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, NULL);
 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, NULL);
 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, NULL);
 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 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 696 {
 697         int size, rc;
 698
 699         size = sizeof(int);
 700         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 701                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 702         if (rc)
 703                 CERROR("Get max cookiesize error rc %d\n", rc);
 704
 705         RETURN(rc);
 706 }
 707
 708 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 709 {
 710         int size, rc;
 711
 712         size = sizeof(int);
 713         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 714                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 715         if (rc)
 716                 CERROR("Get default cookiesize error rc %d\n", rc);
 717
 718         RETURN(rc);
 719 }
 720
 721 static void ll_dump_inode(struct inode *inode)
 722 {
 723         struct ll_d_hlist_node *tmp;
 724         int dentry_count = 0;
 725
 726         LASSERT(inode != NULL);
 727
 728         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 729                 dentry_count++;
 730
 731         CERROR("%s: inode %p dump: dev=%s fid="DFID
 732                " mode=%o count=%u, %d dentries\n",
 733                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 734                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 735                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 736 }
 737
 738 void lustre_dump_dentry(struct dentry *dentry, int recur)
 739 {
 740         struct list_head *tmp;
 741         int subdirs = 0;
 742
 743         LASSERT(dentry != NULL);
 744
 745         list_for_each(tmp, &dentry->d_subdirs)
 746                 subdirs++;
 747
 748         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 749                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 750                dentry->d_name.len, dentry->d_name.name,
 751                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 752                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 753                dentry->d_flags, dentry->d_fsdata, subdirs);
 754         if (dentry->d_inode != NULL)
 755                 ll_dump_inode(dentry->d_inode);
 756
 757         if (recur == 0)
 758                 return;
 759
 760         list_for_each(tmp, &dentry->d_subdirs) {
 761                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 762                 lustre_dump_dentry(d, recur - 1);
 763         }
 764 }
 765
 766 static void client_common_put_super(struct super_block *sb)
 767 {
 768         struct ll_sb_info *sbi = ll_s2sbi(sb);
 769         ENTRY;
 770
 771 #ifdef CONFIG_FS_POSIX_ACL
 772         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 773                 et_fini(&sbi->ll_et);
 774                 rct_fini(&sbi->ll_rct);
 775         }
 776 #endif
 777
 778         cl_sb_fini(sb);
 779
 780         list_del(&sbi->ll_conn_chain);
 781
 782         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 783         obd_disconnect(sbi->ll_dt_exp);
 784         sbi->ll_dt_exp = NULL;
 785
 786         lprocfs_unregister_mountpoint(sbi);
 787
 788         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 789         obd_disconnect(sbi->ll_md_exp);
 790         sbi->ll_md_exp = NULL;
 791
 792         EXIT;
 793 }
 794
 795 void ll_kill_super(struct super_block *sb)
 796 {
 797         struct ll_sb_info *sbi;
 798         ENTRY;
 799
 800         /* not init sb ?*/
 801         if (!(sb->s_flags & MS_ACTIVE))
 802                 return;
 803
 804         sbi = ll_s2sbi(sb);
 805         /* we need restore s_dev from changed for clustred NFS before put_super
 806          * because new kernels have cached s_dev and change sb->s_dev in
 807          * put_super not affected real removing devices */
 808         if (sbi) {
 809                 sb->s_dev = sbi->ll_sdev_orig;
 810                 sbi->ll_umounting = 1;
 811
 812                 /* wait running statahead threads to quit */
 813                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 814                         set_current_state(TASK_UNINTERRUPTIBLE);
 815                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 816                 }
 817         }
 818
 819         EXIT;
 820 }
 821
 822 static inline int ll_set_opt(const char *opt, char *data, int fl)
 823 {
 824         if (strncmp(opt, data, strlen(opt)) != 0)
 825                 return(0);
 826         else
 827                 return(fl);
 828 }
 829
 830 /* non-client-specific mount options are parsed in lmd_parse */
 831 static int ll_options(char *options, int *flags)
 832 {
 833         int tmp;
 834         char *s1 = options, *s2;
 835         ENTRY;
 836
 837         if (!options)
 838                 RETURN(0);
 839
 840         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 841
 842         while (*s1) {
 843                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 844                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 845                 if (tmp) {
 846                         *flags |= tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 855                 if (tmp) {
 856                         *flags |= tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 860                 if (tmp) {
 861                         *flags &= ~tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 865                 if (tmp) {
 866                         *flags |= tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 870                 if (tmp) {
 871                         *flags &= ~tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 875                 if (tmp) {
 876                         *flags |= tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889
 890                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 891                 if (tmp) {
 892                         *flags |= tmp;
 893                         goto next;
 894                 }
 895                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 896                 if (tmp) {
 897                         *flags &= ~tmp;
 898                         goto next;
 899                 }
 900                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 901                 if (tmp) {
 902                         *flags |= tmp;
 903                         goto next;
 904                 }
 905                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 906                 if (tmp) {
 907                         *flags &= ~tmp;
 908                         goto next;
 909                 }
 910                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 911                 if (tmp) {
 912                         *flags |= tmp;
 913                         goto next;
 914                 }
 915                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 916                 if (tmp) {
 917                         *flags &= ~tmp;
 918                         goto next;
 919                 }
 920                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 921                 if (tmp) {
 922                         *flags |= tmp;
 923                         goto next;
 924                 }
 925                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 926                 if (tmp) {
 927                         *flags |= tmp;
 928                         goto next;
 929                 }
 930                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 931                 if (tmp) {
 932                         *flags &= ~tmp;
 933                         goto next;
 934                 }
 935                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 936                                    s1);
 937                 RETURN(-EINVAL);
 938
 939 next:
 940                 /* Find next opt */
 941                 s2 = strchr(s1, ',');
 942                 if (s2 == NULL)
 943                         break;
 944                 s1 = s2 + 1;
 945         }
 946         RETURN(0);
 947 }
 948
 949 void ll_lli_init(struct ll_inode_info *lli)
 950 {
 951         lli->lli_inode_magic = LLI_INODE_MAGIC;
 952         lli->lli_flags = 0;
 953         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 954         spin_lock_init(&lli->lli_lock);
 955         lli->lli_posix_acl = NULL;
 956         lli->lli_remote_perms = NULL;
 957         mutex_init(&lli->lli_rmtperm_mutex);
 958         /* Do not set lli_fid, it has been initialized already. */
 959         fid_zero(&lli->lli_pfid);
 960         INIT_LIST_HEAD(&lli->lli_oss_capas);
 961         atomic_set(&lli->lli_open_count, 0);
 962         lli->lli_mds_capa = NULL;
 963         lli->lli_rmtperm_time = 0;
 964         lli->lli_mds_read_och = NULL;
 965         lli->lli_mds_write_och = NULL;
 966         lli->lli_mds_exec_och = NULL;
 967         lli->lli_open_fd_read_count = 0;
 968         lli->lli_open_fd_write_count = 0;
 969         lli->lli_open_fd_exec_count = 0;
 970         mutex_init(&lli->lli_och_mutex);
 971         spin_lock_init(&lli->lli_agl_lock);
 972         spin_lock_init(&lli->lli_layout_lock);
 973         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 974         lli->lli_clob = NULL;
 975
 976         init_rwsem(&lli->lli_xattrs_list_rwsem);
 977         mutex_init(&lli->lli_xattrs_enq_lock);
 978
 979         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 980         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 981                 mutex_init(&lli->lli_readdir_mutex);
 982                 lli->lli_opendir_key = NULL;
 983                 lli->lli_sai = NULL;
 984                 spin_lock_init(&lli->lli_sa_lock);
 985                 lli->lli_opendir_pid = 0;
 986                 lli->lli_sa_enabled = 0;
 987                 lli->lli_def_stripe_offset = -1;
 988         } else {
 989                 mutex_init(&lli->lli_size_mutex);
 990                 lli->lli_symlink_name = NULL;
 991                 init_rwsem(&lli->lli_trunc_sem);
 992                 range_lock_tree_init(&lli->lli_write_tree);
 993                 init_rwsem(&lli->lli_glimpse_sem);
 994                 lli->lli_glimpse_time = 0;
 995                 INIT_LIST_HEAD(&lli->lli_agl_list);
 996                 lli->lli_agl_index = 0;
 997                 lli->lli_async_rc = 0;
 998         }
 999         mutex_init(&lli->lli_layout_mutex);
1000 }
1001
1002 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1003 {
1004         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1005
1006         bdi->name = "lustre";
1007         return bdi_register(bdi, NULL, "lustre-%d",
1008                             atomic_inc_return(&ll_bdi_num));
1009 }
1010
1011 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1012 {
1013         struct lustre_profile *lprof = NULL;
1014         struct lustre_sb_info *lsi = s2lsi(sb);
1015         struct ll_sb_info *sbi;
1016         char  *dt = NULL, *md = NULL;
1017         char  *profilenm = get_profile_name(sb);
1018         struct config_llog_instance *cfg;
1019         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1020         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1021         int    err;
1022         ENTRY;
1023
1024         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1025
1026         OBD_ALLOC_PTR(cfg);
1027         if (cfg == NULL)
1028                 RETURN(-ENOMEM);
1029
1030         try_module_get(THIS_MODULE);
1031
1032         /* client additional sb info */
1033         lsi->lsi_llsbi = sbi = ll_init_sbi();
1034         if (!sbi) {
1035                 module_put(THIS_MODULE);
1036                 OBD_FREE_PTR(cfg);
1037                 RETURN(-ENOMEM);
1038         }
1039
1040         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1041         if (err)
1042                 GOTO(out_free, err);
1043
1044         err = bdi_init(&lsi->lsi_bdi);
1045         if (err)
1046                 GOTO(out_free, err);
1047         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1048         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1049         err = ll_bdi_register(&lsi->lsi_bdi);
1050         if (err)
1051                 GOTO(out_free, err);
1052
1053         sb->s_bdi = &lsi->lsi_bdi;
1054 #ifndef HAVE_DCACHE_LOCK
1055         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1056         sb->s_d_op = &ll_d_ops;
1057 #endif
1058
1059         /* Generate a string unique to this super, in case some joker tries
1060            to mount the same fs at two mount points.
1061            Use the address of the super itself.*/
1062         cfg->cfg_instance = sb;
1063         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1064         cfg->cfg_callback = class_config_llog_handler;
1065         /* set up client obds */
1066         err = lustre_process_log(sb, profilenm, cfg);
1067         if (err < 0)
1068                 GOTO(out_free, err);
1069
1070         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1071         lprof = class_get_profile(profilenm);
1072         if (lprof == NULL) {
1073                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1074                                    " read from the MGS.  Does that filesystem "
1075                                    "exist?\n", profilenm);
1076                 GOTO(out_free, err = -EINVAL);
1077         }
1078         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1079                lprof->lp_md, lprof->lp_dt);
1080
1081         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1082         if (!dt)
1083                 GOTO(out_free, err = -ENOMEM);
1084         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1085
1086         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1087         if (!md)
1088                 GOTO(out_free, err = -ENOMEM);
1089         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1090
1091         /* connections, registrations, sb setup */
1092         err = client_common_fill_super(sb, md, dt, mnt);
1093         if (err < 0)
1094                 GOTO(out_free, err);
1095
1096         sbi->ll_client_common_fill_super_succeeded = 1;
1097
1098 out_free:
1099         if (md)
1100                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1101         if (dt)
1102                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1103         if (err)
1104                 ll_put_super(sb);
1105         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1106                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1107
1108         OBD_FREE_PTR(cfg);
1109         RETURN(err);
1110 } /* ll_fill_super */
1111
1112 void ll_put_super(struct super_block *sb)
1113 {
1114         struct config_llog_instance cfg, params_cfg;
1115         struct obd_device *obd;
1116         struct lustre_sb_info *lsi = s2lsi(sb);
1117         struct ll_sb_info *sbi = ll_s2sbi(sb);
1118         char *profilenm = get_profile_name(sb);
1119         long ccc_count;
1120         int next, force = 1, rc = 0;
1121         ENTRY;
1122
1123         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1124
1125         ll_print_capa_stat(sbi);
1126
1127         cfg.cfg_instance = sb;
1128         lustre_end_log(sb, profilenm, &cfg);
1129
1130         params_cfg.cfg_instance = sb;
1131         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1132
1133         if (sbi->ll_md_exp) {
1134                 obd = class_exp2obd(sbi->ll_md_exp);
1135                 if (obd)
1136                         force = obd->obd_force;
1137         }
1138
1139         /* Wait for unstable pages to be committed to stable storage */
1140         if (force == 0) {
1141                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1142                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1143                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1144                         &lwi);
1145         }
1146
1147         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1148         if (force == 0 && rc != -EINTR)
1149                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1150
1151
1152         /* We need to set force before the lov_disconnect in
1153            lustre_common_put_super, since l_d cleans up osc's as well. */
1154         if (force) {
1155                 next = 0;
1156                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1157                                                      &next)) != NULL) {
1158                         obd->obd_force = force;
1159                 }
1160         }
1161
1162         if (sbi->ll_client_common_fill_super_succeeded) {
1163                 /* Only if client_common_fill_super succeeded */
1164                 client_common_put_super(sb);
1165         }
1166
1167         next = 0;
1168         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1169                 class_manual_cleanup(obd);
1170         }
1171
1172         if (sbi->ll_flags & LL_SBI_VERBOSE)
1173                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1174
1175         if (profilenm)
1176                 class_del_profile(profilenm);
1177
1178         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1179                 bdi_destroy(&lsi->lsi_bdi);
1180                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1181         }
1182
1183         ll_free_sbi(sb);
1184         lsi->lsi_llsbi = NULL;
1185
1186         lustre_common_put_super(sb);
1187
1188         cl_env_cache_purge(~0);
1189
1190         module_put(THIS_MODULE);
1191
1192         EXIT;
1193 } /* client_put_super */
1194
1195 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1196 {
1197         struct inode *inode = NULL;
1198
1199         /* NOTE: we depend on atomic igrab() -bzzz */
1200         lock_res_and_lock(lock);
1201         if (lock->l_resource->lr_lvb_inode) {
1202                 struct ll_inode_info * lli;
1203                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1204                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1205                         inode = igrab(lock->l_resource->lr_lvb_inode);
1206                 } else {
1207                         inode = lock->l_resource->lr_lvb_inode;
1208                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1209                                          D_WARNING, lock, "lr_lvb_inode %p is "
1210                                          "bogus: magic %08x",
1211                                          lock->l_resource->lr_lvb_inode,
1212                                          lli->lli_inode_magic);
1213                         inode = NULL;
1214                 }
1215         }
1216         unlock_res_and_lock(lock);
1217         return inode;
1218 }
1219
1220 static void ll_dir_clear_lsm_md(struct inode *inode)
1221 {
1222         struct ll_inode_info *lli = ll_i2info(inode);
1223
1224         LASSERT(S_ISDIR(inode->i_mode));
1225
1226         if (lli->lli_lsm_md != NULL) {
1227                 lmv_free_memmd(lli->lli_lsm_md);
1228                 lli->lli_lsm_md = NULL;
1229         }
1230 }
1231
1232 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1233                                       const struct lu_fid *fid,
1234                                       struct lustre_md *md)
1235 {
1236         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1237         struct mdt_body         *body = md->body;
1238         struct inode            *inode;
1239         ino_t                   ino;
1240         ENTRY;
1241
1242         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1243         inode = iget_locked(sb, ino);
1244         if (inode == NULL) {
1245                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1246                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1247                 RETURN(ERR_PTR(-ENOENT));
1248         }
1249
1250         if (inode->i_state & I_NEW) {
1251                 struct ll_inode_info *lli = ll_i2info(inode);
1252                 struct lmv_stripe_md *lsm = md->lmv;
1253
1254                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1255                                 (body->mbo_mode & S_IFMT);
1256                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1257                          PFID(fid));
1258
1259                 LTIME_S(inode->i_mtime) = 0;
1260                 LTIME_S(inode->i_atime) = 0;
1261                 LTIME_S(inode->i_ctime) = 0;
1262                 inode->i_rdev = 0;
1263
1264                 /* initializing backing dev info. */
1265                 inode->i_mapping->backing_dev_info =
1266                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1267                 inode->i_op = &ll_dir_inode_operations;
1268                 inode->i_fop = &ll_dir_operations;
1269                 lli->lli_fid = *fid;
1270                 ll_lli_init(lli);
1271
1272                 LASSERT(lsm != NULL);
1273                 /* master object FID */
1274                 lli->lli_pfid = body->mbo_fid1;
1275                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1276                        lli, PFID(fid), PFID(&lli->lli_pfid));
1277                 unlock_new_inode(inode);
1278         }
1279
1280         RETURN(inode);
1281 }
1282
1283 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1284 {
1285         struct lu_fid *fid;
1286         struct lmv_stripe_md *lsm = md->lmv;
1287         int i;
1288
1289         LASSERT(lsm != NULL);
1290         /* XXX sigh, this lsm_root initialization should be in
1291          * LMV layer, but it needs ll_iget right now, so we
1292          * put this here right now. */
1293         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1294                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1295                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1296                 /* Unfortunately ll_iget will call ll_update_inode,
1297                  * where the initialization of slave inode is slightly
1298                  * different, so it reset lsm_md to NULL to avoid
1299                  * initializing lsm for slave inode. */
1300                 /* For migrating inode, master stripe and master object will
1301                  * be same, so we only need assign this inode */
1302                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1303                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1304                 else
1305                         lsm->lsm_md_oinfo[i].lmo_root =
1306                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1307
1308                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1309                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1310
1311                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1312                         return rc;
1313                 }
1314         }
1315
1316         return 0;
1317 }
1318
1319 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1320                                 const struct lmv_stripe_md *lsm_md2)
1321 {
1322         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1323                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1324                lsm_md1->lsm_md_master_mdt_index ==
1325                                         lsm_md2->lsm_md_master_mdt_index &&
1326                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1327                lsm_md1->lsm_md_layout_version ==
1328                                         lsm_md2->lsm_md_layout_version &&
1329                strcmp(lsm_md1->lsm_md_pool_name,
1330                       lsm_md2->lsm_md_pool_name) == 0;
1331 }
1332
1333 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1334 {
1335         struct ll_inode_info *lli = ll_i2info(inode);
1336         struct lmv_stripe_md *lsm = md->lmv;
1337         int     rc;
1338         ENTRY;
1339
1340         LASSERT(S_ISDIR(inode->i_mode));
1341         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1342                PFID(ll_inode2fid(inode)));
1343
1344         /* no striped information from request. */
1345         if (lsm == NULL) {
1346                 if (lli->lli_lsm_md == NULL) {
1347                         RETURN(0);
1348                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1349                                                 LMV_HASH_FLAG_MIGRATION) {
1350                         /* migration is done, the temporay MIGRATE layout has
1351                          * been removed */
1352                         CDEBUG(D_INODE, DFID" finish migration.\n",
1353                                PFID(ll_inode2fid(inode)));
1354                         lmv_free_memmd(lli->lli_lsm_md);
1355                         lli->lli_lsm_md = NULL;
1356                         RETURN(0);
1357                 } else {
1358                         /* The lustre_md from req does not include stripeEA,
1359                          * see ll_md_setattr */
1360                         RETURN(0);
1361                 }
1362         }
1363
1364         /* set the directory layout */
1365         if (lli->lli_lsm_md == NULL) {
1366
1367                 rc = ll_init_lsm_md(inode, md);
1368                 if (rc != 0)
1369                         RETURN(rc);
1370
1371                 lli->lli_lsm_md = lsm;
1372                 /* set lsm_md to NULL, so the following free lustre_md
1373                  * will not free this lsm */
1374                 md->lmv = NULL;
1375                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1376                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1377                 RETURN(0);
1378         }
1379
1380         /* Compare the old and new stripe information */
1381         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1382                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1383                 int                     idx;
1384
1385                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1386                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1387                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1388                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1389                        inode, lsm, old_lsm,
1390                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1391                        lsm->lsm_md_stripe_count,
1392                        old_lsm->lsm_md_stripe_count,
1393                        lsm->lsm_md_master_mdt_index,
1394                        old_lsm->lsm_md_master_mdt_index,
1395                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1396                        lsm->lsm_md_layout_version,
1397                        old_lsm->lsm_md_layout_version,
1398                        lsm->lsm_md_pool_name,
1399                        old_lsm->lsm_md_pool_name);
1400
1401                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1402                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1403                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1404                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1405                 }
1406
1407                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1408                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1409                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1410                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1411                 }
1412
1413                 RETURN(-EIO);
1414         }
1415
1416         RETURN(0);
1417 }
1418
1419 void ll_clear_inode(struct inode *inode)
1420 {
1421         struct ll_inode_info *lli = ll_i2info(inode);
1422         struct ll_sb_info *sbi = ll_i2sbi(inode);
1423         ENTRY;
1424
1425         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1426                PFID(ll_inode2fid(inode)), inode);
1427
1428         if (S_ISDIR(inode->i_mode)) {
1429                 /* these should have been cleared in ll_file_release */
1430                 LASSERT(lli->lli_opendir_key == NULL);
1431                 LASSERT(lli->lli_sai == NULL);
1432                 LASSERT(lli->lli_opendir_pid == 0);
1433         }
1434
1435         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1436
1437         LASSERT(!lli->lli_open_fd_write_count);
1438         LASSERT(!lli->lli_open_fd_read_count);
1439         LASSERT(!lli->lli_open_fd_exec_count);
1440
1441         if (lli->lli_mds_write_och)
1442                 ll_md_real_close(inode, FMODE_WRITE);
1443         if (lli->lli_mds_exec_och)
1444                 ll_md_real_close(inode, FMODE_EXEC);
1445         if (lli->lli_mds_read_och)
1446                 ll_md_real_close(inode, FMODE_READ);
1447
1448         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1449                 OBD_FREE(lli->lli_symlink_name,
1450                          strlen(lli->lli_symlink_name) + 1);
1451                 lli->lli_symlink_name = NULL;
1452         }
1453
1454         ll_xattr_cache_destroy(inode);
1455
1456         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1457                 LASSERT(lli->lli_posix_acl == NULL);
1458                 if (lli->lli_remote_perms) {
1459                         free_rmtperm_hash(lli->lli_remote_perms);
1460                         lli->lli_remote_perms = NULL;
1461                 }
1462         }
1463 #ifdef CONFIG_FS_POSIX_ACL
1464         else if (lli->lli_posix_acl) {
1465                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1466                 LASSERT(lli->lli_remote_perms == NULL);
1467                 posix_acl_release(lli->lli_posix_acl);
1468                 lli->lli_posix_acl = NULL;
1469         }
1470 #endif
1471         lli->lli_inode_magic = LLI_INODE_DEAD;
1472
1473         ll_clear_inode_capas(inode);
1474         if (S_ISDIR(inode->i_mode))
1475                 ll_dir_clear_lsm_md(inode);
1476         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1477                 LASSERT(list_empty(&lli->lli_agl_list));
1478
1479         /*
1480          * XXX This has to be done before lsm is freed below, because
1481          * cl_object still uses inode lsm.
1482          */
1483         cl_inode_fini(inode);
1484
1485         EXIT;
1486 }
1487
1488 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1489 {
1490         struct lustre_md md;
1491         struct inode *inode = dentry->d_inode;
1492         struct ll_sb_info *sbi = ll_i2sbi(inode);
1493         struct ptlrpc_request *request = NULL;
1494         int rc, ia_valid;
1495         ENTRY;
1496
1497         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1498                                      LUSTRE_OPC_ANY, NULL);
1499         if (IS_ERR(op_data))
1500                 RETURN(PTR_ERR(op_data));
1501
1502         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1503         if (rc) {
1504                 ptlrpc_req_finished(request);
1505                 if (rc == -ENOENT) {
1506                         clear_nlink(inode);
1507                         /* Unlinked special device node? Or just a race?
1508                          * Pretend we done everything. */
1509                         if (!S_ISREG(inode->i_mode) &&
1510                             !S_ISDIR(inode->i_mode)) {
1511                                 ia_valid = op_data->op_attr.ia_valid;
1512                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1513                                 rc = simple_setattr(dentry, &op_data->op_attr);
1514                                 op_data->op_attr.ia_valid = ia_valid;
1515                         }
1516                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1517                         CERROR("md_setattr fails: rc = %d\n", rc);
1518                 }
1519                 RETURN(rc);
1520         }
1521
1522         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1523                               sbi->ll_md_exp, &md);
1524         if (rc) {
1525                 ptlrpc_req_finished(request);
1526                 RETURN(rc);
1527         }
1528
1529         ia_valid = op_data->op_attr.ia_valid;
1530         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1531          * cache is not cleared yet. */
1532         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1533         rc = simple_setattr(dentry, &op_data->op_attr);
1534         op_data->op_attr.ia_valid = ia_valid;
1535
1536         rc = ll_update_inode(inode, &md);
1537         ptlrpc_req_finished(request);
1538
1539         RETURN(rc);
1540 }
1541
1542 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1543 {
1544         struct obd_capa *capa;
1545         int rc;
1546
1547         if (attr->ia_valid & ATTR_SIZE)
1548                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1549         else
1550                 capa = ll_mdscapa_get(inode);
1551
1552         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1553
1554         if (attr->ia_valid & ATTR_SIZE)
1555                 ll_truncate_free_capa(capa);
1556         else
1557                 capa_put(capa);
1558
1559         return rc;
1560 }
1561
1562 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1563  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1564  * keep these values until such a time that objects are allocated for it.
1565  * We do the MDS operations first, as it is checking permissions for us.
1566  * We don't to the MDS RPC if there is nothing that we want to store there,
1567  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1568  * going to do an RPC anyways.
1569  *
1570  * If we are doing a truncate, we will send the mtime and ctime updates
1571  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1572  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1573  * at the same time.
1574  *
1575  * In case of HSMimport, we only set attr on MDS.
1576  */
1577 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1578 {
1579         struct inode *inode = dentry->d_inode;
1580         struct ll_inode_info *lli = ll_i2info(inode);
1581         struct md_op_data *op_data = NULL;
1582         bool file_is_released = false;
1583         int rc = 0;
1584         ENTRY;
1585
1586         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1587                "valid %x, hsm_import %d\n",
1588                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1589                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1590                hsm_import);
1591
1592         if (attr->ia_valid & ATTR_SIZE) {
1593                 /* Check new size against VFS/VM file size limit and rlimit */
1594                 rc = inode_newsize_ok(inode, attr->ia_size);
1595                 if (rc)
1596                         RETURN(rc);
1597
1598                 /* The maximum Lustre file size is variable, based on the
1599                  * OST maximum object size and number of stripes.  This
1600                  * needs another check in addition to the VFS check above. */
1601                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1602                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1603                                PFID(&lli->lli_fid), attr->ia_size,
1604                                ll_file_maxbytes(inode));
1605                         RETURN(-EFBIG);
1606                 }
1607
1608                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1609         }
1610
1611         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1612         if (attr->ia_valid & TIMES_SET_FLAGS) {
1613                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1614                     !cfs_capable(CFS_CAP_FOWNER))
1615                         RETURN(-EPERM);
1616         }
1617
1618         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1619         if (attr->ia_valid & ATTR_CTIME) {
1620                 attr->ia_ctime = CFS_CURRENT_TIME;
1621                 attr->ia_valid |= ATTR_CTIME_SET;
1622         }
1623         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1624             (attr->ia_valid & ATTR_ATIME)) {
1625                 attr->ia_atime = CFS_CURRENT_TIME;
1626                 attr->ia_valid |= ATTR_ATIME_SET;
1627         }
1628         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1629             (attr->ia_valid & ATTR_MTIME)) {
1630                 attr->ia_mtime = CFS_CURRENT_TIME;
1631                 attr->ia_valid |= ATTR_MTIME_SET;
1632         }
1633
1634         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1635                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1636                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1637                        cfs_time_current_sec());
1638
1639         /* We always do an MDS RPC, even if we're only changing the size;
1640          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1641
1642         OBD_ALLOC_PTR(op_data);
1643         if (op_data == NULL)
1644                 RETURN(-ENOMEM);
1645
1646         if (!S_ISDIR(inode->i_mode)) {
1647                 if (attr->ia_valid & ATTR_SIZE)
1648                         inode_dio_write_done(inode);
1649                 mutex_unlock(&inode->i_mutex);
1650         }
1651
1652         /* truncate on a released file must failed with -ENODATA,
1653          * so size must not be set on MDS for released file
1654          * but other attributes must be set
1655          */
1656         if (S_ISREG(inode->i_mode)) {
1657                 struct cl_layout cl = {
1658                         .cl_is_released = false,
1659                 };
1660                 struct lu_env *env;
1661                 int refcheck;
1662                 __u32 gen;
1663
1664                 rc = ll_layout_refresh(inode, &gen);
1665                 if (rc < 0)
1666                         GOTO(out, rc);
1667
1668                 /* XXX: the only place we need to know the layout type,
1669                  * this will be removed by a later patch. -Jinshan */
1670                 env = cl_env_get(&refcheck);
1671                 if (IS_ERR(env))
1672                         GOTO(out, rc = PTR_ERR(env));
1673
1674                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1675                 cl_env_put(env, &refcheck);
1676                 if (rc < 0)
1677                         GOTO(out, rc);
1678
1679                 file_is_released = cl.cl_is_released;
1680
1681                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1682                         if (file_is_released) {
1683                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1684                                 if (rc < 0)
1685                                         GOTO(out, rc);
1686
1687                                 file_is_released = false;
1688                                 ll_layout_refresh(inode, &gen);
1689                         }
1690
1691                         /* If we are changing file size, file content is
1692                          * modified, flag it. */
1693                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1694                         spin_lock(&lli->lli_lock);
1695                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1696                         spin_unlock(&lli->lli_lock);
1697                         op_data->op_bias |= MDS_DATA_MODIFIED;
1698                 }
1699         }
1700
1701         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1702
1703         rc = ll_md_setattr(dentry, op_data);
1704         if (rc)
1705                 GOTO(out, rc);
1706
1707         /* RPC to MDT is sent, cancel data modification flag */
1708         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1709                 spin_lock(&lli->lli_lock);
1710                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1711                 spin_unlock(&lli->lli_lock);
1712         }
1713
1714         if (!S_ISREG(inode->i_mode) || file_is_released)
1715                 GOTO(out, rc = 0);
1716
1717         if (attr->ia_valid & (ATTR_SIZE |
1718                               ATTR_ATIME | ATTR_ATIME_SET |
1719                               ATTR_MTIME | ATTR_MTIME_SET)) {
1720                 /* For truncate and utimes sending attributes to OSTs, setting
1721                  * mtime/atime to the past will be performed under PW [0:EOF]
1722                  * extent lock (new_size:EOF for truncate).  It may seem
1723                  * excessive to send mtime/atime updates to OSTs when not
1724                  * setting times to past, but it is necessary due to possible
1725                  * time de-synchronization between MDT inode and OST objects */
1726                 if (attr->ia_valid & ATTR_SIZE)
1727                         down_write(&lli->lli_trunc_sem);
1728                 rc = ll_setattr_ost(inode, attr);
1729                 if (attr->ia_valid & ATTR_SIZE)
1730                         up_write(&lli->lli_trunc_sem);
1731         }
1732         EXIT;
1733 out:
1734         if (op_data != NULL)
1735                 ll_finish_md_op_data(op_data);
1736
1737         if (!S_ISDIR(inode->i_mode)) {
1738                 mutex_lock(&inode->i_mutex);
1739                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1740                         inode_dio_wait(inode);
1741         }
1742
1743         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1744                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1745
1746         return rc;
1747 }
1748
1749 int ll_setattr(struct dentry *de, struct iattr *attr)
1750 {
1751         int mode = de->d_inode->i_mode;
1752
1753         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1754                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1755                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1756
1757         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1758                                (ATTR_SIZE|ATTR_MODE)) &&
1759             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1760              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1761               !(attr->ia_mode & S_ISGID))))
1762                 attr->ia_valid |= ATTR_FORCE;
1763
1764         if ((attr->ia_valid & ATTR_MODE) &&
1765             (mode & S_ISUID) &&
1766             !(attr->ia_mode & S_ISUID) &&
1767             !(attr->ia_valid & ATTR_KILL_SUID))
1768                 attr->ia_valid |= ATTR_KILL_SUID;
1769
1770         if ((attr->ia_valid & ATTR_MODE) &&
1771             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1772             !(attr->ia_mode & S_ISGID) &&
1773             !(attr->ia_valid & ATTR_KILL_SGID))
1774                 attr->ia_valid |= ATTR_KILL_SGID;
1775
1776         return ll_setattr_raw(de, attr, false);
1777 }
1778
1779 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1780                        __u64 max_age, __u32 flags)
1781 {
1782         struct ll_sb_info *sbi = ll_s2sbi(sb);
1783         struct obd_statfs obd_osfs;
1784         int rc;
1785         ENTRY;
1786
1787         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1788         if (rc) {
1789                 CERROR("md_statfs fails: rc = %d\n", rc);
1790                 RETURN(rc);
1791         }
1792
1793         osfs->os_type = sb->s_magic;
1794
1795         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1796                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1797
1798         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1799                 flags |= OBD_STATFS_NODELAY;
1800
1801         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1802         if (rc) {
1803                 CERROR("obd_statfs fails: rc = %d\n", rc);
1804                 RETURN(rc);
1805         }
1806
1807         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1808                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1809                obd_osfs.os_files);
1810
1811         osfs->os_bsize = obd_osfs.os_bsize;
1812         osfs->os_blocks = obd_osfs.os_blocks;
1813         osfs->os_bfree = obd_osfs.os_bfree;
1814         osfs->os_bavail = obd_osfs.os_bavail;
1815
1816         /* If we don't have as many objects free on the OST as inodes
1817          * on the MDS, we reduce the total number of inodes to
1818          * compensate, so that the "inodes in use" number is correct.
1819          */
1820         if (obd_osfs.os_ffree < osfs->os_ffree) {
1821                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1822                         obd_osfs.os_ffree;
1823                 osfs->os_ffree = obd_osfs.os_ffree;
1824         }
1825
1826         RETURN(rc);
1827 }
1828 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1829 {
1830         struct super_block *sb = de->d_sb;
1831         struct obd_statfs osfs;
1832         __u64 fsid = huge_encode_dev(sb->s_dev);
1833         int rc;
1834
1835         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1836         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1837
1838         /* Some amount of caching on the client is allowed */
1839         rc = ll_statfs_internal(sb, &osfs,
1840                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1841                                 0);
1842         if (rc)
1843                 return rc;
1844
1845         statfs_unpack(sfs, &osfs);
1846
1847         /* We need to downshift for all 32-bit kernels, because we can't
1848          * tell if the kernel is being called via sys_statfs64() or not.
1849          * Stop before overflowing f_bsize - in which case it is better
1850          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1851         if (sizeof(long) < 8) {
1852                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1853                         sfs->f_bsize <<= 1;
1854
1855                         osfs.os_blocks >>= 1;
1856                         osfs.os_bfree >>= 1;
1857                         osfs.os_bavail >>= 1;
1858                 }
1859         }
1860
1861         sfs->f_blocks = osfs.os_blocks;
1862         sfs->f_bfree = osfs.os_bfree;
1863         sfs->f_bavail = osfs.os_bavail;
1864         sfs->f_fsid.val[0] = (__u32)fsid;
1865         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1866         return 0;
1867 }
1868
1869 void ll_inode_size_lock(struct inode *inode)
1870 {
1871         struct ll_inode_info *lli;
1872
1873         LASSERT(!S_ISDIR(inode->i_mode));
1874
1875         lli = ll_i2info(inode);
1876         mutex_lock(&lli->lli_size_mutex);
1877 }
1878
1879 void ll_inode_size_unlock(struct inode *inode)
1880 {
1881         struct ll_inode_info *lli;
1882
1883         lli = ll_i2info(inode);
1884         mutex_unlock(&lli->lli_size_mutex);
1885 }
1886
1887 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1888 {
1889         struct ll_inode_info *lli = ll_i2info(inode);
1890         struct mdt_body *body = md->body;
1891         struct lov_stripe_md *lsm = md->lsm;
1892         struct ll_sb_info *sbi = ll_i2sbi(inode);
1893
1894         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1895         if (lsm != NULL) {
1896                 cl_file_inode_init(inode, md);
1897
1898                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1899                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1900                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1901         }
1902
1903         if (S_ISDIR(inode->i_mode)) {
1904                 int     rc;
1905
1906                 rc = ll_update_lsm_md(inode, md);
1907                 if (rc != 0)
1908                         return rc;
1909         }
1910
1911         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1912                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1913                         ll_update_remote_perm(inode, md->remote_perm);
1914         }
1915 #ifdef CONFIG_FS_POSIX_ACL
1916         else if (body->mbo_valid & OBD_MD_FLACL) {
1917                 spin_lock(&lli->lli_lock);
1918                 if (lli->lli_posix_acl)
1919                         posix_acl_release(lli->lli_posix_acl);
1920                 lli->lli_posix_acl = md->posix_acl;
1921                 spin_unlock(&lli->lli_lock);
1922         }
1923 #endif
1924         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1925                                         sbi->ll_flags & LL_SBI_32BIT_API);
1926         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1927
1928         if (body->mbo_valid & OBD_MD_FLATIME) {
1929                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1930                         LTIME_S(inode->i_atime) = body->mbo_atime;
1931                 lli->lli_atime = body->mbo_atime;
1932         }
1933
1934         if (body->mbo_valid & OBD_MD_FLMTIME) {
1935                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1936                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1937                                "to "LPU64"\n", inode->i_ino,
1938                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1939                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1940                 }
1941                 lli->lli_mtime = body->mbo_mtime;
1942         }
1943
1944         if (body->mbo_valid & OBD_MD_FLCTIME) {
1945                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1946                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1947                 lli->lli_ctime = body->mbo_ctime;
1948         }
1949
1950         if (body->mbo_valid & OBD_MD_FLMODE)
1951                 inode->i_mode = (inode->i_mode & S_IFMT) |
1952                                 (body->mbo_mode & ~S_IFMT);
1953
1954         if (body->mbo_valid & OBD_MD_FLTYPE)
1955                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1956                                 (body->mbo_mode & S_IFMT);
1957
1958         LASSERT(inode->i_mode != 0);
1959         if (S_ISREG(inode->i_mode))
1960                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1961                                        LL_MAX_BLKSIZE_BITS);
1962         else
1963                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1964
1965         if (body->mbo_valid & OBD_MD_FLUID)
1966                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1967         if (body->mbo_valid & OBD_MD_FLGID)
1968                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1969         if (body->mbo_valid & OBD_MD_FLFLAGS)
1970                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1971         if (body->mbo_valid & OBD_MD_FLNLINK)
1972                 set_nlink(inode, body->mbo_nlink);
1973         if (body->mbo_valid & OBD_MD_FLRDEV)
1974                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1975
1976         if (body->mbo_valid & OBD_MD_FLID) {
1977                 /* FID shouldn't be changed! */
1978                 if (fid_is_sane(&lli->lli_fid)) {
1979                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1980                                  "Trying to change FID "DFID
1981                                  " to the "DFID", inode "DFID"(%p)\n",
1982                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1983                                  PFID(ll_inode2fid(inode)), inode);
1984                 } else {
1985                         lli->lli_fid = body->mbo_fid1;
1986                 }
1987         }
1988
1989         LASSERT(fid_seq(&lli->lli_fid) != 0);
1990
1991         if (body->mbo_valid & OBD_MD_FLSIZE) {
1992                 i_size_write(inode, body->mbo_size);
1993
1994                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1995                        PFID(ll_inode2fid(inode)),
1996                        (unsigned long long)body->mbo_size);
1997
1998                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1999                         inode->i_blocks = body->mbo_blocks;
2000         }
2001
2002         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2003                 LASSERT(md->mds_capa);
2004                 ll_add_capa(inode, md->mds_capa);
2005         }
2006
2007         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2008                 LASSERT(md->oss_capa);
2009                 ll_add_capa(inode, md->oss_capa);
2010         }
2011
2012         if (body->mbo_valid & OBD_MD_TSTATE) {
2013                 if (body->mbo_t_state & MS_RESTORE)
2014                         lli->lli_flags |= LLIF_FILE_RESTORING;
2015         }
2016
2017         return 0;
2018 }
2019
2020 int ll_read_inode2(struct inode *inode, void *opaque)
2021 {
2022         struct lustre_md *md = opaque;
2023         struct ll_inode_info *lli = ll_i2info(inode);
2024         int     rc;
2025         ENTRY;
2026
2027         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2028                PFID(&lli->lli_fid), inode);
2029
2030         /* Core attributes from the MDS first.  This is a new inode, and
2031          * the VFS doesn't zero times in the core inode so we have to do
2032          * it ourselves.  They will be overwritten by either MDS or OST
2033          * attributes - we just need to make sure they aren't newer. */
2034         LTIME_S(inode->i_mtime) = 0;
2035         LTIME_S(inode->i_atime) = 0;
2036         LTIME_S(inode->i_ctime) = 0;
2037         inode->i_rdev = 0;
2038         rc = ll_update_inode(inode, md);
2039         if (rc != 0)
2040                 RETURN(rc);
2041
2042         /* OIDEBUG(inode); */
2043
2044         /* initializing backing dev info. */
2045         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2046
2047
2048         if (S_ISREG(inode->i_mode)) {
2049                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2050                 inode->i_op = &ll_file_inode_operations;
2051                 inode->i_fop = sbi->ll_fop;
2052                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2053                 EXIT;
2054         } else if (S_ISDIR(inode->i_mode)) {
2055                 inode->i_op = &ll_dir_inode_operations;
2056                 inode->i_fop = &ll_dir_operations;
2057                 EXIT;
2058         } else if (S_ISLNK(inode->i_mode)) {
2059                 inode->i_op = &ll_fast_symlink_inode_operations;
2060                 EXIT;
2061         } else {
2062                 inode->i_op = &ll_special_inode_operations;
2063
2064                 init_special_inode(inode, inode->i_mode,
2065                                    inode->i_rdev);
2066
2067                 EXIT;
2068         }
2069
2070         return 0;
2071 }
2072
2073 void ll_delete_inode(struct inode *inode)
2074 {
2075         struct ll_inode_info *lli = ll_i2info(inode);
2076         ENTRY;
2077
2078         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2079                 /* It is last chance to write out dirty pages,
2080                  * otherwise we may lose data while umount */
2081                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2082
2083         truncate_inode_pages_final(&inode->i_data);
2084
2085         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2086                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2087                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2088
2089 #ifdef HAVE_SBOPS_EVICT_INODE
2090         ll_clear_inode(inode);
2091 #endif
2092         clear_inode(inode);
2093
2094         EXIT;
2095 }
2096
2097 int ll_iocontrol(struct inode *inode, struct file *file,
2098                  unsigned int cmd, unsigned long arg)
2099 {
2100         struct ll_sb_info *sbi = ll_i2sbi(inode);
2101         struct ptlrpc_request *req = NULL;
2102         int rc, flags = 0;
2103         ENTRY;
2104
2105         switch(cmd) {
2106         case FSFILT_IOC_GETFLAGS: {
2107                 struct mdt_body *body;
2108                 struct md_op_data *op_data;
2109
2110                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2111                                              0, 0, LUSTRE_OPC_ANY,
2112                                              NULL);
2113                 if (IS_ERR(op_data))
2114                         RETURN(PTR_ERR(op_data));
2115
2116                 op_data->op_valid = OBD_MD_FLFLAGS;
2117                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2118                 ll_finish_md_op_data(op_data);
2119                 if (rc) {
2120                         CERROR("%s: failure inode "DFID": rc = %d\n",
2121                                sbi->ll_md_exp->exp_obd->obd_name,
2122                                PFID(ll_inode2fid(inode)), rc);
2123                         RETURN(-abs(rc));
2124                 }
2125
2126                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2127
2128                 flags = body->mbo_flags;
2129
2130                 ptlrpc_req_finished(req);
2131
2132                 RETURN(put_user(flags, (int __user *)arg));
2133         }
2134         case FSFILT_IOC_SETFLAGS: {
2135                 struct iattr *attr;
2136                 struct md_op_data *op_data;
2137                 struct cl_object *obj;
2138                 struct obd_capa *capa;
2139
2140                 if (get_user(flags, (int __user *)arg))
2141                         RETURN(-EFAULT);
2142
2143                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2144                                              LUSTRE_OPC_ANY, NULL);
2145                 if (IS_ERR(op_data))
2146                         RETURN(PTR_ERR(op_data));
2147
2148                 op_data->op_attr_flags = flags;
2149                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2150                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2151                 ll_finish_md_op_data(op_data);
2152                 ptlrpc_req_finished(req);
2153                 if (rc)
2154                         RETURN(rc);
2155
2156                 inode->i_flags = ll_ext_to_inode_flags(flags);
2157
2158                 obj = ll_i2info(inode)->lli_clob;
2159                 if (obj == NULL)
2160                         RETURN(0);
2161
2162                 OBD_ALLOC_PTR(attr);
2163                 if (attr == NULL)
2164                         RETURN(-ENOMEM);
2165
2166                 attr->ia_valid = ATTR_ATTR_FLAG;
2167
2168                 capa = ll_mdscapa_get(inode);
2169                 rc = cl_setattr_ost(obj, attr, flags, capa);
2170                 capa_put(capa);
2171
2172                 OBD_FREE_PTR(attr);
2173                 RETURN(rc);
2174         }
2175         default:
2176                 RETURN(-ENOSYS);
2177         }
2178
2179         RETURN(0);
2180 }
2181
2182 int ll_flush_ctx(struct inode *inode)
2183 {
2184         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2185
2186         CDEBUG(D_SEC, "flush context for user %d\n",
2187                from_kuid(&init_user_ns, current_uid()));
2188
2189         obd_set_info_async(NULL, sbi->ll_md_exp,
2190                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2191                            0, NULL, NULL);
2192         obd_set_info_async(NULL, sbi->ll_dt_exp,
2193                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2194                            0, NULL, NULL);
2195         return 0;
2196 }
2197
2198 /* umount -f client means force down, don't save state */
2199 void ll_umount_begin(struct super_block *sb)
2200 {
2201         struct ll_sb_info *sbi = ll_s2sbi(sb);
2202         struct obd_device *obd;
2203         struct obd_ioctl_data *ioc_data;
2204         ENTRY;
2205
2206         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2207                sb->s_count, atomic_read(&sb->s_active));
2208
2209         obd = class_exp2obd(sbi->ll_md_exp);
2210         if (obd == NULL) {
2211                 CERROR("Invalid MDC connection handle "LPX64"\n",
2212                        sbi->ll_md_exp->exp_handle.h_cookie);
2213                 EXIT;
2214                 return;
2215         }
2216         obd->obd_force = 1;
2217
2218         obd = class_exp2obd(sbi->ll_dt_exp);
2219         if (obd == NULL) {
2220                 CERROR("Invalid LOV connection handle "LPX64"\n",
2221                        sbi->ll_dt_exp->exp_handle.h_cookie);
2222                 EXIT;
2223                 return;
2224         }
2225         obd->obd_force = 1;
2226
2227         OBD_ALLOC_PTR(ioc_data);
2228         if (ioc_data) {
2229                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2230                               sizeof *ioc_data, ioc_data, NULL);
2231
2232                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2233                               sizeof *ioc_data, ioc_data, NULL);
2234
2235                 OBD_FREE_PTR(ioc_data);
2236         }
2237
2238         /* Really, we'd like to wait until there are no requests outstanding,
2239          * and then continue.  For now, we just invalidate the requests,
2240          * schedule() and sleep one second if needed, and hope.
2241          */
2242         schedule();
2243         EXIT;
2244 }
2245
2246 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2247 {
2248         struct ll_sb_info *sbi = ll_s2sbi(sb);
2249         char *profilenm = get_profile_name(sb);
2250         int err;
2251         __u32 read_only;
2252
2253         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2254                 read_only = *flags & MS_RDONLY;
2255                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2256                                          sizeof(KEY_READ_ONLY),
2257                                          KEY_READ_ONLY, sizeof(read_only),
2258                                          &read_only, NULL);
2259                 if (err) {
2260                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2261                                       profilenm, read_only ?
2262                                       "read-only" : "read-write", err);
2263                         return err;
2264                 }
2265
2266                 if (read_only)
2267                         sb->s_flags |= MS_RDONLY;
2268                 else
2269                         sb->s_flags &= ~MS_RDONLY;
2270
2271                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2272                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2273                                       read_only ?  "read-only" : "read-write");
2274         }
2275         return 0;
2276 }
2277
2278 /**
2279  * Cleanup the open handle that is cached on MDT-side.
2280  *
2281  * For open case, the client side open handling thread may hit error
2282  * after the MDT grant the open. Under such case, the client should
2283  * send close RPC to the MDT as cleanup; otherwise, the open handle
2284  * on the MDT will be leaked there until the client umount or evicted.
2285  *
2286  * In further, if someone unlinked the file, because the open handle
2287  * holds the reference on such file/object, then it will block the
2288  * subsequent threads that want to locate such object via FID.
2289  *
2290  * \param[in] sb        super block for this file-system
2291  * \param[in] open_req  pointer to the original open request
2292  */
2293 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2294 {
2295         struct mdt_body                 *body;
2296         struct md_op_data               *op_data;
2297         struct ptlrpc_request           *close_req = NULL;
2298         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2299         ENTRY;
2300
2301         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2302         OBD_ALLOC_PTR(op_data);
2303         if (op_data == NULL) {
2304                 CWARN("%s: cannot allocate op_data to release open handle for "
2305                       DFID"\n",
2306                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2307
2308                 RETURN_EXIT;
2309         }
2310
2311         op_data->op_fid1 = body->mbo_fid1;
2312         op_data->op_handle = body->mbo_handle;
2313         op_data->op_mod_time = cfs_time_current_sec();
2314         md_close(exp, op_data, NULL, &close_req);
2315         ptlrpc_req_finished(close_req);
2316         ll_finish_md_op_data(op_data);
2317
2318         EXIT;
2319 }
2320
2321 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2322                   struct super_block *sb, struct lookup_intent *it)
2323 {
2324         struct ll_sb_info *sbi = NULL;
2325         struct lustre_md md = { NULL };
2326         int rc;
2327         ENTRY;
2328
2329         LASSERT(*inode || sb);
2330         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2331         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2332                               sbi->ll_md_exp, &md);
2333         if (rc != 0)
2334                 GOTO(cleanup, rc);
2335
2336         if (*inode) {
2337                 rc = ll_update_inode(*inode, &md);
2338                 if (rc != 0)
2339                         GOTO(out, rc);
2340         } else {
2341                 LASSERT(sb != NULL);
2342
2343                 /*
2344                  * At this point server returns to client's same fid as client
2345                  * generated for creating. So using ->fid1 is okay here.
2346                  */
2347                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2348
2349                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2350                                              sbi->ll_flags & LL_SBI_32BIT_API),
2351                                  &md);
2352                 if (IS_ERR(*inode)) {
2353 #ifdef CONFIG_FS_POSIX_ACL
2354                         if (md.posix_acl) {
2355                                 posix_acl_release(md.posix_acl);
2356                                 md.posix_acl = NULL;
2357                         }
2358 #endif
2359                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2360                         *inode = NULL;
2361                         CERROR("new_inode -fatal: rc %d\n", rc);
2362                         GOTO(out, rc);
2363                 }
2364         }
2365
2366         /* Handling piggyback layout lock.
2367          * Layout lock can be piggybacked by getattr and open request.
2368          * The lsm can be applied to inode only if it comes with a layout lock
2369          * otherwise correct layout may be overwritten, for example:
2370          * 1. proc1: mdt returns a lsm but not granting layout
2371          * 2. layout was changed by another client
2372          * 3. proc2: refresh layout and layout lock granted
2373          * 4. proc1: to apply a stale layout */
2374         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2375                 struct lustre_handle lockh;
2376                 struct ldlm_lock *lock;
2377
2378                 lockh.cookie = it->d.lustre.it_lock_handle;
2379                 lock = ldlm_handle2lock(&lockh);
2380                 LASSERT(lock != NULL);
2381                 if (ldlm_has_layout(lock)) {
2382                         struct cl_object_conf conf;
2383
2384                         memset(&conf, 0, sizeof(conf));
2385                         conf.coc_opc = OBJECT_CONF_SET;
2386                         conf.coc_inode = *inode;
2387                         conf.coc_lock = lock;
2388                         conf.u.coc_md = &md;
2389                         (void)ll_layout_conf(*inode, &conf);
2390                 }
2391                 LDLM_LOCK_PUT(lock);
2392         }
2393
2394         GOTO(out, rc = 0);
2395
2396 out:
2397         if (md.lsm != NULL)
2398                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2399         md_free_lustre_md(sbi->ll_md_exp, &md);
2400
2401 cleanup:
2402         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2403                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2404
2405         return rc;
2406 }
2407
2408 int ll_obd_statfs(struct inode *inode, void __user *arg)
2409 {
2410         struct ll_sb_info *sbi = NULL;
2411         struct obd_export *exp;
2412         char *buf = NULL;
2413         struct obd_ioctl_data *data = NULL;
2414         __u32 type;
2415         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2416         int len = 0, rc;
2417
2418         if (!inode || !(sbi = ll_i2sbi(inode)))
2419                 GOTO(out_statfs, rc = -EINVAL);
2420
2421         rc = obd_ioctl_getdata(&buf, &len, arg);
2422         if (rc)
2423                 GOTO(out_statfs, rc);
2424
2425         data = (void*)buf;
2426         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2427             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2428                 GOTO(out_statfs, rc = -EINVAL);
2429
2430         if (data->ioc_inllen1 != sizeof(__u32) ||
2431             data->ioc_inllen2 != sizeof(__u32) ||
2432             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2433             data->ioc_plen2 != sizeof(struct obd_uuid))
2434                 GOTO(out_statfs, rc = -EINVAL);
2435
2436         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2437         if (type & LL_STATFS_LMV)
2438                 exp = sbi->ll_md_exp;
2439         else if (type & LL_STATFS_LOV)
2440                 exp = sbi->ll_dt_exp;
2441         else
2442                 GOTO(out_statfs, rc = -ENODEV);
2443
2444         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2445         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2446         if (rc)
2447                 GOTO(out_statfs, rc);
2448 out_statfs:
2449         if (buf)
2450                 obd_ioctl_freedata(buf, len);
2451         return rc;
2452 }
2453
2454 int ll_process_config(struct lustre_cfg *lcfg)
2455 {
2456         struct super_block *sb;
2457         unsigned long x;
2458         int rc = 0;
2459         char *ptr;
2460
2461         /* The instance name contains the sb: lustre-client-aacfe000 */
2462         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2463         if (!ptr || !*(++ptr))
2464                 return -EINVAL;
2465         if (sscanf(ptr, "%lx", &x) != 1)
2466                 return -EINVAL;
2467         sb = (struct super_block *)x;
2468         /* This better be a real Lustre superblock! */
2469         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2470
2471         /* Note we have not called client_common_fill_super yet, so
2472            proc fns must be able to handle that! */
2473         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2474                                       lcfg, sb);
2475         if (rc > 0)
2476                 rc = 0;
2477         return rc;
2478 }
2479
2480 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2481 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2482                                        struct inode *i1, struct inode *i2,
2483                                        const char *name, size_t namelen,
2484                                        __u32 mode, __u32 opc, void *data)
2485 {
2486         LASSERT(i1 != NULL);
2487
2488         if (name == NULL) {
2489                 /* Do not reuse namelen for something else. */
2490                 if (namelen != 0)
2491                         return ERR_PTR(-EINVAL);
2492         } else {
2493                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2494                         return ERR_PTR(-ENAMETOOLONG);
2495
2496                 if (!lu_name_is_valid_2(name, namelen))
2497                         return ERR_PTR(-EINVAL);
2498         }
2499
2500         if (op_data == NULL)
2501                 OBD_ALLOC_PTR(op_data);
2502
2503         if (op_data == NULL)
2504                 return ERR_PTR(-ENOMEM);
2505
2506         ll_i2gids(op_data->op_suppgids, i1, i2);
2507         op_data->op_fid1 = *ll_inode2fid(i1);
2508         op_data->op_capa1 = ll_mdscapa_get(i1);
2509         op_data->op_default_stripe_offset = -1;
2510         if (S_ISDIR(i1->i_mode)) {
2511                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2512                 if (opc == LUSTRE_OPC_MKDIR)
2513                         op_data->op_default_stripe_offset =
2514                                    ll_i2info(i1)->lli_def_stripe_offset;
2515         }
2516
2517         if (i2) {
2518                 op_data->op_fid2 = *ll_inode2fid(i2);
2519                 op_data->op_capa2 = ll_mdscapa_get(i2);
2520                 if (S_ISDIR(i2->i_mode))
2521                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2522         } else {
2523                 fid_zero(&op_data->op_fid2);
2524                 op_data->op_capa2 = NULL;
2525         }
2526
2527         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2528                 op_data->op_cli_flags |= CLI_HASH64;
2529
2530         if (ll_need_32bit_api(ll_i2sbi(i1)))
2531                 op_data->op_cli_flags |= CLI_API32;
2532
2533         op_data->op_name = name;
2534         op_data->op_namelen = namelen;
2535         op_data->op_mode = mode;
2536         op_data->op_mod_time = cfs_time_current_sec();
2537         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2538         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2539         op_data->op_cap = cfs_curproc_cap_pack();
2540         op_data->op_bias = 0;
2541         op_data->op_cli_flags = 0;
2542         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2543              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2544                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2545         } else {
2546                 op_data->op_mds = 0;
2547         }
2548         op_data->op_data = data;
2549
2550         /* When called by ll_setattr_raw, file is i1. */
2551         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2552                 op_data->op_bias |= MDS_DATA_MODIFIED;
2553
2554         return op_data;
2555 }
2556
2557 void ll_finish_md_op_data(struct md_op_data *op_data)
2558 {
2559         capa_put(op_data->op_capa1);
2560         capa_put(op_data->op_capa2);
2561         OBD_FREE_PTR(op_data);
2562 }
2563
2564 #ifdef HAVE_SUPEROPS_USE_DENTRY
2565 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2566 #else
2567 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2568 #endif
2569 {
2570         struct ll_sb_info *sbi;
2571
2572 #ifdef HAVE_SUPEROPS_USE_DENTRY
2573         LASSERT((seq != NULL) && (dentry != NULL));
2574         sbi = ll_s2sbi(dentry->d_sb);
2575 #else
2576         LASSERT((seq != NULL) && (vfs != NULL));
2577         sbi = ll_s2sbi(vfs->mnt_sb);
2578 #endif
2579
2580         if (sbi->ll_flags & LL_SBI_NOLCK)
2581                 seq_puts(seq, ",nolock");
2582
2583         if (sbi->ll_flags & LL_SBI_FLOCK)
2584                 seq_puts(seq, ",flock");
2585
2586         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2587                 seq_puts(seq, ",localflock");
2588
2589         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2590                 seq_puts(seq, ",user_xattr");
2591
2592         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2593                 seq_puts(seq, ",lazystatfs");
2594
2595         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2596                 seq_puts(seq, ",user_fid2path");
2597
2598         RETURN(0);
2599 }
2600
2601 /**
2602  * Get obd name by cmd, and copy out to user space
2603  */
2604 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2605 {
2606         struct ll_sb_info *sbi = ll_i2sbi(inode);
2607         struct obd_device *obd;
2608         ENTRY;
2609
2610         if (cmd == OBD_IOC_GETDTNAME)
2611                 obd = class_exp2obd(sbi->ll_dt_exp);
2612         else if (cmd == OBD_IOC_GETMDNAME)
2613                 obd = class_exp2obd(sbi->ll_md_exp);
2614         else
2615                 RETURN(-EINVAL);
2616
2617         if (!obd)
2618                 RETURN(-ENOENT);
2619
2620         if (copy_to_user((void __user *)arg, obd->obd_name,
2621                          strlen(obd->obd_name) + 1))
2622                 RETURN(-EFAULT);
2623
2624         RETURN(0);
2625 }
2626
2627 /**
2628  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2629  * fsname will be returned in this buffer; otherwise, a static buffer will be
2630  * used to store the fsname and returned to caller.
2631  */
2632 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2633 {
2634         static char fsname_static[MTI_NAME_MAXLEN];
2635         struct lustre_sb_info *lsi = s2lsi(sb);
2636         char *ptr;
2637         int len;
2638
2639         if (buf == NULL) {
2640                 /* this means the caller wants to use static buffer
2641                  * and it doesn't care about race. Usually this is
2642                  * in error reporting path */
2643                 buf = fsname_static;
2644                 buflen = sizeof(fsname_static);
2645         }
2646
2647         len = strlen(lsi->lsi_lmd->lmd_profile);
2648         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2649         if (ptr && (strcmp(ptr, "-client") == 0))
2650                 len -= 7;
2651
2652         if (unlikely(len >= buflen))
2653                 len = buflen - 1;
2654         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2655         buf[len] = '\0';
2656
2657         return buf;
2658 }
2659
2660 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2661 {
2662         char *path = NULL;
2663
2664         struct path p;
2665
2666         p.dentry = dentry;
2667         p.mnt = current->fs->root.mnt;
2668         path_get(&p);
2669         path = d_path(&p, buf, bufsize);
2670         path_put(&p);
2671         return path;
2672 }
2673
2674 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2675 {
2676         char *buf, *path = NULL;
2677         struct dentry *dentry = NULL;
2678         struct inode *inode = page->mapping->host;
2679
2680         /* this can be called inside spin lock so use GFP_ATOMIC. */
2681         buf = (char *)__get_free_page(GFP_ATOMIC);
2682         if (buf != NULL) {
2683                 dentry = d_find_alias(page->mapping->host);
2684                 if (dentry != NULL)
2685                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2686         }
2687
2688         CDEBUG(D_WARNING,
2689                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2690                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2691                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2692                PFID(ll_inode2fid(inode)),
2693                (path && !IS_ERR(path)) ? path : "", ioret);
2694
2695         if (dentry != NULL)
2696                 dput(dentry);
2697
2698         if (buf != NULL)
2699                 free_page((unsigned long)buf);
2700 }
2701
2702 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2703                         struct lov_user_md **kbuf)
2704 {
2705         struct lov_user_md      lum;
2706         ssize_t                 lum_size;
2707         ENTRY;
2708
2709         if (copy_from_user(&lum, md, sizeof(lum)))
2710                 RETURN(-EFAULT);
2711
2712         lum_size = ll_lov_user_md_size(&lum);
2713         if (lum_size < 0)
2714                 RETURN(lum_size);
2715
2716         OBD_ALLOC(*kbuf, lum_size);
2717         if (*kbuf == NULL)
2718                 RETURN(-ENOMEM);
2719
2720         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2721                 OBD_FREE(*kbuf, lum_size);
2722                 RETURN(-EFAULT);
2723         }
2724
2725         RETURN(lum_size);
2726 }
2727
2728 /*
2729  * Compute llite root squash state after a change of root squash
2730  * configuration setting or add/remove of a lnet nid
2731  */
2732 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2733 {
2734         struct root_squash_info *squash = &sbi->ll_squash;
2735         int i;
2736         bool matched;
2737         lnet_process_id_t id;
2738
2739         /* Update norootsquash flag */
2740         down_write(&squash->rsi_sem);
2741         if (list_empty(&squash->rsi_nosquash_nids))
2742                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2743         else {
2744                 /* Do not apply root squash as soon as one of our NIDs is
2745                  * in the nosquash_nids list */
2746                 matched = false;
2747                 i = 0;
2748                 while (LNetGetId(i++, &id) != -ENOENT) {
2749                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2750                                 continue;
2751                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2752                                 matched = true;
2753                                 break;
2754                         }
2755                 }
2756                 if (matched)
2757                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2758                 else
2759                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2760         }
2761         up_write(&squash->rsi_sem);
2762 }
2763
2764 /**
2765  * Parse linkea content to extract information about a given hardlink
2766  *
2767  * \param[in]   ldata      - Initialized linkea data
2768  * \param[in]   linkno     - Link identifier
2769  * \param[out]  parent_fid - The entry's parent FID
2770  * \param[out]  ln         - Entry name destination buffer
2771  *
2772  * \retval 0 on success
2773  * \retval Appropriate negative error code on failure
2774  */
2775 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2776                             struct lu_fid *parent_fid, struct lu_name *ln)
2777 {
2778         unsigned int    idx;
2779         int             rc;
2780         ENTRY;
2781
2782         rc = linkea_init(ldata);
2783         if (rc < 0)
2784                 RETURN(rc);
2785
2786         if (linkno >= ldata->ld_leh->leh_reccount)
2787                 /* beyond last link */
2788                 RETURN(-ENODATA);
2789
2790         linkea_first_entry(ldata);
2791         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2792                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2793                                     parent_fid);
2794                 if (idx == linkno)
2795                         break;
2796
2797                 linkea_next_entry(ldata);
2798         }
2799
2800         if (idx < linkno)
2801                 RETURN(-ENODATA);
2802
2803         RETURN(0);
2804 }
2805
2806 /**
2807  * Get parent FID and name of an identified link. Operation is performed for
2808  * a given link number, letting the caller iterate over linkno to list one or
2809  * all links of an entry.
2810  *
2811  * \param[in]     file - File descriptor against which to perform the operation
2812  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2813  *                       on and the available size. It is eventually filled with
2814  *                       the requested information or left untouched on error
2815  *
2816  * \retval - 0 on success
2817  * \retval - Appropriate negative error code on failure
2818  */
2819 int ll_getparent(struct file *file, struct getparent __user *arg)
2820 {
2821         struct dentry           *dentry = file->f_dentry;
2822         struct inode            *inode = file->f_dentry->d_inode;
2823         struct linkea_data      *ldata;
2824         struct lu_buf            buf = LU_BUF_NULL;
2825         struct lu_name           ln;
2826         struct lu_fid            parent_fid;
2827         __u32                    linkno;
2828         __u32                    name_size;
2829         int                      rc;
2830
2831         ENTRY;
2832
2833         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2834             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2835                 RETURN(-EPERM);
2836
2837         if (get_user(name_size, &arg->gp_name_size))
2838                 RETURN(-EFAULT);
2839
2840         if (get_user(linkno, &arg->gp_linkno))
2841                 RETURN(-EFAULT);
2842
2843         if (name_size > PATH_MAX)
2844                 RETURN(-EINVAL);
2845
2846         OBD_ALLOC(ldata, sizeof(*ldata));
2847         if (ldata == NULL)
2848                 RETURN(-ENOMEM);
2849
2850         rc = linkea_data_new(ldata, &buf);
2851         if (rc < 0)
2852                 GOTO(ldata_free, rc);
2853
2854         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2855         if (rc < 0)
2856                 GOTO(lb_free, rc);
2857
2858         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2859         if (rc < 0)
2860                 GOTO(lb_free, rc);
2861
2862         if (ln.ln_namelen >= name_size)
2863                 GOTO(lb_free, rc = -EOVERFLOW);
2864
2865         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2866                 GOTO(lb_free, rc = -EFAULT);
2867
2868         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2869                 GOTO(lb_free, rc = -EFAULT);
2870
2871         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2872                 GOTO(lb_free, rc = -EFAULT);
2873
2874 lb_free:
2875         lu_buf_free(&buf);
2876 ldata_free:
2877         OBD_FREE(ldata, sizeof(*ldata));
2878
2879         RETURN(rc);
2880 }