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, 2017, 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/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/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_running_max = LL_SA_RUNNING_DEF;
 127         sbi->ll_sa_max = LL_SA_RPC_DEF;
 128         atomic_set(&sbi->ll_sa_total, 0);
 129         atomic_set(&sbi->ll_sa_wrong, 0);
 130         atomic_set(&sbi->ll_sa_running, 0);
 131         atomic_set(&sbi->ll_agl_total, 0);
 132         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 133         sbi->ll_flags |= LL_SBI_FAST_READ;
 134         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 135
 136         /* root squash */
 137         sbi->ll_squash.rsi_uid = 0;
 138         sbi->ll_squash.rsi_gid = 0;
 139         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 140         init_rwsem(&sbi->ll_squash.rsi_sem);
 141
 142         RETURN(sbi);
 143 }
 144
 145 static void ll_free_sbi(struct super_block *sb)
 146 {
 147         struct ll_sb_info *sbi = ll_s2sbi(sb);
 148         ENTRY;
 149
 150         if (sbi != NULL) {
 151                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 152                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 153                 if (sbi->ll_cache != NULL) {
 154                         cl_cache_decref(sbi->ll_cache);
 155                         sbi->ll_cache = NULL;
 156                 }
 157                 OBD_FREE(sbi, sizeof(*sbi));
 158         }
 159         EXIT;
 160 }
 161
 162 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 163 {
 164         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 165                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 166 }
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = NULL;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_statfs *osfs = NULL;
 175         struct ptlrpc_request *request = NULL;
 176         struct obd_connect_data *data = NULL;
 177         struct obd_uuid *uuid;
 178         struct md_op_data *op_data;
 179         struct lustre_md lmd;
 180         u64 valid;
 181         int size, err, checksum;
 182         ENTRY;
 183
 184         obd = class_name2obd(md);
 185         if (!obd) {
 186                 CERROR("MD %s: not setup or attached\n", md);
 187                 RETURN(-EINVAL);
 188         }
 189
 190         OBD_ALLOC_PTR(data);
 191         if (data == NULL)
 192                 RETURN(-ENOMEM);
 193
 194         OBD_ALLOC_PTR(osfs);
 195         if (osfs == NULL) {
 196                 OBD_FREE_PTR(data);
 197                 RETURN(-ENOMEM);
 198         }
 199
 200         /* pass client page size via ocd_grant_blkbits, the server should report
 201          * back its backend blocksize for grant calculation purpose */
 202         data->ocd_grant_blkbits = PAGE_SHIFT;
 203
 204         /* indicate MDT features supported by this client */
 205         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 206                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 207                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 208                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 209                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 210                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 211                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 212                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 213                                   OBD_CONNECT_64BITHASH |
 214                                   OBD_CONNECT_EINPROGRESS |
 215                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 216                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 217                                   OBD_CONNECT_MAX_EASIZE |
 218                                   OBD_CONNECT_FLOCK_DEAD |
 219                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 220                                   OBD_CONNECT_OPEN_BY_FID |
 221                                   OBD_CONNECT_DIR_STRIPE |
 222                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 223                                   OBD_CONNECT_SUBTREE |
 224                                   OBD_CONNECT_MULTIMODRPCS |
 225                                   OBD_CONNECT_GRANT_PARAM |
 226                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 227
 228         data->ocd_connect_flags2 = OBD_CONNECT2_FLR | OBD_CONNECT2_LOCK_CONVERT;
 229
 230 #ifdef HAVE_LRU_RESIZE_SUPPORT
 231         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 232                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 233 #endif
 234 #ifdef CONFIG_FS_POSIX_ACL
 235         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 236                                    OBD_CONNECT_LARGE_ACL;
 237 #endif
 238
 239         data->ocd_cksum_types = obd_cksum_types_supported_client();
 240
 241         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 242                 /* flag mdc connection as lightweight, only used for test
 243                  * purpose, use with care */
 244                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 245
 246         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 247         data->ocd_version = LUSTRE_VERSION_CODE;
 248
 249         if (sb->s_flags & MS_RDONLY)
 250                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 251         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 252                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 253
 254 #ifdef MS_NOSEC
 255         /* Setting this indicates we correctly support S_NOSEC (See kernel
 256          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 257          */
 258         sb->s_flags |= MS_NOSEC;
 259 #endif
 260
 261         if (sbi->ll_flags & LL_SBI_FLOCK)
 262                 sbi->ll_fop = &ll_file_operations_flock;
 263         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 264                 sbi->ll_fop = &ll_file_operations;
 265         else
 266                 sbi->ll_fop = &ll_file_operations_noflock;
 267
 268         /* always ping even if server suppress_pings */
 269         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 270                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 271
 272 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
 273         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 274 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 275
 276         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 277
 278         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 279         if (err == -EBUSY) {
 280                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 281                                    "recovery, of which this client is not a "
 282                                    "part. Please wait for recovery to complete,"
 283                                    " abort, or time out.\n", md);
 284                 GOTO(out, err);
 285         } else if (err) {
 286                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 287                 GOTO(out, err);
 288         }
 289
 290         sbi->ll_md_exp->exp_connect_data = *data;
 291
 292         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 293                            LUSTRE_SEQ_METADATA);
 294         if (err) {
 295                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 296                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 297                 GOTO(out_md, err);
 298         }
 299
 300         /* For mount, we only need fs info from MDT0, and also in DNE, it
 301          * can make sure the client can be mounted as long as MDT0 is
 302          * avaible */
 303         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 304                         ktime_get_seconds() -OBD_STATFS_CACHE_SECONDS,
 305                         OBD_STATFS_FOR_MDT0);
 306         if (err)
 307                 GOTO(out_md_fid, err);
 308
 309         /* This needs to be after statfs to ensure connect has finished.
 310          * Note that "data" does NOT contain the valid connect reply.
 311          * If connecting to a 1.8 server there will be no LMV device, so
 312          * we can access the MDC export directly and exp_connect_flags will
 313          * be non-zero, but if accessing an upgraded 2.1 server it will
 314          * have the correct flags filled in.
 315          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 316         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 317         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 318             valid != CLIENT_CONNECT_MDT_REQD) {
 319                 char *buf;
 320
 321                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 322                 obd_connect_flags2str(buf, PAGE_SIZE,
 323                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 324                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 325                                    "feature(s) needed for correct operation "
 326                                    "of this client (%s). Please upgrade "
 327                                    "server or downgrade client.\n",
 328                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 329                 OBD_FREE(buf, PAGE_SIZE);
 330                 GOTO(out_md_fid, err = -EPROTO);
 331         }
 332
 333         size = sizeof(*data);
 334         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 335                            KEY_CONN_DATA,  &size, data);
 336         if (err) {
 337                 CERROR("%s: Get connect data failed: rc = %d\n",
 338                        sbi->ll_md_exp->exp_obd->obd_name, err);
 339                 GOTO(out_md_fid, err);
 340         }
 341
 342         LASSERT(osfs->os_bsize);
 343         sb->s_blocksize = osfs->os_bsize;
 344         sb->s_blocksize_bits = log2(osfs->os_bsize);
 345         sb->s_magic = LL_SUPER_MAGIC;
 346         sb->s_maxbytes = MAX_LFS_FILESIZE;
 347         sbi->ll_namelen = osfs->os_namelen;
 348         sbi->ll_mnt.mnt = current->fs->root.mnt;
 349
 350         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 351             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 352                 LCONSOLE_INFO("Disabling user_xattr feature because "
 353                               "it is not supported on the server\n");
 354                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 355         }
 356
 357         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 358 #ifdef MS_POSIXACL
 359                 sb->s_flags |= MS_POSIXACL;
 360 #endif
 361                 sbi->ll_flags |= LL_SBI_ACL;
 362         } else {
 363                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 364 #ifdef MS_POSIXACL
 365                 sb->s_flags &= ~MS_POSIXACL;
 366 #endif
 367                 sbi->ll_flags &= ~LL_SBI_ACL;
 368         }
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 371                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 372
 373         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 374                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 375
 376         if (obd_connect_has_secctx(data))
 377                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 378
 379         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 380                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 381                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 382                                       "unknown maximum xattr size.\n", dt);
 383                 } else if (!sbi->ll_xattr_cache_set) {
 384                         /* If xattr_cache is already set (no matter 0 or 1)
 385                          * during processing llog, it won't be enabled here. */
 386                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 387                         sbi->ll_xattr_cache_enabled = 1;
 388                 }
 389         }
 390
 391         obd = class_name2obd(dt);
 392         if (!obd) {
 393                 CERROR("DT %s: not setup or attached\n", dt);
 394                 GOTO(out_md_fid, err = -ENODEV);
 395         }
 396
 397         /* pass client page size via ocd_grant_blkbits, the server should report
 398          * back its backend blocksize for grant calculation purpose */
 399         data->ocd_grant_blkbits = PAGE_SHIFT;
 400
 401         /* indicate OST features supported by this client */
 402         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 403                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 404                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 405                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 406                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 407                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 408                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 409                                   OBD_CONNECT_EINPROGRESS |
 410                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 411                                   OBD_CONNECT_LAYOUTLOCK |
 412                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 413                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 414                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 415
 416 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 417  * can interoperate with an older version of lockahead which was released prior
 418  * to landing in master. This support will be dropped when 2.13 development
 419  * starts.  At the point, we should not just drop the connect flag (below), we
 420  * should also remove the support in the code.
 421  *
 422  * Removing it means a few things:
 423  * 1. Remove this section here
 424  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 425  * 3. Remove function exp_connect_lockahead_old
 426  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 427  * */
 428 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 429         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 430 #endif
 431
 432         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 433
 434         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 435                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 436
 437         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 438          * disabled by default, because it can still be enabled on the
 439          * fly via /proc. As a consequence, we still need to come to an
 440          * agreement on the supported algorithms at connect time */
 441         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 442
 443         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 444                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 445         else
 446                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 447
 448 #ifdef HAVE_LRU_RESIZE_SUPPORT
 449         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 450 #endif
 451         /* always ping even if server suppress_pings */
 452         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 453                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 454
 455         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 456                "ocd_grant: %d\n", data->ocd_connect_flags,
 457                data->ocd_version, data->ocd_grant);
 458
 459         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 460         obd->obd_upcall.onu_upcall = cl_ocd_update;
 461
 462         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 463
 464         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 465                           NULL);
 466         if (err == -EBUSY) {
 467                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 468                                    "recovery, of which this client is not a "
 469                                    "part.  Please wait for recovery to "
 470                                    "complete, abort, or time out.\n", dt);
 471                 GOTO(out_md, err);
 472         } else if (err) {
 473                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 474                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 475                 GOTO(out_md, err);
 476         }
 477
 478         sbi->ll_dt_exp->exp_connect_data = *data;
 479
 480         /* Don't change value if it was specified in the config log */
 481         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 482                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 483                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 484                               (data->ocd_brw_size >> PAGE_SHIFT));
 485
 486         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 487                            LUSTRE_SEQ_METADATA);
 488         if (err) {
 489                 CERROR("%s: Can't init data layer FID infrastructure, "
 490                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 491                 GOTO(out_dt, err);
 492         }
 493
 494         mutex_lock(&sbi->ll_lco.lco_lock);
 495         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 496         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 497         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 498         mutex_unlock(&sbi->ll_lco.lco_lock);
 499
 500         fid_zero(&sbi->ll_root_fid);
 501         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 502                            &sbi->ll_root_fid);
 503         if (err) {
 504                 CERROR("cannot mds_connect: rc = %d\n", err);
 505                 GOTO(out_lock_cn_cb, err);
 506         }
 507         if (!fid_is_sane(&sbi->ll_root_fid)) {
 508                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 509                        sbi->ll_md_exp->exp_obd->obd_name,
 510                        PFID(&sbi->ll_root_fid));
 511                 GOTO(out_lock_cn_cb, err = -EINVAL);
 512         }
 513         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 514
 515         sb->s_op = &lustre_super_operations;
 516 #ifdef HAVE_XATTR_HANDLER_FLAGS
 517         sb->s_xattr = ll_xattr_handlers;
 518 #endif
 519 #if THREAD_SIZE >= 8192 /*b=17630*/
 520         sb->s_export_op = &lustre_export_operations;
 521 #endif
 522
 523         /* make root inode
 524          * XXX: move this to after cbd setup? */
 525         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 526         if (sbi->ll_flags & LL_SBI_ACL)
 527                 valid |= OBD_MD_FLACL;
 528
 529         OBD_ALLOC_PTR(op_data);
 530         if (op_data == NULL)
 531                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 532
 533         op_data->op_fid1 = sbi->ll_root_fid;
 534         op_data->op_mode = 0;
 535         op_data->op_valid = valid;
 536
 537         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 538
 539         OBD_FREE_PTR(op_data);
 540         if (err) {
 541                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 542                        sbi->ll_md_exp->exp_obd->obd_name, err);
 543                 GOTO(out_lock_cn_cb, err);
 544         }
 545
 546         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 547                                sbi->ll_md_exp, &lmd);
 548         if (err) {
 549                 CERROR("failed to understand root inode md: rc = %d\n", err);
 550                 ptlrpc_req_finished(request);
 551                 GOTO(out_lock_cn_cb, err);
 552         }
 553
 554         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 555         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 556                                             sbi->ll_flags & LL_SBI_32BIT_API),
 557                        &lmd);
 558         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 559         ptlrpc_req_finished(request);
 560
 561         if (IS_ERR(root)) {
 562 #ifdef CONFIG_FS_POSIX_ACL
 563                 if (lmd.posix_acl) {
 564                         posix_acl_release(lmd.posix_acl);
 565                         lmd.posix_acl = NULL;
 566                 }
 567 #endif
 568                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 569                 root = NULL;
 570                 CERROR("lustre_lite: bad iget4 for root\n");
 571                 GOTO(out_root, err);
 572         }
 573
 574         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 575         if (sbi->ll_checksum_set) {
 576                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 577                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 578                                          sizeof(checksum), &checksum, NULL);
 579                 if (err) {
 580                         CERROR("%s: Set checksum failed: rc = %d\n",
 581                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 582                         GOTO(out_root, err);
 583                 }
 584         }
 585         cl_sb_init(sb);
 586
 587         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 588                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 589                                  sbi->ll_cache, NULL);
 590         if (err) {
 591                 CERROR("%s: Set cache_set failed: rc = %d\n",
 592                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 593                 GOTO(out_root, err);
 594         }
 595
 596         sb->s_root = d_make_root(root);
 597         if (sb->s_root == NULL) {
 598                 CERROR("%s: can't make root dentry\n",
 599                         ll_get_fsname(sb, NULL, 0));
 600                 GOTO(out_root, err = -ENOMEM);
 601         }
 602 #ifdef HAVE_DCACHE_LOCK
 603         sb->s_root->d_op = &ll_d_ops;
 604 #endif
 605
 606         sbi->ll_sdev_orig = sb->s_dev;
 607
 608         /* We set sb->s_dev equal on all lustre clients in order to support
 609          * NFS export clustering.  NFSD requires that the FSID be the same
 610          * on all clients. */
 611         /* s_dev is also used in lt_compare() to compare two fs, but that is
 612          * only a node-local comparison. */
 613         uuid = obd_get_uuid(sbi->ll_md_exp);
 614         if (uuid != NULL)
 615                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 616
 617         if (data != NULL)
 618                 OBD_FREE_PTR(data);
 619         if (osfs != NULL)
 620                 OBD_FREE_PTR(osfs);
 621
 622         if (sbi->ll_proc_root != NULL) {
 623                 err = lprocfs_ll_register_obd(sb, dt);
 624                 if (err < 0) {
 625                         CERROR("%s: could not register %s in llite: rc = %d\n",
 626                                dt, ll_get_fsname(sb, NULL, 0), err);
 627                         err = 0;
 628                 }
 629                 err = lprocfs_ll_register_obd(sb, md);
 630                 if (err < 0) {
 631                         CERROR("%s: could not register %s in llite: rc = %d\n",
 632                                md, ll_get_fsname(sb, NULL, 0), err);
 633                         err = 0;
 634                 }
 635         }
 636
 637         RETURN(err);
 638 out_root:
 639         if (root)
 640                 iput(root);
 641 out_lock_cn_cb:
 642         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 643 out_dt:
 644         obd_disconnect(sbi->ll_dt_exp);
 645         sbi->ll_dt_exp = NULL;
 646 out_md_fid:
 647         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 648 out_md:
 649         obd_disconnect(sbi->ll_md_exp);
 650         sbi->ll_md_exp = NULL;
 651 out:
 652         if (data != NULL)
 653                 OBD_FREE_PTR(data);
 654         if (osfs != NULL)
 655                 OBD_FREE_PTR(osfs);
 656         return err;
 657 }
 658
 659 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 660 {
 661         int size, rc;
 662
 663         size = sizeof(*lmmsize);
 664         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 665                           KEY_MAX_EASIZE, &size, lmmsize);
 666         if (rc != 0) {
 667                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 668                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 669                 RETURN(rc);
 670         }
 671
 672         size = sizeof(int);
 673         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 674                           KEY_MAX_EASIZE, &size, lmmsize);
 675         if (rc)
 676                 CERROR("Get max mdsize error rc %d\n", rc);
 677
 678         RETURN(rc);
 679 }
 680
 681 /**
 682  * Get the value of the default_easize parameter.
 683  *
 684  * \see client_obd::cl_default_mds_easize
 685  *
 686  * \param[in] sbi       superblock info for this filesystem
 687  * \param[out] lmmsize  pointer to storage location for value
 688  *
 689  * \retval 0            on success
 690  * \retval negative     negated errno on failure
 691  */
 692 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 693 {
 694         int size, rc;
 695
 696         size = sizeof(int);
 697         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 698                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 699         if (rc)
 700                 CERROR("Get default mdsize error rc %d\n", rc);
 701
 702         RETURN(rc);
 703 }
 704
 705 /**
 706  * Set the default_easize parameter to the given value.
 707  *
 708  * \see client_obd::cl_default_mds_easize
 709  *
 710  * \param[in] sbi       superblock info for this filesystem
 711  * \param[in] lmmsize   the size to set
 712  *
 713  * \retval 0            on success
 714  * \retval negative     negated errno on failure
 715  */
 716 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 717 {
 718         int rc;
 719
 720         if (lmmsize < sizeof(struct lov_mds_md) ||
 721             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 722                 return -EINVAL;
 723
 724         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 725                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 726                                 sizeof(int), &lmmsize, NULL);
 727
 728         RETURN(rc);
 729 }
 730
 731 static void client_common_put_super(struct super_block *sb)
 732 {
 733         struct ll_sb_info *sbi = ll_s2sbi(sb);
 734         ENTRY;
 735
 736         cl_sb_fini(sb);
 737
 738         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 739         obd_disconnect(sbi->ll_dt_exp);
 740         sbi->ll_dt_exp = NULL;
 741
 742         ll_debugfs_unregister_super(sb);
 743
 744         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 745         obd_disconnect(sbi->ll_md_exp);
 746         sbi->ll_md_exp = NULL;
 747
 748         EXIT;
 749 }
 750
 751 void ll_kill_super(struct super_block *sb)
 752 {
 753         struct ll_sb_info *sbi;
 754         ENTRY;
 755
 756         /* not init sb ?*/
 757         if (!(sb->s_flags & MS_ACTIVE))
 758                 return;
 759
 760         sbi = ll_s2sbi(sb);
 761         /* we need restore s_dev from changed for clustred NFS before put_super
 762          * because new kernels have cached s_dev and change sb->s_dev in
 763          * put_super not affected real removing devices */
 764         if (sbi) {
 765                 sb->s_dev = sbi->ll_sdev_orig;
 766                 sbi->ll_umounting = 1;
 767
 768                 /* wait running statahead threads to quit */
 769                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 770                         set_current_state(TASK_UNINTERRUPTIBLE);
 771                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 772                 }
 773         }
 774
 775         EXIT;
 776 }
 777
 778 static inline int ll_set_opt(const char *opt, char *data, int fl)
 779 {
 780         if (strncmp(opt, data, strlen(opt)) != 0)
 781                 return 0;
 782         else
 783                 return fl;
 784 }
 785
 786 /* non-client-specific mount options are parsed in lmd_parse */
 787 static int ll_options(char *options, struct ll_sb_info *sbi)
 788 {
 789         int tmp;
 790         char *s1 = options, *s2;
 791         int *flags = &sbi->ll_flags;
 792         ENTRY;
 793
 794         if (!options)
 795                 RETURN(0);
 796
 797         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 798
 799         while (*s1) {
 800                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 801                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 827                 if (tmp) {
 828                         *flags &= ~tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("context", s1, 1);
 832                 if (tmp)
 833                         goto next;
 834                 tmp = ll_set_opt("fscontext", s1, 1);
 835                 if (tmp)
 836                         goto next;
 837                 tmp = ll_set_opt("defcontext", s1, 1);
 838                 if (tmp)
 839                         goto next;
 840                 tmp = ll_set_opt("rootcontext", s1, 1);
 841                 if (tmp)
 842                         goto next;
 843                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 844                 if (tmp) {
 845                         *flags |= tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 849                 if (tmp) {
 850                         *flags &= ~tmp;
 851                         goto next;
 852                 }
 853
 854                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 855                 if (tmp) {
 856                         *flags |= tmp;
 857                         sbi->ll_checksum_set = 1;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         sbi->ll_checksum_set = 1;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 877                 if (tmp) {
 878                         *flags |= tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 882                 if (tmp) {
 883                         *flags &= ~tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 887                 if (tmp) {
 888                         *flags |= tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 897                 if (tmp) {
 898                         *flags &= ~tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 907                                    s1);
 908                 RETURN(-EINVAL);
 909
 910 next:
 911                 /* Find next opt */
 912                 s2 = strchr(s1, ',');
 913                 if (s2 == NULL)
 914                         break;
 915                 s1 = s2 + 1;
 916         }
 917         RETURN(0);
 918 }
 919
 920 void ll_lli_init(struct ll_inode_info *lli)
 921 {
 922         lli->lli_inode_magic = LLI_INODE_MAGIC;
 923         lli->lli_flags = 0;
 924         spin_lock_init(&lli->lli_lock);
 925         lli->lli_posix_acl = NULL;
 926         /* Do not set lli_fid, it has been initialized already. */
 927         fid_zero(&lli->lli_pfid);
 928         lli->lli_mds_read_och = NULL;
 929         lli->lli_mds_write_och = NULL;
 930         lli->lli_mds_exec_och = NULL;
 931         lli->lli_open_fd_read_count = 0;
 932         lli->lli_open_fd_write_count = 0;
 933         lli->lli_open_fd_exec_count = 0;
 934         mutex_init(&lli->lli_och_mutex);
 935         spin_lock_init(&lli->lli_agl_lock);
 936         spin_lock_init(&lli->lli_layout_lock);
 937         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 938         lli->lli_clob = NULL;
 939
 940         init_rwsem(&lli->lli_xattrs_list_rwsem);
 941         mutex_init(&lli->lli_xattrs_enq_lock);
 942
 943         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 944         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 945                 mutex_init(&lli->lli_readdir_mutex);
 946                 lli->lli_opendir_key = NULL;
 947                 lli->lli_sai = NULL;
 948                 spin_lock_init(&lli->lli_sa_lock);
 949                 lli->lli_opendir_pid = 0;
 950                 lli->lli_sa_enabled = 0;
 951                 lli->lli_def_stripe_offset = -1;
 952         } else {
 953                 mutex_init(&lli->lli_size_mutex);
 954                 lli->lli_symlink_name = NULL;
 955                 init_rwsem(&lli->lli_trunc_sem);
 956                 range_lock_tree_init(&lli->lli_write_tree);
 957                 init_rwsem(&lli->lli_glimpse_sem);
 958                 lli->lli_glimpse_time = ktime_set(0, 0);
 959                 INIT_LIST_HEAD(&lli->lli_agl_list);
 960                 lli->lli_agl_index = 0;
 961                 lli->lli_async_rc = 0;
 962         }
 963         mutex_init(&lli->lli_layout_mutex);
 964         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 965 }
 966
 967 #define MAX_STRING_SIZE 128
 968
 969 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 970
 971 #define LSI_BDI_INITIALIZED     0x00400000
 972
 973 #ifndef HAVE_BDI_CAP_MAP_COPY
 974 # define BDI_CAP_MAP_COPY       0
 975 #endif
 976
 977 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 978 {
 979         struct  lustre_sb_info *lsi = s2lsi(sb);
 980         char buf[MAX_STRING_SIZE];
 981         va_list args;
 982         int err;
 983
 984         err = bdi_init(&lsi->lsi_bdi);
 985         if (err)
 986                 return err;
 987
 988         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 989         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 990         lsi->lsi_bdi.name = "lustre";
 991         va_start(args, fmt);
 992         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
 993         va_end(args);
 994         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
 995         va_end(args);
 996         if (!err)
 997                 sb->s_bdi = &lsi->lsi_bdi;
 998
 999         return err;
1000 }
1001 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1002
1003 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1004 {
1005         struct  lustre_profile *lprof = NULL;
1006         struct  lustre_sb_info *lsi = s2lsi(sb);
1007         struct  ll_sb_info *sbi = NULL;
1008         char    *dt = NULL, *md = NULL;
1009         char    *profilenm = get_profile_name(sb);
1010         struct config_llog_instance *cfg;
1011         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1012         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1013         char name[MAX_STRING_SIZE];
1014         int md_len = 0;
1015         int dt_len = 0;
1016         char *ptr;
1017         int len;
1018         int err;
1019
1020         ENTRY;
1021         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1022
1023         try_module_get(THIS_MODULE);
1024
1025         OBD_ALLOC_PTR(cfg);
1026         if (cfg == NULL)
1027                 GOTO(out_free, err = -ENOMEM);
1028
1029         /* client additional sb info */
1030         lsi->lsi_llsbi = sbi = ll_init_sbi();
1031         if (!sbi)
1032                 GOTO(out_free, err = -ENOMEM);
1033
1034         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1035         if (err)
1036                 GOTO(out_free, err);
1037
1038         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1039         if (err)
1040                 GOTO(out_free, err);
1041
1042 #ifndef HAVE_DCACHE_LOCK
1043         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1044         sb->s_d_op = &ll_d_ops;
1045 #endif
1046         /* Get fsname */
1047         len = strlen(lsi->lsi_lmd->lmd_profile);
1048         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1049         if (ptr && (strcmp(ptr, "-client") == 0))
1050                 len -= 7;
1051
1052         /* Mount info */
1053         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1054                  lsi->lsi_lmd->lmd_profile, sb);
1055
1056         /* Call ll_debugfs_register_super() before lustre_process_log()
1057          * so that "llite.*.*" params can be processed correctly.
1058          */
1059         err = ll_debugfs_register_super(sb, name);
1060         if (err < 0) {
1061                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1062                        ll_get_fsname(sb, NULL, 0), err);
1063                 err = 0;
1064         }
1065
1066         /* Generate a string unique to this super, in case some joker tries
1067            to mount the same fs at two mount points.
1068            Use the address of the super itself.*/
1069         cfg->cfg_instance = sb;
1070         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1071         cfg->cfg_callback = class_config_llog_handler;
1072         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1073         /* set up client obds */
1074         err = lustre_process_log(sb, profilenm, cfg);
1075         if (err < 0)
1076                 GOTO(out_proc, err);
1077
1078         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1079         lprof = class_get_profile(profilenm);
1080         if (lprof == NULL) {
1081                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1082                                    " read from the MGS.  Does that filesystem "
1083                                    "exist?\n", profilenm);
1084                 GOTO(out_proc, err = -EINVAL);
1085         }
1086         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1087                lprof->lp_md, lprof->lp_dt);
1088
1089         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1090         OBD_ALLOC(dt, dt_len);
1091         if (!dt)
1092                 GOTO(out_proc, err = -ENOMEM);
1093         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1094
1095         md_len = strlen(lprof->lp_md) + instlen + 2;
1096         OBD_ALLOC(md, md_len);
1097         if (!md)
1098                 GOTO(out_proc, err = -ENOMEM);
1099         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1100
1101         /* connections, registrations, sb setup */
1102         err = client_common_fill_super(sb, md, dt, mnt);
1103         if (err < 0)
1104                 GOTO(out_proc, err);
1105
1106         sbi->ll_client_common_fill_super_succeeded = 1;
1107
1108 out_proc:
1109         if (err < 0)
1110                 ll_debugfs_unregister_super(sb);
1111 out_free:
1112         if (md)
1113                 OBD_FREE(md, md_len);
1114         if (dt)
1115                 OBD_FREE(dt, dt_len);
1116         if (lprof != NULL)
1117                 class_put_profile(lprof);
1118         if (cfg)
1119                 OBD_FREE_PTR(cfg);
1120         if (err)
1121                 ll_put_super(sb);
1122         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1123                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1124         RETURN(err);
1125 } /* ll_fill_super */
1126
1127 void ll_put_super(struct super_block *sb)
1128 {
1129         struct config_llog_instance cfg, params_cfg;
1130         struct obd_device *obd;
1131         struct lustre_sb_info *lsi = s2lsi(sb);
1132         struct ll_sb_info *sbi = ll_s2sbi(sb);
1133         char *profilenm = get_profile_name(sb);
1134         long ccc_count;
1135         int next, force = 1, rc = 0;
1136         ENTRY;
1137
1138         if (!sbi)
1139                 GOTO(out_no_sbi, 0);
1140
1141         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1142
1143         cfg.cfg_instance = sb;
1144         lustre_end_log(sb, profilenm, &cfg);
1145
1146         params_cfg.cfg_instance = sb;
1147         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1148
1149         if (sbi->ll_md_exp) {
1150                 obd = class_exp2obd(sbi->ll_md_exp);
1151                 if (obd)
1152                         force = obd->obd_force;
1153         }
1154
1155         /* Wait for unstable pages to be committed to stable storage */
1156         if (force == 0) {
1157                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1158                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1159                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1160                         &lwi);
1161         }
1162
1163         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1164         if (force == 0 && rc != -EINTR)
1165                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1166
1167
1168         /* We need to set force before the lov_disconnect in
1169            lustre_common_put_super, since l_d cleans up osc's as well. */
1170         if (force) {
1171                 next = 0;
1172                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1173                                                      &next)) != NULL) {
1174                         obd->obd_force = force;
1175                 }
1176         }
1177
1178         if (sbi->ll_client_common_fill_super_succeeded) {
1179                 /* Only if client_common_fill_super succeeded */
1180                 client_common_put_super(sb);
1181         }
1182
1183         next = 0;
1184         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1185                 class_manual_cleanup(obd);
1186         }
1187
1188         if (sbi->ll_flags & LL_SBI_VERBOSE)
1189                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1190
1191         if (profilenm)
1192                 class_del_profile(profilenm);
1193
1194 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1195         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1196                 bdi_destroy(&lsi->lsi_bdi);
1197                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1198         }
1199 #endif
1200
1201         ll_free_sbi(sb);
1202         lsi->lsi_llsbi = NULL;
1203 out_no_sbi:
1204         lustre_common_put_super(sb);
1205
1206         cl_env_cache_purge(~0);
1207
1208         module_put(THIS_MODULE);
1209
1210         EXIT;
1211 } /* client_put_super */
1212
1213 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1214 {
1215         struct inode *inode = NULL;
1216
1217         /* NOTE: we depend on atomic igrab() -bzzz */
1218         lock_res_and_lock(lock);
1219         if (lock->l_resource->lr_lvb_inode) {
1220                 struct ll_inode_info * lli;
1221                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1222                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1223                         inode = igrab(lock->l_resource->lr_lvb_inode);
1224                 } else {
1225                         inode = lock->l_resource->lr_lvb_inode;
1226                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1227                                          D_WARNING, lock, "lr_lvb_inode %p is "
1228                                          "bogus: magic %08x",
1229                                          lock->l_resource->lr_lvb_inode,
1230                                          lli->lli_inode_magic);
1231                         inode = NULL;
1232                 }
1233         }
1234         unlock_res_and_lock(lock);
1235         return inode;
1236 }
1237
1238 void ll_dir_clear_lsm_md(struct inode *inode)
1239 {
1240         struct ll_inode_info *lli = ll_i2info(inode);
1241
1242         LASSERT(S_ISDIR(inode->i_mode));
1243
1244         if (lli->lli_lsm_md != NULL) {
1245                 lmv_free_memmd(lli->lli_lsm_md);
1246                 lli->lli_lsm_md = NULL;
1247         }
1248 }
1249
1250 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1251                                       const struct lu_fid *fid,
1252                                       struct lustre_md *md)
1253 {
1254         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1255         struct mdt_body         *body = md->body;
1256         struct inode            *inode;
1257         ino_t                   ino;
1258         ENTRY;
1259
1260         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1261         inode = iget_locked(sb, ino);
1262         if (inode == NULL) {
1263                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1264                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1265                 RETURN(ERR_PTR(-ENOENT));
1266         }
1267
1268         if (inode->i_state & I_NEW) {
1269                 struct ll_inode_info *lli = ll_i2info(inode);
1270                 struct lmv_stripe_md *lsm = md->lmv;
1271
1272                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1273                                 (body->mbo_mode & S_IFMT);
1274                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1275                          PFID(fid));
1276
1277                 LTIME_S(inode->i_mtime) = 0;
1278                 LTIME_S(inode->i_atime) = 0;
1279                 LTIME_S(inode->i_ctime) = 0;
1280                 inode->i_rdev = 0;
1281
1282 #ifdef HAVE_BACKING_DEV_INFO
1283                 /* initializing backing dev info. */
1284                 inode->i_mapping->backing_dev_info =
1285                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1286 #endif
1287                 inode->i_op = &ll_dir_inode_operations;
1288                 inode->i_fop = &ll_dir_operations;
1289                 lli->lli_fid = *fid;
1290                 ll_lli_init(lli);
1291
1292                 LASSERT(lsm != NULL);
1293                 /* master object FID */
1294                 lli->lli_pfid = body->mbo_fid1;
1295                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1296                        lli, PFID(fid), PFID(&lli->lli_pfid));
1297                 unlock_new_inode(inode);
1298         }
1299
1300         RETURN(inode);
1301 }
1302
1303 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1304 {
1305         struct lu_fid *fid;
1306         struct lmv_stripe_md *lsm = md->lmv;
1307         int i;
1308
1309         LASSERT(lsm != NULL);
1310         /* XXX sigh, this lsm_root initialization should be in
1311          * LMV layer, but it needs ll_iget right now, so we
1312          * put this here right now. */
1313         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1314                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1315                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1316                 /* Unfortunately ll_iget will call ll_update_inode,
1317                  * where the initialization of slave inode is slightly
1318                  * different, so it reset lsm_md to NULL to avoid
1319                  * initializing lsm for slave inode. */
1320                 /* For migrating inode, master stripe and master object will
1321                  * be same, so we only need assign this inode */
1322                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1323                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1324                 else
1325                         lsm->lsm_md_oinfo[i].lmo_root =
1326                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1327
1328                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1329                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1330
1331                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1332                         return rc;
1333                 }
1334         }
1335
1336         return 0;
1337 }
1338
1339 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1340                                 const struct lmv_stripe_md *lsm_md2)
1341 {
1342         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1343                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1344                lsm_md1->lsm_md_master_mdt_index ==
1345                                         lsm_md2->lsm_md_master_mdt_index &&
1346                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1347                lsm_md1->lsm_md_layout_version ==
1348                                         lsm_md2->lsm_md_layout_version &&
1349                strcmp(lsm_md1->lsm_md_pool_name,
1350                       lsm_md2->lsm_md_pool_name) == 0;
1351 }
1352
1353 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1354 {
1355         struct ll_inode_info *lli = ll_i2info(inode);
1356         struct lmv_stripe_md *lsm = md->lmv;
1357         int     rc;
1358         ENTRY;
1359
1360         LASSERT(S_ISDIR(inode->i_mode));
1361         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1362                PFID(ll_inode2fid(inode)));
1363
1364         /* no striped information from request. */
1365         if (lsm == NULL) {
1366                 if (lli->lli_lsm_md == NULL) {
1367                         RETURN(0);
1368                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1369                                                 LMV_HASH_FLAG_MIGRATION) {
1370                         /* migration is done, the temporay MIGRATE layout has
1371                          * been removed */
1372                         CDEBUG(D_INODE, DFID" finish migration.\n",
1373                                PFID(ll_inode2fid(inode)));
1374                         lmv_free_memmd(lli->lli_lsm_md);
1375                         lli->lli_lsm_md = NULL;
1376                         RETURN(0);
1377                 } else {
1378                         /* The lustre_md from req does not include stripeEA,
1379                          * see ll_md_setattr */
1380                         RETURN(0);
1381                 }
1382         }
1383
1384         /* set the directory layout */
1385         if (lli->lli_lsm_md == NULL) {
1386                 struct cl_attr  *attr;
1387
1388                 rc = ll_init_lsm_md(inode, md);
1389                 if (rc != 0)
1390                         RETURN(rc);
1391
1392                 /* set md->lmv to NULL, so the following free lustre_md
1393                  * will not free this lsm */
1394                 md->lmv = NULL;
1395                 lli->lli_lsm_md = lsm;
1396
1397                 OBD_ALLOC_PTR(attr);
1398                 if (attr == NULL)
1399                         RETURN(-ENOMEM);
1400
1401                 /* validate the lsm */
1402                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1403                                    ll_md_blocking_ast);
1404                 if (rc != 0) {
1405                         OBD_FREE_PTR(attr);
1406                         RETURN(rc);
1407                 }
1408
1409                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1410                         md->body->mbo_nlink = attr->cat_nlink;
1411                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1412                         md->body->mbo_size = attr->cat_size;
1413                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1414                         md->body->mbo_atime = attr->cat_atime;
1415                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1416                         md->body->mbo_ctime = attr->cat_ctime;
1417                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1418                         md->body->mbo_mtime = attr->cat_mtime;
1419
1420                 OBD_FREE_PTR(attr);
1421
1422                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1423                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1424                 RETURN(0);
1425         }
1426
1427         /* Compare the old and new stripe information */
1428         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1429                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1430                 int                     idx;
1431
1432                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1433                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1434                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1435                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1436                        inode, lsm, old_lsm,
1437                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1438                        lsm->lsm_md_stripe_count,
1439                        old_lsm->lsm_md_stripe_count,
1440                        lsm->lsm_md_master_mdt_index,
1441                        old_lsm->lsm_md_master_mdt_index,
1442                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1443                        lsm->lsm_md_layout_version,
1444                        old_lsm->lsm_md_layout_version,
1445                        lsm->lsm_md_pool_name,
1446                        old_lsm->lsm_md_pool_name);
1447
1448                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1449                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1450                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1451                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1452                 }
1453
1454                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1455                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1456                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1457                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1458                 }
1459
1460                 RETURN(-EIO);
1461         }
1462
1463         RETURN(0);
1464 }
1465
1466 void ll_clear_inode(struct inode *inode)
1467 {
1468         struct ll_inode_info *lli = ll_i2info(inode);
1469         struct ll_sb_info *sbi = ll_i2sbi(inode);
1470         ENTRY;
1471
1472         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1473                PFID(ll_inode2fid(inode)), inode);
1474
1475         if (S_ISDIR(inode->i_mode)) {
1476                 /* these should have been cleared in ll_file_release */
1477                 LASSERT(lli->lli_opendir_key == NULL);
1478                 LASSERT(lli->lli_sai == NULL);
1479                 LASSERT(lli->lli_opendir_pid == 0);
1480         }
1481
1482         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1483
1484         LASSERT(!lli->lli_open_fd_write_count);
1485         LASSERT(!lli->lli_open_fd_read_count);
1486         LASSERT(!lli->lli_open_fd_exec_count);
1487
1488         if (lli->lli_mds_write_och)
1489                 ll_md_real_close(inode, FMODE_WRITE);
1490         if (lli->lli_mds_exec_och)
1491                 ll_md_real_close(inode, FMODE_EXEC);
1492         if (lli->lli_mds_read_och)
1493                 ll_md_real_close(inode, FMODE_READ);
1494
1495         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1496                 OBD_FREE(lli->lli_symlink_name,
1497                          strlen(lli->lli_symlink_name) + 1);
1498                 lli->lli_symlink_name = NULL;
1499         }
1500
1501         ll_xattr_cache_destroy(inode);
1502
1503 #ifdef CONFIG_FS_POSIX_ACL
1504         forget_all_cached_acls(inode);
1505         if (lli->lli_posix_acl) {
1506                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1507                 posix_acl_release(lli->lli_posix_acl);
1508                 lli->lli_posix_acl = NULL;
1509         }
1510 #endif
1511         lli->lli_inode_magic = LLI_INODE_DEAD;
1512
1513         if (S_ISDIR(inode->i_mode))
1514                 ll_dir_clear_lsm_md(inode);
1515         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1516                 LASSERT(list_empty(&lli->lli_agl_list));
1517
1518         /*
1519          * XXX This has to be done before lsm is freed below, because
1520          * cl_object still uses inode lsm.
1521          */
1522         cl_inode_fini(inode);
1523
1524         EXIT;
1525 }
1526
1527 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1528 {
1529         struct lustre_md md;
1530         struct inode *inode = dentry->d_inode;
1531         struct ll_sb_info *sbi = ll_i2sbi(inode);
1532         struct ptlrpc_request *request = NULL;
1533         int rc, ia_valid;
1534         ENTRY;
1535
1536         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1537                                      LUSTRE_OPC_ANY, NULL);
1538         if (IS_ERR(op_data))
1539                 RETURN(PTR_ERR(op_data));
1540
1541         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1542         if (rc) {
1543                 ptlrpc_req_finished(request);
1544                 if (rc == -ENOENT) {
1545                         clear_nlink(inode);
1546                         /* Unlinked special device node? Or just a race?
1547                          * Pretend we done everything. */
1548                         if (!S_ISREG(inode->i_mode) &&
1549                             !S_ISDIR(inode->i_mode)) {
1550                                 ia_valid = op_data->op_attr.ia_valid;
1551                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1552                                 rc = simple_setattr(dentry, &op_data->op_attr);
1553                                 op_data->op_attr.ia_valid = ia_valid;
1554                         }
1555                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1556                         CERROR("md_setattr fails: rc = %d\n", rc);
1557                 }
1558                 RETURN(rc);
1559         }
1560
1561         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1562                               sbi->ll_md_exp, &md);
1563         if (rc) {
1564                 ptlrpc_req_finished(request);
1565                 RETURN(rc);
1566         }
1567
1568         ia_valid = op_data->op_attr.ia_valid;
1569         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1570          * cache is not cleared yet. */
1571         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1572         if (S_ISREG(inode->i_mode))
1573                 inode_lock(inode);
1574         rc = simple_setattr(dentry, &op_data->op_attr);
1575         if (S_ISREG(inode->i_mode))
1576                 inode_unlock(inode);
1577         op_data->op_attr.ia_valid = ia_valid;
1578
1579         rc = ll_update_inode(inode, &md);
1580         ptlrpc_req_finished(request);
1581
1582         RETURN(rc);
1583 }
1584
1585 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1586  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1587  * keep these values until such a time that objects are allocated for it.
1588  * We do the MDS operations first, as it is checking permissions for us.
1589  * We don't to the MDS RPC if there is nothing that we want to store there,
1590  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1591  * going to do an RPC anyways.
1592  *
1593  * If we are doing a truncate, we will send the mtime and ctime updates
1594  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1595  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1596  * at the same time.
1597  *
1598  * In case of HSMimport, we only set attr on MDS.
1599  */
1600 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1601 {
1602         struct inode *inode = dentry->d_inode;
1603         struct ll_inode_info *lli = ll_i2info(inode);
1604         struct md_op_data *op_data = NULL;
1605         int rc = 0;
1606         ENTRY;
1607
1608         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1609                "valid %x, hsm_import %d\n",
1610                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1611                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1612                hsm_import);
1613
1614         if (attr->ia_valid & ATTR_SIZE) {
1615                 /* Check new size against VFS/VM file size limit and rlimit */
1616                 rc = inode_newsize_ok(inode, attr->ia_size);
1617                 if (rc)
1618                         RETURN(rc);
1619
1620                 /* The maximum Lustre file size is variable, based on the
1621                  * OST maximum object size and number of stripes.  This
1622                  * needs another check in addition to the VFS check above. */
1623                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1624                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1625                                PFID(&lli->lli_fid), attr->ia_size,
1626                                ll_file_maxbytes(inode));
1627                         RETURN(-EFBIG);
1628                 }
1629
1630                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1631         }
1632
1633         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1634         if (attr->ia_valid & TIMES_SET_FLAGS) {
1635                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1636                     !cfs_capable(CFS_CAP_FOWNER))
1637                         RETURN(-EPERM);
1638         }
1639
1640         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1641         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1642             (attr->ia_valid & ATTR_CTIME)) {
1643                 attr->ia_ctime = current_time(inode);
1644                 attr->ia_valid |= ATTR_CTIME_SET;
1645         }
1646         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1647             (attr->ia_valid & ATTR_ATIME)) {
1648                 attr->ia_atime = current_time(inode);
1649                 attr->ia_valid |= ATTR_ATIME_SET;
1650         }
1651         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1652             (attr->ia_valid & ATTR_MTIME)) {
1653                 attr->ia_mtime = current_time(inode);
1654                 attr->ia_valid |= ATTR_MTIME_SET;
1655         }
1656
1657         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1658                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1659                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1660                        (s64)ktime_get_real_seconds());
1661
1662         if (S_ISREG(inode->i_mode)) {
1663                 if (attr->ia_valid & ATTR_SIZE)
1664                         inode_dio_write_done(inode);
1665                 inode_unlock(inode);
1666         }
1667
1668         /* We always do an MDS RPC, even if we're only changing the size;
1669          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1670
1671         OBD_ALLOC_PTR(op_data);
1672         if (op_data == NULL)
1673                 GOTO(out, rc = -ENOMEM);
1674
1675         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1676                 /* If we are changing file size, file content is
1677                  * modified, flag it. */
1678                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1679                 op_data->op_bias |= MDS_DATA_MODIFIED;
1680                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1681         }
1682
1683         op_data->op_attr = *attr;
1684
1685         rc = ll_md_setattr(dentry, op_data);
1686         if (rc)
1687                 GOTO(out, rc);
1688
1689         if (!S_ISREG(inode->i_mode) || hsm_import)
1690                 GOTO(out, rc = 0);
1691
1692         if (attr->ia_valid & (ATTR_SIZE |
1693                               ATTR_ATIME | ATTR_ATIME_SET |
1694                               ATTR_MTIME | ATTR_MTIME_SET |
1695                               ATTR_CTIME | ATTR_CTIME_SET)) {
1696                 /* For truncate and utimes sending attributes to OSTs, setting
1697                  * mtime/atime to the past will be performed under PW [0:EOF]
1698                  * extent lock (new_size:EOF for truncate).  It may seem
1699                  * excessive to send mtime/atime updates to OSTs when not
1700                  * setting times to past, but it is necessary due to possible
1701                  * time de-synchronization between MDT inode and OST objects */
1702                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1703         }
1704
1705         /* If the file was restored, it needs to set dirty flag.
1706          *
1707          * We've already sent MDS_DATA_MODIFIED flag in
1708          * ll_md_setattr() for truncate. However, the MDT refuses to
1709          * set the HS_DIRTY flag on released files, so we have to set
1710          * it again if the file has been restored. Please check how
1711          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1712          *
1713          * Please notice that if the file is not released, the previous
1714          * MDS_DATA_MODIFIED has taken effect and usually
1715          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1716          * This way we can save an RPC for common open + trunc
1717          * operation. */
1718         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1719                 struct hsm_state_set hss = {
1720                         .hss_valid = HSS_SETMASK,
1721                         .hss_setmask = HS_DIRTY,
1722                 };
1723                 int rc2;
1724
1725                 rc2 = ll_hsm_state_set(inode, &hss);
1726                 /* truncate and write can happen at the same time, so that
1727                  * the file can be set modified even though the file is not
1728                  * restored from released state, and ll_hsm_state_set() is
1729                  * not applicable for the file, and rc2 < 0 is normal in this
1730                  * case. */
1731                 if (rc2 < 0)
1732                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1733                                PFID(ll_inode2fid(inode)), rc2);
1734         }
1735
1736         EXIT;
1737 out:
1738         if (op_data != NULL)
1739                 ll_finish_md_op_data(op_data);
1740
1741         if (S_ISREG(inode->i_mode)) {
1742                 inode_lock(inode);
1743                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1744                         inode_dio_wait(inode);
1745                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1746                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1747                  * inode flags, so there is a gap where S_NOSEC is not set.
1748                  * This can cause a writer to take the i_mutex unnecessarily,
1749                  * but this is safe to do and should be rare. */
1750                 inode_has_no_xattr(inode);
1751         }
1752
1753         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1754                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1755
1756         return rc;
1757 }
1758
1759 int ll_setattr(struct dentry *de, struct iattr *attr)
1760 {
1761         int mode = de->d_inode->i_mode;
1762
1763         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1764                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1765                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1766
1767         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1768                                (ATTR_SIZE|ATTR_MODE)) &&
1769             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1770              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1771               !(attr->ia_mode & S_ISGID))))
1772                 attr->ia_valid |= ATTR_FORCE;
1773
1774         if ((attr->ia_valid & ATTR_MODE) &&
1775             (mode & S_ISUID) &&
1776             !(attr->ia_mode & S_ISUID) &&
1777             !(attr->ia_valid & ATTR_KILL_SUID))
1778                 attr->ia_valid |= ATTR_KILL_SUID;
1779
1780         if ((attr->ia_valid & ATTR_MODE) &&
1781             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1782             !(attr->ia_mode & S_ISGID) &&
1783             !(attr->ia_valid & ATTR_KILL_SGID))
1784                 attr->ia_valid |= ATTR_KILL_SGID;
1785
1786         /* avoid polluted from ATTR_TIMES_SET,
1787          * projid is not expected to be set here */
1788         attr->ia_valid &= ~MDS_ATTR_PROJID;
1789
1790         return ll_setattr_raw(de, attr, false);
1791 }
1792
1793 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1794                        u32 flags)
1795 {
1796         struct obd_statfs obd_osfs;
1797         time64_t max_age;
1798         int rc;
1799
1800         ENTRY;
1801         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1802
1803         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1804         if (rc) {
1805                 CERROR("md_statfs fails: rc = %d\n", rc);
1806                 RETURN(rc);
1807         }
1808
1809         osfs->os_type = LL_SUPER_MAGIC;
1810
1811         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1812                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1813
1814         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1815                 flags |= OBD_STATFS_NODELAY;
1816
1817         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1818         if (rc) {
1819                 CERROR("obd_statfs fails: rc = %d\n", rc);
1820                 RETURN(rc);
1821         }
1822
1823         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1824                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1825                obd_osfs.os_files);
1826
1827         osfs->os_bsize = obd_osfs.os_bsize;
1828         osfs->os_blocks = obd_osfs.os_blocks;
1829         osfs->os_bfree = obd_osfs.os_bfree;
1830         osfs->os_bavail = obd_osfs.os_bavail;
1831
1832         /* If we don't have as many objects free on the OST as inodes
1833          * on the MDS, we reduce the total number of inodes to
1834          * compensate, so that the "inodes in use" number is correct.
1835          */
1836         if (obd_osfs.os_ffree < osfs->os_ffree) {
1837                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1838                         obd_osfs.os_ffree;
1839                 osfs->os_ffree = obd_osfs.os_ffree;
1840         }
1841
1842         RETURN(rc);
1843 }
1844 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1845 {
1846         struct super_block *sb = de->d_sb;
1847         struct obd_statfs osfs;
1848         __u64 fsid = huge_encode_dev(sb->s_dev);
1849         int rc;
1850
1851         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1852         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1853
1854         /* Some amount of caching on the client is allowed */
1855         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, 0);
1856         if (rc)
1857                 return rc;
1858
1859         statfs_unpack(sfs, &osfs);
1860
1861         /* We need to downshift for all 32-bit kernels, because we can't
1862          * tell if the kernel is being called via sys_statfs64() or not.
1863          * Stop before overflowing f_bsize - in which case it is better
1864          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1865         if (sizeof(long) < 8) {
1866                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1867                         sfs->f_bsize <<= 1;
1868
1869                         osfs.os_blocks >>= 1;
1870                         osfs.os_bfree >>= 1;
1871                         osfs.os_bavail >>= 1;
1872                 }
1873         }
1874
1875         sfs->f_blocks = osfs.os_blocks;
1876         sfs->f_bfree = osfs.os_bfree;
1877         sfs->f_bavail = osfs.os_bavail;
1878         sfs->f_fsid.val[0] = (__u32)fsid;
1879         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1880         return 0;
1881 }
1882
1883 void ll_inode_size_lock(struct inode *inode)
1884 {
1885         struct ll_inode_info *lli;
1886
1887         LASSERT(!S_ISDIR(inode->i_mode));
1888
1889         lli = ll_i2info(inode);
1890         mutex_lock(&lli->lli_size_mutex);
1891 }
1892
1893 void ll_inode_size_unlock(struct inode *inode)
1894 {
1895         struct ll_inode_info *lli;
1896
1897         lli = ll_i2info(inode);
1898         mutex_unlock(&lli->lli_size_mutex);
1899 }
1900
1901 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1902 {
1903         struct ll_inode_info *lli = ll_i2info(inode);
1904         struct mdt_body *body = md->body;
1905         struct ll_sb_info *sbi = ll_i2sbi(inode);
1906         int rc = 0;
1907
1908         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1909                 rc = cl_file_inode_init(inode, md);
1910                 if (rc)
1911                         return rc;
1912         }
1913
1914         if (S_ISDIR(inode->i_mode)) {
1915                 rc = ll_update_lsm_md(inode, md);
1916                 if (rc != 0)
1917                         return rc;
1918         }
1919
1920 #ifdef CONFIG_FS_POSIX_ACL
1921         if (body->mbo_valid & OBD_MD_FLACL) {
1922                 spin_lock(&lli->lli_lock);
1923                 if (lli->lli_posix_acl)
1924                         posix_acl_release(lli->lli_posix_acl);
1925                 lli->lli_posix_acl = md->posix_acl;
1926                 spin_unlock(&lli->lli_lock);
1927         }
1928 #endif
1929         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1930                                         sbi->ll_flags & LL_SBI_32BIT_API);
1931         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1932
1933         if (body->mbo_valid & OBD_MD_FLATIME) {
1934                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1935                         LTIME_S(inode->i_atime) = body->mbo_atime;
1936                 lli->lli_atime = body->mbo_atime;
1937         }
1938
1939         if (body->mbo_valid & OBD_MD_FLMTIME) {
1940                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1941                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1942                                "to %llu\n", inode->i_ino,
1943                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1944                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1945                 }
1946                 lli->lli_mtime = body->mbo_mtime;
1947         }
1948
1949         if (body->mbo_valid & OBD_MD_FLCTIME) {
1950                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1951                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1952                 lli->lli_ctime = body->mbo_ctime;
1953         }
1954
1955         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1956         if (body->mbo_valid & OBD_MD_FLFLAGS)
1957                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1958         if (body->mbo_valid & OBD_MD_FLMODE)
1959                 inode->i_mode = (inode->i_mode & S_IFMT) |
1960                                 (body->mbo_mode & ~S_IFMT);
1961
1962         if (body->mbo_valid & OBD_MD_FLTYPE)
1963                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1964                                 (body->mbo_mode & S_IFMT);
1965
1966         LASSERT(inode->i_mode != 0);
1967         if (S_ISREG(inode->i_mode))
1968                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1969                                        LL_MAX_BLKSIZE_BITS);
1970         else
1971                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1972
1973         if (body->mbo_valid & OBD_MD_FLUID)
1974                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1975         if (body->mbo_valid & OBD_MD_FLGID)
1976                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1977         if (body->mbo_valid & OBD_MD_FLPROJID)
1978                 lli->lli_projid = body->mbo_projid;
1979         if (body->mbo_valid & OBD_MD_FLNLINK)
1980                 set_nlink(inode, body->mbo_nlink);
1981         if (body->mbo_valid & OBD_MD_FLRDEV)
1982                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1983
1984         if (body->mbo_valid & OBD_MD_FLID) {
1985                 /* FID shouldn't be changed! */
1986                 if (fid_is_sane(&lli->lli_fid)) {
1987                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1988                                  "Trying to change FID "DFID
1989                                  " to the "DFID", inode "DFID"(%p)\n",
1990                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1991                                  PFID(ll_inode2fid(inode)), inode);
1992                 } else {
1993                         lli->lli_fid = body->mbo_fid1;
1994                 }
1995         }
1996
1997         LASSERT(fid_seq(&lli->lli_fid) != 0);
1998
1999         if (body->mbo_valid & OBD_MD_FLSIZE) {
2000                 i_size_write(inode, body->mbo_size);
2001
2002                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2003                        PFID(ll_inode2fid(inode)),
2004                        (unsigned long long)body->mbo_size);
2005
2006                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2007                         inode->i_blocks = body->mbo_blocks;
2008         }
2009
2010         if (body->mbo_valid & OBD_MD_TSTATE) {
2011                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2012                  * clear it when done to ensure to start again
2013                  * glimpsing updated attrs
2014                  */
2015                 if (body->mbo_t_state & MS_RESTORE)
2016                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2017                 else
2018                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2019         }
2020
2021         return 0;
2022 }
2023
2024 int ll_read_inode2(struct inode *inode, void *opaque)
2025 {
2026         struct lustre_md *md = opaque;
2027         struct ll_inode_info *lli = ll_i2info(inode);
2028         int     rc;
2029         ENTRY;
2030
2031         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2032                PFID(&lli->lli_fid), inode);
2033
2034         /* Core attributes from the MDS first.  This is a new inode, and
2035          * the VFS doesn't zero times in the core inode so we have to do
2036          * it ourselves.  They will be overwritten by either MDS or OST
2037          * attributes - we just need to make sure they aren't newer. */
2038         LTIME_S(inode->i_mtime) = 0;
2039         LTIME_S(inode->i_atime) = 0;
2040         LTIME_S(inode->i_ctime) = 0;
2041         inode->i_rdev = 0;
2042         rc = ll_update_inode(inode, md);
2043         if (rc != 0)
2044                 RETURN(rc);
2045
2046         /* OIDEBUG(inode); */
2047
2048 #ifdef HAVE_BACKING_DEV_INFO
2049         /* initializing backing dev info. */
2050         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2051 #endif
2052         if (S_ISREG(inode->i_mode)) {
2053                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2054                 inode->i_op = &ll_file_inode_operations;
2055                 inode->i_fop = sbi->ll_fop;
2056                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2057                 EXIT;
2058         } else if (S_ISDIR(inode->i_mode)) {
2059                 inode->i_op = &ll_dir_inode_operations;
2060                 inode->i_fop = &ll_dir_operations;
2061                 EXIT;
2062         } else if (S_ISLNK(inode->i_mode)) {
2063                 inode->i_op = &ll_fast_symlink_inode_operations;
2064                 EXIT;
2065         } else {
2066                 inode->i_op = &ll_special_inode_operations;
2067
2068                 init_special_inode(inode, inode->i_mode,
2069                                    inode->i_rdev);
2070
2071                 EXIT;
2072         }
2073
2074         return 0;
2075 }
2076
2077 void ll_delete_inode(struct inode *inode)
2078 {
2079         struct ll_inode_info *lli = ll_i2info(inode);
2080         ENTRY;
2081
2082         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2083                 /* It is last chance to write out dirty pages,
2084                  * otherwise we may lose data while umount */
2085                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2086
2087         truncate_inode_pages_final(&inode->i_data);
2088
2089         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2090                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2091                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2092
2093 #ifdef HAVE_SBOPS_EVICT_INODE
2094         ll_clear_inode(inode);
2095 #endif
2096         clear_inode(inode);
2097
2098         EXIT;
2099 }
2100
2101 int ll_iocontrol(struct inode *inode, struct file *file,
2102                  unsigned int cmd, unsigned long arg)
2103 {
2104         struct ll_sb_info *sbi = ll_i2sbi(inode);
2105         struct ptlrpc_request *req = NULL;
2106         int rc, flags = 0;
2107         ENTRY;
2108
2109         switch (cmd) {
2110         case FS_IOC_GETFLAGS: {
2111                 struct mdt_body *body;
2112                 struct md_op_data *op_data;
2113
2114                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2115                                              0, 0, LUSTRE_OPC_ANY,
2116                                              NULL);
2117                 if (IS_ERR(op_data))
2118                         RETURN(PTR_ERR(op_data));
2119
2120                 op_data->op_valid = OBD_MD_FLFLAGS;
2121                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2122                 ll_finish_md_op_data(op_data);
2123                 if (rc) {
2124                         CERROR("%s: failure inode "DFID": rc = %d\n",
2125                                sbi->ll_md_exp->exp_obd->obd_name,
2126                                PFID(ll_inode2fid(inode)), rc);
2127                         RETURN(-abs(rc));
2128                 }
2129
2130                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2131
2132                 flags = body->mbo_flags;
2133
2134                 ptlrpc_req_finished(req);
2135
2136                 RETURN(put_user(flags, (int __user *)arg));
2137         }
2138         case FS_IOC_SETFLAGS: {
2139                 struct iattr *attr;
2140                 struct md_op_data *op_data;
2141                 struct cl_object *obj;
2142
2143                 if (get_user(flags, (int __user *)arg))
2144                         RETURN(-EFAULT);
2145
2146                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2147                                              LUSTRE_OPC_ANY, NULL);
2148                 if (IS_ERR(op_data))
2149                         RETURN(PTR_ERR(op_data));
2150
2151                 op_data->op_attr_flags = flags;
2152                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2153                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2154                 ll_finish_md_op_data(op_data);
2155                 ptlrpc_req_finished(req);
2156                 if (rc)
2157                         RETURN(rc);
2158
2159                 inode->i_flags = ll_ext_to_inode_flags(flags);
2160
2161                 obj = ll_i2info(inode)->lli_clob;
2162                 if (obj == NULL)
2163                         RETURN(0);
2164
2165                 OBD_ALLOC_PTR(attr);
2166                 if (attr == NULL)
2167                         RETURN(-ENOMEM);
2168
2169                 attr->ia_valid = ATTR_ATTR_FLAG;
2170                 rc = cl_setattr_ost(obj, attr, flags);
2171
2172                 OBD_FREE_PTR(attr);
2173                 RETURN(rc);
2174         }
2175         default:
2176                 RETURN(-ENOSYS);
2177         }
2178
2179         RETURN(0);
2180 }
2181
2182 int ll_flush_ctx(struct inode *inode)
2183 {
2184         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2185
2186         CDEBUG(D_SEC, "flush context for user %d\n",
2187                from_kuid(&init_user_ns, current_uid()));
2188
2189         obd_set_info_async(NULL, sbi->ll_md_exp,
2190                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2191                            0, NULL, NULL);
2192         obd_set_info_async(NULL, sbi->ll_dt_exp,
2193                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2194                            0, NULL, NULL);
2195         return 0;
2196 }
2197
2198 /* umount -f client means force down, don't save state */
2199 void ll_umount_begin(struct super_block *sb)
2200 {
2201         struct ll_sb_info *sbi = ll_s2sbi(sb);
2202         struct obd_device *obd;
2203         struct obd_ioctl_data *ioc_data;
2204         struct l_wait_info lwi;
2205         wait_queue_head_t waitq;
2206         ENTRY;
2207
2208         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2209                sb->s_count, atomic_read(&sb->s_active));
2210
2211         obd = class_exp2obd(sbi->ll_md_exp);
2212         if (obd == NULL) {
2213                 CERROR("Invalid MDC connection handle %#llx\n",
2214                        sbi->ll_md_exp->exp_handle.h_cookie);
2215                 EXIT;
2216                 return;
2217         }
2218         obd->obd_force = 1;
2219
2220         obd = class_exp2obd(sbi->ll_dt_exp);
2221         if (obd == NULL) {
2222                 CERROR("Invalid LOV connection handle %#llx\n",
2223                        sbi->ll_dt_exp->exp_handle.h_cookie);
2224                 EXIT;
2225                 return;
2226         }
2227         obd->obd_force = 1;
2228
2229         OBD_ALLOC_PTR(ioc_data);
2230         if (ioc_data) {
2231                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2232                               sizeof *ioc_data, ioc_data, NULL);
2233
2234                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2235                               sizeof *ioc_data, ioc_data, NULL);
2236
2237                 OBD_FREE_PTR(ioc_data);
2238         }
2239
2240         /* Really, we'd like to wait until there are no requests outstanding,
2241          * and then continue.  For now, we just periodically checking for vfs
2242          * to decrement mnt_cnt and hope to finish it within 10sec.
2243          */
2244         init_waitqueue_head(&waitq);
2245         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2246                                    cfs_time_seconds(1), NULL, NULL);
2247         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2248
2249         EXIT;
2250 }
2251
2252 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2253 {
2254         struct ll_sb_info *sbi = ll_s2sbi(sb);
2255         char *profilenm = get_profile_name(sb);
2256         int err;
2257         __u32 read_only;
2258
2259         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2260                 read_only = *flags & MS_RDONLY;
2261                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2262                                          sizeof(KEY_READ_ONLY),
2263                                          KEY_READ_ONLY, sizeof(read_only),
2264                                          &read_only, NULL);
2265                 if (err) {
2266                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2267                                       profilenm, read_only ?
2268                                       "read-only" : "read-write", err);
2269                         return err;
2270                 }
2271
2272                 if (read_only)
2273                         sb->s_flags |= MS_RDONLY;
2274                 else
2275                         sb->s_flags &= ~MS_RDONLY;
2276
2277                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2278                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2279                                       read_only ?  "read-only" : "read-write");
2280         }
2281         return 0;
2282 }
2283
2284 /**
2285  * Cleanup the open handle that is cached on MDT-side.
2286  *
2287  * For open case, the client side open handling thread may hit error
2288  * after the MDT grant the open. Under such case, the client should
2289  * send close RPC to the MDT as cleanup; otherwise, the open handle
2290  * on the MDT will be leaked there until the client umount or evicted.
2291  *
2292  * In further, if someone unlinked the file, because the open handle
2293  * holds the reference on such file/object, then it will block the
2294  * subsequent threads that want to locate such object via FID.
2295  *
2296  * \param[in] sb        super block for this file-system
2297  * \param[in] open_req  pointer to the original open request
2298  */
2299 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2300 {
2301         struct mdt_body                 *body;
2302         struct md_op_data               *op_data;
2303         struct ptlrpc_request           *close_req = NULL;
2304         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2305         ENTRY;
2306
2307         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2308         OBD_ALLOC_PTR(op_data);
2309         if (op_data == NULL) {
2310                 CWARN("%s: cannot allocate op_data to release open handle for "
2311                       DFID"\n",
2312                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2313
2314                 RETURN_EXIT;
2315         }
2316
2317         op_data->op_fid1 = body->mbo_fid1;
2318         op_data->op_handle = body->mbo_handle;
2319         op_data->op_mod_time = ktime_get_real_seconds();
2320         md_close(exp, op_data, NULL, &close_req);
2321         ptlrpc_req_finished(close_req);
2322         ll_finish_md_op_data(op_data);
2323
2324         EXIT;
2325 }
2326
2327 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2328                   struct super_block *sb, struct lookup_intent *it)
2329 {
2330         struct ll_sb_info *sbi = NULL;
2331         struct lustre_md md = { NULL };
2332         int rc;
2333         ENTRY;
2334
2335         LASSERT(*inode || sb);
2336         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2337         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2338                               sbi->ll_md_exp, &md);
2339         if (rc != 0)
2340                 GOTO(cleanup, rc);
2341
2342         if (*inode) {
2343                 rc = ll_update_inode(*inode, &md);
2344                 if (rc != 0)
2345                         GOTO(out, rc);
2346         } else {
2347                 LASSERT(sb != NULL);
2348
2349                 /*
2350                  * At this point server returns to client's same fid as client
2351                  * generated for creating. So using ->fid1 is okay here.
2352                  */
2353                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2354                         CERROR("%s: Fid is insane "DFID"\n",
2355                                 ll_get_fsname(sb, NULL, 0),
2356                                 PFID(&md.body->mbo_fid1));
2357                         GOTO(out, rc = -EINVAL);
2358                 }
2359
2360                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2361                                              sbi->ll_flags & LL_SBI_32BIT_API),
2362                                  &md);
2363                 if (IS_ERR(*inode)) {
2364 #ifdef CONFIG_FS_POSIX_ACL
2365                         if (md.posix_acl) {
2366                                 posix_acl_release(md.posix_acl);
2367                                 md.posix_acl = NULL;
2368                         }
2369 #endif
2370                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2371                         *inode = NULL;
2372                         CERROR("new_inode -fatal: rc %d\n", rc);
2373                         GOTO(out, rc);
2374                 }
2375         }
2376
2377         /* Handling piggyback layout lock.
2378          * Layout lock can be piggybacked by getattr and open request.
2379          * The lsm can be applied to inode only if it comes with a layout lock
2380          * otherwise correct layout may be overwritten, for example:
2381          * 1. proc1: mdt returns a lsm but not granting layout
2382          * 2. layout was changed by another client
2383          * 3. proc2: refresh layout and layout lock granted
2384          * 4. proc1: to apply a stale layout */
2385         if (it != NULL && it->it_lock_mode != 0) {
2386                 struct lustre_handle lockh;
2387                 struct ldlm_lock *lock;
2388
2389                 lockh.cookie = it->it_lock_handle;
2390                 lock = ldlm_handle2lock(&lockh);
2391                 LASSERT(lock != NULL);
2392                 if (ldlm_has_layout(lock)) {
2393                         struct cl_object_conf conf;
2394
2395                         memset(&conf, 0, sizeof(conf));
2396                         conf.coc_opc = OBJECT_CONF_SET;
2397                         conf.coc_inode = *inode;
2398                         conf.coc_lock = lock;
2399                         conf.u.coc_layout = md.layout;
2400                         (void)ll_layout_conf(*inode, &conf);
2401                 }
2402                 LDLM_LOCK_PUT(lock);
2403         }
2404
2405         GOTO(out, rc = 0);
2406
2407 out:
2408         md_free_lustre_md(sbi->ll_md_exp, &md);
2409
2410 cleanup:
2411         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2412                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2413
2414         return rc;
2415 }
2416
2417 int ll_obd_statfs(struct inode *inode, void __user *arg)
2418 {
2419         struct ll_sb_info *sbi = NULL;
2420         struct obd_export *exp;
2421         char *buf = NULL;
2422         struct obd_ioctl_data *data = NULL;
2423         __u32 type;
2424         int len = 0, rc;
2425
2426         if (!inode || !(sbi = ll_i2sbi(inode)))
2427                 GOTO(out_statfs, rc = -EINVAL);
2428
2429         rc = obd_ioctl_getdata(&buf, &len, arg);
2430         if (rc)
2431                 GOTO(out_statfs, rc);
2432
2433         data = (void*)buf;
2434         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2435             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2436                 GOTO(out_statfs, rc = -EINVAL);
2437
2438         if (data->ioc_inllen1 != sizeof(__u32) ||
2439             data->ioc_inllen2 != sizeof(__u32) ||
2440             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2441             data->ioc_plen2 != sizeof(struct obd_uuid))
2442                 GOTO(out_statfs, rc = -EINVAL);
2443
2444         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2445         if (type & LL_STATFS_LMV)
2446                 exp = sbi->ll_md_exp;
2447         else if (type & LL_STATFS_LOV)
2448                 exp = sbi->ll_dt_exp;
2449         else
2450                 GOTO(out_statfs, rc = -ENODEV);
2451
2452         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2453         if (rc)
2454                 GOTO(out_statfs, rc);
2455 out_statfs:
2456         OBD_FREE_LARGE(buf, len);
2457         return rc;
2458 }
2459
2460 /* this function prepares md_op_data hint for passing it down to MD stack. */
2461 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2462                                       struct inode *i1, struct inode *i2,
2463                                       const char *name, size_t namelen,
2464                                       __u32 mode, __u32 opc, void *data)
2465 {
2466         LASSERT(i1 != NULL);
2467
2468         if (name == NULL) {
2469                 /* Do not reuse namelen for something else. */
2470                 if (namelen != 0)
2471                         return ERR_PTR(-EINVAL);
2472         } else {
2473                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2474                         return ERR_PTR(-ENAMETOOLONG);
2475
2476                 if (!lu_name_is_valid_2(name, namelen))
2477                         return ERR_PTR(-EINVAL);
2478         }
2479
2480         if (op_data == NULL)
2481                 OBD_ALLOC_PTR(op_data);
2482
2483         if (op_data == NULL)
2484                 return ERR_PTR(-ENOMEM);
2485
2486         ll_i2gids(op_data->op_suppgids, i1, i2);
2487         op_data->op_fid1 = *ll_inode2fid(i1);
2488         op_data->op_default_stripe_offset = -1;
2489         if (S_ISDIR(i1->i_mode)) {
2490                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2491                 if (opc == LUSTRE_OPC_MKDIR)
2492                         op_data->op_default_stripe_offset =
2493                                    ll_i2info(i1)->lli_def_stripe_offset;
2494         }
2495
2496         if (i2) {
2497                 op_data->op_fid2 = *ll_inode2fid(i2);
2498                 if (S_ISDIR(i2->i_mode))
2499                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2500         } else {
2501                 fid_zero(&op_data->op_fid2);
2502         }
2503
2504         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2505                 op_data->op_cli_flags |= CLI_HASH64;
2506
2507         if (ll_need_32bit_api(ll_i2sbi(i1)))
2508                 op_data->op_cli_flags |= CLI_API32;
2509
2510         op_data->op_name = name;
2511         op_data->op_namelen = namelen;
2512         op_data->op_mode = mode;
2513         op_data->op_mod_time = ktime_get_real_seconds();
2514         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2515         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2516         op_data->op_cap = cfs_curproc_cap_pack();
2517         op_data->op_mds = 0;
2518         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2519              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2520                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2521         }
2522         op_data->op_data = data;
2523
2524         return op_data;
2525 }
2526
2527 void ll_finish_md_op_data(struct md_op_data *op_data)
2528 {
2529         security_release_secctx(op_data->op_file_secctx,
2530                                 op_data->op_file_secctx_size);
2531         OBD_FREE_PTR(op_data);
2532 }
2533
2534 #ifdef HAVE_SUPEROPS_USE_DENTRY
2535 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2536 #else
2537 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2538 #endif
2539 {
2540         struct ll_sb_info *sbi;
2541
2542 #ifdef HAVE_SUPEROPS_USE_DENTRY
2543         LASSERT((seq != NULL) && (dentry != NULL));
2544         sbi = ll_s2sbi(dentry->d_sb);
2545 #else
2546         LASSERT((seq != NULL) && (vfs != NULL));
2547         sbi = ll_s2sbi(vfs->mnt_sb);
2548 #endif
2549
2550         if (sbi->ll_flags & LL_SBI_NOLCK)
2551                 seq_puts(seq, ",nolock");
2552
2553         if (sbi->ll_flags & LL_SBI_FLOCK)
2554                 seq_puts(seq, ",flock");
2555
2556         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2557                 seq_puts(seq, ",localflock");
2558
2559         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2560                 seq_puts(seq, ",user_xattr");
2561
2562         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2563                 seq_puts(seq, ",lazystatfs");
2564
2565         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2566                 seq_puts(seq, ",user_fid2path");
2567
2568         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2569                 seq_puts(seq, ",always_ping");
2570
2571         RETURN(0);
2572 }
2573
2574 /**
2575  * Get obd name by cmd, and copy out to user space
2576  */
2577 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2578 {
2579         struct ll_sb_info *sbi = ll_i2sbi(inode);
2580         struct obd_device *obd;
2581         ENTRY;
2582
2583         if (cmd == OBD_IOC_GETDTNAME)
2584                 obd = class_exp2obd(sbi->ll_dt_exp);
2585         else if (cmd == OBD_IOC_GETMDNAME)
2586                 obd = class_exp2obd(sbi->ll_md_exp);
2587         else
2588                 RETURN(-EINVAL);
2589
2590         if (!obd)
2591                 RETURN(-ENOENT);
2592
2593         if (copy_to_user((void __user *)arg, obd->obd_name,
2594                          strlen(obd->obd_name) + 1))
2595                 RETURN(-EFAULT);
2596
2597         RETURN(0);
2598 }
2599
2600 /**
2601  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2602  * fsname will be returned in this buffer; otherwise, a static buffer will be
2603  * used to store the fsname and returned to caller.
2604  */
2605 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2606 {
2607         static char fsname_static[MTI_NAME_MAXLEN];
2608         struct lustre_sb_info *lsi = s2lsi(sb);
2609         char *ptr;
2610         int len;
2611
2612         if (buf == NULL) {
2613                 /* this means the caller wants to use static buffer
2614                  * and it doesn't care about race. Usually this is
2615                  * in error reporting path */
2616                 buf = fsname_static;
2617                 buflen = sizeof(fsname_static);
2618         }
2619
2620         len = strlen(lsi->lsi_lmd->lmd_profile);
2621         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2622         if (ptr && (strcmp(ptr, "-client") == 0))
2623                 len -= 7;
2624
2625         if (unlikely(len >= buflen))
2626                 len = buflen - 1;
2627         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2628         buf[len] = '\0';
2629
2630         return buf;
2631 }
2632
2633 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2634 {
2635         char *path = NULL;
2636
2637         struct path p;
2638
2639         p.dentry = dentry;
2640         p.mnt = current->fs->root.mnt;
2641         path_get(&p);
2642         path = d_path(&p, buf, bufsize);
2643         path_put(&p);
2644         return path;
2645 }
2646
2647 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2648 {
2649         char *buf, *path = NULL;
2650         struct dentry *dentry = NULL;
2651         struct inode *inode = page->mapping->host;
2652
2653         /* this can be called inside spin lock so use GFP_ATOMIC. */
2654         buf = (char *)__get_free_page(GFP_ATOMIC);
2655         if (buf != NULL) {
2656                 dentry = d_find_alias(page->mapping->host);
2657                 if (dentry != NULL)
2658                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2659         }
2660
2661         CDEBUG(D_WARNING,
2662                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2663                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2664                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2665                PFID(ll_inode2fid(inode)),
2666                (path && !IS_ERR(path)) ? path : "", ioret);
2667
2668         if (dentry != NULL)
2669                 dput(dentry);
2670
2671         if (buf != NULL)
2672                 free_page((unsigned long)buf);
2673 }
2674
2675 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2676                         struct lov_user_md **kbuf)
2677 {
2678         struct lov_user_md      lum;
2679         ssize_t                 lum_size;
2680         ENTRY;
2681
2682         if (copy_from_user(&lum, md, sizeof(lum)))
2683                 RETURN(-EFAULT);
2684
2685         lum_size = ll_lov_user_md_size(&lum);
2686         if (lum_size < 0)
2687                 RETURN(lum_size);
2688
2689         OBD_ALLOC(*kbuf, lum_size);
2690         if (*kbuf == NULL)
2691                 RETURN(-ENOMEM);
2692
2693         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2694                 OBD_FREE(*kbuf, lum_size);
2695                 RETURN(-EFAULT);
2696         }
2697
2698         RETURN(lum_size);
2699 }
2700
2701 /*
2702  * Compute llite root squash state after a change of root squash
2703  * configuration setting or add/remove of a lnet nid
2704  */
2705 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2706 {
2707         struct root_squash_info *squash = &sbi->ll_squash;
2708         int i;
2709         bool matched;
2710         struct lnet_process_id id;
2711
2712         /* Update norootsquash flag */
2713         down_write(&squash->rsi_sem);
2714         if (list_empty(&squash->rsi_nosquash_nids))
2715                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2716         else {
2717                 /* Do not apply root squash as soon as one of our NIDs is
2718                  * in the nosquash_nids list */
2719                 matched = false;
2720                 i = 0;
2721                 while (LNetGetId(i++, &id) != -ENOENT) {
2722                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2723                                 continue;
2724                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2725                                 matched = true;
2726                                 break;
2727                         }
2728                 }
2729                 if (matched)
2730                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2731                 else
2732                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2733         }
2734         up_write(&squash->rsi_sem);
2735 }
2736
2737 /**
2738  * Parse linkea content to extract information about a given hardlink
2739  *
2740  * \param[in]   ldata      - Initialized linkea data
2741  * \param[in]   linkno     - Link identifier
2742  * \param[out]  parent_fid - The entry's parent FID
2743  * \param[out]  ln         - Entry name destination buffer
2744  *
2745  * \retval 0 on success
2746  * \retval Appropriate negative error code on failure
2747  */
2748 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2749                             struct lu_fid *parent_fid, struct lu_name *ln)
2750 {
2751         unsigned int    idx;
2752         int             rc;
2753         ENTRY;
2754
2755         rc = linkea_init_with_rec(ldata);
2756         if (rc < 0)
2757                 RETURN(rc);
2758
2759         if (linkno >= ldata->ld_leh->leh_reccount)
2760                 /* beyond last link */
2761                 RETURN(-ENODATA);
2762
2763         linkea_first_entry(ldata);
2764         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2765                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2766                                     parent_fid);
2767                 if (idx == linkno)
2768                         break;
2769
2770                 linkea_next_entry(ldata);
2771         }
2772
2773         if (idx < linkno)
2774                 RETURN(-ENODATA);
2775
2776         RETURN(0);
2777 }
2778
2779 /**
2780  * Get parent FID and name of an identified link. Operation is performed for
2781  * a given link number, letting the caller iterate over linkno to list one or
2782  * all links of an entry.
2783  *
2784  * \param[in]     file - File descriptor against which to perform the operation
2785  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2786  *                       on and the available size. It is eventually filled with
2787  *                       the requested information or left untouched on error
2788  *
2789  * \retval - 0 on success
2790  * \retval - Appropriate negative error code on failure
2791  */
2792 int ll_getparent(struct file *file, struct getparent __user *arg)
2793 {
2794         struct inode            *inode = file_inode(file);
2795         struct linkea_data      *ldata;
2796         struct lu_buf            buf = LU_BUF_NULL;
2797         struct lu_name           ln;
2798         struct lu_fid            parent_fid;
2799         __u32                    linkno;
2800         __u32                    name_size;
2801         int                      rc;
2802
2803         ENTRY;
2804
2805         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2806             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2807                 RETURN(-EPERM);
2808
2809         if (get_user(name_size, &arg->gp_name_size))
2810                 RETURN(-EFAULT);
2811
2812         if (get_user(linkno, &arg->gp_linkno))
2813                 RETURN(-EFAULT);
2814
2815         if (name_size > PATH_MAX)
2816                 RETURN(-EINVAL);
2817
2818         OBD_ALLOC(ldata, sizeof(*ldata));
2819         if (ldata == NULL)
2820                 RETURN(-ENOMEM);
2821
2822         rc = linkea_data_new(ldata, &buf);
2823         if (rc < 0)
2824                 GOTO(ldata_free, rc);
2825
2826 #ifdef HAVE_XATTR_HANDLER_FLAGS
2827         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2828                            buf.lb_len, OBD_MD_FLXATTR);
2829 #else
2830         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2831                          buf.lb_len);
2832 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2833         if (rc < 0)
2834                 GOTO(lb_free, rc);
2835
2836         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2837         if (rc < 0)
2838                 GOTO(lb_free, rc);
2839
2840         if (ln.ln_namelen >= name_size)
2841                 GOTO(lb_free, rc = -EOVERFLOW);
2842
2843         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2844                 GOTO(lb_free, rc = -EFAULT);
2845
2846         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2847                 GOTO(lb_free, rc = -EFAULT);
2848
2849         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2850                 GOTO(lb_free, rc = -EFAULT);
2851
2852 lb_free:
2853         lu_buf_free(&buf);
2854 ldata_free:
2855         OBD_FREE(ldata, sizeof(*ldata));
2856
2857         RETURN(rc);
2858 }