lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67 #else
  68 extern struct address_space_operations_ext ll_aops;
  69 extern struct address_space_operations_ext ll_dir_aops;
  70 #endif
  71
  72 #ifndef log2
  73 #define log2(n) cfs_ffz(~(n))
  74 #endif
  75
  76 static struct ll_sb_info *ll_init_sbi(void)
  77 {
  78         struct ll_sb_info *sbi = NULL;
  79         unsigned long pages;
  80         struct sysinfo si;
  81         class_uuid_t uuid;
  82         int i;
  83         ENTRY;
  84
  85         OBD_ALLOC(sbi, sizeof(*sbi));
  86         if (!sbi)
  87                 RETURN(NULL);
  88
  89         cfs_spin_lock_init(&sbi->ll_lock);
  90         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  91         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  92         cfs_spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  98 #ifdef HAVE_BGL_SUPPORT
  99                 sbi->ll_async_page_max = pages / 4;
 100 #else
 101                 sbi->ll_async_page_max = pages / 2;
 102 #endif
 103         } else {
 104                 sbi->ll_async_page_max = (pages / 4) * 3;
 105         }
 106
 107         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 108                                            SBI_DEFAULT_READAHEAD_MAX);
 109         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 110         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 111                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 112         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 113         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 114
 115         ll_generate_random_uuid(uuid);
 116         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 117         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 118
 119         cfs_spin_lock(&ll_sb_lock);
 120         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 121         cfs_spin_unlock(&ll_sb_lock);
 122
 123         sbi->ll_flags |= LL_SBI_VERBOSE;
 124 #ifdef ENABLE_CHECKSUM
 125         sbi->ll_flags |= LL_SBI_CHECKSUM;
 126 #endif
 127
 128 #ifdef HAVE_LRU_RESIZE_SUPPORT
 129         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 130 #endif
 131
 132         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 133                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 134                                    pp_r_hist.oh_lock);
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_w_hist.oh_lock);
 137         }
 138
 139         /* metadata statahead is enabled by default */
 140         sbi->ll_sa_max = LL_SA_RPC_DEF;
 141         cfs_atomic_set(&sbi->ll_sa_total, 0);
 142         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 143         cfs_atomic_set(&sbi->ll_agl_total, 0);
 144         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         if (proc_lustre_fs_root) {
 202                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 203                                                   dt, md);
 204                 if (err < 0)
 205                         CERROR("could not register mount in /proc/fs/lustre\n");
 206         }
 207
 208         /* indicate the features supported by this client */
 209         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 210                                   OBD_CONNECT_ATTRFID  |
 211                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 212                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 213                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 214                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 215                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 216                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 217                                   OBD_CONNECT_EINPROGRESS |
 218                                   OBD_CONNECT_JOBSTATS;
 219
 220         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 221                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 222
 223 #ifdef HAVE_LRU_RESIZE_SUPPORT
 224         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 225                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 226 #endif
 227 #ifdef CONFIG_FS_POSIX_ACL
 228         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 229 #endif
 230         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 231         data->ocd_version = LUSTRE_VERSION_CODE;
 232
 233         if (sb->s_flags & MS_RDONLY)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 235         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 236                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 237
 238 #ifdef HAVE_MS_FLOCK_LOCK
 239         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 240         sb->s_flags |= MS_FLOCK_LOCK;
 241 #endif
 242 #ifdef MS_HAS_NEW_AOPS
 243         sb->s_flags |= MS_HAS_NEW_AOPS;
 244 #endif
 245
 246         if (sbi->ll_flags & LL_SBI_FLOCK)
 247                 sbi->ll_fop = &ll_file_operations_flock;
 248         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 249                 sbi->ll_fop = &ll_file_operations;
 250         else
 251                 sbi->ll_fop = &ll_file_operations_noflock;
 252
 253         /* real client */
 254         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 255         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 256                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 257
 258         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 259
 260         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 261         if (err == -EBUSY) {
 262                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 263                                    "recovery, of which this client is not a "
 264                                    "part. Please wait for recovery to complete,"
 265                                    " abort, or time out.\n", md);
 266                 GOTO(out, err);
 267         } else if (err) {
 268                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 269                 GOTO(out, err);
 270         }
 271
 272         err = obd_fid_init(sbi->ll_md_exp);
 273         if (err) {
 274                 CERROR("Can't init metadata layer FID infrastructure, "
 275                        "rc %d\n", err);
 276                 GOTO(out_md, err);
 277         }
 278
 279         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 280                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 281         if (err)
 282                 GOTO(out_md_fid, err);
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 292         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 297                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, CFS_PAGE_SIZE);
 305                 GOTO(out_md, err = -EPROTO);
 306         }
 307
 308         size = sizeof(*data);
 309         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 310                            KEY_CONN_DATA,  &size, data, NULL);
 311         if (err) {
 312                 CERROR("Get connect data failed: %d \n", err);
 313                 GOTO(out_md, err);
 314         }
 315
 316         LASSERT(osfs->os_bsize);
 317         sb->s_blocksize = osfs->os_bsize;
 318         sb->s_blocksize_bits = log2(osfs->os_bsize);
 319         sb->s_magic = LL_SUPER_MAGIC;
 320         sb->s_maxbytes = MAX_LFS_FILESIZE;
 321         sbi->ll_namelen = osfs->os_namelen;
 322         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 323
 324         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 325             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 326                 LCONSOLE_INFO("Disabling user_xattr feature because "
 327                               "it is not supported on the server\n");
 328                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 332 #ifdef MS_POSIXACL
 333                 sb->s_flags |= MS_POSIXACL;
 334 #endif
 335                 sbi->ll_flags |= LL_SBI_ACL;
 336         } else {
 337                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 338 #ifdef MS_POSIXACL
 339                 sb->s_flags &= ~MS_POSIXACL;
 340 #endif
 341                 sbi->ll_flags &= ~LL_SBI_ACL;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 345                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 346                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 347                         LCONSOLE_INFO("client is set as remote by default.\n");
 348                 }
 349         } else {
 350                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 351                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 352                         LCONSOLE_INFO("client claims to be remote, but server "
 353                                       "rejected, forced to be local.\n");
 354                 }
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 358                 LCONSOLE_INFO("client enabled MDS capability!\n");
 359                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 360         }
 361
 362         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 363                 LCONSOLE_INFO("client enabled OSS capability!\n");
 364                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 368                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 371                 sbi->ll_md_brw_size = data->ocd_brw_size;
 372         else
 373                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 374
 375         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 376                 LCONSOLE_INFO("Layout lock feature supported.\n");
 377                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 378         }
 379
 380         obd = class_name2obd(dt);
 381         if (!obd) {
 382                 CERROR("DT %s: not setup or attached\n", dt);
 383                 GOTO(out_md_fid, err = -ENODEV);
 384         }
 385
 386         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 387                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 388                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 389                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 390                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 391                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 392                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 393                                   OBD_CONNECT_MAXBYTES |
 394                                   OBD_CONNECT_EINPROGRESS |
 395                                   OBD_CONNECT_JOBSTATS;
 396
 397         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 398                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 399
 400         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 401                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 402                  * disabled by default, because it can still be enabled on the
 403                  * fly via /proc. As a consequence, we still need to come to an
 404                  * agreement on the supported algorithms at connect time */
 405                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 406
 407                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 408                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 409                 else
 410                         data->ocd_cksum_types = cksum_types_supported_client();
 411         }
 412
 413 #ifdef HAVE_LRU_RESIZE_SUPPORT
 414         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 415 #endif
 416         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 417                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 418
 419         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 420                "ocd_grant: %d\n", data->ocd_connect_flags,
 421                data->ocd_version, data->ocd_grant);
 422
 423         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 424         obd->obd_upcall.onu_upcall = cl_ocd_update;
 425
 426         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 427
 428         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 429         if (err == -EBUSY) {
 430                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 431                                    "recovery, of which this client is not a "
 432                                    "part.  Please wait for recovery to "
 433                                    "complete, abort, or time out.\n", dt);
 434                 GOTO(out_md_fid, err);
 435         } else if (err) {
 436                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 437                 GOTO(out_md_fid, err);
 438         }
 439
 440         err = obd_fid_init(sbi->ll_dt_exp);
 441         if (err) {
 442                 CERROR("Can't init data layer FID infrastructure, "
 443                        "rc %d\n", err);
 444                 GOTO(out_dt, err);
 445         }
 446
 447         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 448         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 449         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 450         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 451         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 452
 453         fid_zero(&sbi->ll_root_fid);
 454         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 455         if (err) {
 456                 CERROR("cannot mds_connect: rc = %d\n", err);
 457                 GOTO(out_lock_cn_cb, err);
 458         }
 459         if (!fid_is_sane(&sbi->ll_root_fid)) {
 460                 CERROR("Invalid root fid during mount\n");
 461                 GOTO(out_lock_cn_cb, err = -EINVAL);
 462         }
 463         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 464
 465         sb->s_op = &lustre_super_operations;
 466 #if THREAD_SIZE >= 8192
 467         sb->s_export_op = &lustre_export_operations;
 468 #endif
 469
 470         /* make root inode
 471          * XXX: move this to after cbd setup? */
 472         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 473         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 474                 valid |= OBD_MD_FLRMTPERM;
 475         else if (sbi->ll_flags & LL_SBI_ACL)
 476                 valid |= OBD_MD_FLACL;
 477
 478         OBD_ALLOC_PTR(op_data);
 479         if (op_data == NULL)
 480                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 481
 482         op_data->op_fid1 = sbi->ll_root_fid;
 483         op_data->op_mode = 0;
 484         op_data->op_capa1 = oc;
 485         op_data->op_valid = valid;
 486
 487         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 488         if (oc)
 489                 capa_put(oc);
 490         OBD_FREE_PTR(op_data);
 491         if (err) {
 492                 CERROR("md_getattr failed for root: rc = %d\n", err);
 493                 GOTO(out_lock_cn_cb, err);
 494         }
 495         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 496                                sbi->ll_md_exp, &lmd);
 497         if (err) {
 498                 CERROR("failed to understand root inode md: rc = %d\n", err);
 499                 ptlrpc_req_finished (request);
 500                 GOTO(out_lock_cn_cb, err);
 501         }
 502
 503         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 504         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 505         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 506         ptlrpc_req_finished(request);
 507
 508         if (root == NULL || IS_ERR(root)) {
 509                 if (lmd.lsm)
 510                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 511 #ifdef CONFIG_FS_POSIX_ACL
 512                 if (lmd.posix_acl) {
 513                         posix_acl_release(lmd.posix_acl);
 514                         lmd.posix_acl = NULL;
 515                 }
 516 #endif
 517                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 518                 root = NULL;
 519                 CERROR("lustre_lite: bad iget4 for root\n");
 520                 GOTO(out_root, err);
 521         }
 522
 523         err = ll_close_thread_start(&sbi->ll_lcq);
 524         if (err) {
 525                 CERROR("cannot start close thread: rc %d\n", err);
 526                 GOTO(out_root, err);
 527         }
 528
 529 #ifdef CONFIG_FS_POSIX_ACL
 530         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 531                 rct_init(&sbi->ll_rct);
 532                 et_init(&sbi->ll_et);
 533         }
 534 #endif
 535
 536         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 537         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 538                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 539                                  NULL);
 540         cl_sb_init(sb);
 541
 542         sb->s_root = d_alloc_root(root);
 543 #ifdef HAVE_DCACHE_LOCK
 544         sb->s_root->d_op = &ll_d_root_ops;
 545 #else
 546         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 547         d_set_d_op(sb->s_root, &ll_d_root_ops);
 548         sb->s_d_op = &ll_d_ops;
 549 #endif
 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(NULL, 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, d_refcount(dentry),
 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_u.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                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 842                 if (tmp) {
 843                         *flags |= tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 847                 if (tmp) {
 848                         *flags &= ~tmp;
 849                         goto next;
 850                 }
 851                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 852                                    s1);
 853                 RETURN(-EINVAL);
 854
 855 next:
 856                 /* Find next opt */
 857                 s2 = strchr(s1, ',');
 858                 if (s2 == NULL)
 859                         break;
 860                 s1 = s2 + 1;
 861         }
 862         RETURN(0);
 863 }
 864
 865 void ll_lli_init(struct ll_inode_info *lli)
 866 {
 867         lli->lli_inode_magic = LLI_INODE_MAGIC;
 868         lli->lli_flags = 0;
 869         lli->lli_ioepoch = 0;
 870         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 871         cfs_spin_lock_init(&lli->lli_lock);
 872         lli->lli_posix_acl = NULL;
 873         lli->lli_remote_perms = NULL;
 874         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 875         /* Do not set lli_fid, it has been initialized already. */
 876         fid_zero(&lli->lli_pfid);
 877         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 878         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 879         cfs_atomic_set(&lli->lli_open_count, 0);
 880         lli->lli_mds_capa = NULL;
 881         lli->lli_rmtperm_time = 0;
 882         lli->lli_pending_och = NULL;
 883         lli->lli_mds_read_och = NULL;
 884         lli->lli_mds_write_och = NULL;
 885         lli->lli_mds_exec_och = NULL;
 886         lli->lli_open_fd_read_count = 0;
 887         lli->lli_open_fd_write_count = 0;
 888         lli->lli_open_fd_exec_count = 0;
 889         cfs_mutex_init(&lli->lli_och_mutex);
 890         cfs_spin_lock_init(&lli->lli_agl_lock);
 891         lli->lli_has_smd = false;
 892         lli->lli_clob = NULL;
 893
 894         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 895         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 896                 cfs_mutex_init(&lli->lli_readdir_mutex);
 897                 lli->lli_opendir_key = NULL;
 898                 lli->lli_sai = NULL;
 899                 lli->lli_sa_pos = 0;
 900                 lli->lli_def_acl = NULL;
 901                 cfs_spin_lock_init(&lli->lli_sa_lock);
 902                 lli->lli_opendir_pid = 0;
 903         } else {
 904                 cfs_sema_init(&lli->lli_size_sem, 1);
 905                 lli->lli_size_sem_owner = NULL;
 906                 lli->lli_symlink_name = NULL;
 907                 cfs_init_rwsem(&lli->lli_trunc_sem);
 908                 cfs_mutex_init(&lli->lli_write_mutex);
 909                 lli->lli_async_rc = 0;
 910                 lli->lli_write_rc = 0;
 911                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 912                 lli->lli_glimpse_time = 0;
 913                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 914                 lli->lli_agl_index = 0;
 915         }
 916         cfs_mutex_init(&lli->lli_layout_mutex);
 917 }
 918
 919 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 920 {
 921 #ifdef HAVE_BDI_REGISTER
 922         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 923
 924 #ifdef HAVE_BDI_NAME
 925         bdi->name = "lustre";
 926 #endif
 927         return bdi_register(bdi, NULL, "lustre-%d",
 928                             atomic_inc_return(&ll_bdi_num));
 929 #else
 930         return 0;
 931 #endif
 932 }
 933
 934 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 935 {
 936         struct lustre_profile *lprof = NULL;
 937         struct lustre_sb_info *lsi = s2lsi(sb);
 938         struct ll_sb_info *sbi;
 939         char  *dt = NULL, *md = NULL;
 940         char  *profilenm = get_profile_name(sb);
 941         struct config_llog_instance *cfg;
 942         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 943         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 944         int    err;
 945         ENTRY;
 946
 947         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 948
 949         OBD_ALLOC_PTR(cfg);
 950         if (cfg == NULL)
 951                 RETURN(-ENOMEM);
 952
 953         cfs_module_get();
 954
 955         /* client additional sb info */
 956         lsi->lsi_llsbi = sbi = ll_init_sbi();
 957         if (!sbi) {
 958                 cfs_module_put(THIS_MODULE);
 959                 OBD_FREE_PTR(cfg);
 960                 RETURN(-ENOMEM);
 961         }
 962
 963         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 964         if (err)
 965                 GOTO(out_free, err);
 966
 967         err = ll_bdi_init(&lsi->lsi_bdi);
 968         if (err)
 969                 GOTO(out_free, err);
 970         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 971         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 972         err = ll_bdi_register(&lsi->lsi_bdi);
 973         if (err)
 974                 GOTO(out_free, err);
 975
 976 #ifdef HAVE_SB_BDI
 977         sb->s_bdi = &lsi->lsi_bdi;
 978 #endif
 979
 980         /* Generate a string unique to this super, in case some joker tries
 981            to mount the same fs at two mount points.
 982            Use the address of the super itself.*/
 983         cfg->cfg_instance = sb;
 984         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 985
 986         /* set up client obds */
 987         err = lustre_process_log(sb, profilenm, cfg);
 988         if (err < 0) {
 989                 CERROR("Unable to process log: %d\n", err);
 990                 GOTO(out_free, err);
 991         }
 992
 993         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 994         lprof = class_get_profile(profilenm);
 995         if (lprof == NULL) {
 996                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 997                                    " read from the MGS.  Does that filesystem "
 998                                    "exist?\n", profilenm);
 999                 GOTO(out_free, err = -EINVAL);
1000         }
1001         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1002                lprof->lp_md, lprof->lp_dt);
1003
1004         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1005         if (!dt)
1006                 GOTO(out_free, err = -ENOMEM);
1007         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1008
1009         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1010         if (!md)
1011                 GOTO(out_free, err = -ENOMEM);
1012         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1013
1014         /* connections, registrations, sb setup */
1015         err = client_common_fill_super(sb, md, dt, mnt);
1016
1017 out_free:
1018         if (md)
1019                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1020         if (dt)
1021                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1022         if (err)
1023                 ll_put_super(sb);
1024         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1025                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1026
1027         OBD_FREE_PTR(cfg);
1028         RETURN(err);
1029 } /* ll_fill_super */
1030
1031
1032 void lu_context_keys_dump(void);
1033
1034 void ll_put_super(struct super_block *sb)
1035 {
1036         struct config_llog_instance cfg;
1037         struct obd_device *obd;
1038         struct lustre_sb_info *lsi = s2lsi(sb);
1039         struct ll_sb_info *sbi = ll_s2sbi(sb);
1040         char *profilenm = get_profile_name(sb);
1041         int force = 1, next;
1042         ENTRY;
1043
1044         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1045
1046         ll_print_capa_stat(sbi);
1047
1048         cfg.cfg_instance = sb;
1049         lustre_end_log(sb, profilenm, &cfg);
1050
1051         if (sbi->ll_md_exp) {
1052                 obd = class_exp2obd(sbi->ll_md_exp);
1053                 if (obd)
1054                         force = obd->obd_force;
1055         }
1056
1057         /* We need to set force before the lov_disconnect in
1058            lustre_common_put_super, since l_d cleans up osc's as well. */
1059         if (force) {
1060                 next = 0;
1061                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1062                                                      &next)) != NULL) {
1063                         obd->obd_force = force;
1064                 }
1065         }
1066
1067         if (sbi->ll_lcq) {
1068                 /* Only if client_common_fill_super succeeded */
1069                 client_common_put_super(sb);
1070         }
1071
1072         next = 0;
1073         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1074                 class_manual_cleanup(obd);
1075         }
1076
1077         if (sbi->ll_flags & LL_SBI_VERBOSE)
1078                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1079
1080         if (profilenm)
1081                 class_del_profile(profilenm);
1082
1083         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1084                 ll_bdi_destroy(&lsi->lsi_bdi);
1085                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1086         }
1087
1088         ll_free_sbi(sb);
1089         lsi->lsi_llsbi = NULL;
1090
1091         lustre_common_put_super(sb);
1092
1093         cl_env_cache_purge(~0);
1094
1095         cfs_module_put(THIS_MODULE);
1096
1097         EXIT;
1098 } /* client_put_super */
1099
1100 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1101 {
1102         struct inode *inode = NULL;
1103         /* NOTE: we depend on atomic igrab() -bzzz */
1104         lock_res_and_lock(lock);
1105         if (lock->l_ast_data) {
1106                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1107                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1108                         inode = igrab(lock->l_ast_data);
1109                 } else {
1110                         inode = lock->l_ast_data;
1111                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1112                                          D_WARNING, lock, "l_ast_data %p is "
1113                                          "bogus: magic %08x", lock->l_ast_data,
1114                                          lli->lli_inode_magic);
1115                         inode = NULL;
1116                 }
1117         }
1118         unlock_res_and_lock(lock);
1119         return inode;
1120 }
1121
1122 static int null_if_equal(struct ldlm_lock *lock, void *data)
1123 {
1124         if (data == lock->l_ast_data) {
1125                 lock->l_ast_data = NULL;
1126
1127                 if (lock->l_req_mode != lock->l_granted_mode)
1128                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1129         }
1130
1131         return LDLM_ITER_CONTINUE;
1132 }
1133
1134 void ll_clear_inode(struct inode *inode)
1135 {
1136         struct ll_inode_info *lli = ll_i2info(inode);
1137         struct ll_sb_info *sbi = ll_i2sbi(inode);
1138         ENTRY;
1139
1140         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1141                inode->i_generation, inode);
1142
1143         if (S_ISDIR(inode->i_mode)) {
1144                 /* these should have been cleared in ll_file_release */
1145                 LASSERT(lli->lli_opendir_key == NULL);
1146                 LASSERT(lli->lli_sai == NULL);
1147                 LASSERT(lli->lli_opendir_pid == 0);
1148         }
1149
1150         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1151         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1152                          null_if_equal, inode);
1153
1154         LASSERT(!lli->lli_open_fd_write_count);
1155         LASSERT(!lli->lli_open_fd_read_count);
1156         LASSERT(!lli->lli_open_fd_exec_count);
1157
1158         if (lli->lli_mds_write_och)
1159                 ll_md_real_close(inode, FMODE_WRITE);
1160         if (lli->lli_mds_exec_och)
1161                 ll_md_real_close(inode, FMODE_EXEC);
1162         if (lli->lli_mds_read_och)
1163                 ll_md_real_close(inode, FMODE_READ);
1164
1165         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1166                 OBD_FREE(lli->lli_symlink_name,
1167                          strlen(lli->lli_symlink_name) + 1);
1168                 lli->lli_symlink_name = NULL;
1169         }
1170
1171         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1172                 LASSERT(lli->lli_posix_acl == NULL);
1173                 if (lli->lli_remote_perms) {
1174                         free_rmtperm_hash(lli->lli_remote_perms);
1175                         lli->lli_remote_perms = NULL;
1176                 }
1177         }
1178 #ifdef CONFIG_FS_POSIX_ACL
1179         else if (lli->lli_posix_acl) {
1180                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1181                 LASSERT(lli->lli_remote_perms == NULL);
1182                 posix_acl_release(lli->lli_posix_acl);
1183                 lli->lli_posix_acl = NULL;
1184         }
1185 #endif
1186         lli->lli_inode_magic = LLI_INODE_DEAD;
1187
1188         ll_clear_inode_capas(inode);
1189         if (!S_ISDIR(inode->i_mode))
1190                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1191
1192         /*
1193          * XXX This has to be done before lsm is freed below, because
1194          * cl_object still uses inode lsm.
1195          */
1196         cl_inode_fini(inode);
1197         lli->lli_has_smd = false;
1198
1199         EXIT;
1200 }
1201
1202 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1203                   struct md_open_data **mod)
1204 {
1205         struct lustre_md md;
1206         struct inode *inode = dentry->d_inode;
1207         struct ll_sb_info *sbi = ll_i2sbi(inode);
1208         struct ptlrpc_request *request = NULL;
1209         int rc, ia_valid;
1210         ENTRY;
1211
1212         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1213                                      LUSTRE_OPC_ANY, NULL);
1214         if (IS_ERR(op_data))
1215                 RETURN(PTR_ERR(op_data));
1216
1217         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1218                         &request, mod);
1219         if (rc) {
1220                 ptlrpc_req_finished(request);
1221                 if (rc == -ENOENT) {
1222                         inode->i_nlink = 0;
1223                         /* Unlinked special device node? Or just a race?
1224                          * Pretend we done everything. */
1225                         if (!S_ISREG(inode->i_mode) &&
1226                             !S_ISDIR(inode->i_mode)) {
1227                                 ia_valid = op_data->op_attr.ia_valid;
1228                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1229                                 rc = simple_setattr(dentry, &op_data->op_attr);
1230                                 op_data->op_attr.ia_valid = ia_valid;
1231                         }
1232                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1233                         CERROR("md_setattr fails: rc = %d\n", rc);
1234                 }
1235                 RETURN(rc);
1236         }
1237
1238         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1239                               sbi->ll_md_exp, &md);
1240         if (rc) {
1241                 ptlrpc_req_finished(request);
1242                 RETURN(rc);
1243         }
1244
1245         /* We want to adjust timestamps.
1246          * If there is at least some data in file, we cleared ATTR_SIZE
1247          * to avoid update size, otherwise it is important to do.(SOM case)
1248          * (bug 6196) */
1249         ia_valid = op_data->op_attr.ia_valid;
1250         /* Since we set ATTR_*_SET flags above, and already done permission
1251          * check, So don't let inode_change_ok() check it again.  */
1252         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1253         rc = simple_setattr(dentry, &op_data->op_attr);
1254         op_data->op_attr.ia_valid = ia_valid;
1255
1256         /* Extract epoch data if obtained. */
1257         op_data->op_handle = md.body->handle;
1258         op_data->op_ioepoch = md.body->ioepoch;
1259
1260         ll_update_inode(inode, &md);
1261         ptlrpc_req_finished(request);
1262
1263         RETURN(rc);
1264 }
1265
1266 /* Close IO epoch and send Size-on-MDS attribute update. */
1267 static int ll_setattr_done_writing(struct inode *inode,
1268                                    struct md_op_data *op_data,
1269                                    struct md_open_data *mod)
1270 {
1271         struct ll_inode_info *lli = ll_i2info(inode);
1272         int rc = 0;
1273         ENTRY;
1274
1275         LASSERT(op_data != NULL);
1276         if (!S_ISREG(inode->i_mode))
1277                 RETURN(0);
1278
1279         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1280                op_data->op_ioepoch, PFID(&lli->lli_fid));
1281
1282         op_data->op_flags = MF_EPOCH_CLOSE;
1283         ll_done_writing_attr(inode, op_data);
1284         ll_pack_inode2opdata(inode, op_data, NULL);
1285
1286         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1287         if (rc == -EAGAIN) {
1288                 /* MDS has instructed us to obtain Size-on-MDS attribute
1289                  * from OSTs and send setattr to back to MDS. */
1290                 rc = ll_som_update(inode, op_data);
1291         } else if (rc) {
1292                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1293                        inode->i_ino, rc);
1294         }
1295         RETURN(rc);
1296 }
1297
1298 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1299 {
1300         struct obd_capa *capa;
1301         int rc;
1302
1303         if (attr->ia_valid & ATTR_SIZE)
1304                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1305         else
1306                 capa = ll_mdscapa_get(inode);
1307
1308         rc = cl_setattr_ost(inode, attr, capa);
1309
1310         if (attr->ia_valid & ATTR_SIZE)
1311                 ll_truncate_free_capa(capa);
1312         else
1313                 capa_put(capa);
1314
1315         return rc;
1316 }
1317
1318 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1319  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1320  * keep these values until such a time that objects are allocated for it.
1321  * We do the MDS operations first, as it is checking permissions for us.
1322  * We don't to the MDS RPC if there is nothing that we want to store there,
1323  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1324  * going to do an RPC anyways.
1325  *
1326  * If we are doing a truncate, we will send the mtime and ctime updates
1327  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1328  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1329  * at the same time.
1330  */
1331 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1332 {
1333         struct lov_stripe_md *lsm;
1334         struct inode *inode = dentry->d_inode;
1335         struct ll_inode_info *lli = ll_i2info(inode);
1336         struct md_op_data *op_data = NULL;
1337         struct md_open_data *mod = NULL;
1338         int ia_valid = attr->ia_valid;
1339         int rc = 0, rc1 = 0;
1340         ENTRY;
1341
1342         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1343                attr->ia_valid);
1344
1345         if (ia_valid & ATTR_SIZE) {
1346                 /* Check new size against VFS/VM file size limit and rlimit */
1347                 rc = inode_newsize_ok(inode, attr->ia_size);
1348                 if (rc)
1349                         RETURN(rc);
1350
1351                 /* The maximum Lustre file size is variable, based on the
1352                  * OST maximum object size and number of stripes.  This
1353                  * needs another check in addition to the VFS check above. */
1354                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1355                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1356                                PFID(&lli->lli_fid), attr->ia_size,
1357                                ll_file_maxbytes(inode));
1358                         RETURN(-EFBIG);
1359                 }
1360
1361                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1362         }
1363
1364         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1365         if (ia_valid & TIMES_SET_FLAGS) {
1366                 if (cfs_curproc_fsuid() != inode->i_uid &&
1367                     !cfs_capable(CFS_CAP_FOWNER))
1368                         RETURN(-EPERM);
1369         }
1370
1371         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1372         if (attr->ia_valid & ATTR_CTIME) {
1373                 attr->ia_ctime = CFS_CURRENT_TIME;
1374                 attr->ia_valid |= ATTR_CTIME_SET;
1375         }
1376         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1377                 attr->ia_atime = CFS_CURRENT_TIME;
1378                 attr->ia_valid |= ATTR_ATIME_SET;
1379         }
1380         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1381                 attr->ia_mtime = CFS_CURRENT_TIME;
1382                 attr->ia_valid |= ATTR_MTIME_SET;
1383         }
1384
1385         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1386                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1387                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1388                        cfs_time_current_sec());
1389
1390         /* We always do an MDS RPC, even if we're only changing the size;
1391          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1392
1393         OBD_ALLOC_PTR(op_data);
1394         if (op_data == NULL)
1395                 RETURN(-ENOMEM);
1396
1397         if (!S_ISDIR(inode->i_mode)) {
1398                 if (ia_valid & ATTR_SIZE)
1399                         UP_WRITE_I_ALLOC_SEM(inode);
1400                 mutex_unlock(&inode->i_mutex);
1401                 cfs_down_write(&lli->lli_trunc_sem);
1402                 mutex_lock(&inode->i_mutex);
1403                 if (ia_valid & ATTR_SIZE)
1404                         DOWN_WRITE_I_ALLOC_SEM(inode);
1405         }
1406
1407         /* We need a steady stripe configuration for setattr to avoid
1408          * confusion. */
1409         lsm = ccc_inode_lsm_get(inode);
1410
1411         /* NB: ATTR_SIZE will only be set after this point if the size
1412          * resides on the MDS, ie, this file has no objects. */
1413         if (lsm != NULL)
1414                 attr->ia_valid &= ~ATTR_SIZE;
1415
1416         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1417
1418         /* Open epoch for truncate. */
1419         if (exp_connect_som(ll_i2mdexp(inode)) &&
1420             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1421                 op_data->op_flags = MF_EPOCH_OPEN;
1422
1423         rc = ll_md_setattr(dentry, op_data, &mod);
1424         if (rc)
1425                 GOTO(out, rc);
1426
1427         ll_ioepoch_open(lli, op_data->op_ioepoch);
1428         if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1429                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1430                 GOTO(out, rc = 0);
1431         }
1432
1433         if (ia_valid & ATTR_SIZE)
1434                 attr->ia_valid |= ATTR_SIZE;
1435         if (ia_valid & (ATTR_SIZE |
1436                         ATTR_ATIME | ATTR_ATIME_SET |
1437                         ATTR_MTIME | ATTR_MTIME_SET))
1438                 /* For truncate and utimes sending attributes to OSTs, setting
1439                  * mtime/atime to the past will be performed under PW [0:EOF]
1440                  * extent lock (new_size:EOF for truncate).  It may seem
1441                  * excessive to send mtime/atime updates to OSTs when not
1442                  * setting times to past, but it is necessary due to possible
1443                  * time de-synchronization between MDT inode and OST objects */
1444                 rc = ll_setattr_ost(inode, attr);
1445         EXIT;
1446 out:
1447         ccc_inode_lsm_put(inode, lsm);
1448         if (op_data) {
1449                 if (op_data->op_ioepoch) {
1450                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1451                         if (!rc)
1452                                 rc = rc1;
1453                 }
1454                 ll_finish_md_op_data(op_data);
1455         }
1456         if (!S_ISDIR(inode->i_mode))
1457                 cfs_up_write(&lli->lli_trunc_sem);
1458
1459         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1460                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1461
1462         return rc;
1463 }
1464
1465 int ll_setattr(struct dentry *de, struct iattr *attr)
1466 {
1467         int mode = de->d_inode->i_mode;
1468
1469         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1470                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1471                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1472
1473         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1474                                (ATTR_SIZE|ATTR_MODE)) &&
1475             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1476              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1477               !(attr->ia_mode & S_ISGID))))
1478                 attr->ia_valid |= ATTR_FORCE;
1479
1480         if ((mode & S_ISUID) &&
1481             !(attr->ia_mode & S_ISUID) &&
1482             !(attr->ia_valid & ATTR_KILL_SUID))
1483                 attr->ia_valid |= ATTR_KILL_SUID;
1484
1485         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1486             !(attr->ia_mode & S_ISGID) &&
1487             !(attr->ia_valid & ATTR_KILL_SGID))
1488                 attr->ia_valid |= ATTR_KILL_SGID;
1489
1490         return ll_setattr_raw(de, attr);
1491 }
1492
1493 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1494                        __u64 max_age, __u32 flags)
1495 {
1496         struct ll_sb_info *sbi = ll_s2sbi(sb);
1497         struct obd_statfs obd_osfs;
1498         int rc;
1499         ENTRY;
1500
1501         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1502         if (rc) {
1503                 CERROR("md_statfs fails: rc = %d\n", rc);
1504                 RETURN(rc);
1505         }
1506
1507         osfs->os_type = sb->s_magic;
1508
1509         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1510                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1511
1512         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1513                 flags |= OBD_STATFS_NODELAY;
1514
1515         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1516         if (rc) {
1517                 CERROR("obd_statfs fails: rc = %d\n", rc);
1518                 RETURN(rc);
1519         }
1520
1521         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1522                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1523                obd_osfs.os_files);
1524
1525         osfs->os_bsize = obd_osfs.os_bsize;
1526         osfs->os_blocks = obd_osfs.os_blocks;
1527         osfs->os_bfree = obd_osfs.os_bfree;
1528         osfs->os_bavail = obd_osfs.os_bavail;
1529
1530         /* If we don't have as many objects free on the OST as inodes
1531          * on the MDS, we reduce the total number of inodes to
1532          * compensate, so that the "inodes in use" number is correct.
1533          */
1534         if (obd_osfs.os_ffree < osfs->os_ffree) {
1535                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1536                         obd_osfs.os_ffree;
1537                 osfs->os_ffree = obd_osfs.os_ffree;
1538         }
1539
1540         RETURN(rc);
1541 }
1542 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1543 {
1544         struct super_block *sb = de->d_sb;
1545         struct obd_statfs osfs;
1546         int rc;
1547
1548         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1549         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1550
1551         /* Some amount of caching on the client is allowed */
1552         rc = ll_statfs_internal(sb, &osfs,
1553                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1554                                 0);
1555         if (rc)
1556                 return rc;
1557
1558         statfs_unpack(sfs, &osfs);
1559
1560         /* We need to downshift for all 32-bit kernels, because we can't
1561          * tell if the kernel is being called via sys_statfs64() or not.
1562          * Stop before overflowing f_bsize - in which case it is better
1563          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1564         if (sizeof(long) < 8) {
1565                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1566                         sfs->f_bsize <<= 1;
1567
1568                         osfs.os_blocks >>= 1;
1569                         osfs.os_bfree >>= 1;
1570                         osfs.os_bavail >>= 1;
1571                 }
1572         }
1573
1574         sfs->f_blocks = osfs.os_blocks;
1575         sfs->f_bfree = osfs.os_bfree;
1576         sfs->f_bavail = osfs.os_bavail;
1577
1578         return 0;
1579 }
1580
1581 void ll_inode_size_lock(struct inode *inode)
1582 {
1583         struct ll_inode_info *lli;
1584
1585         LASSERT(!S_ISDIR(inode->i_mode));
1586
1587         lli = ll_i2info(inode);
1588         LASSERT(lli->lli_size_sem_owner != current);
1589         cfs_down(&lli->lli_size_sem);
1590         LASSERT(lli->lli_size_sem_owner == NULL);
1591         lli->lli_size_sem_owner = current;
1592 }
1593
1594 void ll_inode_size_unlock(struct inode *inode)
1595 {
1596         struct ll_inode_info *lli;
1597
1598         lli = ll_i2info(inode);
1599         LASSERT(lli->lli_size_sem_owner == current);
1600         lli->lli_size_sem_owner = NULL;
1601         cfs_up(&lli->lli_size_sem);
1602 }
1603
1604 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1605 {
1606         struct ll_inode_info *lli = ll_i2info(inode);
1607         struct mdt_body *body = md->body;
1608         struct lov_stripe_md *lsm = md->lsm;
1609         struct ll_sb_info *sbi = ll_i2sbi(inode);
1610
1611         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1612         if (lsm != NULL) {
1613                 LASSERT(S_ISREG(inode->i_mode));
1614                 cfs_mutex_lock(&lli->lli_och_mutex);
1615                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1616                                 lsm, inode->i_ino, inode->i_generation, inode);
1617                 /* cl_file_inode_init must go before lli_has_smd or a race
1618                  * is possible where client thinks the file has stripes,
1619                  * but lov raid0 is not setup yet and parallel e.g.
1620                  * glimpse would try to use uninitialized lov */
1621                 if (cl_file_inode_init(inode, md) == 0)
1622                         lli->lli_has_smd = true;
1623                 cfs_mutex_unlock(&lli->lli_och_mutex);
1624
1625                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1626                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1627                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1628                 if (md->lsm != NULL)
1629                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1630         }
1631
1632         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1633                 if (body->valid & OBD_MD_FLRMTPERM)
1634                         ll_update_remote_perm(inode, md->remote_perm);
1635         }
1636 #ifdef CONFIG_FS_POSIX_ACL
1637         else if (body->valid & OBD_MD_FLACL) {
1638                 cfs_spin_lock(&lli->lli_lock);
1639                 if (lli->lli_posix_acl)
1640                         posix_acl_release(lli->lli_posix_acl);
1641                 lli->lli_posix_acl = md->posix_acl;
1642                 cfs_spin_unlock(&lli->lli_lock);
1643         }
1644 #endif
1645         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1646         inode->i_generation = cl_fid_build_gen(&body->fid1);
1647
1648         if (body->valid & OBD_MD_FLATIME) {
1649                 if (body->atime > LTIME_S(inode->i_atime))
1650                         LTIME_S(inode->i_atime) = body->atime;
1651                 lli->lli_lvb.lvb_atime = body->atime;
1652         }
1653         if (body->valid & OBD_MD_FLMTIME) {
1654                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1655                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1656                                "to "LPU64"\n", inode->i_ino,
1657                                LTIME_S(inode->i_mtime), body->mtime);
1658                         LTIME_S(inode->i_mtime) = body->mtime;
1659                 }
1660                 lli->lli_lvb.lvb_mtime = body->mtime;
1661         }
1662         if (body->valid & OBD_MD_FLCTIME) {
1663                 if (body->ctime > LTIME_S(inode->i_ctime))
1664                         LTIME_S(inode->i_ctime) = body->ctime;
1665                 lli->lli_lvb.lvb_ctime = body->ctime;
1666         }
1667         if (body->valid & OBD_MD_FLMODE)
1668                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1669         if (body->valid & OBD_MD_FLTYPE)
1670                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1671         LASSERT(inode->i_mode != 0);
1672         if (S_ISREG(inode->i_mode)) {
1673                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1674         } else {
1675                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1676         }
1677         if (body->valid & OBD_MD_FLUID)
1678                 inode->i_uid = body->uid;
1679         if (body->valid & OBD_MD_FLGID)
1680                 inode->i_gid = body->gid;
1681         if (body->valid & OBD_MD_FLFLAGS)
1682                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1683         if (body->valid & OBD_MD_FLNLINK)
1684                 inode->i_nlink = body->nlink;
1685         if (body->valid & OBD_MD_FLRDEV)
1686                 inode->i_rdev = old_decode_dev(body->rdev);
1687
1688         if (body->valid & OBD_MD_FLID) {
1689                 /* FID shouldn't be changed! */
1690                 if (fid_is_sane(&lli->lli_fid)) {
1691                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1692                                  "Trying to change FID "DFID
1693                                  " to the "DFID", inode %lu/%u(%p)\n",
1694                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1695                                  inode->i_ino, inode->i_generation, inode);
1696                 } else
1697                         lli->lli_fid = body->fid1;
1698         }
1699
1700         LASSERT(fid_seq(&lli->lli_fid) != 0);
1701
1702         if (body->valid & OBD_MD_FLSIZE) {
1703                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1704                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1705                         struct lustre_handle lockh;
1706                         ldlm_mode_t mode;
1707
1708                         /* As it is possible a blocking ast has been processed
1709                          * by this time, we need to check there is an UPDATE
1710                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1711                          * it. */
1712                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1713                                                &lockh);
1714                         if (mode) {
1715                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1716                                                       LLIF_EPOCH_PENDING |
1717                                                       LLIF_SOM_DIRTY)) {
1718                                         CERROR("ino %lu flags %u still has "
1719                                                "size authority! do not trust "
1720                                                "the size got from MDS\n",
1721                                                inode->i_ino, lli->lli_flags);
1722                                 } else {
1723                                         /* Use old size assignment to avoid
1724                                          * deadlock bz14138 & bz14326 */
1725                                         i_size_write(inode, body->size);
1726                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1727                                 }
1728                                 ldlm_lock_decref(&lockh, mode);
1729                         }
1730                 } else {
1731                         /* Use old size assignment to avoid
1732                          * deadlock bz14138 & bz14326 */
1733                         i_size_write(inode, body->size);
1734
1735                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1736                                inode->i_ino, (unsigned long long)body->size);
1737                 }
1738
1739                 if (body->valid & OBD_MD_FLBLOCKS)
1740                         inode->i_blocks = body->blocks;
1741         }
1742
1743         if (body->valid & OBD_MD_FLMDSCAPA) {
1744                 LASSERT(md->mds_capa);
1745                 ll_add_capa(inode, md->mds_capa);
1746         }
1747         if (body->valid & OBD_MD_FLOSSCAPA) {
1748                 LASSERT(md->oss_capa);
1749                 ll_add_capa(inode, md->oss_capa);
1750         }
1751 }
1752
1753 void ll_read_inode2(struct inode *inode, void *opaque)
1754 {
1755         struct lustre_md *md = opaque;
1756         struct ll_inode_info *lli = ll_i2info(inode);
1757         ENTRY;
1758
1759         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1760                PFID(&lli->lli_fid), inode);
1761
1762         LASSERT(!lli->lli_has_smd);
1763
1764         /* Core attributes from the MDS first.  This is a new inode, and
1765          * the VFS doesn't zero times in the core inode so we have to do
1766          * it ourselves.  They will be overwritten by either MDS or OST
1767          * attributes - we just need to make sure they aren't newer. */
1768         LTIME_S(inode->i_mtime) = 0;
1769         LTIME_S(inode->i_atime) = 0;
1770         LTIME_S(inode->i_ctime) = 0;
1771         inode->i_rdev = 0;
1772         ll_update_inode(inode, md);
1773
1774         /* OIDEBUG(inode); */
1775
1776         /* initializing backing dev info. */
1777         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1778
1779
1780         if (S_ISREG(inode->i_mode)) {
1781                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1782                 inode->i_op = &ll_file_inode_operations;
1783                 inode->i_fop = sbi->ll_fop;
1784                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1785                 EXIT;
1786         } else if (S_ISDIR(inode->i_mode)) {
1787                 inode->i_op = &ll_dir_inode_operations;
1788                 inode->i_fop = &ll_dir_operations;
1789                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1790                 EXIT;
1791         } else if (S_ISLNK(inode->i_mode)) {
1792                 inode->i_op = &ll_fast_symlink_inode_operations;
1793                 EXIT;
1794         } else {
1795                 inode->i_op = &ll_special_inode_operations;
1796
1797                 init_special_inode(inode, inode->i_mode,
1798                                    kdev_t_to_nr(inode->i_rdev));
1799
1800                 EXIT;
1801         }
1802 }
1803
1804 void ll_delete_inode(struct inode *inode)
1805 {
1806         struct cl_inode_info *lli = cl_i2info(inode);
1807         struct ll_sb_info    *sbi = ll_i2sbi(inode);
1808         int rc;
1809         ENTRY;
1810
1811         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1812         if (rc)
1813                 CERROR("fid_delete() failed, rc %d\n", rc);
1814
1815         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1816                 /* discard all dirty pages before truncating them, required by
1817                  * osc_extent implementation at LU-1030. */
1818                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1819
1820         truncate_inode_pages(&inode->i_data, 0);
1821
1822         /* Workaround for LU-118 */
1823         if (inode->i_data.nrpages) {
1824                 TREE_READ_LOCK_IRQ(&inode->i_data);
1825                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1826                 LASSERTF(inode->i_data.nrpages == 0,
1827                          "inode=%lu/%u(%p) nrpages=%lu, see "
1828                          "http://jira.whamcloud.com/browse/LU-118\n",
1829                          inode->i_ino, inode->i_generation, inode,
1830                          inode->i_data.nrpages);
1831         }
1832         /* Workaround end */
1833
1834 #ifdef HAVE_SBOPS_EVICT_INODE
1835         ll_clear_inode(inode);
1836         end_writeback(inode);
1837 #else
1838         clear_inode(inode);
1839 #endif
1840
1841         EXIT;
1842 }
1843
1844 int ll_iocontrol(struct inode *inode, struct file *file,
1845                  unsigned int cmd, unsigned long arg)
1846 {
1847         struct ll_sb_info *sbi = ll_i2sbi(inode);
1848         struct ptlrpc_request *req = NULL;
1849         int rc, flags = 0;
1850         ENTRY;
1851
1852         switch(cmd) {
1853         case FSFILT_IOC_GETFLAGS: {
1854                 struct mdt_body *body;
1855                 struct md_op_data *op_data;
1856
1857                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1858                                              0, 0, LUSTRE_OPC_ANY,
1859                                              NULL);
1860                 if (IS_ERR(op_data))
1861                         RETURN(PTR_ERR(op_data));
1862
1863                 op_data->op_valid = OBD_MD_FLFLAGS;
1864                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1865                 ll_finish_md_op_data(op_data);
1866                 if (rc) {
1867                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1868                         RETURN(-abs(rc));
1869                 }
1870
1871                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1872
1873                 flags = body->flags;
1874
1875                 ptlrpc_req_finished(req);
1876
1877                 RETURN(put_user(flags, (int *)arg));
1878         }
1879         case FSFILT_IOC_SETFLAGS: {
1880                 struct lov_stripe_md *lsm;
1881                 struct obd_info oinfo = { { { 0 } } };
1882                 struct md_op_data *op_data;
1883
1884                 if (get_user(flags, (int *)arg))
1885                         RETURN(-EFAULT);
1886
1887                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1888                                              LUSTRE_OPC_ANY, NULL);
1889                 if (IS_ERR(op_data))
1890                         RETURN(PTR_ERR(op_data));
1891
1892                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1893                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1894                 rc = md_setattr(sbi->ll_md_exp, op_data,
1895                                 NULL, 0, NULL, 0, &req, NULL);
1896                 ll_finish_md_op_data(op_data);
1897                 ptlrpc_req_finished(req);
1898                 if (rc)
1899                         RETURN(rc);
1900
1901                 OBDO_ALLOC(oinfo.oi_oa);
1902                 if (!oinfo.oi_oa)
1903                         RETURN(-ENOMEM);
1904
1905                 lsm = ccc_inode_lsm_get(inode);
1906                 if (lsm == NULL) {
1907                         inode->i_flags = ll_ext_to_inode_flags(flags);
1908                         OBDO_FREE(oinfo.oi_oa);
1909                         RETURN(0);
1910                 }
1911
1912                 oinfo.oi_md = lsm;
1913                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1914                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1915                 oinfo.oi_oa->o_flags = flags;
1916                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1917                                        OBD_MD_FLGROUP;
1918                 oinfo.oi_capa = ll_mdscapa_get(inode);
1919                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1920                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1921                 capa_put(oinfo.oi_capa);
1922                 OBDO_FREE(oinfo.oi_oa);
1923                 ccc_inode_lsm_put(inode, lsm);
1924
1925                 if (rc && rc != -EPERM && rc != -EACCES)
1926                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1927
1928                 RETURN(rc);
1929         }
1930         default:
1931                 RETURN(-ENOSYS);
1932         }
1933
1934         RETURN(0);
1935 }
1936
1937 int ll_flush_ctx(struct inode *inode)
1938 {
1939         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1940
1941         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1942
1943         obd_set_info_async(NULL, sbi->ll_md_exp,
1944                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1945                            0, NULL, NULL);
1946         obd_set_info_async(NULL, sbi->ll_dt_exp,
1947                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1948                            0, NULL, NULL);
1949         return 0;
1950 }
1951
1952 /* umount -f client means force down, don't save state */
1953 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1954 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1955 {
1956         struct super_block *sb = vfsmnt->mnt_sb;
1957 #else
1958 void ll_umount_begin(struct super_block *sb)
1959 {
1960 #endif
1961         struct lustre_sb_info *lsi = s2lsi(sb);
1962         struct ll_sb_info *sbi = ll_s2sbi(sb);
1963         struct obd_device *obd;
1964         struct obd_ioctl_data *ioc_data;
1965         ENTRY;
1966
1967 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1968         if (!(flags & MNT_FORCE)) {
1969                 EXIT;
1970                 return;
1971         }
1972 #endif
1973
1974         /* Tell the MGC we got umount -f */
1975         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1976
1977         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1978                sb->s_count, atomic_read(&sb->s_active));
1979
1980         obd = class_exp2obd(sbi->ll_md_exp);
1981         if (obd == NULL) {
1982                 CERROR("Invalid MDC connection handle "LPX64"\n",
1983                        sbi->ll_md_exp->exp_handle.h_cookie);
1984                 EXIT;
1985                 return;
1986         }
1987         obd->obd_force = 1;
1988
1989         obd = class_exp2obd(sbi->ll_dt_exp);
1990         if (obd == NULL) {
1991                 CERROR("Invalid LOV connection handle "LPX64"\n",
1992                        sbi->ll_dt_exp->exp_handle.h_cookie);
1993                 EXIT;
1994                 return;
1995         }
1996         obd->obd_force = 1;
1997
1998         OBD_ALLOC_PTR(ioc_data);
1999         if (ioc_data) {
2000                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2001                               sizeof ioc_data, ioc_data, NULL);
2002
2003                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2004                               sizeof ioc_data, ioc_data, NULL);
2005
2006                 OBD_FREE_PTR(ioc_data);
2007         }
2008
2009
2010         /* Really, we'd like to wait until there are no requests outstanding,
2011          * and then continue.  For now, we just invalidate the requests,
2012          * schedule() and sleep one second if needed, and hope.
2013          */
2014         cfs_schedule();
2015 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2016         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2017                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2018                                                    cfs_time_seconds(1));
2019                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2020                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2021                                       "may try to umount it a bit later\n",
2022                                       atomic_read(&vfsmnt->mnt_count));
2023         }
2024 #endif
2025
2026         EXIT;
2027 }
2028
2029 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2030 {
2031         struct ll_sb_info *sbi = ll_s2sbi(sb);
2032         char *profilenm = get_profile_name(sb);
2033         int err;
2034         __u32 read_only;
2035
2036         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2037                 read_only = *flags & MS_RDONLY;
2038                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2039                                          sizeof(KEY_READ_ONLY),
2040                                          KEY_READ_ONLY, sizeof(read_only),
2041                                          &read_only, NULL);
2042                 if (err) {
2043                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2044                                       profilenm, read_only ?
2045                                       "read-only" : "read-write", err);
2046                         return err;
2047                 }
2048
2049                 if (read_only)
2050                         sb->s_flags |= MS_RDONLY;
2051                 else
2052                         sb->s_flags &= ~MS_RDONLY;
2053
2054                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2055                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2056                                       read_only ?  "read-only" : "read-write");
2057         }
2058         return 0;
2059 }
2060
2061 int ll_prep_inode(struct inode **inode,
2062                   struct ptlrpc_request *req,
2063                   struct super_block *sb)
2064 {
2065         struct ll_sb_info *sbi = NULL;
2066         struct lustre_md md;
2067         __u64 ibits;
2068         int rc;
2069         ENTRY;
2070
2071         LASSERT(*inode || sb);
2072         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2073         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2074                               sbi->ll_md_exp, &md);
2075         if (rc)
2076                 RETURN(rc);
2077
2078         if (*inode) {
2079                 ll_update_inode(*inode, &md);
2080         } else {
2081                 LASSERT(sb != NULL);
2082
2083                 /*
2084                  * At this point server returns to client's same fid as client
2085                  * generated for creating. So using ->fid1 is okay here.
2086                  */
2087                 LASSERT(fid_is_sane(&md.body->fid1));
2088
2089                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2090                 if (*inode == NULL || IS_ERR(*inode)) {
2091                         if (md.lsm)
2092                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2093 #ifdef CONFIG_FS_POSIX_ACL
2094                         if (md.posix_acl) {
2095                                 posix_acl_release(md.posix_acl);
2096                                 md.posix_acl = NULL;
2097                         }
2098 #endif
2099                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2100                         *inode = NULL;
2101                         CERROR("new_inode -fatal: rc %d\n", rc);
2102                         GOTO(out, rc);
2103                 }
2104         }
2105
2106         /* sanity check for LAYOUT lock. */
2107         ibits = MDS_INODELOCK_LAYOUT;
2108         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2109             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2110                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2111                         ll_get_fsname(*inode), PFID(ll_inode2fid(*inode)),
2112                         *inode);
2113         }
2114
2115 out:
2116         md_free_lustre_md(sbi->ll_md_exp, &md);
2117         RETURN(rc);
2118 }
2119
2120 int ll_obd_statfs(struct inode *inode, void *arg)
2121 {
2122         struct ll_sb_info *sbi = NULL;
2123         struct obd_export *exp;
2124         char *buf = NULL;
2125         struct obd_ioctl_data *data = NULL;
2126         __u32 type;
2127         int len = 0, rc;
2128
2129         if (!inode || !(sbi = ll_i2sbi(inode)))
2130                 GOTO(out_statfs, rc = -EINVAL);
2131
2132         rc = obd_ioctl_getdata(&buf, &len, arg);
2133         if (rc)
2134                 GOTO(out_statfs, rc);
2135
2136         data = (void*)buf;
2137         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2138             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2139                 GOTO(out_statfs, rc = -EINVAL);
2140
2141         if (data->ioc_inllen1 != sizeof(__u32) ||
2142             data->ioc_inllen2 != sizeof(__u32) ||
2143             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2144             data->ioc_plen2 != sizeof(struct obd_uuid))
2145                 GOTO(out_statfs, rc = -EINVAL);
2146
2147         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2148         if (type == LL_STATFS_LMV)
2149                 exp = sbi->ll_md_exp;
2150         else if (type == LL_STATFS_LOV)
2151                 exp = sbi->ll_dt_exp;
2152         else
2153                 GOTO(out_statfs, rc = -ENODEV);
2154
2155         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2156         if (rc)
2157                 GOTO(out_statfs, rc);
2158 out_statfs:
2159         if (buf)
2160                 obd_ioctl_freedata(buf, len);
2161         return rc;
2162 }
2163
2164 int ll_process_config(struct lustre_cfg *lcfg)
2165 {
2166         char *ptr;
2167         void *sb;
2168         struct lprocfs_static_vars lvars;
2169         unsigned long x;
2170         int rc = 0;
2171
2172         lprocfs_llite_init_vars(&lvars);
2173
2174         /* The instance name contains the sb: lustre-client-aacfe000 */
2175         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2176         if (!ptr || !*(++ptr))
2177                 return -EINVAL;
2178         if (sscanf(ptr, "%lx", &x) != 1)
2179                 return -EINVAL;
2180         sb = (void *)x;
2181         /* This better be a real Lustre superblock! */
2182         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2183
2184         /* Note we have not called client_common_fill_super yet, so
2185            proc fns must be able to handle that! */
2186         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2187                                       lcfg, sb);
2188         if (rc > 0)
2189                 rc = 0;
2190         return(rc);
2191 }
2192
2193 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2194 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2195                                        struct inode *i1, struct inode *i2,
2196                                        const char *name, int namelen,
2197                                        int mode, __u32 opc, void *data)
2198 {
2199         LASSERT(i1 != NULL);
2200
2201         if (namelen > ll_i2sbi(i1)->ll_namelen)
2202                 return ERR_PTR(-ENAMETOOLONG);
2203
2204         if (op_data == NULL)
2205                 OBD_ALLOC_PTR(op_data);
2206
2207         if (op_data == NULL)
2208                 return ERR_PTR(-ENOMEM);
2209
2210         ll_i2gids(op_data->op_suppgids, i1, i2);
2211         op_data->op_fid1 = *ll_inode2fid(i1);
2212         op_data->op_capa1 = ll_mdscapa_get(i1);
2213
2214         if (i2) {
2215                 op_data->op_fid2 = *ll_inode2fid(i2);
2216                 op_data->op_capa2 = ll_mdscapa_get(i2);
2217         } else {
2218                 fid_zero(&op_data->op_fid2);
2219                 op_data->op_capa2 = NULL;
2220         }
2221
2222         op_data->op_name = name;
2223         op_data->op_namelen = namelen;
2224         op_data->op_mode = mode;
2225         op_data->op_mod_time = cfs_time_current_sec();
2226         op_data->op_fsuid = cfs_curproc_fsuid();
2227         op_data->op_fsgid = cfs_curproc_fsgid();
2228         op_data->op_cap = cfs_curproc_cap_pack();
2229         op_data->op_bias = MDS_CHECK_SPLIT;
2230         op_data->op_opc = opc;
2231         op_data->op_mds = 0;
2232         op_data->op_data = data;
2233
2234         /* If the file is being opened after mknod() (normally due to NFS)
2235          * try to use the default stripe data from parent directory for
2236          * allocating OST objects.  Try to pass the parent FID to MDS. */
2237         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2238             !ll_i2info(i2)->lli_has_smd) {
2239                 struct ll_inode_info *lli = ll_i2info(i2);
2240
2241                 cfs_spin_lock(&lli->lli_lock);
2242                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2243                         op_data->op_fid1 = lli->lli_pfid;
2244                 cfs_spin_unlock(&lli->lli_lock);
2245                 /** We ignore parent's capability temporary. */
2246         }
2247
2248         return op_data;
2249 }
2250
2251 void ll_finish_md_op_data(struct md_op_data *op_data)
2252 {
2253         capa_put(op_data->op_capa1);
2254         capa_put(op_data->op_capa2);
2255         OBD_FREE_PTR(op_data);
2256 }
2257
2258 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2259 {
2260         struct ll_sb_info *sbi;
2261
2262         LASSERT((seq != NULL) && (vfs != NULL));
2263         sbi = ll_s2sbi(vfs->mnt_sb);
2264
2265         if (sbi->ll_flags & LL_SBI_NOLCK)
2266                 seq_puts(seq, ",nolock");
2267
2268         if (sbi->ll_flags & LL_SBI_FLOCK)
2269                 seq_puts(seq, ",flock");
2270
2271         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2272                 seq_puts(seq, ",localflock");
2273
2274         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2275                 seq_puts(seq, ",user_xattr");
2276
2277         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2278                 seq_puts(seq, ",lazystatfs");
2279
2280         RETURN(0);
2281 }
2282
2283 /**
2284  * Get obd name by cmd, and copy out to user space
2285  */
2286 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2287 {
2288         struct ll_sb_info *sbi = ll_i2sbi(inode);
2289         struct obd_device *obd;
2290         ENTRY;
2291
2292         if (cmd == OBD_IOC_GETDTNAME)
2293                 obd = class_exp2obd(sbi->ll_dt_exp);
2294         else if (cmd == OBD_IOC_GETMDNAME)
2295                 obd = class_exp2obd(sbi->ll_md_exp);
2296         else
2297                 RETURN(-EINVAL);
2298
2299         if (!obd)
2300                 RETURN(-ENOENT);
2301
2302         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2303                              strlen(obd->obd_name) + 1))
2304                 RETURN(-EFAULT);
2305
2306         RETURN(0);
2307 }