 528         }
 529 #endif
 530
 531         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 532         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 533                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 534                                  NULL);
 535         cl_sb_init(sb);
 536
 537         sb->s_root = d_alloc_root(root);
 538 #ifdef HAVE_DCACHE_LOCK
 539         sb->s_root->d_op = &ll_d_root_ops;
 540 #else
 541         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 542         d_set_d_op(sb->s_root, &ll_d_root_ops);
 543         sb->s_d_op = &ll_d_ops;
 544 #endif
 545
 546         sbi->ll_sdev_orig = sb->s_dev;
 547
 548         /* We set sb->s_dev equal on all lustre clients in order to support
 549          * NFS export clustering.  NFSD requires that the FSID be the same
 550          * on all clients. */
 551         /* s_dev is also used in lt_compare() to compare two fs, but that is
 552          * only a node-local comparison. */
 553         uuid = obd_get_uuid(sbi->ll_md_exp);
 554         if (uuid != NULL)
 555                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 556         sbi->ll_mnt = mnt;
 557
 558         if (data != NULL)
 559                 OBD_FREE_PTR(data);
 560         if (osfs != NULL)
 561                 OBD_FREE_PTR(osfs);
 562
 563         RETURN(err);
 564 out_root:
 565         if (root)
 566                 iput(root);
 567 out_lock_cn_cb:
 568         obd_fid_fini(sbi->ll_dt_exp);
 569 out_dt:
 570         obd_disconnect(sbi->ll_dt_exp);
 571         sbi->ll_dt_exp = NULL;
 572 out_md_fid:
 573         obd_fid_fini(sbi->ll_md_exp);
 574 out_md:
 575         obd_disconnect(sbi->ll_md_exp);
 576         sbi->ll_md_exp = NULL;
 577 out:
 578         if (data != NULL)
 579                 OBD_FREE_PTR(data);
 580         if (osfs != NULL)
 581                 OBD_FREE_PTR(osfs);
 582         lprocfs_unregister_mountpoint(sbi);
 583         return err;
 584 }
 585
 586 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 587 {
 588         int size, rc;
 589
 590         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 591         size = sizeof(int);
 592         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 593                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 594         if (rc)
 595                 CERROR("Get max mdsize error rc %d \n", rc);
 596
 597         RETURN(rc);
 598 }
 599
 600 void ll_dump_inode(struct inode *inode)
 601 {
 602         struct list_head *tmp;
 603         int dentry_count = 0;
 604
 605         LASSERT(inode != NULL);
 606
 607         list_for_each(tmp, &inode->i_dentry)
 608                 dentry_count++;
 609
 610         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 611                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 612                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 613 }
 614
 615 void lustre_dump_dentry(struct dentry *dentry, int recur)
 616 {
 617         struct list_head *tmp;
 618         int subdirs = 0;
 619
 620         LASSERT(dentry != NULL);
 621
 622         list_for_each(tmp, &dentry->d_subdirs)
 623                 subdirs++;
 624
 625         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 626                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 627                dentry->d_name.len, dentry->d_name.name,
 628                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 629                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 630                dentry->d_flags, dentry->d_fsdata, subdirs);
 631         if (dentry->d_inode != NULL)
 632                 ll_dump_inode(dentry->d_inode);
 633
 634         if (recur == 0)
 635                 return;
 636
 637        list_for_each(tmp, &dentry->d_subdirs) {
 638                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 639                 lustre_dump_dentry(d, recur - 1);
 640         }
 641 }
 642
 643 void client_common_put_super(struct super_block *sb)
 644 {
 645         struct ll_sb_info *sbi = ll_s2sbi(sb);
 646         ENTRY;
 647
 648 #ifdef CONFIG_FS_POSIX_ACL
 649         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 650                 et_fini(&sbi->ll_et);
 651                 rct_fini(&sbi->ll_rct);
 652         }
 653 #endif
 654
 655         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 656
 657         ll_close_thread_shutdown(sbi->ll_lcq);
 658
 659         cl_sb_fini(sb);
 660
 661         cfs_list_del(&sbi->ll_conn_chain);
 662
 663         obd_fid_fini(sbi->ll_dt_exp);
 664         obd_disconnect(sbi->ll_dt_exp);
 665         sbi->ll_dt_exp = NULL;
 666
 667         lprocfs_unregister_mountpoint(sbi);
 668
 669         obd_fid_fini(sbi->ll_md_exp);
 670         obd_disconnect(sbi->ll_md_exp);
 671         sbi->ll_md_exp = NULL;
 672
 673         EXIT;
 674 }
 675
 676 void ll_kill_super(struct super_block *sb)
 677 {
 678         struct ll_sb_info *sbi;
 679
 680         ENTRY;
 681
 682         /* not init sb ?*/
 683         if (!(sb->s_flags & MS_ACTIVE))
 684                 return;
 685
 686         sbi = ll_s2sbi(sb);
 687         /* we need restore s_dev from changed for clustred NFS before put_super
 688          * because new kernels have cached s_dev and change sb->s_dev in
 689          * put_super not affected real removing devices */
 690         if (sbi)
 691                 sb->s_dev = sbi->ll_sdev_orig;
 692         EXIT;
 693 }
 694
 695 char *ll_read_opt(const char *opt, char *data)
 696 {
 697         char *value;
 698         char *retval;
 699         ENTRY;
 700
 701         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 702         if (strncmp(opt, data, strlen(opt)))
 703                 RETURN(NULL);
 704         if ((value = strchr(data, '=')) == NULL)
 705                 RETURN(NULL);
 706
 707         value++;
 708         OBD_ALLOC(retval, strlen(value) + 1);
 709         if (!retval) {
 710                 CERROR("out of memory!\n");
 711                 RETURN(NULL);
 712         }
 713
 714         memcpy(retval, value, strlen(value)+1);
 715         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 716         RETURN(retval);
 717 }
 718
 719 static inline int ll_set_opt(const char *opt, char *data, int fl)
 720 {
 721         if (strncmp(opt, data, strlen(opt)) != 0)
 722                 return(0);
 723         else
 724                 return(fl);
 725 }
 726
 727 /* non-client-specific mount options are parsed in lmd_parse */
 728 static int ll_options(char *options, int *flags)
 729 {
 730         int tmp;
 731         char *s1 = options, *s2;
 732         ENTRY;
 733
 734         if (!options)
 735                 RETURN(0);
 736
 737         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 738
 739         while (*s1) {
 740                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 741                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 742                 if (tmp) {
 743                         *flags |= tmp;
 744                         goto next;
 745                 }
 746                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 747                 if (tmp) {
 748                         *flags |= tmp;
 749                         goto next;
 750                 }
 751                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 752                 if (tmp) {
 753                         *flags |= tmp;
 754                         goto next;
 755                 }
 756                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 757                 if (tmp) {
 758                         *flags &= ~tmp;
 759                         goto next;
 760                 }
 761                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 767                 if (tmp) {
 768                         *flags &= ~tmp;
 769                         goto next;
 770                 }
 771 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 772                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 773                 if (tmp) {
 774                         /* Ignore deprecated mount option.  The client will
 775                          * always try to mount with ACL support, whether this
 776                          * is used depends on whether server supports it. */
 777                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 778                                                   "mount option 'acl'.\n");
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 782                 if (tmp) {
 783                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 784                                                   "mount option 'noacl'.\n");
 785                         goto next;
 786                 }
 787 #else
 788 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 789 #endif
 790                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 791                 if (tmp) {
 792                         *flags |= tmp;
 793                         goto next;
 794                 }
 795
 796                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 802                 if (tmp) {
 803                         *flags &= ~tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 812                 if (tmp) {
 813                         *flags &= ~tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 817                 if (tmp) {
 818                         *flags |= tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 822                 if (tmp) {
 823                         *flags &= ~tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 832                 if (tmp) {
 833                         *flags |= tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 847                                    s1);
 848                 RETURN(-EINVAL);
 849
 850 next:
 851                 /* Find next opt */
 852                 s2 = strchr(s1, ',');
 853                 if (s2 == NULL)
 854                         break;
 855                 s1 = s2 + 1;
 856         }
 857         RETURN(0);
 858 }
 859
 860 void ll_lli_init(struct ll_inode_info *lli)
 861 {
 862         lli->lli_inode_magic = LLI_INODE_MAGIC;
 863         lli->lli_flags = 0;
 864         lli->lli_ioepoch = 0;
 865         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 866         cfs_spin_lock_init(&lli->lli_lock);
 867         lli->lli_posix_acl = NULL;
 868         lli->lli_remote_perms = NULL;
 869         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 870         /* Do not set lli_fid, it has been initialized already. */
 871         fid_zero(&lli->lli_pfid);
 872         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 873         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 874         cfs_atomic_set(&lli->lli_open_count, 0);
 875         lli->lli_mds_capa = NULL;
 876         lli->lli_rmtperm_time = 0;
 877         lli->lli_pending_och = NULL;
 878         lli->lli_mds_read_och = NULL;
 879         lli->lli_mds_write_och = NULL;
 880         lli->lli_mds_exec_och = NULL;
 881         lli->lli_open_fd_read_count = 0;
 882         lli->lli_open_fd_write_count = 0;
 883         lli->lli_open_fd_exec_count = 0;
 884         cfs_mutex_init(&lli->lli_och_mutex);
 885         cfs_spin_lock_init(&lli->lli_agl_lock);
 886         lli->lli_smd = NULL;
 887         lli->lli_clob = NULL;
 888
 889         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 890         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 891                 cfs_mutex_init(&lli->lli_readdir_mutex);
 892                 lli->lli_opendir_key = NULL;
 893                 lli->lli_sai = NULL;
 894                 lli->lli_sa_pos = 0;
 895                 lli->lli_def_acl = NULL;
 896                 cfs_spin_lock_init(&lli->lli_sa_lock);
 897                 lli->lli_opendir_pid = 0;
 898         } else {
 899                 cfs_sema_init(&lli->lli_size_sem, 1);
 900                 lli->lli_size_sem_owner = NULL;
 901                 lli->lli_symlink_name = NULL;
 902                 cfs_init_rwsem(&lli->lli_trunc_sem);
 903                 cfs_mutex_init(&lli->lli_write_mutex);
 904                 lli->lli_async_rc = 0;
 905                 lli->lli_write_rc = 0;
 906                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 907                 lli->lli_glimpse_time = 0;
 908                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 909                 lli->lli_agl_index = 0;
 910         }
 911 }
 912
 913 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 914 {
 915 #ifdef HAVE_BDI_REGISTER
 916         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 917
 918 #ifdef HAVE_BDI_NAME
 919         bdi->name = "lustre";
 920 #endif
 921         return bdi_register(bdi, NULL, "lustre-%d",
 922                             atomic_inc_return(&ll_bdi_num));
 923 #else
 924         return 0;
 925 #endif
 926 }
 927
 928 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 929 {
 930         struct lustre_profile *lprof = NULL;
 931         struct lustre_sb_info *lsi = s2lsi(sb);
 932         struct ll_sb_info *sbi;
 933         char  *dt = NULL, *md = NULL;
 934         char  *profilenm = get_profile_name(sb);
 935         struct config_llog_instance *cfg;
 936         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 937         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 938         int    err;
 939         ENTRY;
 940
 941         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 942
 943         OBD_ALLOC_PTR(cfg);
 944         if (cfg == NULL)
 945                 RETURN(-ENOMEM);
 946
 947         cfs_module_get();
 948
 949         /* client additional sb info */
 950         lsi->lsi_llsbi = sbi = ll_init_sbi();
 951         if (!sbi) {
 952                 cfs_module_put(THIS_MODULE);
 953                 OBD_FREE_PTR(cfg);
 954                 RETURN(-ENOMEM);
 955         }
 956
 957         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 958         if (err)
 959                 GOTO(out_free, err);
 960
 961         err = ll_bdi_init(&lsi->lsi_bdi);
 962         if (err)
 963                 GOTO(out_free, err);
 964         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 965         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 966         err = ll_bdi_register(&lsi->lsi_bdi);
 967         if (err)
 968                 GOTO(out_free, err);
 969
 970 #ifdef HAVE_SB_BDI
 971         sb->s_bdi = &lsi->lsi_bdi;
 972 #endif
 973
 974         /* Generate a string unique to this super, in case some joker tries
 975            to mount the same fs at two mount points.
 976            Use the address of the super itself.*/
 977         cfg->cfg_instance = sb;
 978         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 979
 980         /* set up client obds */
 981         err = lustre_process_log(sb, profilenm, cfg);
 982         if (err < 0) {
 983                 CERROR("Unable to process log: %d\n", err);
 984                 GOTO(out_free, err);
 985         }
 986
 987         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 988         lprof = class_get_profile(profilenm);
 989         if (lprof == NULL) {
 990                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 991                                    " read from the MGS.  Does that filesystem "
 992                                    "exist?\n", profilenm);
 993                 GOTO(out_free, err = -EINVAL);
 994         }
 995         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 996                lprof->lp_md, lprof->lp_dt);
 997
 998         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 999         if (!dt)
1000                 GOTO(out_free, err = -ENOMEM);
1001         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1002
1003         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1004         if (!md)
1005                 GOTO(out_free, err = -ENOMEM);
1006         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1007
1008         /* connections, registrations, sb setup */
1009         err = client_common_fill_super(sb, md, dt, mnt);
1010
1011 out_free:
1012         if (md)
1013                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1014         if (dt)
1015                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1016         if (err)
1017                 ll_put_super(sb);
1018         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1019                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1020
1021         OBD_FREE_PTR(cfg);
1022         RETURN(err);
1023 } /* ll_fill_super */
1024
1025
1026 void lu_context_keys_dump(void);
1027
1028 void ll_put_super(struct super_block *sb)
1029 {
1030         struct config_llog_instance cfg;
1031         struct obd_device *obd;
1032         struct lustre_sb_info *lsi = s2lsi(sb);
1033         struct ll_sb_info *sbi = ll_s2sbi(sb);
1034         char *profilenm = get_profile_name(sb);
1035         int force = 1, next;
1036         ENTRY;
1037
1038         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1039
1040         ll_print_capa_stat(sbi);
1041
1042         cfg.cfg_instance = sb;
1043         lustre_end_log(sb, profilenm, &cfg);
1044
1045         if (sbi->ll_md_exp) {
1046                 obd = class_exp2obd(sbi->ll_md_exp);
1047                 if (obd)
1048                         force = obd->obd_force;
1049         }
1050
1051         /* We need to set force before the lov_disconnect in
1052            lustre_common_put_super, since l_d cleans up osc's as well. */
1053         if (force) {
1054                 next = 0;
1055                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1056                                                      &next)) != NULL) {
1057                         obd->obd_force = force;
1058                 }
1059         }
1060
1061         if (sbi->ll_lcq) {
1062                 /* Only if client_common_fill_super succeeded */
1063                 client_common_put_super(sb);
1064         }
1065
1066         next = 0;
1067         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1068                 class_manual_cleanup(obd);
1069         }
1070
1071         if (sbi->ll_flags & LL_SBI_VERBOSE)
1072                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1073
1074         if (profilenm)
1075                 class_del_profile(profilenm);
1076
1077         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1078                 ll_bdi_destroy(&lsi->lsi_bdi);
1079                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1080         }
1081
1082         ll_free_sbi(sb);
1083         lsi->lsi_llsbi = NULL;
1084
1085         lustre_common_put_super(sb);
1086
1087         cl_env_cache_purge(~0);
1088
1089         cfs_module_put(THIS_MODULE);
1090
1091         EXIT;
1092 } /* client_put_super */
1093
1094 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1095 {
1096         struct inode *inode = NULL;
1097         /* NOTE: we depend on atomic igrab() -bzzz */
1098         lock_res_and_lock(lock);
1099         if (lock->l_ast_data) {
1100                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1101                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1102                         inode = igrab(lock->l_ast_data);
1103                 } else {
1104                         inode = lock->l_ast_data;
1105                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1106                                          D_WARNING, lock, "l_ast_data %p is "
1107                                          "bogus: magic %08x", lock->l_ast_data,
1108                                          lli->lli_inode_magic);
1109                         inode = NULL;
1110                 }
1111         }
1112         unlock_res_and_lock(lock);
1113         return inode;
1114 }
1115
1116 static int null_if_equal(struct ldlm_lock *lock, void *data)
1117 {
1118         if (data == lock->l_ast_data) {
1119                 lock->l_ast_data = NULL;
1120
1121                 if (lock->l_req_mode != lock->l_granted_mode)
1122                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1123         }
1124
1125         return LDLM_ITER_CONTINUE;
1126 }
1127
1128 void ll_clear_inode(struct inode *inode)
1129 {
1130         struct ll_inode_info *lli = ll_i2info(inode);
1131         struct ll_sb_info *sbi = ll_i2sbi(inode);
1132         ENTRY;
1133
1134         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1135                inode->i_generation, inode);
1136
1137         if (S_ISDIR(inode->i_mode)) {
1138                 /* these should have been cleared in ll_file_release */
1139                 LASSERT(lli->lli_opendir_key == NULL);
1140                 LASSERT(lli->lli_sai == NULL);
1141                 LASSERT(lli->lli_opendir_pid == 0);
1142         }
1143
1144         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1145         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1146                          null_if_equal, inode);
1147
1148         LASSERT(!lli->lli_open_fd_write_count);
1149         LASSERT(!lli->lli_open_fd_read_count);
1150         LASSERT(!lli->lli_open_fd_exec_count);
1151
1152         if (lli->lli_mds_write_och)
1153                 ll_md_real_close(inode, FMODE_WRITE);
1154         if (lli->lli_mds_exec_och)
1155                 ll_md_real_close(inode, FMODE_EXEC);
1156         if (lli->lli_mds_read_och)
1157                 ll_md_real_close(inode, FMODE_READ);
1158
1159         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1160                 OBD_FREE(lli->lli_symlink_name,
1161                          strlen(lli->lli_symlink_name) + 1);
1162                 lli->lli_symlink_name = NULL;
1163         }
1164
1165         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1166                 LASSERT(lli->lli_posix_acl == NULL);
1167                 if (lli->lli_remote_perms) {
1168                         free_rmtperm_hash(lli->lli_remote_perms);
1169                         lli->lli_remote_perms = NULL;
1170                 }
1171         }
1172 #ifdef CONFIG_FS_POSIX_ACL
1173         else if (lli->lli_posix_acl) {
1174                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1175                 LASSERT(lli->lli_remote_perms == NULL);
1176                 posix_acl_release(lli->lli_posix_acl);
1177                 lli->lli_posix_acl = NULL;
1178         }
1179 #endif
1180         lli->lli_inode_magic = LLI_INODE_DEAD;
1181
1182         ll_clear_inode_capas(inode);
1183         if (!S_ISDIR(inode->i_mode))
1184                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1185
1186         /*
1187          * XXX This has to be done before lsm is freed below, because
1188          * cl_object still uses inode lsm.
1189          */
1190         cl_inode_fini(inode);
1191
1192         if (lli->lli_smd) {
1193                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1194                 lli->lli_smd = NULL;
1195         }
1196
1197
1198         EXIT;
1199 }
1200
1201 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1202                   struct md_open_data **mod)
1203 {
1204         struct lustre_md md;
1205         struct inode *inode = dentry->d_inode;
1206         struct ll_sb_info *sbi = ll_i2sbi(inode);
1207         struct ptlrpc_request *request = NULL;
1208         int rc, ia_valid;
1209         ENTRY;
1210
1211         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1212                                      LUSTRE_OPC_ANY, NULL);
1213         if (IS_ERR(op_data))
1214                 RETURN(PTR_ERR(op_data));
1215
1216         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1217                         &request, mod);
1218         if (rc) {
1219                 ptlrpc_req_finished(request);
1220                 if (rc == -ENOENT) {
1221                         inode->i_nlink = 0;
1222                         /* Unlinked special device node? Or just a race?
1223                          * Pretend we done everything. */
1224                         if (!S_ISREG(inode->i_mode) &&
1225                             !S_ISDIR(inode->i_mode)) {
1226                                 ia_valid = op_data->op_attr.ia_valid;
1227                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1228                                 rc = simple_setattr(dentry, &op_data->op_attr);
1229                                 op_data->op_attr.ia_valid = ia_valid;
1230                         }
1231                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1232                         CERROR("md_setattr fails: rc = %d\n", rc);
1233                 }
1234                 RETURN(rc);
1235         }
1236
1237         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1238                               sbi->ll_md_exp, &md);
1239         if (rc) {
1240                 ptlrpc_req_finished(request);
1241                 RETURN(rc);
1242         }
1243
1244         /* We want to adjust timestamps.
1245          * If there is at least some data in file, we cleared ATTR_SIZE
1246          * to avoid update size, otherwise it is important to do.(SOM case)
1247          * (bug 6196) */
1248         ia_valid = op_data->op_attr.ia_valid;
1249         /* Since we set ATTR_*_SET flags above, and already done permission
1250          * check, So don't let inode_change_ok() check it again.  */
1251         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252         rc = simple_setattr(dentry, &op_data->op_attr);
1253         op_data->op_attr.ia_valid = ia_valid;
1254
1255         /* Extract epoch data if obtained. */
1256         op_data->op_handle = md.body->handle;
1257         op_data->op_ioepoch = md.body->ioepoch;
1258
1259         ll_update_inode(inode, &md);
1260         ptlrpc_req_finished(request);
1261
1262         RETURN(rc);
1263 }
1264
1265 /* Close IO epoch and send Size-on-MDS attribute update. */
1266 static int ll_setattr_done_writing(struct inode *inode,
1267                                    struct md_op_data *op_data,
1268                                    struct md_open_data *mod)
1269 {
1270         struct ll_inode_info *lli = ll_i2info(inode);
1271         int rc = 0;
1272         ENTRY;
1273
1274         LASSERT(op_data != NULL);
1275         if (!S_ISREG(inode->i_mode))
1276                 RETURN(0);
1277
1278         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1279                op_data->op_ioepoch, PFID(&lli->lli_fid));
1280
1281         op_data->op_flags = MF_EPOCH_CLOSE;
1282         ll_done_writing_attr(inode, op_data);
1283         ll_pack_inode2opdata(inode, op_data, NULL);
1284
1285         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1286         if (rc == -EAGAIN) {
1287                 /* MDS has instructed us to obtain Size-on-MDS attribute
1288                  * from OSTs and send setattr to back to MDS. */
1289                 rc = ll_som_update(inode, op_data);
1290         } else if (rc) {
1291                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1292                        inode->i_ino, rc);
1293         }
1294         RETURN(rc);
1295 }
1296
1297 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1298 {
1299         struct obd_capa *capa;
1300         int rc;
1301
1302         if (attr->ia_valid & ATTR_SIZE)
1303                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1304         else
1305                 capa = ll_mdscapa_get(inode);
1306
1307         rc = cl_setattr_ost(inode, attr, capa);
1308
1309         if (attr->ia_valid & ATTR_SIZE)
1310                 ll_truncate_free_capa(capa);
1311         else
1312                 capa_put(capa);
1313
1314         return rc;
1315 }
1316
1317 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1318  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1319  * keep these values until such a time that objects are allocated for it.
1320  * We do the MDS operations first, as it is checking permissions for us.
1321  * We don't to the MDS RPC if there is nothing that we want to store there,
1322  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1323  * going to do an RPC anyways.
1324  *
1325  * If we are doing a truncate, we will send the mtime and ctime updates
1326  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1327  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1328  * at the same time.
1329  */
1330 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1331 {
1332         struct inode *inode = dentry->d_inode;
1333         struct ll_inode_info *lli = ll_i2info(inode);
1334         struct lov_stripe_md *lsm = lli->lli_smd;
1335         struct md_op_data *op_data = NULL;
1336         struct md_open_data *mod = NULL;
1337         int ia_valid = attr->ia_valid;
1338         int rc = 0, rc1 = 0;
1339         ENTRY;
1340
1341         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1342                attr->ia_valid);
1343
1344         if (ia_valid & ATTR_SIZE) {
1345                 /* Check new size against VFS/VM file size limit and rlimit */
1346                 rc = inode_newsize_ok(inode, attr->ia_size);
1347                 if (rc)
1348                         RETURN(rc);
1349
1350                 /* The maximum Lustre file size is variable, based on the
1351                  * OST maximum object size and number of stripes.  This
1352                  * needs another check in addition to the VFS check above. */
1353                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1354                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1355                                PFID(&lli->lli_fid), attr->ia_size,
1356                                ll_file_maxbytes(inode));
1357                         RETURN(-EFBIG);
1358                 }
1359
1360                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1361         }
1362
1363         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1364         if (ia_valid & TIMES_SET_FLAGS) {
1365                 if (cfs_curproc_fsuid() != inode->i_uid &&
1366                     !cfs_capable(CFS_CAP_FOWNER))
1367                         RETURN(-EPERM);
1368         }
1369
1370         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1371         if (attr->ia_valid & ATTR_CTIME) {
1372                 attr->ia_ctime = CFS_CURRENT_TIME;
1373                 attr->ia_valid |= ATTR_CTIME_SET;
1374         }
1375         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1376                 attr->ia_atime = CFS_CURRENT_TIME;
1377                 attr->ia_valid |= ATTR_ATIME_SET;
1378         }
1379         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1380                 attr->ia_mtime = CFS_CURRENT_TIME;
1381                 attr->ia_valid |= ATTR_MTIME_SET;
1382         }
1383
1384         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1385                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1386                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1387                        cfs_time_current_sec());
1388
1389         /* NB: ATTR_SIZE will only be set after this point if the size
1390          * resides on the MDS, ie, this file has no objects. */
1391         if (lsm)
1392                 attr->ia_valid &= ~ATTR_SIZE;
1393
1394         /* We always do an MDS RPC, even if we're only changing the size;
1395          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1396
1397         OBD_ALLOC_PTR(op_data);
1398         if (op_data == NULL)
1399                 RETURN(-ENOMEM);
1400
1401         if (!S_ISDIR(inode->i_mode)) {
1402                 if (ia_valid & ATTR_SIZE)
1403                         UP_WRITE_I_ALLOC_SEM(inode);
1404                 UNLOCK_INODE_MUTEX(inode);
1405                 cfs_down_write(&lli->lli_trunc_sem);
1406                 LOCK_INODE_MUTEX(inode);
1407                 if (ia_valid & ATTR_SIZE)
1408                         DOWN_WRITE_I_ALLOC_SEM(inode);
1409         }
1410
1411         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1412
1413         /* Open epoch for truncate. */
1414         if (exp_connect_som(ll_i2mdexp(inode)) &&
1415             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1416                 op_data->op_flags = MF_EPOCH_OPEN;
1417
1418         rc = ll_md_setattr(dentry, op_data, &mod);
1419         if (rc)
1420                 GOTO(out, rc);
1421
1422         ll_ioepoch_open(lli, op_data->op_ioepoch);
1423         if (!lsm || !S_ISREG(inode->i_mode)) {
1424                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1425                 GOTO(out, rc = 0);
1426         }
1427
1428         if (ia_valid & ATTR_SIZE)
1429                 attr->ia_valid |= ATTR_SIZE;
1430         if (ia_valid & (ATTR_SIZE |
1431                         ATTR_ATIME | ATTR_ATIME_SET |
1432                         ATTR_MTIME | ATTR_MTIME_SET))
1433                 /* For truncate and utimes sending attributes to OSTs, setting
1434                  * mtime/atime to the past will be performed under PW [0:EOF]
1435                  * extent lock (new_size:EOF for truncate).  It may seem
1436                  * excessive to send mtime/atime updates to OSTs when not
1437                  * setting times to past, but it is necessary due to possible
1438                  * time de-synchronization between MDT inode and OST objects */
1439                 rc = ll_setattr_ost(inode, attr);
1440         EXIT;
1441 out:
1442         if (op_data) {
1443                 if (op_data->op_ioepoch) {
1444                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1445                         if (!rc)
1446                                 rc = rc1;
1447                 }
1448                 ll_finish_md_op_data(op_data);
1449         }
1450         if (!S_ISDIR(inode->i_mode))
1451                 cfs_up_write(&lli->lli_trunc_sem);
1452
1453         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1454                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1455
1456         return rc;
1457 }
1458
1459 int ll_setattr(struct dentry *de, struct iattr *attr)
1460 {
1461         int mode = de->d_inode->i_mode;
1462
1463         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1464                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1465                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1466
1467         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1468                                (ATTR_SIZE|ATTR_MODE)) &&
1469             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1470              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1471               !(attr->ia_mode & S_ISGID))))
1472                 attr->ia_valid |= ATTR_FORCE;
1473
1474         if ((mode & S_ISUID) &&
1475             !(attr->ia_mode & S_ISUID) &&
1476             !(attr->ia_valid & ATTR_KILL_SUID))
1477                 attr->ia_valid |= ATTR_KILL_SUID;
1478
1479         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1480             !(attr->ia_mode & S_ISGID) &&
1481             !(attr->ia_valid & ATTR_KILL_SGID))
1482                 attr->ia_valid |= ATTR_KILL_SGID;
1483
1484         return ll_setattr_raw(de, attr);
1485 }
1486
1487 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1488                        __u64 max_age, __u32 flags)
1489 {
1490         struct ll_sb_info *sbi = ll_s2sbi(sb);
1491         struct obd_statfs obd_osfs;
1492         int rc;
1493         ENTRY;
1494
1495         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1496         if (rc) {
1497                 CERROR("md_statfs fails: rc = %d\n", rc);
1498                 RETURN(rc);
1499         }
1500
1501         osfs->os_type = sb->s_magic;
1502
1503         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1504                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1505
1506         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1507                 flags |= OBD_STATFS_NODELAY;
1508
1509         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1510         if (rc) {
1511                 CERROR("obd_statfs fails: rc = %d\n", rc);
1512                 RETURN(rc);
1513         }
1514
1515         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1516                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1517                obd_osfs.os_files);
1518
1519         osfs->os_bsize = obd_osfs.os_bsize;
1520         osfs->os_blocks = obd_osfs.os_blocks;
1521         osfs->os_bfree = obd_osfs.os_bfree;
1522         osfs->os_bavail = obd_osfs.os_bavail;
1523
1524         /* If we don't have as many objects free on the OST as inodes
1525          * on the MDS, we reduce the total number of inodes to
1526          * compensate, so that the "inodes in use" number is correct.
1527          */
1528         if (obd_osfs.os_ffree < osfs->os_ffree) {
1529                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1530                         obd_osfs.os_ffree;
1531                 osfs->os_ffree = obd_osfs.os_ffree;
1532         }
1533
1534         RETURN(rc);
1535 }
1536 #ifndef HAVE_STATFS_DENTRY_PARAM
1537 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1538 {
1539 #else
1540 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1541 {
1542         struct super_block *sb = de->d_sb;
1543 #endif
1544         struct obd_statfs osfs;
1545         int rc;
1546
1547         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1548         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1549
1550         /* Some amount of caching on the client is allowed */
1551         rc = ll_statfs_internal(sb, &osfs,
1552                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1553                                 0);
1554         if (rc)
1555                 return rc;
1556
1557         statfs_unpack(sfs, &osfs);
1558
1559         /* We need to downshift for all 32-bit kernels, because we can't
1560          * tell if the kernel is being called via sys_statfs64() or not.
1561          * Stop before overflowing f_bsize - in which case it is better
1562          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1563         if (sizeof(long) < 8) {
1564                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1565                         sfs->f_bsize <<= 1;
1566
1567                         osfs.os_blocks >>= 1;
1568                         osfs.os_bfree >>= 1;
1569                         osfs.os_bavail >>= 1;
1570                 }
1571         }
1572
1573         sfs->f_blocks = osfs.os_blocks;
1574         sfs->f_bfree = osfs.os_bfree;
1575         sfs->f_bavail = osfs.os_bavail;
1576
1577         return 0;
1578 }
1579
1580 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1581 {
1582         struct ll_inode_info *lli;
1583         struct lov_stripe_md *lsm;
1584
1585         LASSERT(!S_ISDIR(inode->i_mode));
1586
1587         lli = ll_i2info(inode);
1588         LASSERT(lli->lli_size_sem_owner != current);
1589         cfs_down(&lli->lli_size_sem);
1590         LASSERT(lli->lli_size_sem_owner == NULL);
1591         lli->lli_size_sem_owner = current;
1592         lsm = lli->lli_smd;
1593         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1594                  lsm, lock_lsm);
1595         if (lock_lsm)
1596                 lov_stripe_lock(lsm);
1597 }
1598
1599 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1600 {
1601         struct ll_inode_info *lli;
1602         struct lov_stripe_md *lsm;
1603
1604         lli = ll_i2info(inode);
1605         lsm = lli->lli_smd;
1606         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1607                  lsm, unlock_lsm);
1608         if (unlock_lsm)
1609                 lov_stripe_unlock(lsm);
1610         LASSERT(lli->lli_size_sem_owner == current);
1611         lli->lli_size_sem_owner = NULL;
1612         cfs_up(&lli->lli_size_sem);
1613 }
1614
1615 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1616 {
1617         struct ll_inode_info *lli = ll_i2info(inode);
1618         struct mdt_body *body = md->body;
1619         struct lov_stripe_md *lsm = md->lsm;
1620         struct ll_sb_info *sbi = ll_i2sbi(inode);
1621
1622         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1623         if (lsm != NULL) {
1624                 LASSERT(S_ISREG(inode->i_mode));
1625
1626                 cfs_mutex_lock(&lli->lli_och_mutex);
1627                 if (lli->lli_smd == NULL) {
1628                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1629                             lsm->lsm_magic != LOV_MAGIC_V3) {
1630                                 dump_lsm(D_ERROR, lsm);
1631                                 LBUG();
1632                         }
1633                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1634                                lsm, inode->i_ino, inode->i_generation, inode);
1635                         /* cl_file_inode_init must go before lli_smd or a race
1636                          * is possible where client thinks the file has stripes,
1637                          * but lov raid0 is not setup yet and parallel e.g.
1638                          * glimpse would try to use uninitialized lov */
1639                         cl_file_inode_init(inode, md);
1640                         cfs_spin_lock(&lli->lli_lock);
1641                         lli->lli_smd = lsm;
1642                         cfs_spin_unlock(&lli->lli_lock);
1643                         cfs_mutex_unlock(&lli->lli_och_mutex);
1644                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1645                         if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1646                                 lli->lli_maxbytes = MAX_LFS_FILESIZE;
1647                 } else {
1648                         cfs_mutex_unlock(&lli->lli_och_mutex);
1649                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1650                                 lli->lli_smd->lsm_stripe_count ==
1651                                 lsm->lsm_stripe_count);
1652                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1653                                 CERROR("lsm mismatch for inode %ld\n",
1654                                        inode->i_ino);
1655                                 CERROR("lli_smd:\n");
1656                                 dump_lsm(D_ERROR, lli->lli_smd);
1657                                 CERROR("lsm:\n");
1658                                 dump_lsm(D_ERROR, lsm);
1659                                 LBUG();
1660                         }
1661                 }
1662                 if (lli->lli_smd != lsm)
1663                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1664         }
1665
1666         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1667                 if (body->valid & OBD_MD_FLRMTPERM)
1668                         ll_update_remote_perm(inode, md->remote_perm);
1669         }
1670 #ifdef CONFIG_FS_POSIX_ACL
1671         else if (body->valid & OBD_MD_FLACL) {
1672                 cfs_spin_lock(&lli->lli_lock);
1673                 if (lli->lli_posix_acl)
1674                         posix_acl_release(lli->lli_posix_acl);
1675                 lli->lli_posix_acl = md->posix_acl;
1676                 cfs_spin_unlock(&lli->lli_lock);
1677         }
1678 #endif
1679         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1680         inode->i_generation = cl_fid_build_gen(&body->fid1);
1681
1682         if (body->valid & OBD_MD_FLATIME) {
1683                 if (body->atime > LTIME_S(inode->i_atime))
1684                         LTIME_S(inode->i_atime) = body->atime;
1685                 lli->lli_lvb.lvb_atime = body->atime;
1686         }
1687         if (body->valid & OBD_MD_FLMTIME) {
1688                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1689                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1690                                "to "LPU64"\n", inode->i_ino,
1691                                LTIME_S(inode->i_mtime), body->mtime);
1692                         LTIME_S(inode->i_mtime) = body->mtime;
1693                 }
1694                 lli->lli_lvb.lvb_mtime = body->mtime;
1695         }
1696         if (body->valid & OBD_MD_FLCTIME) {
1697                 if (body->ctime > LTIME_S(inode->i_ctime))
1698                         LTIME_S(inode->i_ctime) = body->ctime;
1699                 lli->lli_lvb.lvb_ctime = body->ctime;
1700         }
1701         if (body->valid & OBD_MD_FLMODE)
1702                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1703         if (body->valid & OBD_MD_FLTYPE)
1704                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1705         LASSERT(inode->i_mode != 0);
1706         if (S_ISREG(inode->i_mode)) {
1707                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1708         } else {
1709                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1710         }
1711 #ifdef HAVE_INODE_BLKSIZE
1712         inode->i_blksize = 1<<inode->i_blkbits;
1713 #endif
1714         if (body->valid & OBD_MD_FLUID)
1715                 inode->i_uid = body->uid;
1716         if (body->valid & OBD_MD_FLGID)
1717                 inode->i_gid = body->gid;
1718         if (body->valid & OBD_MD_FLFLAGS)
1719                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1720         if (body->valid & OBD_MD_FLNLINK)
1721                 inode->i_nlink = body->nlink;
1722         if (body->valid & OBD_MD_FLRDEV)
1723                 inode->i_rdev = old_decode_dev(body->rdev);
1724
1725         if (body->valid & OBD_MD_FLID) {
1726                 /* FID shouldn't be changed! */
1727                 if (fid_is_sane(&lli->lli_fid)) {
1728                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1729                                  "Trying to change FID "DFID
1730                                  " to the "DFID", inode %lu/%u(%p)\n",
1731                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1732                                  inode->i_ino, inode->i_generation, inode);
1733                 } else
1734                         lli->lli_fid = body->fid1;
1735         }
1736
1737         LASSERT(fid_seq(&lli->lli_fid) != 0);
1738
1739         if (body->valid & OBD_MD_FLSIZE) {
1740                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1741                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1742                         struct lustre_handle lockh;
1743                         ldlm_mode_t mode;
1744
1745                         /* As it is possible a blocking ast has been processed
1746                          * by this time, we need to check there is an UPDATE
1747                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1748                          * it. */
1749                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1750                                                &lockh);
1751                         if (mode) {
1752                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1753                                                       LLIF_EPOCH_PENDING |
1754                                                       LLIF_SOM_DIRTY)) {
1755                                         CERROR("ino %lu flags %u still has "
1756                                                "size authority! do not trust "
1757                                                "the size got from MDS\n",
1758                                                inode->i_ino, lli->lli_flags);
1759                                 } else {
1760                                         /* Use old size assignment to avoid
1761                                          * deadlock bz14138 & bz14326 */
1762                                         i_size_write(inode, body->size);
1763                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1764                                 }
1765                                 ldlm_lock_decref(&lockh, mode);
1766                         }
1767                 } else {
1768                         /* Use old size assignment to avoid
1769                          * deadlock bz14138 & bz14326 */
1770                         i_size_write(inode, body->size);
1771
1772                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1773                                inode->i_ino, (unsigned long long)body->size);
1774                 }
1775
1776                 if (body->valid & OBD_MD_FLBLOCKS)
1777                         inode->i_blocks = body->blocks;
1778         }
1779
1780         if (body->valid & OBD_MD_FLMDSCAPA) {
1781                 LASSERT(md->mds_capa);
1782                 ll_add_capa(inode, md->mds_capa);
1783         }
1784         if (body->valid & OBD_MD_FLOSSCAPA) {
1785                 LASSERT(md->oss_capa);
1786                 ll_add_capa(inode, md->oss_capa);
1787         }
1788 }
1789
1790 void ll_read_inode2(struct inode *inode, void *opaque)
1791 {
1792         struct lustre_md *md = opaque;
1793         struct ll_inode_info *lli = ll_i2info(inode);
1794         ENTRY;
1795
1796         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1797                PFID(&lli->lli_fid), inode);
1798
1799         LASSERT(!lli->lli_smd);
1800
1801         /* Core attributes from the MDS first.  This is a new inode, and
1802          * the VFS doesn't zero times in the core inode so we have to do
1803          * it ourselves.  They will be overwritten by either MDS or OST
1804          * attributes - we just need to make sure they aren't newer. */
1805         LTIME_S(inode->i_mtime) = 0;
1806         LTIME_S(inode->i_atime) = 0;
1807         LTIME_S(inode->i_ctime) = 0;
1808         inode->i_rdev = 0;
1809         ll_update_inode(inode, md);
1810
1811         /* OIDEBUG(inode); */
1812
1813         /* initializing backing dev info. */
1814         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1815
1816
1817         if (S_ISREG(inode->i_mode)) {
1818                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1819                 inode->i_op = &ll_file_inode_operations;
1820                 inode->i_fop = sbi->ll_fop;
1821                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1822                 EXIT;
1823         } else if (S_ISDIR(inode->i_mode)) {
1824                 inode->i_op = &ll_dir_inode_operations;
1825                 inode->i_fop = &ll_dir_operations;
1826                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1827                 EXIT;
1828         } else if (S_ISLNK(inode->i_mode)) {
1829                 inode->i_op = &ll_fast_symlink_inode_operations;
1830                 EXIT;
1831         } else {
1832                 inode->i_op = &ll_special_inode_operations;
1833
1834                 init_special_inode(inode, inode->i_mode,
1835                                    kdev_t_to_nr(inode->i_rdev));
1836
1837                 EXIT;
1838         }
1839 }
1840
1841 void ll_delete_inode(struct inode *inode)
1842 {
1843         struct cl_inode_info *lli = cl_i2info(inode);
1844         struct ll_sb_info    *sbi = ll_i2sbi(inode);
1845         int rc;
1846         ENTRY;
1847
1848         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1849         if (rc)
1850                 CERROR("fid_delete() failed, rc %d\n", rc);
1851
1852         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1853                 /* discard all dirty pages before truncating them, required by
1854                  * osc_extent implementation at LU-1030. */
1855                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1856
1857         truncate_inode_pages(&inode->i_data, 0);
1858
1859         /* Workaround for LU-118 */
1860         if (inode->i_data.nrpages) {
1861                 TREE_READ_LOCK_IRQ(&inode->i_data);
1862                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1863                 LASSERTF(inode->i_data.nrpages == 0,
1864                          "inode=%lu/%u(%p) nrpages=%lu, see "
1865                          "http://jira.whamcloud.com/browse/LU-118\n",
1866                          inode->i_ino, inode->i_generation, inode,
1867                          inode->i_data.nrpages);
1868         }
1869         /* Workaround end */
1870
1871 #ifdef HAVE_SBOPS_EVICT_INODE
1872         ll_clear_inode(inode);
1873         end_writeback(inode);
1874 #else
1875         clear_inode(inode);
1876 #endif
1877
1878         EXIT;
1879 }
1880
1881 int ll_iocontrol(struct inode *inode, struct file *file,
1882                  unsigned int cmd, unsigned long arg)
1883 {
1884         struct ll_sb_info *sbi = ll_i2sbi(inode);
1885         struct ptlrpc_request *req = NULL;
1886         int rc, flags = 0;
1887         ENTRY;
1888
1889         switch(cmd) {
1890         case FSFILT_IOC_GETFLAGS: {
1891                 struct mdt_body *body;
1892                 struct md_op_data *op_data;
1893
1894                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1895                                              0, 0, LUSTRE_OPC_ANY,
1896                                              NULL);
1897                 if (IS_ERR(op_data))
1898                         RETURN(PTR_ERR(op_data));
1899
1900                 op_data->op_valid = OBD_MD_FLFLAGS;
1901                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1902                 ll_finish_md_op_data(op_data);
1903                 if (rc) {
1904                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1905                         RETURN(-abs(rc));
1906                 }
1907
1908                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1909
1910                 flags = body->flags;
1911
1912                 ptlrpc_req_finished(req);
1913
1914                 RETURN(put_user(flags, (int *)arg));
1915         }
1916         case FSFILT_IOC_SETFLAGS: {
1917                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1918                 struct obd_info oinfo = { { { 0 } } };
1919                 struct md_op_data *op_data;
1920
1921                 if (get_user(flags, (int *)arg))
1922                         RETURN(-EFAULT);
1923
1924                 oinfo.oi_md = lsm;
1925                 OBDO_ALLOC(oinfo.oi_oa);
1926                 if (!oinfo.oi_oa)
1927                         RETURN(-ENOMEM);
1928
1929                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1930                                              LUSTRE_OPC_ANY, NULL);
1931                 if (IS_ERR(op_data))
1932                         RETURN(PTR_ERR(op_data));
1933
1934                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1935                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1936                 rc = md_setattr(sbi->ll_md_exp, op_data,
1937                                 NULL, 0, NULL, 0, &req, NULL);
1938                 ll_finish_md_op_data(op_data);
1939                 ptlrpc_req_finished(req);
1940                 if (rc) {
1941                         OBDO_FREE(oinfo.oi_oa);
1942                         RETURN(rc);
1943                 }
1944
1945                 if (lsm == NULL) {
1946                         OBDO_FREE(oinfo.oi_oa);
1947                         GOTO(update_cache, rc);
1948                 }
1949
1950                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1951                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1952                 oinfo.oi_oa->o_flags = flags;
1953                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1954                                        OBD_MD_FLGROUP;
1955                 oinfo.oi_capa = ll_mdscapa_get(inode);
1956                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1957                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1958                 capa_put(oinfo.oi_capa);
1959                 OBDO_FREE(oinfo.oi_oa);
1960                 if (rc) {
1961                         if (rc != -EPERM && rc != -EACCES)
1962                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1963                         RETURN(rc);
1964                 }
1965
1966                 EXIT;
1967 update_cache:
1968                 inode->i_flags = ll_ext_to_inode_flags(flags);
1969                 return 0;
1970         }
1971         default:
1972                 RETURN(-ENOSYS);
1973         }
1974
1975         RETURN(0);
1976 }
1977
1978 int ll_flush_ctx(struct inode *inode)
1979 {
1980         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1981
1982         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1983
1984         obd_set_info_async(NULL, sbi->ll_md_exp,
1985                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1986                            0, NULL, NULL);
1987         obd_set_info_async(NULL, sbi->ll_dt_exp,
1988                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1989                            0, NULL, NULL);
1990         return 0;
1991 }
1992
1993 /* umount -f client means force down, don't save state */
1994 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1995 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1996 {
1997         struct super_block *sb = vfsmnt->mnt_sb;
1998 #else
1999 void ll_umount_begin(struct super_block *sb)
2000 {
2001 #endif
2002         struct lustre_sb_info *lsi = s2lsi(sb);
2003         struct ll_sb_info *sbi = ll_s2sbi(sb);
2004         struct obd_device *obd;
2005         struct obd_ioctl_data *ioc_data;
2006         ENTRY;
2007
2008 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2009         if (!(flags & MNT_FORCE)) {
2010                 EXIT;
2011                 return;
2012         }
2013 #endif
2014
2015         /* Tell the MGC we got umount -f */
2016         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2017
2018         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2019                sb->s_count, atomic_read(&sb->s_active));
2020
2021         obd = class_exp2obd(sbi->ll_md_exp);
2022         if (obd == NULL) {
2023                 CERROR("Invalid MDC connection handle "LPX64"\n",
2024                        sbi->ll_md_exp->exp_handle.h_cookie);
2025                 EXIT;
2026                 return;
2027         }
2028         obd->obd_force = 1;
2029
2030         obd = class_exp2obd(sbi->ll_dt_exp);
2031         if (obd == NULL) {
2032                 CERROR("Invalid LOV connection handle "LPX64"\n",
2033                        sbi->ll_dt_exp->exp_handle.h_cookie);
2034                 EXIT;
2035                 return;
2036         }
2037         obd->obd_force = 1;
2038
2039         OBD_ALLOC_PTR(ioc_data);
2040         if (ioc_data) {
2041                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2042                               sizeof ioc_data, ioc_data, NULL);
2043
2044                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2045                               sizeof ioc_data, ioc_data, NULL);
2046
2047                 OBD_FREE_PTR(ioc_data);
2048         }
2049
2050
2051         /* Really, we'd like to wait until there are no requests outstanding,
2052          * and then continue.  For now, we just invalidate the requests,
2053          * schedule() and sleep one second if needed, and hope.
2054          */
2055         cfs_schedule();
2056 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2057         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2058                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2059                                                    cfs_time_seconds(1));
2060                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2061                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2062                                       "may try to umount it a bit later\n",
2063                                       atomic_read(&vfsmnt->mnt_count));
2064         }
2065 #endif
2066
2067         EXIT;
2068 }
2069
2070 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2071 {
2072         struct ll_sb_info *sbi = ll_s2sbi(sb);
2073         char *profilenm = get_profile_name(sb);
2074         int err;
2075         __u32 read_only;
2076
2077         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2078                 read_only = *flags & MS_RDONLY;
2079                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2080                                          sizeof(KEY_READ_ONLY),
2081                                          KEY_READ_ONLY, sizeof(read_only),
2082                                          &read_only, NULL);
2083                 if (err) {
2084                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2085                                       profilenm, read_only ?
2086                                       "read-only" : "read-write", err);
2087                         return err;
2088                 }
2089
2090                 if (read_only)
2091                         sb->s_flags |= MS_RDONLY;
2092                 else
2093                         sb->s_flags &= ~MS_RDONLY;
2094
2095                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2096                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2097                                       read_only ?  "read-only" : "read-write");
2098         }
2099         return 0;
2100 }
2101
2102 int ll_prep_inode(struct inode **inode,
2103                   struct ptlrpc_request *req,
2104                   struct super_block *sb)
2105 {
2106         struct ll_sb_info *sbi = NULL;
2107         struct lustre_md md;
2108         int rc;
2109         ENTRY;
2110
2111         LASSERT(*inode || sb);
2112         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2113         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2114                               sbi->ll_md_exp, &md);
2115         if (rc)
2116                 RETURN(rc);
2117
2118         if (*inode) {
2119                 ll_update_inode(*inode, &md);
2120         } else {
2121                 LASSERT(sb != NULL);
2122
2123                 /*
2124                  * At this point server returns to client's same fid as client
2125                  * generated for creating. So using ->fid1 is okay here.
2126                  */
2127                 LASSERT(fid_is_sane(&md.body->fid1));
2128
2129                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2130                 if (*inode == NULL || IS_ERR(*inode)) {
2131                         if (md.lsm)
2132                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2133 #ifdef CONFIG_FS_POSIX_ACL
2134                         if (md.posix_acl) {
2135                                 posix_acl_release(md.posix_acl);
2136                                 md.posix_acl = NULL;
2137                         }
2138 #endif
2139                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2140                         *inode = NULL;
2141                         CERROR("new_inode -fatal: rc %d\n", rc);
2142                         GOTO(out, rc);
2143                 }
2144         }
2145
2146 out:
2147         md_free_lustre_md(sbi->ll_md_exp, &md);
2148         RETURN(rc);
2149 }
2150
2151 int ll_obd_statfs(struct inode *inode, void *arg)
2152 {
2153         struct ll_sb_info *sbi = NULL;
2154         struct obd_export *exp;
2155         char *buf = NULL;
2156         struct obd_ioctl_data *data = NULL;
2157         __u32 type;
2158         int len = 0, rc;
2159
2160         if (!inode || !(sbi = ll_i2sbi(inode)))
2161                 GOTO(out_statfs, rc = -EINVAL);
2162
2163         rc = obd_ioctl_getdata(&buf, &len, arg);
2164         if (rc)
2165                 GOTO(out_statfs, rc);
2166
2167         data = (void*)buf;
2168         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2169             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2170                 GOTO(out_statfs, rc = -EINVAL);
2171
2172         if (data->ioc_inllen1 != sizeof(__u32) ||
2173             data->ioc_inllen2 != sizeof(__u32) ||
2174             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2175             data->ioc_plen2 != sizeof(struct obd_uuid))
2176                 GOTO(out_statfs, rc = -EINVAL);
2177
2178         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2179         if (type == LL_STATFS_LMV)
2180                 exp = sbi->ll_md_exp;
2181         else if (type == LL_STATFS_LOV)
2182                 exp = sbi->ll_dt_exp;
2183         else
2184                 GOTO(out_statfs, rc = -ENODEV);
2185
2186         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2187         if (rc)
2188                 GOTO(out_statfs, rc);
2189 out_statfs:
2190         if (buf)
2191                 obd_ioctl_freedata(buf, len);
2192         return rc;
2193 }
2194
2195 int ll_process_config(struct lustre_cfg *lcfg)
2196 {
2197         char *ptr;
2198         void *sb;
2199         struct lprocfs_static_vars lvars;
2200         unsigned long x;
2201         int rc = 0;
2202
2203         lprocfs_llite_init_vars(&lvars);
2204
2205         /* The instance name contains the sb: lustre-client-aacfe000 */
2206         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2207         if (!ptr || !*(++ptr))
2208                 return -EINVAL;
2209         if (sscanf(ptr, "%lx", &x) != 1)
2210                 return -EINVAL;
2211         sb = (void *)x;
2212         /* This better be a real Lustre superblock! */
2213         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2214
2215         /* Note we have not called client_common_fill_super yet, so
2216            proc fns must be able to handle that! */
2217         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2218                                       lcfg, sb);
2219         if (rc > 0)
2220                 rc = 0;
2221         return(rc);
2222 }
2223
2224 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2225 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2226                                        struct inode *i1, struct inode *i2,
2227                                        const char *name, int namelen,
2228                                        int mode, __u32 opc, void *data)
2229 {
2230         LASSERT(i1 != NULL);
2231
2232         if (namelen > ll_i2sbi(i1)->ll_namelen)
2233                 return ERR_PTR(-ENAMETOOLONG);
2234
2235         if (op_data == NULL)
2236                 OBD_ALLOC_PTR(op_data);
2237
2238         if (op_data == NULL)
2239                 return ERR_PTR(-ENOMEM);
2240
2241         ll_i2gids(op_data->op_suppgids, i1, i2);
2242         op_data->op_fid1 = *ll_inode2fid(i1);
2243         op_data->op_capa1 = ll_mdscapa_get(i1);
2244
2245         if (i2) {
2246                 op_data->op_fid2 = *ll_inode2fid(i2);
2247                 op_data->op_capa2 = ll_mdscapa_get(i2);
2248         } else {
2249                 fid_zero(&op_data->op_fid2);
2250                 op_data->op_capa2 = NULL;
2251         }
2252
2253         op_data->op_name = name;
2254         op_data->op_namelen = namelen;
2255         op_data->op_mode = mode;
2256         op_data->op_mod_time = cfs_time_current_sec();
2257         op_data->op_fsuid = cfs_curproc_fsuid();
2258         op_data->op_fsgid = cfs_curproc_fsgid();
2259         op_data->op_cap = cfs_curproc_cap_pack();
2260         op_data->op_bias = MDS_CHECK_SPLIT;
2261         op_data->op_opc = opc;
2262         op_data->op_mds = 0;
2263         op_data->op_data = data;
2264
2265         /* If the file is being opened after mknod() (normally due to NFS)
2266          * try to use the default stripe data from parent directory for
2267          * allocating OST objects.  Try to pass the parent FID to MDS. */
2268         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2269             ll_i2info(i2)->lli_smd == NULL) {
2270                 struct ll_inode_info *lli = ll_i2info(i2);
2271
2272                 cfs_spin_lock(&lli->lli_lock);
2273                 if (likely(lli->lli_smd == NULL &&
2274                            !fid_is_zero(&lli->lli_pfid)))
2275                         op_data->op_fid1 = lli->lli_pfid;
2276                 cfs_spin_unlock(&lli->lli_lock);
2277                 /** We ignore parent's capability temporary. */
2278         }
2279
2280         return op_data;
2281 }
2282
2283 void ll_finish_md_op_data(struct md_op_data *op_data)
2284 {
2285         capa_put(op_data->op_capa1);
2286         capa_put(op_data->op_capa2);
2287         OBD_FREE_PTR(op_data);
2288 }
2289
2290 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2291 {
2292         struct ll_sb_info *sbi;
2293
2294         LASSERT((seq != NULL) && (vfs != NULL));
2295         sbi = ll_s2sbi(vfs->mnt_sb);
2296
2297         if (sbi->ll_flags & LL_SBI_NOLCK)
2298                 seq_puts(seq, ",nolock");
2299
2300         if (sbi->ll_flags & LL_SBI_FLOCK)
2301                 seq_puts(seq, ",flock");
2302
2303         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2304                 seq_puts(seq, ",localflock");
2305
2306         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2307                 seq_puts(seq, ",user_xattr");
2308
2309         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2310                 seq_puts(seq, ",lazystatfs");
2311
2312         RETURN(0);
2313 }
2314
2315 /**
2316  * Get obd name by cmd, and copy out to user space
2317  */
2318 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2319 {
2320         struct ll_sb_info *sbi = ll_i2sbi(inode);
2321         struct obd_device *obd;
2322         ENTRY;
2323
2324         if (cmd == OBD_IOC_GETDTNAME)
2325                 obd = class_exp2obd(sbi->ll_dt_exp);
2326         else if (cmd == OBD_IOC_GETMDNAME)
2327                 obd = class_exp2obd(sbi->ll_md_exp);
2328         else
2329                 RETURN(-EINVAL);
2330
2331         if (!obd)
2332                 RETURN(-ENOENT);
2333
2334         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2335                              strlen(obd->obd_name) + 1))
2336                 RETURN(-EFAULT);
2337
2338         RETURN(0);
2339 }