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