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         lli->lli_has_smd = false;
 968         spin_lock_init(&lli->lli_layout_lock);
 969         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 970         lli->lli_clob = NULL;
 971
 972         init_rwsem(&lli->lli_xattrs_list_rwsem);
 973         mutex_init(&lli->lli_xattrs_enq_lock);
 974
 975         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 976         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 977                 mutex_init(&lli->lli_readdir_mutex);
 978                 lli->lli_opendir_key = NULL;
 979                 lli->lli_sai = NULL;
 980                 spin_lock_init(&lli->lli_sa_lock);
 981                 lli->lli_opendir_pid = 0;
 982                 lli->lli_sa_enabled = 0;
 983                 lli->lli_def_stripe_offset = -1;
 984         } else {
 985                 mutex_init(&lli->lli_size_mutex);
 986                 lli->lli_symlink_name = NULL;
 987                 init_rwsem(&lli->lli_trunc_sem);
 988                 range_lock_tree_init(&lli->lli_write_tree);
 989                 init_rwsem(&lli->lli_glimpse_sem);
 990                 lli->lli_glimpse_time = 0;
 991                 INIT_LIST_HEAD(&lli->lli_agl_list);
 992                 lli->lli_agl_index = 0;
 993                 lli->lli_async_rc = 0;
 994         }
 995         mutex_init(&lli->lli_layout_mutex);
 996 }
 997
 998 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 999 {
1000         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1001
1002         bdi->name = "lustre";
1003         return bdi_register(bdi, NULL, "lustre-%d",
1004                             atomic_inc_return(&ll_bdi_num));
1005 }
1006
1007 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1008 {
1009         struct lustre_profile *lprof = NULL;
1010         struct lustre_sb_info *lsi = s2lsi(sb);
1011         struct ll_sb_info *sbi;
1012         char  *dt = NULL, *md = NULL;
1013         char  *profilenm = get_profile_name(sb);
1014         struct config_llog_instance *cfg;
1015         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1016         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1017         int    err;
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1021
1022         OBD_ALLOC_PTR(cfg);
1023         if (cfg == NULL)
1024                 RETURN(-ENOMEM);
1025
1026         try_module_get(THIS_MODULE);
1027
1028         /* client additional sb info */
1029         lsi->lsi_llsbi = sbi = ll_init_sbi();
1030         if (!sbi) {
1031                 module_put(THIS_MODULE);
1032                 OBD_FREE_PTR(cfg);
1033                 RETURN(-ENOMEM);
1034         }
1035
1036         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1037         if (err)
1038                 GOTO(out_free, err);
1039
1040         err = bdi_init(&lsi->lsi_bdi);
1041         if (err)
1042                 GOTO(out_free, err);
1043         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1044         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1045         err = ll_bdi_register(&lsi->lsi_bdi);
1046         if (err)
1047                 GOTO(out_free, err);
1048
1049         sb->s_bdi = &lsi->lsi_bdi;
1050 #ifndef HAVE_DCACHE_LOCK
1051         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1052         sb->s_d_op = &ll_d_ops;
1053 #endif
1054
1055         /* Generate a string unique to this super, in case some joker tries
1056            to mount the same fs at two mount points.
1057            Use the address of the super itself.*/
1058         cfg->cfg_instance = sb;
1059         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1060         cfg->cfg_callback = class_config_llog_handler;
1061         /* set up client obds */
1062         err = lustre_process_log(sb, profilenm, cfg);
1063         if (err < 0)
1064                 GOTO(out_free, err);
1065
1066         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1067         lprof = class_get_profile(profilenm);
1068         if (lprof == NULL) {
1069                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1070                                    " read from the MGS.  Does that filesystem "
1071                                    "exist?\n", profilenm);
1072                 GOTO(out_free, err = -EINVAL);
1073         }
1074         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1075                lprof->lp_md, lprof->lp_dt);
1076
1077         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1078         if (!dt)
1079                 GOTO(out_free, err = -ENOMEM);
1080         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1081
1082         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1083         if (!md)
1084                 GOTO(out_free, err = -ENOMEM);
1085         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1086
1087         /* connections, registrations, sb setup */
1088         err = client_common_fill_super(sb, md, dt, mnt);
1089         if (err < 0)
1090                 GOTO(out_free, err);
1091
1092         sbi->ll_client_common_fill_super_succeeded = 1;
1093
1094 out_free:
1095         if (md)
1096                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1097         if (dt)
1098                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1099         if (err)
1100                 ll_put_super(sb);
1101         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1102                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1103
1104         OBD_FREE_PTR(cfg);
1105         RETURN(err);
1106 } /* ll_fill_super */
1107
1108 void ll_put_super(struct super_block *sb)
1109 {
1110         struct config_llog_instance cfg, params_cfg;
1111         struct obd_device *obd;
1112         struct lustre_sb_info *lsi = s2lsi(sb);
1113         struct ll_sb_info *sbi = ll_s2sbi(sb);
1114         char *profilenm = get_profile_name(sb);
1115         long ccc_count;
1116         int next, force = 1, rc = 0;
1117         ENTRY;
1118
1119         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1120
1121         ll_print_capa_stat(sbi);
1122
1123         cfg.cfg_instance = sb;
1124         lustre_end_log(sb, profilenm, &cfg);
1125
1126         params_cfg.cfg_instance = sb;
1127         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1128
1129         if (sbi->ll_md_exp) {
1130                 obd = class_exp2obd(sbi->ll_md_exp);
1131                 if (obd)
1132                         force = obd->obd_force;
1133         }
1134
1135         /* Wait for unstable pages to be committed to stable storage */
1136         if (force == 0) {
1137                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1138                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1139                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1140                         &lwi);
1141         }
1142
1143         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1144         if (force == 0 && rc != -EINTR)
1145                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1146
1147
1148         /* We need to set force before the lov_disconnect in
1149            lustre_common_put_super, since l_d cleans up osc's as well. */
1150         if (force) {
1151                 next = 0;
1152                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1153                                                      &next)) != NULL) {
1154                         obd->obd_force = force;
1155                 }
1156         }
1157
1158         if (sbi->ll_client_common_fill_super_succeeded) {
1159                 /* Only if client_common_fill_super succeeded */
1160                 client_common_put_super(sb);
1161         }
1162
1163         next = 0;
1164         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1165                 class_manual_cleanup(obd);
1166         }
1167
1168         if (sbi->ll_flags & LL_SBI_VERBOSE)
1169                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1170
1171         if (profilenm)
1172                 class_del_profile(profilenm);
1173
1174         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1175                 bdi_destroy(&lsi->lsi_bdi);
1176                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1177         }
1178
1179         ll_free_sbi(sb);
1180         lsi->lsi_llsbi = NULL;
1181
1182         lustre_common_put_super(sb);
1183
1184         cl_env_cache_purge(~0);
1185
1186         module_put(THIS_MODULE);
1187
1188         EXIT;
1189 } /* client_put_super */
1190
1191 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1192 {
1193         struct inode *inode = NULL;
1194
1195         /* NOTE: we depend on atomic igrab() -bzzz */
1196         lock_res_and_lock(lock);
1197         if (lock->l_resource->lr_lvb_inode) {
1198                 struct ll_inode_info * lli;
1199                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1200                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1201                         inode = igrab(lock->l_resource->lr_lvb_inode);
1202                 } else {
1203                         inode = lock->l_resource->lr_lvb_inode;
1204                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1205                                          D_WARNING, lock, "lr_lvb_inode %p is "
1206                                          "bogus: magic %08x",
1207                                          lock->l_resource->lr_lvb_inode,
1208                                          lli->lli_inode_magic);
1209                         inode = NULL;
1210                 }
1211         }
1212         unlock_res_and_lock(lock);
1213         return inode;
1214 }
1215
1216 static void ll_dir_clear_lsm_md(struct inode *inode)
1217 {
1218         struct ll_inode_info *lli = ll_i2info(inode);
1219
1220         LASSERT(S_ISDIR(inode->i_mode));
1221
1222         if (lli->lli_lsm_md != NULL) {
1223                 lmv_free_memmd(lli->lli_lsm_md);
1224                 lli->lli_lsm_md = NULL;
1225         }
1226 }
1227
1228 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1229                                       const struct lu_fid *fid,
1230                                       struct lustre_md *md)
1231 {
1232         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1233         struct mdt_body         *body = md->body;
1234         struct inode            *inode;
1235         ino_t                   ino;
1236         ENTRY;
1237
1238         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1239         inode = iget_locked(sb, ino);
1240         if (inode == NULL) {
1241                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1242                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1243                 RETURN(ERR_PTR(-ENOENT));
1244         }
1245
1246         if (inode->i_state & I_NEW) {
1247                 struct ll_inode_info *lli = ll_i2info(inode);
1248                 struct lmv_stripe_md *lsm = md->lmv;
1249
1250                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1251                                 (body->mbo_mode & S_IFMT);
1252                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1253                          PFID(fid));
1254
1255                 LTIME_S(inode->i_mtime) = 0;
1256                 LTIME_S(inode->i_atime) = 0;
1257                 LTIME_S(inode->i_ctime) = 0;
1258                 inode->i_rdev = 0;
1259
1260                 /* initializing backing dev info. */
1261                 inode->i_mapping->backing_dev_info =
1262                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1263                 inode->i_op = &ll_dir_inode_operations;
1264                 inode->i_fop = &ll_dir_operations;
1265                 lli->lli_fid = *fid;
1266                 ll_lli_init(lli);
1267
1268                 LASSERT(lsm != NULL);
1269                 /* master object FID */
1270                 lli->lli_pfid = body->mbo_fid1;
1271                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1272                        lli, PFID(fid), PFID(&lli->lli_pfid));
1273                 unlock_new_inode(inode);
1274         }
1275
1276         RETURN(inode);
1277 }
1278
1279 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1280 {
1281         struct lu_fid *fid;
1282         struct lmv_stripe_md *lsm = md->lmv;
1283         int i;
1284
1285         LASSERT(lsm != NULL);
1286         /* XXX sigh, this lsm_root initialization should be in
1287          * LMV layer, but it needs ll_iget right now, so we
1288          * put this here right now. */
1289         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1290                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1291                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1292                 /* Unfortunately ll_iget will call ll_update_inode,
1293                  * where the initialization of slave inode is slightly
1294                  * different, so it reset lsm_md to NULL to avoid
1295                  * initializing lsm for slave inode. */
1296                 /* For migrating inode, master stripe and master object will
1297                  * be same, so we only need assign this inode */
1298                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1299                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1300                 else
1301                         lsm->lsm_md_oinfo[i].lmo_root =
1302                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1303
1304                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1305                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1306
1307                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1308                         return rc;
1309                 }
1310         }
1311
1312         return 0;
1313 }
1314
1315 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1316                                 const struct lmv_stripe_md *lsm_md2)
1317 {
1318         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1319                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1320                lsm_md1->lsm_md_master_mdt_index ==
1321                                         lsm_md2->lsm_md_master_mdt_index &&
1322                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1323                lsm_md1->lsm_md_layout_version ==
1324                                         lsm_md2->lsm_md_layout_version &&
1325                strcmp(lsm_md1->lsm_md_pool_name,
1326                       lsm_md2->lsm_md_pool_name) == 0;
1327 }
1328
1329 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1330 {
1331         struct ll_inode_info *lli = ll_i2info(inode);
1332         struct lmv_stripe_md *lsm = md->lmv;
1333         int     rc;
1334         ENTRY;
1335
1336         LASSERT(S_ISDIR(inode->i_mode));
1337         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1338                PFID(ll_inode2fid(inode)));
1339
1340         /* no striped information from request. */
1341         if (lsm == NULL) {
1342                 if (lli->lli_lsm_md == NULL) {
1343                         RETURN(0);
1344                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1345                                                 LMV_HASH_FLAG_MIGRATION) {
1346                         /* migration is done, the temporay MIGRATE layout has
1347                          * been removed */
1348                         CDEBUG(D_INODE, DFID" finish migration.\n",
1349                                PFID(ll_inode2fid(inode)));
1350                         lmv_free_memmd(lli->lli_lsm_md);
1351                         lli->lli_lsm_md = NULL;
1352                         RETURN(0);
1353                 } else {
1354                         /* The lustre_md from req does not include stripeEA,
1355                          * see ll_md_setattr */
1356                         RETURN(0);
1357                 }
1358         }
1359
1360         /* set the directory layout */
1361         if (lli->lli_lsm_md == NULL) {
1362
1363                 rc = ll_init_lsm_md(inode, md);
1364                 if (rc != 0)
1365                         RETURN(rc);
1366
1367                 lli->lli_lsm_md = lsm;
1368                 /* set lsm_md to NULL, so the following free lustre_md
1369                  * will not free this lsm */
1370                 md->lmv = NULL;
1371                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1372                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1373                 RETURN(0);
1374         }
1375
1376         /* Compare the old and new stripe information */
1377         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1378                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1379                 int                     idx;
1380
1381                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1382                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1383                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1384                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1385                        inode, lsm, old_lsm,
1386                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1387                        lsm->lsm_md_stripe_count,
1388                        old_lsm->lsm_md_stripe_count,
1389                        lsm->lsm_md_master_mdt_index,
1390                        old_lsm->lsm_md_master_mdt_index,
1391                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1392                        lsm->lsm_md_layout_version,
1393                        old_lsm->lsm_md_layout_version,
1394                        lsm->lsm_md_pool_name,
1395                        old_lsm->lsm_md_pool_name);
1396
1397                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1398                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1399                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1400                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1401                 }
1402
1403                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1404                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1405                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1406                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1407                 }
1408
1409                 RETURN(-EIO);
1410         }
1411
1412         RETURN(0);
1413 }
1414
1415 void ll_clear_inode(struct inode *inode)
1416 {
1417         struct ll_inode_info *lli = ll_i2info(inode);
1418         struct ll_sb_info *sbi = ll_i2sbi(inode);
1419         ENTRY;
1420
1421         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1422                PFID(ll_inode2fid(inode)), inode);
1423
1424         if (S_ISDIR(inode->i_mode)) {
1425                 /* these should have been cleared in ll_file_release */
1426                 LASSERT(lli->lli_opendir_key == NULL);
1427                 LASSERT(lli->lli_sai == NULL);
1428                 LASSERT(lli->lli_opendir_pid == 0);
1429         }
1430
1431         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1432
1433         LASSERT(!lli->lli_open_fd_write_count);
1434         LASSERT(!lli->lli_open_fd_read_count);
1435         LASSERT(!lli->lli_open_fd_exec_count);
1436
1437         if (lli->lli_mds_write_och)
1438                 ll_md_real_close(inode, FMODE_WRITE);
1439         if (lli->lli_mds_exec_och)
1440                 ll_md_real_close(inode, FMODE_EXEC);
1441         if (lli->lli_mds_read_och)
1442                 ll_md_real_close(inode, FMODE_READ);
1443
1444         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1445                 OBD_FREE(lli->lli_symlink_name,
1446                          strlen(lli->lli_symlink_name) + 1);
1447                 lli->lli_symlink_name = NULL;
1448         }
1449
1450         ll_xattr_cache_destroy(inode);
1451
1452         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1453                 LASSERT(lli->lli_posix_acl == NULL);
1454                 if (lli->lli_remote_perms) {
1455                         free_rmtperm_hash(lli->lli_remote_perms);
1456                         lli->lli_remote_perms = NULL;
1457                 }
1458         }
1459 #ifdef CONFIG_FS_POSIX_ACL
1460         else if (lli->lli_posix_acl) {
1461                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1462                 LASSERT(lli->lli_remote_perms == NULL);
1463                 posix_acl_release(lli->lli_posix_acl);
1464                 lli->lli_posix_acl = NULL;
1465         }
1466 #endif
1467         lli->lli_inode_magic = LLI_INODE_DEAD;
1468
1469         ll_clear_inode_capas(inode);
1470         if (S_ISDIR(inode->i_mode))
1471                 ll_dir_clear_lsm_md(inode);
1472         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1473                 LASSERT(list_empty(&lli->lli_agl_list));
1474
1475         /*
1476          * XXX This has to be done before lsm is freed below, because
1477          * cl_object still uses inode lsm.
1478          */
1479         cl_inode_fini(inode);
1480         lli->lli_has_smd = false;
1481
1482         EXIT;
1483 }
1484
1485 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1486 {
1487         struct lustre_md md;
1488         struct inode *inode = dentry->d_inode;
1489         struct ll_sb_info *sbi = ll_i2sbi(inode);
1490         struct ptlrpc_request *request = NULL;
1491         int rc, ia_valid;
1492         ENTRY;
1493
1494         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1495                                      LUSTRE_OPC_ANY, NULL);
1496         if (IS_ERR(op_data))
1497                 RETURN(PTR_ERR(op_data));
1498
1499         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1500         if (rc) {
1501                 ptlrpc_req_finished(request);
1502                 if (rc == -ENOENT) {
1503                         clear_nlink(inode);
1504                         /* Unlinked special device node? Or just a race?
1505                          * Pretend we done everything. */
1506                         if (!S_ISREG(inode->i_mode) &&
1507                             !S_ISDIR(inode->i_mode)) {
1508                                 ia_valid = op_data->op_attr.ia_valid;
1509                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1510                                 rc = simple_setattr(dentry, &op_data->op_attr);
1511                                 op_data->op_attr.ia_valid = ia_valid;
1512                         }
1513                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1514                         CERROR("md_setattr fails: rc = %d\n", rc);
1515                 }
1516                 RETURN(rc);
1517         }
1518
1519         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1520                               sbi->ll_md_exp, &md);
1521         if (rc) {
1522                 ptlrpc_req_finished(request);
1523                 RETURN(rc);
1524         }
1525
1526         ia_valid = op_data->op_attr.ia_valid;
1527         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1528          * cache is not cleared yet. */
1529         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1530         rc = simple_setattr(dentry, &op_data->op_attr);
1531         op_data->op_attr.ia_valid = ia_valid;
1532
1533         rc = ll_update_inode(inode, &md);
1534         ptlrpc_req_finished(request);
1535
1536         RETURN(rc);
1537 }
1538
1539 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1540 {
1541         struct obd_capa *capa;
1542         int rc;
1543
1544         if (attr->ia_valid & ATTR_SIZE)
1545                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1546         else
1547                 capa = ll_mdscapa_get(inode);
1548
1549         rc = cl_setattr_ost(ll_i2info(inode)->lli_clob, attr, 0, capa);
1550
1551         if (attr->ia_valid & ATTR_SIZE)
1552                 ll_truncate_free_capa(capa);
1553         else
1554                 capa_put(capa);
1555
1556         return rc;
1557 }
1558
1559 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1560  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1561  * keep these values until such a time that objects are allocated for it.
1562  * We do the MDS operations first, as it is checking permissions for us.
1563  * We don't to the MDS RPC if there is nothing that we want to store there,
1564  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1565  * going to do an RPC anyways.
1566  *
1567  * If we are doing a truncate, we will send the mtime and ctime updates
1568  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1569  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1570  * at the same time.
1571  *
1572  * In case of HSMimport, we only set attr on MDS.
1573  */
1574 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1575 {
1576         struct inode *inode = dentry->d_inode;
1577         struct ll_inode_info *lli = ll_i2info(inode);
1578         struct md_op_data *op_data = NULL;
1579         bool file_is_released = false;
1580         int rc = 0;
1581         ENTRY;
1582
1583         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1584                "valid %x, hsm_import %d\n",
1585                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1586                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1587                hsm_import);
1588
1589         if (attr->ia_valid & ATTR_SIZE) {
1590                 /* Check new size against VFS/VM file size limit and rlimit */
1591                 rc = inode_newsize_ok(inode, attr->ia_size);
1592                 if (rc)
1593                         RETURN(rc);
1594
1595                 /* The maximum Lustre file size is variable, based on the
1596                  * OST maximum object size and number of stripes.  This
1597                  * needs another check in addition to the VFS check above. */
1598                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1599                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1600                                PFID(&lli->lli_fid), attr->ia_size,
1601                                ll_file_maxbytes(inode));
1602                         RETURN(-EFBIG);
1603                 }
1604
1605                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1606         }
1607
1608         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1609         if (attr->ia_valid & TIMES_SET_FLAGS) {
1610                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1611                     !cfs_capable(CFS_CAP_FOWNER))
1612                         RETURN(-EPERM);
1613         }
1614
1615         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1616         if (attr->ia_valid & ATTR_CTIME) {
1617                 attr->ia_ctime = CFS_CURRENT_TIME;
1618                 attr->ia_valid |= ATTR_CTIME_SET;
1619         }
1620         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1621             (attr->ia_valid & ATTR_ATIME)) {
1622                 attr->ia_atime = CFS_CURRENT_TIME;
1623                 attr->ia_valid |= ATTR_ATIME_SET;
1624         }
1625         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1626             (attr->ia_valid & ATTR_MTIME)) {
1627                 attr->ia_mtime = CFS_CURRENT_TIME;
1628                 attr->ia_valid |= ATTR_MTIME_SET;
1629         }
1630
1631         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1632                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1633                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1634                        cfs_time_current_sec());
1635
1636         /* We always do an MDS RPC, even if we're only changing the size;
1637          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1638
1639         OBD_ALLOC_PTR(op_data);
1640         if (op_data == NULL)
1641                 RETURN(-ENOMEM);
1642
1643         if (!S_ISDIR(inode->i_mode)) {
1644                 if (attr->ia_valid & ATTR_SIZE)
1645                         inode_dio_write_done(inode);
1646                 mutex_unlock(&inode->i_mutex);
1647         }
1648
1649         /* truncate on a released file must failed with -ENODATA,
1650          * so size must not be set on MDS for released file
1651          * but other attributes must be set
1652          */
1653         if (S_ISREG(inode->i_mode)) {
1654                 struct cl_layout cl = {
1655                         .cl_is_released = false,
1656                 };
1657                 struct lu_env *env;
1658                 int refcheck;
1659                 __u32 gen;
1660
1661                 rc = ll_layout_refresh(inode, &gen);
1662                 if (rc < 0)
1663                         GOTO(out, rc);
1664
1665                 /* XXX: the only place we need to know the layout type,
1666                  * this will be removed by a later patch. -Jinshan */
1667                 env = cl_env_get(&refcheck);
1668                 if (IS_ERR(env))
1669                         GOTO(out, rc = PTR_ERR(env));
1670
1671                 rc = cl_object_layout_get(env, lli->lli_clob, &cl);
1672                 cl_env_put(env, &refcheck);
1673                 if (rc < 0)
1674                         GOTO(out, rc);
1675
1676                 file_is_released = cl.cl_is_released;
1677
1678                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1679                         if (file_is_released) {
1680                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1681                                 if (rc < 0)
1682                                         GOTO(out, rc);
1683
1684                                 file_is_released = false;
1685                                 ll_layout_refresh(inode, &gen);
1686                         }
1687
1688                         /* If we are changing file size, file content is
1689                          * modified, flag it. */
1690                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1691                         spin_lock(&lli->lli_lock);
1692                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1693                         spin_unlock(&lli->lli_lock);
1694                         op_data->op_bias |= MDS_DATA_MODIFIED;
1695                 }
1696         }
1697
1698         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1699
1700         rc = ll_md_setattr(dentry, op_data);
1701         if (rc)
1702                 GOTO(out, rc);
1703
1704         /* RPC to MDT is sent, cancel data modification flag */
1705         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1706                 spin_lock(&lli->lli_lock);
1707                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1708                 spin_unlock(&lli->lli_lock);
1709         }
1710
1711         if (!S_ISREG(inode->i_mode) || file_is_released)
1712                 GOTO(out, rc = 0);
1713
1714         if (attr->ia_valid & (ATTR_SIZE |
1715                               ATTR_ATIME | ATTR_ATIME_SET |
1716                               ATTR_MTIME | ATTR_MTIME_SET)) {
1717                 /* For truncate and utimes sending attributes to OSTs, setting
1718                  * mtime/atime to the past will be performed under PW [0:EOF]
1719                  * extent lock (new_size:EOF for truncate).  It may seem
1720                  * excessive to send mtime/atime updates to OSTs when not
1721                  * setting times to past, but it is necessary due to possible
1722                  * time de-synchronization between MDT inode and OST objects */
1723                 if (attr->ia_valid & ATTR_SIZE)
1724                         down_write(&lli->lli_trunc_sem);
1725                 rc = ll_setattr_ost(inode, attr);
1726                 if (attr->ia_valid & ATTR_SIZE)
1727                         up_write(&lli->lli_trunc_sem);
1728         }
1729         EXIT;
1730 out:
1731         if (op_data != NULL)
1732                 ll_finish_md_op_data(op_data);
1733
1734         if (!S_ISDIR(inode->i_mode)) {
1735                 mutex_lock(&inode->i_mutex);
1736                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1737                         inode_dio_wait(inode);
1738         }
1739
1740         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1741                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1742
1743         return rc;
1744 }
1745
1746 int ll_setattr(struct dentry *de, struct iattr *attr)
1747 {
1748         int mode = de->d_inode->i_mode;
1749
1750         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1751                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1752                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1753
1754         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1755                                (ATTR_SIZE|ATTR_MODE)) &&
1756             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1757              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1758               !(attr->ia_mode & S_ISGID))))
1759                 attr->ia_valid |= ATTR_FORCE;
1760
1761         if ((attr->ia_valid & ATTR_MODE) &&
1762             (mode & S_ISUID) &&
1763             !(attr->ia_mode & S_ISUID) &&
1764             !(attr->ia_valid & ATTR_KILL_SUID))
1765                 attr->ia_valid |= ATTR_KILL_SUID;
1766
1767         if ((attr->ia_valid & ATTR_MODE) &&
1768             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1769             !(attr->ia_mode & S_ISGID) &&
1770             !(attr->ia_valid & ATTR_KILL_SGID))
1771                 attr->ia_valid |= ATTR_KILL_SGID;
1772
1773         return ll_setattr_raw(de, attr, false);
1774 }
1775
1776 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1777                        __u64 max_age, __u32 flags)
1778 {
1779         struct ll_sb_info *sbi = ll_s2sbi(sb);
1780         struct obd_statfs obd_osfs;
1781         int rc;
1782         ENTRY;
1783
1784         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1785         if (rc) {
1786                 CERROR("md_statfs fails: rc = %d\n", rc);
1787                 RETURN(rc);
1788         }
1789
1790         osfs->os_type = sb->s_magic;
1791
1792         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1793                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1794
1795         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1796                 flags |= OBD_STATFS_NODELAY;
1797
1798         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1799         if (rc) {
1800                 CERROR("obd_statfs fails: rc = %d\n", rc);
1801                 RETURN(rc);
1802         }
1803
1804         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1805                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1806                obd_osfs.os_files);
1807
1808         osfs->os_bsize = obd_osfs.os_bsize;
1809         osfs->os_blocks = obd_osfs.os_blocks;
1810         osfs->os_bfree = obd_osfs.os_bfree;
1811         osfs->os_bavail = obd_osfs.os_bavail;
1812
1813         /* If we don't have as many objects free on the OST as inodes
1814          * on the MDS, we reduce the total number of inodes to
1815          * compensate, so that the "inodes in use" number is correct.
1816          */
1817         if (obd_osfs.os_ffree < osfs->os_ffree) {
1818                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1819                         obd_osfs.os_ffree;
1820                 osfs->os_ffree = obd_osfs.os_ffree;
1821         }
1822
1823         RETURN(rc);
1824 }
1825 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1826 {
1827         struct super_block *sb = de->d_sb;
1828         struct obd_statfs osfs;
1829         __u64 fsid = huge_encode_dev(sb->s_dev);
1830         int rc;
1831
1832         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1833         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1834
1835         /* Some amount of caching on the client is allowed */
1836         rc = ll_statfs_internal(sb, &osfs,
1837                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1838                                 0);
1839         if (rc)
1840                 return rc;
1841
1842         statfs_unpack(sfs, &osfs);
1843
1844         /* We need to downshift for all 32-bit kernels, because we can't
1845          * tell if the kernel is being called via sys_statfs64() or not.
1846          * Stop before overflowing f_bsize - in which case it is better
1847          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1848         if (sizeof(long) < 8) {
1849                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1850                         sfs->f_bsize <<= 1;
1851
1852                         osfs.os_blocks >>= 1;
1853                         osfs.os_bfree >>= 1;
1854                         osfs.os_bavail >>= 1;
1855                 }
1856         }
1857
1858         sfs->f_blocks = osfs.os_blocks;
1859         sfs->f_bfree = osfs.os_bfree;
1860         sfs->f_bavail = osfs.os_bavail;
1861         sfs->f_fsid.val[0] = (__u32)fsid;
1862         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1863         return 0;
1864 }
1865
1866 void ll_inode_size_lock(struct inode *inode)
1867 {
1868         struct ll_inode_info *lli;
1869
1870         LASSERT(!S_ISDIR(inode->i_mode));
1871
1872         lli = ll_i2info(inode);
1873         mutex_lock(&lli->lli_size_mutex);
1874 }
1875
1876 void ll_inode_size_unlock(struct inode *inode)
1877 {
1878         struct ll_inode_info *lli;
1879
1880         lli = ll_i2info(inode);
1881         mutex_unlock(&lli->lli_size_mutex);
1882 }
1883
1884 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1885 {
1886         struct ll_inode_info *lli = ll_i2info(inode);
1887         struct mdt_body *body = md->body;
1888         struct lov_stripe_md *lsm = md->lsm;
1889         struct ll_sb_info *sbi = ll_i2sbi(inode);
1890
1891         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1892         if (lsm != NULL) {
1893                 if (!lli->lli_has_smd &&
1894                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1895                         cl_file_inode_init(inode, md);
1896
1897                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1898                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1899                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1900         }
1901
1902         if (S_ISDIR(inode->i_mode)) {
1903                 int     rc;
1904
1905                 rc = ll_update_lsm_md(inode, md);
1906                 if (rc != 0)
1907                         return rc;
1908         }
1909
1910         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1911                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1912                         ll_update_remote_perm(inode, md->remote_perm);
1913         }
1914 #ifdef CONFIG_FS_POSIX_ACL
1915         else if (body->mbo_valid & OBD_MD_FLACL) {
1916                 spin_lock(&lli->lli_lock);
1917                 if (lli->lli_posix_acl)
1918                         posix_acl_release(lli->lli_posix_acl);
1919                 lli->lli_posix_acl = md->posix_acl;
1920                 spin_unlock(&lli->lli_lock);
1921         }
1922 #endif
1923         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1924                                         sbi->ll_flags & LL_SBI_32BIT_API);
1925         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1926
1927         if (body->mbo_valid & OBD_MD_FLATIME) {
1928                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1929                         LTIME_S(inode->i_atime) = body->mbo_atime;
1930                 lli->lli_atime = body->mbo_atime;
1931         }
1932
1933         if (body->mbo_valid & OBD_MD_FLMTIME) {
1934                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1935                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1936                                "to "LPU64"\n", inode->i_ino,
1937                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1938                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1939                 }
1940                 lli->lli_mtime = body->mbo_mtime;
1941         }
1942
1943         if (body->mbo_valid & OBD_MD_FLCTIME) {
1944                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1945                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1946                 lli->lli_ctime = body->mbo_ctime;
1947         }
1948
1949         if (body->mbo_valid & OBD_MD_FLMODE)
1950                 inode->i_mode = (inode->i_mode & S_IFMT) |
1951                                 (body->mbo_mode & ~S_IFMT);
1952
1953         if (body->mbo_valid & OBD_MD_FLTYPE)
1954                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1955                                 (body->mbo_mode & S_IFMT);
1956
1957         LASSERT(inode->i_mode != 0);
1958         if (S_ISREG(inode->i_mode))
1959                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1960                                        LL_MAX_BLKSIZE_BITS);
1961         else
1962                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1963
1964         if (body->mbo_valid & OBD_MD_FLUID)
1965                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1966         if (body->mbo_valid & OBD_MD_FLGID)
1967                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1968         if (body->mbo_valid & OBD_MD_FLFLAGS)
1969                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1970         if (body->mbo_valid & OBD_MD_FLNLINK)
1971                 set_nlink(inode, body->mbo_nlink);
1972         if (body->mbo_valid & OBD_MD_FLRDEV)
1973                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1974
1975         if (body->mbo_valid & OBD_MD_FLID) {
1976                 /* FID shouldn't be changed! */
1977                 if (fid_is_sane(&lli->lli_fid)) {
1978                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1979                                  "Trying to change FID "DFID
1980                                  " to the "DFID", inode "DFID"(%p)\n",
1981                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1982                                  PFID(ll_inode2fid(inode)), inode);
1983                 } else {
1984                         lli->lli_fid = body->mbo_fid1;
1985                 }
1986         }
1987
1988         LASSERT(fid_seq(&lli->lli_fid) != 0);
1989
1990         if (body->mbo_valid & OBD_MD_FLSIZE) {
1991                 i_size_write(inode, body->mbo_size);
1992
1993                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1994                        PFID(ll_inode2fid(inode)),
1995                        (unsigned long long)body->mbo_size);
1996
1997                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1998                         inode->i_blocks = body->mbo_blocks;
1999         }
2000
2001         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2002                 LASSERT(md->mds_capa);
2003                 ll_add_capa(inode, md->mds_capa);
2004         }
2005
2006         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2007                 LASSERT(md->oss_capa);
2008                 ll_add_capa(inode, md->oss_capa);
2009         }
2010
2011         if (body->mbo_valid & OBD_MD_TSTATE) {
2012                 if (body->mbo_t_state & MS_RESTORE)
2013                         lli->lli_flags |= LLIF_FILE_RESTORING;
2014         }
2015
2016         return 0;
2017 }
2018
2019 int ll_read_inode2(struct inode *inode, void *opaque)
2020 {
2021         struct lustre_md *md = opaque;
2022         struct ll_inode_info *lli = ll_i2info(inode);
2023         int     rc;
2024         ENTRY;
2025
2026         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2027                PFID(&lli->lli_fid), inode);
2028
2029         LASSERT(!lli->lli_has_smd);
2030
2031         /* Core attributes from the MDS first.  This is a new inode, and
2032          * the VFS doesn't zero times in the core inode so we have to do
2033          * it ourselves.  They will be overwritten by either MDS or OST
2034          * attributes - we just need to make sure they aren't newer. */
2035         LTIME_S(inode->i_mtime) = 0;
2036         LTIME_S(inode->i_atime) = 0;
2037         LTIME_S(inode->i_ctime) = 0;
2038         inode->i_rdev = 0;
2039         rc = ll_update_inode(inode, md);
2040         if (rc != 0)
2041                 RETURN(rc);
2042
2043         /* OIDEBUG(inode); */
2044
2045         /* initializing backing dev info. */
2046         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2047
2048
2049         if (S_ISREG(inode->i_mode)) {
2050                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2051                 inode->i_op = &ll_file_inode_operations;
2052                 inode->i_fop = sbi->ll_fop;
2053                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2054                 EXIT;
2055         } else if (S_ISDIR(inode->i_mode)) {
2056                 inode->i_op = &ll_dir_inode_operations;
2057                 inode->i_fop = &ll_dir_operations;
2058                 EXIT;
2059         } else if (S_ISLNK(inode->i_mode)) {
2060                 inode->i_op = &ll_fast_symlink_inode_operations;
2061                 EXIT;
2062         } else {
2063                 inode->i_op = &ll_special_inode_operations;
2064
2065                 init_special_inode(inode, inode->i_mode,
2066                                    inode->i_rdev);
2067
2068                 EXIT;
2069         }
2070
2071         return 0;
2072 }
2073
2074 void ll_delete_inode(struct inode *inode)
2075 {
2076         struct ll_inode_info *lli = ll_i2info(inode);
2077         ENTRY;
2078
2079         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2080                 /* It is last chance to write out dirty pages,
2081                  * otherwise we may lose data while umount */
2082                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2083
2084         truncate_inode_pages_final(&inode->i_data);
2085
2086         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2087                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2088                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2089
2090 #ifdef HAVE_SBOPS_EVICT_INODE
2091         ll_clear_inode(inode);
2092 #endif
2093         clear_inode(inode);
2094
2095         EXIT;
2096 }
2097
2098 int ll_iocontrol(struct inode *inode, struct file *file,
2099                  unsigned int cmd, unsigned long arg)
2100 {
2101         struct ll_sb_info *sbi = ll_i2sbi(inode);
2102         struct ptlrpc_request *req = NULL;
2103         int rc, flags = 0;
2104         ENTRY;
2105
2106         switch(cmd) {
2107         case FSFILT_IOC_GETFLAGS: {
2108                 struct mdt_body *body;
2109                 struct md_op_data *op_data;
2110
2111                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2112                                              0, 0, LUSTRE_OPC_ANY,
2113                                              NULL);
2114                 if (IS_ERR(op_data))
2115                         RETURN(PTR_ERR(op_data));
2116
2117                 op_data->op_valid = OBD_MD_FLFLAGS;
2118                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2119                 ll_finish_md_op_data(op_data);
2120                 if (rc) {
2121                         CERROR("%s: failure inode "DFID": rc = %d\n",
2122                                sbi->ll_md_exp->exp_obd->obd_name,
2123                                PFID(ll_inode2fid(inode)), rc);
2124                         RETURN(-abs(rc));
2125                 }
2126
2127                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2128
2129                 flags = body->mbo_flags;
2130
2131                 ptlrpc_req_finished(req);
2132
2133                 RETURN(put_user(flags, (int __user *)arg));
2134         }
2135         case FSFILT_IOC_SETFLAGS: {
2136                 struct iattr *attr;
2137                 struct md_op_data *op_data;
2138                 struct cl_object *obj;
2139                 struct obd_capa *capa;
2140
2141                 if (get_user(flags, (int __user *)arg))
2142                         RETURN(-EFAULT);
2143
2144                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2145                                              LUSTRE_OPC_ANY, NULL);
2146                 if (IS_ERR(op_data))
2147                         RETURN(PTR_ERR(op_data));
2148
2149                 op_data->op_attr_flags = flags;
2150                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2151                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2152                 ll_finish_md_op_data(op_data);
2153                 ptlrpc_req_finished(req);
2154                 if (rc)
2155                         RETURN(rc);
2156
2157                 inode->i_flags = ll_ext_to_inode_flags(flags);
2158
2159                 obj = ll_i2info(inode)->lli_clob;
2160                 if (obj == NULL)
2161                         RETURN(0);
2162
2163                 OBD_ALLOC_PTR(attr);
2164                 if (attr == NULL)
2165                         RETURN(-ENOMEM);
2166
2167                 attr->ia_valid = ATTR_ATTR_FLAG;
2168
2169                 capa = ll_mdscapa_get(inode);
2170                 rc = cl_setattr_ost(obj, attr, flags, capa);
2171                 capa_put(capa);
2172
2173                 OBD_FREE_PTR(attr);
2174                 RETURN(rc);
2175         }
2176         default:
2177                 RETURN(-ENOSYS);
2178         }
2179
2180         RETURN(0);
2181 }
2182
2183 int ll_flush_ctx(struct inode *inode)
2184 {
2185         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2186
2187         CDEBUG(D_SEC, "flush context for user %d\n",
2188                from_kuid(&init_user_ns, current_uid()));
2189
2190         obd_set_info_async(NULL, sbi->ll_md_exp,
2191                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2192                            0, NULL, NULL);
2193         obd_set_info_async(NULL, sbi->ll_dt_exp,
2194                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2195                            0, NULL, NULL);
2196         return 0;
2197 }
2198
2199 /* umount -f client means force down, don't save state */
2200 void ll_umount_begin(struct super_block *sb)
2201 {
2202         struct ll_sb_info *sbi = ll_s2sbi(sb);
2203         struct obd_device *obd;
2204         struct obd_ioctl_data *ioc_data;
2205         ENTRY;
2206
2207         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2208                sb->s_count, atomic_read(&sb->s_active));
2209
2210         obd = class_exp2obd(sbi->ll_md_exp);
2211         if (obd == NULL) {
2212                 CERROR("Invalid MDC connection handle "LPX64"\n",
2213                        sbi->ll_md_exp->exp_handle.h_cookie);
2214                 EXIT;
2215                 return;
2216         }
2217         obd->obd_force = 1;
2218
2219         obd = class_exp2obd(sbi->ll_dt_exp);
2220         if (obd == NULL) {
2221                 CERROR("Invalid LOV connection handle "LPX64"\n",
2222                        sbi->ll_dt_exp->exp_handle.h_cookie);
2223                 EXIT;
2224                 return;
2225         }
2226         obd->obd_force = 1;
2227
2228         OBD_ALLOC_PTR(ioc_data);
2229         if (ioc_data) {
2230                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2231                               sizeof *ioc_data, ioc_data, NULL);
2232
2233                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2234                               sizeof *ioc_data, ioc_data, NULL);
2235
2236                 OBD_FREE_PTR(ioc_data);
2237         }
2238
2239         /* Really, we'd like to wait until there are no requests outstanding,
2240          * and then continue.  For now, we just invalidate the requests,
2241          * schedule() and sleep one second if needed, and hope.
2242          */
2243         schedule();
2244         EXIT;
2245 }
2246
2247 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2248 {
2249         struct ll_sb_info *sbi = ll_s2sbi(sb);
2250         char *profilenm = get_profile_name(sb);
2251         int err;
2252         __u32 read_only;
2253
2254         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2255                 read_only = *flags & MS_RDONLY;
2256                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2257                                          sizeof(KEY_READ_ONLY),
2258                                          KEY_READ_ONLY, sizeof(read_only),
2259                                          &read_only, NULL);
2260                 if (err) {
2261                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2262                                       profilenm, read_only ?
2263                                       "read-only" : "read-write", err);
2264                         return err;
2265                 }
2266
2267                 if (read_only)
2268                         sb->s_flags |= MS_RDONLY;
2269                 else
2270                         sb->s_flags &= ~MS_RDONLY;
2271
2272                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2273                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2274                                       read_only ?  "read-only" : "read-write");
2275         }
2276         return 0;
2277 }
2278
2279 /**
2280  * Cleanup the open handle that is cached on MDT-side.
2281  *
2282  * For open case, the client side open handling thread may hit error
2283  * after the MDT grant the open. Under such case, the client should
2284  * send close RPC to the MDT as cleanup; otherwise, the open handle
2285  * on the MDT will be leaked there until the client umount or evicted.
2286  *
2287  * In further, if someone unlinked the file, because the open handle
2288  * holds the reference on such file/object, then it will block the
2289  * subsequent threads that want to locate such object via FID.
2290  *
2291  * \param[in] sb        super block for this file-system
2292  * \param[in] open_req  pointer to the original open request
2293  */
2294 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2295 {
2296         struct mdt_body                 *body;
2297         struct md_op_data               *op_data;
2298         struct ptlrpc_request           *close_req = NULL;
2299         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2300         ENTRY;
2301
2302         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2303         OBD_ALLOC_PTR(op_data);
2304         if (op_data == NULL) {
2305                 CWARN("%s: cannot allocate op_data to release open handle for "
2306                       DFID"\n",
2307                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2308
2309                 RETURN_EXIT;
2310         }
2311
2312         op_data->op_fid1 = body->mbo_fid1;
2313         op_data->op_handle = body->mbo_handle;
2314         op_data->op_mod_time = cfs_time_current_sec();
2315         md_close(exp, op_data, NULL, &close_req);
2316         ptlrpc_req_finished(close_req);
2317         ll_finish_md_op_data(op_data);
2318
2319         EXIT;
2320 }
2321
2322 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2323                   struct super_block *sb, struct lookup_intent *it)
2324 {
2325         struct ll_sb_info *sbi = NULL;
2326         struct lustre_md md = { NULL };
2327         int rc;
2328         ENTRY;
2329
2330         LASSERT(*inode || sb);
2331         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2332         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2333                               sbi->ll_md_exp, &md);
2334         if (rc != 0)
2335                 GOTO(cleanup, rc);
2336
2337         if (*inode) {
2338                 rc = ll_update_inode(*inode, &md);
2339                 if (rc != 0)
2340                         GOTO(out, rc);
2341         } else {
2342                 LASSERT(sb != NULL);
2343
2344                 /*
2345                  * At this point server returns to client's same fid as client
2346                  * generated for creating. So using ->fid1 is okay here.
2347                  */
2348                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2349
2350                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2351                                              sbi->ll_flags & LL_SBI_32BIT_API),
2352                                  &md);
2353                 if (IS_ERR(*inode)) {
2354 #ifdef CONFIG_FS_POSIX_ACL
2355                         if (md.posix_acl) {
2356                                 posix_acl_release(md.posix_acl);
2357                                 md.posix_acl = NULL;
2358                         }
2359 #endif
2360                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2361                         *inode = NULL;
2362                         CERROR("new_inode -fatal: rc %d\n", rc);
2363                         GOTO(out, rc);
2364                 }
2365         }
2366
2367         /* Handling piggyback layout lock.
2368          * Layout lock can be piggybacked by getattr and open request.
2369          * The lsm can be applied to inode only if it comes with a layout lock
2370          * otherwise correct layout may be overwritten, for example:
2371          * 1. proc1: mdt returns a lsm but not granting layout
2372          * 2. layout was changed by another client
2373          * 3. proc2: refresh layout and layout lock granted
2374          * 4. proc1: to apply a stale layout */
2375         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2376                 struct lustre_handle lockh;
2377                 struct ldlm_lock *lock;
2378
2379                 lockh.cookie = it->d.lustre.it_lock_handle;
2380                 lock = ldlm_handle2lock(&lockh);
2381                 LASSERT(lock != NULL);
2382                 if (ldlm_has_layout(lock)) {
2383                         struct cl_object_conf conf;
2384
2385                         memset(&conf, 0, sizeof(conf));
2386                         conf.coc_opc = OBJECT_CONF_SET;
2387                         conf.coc_inode = *inode;
2388                         conf.coc_lock = lock;
2389                         conf.u.coc_md = &md;
2390                         (void)ll_layout_conf(*inode, &conf);
2391                 }
2392                 LDLM_LOCK_PUT(lock);
2393         }
2394
2395         GOTO(out, rc = 0);
2396
2397 out:
2398         if (md.lsm != NULL)
2399                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2400         md_free_lustre_md(sbi->ll_md_exp, &md);
2401
2402 cleanup:
2403         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2404                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2405
2406         return rc;
2407 }
2408
2409 int ll_obd_statfs(struct inode *inode, void __user *arg)
2410 {
2411         struct ll_sb_info *sbi = NULL;
2412         struct obd_export *exp;
2413         char *buf = NULL;
2414         struct obd_ioctl_data *data = NULL;
2415         __u32 type;
2416         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2417         int len = 0, rc;
2418
2419         if (!inode || !(sbi = ll_i2sbi(inode)))
2420                 GOTO(out_statfs, rc = -EINVAL);
2421
2422         rc = obd_ioctl_getdata(&buf, &len, arg);
2423         if (rc)
2424                 GOTO(out_statfs, rc);
2425
2426         data = (void*)buf;
2427         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2428             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2429                 GOTO(out_statfs, rc = -EINVAL);
2430
2431         if (data->ioc_inllen1 != sizeof(__u32) ||
2432             data->ioc_inllen2 != sizeof(__u32) ||
2433             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2434             data->ioc_plen2 != sizeof(struct obd_uuid))
2435                 GOTO(out_statfs, rc = -EINVAL);
2436
2437         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2438         if (type & LL_STATFS_LMV)
2439                 exp = sbi->ll_md_exp;
2440         else if (type & LL_STATFS_LOV)
2441                 exp = sbi->ll_dt_exp;
2442         else
2443                 GOTO(out_statfs, rc = -ENODEV);
2444
2445         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2446         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2447         if (rc)
2448                 GOTO(out_statfs, rc);
2449 out_statfs:
2450         if (buf)
2451                 obd_ioctl_freedata(buf, len);
2452         return rc;
2453 }
2454
2455 int ll_process_config(struct lustre_cfg *lcfg)
2456 {
2457         struct super_block *sb;
2458         unsigned long x;
2459         int rc = 0;
2460         char *ptr;
2461
2462         /* The instance name contains the sb: lustre-client-aacfe000 */
2463         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2464         if (!ptr || !*(++ptr))
2465                 return -EINVAL;
2466         if (sscanf(ptr, "%lx", &x) != 1)
2467                 return -EINVAL;
2468         sb = (struct super_block *)x;
2469         /* This better be a real Lustre superblock! */
2470         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2471
2472         /* Note we have not called client_common_fill_super yet, so
2473            proc fns must be able to handle that! */
2474         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2475                                       lcfg, sb);
2476         if (rc > 0)
2477                 rc = 0;
2478         return rc;
2479 }
2480
2481 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2482 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2483                                        struct inode *i1, struct inode *i2,
2484                                        const char *name, size_t namelen,
2485                                        __u32 mode, __u32 opc, void *data)
2486 {
2487         LASSERT(i1 != NULL);
2488
2489         if (name == NULL) {
2490                 /* Do not reuse namelen for something else. */
2491                 if (namelen != 0)
2492                         return ERR_PTR(-EINVAL);
2493         } else {
2494                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2495                         return ERR_PTR(-ENAMETOOLONG);
2496
2497                 if (!lu_name_is_valid_2(name, namelen))
2498                         return ERR_PTR(-EINVAL);
2499         }
2500
2501         if (op_data == NULL)
2502                 OBD_ALLOC_PTR(op_data);
2503
2504         if (op_data == NULL)
2505                 return ERR_PTR(-ENOMEM);
2506
2507         ll_i2gids(op_data->op_suppgids, i1, i2);
2508         op_data->op_fid1 = *ll_inode2fid(i1);
2509         op_data->op_capa1 = ll_mdscapa_get(i1);
2510         op_data->op_default_stripe_offset = -1;
2511         if (S_ISDIR(i1->i_mode)) {
2512                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2513                 if (opc == LUSTRE_OPC_MKDIR)
2514                         op_data->op_default_stripe_offset =
2515                                    ll_i2info(i1)->lli_def_stripe_offset;
2516         }
2517
2518         if (i2) {
2519                 op_data->op_fid2 = *ll_inode2fid(i2);
2520                 op_data->op_capa2 = ll_mdscapa_get(i2);
2521                 if (S_ISDIR(i2->i_mode))
2522                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2523         } else {
2524                 fid_zero(&op_data->op_fid2);
2525                 op_data->op_capa2 = NULL;
2526         }
2527
2528         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2529                 op_data->op_cli_flags |= CLI_HASH64;
2530
2531         if (ll_need_32bit_api(ll_i2sbi(i1)))
2532                 op_data->op_cli_flags |= CLI_API32;
2533
2534         op_data->op_name = name;
2535         op_data->op_namelen = namelen;
2536         op_data->op_mode = mode;
2537         op_data->op_mod_time = cfs_time_current_sec();
2538         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2539         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2540         op_data->op_cap = cfs_curproc_cap_pack();
2541         op_data->op_bias = 0;
2542         op_data->op_cli_flags = 0;
2543         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2544              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2545                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2546         } else {
2547                 op_data->op_mds = 0;
2548         }
2549         op_data->op_data = data;
2550
2551         /* When called by ll_setattr_raw, file is i1. */
2552         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2553                 op_data->op_bias |= MDS_DATA_MODIFIED;
2554
2555         return op_data;
2556 }
2557
2558 void ll_finish_md_op_data(struct md_op_data *op_data)
2559 {
2560         capa_put(op_data->op_capa1);
2561         capa_put(op_data->op_capa2);
2562         OBD_FREE_PTR(op_data);
2563 }
2564
2565 #ifdef HAVE_SUPEROPS_USE_DENTRY
2566 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2567 #else
2568 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2569 #endif
2570 {
2571         struct ll_sb_info *sbi;
2572
2573 #ifdef HAVE_SUPEROPS_USE_DENTRY
2574         LASSERT((seq != NULL) && (dentry != NULL));
2575         sbi = ll_s2sbi(dentry->d_sb);
2576 #else
2577         LASSERT((seq != NULL) && (vfs != NULL));
2578         sbi = ll_s2sbi(vfs->mnt_sb);
2579 #endif
2580
2581         if (sbi->ll_flags & LL_SBI_NOLCK)
2582                 seq_puts(seq, ",nolock");
2583
2584         if (sbi->ll_flags & LL_SBI_FLOCK)
2585                 seq_puts(seq, ",flock");
2586
2587         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2588                 seq_puts(seq, ",localflock");
2589
2590         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2591                 seq_puts(seq, ",user_xattr");
2592
2593         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2594                 seq_puts(seq, ",lazystatfs");
2595
2596         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2597                 seq_puts(seq, ",user_fid2path");
2598
2599         RETURN(0);
2600 }
2601
2602 /**
2603  * Get obd name by cmd, and copy out to user space
2604  */
2605 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2606 {
2607         struct ll_sb_info *sbi = ll_i2sbi(inode);
2608         struct obd_device *obd;
2609         ENTRY;
2610
2611         if (cmd == OBD_IOC_GETDTNAME)
2612                 obd = class_exp2obd(sbi->ll_dt_exp);
2613         else if (cmd == OBD_IOC_GETMDNAME)
2614                 obd = class_exp2obd(sbi->ll_md_exp);
2615         else
2616                 RETURN(-EINVAL);
2617
2618         if (!obd)
2619                 RETURN(-ENOENT);
2620
2621         if (copy_to_user((void __user *)arg, obd->obd_name,
2622                          strlen(obd->obd_name) + 1))
2623                 RETURN(-EFAULT);
2624
2625         RETURN(0);
2626 }
2627
2628 /**
2629  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2630  * fsname will be returned in this buffer; otherwise, a static buffer will be
2631  * used to store the fsname and returned to caller.
2632  */
2633 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2634 {
2635         static char fsname_static[MTI_NAME_MAXLEN];
2636         struct lustre_sb_info *lsi = s2lsi(sb);
2637         char *ptr;
2638         int len;
2639
2640         if (buf == NULL) {
2641                 /* this means the caller wants to use static buffer
2642                  * and it doesn't care about race. Usually this is
2643                  * in error reporting path */
2644                 buf = fsname_static;
2645                 buflen = sizeof(fsname_static);
2646         }
2647
2648         len = strlen(lsi->lsi_lmd->lmd_profile);
2649         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2650         if (ptr && (strcmp(ptr, "-client") == 0))
2651                 len -= 7;
2652
2653         if (unlikely(len >= buflen))
2654                 len = buflen - 1;
2655         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2656         buf[len] = '\0';
2657
2658         return buf;
2659 }
2660
2661 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2662 {
2663         char *path = NULL;
2664
2665         struct path p;
2666
2667         p.dentry = dentry;
2668         p.mnt = current->fs->root.mnt;
2669         path_get(&p);
2670         path = d_path(&p, buf, bufsize);
2671         path_put(&p);
2672         return path;
2673 }
2674
2675 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2676 {
2677         char *buf, *path = NULL;
2678         struct dentry *dentry = NULL;
2679         struct inode *inode = page->mapping->host;
2680
2681         /* this can be called inside spin lock so use GFP_ATOMIC. */
2682         buf = (char *)__get_free_page(GFP_ATOMIC);
2683         if (buf != NULL) {
2684                 dentry = d_find_alias(page->mapping->host);
2685                 if (dentry != NULL)
2686                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2687         }
2688
2689         CDEBUG(D_WARNING,
2690                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2691                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2692                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2693                PFID(ll_inode2fid(inode)),
2694                (path && !IS_ERR(path)) ? path : "", ioret);
2695
2696         if (dentry != NULL)
2697                 dput(dentry);
2698
2699         if (buf != NULL)
2700                 free_page((unsigned long)buf);
2701 }
2702
2703 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2704                         struct lov_user_md **kbuf)
2705 {
2706         struct lov_user_md      lum;
2707         ssize_t                 lum_size;
2708         ENTRY;
2709
2710         if (copy_from_user(&lum, md, sizeof(lum)))
2711                 RETURN(-EFAULT);
2712
2713         lum_size = ll_lov_user_md_size(&lum);
2714         if (lum_size < 0)
2715                 RETURN(lum_size);
2716
2717         OBD_ALLOC(*kbuf, lum_size);
2718         if (*kbuf == NULL)
2719                 RETURN(-ENOMEM);
2720
2721         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2722                 OBD_FREE(*kbuf, lum_size);
2723                 RETURN(-EFAULT);
2724         }
2725
2726         RETURN(lum_size);
2727 }
2728
2729 /*
2730  * Compute llite root squash state after a change of root squash
2731  * configuration setting or add/remove of a lnet nid
2732  */
2733 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2734 {
2735         struct root_squash_info *squash = &sbi->ll_squash;
2736         int i;
2737         bool matched;
2738         lnet_process_id_t id;
2739
2740         /* Update norootsquash flag */
2741         down_write(&squash->rsi_sem);
2742         if (list_empty(&squash->rsi_nosquash_nids))
2743                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2744         else {
2745                 /* Do not apply root squash as soon as one of our NIDs is
2746                  * in the nosquash_nids list */
2747                 matched = false;
2748                 i = 0;
2749                 while (LNetGetId(i++, &id) != -ENOENT) {
2750                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2751                                 continue;
2752                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2753                                 matched = true;
2754                                 break;
2755                         }
2756                 }
2757                 if (matched)
2758                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2759                 else
2760                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2761         }
2762         up_write(&squash->rsi_sem);
2763 }
2764
2765 /**
2766  * Parse linkea content to extract information about a given hardlink
2767  *
2768  * \param[in]   ldata      - Initialized linkea data
2769  * \param[in]   linkno     - Link identifier
2770  * \param[out]  parent_fid - The entry's parent FID
2771  * \param[out]  ln         - Entry name destination buffer
2772  *
2773  * \retval 0 on success
2774  * \retval Appropriate negative error code on failure
2775  */
2776 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2777                             struct lu_fid *parent_fid, struct lu_name *ln)
2778 {
2779         unsigned int    idx;
2780         int             rc;
2781         ENTRY;
2782
2783         rc = linkea_init(ldata);
2784         if (rc < 0)
2785                 RETURN(rc);
2786
2787         if (linkno >= ldata->ld_leh->leh_reccount)
2788                 /* beyond last link */
2789                 RETURN(-ENODATA);
2790
2791         linkea_first_entry(ldata);
2792         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2793                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2794                                     parent_fid);
2795                 if (idx == linkno)
2796                         break;
2797
2798                 linkea_next_entry(ldata);
2799         }
2800
2801         if (idx < linkno)
2802                 RETURN(-ENODATA);
2803
2804         RETURN(0);
2805 }
2806
2807 /**
2808  * Get parent FID and name of an identified link. Operation is performed for
2809  * a given link number, letting the caller iterate over linkno to list one or
2810  * all links of an entry.
2811  *
2812  * \param[in]     file - File descriptor against which to perform the operation
2813  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2814  *                       on and the available size. It is eventually filled with
2815  *                       the requested information or left untouched on error
2816  *
2817  * \retval - 0 on success
2818  * \retval - Appropriate negative error code on failure
2819  */
2820 int ll_getparent(struct file *file, struct getparent __user *arg)
2821 {
2822         struct dentry           *dentry = file->f_dentry;
2823         struct inode            *inode = file->f_dentry->d_inode;
2824         struct linkea_data      *ldata;
2825         struct lu_buf            buf = LU_BUF_NULL;
2826         struct lu_name           ln;
2827         struct lu_fid            parent_fid;
2828         __u32                    linkno;
2829         __u32                    name_size;
2830         int                      rc;
2831
2832         ENTRY;
2833
2834         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2835             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2836                 RETURN(-EPERM);
2837
2838         if (get_user(name_size, &arg->gp_name_size))
2839                 RETURN(-EFAULT);
2840
2841         if (get_user(linkno, &arg->gp_linkno))
2842                 RETURN(-EFAULT);
2843
2844         if (name_size > PATH_MAX)
2845                 RETURN(-EINVAL);
2846
2847         OBD_ALLOC(ldata, sizeof(*ldata));
2848         if (ldata == NULL)
2849                 RETURN(-ENOMEM);
2850
2851         rc = linkea_data_new(ldata, &buf);
2852         if (rc < 0)
2853                 GOTO(ldata_free, rc);
2854
2855         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2856         if (rc < 0)
2857                 GOTO(lb_free, rc);
2858
2859         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2860         if (rc < 0)
2861                 GOTO(lb_free, rc);
2862
2863         if (ln.ln_namelen >= name_size)
2864                 GOTO(lb_free, rc = -EOVERFLOW);
2865
2866         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2867                 GOTO(lb_free, rc = -EFAULT);
2868
2869         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2870                 GOTO(lb_free, rc = -EFAULT);
2871
2872         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2873                 GOTO(lb_free, rc = -EFAULT);
2874
2875 lb_free:
2876         lu_buf_free(&buf);
2877 ldata_free:
2878         OBD_FREE(ldata, sizeof(*ldata));
2879
2880         RETURN(rc);
2881 }