lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/llite_lib.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/module.h>
  47 #include <linux/types.h>
  48 #include <linux/version.h>
  49 #include <linux/mm.h>
  50
  51 #include <lustre_lite.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 cfs_mem_cache_t *ll_file_data_slab;
  63
  64 CFS_LIST_HEAD(ll_super_blocks);
  65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  66
  67 #ifndef MS_HAS_NEW_AOPS
  68 extern struct address_space_operations ll_aops;
  69 extern struct address_space_operations ll_dir_aops;
  70 #else
  71 extern struct address_space_operations_ext ll_aops;
  72 extern struct address_space_operations_ext ll_dir_aops;
  73 #endif
  74
  75 #ifndef log2
  76 #define log2(n) cfs_ffz(~(n))
  77 #endif
  78
  79 static struct ll_sb_info *ll_init_sbi(void)
  80 {
  81         struct ll_sb_info *sbi = NULL;
  82         unsigned long pages;
  83         struct sysinfo si;
  84         class_uuid_t uuid;
  85         int i;
  86         ENTRY;
  87
  88         OBD_ALLOC(sbi, sizeof(*sbi));
  89         if (!sbi)
  90                 RETURN(NULL);
  91
  92         cfs_spin_lock_init(&sbi->ll_lock);
  93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  95         cfs_spin_lock_init(&sbi->ll_process_lock);
  96         sbi->ll_rw_stats_on = 0;
  97
  98         si_meminfo(&si);
  99         pages = si.totalram - si.totalhigh;
 100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 101 #ifdef HAVE_BGL_SUPPORT
 102                 sbi->ll_async_page_max = pages / 4;
 103 #else
 104                 sbi->ll_async_page_max = pages / 2;
 105 #endif
 106         } else {
 107                 sbi->ll_async_page_max = (pages / 4) * 3;
 108         }
 109
 110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 111                                            SBI_DEFAULT_READAHEAD_MAX);
 112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 117
 118         ll_generate_random_uuid(uuid);
 119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 121
 122         cfs_spin_lock(&ll_sb_lock);
 123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 124         cfs_spin_unlock(&ll_sb_lock);
 125
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_r_hist.oh_lock);
 137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 138                                    pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         cfs_atomic_set(&sbi->ll_sa_total, 0);
 144         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 145
 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
 218         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 219                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 220
 221 #ifdef HAVE_LRU_RESIZE_SUPPORT
 222         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 223                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 224 #endif
 225 #ifdef CONFIG_FS_POSIX_ACL
 226         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 227 #endif
 228         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 229         data->ocd_version = LUSTRE_VERSION_CODE;
 230
 231         if (sb->s_flags & MS_RDONLY)
 232                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 233         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 234                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 235
 236 #ifdef HAVE_MS_FLOCK_LOCK
 237         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 238         sb->s_flags |= MS_FLOCK_LOCK;
 239 #endif
 240 #ifdef MS_HAS_NEW_AOPS
 241         sb->s_flags |= MS_HAS_NEW_AOPS;
 242 #endif
 243
 244         if (sbi->ll_flags & LL_SBI_FLOCK)
 245                 sbi->ll_fop = &ll_file_operations_flock;
 246         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 247                 sbi->ll_fop = &ll_file_operations;
 248         else
 249                 sbi->ll_fop = &ll_file_operations_noflock;
 250
 251         /* real client */
 252         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 253         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 254                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 255
 256         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 257
 258         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 259         if (err == -EBUSY) {
 260                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 261                                    "recovery, of which this client is not a "
 262                                    "part. Please wait for recovery to complete,"
 263                                    " abort, or time out.\n", md);
 264                 GOTO(out, err);
 265         } else if (err) {
 266                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 267                 GOTO(out, err);
 268         }
 269
 270         err = obd_fid_init(sbi->ll_md_exp);
 271         if (err) {
 272                 CERROR("Can't init metadata layer FID infrastructure, "
 273                        "rc %d\n", err);
 274                 GOTO(out_md, err);
 275         }
 276
 277         err = obd_statfs(obd, osfs,
 278                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 279         if (err)
 280                 GOTO(out_md_fid, err);
 281
 282         /* This needs to be after statfs to ensure connect has finished.
 283          * Note that "data" does NOT contain the valid connect reply.
 284          * If connecting to a 1.8 server there will be no LMV device, so
 285          * we can access the MDC export directly and exp_connect_flags will
 286          * be non-zero, but if accessing an upgraded 2.1 server it will
 287          * have the correct flags filled in.
 288          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 289         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 290         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 291             valid != CLIENT_CONNECT_MDT_REQD) {
 292                 char *buf;
 293
 294                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 295                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 296                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 297                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 298                                    "feature(s) needed for correct operation "
 299                                    "of this client (%s). Please upgrade "
 300                                    "server or downgrade client.\n",
 301                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 302                 OBD_FREE(buf, CFS_PAGE_SIZE);
 303                 GOTO(out_md, err = -EPROTO);
 304         }
 305
 306         size = sizeof(*data);
 307         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 308                            KEY_CONN_DATA,  &size, data, NULL);
 309         if (err) {
 310                 CERROR("Get connect data failed: %d \n", err);
 311                 GOTO(out_md, err);
 312         }
 313
 314         LASSERT(osfs->os_bsize);
 315         sb->s_blocksize = osfs->os_bsize;
 316         sb->s_blocksize_bits = log2(osfs->os_bsize);
 317         sb->s_magic = LL_SUPER_MAGIC;
 318
 319         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 320          *         retval = in_file->f_op->sendfile(...);
 321          *         if (*ppos > max)
 322          *                 retval = -EOVERFLOW;
 323          *
 324          * it will check if *ppos is greater than max. However, max equals to
 325          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 326          * has been defined as a signed long long integer in linux kernel. */
 327 #if BITS_PER_LONG == 64
 328         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 329 #else
 330         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 331 #endif
 332         sbi->ll_namelen = osfs->os_namelen;
 333         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 334
 335         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 336             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 337                 LCONSOLE_INFO("Disabling user_xattr feature because "
 338                               "it is not supported on the server\n");
 339                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 343 #ifdef MS_POSIXACL
 344                 sb->s_flags |= MS_POSIXACL;
 345 #endif
 346                 sbi->ll_flags |= LL_SBI_ACL;
 347         } else {
 348                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 349 #ifdef MS_POSIXACL
 350                 sb->s_flags &= ~MS_POSIXACL;
 351 #endif
 352                 sbi->ll_flags &= ~LL_SBI_ACL;
 353         }
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 356                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 357                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 358                         LCONSOLE_INFO("client is set as remote by default.\n");
 359                 }
 360         } else {
 361                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 362                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 363                         LCONSOLE_INFO("client claims to be remote, but server "
 364                                       "rejected, forced to be local.\n");
 365                 }
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 369                 LCONSOLE_INFO("client enabled MDS capability!\n");
 370                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 374                 LCONSOLE_INFO("client enabled OSS capability!\n");
 375                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 376         }
 377
 378         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 379                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 382                 sbi->ll_md_brw_size = data->ocd_brw_size;
 383         else
 384                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 385
 386         obd = class_name2obd(dt);
 387         if (!obd) {
 388                 CERROR("DT %s: not setup or attached\n", dt);
 389                 GOTO(out_md_fid, err = -ENODEV);
 390         }
 391
 392         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 393                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 394                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 395                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 396                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 397                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 398                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 399                                   OBD_CONNECT_MAXBYTES;
 400
 401         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 402                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 403
 404         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 405                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 406                  * disabled by default, because it can still be enabled on the
 407                  * fly via /proc. As a consequence, we still need to come to an
 408                  * agreement on the supported algorithms at connect time */
 409                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 410
 411                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 412                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 413                 else
 414                         data->ocd_cksum_types = cksum_types_supported();
 415         }
 416
 417 #ifdef HAVE_LRU_RESIZE_SUPPORT
 418         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 419 #endif
 420         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 421                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 422
 423         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 424                "ocd_grant: %d\n", data->ocd_connect_flags,
 425                data->ocd_version, data->ocd_grant);
 426
 427         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 428         obd->obd_upcall.onu_upcall = cl_ocd_update;
 429
 430         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 431
 432         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 433         if (err == -EBUSY) {
 434                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 435                                    "recovery, of which this client is not a "
 436                                    "part.  Please wait for recovery to "
 437                                    "complete, abort, or time out.\n", dt);
 438                 GOTO(out_md_fid, err);
 439         } else if (err) {
 440                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 441                 GOTO(out_md_fid, err);
 442         }
 443
 444         err = obd_fid_init(sbi->ll_dt_exp);
 445         if (err) {
 446                 CERROR("Can't init data layer FID infrastructure, "
 447                        "rc %d\n", err);
 448                 GOTO(out_dt, err);
 449         }
 450
 451         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 452         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 453         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 454         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 455         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 456
 457         fid_zero(&sbi->ll_root_fid);
 458         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 459         if (err) {
 460                 CERROR("cannot mds_connect: rc = %d\n", err);
 461                 GOTO(out_lock_cn_cb, err);
 462         }
 463         if (!fid_is_sane(&sbi->ll_root_fid)) {
 464                 CERROR("Invalid root fid during mount\n");
 465                 GOTO(out_lock_cn_cb, err = -EINVAL);
 466         }
 467         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 468
 469         sb->s_op = &lustre_super_operations;
 470 #if THREAD_SIZE >= 8192
 471         sb->s_export_op = &lustre_export_operations;
 472 #endif
 473
 474         /* make root inode
 475          * XXX: move this to after cbd setup? */
 476         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 477         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 478                 valid |= OBD_MD_FLRMTPERM;
 479         else if (sbi->ll_flags & LL_SBI_ACL)
 480                 valid |= OBD_MD_FLACL;
 481
 482         OBD_ALLOC_PTR(op_data);
 483         if (op_data == NULL)
 484                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 485
 486         op_data->op_fid1 = sbi->ll_root_fid;
 487         op_data->op_mode = 0;
 488         op_data->op_capa1 = oc;
 489         op_data->op_valid = valid;
 490
 491         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 492         if (oc)
 493                 capa_put(oc);
 494         OBD_FREE_PTR(op_data);
 495         if (err) {
 496                 CERROR("md_getattr failed for root: rc = %d\n", err);
 497                 GOTO(out_lock_cn_cb, err);
 498         }
 499         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 500                                sbi->ll_md_exp, &lmd);
 501         if (err) {
 502                 CERROR("failed to understand root inode md: rc = %d\n", err);
 503                 ptlrpc_req_finished (request);
 504                 GOTO(out_lock_cn_cb, err);
 505         }
 506
 507         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 508         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 509         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 510         ptlrpc_req_finished(request);
 511
 512         if (root == NULL || IS_ERR(root)) {
 513                 if (lmd.lsm)
 514                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 515 #ifdef CONFIG_FS_POSIX_ACL
 516                 if (lmd.posix_acl) {
 517                         posix_acl_release(lmd.posix_acl);
 518                         lmd.posix_acl = NULL;
 519                 }
 520 #endif
 521                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 522                 root = NULL;
 523                 CERROR("lustre_lite: bad iget4 for root\n");
 524                 GOTO(out_root, err);
 525         }
 526
 527         err = ll_close_thread_start(&sbi->ll_lcq);
 528         if (err) {
 529                 CERROR("cannot start close thread: rc %d\n", err);
 530                 GOTO(out_root, err);
 531         }
 532
 533 #ifdef CONFIG_FS_POSIX_ACL
 534         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 535                 rct_init(&sbi->ll_rct);
 536                 et_init(&sbi->ll_et);
 537         }
 538 #endif
 539
 540         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 541         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 542                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 543                                  NULL);
 544         cl_sb_init(sb);
 545
 546         sb->s_root = d_alloc_root(root);
 547         sb->s_root->d_op = &ll_d_root_ops;
 548
 549         sbi->ll_sdev_orig = sb->s_dev;
 550
 551         /* We set sb->s_dev equal on all lustre clients in order to support
 552          * NFS export clustering.  NFSD requires that the FSID be the same
 553          * on all clients. */
 554         /* s_dev is also used in lt_compare() to compare two fs, but that is
 555          * only a node-local comparison. */
 556         uuid = obd_get_uuid(sbi->ll_md_exp);
 557         if (uuid != NULL)
 558                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 559         sbi->ll_mnt = mnt;
 560
 561         if (data != NULL)
 562                 OBD_FREE_PTR(data);
 563         if (osfs != NULL)
 564                 OBD_FREE_PTR(osfs);
 565
 566         RETURN(err);
 567 out_root:
 568         if (root)
 569                 iput(root);
 570 out_lock_cn_cb:
 571         obd_fid_fini(sbi->ll_dt_exp);
 572 out_dt:
 573         obd_disconnect(sbi->ll_dt_exp);
 574         sbi->ll_dt_exp = NULL;
 575 out_md_fid:
 576         obd_fid_fini(sbi->ll_md_exp);
 577 out_md:
 578         obd_disconnect(sbi->ll_md_exp);
 579         sbi->ll_md_exp = NULL;
 580 out:
 581         if (data != NULL)
 582                 OBD_FREE_PTR(data);
 583         if (osfs != NULL)
 584                 OBD_FREE_PTR(osfs);
 585         lprocfs_unregister_mountpoint(sbi);
 586         return err;
 587 }
 588
 589 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 590 {
 591         int size, rc;
 592
 593         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 594         size = sizeof(int);
 595         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 596                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 597         if (rc)
 598                 CERROR("Get max mdsize error rc %d \n", rc);
 599
 600         RETURN(rc);
 601 }
 602
 603 void ll_dump_inode(struct inode *inode)
 604 {
 605         struct list_head *tmp;
 606         int dentry_count = 0;
 607
 608         LASSERT(inode != NULL);
 609
 610         list_for_each(tmp, &inode->i_dentry)
 611                 dentry_count++;
 612
 613         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 614                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 615                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 616 }
 617
 618 void lustre_dump_dentry(struct dentry *dentry, int recur)
 619 {
 620         struct list_head *tmp;
 621         int subdirs = 0;
 622
 623         LASSERT(dentry != NULL);
 624
 625         list_for_each(tmp, &dentry->d_subdirs)
 626                 subdirs++;
 627
 628         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 629                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 630                dentry->d_name.len, dentry->d_name.name,
 631                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 632                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 633                dentry->d_flags, dentry->d_fsdata, subdirs);
 634         if (dentry->d_inode != NULL)
 635                 ll_dump_inode(dentry->d_inode);
 636
 637         if (recur == 0)
 638                 return;
 639
 640        list_for_each(tmp, &dentry->d_subdirs) {
 641                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 642                 lustre_dump_dentry(d, recur - 1);
 643         }
 644 }
 645
 646 void client_common_put_super(struct super_block *sb)
 647 {
 648         struct ll_sb_info *sbi = ll_s2sbi(sb);
 649         ENTRY;
 650
 651 #ifdef CONFIG_FS_POSIX_ACL
 652         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 653                 et_fini(&sbi->ll_et);
 654                 rct_fini(&sbi->ll_rct);
 655         }
 656 #endif
 657
 658         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 659
 660         ll_close_thread_shutdown(sbi->ll_lcq);
 661
 662         cl_sb_fini(sb);
 663
 664         cfs_list_del(&sbi->ll_conn_chain);
 665
 666         obd_fid_fini(sbi->ll_dt_exp);
 667         obd_disconnect(sbi->ll_dt_exp);
 668         sbi->ll_dt_exp = NULL;
 669
 670         lprocfs_unregister_mountpoint(sbi);
 671
 672         obd_fid_fini(sbi->ll_md_exp);
 673         obd_disconnect(sbi->ll_md_exp);
 674         sbi->ll_md_exp = NULL;
 675
 676         EXIT;
 677 }
 678
 679 void ll_kill_super(struct super_block *sb)
 680 {
 681         struct ll_sb_info *sbi;
 682
 683         ENTRY;
 684
 685         /* not init sb ?*/
 686         if (!(sb->s_flags & MS_ACTIVE))
 687                 return;
 688
 689         sbi = ll_s2sbi(sb);
 690         /* we need restore s_dev from changed for clustred NFS before put_super
 691          * because new kernels have cached s_dev and change sb->s_dev in
 692          * put_super not affected real removing devices */
 693         if (sbi)
 694                 sb->s_dev = sbi->ll_sdev_orig;
 695         EXIT;
 696 }
 697
 698 char *ll_read_opt(const char *opt, char *data)
 699 {
 700         char *value;
 701         char *retval;
 702         ENTRY;
 703
 704         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 705         if (strncmp(opt, data, strlen(opt)))
 706                 RETURN(NULL);
 707         if ((value = strchr(data, '=')) == NULL)
 708                 RETURN(NULL);
 709
 710         value++;
 711         OBD_ALLOC(retval, strlen(value) + 1);
 712         if (!retval) {
 713                 CERROR("out of memory!\n");
 714                 RETURN(NULL);
 715         }
 716
 717         memcpy(retval, value, strlen(value)+1);
 718         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 719         RETURN(retval);
 720 }
 721
 722 static inline int ll_set_opt(const char *opt, char *data, int fl)
 723 {
 724         if (strncmp(opt, data, strlen(opt)) != 0)
 725                 return(0);
 726         else
 727                 return(fl);
 728 }
 729
 730 /* non-client-specific mount options are parsed in lmd_parse */
 731 static int ll_options(char *options, int *flags)
 732 {
 733         int tmp;
 734         char *s1 = options, *s2;
 735         ENTRY;
 736
 737         if (!options)
 738                 RETURN(0);
 739
 740         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 741
 742         while (*s1) {
 743                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 744                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 745                 if (tmp) {
 746                         *flags |= tmp;
 747                         goto next;
 748                 }
 749                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 750                 if (tmp) {
 751                         *flags |= tmp;
 752                         goto next;
 753                 }
 754                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 755                 if (tmp) {
 756                         *flags |= tmp;
 757                         goto next;
 758                 }
 759                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 760                 if (tmp) {
 761                         *flags &= ~tmp;
 762                         goto next;
 763                 }
 764                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 765                 if (tmp) {
 766                         *flags |= tmp;
 767                         goto next;
 768                 }
 769                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 770                 if (tmp) {
 771                         *flags &= ~tmp;
 772                         goto next;
 773                 }
 774 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 775                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 776                 if (tmp) {
 777                         /* Ignore deprecated mount option.  The client will
 778                          * always try to mount with ACL support, whether this
 779                          * is used depends on whether server supports it. */
 780                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 781                                                   "mount option 'acl'.\n");
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 785                 if (tmp) {
 786                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 787                                                   "mount option 'noacl'.\n");
 788                         goto next;
 789                 }
 790 #else
 791 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 792 #endif
 793                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 794                 if (tmp) {
 795                         *flags |= tmp;
 796                         goto next;
 797                 }
 798
 799                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 805                 if (tmp) {
 806                         *flags &= ~tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 810                 if (tmp) {
 811                         *flags |= tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 815                 if (tmp) {
 816                         *flags &= ~tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 825                 if (tmp) {
 826                         *flags &= ~tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 835                 if (tmp) {
 836                         *flags |= tmp;
 837                         goto next;
 838                 }
 839
 840                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 841                                    s1);
 842                 RETURN(-EINVAL);
 843
 844 next:
 845                 /* Find next opt */
 846                 s2 = strchr(s1, ',');
 847                 if (s2 == NULL)
 848                         break;
 849                 s1 = s2 + 1;
 850         }
 851         RETURN(0);
 852 }
 853
 854 void ll_lli_init(struct ll_inode_info *lli)
 855 {
 856         lli->lli_inode_magic = LLI_INODE_MAGIC;
 857         lli->lli_flags = 0;
 858         lli->lli_ioepoch = 0;
 859         cfs_spin_lock_init(&lli->lli_lock);
 860         lli->lli_posix_acl = NULL;
 861         lli->lli_remote_perms = NULL;
 862         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 863         /* Do not set lli_fid, it has been initialized already. */
 864         fid_zero(&lli->lli_pfid);
 865         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 866         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 867         cfs_atomic_set(&lli->lli_open_count, 0);
 868         lli->lli_mds_capa = NULL;
 869         lli->lli_rmtperm_time = 0;
 870         lli->lli_pending_och = NULL;
 871         lli->lli_mds_read_och = NULL;
 872         lli->lli_mds_write_och = NULL;
 873         lli->lli_mds_exec_och = NULL;
 874         lli->lli_open_fd_read_count = 0;
 875         lli->lli_open_fd_write_count = 0;
 876         lli->lli_open_fd_exec_count = 0;
 877         cfs_sema_init(&lli->lli_och_sem, 1);
 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_sema_init(&lli->lli_readdir_sem, 1);
 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                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 895                 cfs_init_rwsem(&lli->lli_trunc_sem);
 896                 cfs_sema_init(&lli->lli_write_sem, 1);
 897                 lli->lli_async_rc = 0;
 898                 lli->lli_write_rc = 0;
 899         }
 900 }
 901
 902 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 903 {
 904 #ifdef HAVE_BDI_REGISTER
 905         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 906
 907 #ifdef HAVE_BDI_NAME
 908         bdi->name = "lustre";
 909 #endif
 910         return bdi_register(bdi, NULL, "lustre-%d",
 911                             atomic_inc_return(&ll_bdi_num));
 912 #else
 913         return 0;
 914 #endif
 915 }
 916
 917 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 918 {
 919         struct lustre_profile *lprof = NULL;
 920         struct lustre_sb_info *lsi = s2lsi(sb);
 921         struct ll_sb_info *sbi;
 922         char  *dt = NULL, *md = NULL;
 923         char  *profilenm = get_profile_name(sb);
 924         struct config_llog_instance *cfg;
 925         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 926         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 927         int    err;
 928         ENTRY;
 929
 930         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 931
 932         OBD_ALLOC_PTR(cfg);
 933         if (cfg == NULL)
 934                 RETURN(-ENOMEM);
 935
 936         cfs_module_get();
 937
 938         /* client additional sb info */
 939         lsi->lsi_llsbi = sbi = ll_init_sbi();
 940         if (!sbi) {
 941                 cfs_module_put(THIS_MODULE);
 942                 OBD_FREE_PTR(cfg);
 943                 RETURN(-ENOMEM);
 944         }
 945
 946         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 947         if (err)
 948                 GOTO(out_free, err);
 949
 950         err = ll_bdi_init(&lsi->lsi_bdi);
 951         if (err)
 952                 GOTO(out_free, err);
 953         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 954         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 955         err = ll_bdi_register(&lsi->lsi_bdi);
 956         if (err)
 957                 GOTO(out_free, err);
 958
 959 #ifdef HAVE_SB_BDI
 960         sb->s_bdi = &lsi->lsi_bdi;
 961 #endif
 962
 963         /* Generate a string unique to this super, in case some joker tries
 964            to mount the same fs at two mount points.
 965            Use the address of the super itself.*/
 966         cfg->cfg_instance = sb;
 967         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 968
 969         /* set up client obds */
 970         err = lustre_process_log(sb, profilenm, cfg);
 971         if (err < 0) {
 972                 CERROR("Unable to process log: %d\n", err);
 973                 GOTO(out_free, err);
 974         }
 975
 976         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 977         lprof = class_get_profile(profilenm);
 978         if (lprof == NULL) {
 979                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 980                                    " read from the MGS.  Does that filesystem "
 981                                    "exist?\n", profilenm);
 982                 GOTO(out_free, err = -EINVAL);
 983         }
 984         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 985                lprof->lp_md, lprof->lp_dt);
 986
 987         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
 988         if (!dt)
 989                 GOTO(out_free, err = -ENOMEM);
 990         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 991
 992         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
 993         if (!md)
 994                 GOTO(out_free, err = -ENOMEM);
 995         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
 996
 997         /* connections, registrations, sb setup */
 998         err = client_common_fill_super(sb, md, dt, mnt);
 999
1000 out_free:
1001         if (md)
1002                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1003         if (dt)
1004                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1005         if (err)
1006                 ll_put_super(sb);
1007         else
1008                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1009
1010         OBD_FREE_PTR(cfg);
1011         RETURN(err);
1012 } /* ll_fill_super */
1013
1014
1015 void lu_context_keys_dump(void);
1016
1017 void ll_put_super(struct super_block *sb)
1018 {
1019         struct config_llog_instance cfg;
1020         struct obd_device *obd;
1021         struct lustre_sb_info *lsi = s2lsi(sb);
1022         struct ll_sb_info *sbi = ll_s2sbi(sb);
1023         char *profilenm = get_profile_name(sb);
1024         int force = 1, next;
1025         ENTRY;
1026
1027         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1028
1029         ll_print_capa_stat(sbi);
1030
1031         cfg.cfg_instance = sb;
1032         lustre_end_log(sb, profilenm, &cfg);
1033
1034         if (sbi->ll_md_exp) {
1035                 obd = class_exp2obd(sbi->ll_md_exp);
1036                 if (obd)
1037                         force = obd->obd_force;
1038         }
1039
1040         /* We need to set force before the lov_disconnect in
1041            lustre_common_put_super, since l_d cleans up osc's as well. */
1042         if (force) {
1043                 next = 0;
1044                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1045                                                      &next)) != NULL) {
1046                         obd->obd_force = force;
1047                 }
1048         }
1049
1050         if (sbi->ll_lcq) {
1051                 /* Only if client_common_fill_super succeeded */
1052                 client_common_put_super(sb);
1053         }
1054
1055         next = 0;
1056         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1057                 class_manual_cleanup(obd);
1058         }
1059
1060         if (profilenm)
1061                 class_del_profile(profilenm);
1062
1063         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1064                 ll_bdi_destroy(&lsi->lsi_bdi);
1065                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1066         }
1067
1068         ll_free_sbi(sb);
1069         lsi->lsi_llsbi = NULL;
1070
1071         lustre_common_put_super(sb);
1072
1073         cl_env_cache_purge(~0);
1074
1075         LCONSOLE_WARN("client %p umount complete\n", cfg.cfg_instance);
1076
1077         cfs_module_put(THIS_MODULE);
1078
1079         EXIT;
1080 } /* client_put_super */
1081
1082 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1083 {
1084         struct inode *inode = NULL;
1085         /* NOTE: we depend on atomic igrab() -bzzz */
1086         lock_res_and_lock(lock);
1087         if (lock->l_ast_data) {
1088                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1089                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1090                         inode = igrab(lock->l_ast_data);
1091                 } else {
1092                         inode = lock->l_ast_data;
1093                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1094                                          D_WARNING, lock, "l_ast_data %p is "
1095                                          "bogus: magic %08x", lock->l_ast_data,
1096                                          lli->lli_inode_magic);
1097                         inode = NULL;
1098                 }
1099         }
1100         unlock_res_and_lock(lock);
1101         return inode;
1102 }
1103
1104 static int null_if_equal(struct ldlm_lock *lock, void *data)
1105 {
1106         if (data == lock->l_ast_data) {
1107                 lock->l_ast_data = NULL;
1108
1109                 if (lock->l_req_mode != lock->l_granted_mode)
1110                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1111         }
1112
1113         return LDLM_ITER_CONTINUE;
1114 }
1115
1116 void ll_clear_inode(struct inode *inode)
1117 {
1118         struct ll_inode_info *lli = ll_i2info(inode);
1119         struct ll_sb_info *sbi = ll_i2sbi(inode);
1120         ENTRY;
1121
1122         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1123                inode->i_generation, inode);
1124
1125         if (S_ISDIR(inode->i_mode)) {
1126                 /* these should have been cleared in ll_file_release */
1127                 LASSERT(lli->lli_opendir_key == NULL);
1128                 LASSERT(lli->lli_sai == NULL);
1129                 LASSERT(lli->lli_opendir_pid == 0);
1130         }
1131
1132         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1133         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1134                          null_if_equal, inode);
1135
1136         LASSERT(!lli->lli_open_fd_write_count);
1137         LASSERT(!lli->lli_open_fd_read_count);
1138         LASSERT(!lli->lli_open_fd_exec_count);
1139
1140         if (lli->lli_mds_write_och)
1141                 ll_md_real_close(inode, FMODE_WRITE);
1142         if (lli->lli_mds_exec_och)
1143                 ll_md_real_close(inode, FMODE_EXEC);
1144         if (lli->lli_mds_read_och)
1145                 ll_md_real_close(inode, FMODE_READ);
1146
1147         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1148                 OBD_FREE(lli->lli_symlink_name,
1149                          strlen(lli->lli_symlink_name) + 1);
1150                 lli->lli_symlink_name = NULL;
1151         }
1152
1153         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1154                 LASSERT(lli->lli_posix_acl == NULL);
1155                 if (lli->lli_remote_perms) {
1156                         free_rmtperm_hash(lli->lli_remote_perms);
1157                         lli->lli_remote_perms = NULL;
1158                 }
1159         }
1160 #ifdef CONFIG_FS_POSIX_ACL
1161         else if (lli->lli_posix_acl) {
1162                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1163                 LASSERT(lli->lli_remote_perms == NULL);
1164                 posix_acl_release(lli->lli_posix_acl);
1165                 lli->lli_posix_acl = NULL;
1166         }
1167 #endif
1168         lli->lli_inode_magic = LLI_INODE_DEAD;
1169
1170         ll_clear_inode_capas(inode);
1171         /*
1172          * XXX This has to be done before lsm is freed below, because
1173          * cl_object still uses inode lsm.
1174          */
1175         cl_inode_fini(inode);
1176
1177         if (lli->lli_smd) {
1178                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1179                 lli->lli_smd = NULL;
1180         }
1181
1182
1183         EXIT;
1184 }
1185
1186 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1187                   struct md_open_data **mod)
1188 {
1189         struct lustre_md md;
1190         struct ll_sb_info *sbi = ll_i2sbi(inode);
1191         struct ptlrpc_request *request = NULL;
1192         int rc;
1193         ENTRY;
1194
1195         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1196                                      LUSTRE_OPC_ANY, NULL);
1197         if (IS_ERR(op_data))
1198                 RETURN(PTR_ERR(op_data));
1199
1200         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1201                         &request, mod);
1202         if (rc) {
1203                 ptlrpc_req_finished(request);
1204                 if (rc == -ENOENT) {
1205                         inode->i_nlink = 0;
1206                         /* Unlinked special device node? Or just a race?
1207                          * Pretend we done everything. */
1208                         if (!S_ISREG(inode->i_mode) &&
1209                             !S_ISDIR(inode->i_mode))
1210                                 rc = inode_setattr(inode, &op_data->op_attr);
1211                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1212                         CERROR("md_setattr fails: rc = %d\n", rc);
1213                 }
1214                 RETURN(rc);
1215         }
1216
1217         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1218                               sbi->ll_md_exp, &md);
1219         if (rc) {
1220                 ptlrpc_req_finished(request);
1221                 RETURN(rc);
1222         }
1223
1224         /* We call inode_setattr to adjust timestamps.
1225          * If there is at least some data in file, we cleared ATTR_SIZE
1226          * above to avoid invoking vmtruncate, otherwise it is important
1227          * to call vmtruncate in inode_setattr to update inode->i_size
1228          * (bug 6196) */
1229         rc = inode_setattr(inode, &op_data->op_attr);
1230
1231         /* Extract epoch data if obtained. */
1232         op_data->op_handle = md.body->handle;
1233         op_data->op_ioepoch = md.body->ioepoch;
1234
1235         ll_update_inode(inode, &md);
1236         ptlrpc_req_finished(request);
1237
1238         RETURN(rc);
1239 }
1240
1241 /* Close IO epoch and send Size-on-MDS attribute update. */
1242 static int ll_setattr_done_writing(struct inode *inode,
1243                                    struct md_op_data *op_data,
1244                                    struct md_open_data *mod)
1245 {
1246         struct ll_inode_info *lli = ll_i2info(inode);
1247         int rc = 0;
1248         ENTRY;
1249
1250         LASSERT(op_data != NULL);
1251         if (!S_ISREG(inode->i_mode))
1252                 RETURN(0);
1253
1254         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1255                op_data->op_ioepoch, PFID(&lli->lli_fid));
1256
1257         op_data->op_flags = MF_EPOCH_CLOSE;
1258         ll_done_writing_attr(inode, op_data);
1259         ll_pack_inode2opdata(inode, op_data, NULL);
1260
1261         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1262         if (rc == -EAGAIN) {
1263                 /* MDS has instructed us to obtain Size-on-MDS attribute
1264                  * from OSTs and send setattr to back to MDS. */
1265                 rc = ll_som_update(inode, op_data);
1266         } else if (rc) {
1267                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1268                        inode->i_ino, rc);
1269         }
1270         RETURN(rc);
1271 }
1272
1273 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1274 {
1275         struct obd_capa *capa;
1276         int rc;
1277
1278         if (attr->ia_valid & ATTR_SIZE)
1279                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1280         else
1281                 capa = ll_mdscapa_get(inode);
1282
1283         rc = cl_setattr_ost(inode, attr, capa);
1284
1285         if (attr->ia_valid & ATTR_SIZE)
1286                 ll_truncate_free_capa(capa);
1287         else
1288                 capa_put(capa);
1289
1290         return rc;
1291 }
1292
1293 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1294  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1295  * keep these values until such a time that objects are allocated for it.
1296  * We do the MDS operations first, as it is checking permissions for us.
1297  * We don't to the MDS RPC if there is nothing that we want to store there,
1298  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1299  * going to do an RPC anyways.
1300  *
1301  * If we are doing a truncate, we will send the mtime and ctime updates
1302  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1303  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1304  * at the same time.
1305  */
1306 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1307 {
1308         struct ll_inode_info *lli = ll_i2info(inode);
1309         struct lov_stripe_md *lsm = lli->lli_smd;
1310         struct md_op_data *op_data = NULL;
1311         struct md_open_data *mod = NULL;
1312         int ia_valid = attr->ia_valid;
1313         int rc = 0, rc1 = 0;
1314         ENTRY;
1315
1316         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1317                attr->ia_valid);
1318         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1319
1320         if (ia_valid & ATTR_SIZE) {
1321                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1322                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1323                                attr->ia_size, ll_file_maxbytes(inode));
1324                         RETURN(-EFBIG);
1325                 }
1326
1327                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1328         }
1329
1330         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1331         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1332                 if (cfs_curproc_fsuid() != inode->i_uid &&
1333                     !cfs_capable(CFS_CAP_FOWNER))
1334                         RETURN(-EPERM);
1335         }
1336
1337         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1338         if (attr->ia_valid & ATTR_CTIME) {
1339                 attr->ia_ctime = CFS_CURRENT_TIME;
1340                 attr->ia_valid |= ATTR_CTIME_SET;
1341         }
1342         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1343                 attr->ia_atime = CFS_CURRENT_TIME;
1344                 attr->ia_valid |= ATTR_ATIME_SET;
1345         }
1346         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1347                 attr->ia_mtime = CFS_CURRENT_TIME;
1348                 attr->ia_valid |= ATTR_MTIME_SET;
1349         }
1350
1351         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1352                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1353                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1354                        cfs_time_current_sec());
1355
1356         /* NB: ATTR_SIZE will only be set after this point if the size
1357          * resides on the MDS, ie, this file has no objects. */
1358         if (lsm)
1359                 attr->ia_valid &= ~ATTR_SIZE;
1360
1361         /* We always do an MDS RPC, even if we're only changing the size;
1362          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1363
1364         OBD_ALLOC_PTR(op_data);
1365         if (op_data == NULL)
1366                 RETURN(-ENOMEM);
1367
1368         UNLOCK_INODE_MUTEX(inode);
1369         if (ia_valid & ATTR_SIZE)
1370                 UP_WRITE_I_ALLOC_SEM(inode);
1371         if (!S_ISDIR(inode->i_mode))
1372                 cfs_down_write(&lli->lli_trunc_sem);
1373         LOCK_INODE_MUTEX(inode);
1374         if (ia_valid & ATTR_SIZE)
1375                 DOWN_WRITE_I_ALLOC_SEM(inode);
1376
1377         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1378
1379         /* Open epoch for truncate. */
1380         if (exp_connect_som(ll_i2mdexp(inode)) &&
1381             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1382                 op_data->op_flags = MF_EPOCH_OPEN;
1383
1384         rc = ll_md_setattr(inode, op_data, &mod);
1385         if (rc)
1386                 GOTO(out, rc);
1387
1388         ll_ioepoch_open(lli, op_data->op_ioepoch);
1389         if (!lsm || !S_ISREG(inode->i_mode)) {
1390                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1391                 GOTO(out, rc = 0);
1392         }
1393
1394         if (ia_valid & ATTR_SIZE)
1395                 attr->ia_valid |= ATTR_SIZE;
1396         if (ia_valid & (ATTR_SIZE |
1397                         ATTR_ATIME | ATTR_ATIME_SET |
1398                         ATTR_MTIME | ATTR_MTIME_SET))
1399                 /* on truncate and utimes send attributes to osts, setting
1400                  * mtime/atime to past will be performed under PW 0:EOF extent
1401                  * lock (new_size:EOF for truncate)
1402                  * it may seem excessive to send mtime/atime updates to osts
1403                  * when not setting times to past, but it is necessary due to
1404                  * possible time de-synchronization */
1405                 rc = ll_setattr_ost(inode, attr);
1406         EXIT;
1407 out:
1408         if (op_data) {
1409                 if (op_data->op_ioepoch)
1410                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1411                 ll_finish_md_op_data(op_data);
1412         }
1413         if (!S_ISDIR(inode->i_mode))
1414                 cfs_up_write(&lli->lli_trunc_sem);
1415         return rc ? rc : rc1;
1416 }
1417
1418 int ll_setattr(struct dentry *de, struct iattr *attr)
1419 {
1420         int mode = de->d_inode->i_mode;
1421
1422         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1423                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1424                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1425
1426         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1427                                (ATTR_SIZE|ATTR_MODE)) &&
1428             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1429              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1430               !(attr->ia_mode & S_ISGID))))
1431                 attr->ia_valid |= ATTR_FORCE;
1432
1433         if ((mode & S_ISUID) &&
1434             !(attr->ia_mode & S_ISUID) &&
1435             !(attr->ia_valid & ATTR_KILL_SUID))
1436                 attr->ia_valid |= ATTR_KILL_SUID;
1437
1438         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1439             !(attr->ia_mode & S_ISGID) &&
1440             !(attr->ia_valid & ATTR_KILL_SGID))
1441                 attr->ia_valid |= ATTR_KILL_SGID;
1442
1443         return ll_setattr_raw(de->d_inode, attr);
1444 }
1445
1446 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1447                        __u64 max_age, __u32 flags)
1448 {
1449         struct ll_sb_info *sbi = ll_s2sbi(sb);
1450         struct obd_statfs obd_osfs;
1451         int rc;
1452         ENTRY;
1453
1454         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1455         if (rc) {
1456                 CERROR("md_statfs fails: rc = %d\n", rc);
1457                 RETURN(rc);
1458         }
1459
1460         osfs->os_type = sb->s_magic;
1461
1462         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1463                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1464
1465         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1466                 flags |= OBD_STATFS_NODELAY;
1467
1468         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1469                               &obd_osfs, max_age, flags);
1470         if (rc) {
1471                 CERROR("obd_statfs fails: rc = %d\n", rc);
1472                 RETURN(rc);
1473         }
1474
1475         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1476                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1477                obd_osfs.os_files);
1478
1479         osfs->os_bsize = obd_osfs.os_bsize;
1480         osfs->os_blocks = obd_osfs.os_blocks;
1481         osfs->os_bfree = obd_osfs.os_bfree;
1482         osfs->os_bavail = obd_osfs.os_bavail;
1483
1484         /* If we don't have as many objects free on the OST as inodes
1485          * on the MDS, we reduce the total number of inodes to
1486          * compensate, so that the "inodes in use" number is correct.
1487          */
1488         if (obd_osfs.os_ffree < osfs->os_ffree) {
1489                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1490                         obd_osfs.os_ffree;
1491                 osfs->os_ffree = obd_osfs.os_ffree;
1492         }
1493
1494         RETURN(rc);
1495 }
1496 #ifndef HAVE_STATFS_DENTRY_PARAM
1497 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1498 {
1499 #else
1500 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1501 {
1502         struct super_block *sb = de->d_sb;
1503 #endif
1504         struct obd_statfs osfs;
1505         int rc;
1506
1507         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1508         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1509
1510         /* Some amount of caching on the client is allowed */
1511         rc = ll_statfs_internal(sb, &osfs,
1512                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1513                                 0);
1514         if (rc)
1515                 return rc;
1516
1517         statfs_unpack(sfs, &osfs);
1518
1519         /* We need to downshift for all 32-bit kernels, because we can't
1520          * tell if the kernel is being called via sys_statfs64() or not.
1521          * Stop before overflowing f_bsize - in which case it is better
1522          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1523         if (sizeof(long) < 8) {
1524                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1525                         sfs->f_bsize <<= 1;
1526
1527                         osfs.os_blocks >>= 1;
1528                         osfs.os_bfree >>= 1;
1529                         osfs.os_bavail >>= 1;
1530                 }
1531         }
1532
1533         sfs->f_blocks = osfs.os_blocks;
1534         sfs->f_bfree = osfs.os_bfree;
1535         sfs->f_bavail = osfs.os_bavail;
1536
1537         return 0;
1538 }
1539
1540 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1541 {
1542         struct ll_inode_info *lli;
1543         struct lov_stripe_md *lsm;
1544
1545         LASSERT(!S_ISDIR(inode->i_mode));
1546
1547         lli = ll_i2info(inode);
1548         LASSERT(lli->lli_size_sem_owner != current);
1549         cfs_down(&lli->lli_size_sem);
1550         LASSERT(lli->lli_size_sem_owner == NULL);
1551         lli->lli_size_sem_owner = current;
1552         lsm = lli->lli_smd;
1553         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1554                  lsm, lock_lsm);
1555         if (lock_lsm)
1556                 lov_stripe_lock(lsm);
1557 }
1558
1559 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1560 {
1561         struct ll_inode_info *lli;
1562         struct lov_stripe_md *lsm;
1563
1564         lli = ll_i2info(inode);
1565         lsm = lli->lli_smd;
1566         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1567                  lsm, unlock_lsm);
1568         if (unlock_lsm)
1569                 lov_stripe_unlock(lsm);
1570         LASSERT(lli->lli_size_sem_owner == current);
1571         lli->lli_size_sem_owner = NULL;
1572         cfs_up(&lli->lli_size_sem);
1573 }
1574
1575 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1576 {
1577         struct ll_inode_info *lli = ll_i2info(inode);
1578         struct mdt_body *body = md->body;
1579         struct lov_stripe_md *lsm = md->lsm;
1580         struct ll_sb_info *sbi = ll_i2sbi(inode);
1581
1582         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1583         if (lsm != NULL) {
1584                 LASSERT(S_ISREG(inode->i_mode));
1585
1586                 cfs_down(&lli->lli_och_sem);
1587                 if (lli->lli_smd == NULL) {
1588                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1589                             lsm->lsm_magic != LOV_MAGIC_V3) {
1590                                 dump_lsm(D_ERROR, lsm);
1591                                 LBUG();
1592                         }
1593                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1594                                lsm, inode->i_ino, inode->i_generation, inode);
1595                         /* cl_file_inode_init must go before lli_smd or a race
1596                          * is possible where client thinks the file has stripes,
1597                          * but lov raid0 is not setup yet and parallel e.g.
1598                          * glimpse would try to use uninitialized lov */
1599                         cl_file_inode_init(inode, md);
1600                         cfs_spin_lock(&lli->lli_lock);
1601                         lli->lli_smd = lsm;
1602                         cfs_spin_unlock(&lli->lli_lock);
1603                         cfs_up(&lli->lli_och_sem);
1604                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1605                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1606                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1607                 } else {
1608                         cfs_up(&lli->lli_och_sem);
1609                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1610                                 lli->lli_smd->lsm_stripe_count ==
1611                                 lsm->lsm_stripe_count);
1612                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1613                                 CERROR("lsm mismatch for inode %ld\n",
1614                                        inode->i_ino);
1615                                 CERROR("lli_smd:\n");
1616                                 dump_lsm(D_ERROR, lli->lli_smd);
1617                                 CERROR("lsm:\n");
1618                                 dump_lsm(D_ERROR, lsm);
1619                                 LBUG();
1620                         }
1621                 }
1622                 if (lli->lli_smd != lsm)
1623                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1624         }
1625
1626         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1627                 if (body->valid & OBD_MD_FLRMTPERM)
1628                         ll_update_remote_perm(inode, md->remote_perm);
1629         }
1630 #ifdef CONFIG_FS_POSIX_ACL
1631         else if (body->valid & OBD_MD_FLACL) {
1632                 cfs_spin_lock(&lli->lli_lock);
1633                 if (lli->lli_posix_acl)
1634                         posix_acl_release(lli->lli_posix_acl);
1635                 lli->lli_posix_acl = md->posix_acl;
1636                 cfs_spin_unlock(&lli->lli_lock);
1637         }
1638 #endif
1639         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1640         inode->i_generation = cl_fid_build_gen(&body->fid1);
1641
1642         if (body->valid & OBD_MD_FLATIME) {
1643                 if (body->atime > LTIME_S(inode->i_atime))
1644                         LTIME_S(inode->i_atime) = body->atime;
1645                 lli->lli_lvb.lvb_atime = body->atime;
1646         }
1647         if (body->valid & OBD_MD_FLMTIME) {
1648                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1649                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1650                                "to "LPU64"\n", inode->i_ino,
1651                                LTIME_S(inode->i_mtime), body->mtime);
1652                         LTIME_S(inode->i_mtime) = body->mtime;
1653                 }
1654                 lli->lli_lvb.lvb_mtime = body->mtime;
1655         }
1656         if (body->valid & OBD_MD_FLCTIME) {
1657                 if (body->ctime > LTIME_S(inode->i_ctime))
1658                         LTIME_S(inode->i_ctime) = body->ctime;
1659                 lli->lli_lvb.lvb_ctime = body->ctime;
1660         }
1661         if (body->valid & OBD_MD_FLMODE)
1662                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1663         if (body->valid & OBD_MD_FLTYPE)
1664                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1665         LASSERT(inode->i_mode != 0);
1666         if (S_ISREG(inode->i_mode)) {
1667                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1668         } else {
1669                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1670         }
1671 #ifdef HAVE_INODE_BLKSIZE
1672         inode->i_blksize = 1<<inode->i_blkbits;
1673 #endif
1674         if (body->valid & OBD_MD_FLUID)
1675                 inode->i_uid = body->uid;
1676         if (body->valid & OBD_MD_FLGID)
1677                 inode->i_gid = body->gid;
1678         if (body->valid & OBD_MD_FLFLAGS)
1679                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1680         if (body->valid & OBD_MD_FLNLINK)
1681                 inode->i_nlink = body->nlink;
1682         if (body->valid & OBD_MD_FLRDEV)
1683                 inode->i_rdev = old_decode_dev(body->rdev);
1684
1685         if (body->valid & OBD_MD_FLID) {
1686                 /* FID shouldn't be changed! */
1687                 if (fid_is_sane(&lli->lli_fid)) {
1688                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1689                                  "Trying to change FID "DFID
1690                                  " to the "DFID", inode %lu/%u(%p)\n",
1691                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1692                                  inode->i_ino, inode->i_generation, inode);
1693                 } else
1694                         lli->lli_fid = body->fid1;
1695         }
1696
1697         LASSERT(fid_seq(&lli->lli_fid) != 0);
1698
1699         if (body->valid & OBD_MD_FLSIZE) {
1700                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1701                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1702                         struct lustre_handle lockh;
1703                         ldlm_mode_t mode;
1704
1705                         /* As it is possible a blocking ast has been processed
1706                          * by this time, we need to check there is an UPDATE
1707                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1708                          * it. */
1709                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1710                                                &lockh);
1711                         if (mode) {
1712                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1713                                                       LLIF_EPOCH_PENDING |
1714                                                       LLIF_SOM_DIRTY)) {
1715                                         CERROR("ino %lu flags %u still has "
1716                                                "size authority! do not trust "
1717                                                "the size got from MDS\n",
1718                                                inode->i_ino, lli->lli_flags);
1719                                 } else {
1720                                         /* Use old size assignment to avoid
1721                                          * deadlock bz14138 & bz14326 */
1722                                         inode->i_size = body->size;
1723                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1724                                 }
1725                                 ldlm_lock_decref(&lockh, mode);
1726                         }
1727                 } else {
1728                         /* Use old size assignment to avoid
1729                          * deadlock bz14138 & bz14326 */
1730                         inode->i_size = body->size;
1731
1732                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1733                                inode->i_ino, (unsigned long long)body->size);
1734                 }
1735
1736                 if (body->valid & OBD_MD_FLBLOCKS)
1737                         inode->i_blocks = body->blocks;
1738         }
1739
1740         if (body->valid & OBD_MD_FLMDSCAPA) {
1741                 LASSERT(md->mds_capa);
1742                 ll_add_capa(inode, md->mds_capa);
1743         }
1744         if (body->valid & OBD_MD_FLOSSCAPA) {
1745                 LASSERT(md->oss_capa);
1746                 ll_add_capa(inode, md->oss_capa);
1747         }
1748 }
1749
1750 void ll_read_inode2(struct inode *inode, void *opaque)
1751 {
1752         struct lustre_md *md = opaque;
1753         struct ll_inode_info *lli = ll_i2info(inode);
1754         ENTRY;
1755
1756         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1757                PFID(&lli->lli_fid), inode);
1758
1759         LASSERT(!lli->lli_smd);
1760
1761         /* Core attributes from the MDS first.  This is a new inode, and
1762          * the VFS doesn't zero times in the core inode so we have to do
1763          * it ourselves.  They will be overwritten by either MDS or OST
1764          * attributes - we just need to make sure they aren't newer. */
1765         LTIME_S(inode->i_mtime) = 0;
1766         LTIME_S(inode->i_atime) = 0;
1767         LTIME_S(inode->i_ctime) = 0;
1768         inode->i_rdev = 0;
1769         ll_update_inode(inode, md);
1770
1771         /* OIDEBUG(inode); */
1772
1773         /* initializing backing dev info. */
1774         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1775
1776
1777         if (S_ISREG(inode->i_mode)) {
1778                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1779                 inode->i_op = &ll_file_inode_operations;
1780                 inode->i_fop = sbi->ll_fop;
1781                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1782                 EXIT;
1783         } else if (S_ISDIR(inode->i_mode)) {
1784                 inode->i_op = &ll_dir_inode_operations;
1785                 inode->i_fop = &ll_dir_operations;
1786                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1787                 EXIT;
1788         } else if (S_ISLNK(inode->i_mode)) {
1789                 inode->i_op = &ll_fast_symlink_inode_operations;
1790                 EXIT;
1791         } else {
1792                 inode->i_op = &ll_special_inode_operations;
1793
1794                 init_special_inode(inode, inode->i_mode,
1795                                    kdev_t_to_nr(inode->i_rdev));
1796
1797                 EXIT;
1798         }
1799 }
1800
1801 void ll_delete_inode(struct inode *inode)
1802 {
1803         struct ll_sb_info *sbi = ll_i2sbi(inode);
1804         int rc;
1805         ENTRY;
1806
1807         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1808         if (rc)
1809                 CERROR("fid_delete() failed, rc %d\n", rc);
1810
1811         truncate_inode_pages(&inode->i_data, 0);
1812
1813         /* Workaround for LU-118 */
1814         if (inode->i_data.nrpages) {
1815                 TREE_READ_LOCK_IRQ(&inode->i_data);
1816                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1817                 LASSERTF(inode->i_data.nrpages == 0,
1818                          "inode=%lu/%u(%p) nrpages=%lu, see "
1819                          "http://jira.whamcloud.com/browse/LU-118\n",
1820                          inode->i_ino, inode->i_generation, inode,
1821                          inode->i_data.nrpages);
1822         }
1823         /* Workaround end */
1824
1825 #ifdef HAVE_SBOPS_EVICT_INODE
1826         ll_clear_inode(inode);
1827         end_writeback(inode);
1828 #else
1829         clear_inode(inode);
1830 #endif
1831
1832         EXIT;
1833 }
1834
1835 int ll_iocontrol(struct inode *inode, struct file *file,
1836                  unsigned int cmd, unsigned long arg)
1837 {
1838         struct ll_sb_info *sbi = ll_i2sbi(inode);
1839         struct ptlrpc_request *req = NULL;
1840         int rc, flags = 0;
1841         ENTRY;
1842
1843         switch(cmd) {
1844         case FSFILT_IOC_GETFLAGS: {
1845                 struct mdt_body *body;
1846                 struct md_op_data *op_data;
1847
1848                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1849                                              0, 0, LUSTRE_OPC_ANY,
1850                                              NULL);
1851                 if (IS_ERR(op_data))
1852                         RETURN(PTR_ERR(op_data));
1853
1854                 op_data->op_valid = OBD_MD_FLFLAGS;
1855                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1856                 ll_finish_md_op_data(op_data);
1857                 if (rc) {
1858                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1859                         RETURN(-abs(rc));
1860                 }
1861
1862                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1863
1864                 flags = body->flags;
1865
1866                 ptlrpc_req_finished(req);
1867
1868                 RETURN(put_user(flags, (int *)arg));
1869         }
1870         case FSFILT_IOC_SETFLAGS: {
1871                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1872                 struct obd_info oinfo = { { { 0 } } };
1873                 struct md_op_data *op_data;
1874
1875                 if (get_user(flags, (int *)arg))
1876                         RETURN(-EFAULT);
1877
1878                 oinfo.oi_md = lsm;
1879                 OBDO_ALLOC(oinfo.oi_oa);
1880                 if (!oinfo.oi_oa)
1881                         RETURN(-ENOMEM);
1882
1883                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1884                                              LUSTRE_OPC_ANY, NULL);
1885                 if (IS_ERR(op_data))
1886                         RETURN(PTR_ERR(op_data));
1887
1888                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1889                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1890                 rc = md_setattr(sbi->ll_md_exp, op_data,
1891                                 NULL, 0, NULL, 0, &req, NULL);
1892                 ll_finish_md_op_data(op_data);
1893                 ptlrpc_req_finished(req);
1894                 if (rc) {
1895                         OBDO_FREE(oinfo.oi_oa);
1896                         RETURN(rc);
1897                 }
1898
1899                 if (lsm == NULL) {
1900                         OBDO_FREE(oinfo.oi_oa);
1901                         GOTO(update_cache, rc);
1902                 }
1903
1904                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1905                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1906                 oinfo.oi_oa->o_flags = flags;
1907                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1908                                        OBD_MD_FLGROUP;
1909                 oinfo.oi_capa = ll_mdscapa_get(inode);
1910                 obdo_from_inode(oinfo.oi_oa, inode,
1911                                 &ll_i2info(inode)->lli_fid, 0);
1912                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1913                 capa_put(oinfo.oi_capa);
1914                 OBDO_FREE(oinfo.oi_oa);
1915                 if (rc) {
1916                         if (rc != -EPERM && rc != -EACCES)
1917                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1918                         RETURN(rc);
1919                 }
1920
1921                 EXIT;
1922 update_cache:
1923                 inode->i_flags = ll_ext_to_inode_flags(flags);
1924                 return 0;
1925         }
1926         default:
1927                 RETURN(-ENOSYS);
1928         }
1929
1930         RETURN(0);
1931 }
1932
1933 int ll_flush_ctx(struct inode *inode)
1934 {
1935         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1936
1937         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1938
1939         obd_set_info_async(sbi->ll_md_exp,
1940                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1941                            0, NULL, NULL);
1942         obd_set_info_async(sbi->ll_dt_exp,
1943                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1944                            0, NULL, NULL);
1945         return 0;
1946 }
1947
1948 /* umount -f client means force down, don't save state */
1949 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1950 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1951 {
1952         struct super_block *sb = vfsmnt->mnt_sb;
1953 #else
1954 void ll_umount_begin(struct super_block *sb)
1955 {
1956 #endif
1957         struct lustre_sb_info *lsi = s2lsi(sb);
1958         struct ll_sb_info *sbi = ll_s2sbi(sb);
1959         struct obd_device *obd;
1960         struct obd_ioctl_data *ioc_data;
1961         ENTRY;
1962
1963 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1964         if (!(flags & MNT_FORCE)) {
1965                 EXIT;
1966                 return;
1967         }
1968 #endif
1969
1970         /* Tell the MGC we got umount -f */
1971         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1972
1973         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1974                sb->s_count, atomic_read(&sb->s_active));
1975
1976         obd = class_exp2obd(sbi->ll_md_exp);
1977         if (obd == NULL) {
1978                 CERROR("Invalid MDC connection handle "LPX64"\n",
1979                        sbi->ll_md_exp->exp_handle.h_cookie);
1980                 EXIT;
1981                 return;
1982         }
1983         obd->obd_force = 1;
1984
1985         obd = class_exp2obd(sbi->ll_dt_exp);
1986         if (obd == NULL) {
1987                 CERROR("Invalid LOV connection handle "LPX64"\n",
1988                        sbi->ll_dt_exp->exp_handle.h_cookie);
1989                 EXIT;
1990                 return;
1991         }
1992         obd->obd_force = 1;
1993
1994         OBD_ALLOC_PTR(ioc_data);
1995         if (ioc_data) {
1996                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1997                               sizeof ioc_data, ioc_data, NULL);
1998
1999                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2000                               sizeof ioc_data, ioc_data, NULL);
2001
2002                 OBD_FREE_PTR(ioc_data);
2003         }
2004
2005
2006         /* Really, we'd like to wait until there are no requests outstanding,
2007          * and then continue.  For now, we just invalidate the requests,
2008          * schedule() and sleep one second if needed, and hope.
2009          */
2010         cfs_schedule();
2011 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2012         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2013                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2014                                                    cfs_time_seconds(1));
2015                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2016                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2017                                       "may try to umount it a bit later\n",
2018                                       atomic_read(&vfsmnt->mnt_count));
2019         }
2020 #endif
2021
2022         EXIT;
2023 }
2024
2025 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2026 {
2027         struct ll_sb_info *sbi = ll_s2sbi(sb);
2028         int err;
2029         __u32 read_only;
2030
2031         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2032                 read_only = *flags & MS_RDONLY;
2033                 err = obd_set_info_async(sbi->ll_md_exp,
2034                                          sizeof(KEY_READ_ONLY),
2035                                          KEY_READ_ONLY, sizeof(read_only),
2036                                          &read_only, NULL);
2037                 if (err) {
2038                         CERROR("Failed to change the read-only flag during "
2039                                "remount: %d\n", err);
2040                         return err;
2041                 }
2042
2043                 if (read_only)
2044                         sb->s_flags |= MS_RDONLY;
2045                 else
2046                         sb->s_flags &= ~MS_RDONLY;
2047         }
2048         return 0;
2049 }
2050
2051 int ll_prep_inode(struct inode **inode,
2052                   struct ptlrpc_request *req,
2053                   struct super_block *sb)
2054 {
2055         struct ll_sb_info *sbi = NULL;
2056         struct lustre_md md;
2057         int rc;
2058         ENTRY;
2059
2060         LASSERT(*inode || sb);
2061         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2062         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2063                               sbi->ll_md_exp, &md);
2064         if (rc)
2065                 RETURN(rc);
2066
2067         if (*inode) {
2068                 ll_update_inode(*inode, &md);
2069         } else {
2070                 LASSERT(sb != NULL);
2071
2072                 /*
2073                  * At this point server returns to client's same fid as client
2074                  * generated for creating. So using ->fid1 is okay here.
2075                  */
2076                 LASSERT(fid_is_sane(&md.body->fid1));
2077
2078                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2079                 if (*inode == NULL || IS_ERR(*inode)) {
2080                         if (md.lsm)
2081                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2082 #ifdef CONFIG_FS_POSIX_ACL
2083                         if (md.posix_acl) {
2084                                 posix_acl_release(md.posix_acl);
2085                                 md.posix_acl = NULL;
2086                         }
2087 #endif
2088                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2089                         *inode = NULL;
2090                         CERROR("new_inode -fatal: rc %d\n", rc);
2091                         GOTO(out, rc);
2092                 }
2093         }
2094
2095 out:
2096         md_free_lustre_md(sbi->ll_md_exp, &md);
2097         RETURN(rc);
2098 }
2099
2100 int ll_obd_statfs(struct inode *inode, void *arg)
2101 {
2102         struct ll_sb_info *sbi = NULL;
2103         struct obd_export *exp;
2104         char *buf = NULL;
2105         struct obd_ioctl_data *data = NULL;
2106         __u32 type;
2107         int len = 0, rc;
2108
2109         if (!inode || !(sbi = ll_i2sbi(inode)))
2110                 GOTO(out_statfs, rc = -EINVAL);
2111
2112         rc = obd_ioctl_getdata(&buf, &len, arg);
2113         if (rc)
2114                 GOTO(out_statfs, rc);
2115
2116         data = (void*)buf;
2117         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2118             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2119                 GOTO(out_statfs, rc = -EINVAL);
2120
2121         if (data->ioc_inllen1 != sizeof(__u32) ||
2122             data->ioc_inllen2 != sizeof(__u32) ||
2123             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2124             data->ioc_plen2 != sizeof(struct obd_uuid))
2125                 GOTO(out_statfs, rc = -EINVAL);
2126
2127         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2128         if (type == LL_STATFS_MDC)
2129                 exp = sbi->ll_md_exp;
2130         else if (type == LL_STATFS_LOV)
2131                 exp = sbi->ll_dt_exp;
2132         else
2133                 GOTO(out_statfs, rc = -ENODEV);
2134
2135         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2136         if (rc)
2137                 GOTO(out_statfs, rc);
2138 out_statfs:
2139         if (buf)
2140                 obd_ioctl_freedata(buf, len);
2141         return rc;
2142 }
2143
2144 int ll_process_config(struct lustre_cfg *lcfg)
2145 {
2146         char *ptr;
2147         void *sb;
2148         struct lprocfs_static_vars lvars;
2149         unsigned long x;
2150         int rc = 0;
2151
2152         lprocfs_llite_init_vars(&lvars);
2153
2154         /* The instance name contains the sb: lustre-client-aacfe000 */
2155         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2156         if (!ptr || !*(++ptr))
2157                 return -EINVAL;
2158         if (sscanf(ptr, "%lx", &x) != 1)
2159                 return -EINVAL;
2160         sb = (void *)x;
2161         /* This better be a real Lustre superblock! */
2162         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2163
2164         /* Note we have not called client_common_fill_super yet, so
2165            proc fns must be able to handle that! */
2166         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2167                                       lcfg, sb);
2168         if (rc > 0)
2169                 rc = 0;
2170         return(rc);
2171 }
2172
2173 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2174 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2175                                        struct inode *i1, struct inode *i2,
2176                                        const char *name, int namelen,
2177                                        int mode, __u32 opc, void *data)
2178 {
2179         LASSERT(i1 != NULL);
2180
2181         if (namelen > ll_i2sbi(i1)->ll_namelen)
2182                 return ERR_PTR(-ENAMETOOLONG);
2183
2184         if (op_data == NULL)
2185                 OBD_ALLOC_PTR(op_data);
2186
2187         if (op_data == NULL)
2188                 return ERR_PTR(-ENOMEM);
2189
2190         ll_i2gids(op_data->op_suppgids, i1, i2);
2191         op_data->op_fid1 = *ll_inode2fid(i1);
2192         op_data->op_capa1 = ll_mdscapa_get(i1);
2193
2194         if (i2) {
2195                 op_data->op_fid2 = *ll_inode2fid(i2);
2196                 op_data->op_capa2 = ll_mdscapa_get(i2);
2197         } else {
2198                 fid_zero(&op_data->op_fid2);
2199                 op_data->op_capa2 = NULL;
2200         }
2201
2202         op_data->op_name = name;
2203         op_data->op_namelen = namelen;
2204         op_data->op_mode = mode;
2205         op_data->op_mod_time = cfs_time_current_sec();
2206         op_data->op_fsuid = cfs_curproc_fsuid();
2207         op_data->op_fsgid = cfs_curproc_fsgid();
2208         op_data->op_cap = cfs_curproc_cap_pack();
2209         op_data->op_bias = MDS_CHECK_SPLIT;
2210         op_data->op_opc = opc;
2211         op_data->op_mds = 0;
2212         op_data->op_data = data;
2213
2214         /* If the file is being opened after mknod() (normally due to NFS)
2215          * try to use the default stripe data from parent directory for
2216          * allocating OST objects.  Try to pass the parent FID to MDS. */
2217         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2218             ll_i2info(i2)->lli_smd == NULL) {
2219                 struct ll_inode_info *lli = ll_i2info(i2);
2220
2221                 cfs_spin_lock(&lli->lli_lock);
2222                 if (likely(lli->lli_smd == NULL &&
2223                            !fid_is_zero(&lli->lli_pfid)))
2224                         op_data->op_fid1 = lli->lli_pfid;
2225                 cfs_spin_unlock(&lli->lli_lock);
2226                 /** We ignore parent's capability temporary. */
2227         }
2228
2229         return op_data;
2230 }
2231
2232 void ll_finish_md_op_data(struct md_op_data *op_data)
2233 {
2234         capa_put(op_data->op_capa1);
2235         capa_put(op_data->op_capa2);
2236         OBD_FREE_PTR(op_data);
2237 }
2238
2239 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2240 {
2241         struct ll_sb_info *sbi;
2242
2243         LASSERT((seq != NULL) && (vfs != NULL));
2244         sbi = ll_s2sbi(vfs->mnt_sb);
2245
2246         if (sbi->ll_flags & LL_SBI_NOLCK)
2247                 seq_puts(seq, ",nolock");
2248
2249         if (sbi->ll_flags & LL_SBI_FLOCK)
2250                 seq_puts(seq, ",flock");
2251
2252         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2253                 seq_puts(seq, ",localflock");
2254
2255         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2256                 seq_puts(seq, ",user_xattr");
2257
2258         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2259                 seq_puts(seq, ",lazystatfs");
2260
2261         RETURN(0);
2262 }