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