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