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, 2014, Intel Corporation.
  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/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #ifdef HAVE_UIDGID_HEADER
  50 # include <linux/uidgid.h>
  51 #endif
  52
  53 #include <lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  67 static DEFINE_SPINLOCK(ll_sb_lock);
  68
  69 #ifndef log2
  70 #define log2(n) ffz(~(n))
  71 #endif
  72
  73 static struct ll_sb_info *ll_init_sbi(void)
  74 {
  75         struct ll_sb_info *sbi = NULL;
  76         unsigned long pages;
  77         unsigned long lru_page_max;
  78         struct sysinfo si;
  79         class_uuid_t uuid;
  80         int i;
  81         ENTRY;
  82
  83         OBD_ALLOC_PTR(sbi);
  84         if (sbi == NULL)
  85                 RETURN(NULL);
  86
  87         spin_lock_init(&sbi->ll_lock);
  88         mutex_init(&sbi->ll_lco.lco_lock);
  89         spin_lock_init(&sbi->ll_pp_extent_lock);
  90         spin_lock_init(&sbi->ll_process_lock);
  91         sbi->ll_rw_stats_on = 0;
  92
  93         si_meminfo(&si);
  94         pages = si.totalram - si.totalhigh;
  95         lru_page_max = pages / 2;
  96
  97         /* initialize ll_cache data */
  98         sbi->ll_cache = cl_cache_init(lru_page_max);
  99         if (sbi->ll_cache == NULL) {
 100                 OBD_FREE(sbi, sizeof(*sbi));
 101                 RETURN(NULL);
 102         }
 103
 104         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 105                                            SBI_DEFAULT_READAHEAD_MAX);
 106         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 107         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 108                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 109         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 110         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 111
 112         ll_generate_random_uuid(uuid);
 113         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 114         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 115
 116         spin_lock(&ll_sb_lock);
 117         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 118         spin_unlock(&ll_sb_lock);
 119
 120         sbi->ll_flags |= LL_SBI_VERBOSE;
 121 #ifdef ENABLE_CHECKSUM
 122         sbi->ll_flags |= LL_SBI_CHECKSUM;
 123 #endif
 124
 125 #ifdef HAVE_LRU_RESIZE_SUPPORT
 126         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 127 #endif
 128
 129         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 130                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 131                                pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 133                                pp_w_hist.oh_lock);
 134         }
 135
 136         /* metadata statahead is enabled by default */
 137         sbi->ll_sa_max = LL_SA_RPC_DEF;
 138         atomic_set(&sbi->ll_sa_total, 0);
 139         atomic_set(&sbi->ll_sa_wrong, 0);
 140         atomic_set(&sbi->ll_sa_running, 0);
 141         atomic_set(&sbi->ll_agl_total, 0);
 142         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 spin_lock(&ll_sb_lock);
 160                 list_del(&sbi->ll_list);
 161                 spin_unlock(&ll_sb_lock);
 162                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 163                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 164                 if (sbi->ll_cache != NULL) {
 165                         cl_cache_decref(sbi->ll_cache);
 166                         sbi->ll_cache = NULL;
 167                 }
 168                 OBD_FREE(sbi, sizeof(*sbi));
 169         }
 170         EXIT;
 171 }
 172
 173 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 174                                     struct vfsmount *mnt)
 175 {
 176         struct inode *root = NULL;
 177         struct ll_sb_info *sbi = ll_s2sbi(sb);
 178         struct obd_device *obd;
 179         struct obd_capa *oc = NULL;
 180         struct obd_statfs *osfs = NULL;
 181         struct ptlrpc_request *request = NULL;
 182         struct obd_connect_data *data = NULL;
 183         struct obd_uuid *uuid;
 184         struct md_op_data *op_data;
 185         struct lustre_md lmd;
 186         u64 valid;
 187         int size, err, checksum;
 188         ENTRY;
 189
 190         obd = class_name2obd(md);
 191         if (!obd) {
 192                 CERROR("MD %s: not setup or attached\n", md);
 193                 RETURN(-EINVAL);
 194         }
 195
 196         OBD_ALLOC_PTR(data);
 197         if (data == NULL)
 198                 RETURN(-ENOMEM);
 199
 200         OBD_ALLOC_PTR(osfs);
 201         if (osfs == NULL) {
 202                 OBD_FREE_PTR(data);
 203                 RETURN(-ENOMEM);
 204         }
 205
 206         /* indicate the features supported by this client */
 207         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 208                                   OBD_CONNECT_ATTRFID  |
 209                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 210                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 211                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 212                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 213                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 214                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 215                                   OBD_CONNECT_EINPROGRESS |
 216                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 217                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 218                                   OBD_CONNECT_MAX_EASIZE |
 219                                   OBD_CONNECT_FLOCK_DEAD |
 220                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 221                                   OBD_CONNECT_OPEN_BY_FID |
 222                                   OBD_CONNECT_DIR_STRIPE;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 230 #endif
 231
 232         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 233                 /* flag mdc connection as lightweight, only used for test
 234                  * purpose, use with care */
 235                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 236
 237         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 238         data->ocd_version = LUSTRE_VERSION_CODE;
 239
 240         if (sb->s_flags & MS_RDONLY)
 241                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 242         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 243                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 244
 245 #ifdef HAVE_MS_FLOCK_LOCK
 246         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 247         sb->s_flags |= MS_FLOCK_LOCK;
 248 #endif
 249 #ifdef MS_HAS_NEW_AOPS
 250         sb->s_flags |= MS_HAS_NEW_AOPS;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* real client */
 261         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 262         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 263                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337
 338         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 339             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 340                 LCONSOLE_INFO("Disabling user_xattr feature because "
 341                               "it is not supported on the server\n");
 342                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 346 #ifdef MS_POSIXACL
 347                 sb->s_flags |= MS_POSIXACL;
 348 #endif
 349                 sbi->ll_flags |= LL_SBI_ACL;
 350         } else {
 351                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 352 #ifdef MS_POSIXACL
 353                 sb->s_flags &= ~MS_POSIXACL;
 354 #endif
 355                 sbi->ll_flags &= ~LL_SBI_ACL;
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 359                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 360                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client is set as remote by default.\n");
 362                 }
 363         } else {
 364                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 365                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 366                         LCONSOLE_INFO("client claims to be remote, but server "
 367                                       "rejected, forced to be local.\n");
 368                 }
 369         }
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 372                 LCONSOLE_INFO("client enabled MDS capability!\n");
 373                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 377                 LCONSOLE_INFO("client enabled OSS capability!\n");
 378                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 379         }
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 382                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 385                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 386         else
 387                 sbi->ll_md_brw_pages = 1;
 388
 389         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391
 392         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 393                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 394                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 395                                       "unknown maximum xattr size.\n", dt);
 396                 } else {
 397                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 398                         sbi->ll_xattr_cache_enabled = 1;
 399                 }
 400         }
 401
 402         obd = class_name2obd(dt);
 403         if (!obd) {
 404                 CERROR("DT %s: not setup or attached\n", dt);
 405                 GOTO(out_md_fid, err = -ENODEV);
 406         }
 407
 408         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 409                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 410                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 411                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 412                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 413                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 414                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 415                                   OBD_CONNECT_MAXBYTES |
 416                                   OBD_CONNECT_EINPROGRESS |
 417                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 418                                   OBD_CONNECT_LAYOUTLOCK |
 419                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 420
 421         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 422                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 423                  * disabled by default, because it can still be enabled on the
 424                  * fly via /proc. As a consequence, we still need to come to an
 425                  * agreement on the supported algorithms at connect time */
 426                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 427
 428                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 429                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 430                 else
 431                         data->ocd_cksum_types = cksum_types_supported_client();
 432         }
 433
 434 #ifdef HAVE_LRU_RESIZE_SUPPORT
 435         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 436 #endif
 437         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 438                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 439
 440         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 441                "ocd_grant: %d\n", data->ocd_connect_flags,
 442                data->ocd_version, data->ocd_grant);
 443
 444         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 445         obd->obd_upcall.onu_upcall = cl_ocd_update;
 446
 447         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 448
 449         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 450                           NULL);
 451         if (err == -EBUSY) {
 452                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 453                                    "recovery, of which this client is not a "
 454                                    "part.  Please wait for recovery to "
 455                                    "complete, abort, or time out.\n", dt);
 456                 GOTO(out_md, err);
 457         } else if (err) {
 458                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 459                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 460                 GOTO(out_md, err);
 461         }
 462
 463         sbi->ll_dt_exp->exp_connect_data = *data;
 464
 465         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 466                            LUSTRE_SEQ_METADATA);
 467         if (err) {
 468                 CERROR("%s: Can't init data layer FID infrastructure, "
 469                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 470                 GOTO(out_dt, err);
 471         }
 472
 473         mutex_lock(&sbi->ll_lco.lco_lock);
 474         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 475         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 476         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 477         mutex_unlock(&sbi->ll_lco.lco_lock);
 478
 479         fid_zero(&sbi->ll_root_fid);
 480         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 481         if (err) {
 482                 CERROR("cannot mds_connect: rc = %d\n", err);
 483                 GOTO(out_lock_cn_cb, err);
 484         }
 485         if (!fid_is_sane(&sbi->ll_root_fid)) {
 486                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 487                        sbi->ll_md_exp->exp_obd->obd_name,
 488                        PFID(&sbi->ll_root_fid));
 489                 GOTO(out_lock_cn_cb, err = -EINVAL);
 490         }
 491         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 492
 493         sb->s_op = &lustre_super_operations;
 494 #if THREAD_SIZE >= 8192 /*b=17630*/
 495         sb->s_export_op = &lustre_export_operations;
 496 #endif
 497
 498         /* make root inode
 499          * XXX: move this to after cbd setup? */
 500         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 501                 OBD_MD_FLMODEASIZE;
 502         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 503                 valid |= OBD_MD_FLRMTPERM;
 504         else if (sbi->ll_flags & LL_SBI_ACL)
 505                 valid |= OBD_MD_FLACL;
 506
 507         OBD_ALLOC_PTR(op_data);
 508         if (op_data == NULL)
 509                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 510
 511         op_data->op_fid1 = sbi->ll_root_fid;
 512         op_data->op_mode = 0;
 513         op_data->op_capa1 = oc;
 514         op_data->op_valid = valid;
 515
 516         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 517         if (oc)
 518                 capa_put(oc);
 519         OBD_FREE_PTR(op_data);
 520         if (err) {
 521                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 522                        sbi->ll_md_exp->exp_obd->obd_name, err);
 523                 GOTO(out_lock_cn_cb, err);
 524         }
 525
 526         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 527                                sbi->ll_md_exp, &lmd);
 528         if (err) {
 529                 CERROR("failed to understand root inode md: rc = %d\n", err);
 530                 ptlrpc_req_finished(request);
 531                 GOTO(out_lock_cn_cb, err);
 532         }
 533
 534         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 535         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 536                                             sbi->ll_flags & LL_SBI_32BIT_API),
 537                        &lmd);
 538         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 539         ptlrpc_req_finished(request);
 540
 541         if (IS_ERR(root)) {
 542                 if (lmd.lsm)
 543                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 544 #ifdef CONFIG_FS_POSIX_ACL
 545                 if (lmd.posix_acl) {
 546                         posix_acl_release(lmd.posix_acl);
 547                         lmd.posix_acl = NULL;
 548                 }
 549 #endif
 550                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 551                 root = NULL;
 552                 CERROR("lustre_lite: bad iget4 for root\n");
 553                 GOTO(out_root, err);
 554         }
 555
 556 #ifdef CONFIG_FS_POSIX_ACL
 557         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 558                 rct_init(&sbi->ll_rct);
 559                 et_init(&sbi->ll_et);
 560         }
 561 #endif
 562
 563         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 564         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 565                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 566                                  NULL);
 567         cl_sb_init(sb);
 568
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 570                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 571                                  sbi->ll_cache, NULL);
 572
 573         sb->s_root = d_make_root(root);
 574         if (sb->s_root == NULL) {
 575                 CERROR("%s: can't make root dentry\n",
 576                         ll_get_fsname(sb, NULL, 0));
 577                 GOTO(out_root, err = -ENOMEM);
 578         }
 579 #ifdef HAVE_DCACHE_LOCK
 580         sb->s_root->d_op = &ll_d_ops;
 581 #endif
 582
 583         sbi->ll_sdev_orig = sb->s_dev;
 584
 585         /* We set sb->s_dev equal on all lustre clients in order to support
 586          * NFS export clustering.  NFSD requires that the FSID be the same
 587          * on all clients. */
 588         /* s_dev is also used in lt_compare() to compare two fs, but that is
 589          * only a node-local comparison. */
 590         uuid = obd_get_uuid(sbi->ll_md_exp);
 591         if (uuid != NULL)
 592                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 593
 594         if (data != NULL)
 595                 OBD_FREE_PTR(data);
 596         if (osfs != NULL)
 597                 OBD_FREE_PTR(osfs);
 598         if (proc_lustre_fs_root != NULL) {
 599                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 600                                                   dt, md);
 601                 if (err < 0) {
 602                         CERROR("%s: could not register mount in lprocfs: "
 603                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 604                         err = 0;
 605                 }
 606         }
 607
 608         RETURN(err);
 609 out_root:
 610         if (root)
 611                 iput(root);
 612 out_lock_cn_cb:
 613         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 614 out_dt:
 615         obd_disconnect(sbi->ll_dt_exp);
 616         sbi->ll_dt_exp = NULL;
 617 out_md_fid:
 618         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 619 out_md:
 620         obd_disconnect(sbi->ll_md_exp);
 621         sbi->ll_md_exp = NULL;
 622 out:
 623         if (data != NULL)
 624                 OBD_FREE_PTR(data);
 625         if (osfs != NULL)
 626                 OBD_FREE_PTR(osfs);
 627         return err;
 628 }
 629
 630 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 631 {
 632         int size, rc;
 633
 634         size = sizeof(*lmmsize);
 635         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 636                           KEY_MAX_EASIZE, &size, lmmsize);
 637         if (rc != 0) {
 638                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 639                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 640                 RETURN(rc);
 641         }
 642
 643         size = sizeof(int);
 644         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 645                           KEY_MAX_EASIZE, &size, lmmsize);
 646         if (rc)
 647                 CERROR("Get max mdsize error rc %d\n", rc);
 648
 649         RETURN(rc);
 650 }
 651
 652 /**
 653  * Get the value of the default_easize parameter.
 654  *
 655  * \see client_obd::cl_default_mds_easize
 656  *
 657  * \param[in] sbi       superblock info for this filesystem
 658  * \param[out] lmmsize  pointer to storage location for value
 659  *
 660  * \retval 0            on success
 661  * \retval negative     negated errno on failure
 662  */
 663 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 664 {
 665         int size, rc;
 666
 667         size = sizeof(int);
 668         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 669                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 670         if (rc)
 671                 CERROR("Get default mdsize error rc %d\n", rc);
 672
 673         RETURN(rc);
 674 }
 675
 676 /**
 677  * Set the default_easize parameter to the given value.
 678  *
 679  * \see client_obd::cl_default_mds_easize
 680  *
 681  * \param[in] sbi       superblock info for this filesystem
 682  * \param[in] lmmsize   the size to set
 683  *
 684  * \retval 0            on success
 685  * \retval negative     negated errno on failure
 686  */
 687 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 688 {
 689         int rc;
 690
 691         if (lmmsize < sizeof(struct lov_mds_md) ||
 692             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 693                 return -EINVAL;
 694
 695         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 696                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 697                                 sizeof(int), &lmmsize, NULL);
 698
 699         RETURN(rc);
 700 }
 701
 702 static void ll_dump_inode(struct inode *inode)
 703 {
 704         struct ll_d_hlist_node *tmp;
 705         int dentry_count = 0;
 706
 707         LASSERT(inode != NULL);
 708
 709         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 710                 dentry_count++;
 711
 712         CERROR("%s: inode %p dump: dev=%s fid="DFID
 713                " mode=%o count=%u, %d dentries\n",
 714                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 715                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 716                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 717 }
 718
 719 void lustre_dump_dentry(struct dentry *dentry, int recur)
 720 {
 721         struct list_head *tmp;
 722         int subdirs = 0;
 723
 724         LASSERT(dentry != NULL);
 725
 726         list_for_each(tmp, &dentry->d_subdirs)
 727                 subdirs++;
 728
 729         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 730                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 731                dentry->d_name.len, dentry->d_name.name,
 732                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 733                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 734                dentry->d_flags, dentry->d_fsdata, subdirs);
 735         if (dentry->d_inode != NULL)
 736                 ll_dump_inode(dentry->d_inode);
 737
 738         if (recur == 0)
 739                 return;
 740
 741         list_for_each(tmp, &dentry->d_subdirs) {
 742                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 743                 lustre_dump_dentry(d, recur - 1);
 744         }
 745 }
 746
 747 static void client_common_put_super(struct super_block *sb)
 748 {
 749         struct ll_sb_info *sbi = ll_s2sbi(sb);
 750         ENTRY;
 751
 752 #ifdef CONFIG_FS_POSIX_ACL
 753         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 754                 et_fini(&sbi->ll_et);
 755                 rct_fini(&sbi->ll_rct);
 756         }
 757 #endif
 758
 759         cl_sb_fini(sb);
 760
 761         list_del(&sbi->ll_conn_chain);
 762
 763         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 764         obd_disconnect(sbi->ll_dt_exp);
 765         sbi->ll_dt_exp = NULL;
 766
 767         lprocfs_unregister_mountpoint(sbi);
 768
 769         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 770         obd_disconnect(sbi->ll_md_exp);
 771         sbi->ll_md_exp = NULL;
 772
 773         EXIT;
 774 }
 775
 776 void ll_kill_super(struct super_block *sb)
 777 {
 778         struct ll_sb_info *sbi;
 779         ENTRY;
 780
 781         /* not init sb ?*/
 782         if (!(sb->s_flags & MS_ACTIVE))
 783                 return;
 784
 785         sbi = ll_s2sbi(sb);
 786         /* we need restore s_dev from changed for clustred NFS before put_super
 787          * because new kernels have cached s_dev and change sb->s_dev in
 788          * put_super not affected real removing devices */
 789         if (sbi) {
 790                 sb->s_dev = sbi->ll_sdev_orig;
 791                 sbi->ll_umounting = 1;
 792
 793                 /* wait running statahead threads to quit */
 794                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 795                         set_current_state(TASK_UNINTERRUPTIBLE);
 796                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 797                 }
 798         }
 799
 800         EXIT;
 801 }
 802
 803 static inline int ll_set_opt(const char *opt, char *data, int fl)
 804 {
 805         if (strncmp(opt, data, strlen(opt)) != 0)
 806                 return(0);
 807         else
 808                 return(fl);
 809 }
 810
 811 /* non-client-specific mount options are parsed in lmd_parse */
 812 static int ll_options(char *options, int *flags)
 813 {
 814         int tmp;
 815         char *s1 = options, *s2;
 816         ENTRY;
 817
 818         if (!options)
 819                 RETURN(0);
 820
 821         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 822
 823         while (*s1) {
 824                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 825                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 826                 if (tmp) {
 827                         *flags |= tmp;
 828                         goto next;
 829                 }
 830                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 841                 if (tmp) {
 842                         *flags &= ~tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 846                 if (tmp) {
 847                         *flags |= tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 851                 if (tmp) {
 852                         *flags &= ~tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags |= tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 866                 if (tmp) {
 867                         *flags &= ~tmp;
 868                         goto next;
 869                 }
 870
 871                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 872                 if (tmp) {
 873                         *flags |= tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 877                 if (tmp) {
 878                         *flags &= ~tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 882                 if (tmp) {
 883                         *flags |= tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 887                 if (tmp) {
 888                         *flags &= ~tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 897                 if (tmp) {
 898                         *flags &= ~tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 907                 if (tmp) {
 908                         *flags |= tmp;
 909                         goto next;
 910                 }
 911                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 912                 if (tmp) {
 913                         *flags &= ~tmp;
 914                         goto next;
 915                 }
 916                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 917                                    s1);
 918                 RETURN(-EINVAL);
 919
 920 next:
 921                 /* Find next opt */
 922                 s2 = strchr(s1, ',');
 923                 if (s2 == NULL)
 924                         break;
 925                 s1 = s2 + 1;
 926         }
 927         RETURN(0);
 928 }
 929
 930 void ll_lli_init(struct ll_inode_info *lli)
 931 {
 932         lli->lli_inode_magic = LLI_INODE_MAGIC;
 933         lli->lli_flags = 0;
 934         spin_lock_init(&lli->lli_lock);
 935         lli->lli_posix_acl = NULL;
 936         lli->lli_remote_perms = NULL;
 937         mutex_init(&lli->lli_rmtperm_mutex);
 938         /* Do not set lli_fid, it has been initialized already. */
 939         fid_zero(&lli->lli_pfid);
 940         INIT_LIST_HEAD(&lli->lli_oss_capas);
 941         atomic_set(&lli->lli_open_count, 0);
 942         lli->lli_mds_capa = NULL;
 943         lli->lli_rmtperm_time = 0;
 944         lli->lli_mds_read_och = NULL;
 945         lli->lli_mds_write_och = NULL;
 946         lli->lli_mds_exec_och = NULL;
 947         lli->lli_open_fd_read_count = 0;
 948         lli->lli_open_fd_write_count = 0;
 949         lli->lli_open_fd_exec_count = 0;
 950         mutex_init(&lli->lli_och_mutex);
 951         spin_lock_init(&lli->lli_agl_lock);
 952         spin_lock_init(&lli->lli_layout_lock);
 953         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 954         lli->lli_clob = NULL;
 955
 956         init_rwsem(&lli->lli_xattrs_list_rwsem);
 957         mutex_init(&lli->lli_xattrs_enq_lock);
 958
 959         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 960         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 961                 mutex_init(&lli->lli_readdir_mutex);
 962                 lli->lli_opendir_key = NULL;
 963                 lli->lli_sai = NULL;
 964                 spin_lock_init(&lli->lli_sa_lock);
 965                 lli->lli_opendir_pid = 0;
 966                 lli->lli_sa_enabled = 0;
 967                 lli->lli_def_stripe_offset = -1;
 968         } else {
 969                 mutex_init(&lli->lli_size_mutex);
 970                 lli->lli_symlink_name = NULL;
 971                 init_rwsem(&lli->lli_trunc_sem);
 972                 range_lock_tree_init(&lli->lli_write_tree);
 973                 init_rwsem(&lli->lli_glimpse_sem);
 974                 lli->lli_glimpse_time = 0;
 975                 INIT_LIST_HEAD(&lli->lli_agl_list);
 976                 lli->lli_agl_index = 0;
 977                 lli->lli_async_rc = 0;
 978         }
 979         mutex_init(&lli->lli_layout_mutex);
 980 }
 981
 982 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 983 {
 984         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 985
 986         bdi->name = "lustre";
 987         return bdi_register(bdi, NULL, "lustre-%d",
 988                             atomic_inc_return(&ll_bdi_num));
 989 }
 990
 991 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 992 {
 993         struct lustre_profile *lprof = NULL;
 994         struct lustre_sb_info *lsi = s2lsi(sb);
 995         struct ll_sb_info *sbi;
 996         char  *dt = NULL, *md = NULL;
 997         char  *profilenm = get_profile_name(sb);
 998         struct config_llog_instance *cfg;
 999         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1000         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1001         int    err;
1002         ENTRY;
1003
1004         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1005
1006         OBD_ALLOC_PTR(cfg);
1007         if (cfg == NULL)
1008                 RETURN(-ENOMEM);
1009
1010         try_module_get(THIS_MODULE);
1011
1012         /* client additional sb info */
1013         lsi->lsi_llsbi = sbi = ll_init_sbi();
1014         if (!sbi) {
1015                 module_put(THIS_MODULE);
1016                 OBD_FREE_PTR(cfg);
1017                 RETURN(-ENOMEM);
1018         }
1019
1020         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1021         if (err)
1022                 GOTO(out_free, err);
1023
1024         err = bdi_init(&lsi->lsi_bdi);
1025         if (err)
1026                 GOTO(out_free, err);
1027         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1028         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1029         err = ll_bdi_register(&lsi->lsi_bdi);
1030         if (err)
1031                 GOTO(out_free, err);
1032
1033         sb->s_bdi = &lsi->lsi_bdi;
1034 #ifndef HAVE_DCACHE_LOCK
1035         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1036         sb->s_d_op = &ll_d_ops;
1037 #endif
1038
1039         /* Generate a string unique to this super, in case some joker tries
1040            to mount the same fs at two mount points.
1041            Use the address of the super itself.*/
1042         cfg->cfg_instance = sb;
1043         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1044         cfg->cfg_callback = class_config_llog_handler;
1045         /* set up client obds */
1046         err = lustre_process_log(sb, profilenm, cfg);
1047         if (err < 0)
1048                 GOTO(out_free, err);
1049
1050         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1051         lprof = class_get_profile(profilenm);
1052         if (lprof == NULL) {
1053                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1054                                    " read from the MGS.  Does that filesystem "
1055                                    "exist?\n", profilenm);
1056                 GOTO(out_free, err = -EINVAL);
1057         }
1058         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1059                lprof->lp_md, lprof->lp_dt);
1060
1061         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1062         if (!dt)
1063                 GOTO(out_free, err = -ENOMEM);
1064         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1065
1066         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1067         if (!md)
1068                 GOTO(out_free, err = -ENOMEM);
1069         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1070
1071         /* connections, registrations, sb setup */
1072         err = client_common_fill_super(sb, md, dt, mnt);
1073         if (err < 0)
1074                 GOTO(out_free, err);
1075
1076         sbi->ll_client_common_fill_super_succeeded = 1;
1077
1078 out_free:
1079         if (md)
1080                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1081         if (dt)
1082                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1083         if (err)
1084                 ll_put_super(sb);
1085         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1086                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1087
1088         OBD_FREE_PTR(cfg);
1089         RETURN(err);
1090 } /* ll_fill_super */
1091
1092 void ll_put_super(struct super_block *sb)
1093 {
1094         struct config_llog_instance cfg, params_cfg;
1095         struct obd_device *obd;
1096         struct lustre_sb_info *lsi = s2lsi(sb);
1097         struct ll_sb_info *sbi = ll_s2sbi(sb);
1098         char *profilenm = get_profile_name(sb);
1099         long ccc_count;
1100         int next, force = 1, rc = 0;
1101         ENTRY;
1102
1103         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1104
1105         ll_print_capa_stat(sbi);
1106
1107         cfg.cfg_instance = sb;
1108         lustre_end_log(sb, profilenm, &cfg);
1109
1110         params_cfg.cfg_instance = sb;
1111         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1112
1113         if (sbi->ll_md_exp) {
1114                 obd = class_exp2obd(sbi->ll_md_exp);
1115                 if (obd)
1116                         force = obd->obd_force;
1117         }
1118
1119         /* Wait for unstable pages to be committed to stable storage */
1120         if (force == 0) {
1121                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1122                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1123                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1124                         &lwi);
1125         }
1126
1127         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1128         if (force == 0 && rc != -EINTR)
1129                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1130
1131
1132         /* We need to set force before the lov_disconnect in
1133            lustre_common_put_super, since l_d cleans up osc's as well. */
1134         if (force) {
1135                 next = 0;
1136                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1137                                                      &next)) != NULL) {
1138                         obd->obd_force = force;
1139                 }
1140         }
1141
1142         if (sbi->ll_client_common_fill_super_succeeded) {
1143                 /* Only if client_common_fill_super succeeded */
1144                 client_common_put_super(sb);
1145         }
1146
1147         next = 0;
1148         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1149                 class_manual_cleanup(obd);
1150         }
1151
1152         if (sbi->ll_flags & LL_SBI_VERBOSE)
1153                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1154
1155         if (profilenm)
1156                 class_del_profile(profilenm);
1157
1158         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1159                 bdi_destroy(&lsi->lsi_bdi);
1160                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1161         }
1162
1163         ll_free_sbi(sb);
1164         lsi->lsi_llsbi = NULL;
1165
1166         lustre_common_put_super(sb);
1167
1168         cl_env_cache_purge(~0);
1169
1170         module_put(THIS_MODULE);
1171
1172         EXIT;
1173 } /* client_put_super */
1174
1175 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1176 {
1177         struct inode *inode = NULL;
1178
1179         /* NOTE: we depend on atomic igrab() -bzzz */
1180         lock_res_and_lock(lock);
1181         if (lock->l_resource->lr_lvb_inode) {
1182                 struct ll_inode_info * lli;
1183                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1184                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1185                         inode = igrab(lock->l_resource->lr_lvb_inode);
1186                 } else {
1187                         inode = lock->l_resource->lr_lvb_inode;
1188                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1189                                          D_WARNING, lock, "lr_lvb_inode %p is "
1190                                          "bogus: magic %08x",
1191                                          lock->l_resource->lr_lvb_inode,
1192                                          lli->lli_inode_magic);
1193                         inode = NULL;
1194                 }
1195         }
1196         unlock_res_and_lock(lock);
1197         return inode;
1198 }
1199
1200 static void ll_dir_clear_lsm_md(struct inode *inode)
1201 {
1202         struct ll_inode_info *lli = ll_i2info(inode);
1203
1204         LASSERT(S_ISDIR(inode->i_mode));
1205
1206         if (lli->lli_lsm_md != NULL) {
1207                 lmv_free_memmd(lli->lli_lsm_md);
1208                 lli->lli_lsm_md = NULL;
1209         }
1210 }
1211
1212 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1213                                       const struct lu_fid *fid,
1214                                       struct lustre_md *md)
1215 {
1216         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1217         struct mdt_body         *body = md->body;
1218         struct inode            *inode;
1219         ino_t                   ino;
1220         ENTRY;
1221
1222         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1223         inode = iget_locked(sb, ino);
1224         if (inode == NULL) {
1225                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1226                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1227                 RETURN(ERR_PTR(-ENOENT));
1228         }
1229
1230         if (inode->i_state & I_NEW) {
1231                 struct ll_inode_info *lli = ll_i2info(inode);
1232                 struct lmv_stripe_md *lsm = md->lmv;
1233
1234                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1235                                 (body->mbo_mode & S_IFMT);
1236                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1237                          PFID(fid));
1238
1239                 LTIME_S(inode->i_mtime) = 0;
1240                 LTIME_S(inode->i_atime) = 0;
1241                 LTIME_S(inode->i_ctime) = 0;
1242                 inode->i_rdev = 0;
1243
1244                 /* initializing backing dev info. */
1245                 inode->i_mapping->backing_dev_info =
1246                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1247                 inode->i_op = &ll_dir_inode_operations;
1248                 inode->i_fop = &ll_dir_operations;
1249                 lli->lli_fid = *fid;
1250                 ll_lli_init(lli);
1251
1252                 LASSERT(lsm != NULL);
1253                 /* master object FID */
1254                 lli->lli_pfid = body->mbo_fid1;
1255                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1256                        lli, PFID(fid), PFID(&lli->lli_pfid));
1257                 unlock_new_inode(inode);
1258         }
1259
1260         RETURN(inode);
1261 }
1262
1263 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1264 {
1265         struct lu_fid *fid;
1266         struct lmv_stripe_md *lsm = md->lmv;
1267         int i;
1268
1269         LASSERT(lsm != NULL);
1270         /* XXX sigh, this lsm_root initialization should be in
1271          * LMV layer, but it needs ll_iget right now, so we
1272          * put this here right now. */
1273         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1274                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1275                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1276                 /* Unfortunately ll_iget will call ll_update_inode,
1277                  * where the initialization of slave inode is slightly
1278                  * different, so it reset lsm_md to NULL to avoid
1279                  * initializing lsm for slave inode. */
1280                 /* For migrating inode, master stripe and master object will
1281                  * be same, so we only need assign this inode */
1282                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1283                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1284                 else
1285                         lsm->lsm_md_oinfo[i].lmo_root =
1286                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1287
1288                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1289                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1290
1291                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1292                         return rc;
1293                 }
1294         }
1295
1296         return 0;
1297 }
1298
1299 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1300                                 const struct lmv_stripe_md *lsm_md2)
1301 {
1302         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1303                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1304                lsm_md1->lsm_md_master_mdt_index ==
1305                                         lsm_md2->lsm_md_master_mdt_index &&
1306                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1307                lsm_md1->lsm_md_layout_version ==
1308                                         lsm_md2->lsm_md_layout_version &&
1309                strcmp(lsm_md1->lsm_md_pool_name,
1310                       lsm_md2->lsm_md_pool_name) == 0;
1311 }
1312
1313 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1314 {
1315         struct ll_inode_info *lli = ll_i2info(inode);
1316         struct lmv_stripe_md *lsm = md->lmv;
1317         int     rc;
1318         ENTRY;
1319
1320         LASSERT(S_ISDIR(inode->i_mode));
1321         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1322                PFID(ll_inode2fid(inode)));
1323
1324         /* no striped information from request. */
1325         if (lsm == NULL) {
1326                 if (lli->lli_lsm_md == NULL) {
1327                         RETURN(0);
1328                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1329                                                 LMV_HASH_FLAG_MIGRATION) {
1330                         /* migration is done, the temporay MIGRATE layout has
1331                          * been removed */
1332                         CDEBUG(D_INODE, DFID" finish migration.\n",
1333                                PFID(ll_inode2fid(inode)));
1334                         lmv_free_memmd(lli->lli_lsm_md);
1335                         lli->lli_lsm_md = NULL;
1336                         RETURN(0);
1337                 } else {
1338                         /* The lustre_md from req does not include stripeEA,
1339                          * see ll_md_setattr */
1340                         RETURN(0);
1341                 }
1342         }
1343
1344         /* set the directory layout */
1345         if (lli->lli_lsm_md == NULL) {
1346
1347                 rc = ll_init_lsm_md(inode, md);
1348                 if (rc != 0)
1349                         RETURN(rc);
1350
1351                 lli->lli_lsm_md = lsm;
1352                 /* set lsm_md to NULL, so the following free lustre_md
1353                  * will not free this lsm */
1354                 md->lmv = NULL;
1355                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1356                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1357                 RETURN(0);
1358         }
1359
1360         /* Compare the old and new stripe information */
1361         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1362                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1363                 int                     idx;
1364
1365                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1366                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1367                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1368                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1369                        inode, lsm, old_lsm,
1370                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1371                        lsm->lsm_md_stripe_count,
1372                        old_lsm->lsm_md_stripe_count,
1373                        lsm->lsm_md_master_mdt_index,
1374                        old_lsm->lsm_md_master_mdt_index,
1375                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1376                        lsm->lsm_md_layout_version,
1377                        old_lsm->lsm_md_layout_version,
1378                        lsm->lsm_md_pool_name,
1379                        old_lsm->lsm_md_pool_name);
1380
1381                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1382                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1383                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1385                 }
1386
1387                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1388                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1389                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1390                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1391                 }
1392
1393                 RETURN(-EIO);
1394         }
1395
1396         RETURN(0);
1397 }
1398
1399 void ll_clear_inode(struct inode *inode)
1400 {
1401         struct ll_inode_info *lli = ll_i2info(inode);
1402         struct ll_sb_info *sbi = ll_i2sbi(inode);
1403         ENTRY;
1404
1405         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1406                PFID(ll_inode2fid(inode)), inode);
1407
1408         if (S_ISDIR(inode->i_mode)) {
1409                 /* these should have been cleared in ll_file_release */
1410                 LASSERT(lli->lli_opendir_key == NULL);
1411                 LASSERT(lli->lli_sai == NULL);
1412                 LASSERT(lli->lli_opendir_pid == 0);
1413         }
1414
1415         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1416
1417         LASSERT(!lli->lli_open_fd_write_count);
1418         LASSERT(!lli->lli_open_fd_read_count);
1419         LASSERT(!lli->lli_open_fd_exec_count);
1420
1421         if (lli->lli_mds_write_och)
1422                 ll_md_real_close(inode, FMODE_WRITE);
1423         if (lli->lli_mds_exec_och)
1424                 ll_md_real_close(inode, FMODE_EXEC);
1425         if (lli->lli_mds_read_och)
1426                 ll_md_real_close(inode, FMODE_READ);
1427
1428         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429                 OBD_FREE(lli->lli_symlink_name,
1430                          strlen(lli->lli_symlink_name) + 1);
1431                 lli->lli_symlink_name = NULL;
1432         }
1433
1434         ll_xattr_cache_destroy(inode);
1435
1436         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1437                 LASSERT(lli->lli_posix_acl == NULL);
1438                 if (lli->lli_remote_perms) {
1439                         free_rmtperm_hash(lli->lli_remote_perms);
1440                         lli->lli_remote_perms = NULL;
1441                 }
1442         }
1443 #ifdef CONFIG_FS_POSIX_ACL
1444         else if (lli->lli_posix_acl) {
1445                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1446                 LASSERT(lli->lli_remote_perms == NULL);
1447                 posix_acl_release(lli->lli_posix_acl);
1448                 lli->lli_posix_acl = NULL;
1449         }
1450 #endif
1451         lli->lli_inode_magic = LLI_INODE_DEAD;
1452
1453         ll_clear_inode_capas(inode);
1454         if (S_ISDIR(inode->i_mode))
1455                 ll_dir_clear_lsm_md(inode);
1456         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1457                 LASSERT(list_empty(&lli->lli_agl_list));
1458
1459         /*
1460          * XXX This has to be done before lsm is freed below, because
1461          * cl_object still uses inode lsm.
1462          */
1463         cl_inode_fini(inode);
1464
1465         EXIT;
1466 }
1467
1468 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1469 {
1470         struct lustre_md md;
1471         struct inode *inode = dentry->d_inode;
1472         struct ll_sb_info *sbi = ll_i2sbi(inode);
1473         struct ptlrpc_request *request = NULL;
1474         int rc, ia_valid;
1475         ENTRY;
1476
1477         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1478                                      LUSTRE_OPC_ANY, NULL);
1479         if (IS_ERR(op_data))
1480                 RETURN(PTR_ERR(op_data));
1481
1482         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1483         if (rc) {
1484                 ptlrpc_req_finished(request);
1485                 if (rc == -ENOENT) {
1486                         clear_nlink(inode);
1487                         /* Unlinked special device node? Or just a race?
1488                          * Pretend we done everything. */
1489                         if (!S_ISREG(inode->i_mode) &&
1490                             !S_ISDIR(inode->i_mode)) {
1491                                 ia_valid = op_data->op_attr.ia_valid;
1492                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1493                                 rc = simple_setattr(dentry, &op_data->op_attr);
1494                                 op_data->op_attr.ia_valid = ia_valid;
1495                         }
1496                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1497                         CERROR("md_setattr fails: rc = %d\n", rc);
1498                 }
1499                 RETURN(rc);
1500         }
1501
1502         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1503                               sbi->ll_md_exp, &md);
1504         if (rc) {
1505                 ptlrpc_req_finished(request);
1506                 RETURN(rc);
1507         }
1508
1509         ia_valid = op_data->op_attr.ia_valid;
1510         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1511          * cache is not cleared yet. */
1512         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1513         rc = simple_setattr(dentry, &op_data->op_attr);
1514         op_data->op_attr.ia_valid = ia_valid;
1515
1516         rc = ll_update_inode(inode, &md);
1517         ptlrpc_req_finished(request);
1518
1519         RETURN(rc);
1520 }
1521
1522 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1523 {
1524         struct obd_capa *capa;
1525         int rc;
1526
1527         if (attr->ia_valid & ATTR_SIZE)
1528                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1529         else
1530                 capa = ll_mdscapa_get(inode);
1531
1532         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1533
1534         if (attr->ia_valid & ATTR_SIZE)
1535                 ll_truncate_free_capa(capa);
1536         else
1537                 capa_put(capa);
1538
1539         return rc;
1540 }
1541
1542 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1543  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1544  * keep these values until such a time that objects are allocated for it.
1545  * We do the MDS operations first, as it is checking permissions for us.
1546  * We don't to the MDS RPC if there is nothing that we want to store there,
1547  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1548  * going to do an RPC anyways.
1549  *
1550  * If we are doing a truncate, we will send the mtime and ctime updates
1551  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1552  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1553  * at the same time.
1554  *
1555  * In case of HSMimport, we only set attr on MDS.
1556  */
1557 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1558 {
1559         struct inode *inode = dentry->d_inode;
1560         struct ll_inode_info *lli = ll_i2info(inode);
1561         struct md_op_data *op_data = NULL;
1562         bool file_is_released = false;
1563         int rc = 0;
1564         ENTRY;
1565
1566         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1567                "valid %x, hsm_import %d\n",
1568                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1569                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1570                hsm_import);
1571
1572         if (attr->ia_valid & ATTR_SIZE) {
1573                 /* Check new size against VFS/VM file size limit and rlimit */
1574                 rc = inode_newsize_ok(inode, attr->ia_size);
1575                 if (rc)
1576                         RETURN(rc);
1577
1578                 /* The maximum Lustre file size is variable, based on the
1579                  * OST maximum object size and number of stripes.  This
1580                  * needs another check in addition to the VFS check above. */
1581                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1582                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1583                                PFID(&lli->lli_fid), attr->ia_size,
1584                                ll_file_maxbytes(inode));
1585                         RETURN(-EFBIG);
1586                 }
1587
1588                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1589         }
1590
1591         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1592         if (attr->ia_valid & TIMES_SET_FLAGS) {
1593                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1594                     !cfs_capable(CFS_CAP_FOWNER))
1595                         RETURN(-EPERM);
1596         }
1597
1598         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1599         if (attr->ia_valid & ATTR_CTIME) {
1600                 attr->ia_ctime = CFS_CURRENT_TIME;
1601                 attr->ia_valid |= ATTR_CTIME_SET;
1602         }
1603         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1604             (attr->ia_valid & ATTR_ATIME)) {
1605                 attr->ia_atime = CFS_CURRENT_TIME;
1606                 attr->ia_valid |= ATTR_ATIME_SET;
1607         }
1608         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1609             (attr->ia_valid & ATTR_MTIME)) {
1610                 attr->ia_mtime = CFS_CURRENT_TIME;
1611                 attr->ia_valid |= ATTR_MTIME_SET;
1612         }
1613
1614         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1615                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1616                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1617                        cfs_time_current_sec());
1618
1619         /* We always do an MDS RPC, even if we're only changing the size;
1620          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1621
1622         OBD_ALLOC_PTR(op_data);
1623         if (op_data == NULL)
1624                 RETURN(-ENOMEM);
1625
1626         if (!S_ISDIR(inode->i_mode)) {
1627                 if (attr->ia_valid & ATTR_SIZE)
1628                         inode_dio_write_done(inode);
1629                 mutex_unlock(&inode->i_mutex);
1630         }
1631
1632         /* truncate on a released file must failed with -ENODATA,
1633          * so size must not be set on MDS for released file
1634          * but other attributes must be set
1635          */
1636         if (S_ISREG(inode->i_mode)) {
1637                 struct cl_layout cl = {
1638                         .cl_is_released = false,
1639                 };
1640                 struct lu_env *env;
1641                 int refcheck;
1642                 __u32 gen;
1643
1644                 rc = ll_layout_refresh(inode, &gen);
1645                 if (rc < 0)
1646                         GOTO(out, rc);
1647
1648                 /* XXX: the only place we need to know the layout type,
1649                  * this will be removed by a later patch. -Jinshan */
1650                 env = cl_env_get(&refcheck);
1651                 if (IS_ERR(env))
1652                         GOTO(out, rc = PTR_ERR(env));
1653
1654                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1655                 cl_env_put(env, &refcheck);
1656                 if (rc < 0)
1657                         GOTO(out, rc);
1658
1659                 file_is_released = cl.cl_is_released;
1660
1661                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1662                         if (file_is_released) {
1663                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1664                                 if (rc < 0)
1665                                         GOTO(out, rc);
1666
1667                                 file_is_released = false;
1668                                 ll_layout_refresh(inode, &gen);
1669                         }
1670
1671                         /* If we are changing file size, file content is
1672                          * modified, flag it. */
1673                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1674                         spin_lock(&lli->lli_lock);
1675                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1676                         spin_unlock(&lli->lli_lock);
1677                         op_data->op_bias |= MDS_DATA_MODIFIED;
1678                 }
1679         }
1680
1681         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1682
1683         rc = ll_md_setattr(dentry, op_data);
1684         if (rc)
1685                 GOTO(out, rc);
1686
1687         /* RPC to MDT is sent, cancel data modification flag */
1688         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1689                 spin_lock(&lli->lli_lock);
1690                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1691                 spin_unlock(&lli->lli_lock);
1692         }
1693
1694         if (!S_ISREG(inode->i_mode) || file_is_released)
1695                 GOTO(out, rc = 0);
1696
1697         if (attr->ia_valid & (ATTR_SIZE |
1698                               ATTR_ATIME | ATTR_ATIME_SET |
1699                               ATTR_MTIME | ATTR_MTIME_SET)) {
1700                 /* For truncate and utimes sending attributes to OSTs, setting
1701                  * mtime/atime to the past will be performed under PW [0:EOF]
1702                  * extent lock (new_size:EOF for truncate).  It may seem
1703                  * excessive to send mtime/atime updates to OSTs when not
1704                  * setting times to past, but it is necessary due to possible
1705                  * time de-synchronization between MDT inode and OST objects */
1706                 if (attr->ia_valid & ATTR_SIZE)
1707                         down_write(&lli->lli_trunc_sem);
1708                 rc = ll_setattr_ost(inode, attr);
1709                 if (attr->ia_valid & ATTR_SIZE)
1710                         up_write(&lli->lli_trunc_sem);
1711         }
1712         EXIT;
1713 out:
1714         if (op_data != NULL)
1715                 ll_finish_md_op_data(op_data);
1716
1717         if (!S_ISDIR(inode->i_mode)) {
1718                 mutex_lock(&inode->i_mutex);
1719                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1720                         inode_dio_wait(inode);
1721         }
1722
1723         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1724                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1725
1726         return rc;
1727 }
1728
1729 int ll_setattr(struct dentry *de, struct iattr *attr)
1730 {
1731         int mode = de->d_inode->i_mode;
1732
1733         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1734                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1735                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1736
1737         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1738                                (ATTR_SIZE|ATTR_MODE)) &&
1739             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1740              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1741               !(attr->ia_mode & S_ISGID))))
1742                 attr->ia_valid |= ATTR_FORCE;
1743
1744         if ((attr->ia_valid & ATTR_MODE) &&
1745             (mode & S_ISUID) &&
1746             !(attr->ia_mode & S_ISUID) &&
1747             !(attr->ia_valid & ATTR_KILL_SUID))
1748                 attr->ia_valid |= ATTR_KILL_SUID;
1749
1750         if ((attr->ia_valid & ATTR_MODE) &&
1751             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1752             !(attr->ia_mode & S_ISGID) &&
1753             !(attr->ia_valid & ATTR_KILL_SGID))
1754                 attr->ia_valid |= ATTR_KILL_SGID;
1755
1756         return ll_setattr_raw(de, attr, false);
1757 }
1758
1759 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1760                        __u64 max_age, __u32 flags)
1761 {
1762         struct ll_sb_info *sbi = ll_s2sbi(sb);
1763         struct obd_statfs obd_osfs;
1764         int rc;
1765         ENTRY;
1766
1767         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1768         if (rc) {
1769                 CERROR("md_statfs fails: rc = %d\n", rc);
1770                 RETURN(rc);
1771         }
1772
1773         osfs->os_type = sb->s_magic;
1774
1775         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1776                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1777
1778         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1779                 flags |= OBD_STATFS_NODELAY;
1780
1781         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1782         if (rc) {
1783                 CERROR("obd_statfs fails: rc = %d\n", rc);
1784                 RETURN(rc);
1785         }
1786
1787         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1788                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1789                obd_osfs.os_files);
1790
1791         osfs->os_bsize = obd_osfs.os_bsize;
1792         osfs->os_blocks = obd_osfs.os_blocks;
1793         osfs->os_bfree = obd_osfs.os_bfree;
1794         osfs->os_bavail = obd_osfs.os_bavail;
1795
1796         /* If we don't have as many objects free on the OST as inodes
1797          * on the MDS, we reduce the total number of inodes to
1798          * compensate, so that the "inodes in use" number is correct.
1799          */
1800         if (obd_osfs.os_ffree < osfs->os_ffree) {
1801                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1802                         obd_osfs.os_ffree;
1803                 osfs->os_ffree = obd_osfs.os_ffree;
1804         }
1805
1806         RETURN(rc);
1807 }
1808 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1809 {
1810         struct super_block *sb = de->d_sb;
1811         struct obd_statfs osfs;
1812         __u64 fsid = huge_encode_dev(sb->s_dev);
1813         int rc;
1814
1815         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1816         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1817
1818         /* Some amount of caching on the client is allowed */
1819         rc = ll_statfs_internal(sb, &osfs,
1820                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1821                                 0);
1822         if (rc)
1823                 return rc;
1824
1825         statfs_unpack(sfs, &osfs);
1826
1827         /* We need to downshift for all 32-bit kernels, because we can't
1828          * tell if the kernel is being called via sys_statfs64() or not.
1829          * Stop before overflowing f_bsize - in which case it is better
1830          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1831         if (sizeof(long) < 8) {
1832                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1833                         sfs->f_bsize <<= 1;
1834
1835                         osfs.os_blocks >>= 1;
1836                         osfs.os_bfree >>= 1;
1837                         osfs.os_bavail >>= 1;
1838                 }
1839         }
1840
1841         sfs->f_blocks = osfs.os_blocks;
1842         sfs->f_bfree = osfs.os_bfree;
1843         sfs->f_bavail = osfs.os_bavail;
1844         sfs->f_fsid.val[0] = (__u32)fsid;
1845         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1846         return 0;
1847 }
1848
1849 void ll_inode_size_lock(struct inode *inode)
1850 {
1851         struct ll_inode_info *lli;
1852
1853         LASSERT(!S_ISDIR(inode->i_mode));
1854
1855         lli = ll_i2info(inode);
1856         mutex_lock(&lli->lli_size_mutex);
1857 }
1858
1859 void ll_inode_size_unlock(struct inode *inode)
1860 {
1861         struct ll_inode_info *lli;
1862
1863         lli = ll_i2info(inode);
1864         mutex_unlock(&lli->lli_size_mutex);
1865 }
1866
1867 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1868 {
1869         struct ll_inode_info *lli = ll_i2info(inode);
1870         struct mdt_body *body = md->body;
1871         struct lov_stripe_md *lsm = md->lsm;
1872         struct ll_sb_info *sbi = ll_i2sbi(inode);
1873
1874         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1875         if (lsm != NULL)
1876                 cl_file_inode_init(inode, md);
1877
1878         if (S_ISDIR(inode->i_mode)) {
1879                 int     rc;
1880
1881                 rc = ll_update_lsm_md(inode, md);
1882                 if (rc != 0)
1883                         return rc;
1884         }
1885
1886         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1887                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1888                         ll_update_remote_perm(inode, md->remote_perm);
1889         }
1890 #ifdef CONFIG_FS_POSIX_ACL
1891         else if (body->mbo_valid & OBD_MD_FLACL) {
1892                 spin_lock(&lli->lli_lock);
1893                 if (lli->lli_posix_acl)
1894                         posix_acl_release(lli->lli_posix_acl);
1895                 lli->lli_posix_acl = md->posix_acl;
1896                 spin_unlock(&lli->lli_lock);
1897         }
1898 #endif
1899         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1900                                         sbi->ll_flags & LL_SBI_32BIT_API);
1901         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1902
1903         if (body->mbo_valid & OBD_MD_FLATIME) {
1904                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1905                         LTIME_S(inode->i_atime) = body->mbo_atime;
1906                 lli->lli_atime = body->mbo_atime;
1907         }
1908
1909         if (body->mbo_valid & OBD_MD_FLMTIME) {
1910                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1911                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1912                                "to "LPU64"\n", inode->i_ino,
1913                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1914                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1915                 }
1916                 lli->lli_mtime = body->mbo_mtime;
1917         }
1918
1919         if (body->mbo_valid & OBD_MD_FLCTIME) {
1920                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1921                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1922                 lli->lli_ctime = body->mbo_ctime;
1923         }
1924
1925         if (body->mbo_valid & OBD_MD_FLMODE)
1926                 inode->i_mode = (inode->i_mode & S_IFMT) |
1927                                 (body->mbo_mode & ~S_IFMT);
1928
1929         if (body->mbo_valid & OBD_MD_FLTYPE)
1930                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1931                                 (body->mbo_mode & S_IFMT);
1932
1933         LASSERT(inode->i_mode != 0);
1934         if (S_ISREG(inode->i_mode))
1935                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1936                                        LL_MAX_BLKSIZE_BITS);
1937         else
1938                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1939
1940         if (body->mbo_valid & OBD_MD_FLUID)
1941                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1942         if (body->mbo_valid & OBD_MD_FLGID)
1943                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1944         if (body->mbo_valid & OBD_MD_FLFLAGS)
1945                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1946         if (body->mbo_valid & OBD_MD_FLNLINK)
1947                 set_nlink(inode, body->mbo_nlink);
1948         if (body->mbo_valid & OBD_MD_FLRDEV)
1949                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1950
1951         if (body->mbo_valid & OBD_MD_FLID) {
1952                 /* FID shouldn't be changed! */
1953                 if (fid_is_sane(&lli->lli_fid)) {
1954                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1955                                  "Trying to change FID "DFID
1956                                  " to the "DFID", inode "DFID"(%p)\n",
1957                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1958                                  PFID(ll_inode2fid(inode)), inode);
1959                 } else {
1960                         lli->lli_fid = body->mbo_fid1;
1961                 }
1962         }
1963
1964         LASSERT(fid_seq(&lli->lli_fid) != 0);
1965
1966         if (body->mbo_valid & OBD_MD_FLSIZE) {
1967                 i_size_write(inode, body->mbo_size);
1968
1969                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1970                        PFID(ll_inode2fid(inode)),
1971                        (unsigned long long)body->mbo_size);
1972
1973                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1974                         inode->i_blocks = body->mbo_blocks;
1975         }
1976
1977         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1978                 LASSERT(md->mds_capa);
1979                 ll_add_capa(inode, md->mds_capa);
1980         }
1981
1982         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
1983                 LASSERT(md->oss_capa);
1984                 ll_add_capa(inode, md->oss_capa);
1985         }
1986
1987         if (body->mbo_valid & OBD_MD_TSTATE) {
1988                 if (body->mbo_t_state & MS_RESTORE)
1989                         lli->lli_flags |= LLIF_FILE_RESTORING;
1990         }
1991
1992         return 0;
1993 }
1994
1995 int ll_read_inode2(struct inode *inode, void *opaque)
1996 {
1997         struct lustre_md *md = opaque;
1998         struct ll_inode_info *lli = ll_i2info(inode);
1999         int     rc;
2000         ENTRY;
2001
2002         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2003                PFID(&lli->lli_fid), inode);
2004
2005         /* Core attributes from the MDS first.  This is a new inode, and
2006          * the VFS doesn't zero times in the core inode so we have to do
2007          * it ourselves.  They will be overwritten by either MDS or OST
2008          * attributes - we just need to make sure they aren't newer. */
2009         LTIME_S(inode->i_mtime) = 0;
2010         LTIME_S(inode->i_atime) = 0;
2011         LTIME_S(inode->i_ctime) = 0;
2012         inode->i_rdev = 0;
2013         rc = ll_update_inode(inode, md);
2014         if (rc != 0)
2015                 RETURN(rc);
2016
2017         /* OIDEBUG(inode); */
2018
2019         /* initializing backing dev info. */
2020         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2021
2022
2023         if (S_ISREG(inode->i_mode)) {
2024                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2025                 inode->i_op = &ll_file_inode_operations;
2026                 inode->i_fop = sbi->ll_fop;
2027                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2028                 EXIT;
2029         } else if (S_ISDIR(inode->i_mode)) {
2030                 inode->i_op = &ll_dir_inode_operations;
2031                 inode->i_fop = &ll_dir_operations;
2032                 EXIT;
2033         } else if (S_ISLNK(inode->i_mode)) {
2034                 inode->i_op = &ll_fast_symlink_inode_operations;
2035                 EXIT;
2036         } else {
2037                 inode->i_op = &ll_special_inode_operations;
2038
2039                 init_special_inode(inode, inode->i_mode,
2040                                    inode->i_rdev);
2041
2042                 EXIT;
2043         }
2044
2045         return 0;
2046 }
2047
2048 void ll_delete_inode(struct inode *inode)
2049 {
2050         struct ll_inode_info *lli = ll_i2info(inode);
2051         ENTRY;
2052
2053         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2054                 /* It is last chance to write out dirty pages,
2055                  * otherwise we may lose data while umount */
2056                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2057
2058         truncate_inode_pages_final(&inode->i_data);
2059
2060         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2061                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2062                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2063
2064 #ifdef HAVE_SBOPS_EVICT_INODE
2065         ll_clear_inode(inode);
2066 #endif
2067         clear_inode(inode);
2068
2069         EXIT;
2070 }
2071
2072 int ll_iocontrol(struct inode *inode, struct file *file,
2073                  unsigned int cmd, unsigned long arg)
2074 {
2075         struct ll_sb_info *sbi = ll_i2sbi(inode);
2076         struct ptlrpc_request *req = NULL;
2077         int rc, flags = 0;
2078         ENTRY;
2079
2080         switch(cmd) {
2081         case FSFILT_IOC_GETFLAGS: {
2082                 struct mdt_body *body;
2083                 struct md_op_data *op_data;
2084
2085                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2086                                              0, 0, LUSTRE_OPC_ANY,
2087                                              NULL);
2088                 if (IS_ERR(op_data))
2089                         RETURN(PTR_ERR(op_data));
2090
2091                 op_data->op_valid = OBD_MD_FLFLAGS;
2092                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2093                 ll_finish_md_op_data(op_data);
2094                 if (rc) {
2095                         CERROR("%s: failure inode "DFID": rc = %d\n",
2096                                sbi->ll_md_exp->exp_obd->obd_name,
2097                                PFID(ll_inode2fid(inode)), rc);
2098                         RETURN(-abs(rc));
2099                 }
2100
2101                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2102
2103                 flags = body->mbo_flags;
2104
2105                 ptlrpc_req_finished(req);
2106
2107                 RETURN(put_user(flags, (int __user *)arg));
2108         }
2109         case FSFILT_IOC_SETFLAGS: {
2110                 struct iattr *attr;
2111                 struct md_op_data *op_data;
2112                 struct cl_object *obj;
2113                 struct obd_capa *capa;
2114
2115                 if (get_user(flags, (int __user *)arg))
2116                         RETURN(-EFAULT);
2117
2118                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2119                                              LUSTRE_OPC_ANY, NULL);
2120                 if (IS_ERR(op_data))
2121                         RETURN(PTR_ERR(op_data));
2122
2123                 op_data->op_attr_flags = flags;
2124                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2125                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2126                 ll_finish_md_op_data(op_data);
2127                 ptlrpc_req_finished(req);
2128                 if (rc)
2129                         RETURN(rc);
2130
2131                 inode->i_flags = ll_ext_to_inode_flags(flags);
2132
2133                 obj = ll_i2info(inode)->lli_clob;
2134                 if (obj == NULL)
2135                         RETURN(0);
2136
2137                 OBD_ALLOC_PTR(attr);
2138                 if (attr == NULL)
2139                         RETURN(-ENOMEM);
2140
2141                 attr->ia_valid = ATTR_ATTR_FLAG;
2142
2143                 capa = ll_mdscapa_get(inode);
2144                 rc = cl_setattr_ost(obj, attr, flags, capa);
2145                 capa_put(capa);
2146
2147                 OBD_FREE_PTR(attr);
2148                 RETURN(rc);
2149         }
2150         default:
2151                 RETURN(-ENOSYS);
2152         }
2153
2154         RETURN(0);
2155 }
2156
2157 int ll_flush_ctx(struct inode *inode)
2158 {
2159         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2160
2161         CDEBUG(D_SEC, "flush context for user %d\n",
2162                from_kuid(&init_user_ns, current_uid()));
2163
2164         obd_set_info_async(NULL, sbi->ll_md_exp,
2165                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2166                            0, NULL, NULL);
2167         obd_set_info_async(NULL, sbi->ll_dt_exp,
2168                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2169                            0, NULL, NULL);
2170         return 0;
2171 }
2172
2173 /* umount -f client means force down, don't save state */
2174 void ll_umount_begin(struct super_block *sb)
2175 {
2176         struct ll_sb_info *sbi = ll_s2sbi(sb);
2177         struct obd_device *obd;
2178         struct obd_ioctl_data *ioc_data;
2179         ENTRY;
2180
2181         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2182                sb->s_count, atomic_read(&sb->s_active));
2183
2184         obd = class_exp2obd(sbi->ll_md_exp);
2185         if (obd == NULL) {
2186                 CERROR("Invalid MDC connection handle "LPX64"\n",
2187                        sbi->ll_md_exp->exp_handle.h_cookie);
2188                 EXIT;
2189                 return;
2190         }
2191         obd->obd_force = 1;
2192
2193         obd = class_exp2obd(sbi->ll_dt_exp);
2194         if (obd == NULL) {
2195                 CERROR("Invalid LOV connection handle "LPX64"\n",
2196                        sbi->ll_dt_exp->exp_handle.h_cookie);
2197                 EXIT;
2198                 return;
2199         }
2200         obd->obd_force = 1;
2201
2202         OBD_ALLOC_PTR(ioc_data);
2203         if (ioc_data) {
2204                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2205                               sizeof *ioc_data, ioc_data, NULL);
2206
2207                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2208                               sizeof *ioc_data, ioc_data, NULL);
2209
2210                 OBD_FREE_PTR(ioc_data);
2211         }
2212
2213         /* Really, we'd like to wait until there are no requests outstanding,
2214          * and then continue.  For now, we just invalidate the requests,
2215          * schedule() and sleep one second if needed, and hope.
2216          */
2217         schedule();
2218         EXIT;
2219 }
2220
2221 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2222 {
2223         struct ll_sb_info *sbi = ll_s2sbi(sb);
2224         char *profilenm = get_profile_name(sb);
2225         int err;
2226         __u32 read_only;
2227
2228         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2229                 read_only = *flags & MS_RDONLY;
2230                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2231                                          sizeof(KEY_READ_ONLY),
2232                                          KEY_READ_ONLY, sizeof(read_only),
2233                                          &read_only, NULL);
2234                 if (err) {
2235                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2236                                       profilenm, read_only ?
2237                                       "read-only" : "read-write", err);
2238                         return err;
2239                 }
2240
2241                 if (read_only)
2242                         sb->s_flags |= MS_RDONLY;
2243                 else
2244                         sb->s_flags &= ~MS_RDONLY;
2245
2246                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2247                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2248                                       read_only ?  "read-only" : "read-write");
2249         }
2250         return 0;
2251 }
2252
2253 /**
2254  * Cleanup the open handle that is cached on MDT-side.
2255  *
2256  * For open case, the client side open handling thread may hit error
2257  * after the MDT grant the open. Under such case, the client should
2258  * send close RPC to the MDT as cleanup; otherwise, the open handle
2259  * on the MDT will be leaked there until the client umount or evicted.
2260  *
2261  * In further, if someone unlinked the file, because the open handle
2262  * holds the reference on such file/object, then it will block the
2263  * subsequent threads that want to locate such object via FID.
2264  *
2265  * \param[in] sb        super block for this file-system
2266  * \param[in] open_req  pointer to the original open request
2267  */
2268 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2269 {
2270         struct mdt_body                 *body;
2271         struct md_op_data               *op_data;
2272         struct ptlrpc_request           *close_req = NULL;
2273         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2274         ENTRY;
2275
2276         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2277         OBD_ALLOC_PTR(op_data);
2278         if (op_data == NULL) {
2279                 CWARN("%s: cannot allocate op_data to release open handle for "
2280                       DFID"\n",
2281                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2282
2283                 RETURN_EXIT;
2284         }
2285
2286         op_data->op_fid1 = body->mbo_fid1;
2287         op_data->op_handle = body->mbo_handle;
2288         op_data->op_mod_time = cfs_time_current_sec();
2289         md_close(exp, op_data, NULL, &close_req);
2290         ptlrpc_req_finished(close_req);
2291         ll_finish_md_op_data(op_data);
2292
2293         EXIT;
2294 }
2295
2296 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2297                   struct super_block *sb, struct lookup_intent *it)
2298 {
2299         struct ll_sb_info *sbi = NULL;
2300         struct lustre_md md = { NULL };
2301         int rc;
2302         ENTRY;
2303
2304         LASSERT(*inode || sb);
2305         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2306         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2307                               sbi->ll_md_exp, &md);
2308         if (rc != 0)
2309                 GOTO(cleanup, rc);
2310
2311         if (*inode) {
2312                 rc = ll_update_inode(*inode, &md);
2313                 if (rc != 0)
2314                         GOTO(out, rc);
2315         } else {
2316                 LASSERT(sb != NULL);
2317
2318                 /*
2319                  * At this point server returns to client's same fid as client
2320                  * generated for creating. So using ->fid1 is okay here.
2321                  */
2322                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2323
2324                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2325                                              sbi->ll_flags & LL_SBI_32BIT_API),
2326                                  &md);
2327                 if (IS_ERR(*inode)) {
2328 #ifdef CONFIG_FS_POSIX_ACL
2329                         if (md.posix_acl) {
2330                                 posix_acl_release(md.posix_acl);
2331                                 md.posix_acl = NULL;
2332                         }
2333 #endif
2334                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2335                         *inode = NULL;
2336                         CERROR("new_inode -fatal: rc %d\n", rc);
2337                         GOTO(out, rc);
2338                 }
2339         }
2340
2341         /* Handling piggyback layout lock.
2342          * Layout lock can be piggybacked by getattr and open request.
2343          * The lsm can be applied to inode only if it comes with a layout lock
2344          * otherwise correct layout may be overwritten, for example:
2345          * 1. proc1: mdt returns a lsm but not granting layout
2346          * 2. layout was changed by another client
2347          * 3. proc2: refresh layout and layout lock granted
2348          * 4. proc1: to apply a stale layout */
2349         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2350                 struct lustre_handle lockh;
2351                 struct ldlm_lock *lock;
2352
2353                 lockh.cookie = it->d.lustre.it_lock_handle;
2354                 lock = ldlm_handle2lock(&lockh);
2355                 LASSERT(lock != NULL);
2356                 if (ldlm_has_layout(lock)) {
2357                         struct cl_object_conf conf;
2358
2359                         memset(&conf, 0, sizeof(conf));
2360                         conf.coc_opc = OBJECT_CONF_SET;
2361                         conf.coc_inode = *inode;
2362                         conf.coc_lock = lock;
2363                         conf.u.coc_md = &md;
2364                         (void)ll_layout_conf(*inode, &conf);
2365                 }
2366                 LDLM_LOCK_PUT(lock);
2367         }
2368
2369         GOTO(out, rc = 0);
2370
2371 out:
2372         if (md.lsm != NULL)
2373                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2374         md_free_lustre_md(sbi->ll_md_exp, &md);
2375
2376 cleanup:
2377         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2378                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2379
2380         return rc;
2381 }
2382
2383 int ll_obd_statfs(struct inode *inode, void __user *arg)
2384 {
2385         struct ll_sb_info *sbi = NULL;
2386         struct obd_export *exp;
2387         char *buf = NULL;
2388         struct obd_ioctl_data *data = NULL;
2389         __u32 type;
2390         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2391         int len = 0, rc;
2392
2393         if (!inode || !(sbi = ll_i2sbi(inode)))
2394                 GOTO(out_statfs, rc = -EINVAL);
2395
2396         rc = obd_ioctl_getdata(&buf, &len, arg);
2397         if (rc)
2398                 GOTO(out_statfs, rc);
2399
2400         data = (void*)buf;
2401         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2402             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2403                 GOTO(out_statfs, rc = -EINVAL);
2404
2405         if (data->ioc_inllen1 != sizeof(__u32) ||
2406             data->ioc_inllen2 != sizeof(__u32) ||
2407             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2408             data->ioc_plen2 != sizeof(struct obd_uuid))
2409                 GOTO(out_statfs, rc = -EINVAL);
2410
2411         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2412         if (type & LL_STATFS_LMV)
2413                 exp = sbi->ll_md_exp;
2414         else if (type & LL_STATFS_LOV)
2415                 exp = sbi->ll_dt_exp;
2416         else
2417                 GOTO(out_statfs, rc = -ENODEV);
2418
2419         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2420         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2421         if (rc)
2422                 GOTO(out_statfs, rc);
2423 out_statfs:
2424         if (buf)
2425                 obd_ioctl_freedata(buf, len);
2426         return rc;
2427 }
2428
2429 int ll_process_config(struct lustre_cfg *lcfg)
2430 {
2431         struct super_block *sb;
2432         unsigned long x;
2433         int rc = 0;
2434         char *ptr;
2435
2436         /* The instance name contains the sb: lustre-client-aacfe000 */
2437         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2438         if (!ptr || !*(++ptr))
2439                 return -EINVAL;
2440         if (sscanf(ptr, "%lx", &x) != 1)
2441                 return -EINVAL;
2442         sb = (struct super_block *)x;
2443         /* This better be a real Lustre superblock! */
2444         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2445
2446         /* Note we have not called client_common_fill_super yet, so
2447            proc fns must be able to handle that! */
2448         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2449                                       lcfg, sb);
2450         if (rc > 0)
2451                 rc = 0;
2452         return rc;
2453 }
2454
2455 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2456 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2457                                        struct inode *i1, struct inode *i2,
2458                                        const char *name, size_t namelen,
2459                                        __u32 mode, __u32 opc, void *data)
2460 {
2461         LASSERT(i1 != NULL);
2462
2463         if (name == NULL) {
2464                 /* Do not reuse namelen for something else. */
2465                 if (namelen != 0)
2466                         return ERR_PTR(-EINVAL);
2467         } else {
2468                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2469                         return ERR_PTR(-ENAMETOOLONG);
2470
2471                 if (!lu_name_is_valid_2(name, namelen))
2472                         return ERR_PTR(-EINVAL);
2473         }
2474
2475         if (op_data == NULL)
2476                 OBD_ALLOC_PTR(op_data);
2477
2478         if (op_data == NULL)
2479                 return ERR_PTR(-ENOMEM);
2480
2481         ll_i2gids(op_data->op_suppgids, i1, i2);
2482         op_data->op_fid1 = *ll_inode2fid(i1);
2483         op_data->op_capa1 = ll_mdscapa_get(i1);
2484         op_data->op_default_stripe_offset = -1;
2485         if (S_ISDIR(i1->i_mode)) {
2486                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2487                 if (opc == LUSTRE_OPC_MKDIR)
2488                         op_data->op_default_stripe_offset =
2489                                    ll_i2info(i1)->lli_def_stripe_offset;
2490         }
2491
2492         if (i2) {
2493                 op_data->op_fid2 = *ll_inode2fid(i2);
2494                 op_data->op_capa2 = ll_mdscapa_get(i2);
2495                 if (S_ISDIR(i2->i_mode))
2496                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2497         } else {
2498                 fid_zero(&op_data->op_fid2);
2499                 op_data->op_capa2 = NULL;
2500         }
2501
2502         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2503                 op_data->op_cli_flags |= CLI_HASH64;
2504
2505         if (ll_need_32bit_api(ll_i2sbi(i1)))
2506                 op_data->op_cli_flags |= CLI_API32;
2507
2508         op_data->op_name = name;
2509         op_data->op_namelen = namelen;
2510         op_data->op_mode = mode;
2511         op_data->op_mod_time = cfs_time_current_sec();
2512         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2513         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2514         op_data->op_cap = cfs_curproc_cap_pack();
2515         op_data->op_bias = 0;
2516         op_data->op_cli_flags = 0;
2517         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2518              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2519                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2520         } else {
2521                 op_data->op_mds = 0;
2522         }
2523         op_data->op_data = data;
2524
2525         /* When called by ll_setattr_raw, file is i1. */
2526         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2527                 op_data->op_bias |= MDS_DATA_MODIFIED;
2528
2529         return op_data;
2530 }
2531
2532 void ll_finish_md_op_data(struct md_op_data *op_data)
2533 {
2534         capa_put(op_data->op_capa1);
2535         capa_put(op_data->op_capa2);
2536         OBD_FREE_PTR(op_data);
2537 }
2538
2539 #ifdef HAVE_SUPEROPS_USE_DENTRY
2540 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2541 #else
2542 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2543 #endif
2544 {
2545         struct ll_sb_info *sbi;
2546
2547 #ifdef HAVE_SUPEROPS_USE_DENTRY
2548         LASSERT((seq != NULL) && (dentry != NULL));
2549         sbi = ll_s2sbi(dentry->d_sb);
2550 #else
2551         LASSERT((seq != NULL) && (vfs != NULL));
2552         sbi = ll_s2sbi(vfs->mnt_sb);
2553 #endif
2554
2555         if (sbi->ll_flags & LL_SBI_NOLCK)
2556                 seq_puts(seq, ",nolock");
2557
2558         if (sbi->ll_flags & LL_SBI_FLOCK)
2559                 seq_puts(seq, ",flock");
2560
2561         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2562                 seq_puts(seq, ",localflock");
2563
2564         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2565                 seq_puts(seq, ",user_xattr");
2566
2567         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2568                 seq_puts(seq, ",lazystatfs");
2569
2570         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2571                 seq_puts(seq, ",user_fid2path");
2572
2573         RETURN(0);
2574 }
2575
2576 /**
2577  * Get obd name by cmd, and copy out to user space
2578  */
2579 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2580 {
2581         struct ll_sb_info *sbi = ll_i2sbi(inode);
2582         struct obd_device *obd;
2583         ENTRY;
2584
2585         if (cmd == OBD_IOC_GETDTNAME)
2586                 obd = class_exp2obd(sbi->ll_dt_exp);
2587         else if (cmd == OBD_IOC_GETMDNAME)
2588                 obd = class_exp2obd(sbi->ll_md_exp);
2589         else
2590                 RETURN(-EINVAL);
2591
2592         if (!obd)
2593                 RETURN(-ENOENT);
2594
2595         if (copy_to_user((void __user *)arg, obd->obd_name,
2596                          strlen(obd->obd_name) + 1))
2597                 RETURN(-EFAULT);
2598
2599         RETURN(0);
2600 }
2601
2602 /**
2603  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2604  * fsname will be returned in this buffer; otherwise, a static buffer will be
2605  * used to store the fsname and returned to caller.
2606  */
2607 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2608 {
2609         static char fsname_static[MTI_NAME_MAXLEN];
2610         struct lustre_sb_info *lsi = s2lsi(sb);
2611         char *ptr;
2612         int len;
2613
2614         if (buf == NULL) {
2615                 /* this means the caller wants to use static buffer
2616                  * and it doesn't care about race. Usually this is
2617                  * in error reporting path */
2618                 buf = fsname_static;
2619                 buflen = sizeof(fsname_static);
2620         }
2621
2622         len = strlen(lsi->lsi_lmd->lmd_profile);
2623         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2624         if (ptr && (strcmp(ptr, "-client") == 0))
2625                 len -= 7;
2626
2627         if (unlikely(len >= buflen))
2628                 len = buflen - 1;
2629         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2630         buf[len] = '\0';
2631
2632         return buf;
2633 }
2634
2635 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2636 {
2637         char *path = NULL;
2638
2639         struct path p;
2640
2641         p.dentry = dentry;
2642         p.mnt = current->fs->root.mnt;
2643         path_get(&p);
2644         path = d_path(&p, buf, bufsize);
2645         path_put(&p);
2646         return path;
2647 }
2648
2649 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2650 {
2651         char *buf, *path = NULL;
2652         struct dentry *dentry = NULL;
2653         struct inode *inode = page->mapping->host;
2654
2655         /* this can be called inside spin lock so use GFP_ATOMIC. */
2656         buf = (char *)__get_free_page(GFP_ATOMIC);
2657         if (buf != NULL) {
2658                 dentry = d_find_alias(page->mapping->host);
2659                 if (dentry != NULL)
2660                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2661         }
2662
2663         CDEBUG(D_WARNING,
2664                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2665                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2666                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2667                PFID(ll_inode2fid(inode)),
2668                (path && !IS_ERR(path)) ? path : "", ioret);
2669
2670         if (dentry != NULL)
2671                 dput(dentry);
2672
2673         if (buf != NULL)
2674                 free_page((unsigned long)buf);
2675 }
2676
2677 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2678                         struct lov_user_md **kbuf)
2679 {
2680         struct lov_user_md      lum;
2681         ssize_t                 lum_size;
2682         ENTRY;
2683
2684         if (copy_from_user(&lum, md, sizeof(lum)))
2685                 RETURN(-EFAULT);
2686
2687         lum_size = ll_lov_user_md_size(&lum);
2688         if (lum_size < 0)
2689                 RETURN(lum_size);
2690
2691         OBD_ALLOC(*kbuf, lum_size);
2692         if (*kbuf == NULL)
2693                 RETURN(-ENOMEM);
2694
2695         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2696                 OBD_FREE(*kbuf, lum_size);
2697                 RETURN(-EFAULT);
2698         }
2699
2700         RETURN(lum_size);
2701 }
2702
2703 /*
2704  * Compute llite root squash state after a change of root squash
2705  * configuration setting or add/remove of a lnet nid
2706  */
2707 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2708 {
2709         struct root_squash_info *squash = &sbi->ll_squash;
2710         int i;
2711         bool matched;
2712         lnet_process_id_t id;
2713
2714         /* Update norootsquash flag */
2715         down_write(&squash->rsi_sem);
2716         if (list_empty(&squash->rsi_nosquash_nids))
2717                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2718         else {
2719                 /* Do not apply root squash as soon as one of our NIDs is
2720                  * in the nosquash_nids list */
2721                 matched = false;
2722                 i = 0;
2723                 while (LNetGetId(i++, &id) != -ENOENT) {
2724                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2725                                 continue;
2726                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2727                                 matched = true;
2728                                 break;
2729                         }
2730                 }
2731                 if (matched)
2732                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2733                 else
2734                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2735         }
2736         up_write(&squash->rsi_sem);
2737 }
2738
2739 /**
2740  * Parse linkea content to extract information about a given hardlink
2741  *
2742  * \param[in]   ldata      - Initialized linkea data
2743  * \param[in]   linkno     - Link identifier
2744  * \param[out]  parent_fid - The entry's parent FID
2745  * \param[out]  ln         - Entry name destination buffer
2746  *
2747  * \retval 0 on success
2748  * \retval Appropriate negative error code on failure
2749  */
2750 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2751                             struct lu_fid *parent_fid, struct lu_name *ln)
2752 {
2753         unsigned int    idx;
2754         int             rc;
2755         ENTRY;
2756
2757         rc = linkea_init(ldata);
2758         if (rc < 0)
2759                 RETURN(rc);
2760
2761         if (linkno >= ldata->ld_leh->leh_reccount)
2762                 /* beyond last link */
2763                 RETURN(-ENODATA);
2764
2765         linkea_first_entry(ldata);
2766         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2767                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2768                                     parent_fid);
2769                 if (idx == linkno)
2770                         break;
2771
2772                 linkea_next_entry(ldata);
2773         }
2774
2775         if (idx < linkno)
2776                 RETURN(-ENODATA);
2777
2778         RETURN(0);
2779 }
2780
2781 /**
2782  * Get parent FID and name of an identified link. Operation is performed for
2783  * a given link number, letting the caller iterate over linkno to list one or
2784  * all links of an entry.
2785  *
2786  * \param[in]     file - File descriptor against which to perform the operation
2787  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2788  *                       on and the available size. It is eventually filled with
2789  *                       the requested information or left untouched on error
2790  *
2791  * \retval - 0 on success
2792  * \retval - Appropriate negative error code on failure
2793  */
2794 int ll_getparent(struct file *file, struct getparent __user *arg)
2795 {
2796         struct dentry           *dentry = file->f_dentry;
2797         struct inode            *inode = file->f_dentry->d_inode;
2798         struct linkea_data      *ldata;
2799         struct lu_buf            buf = LU_BUF_NULL;
2800         struct lu_name           ln;
2801         struct lu_fid            parent_fid;
2802         __u32                    linkno;
2803         __u32                    name_size;
2804         int                      rc;
2805
2806         ENTRY;
2807
2808         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2809             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2810                 RETURN(-EPERM);
2811
2812         if (get_user(name_size, &arg->gp_name_size))
2813                 RETURN(-EFAULT);
2814
2815         if (get_user(linkno, &arg->gp_linkno))
2816                 RETURN(-EFAULT);
2817
2818         if (name_size > PATH_MAX)
2819                 RETURN(-EINVAL);
2820
2821         OBD_ALLOC(ldata, sizeof(*ldata));
2822         if (ldata == NULL)
2823                 RETURN(-ENOMEM);
2824
2825         rc = linkea_data_new(ldata, &buf);
2826         if (rc < 0)
2827                 GOTO(ldata_free, rc);
2828
2829         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2830         if (rc < 0)
2831                 GOTO(lb_free, rc);
2832
2833         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2834         if (rc < 0)
2835                 GOTO(lb_free, rc);
2836
2837         if (ln.ln_namelen >= name_size)
2838                 GOTO(lb_free, rc = -EOVERFLOW);
2839
2840         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2841                 GOTO(lb_free, rc = -EFAULT);
2842
2843         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2844                 GOTO(lb_free, rc = -EFAULT);
2845
2846         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2847                 GOTO(lb_free, rc = -EFAULT);
2848
2849 lb_free:
2850         lu_buf_free(&buf);
2851 ldata_free:
2852         OBD_FREE(ldata, sizeof(*ldata));
2853
2854         RETURN(rc);
2855 }