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
  49 #include <lustre_ioctl.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         sbi->ll_cache = cl_cache_init(lru_page_max);
  95         if (sbi->ll_cache == NULL) {
  96                 OBD_FREE(sbi, sizeof(*sbi));
  97                 RETURN(NULL);
  98         }
  99
 100         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 101                                            SBI_DEFAULT_READAHEAD_MAX);
 102         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 103         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 104                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 105         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 106         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 107
 108         ll_generate_random_uuid(uuid);
 109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 111
 112         spin_lock(&ll_sb_lock);
 113         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 114         spin_unlock(&ll_sb_lock);
 115
 116         sbi->ll_flags |= LL_SBI_VERBOSE;
 117 #ifdef ENABLE_CHECKSUM
 118         sbi->ll_flags |= LL_SBI_CHECKSUM;
 119 #endif
 120
 121 #ifdef HAVE_LRU_RESIZE_SUPPORT
 122         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 123 #endif
 124
 125         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 126                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 127                                pp_r_hist.oh_lock);
 128                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 129                                pp_w_hist.oh_lock);
 130         }
 131
 132         /* metadata statahead is enabled by default */
 133         sbi->ll_sa_max = LL_SA_RPC_DEF;
 134         atomic_set(&sbi->ll_sa_total, 0);
 135         atomic_set(&sbi->ll_sa_wrong, 0);
 136         atomic_set(&sbi->ll_sa_running, 0);
 137         atomic_set(&sbi->ll_agl_total, 0);
 138         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 139
 140         /* root squash */
 141         sbi->ll_squash.rsi_uid = 0;
 142         sbi->ll_squash.rsi_gid = 0;
 143         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 144         init_rwsem(&sbi->ll_squash.rsi_sem);
 145
 146         RETURN(sbi);
 147 }
 148
 149 static void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 spin_lock(&ll_sb_lock);
 156                 list_del(&sbi->ll_list);
 157                 spin_unlock(&ll_sb_lock);
 158                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 159                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 160                 if (sbi->ll_cache != NULL) {
 161                         cl_cache_decref(sbi->ll_cache);
 162                         sbi->ll_cache = NULL;
 163                 }
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 170                                     struct vfsmount *mnt)
 171 {
 172         struct inode *root = NULL;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct obd_capa *oc = NULL;
 176         struct obd_statfs *osfs = NULL;
 177         struct ptlrpc_request *request = NULL;
 178         struct obd_connect_data *data = NULL;
 179         struct obd_uuid *uuid;
 180         struct md_op_data *op_data;
 181         struct lustre_md lmd;
 182         u64 valid;
 183         int size, err, checksum;
 184         ENTRY;
 185
 186         obd = class_name2obd(md);
 187         if (!obd) {
 188                 CERROR("MD %s: not setup or attached\n", md);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         OBD_ALLOC_PTR(data);
 193         if (data == NULL)
 194                 RETURN(-ENOMEM);
 195
 196         OBD_ALLOC_PTR(osfs);
 197         if (osfs == NULL) {
 198                 OBD_FREE_PTR(data);
 199                 RETURN(-ENOMEM);
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 211                                   OBD_CONNECT_EINPROGRESS |
 212                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 213                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 214                                   OBD_CONNECT_MAX_EASIZE |
 215                                   OBD_CONNECT_FLOCK_DEAD |
 216                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 217                                   OBD_CONNECT_OPEN_BY_FID |
 218                                   OBD_CONNECT_DIR_STRIPE;
 219
 220 #ifdef HAVE_LRU_RESIZE_SUPPORT
 221         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 222                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 223 #endif
 224 #ifdef CONFIG_FS_POSIX_ACL
 225         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 226 #endif
 227
 228         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 229                 /* flag mdc connection as lightweight, only used for test
 230                  * purpose, use with care */
 231                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 232
 233         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 234         data->ocd_version = LUSTRE_VERSION_CODE;
 235
 236         if (sb->s_flags & MS_RDONLY)
 237                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 238         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 239                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 240
 241 #ifdef HAVE_MS_FLOCK_LOCK
 242         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 243         sb->s_flags |= MS_FLOCK_LOCK;
 244 #endif
 245 #ifdef MS_HAS_NEW_AOPS
 246         sb->s_flags |= MS_HAS_NEW_AOPS;
 247 #endif
 248
 249         if (sbi->ll_flags & LL_SBI_FLOCK)
 250                 sbi->ll_fop = &ll_file_operations_flock;
 251         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 252                 sbi->ll_fop = &ll_file_operations;
 253         else
 254                 sbi->ll_fop = &ll_file_operations_noflock;
 255
 256         /* real client */
 257         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 258         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 259                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 260
 261         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 262
 263         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 264         if (err == -EBUSY) {
 265                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 266                                    "recovery, of which this client is not a "
 267                                    "part. Please wait for recovery to complete,"
 268                                    " abort, or time out.\n", md);
 269                 GOTO(out, err);
 270         } else if (err) {
 271                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 272                 GOTO(out, err);
 273         }
 274
 275         sbi->ll_md_exp->exp_connect_data = *data;
 276
 277         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 278                            LUSTRE_SEQ_METADATA);
 279         if (err) {
 280                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 281                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 282                 GOTO(out_md, err);
 283         }
 284
 285         /* For mount, we only need fs info from MDT0, and also in DNE, it
 286          * can make sure the client can be mounted as long as MDT0 is
 287          * avaible */
 288         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 289                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 290                         OBD_STATFS_FOR_MDT0);
 291         if (err)
 292                 GOTO(out_md_fid, err);
 293
 294         /* This needs to be after statfs to ensure connect has finished.
 295          * Note that "data" does NOT contain the valid connect reply.
 296          * If connecting to a 1.8 server there will be no LMV device, so
 297          * we can access the MDC export directly and exp_connect_flags will
 298          * be non-zero, but if accessing an upgraded 2.1 server it will
 299          * have the correct flags filled in.
 300          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 301         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 302         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 303             valid != CLIENT_CONNECT_MDT_REQD) {
 304                 char *buf;
 305
 306                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 307                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 308                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 309                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 310                                    "feature(s) needed for correct operation "
 311                                    "of this client (%s). Please upgrade "
 312                                    "server or downgrade client.\n",
 313                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 314                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 315                 GOTO(out_md_fid, err = -EPROTO);
 316         }
 317
 318         size = sizeof(*data);
 319         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 320                            KEY_CONN_DATA,  &size, data, NULL);
 321         if (err) {
 322                 CERROR("%s: Get connect data failed: rc = %d\n",
 323                        sbi->ll_md_exp->exp_obd->obd_name, err);
 324                 GOTO(out_md_fid, err);
 325         }
 326
 327         LASSERT(osfs->os_bsize);
 328         sb->s_blocksize = osfs->os_bsize;
 329         sb->s_blocksize_bits = log2(osfs->os_bsize);
 330         sb->s_magic = LL_SUPER_MAGIC;
 331         sb->s_maxbytes = MAX_LFS_FILESIZE;
 332         sbi->ll_namelen = osfs->os_namelen;
 333         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 334
 335         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 336             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 337                 LCONSOLE_INFO("Disabling user_xattr feature because "
 338                               "it is not supported on the server\n");
 339                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 343 #ifdef MS_POSIXACL
 344                 sb->s_flags |= MS_POSIXACL;
 345 #endif
 346                 sbi->ll_flags |= LL_SBI_ACL;
 347         } else {
 348                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 349 #ifdef MS_POSIXACL
 350                 sb->s_flags &= ~MS_POSIXACL;
 351 #endif
 352                 sbi->ll_flags &= ~LL_SBI_ACL;
 353         }
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 356                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 357                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 358                         LCONSOLE_INFO("client is set as remote by default.\n");
 359                 }
 360         } else {
 361                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 362                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 363                         LCONSOLE_INFO("client claims to be remote, but server "
 364                                       "rejected, forced to be local.\n");
 365                 }
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 369                 LCONSOLE_INFO("client enabled MDS capability!\n");
 370                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 371         }
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 374                 LCONSOLE_INFO("client enabled OSS capability!\n");
 375                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 376         }
 377
 378         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 379                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 382                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 383         else
 384                 sbi->ll_md_brw_pages = 1;
 385
 386         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 387                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 388
 389         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 390                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 391                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 392                                       "unknown maximum xattr size.\n", dt);
 393                 } else {
 394                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 395                         sbi->ll_xattr_cache_enabled = 1;
 396                 }
 397         }
 398
 399         obd = class_name2obd(dt);
 400         if (!obd) {
 401                 CERROR("DT %s: not setup or attached\n", dt);
 402                 GOTO(out_md_fid, err = -ENODEV);
 403         }
 404
 405         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 406                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 407                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 408                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 409                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 410                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 411                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 412                                   OBD_CONNECT_MAXBYTES |
 413                                   OBD_CONNECT_EINPROGRESS |
 414                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 415                                   OBD_CONNECT_LAYOUTLOCK |
 416                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 417
 418         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 419                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 420                  * disabled by default, because it can still be enabled on the
 421                  * fly via /proc. As a consequence, we still need to come to an
 422                  * agreement on the supported algorithms at connect time */
 423                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 424
 425                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 426                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 427                 else
 428                         data->ocd_cksum_types = cksum_types_supported_client();
 429         }
 430
 431 #ifdef HAVE_LRU_RESIZE_SUPPORT
 432         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 433 #endif
 434         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 435                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 436
 437         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 438                "ocd_grant: %d\n", data->ocd_connect_flags,
 439                data->ocd_version, data->ocd_grant);
 440
 441         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 442         obd->obd_upcall.onu_upcall = cl_ocd_update;
 443
 444         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 445
 446         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 447                           NULL);
 448         if (err == -EBUSY) {
 449                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 450                                    "recovery, of which this client is not a "
 451                                    "part.  Please wait for recovery to "
 452                                    "complete, abort, or time out.\n", dt);
 453                 GOTO(out_md, err);
 454         } else if (err) {
 455                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 456                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 457                 GOTO(out_md, err);
 458         }
 459
 460         sbi->ll_dt_exp->exp_connect_data = *data;
 461
 462         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 463                            LUSTRE_SEQ_METADATA);
 464         if (err) {
 465                 CERROR("%s: Can't init data layer FID infrastructure, "
 466                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 467                 GOTO(out_dt, err);
 468         }
 469
 470         mutex_lock(&sbi->ll_lco.lco_lock);
 471         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 472         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 473         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 474         mutex_unlock(&sbi->ll_lco.lco_lock);
 475
 476         fid_zero(&sbi->ll_root_fid);
 477         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 478         if (err) {
 479                 CERROR("cannot mds_connect: rc = %d\n", err);
 480                 GOTO(out_lock_cn_cb, err);
 481         }
 482         if (!fid_is_sane(&sbi->ll_root_fid)) {
 483                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 484                        sbi->ll_md_exp->exp_obd->obd_name,
 485                        PFID(&sbi->ll_root_fid));
 486                 GOTO(out_lock_cn_cb, err = -EINVAL);
 487         }
 488         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 489
 490         sb->s_op = &lustre_super_operations;
 491 #if THREAD_SIZE >= 8192 /*b=17630*/
 492         sb->s_export_op = &lustre_export_operations;
 493 #endif
 494
 495         /* make root inode
 496          * XXX: move this to after cbd setup? */
 497         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 498                 OBD_MD_FLMODEASIZE;
 499         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 500                 valid |= OBD_MD_FLRMTPERM;
 501         else if (sbi->ll_flags & LL_SBI_ACL)
 502                 valid |= OBD_MD_FLACL;
 503
 504         OBD_ALLOC_PTR(op_data);
 505         if (op_data == NULL)
 506                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 507
 508         op_data->op_fid1 = sbi->ll_root_fid;
 509         op_data->op_mode = 0;
 510         op_data->op_capa1 = oc;
 511         op_data->op_valid = valid;
 512
 513         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 514         if (oc)
 515                 capa_put(oc);
 516         OBD_FREE_PTR(op_data);
 517         if (err) {
 518                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 519                        sbi->ll_md_exp->exp_obd->obd_name, err);
 520                 GOTO(out_lock_cn_cb, err);
 521         }
 522
 523         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 524                                sbi->ll_md_exp, &lmd);
 525         if (err) {
 526                 CERROR("failed to understand root inode md: rc = %d\n", err);
 527                 ptlrpc_req_finished(request);
 528                 GOTO(out_lock_cn_cb, err);
 529         }
 530
 531         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 532         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 533                                             sbi->ll_flags & LL_SBI_32BIT_API),
 534                        &lmd);
 535         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 536         ptlrpc_req_finished(request);
 537
 538         if (IS_ERR(root)) {
 539                 if (lmd.lsm)
 540                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 541 #ifdef CONFIG_FS_POSIX_ACL
 542                 if (lmd.posix_acl) {
 543                         posix_acl_release(lmd.posix_acl);
 544                         lmd.posix_acl = NULL;
 545                 }
 546 #endif
 547                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 548                 root = NULL;
 549                 CERROR("lustre_lite: bad iget4 for root\n");
 550                 GOTO(out_root, err);
 551         }
 552
 553 #ifdef CONFIG_FS_POSIX_ACL
 554         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 555                 rct_init(&sbi->ll_rct);
 556                 et_init(&sbi->ll_et);
 557         }
 558 #endif
 559
 560         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 561         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 562                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 563                                  NULL);
 564         cl_sb_init(sb);
 565
 566         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 567                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 568                                  sbi->ll_cache, NULL);
 569
 570         sb->s_root = d_make_root(root);
 571         if (sb->s_root == NULL) {
 572                 CERROR("%s: can't make root dentry\n",
 573                         ll_get_fsname(sb, NULL, 0));
 574                 GOTO(out_root, err = -ENOMEM);
 575         }
 576 #ifdef HAVE_DCACHE_LOCK
 577         sb->s_root->d_op = &ll_d_ops;
 578 #endif
 579
 580         sbi->ll_sdev_orig = sb->s_dev;
 581
 582         /* We set sb->s_dev equal on all lustre clients in order to support
 583          * NFS export clustering.  NFSD requires that the FSID be the same
 584          * on all clients. */
 585         /* s_dev is also used in lt_compare() to compare two fs, but that is
 586          * only a node-local comparison. */
 587         uuid = obd_get_uuid(sbi->ll_md_exp);
 588         if (uuid != NULL)
 589                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 590
 591         if (data != NULL)
 592                 OBD_FREE_PTR(data);
 593         if (osfs != NULL)
 594                 OBD_FREE_PTR(osfs);
 595         if (proc_lustre_fs_root != NULL) {
 596                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 597                                                   dt, md);
 598                 if (err < 0) {
 599                         CERROR("%s: could not register mount in lprocfs: "
 600                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 601                         err = 0;
 602                 }
 603         }
 604
 605         RETURN(err);
 606 out_root:
 607         if (root)
 608                 iput(root);
 609 out_lock_cn_cb:
 610         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 611 out_dt:
 612         obd_disconnect(sbi->ll_dt_exp);
 613         sbi->ll_dt_exp = NULL;
 614 out_md_fid:
 615         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 616 out_md:
 617         obd_disconnect(sbi->ll_md_exp);
 618         sbi->ll_md_exp = NULL;
 619 out:
 620         if (data != NULL)
 621                 OBD_FREE_PTR(data);
 622         if (osfs != NULL)
 623                 OBD_FREE_PTR(osfs);
 624         return err;
 625 }
 626
 627 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 628 {
 629         int size, rc;
 630
 631         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 632         size = sizeof(int);
 633         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 634                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 635         if (rc)
 636                 CERROR("Get max mdsize error rc %d\n", rc);
 637
 638         RETURN(rc);
 639 }
 640
 641 /**
 642  * Get the value of the default_easize parameter.
 643  *
 644  * \see client_obd::cl_default_mds_easize
 645  *
 646  * \param[in] sbi       superblock info for this filesystem
 647  * \param[out] lmmsize  pointer to storage location for value
 648  *
 649  * \retval 0            on success
 650  * \retval negative     negated errno on failure
 651  */
 652 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 653 {
 654         int size, rc;
 655
 656         size = sizeof(int);
 657         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 658                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 659         if (rc)
 660                 CERROR("Get default mdsize error rc %d\n", rc);
 661
 662         RETURN(rc);
 663 }
 664
 665 /**
 666  * Set the default_easize parameter to the given value.
 667  *
 668  * \see client_obd::cl_default_mds_easize
 669  *
 670  * \param[in] sbi       superblock info for this filesystem
 671  * \param[in] lmmsize   the size to set
 672  *
 673  * \retval 0            on success
 674  * \retval negative     negated errno on failure
 675  */
 676 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 677 {
 678         int rc;
 679
 680         if (lmmsize < sizeof(struct lov_mds_md) ||
 681             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 682                 return -EINVAL;
 683
 684         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 685                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 686                                 sizeof(int), &lmmsize, NULL);
 687
 688         RETURN(rc);
 689 }
 690
 691 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 692 {
 693         int size, rc;
 694
 695         size = sizeof(int);
 696         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 697                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 698         if (rc)
 699                 CERROR("Get max cookiesize error rc %d\n", rc);
 700
 701         RETURN(rc);
 702 }
 703
 704 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 705 {
 706         int size, rc;
 707
 708         size = sizeof(int);
 709         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 710                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 711         if (rc)
 712                 CERROR("Get default cookiesize error rc %d\n", rc);
 713
 714         RETURN(rc);
 715 }
 716
 717 static void ll_dump_inode(struct inode *inode)
 718 {
 719         struct ll_d_hlist_node *tmp;
 720         int dentry_count = 0;
 721
 722         LASSERT(inode != NULL);
 723
 724         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 725                 dentry_count++;
 726
 727         CERROR("%s: inode %p dump: dev=%s fid="DFID
 728                " mode=%o count=%u, %d dentries\n",
 729                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 730                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 731                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 732 }
 733
 734 void lustre_dump_dentry(struct dentry *dentry, int recur)
 735 {
 736         struct list_head *tmp;
 737         int subdirs = 0;
 738
 739         LASSERT(dentry != NULL);
 740
 741         list_for_each(tmp, &dentry->d_subdirs)
 742                 subdirs++;
 743
 744         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 745                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 746                dentry->d_name.len, dentry->d_name.name,
 747                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 748                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 749                dentry->d_flags, dentry->d_fsdata, subdirs);
 750         if (dentry->d_inode != NULL)
 751                 ll_dump_inode(dentry->d_inode);
 752
 753         if (recur == 0)
 754                 return;
 755
 756         list_for_each(tmp, &dentry->d_subdirs) {
 757                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 758                 lustre_dump_dentry(d, recur - 1);
 759         }
 760 }
 761
 762 static void client_common_put_super(struct super_block *sb)
 763 {
 764         struct ll_sb_info *sbi = ll_s2sbi(sb);
 765         ENTRY;
 766
 767 #ifdef CONFIG_FS_POSIX_ACL
 768         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 769                 et_fini(&sbi->ll_et);
 770                 rct_fini(&sbi->ll_rct);
 771         }
 772 #endif
 773
 774         cl_sb_fini(sb);
 775
 776         list_del(&sbi->ll_conn_chain);
 777
 778         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 779         obd_disconnect(sbi->ll_dt_exp);
 780         sbi->ll_dt_exp = NULL;
 781
 782         lprocfs_unregister_mountpoint(sbi);
 783
 784         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 785         obd_disconnect(sbi->ll_md_exp);
 786         sbi->ll_md_exp = NULL;
 787
 788         EXIT;
 789 }
 790
 791 void ll_kill_super(struct super_block *sb)
 792 {
 793         struct ll_sb_info *sbi;
 794         ENTRY;
 795
 796         /* not init sb ?*/
 797         if (!(sb->s_flags & MS_ACTIVE))
 798                 return;
 799
 800         sbi = ll_s2sbi(sb);
 801         /* we need restore s_dev from changed for clustred NFS before put_super
 802          * because new kernels have cached s_dev and change sb->s_dev in
 803          * put_super not affected real removing devices */
 804         if (sbi) {
 805                 sb->s_dev = sbi->ll_sdev_orig;
 806                 sbi->ll_umounting = 1;
 807
 808                 /* wait running statahead threads to quit */
 809                 while (atomic_read(&sbi->ll_sa_running) > 0)
 810                         schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
 811                                 msecs_to_jiffies(MSEC_PER_SEC >> 3));
 812         }
 813
 814         EXIT;
 815 }
 816
 817 static inline int ll_set_opt(const char *opt, char *data, int fl)
 818 {
 819         if (strncmp(opt, data, strlen(opt)) != 0)
 820                 return(0);
 821         else
 822                 return(fl);
 823 }
 824
 825 /* non-client-specific mount options are parsed in lmd_parse */
 826 static int ll_options(char *options, int *flags)
 827 {
 828         int tmp;
 829         char *s1 = options, *s2;
 830         ENTRY;
 831
 832         if (!options)
 833                 RETURN(0);
 834
 835         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 836
 837         while (*s1) {
 838                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 839                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 840                 if (tmp) {
 841                         *flags |= tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 845                 if (tmp) {
 846                         *flags |= tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 855                 if (tmp) {
 856                         *flags &= ~tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 860                 if (tmp) {
 861                         *flags |= tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 865                 if (tmp) {
 866                         *flags &= ~tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 875                 if (tmp) {
 876                         *flags |= tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 880                 if (tmp) {
 881                         *flags &= ~tmp;
 882                         goto next;
 883                 }
 884
 885                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 886                 if (tmp) {
 887                         *flags |= tmp;
 888                         goto next;
 889                 }
 890                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 891                 if (tmp) {
 892                         *flags &= ~tmp;
 893                         goto next;
 894                 }
 895                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 896                 if (tmp) {
 897                         *flags |= tmp;
 898                         goto next;
 899                 }
 900                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 901                 if (tmp) {
 902                         *flags &= ~tmp;
 903                         goto next;
 904                 }
 905                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 906                 if (tmp) {
 907                         *flags |= tmp;
 908                         goto next;
 909                 }
 910                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 911                 if (tmp) {
 912                         *flags &= ~tmp;
 913                         goto next;
 914                 }
 915                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 916                 if (tmp) {
 917                         *flags |= tmp;
 918                         goto next;
 919                 }
 920                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 921                 if (tmp) {
 922                         *flags |= tmp;
 923                         goto next;
 924                 }
 925                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 926                 if (tmp) {
 927                         *flags &= ~tmp;
 928                         goto next;
 929                 }
 930                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 931                                    s1);
 932                 RETURN(-EINVAL);
 933
 934 next:
 935                 /* Find next opt */
 936                 s2 = strchr(s1, ',');
 937                 if (s2 == NULL)
 938                         break;
 939                 s1 = s2 + 1;
 940         }
 941         RETURN(0);
 942 }
 943
 944 void ll_lli_init(struct ll_inode_info *lli)
 945 {
 946         lli->lli_inode_magic = LLI_INODE_MAGIC;
 947         lli->lli_flags = 0;
 948         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 949         spin_lock_init(&lli->lli_lock);
 950         lli->lli_posix_acl = NULL;
 951         lli->lli_remote_perms = NULL;
 952         mutex_init(&lli->lli_rmtperm_mutex);
 953         /* Do not set lli_fid, it has been initialized already. */
 954         fid_zero(&lli->lli_pfid);
 955         INIT_LIST_HEAD(&lli->lli_oss_capas);
 956         atomic_set(&lli->lli_open_count, 0);
 957         lli->lli_mds_capa = NULL;
 958         lli->lli_rmtperm_time = 0;
 959         lli->lli_mds_read_och = NULL;
 960         lli->lli_mds_write_och = NULL;
 961         lli->lli_mds_exec_och = NULL;
 962         lli->lli_open_fd_read_count = 0;
 963         lli->lli_open_fd_write_count = 0;
 964         lli->lli_open_fd_exec_count = 0;
 965         mutex_init(&lli->lli_och_mutex);
 966         spin_lock_init(&lli->lli_agl_lock);
 967         spin_lock_init(&lli->lli_layout_lock);
 968         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 969         lli->lli_clob = NULL;
 970
 971         init_rwsem(&lli->lli_xattrs_list_rwsem);
 972         mutex_init(&lli->lli_xattrs_enq_lock);
 973
 974         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 975         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 976                 mutex_init(&lli->lli_readdir_mutex);
 977                 lli->lli_opendir_key = NULL;
 978                 lli->lli_sai = NULL;
 979                 spin_lock_init(&lli->lli_sa_lock);
 980                 lli->lli_opendir_pid = 0;
 981                 lli->lli_sa_enabled = 0;
 982                 lli->lli_def_stripe_offset = -1;
 983         } else {
 984                 mutex_init(&lli->lli_size_mutex);
 985                 lli->lli_symlink_name = NULL;
 986                 init_rwsem(&lli->lli_trunc_sem);
 987                 range_lock_tree_init(&lli->lli_write_tree);
 988                 init_rwsem(&lli->lli_glimpse_sem);
 989                 lli->lli_glimpse_time = 0;
 990                 INIT_LIST_HEAD(&lli->lli_agl_list);
 991                 lli->lli_agl_index = 0;
 992                 lli->lli_async_rc = 0;
 993         }
 994         mutex_init(&lli->lli_layout_mutex);
 995 }
 996
 997 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 998 {
 999         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1000
1001         bdi->name = "lustre";
1002         return bdi_register(bdi, NULL, "lustre-%d",
1003                             atomic_inc_return(&ll_bdi_num));
1004 }
1005
1006 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1007 {
1008         struct lustre_profile *lprof = NULL;
1009         struct lustre_sb_info *lsi = s2lsi(sb);
1010         struct ll_sb_info *sbi;
1011         char  *dt = NULL, *md = NULL;
1012         char  *profilenm = get_profile_name(sb);
1013         struct config_llog_instance *cfg;
1014         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1015         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1016         int    err;
1017         ENTRY;
1018
1019         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1020
1021         OBD_ALLOC_PTR(cfg);
1022         if (cfg == NULL)
1023                 RETURN(-ENOMEM);
1024
1025         try_module_get(THIS_MODULE);
1026
1027         /* client additional sb info */
1028         lsi->lsi_llsbi = sbi = ll_init_sbi();
1029         if (!sbi) {
1030                 module_put(THIS_MODULE);
1031                 OBD_FREE_PTR(cfg);
1032                 RETURN(-ENOMEM);
1033         }
1034
1035         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1036         if (err)
1037                 GOTO(out_free, err);
1038
1039         err = bdi_init(&lsi->lsi_bdi);
1040         if (err)
1041                 GOTO(out_free, err);
1042         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1043         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1044         err = ll_bdi_register(&lsi->lsi_bdi);
1045         if (err)
1046                 GOTO(out_free, err);
1047
1048         sb->s_bdi = &lsi->lsi_bdi;
1049 #ifndef HAVE_DCACHE_LOCK
1050         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1051         sb->s_d_op = &ll_d_ops;
1052 #endif
1053
1054         /* Generate a string unique to this super, in case some joker tries
1055            to mount the same fs at two mount points.
1056            Use the address of the super itself.*/
1057         cfg->cfg_instance = sb;
1058         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1059         cfg->cfg_callback = class_config_llog_handler;
1060         /* set up client obds */
1061         err = lustre_process_log(sb, profilenm, cfg);
1062         if (err < 0)
1063                 GOTO(out_free, err);
1064
1065         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1066         lprof = class_get_profile(profilenm);
1067         if (lprof == NULL) {
1068                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1069                                    " read from the MGS.  Does that filesystem "
1070                                    "exist?\n", profilenm);
1071                 GOTO(out_free, err = -EINVAL);
1072         }
1073         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1074                lprof->lp_md, lprof->lp_dt);
1075
1076         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1077         if (!dt)
1078                 GOTO(out_free, err = -ENOMEM);
1079         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1080
1081         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1082         if (!md)
1083                 GOTO(out_free, err = -ENOMEM);
1084         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1085
1086         /* connections, registrations, sb setup */
1087         err = client_common_fill_super(sb, md, dt, mnt);
1088         if (err < 0)
1089                 GOTO(out_free, err);
1090
1091         sbi->ll_client_common_fill_super_succeeded = 1;
1092
1093 out_free:
1094         if (md)
1095                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1096         if (dt)
1097                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1098         if (err)
1099                 ll_put_super(sb);
1100         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1101                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1102
1103         OBD_FREE_PTR(cfg);
1104         RETURN(err);
1105 } /* ll_fill_super */
1106
1107 void ll_put_super(struct super_block *sb)
1108 {
1109         struct config_llog_instance cfg, params_cfg;
1110         struct obd_device *obd;
1111         struct lustre_sb_info *lsi = s2lsi(sb);
1112         struct ll_sb_info *sbi = ll_s2sbi(sb);
1113         char *profilenm = get_profile_name(sb);
1114         long ccc_count;
1115         int next, force = 1, rc = 0;
1116         ENTRY;
1117
1118         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1119
1120         ll_print_capa_stat(sbi);
1121
1122         cfg.cfg_instance = sb;
1123         lustre_end_log(sb, profilenm, &cfg);
1124
1125         params_cfg.cfg_instance = sb;
1126         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1127
1128         if (sbi->ll_md_exp) {
1129                 obd = class_exp2obd(sbi->ll_md_exp);
1130                 if (obd)
1131                         force = obd->obd_force;
1132         }
1133
1134         /* Wait for unstable pages to be committed to stable storage */
1135         if (force == 0) {
1136                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1137                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1138                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1139                         &lwi);
1140         }
1141
1142         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1143         if (force == 0 && rc != -EINTR)
1144                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1145
1146
1147         /* We need to set force before the lov_disconnect in
1148            lustre_common_put_super, since l_d cleans up osc's as well. */
1149         if (force) {
1150                 next = 0;
1151                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1152                                                      &next)) != NULL) {
1153                         obd->obd_force = force;
1154                 }
1155         }
1156
1157         if (sbi->ll_client_common_fill_super_succeeded) {
1158                 /* Only if client_common_fill_super succeeded */
1159                 client_common_put_super(sb);
1160         }
1161
1162         next = 0;
1163         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1164                 class_manual_cleanup(obd);
1165         }
1166
1167         if (sbi->ll_flags & LL_SBI_VERBOSE)
1168                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1169
1170         if (profilenm)
1171                 class_del_profile(profilenm);
1172
1173         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1174                 bdi_destroy(&lsi->lsi_bdi);
1175                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1176         }
1177
1178         ll_free_sbi(sb);
1179         lsi->lsi_llsbi = NULL;
1180
1181         lustre_common_put_super(sb);
1182
1183         cl_env_cache_purge(~0);
1184
1185         module_put(THIS_MODULE);
1186
1187         EXIT;
1188 } /* client_put_super */
1189
1190 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1191 {
1192         struct inode *inode = NULL;
1193
1194         /* NOTE: we depend on atomic igrab() -bzzz */
1195         lock_res_and_lock(lock);
1196         if (lock->l_resource->lr_lvb_inode) {
1197                 struct ll_inode_info * lli;
1198                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1199                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1200                         inode = igrab(lock->l_resource->lr_lvb_inode);
1201                 } else {
1202                         inode = lock->l_resource->lr_lvb_inode;
1203                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1204                                          D_WARNING, lock, "lr_lvb_inode %p is "
1205                                          "bogus: magic %08x",
1206                                          lock->l_resource->lr_lvb_inode,
1207                                          lli->lli_inode_magic);
1208                         inode = NULL;
1209                 }
1210         }
1211         unlock_res_and_lock(lock);
1212         return inode;
1213 }
1214
1215 static void ll_dir_clear_lsm_md(struct inode *inode)
1216 {
1217         struct ll_inode_info *lli = ll_i2info(inode);
1218
1219         LASSERT(S_ISDIR(inode->i_mode));
1220
1221         if (lli->lli_lsm_md != NULL) {
1222                 lmv_free_memmd(lli->lli_lsm_md);
1223                 lli->lli_lsm_md = NULL;
1224         }
1225 }
1226
1227 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1228                                       const struct lu_fid *fid,
1229                                       struct lustre_md *md)
1230 {
1231         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1232         struct mdt_body         *body = md->body;
1233         struct inode            *inode;
1234         ino_t                   ino;
1235         ENTRY;
1236
1237         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1238         inode = iget_locked(sb, ino);
1239         if (inode == NULL) {
1240                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1241                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1242                 RETURN(ERR_PTR(-ENOENT));
1243         }
1244
1245         if (inode->i_state & I_NEW) {
1246                 struct ll_inode_info *lli = ll_i2info(inode);
1247                 struct lmv_stripe_md *lsm = md->lmv;
1248
1249                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1250                                 (body->mbo_mode & S_IFMT);
1251                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1252                          PFID(fid));
1253
1254                 LTIME_S(inode->i_mtime) = 0;
1255                 LTIME_S(inode->i_atime) = 0;
1256                 LTIME_S(inode->i_ctime) = 0;
1257                 inode->i_rdev = 0;
1258
1259                 /* initializing backing dev info. */
1260                 inode->i_mapping->backing_dev_info =
1261                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1262                 inode->i_op = &ll_dir_inode_operations;
1263                 inode->i_fop = &ll_dir_operations;
1264                 lli->lli_fid = *fid;
1265                 ll_lli_init(lli);
1266
1267                 LASSERT(lsm != NULL);
1268                 /* master object FID */
1269                 lli->lli_pfid = body->mbo_fid1;
1270                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1271                        lli, PFID(fid), PFID(&lli->lli_pfid));
1272                 unlock_new_inode(inode);
1273         }
1274
1275         RETURN(inode);
1276 }
1277
1278 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1279 {
1280         struct lu_fid *fid;
1281         struct lmv_stripe_md *lsm = md->lmv;
1282         int i;
1283
1284         LASSERT(lsm != NULL);
1285         /* XXX sigh, this lsm_root initialization should be in
1286          * LMV layer, but it needs ll_iget right now, so we
1287          * put this here right now. */
1288         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1289                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1290                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1291                 /* Unfortunately ll_iget will call ll_update_inode,
1292                  * where the initialization of slave inode is slightly
1293                  * different, so it reset lsm_md to NULL to avoid
1294                  * initializing lsm for slave inode. */
1295                 /* For migrating inode, master stripe and master object will
1296                  * be same, so we only need assign this inode */
1297                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1298                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1299                 else
1300                         lsm->lsm_md_oinfo[i].lmo_root =
1301                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1302
1303                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1304                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1305
1306                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1307                         return rc;
1308                 }
1309         }
1310
1311         return 0;
1312 }
1313
1314 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1315                                 const struct lmv_stripe_md *lsm_md2)
1316 {
1317         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1318                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1319                lsm_md1->lsm_md_master_mdt_index ==
1320                                         lsm_md2->lsm_md_master_mdt_index &&
1321                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1322                lsm_md1->lsm_md_layout_version ==
1323                                         lsm_md2->lsm_md_layout_version &&
1324                strcmp(lsm_md1->lsm_md_pool_name,
1325                       lsm_md2->lsm_md_pool_name) == 0;
1326 }
1327
1328 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1329 {
1330         struct ll_inode_info *lli = ll_i2info(inode);
1331         struct lmv_stripe_md *lsm = md->lmv;
1332         int     rc;
1333         ENTRY;
1334
1335         LASSERT(S_ISDIR(inode->i_mode));
1336         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1337                PFID(ll_inode2fid(inode)));
1338
1339         /* no striped information from request. */
1340         if (lsm == NULL) {
1341                 if (lli->lli_lsm_md == NULL) {
1342                         RETURN(0);
1343                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1344                                                 LMV_HASH_FLAG_MIGRATION) {
1345                         /* migration is done, the temporay MIGRATE layout has
1346                          * been removed */
1347                         CDEBUG(D_INODE, DFID" finish migration.\n",
1348                                PFID(ll_inode2fid(inode)));
1349                         lmv_free_memmd(lli->lli_lsm_md);
1350                         lli->lli_lsm_md = NULL;
1351                         RETURN(0);
1352                 } else {
1353                         /* The lustre_md from req does not include stripeEA,
1354                          * see ll_md_setattr */
1355                         RETURN(0);
1356                 }
1357         }
1358
1359         /* set the directory layout */
1360         if (lli->lli_lsm_md == NULL) {
1361
1362                 rc = ll_init_lsm_md(inode, md);
1363                 if (rc != 0)
1364                         RETURN(rc);
1365
1366                 lli->lli_lsm_md = lsm;
1367                 /* set lsm_md to NULL, so the following free lustre_md
1368                  * will not free this lsm */
1369                 md->lmv = NULL;
1370                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1371                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1372                 RETURN(0);
1373         }
1374
1375         /* Compare the old and new stripe information */
1376         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1377                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1378                 int                     idx;
1379
1380                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1381                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1382                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1383                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1384                        inode, lsm, old_lsm,
1385                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1386                        lsm->lsm_md_stripe_count,
1387                        old_lsm->lsm_md_stripe_count,
1388                        lsm->lsm_md_master_mdt_index,
1389                        old_lsm->lsm_md_master_mdt_index,
1390                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1391                        lsm->lsm_md_layout_version,
1392                        old_lsm->lsm_md_layout_version,
1393                        lsm->lsm_md_pool_name,
1394                        old_lsm->lsm_md_pool_name);
1395
1396                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1397                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1398                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1399                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1400                 }
1401
1402                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1403                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1404                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1405                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1406                 }
1407
1408                 RETURN(-EIO);
1409         }
1410
1411         RETURN(0);
1412 }
1413
1414 void ll_clear_inode(struct inode *inode)
1415 {
1416         struct ll_inode_info *lli = ll_i2info(inode);
1417         struct ll_sb_info *sbi = ll_i2sbi(inode);
1418         ENTRY;
1419
1420         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1421                PFID(ll_inode2fid(inode)), inode);
1422
1423         if (S_ISDIR(inode->i_mode)) {
1424                 /* these should have been cleared in ll_file_release */
1425                 LASSERT(lli->lli_opendir_key == NULL);
1426                 LASSERT(lli->lli_sai == NULL);
1427                 LASSERT(lli->lli_opendir_pid == 0);
1428         }
1429
1430         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1431
1432         LASSERT(!lli->lli_open_fd_write_count);
1433         LASSERT(!lli->lli_open_fd_read_count);
1434         LASSERT(!lli->lli_open_fd_exec_count);
1435
1436         if (lli->lli_mds_write_och)
1437                 ll_md_real_close(inode, FMODE_WRITE);
1438         if (lli->lli_mds_exec_och)
1439                 ll_md_real_close(inode, FMODE_EXEC);
1440         if (lli->lli_mds_read_och)
1441                 ll_md_real_close(inode, FMODE_READ);
1442
1443         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1444                 OBD_FREE(lli->lli_symlink_name,
1445                          strlen(lli->lli_symlink_name) + 1);
1446                 lli->lli_symlink_name = NULL;
1447         }
1448
1449         ll_xattr_cache_destroy(inode);
1450
1451         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1452                 LASSERT(lli->lli_posix_acl == NULL);
1453                 if (lli->lli_remote_perms) {
1454                         free_rmtperm_hash(lli->lli_remote_perms);
1455                         lli->lli_remote_perms = NULL;
1456                 }
1457         }
1458 #ifdef CONFIG_FS_POSIX_ACL
1459         else if (lli->lli_posix_acl) {
1460                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1461                 LASSERT(lli->lli_remote_perms == NULL);
1462                 posix_acl_release(lli->lli_posix_acl);
1463                 lli->lli_posix_acl = NULL;
1464         }
1465 #endif
1466         lli->lli_inode_magic = LLI_INODE_DEAD;
1467
1468         ll_clear_inode_capas(inode);
1469         if (S_ISDIR(inode->i_mode))
1470                 ll_dir_clear_lsm_md(inode);
1471         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1472                 LASSERT(list_empty(&lli->lli_agl_list));
1473
1474         /*
1475          * XXX This has to be done before lsm is freed below, because
1476          * cl_object still uses inode lsm.
1477          */
1478         cl_inode_fini(inode);
1479
1480         EXIT;
1481 }
1482
1483 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1484 {
1485         struct lustre_md md;
1486         struct inode *inode = dentry->d_inode;
1487         struct ll_sb_info *sbi = ll_i2sbi(inode);
1488         struct ptlrpc_request *request = NULL;
1489         int rc, ia_valid;
1490         ENTRY;
1491
1492         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1493                                      LUSTRE_OPC_ANY, NULL);
1494         if (IS_ERR(op_data))
1495                 RETURN(PTR_ERR(op_data));
1496
1497         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1498         if (rc) {
1499                 ptlrpc_req_finished(request);
1500                 if (rc == -ENOENT) {
1501                         clear_nlink(inode);
1502                         /* Unlinked special device node? Or just a race?
1503                          * Pretend we done everything. */
1504                         if (!S_ISREG(inode->i_mode) &&
1505                             !S_ISDIR(inode->i_mode)) {
1506                                 ia_valid = op_data->op_attr.ia_valid;
1507                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1508                                 rc = simple_setattr(dentry, &op_data->op_attr);
1509                                 op_data->op_attr.ia_valid = ia_valid;
1510                         }
1511                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1512                         CERROR("md_setattr fails: rc = %d\n", rc);
1513                 }
1514                 RETURN(rc);
1515         }
1516
1517         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1518                               sbi->ll_md_exp, &md);
1519         if (rc) {
1520                 ptlrpc_req_finished(request);
1521                 RETURN(rc);
1522         }
1523
1524         ia_valid = op_data->op_attr.ia_valid;
1525         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1526          * cache is not cleared yet. */
1527         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1528         rc = simple_setattr(dentry, &op_data->op_attr);
1529         op_data->op_attr.ia_valid = ia_valid;
1530
1531         rc = ll_update_inode(inode, &md);
1532         ptlrpc_req_finished(request);
1533
1534         RETURN(rc);
1535 }
1536
1537 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1538 {
1539         struct obd_capa *capa;
1540         int rc;
1541
1542         if (attr->ia_valid & ATTR_SIZE)
1543                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1544         else
1545                 capa = ll_mdscapa_get(inode);
1546
1547         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1548
1549         if (attr->ia_valid & ATTR_SIZE)
1550                 ll_truncate_free_capa(capa);
1551         else
1552                 capa_put(capa);
1553
1554         return rc;
1555 }
1556
1557 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1558  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1559  * keep these values until such a time that objects are allocated for it.
1560  * We do the MDS operations first, as it is checking permissions for us.
1561  * We don't to the MDS RPC if there is nothing that we want to store there,
1562  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1563  * going to do an RPC anyways.
1564  *
1565  * If we are doing a truncate, we will send the mtime and ctime updates
1566  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1567  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1568  * at the same time.
1569  *
1570  * In case of HSMimport, we only set attr on MDS.
1571  */
1572 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1573 {
1574         struct inode *inode = dentry->d_inode;
1575         struct ll_inode_info *lli = ll_i2info(inode);
1576         struct md_op_data *op_data = NULL;
1577         bool file_is_released = false;
1578         int rc = 0;
1579         ENTRY;
1580
1581         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1582                "valid %x, hsm_import %d\n",
1583                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1584                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1585                hsm_import);
1586
1587         if (attr->ia_valid & ATTR_SIZE) {
1588                 /* Check new size against VFS/VM file size limit and rlimit */
1589                 rc = inode_newsize_ok(inode, attr->ia_size);
1590                 if (rc)
1591                         RETURN(rc);
1592
1593                 /* The maximum Lustre file size is variable, based on the
1594                  * OST maximum object size and number of stripes.  This
1595                  * needs another check in addition to the VFS check above. */
1596                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1597                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1598                                PFID(&lli->lli_fid), attr->ia_size,
1599                                ll_file_maxbytes(inode));
1600                         RETURN(-EFBIG);
1601                 }
1602
1603                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1604         }
1605
1606         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1607         if (attr->ia_valid & TIMES_SET_FLAGS) {
1608                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1609                     !cfs_capable(CFS_CAP_FOWNER))
1610                         RETURN(-EPERM);
1611         }
1612
1613         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1614         if (attr->ia_valid & ATTR_CTIME) {
1615                 attr->ia_ctime = CFS_CURRENT_TIME;
1616                 attr->ia_valid |= ATTR_CTIME_SET;
1617         }
1618         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1619             (attr->ia_valid & ATTR_ATIME)) {
1620                 attr->ia_atime = CFS_CURRENT_TIME;
1621                 attr->ia_valid |= ATTR_ATIME_SET;
1622         }
1623         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1624             (attr->ia_valid & ATTR_MTIME)) {
1625                 attr->ia_mtime = CFS_CURRENT_TIME;
1626                 attr->ia_valid |= ATTR_MTIME_SET;
1627         }
1628
1629         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1630                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1631                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1632                        cfs_time_current_sec());
1633
1634         /* We always do an MDS RPC, even if we're only changing the size;
1635          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1636
1637         OBD_ALLOC_PTR(op_data);
1638         if (op_data == NULL)
1639                 RETURN(-ENOMEM);
1640
1641         if (!S_ISDIR(inode->i_mode)) {
1642                 if (attr->ia_valid & ATTR_SIZE)
1643                         inode_dio_write_done(inode);
1644                 mutex_unlock(&inode->i_mutex);
1645         }
1646
1647         /* truncate on a released file must failed with -ENODATA,
1648          * so size must not be set on MDS for released file
1649          * but other attributes must be set
1650          */
1651         if (S_ISREG(inode->i_mode)) {
1652                 struct cl_layout cl = {
1653                         .cl_is_released = false,
1654                 };
1655                 struct lu_env *env;
1656                 int refcheck;
1657                 __u32 gen;
1658
1659                 rc = ll_layout_refresh(inode, &gen);
1660                 if (rc < 0)
1661                         GOTO(out, rc);
1662
1663                 /* XXX: the only place we need to know the layout type,
1664                  * this will be removed by a later patch. -Jinshan */
1665                 env = cl_env_get(&refcheck);
1666                 if (IS_ERR(env))
1667                         GOTO(out, rc = PTR_ERR(env));
1668
1669                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1670                 cl_env_put(env, &refcheck);
1671                 if (rc < 0)
1672                         GOTO(out, rc);
1673
1674                 file_is_released = cl.cl_is_released;
1675
1676                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1677                         if (file_is_released) {
1678                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1679                                 if (rc < 0)
1680                                         GOTO(out, rc);
1681
1682                                 file_is_released = false;
1683                                 ll_layout_refresh(inode, &gen);
1684                         }
1685
1686                         /* If we are changing file size, file content is
1687                          * modified, flag it. */
1688                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1689                         spin_lock(&lli->lli_lock);
1690                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1691                         spin_unlock(&lli->lli_lock);
1692                         op_data->op_bias |= MDS_DATA_MODIFIED;
1693                 }
1694         }
1695
1696         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1697
1698         rc = ll_md_setattr(dentry, op_data);
1699         if (rc)
1700                 GOTO(out, rc);
1701
1702         /* RPC to MDT is sent, cancel data modification flag */
1703         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1704                 spin_lock(&lli->lli_lock);
1705                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1706                 spin_unlock(&lli->lli_lock);
1707         }
1708
1709         if (!S_ISREG(inode->i_mode) || file_is_released)
1710                 GOTO(out, rc = 0);
1711
1712         if (attr->ia_valid & (ATTR_SIZE |
1713                               ATTR_ATIME | ATTR_ATIME_SET |
1714                               ATTR_MTIME | ATTR_MTIME_SET)) {
1715                 /* For truncate and utimes sending attributes to OSTs, setting
1716                  * mtime/atime to the past will be performed under PW [0:EOF]
1717                  * extent lock (new_size:EOF for truncate).  It may seem
1718                  * excessive to send mtime/atime updates to OSTs when not
1719                  * setting times to past, but it is necessary due to possible
1720                  * time de-synchronization between MDT inode and OST objects */
1721                 if (attr->ia_valid & ATTR_SIZE)
1722                         down_write(&lli->lli_trunc_sem);
1723                 rc = ll_setattr_ost(inode, attr);
1724                 if (attr->ia_valid & ATTR_SIZE)
1725                         up_write(&lli->lli_trunc_sem);
1726         }
1727         EXIT;
1728 out:
1729         if (op_data != NULL)
1730                 ll_finish_md_op_data(op_data);
1731
1732         if (!S_ISDIR(inode->i_mode)) {
1733                 mutex_lock(&inode->i_mutex);
1734                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1735                         inode_dio_wait(inode);
1736         }
1737
1738         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1739                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1740
1741         return rc;
1742 }
1743
1744 int ll_setattr(struct dentry *de, struct iattr *attr)
1745 {
1746         int mode = de->d_inode->i_mode;
1747
1748         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1749                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1750                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1751
1752         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1753                                (ATTR_SIZE|ATTR_MODE)) &&
1754             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1755              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1756               !(attr->ia_mode & S_ISGID))))
1757                 attr->ia_valid |= ATTR_FORCE;
1758
1759         if ((attr->ia_valid & ATTR_MODE) &&
1760             (mode & S_ISUID) &&
1761             !(attr->ia_mode & S_ISUID) &&
1762             !(attr->ia_valid & ATTR_KILL_SUID))
1763                 attr->ia_valid |= ATTR_KILL_SUID;
1764
1765         if ((attr->ia_valid & ATTR_MODE) &&
1766             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1767             !(attr->ia_mode & S_ISGID) &&
1768             !(attr->ia_valid & ATTR_KILL_SGID))
1769                 attr->ia_valid |= ATTR_KILL_SGID;
1770
1771         return ll_setattr_raw(de, attr, false);
1772 }
1773
1774 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1775                        __u64 max_age, __u32 flags)
1776 {
1777         struct ll_sb_info *sbi = ll_s2sbi(sb);
1778         struct obd_statfs obd_osfs;
1779         int rc;
1780         ENTRY;
1781
1782         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1783         if (rc) {
1784                 CERROR("md_statfs fails: rc = %d\n", rc);
1785                 RETURN(rc);
1786         }
1787
1788         osfs->os_type = sb->s_magic;
1789
1790         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1791                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1792
1793         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1794                 flags |= OBD_STATFS_NODELAY;
1795
1796         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1797         if (rc) {
1798                 CERROR("obd_statfs fails: rc = %d\n", rc);
1799                 RETURN(rc);
1800         }
1801
1802         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1803                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1804                obd_osfs.os_files);
1805
1806         osfs->os_bsize = obd_osfs.os_bsize;
1807         osfs->os_blocks = obd_osfs.os_blocks;
1808         osfs->os_bfree = obd_osfs.os_bfree;
1809         osfs->os_bavail = obd_osfs.os_bavail;
1810
1811         /* If we don't have as many objects free on the OST as inodes
1812          * on the MDS, we reduce the total number of inodes to
1813          * compensate, so that the "inodes in use" number is correct.
1814          */
1815         if (obd_osfs.os_ffree < osfs->os_ffree) {
1816                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1817                         obd_osfs.os_ffree;
1818                 osfs->os_ffree = obd_osfs.os_ffree;
1819         }
1820
1821         RETURN(rc);
1822 }
1823 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1824 {
1825         struct super_block *sb = de->d_sb;
1826         struct obd_statfs osfs;
1827         __u64 fsid = huge_encode_dev(sb->s_dev);
1828         int rc;
1829
1830         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1831         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1832
1833         /* Some amount of caching on the client is allowed */
1834         rc = ll_statfs_internal(sb, &osfs,
1835                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1836                                 0);
1837         if (rc)
1838                 return rc;
1839
1840         statfs_unpack(sfs, &osfs);
1841
1842         /* We need to downshift for all 32-bit kernels, because we can't
1843          * tell if the kernel is being called via sys_statfs64() or not.
1844          * Stop before overflowing f_bsize - in which case it is better
1845          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1846         if (sizeof(long) < 8) {
1847                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1848                         sfs->f_bsize <<= 1;
1849
1850                         osfs.os_blocks >>= 1;
1851                         osfs.os_bfree >>= 1;
1852                         osfs.os_bavail >>= 1;
1853                 }
1854         }
1855
1856         sfs->f_blocks = osfs.os_blocks;
1857         sfs->f_bfree = osfs.os_bfree;
1858         sfs->f_bavail = osfs.os_bavail;
1859         sfs->f_fsid.val[0] = (__u32)fsid;
1860         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1861         return 0;
1862 }
1863
1864 void ll_inode_size_lock(struct inode *inode)
1865 {
1866         struct ll_inode_info *lli;
1867
1868         LASSERT(!S_ISDIR(inode->i_mode));
1869
1870         lli = ll_i2info(inode);
1871         mutex_lock(&lli->lli_size_mutex);
1872 }
1873
1874 void ll_inode_size_unlock(struct inode *inode)
1875 {
1876         struct ll_inode_info *lli;
1877
1878         lli = ll_i2info(inode);
1879         mutex_unlock(&lli->lli_size_mutex);
1880 }
1881
1882 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1883 {
1884         struct ll_inode_info *lli = ll_i2info(inode);
1885         struct mdt_body *body = md->body;
1886         struct lov_stripe_md *lsm = md->lsm;
1887         struct ll_sb_info *sbi = ll_i2sbi(inode);
1888
1889         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1890         if (lsm != NULL) {
1891                 cl_file_inode_init(inode, md);
1892
1893                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1894                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1895                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1896         }
1897
1898         if (S_ISDIR(inode->i_mode)) {
1899                 int     rc;
1900
1901                 rc = ll_update_lsm_md(inode, md);
1902                 if (rc != 0)
1903                         return rc;
1904         }
1905
1906         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1907                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1908                         ll_update_remote_perm(inode, md->remote_perm);
1909         }
1910 #ifdef CONFIG_FS_POSIX_ACL
1911         else if (body->mbo_valid & OBD_MD_FLACL) {
1912                 spin_lock(&lli->lli_lock);
1913                 if (lli->lli_posix_acl)
1914                         posix_acl_release(lli->lli_posix_acl);
1915                 lli->lli_posix_acl = md->posix_acl;
1916                 spin_unlock(&lli->lli_lock);
1917         }
1918 #endif
1919         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1920                                         sbi->ll_flags & LL_SBI_32BIT_API);
1921         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1922
1923         if (body->mbo_valid & OBD_MD_FLATIME) {
1924                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1925                         LTIME_S(inode->i_atime) = body->mbo_atime;
1926                 lli->lli_atime = body->mbo_atime;
1927         }
1928
1929         if (body->mbo_valid & OBD_MD_FLMTIME) {
1930                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1931                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1932                                "to "LPU64"\n", inode->i_ino,
1933                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1934                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1935                 }
1936                 lli->lli_mtime = body->mbo_mtime;
1937         }
1938
1939         if (body->mbo_valid & OBD_MD_FLCTIME) {
1940                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1941                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1942                 lli->lli_ctime = body->mbo_ctime;
1943         }
1944
1945         if (body->mbo_valid & OBD_MD_FLMODE)
1946                 inode->i_mode = (inode->i_mode & S_IFMT) |
1947                                 (body->mbo_mode & ~S_IFMT);
1948
1949         if (body->mbo_valid & OBD_MD_FLTYPE)
1950                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1951                                 (body->mbo_mode & S_IFMT);
1952
1953         LASSERT(inode->i_mode != 0);
1954         if (S_ISREG(inode->i_mode))
1955                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1956                                        LL_MAX_BLKSIZE_BITS);
1957         else
1958                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1959
1960         if (body->mbo_valid & OBD_MD_FLUID)
1961                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1962         if (body->mbo_valid & OBD_MD_FLGID)
1963                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1964         if (body->mbo_valid & OBD_MD_FLFLAGS)
1965                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1966         if (body->mbo_valid & OBD_MD_FLNLINK)
1967                 set_nlink(inode, body->mbo_nlink);
1968         if (body->mbo_valid & OBD_MD_FLRDEV)
1969                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1970
1971         if (body->mbo_valid & OBD_MD_FLID) {
1972                 /* FID shouldn't be changed! */
1973                 if (fid_is_sane(&lli->lli_fid)) {
1974                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1975                                  "Trying to change FID "DFID
1976                                  " to the "DFID", inode "DFID"(%p)\n",
1977                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1978                                  PFID(ll_inode2fid(inode)), inode);
1979                 } else {
1980                         lli->lli_fid = body->mbo_fid1;
1981                 }
1982         }
1983
1984         LASSERT(fid_seq(&lli->lli_fid) != 0);
1985
1986         if (body->mbo_valid & OBD_MD_FLSIZE) {
1987                 i_size_write(inode, body->mbo_size);
1988
1989                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1990                        PFID(ll_inode2fid(inode)),
1991                        (unsigned long long)body->mbo_size);
1992
1993                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1994                         inode->i_blocks = body->mbo_blocks;
1995         }
1996
1997         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
1998                 LASSERT(md->mds_capa);
1999                 ll_add_capa(inode, md->mds_capa);
2000         }
2001
2002         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2003                 LASSERT(md->oss_capa);
2004                 ll_add_capa(inode, md->oss_capa);
2005         }
2006
2007         if (body->mbo_valid & OBD_MD_TSTATE) {
2008                 if (body->mbo_t_state & MS_RESTORE)
2009                         lli->lli_flags |= LLIF_FILE_RESTORING;
2010         }
2011
2012         return 0;
2013 }
2014
2015 int ll_read_inode2(struct inode *inode, void *opaque)
2016 {
2017         struct lustre_md *md = opaque;
2018         struct ll_inode_info *lli = ll_i2info(inode);
2019         int     rc;
2020         ENTRY;
2021
2022         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2023                PFID(&lli->lli_fid), inode);
2024
2025         /* Core attributes from the MDS first.  This is a new inode, and
2026          * the VFS doesn't zero times in the core inode so we have to do
2027          * it ourselves.  They will be overwritten by either MDS or OST
2028          * attributes - we just need to make sure they aren't newer. */
2029         LTIME_S(inode->i_mtime) = 0;
2030         LTIME_S(inode->i_atime) = 0;
2031         LTIME_S(inode->i_ctime) = 0;
2032         inode->i_rdev = 0;
2033         rc = ll_update_inode(inode, md);
2034         if (rc != 0)
2035                 RETURN(rc);
2036
2037         /* OIDEBUG(inode); */
2038
2039         /* initializing backing dev info. */
2040         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2041
2042
2043         if (S_ISREG(inode->i_mode)) {
2044                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2045                 inode->i_op = &ll_file_inode_operations;
2046                 inode->i_fop = sbi->ll_fop;
2047                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2048                 EXIT;
2049         } else if (S_ISDIR(inode->i_mode)) {
2050                 inode->i_op = &ll_dir_inode_operations;
2051                 inode->i_fop = &ll_dir_operations;
2052                 EXIT;
2053         } else if (S_ISLNK(inode->i_mode)) {
2054                 inode->i_op = &ll_fast_symlink_inode_operations;
2055                 EXIT;
2056         } else {
2057                 inode->i_op = &ll_special_inode_operations;
2058
2059                 init_special_inode(inode, inode->i_mode,
2060                                    inode->i_rdev);
2061
2062                 EXIT;
2063         }
2064
2065         return 0;
2066 }
2067
2068 void ll_delete_inode(struct inode *inode)
2069 {
2070         struct ll_inode_info *lli = ll_i2info(inode);
2071         ENTRY;
2072
2073         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2074                 /* It is last chance to write out dirty pages,
2075                  * otherwise we may lose data while umount */
2076                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2077
2078         truncate_inode_pages_final(&inode->i_data);
2079
2080         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2081                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2082                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2083
2084 #ifdef HAVE_SBOPS_EVICT_INODE
2085         ll_clear_inode(inode);
2086 #endif
2087         clear_inode(inode);
2088
2089         EXIT;
2090 }
2091
2092 int ll_iocontrol(struct inode *inode, struct file *file,
2093                  unsigned int cmd, unsigned long arg)
2094 {
2095         struct ll_sb_info *sbi = ll_i2sbi(inode);
2096         struct ptlrpc_request *req = NULL;
2097         int rc, flags = 0;
2098         ENTRY;
2099
2100         switch(cmd) {
2101         case FSFILT_IOC_GETFLAGS: {
2102                 struct mdt_body *body;
2103                 struct md_op_data *op_data;
2104
2105                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2106                                              0, 0, LUSTRE_OPC_ANY,
2107                                              NULL);
2108                 if (IS_ERR(op_data))
2109                         RETURN(PTR_ERR(op_data));
2110
2111                 op_data->op_valid = OBD_MD_FLFLAGS;
2112                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2113                 ll_finish_md_op_data(op_data);
2114                 if (rc) {
2115                         CERROR("%s: failure inode "DFID": rc = %d\n",
2116                                sbi->ll_md_exp->exp_obd->obd_name,
2117                                PFID(ll_inode2fid(inode)), rc);
2118                         RETURN(-abs(rc));
2119                 }
2120
2121                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2122
2123                 flags = body->mbo_flags;
2124
2125                 ptlrpc_req_finished(req);
2126
2127                 RETURN(put_user(flags, (int __user *)arg));
2128         }
2129         case FSFILT_IOC_SETFLAGS: {
2130                 struct iattr *attr;
2131                 struct md_op_data *op_data;
2132                 struct cl_object *obj;
2133                 struct obd_capa *capa;
2134
2135                 if (get_user(flags, (int __user *)arg))
2136                         RETURN(-EFAULT);
2137
2138                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2139                                              LUSTRE_OPC_ANY, NULL);
2140                 if (IS_ERR(op_data))
2141                         RETURN(PTR_ERR(op_data));
2142
2143                 op_data->op_attr_flags = flags;
2144                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2145                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2146                 ll_finish_md_op_data(op_data);
2147                 ptlrpc_req_finished(req);
2148                 if (rc)
2149                         RETURN(rc);
2150
2151                 inode->i_flags = ll_ext_to_inode_flags(flags);
2152
2153                 obj = ll_i2info(inode)->lli_clob;
2154                 if (obj == NULL)
2155                         RETURN(0);
2156
2157                 OBD_ALLOC_PTR(attr);
2158                 if (attr == NULL)
2159                         RETURN(-ENOMEM);
2160
2161                 attr->ia_valid = ATTR_ATTR_FLAG;
2162
2163                 capa = ll_mdscapa_get(inode);
2164                 rc = cl_setattr_ost(obj, attr, flags, capa);
2165                 capa_put(capa);
2166
2167                 OBD_FREE_PTR(attr);
2168                 RETURN(rc);
2169         }
2170         default:
2171                 RETURN(-ENOSYS);
2172         }
2173
2174         RETURN(0);
2175 }
2176
2177 int ll_flush_ctx(struct inode *inode)
2178 {
2179         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2180
2181         CDEBUG(D_SEC, "flush context for user %d\n",
2182                from_kuid(&init_user_ns, current_uid()));
2183
2184         obd_set_info_async(NULL, sbi->ll_md_exp,
2185                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2186                            0, NULL, NULL);
2187         obd_set_info_async(NULL, sbi->ll_dt_exp,
2188                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2189                            0, NULL, NULL);
2190         return 0;
2191 }
2192
2193 /* umount -f client means force down, don't save state */
2194 void ll_umount_begin(struct super_block *sb)
2195 {
2196         struct ll_sb_info *sbi = ll_s2sbi(sb);
2197         struct obd_device *obd;
2198         struct obd_ioctl_data *ioc_data;
2199         ENTRY;
2200
2201         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2202                sb->s_count, atomic_read(&sb->s_active));
2203
2204         obd = class_exp2obd(sbi->ll_md_exp);
2205         if (obd == NULL) {
2206                 CERROR("Invalid MDC connection handle "LPX64"\n",
2207                        sbi->ll_md_exp->exp_handle.h_cookie);
2208                 EXIT;
2209                 return;
2210         }
2211         obd->obd_force = 1;
2212
2213         obd = class_exp2obd(sbi->ll_dt_exp);
2214         if (obd == NULL) {
2215                 CERROR("Invalid LOV connection handle "LPX64"\n",
2216                        sbi->ll_dt_exp->exp_handle.h_cookie);
2217                 EXIT;
2218                 return;
2219         }
2220         obd->obd_force = 1;
2221
2222         OBD_ALLOC_PTR(ioc_data);
2223         if (ioc_data) {
2224                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2225                               sizeof *ioc_data, ioc_data, NULL);
2226
2227                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2228                               sizeof *ioc_data, ioc_data, NULL);
2229
2230                 OBD_FREE_PTR(ioc_data);
2231         }
2232
2233         /* Really, we'd like to wait until there are no requests outstanding,
2234          * and then continue.  For now, we just invalidate the requests,
2235          * schedule() and sleep one second if needed, and hope.
2236          */
2237         schedule();
2238         EXIT;
2239 }
2240
2241 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2242 {
2243         struct ll_sb_info *sbi = ll_s2sbi(sb);
2244         char *profilenm = get_profile_name(sb);
2245         int err;
2246         __u32 read_only;
2247
2248         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2249                 read_only = *flags & MS_RDONLY;
2250                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2251                                          sizeof(KEY_READ_ONLY),
2252                                          KEY_READ_ONLY, sizeof(read_only),
2253                                          &read_only, NULL);
2254                 if (err) {
2255                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2256                                       profilenm, read_only ?
2257                                       "read-only" : "read-write", err);
2258                         return err;
2259                 }
2260
2261                 if (read_only)
2262                         sb->s_flags |= MS_RDONLY;
2263                 else
2264                         sb->s_flags &= ~MS_RDONLY;
2265
2266                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2267                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2268                                       read_only ?  "read-only" : "read-write");
2269         }
2270         return 0;
2271 }
2272
2273 /**
2274  * Cleanup the open handle that is cached on MDT-side.
2275  *
2276  * For open case, the client side open handling thread may hit error
2277  * after the MDT grant the open. Under such case, the client should
2278  * send close RPC to the MDT as cleanup; otherwise, the open handle
2279  * on the MDT will be leaked there until the client umount or evicted.
2280  *
2281  * In further, if someone unlinked the file, because the open handle
2282  * holds the reference on such file/object, then it will block the
2283  * subsequent threads that want to locate such object via FID.
2284  *
2285  * \param[in] sb        super block for this file-system
2286  * \param[in] open_req  pointer to the original open request
2287  */
2288 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2289 {
2290         struct mdt_body                 *body;
2291         struct md_op_data               *op_data;
2292         struct ptlrpc_request           *close_req = NULL;
2293         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2294         ENTRY;
2295
2296         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2297         OBD_ALLOC_PTR(op_data);
2298         if (op_data == NULL) {
2299                 CWARN("%s: cannot allocate op_data to release open handle for "
2300                       DFID"\n",
2301                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2302
2303                 RETURN_EXIT;
2304         }
2305
2306         op_data->op_fid1 = body->mbo_fid1;
2307         op_data->op_handle = body->mbo_handle;
2308         op_data->op_mod_time = cfs_time_current_sec();
2309         md_close(exp, op_data, NULL, &close_req);
2310         ptlrpc_req_finished(close_req);
2311         ll_finish_md_op_data(op_data);
2312
2313         EXIT;
2314 }
2315
2316 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2317                   struct super_block *sb, struct lookup_intent *it)
2318 {
2319         struct ll_sb_info *sbi = NULL;
2320         struct lustre_md md = { NULL };
2321         int rc;
2322         ENTRY;
2323
2324         LASSERT(*inode || sb);
2325         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2326         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2327                               sbi->ll_md_exp, &md);
2328         if (rc != 0)
2329                 GOTO(cleanup, rc);
2330
2331         if (*inode) {
2332                 rc = ll_update_inode(*inode, &md);
2333                 if (rc != 0)
2334                         GOTO(out, rc);
2335         } else {
2336                 LASSERT(sb != NULL);
2337
2338                 /*
2339                  * At this point server returns to client's same fid as client
2340                  * generated for creating. So using ->fid1 is okay here.
2341                  */
2342                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2343
2344                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2345                                              sbi->ll_flags & LL_SBI_32BIT_API),
2346                                  &md);
2347                 if (IS_ERR(*inode)) {
2348 #ifdef CONFIG_FS_POSIX_ACL
2349                         if (md.posix_acl) {
2350                                 posix_acl_release(md.posix_acl);
2351                                 md.posix_acl = NULL;
2352                         }
2353 #endif
2354                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2355                         *inode = NULL;
2356                         CERROR("new_inode -fatal: rc %d\n", rc);
2357                         GOTO(out, rc);
2358                 }
2359         }
2360
2361         /* Handling piggyback layout lock.
2362          * Layout lock can be piggybacked by getattr and open request.
2363          * The lsm can be applied to inode only if it comes with a layout lock
2364          * otherwise correct layout may be overwritten, for example:
2365          * 1. proc1: mdt returns a lsm but not granting layout
2366          * 2. layout was changed by another client
2367          * 3. proc2: refresh layout and layout lock granted
2368          * 4. proc1: to apply a stale layout */
2369         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2370                 struct lustre_handle lockh;
2371                 struct ldlm_lock *lock;
2372
2373                 lockh.cookie = it->d.lustre.it_lock_handle;
2374                 lock = ldlm_handle2lock(&lockh);
2375                 LASSERT(lock != NULL);
2376                 if (ldlm_has_layout(lock)) {
2377                         struct cl_object_conf conf;
2378
2379                         memset(&conf, 0, sizeof(conf));
2380                         conf.coc_opc = OBJECT_CONF_SET;
2381                         conf.coc_inode = *inode;
2382                         conf.coc_lock = lock;
2383                         conf.u.coc_md = &md;
2384                         (void)ll_layout_conf(*inode, &conf);
2385                 }
2386                 LDLM_LOCK_PUT(lock);
2387         }
2388
2389         GOTO(out, rc = 0);
2390
2391 out:
2392         if (md.lsm != NULL)
2393                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2394         md_free_lustre_md(sbi->ll_md_exp, &md);
2395
2396 cleanup:
2397         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2398                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2399
2400         return rc;
2401 }
2402
2403 int ll_obd_statfs(struct inode *inode, void __user *arg)
2404 {
2405         struct ll_sb_info *sbi = NULL;
2406         struct obd_export *exp;
2407         char *buf = NULL;
2408         struct obd_ioctl_data *data = NULL;
2409         __u32 type;
2410         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2411         int len = 0, rc;
2412
2413         if (!inode || !(sbi = ll_i2sbi(inode)))
2414                 GOTO(out_statfs, rc = -EINVAL);
2415
2416         rc = obd_ioctl_getdata(&buf, &len, arg);
2417         if (rc)
2418                 GOTO(out_statfs, rc);
2419
2420         data = (void*)buf;
2421         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2422             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2423                 GOTO(out_statfs, rc = -EINVAL);
2424
2425         if (data->ioc_inllen1 != sizeof(__u32) ||
2426             data->ioc_inllen2 != sizeof(__u32) ||
2427             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2428             data->ioc_plen2 != sizeof(struct obd_uuid))
2429                 GOTO(out_statfs, rc = -EINVAL);
2430
2431         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2432         if (type & LL_STATFS_LMV)
2433                 exp = sbi->ll_md_exp;
2434         else if (type & LL_STATFS_LOV)
2435                 exp = sbi->ll_dt_exp;
2436         else
2437                 GOTO(out_statfs, rc = -ENODEV);
2438
2439         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2440         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2441         if (rc)
2442                 GOTO(out_statfs, rc);
2443 out_statfs:
2444         if (buf)
2445                 obd_ioctl_freedata(buf, len);
2446         return rc;
2447 }
2448
2449 int ll_process_config(struct lustre_cfg *lcfg)
2450 {
2451         struct super_block *sb;
2452         unsigned long x;
2453         int rc = 0;
2454         char *ptr;
2455
2456         /* The instance name contains the sb: lustre-client-aacfe000 */
2457         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2458         if (!ptr || !*(++ptr))
2459                 return -EINVAL;
2460         if (sscanf(ptr, "%lx", &x) != 1)
2461                 return -EINVAL;
2462         sb = (struct super_block *)x;
2463         /* This better be a real Lustre superblock! */
2464         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2465
2466         /* Note we have not called client_common_fill_super yet, so
2467            proc fns must be able to handle that! */
2468         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2469                                       lcfg, sb);
2470         if (rc > 0)
2471                 rc = 0;
2472         return rc;
2473 }
2474
2475 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2476 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2477                                        struct inode *i1, struct inode *i2,
2478                                        const char *name, size_t namelen,
2479                                        __u32 mode, __u32 opc, void *data)
2480 {
2481         LASSERT(i1 != NULL);
2482
2483         if (name == NULL) {
2484                 /* Do not reuse namelen for something else. */
2485                 if (namelen != 0)
2486                         return ERR_PTR(-EINVAL);
2487         } else {
2488                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2489                         return ERR_PTR(-ENAMETOOLONG);
2490
2491                 if (!lu_name_is_valid_2(name, namelen))
2492                         return ERR_PTR(-EINVAL);
2493         }
2494
2495         if (op_data == NULL)
2496                 OBD_ALLOC_PTR(op_data);
2497
2498         if (op_data == NULL)
2499                 return ERR_PTR(-ENOMEM);
2500
2501         ll_i2gids(op_data->op_suppgids, i1, i2);
2502         op_data->op_fid1 = *ll_inode2fid(i1);
2503         op_data->op_capa1 = ll_mdscapa_get(i1);
2504         op_data->op_default_stripe_offset = -1;
2505         if (S_ISDIR(i1->i_mode)) {
2506                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2507                 if (opc == LUSTRE_OPC_MKDIR)
2508                         op_data->op_default_stripe_offset =
2509                                    ll_i2info(i1)->lli_def_stripe_offset;
2510         }
2511
2512         if (i2) {
2513                 op_data->op_fid2 = *ll_inode2fid(i2);
2514                 op_data->op_capa2 = ll_mdscapa_get(i2);
2515                 if (S_ISDIR(i2->i_mode))
2516                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2517         } else {
2518                 fid_zero(&op_data->op_fid2);
2519                 op_data->op_capa2 = NULL;
2520         }
2521
2522         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2523                 op_data->op_cli_flags |= CLI_HASH64;
2524
2525         if (ll_need_32bit_api(ll_i2sbi(i1)))
2526                 op_data->op_cli_flags |= CLI_API32;
2527
2528         op_data->op_name = name;
2529         op_data->op_namelen = namelen;
2530         op_data->op_mode = mode;
2531         op_data->op_mod_time = cfs_time_current_sec();
2532         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2533         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2534         op_data->op_cap = cfs_curproc_cap_pack();
2535         op_data->op_bias = 0;
2536         op_data->op_cli_flags = 0;
2537         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2538              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2539                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2540         } else {
2541                 op_data->op_mds = 0;
2542         }
2543         op_data->op_data = data;
2544
2545         /* When called by ll_setattr_raw, file is i1. */
2546         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2547                 op_data->op_bias |= MDS_DATA_MODIFIED;
2548
2549         return op_data;
2550 }
2551
2552 void ll_finish_md_op_data(struct md_op_data *op_data)
2553 {
2554         capa_put(op_data->op_capa1);
2555         capa_put(op_data->op_capa2);
2556         OBD_FREE_PTR(op_data);
2557 }
2558
2559 #ifdef HAVE_SUPEROPS_USE_DENTRY
2560 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2561 #else
2562 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2563 #endif
2564 {
2565         struct ll_sb_info *sbi;
2566
2567 #ifdef HAVE_SUPEROPS_USE_DENTRY
2568         LASSERT((seq != NULL) && (dentry != NULL));
2569         sbi = ll_s2sbi(dentry->d_sb);
2570 #else
2571         LASSERT((seq != NULL) && (vfs != NULL));
2572         sbi = ll_s2sbi(vfs->mnt_sb);
2573 #endif
2574
2575         if (sbi->ll_flags & LL_SBI_NOLCK)
2576                 seq_puts(seq, ",nolock");
2577
2578         if (sbi->ll_flags & LL_SBI_FLOCK)
2579                 seq_puts(seq, ",flock");
2580
2581         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2582                 seq_puts(seq, ",localflock");
2583
2584         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2585                 seq_puts(seq, ",user_xattr");
2586
2587         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2588                 seq_puts(seq, ",lazystatfs");
2589
2590         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2591                 seq_puts(seq, ",user_fid2path");
2592
2593         RETURN(0);
2594 }
2595
2596 /**
2597  * Get obd name by cmd, and copy out to user space
2598  */
2599 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2600 {
2601         struct ll_sb_info *sbi = ll_i2sbi(inode);
2602         struct obd_device *obd;
2603         ENTRY;
2604
2605         if (cmd == OBD_IOC_GETDTNAME)
2606                 obd = class_exp2obd(sbi->ll_dt_exp);
2607         else if (cmd == OBD_IOC_GETMDNAME)
2608                 obd = class_exp2obd(sbi->ll_md_exp);
2609         else
2610                 RETURN(-EINVAL);
2611
2612         if (!obd)
2613                 RETURN(-ENOENT);
2614
2615         if (copy_to_user((void __user *)arg, obd->obd_name,
2616                          strlen(obd->obd_name) + 1))
2617                 RETURN(-EFAULT);
2618
2619         RETURN(0);
2620 }
2621
2622 /**
2623  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2624  * fsname will be returned in this buffer; otherwise, a static buffer will be
2625  * used to store the fsname and returned to caller.
2626  */
2627 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2628 {
2629         static char fsname_static[MTI_NAME_MAXLEN];
2630         struct lustre_sb_info *lsi = s2lsi(sb);
2631         char *ptr;
2632         int len;
2633
2634         if (buf == NULL) {
2635                 /* this means the caller wants to use static buffer
2636                  * and it doesn't care about race. Usually this is
2637                  * in error reporting path */
2638                 buf = fsname_static;
2639                 buflen = sizeof(fsname_static);
2640         }
2641
2642         len = strlen(lsi->lsi_lmd->lmd_profile);
2643         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2644         if (ptr && (strcmp(ptr, "-client") == 0))
2645                 len -= 7;
2646
2647         if (unlikely(len >= buflen))
2648                 len = buflen - 1;
2649         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2650         buf[len] = '\0';
2651
2652         return buf;
2653 }
2654
2655 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2656 {
2657         char *path = NULL;
2658
2659         struct path p;
2660
2661         p.dentry = dentry;
2662         p.mnt = current->fs->root.mnt;
2663         path_get(&p);
2664         path = d_path(&p, buf, bufsize);
2665         path_put(&p);
2666         return path;
2667 }
2668
2669 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2670 {
2671         char *buf, *path = NULL;
2672         struct dentry *dentry = NULL;
2673         struct inode *inode = page->mapping->host;
2674
2675         /* this can be called inside spin lock so use GFP_ATOMIC. */
2676         buf = (char *)__get_free_page(GFP_ATOMIC);
2677         if (buf != NULL) {
2678                 dentry = d_find_alias(page->mapping->host);
2679                 if (dentry != NULL)
2680                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2681         }
2682
2683         CDEBUG(D_WARNING,
2684                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2685                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2686                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2687                PFID(ll_inode2fid(inode)),
2688                (path && !IS_ERR(path)) ? path : "", ioret);
2689
2690         if (dentry != NULL)
2691                 dput(dentry);
2692
2693         if (buf != NULL)
2694                 free_page((unsigned long)buf);
2695 }
2696
2697 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2698                         struct lov_user_md **kbuf)
2699 {
2700         struct lov_user_md      lum;
2701         ssize_t                 lum_size;
2702         ENTRY;
2703
2704         if (copy_from_user(&lum, md, sizeof(lum)))
2705                 RETURN(-EFAULT);
2706
2707         lum_size = ll_lov_user_md_size(&lum);
2708         if (lum_size < 0)
2709                 RETURN(lum_size);
2710
2711         OBD_ALLOC(*kbuf, lum_size);
2712         if (*kbuf == NULL)
2713                 RETURN(-ENOMEM);
2714
2715         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2716                 OBD_FREE(*kbuf, lum_size);
2717                 RETURN(-EFAULT);
2718         }
2719
2720         RETURN(lum_size);
2721 }
2722
2723 /*
2724  * Compute llite root squash state after a change of root squash
2725  * configuration setting or add/remove of a lnet nid
2726  */
2727 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2728 {
2729         struct root_squash_info *squash = &sbi->ll_squash;
2730         int i;
2731         bool matched;
2732         lnet_process_id_t id;
2733
2734         /* Update norootsquash flag */
2735         down_write(&squash->rsi_sem);
2736         if (list_empty(&squash->rsi_nosquash_nids))
2737                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2738         else {
2739                 /* Do not apply root squash as soon as one of our NIDs is
2740                  * in the nosquash_nids list */
2741                 matched = false;
2742                 i = 0;
2743                 while (LNetGetId(i++, &id) != -ENOENT) {
2744                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2745                                 continue;
2746                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2747                                 matched = true;
2748                                 break;
2749                         }
2750                 }
2751                 if (matched)
2752                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2753                 else
2754                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2755         }
2756         up_write(&squash->rsi_sem);
2757 }
2758
2759 /**
2760  * Parse linkea content to extract information about a given hardlink
2761  *
2762  * \param[in]   ldata      - Initialized linkea data
2763  * \param[in]   linkno     - Link identifier
2764  * \param[out]  parent_fid - The entry's parent FID
2765  * \param[out]  ln         - Entry name destination buffer
2766  *
2767  * \retval 0 on success
2768  * \retval Appropriate negative error code on failure
2769  */
2770 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2771                             struct lu_fid *parent_fid, struct lu_name *ln)
2772 {
2773         unsigned int    idx;
2774         int             rc;
2775         ENTRY;
2776
2777         rc = linkea_init(ldata);
2778         if (rc < 0)
2779                 RETURN(rc);
2780
2781         if (linkno >= ldata->ld_leh->leh_reccount)
2782                 /* beyond last link */
2783                 RETURN(-ENODATA);
2784
2785         linkea_first_entry(ldata);
2786         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2787                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2788                                     parent_fid);
2789                 if (idx == linkno)
2790                         break;
2791
2792                 linkea_next_entry(ldata);
2793         }
2794
2795         if (idx < linkno)
2796                 RETURN(-ENODATA);
2797
2798         RETURN(0);
2799 }
2800
2801 /**
2802  * Get parent FID and name of an identified link. Operation is performed for
2803  * a given link number, letting the caller iterate over linkno to list one or
2804  * all links of an entry.
2805  *
2806  * \param[in]     file - File descriptor against which to perform the operation
2807  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2808  *                       on and the available size. It is eventually filled with
2809  *                       the requested information or left untouched on error
2810  *
2811  * \retval - 0 on success
2812  * \retval - Appropriate negative error code on failure
2813  */
2814 int ll_getparent(struct file *file, struct getparent __user *arg)
2815 {
2816         struct dentry           *dentry = file->f_dentry;
2817         struct inode            *inode = file->f_dentry->d_inode;
2818         struct linkea_data      *ldata;
2819         struct lu_buf            buf = LU_BUF_NULL;
2820         struct lu_name           ln;
2821         struct lu_fid            parent_fid;
2822         __u32                    linkno;
2823         __u32                    name_size;
2824         int                      rc;
2825
2826         ENTRY;
2827
2828         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2829             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2830                 RETURN(-EPERM);
2831
2832         if (get_user(name_size, &arg->gp_name_size))
2833                 RETURN(-EFAULT);
2834
2835         if (get_user(linkno, &arg->gp_linkno))
2836                 RETURN(-EFAULT);
2837
2838         if (name_size > PATH_MAX)
2839                 RETURN(-EINVAL);
2840
2841         OBD_ALLOC(ldata, sizeof(*ldata));
2842         if (ldata == NULL)
2843                 RETURN(-ENOMEM);
2844
2845         rc = linkea_data_new(ldata, &buf);
2846         if (rc < 0)
2847                 GOTO(ldata_free, rc);
2848
2849         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2850         if (rc < 0)
2851                 GOTO(lb_free, rc);
2852
2853         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2854         if (rc < 0)
2855                 GOTO(lb_free, rc);
2856
2857         if (ln.ln_namelen >= name_size)
2858                 GOTO(lb_free, rc = -EOVERFLOW);
2859
2860         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2861                 GOTO(lb_free, rc = -EFAULT);
2862
2863         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2864                 GOTO(lb_free, rc = -EFAULT);
2865
2866         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2867                 GOTO(lb_free, rc = -EFAULT);
2868
2869 lb_free:
2870         lu_buf_free(&buf);
2871 ldata_free:
2872         OBD_FREE(ldata, sizeof(*ldata));
2873
2874         RETURN(rc);
2875 }