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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <uapi/linux/lustre_ioctl.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <uapi/linux/lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 struct kmem_cache *ll_file_data_slab;
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC_PTR(sbi);
  79         if (sbi == NULL)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         sbi->ll_cache = cl_cache_init(lru_page_max);
  94         if (sbi->ll_cache == NULL) {
  95                 OBD_FREE(sbi, sizeof(*sbi));
  96                 RETURN(NULL);
  97         }
  98
  99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 100                                            SBI_DEFAULT_READAHEAD_MAX);
 101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 103
 104         ll_generate_random_uuid(uuid);
 105         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 106         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 107
 108         sbi->ll_flags |= LL_SBI_VERBOSE;
 109 #ifdef ENABLE_CHECKSUM
 110         sbi->ll_flags |= LL_SBI_CHECKSUM;
 111 #endif
 112
 113 #ifdef HAVE_LRU_RESIZE_SUPPORT
 114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 115 #endif
 116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 117
 118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 120                                pp_r_hist.oh_lock);
 121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 122                                pp_w_hist.oh_lock);
 123         }
 124
 125         /* metadata statahead is enabled by default */
 126         sbi->ll_sa_max = LL_SA_RPC_DEF;
 127         atomic_set(&sbi->ll_sa_total, 0);
 128         atomic_set(&sbi->ll_sa_wrong, 0);
 129         atomic_set(&sbi->ll_sa_running, 0);
 130         atomic_set(&sbi->ll_agl_total, 0);
 131         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 132         sbi->ll_flags |= LL_SBI_FAST_READ;
 133
 134         /* root squash */
 135         sbi->ll_squash.rsi_uid = 0;
 136         sbi->ll_squash.rsi_gid = 0;
 137         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 138         init_rwsem(&sbi->ll_squash.rsi_sem);
 139
 140         RETURN(sbi);
 141 }
 142
 143 static void ll_free_sbi(struct super_block *sb)
 144 {
 145         struct ll_sb_info *sbi = ll_s2sbi(sb);
 146         ENTRY;
 147
 148         if (sbi != NULL) {
 149                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 150                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 151                 if (sbi->ll_cache != NULL) {
 152                         cl_cache_decref(sbi->ll_cache);
 153                         sbi->ll_cache = NULL;
 154                 }
 155                 OBD_FREE(sbi, sizeof(*sbi));
 156         }
 157         EXIT;
 158 }
 159
 160 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 161 {
 162         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 163                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 164 }
 165
 166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 167                                     struct vfsmount *mnt)
 168 {
 169         struct inode *root = NULL;
 170         struct ll_sb_info *sbi = ll_s2sbi(sb);
 171         struct obd_device *obd;
 172         struct obd_statfs *osfs = NULL;
 173         struct ptlrpc_request *request = NULL;
 174         struct obd_connect_data *data = NULL;
 175         struct obd_uuid *uuid;
 176         struct md_op_data *op_data;
 177         struct lustre_md lmd;
 178         u64 valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         OBD_ALLOC_PTR(osfs);
 193         if (osfs == NULL) {
 194                 OBD_FREE_PTR(data);
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         /* indicate the features supported by this client */
 199         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 200                                   OBD_CONNECT_ATTRFID  |
 201                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 202                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 203                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 204                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 205                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 206                                   OBD_CONNECT_64BITHASH |
 207                                   OBD_CONNECT_EINPROGRESS |
 208                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 209                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 210                                   OBD_CONNECT_MAX_EASIZE |
 211                                   OBD_CONNECT_FLOCK_DEAD |
 212                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 213                                   OBD_CONNECT_OPEN_BY_FID |
 214                                   OBD_CONNECT_DIR_STRIPE |
 215                                   OBD_CONNECT_BULK_MBITS |
 216                                   OBD_CONNECT_SUBTREE |
 217                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_MULTIMODRPCS;
 218
 219         data->ocd_connect_flags2 = 0;
 220
 221 #ifdef HAVE_LRU_RESIZE_SUPPORT
 222         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 223                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 224 #endif
 225 #ifdef CONFIG_FS_POSIX_ACL
 226         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 227                                    OBD_CONNECT_LARGE_ACL;
 228 #endif
 229
 230         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 231                 /* flag mdc connection as lightweight, only used for test
 232                  * purpose, use with care */
 233                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 234
 235         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 236         data->ocd_version = LUSTRE_VERSION_CODE;
 237
 238         if (sb->s_flags & MS_RDONLY)
 239                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 240         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 241                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 242
 243 #ifdef MS_NOSEC
 244         /* Setting this indicates we correctly support S_NOSEC (See kernel
 245          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 246          */
 247         sb->s_flags |= MS_NOSEC;
 248 #endif
 249
 250         if (sbi->ll_flags & LL_SBI_FLOCK)
 251                 sbi->ll_fop = &ll_file_operations_flock;
 252         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 253                 sbi->ll_fop = &ll_file_operations;
 254         else
 255                 sbi->ll_fop = &ll_file_operations_noflock;
 256
 257         /* always ping even if server suppress_pings */
 258         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 259                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 260
 261 #ifdef HAVE_SECURITY_DENTRY_INIT_SECURITY
 262         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 263 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337         sbi->ll_mnt.mnt = current->fs->root.mnt;
 338
 339         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 340             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 341                 LCONSOLE_INFO("Disabling user_xattr feature because "
 342                               "it is not supported on the server\n");
 343                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags |= MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags |= LL_SBI_ACL;
 351         } else {
 352                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 353 #ifdef MS_POSIXACL
 354                 sb->s_flags &= ~MS_POSIXACL;
 355 #endif
 356                 sbi->ll_flags &= ~LL_SBI_ACL;
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 360                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 361
 362         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 363                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 364
 365         if (obd_connect_has_secctx(data))
 366                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 367
 368         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 369                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 370                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 371                                       "unknown maximum xattr size.\n", dt);
 372                 } else if (!sbi->ll_xattr_cache_set) {
 373                         /* If xattr_cache is already set (no matter 0 or 1)
 374                          * during processing llog, it won't be enabled here. */
 375                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 376                         sbi->ll_xattr_cache_enabled = 1;
 377                 }
 378         }
 379
 380         obd = class_name2obd(dt);
 381         if (!obd) {
 382                 CERROR("DT %s: not setup or attached\n", dt);
 383                 GOTO(out_md_fid, err = -ENODEV);
 384         }
 385
 386         /* pass client page size via ocd_grant_blkbits, the server should report
 387          * back its backend blocksize for grant calculation purpose */
 388         data->ocd_grant_blkbits = PAGE_SHIFT;
 389
 390         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 391                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 392                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 393                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 394                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 395                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 396                                   OBD_CONNECT_64BITHASH | 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         data->ocd_connect_flags2 = 0;
 404
 405         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 406                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 407
 408         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 409          * disabled by default, because it can still be enabled on the
 410          * fly via /proc. As a consequence, we still need to come to an
 411          * agreement on the supported algorithms at connect time */
 412         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 413
 414         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 415                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 416         else
 417                 data->ocd_cksum_types = cksum_types_supported_client();
 418
 419 #ifdef HAVE_LRU_RESIZE_SUPPORT
 420         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 421 #endif
 422         /* always ping even if server suppress_pings */
 423         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 424                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 425
 426         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 427                "ocd_grant: %d\n", data->ocd_connect_flags,
 428                data->ocd_version, data->ocd_grant);
 429
 430         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 431         obd->obd_upcall.onu_upcall = cl_ocd_update;
 432
 433         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 434
 435         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 436                           NULL);
 437         if (err == -EBUSY) {
 438                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 439                                    "recovery, of which this client is not a "
 440                                    "part.  Please wait for recovery to "
 441                                    "complete, abort, or time out.\n", dt);
 442                 GOTO(out_md, err);
 443         } else if (err) {
 444                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 445                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 446                 GOTO(out_md, err);
 447         }
 448
 449         sbi->ll_dt_exp->exp_connect_data = *data;
 450
 451         /* Don't change value if it was specified in the config log */
 452         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 453                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 454                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 455                               (data->ocd_brw_size >> PAGE_SHIFT));
 456
 457         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 458                            LUSTRE_SEQ_METADATA);
 459         if (err) {
 460                 CERROR("%s: Can't init data layer FID infrastructure, "
 461                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 462                 GOTO(out_dt, err);
 463         }
 464
 465         mutex_lock(&sbi->ll_lco.lco_lock);
 466         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 467         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 468         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 469         mutex_unlock(&sbi->ll_lco.lco_lock);
 470
 471         fid_zero(&sbi->ll_root_fid);
 472         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 473                            &sbi->ll_root_fid);
 474         if (err) {
 475                 CERROR("cannot mds_connect: rc = %d\n", err);
 476                 GOTO(out_lock_cn_cb, err);
 477         }
 478         if (!fid_is_sane(&sbi->ll_root_fid)) {
 479                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 480                        sbi->ll_md_exp->exp_obd->obd_name,
 481                        PFID(&sbi->ll_root_fid));
 482                 GOTO(out_lock_cn_cb, err = -EINVAL);
 483         }
 484         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 485
 486         sb->s_op = &lustre_super_operations;
 487 #ifdef HAVE_XATTR_HANDLER_FLAGS
 488         sb->s_xattr = ll_xattr_handlers;
 489 #endif
 490 #if THREAD_SIZE >= 8192 /*b=17630*/
 491         sb->s_export_op = &lustre_export_operations;
 492 #endif
 493
 494         /* make root inode
 495          * XXX: move this to after cbd setup? */
 496         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 497         if (sbi->ll_flags & LL_SBI_ACL)
 498                 valid |= OBD_MD_FLACL;
 499
 500         OBD_ALLOC_PTR(op_data);
 501         if (op_data == NULL)
 502                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 503
 504         op_data->op_fid1 = sbi->ll_root_fid;
 505         op_data->op_mode = 0;
 506         op_data->op_valid = valid;
 507
 508         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 509
 510         OBD_FREE_PTR(op_data);
 511         if (err) {
 512                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 513                        sbi->ll_md_exp->exp_obd->obd_name, err);
 514                 GOTO(out_lock_cn_cb, err);
 515         }
 516
 517         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 518                                sbi->ll_md_exp, &lmd);
 519         if (err) {
 520                 CERROR("failed to understand root inode md: rc = %d\n", err);
 521                 ptlrpc_req_finished(request);
 522                 GOTO(out_lock_cn_cb, err);
 523         }
 524
 525         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 526         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 527                                             sbi->ll_flags & LL_SBI_32BIT_API),
 528                        &lmd);
 529         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 530         ptlrpc_req_finished(request);
 531
 532         if (IS_ERR(root)) {
 533 #ifdef CONFIG_FS_POSIX_ACL
 534                 if (lmd.posix_acl) {
 535                         posix_acl_release(lmd.posix_acl);
 536                         lmd.posix_acl = NULL;
 537                 }
 538 #endif
 539                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 540                 root = NULL;
 541                 CERROR("lustre_lite: bad iget4 for root\n");
 542                 GOTO(out_root, err);
 543         }
 544
 545         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 546         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 547                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 548                                  NULL);
 549         if (err) {
 550                 CERROR("%s: Set checksum failed: rc = %d\n",
 551                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 552                 GOTO(out_root, err);
 553         }
 554         cl_sb_init(sb);
 555
 556         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 557                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 558                                  sbi->ll_cache, NULL);
 559         if (err) {
 560                 CERROR("%s: Set cache_set failed: rc = %d\n",
 561                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 562                 GOTO(out_root, err);
 563         }
 564
 565         sb->s_root = d_make_root(root);
 566         if (sb->s_root == NULL) {
 567                 CERROR("%s: can't make root dentry\n",
 568                         ll_get_fsname(sb, NULL, 0));
 569                 GOTO(out_root, err = -ENOMEM);
 570         }
 571 #ifdef HAVE_DCACHE_LOCK
 572         sb->s_root->d_op = &ll_d_ops;
 573 #endif
 574
 575         sbi->ll_sdev_orig = sb->s_dev;
 576
 577         /* We set sb->s_dev equal on all lustre clients in order to support
 578          * NFS export clustering.  NFSD requires that the FSID be the same
 579          * on all clients. */
 580         /* s_dev is also used in lt_compare() to compare two fs, but that is
 581          * only a node-local comparison. */
 582         uuid = obd_get_uuid(sbi->ll_md_exp);
 583         if (uuid != NULL)
 584                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 585
 586         if (data != NULL)
 587                 OBD_FREE_PTR(data);
 588         if (osfs != NULL)
 589                 OBD_FREE_PTR(osfs);
 590
 591         if (sbi->ll_proc_root != NULL) {
 592                 err = lprocfs_ll_register_obd(sb, dt);
 593                 if (err < 0) {
 594                         CERROR("%s: could not register %s in llite: rc = %d\n",
 595                                dt, ll_get_fsname(sb, NULL, 0), err);
 596                         err = 0;
 597                 }
 598                 err = lprocfs_ll_register_obd(sb, md);
 599                 if (err < 0) {
 600                         CERROR("%s: could not register %s in llite: rc = %d\n",
 601                                md, ll_get_fsname(sb, NULL, 0), err);
 602                         err = 0;
 603                 }
 604         }
 605
 606         RETURN(err);
 607 out_root:
 608         if (root)
 609                 iput(root);
 610 out_lock_cn_cb:
 611         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 612 out_dt:
 613         obd_disconnect(sbi->ll_dt_exp);
 614         sbi->ll_dt_exp = NULL;
 615 out_md_fid:
 616         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 617 out_md:
 618         obd_disconnect(sbi->ll_md_exp);
 619         sbi->ll_md_exp = NULL;
 620 out:
 621         if (data != NULL)
 622                 OBD_FREE_PTR(data);
 623         if (osfs != NULL)
 624                 OBD_FREE_PTR(osfs);
 625         return err;
 626 }
 627
 628 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 629 {
 630         int size, rc;
 631
 632         size = sizeof(*lmmsize);
 633         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 634                           KEY_MAX_EASIZE, &size, lmmsize);
 635         if (rc != 0) {
 636                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 637                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 638                 RETURN(rc);
 639         }
 640
 641         size = sizeof(int);
 642         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 643                           KEY_MAX_EASIZE, &size, lmmsize);
 644         if (rc)
 645                 CERROR("Get max mdsize error rc %d\n", rc);
 646
 647         RETURN(rc);
 648 }
 649
 650 /**
 651  * Get the value of the default_easize parameter.
 652  *
 653  * \see client_obd::cl_default_mds_easize
 654  *
 655  * \param[in] sbi       superblock info for this filesystem
 656  * \param[out] lmmsize  pointer to storage location for value
 657  *
 658  * \retval 0            on success
 659  * \retval negative     negated errno on failure
 660  */
 661 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 662 {
 663         int size, rc;
 664
 665         size = sizeof(int);
 666         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 667                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 668         if (rc)
 669                 CERROR("Get default mdsize error rc %d\n", rc);
 670
 671         RETURN(rc);
 672 }
 673
 674 /**
 675  * Set the default_easize parameter to the given value.
 676  *
 677  * \see client_obd::cl_default_mds_easize
 678  *
 679  * \param[in] sbi       superblock info for this filesystem
 680  * \param[in] lmmsize   the size to set
 681  *
 682  * \retval 0            on success
 683  * \retval negative     negated errno on failure
 684  */
 685 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 686 {
 687         int rc;
 688
 689         if (lmmsize < sizeof(struct lov_mds_md) ||
 690             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 691                 return -EINVAL;
 692
 693         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 694                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 695                                 sizeof(int), &lmmsize, NULL);
 696
 697         RETURN(rc);
 698 }
 699
 700 static void client_common_put_super(struct super_block *sb)
 701 {
 702         struct ll_sb_info *sbi = ll_s2sbi(sb);
 703         ENTRY;
 704
 705         cl_sb_fini(sb);
 706
 707         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 708         obd_disconnect(sbi->ll_dt_exp);
 709         sbi->ll_dt_exp = NULL;
 710
 711         lprocfs_ll_unregister_mountpoint(sbi);
 712
 713         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 714         obd_disconnect(sbi->ll_md_exp);
 715         sbi->ll_md_exp = NULL;
 716
 717         EXIT;
 718 }
 719
 720 void ll_kill_super(struct super_block *sb)
 721 {
 722         struct ll_sb_info *sbi;
 723         ENTRY;
 724
 725         /* not init sb ?*/
 726         if (!(sb->s_flags & MS_ACTIVE))
 727                 return;
 728
 729         sbi = ll_s2sbi(sb);
 730         /* we need restore s_dev from changed for clustred NFS before put_super
 731          * because new kernels have cached s_dev and change sb->s_dev in
 732          * put_super not affected real removing devices */
 733         if (sbi) {
 734                 sb->s_dev = sbi->ll_sdev_orig;
 735                 sbi->ll_umounting = 1;
 736
 737                 /* wait running statahead threads to quit */
 738                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 739                         set_current_state(TASK_UNINTERRUPTIBLE);
 740                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 741                 }
 742         }
 743
 744         EXIT;
 745 }
 746
 747 static inline int ll_set_opt(const char *opt, char *data, int fl)
 748 {
 749         if (strncmp(opt, data, strlen(opt)) != 0)
 750                 return(0);
 751         else
 752                 return(fl);
 753 }
 754
 755 /* non-client-specific mount options are parsed in lmd_parse */
 756 static int ll_options(char *options, int *flags)
 757 {
 758         int tmp;
 759         char *s1 = options, *s2;
 760         ENTRY;
 761
 762         if (!options)
 763                 RETURN(0);
 764
 765         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 766
 767         while (*s1) {
 768                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 769                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 770                 if (tmp) {
 771                         *flags |= tmp;
 772                         goto next;
 773                 }
 774                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 775                 if (tmp) {
 776                         *flags |= tmp;
 777                         goto next;
 778                 }
 779                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 780                 if (tmp) {
 781                         *flags |= tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 785                 if (tmp) {
 786                         *flags &= ~tmp;
 787                         goto next;
 788                 }
 789                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 795                 if (tmp) {
 796                         *flags &= ~tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("context", s1, 1);
 800                 if (tmp)
 801                         goto next;
 802                 tmp = ll_set_opt("fscontext", s1, 1);
 803                 if (tmp)
 804                         goto next;
 805                 tmp = ll_set_opt("defcontext", s1, 1);
 806                 if (tmp)
 807                         goto next;
 808                 tmp = ll_set_opt("rootcontext", s1, 1);
 809                 if (tmp)
 810                         goto next;
 811                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821
 822                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 828                 if (tmp) {
 829                         *flags &= ~tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 838                 if (tmp) {
 839                         *flags &= ~tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 873                                    s1);
 874                 RETURN(-EINVAL);
 875
 876 next:
 877                 /* Find next opt */
 878                 s2 = strchr(s1, ',');
 879                 if (s2 == NULL)
 880                         break;
 881                 s1 = s2 + 1;
 882         }
 883         RETURN(0);
 884 }
 885
 886 void ll_lli_init(struct ll_inode_info *lli)
 887 {
 888         lli->lli_inode_magic = LLI_INODE_MAGIC;
 889         lli->lli_flags = 0;
 890         spin_lock_init(&lli->lli_lock);
 891         lli->lli_posix_acl = NULL;
 892         /* Do not set lli_fid, it has been initialized already. */
 893         fid_zero(&lli->lli_pfid);
 894         lli->lli_mds_read_och = NULL;
 895         lli->lli_mds_write_och = NULL;
 896         lli->lli_mds_exec_och = NULL;
 897         lli->lli_open_fd_read_count = 0;
 898         lli->lli_open_fd_write_count = 0;
 899         lli->lli_open_fd_exec_count = 0;
 900         mutex_init(&lli->lli_och_mutex);
 901         spin_lock_init(&lli->lli_agl_lock);
 902         spin_lock_init(&lli->lli_layout_lock);
 903         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 904         lli->lli_clob = NULL;
 905
 906         init_rwsem(&lli->lli_xattrs_list_rwsem);
 907         mutex_init(&lli->lli_xattrs_enq_lock);
 908
 909         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 910         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 911                 mutex_init(&lli->lli_readdir_mutex);
 912                 lli->lli_opendir_key = NULL;
 913                 lli->lli_sai = NULL;
 914                 spin_lock_init(&lli->lli_sa_lock);
 915                 lli->lli_opendir_pid = 0;
 916                 lli->lli_sa_enabled = 0;
 917                 lli->lli_def_stripe_offset = -1;
 918         } else {
 919                 mutex_init(&lli->lli_size_mutex);
 920                 lli->lli_symlink_name = NULL;
 921                 init_rwsem(&lli->lli_trunc_sem);
 922                 range_lock_tree_init(&lli->lli_write_tree);
 923                 init_rwsem(&lli->lli_glimpse_sem);
 924                 lli->lli_glimpse_time = 0;
 925                 INIT_LIST_HEAD(&lli->lli_agl_list);
 926                 lli->lli_agl_index = 0;
 927                 lli->lli_async_rc = 0;
 928         }
 929         mutex_init(&lli->lli_layout_mutex);
 930         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 931 }
 932
 933 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 934 {
 935         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 936
 937         bdi->name = "lustre";
 938         return bdi_register(bdi, NULL, "lustre-%d",
 939                             atomic_inc_return(&ll_bdi_num));
 940 }
 941
 942 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 943 {
 944         struct  lustre_profile *lprof = NULL;
 945         struct  lustre_sb_info *lsi = s2lsi(sb);
 946         struct  ll_sb_info *sbi;
 947         char    *dt = NULL, *md = NULL;
 948         char    *profilenm = get_profile_name(sb);
 949         struct config_llog_instance *cfg;
 950         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 951         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 952         int     md_len = 0;
 953         int     dt_len = 0;
 954         int     err;
 955         ENTRY;
 956
 957         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 958
 959         OBD_ALLOC_PTR(cfg);
 960         if (cfg == NULL)
 961                 RETURN(-ENOMEM);
 962
 963         try_module_get(THIS_MODULE);
 964
 965         /* client additional sb info */
 966         lsi->lsi_llsbi = sbi = ll_init_sbi();
 967         if (!sbi) {
 968                 module_put(THIS_MODULE);
 969                 OBD_FREE_PTR(cfg);
 970                 RETURN(-ENOMEM);
 971         }
 972
 973         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 974         if (err)
 975                 GOTO(out_free, err);
 976
 977         err = bdi_init(&lsi->lsi_bdi);
 978         if (err)
 979                 GOTO(out_free, err);
 980         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 981 #ifdef HAVE_BDI_CAP_MAP_COPY
 982         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 983 #else
 984         lsi->lsi_bdi.capabilities = 0;
 985 #endif
 986         err = ll_bdi_register(&lsi->lsi_bdi);
 987         if (err)
 988                 GOTO(out_free, err);
 989
 990         sb->s_bdi = &lsi->lsi_bdi;
 991 #ifndef HAVE_DCACHE_LOCK
 992         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 993         sb->s_d_op = &ll_d_ops;
 994 #endif
 995
 996         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
 997          * so that "llite.*.*" params can be processed correctly. */
 998         if (proc_lustre_fs_root != NULL) {
 999                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1000                 if (err < 0) {
1001                         CERROR("%s: could not register mountpoint in llite: "
1002                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1003                         err = 0;
1004                 }
1005         }
1006
1007         /* Generate a string unique to this super, in case some joker tries
1008            to mount the same fs at two mount points.
1009            Use the address of the super itself.*/
1010         cfg->cfg_instance = sb;
1011         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1012         cfg->cfg_callback = class_config_llog_handler;
1013         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1014         /* set up client obds */
1015         err = lustre_process_log(sb, profilenm, cfg);
1016         if (err < 0)
1017                 GOTO(out_proc, err);
1018
1019         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1020         lprof = class_get_profile(profilenm);
1021         if (lprof == NULL) {
1022                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1023                                    " read from the MGS.  Does that filesystem "
1024                                    "exist?\n", profilenm);
1025                 GOTO(out_proc, err = -EINVAL);
1026         }
1027         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1028                lprof->lp_md, lprof->lp_dt);
1029
1030         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1031         OBD_ALLOC(dt, dt_len);
1032         if (!dt)
1033                 GOTO(out_proc, err = -ENOMEM);
1034         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1035
1036         md_len = strlen(lprof->lp_md) + instlen + 2;
1037         OBD_ALLOC(md, md_len);
1038         if (!md)
1039                 GOTO(out_proc, err = -ENOMEM);
1040         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1041
1042         /* connections, registrations, sb setup */
1043         err = client_common_fill_super(sb, md, dt, mnt);
1044         if (err < 0)
1045                 GOTO(out_proc, err);
1046
1047         sbi->ll_client_common_fill_super_succeeded = 1;
1048
1049 out_proc:
1050         if (err < 0)
1051                 lprocfs_ll_unregister_mountpoint(sbi);
1052 out_free:
1053         if (md)
1054                 OBD_FREE(md, md_len);
1055         if (dt)
1056                 OBD_FREE(dt, dt_len);
1057         if (lprof != NULL)
1058                 class_put_profile(lprof);
1059         if (err)
1060                 ll_put_super(sb);
1061         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1062                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1063
1064         OBD_FREE_PTR(cfg);
1065         RETURN(err);
1066 } /* ll_fill_super */
1067
1068 void ll_put_super(struct super_block *sb)
1069 {
1070         struct config_llog_instance cfg, params_cfg;
1071         struct obd_device *obd;
1072         struct lustre_sb_info *lsi = s2lsi(sb);
1073         struct ll_sb_info *sbi = ll_s2sbi(sb);
1074         char *profilenm = get_profile_name(sb);
1075         long ccc_count;
1076         int next, force = 1, rc = 0;
1077         ENTRY;
1078
1079         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1080
1081         cfg.cfg_instance = sb;
1082         lustre_end_log(sb, profilenm, &cfg);
1083
1084         params_cfg.cfg_instance = sb;
1085         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1086
1087         if (sbi->ll_md_exp) {
1088                 obd = class_exp2obd(sbi->ll_md_exp);
1089                 if (obd)
1090                         force = obd->obd_force;
1091         }
1092
1093         /* Wait for unstable pages to be committed to stable storage */
1094         if (force == 0) {
1095                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1096                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1097                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1098                         &lwi);
1099         }
1100
1101         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1102         if (force == 0 && rc != -EINTR)
1103                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1104
1105
1106         /* We need to set force before the lov_disconnect in
1107            lustre_common_put_super, since l_d cleans up osc's as well. */
1108         if (force) {
1109                 next = 0;
1110                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1111                                                      &next)) != NULL) {
1112                         obd->obd_force = force;
1113                 }
1114         }
1115
1116         if (sbi->ll_client_common_fill_super_succeeded) {
1117                 /* Only if client_common_fill_super succeeded */
1118                 client_common_put_super(sb);
1119         }
1120
1121         next = 0;
1122         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1123                 class_manual_cleanup(obd);
1124         }
1125
1126         if (sbi->ll_flags & LL_SBI_VERBOSE)
1127                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1128
1129         if (profilenm)
1130                 class_del_profile(profilenm);
1131
1132         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1133                 bdi_destroy(&lsi->lsi_bdi);
1134                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1135         }
1136
1137         ll_free_sbi(sb);
1138         lsi->lsi_llsbi = NULL;
1139
1140         lustre_common_put_super(sb);
1141
1142         cl_env_cache_purge(~0);
1143
1144         module_put(THIS_MODULE);
1145
1146         EXIT;
1147 } /* client_put_super */
1148
1149 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1150 {
1151         struct inode *inode = NULL;
1152
1153         /* NOTE: we depend on atomic igrab() -bzzz */
1154         lock_res_and_lock(lock);
1155         if (lock->l_resource->lr_lvb_inode) {
1156                 struct ll_inode_info * lli;
1157                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1158                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1159                         inode = igrab(lock->l_resource->lr_lvb_inode);
1160                 } else {
1161                         inode = lock->l_resource->lr_lvb_inode;
1162                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1163                                          D_WARNING, lock, "lr_lvb_inode %p is "
1164                                          "bogus: magic %08x",
1165                                          lock->l_resource->lr_lvb_inode,
1166                                          lli->lli_inode_magic);
1167                         inode = NULL;
1168                 }
1169         }
1170         unlock_res_and_lock(lock);
1171         return inode;
1172 }
1173
1174 void ll_dir_clear_lsm_md(struct inode *inode)
1175 {
1176         struct ll_inode_info *lli = ll_i2info(inode);
1177
1178         LASSERT(S_ISDIR(inode->i_mode));
1179
1180         if (lli->lli_lsm_md != NULL) {
1181                 lmv_free_memmd(lli->lli_lsm_md);
1182                 lli->lli_lsm_md = NULL;
1183         }
1184 }
1185
1186 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1187                                       const struct lu_fid *fid,
1188                                       struct lustre_md *md)
1189 {
1190         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1191         struct mdt_body         *body = md->body;
1192         struct inode            *inode;
1193         ino_t                   ino;
1194         ENTRY;
1195
1196         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1197         inode = iget_locked(sb, ino);
1198         if (inode == NULL) {
1199                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1200                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1201                 RETURN(ERR_PTR(-ENOENT));
1202         }
1203
1204         if (inode->i_state & I_NEW) {
1205                 struct ll_inode_info *lli = ll_i2info(inode);
1206                 struct lmv_stripe_md *lsm = md->lmv;
1207
1208                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1209                                 (body->mbo_mode & S_IFMT);
1210                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1211                          PFID(fid));
1212
1213                 LTIME_S(inode->i_mtime) = 0;
1214                 LTIME_S(inode->i_atime) = 0;
1215                 LTIME_S(inode->i_ctime) = 0;
1216                 inode->i_rdev = 0;
1217
1218 #ifdef HAVE_BACKING_DEV_INFO
1219                 /* initializing backing dev info. */
1220                 inode->i_mapping->backing_dev_info =
1221                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1222 #endif
1223                 inode->i_op = &ll_dir_inode_operations;
1224                 inode->i_fop = &ll_dir_operations;
1225                 lli->lli_fid = *fid;
1226                 ll_lli_init(lli);
1227
1228                 LASSERT(lsm != NULL);
1229                 /* master object FID */
1230                 lli->lli_pfid = body->mbo_fid1;
1231                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1232                        lli, PFID(fid), PFID(&lli->lli_pfid));
1233                 unlock_new_inode(inode);
1234         }
1235
1236         RETURN(inode);
1237 }
1238
1239 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1240 {
1241         struct lu_fid *fid;
1242         struct lmv_stripe_md *lsm = md->lmv;
1243         int i;
1244
1245         LASSERT(lsm != NULL);
1246         /* XXX sigh, this lsm_root initialization should be in
1247          * LMV layer, but it needs ll_iget right now, so we
1248          * put this here right now. */
1249         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1250                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1251                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1252                 /* Unfortunately ll_iget will call ll_update_inode,
1253                  * where the initialization of slave inode is slightly
1254                  * different, so it reset lsm_md to NULL to avoid
1255                  * initializing lsm for slave inode. */
1256                 /* For migrating inode, master stripe and master object will
1257                  * be same, so we only need assign this inode */
1258                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1259                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1260                 else
1261                         lsm->lsm_md_oinfo[i].lmo_root =
1262                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1263
1264                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1265                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1266
1267                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1268                         return rc;
1269                 }
1270         }
1271
1272         return 0;
1273 }
1274
1275 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1276                                 const struct lmv_stripe_md *lsm_md2)
1277 {
1278         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1279                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1280                lsm_md1->lsm_md_master_mdt_index ==
1281                                         lsm_md2->lsm_md_master_mdt_index &&
1282                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1283                lsm_md1->lsm_md_layout_version ==
1284                                         lsm_md2->lsm_md_layout_version &&
1285                strcmp(lsm_md1->lsm_md_pool_name,
1286                       lsm_md2->lsm_md_pool_name) == 0;
1287 }
1288
1289 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1290 {
1291         struct ll_inode_info *lli = ll_i2info(inode);
1292         struct lmv_stripe_md *lsm = md->lmv;
1293         int     rc;
1294         ENTRY;
1295
1296         LASSERT(S_ISDIR(inode->i_mode));
1297         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1298                PFID(ll_inode2fid(inode)));
1299
1300         /* no striped information from request. */
1301         if (lsm == NULL) {
1302                 if (lli->lli_lsm_md == NULL) {
1303                         RETURN(0);
1304                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1305                                                 LMV_HASH_FLAG_MIGRATION) {
1306                         /* migration is done, the temporay MIGRATE layout has
1307                          * been removed */
1308                         CDEBUG(D_INODE, DFID" finish migration.\n",
1309                                PFID(ll_inode2fid(inode)));
1310                         lmv_free_memmd(lli->lli_lsm_md);
1311                         lli->lli_lsm_md = NULL;
1312                         RETURN(0);
1313                 } else {
1314                         /* The lustre_md from req does not include stripeEA,
1315                          * see ll_md_setattr */
1316                         RETURN(0);
1317                 }
1318         }
1319
1320         /* set the directory layout */
1321         if (lli->lli_lsm_md == NULL) {
1322                 struct cl_attr  *attr;
1323
1324                 rc = ll_init_lsm_md(inode, md);
1325                 if (rc != 0)
1326                         RETURN(rc);
1327
1328                 /* set md->lmv to NULL, so the following free lustre_md
1329                  * will not free this lsm */
1330                 md->lmv = NULL;
1331                 lli->lli_lsm_md = lsm;
1332
1333                 OBD_ALLOC_PTR(attr);
1334                 if (attr == NULL)
1335                         RETURN(-ENOMEM);
1336
1337                 /* validate the lsm */
1338                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1339                                    ll_md_blocking_ast);
1340                 if (rc != 0) {
1341                         OBD_FREE_PTR(attr);
1342                         RETURN(rc);
1343                 }
1344
1345                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1346                         md->body->mbo_nlink = attr->cat_nlink;
1347                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1348                         md->body->mbo_size = attr->cat_size;
1349                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1350                         md->body->mbo_atime = attr->cat_atime;
1351                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1352                         md->body->mbo_ctime = attr->cat_ctime;
1353                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1354                         md->body->mbo_mtime = attr->cat_mtime;
1355
1356                 OBD_FREE_PTR(attr);
1357
1358                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1359                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1360                 RETURN(0);
1361         }
1362
1363         /* Compare the old and new stripe information */
1364         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1366                 int                     idx;
1367
1368                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1369                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1370                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1371                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1372                        inode, lsm, old_lsm,
1373                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1374                        lsm->lsm_md_stripe_count,
1375                        old_lsm->lsm_md_stripe_count,
1376                        lsm->lsm_md_master_mdt_index,
1377                        old_lsm->lsm_md_master_mdt_index,
1378                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1379                        lsm->lsm_md_layout_version,
1380                        old_lsm->lsm_md_layout_version,
1381                        lsm->lsm_md_pool_name,
1382                        old_lsm->lsm_md_pool_name);
1383
1384                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1385                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1386                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1387                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1388                 }
1389
1390                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1391                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1392                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1393                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1394                 }
1395
1396                 RETURN(-EIO);
1397         }
1398
1399         RETURN(0);
1400 }
1401
1402 void ll_clear_inode(struct inode *inode)
1403 {
1404         struct ll_inode_info *lli = ll_i2info(inode);
1405         struct ll_sb_info *sbi = ll_i2sbi(inode);
1406         ENTRY;
1407
1408         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1409                PFID(ll_inode2fid(inode)), inode);
1410
1411         if (S_ISDIR(inode->i_mode)) {
1412                 /* these should have been cleared in ll_file_release */
1413                 LASSERT(lli->lli_opendir_key == NULL);
1414                 LASSERT(lli->lli_sai == NULL);
1415                 LASSERT(lli->lli_opendir_pid == 0);
1416         }
1417
1418         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1419
1420         LASSERT(!lli->lli_open_fd_write_count);
1421         LASSERT(!lli->lli_open_fd_read_count);
1422         LASSERT(!lli->lli_open_fd_exec_count);
1423
1424         if (lli->lli_mds_write_och)
1425                 ll_md_real_close(inode, FMODE_WRITE);
1426         if (lli->lli_mds_exec_och)
1427                 ll_md_real_close(inode, FMODE_EXEC);
1428         if (lli->lli_mds_read_och)
1429                 ll_md_real_close(inode, FMODE_READ);
1430
1431         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1432                 OBD_FREE(lli->lli_symlink_name,
1433                          strlen(lli->lli_symlink_name) + 1);
1434                 lli->lli_symlink_name = NULL;
1435         }
1436
1437         ll_xattr_cache_destroy(inode);
1438
1439 #ifdef CONFIG_FS_POSIX_ACL
1440         forget_all_cached_acls(inode);
1441         if (lli->lli_posix_acl) {
1442                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1443                 posix_acl_release(lli->lli_posix_acl);
1444                 lli->lli_posix_acl = NULL;
1445         }
1446 #endif
1447         lli->lli_inode_magic = LLI_INODE_DEAD;
1448
1449         if (S_ISDIR(inode->i_mode))
1450                 ll_dir_clear_lsm_md(inode);
1451         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1452                 LASSERT(list_empty(&lli->lli_agl_list));
1453
1454         /*
1455          * XXX This has to be done before lsm is freed below, because
1456          * cl_object still uses inode lsm.
1457          */
1458         cl_inode_fini(inode);
1459
1460         EXIT;
1461 }
1462
1463 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1464 {
1465         struct lustre_md md;
1466         struct inode *inode = dentry->d_inode;
1467         struct ll_sb_info *sbi = ll_i2sbi(inode);
1468         struct ptlrpc_request *request = NULL;
1469         int rc, ia_valid;
1470         ENTRY;
1471
1472         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1473                                      LUSTRE_OPC_ANY, NULL);
1474         if (IS_ERR(op_data))
1475                 RETURN(PTR_ERR(op_data));
1476
1477         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1478         if (rc) {
1479                 ptlrpc_req_finished(request);
1480                 if (rc == -ENOENT) {
1481                         clear_nlink(inode);
1482                         /* Unlinked special device node? Or just a race?
1483                          * Pretend we done everything. */
1484                         if (!S_ISREG(inode->i_mode) &&
1485                             !S_ISDIR(inode->i_mode)) {
1486                                 ia_valid = op_data->op_attr.ia_valid;
1487                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1488                                 rc = simple_setattr(dentry, &op_data->op_attr);
1489                                 op_data->op_attr.ia_valid = ia_valid;
1490                         }
1491                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1492                         CERROR("md_setattr fails: rc = %d\n", rc);
1493                 }
1494                 RETURN(rc);
1495         }
1496
1497         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1498                               sbi->ll_md_exp, &md);
1499         if (rc) {
1500                 ptlrpc_req_finished(request);
1501                 RETURN(rc);
1502         }
1503
1504         ia_valid = op_data->op_attr.ia_valid;
1505         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1506          * cache is not cleared yet. */
1507         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1508         if (S_ISREG(inode->i_mode))
1509                 inode_lock(inode);
1510         rc = simple_setattr(dentry, &op_data->op_attr);
1511         if (S_ISREG(inode->i_mode))
1512                 inode_unlock(inode);
1513         op_data->op_attr.ia_valid = ia_valid;
1514
1515         rc = ll_update_inode(inode, &md);
1516         ptlrpc_req_finished(request);
1517
1518         RETURN(rc);
1519 }
1520
1521 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1522  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1523  * keep these values until such a time that objects are allocated for it.
1524  * We do the MDS operations first, as it is checking permissions for us.
1525  * We don't to the MDS RPC if there is nothing that we want to store there,
1526  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1527  * going to do an RPC anyways.
1528  *
1529  * If we are doing a truncate, we will send the mtime and ctime updates
1530  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1531  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1532  * at the same time.
1533  *
1534  * In case of HSMimport, we only set attr on MDS.
1535  */
1536 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1537 {
1538         struct inode *inode = dentry->d_inode;
1539         struct ll_inode_info *lli = ll_i2info(inode);
1540         struct md_op_data *op_data = NULL;
1541         int rc = 0;
1542         ENTRY;
1543
1544         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1545                "valid %x, hsm_import %d\n",
1546                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1547                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1548                hsm_import);
1549
1550         if (attr->ia_valid & ATTR_SIZE) {
1551                 /* Check new size against VFS/VM file size limit and rlimit */
1552                 rc = inode_newsize_ok(inode, attr->ia_size);
1553                 if (rc)
1554                         RETURN(rc);
1555
1556                 /* The maximum Lustre file size is variable, based on the
1557                  * OST maximum object size and number of stripes.  This
1558                  * needs another check in addition to the VFS check above. */
1559                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1560                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1561                                PFID(&lli->lli_fid), attr->ia_size,
1562                                ll_file_maxbytes(inode));
1563                         RETURN(-EFBIG);
1564                 }
1565
1566                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1567         }
1568
1569         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1570         if (attr->ia_valid & TIMES_SET_FLAGS) {
1571                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1572                     !cfs_capable(CFS_CAP_FOWNER))
1573                         RETURN(-EPERM);
1574         }
1575
1576         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1577         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1578             (attr->ia_valid & ATTR_CTIME)) {
1579                 attr->ia_ctime = CURRENT_TIME;
1580                 attr->ia_valid |= ATTR_CTIME_SET;
1581         }
1582         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1583             (attr->ia_valid & ATTR_ATIME)) {
1584                 attr->ia_atime = CURRENT_TIME;
1585                 attr->ia_valid |= ATTR_ATIME_SET;
1586         }
1587         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1588             (attr->ia_valid & ATTR_MTIME)) {
1589                 attr->ia_mtime = CURRENT_TIME;
1590                 attr->ia_valid |= ATTR_MTIME_SET;
1591         }
1592
1593         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1594                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1595                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1596                        (s64)ktime_get_real_seconds());
1597
1598         if (S_ISREG(inode->i_mode)) {
1599                 if (attr->ia_valid & ATTR_SIZE)
1600                         inode_dio_write_done(inode);
1601                 inode_unlock(inode);
1602         }
1603
1604         /* We always do an MDS RPC, even if we're only changing the size;
1605          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1606
1607         OBD_ALLOC_PTR(op_data);
1608         if (op_data == NULL)
1609                 GOTO(out, rc = -ENOMEM);
1610
1611         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1612                 /* If we are changing file size, file content is
1613                  * modified, flag it. */
1614                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1615                 op_data->op_bias |= MDS_DATA_MODIFIED;
1616                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1617         }
1618
1619         op_data->op_attr = *attr;
1620
1621         rc = ll_md_setattr(dentry, op_data);
1622         if (rc)
1623                 GOTO(out, rc);
1624
1625         if (!S_ISREG(inode->i_mode) || hsm_import)
1626                 GOTO(out, rc = 0);
1627
1628         if (attr->ia_valid & (ATTR_SIZE |
1629                               ATTR_ATIME | ATTR_ATIME_SET |
1630                               ATTR_MTIME | ATTR_MTIME_SET |
1631                               ATTR_CTIME | ATTR_CTIME_SET)) {
1632                 /* For truncate and utimes sending attributes to OSTs, setting
1633                  * mtime/atime to the past will be performed under PW [0:EOF]
1634                  * extent lock (new_size:EOF for truncate).  It may seem
1635                  * excessive to send mtime/atime updates to OSTs when not
1636                  * setting times to past, but it is necessary due to possible
1637                  * time de-synchronization between MDT inode and OST objects */
1638                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1639         }
1640
1641         /* If the file was restored, it needs to set dirty flag.
1642          *
1643          * We've already sent MDS_DATA_MODIFIED flag in
1644          * ll_md_setattr() for truncate. However, the MDT refuses to
1645          * set the HS_DIRTY flag on released files, so we have to set
1646          * it again if the file has been restored. Please check how
1647          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1648          *
1649          * Please notice that if the file is not released, the previous
1650          * MDS_DATA_MODIFIED has taken effect and usually
1651          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1652          * This way we can save an RPC for common open + trunc
1653          * operation. */
1654         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1655                 struct hsm_state_set hss = {
1656                         .hss_valid = HSS_SETMASK,
1657                         .hss_setmask = HS_DIRTY,
1658                 };
1659                 int rc2;
1660
1661                 rc2 = ll_hsm_state_set(inode, &hss);
1662                 /* truncate and write can happen at the same time, so that
1663                  * the file can be set modified even though the file is not
1664                  * restored from released state, and ll_hsm_state_set() is
1665                  * not applicable for the file, and rc2 < 0 is normal in this
1666                  * case. */
1667                 if (rc2 < 0)
1668                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1669                                PFID(ll_inode2fid(inode)), rc2);
1670         }
1671
1672         EXIT;
1673 out:
1674         if (op_data != NULL)
1675                 ll_finish_md_op_data(op_data);
1676
1677         if (S_ISREG(inode->i_mode)) {
1678                 inode_lock(inode);
1679                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1680                         inode_dio_wait(inode);
1681                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1682                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1683                  * inode flags, so there is a gap where S_NOSEC is not set.
1684                  * This can cause a writer to take the i_mutex unnecessarily,
1685                  * but this is safe to do and should be rare. */
1686                 inode_has_no_xattr(inode);
1687         }
1688
1689         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1690                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1691
1692         return rc;
1693 }
1694
1695 int ll_setattr(struct dentry *de, struct iattr *attr)
1696 {
1697         int mode = de->d_inode->i_mode;
1698
1699         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1700                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1701                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1702
1703         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1704                                (ATTR_SIZE|ATTR_MODE)) &&
1705             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1706              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1707               !(attr->ia_mode & S_ISGID))))
1708                 attr->ia_valid |= ATTR_FORCE;
1709
1710         if ((attr->ia_valid & ATTR_MODE) &&
1711             (mode & S_ISUID) &&
1712             !(attr->ia_mode & S_ISUID) &&
1713             !(attr->ia_valid & ATTR_KILL_SUID))
1714                 attr->ia_valid |= ATTR_KILL_SUID;
1715
1716         if ((attr->ia_valid & ATTR_MODE) &&
1717             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1718             !(attr->ia_mode & S_ISGID) &&
1719             !(attr->ia_valid & ATTR_KILL_SGID))
1720                 attr->ia_valid |= ATTR_KILL_SGID;
1721
1722         /* avoid polluted from ATTR_TIMES_SET,
1723          * projid is not expected to be set here */
1724         attr->ia_valid &= ~MDS_ATTR_PROJID;
1725
1726         return ll_setattr_raw(de, attr, false);
1727 }
1728
1729 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1730                        __u64 max_age, __u32 flags)
1731 {
1732         struct ll_sb_info *sbi = ll_s2sbi(sb);
1733         struct obd_statfs obd_osfs;
1734         int rc;
1735         ENTRY;
1736
1737         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1738         if (rc) {
1739                 CERROR("md_statfs fails: rc = %d\n", rc);
1740                 RETURN(rc);
1741         }
1742
1743         osfs->os_type = sb->s_magic;
1744
1745         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1746                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1747
1748         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1749                 flags |= OBD_STATFS_NODELAY;
1750
1751         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1752         if (rc) {
1753                 CERROR("obd_statfs fails: rc = %d\n", rc);
1754                 RETURN(rc);
1755         }
1756
1757         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1758                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1759                obd_osfs.os_files);
1760
1761         osfs->os_bsize = obd_osfs.os_bsize;
1762         osfs->os_blocks = obd_osfs.os_blocks;
1763         osfs->os_bfree = obd_osfs.os_bfree;
1764         osfs->os_bavail = obd_osfs.os_bavail;
1765
1766         /* If we don't have as many objects free on the OST as inodes
1767          * on the MDS, we reduce the total number of inodes to
1768          * compensate, so that the "inodes in use" number is correct.
1769          */
1770         if (obd_osfs.os_ffree < osfs->os_ffree) {
1771                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1772                         obd_osfs.os_ffree;
1773                 osfs->os_ffree = obd_osfs.os_ffree;
1774         }
1775
1776         RETURN(rc);
1777 }
1778 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1779 {
1780         struct super_block *sb = de->d_sb;
1781         struct obd_statfs osfs;
1782         __u64 fsid = huge_encode_dev(sb->s_dev);
1783         int rc;
1784
1785         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1786         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1787
1788         /* Some amount of caching on the client is allowed */
1789         rc = ll_statfs_internal(sb, &osfs,
1790                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1791                                 0);
1792         if (rc)
1793                 return rc;
1794
1795         statfs_unpack(sfs, &osfs);
1796
1797         /* We need to downshift for all 32-bit kernels, because we can't
1798          * tell if the kernel is being called via sys_statfs64() or not.
1799          * Stop before overflowing f_bsize - in which case it is better
1800          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1801         if (sizeof(long) < 8) {
1802                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1803                         sfs->f_bsize <<= 1;
1804
1805                         osfs.os_blocks >>= 1;
1806                         osfs.os_bfree >>= 1;
1807                         osfs.os_bavail >>= 1;
1808                 }
1809         }
1810
1811         sfs->f_blocks = osfs.os_blocks;
1812         sfs->f_bfree = osfs.os_bfree;
1813         sfs->f_bavail = osfs.os_bavail;
1814         sfs->f_fsid.val[0] = (__u32)fsid;
1815         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1816         return 0;
1817 }
1818
1819 void ll_inode_size_lock(struct inode *inode)
1820 {
1821         struct ll_inode_info *lli;
1822
1823         LASSERT(!S_ISDIR(inode->i_mode));
1824
1825         lli = ll_i2info(inode);
1826         mutex_lock(&lli->lli_size_mutex);
1827 }
1828
1829 void ll_inode_size_unlock(struct inode *inode)
1830 {
1831         struct ll_inode_info *lli;
1832
1833         lli = ll_i2info(inode);
1834         mutex_unlock(&lli->lli_size_mutex);
1835 }
1836
1837 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1838 {
1839         struct ll_inode_info *lli = ll_i2info(inode);
1840         struct mdt_body *body = md->body;
1841         struct ll_sb_info *sbi = ll_i2sbi(inode);
1842         int rc = 0;
1843
1844         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1845                 rc = cl_file_inode_init(inode, md);
1846                 if (rc)
1847                         return rc;
1848         }
1849
1850         if (S_ISDIR(inode->i_mode)) {
1851                 rc = ll_update_lsm_md(inode, md);
1852                 if (rc != 0)
1853                         return rc;
1854         }
1855
1856 #ifdef CONFIG_FS_POSIX_ACL
1857         if (body->mbo_valid & OBD_MD_FLACL) {
1858                 spin_lock(&lli->lli_lock);
1859                 if (lli->lli_posix_acl)
1860                         posix_acl_release(lli->lli_posix_acl);
1861                 lli->lli_posix_acl = md->posix_acl;
1862                 spin_unlock(&lli->lli_lock);
1863         }
1864 #endif
1865         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1866                                         sbi->ll_flags & LL_SBI_32BIT_API);
1867         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1868
1869         if (body->mbo_valid & OBD_MD_FLATIME) {
1870                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1871                         LTIME_S(inode->i_atime) = body->mbo_atime;
1872                 lli->lli_atime = body->mbo_atime;
1873         }
1874
1875         if (body->mbo_valid & OBD_MD_FLMTIME) {
1876                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1877                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1878                                "to %llu\n", inode->i_ino,
1879                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1880                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1881                 }
1882                 lli->lli_mtime = body->mbo_mtime;
1883         }
1884
1885         if (body->mbo_valid & OBD_MD_FLCTIME) {
1886                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1887                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1888                 lli->lli_ctime = body->mbo_ctime;
1889         }
1890
1891         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1892         if (body->mbo_valid & OBD_MD_FLFLAGS)
1893                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1894         if (body->mbo_valid & OBD_MD_FLMODE)
1895                 inode->i_mode = (inode->i_mode & S_IFMT) |
1896                                 (body->mbo_mode & ~S_IFMT);
1897
1898         if (body->mbo_valid & OBD_MD_FLTYPE)
1899                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1900                                 (body->mbo_mode & S_IFMT);
1901
1902         LASSERT(inode->i_mode != 0);
1903         if (S_ISREG(inode->i_mode))
1904                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1905                                        LL_MAX_BLKSIZE_BITS);
1906         else
1907                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1908
1909         if (body->mbo_valid & OBD_MD_FLUID)
1910                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1911         if (body->mbo_valid & OBD_MD_FLGID)
1912                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1913         if (body->mbo_valid & OBD_MD_FLPROJID)
1914                 lli->lli_projid = body->mbo_projid;
1915         if (body->mbo_valid & OBD_MD_FLNLINK)
1916                 set_nlink(inode, body->mbo_nlink);
1917         if (body->mbo_valid & OBD_MD_FLRDEV)
1918                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1919
1920         if (body->mbo_valid & OBD_MD_FLID) {
1921                 /* FID shouldn't be changed! */
1922                 if (fid_is_sane(&lli->lli_fid)) {
1923                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1924                                  "Trying to change FID "DFID
1925                                  " to the "DFID", inode "DFID"(%p)\n",
1926                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1927                                  PFID(ll_inode2fid(inode)), inode);
1928                 } else {
1929                         lli->lli_fid = body->mbo_fid1;
1930                 }
1931         }
1932
1933         LASSERT(fid_seq(&lli->lli_fid) != 0);
1934
1935         if (body->mbo_valid & OBD_MD_FLSIZE) {
1936                 i_size_write(inode, body->mbo_size);
1937
1938                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1939                        PFID(ll_inode2fid(inode)),
1940                        (unsigned long long)body->mbo_size);
1941
1942                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1943                         inode->i_blocks = body->mbo_blocks;
1944         }
1945
1946         if (body->mbo_valid & OBD_MD_TSTATE) {
1947                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1948                  * clear it when done to ensure to start again
1949                  * glimpsing updated attrs
1950                  */
1951                 if (body->mbo_t_state & MS_RESTORE)
1952                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1953                 else
1954                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1955         }
1956
1957         return 0;
1958 }
1959
1960 int ll_read_inode2(struct inode *inode, void *opaque)
1961 {
1962         struct lustre_md *md = opaque;
1963         struct ll_inode_info *lli = ll_i2info(inode);
1964         int     rc;
1965         ENTRY;
1966
1967         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1968                PFID(&lli->lli_fid), inode);
1969
1970         /* Core attributes from the MDS first.  This is a new inode, and
1971          * the VFS doesn't zero times in the core inode so we have to do
1972          * it ourselves.  They will be overwritten by either MDS or OST
1973          * attributes - we just need to make sure they aren't newer. */
1974         LTIME_S(inode->i_mtime) = 0;
1975         LTIME_S(inode->i_atime) = 0;
1976         LTIME_S(inode->i_ctime) = 0;
1977         inode->i_rdev = 0;
1978         rc = ll_update_inode(inode, md);
1979         if (rc != 0)
1980                 RETURN(rc);
1981
1982         /* OIDEBUG(inode); */
1983
1984 #ifdef HAVE_BACKING_DEV_INFO
1985         /* initializing backing dev info. */
1986         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1987 #endif
1988         if (S_ISREG(inode->i_mode)) {
1989                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1990                 inode->i_op = &ll_file_inode_operations;
1991                 inode->i_fop = sbi->ll_fop;
1992                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1993                 EXIT;
1994         } else if (S_ISDIR(inode->i_mode)) {
1995                 inode->i_op = &ll_dir_inode_operations;
1996                 inode->i_fop = &ll_dir_operations;
1997                 EXIT;
1998         } else if (S_ISLNK(inode->i_mode)) {
1999                 inode->i_op = &ll_fast_symlink_inode_operations;
2000                 EXIT;
2001         } else {
2002                 inode->i_op = &ll_special_inode_operations;
2003
2004                 init_special_inode(inode, inode->i_mode,
2005                                    inode->i_rdev);
2006
2007                 EXIT;
2008         }
2009
2010         return 0;
2011 }
2012
2013 void ll_delete_inode(struct inode *inode)
2014 {
2015         struct ll_inode_info *lli = ll_i2info(inode);
2016         ENTRY;
2017
2018         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2019                 /* It is last chance to write out dirty pages,
2020                  * otherwise we may lose data while umount */
2021                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2022
2023         truncate_inode_pages_final(&inode->i_data);
2024
2025         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2026                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2027                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2028
2029 #ifdef HAVE_SBOPS_EVICT_INODE
2030         ll_clear_inode(inode);
2031 #endif
2032         clear_inode(inode);
2033
2034         EXIT;
2035 }
2036
2037 int ll_iocontrol(struct inode *inode, struct file *file,
2038                  unsigned int cmd, unsigned long arg)
2039 {
2040         struct ll_sb_info *sbi = ll_i2sbi(inode);
2041         struct ptlrpc_request *req = NULL;
2042         int rc, flags = 0;
2043         ENTRY;
2044
2045         switch(cmd) {
2046         case FSFILT_IOC_GETFLAGS: {
2047                 struct mdt_body *body;
2048                 struct md_op_data *op_data;
2049
2050                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2051                                              0, 0, LUSTRE_OPC_ANY,
2052                                              NULL);
2053                 if (IS_ERR(op_data))
2054                         RETURN(PTR_ERR(op_data));
2055
2056                 op_data->op_valid = OBD_MD_FLFLAGS;
2057                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2058                 ll_finish_md_op_data(op_data);
2059                 if (rc) {
2060                         CERROR("%s: failure inode "DFID": rc = %d\n",
2061                                sbi->ll_md_exp->exp_obd->obd_name,
2062                                PFID(ll_inode2fid(inode)), rc);
2063                         RETURN(-abs(rc));
2064                 }
2065
2066                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2067
2068                 flags = body->mbo_flags;
2069
2070                 ptlrpc_req_finished(req);
2071
2072                 RETURN(put_user(flags, (int __user *)arg));
2073         }
2074         case FSFILT_IOC_SETFLAGS: {
2075                 struct iattr *attr;
2076                 struct md_op_data *op_data;
2077                 struct cl_object *obj;
2078
2079                 if (get_user(flags, (int __user *)arg))
2080                         RETURN(-EFAULT);
2081
2082                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2083                                              LUSTRE_OPC_ANY, NULL);
2084                 if (IS_ERR(op_data))
2085                         RETURN(PTR_ERR(op_data));
2086
2087                 op_data->op_attr_flags = flags;
2088                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2089                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2090                 ll_finish_md_op_data(op_data);
2091                 ptlrpc_req_finished(req);
2092                 if (rc)
2093                         RETURN(rc);
2094
2095                 inode->i_flags = ll_ext_to_inode_flags(flags);
2096
2097                 obj = ll_i2info(inode)->lli_clob;
2098                 if (obj == NULL)
2099                         RETURN(0);
2100
2101                 OBD_ALLOC_PTR(attr);
2102                 if (attr == NULL)
2103                         RETURN(-ENOMEM);
2104
2105                 attr->ia_valid = ATTR_ATTR_FLAG;
2106                 rc = cl_setattr_ost(obj, attr, flags);
2107
2108                 OBD_FREE_PTR(attr);
2109                 RETURN(rc);
2110         }
2111         default:
2112                 RETURN(-ENOSYS);
2113         }
2114
2115         RETURN(0);
2116 }
2117
2118 int ll_flush_ctx(struct inode *inode)
2119 {
2120         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2121
2122         CDEBUG(D_SEC, "flush context for user %d\n",
2123                from_kuid(&init_user_ns, current_uid()));
2124
2125         obd_set_info_async(NULL, sbi->ll_md_exp,
2126                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2127                            0, NULL, NULL);
2128         obd_set_info_async(NULL, sbi->ll_dt_exp,
2129                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2130                            0, NULL, NULL);
2131         return 0;
2132 }
2133
2134 /* umount -f client means force down, don't save state */
2135 void ll_umount_begin(struct super_block *sb)
2136 {
2137         struct ll_sb_info *sbi = ll_s2sbi(sb);
2138         struct obd_device *obd;
2139         struct obd_ioctl_data *ioc_data;
2140         struct l_wait_info lwi;
2141         wait_queue_head_t waitq;
2142         ENTRY;
2143
2144         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2145                sb->s_count, atomic_read(&sb->s_active));
2146
2147         obd = class_exp2obd(sbi->ll_md_exp);
2148         if (obd == NULL) {
2149                 CERROR("Invalid MDC connection handle %#llx\n",
2150                        sbi->ll_md_exp->exp_handle.h_cookie);
2151                 EXIT;
2152                 return;
2153         }
2154         obd->obd_force = 1;
2155
2156         obd = class_exp2obd(sbi->ll_dt_exp);
2157         if (obd == NULL) {
2158                 CERROR("Invalid LOV connection handle %#llx\n",
2159                        sbi->ll_dt_exp->exp_handle.h_cookie);
2160                 EXIT;
2161                 return;
2162         }
2163         obd->obd_force = 1;
2164
2165         OBD_ALLOC_PTR(ioc_data);
2166         if (ioc_data) {
2167                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2168                               sizeof *ioc_data, ioc_data, NULL);
2169
2170                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2171                               sizeof *ioc_data, ioc_data, NULL);
2172
2173                 OBD_FREE_PTR(ioc_data);
2174         }
2175
2176         /* Really, we'd like to wait until there are no requests outstanding,
2177          * and then continue.  For now, we just periodically checking for vfs
2178          * to decrement mnt_cnt and hope to finish it within 10sec.
2179          */
2180         init_waitqueue_head(&waitq);
2181         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2182                                    cfs_time_seconds(1), NULL, NULL);
2183         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2184
2185         EXIT;
2186 }
2187
2188 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2189 {
2190         struct ll_sb_info *sbi = ll_s2sbi(sb);
2191         char *profilenm = get_profile_name(sb);
2192         int err;
2193         __u32 read_only;
2194
2195         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2196                 read_only = *flags & MS_RDONLY;
2197                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2198                                          sizeof(KEY_READ_ONLY),
2199                                          KEY_READ_ONLY, sizeof(read_only),
2200                                          &read_only, NULL);
2201                 if (err) {
2202                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2203                                       profilenm, read_only ?
2204                                       "read-only" : "read-write", err);
2205                         return err;
2206                 }
2207
2208                 if (read_only)
2209                         sb->s_flags |= MS_RDONLY;
2210                 else
2211                         sb->s_flags &= ~MS_RDONLY;
2212
2213                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2214                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2215                                       read_only ?  "read-only" : "read-write");
2216         }
2217         return 0;
2218 }
2219
2220 /**
2221  * Cleanup the open handle that is cached on MDT-side.
2222  *
2223  * For open case, the client side open handling thread may hit error
2224  * after the MDT grant the open. Under such case, the client should
2225  * send close RPC to the MDT as cleanup; otherwise, the open handle
2226  * on the MDT will be leaked there until the client umount or evicted.
2227  *
2228  * In further, if someone unlinked the file, because the open handle
2229  * holds the reference on such file/object, then it will block the
2230  * subsequent threads that want to locate such object via FID.
2231  *
2232  * \param[in] sb        super block for this file-system
2233  * \param[in] open_req  pointer to the original open request
2234  */
2235 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2236 {
2237         struct mdt_body                 *body;
2238         struct md_op_data               *op_data;
2239         struct ptlrpc_request           *close_req = NULL;
2240         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2241         ENTRY;
2242
2243         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2244         OBD_ALLOC_PTR(op_data);
2245         if (op_data == NULL) {
2246                 CWARN("%s: cannot allocate op_data to release open handle for "
2247                       DFID"\n",
2248                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2249
2250                 RETURN_EXIT;
2251         }
2252
2253         op_data->op_fid1 = body->mbo_fid1;
2254         op_data->op_handle = body->mbo_handle;
2255         op_data->op_mod_time = ktime_get_real_seconds();
2256         md_close(exp, op_data, NULL, &close_req);
2257         ptlrpc_req_finished(close_req);
2258         ll_finish_md_op_data(op_data);
2259
2260         EXIT;
2261 }
2262
2263 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2264                   struct super_block *sb, struct lookup_intent *it)
2265 {
2266         struct ll_sb_info *sbi = NULL;
2267         struct lustre_md md = { NULL };
2268         int rc;
2269         ENTRY;
2270
2271         LASSERT(*inode || sb);
2272         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2273         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2274                               sbi->ll_md_exp, &md);
2275         if (rc != 0)
2276                 GOTO(cleanup, rc);
2277
2278         if (*inode) {
2279                 rc = ll_update_inode(*inode, &md);
2280                 if (rc != 0)
2281                         GOTO(out, rc);
2282         } else {
2283                 LASSERT(sb != NULL);
2284
2285                 /*
2286                  * At this point server returns to client's same fid as client
2287                  * generated for creating. So using ->fid1 is okay here.
2288                  */
2289                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2290                         CERROR("%s: Fid is insane "DFID"\n",
2291                                 ll_get_fsname(sb, NULL, 0),
2292                                 PFID(&md.body->mbo_fid1));
2293                         GOTO(out, rc = -EINVAL);
2294                 }
2295
2296                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2297                                              sbi->ll_flags & LL_SBI_32BIT_API),
2298                                  &md);
2299                 if (IS_ERR(*inode)) {
2300 #ifdef CONFIG_FS_POSIX_ACL
2301                         if (md.posix_acl) {
2302                                 posix_acl_release(md.posix_acl);
2303                                 md.posix_acl = NULL;
2304                         }
2305 #endif
2306                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2307                         *inode = NULL;
2308                         CERROR("new_inode -fatal: rc %d\n", rc);
2309                         GOTO(out, rc);
2310                 }
2311         }
2312
2313         /* Handling piggyback layout lock.
2314          * Layout lock can be piggybacked by getattr and open request.
2315          * The lsm can be applied to inode only if it comes with a layout lock
2316          * otherwise correct layout may be overwritten, for example:
2317          * 1. proc1: mdt returns a lsm but not granting layout
2318          * 2. layout was changed by another client
2319          * 3. proc2: refresh layout and layout lock granted
2320          * 4. proc1: to apply a stale layout */
2321         if (it != NULL && it->it_lock_mode != 0) {
2322                 struct lustre_handle lockh;
2323                 struct ldlm_lock *lock;
2324
2325                 lockh.cookie = it->it_lock_handle;
2326                 lock = ldlm_handle2lock(&lockh);
2327                 LASSERT(lock != NULL);
2328                 if (ldlm_has_layout(lock)) {
2329                         struct cl_object_conf conf;
2330
2331                         memset(&conf, 0, sizeof(conf));
2332                         conf.coc_opc = OBJECT_CONF_SET;
2333                         conf.coc_inode = *inode;
2334                         conf.coc_lock = lock;
2335                         conf.u.coc_layout = md.layout;
2336                         (void)ll_layout_conf(*inode, &conf);
2337                 }
2338                 LDLM_LOCK_PUT(lock);
2339         }
2340
2341         GOTO(out, rc = 0);
2342
2343 out:
2344         md_free_lustre_md(sbi->ll_md_exp, &md);
2345
2346 cleanup:
2347         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2348                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2349
2350         return rc;
2351 }
2352
2353 int ll_obd_statfs(struct inode *inode, void __user *arg)
2354 {
2355         struct ll_sb_info *sbi = NULL;
2356         struct obd_export *exp;
2357         char *buf = NULL;
2358         struct obd_ioctl_data *data = NULL;
2359         __u32 type;
2360         int len = 0, rc;
2361
2362         if (!inode || !(sbi = ll_i2sbi(inode)))
2363                 GOTO(out_statfs, rc = -EINVAL);
2364
2365         rc = obd_ioctl_getdata(&buf, &len, arg);
2366         if (rc)
2367                 GOTO(out_statfs, rc);
2368
2369         data = (void*)buf;
2370         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2371             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2372                 GOTO(out_statfs, rc = -EINVAL);
2373
2374         if (data->ioc_inllen1 != sizeof(__u32) ||
2375             data->ioc_inllen2 != sizeof(__u32) ||
2376             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2377             data->ioc_plen2 != sizeof(struct obd_uuid))
2378                 GOTO(out_statfs, rc = -EINVAL);
2379
2380         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2381         if (type & LL_STATFS_LMV)
2382                 exp = sbi->ll_md_exp;
2383         else if (type & LL_STATFS_LOV)
2384                 exp = sbi->ll_dt_exp;
2385         else
2386                 GOTO(out_statfs, rc = -ENODEV);
2387
2388         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2389         if (rc)
2390                 GOTO(out_statfs, rc);
2391 out_statfs:
2392         OBD_FREE_LARGE(buf, len);
2393         return rc;
2394 }
2395
2396 int ll_process_config(struct lustre_cfg *lcfg)
2397 {
2398         struct super_block *sb;
2399         unsigned long x;
2400         int rc = 0;
2401         char *ptr;
2402
2403         /* The instance name contains the sb: lustre-client-aacfe000 */
2404         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2405         if (!ptr || !*(++ptr))
2406                 return -EINVAL;
2407         if (sscanf(ptr, "%lx", &x) != 1)
2408                 return -EINVAL;
2409         sb = (struct super_block *)x;
2410         /* This better be a real Lustre superblock! */
2411         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2412
2413         /* Note we have not called client_common_fill_super yet, so
2414            proc fns must be able to handle that! */
2415         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2416                                       lcfg, sb);
2417         if (rc > 0)
2418                 rc = 0;
2419         return rc;
2420 }
2421
2422 /* this function prepares md_op_data hint for passing it down to MD stack. */
2423 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2424                                       struct inode *i1, struct inode *i2,
2425                                       const char *name, size_t namelen,
2426                                       __u32 mode, __u32 opc, void *data)
2427 {
2428         LASSERT(i1 != NULL);
2429
2430         if (name == NULL) {
2431                 /* Do not reuse namelen for something else. */
2432                 if (namelen != 0)
2433                         return ERR_PTR(-EINVAL);
2434         } else {
2435                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2436                         return ERR_PTR(-ENAMETOOLONG);
2437
2438                 if (!lu_name_is_valid_2(name, namelen))
2439                         return ERR_PTR(-EINVAL);
2440         }
2441
2442         if (op_data == NULL)
2443                 OBD_ALLOC_PTR(op_data);
2444
2445         if (op_data == NULL)
2446                 return ERR_PTR(-ENOMEM);
2447
2448         ll_i2gids(op_data->op_suppgids, i1, i2);
2449         op_data->op_fid1 = *ll_inode2fid(i1);
2450         op_data->op_default_stripe_offset = -1;
2451         if (S_ISDIR(i1->i_mode)) {
2452                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2453                 if (opc == LUSTRE_OPC_MKDIR)
2454                         op_data->op_default_stripe_offset =
2455                                    ll_i2info(i1)->lli_def_stripe_offset;
2456         }
2457
2458         if (i2) {
2459                 op_data->op_fid2 = *ll_inode2fid(i2);
2460                 if (S_ISDIR(i2->i_mode))
2461                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2462         } else {
2463                 fid_zero(&op_data->op_fid2);
2464         }
2465
2466         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2467                 op_data->op_cli_flags |= CLI_HASH64;
2468
2469         if (ll_need_32bit_api(ll_i2sbi(i1)))
2470                 op_data->op_cli_flags |= CLI_API32;
2471
2472         op_data->op_name = name;
2473         op_data->op_namelen = namelen;
2474         op_data->op_mode = mode;
2475         op_data->op_mod_time = cfs_time_current_sec();
2476         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2477         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2478         op_data->op_cap = cfs_curproc_cap_pack();
2479         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2480              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2481                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2482         } else {
2483                 op_data->op_mds = 0;
2484         }
2485         op_data->op_data = data;
2486
2487         return op_data;
2488 }
2489
2490 void ll_finish_md_op_data(struct md_op_data *op_data)
2491 {
2492         security_release_secctx(op_data->op_file_secctx,
2493                                 op_data->op_file_secctx_size);
2494         OBD_FREE_PTR(op_data);
2495 }
2496
2497 #ifdef HAVE_SUPEROPS_USE_DENTRY
2498 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2499 #else
2500 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2501 #endif
2502 {
2503         struct ll_sb_info *sbi;
2504
2505 #ifdef HAVE_SUPEROPS_USE_DENTRY
2506         LASSERT((seq != NULL) && (dentry != NULL));
2507         sbi = ll_s2sbi(dentry->d_sb);
2508 #else
2509         LASSERT((seq != NULL) && (vfs != NULL));
2510         sbi = ll_s2sbi(vfs->mnt_sb);
2511 #endif
2512
2513         if (sbi->ll_flags & LL_SBI_NOLCK)
2514                 seq_puts(seq, ",nolock");
2515
2516         if (sbi->ll_flags & LL_SBI_FLOCK)
2517                 seq_puts(seq, ",flock");
2518
2519         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2520                 seq_puts(seq, ",localflock");
2521
2522         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2523                 seq_puts(seq, ",user_xattr");
2524
2525         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2526                 seq_puts(seq, ",lazystatfs");
2527
2528         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2529                 seq_puts(seq, ",user_fid2path");
2530
2531         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2532                 seq_puts(seq, ",always_ping");
2533
2534         RETURN(0);
2535 }
2536
2537 /**
2538  * Get obd name by cmd, and copy out to user space
2539  */
2540 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2541 {
2542         struct ll_sb_info *sbi = ll_i2sbi(inode);
2543         struct obd_device *obd;
2544         ENTRY;
2545
2546         if (cmd == OBD_IOC_GETDTNAME)
2547                 obd = class_exp2obd(sbi->ll_dt_exp);
2548         else if (cmd == OBD_IOC_GETMDNAME)
2549                 obd = class_exp2obd(sbi->ll_md_exp);
2550         else
2551                 RETURN(-EINVAL);
2552
2553         if (!obd)
2554                 RETURN(-ENOENT);
2555
2556         if (copy_to_user((void __user *)arg, obd->obd_name,
2557                          strlen(obd->obd_name) + 1))
2558                 RETURN(-EFAULT);
2559
2560         RETURN(0);
2561 }
2562
2563 /**
2564  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2565  * fsname will be returned in this buffer; otherwise, a static buffer will be
2566  * used to store the fsname and returned to caller.
2567  */
2568 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2569 {
2570         static char fsname_static[MTI_NAME_MAXLEN];
2571         struct lustre_sb_info *lsi = s2lsi(sb);
2572         char *ptr;
2573         int len;
2574
2575         if (buf == NULL) {
2576                 /* this means the caller wants to use static buffer
2577                  * and it doesn't care about race. Usually this is
2578                  * in error reporting path */
2579                 buf = fsname_static;
2580                 buflen = sizeof(fsname_static);
2581         }
2582
2583         len = strlen(lsi->lsi_lmd->lmd_profile);
2584         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2585         if (ptr && (strcmp(ptr, "-client") == 0))
2586                 len -= 7;
2587
2588         if (unlikely(len >= buflen))
2589                 len = buflen - 1;
2590         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2591         buf[len] = '\0';
2592
2593         return buf;
2594 }
2595
2596 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2597 {
2598         char *path = NULL;
2599
2600         struct path p;
2601
2602         p.dentry = dentry;
2603         p.mnt = current->fs->root.mnt;
2604         path_get(&p);
2605         path = d_path(&p, buf, bufsize);
2606         path_put(&p);
2607         return path;
2608 }
2609
2610 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2611 {
2612         char *buf, *path = NULL;
2613         struct dentry *dentry = NULL;
2614         struct inode *inode = page->mapping->host;
2615
2616         /* this can be called inside spin lock so use GFP_ATOMIC. */
2617         buf = (char *)__get_free_page(GFP_ATOMIC);
2618         if (buf != NULL) {
2619                 dentry = d_find_alias(page->mapping->host);
2620                 if (dentry != NULL)
2621                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2622         }
2623
2624         CDEBUG(D_WARNING,
2625                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2626                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2627                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2628                PFID(ll_inode2fid(inode)),
2629                (path && !IS_ERR(path)) ? path : "", ioret);
2630
2631         if (dentry != NULL)
2632                 dput(dentry);
2633
2634         if (buf != NULL)
2635                 free_page((unsigned long)buf);
2636 }
2637
2638 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2639                         struct lov_user_md **kbuf)
2640 {
2641         struct lov_user_md      lum;
2642         ssize_t                 lum_size;
2643         ENTRY;
2644
2645         if (copy_from_user(&lum, md, sizeof(lum)))
2646                 RETURN(-EFAULT);
2647
2648         lum_size = ll_lov_user_md_size(&lum);
2649         if (lum_size < 0)
2650                 RETURN(lum_size);
2651
2652         OBD_ALLOC(*kbuf, lum_size);
2653         if (*kbuf == NULL)
2654                 RETURN(-ENOMEM);
2655
2656         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2657                 OBD_FREE(*kbuf, lum_size);
2658                 RETURN(-EFAULT);
2659         }
2660
2661         RETURN(lum_size);
2662 }
2663
2664 /*
2665  * Compute llite root squash state after a change of root squash
2666  * configuration setting or add/remove of a lnet nid
2667  */
2668 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2669 {
2670         struct root_squash_info *squash = &sbi->ll_squash;
2671         int i;
2672         bool matched;
2673         struct lnet_process_id id;
2674
2675         /* Update norootsquash flag */
2676         down_write(&squash->rsi_sem);
2677         if (list_empty(&squash->rsi_nosquash_nids))
2678                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2679         else {
2680                 /* Do not apply root squash as soon as one of our NIDs is
2681                  * in the nosquash_nids list */
2682                 matched = false;
2683                 i = 0;
2684                 while (LNetGetId(i++, &id) != -ENOENT) {
2685                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2686                                 continue;
2687                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2688                                 matched = true;
2689                                 break;
2690                         }
2691                 }
2692                 if (matched)
2693                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2694                 else
2695                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2696         }
2697         up_write(&squash->rsi_sem);
2698 }
2699
2700 /**
2701  * Parse linkea content to extract information about a given hardlink
2702  *
2703  * \param[in]   ldata      - Initialized linkea data
2704  * \param[in]   linkno     - Link identifier
2705  * \param[out]  parent_fid - The entry's parent FID
2706  * \param[out]  ln         - Entry name destination buffer
2707  *
2708  * \retval 0 on success
2709  * \retval Appropriate negative error code on failure
2710  */
2711 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2712                             struct lu_fid *parent_fid, struct lu_name *ln)
2713 {
2714         unsigned int    idx;
2715         int             rc;
2716         ENTRY;
2717
2718         rc = linkea_init_with_rec(ldata);
2719         if (rc < 0)
2720                 RETURN(rc);
2721
2722         if (linkno >= ldata->ld_leh->leh_reccount)
2723                 /* beyond last link */
2724                 RETURN(-ENODATA);
2725
2726         linkea_first_entry(ldata);
2727         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2728                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2729                                     parent_fid);
2730                 if (idx == linkno)
2731                         break;
2732
2733                 linkea_next_entry(ldata);
2734         }
2735
2736         if (idx < linkno)
2737                 RETURN(-ENODATA);
2738
2739         RETURN(0);
2740 }
2741
2742 /**
2743  * Get parent FID and name of an identified link. Operation is performed for
2744  * a given link number, letting the caller iterate over linkno to list one or
2745  * all links of an entry.
2746  *
2747  * \param[in]     file - File descriptor against which to perform the operation
2748  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2749  *                       on and the available size. It is eventually filled with
2750  *                       the requested information or left untouched on error
2751  *
2752  * \retval - 0 on success
2753  * \retval - Appropriate negative error code on failure
2754  */
2755 int ll_getparent(struct file *file, struct getparent __user *arg)
2756 {
2757         struct inode            *inode = file_inode(file);
2758         struct linkea_data      *ldata;
2759         struct lu_buf            buf = LU_BUF_NULL;
2760         struct lu_name           ln;
2761         struct lu_fid            parent_fid;
2762         __u32                    linkno;
2763         __u32                    name_size;
2764         int                      rc;
2765
2766         ENTRY;
2767
2768         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2769             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2770                 RETURN(-EPERM);
2771
2772         if (get_user(name_size, &arg->gp_name_size))
2773                 RETURN(-EFAULT);
2774
2775         if (get_user(linkno, &arg->gp_linkno))
2776                 RETURN(-EFAULT);
2777
2778         if (name_size > PATH_MAX)
2779                 RETURN(-EINVAL);
2780
2781         OBD_ALLOC(ldata, sizeof(*ldata));
2782         if (ldata == NULL)
2783                 RETURN(-ENOMEM);
2784
2785         rc = linkea_data_new(ldata, &buf);
2786         if (rc < 0)
2787                 GOTO(ldata_free, rc);
2788
2789 #ifdef HAVE_XATTR_HANDLER_FLAGS
2790         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2791                            buf.lb_len, OBD_MD_FLXATTR);
2792 #else
2793         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2794                          buf.lb_len);
2795 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2796         if (rc < 0)
2797                 GOTO(lb_free, rc);
2798
2799         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2800         if (rc < 0)
2801                 GOTO(lb_free, rc);
2802
2803         if (ln.ln_namelen >= name_size)
2804                 GOTO(lb_free, rc = -EOVERFLOW);
2805
2806         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2807                 GOTO(lb_free, rc = -EFAULT);
2808
2809         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2810                 GOTO(lb_free, rc = -EFAULT);
2811
2812         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2813                 GOTO(lb_free, rc = -EFAULT);
2814
2815 lb_free:
2816         lu_buf_free(&buf);
2817 ldata_free:
2818         OBD_FREE(ldata, sizeof(*ldata));
2819
2820         RETURN(rc);
2821 }