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, 2015, 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/time.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49 #include <linux/user_namespace.h>
  50 #ifdef HAVE_UIDGID_HEADER
  51 # include <linux/uidgid.h>
  52 #endif
  53
  54 #include <lustre_ioctl.h>
  55 #include <lustre_ha.h>
  56 #include <lustre_dlm.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_disk.h>
  59 #include <lustre_param.h>
  60 #include <lustre_log.h>
  61 #include <cl_object.h>
  62 #include <obd_cksum.h>
  63 #include "llite_internal.h"
  64
  65 struct kmem_cache *ll_file_data_slab;
  66
  67 #ifndef log2
  68 #define log2(n) ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         unsigned long lru_page_max;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         mutex_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         lru_page_max = pages / 2;
  94
  95         /* initialize ll_cache data */
  96         sbi->ll_cache = cl_cache_init(lru_page_max);
  97         if (sbi->ll_cache == NULL) {
  98                 OBD_FREE(sbi, sizeof(*sbi));
  99                 RETURN(NULL);
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 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         sbi->ll_flags |= LL_SBI_VERBOSE;
 113 #ifdef ENABLE_CHECKSUM
 114         sbi->ll_flags |= LL_SBI_CHECKSUM;
 115 #endif
 116
 117 #ifdef HAVE_LRU_RESIZE_SUPPORT
 118         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 119 #endif
 120
 121         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_r_hist.oh_lock);
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_w_hist.oh_lock);
 126         }
 127
 128         /* metadata statahead is enabled by default */
 129         sbi->ll_sa_max = LL_SA_RPC_DEF;
 130         atomic_set(&sbi->ll_sa_total, 0);
 131         atomic_set(&sbi->ll_sa_wrong, 0);
 132         atomic_set(&sbi->ll_sa_running, 0);
 133         atomic_set(&sbi->ll_agl_total, 0);
 134         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 135         sbi->ll_flags |= LL_SBI_FAST_READ;
 136
 137         /* root squash */
 138         sbi->ll_squash.rsi_uid = 0;
 139         sbi->ll_squash.rsi_gid = 0;
 140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 141         init_rwsem(&sbi->ll_squash.rsi_sem);
 142
 143         RETURN(sbi);
 144 }
 145
 146 static void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 154                 if (sbi->ll_cache != NULL) {
 155                         cl_cache_decref(sbi->ll_cache);
 156                         sbi->ll_cache = NULL;
 157                 }
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 164                                     struct vfsmount *mnt)
 165 {
 166         struct inode *root = NULL;
 167         struct ll_sb_info *sbi = ll_s2sbi(sb);
 168         struct obd_device *obd;
 169         struct obd_statfs *osfs = NULL;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         u64 valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         OBD_ALLOC_PTR(osfs);
 190         if (osfs == NULL) {
 191                 OBD_FREE_PTR(data);
 192                 RETURN(-ENOMEM);
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 204                                   OBD_CONNECT_EINPROGRESS |
 205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 207                                   OBD_CONNECT_MAX_EASIZE |
 208                                   OBD_CONNECT_FLOCK_DEAD |
 209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 210                                   OBD_CONNECT_OPEN_BY_FID |
 211                                   OBD_CONNECT_DIR_STRIPE |
 212                                   OBD_CONNECT_BULK_MBITS |
 213                                   OBD_CONNECT_SUBTREE;
 214
 215 #ifdef HAVE_LRU_RESIZE_SUPPORT
 216         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 217                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 218 #endif
 219 #ifdef CONFIG_FS_POSIX_ACL
 220         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 221 #endif
 222
 223         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 224                 /* flag mdc connection as lightweight, only used for test
 225                  * purpose, use with care */
 226                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 227
 228         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 229         data->ocd_version = LUSTRE_VERSION_CODE;
 230
 231         if (sb->s_flags & MS_RDONLY)
 232                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 233         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 234                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 235
 236         if (sbi->ll_flags & LL_SBI_FLOCK)
 237                 sbi->ll_fop = &ll_file_operations_flock;
 238         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 239                 sbi->ll_fop = &ll_file_operations;
 240         else
 241                 sbi->ll_fop = &ll_file_operations_noflock;
 242
 243         /* real client */
 244         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 245         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 246                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 247
 248         /* always ping even if server suppress_pings */
 249         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 250                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 251
 252         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 253
 254         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 255         if (err == -EBUSY) {
 256                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 257                                    "recovery, of which this client is not a "
 258                                    "part. Please wait for recovery to complete,"
 259                                    " abort, or time out.\n", md);
 260                 GOTO(out, err);
 261         } else if (err) {
 262                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 263                 GOTO(out, err);
 264         }
 265
 266         sbi->ll_md_exp->exp_connect_data = *data;
 267
 268         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 269                            LUSTRE_SEQ_METADATA);
 270         if (err) {
 271                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 272                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 273                 GOTO(out_md, err);
 274         }
 275
 276         /* For mount, we only need fs info from MDT0, and also in DNE, it
 277          * can make sure the client can be mounted as long as MDT0 is
 278          * avaible */
 279         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 280                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 281                         OBD_STATFS_FOR_MDT0);
 282         if (err)
 283                 GOTO(out_md_fid, err);
 284
 285         /* This needs to be after statfs to ensure connect has finished.
 286          * Note that "data" does NOT contain the valid connect reply.
 287          * If connecting to a 1.8 server there will be no LMV device, so
 288          * we can access the MDC export directly and exp_connect_flags will
 289          * be non-zero, but if accessing an upgraded 2.1 server it will
 290          * have the correct flags filled in.
 291          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 292         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 293         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 294             valid != CLIENT_CONNECT_MDT_REQD) {
 295                 char *buf;
 296
 297                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 298                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 299                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 300                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 301                                    "feature(s) needed for correct operation "
 302                                    "of this client (%s). Please upgrade "
 303                                    "server or downgrade client.\n",
 304                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 305                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 306                 GOTO(out_md_fid, err = -EPROTO);
 307         }
 308
 309         size = sizeof(*data);
 310         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 311                            KEY_CONN_DATA,  &size, data);
 312         if (err) {
 313                 CERROR("%s: Get connect data failed: rc = %d\n",
 314                        sbi->ll_md_exp->exp_obd->obd_name, err);
 315                 GOTO(out_md_fid, err);
 316         }
 317
 318         LASSERT(osfs->os_bsize);
 319         sb->s_blocksize = osfs->os_bsize;
 320         sb->s_blocksize_bits = log2(osfs->os_bsize);
 321         sb->s_magic = LL_SUPER_MAGIC;
 322         sb->s_maxbytes = MAX_LFS_FILESIZE;
 323         sbi->ll_namelen = osfs->os_namelen;
 324
 325         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 326             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 327                 LCONSOLE_INFO("Disabling user_xattr feature because "
 328                               "it is not supported on the server\n");
 329                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 330         }
 331
 332         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 333 #ifdef MS_POSIXACL
 334                 sb->s_flags |= MS_POSIXACL;
 335 #endif
 336                 sbi->ll_flags |= LL_SBI_ACL;
 337         } else {
 338                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags &= ~MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags &= ~LL_SBI_ACL;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 346                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 347                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 348                         LCONSOLE_INFO("client is set as remote by default.\n");
 349                 }
 350         } else {
 351                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 352                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 353                         LCONSOLE_INFO("client claims to be remote, but server "
 354                                       "rejected, forced to be local.\n");
 355                 }
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 359                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 362                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 363         else
 364                 sbi->ll_md_brw_pages = 1;
 365
 366         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 367                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 368
 369         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 370                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 371                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 372                                       "unknown maximum xattr size.\n", dt);
 373                 } else {
 374                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 375                         sbi->ll_xattr_cache_enabled = 1;
 376                 }
 377         }
 378
 379         obd = class_name2obd(dt);
 380         if (!obd) {
 381                 CERROR("DT %s: not setup or attached\n", dt);
 382                 GOTO(out_md_fid, err = -ENODEV);
 383         }
 384
 385         /* pass client page size via ocd_grant_blkbits, the server should report
 386          * back its backend blocksize for grant calculation purpose */
 387         data->ocd_grant_blkbits = PAGE_SHIFT;
 388
 389         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 390                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 391                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 392                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 393                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 394                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 395                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 396                                   OBD_CONNECT_MAXBYTES |
 397                                   OBD_CONNECT_EINPROGRESS |
 398                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 399                                   OBD_CONNECT_LAYOUTLOCK |
 400                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 401                                   OBD_CONNECT_BULK_MBITS;
 402
 403         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 404                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 405
 406         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 407                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 408                  * disabled by default, because it can still be enabled on the
 409                  * fly via /proc. As a consequence, we still need to come to an
 410                  * agreement on the supported algorithms at connect time */
 411                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 412
 413                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 414                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 415                 else
 416                         data->ocd_cksum_types = cksum_types_supported_client();
 417         }
 418
 419 #ifdef HAVE_LRU_RESIZE_SUPPORT
 420         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 421 #endif
 422         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 423                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 424
 425         /* always ping even if server suppress_pings */
 426         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 427                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 428
 429         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 430                "ocd_grant: %d\n", data->ocd_connect_flags,
 431                data->ocd_version, data->ocd_grant);
 432
 433         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 434         obd->obd_upcall.onu_upcall = cl_ocd_update;
 435
 436         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 437
 438         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 439                           NULL);
 440         if (err == -EBUSY) {
 441                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 442                                    "recovery, of which this client is not a "
 443                                    "part.  Please wait for recovery to "
 444                                    "complete, abort, or time out.\n", dt);
 445                 GOTO(out_md, err);
 446         } else if (err) {
 447                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 448                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 449                 GOTO(out_md, err);
 450         }
 451
 452         sbi->ll_dt_exp->exp_connect_data = *data;
 453
 454         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 455                            LUSTRE_SEQ_METADATA);
 456         if (err) {
 457                 CERROR("%s: Can't init data layer FID infrastructure, "
 458                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 459                 GOTO(out_dt, err);
 460         }
 461
 462         mutex_lock(&sbi->ll_lco.lco_lock);
 463         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 464         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 465         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 466         mutex_unlock(&sbi->ll_lco.lco_lock);
 467
 468         fid_zero(&sbi->ll_root_fid);
 469         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 470                            &sbi->ll_root_fid);
 471         if (err) {
 472                 CERROR("cannot mds_connect: rc = %d\n", err);
 473                 GOTO(out_lock_cn_cb, err);
 474         }
 475         if (!fid_is_sane(&sbi->ll_root_fid)) {
 476                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 477                        sbi->ll_md_exp->exp_obd->obd_name,
 478                        PFID(&sbi->ll_root_fid));
 479                 GOTO(out_lock_cn_cb, err = -EINVAL);
 480         }
 481         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 482
 483         sb->s_op = &lustre_super_operations;
 484 #if THREAD_SIZE >= 8192 /*b=17630*/
 485         sb->s_export_op = &lustre_export_operations;
 486 #endif
 487
 488         /* make root inode
 489          * XXX: move this to after cbd setup? */
 490         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 491         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 492                 valid |= OBD_MD_FLRMTPERM;
 493         else if (sbi->ll_flags & LL_SBI_ACL)
 494                 valid |= OBD_MD_FLACL;
 495
 496         OBD_ALLOC_PTR(op_data);
 497         if (op_data == NULL)
 498                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 499
 500         op_data->op_fid1 = sbi->ll_root_fid;
 501         op_data->op_mode = 0;
 502         op_data->op_valid = valid;
 503
 504         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 505
 506         OBD_FREE_PTR(op_data);
 507         if (err) {
 508                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 509                        sbi->ll_md_exp->exp_obd->obd_name, err);
 510                 GOTO(out_lock_cn_cb, err);
 511         }
 512
 513         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 514                                sbi->ll_md_exp, &lmd);
 515         if (err) {
 516                 CERROR("failed to understand root inode md: rc = %d\n", err);
 517                 ptlrpc_req_finished(request);
 518                 GOTO(out_lock_cn_cb, err);
 519         }
 520
 521         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 522         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 523                                             sbi->ll_flags & LL_SBI_32BIT_API),
 524                        &lmd);
 525         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 526         ptlrpc_req_finished(request);
 527
 528         if (IS_ERR(root)) {
 529 #ifdef CONFIG_FS_POSIX_ACL
 530                 if (lmd.posix_acl) {
 531                         posix_acl_release(lmd.posix_acl);
 532                         lmd.posix_acl = NULL;
 533                 }
 534 #endif
 535                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 536                 root = NULL;
 537                 CERROR("lustre_lite: bad iget4 for root\n");
 538                 GOTO(out_root, err);
 539         }
 540
 541 #ifdef CONFIG_FS_POSIX_ACL
 542         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 543                 rct_init(&sbi->ll_rct);
 544                 et_init(&sbi->ll_et);
 545         }
 546 #endif
 547
 548         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 549         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 550                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 551                                  NULL);
 552         cl_sb_init(sb);
 553
 554         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 555                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 556                                  sbi->ll_cache, NULL);
 557
 558         sb->s_root = d_make_root(root);
 559         if (sb->s_root == NULL) {
 560                 CERROR("%s: can't make root dentry\n",
 561                         ll_get_fsname(sb, NULL, 0));
 562                 GOTO(out_root, err = -ENOMEM);
 563         }
 564 #ifdef HAVE_DCACHE_LOCK
 565         sb->s_root->d_op = &ll_d_ops;
 566 #endif
 567
 568         sbi->ll_sdev_orig = sb->s_dev;
 569
 570         /* We set sb->s_dev equal on all lustre clients in order to support
 571          * NFS export clustering.  NFSD requires that the FSID be the same
 572          * on all clients. */
 573         /* s_dev is also used in lt_compare() to compare two fs, but that is
 574          * only a node-local comparison. */
 575         uuid = obd_get_uuid(sbi->ll_md_exp);
 576         if (uuid != NULL)
 577                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 578
 579         if (data != NULL)
 580                 OBD_FREE_PTR(data);
 581         if (osfs != NULL)
 582                 OBD_FREE_PTR(osfs);
 583         if (proc_lustre_fs_root != NULL) {
 584                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 585                                                   dt, md);
 586                 if (err < 0) {
 587                         CERROR("%s: could not register mount in lprocfs: "
 588                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 589                         err = 0;
 590                 }
 591         }
 592
 593         RETURN(err);
 594 out_root:
 595         if (root)
 596                 iput(root);
 597 out_lock_cn_cb:
 598         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 599 out_dt:
 600         obd_disconnect(sbi->ll_dt_exp);
 601         sbi->ll_dt_exp = NULL;
 602 out_md_fid:
 603         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 604 out_md:
 605         obd_disconnect(sbi->ll_md_exp);
 606         sbi->ll_md_exp = NULL;
 607 out:
 608         if (data != NULL)
 609                 OBD_FREE_PTR(data);
 610         if (osfs != NULL)
 611                 OBD_FREE_PTR(osfs);
 612         return err;
 613 }
 614
 615 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 616 {
 617         int size, rc;
 618
 619         size = sizeof(*lmmsize);
 620         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 621                           KEY_MAX_EASIZE, &size, lmmsize);
 622         if (rc != 0) {
 623                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 624                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 625                 RETURN(rc);
 626         }
 627
 628         size = sizeof(int);
 629         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 630                           KEY_MAX_EASIZE, &size, lmmsize);
 631         if (rc)
 632                 CERROR("Get max mdsize error rc %d\n", rc);
 633
 634         RETURN(rc);
 635 }
 636
 637 /**
 638  * Get the value of the default_easize parameter.
 639  *
 640  * \see client_obd::cl_default_mds_easize
 641  *
 642  * \param[in] sbi       superblock info for this filesystem
 643  * \param[out] lmmsize  pointer to storage location for value
 644  *
 645  * \retval 0            on success
 646  * \retval negative     negated errno on failure
 647  */
 648 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 649 {
 650         int size, rc;
 651
 652         size = sizeof(int);
 653         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 654                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 655         if (rc)
 656                 CERROR("Get default mdsize error rc %d\n", rc);
 657
 658         RETURN(rc);
 659 }
 660
 661 /**
 662  * Set the default_easize parameter to the given value.
 663  *
 664  * \see client_obd::cl_default_mds_easize
 665  *
 666  * \param[in] sbi       superblock info for this filesystem
 667  * \param[in] lmmsize   the size to set
 668  *
 669  * \retval 0            on success
 670  * \retval negative     negated errno on failure
 671  */
 672 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 673 {
 674         int rc;
 675
 676         if (lmmsize < sizeof(struct lov_mds_md) ||
 677             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 678                 return -EINVAL;
 679
 680         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 681                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 682                                 sizeof(int), &lmmsize, NULL);
 683
 684         RETURN(rc);
 685 }
 686
 687 static void ll_dump_inode(struct inode *inode)
 688 {
 689         struct ll_d_hlist_node *tmp;
 690         int dentry_count = 0;
 691
 692         LASSERT(inode != NULL);
 693
 694         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 695                 dentry_count++;
 696
 697         CERROR("%s: inode %p dump: dev=%s fid="DFID
 698                " mode=%o count=%u, %d dentries\n",
 699                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 700                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 701                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 702 }
 703
 704 void lustre_dump_dentry(struct dentry *dentry, int recur)
 705 {
 706         struct list_head *tmp;
 707         int subdirs = 0;
 708
 709         LASSERT(dentry != NULL);
 710
 711         list_for_each(tmp, &dentry->d_subdirs)
 712                 subdirs++;
 713
 714         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 715                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 716                dentry->d_name.len, dentry->d_name.name,
 717                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 718                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 719                dentry->d_flags, dentry->d_fsdata, subdirs);
 720         if (dentry->d_inode != NULL)
 721                 ll_dump_inode(dentry->d_inode);
 722
 723         if (recur == 0)
 724                 return;
 725
 726         list_for_each(tmp, &dentry->d_subdirs) {
 727                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 728                 lustre_dump_dentry(d, recur - 1);
 729         }
 730 }
 731
 732 static void client_common_put_super(struct super_block *sb)
 733 {
 734         struct ll_sb_info *sbi = ll_s2sbi(sb);
 735         ENTRY;
 736
 737 #ifdef CONFIG_FS_POSIX_ACL
 738         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 739                 et_fini(&sbi->ll_et);
 740                 rct_fini(&sbi->ll_rct);
 741         }
 742 #endif
 743
 744         cl_sb_fini(sb);
 745
 746         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 747         obd_disconnect(sbi->ll_dt_exp);
 748         sbi->ll_dt_exp = NULL;
 749
 750         lprocfs_unregister_mountpoint(sbi);
 751
 752         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 753         obd_disconnect(sbi->ll_md_exp);
 754         sbi->ll_md_exp = NULL;
 755
 756         EXIT;
 757 }
 758
 759 void ll_kill_super(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi;
 762         ENTRY;
 763
 764         /* not init sb ?*/
 765         if (!(sb->s_flags & MS_ACTIVE))
 766                 return;
 767
 768         sbi = ll_s2sbi(sb);
 769         /* we need restore s_dev from changed for clustred NFS before put_super
 770          * because new kernels have cached s_dev and change sb->s_dev in
 771          * put_super not affected real removing devices */
 772         if (sbi) {
 773                 sb->s_dev = sbi->ll_sdev_orig;
 774                 sbi->ll_umounting = 1;
 775
 776                 /* wait running statahead threads to quit */
 777                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 778                         set_current_state(TASK_UNINTERRUPTIBLE);
 779                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 780                 }
 781         }
 782
 783         EXIT;
 784 }
 785
 786 static inline int ll_set_opt(const char *opt, char *data, int fl)
 787 {
 788         if (strncmp(opt, data, strlen(opt)) != 0)
 789                 return(0);
 790         else
 791                 return(fl);
 792 }
 793
 794 /* non-client-specific mount options are parsed in lmd_parse */
 795 static int ll_options(char *options, int *flags)
 796 {
 797         int tmp;
 798         char *s1 = options, *s2;
 799         ENTRY;
 800
 801         if (!options)
 802                 RETURN(0);
 803
 804         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 805
 806         while (*s1) {
 807                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 808                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 809                 if (tmp) {
 810                         *flags |= tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("context", s1, 1);
 839                 if (tmp)
 840                         goto next;
 841                 tmp = ll_set_opt("fscontext", s1, 1);
 842                 if (tmp)
 843                         goto next;
 844                 tmp = ll_set_opt("defcontext", s1, 1);
 845                 if (tmp)
 846                         goto next;
 847                 tmp = ll_set_opt("rootcontext", s1, 1);
 848                 if (tmp)
 849                         goto next;
 850                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865
 866                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 892                 if (tmp) {
 893                         *flags &= ~tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 897                 if (tmp) {
 898                         *flags |= tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 907                 if (tmp) {
 908                         *flags &= ~tmp;
 909                         goto next;
 910                 }
 911                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 912                 if (tmp) {
 913                         *flags |= tmp;
 914                         goto next;
 915                 }
 916                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 917                                    s1);
 918                 RETURN(-EINVAL);
 919
 920 next:
 921                 /* Find next opt */
 922                 s2 = strchr(s1, ',');
 923                 if (s2 == NULL)
 924                         break;
 925                 s1 = s2 + 1;
 926         }
 927         RETURN(0);
 928 }
 929
 930 void ll_lli_init(struct ll_inode_info *lli)
 931 {
 932         lli->lli_inode_magic = LLI_INODE_MAGIC;
 933         lli->lli_flags = 0;
 934         spin_lock_init(&lli->lli_lock);
 935         lli->lli_posix_acl = NULL;
 936         lli->lli_remote_perms = NULL;
 937         mutex_init(&lli->lli_rmtperm_mutex);
 938         /* Do not set lli_fid, it has been initialized already. */
 939         fid_zero(&lli->lli_pfid);
 940         atomic_set(&lli->lli_open_count, 0);
 941         lli->lli_rmtperm_time = 0;
 942         lli->lli_mds_read_och = NULL;
 943         lli->lli_mds_write_och = NULL;
 944         lli->lli_mds_exec_och = NULL;
 945         lli->lli_open_fd_read_count = 0;
 946         lli->lli_open_fd_write_count = 0;
 947         lli->lli_open_fd_exec_count = 0;
 948         mutex_init(&lli->lli_och_mutex);
 949         spin_lock_init(&lli->lli_agl_lock);
 950         spin_lock_init(&lli->lli_layout_lock);
 951         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 952         lli->lli_clob = NULL;
 953
 954         init_rwsem(&lli->lli_xattrs_list_rwsem);
 955         mutex_init(&lli->lli_xattrs_enq_lock);
 956
 957         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 958         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 959                 mutex_init(&lli->lli_readdir_mutex);
 960                 lli->lli_opendir_key = NULL;
 961                 lli->lli_sai = NULL;
 962                 spin_lock_init(&lli->lli_sa_lock);
 963                 lli->lli_opendir_pid = 0;
 964                 lli->lli_sa_enabled = 0;
 965                 lli->lli_def_stripe_offset = -1;
 966         } else {
 967                 mutex_init(&lli->lli_size_mutex);
 968                 lli->lli_symlink_name = NULL;
 969                 init_rwsem(&lli->lli_trunc_sem);
 970                 range_lock_tree_init(&lli->lli_write_tree);
 971                 init_rwsem(&lli->lli_glimpse_sem);
 972                 lli->lli_glimpse_time = 0;
 973                 INIT_LIST_HEAD(&lli->lli_agl_list);
 974                 lli->lli_agl_index = 0;
 975                 lli->lli_async_rc = 0;
 976         }
 977         mutex_init(&lli->lli_layout_mutex);
 978 }
 979
 980 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 981 {
 982         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 983
 984         bdi->name = "lustre";
 985         return bdi_register(bdi, NULL, "lustre-%d",
 986                             atomic_inc_return(&ll_bdi_num));
 987 }
 988
 989 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 990 {
 991         struct lustre_profile *lprof = NULL;
 992         struct lustre_sb_info *lsi = s2lsi(sb);
 993         struct ll_sb_info *sbi;
 994         char  *dt = NULL, *md = NULL;
 995         char  *profilenm = get_profile_name(sb);
 996         struct config_llog_instance *cfg;
 997         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 998         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 999         int    err;
1000         ENTRY;
1001
1002         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1003
1004         OBD_ALLOC_PTR(cfg);
1005         if (cfg == NULL)
1006                 RETURN(-ENOMEM);
1007
1008         try_module_get(THIS_MODULE);
1009
1010         /* client additional sb info */
1011         lsi->lsi_llsbi = sbi = ll_init_sbi();
1012         if (!sbi) {
1013                 module_put(THIS_MODULE);
1014                 OBD_FREE_PTR(cfg);
1015                 RETURN(-ENOMEM);
1016         }
1017
1018         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1019         if (err)
1020                 GOTO(out_free, err);
1021
1022         err = bdi_init(&lsi->lsi_bdi);
1023         if (err)
1024                 GOTO(out_free, err);
1025         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1026 #ifdef HAVE_BDI_CAP_MAP_COPY
1027         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1028 #else
1029         lsi->lsi_bdi.capabilities = 0;
1030 #endif
1031         err = ll_bdi_register(&lsi->lsi_bdi);
1032         if (err)
1033                 GOTO(out_free, err);
1034
1035         sb->s_bdi = &lsi->lsi_bdi;
1036 #ifndef HAVE_DCACHE_LOCK
1037         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1038         sb->s_d_op = &ll_d_ops;
1039 #endif
1040
1041         /* Generate a string unique to this super, in case some joker tries
1042            to mount the same fs at two mount points.
1043            Use the address of the super itself.*/
1044         cfg->cfg_instance = sb;
1045         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1046         cfg->cfg_callback = class_config_llog_handler;
1047         /* set up client obds */
1048         err = lustre_process_log(sb, profilenm, cfg);
1049         if (err < 0)
1050                 GOTO(out_free, err);
1051
1052         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1053         lprof = class_get_profile(profilenm);
1054         if (lprof == NULL) {
1055                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1056                                    " read from the MGS.  Does that filesystem "
1057                                    "exist?\n", profilenm);
1058                 GOTO(out_free, err = -EINVAL);
1059         }
1060         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1061                lprof->lp_md, lprof->lp_dt);
1062
1063         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1064         if (!dt)
1065                 GOTO(out_free, err = -ENOMEM);
1066         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1067
1068         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1069         if (!md)
1070                 GOTO(out_free, err = -ENOMEM);
1071         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1072
1073         /* connections, registrations, sb setup */
1074         err = client_common_fill_super(sb, md, dt, mnt);
1075         if (err < 0)
1076                 GOTO(out_free, err);
1077
1078         sbi->ll_client_common_fill_super_succeeded = 1;
1079
1080 out_free:
1081         if (md)
1082                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1083         if (dt)
1084                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1085         if (lprof != NULL)
1086                 class_put_profile(lprof);
1087         if (err)
1088                 ll_put_super(sb);
1089         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1090                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1091
1092         OBD_FREE_PTR(cfg);
1093         RETURN(err);
1094 } /* ll_fill_super */
1095
1096 void ll_put_super(struct super_block *sb)
1097 {
1098         struct config_llog_instance cfg, params_cfg;
1099         struct obd_device *obd;
1100         struct lustre_sb_info *lsi = s2lsi(sb);
1101         struct ll_sb_info *sbi = ll_s2sbi(sb);
1102         char *profilenm = get_profile_name(sb);
1103         long ccc_count;
1104         int next, force = 1, rc = 0;
1105         ENTRY;
1106
1107         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1108
1109         cfg.cfg_instance = sb;
1110         lustre_end_log(sb, profilenm, &cfg);
1111
1112         params_cfg.cfg_instance = sb;
1113         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1114
1115         if (sbi->ll_md_exp) {
1116                 obd = class_exp2obd(sbi->ll_md_exp);
1117                 if (obd)
1118                         force = obd->obd_force;
1119         }
1120
1121         /* Wait for unstable pages to be committed to stable storage */
1122         if (force == 0) {
1123                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1124                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1125                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1126                         &lwi);
1127         }
1128
1129         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1130         if (force == 0 && rc != -EINTR)
1131                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1132
1133
1134         /* We need to set force before the lov_disconnect in
1135            lustre_common_put_super, since l_d cleans up osc's as well. */
1136         if (force) {
1137                 next = 0;
1138                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1139                                                      &next)) != NULL) {
1140                         obd->obd_force = force;
1141                 }
1142         }
1143
1144         if (sbi->ll_client_common_fill_super_succeeded) {
1145                 /* Only if client_common_fill_super succeeded */
1146                 client_common_put_super(sb);
1147         }
1148
1149         next = 0;
1150         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1151                 class_manual_cleanup(obd);
1152         }
1153
1154         if (sbi->ll_flags & LL_SBI_VERBOSE)
1155                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1156
1157         if (profilenm)
1158                 class_del_profile(profilenm);
1159
1160         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1161                 bdi_destroy(&lsi->lsi_bdi);
1162                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1163         }
1164
1165         ll_free_sbi(sb);
1166         lsi->lsi_llsbi = NULL;
1167
1168         lustre_common_put_super(sb);
1169
1170         cl_env_cache_purge(~0);
1171
1172         module_put(THIS_MODULE);
1173
1174         EXIT;
1175 } /* client_put_super */
1176
1177 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1178 {
1179         struct inode *inode = NULL;
1180
1181         /* NOTE: we depend on atomic igrab() -bzzz */
1182         lock_res_and_lock(lock);
1183         if (lock->l_resource->lr_lvb_inode) {
1184                 struct ll_inode_info * lli;
1185                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1186                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1187                         inode = igrab(lock->l_resource->lr_lvb_inode);
1188                 } else {
1189                         inode = lock->l_resource->lr_lvb_inode;
1190                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1191                                          D_WARNING, lock, "lr_lvb_inode %p is "
1192                                          "bogus: magic %08x",
1193                                          lock->l_resource->lr_lvb_inode,
1194                                          lli->lli_inode_magic);
1195                         inode = NULL;
1196                 }
1197         }
1198         unlock_res_and_lock(lock);
1199         return inode;
1200 }
1201
1202 void ll_dir_clear_lsm_md(struct inode *inode)
1203 {
1204         struct ll_inode_info *lli = ll_i2info(inode);
1205
1206         LASSERT(S_ISDIR(inode->i_mode));
1207
1208         if (lli->lli_lsm_md != NULL) {
1209                 lmv_free_memmd(lli->lli_lsm_md);
1210                 lli->lli_lsm_md = NULL;
1211         }
1212 }
1213
1214 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1215                                       const struct lu_fid *fid,
1216                                       struct lustre_md *md)
1217 {
1218         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1219         struct mdt_body         *body = md->body;
1220         struct inode            *inode;
1221         ino_t                   ino;
1222         ENTRY;
1223
1224         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1225         inode = iget_locked(sb, ino);
1226         if (inode == NULL) {
1227                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1228                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1229                 RETURN(ERR_PTR(-ENOENT));
1230         }
1231
1232         if (inode->i_state & I_NEW) {
1233                 struct ll_inode_info *lli = ll_i2info(inode);
1234                 struct lmv_stripe_md *lsm = md->lmv;
1235
1236                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1237                                 (body->mbo_mode & S_IFMT);
1238                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1239                          PFID(fid));
1240
1241                 LTIME_S(inode->i_mtime) = 0;
1242                 LTIME_S(inode->i_atime) = 0;
1243                 LTIME_S(inode->i_ctime) = 0;
1244                 inode->i_rdev = 0;
1245
1246 #ifdef HAVE_BACKING_DEV_INFO
1247                 /* initializing backing dev info. */
1248                 inode->i_mapping->backing_dev_info =
1249                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1250 #endif
1251                 inode->i_op = &ll_dir_inode_operations;
1252                 inode->i_fop = &ll_dir_operations;
1253                 lli->lli_fid = *fid;
1254                 ll_lli_init(lli);
1255
1256                 LASSERT(lsm != NULL);
1257                 /* master object FID */
1258                 lli->lli_pfid = body->mbo_fid1;
1259                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1260                        lli, PFID(fid), PFID(&lli->lli_pfid));
1261                 unlock_new_inode(inode);
1262         }
1263
1264         RETURN(inode);
1265 }
1266
1267 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1268 {
1269         struct lu_fid *fid;
1270         struct lmv_stripe_md *lsm = md->lmv;
1271         int i;
1272
1273         LASSERT(lsm != NULL);
1274         /* XXX sigh, this lsm_root initialization should be in
1275          * LMV layer, but it needs ll_iget right now, so we
1276          * put this here right now. */
1277         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1278                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1279                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1280                 /* Unfortunately ll_iget will call ll_update_inode,
1281                  * where the initialization of slave inode is slightly
1282                  * different, so it reset lsm_md to NULL to avoid
1283                  * initializing lsm for slave inode. */
1284                 /* For migrating inode, master stripe and master object will
1285                  * be same, so we only need assign this inode */
1286                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1287                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1288                 else
1289                         lsm->lsm_md_oinfo[i].lmo_root =
1290                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1291
1292                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1293                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1294
1295                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1296                         return rc;
1297                 }
1298         }
1299
1300         return 0;
1301 }
1302
1303 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1304                                 const struct lmv_stripe_md *lsm_md2)
1305 {
1306         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1307                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1308                lsm_md1->lsm_md_master_mdt_index ==
1309                                         lsm_md2->lsm_md_master_mdt_index &&
1310                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1311                lsm_md1->lsm_md_layout_version ==
1312                                         lsm_md2->lsm_md_layout_version &&
1313                strcmp(lsm_md1->lsm_md_pool_name,
1314                       lsm_md2->lsm_md_pool_name) == 0;
1315 }
1316
1317 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1318 {
1319         struct ll_inode_info *lli = ll_i2info(inode);
1320         struct lmv_stripe_md *lsm = md->lmv;
1321         int     rc;
1322         ENTRY;
1323
1324         LASSERT(S_ISDIR(inode->i_mode));
1325         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1326                PFID(ll_inode2fid(inode)));
1327
1328         /* no striped information from request. */
1329         if (lsm == NULL) {
1330                 if (lli->lli_lsm_md == NULL) {
1331                         RETURN(0);
1332                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1333                                                 LMV_HASH_FLAG_MIGRATION) {
1334                         /* migration is done, the temporay MIGRATE layout has
1335                          * been removed */
1336                         CDEBUG(D_INODE, DFID" finish migration.\n",
1337                                PFID(ll_inode2fid(inode)));
1338                         lmv_free_memmd(lli->lli_lsm_md);
1339                         lli->lli_lsm_md = NULL;
1340                         RETURN(0);
1341                 } else {
1342                         /* The lustre_md from req does not include stripeEA,
1343                          * see ll_md_setattr */
1344                         RETURN(0);
1345                 }
1346         }
1347
1348         /* set the directory layout */
1349         if (lli->lli_lsm_md == NULL) {
1350                 struct cl_attr  *attr;
1351
1352                 rc = ll_init_lsm_md(inode, md);
1353                 if (rc != 0)
1354                         RETURN(rc);
1355
1356                 /* set md->lmv to NULL, so the following free lustre_md
1357                  * will not free this lsm */
1358                 md->lmv = NULL;
1359                 lli->lli_lsm_md = lsm;
1360
1361                 OBD_ALLOC_PTR(attr);
1362                 if (attr == NULL)
1363                         RETURN(-ENOMEM);
1364
1365                 /* validate the lsm */
1366                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1367                                    ll_md_blocking_ast);
1368                 if (rc != 0) {
1369                         OBD_FREE_PTR(attr);
1370                         RETURN(rc);
1371                 }
1372
1373                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1374                         md->body->mbo_nlink = attr->cat_nlink;
1375                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1376                         md->body->mbo_size = attr->cat_size;
1377                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1378                         md->body->mbo_atime = attr->cat_atime;
1379                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1380                         md->body->mbo_ctime = attr->cat_ctime;
1381                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1382                         md->body->mbo_mtime = attr->cat_mtime;
1383
1384                 OBD_FREE_PTR(attr);
1385
1386                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1387                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1388                 RETURN(0);
1389         }
1390
1391         /* Compare the old and new stripe information */
1392         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1393                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1394                 int                     idx;
1395
1396                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1397                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1398                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1399                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1400                        inode, lsm, old_lsm,
1401                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1402                        lsm->lsm_md_stripe_count,
1403                        old_lsm->lsm_md_stripe_count,
1404                        lsm->lsm_md_master_mdt_index,
1405                        old_lsm->lsm_md_master_mdt_index,
1406                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1407                        lsm->lsm_md_layout_version,
1408                        old_lsm->lsm_md_layout_version,
1409                        lsm->lsm_md_pool_name,
1410                        old_lsm->lsm_md_pool_name);
1411
1412                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1413                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1414                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1415                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1416                 }
1417
1418                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1419                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1420                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1421                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1422                 }
1423
1424                 RETURN(-EIO);
1425         }
1426
1427         RETURN(0);
1428 }
1429
1430 void ll_clear_inode(struct inode *inode)
1431 {
1432         struct ll_inode_info *lli = ll_i2info(inode);
1433         struct ll_sb_info *sbi = ll_i2sbi(inode);
1434         ENTRY;
1435
1436         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1437                PFID(ll_inode2fid(inode)), inode);
1438
1439         if (S_ISDIR(inode->i_mode)) {
1440                 /* these should have been cleared in ll_file_release */
1441                 LASSERT(lli->lli_opendir_key == NULL);
1442                 LASSERT(lli->lli_sai == NULL);
1443                 LASSERT(lli->lli_opendir_pid == 0);
1444         }
1445
1446         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1447
1448         LASSERT(!lli->lli_open_fd_write_count);
1449         LASSERT(!lli->lli_open_fd_read_count);
1450         LASSERT(!lli->lli_open_fd_exec_count);
1451
1452         if (lli->lli_mds_write_och)
1453                 ll_md_real_close(inode, FMODE_WRITE);
1454         if (lli->lli_mds_exec_och)
1455                 ll_md_real_close(inode, FMODE_EXEC);
1456         if (lli->lli_mds_read_och)
1457                 ll_md_real_close(inode, FMODE_READ);
1458
1459         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1460                 OBD_FREE(lli->lli_symlink_name,
1461                          strlen(lli->lli_symlink_name) + 1);
1462                 lli->lli_symlink_name = NULL;
1463         }
1464
1465         ll_xattr_cache_destroy(inode);
1466
1467         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1468                 LASSERT(lli->lli_posix_acl == NULL);
1469                 if (lli->lli_remote_perms) {
1470                         free_rmtperm_hash(lli->lli_remote_perms);
1471                         lli->lli_remote_perms = NULL;
1472                 }
1473         }
1474 #ifdef CONFIG_FS_POSIX_ACL
1475         else if (lli->lli_posix_acl) {
1476                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1477                 LASSERT(lli->lli_remote_perms == NULL);
1478                 posix_acl_release(lli->lli_posix_acl);
1479                 lli->lli_posix_acl = NULL;
1480         }
1481 #endif
1482         lli->lli_inode_magic = LLI_INODE_DEAD;
1483
1484         if (S_ISDIR(inode->i_mode))
1485                 ll_dir_clear_lsm_md(inode);
1486         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1487                 LASSERT(list_empty(&lli->lli_agl_list));
1488
1489         /*
1490          * XXX This has to be done before lsm is freed below, because
1491          * cl_object still uses inode lsm.
1492          */
1493         cl_inode_fini(inode);
1494
1495         EXIT;
1496 }
1497
1498 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1499 {
1500         struct lustre_md md;
1501         struct inode *inode = dentry->d_inode;
1502         struct ll_sb_info *sbi = ll_i2sbi(inode);
1503         struct ptlrpc_request *request = NULL;
1504         int rc, ia_valid;
1505         ENTRY;
1506
1507         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1508                                      LUSTRE_OPC_ANY, NULL);
1509         if (IS_ERR(op_data))
1510                 RETURN(PTR_ERR(op_data));
1511
1512         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1513         if (rc) {
1514                 ptlrpc_req_finished(request);
1515                 if (rc == -ENOENT) {
1516                         clear_nlink(inode);
1517                         /* Unlinked special device node? Or just a race?
1518                          * Pretend we done everything. */
1519                         if (!S_ISREG(inode->i_mode) &&
1520                             !S_ISDIR(inode->i_mode)) {
1521                                 ia_valid = op_data->op_attr.ia_valid;
1522                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1523                                 rc = simple_setattr(dentry, &op_data->op_attr);
1524                                 op_data->op_attr.ia_valid = ia_valid;
1525                         }
1526                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1527                         CERROR("md_setattr fails: rc = %d\n", rc);
1528                 }
1529                 RETURN(rc);
1530         }
1531
1532         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1533                               sbi->ll_md_exp, &md);
1534         if (rc) {
1535                 ptlrpc_req_finished(request);
1536                 RETURN(rc);
1537         }
1538
1539         ia_valid = op_data->op_attr.ia_valid;
1540         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1541          * cache is not cleared yet. */
1542         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1543         if (S_ISREG(inode->i_mode))
1544                 mutex_lock(&inode->i_mutex);
1545         rc = simple_setattr(dentry, &op_data->op_attr);
1546         if (S_ISREG(inode->i_mode))
1547                 mutex_unlock(&inode->i_mutex);
1548         op_data->op_attr.ia_valid = ia_valid;
1549
1550         rc = ll_update_inode(inode, &md);
1551         ptlrpc_req_finished(request);
1552
1553         RETURN(rc);
1554 }
1555
1556 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1557  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1558  * keep these values until such a time that objects are allocated for it.
1559  * We do the MDS operations first, as it is checking permissions for us.
1560  * We don't to the MDS RPC if there is nothing that we want to store there,
1561  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1562  * going to do an RPC anyways.
1563  *
1564  * If we are doing a truncate, we will send the mtime and ctime updates
1565  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1566  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1567  * at the same time.
1568  *
1569  * In case of HSMimport, we only set attr on MDS.
1570  */
1571 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1572 {
1573         struct inode *inode = dentry->d_inode;
1574         struct ll_inode_info *lli = ll_i2info(inode);
1575         struct md_op_data *op_data = NULL;
1576         int rc = 0;
1577         ENTRY;
1578
1579         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1580                "valid %x, hsm_import %d\n",
1581                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1582                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1583                hsm_import);
1584
1585         if (attr->ia_valid & ATTR_SIZE) {
1586                 /* Check new size against VFS/VM file size limit and rlimit */
1587                 rc = inode_newsize_ok(inode, attr->ia_size);
1588                 if (rc)
1589                         RETURN(rc);
1590
1591                 /* The maximum Lustre file size is variable, based on the
1592                  * OST maximum object size and number of stripes.  This
1593                  * needs another check in addition to the VFS check above. */
1594                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1595                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1596                                PFID(&lli->lli_fid), attr->ia_size,
1597                                ll_file_maxbytes(inode));
1598                         RETURN(-EFBIG);
1599                 }
1600
1601                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1602         }
1603
1604         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1605         if (attr->ia_valid & TIMES_SET_FLAGS) {
1606                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1607                     !cfs_capable(CFS_CAP_FOWNER))
1608                         RETURN(-EPERM);
1609         }
1610
1611         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1612         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1613             (attr->ia_valid & ATTR_CTIME)) {
1614                 attr->ia_ctime = CURRENT_TIME;
1615                 attr->ia_valid |= ATTR_CTIME_SET;
1616         }
1617         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1618             (attr->ia_valid & ATTR_ATIME)) {
1619                 attr->ia_atime = CURRENT_TIME;
1620                 attr->ia_valid |= ATTR_ATIME_SET;
1621         }
1622         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1623             (attr->ia_valid & ATTR_MTIME)) {
1624                 attr->ia_mtime = CURRENT_TIME;
1625                 attr->ia_valid |= ATTR_MTIME_SET;
1626         }
1627
1628         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1629                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1630                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1631                        cfs_time_current_sec());
1632
1633         if (S_ISREG(inode->i_mode)) {
1634                 if (attr->ia_valid & ATTR_SIZE)
1635                         inode_dio_write_done(inode);
1636                 mutex_unlock(&inode->i_mutex);
1637         }
1638
1639         /* We always do an MDS RPC, even if we're only changing the size;
1640          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1641
1642         OBD_ALLOC_PTR(op_data);
1643         if (op_data == NULL)
1644                 GOTO(out, rc = -ENOMEM);
1645
1646         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1647                 /* If we are changing file size, file content is
1648                  * modified, flag it. */
1649                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1650                 op_data->op_bias |= MDS_DATA_MODIFIED;
1651                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1652         }
1653
1654         op_data->op_attr = *attr;
1655
1656         rc = ll_md_setattr(dentry, op_data);
1657         if (rc)
1658                 GOTO(out, rc);
1659
1660         if (!S_ISREG(inode->i_mode) || hsm_import)
1661                 GOTO(out, rc = 0);
1662
1663         if (attr->ia_valid & (ATTR_SIZE |
1664                               ATTR_ATIME | ATTR_ATIME_SET |
1665                               ATTR_MTIME | ATTR_MTIME_SET |
1666                               ATTR_CTIME | ATTR_CTIME_SET)) {
1667                 /* For truncate and utimes sending attributes to OSTs, setting
1668                  * mtime/atime to the past will be performed under PW [0:EOF]
1669                  * extent lock (new_size:EOF for truncate).  It may seem
1670                  * excessive to send mtime/atime updates to OSTs when not
1671                  * setting times to past, but it is necessary due to possible
1672                  * time de-synchronization between MDT inode and OST objects */
1673                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1674         }
1675
1676         /* If the file was restored, it needs to set dirty flag.
1677          *
1678          * We've already sent MDS_DATA_MODIFIED flag in
1679          * ll_md_setattr() for truncate. However, the MDT refuses to
1680          * set the HS_DIRTY flag on released files, so we have to set
1681          * it again if the file has been restored. Please check how
1682          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1683          *
1684          * Please notice that if the file is not released, the previous
1685          * MDS_DATA_MODIFIED has taken effect and usually
1686          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1687          * This way we can save an RPC for common open + trunc
1688          * operation. */
1689         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1690                 struct hsm_state_set hss = {
1691                         .hss_valid = HSS_SETMASK,
1692                         .hss_setmask = HS_DIRTY,
1693                 };
1694                 int rc2;
1695
1696                 rc2 = ll_hsm_state_set(inode, &hss);
1697                 /* truncate and write can happen at the same time, so that
1698                  * the file can be set modified even though the file is not
1699                  * restored from released state, and ll_hsm_state_set() is
1700                  * not applicable for the file, and rc2 < 0 is normal in this
1701                  * case. */
1702                 if (rc2 < 0)
1703                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1704                                PFID(ll_inode2fid(inode)), rc2);
1705         }
1706
1707         EXIT;
1708 out:
1709         if (op_data != NULL)
1710                 ll_finish_md_op_data(op_data);
1711
1712         if (S_ISREG(inode->i_mode)) {
1713                 mutex_lock(&inode->i_mutex);
1714                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1715                         inode_dio_wait(inode);
1716         }
1717
1718         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1719                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1720
1721         return rc;
1722 }
1723
1724 int ll_setattr(struct dentry *de, struct iattr *attr)
1725 {
1726         int mode = de->d_inode->i_mode;
1727
1728         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1729                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1730                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1731
1732         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1733                                (ATTR_SIZE|ATTR_MODE)) &&
1734             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1735              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1736               !(attr->ia_mode & S_ISGID))))
1737                 attr->ia_valid |= ATTR_FORCE;
1738
1739         if ((attr->ia_valid & ATTR_MODE) &&
1740             (mode & S_ISUID) &&
1741             !(attr->ia_mode & S_ISUID) &&
1742             !(attr->ia_valid & ATTR_KILL_SUID))
1743                 attr->ia_valid |= ATTR_KILL_SUID;
1744
1745         if ((attr->ia_valid & ATTR_MODE) &&
1746             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1747             !(attr->ia_mode & S_ISGID) &&
1748             !(attr->ia_valid & ATTR_KILL_SGID))
1749                 attr->ia_valid |= ATTR_KILL_SGID;
1750
1751         return ll_setattr_raw(de, attr, false);
1752 }
1753
1754 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1755                        __u64 max_age, __u32 flags)
1756 {
1757         struct ll_sb_info *sbi = ll_s2sbi(sb);
1758         struct obd_statfs obd_osfs;
1759         int rc;
1760         ENTRY;
1761
1762         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1763         if (rc) {
1764                 CERROR("md_statfs fails: rc = %d\n", rc);
1765                 RETURN(rc);
1766         }
1767
1768         osfs->os_type = sb->s_magic;
1769
1770         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1771                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1772
1773         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1774                 flags |= OBD_STATFS_NODELAY;
1775
1776         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1777         if (rc) {
1778                 CERROR("obd_statfs fails: rc = %d\n", rc);
1779                 RETURN(rc);
1780         }
1781
1782         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1783                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1784                obd_osfs.os_files);
1785
1786         osfs->os_bsize = obd_osfs.os_bsize;
1787         osfs->os_blocks = obd_osfs.os_blocks;
1788         osfs->os_bfree = obd_osfs.os_bfree;
1789         osfs->os_bavail = obd_osfs.os_bavail;
1790
1791         /* If we don't have as many objects free on the OST as inodes
1792          * on the MDS, we reduce the total number of inodes to
1793          * compensate, so that the "inodes in use" number is correct.
1794          */
1795         if (obd_osfs.os_ffree < osfs->os_ffree) {
1796                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1797                         obd_osfs.os_ffree;
1798                 osfs->os_ffree = obd_osfs.os_ffree;
1799         }
1800
1801         RETURN(rc);
1802 }
1803 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1804 {
1805         struct super_block *sb = de->d_sb;
1806         struct obd_statfs osfs;
1807         __u64 fsid = huge_encode_dev(sb->s_dev);
1808         int rc;
1809
1810         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1811         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1812
1813         /* Some amount of caching on the client is allowed */
1814         rc = ll_statfs_internal(sb, &osfs,
1815                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1816                                 0);
1817         if (rc)
1818                 return rc;
1819
1820         statfs_unpack(sfs, &osfs);
1821
1822         /* We need to downshift for all 32-bit kernels, because we can't
1823          * tell if the kernel is being called via sys_statfs64() or not.
1824          * Stop before overflowing f_bsize - in which case it is better
1825          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1826         if (sizeof(long) < 8) {
1827                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1828                         sfs->f_bsize <<= 1;
1829
1830                         osfs.os_blocks >>= 1;
1831                         osfs.os_bfree >>= 1;
1832                         osfs.os_bavail >>= 1;
1833                 }
1834         }
1835
1836         sfs->f_blocks = osfs.os_blocks;
1837         sfs->f_bfree = osfs.os_bfree;
1838         sfs->f_bavail = osfs.os_bavail;
1839         sfs->f_fsid.val[0] = (__u32)fsid;
1840         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1841         return 0;
1842 }
1843
1844 void ll_inode_size_lock(struct inode *inode)
1845 {
1846         struct ll_inode_info *lli;
1847
1848         LASSERT(!S_ISDIR(inode->i_mode));
1849
1850         lli = ll_i2info(inode);
1851         mutex_lock(&lli->lli_size_mutex);
1852 }
1853
1854 void ll_inode_size_unlock(struct inode *inode)
1855 {
1856         struct ll_inode_info *lli;
1857
1858         lli = ll_i2info(inode);
1859         mutex_unlock(&lli->lli_size_mutex);
1860 }
1861
1862 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1863 {
1864         struct ll_inode_info *lli = ll_i2info(inode);
1865         struct mdt_body *body = md->body;
1866         struct ll_sb_info *sbi = ll_i2sbi(inode);
1867
1868         if (body->mbo_valid & OBD_MD_FLEASIZE)
1869                 cl_file_inode_init(inode, md);
1870
1871         if (S_ISDIR(inode->i_mode)) {
1872                 int     rc;
1873
1874                 rc = ll_update_lsm_md(inode, md);
1875                 if (rc != 0)
1876                         return rc;
1877         }
1878
1879         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1880                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1881                         ll_update_remote_perm(inode, md->remote_perm);
1882         }
1883 #ifdef CONFIG_FS_POSIX_ACL
1884         else if (body->mbo_valid & OBD_MD_FLACL) {
1885                 spin_lock(&lli->lli_lock);
1886                 if (lli->lli_posix_acl)
1887                         posix_acl_release(lli->lli_posix_acl);
1888                 lli->lli_posix_acl = md->posix_acl;
1889                 spin_unlock(&lli->lli_lock);
1890         }
1891 #endif
1892         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1893                                         sbi->ll_flags & LL_SBI_32BIT_API);
1894         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1895
1896         if (body->mbo_valid & OBD_MD_FLATIME) {
1897                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1898                         LTIME_S(inode->i_atime) = body->mbo_atime;
1899                 lli->lli_atime = body->mbo_atime;
1900         }
1901
1902         if (body->mbo_valid & OBD_MD_FLMTIME) {
1903                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1904                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1905                                "to "LPU64"\n", inode->i_ino,
1906                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1907                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1908                 }
1909                 lli->lli_mtime = body->mbo_mtime;
1910         }
1911
1912         if (body->mbo_valid & OBD_MD_FLCTIME) {
1913                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1914                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1915                 lli->lli_ctime = body->mbo_ctime;
1916         }
1917
1918         if (body->mbo_valid & OBD_MD_FLMODE)
1919                 inode->i_mode = (inode->i_mode & S_IFMT) |
1920                                 (body->mbo_mode & ~S_IFMT);
1921
1922         if (body->mbo_valid & OBD_MD_FLTYPE)
1923                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1924                                 (body->mbo_mode & S_IFMT);
1925
1926         LASSERT(inode->i_mode != 0);
1927         if (S_ISREG(inode->i_mode))
1928                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1929                                        LL_MAX_BLKSIZE_BITS);
1930         else
1931                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1932
1933         if (body->mbo_valid & OBD_MD_FLUID)
1934                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1935         if (body->mbo_valid & OBD_MD_FLGID)
1936                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1937         if (body->mbo_valid & OBD_MD_FLFLAGS)
1938                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1939         if (body->mbo_valid & OBD_MD_FLNLINK)
1940                 set_nlink(inode, body->mbo_nlink);
1941         if (body->mbo_valid & OBD_MD_FLRDEV)
1942                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1943
1944         if (body->mbo_valid & OBD_MD_FLID) {
1945                 /* FID shouldn't be changed! */
1946                 if (fid_is_sane(&lli->lli_fid)) {
1947                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1948                                  "Trying to change FID "DFID
1949                                  " to the "DFID", inode "DFID"(%p)\n",
1950                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1951                                  PFID(ll_inode2fid(inode)), inode);
1952                 } else {
1953                         lli->lli_fid = body->mbo_fid1;
1954                 }
1955         }
1956
1957         LASSERT(fid_seq(&lli->lli_fid) != 0);
1958
1959         if (body->mbo_valid & OBD_MD_FLSIZE) {
1960                 i_size_write(inode, body->mbo_size);
1961
1962                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1963                        PFID(ll_inode2fid(inode)),
1964                        (unsigned long long)body->mbo_size);
1965
1966                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1967                         inode->i_blocks = body->mbo_blocks;
1968         }
1969
1970         if (body->mbo_valid & OBD_MD_TSTATE) {
1971                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1972                  * clear it when done to ensure to start again
1973                  * glimpsing updated attrs
1974                  */
1975                 if (body->mbo_t_state & MS_RESTORE)
1976                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1977                 else
1978                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1979         }
1980
1981         return 0;
1982 }
1983
1984 int ll_read_inode2(struct inode *inode, void *opaque)
1985 {
1986         struct lustre_md *md = opaque;
1987         struct ll_inode_info *lli = ll_i2info(inode);
1988         int     rc;
1989         ENTRY;
1990
1991         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1992                PFID(&lli->lli_fid), inode);
1993
1994         /* Core attributes from the MDS first.  This is a new inode, and
1995          * the VFS doesn't zero times in the core inode so we have to do
1996          * it ourselves.  They will be overwritten by either MDS or OST
1997          * attributes - we just need to make sure they aren't newer. */
1998         LTIME_S(inode->i_mtime) = 0;
1999         LTIME_S(inode->i_atime) = 0;
2000         LTIME_S(inode->i_ctime) = 0;
2001         inode->i_rdev = 0;
2002         rc = ll_update_inode(inode, md);
2003         if (rc != 0)
2004                 RETURN(rc);
2005
2006         /* OIDEBUG(inode); */
2007
2008 #ifdef HAVE_BACKING_DEV_INFO
2009         /* initializing backing dev info. */
2010         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2011 #endif
2012         if (S_ISREG(inode->i_mode)) {
2013                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2014                 inode->i_op = &ll_file_inode_operations;
2015                 inode->i_fop = sbi->ll_fop;
2016                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2017                 EXIT;
2018         } else if (S_ISDIR(inode->i_mode)) {
2019                 inode->i_op = &ll_dir_inode_operations;
2020                 inode->i_fop = &ll_dir_operations;
2021                 EXIT;
2022         } else if (S_ISLNK(inode->i_mode)) {
2023                 inode->i_op = &ll_fast_symlink_inode_operations;
2024                 EXIT;
2025         } else {
2026                 inode->i_op = &ll_special_inode_operations;
2027
2028                 init_special_inode(inode, inode->i_mode,
2029                                    inode->i_rdev);
2030
2031                 EXIT;
2032         }
2033
2034         return 0;
2035 }
2036
2037 void ll_delete_inode(struct inode *inode)
2038 {
2039         struct ll_inode_info *lli = ll_i2info(inode);
2040         ENTRY;
2041
2042         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2043                 /* It is last chance to write out dirty pages,
2044                  * otherwise we may lose data while umount */
2045                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2046
2047         truncate_inode_pages_final(&inode->i_data);
2048
2049         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2050                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2051                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2052
2053 #ifdef HAVE_SBOPS_EVICT_INODE
2054         ll_clear_inode(inode);
2055 #endif
2056         clear_inode(inode);
2057
2058         EXIT;
2059 }
2060
2061 int ll_iocontrol(struct inode *inode, struct file *file,
2062                  unsigned int cmd, unsigned long arg)
2063 {
2064         struct ll_sb_info *sbi = ll_i2sbi(inode);
2065         struct ptlrpc_request *req = NULL;
2066         int rc, flags = 0;
2067         ENTRY;
2068
2069         switch(cmd) {
2070         case FSFILT_IOC_GETFLAGS: {
2071                 struct mdt_body *body;
2072                 struct md_op_data *op_data;
2073
2074                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2075                                              0, 0, LUSTRE_OPC_ANY,
2076                                              NULL);
2077                 if (IS_ERR(op_data))
2078                         RETURN(PTR_ERR(op_data));
2079
2080                 op_data->op_valid = OBD_MD_FLFLAGS;
2081                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2082                 ll_finish_md_op_data(op_data);
2083                 if (rc) {
2084                         CERROR("%s: failure inode "DFID": rc = %d\n",
2085                                sbi->ll_md_exp->exp_obd->obd_name,
2086                                PFID(ll_inode2fid(inode)), rc);
2087                         RETURN(-abs(rc));
2088                 }
2089
2090                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2091
2092                 flags = body->mbo_flags;
2093
2094                 ptlrpc_req_finished(req);
2095
2096                 RETURN(put_user(flags, (int __user *)arg));
2097         }
2098         case FSFILT_IOC_SETFLAGS: {
2099                 struct iattr *attr;
2100                 struct md_op_data *op_data;
2101                 struct cl_object *obj;
2102
2103                 if (get_user(flags, (int __user *)arg))
2104                         RETURN(-EFAULT);
2105
2106                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2107                                              LUSTRE_OPC_ANY, NULL);
2108                 if (IS_ERR(op_data))
2109                         RETURN(PTR_ERR(op_data));
2110
2111                 op_data->op_attr_flags = flags;
2112                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2113                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2114                 ll_finish_md_op_data(op_data);
2115                 ptlrpc_req_finished(req);
2116                 if (rc)
2117                         RETURN(rc);
2118
2119                 inode->i_flags = ll_ext_to_inode_flags(flags);
2120
2121                 obj = ll_i2info(inode)->lli_clob;
2122                 if (obj == NULL)
2123                         RETURN(0);
2124
2125                 OBD_ALLOC_PTR(attr);
2126                 if (attr == NULL)
2127                         RETURN(-ENOMEM);
2128
2129                 attr->ia_valid = ATTR_ATTR_FLAG;
2130                 rc = cl_setattr_ost(obj, attr, flags);
2131
2132                 OBD_FREE_PTR(attr);
2133                 RETURN(rc);
2134         }
2135         default:
2136                 RETURN(-ENOSYS);
2137         }
2138
2139         RETURN(0);
2140 }
2141
2142 int ll_flush_ctx(struct inode *inode)
2143 {
2144         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2145
2146         CDEBUG(D_SEC, "flush context for user %d\n",
2147                from_kuid(&init_user_ns, current_uid()));
2148
2149         obd_set_info_async(NULL, sbi->ll_md_exp,
2150                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2151                            0, NULL, NULL);
2152         obd_set_info_async(NULL, sbi->ll_dt_exp,
2153                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2154                            0, NULL, NULL);
2155         return 0;
2156 }
2157
2158 /* umount -f client means force down, don't save state */
2159 void ll_umount_begin(struct super_block *sb)
2160 {
2161         struct ll_sb_info *sbi = ll_s2sbi(sb);
2162         struct obd_device *obd;
2163         struct obd_ioctl_data *ioc_data;
2164         ENTRY;
2165
2166         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2167                sb->s_count, atomic_read(&sb->s_active));
2168
2169         obd = class_exp2obd(sbi->ll_md_exp);
2170         if (obd == NULL) {
2171                 CERROR("Invalid MDC connection handle "LPX64"\n",
2172                        sbi->ll_md_exp->exp_handle.h_cookie);
2173                 EXIT;
2174                 return;
2175         }
2176         obd->obd_force = 1;
2177
2178         obd = class_exp2obd(sbi->ll_dt_exp);
2179         if (obd == NULL) {
2180                 CERROR("Invalid LOV connection handle "LPX64"\n",
2181                        sbi->ll_dt_exp->exp_handle.h_cookie);
2182                 EXIT;
2183                 return;
2184         }
2185         obd->obd_force = 1;
2186
2187         OBD_ALLOC_PTR(ioc_data);
2188         if (ioc_data) {
2189                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2190                               sizeof *ioc_data, ioc_data, NULL);
2191
2192                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2193                               sizeof *ioc_data, ioc_data, NULL);
2194
2195                 OBD_FREE_PTR(ioc_data);
2196         }
2197
2198         /* Really, we'd like to wait until there are no requests outstanding,
2199          * and then continue.  For now, we just invalidate the requests,
2200          * schedule() and sleep one second if needed, and hope.
2201          */
2202         schedule();
2203         EXIT;
2204 }
2205
2206 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2207 {
2208         struct ll_sb_info *sbi = ll_s2sbi(sb);
2209         char *profilenm = get_profile_name(sb);
2210         int err;
2211         __u32 read_only;
2212
2213         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2214                 read_only = *flags & MS_RDONLY;
2215                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2216                                          sizeof(KEY_READ_ONLY),
2217                                          KEY_READ_ONLY, sizeof(read_only),
2218                                          &read_only, NULL);
2219                 if (err) {
2220                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2221                                       profilenm, read_only ?
2222                                       "read-only" : "read-write", err);
2223                         return err;
2224                 }
2225
2226                 if (read_only)
2227                         sb->s_flags |= MS_RDONLY;
2228                 else
2229                         sb->s_flags &= ~MS_RDONLY;
2230
2231                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2232                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2233                                       read_only ?  "read-only" : "read-write");
2234         }
2235         return 0;
2236 }
2237
2238 /**
2239  * Cleanup the open handle that is cached on MDT-side.
2240  *
2241  * For open case, the client side open handling thread may hit error
2242  * after the MDT grant the open. Under such case, the client should
2243  * send close RPC to the MDT as cleanup; otherwise, the open handle
2244  * on the MDT will be leaked there until the client umount or evicted.
2245  *
2246  * In further, if someone unlinked the file, because the open handle
2247  * holds the reference on such file/object, then it will block the
2248  * subsequent threads that want to locate such object via FID.
2249  *
2250  * \param[in] sb        super block for this file-system
2251  * \param[in] open_req  pointer to the original open request
2252  */
2253 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2254 {
2255         struct mdt_body                 *body;
2256         struct md_op_data               *op_data;
2257         struct ptlrpc_request           *close_req = NULL;
2258         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2259         ENTRY;
2260
2261         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2262         OBD_ALLOC_PTR(op_data);
2263         if (op_data == NULL) {
2264                 CWARN("%s: cannot allocate op_data to release open handle for "
2265                       DFID"\n",
2266                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2267
2268                 RETURN_EXIT;
2269         }
2270
2271         op_data->op_fid1 = body->mbo_fid1;
2272         op_data->op_handle = body->mbo_handle;
2273         op_data->op_mod_time = cfs_time_current_sec();
2274         md_close(exp, op_data, NULL, &close_req);
2275         ptlrpc_req_finished(close_req);
2276         ll_finish_md_op_data(op_data);
2277
2278         EXIT;
2279 }
2280
2281 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2282                   struct super_block *sb, struct lookup_intent *it)
2283 {
2284         struct ll_sb_info *sbi = NULL;
2285         struct lustre_md md = { NULL };
2286         int rc;
2287         ENTRY;
2288
2289         LASSERT(*inode || sb);
2290         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2291         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2292                               sbi->ll_md_exp, &md);
2293         if (rc != 0)
2294                 GOTO(cleanup, rc);
2295
2296         if (*inode) {
2297                 rc = ll_update_inode(*inode, &md);
2298                 if (rc != 0)
2299                         GOTO(out, rc);
2300         } else {
2301                 LASSERT(sb != NULL);
2302
2303                 /*
2304                  * At this point server returns to client's same fid as client
2305                  * generated for creating. So using ->fid1 is okay here.
2306                  */
2307                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2308                         CERROR("%s: Fid is insane "DFID"\n",
2309                                 ll_get_fsname(sb, NULL, 0),
2310                                 PFID(&md.body->mbo_fid1));
2311                         GOTO(out, rc = -EINVAL);
2312                 }
2313
2314                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2315                                              sbi->ll_flags & LL_SBI_32BIT_API),
2316                                  &md);
2317                 if (IS_ERR(*inode)) {
2318 #ifdef CONFIG_FS_POSIX_ACL
2319                         if (md.posix_acl) {
2320                                 posix_acl_release(md.posix_acl);
2321                                 md.posix_acl = NULL;
2322                         }
2323 #endif
2324                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2325                         *inode = NULL;
2326                         CERROR("new_inode -fatal: rc %d\n", rc);
2327                         GOTO(out, rc);
2328                 }
2329         }
2330
2331         /* Handling piggyback layout lock.
2332          * Layout lock can be piggybacked by getattr and open request.
2333          * The lsm can be applied to inode only if it comes with a layout lock
2334          * otherwise correct layout may be overwritten, for example:
2335          * 1. proc1: mdt returns a lsm but not granting layout
2336          * 2. layout was changed by another client
2337          * 3. proc2: refresh layout and layout lock granted
2338          * 4. proc1: to apply a stale layout */
2339         if (it != NULL && it->it_lock_mode != 0) {
2340                 struct lustre_handle lockh;
2341                 struct ldlm_lock *lock;
2342
2343                 lockh.cookie = it->it_lock_handle;
2344                 lock = ldlm_handle2lock(&lockh);
2345                 LASSERT(lock != NULL);
2346                 if (ldlm_has_layout(lock)) {
2347                         struct cl_object_conf conf;
2348
2349                         memset(&conf, 0, sizeof(conf));
2350                         conf.coc_opc = OBJECT_CONF_SET;
2351                         conf.coc_inode = *inode;
2352                         conf.coc_lock = lock;
2353                         conf.u.coc_layout = md.layout;
2354                         (void)ll_layout_conf(*inode, &conf);
2355                 }
2356                 LDLM_LOCK_PUT(lock);
2357         }
2358
2359         GOTO(out, rc = 0);
2360
2361 out:
2362         md_free_lustre_md(sbi->ll_md_exp, &md);
2363
2364 cleanup:
2365         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2366                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2367
2368         return rc;
2369 }
2370
2371 int ll_obd_statfs(struct inode *inode, void __user *arg)
2372 {
2373         struct ll_sb_info *sbi = NULL;
2374         struct obd_export *exp;
2375         char *buf = NULL;
2376         struct obd_ioctl_data *data = NULL;
2377         __u32 type;
2378         int len = 0, rc;
2379
2380         if (!inode || !(sbi = ll_i2sbi(inode)))
2381                 GOTO(out_statfs, rc = -EINVAL);
2382
2383         rc = obd_ioctl_getdata(&buf, &len, arg);
2384         if (rc)
2385                 GOTO(out_statfs, rc);
2386
2387         data = (void*)buf;
2388         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2389             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2390                 GOTO(out_statfs, rc = -EINVAL);
2391
2392         if (data->ioc_inllen1 != sizeof(__u32) ||
2393             data->ioc_inllen2 != sizeof(__u32) ||
2394             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2395             data->ioc_plen2 != sizeof(struct obd_uuid))
2396                 GOTO(out_statfs, rc = -EINVAL);
2397
2398         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2399         if (type & LL_STATFS_LMV)
2400                 exp = sbi->ll_md_exp;
2401         else if (type & LL_STATFS_LOV)
2402                 exp = sbi->ll_dt_exp;
2403         else
2404                 GOTO(out_statfs, rc = -ENODEV);
2405
2406         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2407         if (rc)
2408                 GOTO(out_statfs, rc);
2409 out_statfs:
2410         if (buf)
2411                 obd_ioctl_freedata(buf, len);
2412         return rc;
2413 }
2414
2415 int ll_process_config(struct lustre_cfg *lcfg)
2416 {
2417         struct super_block *sb;
2418         unsigned long x;
2419         int rc = 0;
2420         char *ptr;
2421
2422         /* The instance name contains the sb: lustre-client-aacfe000 */
2423         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2424         if (!ptr || !*(++ptr))
2425                 return -EINVAL;
2426         if (sscanf(ptr, "%lx", &x) != 1)
2427                 return -EINVAL;
2428         sb = (struct super_block *)x;
2429         /* This better be a real Lustre superblock! */
2430         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2431
2432         /* Note we have not called client_common_fill_super yet, so
2433            proc fns must be able to handle that! */
2434         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2435                                       lcfg, sb);
2436         if (rc > 0)
2437                 rc = 0;
2438         return rc;
2439 }
2440
2441 /* this function prepares md_op_data hint for passing it down to MD stack. */
2442 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2443                                       struct inode *i1, struct inode *i2,
2444                                       const char *name, size_t namelen,
2445                                       __u32 mode, __u32 opc, void *data)
2446 {
2447         LASSERT(i1 != NULL);
2448
2449         if (name == NULL) {
2450                 /* Do not reuse namelen for something else. */
2451                 if (namelen != 0)
2452                         return ERR_PTR(-EINVAL);
2453         } else {
2454                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2455                         return ERR_PTR(-ENAMETOOLONG);
2456
2457                 if (!lu_name_is_valid_2(name, namelen))
2458                         return ERR_PTR(-EINVAL);
2459         }
2460
2461         if (op_data == NULL)
2462                 OBD_ALLOC_PTR(op_data);
2463
2464         if (op_data == NULL)
2465                 return ERR_PTR(-ENOMEM);
2466
2467         ll_i2gids(op_data->op_suppgids, i1, i2);
2468         op_data->op_fid1 = *ll_inode2fid(i1);
2469         op_data->op_default_stripe_offset = -1;
2470         if (S_ISDIR(i1->i_mode)) {
2471                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2472                 if (opc == LUSTRE_OPC_MKDIR)
2473                         op_data->op_default_stripe_offset =
2474                                    ll_i2info(i1)->lli_def_stripe_offset;
2475         }
2476
2477         if (i2) {
2478                 op_data->op_fid2 = *ll_inode2fid(i2);
2479                 if (S_ISDIR(i2->i_mode))
2480                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2481         } else {
2482                 fid_zero(&op_data->op_fid2);
2483         }
2484
2485         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2486                 op_data->op_cli_flags |= CLI_HASH64;
2487
2488         if (ll_need_32bit_api(ll_i2sbi(i1)))
2489                 op_data->op_cli_flags |= CLI_API32;
2490
2491         op_data->op_name = name;
2492         op_data->op_namelen = namelen;
2493         op_data->op_mode = mode;
2494         op_data->op_mod_time = cfs_time_current_sec();
2495         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2496         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2497         op_data->op_cap = cfs_curproc_cap_pack();
2498         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2499              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2500                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2501         } else {
2502                 op_data->op_mds = 0;
2503         }
2504         op_data->op_data = data;
2505
2506         return op_data;
2507 }
2508
2509 void ll_finish_md_op_data(struct md_op_data *op_data)
2510 {
2511         OBD_FREE_PTR(op_data);
2512 }
2513
2514 #ifdef HAVE_SUPEROPS_USE_DENTRY
2515 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2516 #else
2517 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2518 #endif
2519 {
2520         struct ll_sb_info *sbi;
2521
2522 #ifdef HAVE_SUPEROPS_USE_DENTRY
2523         LASSERT((seq != NULL) && (dentry != NULL));
2524         sbi = ll_s2sbi(dentry->d_sb);
2525 #else
2526         LASSERT((seq != NULL) && (vfs != NULL));
2527         sbi = ll_s2sbi(vfs->mnt_sb);
2528 #endif
2529
2530         if (sbi->ll_flags & LL_SBI_NOLCK)
2531                 seq_puts(seq, ",nolock");
2532
2533         if (sbi->ll_flags & LL_SBI_FLOCK)
2534                 seq_puts(seq, ",flock");
2535
2536         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2537                 seq_puts(seq, ",localflock");
2538
2539         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2540                 seq_puts(seq, ",user_xattr");
2541
2542         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2543                 seq_puts(seq, ",lazystatfs");
2544
2545         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2546                 seq_puts(seq, ",user_fid2path");
2547
2548         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2549                 seq_puts(seq, ",always_ping");
2550
2551         RETURN(0);
2552 }
2553
2554 /**
2555  * Get obd name by cmd, and copy out to user space
2556  */
2557 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2558 {
2559         struct ll_sb_info *sbi = ll_i2sbi(inode);
2560         struct obd_device *obd;
2561         ENTRY;
2562
2563         if (cmd == OBD_IOC_GETDTNAME)
2564                 obd = class_exp2obd(sbi->ll_dt_exp);
2565         else if (cmd == OBD_IOC_GETMDNAME)
2566                 obd = class_exp2obd(sbi->ll_md_exp);
2567         else
2568                 RETURN(-EINVAL);
2569
2570         if (!obd)
2571                 RETURN(-ENOENT);
2572
2573         if (copy_to_user((void __user *)arg, obd->obd_name,
2574                          strlen(obd->obd_name) + 1))
2575                 RETURN(-EFAULT);
2576
2577         RETURN(0);
2578 }
2579
2580 /**
2581  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2582  * fsname will be returned in this buffer; otherwise, a static buffer will be
2583  * used to store the fsname and returned to caller.
2584  */
2585 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2586 {
2587         static char fsname_static[MTI_NAME_MAXLEN];
2588         struct lustre_sb_info *lsi = s2lsi(sb);
2589         char *ptr;
2590         int len;
2591
2592         if (buf == NULL) {
2593                 /* this means the caller wants to use static buffer
2594                  * and it doesn't care about race. Usually this is
2595                  * in error reporting path */
2596                 buf = fsname_static;
2597                 buflen = sizeof(fsname_static);
2598         }
2599
2600         len = strlen(lsi->lsi_lmd->lmd_profile);
2601         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2602         if (ptr && (strcmp(ptr, "-client") == 0))
2603                 len -= 7;
2604
2605         if (unlikely(len >= buflen))
2606                 len = buflen - 1;
2607         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2608         buf[len] = '\0';
2609
2610         return buf;
2611 }
2612
2613 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2614 {
2615         char *path = NULL;
2616
2617         struct path p;
2618
2619         p.dentry = dentry;
2620         p.mnt = current->fs->root.mnt;
2621         path_get(&p);
2622         path = d_path(&p, buf, bufsize);
2623         path_put(&p);
2624         return path;
2625 }
2626
2627 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2628 {
2629         char *buf, *path = NULL;
2630         struct dentry *dentry = NULL;
2631         struct inode *inode = page->mapping->host;
2632
2633         /* this can be called inside spin lock so use GFP_ATOMIC. */
2634         buf = (char *)__get_free_page(GFP_ATOMIC);
2635         if (buf != NULL) {
2636                 dentry = d_find_alias(page->mapping->host);
2637                 if (dentry != NULL)
2638                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2639         }
2640
2641         CDEBUG(D_WARNING,
2642                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2643                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2644                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2645                PFID(ll_inode2fid(inode)),
2646                (path && !IS_ERR(path)) ? path : "", ioret);
2647
2648         if (dentry != NULL)
2649                 dput(dentry);
2650
2651         if (buf != NULL)
2652                 free_page((unsigned long)buf);
2653 }
2654
2655 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2656                         struct lov_user_md **kbuf)
2657 {
2658         struct lov_user_md      lum;
2659         ssize_t                 lum_size;
2660         ENTRY;
2661
2662         if (copy_from_user(&lum, md, sizeof(lum)))
2663                 RETURN(-EFAULT);
2664
2665         lum_size = ll_lov_user_md_size(&lum);
2666         if (lum_size < 0)
2667                 RETURN(lum_size);
2668
2669         OBD_ALLOC(*kbuf, lum_size);
2670         if (*kbuf == NULL)
2671                 RETURN(-ENOMEM);
2672
2673         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2674                 OBD_FREE(*kbuf, lum_size);
2675                 RETURN(-EFAULT);
2676         }
2677
2678         RETURN(lum_size);
2679 }
2680
2681 /*
2682  * Compute llite root squash state after a change of root squash
2683  * configuration setting or add/remove of a lnet nid
2684  */
2685 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2686 {
2687         struct root_squash_info *squash = &sbi->ll_squash;
2688         int i;
2689         bool matched;
2690         lnet_process_id_t id;
2691
2692         /* Update norootsquash flag */
2693         down_write(&squash->rsi_sem);
2694         if (list_empty(&squash->rsi_nosquash_nids))
2695                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2696         else {
2697                 /* Do not apply root squash as soon as one of our NIDs is
2698                  * in the nosquash_nids list */
2699                 matched = false;
2700                 i = 0;
2701                 while (LNetGetId(i++, &id) != -ENOENT) {
2702                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2703                                 continue;
2704                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2705                                 matched = true;
2706                                 break;
2707                         }
2708                 }
2709                 if (matched)
2710                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2711                 else
2712                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2713         }
2714         up_write(&squash->rsi_sem);
2715 }
2716
2717 /**
2718  * Parse linkea content to extract information about a given hardlink
2719  *
2720  * \param[in]   ldata      - Initialized linkea data
2721  * \param[in]   linkno     - Link identifier
2722  * \param[out]  parent_fid - The entry's parent FID
2723  * \param[out]  ln         - Entry name destination buffer
2724  *
2725  * \retval 0 on success
2726  * \retval Appropriate negative error code on failure
2727  */
2728 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2729                             struct lu_fid *parent_fid, struct lu_name *ln)
2730 {
2731         unsigned int    idx;
2732         int             rc;
2733         ENTRY;
2734
2735         rc = linkea_init(ldata);
2736         if (rc < 0)
2737                 RETURN(rc);
2738
2739         if (linkno >= ldata->ld_leh->leh_reccount)
2740                 /* beyond last link */
2741                 RETURN(-ENODATA);
2742
2743         linkea_first_entry(ldata);
2744         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2745                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2746                                     parent_fid);
2747                 if (idx == linkno)
2748                         break;
2749
2750                 linkea_next_entry(ldata);
2751         }
2752
2753         if (idx < linkno)
2754                 RETURN(-ENODATA);
2755
2756         RETURN(0);
2757 }
2758
2759 /**
2760  * Get parent FID and name of an identified link. Operation is performed for
2761  * a given link number, letting the caller iterate over linkno to list one or
2762  * all links of an entry.
2763  *
2764  * \param[in]     file - File descriptor against which to perform the operation
2765  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2766  *                       on and the available size. It is eventually filled with
2767  *                       the requested information or left untouched on error
2768  *
2769  * \retval - 0 on success
2770  * \retval - Appropriate negative error code on failure
2771  */
2772 int ll_getparent(struct file *file, struct getparent __user *arg)
2773 {
2774         struct dentry           *dentry = file->f_path.dentry;
2775         struct inode            *inode = dentry->d_inode;
2776         struct linkea_data      *ldata;
2777         struct lu_buf            buf = LU_BUF_NULL;
2778         struct lu_name           ln;
2779         struct lu_fid            parent_fid;
2780         __u32                    linkno;
2781         __u32                    name_size;
2782         int                      rc;
2783
2784         ENTRY;
2785
2786         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2787             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2788                 RETURN(-EPERM);
2789
2790         if (get_user(name_size, &arg->gp_name_size))
2791                 RETURN(-EFAULT);
2792
2793         if (get_user(linkno, &arg->gp_linkno))
2794                 RETURN(-EFAULT);
2795
2796         if (name_size > PATH_MAX)
2797                 RETURN(-EINVAL);
2798
2799         OBD_ALLOC(ldata, sizeof(*ldata));
2800         if (ldata == NULL)
2801                 RETURN(-ENOMEM);
2802
2803         rc = linkea_data_new(ldata, &buf);
2804         if (rc < 0)
2805                 GOTO(ldata_free, rc);
2806
2807         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2808         if (rc < 0)
2809                 GOTO(lb_free, rc);
2810
2811         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2812         if (rc < 0)
2813                 GOTO(lb_free, rc);
2814
2815         if (ln.ln_namelen >= name_size)
2816                 GOTO(lb_free, rc = -EOVERFLOW);
2817
2818         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2819                 GOTO(lb_free, rc = -EFAULT);
2820
2821         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2822                 GOTO(lb_free, rc = -EFAULT);
2823
2824         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2825                 GOTO(lb_free, rc = -EFAULT);
2826
2827 lb_free:
2828         lu_buf_free(&buf);
2829 ldata_free:
2830         OBD_FREE(ldata, sizeof(*ldata));
2831
2832         RETURN(rc);
2833 }