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 #ifdef ENABLE_FLOCK
 113         sbi->ll_flags |= LL_SBI_FLOCK;
 114 #endif
 115
 116 #ifdef HAVE_LRU_RESIZE_SUPPORT
 117         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 118 #endif
 119         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 120
 121         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 122                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 123                                pp_r_hist.oh_lock);
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_w_hist.oh_lock);
 126         }
 127
 128         /* metadata statahead is enabled by default */
 129         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136         sbi->ll_flags |= LL_SBI_FAST_READ;
 137         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 138
 139         /* root squash */
 140         sbi->ll_squash.rsi_uid = 0;
 141         sbi->ll_squash.rsi_gid = 0;
 142         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 143         init_rwsem(&sbi->ll_squash.rsi_sem);
 144
 145         /* Per-filesystem file heat */
 146         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 147         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 148         RETURN(sbi);
 149 }
 150
 151 static void ll_free_sbi(struct super_block *sb)
 152 {
 153         struct ll_sb_info *sbi = ll_s2sbi(sb);
 154         ENTRY;
 155
 156         if (sbi != NULL) {
 157                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 158                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 159                 if (sbi->ll_cache != NULL) {
 160                         cl_cache_decref(sbi->ll_cache);
 161                         sbi->ll_cache = NULL;
 162                 }
 163                 OBD_FREE(sbi, sizeof(*sbi));
 164         }
 165         EXIT;
 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_statfs *osfs = NULL;
 174         struct ptlrpc_request *request = NULL;
 175         struct obd_connect_data *data = NULL;
 176         struct obd_uuid *uuid;
 177         struct md_op_data *op_data;
 178         struct lustre_md lmd;
 179         u64 valid;
 180         int size, err, checksum;
 181
 182         ENTRY;
 183         sbi->ll_md_obd = class_name2obd(md);
 184         if (!sbi->ll_md_obd) {
 185                 CERROR("MD %s: not setup or attached\n", md);
 186                 RETURN(-EINVAL);
 187         }
 188
 189         OBD_ALLOC_PTR(data);
 190         if (data == NULL)
 191                 RETURN(-ENOMEM);
 192
 193         OBD_ALLOC_PTR(osfs);
 194         if (osfs == NULL) {
 195                 OBD_FREE_PTR(data);
 196                 RETURN(-ENOMEM);
 197         }
 198
 199         /* pass client page size via ocd_grant_blkbits, the server should report
 200          * back its backend blocksize for grant calculation purpose */
 201         data->ocd_grant_blkbits = PAGE_SHIFT;
 202
 203         /* indicate MDT features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 208                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 209                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 210                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 211                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 212                                   OBD_CONNECT_64BITHASH |
 213                                   OBD_CONNECT_EINPROGRESS |
 214                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 215                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 216                                   OBD_CONNECT_MAX_EASIZE |
 217                                   OBD_CONNECT_FLOCK_DEAD |
 218                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 219                                   OBD_CONNECT_OPEN_BY_FID |
 220                                   OBD_CONNECT_DIR_STRIPE |
 221                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 222                                   OBD_CONNECT_SUBTREE |
 223                                   OBD_CONNECT_MULTIMODRPCS |
 224                                   OBD_CONNECT_GRANT_PARAM |
 225                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 226
 227         data->ocd_connect_flags2 = OBD_CONNECT2_FLR |
 228                                    OBD_CONNECT2_LOCK_CONVERT |
 229                                    OBD_CONNECT2_DIR_MIGRATE |
 230                                    OBD_CONNECT2_SUM_STATFS |
 231                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 232                                    OBD_CONNECT2_LSOM;
 233
 234 #ifdef HAVE_LRU_RESIZE_SUPPORT
 235         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 236                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 237 #endif
 238 #ifdef CONFIG_FS_POSIX_ACL
 239         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 240                                    OBD_CONNECT_LARGE_ACL;
 241 #endif
 242
 243         data->ocd_cksum_types = obd_cksum_types_supported_client();
 244
 245         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 246                 /* flag mdc connection as lightweight, only used for test
 247                  * purpose, use with care */
 248                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 249
 250         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 251         data->ocd_version = LUSTRE_VERSION_CODE;
 252
 253         if (sb->s_flags & MS_RDONLY)
 254                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 255         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 256                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 257
 258 #ifdef MS_NOSEC
 259         /* Setting this indicates we correctly support S_NOSEC (See kernel
 260          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 261          */
 262         sb->s_flags |= MS_NOSEC;
 263 #endif
 264
 265         if (sbi->ll_flags & LL_SBI_FLOCK)
 266                 sbi->ll_fop = &ll_file_operations_flock;
 267         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 268                 sbi->ll_fop = &ll_file_operations;
 269         else
 270                 sbi->ll_fop = &ll_file_operations_noflock;
 271
 272         /* always ping even if server suppress_pings */
 273         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 274                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 275
 276         obd_connect_set_secctx(data);
 277
 278 #if defined(CONFIG_SECURITY)
 279         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 280 #endif
 281
 282         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 283
 284         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 285                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 286         if (err == -EBUSY) {
 287                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 288                                    "recovery, of which this client is not a "
 289                                    "part. Please wait for recovery to complete,"
 290                                    " abort, or time out.\n", md);
 291                 GOTO(out, err);
 292         } else if (err) {
 293                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 294                 GOTO(out, err);
 295         }
 296
 297         sbi->ll_md_exp->exp_connect_data = *data;
 298
 299         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 300                            LUSTRE_SEQ_METADATA);
 301         if (err) {
 302                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 303                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 304                 GOTO(out_md, err);
 305         }
 306
 307         /* For mount, we only need fs info from MDT0, and also in DNE, it
 308          * can make sure the client can be mounted as long as MDT0 is
 309          * avaible */
 310         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 311                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 312                         OBD_STATFS_FOR_MDT0);
 313         if (err)
 314                 GOTO(out_md_fid, err);
 315
 316         /* This needs to be after statfs to ensure connect has finished.
 317          * Note that "data" does NOT contain the valid connect reply.
 318          * If connecting to a 1.8 server there will be no LMV device, so
 319          * we can access the MDC export directly and exp_connect_flags will
 320          * be non-zero, but if accessing an upgraded 2.1 server it will
 321          * have the correct flags filled in.
 322          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 323         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 324         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 325             valid != CLIENT_CONNECT_MDT_REQD) {
 326                 char *buf;
 327
 328                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 329                 obd_connect_flags2str(buf, PAGE_SIZE,
 330                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 331                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 332                                    "feature(s) needed for correct operation "
 333                                    "of this client (%s). Please upgrade "
 334                                    "server or downgrade client.\n",
 335                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 336                 OBD_FREE(buf, PAGE_SIZE);
 337                 GOTO(out_md_fid, err = -EPROTO);
 338         }
 339
 340         size = sizeof(*data);
 341         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 342                            KEY_CONN_DATA,  &size, data);
 343         if (err) {
 344                 CERROR("%s: Get connect data failed: rc = %d\n",
 345                        sbi->ll_md_exp->exp_obd->obd_name, err);
 346                 GOTO(out_md_fid, err);
 347         }
 348
 349         LASSERT(osfs->os_bsize);
 350         sb->s_blocksize = osfs->os_bsize;
 351         sb->s_blocksize_bits = log2(osfs->os_bsize);
 352         sb->s_magic = LL_SUPER_MAGIC;
 353         sb->s_maxbytes = MAX_LFS_FILESIZE;
 354         sbi->ll_namelen = osfs->os_namelen;
 355         sbi->ll_mnt.mnt = current->fs->root.mnt;
 356
 357         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 358             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 359                 LCONSOLE_INFO("Disabling user_xattr feature because "
 360                               "it is not supported on the server\n");
 361                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 365 #ifdef MS_POSIXACL
 366                 sb->s_flags |= MS_POSIXACL;
 367 #endif
 368                 sbi->ll_flags |= LL_SBI_ACL;
 369         } else {
 370                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 371 #ifdef MS_POSIXACL
 372                 sb->s_flags &= ~MS_POSIXACL;
 373 #endif
 374                 sbi->ll_flags &= ~LL_SBI_ACL;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 378                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 381                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 382
 383         if (obd_connect_has_secctx(data))
 384                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 385
 386         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 387                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 388                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 389                                       "unknown maximum xattr size.\n", dt);
 390                 } else if (!sbi->ll_xattr_cache_set) {
 391                         /* If xattr_cache is already set (no matter 0 or 1)
 392                          * during processing llog, it won't be enabled here. */
 393                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 394                         sbi->ll_xattr_cache_enabled = 1;
 395                 }
 396         }
 397
 398         sbi->ll_dt_obd = class_name2obd(dt);
 399         if (!sbi->ll_dt_obd) {
 400                 CERROR("DT %s: not setup or attached\n", dt);
 401                 GOTO(out_md_fid, err = -ENODEV);
 402         }
 403
 404         /* pass client page size via ocd_grant_blkbits, the server should report
 405          * back its backend blocksize for grant calculation purpose */
 406         data->ocd_grant_blkbits = PAGE_SHIFT;
 407
 408         /* indicate OST features supported by this client */
 409         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 410                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 411                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 412                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 413                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 414                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 415                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 416                                   OBD_CONNECT_EINPROGRESS |
 417                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 418                                   OBD_CONNECT_LAYOUTLOCK |
 419                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 420                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 421                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 422
 423 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 424  * can interoperate with an older version of lockahead which was released prior
 425  * to landing in master. This support will be dropped when 2.13 development
 426  * starts.  At the point, we should not just drop the connect flag (below), we
 427  * should also remove the support in the code.
 428  *
 429  * Removing it means a few things:
 430  * 1. Remove this section here
 431  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 432  * 3. Remove function exp_connect_lockahead_old
 433  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 434  * */
 435 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 436         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 437 #endif
 438
 439         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 440
 441         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 442                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 443
 444         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 445          * disabled by default, because it can still be enabled on the
 446          * fly via /sys. As a consequence, we still need to come to an
 447          * agreement on the supported algorithms at connect time
 448          */
 449         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 450
 451         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 452                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 453         else
 454                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 455
 456 #ifdef HAVE_LRU_RESIZE_SUPPORT
 457         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 458 #endif
 459         /* always ping even if server suppress_pings */
 460         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 461                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 462
 463         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 464                "ocd_grant: %d\n", data->ocd_connect_flags,
 465                data->ocd_version, data->ocd_grant);
 466
 467         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 468         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 469
 470         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 471
 472         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 473                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 474         if (err == -EBUSY) {
 475                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 476                                    "recovery, of which this client is not a "
 477                                    "part.  Please wait for recovery to "
 478                                    "complete, abort, or time out.\n", dt);
 479                 GOTO(out_md, err);
 480         } else if (err) {
 481                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 482                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 483                 GOTO(out_md, err);
 484         }
 485
 486         sbi->ll_dt_exp->exp_connect_data = *data;
 487
 488         /* Don't change value if it was specified in the config log */
 489         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 490                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 491                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 492                               (data->ocd_brw_size >> PAGE_SHIFT));
 493
 494         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 495                            LUSTRE_SEQ_METADATA);
 496         if (err) {
 497                 CERROR("%s: Can't init data layer FID infrastructure, "
 498                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 499                 GOTO(out_dt, err);
 500         }
 501
 502         mutex_lock(&sbi->ll_lco.lco_lock);
 503         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 504         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 505         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 506         mutex_unlock(&sbi->ll_lco.lco_lock);
 507
 508         fid_zero(&sbi->ll_root_fid);
 509         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 510                            &sbi->ll_root_fid);
 511         if (err) {
 512                 CERROR("cannot mds_connect: rc = %d\n", err);
 513                 GOTO(out_lock_cn_cb, err);
 514         }
 515         if (!fid_is_sane(&sbi->ll_root_fid)) {
 516                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 517                        sbi->ll_md_exp->exp_obd->obd_name,
 518                        PFID(&sbi->ll_root_fid));
 519                 GOTO(out_lock_cn_cb, err = -EINVAL);
 520         }
 521         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 522
 523         sb->s_op = &lustre_super_operations;
 524 #ifdef HAVE_XATTR_HANDLER_FLAGS
 525         sb->s_xattr = ll_xattr_handlers;
 526 #endif
 527 #if THREAD_SIZE >= 8192 /*b=17630*/
 528         sb->s_export_op = &lustre_export_operations;
 529 #endif
 530
 531         /* make root inode
 532          * XXX: move this to after cbd setup? */
 533         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 534         if (sbi->ll_flags & LL_SBI_ACL)
 535                 valid |= OBD_MD_FLACL;
 536
 537         OBD_ALLOC_PTR(op_data);
 538         if (op_data == NULL)
 539                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 540
 541         op_data->op_fid1 = sbi->ll_root_fid;
 542         op_data->op_mode = 0;
 543         op_data->op_valid = valid;
 544
 545         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 546
 547         OBD_FREE_PTR(op_data);
 548         if (err) {
 549                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 550                        sbi->ll_md_exp->exp_obd->obd_name, err);
 551                 GOTO(out_lock_cn_cb, err);
 552         }
 553
 554         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 555                                sbi->ll_md_exp, &lmd);
 556         if (err) {
 557                 CERROR("failed to understand root inode md: rc = %d\n", err);
 558                 ptlrpc_req_finished(request);
 559                 GOTO(out_lock_cn_cb, err);
 560         }
 561
 562         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 563         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 564                                             sbi->ll_flags & LL_SBI_32BIT_API),
 565                        &lmd);
 566         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 567         ptlrpc_req_finished(request);
 568
 569         if (IS_ERR(root)) {
 570 #ifdef CONFIG_FS_POSIX_ACL
 571                 if (lmd.posix_acl) {
 572                         posix_acl_release(lmd.posix_acl);
 573                         lmd.posix_acl = NULL;
 574                 }
 575 #endif
 576                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 577                 root = NULL;
 578                 CERROR("lustre_lite: bad iget4 for root\n");
 579                 GOTO(out_root, err);
 580         }
 581
 582         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 583         if (sbi->ll_checksum_set) {
 584                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 585                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 586                                          sizeof(checksum), &checksum, NULL);
 587                 if (err) {
 588                         CERROR("%s: Set checksum failed: rc = %d\n",
 589                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 590                         GOTO(out_root, err);
 591                 }
 592         }
 593         cl_sb_init(sb);
 594
 595         sb->s_root = d_make_root(root);
 596         if (sb->s_root == NULL) {
 597                 CERROR("%s: can't make root dentry\n",
 598                         ll_get_fsname(sb, NULL, 0));
 599                 GOTO(out_root, err = -ENOMEM);
 600         }
 601 #ifdef HAVE_DCACHE_LOCK
 602         sb->s_root->d_op = &ll_d_ops;
 603 #endif
 604
 605         sbi->ll_sdev_orig = sb->s_dev;
 606
 607         /* We set sb->s_dev equal on all lustre clients in order to support
 608          * NFS export clustering.  NFSD requires that the FSID be the same
 609          * on all clients. */
 610         /* s_dev is also used in lt_compare() to compare two fs, but that is
 611          * only a node-local comparison. */
 612         uuid = obd_get_uuid(sbi->ll_md_exp);
 613         if (uuid != NULL)
 614                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 615
 616         if (data != NULL)
 617                 OBD_FREE_PTR(data);
 618         if (osfs != NULL)
 619                 OBD_FREE_PTR(osfs);
 620
 621         if (sbi->ll_dt_obd) {
 622                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 623                                         &sbi->ll_dt_obd->obd_kset.kobj,
 624                                         sbi->ll_dt_obd->obd_type->typ_name);
 625                 if (err < 0) {
 626                         CERROR("%s: could not register %s in llite: rc = %d\n",
 627                                dt, ll_get_fsname(sb, NULL, 0), err);
 628                         err = 0;
 629                 }
 630         }
 631
 632         if (sbi->ll_md_obd) {
 633                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 634                                         &sbi->ll_md_obd->obd_kset.kobj,
 635                                         sbi->ll_md_obd->obd_type->typ_name);
 636                 if (err < 0) {
 637                         CERROR("%s: could not register %s in llite: rc = %d\n",
 638                                md, ll_get_fsname(sb, NULL, 0), err);
 639                         err = 0;
 640                 }
 641         }
 642
 643         RETURN(err);
 644 out_root:
 645         if (root)
 646                 iput(root);
 647 out_lock_cn_cb:
 648         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 649 out_dt:
 650         obd_disconnect(sbi->ll_dt_exp);
 651         sbi->ll_dt_exp = NULL;
 652         sbi->ll_dt_obd = NULL;
 653 out_md_fid:
 654         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 655 out_md:
 656         obd_disconnect(sbi->ll_md_exp);
 657         sbi->ll_md_exp = NULL;
 658         sbi->ll_md_obd = NULL;
 659 out:
 660         if (data != NULL)
 661                 OBD_FREE_PTR(data);
 662         if (osfs != NULL)
 663                 OBD_FREE_PTR(osfs);
 664         return err;
 665 }
 666
 667 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 668 {
 669         int size, rc;
 670
 671         size = sizeof(*lmmsize);
 672         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 673                           KEY_MAX_EASIZE, &size, lmmsize);
 674         if (rc != 0) {
 675                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 676                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 677                 RETURN(rc);
 678         }
 679
 680         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 681
 682         size = sizeof(int);
 683         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 684                           KEY_MAX_EASIZE, &size, lmmsize);
 685         if (rc)
 686                 CERROR("Get max mdsize error rc %d\n", rc);
 687
 688         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 689
 690         RETURN(rc);
 691 }
 692
 693 /**
 694  * Get the value of the default_easize parameter.
 695  *
 696  * \see client_obd::cl_default_mds_easize
 697  *
 698  * \param[in] sbi       superblock info for this filesystem
 699  * \param[out] lmmsize  pointer to storage location for value
 700  *
 701  * \retval 0            on success
 702  * \retval negative     negated errno on failure
 703  */
 704 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 705 {
 706         int size, rc;
 707
 708         size = sizeof(int);
 709         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 710                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 711         if (rc)
 712                 CERROR("Get default mdsize error rc %d\n", rc);
 713
 714         RETURN(rc);
 715 }
 716
 717 /**
 718  * Set the default_easize parameter to the given value.
 719  *
 720  * \see client_obd::cl_default_mds_easize
 721  *
 722  * \param[in] sbi       superblock info for this filesystem
 723  * \param[in] lmmsize   the size to set
 724  *
 725  * \retval 0            on success
 726  * \retval negative     negated errno on failure
 727  */
 728 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 729 {
 730         int rc;
 731
 732         if (lmmsize < sizeof(struct lov_mds_md) ||
 733             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 734                 return -EINVAL;
 735
 736         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 737                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 738                                 sizeof(int), &lmmsize, NULL);
 739
 740         RETURN(rc);
 741 }
 742
 743 static void client_common_put_super(struct super_block *sb)
 744 {
 745         struct ll_sb_info *sbi = ll_s2sbi(sb);
 746         ENTRY;
 747
 748         cl_sb_fini(sb);
 749
 750         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 751         obd_disconnect(sbi->ll_dt_exp);
 752         sbi->ll_dt_exp = NULL;
 753
 754         ll_debugfs_unregister_super(sb);
 755
 756         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 757         obd_disconnect(sbi->ll_md_exp);
 758         sbi->ll_md_exp = NULL;
 759
 760         EXIT;
 761 }
 762
 763 void ll_kill_super(struct super_block *sb)
 764 {
 765         struct ll_sb_info *sbi;
 766         ENTRY;
 767
 768         /* not init sb ?*/
 769         if (!(sb->s_flags & MS_ACTIVE))
 770                 return;
 771
 772         sbi = ll_s2sbi(sb);
 773         /* we need restore s_dev from changed for clustred NFS before put_super
 774          * because new kernels have cached s_dev and change sb->s_dev in
 775          * put_super not affected real removing devices */
 776         if (sbi) {
 777                 sb->s_dev = sbi->ll_sdev_orig;
 778                 sbi->ll_umounting = 1;
 779
 780                 /* wait running statahead threads to quit */
 781                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 782                         set_current_state(TASK_UNINTERRUPTIBLE);
 783                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 784                 }
 785         }
 786
 787         EXIT;
 788 }
 789
 790 static inline int ll_set_opt(const char *opt, char *data, int fl)
 791 {
 792         if (strncmp(opt, data, strlen(opt)) != 0)
 793                 return 0;
 794         else
 795                 return fl;
 796 }
 797
 798 /* non-client-specific mount options are parsed in lmd_parse */
 799 static int ll_options(char *options, struct ll_sb_info *sbi)
 800 {
 801         int tmp;
 802         char *s1 = options, *s2;
 803         int *flags = &sbi->ll_flags;
 804         ENTRY;
 805
 806         if (!options)
 807                 RETURN(0);
 808
 809         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 810
 811         while (*s1) {
 812                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 813                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 829                 if (tmp) {
 830                         *flags &= ~tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags |= tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 839                 if (tmp) {
 840                         *flags &= ~tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("context", s1, 1);
 844                 if (tmp)
 845                         goto next;
 846                 tmp = ll_set_opt("fscontext", s1, 1);
 847                 if (tmp)
 848                         goto next;
 849                 tmp = ll_set_opt("defcontext", s1, 1);
 850                 if (tmp)
 851                         goto next;
 852                 tmp = ll_set_opt("rootcontext", s1, 1);
 853                 if (tmp)
 854                         goto next;
 855                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865
 866                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         sbi->ll_checksum_set = 1;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         sbi->ll_checksum_set = 1;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 894                 if (tmp) {
 895                         *flags &= ~tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 904                 if (tmp) {
 905                         *flags |= tmp;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 909                 if (tmp) {
 910                         *flags &= ~tmp;
 911                         goto next;
 912                 }
 913                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 914                 if (tmp) {
 915                         *flags |= tmp;
 916                         goto next;
 917                 }
 918                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 919                                    s1);
 920                 RETURN(-EINVAL);
 921
 922 next:
 923                 /* Find next opt */
 924                 s2 = strchr(s1, ',');
 925                 if (s2 == NULL)
 926                         break;
 927                 s1 = s2 + 1;
 928         }
 929         RETURN(0);
 930 }
 931
 932 void ll_lli_init(struct ll_inode_info *lli)
 933 {
 934         lli->lli_inode_magic = LLI_INODE_MAGIC;
 935         lli->lli_flags = 0;
 936         spin_lock_init(&lli->lli_lock);
 937         lli->lli_posix_acl = NULL;
 938         /* Do not set lli_fid, it has been initialized already. */
 939         fid_zero(&lli->lli_pfid);
 940         lli->lli_mds_read_och = NULL;
 941         lli->lli_mds_write_och = NULL;
 942         lli->lli_mds_exec_och = NULL;
 943         lli->lli_open_fd_read_count = 0;
 944         lli->lli_open_fd_write_count = 0;
 945         lli->lli_open_fd_exec_count = 0;
 946         mutex_init(&lli->lli_och_mutex);
 947         spin_lock_init(&lli->lli_agl_lock);
 948         spin_lock_init(&lli->lli_layout_lock);
 949         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 950         lli->lli_clob = NULL;
 951
 952         init_rwsem(&lli->lli_xattrs_list_rwsem);
 953         mutex_init(&lli->lli_xattrs_enq_lock);
 954
 955         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 956         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 957                 mutex_init(&lli->lli_readdir_mutex);
 958                 lli->lli_opendir_key = NULL;
 959                 lli->lli_sai = NULL;
 960                 spin_lock_init(&lli->lli_sa_lock);
 961                 lli->lli_opendir_pid = 0;
 962                 lli->lli_sa_enabled = 0;
 963                 lli->lli_def_stripe_offset = -1;
 964                 init_rwsem(&lli->lli_lsm_sem);
 965         } else {
 966                 mutex_init(&lli->lli_size_mutex);
 967                 lli->lli_symlink_name = NULL;
 968                 init_rwsem(&lli->lli_trunc_sem);
 969                 range_lock_tree_init(&lli->lli_write_tree);
 970                 init_rwsem(&lli->lli_glimpse_sem);
 971                 lli->lli_glimpse_time = ktime_set(0, 0);
 972                 INIT_LIST_HEAD(&lli->lli_agl_list);
 973                 lli->lli_agl_index = 0;
 974                 lli->lli_async_rc = 0;
 975                 spin_lock_init(&lli->lli_heat_lock);
 976                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
 977                 lli->lli_heat_flags = 0;
 978         }
 979         mutex_init(&lli->lli_layout_mutex);
 980         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 981 }
 982
 983 #define MAX_STRING_SIZE 128
 984
 985 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 986
 987 #define LSI_BDI_INITIALIZED     0x00400000
 988
 989 #ifndef HAVE_BDI_CAP_MAP_COPY
 990 # define BDI_CAP_MAP_COPY       0
 991 #endif
 992
 993 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 994 {
 995         struct  lustre_sb_info *lsi = s2lsi(sb);
 996         char buf[MAX_STRING_SIZE];
 997         va_list args;
 998         int err;
 999
1000         err = bdi_init(&lsi->lsi_bdi);
1001         if (err)
1002                 return err;
1003
1004         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1005         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1006         lsi->lsi_bdi.name = "lustre";
1007         va_start(args, fmt);
1008         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1009         va_end(args);
1010         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1011         va_end(args);
1012         if (!err)
1013                 sb->s_bdi = &lsi->lsi_bdi;
1014
1015         return err;
1016 }
1017 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1018
1019 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1020 {
1021         struct  lustre_profile *lprof = NULL;
1022         struct  lustre_sb_info *lsi = s2lsi(sb);
1023         struct  ll_sb_info *sbi = NULL;
1024         char    *dt = NULL, *md = NULL;
1025         char    *profilenm = get_profile_name(sb);
1026         struct config_llog_instance *cfg;
1027         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1028         const int instlen = 16 + 2;
1029         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1030         char name[MAX_STRING_SIZE];
1031         int md_len = 0;
1032         int dt_len = 0;
1033         char *ptr;
1034         int len;
1035         int err;
1036
1037         ENTRY;
1038         /* for ASLR, to map between cfg_instance and hashed ptr */
1039         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1040                profilenm, cfg_instance, sb);
1041
1042         try_module_get(THIS_MODULE);
1043
1044         OBD_ALLOC_PTR(cfg);
1045         if (cfg == NULL)
1046                 GOTO(out_free_cfg, err = -ENOMEM);
1047
1048         /* client additional sb info */
1049         lsi->lsi_llsbi = sbi = ll_init_sbi();
1050         if (!sbi)
1051                 GOTO(out_free_cfg, err = -ENOMEM);
1052
1053         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1054         if (err)
1055                 GOTO(out_free_cfg, err);
1056
1057         err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1058         if (err)
1059                 GOTO(out_free_cfg, err);
1060
1061 #ifndef HAVE_DCACHE_LOCK
1062         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1063         sb->s_d_op = &ll_d_ops;
1064 #endif
1065         /* Get fsname */
1066         len = strlen(profilenm);
1067         ptr = strrchr(profilenm, '-');
1068         if (ptr && (strcmp(ptr, "-client") == 0))
1069                 len -= 7;
1070
1071         /* Mount info */
1072         snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1073                  profilenm, cfg_instance);
1074
1075         /* Call ll_debugfs_register_super() before lustre_process_log()
1076          * so that "llite.*.*" params can be processed correctly.
1077          */
1078         err = ll_debugfs_register_super(sb, name);
1079         if (err < 0) {
1080                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1081                        ll_get_fsname(sb, NULL, 0), err);
1082                 err = 0;
1083         }
1084
1085         /* The cfg_instance is a value unique to this super, in case some
1086          * joker tries to mount the same fs at two mount points.
1087          */
1088         cfg->cfg_instance = cfg_instance;
1089         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1090         cfg->cfg_callback = class_config_llog_handler;
1091         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1092         /* set up client obds */
1093         err = lustre_process_log(sb, profilenm, cfg);
1094         if (err < 0)
1095                 GOTO(out_debugfs, err);
1096
1097         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1098         lprof = class_get_profile(profilenm);
1099         if (lprof == NULL) {
1100                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1101                                    " read from the MGS.  Does that filesystem "
1102                                    "exist?\n", profilenm);
1103                 GOTO(out_debugfs, err = -EINVAL);
1104         }
1105         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1106                lprof->lp_md, lprof->lp_dt);
1107
1108         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1109         OBD_ALLOC(dt, dt_len);
1110         if (!dt)
1111                 GOTO(out_profile, err = -ENOMEM);
1112         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1113
1114         md_len = strlen(lprof->lp_md) + instlen + 2;
1115         OBD_ALLOC(md, md_len);
1116         if (!md)
1117                 GOTO(out_free_dt, err = -ENOMEM);
1118         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1119
1120         /* connections, registrations, sb setup */
1121         err = client_common_fill_super(sb, md, dt, mnt);
1122         if (err < 0)
1123                 GOTO(out_free_md, err);
1124
1125         sbi->ll_client_common_fill_super_succeeded = 1;
1126
1127 out_free_md:
1128         if (md)
1129                 OBD_FREE(md, md_len);
1130 out_free_dt:
1131         if (dt)
1132                 OBD_FREE(dt, dt_len);
1133 out_profile:
1134         if (lprof)
1135                 class_put_profile(lprof);
1136 out_debugfs:
1137         if (err < 0)
1138                 ll_debugfs_unregister_super(sb);
1139 out_free_cfg:
1140         if (cfg)
1141                 OBD_FREE_PTR(cfg);
1142
1143         if (err)
1144                 ll_put_super(sb);
1145         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1146                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1147         RETURN(err);
1148 } /* ll_fill_super */
1149
1150 void ll_put_super(struct super_block *sb)
1151 {
1152         struct config_llog_instance cfg, params_cfg;
1153         struct obd_device *obd;
1154         struct lustre_sb_info *lsi = s2lsi(sb);
1155         struct ll_sb_info *sbi = ll_s2sbi(sb);
1156         char *profilenm = get_profile_name(sb);
1157         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1158         long ccc_count;
1159         int next, force = 1, rc = 0;
1160         ENTRY;
1161
1162         if (!sbi)
1163                 GOTO(out_no_sbi, 0);
1164
1165         /* Should replace instance_id with something better for ASLR */
1166         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1167                profilenm, cfg_instance, sb);
1168
1169         cfg.cfg_instance = cfg_instance;
1170         lustre_end_log(sb, profilenm, &cfg);
1171
1172         params_cfg.cfg_instance = cfg_instance;
1173         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1174
1175         if (sbi->ll_md_exp) {
1176                 obd = class_exp2obd(sbi->ll_md_exp);
1177                 if (obd)
1178                         force = obd->obd_force;
1179         }
1180
1181         /* Wait for unstable pages to be committed to stable storage */
1182         if (force == 0) {
1183                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1184                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1185                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1186                         &lwi);
1187         }
1188
1189         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1190         if (force == 0 && rc != -EINTR)
1191                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1192
1193         /* We need to set force before the lov_disconnect in
1194            lustre_common_put_super, since l_d cleans up osc's as well. */
1195         if (force) {
1196                 next = 0;
1197                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1198                                                      &next)) != NULL) {
1199                         obd->obd_force = force;
1200                 }
1201         }
1202
1203         if (sbi->ll_client_common_fill_super_succeeded) {
1204                 /* Only if client_common_fill_super succeeded */
1205                 client_common_put_super(sb);
1206         }
1207
1208         next = 0;
1209         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1210                 class_manual_cleanup(obd);
1211         }
1212
1213         if (sbi->ll_flags & LL_SBI_VERBOSE)
1214                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1215
1216         if (profilenm)
1217                 class_del_profile(profilenm);
1218
1219 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1220         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1221                 bdi_destroy(&lsi->lsi_bdi);
1222                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1223         }
1224 #endif
1225
1226         ll_free_sbi(sb);
1227         lsi->lsi_llsbi = NULL;
1228 out_no_sbi:
1229         lustre_common_put_super(sb);
1230
1231         cl_env_cache_purge(~0);
1232
1233         module_put(THIS_MODULE);
1234
1235         EXIT;
1236 } /* client_put_super */
1237
1238 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1239 {
1240         struct inode *inode = NULL;
1241
1242         /* NOTE: we depend on atomic igrab() -bzzz */
1243         lock_res_and_lock(lock);
1244         if (lock->l_resource->lr_lvb_inode) {
1245                 struct ll_inode_info * lli;
1246                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1247                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1248                         inode = igrab(lock->l_resource->lr_lvb_inode);
1249                 } else {
1250                         inode = lock->l_resource->lr_lvb_inode;
1251                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1252                                          D_WARNING, lock, "lr_lvb_inode %p is "
1253                                          "bogus: magic %08x",
1254                                          lock->l_resource->lr_lvb_inode,
1255                                          lli->lli_inode_magic);
1256                         inode = NULL;
1257                 }
1258         }
1259         unlock_res_and_lock(lock);
1260         return inode;
1261 }
1262
1263 void ll_dir_clear_lsm_md(struct inode *inode)
1264 {
1265         struct ll_inode_info *lli = ll_i2info(inode);
1266
1267         LASSERT(S_ISDIR(inode->i_mode));
1268
1269         if (lli->lli_lsm_md != NULL) {
1270                 lmv_free_memmd(lli->lli_lsm_md);
1271                 lli->lli_lsm_md = NULL;
1272         }
1273 }
1274
1275 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1276                                       const struct lu_fid *fid,
1277                                       struct lustre_md *md)
1278 {
1279         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1280         struct mdt_body         *body = md->body;
1281         struct inode            *inode;
1282         ino_t                   ino;
1283         ENTRY;
1284
1285         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1286         inode = iget_locked(sb, ino);
1287         if (inode == NULL) {
1288                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1289                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1290                 RETURN(ERR_PTR(-ENOENT));
1291         }
1292
1293         if (inode->i_state & I_NEW) {
1294                 struct ll_inode_info *lli = ll_i2info(inode);
1295                 struct lmv_stripe_md *lsm = md->lmv;
1296
1297                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1298                                 (body->mbo_mode & S_IFMT);
1299                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1300                          PFID(fid));
1301
1302                 inode->i_mtime.tv_sec = 0;
1303                 inode->i_atime.tv_sec = 0;
1304                 inode->i_ctime.tv_sec = 0;
1305                 inode->i_rdev = 0;
1306
1307 #ifdef HAVE_BACKING_DEV_INFO
1308                 /* initializing backing dev info. */
1309                 inode->i_mapping->backing_dev_info =
1310                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1311 #endif
1312                 inode->i_op = &ll_dir_inode_operations;
1313                 inode->i_fop = &ll_dir_operations;
1314                 lli->lli_fid = *fid;
1315                 ll_lli_init(lli);
1316
1317                 LASSERT(lsm != NULL);
1318                 /* master object FID */
1319                 lli->lli_pfid = body->mbo_fid1;
1320                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1321                        lli, PFID(fid), PFID(&lli->lli_pfid));
1322                 unlock_new_inode(inode);
1323         }
1324
1325         RETURN(inode);
1326 }
1327
1328 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1329 {
1330         struct lu_fid *fid;
1331         struct lmv_stripe_md *lsm = md->lmv;
1332         struct ll_inode_info *lli = ll_i2info(inode);
1333         int i;
1334
1335         LASSERT(lsm != NULL);
1336
1337         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1338                 ll_get_fsname(inode->i_sb, NULL, 0),
1339                 PFID(&lli->lli_fid));
1340         lsm_md_dump(D_INODE, lsm);
1341
1342         /* XXX sigh, this lsm_root initialization should be in
1343          * LMV layer, but it needs ll_iget right now, so we
1344          * put this here right now. */
1345         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1346                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1347                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1348                 /* Unfortunately ll_iget will call ll_update_inode,
1349                  * where the initialization of slave inode is slightly
1350                  * different, so it reset lsm_md to NULL to avoid
1351                  * initializing lsm for slave inode. */
1352                 lsm->lsm_md_oinfo[i].lmo_root =
1353                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1354                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1355                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1356
1357                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1358                         while (i-- > 0) {
1359                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1360                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1361                         }
1362                         return rc;
1363                 }
1364         }
1365
1366         lli->lli_lsm_md = lsm;
1367
1368         return 0;
1369 }
1370
1371 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1372 {
1373         struct ll_inode_info *lli = ll_i2info(inode);
1374         struct lmv_stripe_md *lsm = md->lmv;
1375         int rc = 0;
1376
1377         ENTRY;
1378
1379         LASSERT(S_ISDIR(inode->i_mode));
1380         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1381                PFID(ll_inode2fid(inode)));
1382
1383         /*
1384          * no striped information from request, lustre_md from req does not
1385          * include stripeEA, see ll_md_setattr()
1386          */
1387         if (!lsm)
1388                 RETURN(0);
1389
1390         /*
1391          * normally dir layout doesn't change, only take read lock to check
1392          * that to avoid blocking other MD operations.
1393          */
1394         if (lli->lli_lsm_md)
1395                 down_read(&lli->lli_lsm_sem);
1396         else
1397                 down_write(&lli->lli_lsm_sem);
1398
1399         /*
1400          * if dir layout mismatch, check whether version is increased, which
1401          * means layout is changed, this happens in dir migration and lfsck.
1402          */
1403         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1404                 if (lsm->lsm_md_layout_version <=
1405                     lli->lli_lsm_md->lsm_md_layout_version) {
1406                         CERROR("%s: "DFID" dir layout mismatch:\n",
1407                                 ll_get_fsname(inode->i_sb, NULL, 0),
1408                                 PFID(&lli->lli_fid));
1409                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1410                         lsm_md_dump(D_ERROR, lsm);
1411                         GOTO(unlock, rc = -EINVAL);
1412                 }
1413
1414                 /* layout changed, switch to write lock */
1415                 up_read(&lli->lli_lsm_sem);
1416                 down_write(&lli->lli_lsm_sem);
1417                 ll_dir_clear_lsm_md(inode);
1418         }
1419
1420         /* set directory layout */
1421         if (!lli->lli_lsm_md) {
1422                 struct cl_attr  *attr;
1423
1424                 rc = ll_init_lsm_md(inode, md);
1425                 up_write(&lli->lli_lsm_sem);
1426                 if (rc != 0)
1427                         RETURN(rc);
1428
1429                 /* set md->lmv to NULL, so the following free lustre_md
1430                  * will not free this lsm */
1431                 md->lmv = NULL;
1432
1433                 /*
1434                  * md_merge_attr() may take long, since lsm is already set,
1435                  * switch to read lock.
1436                  */
1437                 down_read(&lli->lli_lsm_sem);
1438
1439                 OBD_ALLOC_PTR(attr);
1440                 if (attr == NULL)
1441                         GOTO(unlock, rc = -ENOMEM);
1442
1443                 /* validate the lsm */
1444                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1445                                    ll_md_blocking_ast);
1446                 if (rc != 0) {
1447                         OBD_FREE_PTR(attr);
1448                         GOTO(unlock, rc);
1449                 }
1450
1451                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1452                         md->body->mbo_nlink = attr->cat_nlink;
1453                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1454                         md->body->mbo_size = attr->cat_size;
1455                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1456                         md->body->mbo_atime = attr->cat_atime;
1457                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1458                         md->body->mbo_ctime = attr->cat_ctime;
1459                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1460                         md->body->mbo_mtime = attr->cat_mtime;
1461
1462                 OBD_FREE_PTR(attr);
1463         }
1464 unlock:
1465         up_read(&lli->lli_lsm_sem);
1466
1467         RETURN(rc);
1468 }
1469
1470 void ll_clear_inode(struct inode *inode)
1471 {
1472         struct ll_inode_info *lli = ll_i2info(inode);
1473         struct ll_sb_info *sbi = ll_i2sbi(inode);
1474         ENTRY;
1475
1476         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1477                PFID(ll_inode2fid(inode)), inode);
1478
1479         if (S_ISDIR(inode->i_mode)) {
1480                 /* these should have been cleared in ll_file_release */
1481                 LASSERT(lli->lli_opendir_key == NULL);
1482                 LASSERT(lli->lli_sai == NULL);
1483                 LASSERT(lli->lli_opendir_pid == 0);
1484         }
1485
1486         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1487
1488         LASSERT(!lli->lli_open_fd_write_count);
1489         LASSERT(!lli->lli_open_fd_read_count);
1490         LASSERT(!lli->lli_open_fd_exec_count);
1491
1492         if (lli->lli_mds_write_och)
1493                 ll_md_real_close(inode, FMODE_WRITE);
1494         if (lli->lli_mds_exec_och)
1495                 ll_md_real_close(inode, FMODE_EXEC);
1496         if (lli->lli_mds_read_och)
1497                 ll_md_real_close(inode, FMODE_READ);
1498
1499         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1500                 OBD_FREE(lli->lli_symlink_name,
1501                          strlen(lli->lli_symlink_name) + 1);
1502                 lli->lli_symlink_name = NULL;
1503         }
1504
1505         ll_xattr_cache_destroy(inode);
1506
1507 #ifdef CONFIG_FS_POSIX_ACL
1508         forget_all_cached_acls(inode);
1509         if (lli->lli_posix_acl) {
1510                 posix_acl_release(lli->lli_posix_acl);
1511                 lli->lli_posix_acl = NULL;
1512         }
1513 #endif
1514         lli->lli_inode_magic = LLI_INODE_DEAD;
1515
1516         if (S_ISDIR(inode->i_mode))
1517                 ll_dir_clear_lsm_md(inode);
1518         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1519                 LASSERT(list_empty(&lli->lli_agl_list));
1520
1521         /*
1522          * XXX This has to be done before lsm is freed below, because
1523          * cl_object still uses inode lsm.
1524          */
1525         cl_inode_fini(inode);
1526
1527         EXIT;
1528 }
1529
1530 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1531 {
1532         struct lustre_md md;
1533         struct inode *inode = dentry->d_inode;
1534         struct ll_sb_info *sbi = ll_i2sbi(inode);
1535         struct ptlrpc_request *request = NULL;
1536         int rc, ia_valid;
1537         ENTRY;
1538
1539         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1540                                      LUSTRE_OPC_ANY, NULL);
1541         if (IS_ERR(op_data))
1542                 RETURN(PTR_ERR(op_data));
1543
1544         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1545         if (rc) {
1546                 ptlrpc_req_finished(request);
1547                 if (rc == -ENOENT) {
1548                         clear_nlink(inode);
1549                         /* Unlinked special device node? Or just a race?
1550                          * Pretend we done everything. */
1551                         if (!S_ISREG(inode->i_mode) &&
1552                             !S_ISDIR(inode->i_mode)) {
1553                                 ia_valid = op_data->op_attr.ia_valid;
1554                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1555                                 rc = simple_setattr(dentry, &op_data->op_attr);
1556                                 op_data->op_attr.ia_valid = ia_valid;
1557                         }
1558                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1559                         CERROR("md_setattr fails: rc = %d\n", rc);
1560                 }
1561                 RETURN(rc);
1562         }
1563
1564         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1565                               sbi->ll_md_exp, &md);
1566         if (rc) {
1567                 ptlrpc_req_finished(request);
1568                 RETURN(rc);
1569         }
1570
1571         ia_valid = op_data->op_attr.ia_valid;
1572         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1573          * cache is not cleared yet. */
1574         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1575         if (S_ISREG(inode->i_mode))
1576                 inode_lock(inode);
1577         rc = simple_setattr(dentry, &op_data->op_attr);
1578         if (S_ISREG(inode->i_mode))
1579                 inode_unlock(inode);
1580         op_data->op_attr.ia_valid = ia_valid;
1581
1582         rc = ll_update_inode(inode, &md);
1583         ptlrpc_req_finished(request);
1584
1585         RETURN(rc);
1586 }
1587
1588 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1589  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1590  * keep these values until such a time that objects are allocated for it.
1591  * We do the MDS operations first, as it is checking permissions for us.
1592  * We don't to the MDS RPC if there is nothing that we want to store there,
1593  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1594  * going to do an RPC anyways.
1595  *
1596  * If we are doing a truncate, we will send the mtime and ctime updates
1597  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1598  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1599  * at the same time.
1600  *
1601  * In case of HSMimport, we only set attr on MDS.
1602  */
1603 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1604                    enum op_xvalid xvalid, bool hsm_import)
1605 {
1606         struct inode *inode = dentry->d_inode;
1607         struct ll_inode_info *lli = ll_i2info(inode);
1608         struct md_op_data *op_data = NULL;
1609         int rc = 0;
1610         ENTRY;
1611
1612         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1613                "valid %x, hsm_import %d\n",
1614                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1615                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1616                hsm_import);
1617
1618         if (attr->ia_valid & ATTR_SIZE) {
1619                 /* Check new size against VFS/VM file size limit and rlimit */
1620                 rc = inode_newsize_ok(inode, attr->ia_size);
1621                 if (rc)
1622                         RETURN(rc);
1623
1624                 /* The maximum Lustre file size is variable, based on the
1625                  * OST maximum object size and number of stripes.  This
1626                  * needs another check in addition to the VFS check above. */
1627                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1628                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1629                                PFID(&lli->lli_fid), attr->ia_size,
1630                                ll_file_maxbytes(inode));
1631                         RETURN(-EFBIG);
1632                 }
1633
1634                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1635         }
1636
1637         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1638         if (attr->ia_valid & TIMES_SET_FLAGS) {
1639                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1640                     !cfs_capable(CFS_CAP_FOWNER))
1641                         RETURN(-EPERM);
1642         }
1643
1644         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1645         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1646              (attr->ia_valid & ATTR_CTIME)) {
1647                 attr->ia_ctime = current_time(inode);
1648                 xvalid |= OP_XVALID_CTIME_SET;
1649         }
1650         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1651             (attr->ia_valid & ATTR_ATIME)) {
1652                 attr->ia_atime = current_time(inode);
1653                 attr->ia_valid |= ATTR_ATIME_SET;
1654         }
1655         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1656             (attr->ia_valid & ATTR_MTIME)) {
1657                 attr->ia_mtime = current_time(inode);
1658                 attr->ia_valid |= ATTR_MTIME_SET;
1659         }
1660
1661         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1662                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1663                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1664                        ktime_get_real_seconds());
1665
1666         if (S_ISREG(inode->i_mode)) {
1667                 if (attr->ia_valid & ATTR_SIZE)
1668                         inode_dio_write_done(inode);
1669                 inode_unlock(inode);
1670         }
1671
1672         /* We always do an MDS RPC, even if we're only changing the size;
1673          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1674
1675         OBD_ALLOC_PTR(op_data);
1676         if (op_data == NULL)
1677                 GOTO(out, rc = -ENOMEM);
1678
1679         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1680                 /* If we are changing file size, file content is
1681                  * modified, flag it.
1682                  */
1683                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1684                 op_data->op_bias |= MDS_DATA_MODIFIED;
1685                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1686         }
1687
1688         if (attr->ia_valid & ATTR_FILE) {
1689                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1690
1691                 if (fd->fd_lease_och)
1692                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1693         }
1694
1695         op_data->op_attr = *attr;
1696         op_data->op_xvalid = xvalid;
1697
1698         rc = ll_md_setattr(dentry, op_data);
1699         if (rc)
1700                 GOTO(out, rc);
1701
1702         if (!S_ISREG(inode->i_mode) || hsm_import)
1703                 GOTO(out, rc = 0);
1704
1705         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1706                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1707             xvalid & OP_XVALID_CTIME_SET) {
1708                 /* For truncate and utimes sending attributes to OSTs, setting
1709                  * mtime/atime to the past will be performed under PW [0:EOF]
1710                  * extent lock (new_size:EOF for truncate).  It may seem
1711                  * excessive to send mtime/atime updates to OSTs when not
1712                  * setting times to past, but it is necessary due to possible
1713                  * time de-synchronization between MDT inode and OST objects
1714                  */
1715                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1716         }
1717
1718         /* If the file was restored, it needs to set dirty flag.
1719          *
1720          * We've already sent MDS_DATA_MODIFIED flag in
1721          * ll_md_setattr() for truncate. However, the MDT refuses to
1722          * set the HS_DIRTY flag on released files, so we have to set
1723          * it again if the file has been restored. Please check how
1724          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1725          *
1726          * Please notice that if the file is not released, the previous
1727          * MDS_DATA_MODIFIED has taken effect and usually
1728          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1729          * This way we can save an RPC for common open + trunc
1730          * operation. */
1731         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1732                 struct hsm_state_set hss = {
1733                         .hss_valid = HSS_SETMASK,
1734                         .hss_setmask = HS_DIRTY,
1735                 };
1736                 int rc2;
1737
1738                 rc2 = ll_hsm_state_set(inode, &hss);
1739                 /* truncate and write can happen at the same time, so that
1740                  * the file can be set modified even though the file is not
1741                  * restored from released state, and ll_hsm_state_set() is
1742                  * not applicable for the file, and rc2 < 0 is normal in this
1743                  * case. */
1744                 if (rc2 < 0)
1745                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1746                                PFID(ll_inode2fid(inode)), rc2);
1747         }
1748
1749         EXIT;
1750 out:
1751         if (op_data != NULL)
1752                 ll_finish_md_op_data(op_data);
1753
1754         if (S_ISREG(inode->i_mode)) {
1755                 inode_lock(inode);
1756                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1757                         inode_dio_wait(inode);
1758                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1759                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1760                  * inode flags, so there is a gap where S_NOSEC is not set.
1761                  * This can cause a writer to take the i_mutex unnecessarily,
1762                  * but this is safe to do and should be rare. */
1763                 inode_has_no_xattr(inode);
1764         }
1765
1766         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1767                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1768
1769         return rc;
1770 }
1771
1772 int ll_setattr(struct dentry *de, struct iattr *attr)
1773 {
1774         int mode = de->d_inode->i_mode;
1775         enum op_xvalid xvalid = 0;
1776
1777         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1778                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1779                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1780
1781         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1782                                (ATTR_SIZE|ATTR_MODE)) &&
1783             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1784              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1785               !(attr->ia_mode & S_ISGID))))
1786                 attr->ia_valid |= ATTR_FORCE;
1787
1788         if ((attr->ia_valid & ATTR_MODE) &&
1789             (mode & S_ISUID) &&
1790             !(attr->ia_mode & S_ISUID) &&
1791             !(attr->ia_valid & ATTR_KILL_SUID))
1792                 attr->ia_valid |= ATTR_KILL_SUID;
1793
1794         if ((attr->ia_valid & ATTR_MODE) &&
1795             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1796             !(attr->ia_mode & S_ISGID) &&
1797             !(attr->ia_valid & ATTR_KILL_SGID))
1798                 attr->ia_valid |= ATTR_KILL_SGID;
1799
1800         return ll_setattr_raw(de, attr, xvalid, false);
1801 }
1802
1803 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1804                        u32 flags)
1805 {
1806         struct obd_statfs obd_osfs = { 0 };
1807         time64_t max_age;
1808         int rc;
1809
1810         ENTRY;
1811         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1812
1813         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1814         if (rc)
1815                 RETURN(rc);
1816
1817         osfs->os_type = LL_SUPER_MAGIC;
1818
1819         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1820               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1821
1822         if (osfs->os_state & OS_STATE_SUM)
1823                 GOTO(out, rc);
1824
1825         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1826                 flags |= OBD_STATFS_NODELAY;
1827
1828         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1829         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1830                 GOTO(out, rc = 0);
1831
1832         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1833                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1834                obd_osfs.os_files);
1835
1836         osfs->os_bsize = obd_osfs.os_bsize;
1837         osfs->os_blocks = obd_osfs.os_blocks;
1838         osfs->os_bfree = obd_osfs.os_bfree;
1839         osfs->os_bavail = obd_osfs.os_bavail;
1840
1841         /* If we have _some_ OSTs, but don't have as many free objects on the
1842          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1843          * to compensate, so that the "inodes in use" number is correct.
1844          * This should be kept in sync with lod_statfs() behaviour.
1845          */
1846         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1847                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1848                                  obd_osfs.os_ffree;
1849                 osfs->os_ffree = obd_osfs.os_ffree;
1850         }
1851
1852 out:
1853         RETURN(rc);
1854 }
1855 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1856 {
1857         struct super_block *sb = de->d_sb;
1858         struct obd_statfs osfs;
1859         __u64 fsid = huge_encode_dev(sb->s_dev);
1860         int rc;
1861
1862         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1863         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1864
1865         /* Some amount of caching on the client is allowed */
1866         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1867         if (rc)
1868                 return rc;
1869
1870         statfs_unpack(sfs, &osfs);
1871
1872         /* We need to downshift for all 32-bit kernels, because we can't
1873          * tell if the kernel is being called via sys_statfs64() or not.
1874          * Stop before overflowing f_bsize - in which case it is better
1875          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1876         if (sizeof(long) < 8) {
1877                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1878                         sfs->f_bsize <<= 1;
1879
1880                         osfs.os_blocks >>= 1;
1881                         osfs.os_bfree >>= 1;
1882                         osfs.os_bavail >>= 1;
1883                 }
1884         }
1885
1886         sfs->f_blocks = osfs.os_blocks;
1887         sfs->f_bfree = osfs.os_bfree;
1888         sfs->f_bavail = osfs.os_bavail;
1889         sfs->f_fsid.val[0] = (__u32)fsid;
1890         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1891         return 0;
1892 }
1893
1894 void ll_inode_size_lock(struct inode *inode)
1895 {
1896         struct ll_inode_info *lli;
1897
1898         LASSERT(!S_ISDIR(inode->i_mode));
1899
1900         lli = ll_i2info(inode);
1901         mutex_lock(&lli->lli_size_mutex);
1902 }
1903
1904 void ll_inode_size_unlock(struct inode *inode)
1905 {
1906         struct ll_inode_info *lli;
1907
1908         lli = ll_i2info(inode);
1909         mutex_unlock(&lli->lli_size_mutex);
1910 }
1911
1912 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1913 {
1914         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1915         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1916                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1917         else
1918                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1919 }
1920
1921 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1922 {
1923         struct ll_inode_info *lli = ll_i2info(inode);
1924         struct mdt_body *body = md->body;
1925         struct ll_sb_info *sbi = ll_i2sbi(inode);
1926         int rc = 0;
1927
1928         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1929                 rc = cl_file_inode_init(inode, md);
1930                 if (rc)
1931                         return rc;
1932         }
1933
1934         if (S_ISDIR(inode->i_mode)) {
1935                 rc = ll_update_lsm_md(inode, md);
1936                 if (rc != 0)
1937                         return rc;
1938         }
1939
1940 #ifdef CONFIG_FS_POSIX_ACL
1941         if (body->mbo_valid & OBD_MD_FLACL) {
1942                 spin_lock(&lli->lli_lock);
1943                 if (lli->lli_posix_acl)
1944                         posix_acl_release(lli->lli_posix_acl);
1945                 lli->lli_posix_acl = md->posix_acl;
1946                 spin_unlock(&lli->lli_lock);
1947         }
1948 #endif
1949         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1950                                         sbi->ll_flags & LL_SBI_32BIT_API);
1951         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1952
1953         if (body->mbo_valid & OBD_MD_FLATIME) {
1954                 if (body->mbo_atime > inode->i_atime.tv_sec)
1955                         inode->i_atime.tv_sec = body->mbo_atime;
1956                 lli->lli_atime = body->mbo_atime;
1957         }
1958
1959         if (body->mbo_valid & OBD_MD_FLMTIME) {
1960                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1961                         CDEBUG(D_INODE,
1962                                "setting ino %lu mtime from %lld to %llu\n",
1963                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
1964                                body->mbo_mtime);
1965                         inode->i_mtime.tv_sec = body->mbo_mtime;
1966                 }
1967                 lli->lli_mtime = body->mbo_mtime;
1968         }
1969
1970         if (body->mbo_valid & OBD_MD_FLCTIME) {
1971                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1972                         inode->i_ctime.tv_sec = body->mbo_ctime;
1973                 lli->lli_ctime = body->mbo_ctime;
1974         }
1975
1976         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1977         if (body->mbo_valid & OBD_MD_FLFLAGS)
1978                 ll_update_inode_flags(inode, body->mbo_flags);
1979         if (body->mbo_valid & OBD_MD_FLMODE)
1980                 inode->i_mode = (inode->i_mode & S_IFMT) |
1981                                 (body->mbo_mode & ~S_IFMT);
1982
1983         if (body->mbo_valid & OBD_MD_FLTYPE)
1984                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1985                                 (body->mbo_mode & S_IFMT);
1986
1987         LASSERT(inode->i_mode != 0);
1988         if (S_ISREG(inode->i_mode))
1989                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1990                                        LL_MAX_BLKSIZE_BITS);
1991         else
1992                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1993
1994         if (body->mbo_valid & OBD_MD_FLUID)
1995                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1996         if (body->mbo_valid & OBD_MD_FLGID)
1997                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1998         if (body->mbo_valid & OBD_MD_FLPROJID)
1999                 lli->lli_projid = body->mbo_projid;
2000         if (body->mbo_valid & OBD_MD_FLNLINK)
2001                 set_nlink(inode, body->mbo_nlink);
2002         if (body->mbo_valid & OBD_MD_FLRDEV)
2003                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2004
2005         if (body->mbo_valid & OBD_MD_FLID) {
2006                 /* FID shouldn't be changed! */
2007                 if (fid_is_sane(&lli->lli_fid)) {
2008                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2009                                  "Trying to change FID "DFID
2010                                  " to the "DFID", inode "DFID"(%p)\n",
2011                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2012                                  PFID(ll_inode2fid(inode)), inode);
2013                 } else {
2014                         lli->lli_fid = body->mbo_fid1;
2015                 }
2016         }
2017
2018         LASSERT(fid_seq(&lli->lli_fid) != 0);
2019
2020         if (body->mbo_valid & OBD_MD_FLSIZE) {
2021                 i_size_write(inode, body->mbo_size);
2022
2023                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2024                        PFID(ll_inode2fid(inode)),
2025                        (unsigned long long)body->mbo_size);
2026
2027                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2028                         inode->i_blocks = body->mbo_blocks;
2029         }
2030
2031         if (body->mbo_valid & OBD_MD_TSTATE) {
2032                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2033                  * clear it when done to ensure to start again
2034                  * glimpsing updated attrs
2035                  */
2036                 if (body->mbo_t_state & MS_RESTORE)
2037                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2038                 else
2039                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2040         }
2041
2042         return 0;
2043 }
2044
2045 int ll_read_inode2(struct inode *inode, void *opaque)
2046 {
2047         struct lustre_md *md = opaque;
2048         struct ll_inode_info *lli = ll_i2info(inode);
2049         int     rc;
2050         ENTRY;
2051
2052         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2053                PFID(&lli->lli_fid), inode);
2054
2055         /* Core attributes from the MDS first.  This is a new inode, and
2056          * the VFS doesn't zero times in the core inode so we have to do
2057          * it ourselves.  They will be overwritten by either MDS or OST
2058          * attributes - we just need to make sure they aren't newer.
2059          */
2060         inode->i_mtime.tv_sec = 0;
2061         inode->i_atime.tv_sec = 0;
2062         inode->i_ctime.tv_sec = 0;
2063         inode->i_rdev = 0;
2064         rc = ll_update_inode(inode, md);
2065         if (rc != 0)
2066                 RETURN(rc);
2067
2068         /* OIDEBUG(inode); */
2069
2070 #ifdef HAVE_BACKING_DEV_INFO
2071         /* initializing backing dev info. */
2072         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2073 #endif
2074         if (S_ISREG(inode->i_mode)) {
2075                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2076                 inode->i_op = &ll_file_inode_operations;
2077                 inode->i_fop = sbi->ll_fop;
2078                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2079                 EXIT;
2080         } else if (S_ISDIR(inode->i_mode)) {
2081                 inode->i_op = &ll_dir_inode_operations;
2082                 inode->i_fop = &ll_dir_operations;
2083                 EXIT;
2084         } else if (S_ISLNK(inode->i_mode)) {
2085                 inode->i_op = &ll_fast_symlink_inode_operations;
2086                 EXIT;
2087         } else {
2088                 inode->i_op = &ll_special_inode_operations;
2089
2090                 init_special_inode(inode, inode->i_mode,
2091                                    inode->i_rdev);
2092
2093                 EXIT;
2094         }
2095
2096         return 0;
2097 }
2098
2099 void ll_delete_inode(struct inode *inode)
2100 {
2101         struct ll_inode_info *lli = ll_i2info(inode);
2102         struct address_space *mapping = &inode->i_data;
2103         unsigned long nrpages;
2104         ENTRY;
2105
2106         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2107                 /* It is last chance to write out dirty pages,
2108                  * otherwise we may lose data while umount */
2109                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2110
2111         truncate_inode_pages_final(mapping);
2112
2113         /* Workaround for LU-118: Note nrpages may not be totally updated when
2114          * truncate_inode_pages() returns, as there can be a page in the process
2115          * of deletion (inside __delete_from_page_cache()) in the specified
2116          * range. Thus mapping->nrpages can be non-zero when this function
2117          * returns even after truncation of the whole mapping.  Only do this if
2118          * npages isn't already zero.
2119          */
2120         nrpages = mapping->nrpages;
2121         if (nrpages) {
2122                 spin_lock_irq(&mapping->tree_lock);
2123                 nrpages = mapping->nrpages;
2124                 spin_unlock_irq(&mapping->tree_lock);
2125         } /* Workaround end */
2126
2127         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2128                  "see https://jira.whamcloud.com/browse/LU-118\n",
2129                  ll_get_fsname(inode->i_sb, NULL, 0),
2130                  PFID(ll_inode2fid(inode)), inode, nrpages);
2131
2132 #ifdef HAVE_SBOPS_EVICT_INODE
2133         ll_clear_inode(inode);
2134 #endif
2135         clear_inode(inode);
2136
2137         EXIT;
2138 }
2139
2140 int ll_iocontrol(struct inode *inode, struct file *file,
2141                  unsigned int cmd, unsigned long arg)
2142 {
2143         struct ll_sb_info *sbi = ll_i2sbi(inode);
2144         struct ptlrpc_request *req = NULL;
2145         int rc, flags = 0;
2146         ENTRY;
2147
2148         switch (cmd) {
2149         case FS_IOC_GETFLAGS: {
2150                 struct mdt_body *body;
2151                 struct md_op_data *op_data;
2152
2153                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2154                                              0, 0, LUSTRE_OPC_ANY,
2155                                              NULL);
2156                 if (IS_ERR(op_data))
2157                         RETURN(PTR_ERR(op_data));
2158
2159                 op_data->op_valid = OBD_MD_FLFLAGS;
2160                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2161                 ll_finish_md_op_data(op_data);
2162                 if (rc) {
2163                         CERROR("%s: failure inode "DFID": rc = %d\n",
2164                                sbi->ll_md_exp->exp_obd->obd_name,
2165                                PFID(ll_inode2fid(inode)), rc);
2166                         RETURN(-abs(rc));
2167                 }
2168
2169                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2170
2171                 flags = body->mbo_flags;
2172
2173                 ptlrpc_req_finished(req);
2174
2175                 RETURN(put_user(flags, (int __user *)arg));
2176         }
2177         case FS_IOC_SETFLAGS: {
2178                 struct iattr *attr;
2179                 struct md_op_data *op_data;
2180                 struct cl_object *obj;
2181                 struct fsxattr fa = { 0 };
2182
2183                 if (get_user(flags, (int __user *)arg))
2184                         RETURN(-EFAULT);
2185
2186                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2187                 if (flags & LUSTRE_PROJINHERIT_FL)
2188                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2189
2190                 rc = ll_ioctl_check_project(inode, &fa);
2191                 if (rc)
2192                         RETURN(rc);
2193
2194                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2195                                              LUSTRE_OPC_ANY, NULL);
2196                 if (IS_ERR(op_data))
2197                         RETURN(PTR_ERR(op_data));
2198
2199                 op_data->op_attr_flags = flags;
2200                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2201                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2202                 ll_finish_md_op_data(op_data);
2203                 ptlrpc_req_finished(req);
2204                 if (rc)
2205                         RETURN(rc);
2206
2207                 ll_update_inode_flags(inode, flags);
2208
2209                 obj = ll_i2info(inode)->lli_clob;
2210                 if (obj == NULL)
2211                         RETURN(0);
2212
2213                 OBD_ALLOC_PTR(attr);
2214                 if (attr == NULL)
2215                         RETURN(-ENOMEM);
2216
2217                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2218
2219                 OBD_FREE_PTR(attr);
2220                 RETURN(rc);
2221         }
2222         default:
2223                 RETURN(-ENOSYS);
2224         }
2225
2226         RETURN(0);
2227 }
2228
2229 int ll_flush_ctx(struct inode *inode)
2230 {
2231         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2232
2233         CDEBUG(D_SEC, "flush context for user %d\n",
2234                from_kuid(&init_user_ns, current_uid()));
2235
2236         obd_set_info_async(NULL, sbi->ll_md_exp,
2237                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2238                            0, NULL, NULL);
2239         obd_set_info_async(NULL, sbi->ll_dt_exp,
2240                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2241                            0, NULL, NULL);
2242         return 0;
2243 }
2244
2245 /* umount -f client means force down, don't save state */
2246 void ll_umount_begin(struct super_block *sb)
2247 {
2248         struct ll_sb_info *sbi = ll_s2sbi(sb);
2249         struct obd_device *obd;
2250         struct obd_ioctl_data *ioc_data;
2251         struct l_wait_info lwi;
2252         wait_queue_head_t waitq;
2253         ENTRY;
2254
2255         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2256                sb->s_count, atomic_read(&sb->s_active));
2257
2258         obd = class_exp2obd(sbi->ll_md_exp);
2259         if (obd == NULL) {
2260                 CERROR("Invalid MDC connection handle %#llx\n",
2261                        sbi->ll_md_exp->exp_handle.h_cookie);
2262                 EXIT;
2263                 return;
2264         }
2265         obd->obd_force = 1;
2266
2267         obd = class_exp2obd(sbi->ll_dt_exp);
2268         if (obd == NULL) {
2269                 CERROR("Invalid LOV connection handle %#llx\n",
2270                        sbi->ll_dt_exp->exp_handle.h_cookie);
2271                 EXIT;
2272                 return;
2273         }
2274         obd->obd_force = 1;
2275
2276         OBD_ALLOC_PTR(ioc_data);
2277         if (ioc_data) {
2278                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2279                               sizeof *ioc_data, ioc_data, NULL);
2280
2281                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2282                               sizeof *ioc_data, ioc_data, NULL);
2283
2284                 OBD_FREE_PTR(ioc_data);
2285         }
2286
2287         /* Really, we'd like to wait until there are no requests outstanding,
2288          * and then continue.  For now, we just periodically checking for vfs
2289          * to decrement mnt_cnt and hope to finish it within 10sec.
2290          */
2291         init_waitqueue_head(&waitq);
2292         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2293                                    cfs_time_seconds(1), NULL, NULL);
2294         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2295
2296         EXIT;
2297 }
2298
2299 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2300 {
2301         struct ll_sb_info *sbi = ll_s2sbi(sb);
2302         char *profilenm = get_profile_name(sb);
2303         int err;
2304         __u32 read_only;
2305
2306         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2307                 read_only = *flags & MS_RDONLY;
2308                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2309                                          sizeof(KEY_READ_ONLY),
2310                                          KEY_READ_ONLY, sizeof(read_only),
2311                                          &read_only, NULL);
2312                 if (err) {
2313                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2314                                       profilenm, read_only ?
2315                                       "read-only" : "read-write", err);
2316                         return err;
2317                 }
2318
2319                 if (read_only)
2320                         sb->s_flags |= MS_RDONLY;
2321                 else
2322                         sb->s_flags &= ~MS_RDONLY;
2323
2324                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2325                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2326                                       read_only ?  "read-only" : "read-write");
2327         }
2328         return 0;
2329 }
2330
2331 /**
2332  * Cleanup the open handle that is cached on MDT-side.
2333  *
2334  * For open case, the client side open handling thread may hit error
2335  * after the MDT grant the open. Under such case, the client should
2336  * send close RPC to the MDT as cleanup; otherwise, the open handle
2337  * on the MDT will be leaked there until the client umount or evicted.
2338  *
2339  * In further, if someone unlinked the file, because the open handle
2340  * holds the reference on such file/object, then it will block the
2341  * subsequent threads that want to locate such object via FID.
2342  *
2343  * \param[in] sb        super block for this file-system
2344  * \param[in] open_req  pointer to the original open request
2345  */
2346 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2347 {
2348         struct mdt_body                 *body;
2349         struct md_op_data               *op_data;
2350         struct ptlrpc_request           *close_req = NULL;
2351         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2352         ENTRY;
2353
2354         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2355         OBD_ALLOC_PTR(op_data);
2356         if (op_data == NULL) {
2357                 CWARN("%s: cannot allocate op_data to release open handle for "
2358                       DFID"\n",
2359                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2360
2361                 RETURN_EXIT;
2362         }
2363
2364         op_data->op_fid1 = body->mbo_fid1;
2365         op_data->op_open_handle = body->mbo_open_handle;
2366         op_data->op_mod_time = ktime_get_real_seconds();
2367         md_close(exp, op_data, NULL, &close_req);
2368         ptlrpc_req_finished(close_req);
2369         ll_finish_md_op_data(op_data);
2370
2371         EXIT;
2372 }
2373
2374 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2375                   struct super_block *sb, struct lookup_intent *it)
2376 {
2377         struct ll_sb_info *sbi = NULL;
2378         struct lustre_md md = { NULL };
2379         int rc;
2380         ENTRY;
2381
2382         LASSERT(*inode || sb);
2383         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2384         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2385                               sbi->ll_md_exp, &md);
2386         if (rc != 0)
2387                 GOTO(cleanup, rc);
2388
2389         if (*inode) {
2390                 rc = ll_update_inode(*inode, &md);
2391                 if (rc != 0)
2392                         GOTO(out, rc);
2393         } else {
2394                 LASSERT(sb != NULL);
2395
2396                 /*
2397                  * At this point server returns to client's same fid as client
2398                  * generated for creating. So using ->fid1 is okay here.
2399                  */
2400                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2401                         CERROR("%s: Fid is insane "DFID"\n",
2402                                 ll_get_fsname(sb, NULL, 0),
2403                                 PFID(&md.body->mbo_fid1));
2404                         GOTO(out, rc = -EINVAL);
2405                 }
2406
2407                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2408                                              sbi->ll_flags & LL_SBI_32BIT_API),
2409                                  &md);
2410                 if (IS_ERR(*inode)) {
2411 #ifdef CONFIG_FS_POSIX_ACL
2412                         if (md.posix_acl) {
2413                                 posix_acl_release(md.posix_acl);
2414                                 md.posix_acl = NULL;
2415                         }
2416 #endif
2417                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2418                         *inode = NULL;
2419                         CERROR("new_inode -fatal: rc %d\n", rc);
2420                         GOTO(out, rc);
2421                 }
2422         }
2423
2424         /* Handling piggyback layout lock.
2425          * Layout lock can be piggybacked by getattr and open request.
2426          * The lsm can be applied to inode only if it comes with a layout lock
2427          * otherwise correct layout may be overwritten, for example:
2428          * 1. proc1: mdt returns a lsm but not granting layout
2429          * 2. layout was changed by another client
2430          * 3. proc2: refresh layout and layout lock granted
2431          * 4. proc1: to apply a stale layout */
2432         if (it != NULL && it->it_lock_mode != 0) {
2433                 struct lustre_handle lockh;
2434                 struct ldlm_lock *lock;
2435
2436                 lockh.cookie = it->it_lock_handle;
2437                 lock = ldlm_handle2lock(&lockh);
2438                 LASSERT(lock != NULL);
2439                 if (ldlm_has_layout(lock)) {
2440                         struct cl_object_conf conf;
2441
2442                         memset(&conf, 0, sizeof(conf));
2443                         conf.coc_opc = OBJECT_CONF_SET;
2444                         conf.coc_inode = *inode;
2445                         conf.coc_lock = lock;
2446                         conf.u.coc_layout = md.layout;
2447                         (void)ll_layout_conf(*inode, &conf);
2448                 }
2449                 LDLM_LOCK_PUT(lock);
2450         }
2451
2452         GOTO(out, rc = 0);
2453
2454 out:
2455         md_free_lustre_md(sbi->ll_md_exp, &md);
2456
2457 cleanup:
2458         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2459                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2460
2461         return rc;
2462 }
2463
2464 int ll_obd_statfs(struct inode *inode, void __user *arg)
2465 {
2466         struct ll_sb_info *sbi = NULL;
2467         struct obd_export *exp;
2468         char *buf = NULL;
2469         struct obd_ioctl_data *data = NULL;
2470         __u32 type;
2471         int len = 0, rc;
2472
2473         if (!inode || !(sbi = ll_i2sbi(inode)))
2474                 GOTO(out_statfs, rc = -EINVAL);
2475
2476         rc = obd_ioctl_getdata(&buf, &len, arg);
2477         if (rc)
2478                 GOTO(out_statfs, rc);
2479
2480         data = (void*)buf;
2481         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2482             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2483                 GOTO(out_statfs, rc = -EINVAL);
2484
2485         if (data->ioc_inllen1 != sizeof(__u32) ||
2486             data->ioc_inllen2 != sizeof(__u32) ||
2487             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2488             data->ioc_plen2 != sizeof(struct obd_uuid))
2489                 GOTO(out_statfs, rc = -EINVAL);
2490
2491         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2492         if (type & LL_STATFS_LMV)
2493                 exp = sbi->ll_md_exp;
2494         else if (type & LL_STATFS_LOV)
2495                 exp = sbi->ll_dt_exp;
2496         else
2497                 GOTO(out_statfs, rc = -ENODEV);
2498
2499         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2500         if (rc)
2501                 GOTO(out_statfs, rc);
2502 out_statfs:
2503         OBD_FREE_LARGE(buf, len);
2504         return rc;
2505 }
2506
2507 /*
2508  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2509  * be called early to avoid deadlock.
2510  */
2511 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2512 {
2513         if (op_data->op_mea2_sem) {
2514                 up_read(op_data->op_mea2_sem);
2515                 op_data->op_mea2_sem = NULL;
2516         }
2517
2518         if (op_data->op_mea1_sem) {
2519                 up_read(op_data->op_mea1_sem);
2520                 op_data->op_mea1_sem = NULL;
2521         }
2522 }
2523
2524 /* this function prepares md_op_data hint for passing it down to MD stack. */
2525 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2526                                       struct inode *i1, struct inode *i2,
2527                                       const char *name, size_t namelen,
2528                                       __u32 mode, __u32 opc, void *data)
2529 {
2530         LASSERT(i1 != NULL);
2531
2532         if (name == NULL) {
2533                 /* Do not reuse namelen for something else. */
2534                 if (namelen != 0)
2535                         return ERR_PTR(-EINVAL);
2536         } else {
2537                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2538                         return ERR_PTR(-ENAMETOOLONG);
2539
2540                 if (!lu_name_is_valid_2(name, namelen))
2541                         return ERR_PTR(-EINVAL);
2542         }
2543
2544         if (op_data == NULL)
2545                 OBD_ALLOC_PTR(op_data);
2546
2547         if (op_data == NULL)
2548                 return ERR_PTR(-ENOMEM);
2549
2550         ll_i2gids(op_data->op_suppgids, i1, i2);
2551         op_data->op_fid1 = *ll_inode2fid(i1);
2552         op_data->op_default_stripe_offset = -1;
2553
2554         if (S_ISDIR(i1->i_mode)) {
2555                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2556                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2557                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2558                 if (opc == LUSTRE_OPC_MKDIR)
2559                         op_data->op_default_stripe_offset =
2560                                    ll_i2info(i1)->lli_def_stripe_offset;
2561         }
2562
2563         if (i2) {
2564                 op_data->op_fid2 = *ll_inode2fid(i2);
2565                 if (S_ISDIR(i2->i_mode)) {
2566                         if (i2 != i1) {
2567                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2568                                 op_data->op_mea2_sem =
2569                                                 &ll_i2info(i2)->lli_lsm_sem;
2570                         }
2571                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2572                 }
2573         } else {
2574                 fid_zero(&op_data->op_fid2);
2575         }
2576
2577         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2578                 op_data->op_cli_flags |= CLI_HASH64;
2579
2580         if (ll_need_32bit_api(ll_i2sbi(i1)))
2581                 op_data->op_cli_flags |= CLI_API32;
2582
2583         op_data->op_name = name;
2584         op_data->op_namelen = namelen;
2585         op_data->op_mode = mode;
2586         op_data->op_mod_time = ktime_get_real_seconds();
2587         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2588         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2589         op_data->op_cap = cfs_curproc_cap_pack();
2590         op_data->op_mds = 0;
2591         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2592              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2593                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2594         }
2595         op_data->op_data = data;
2596
2597         return op_data;
2598 }
2599
2600 void ll_finish_md_op_data(struct md_op_data *op_data)
2601 {
2602         ll_unlock_md_op_lsm(op_data);
2603         security_release_secctx(op_data->op_file_secctx,
2604                                 op_data->op_file_secctx_size);
2605         OBD_FREE_PTR(op_data);
2606 }
2607
2608 #ifdef HAVE_SUPEROPS_USE_DENTRY
2609 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2610 #else
2611 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2612 #endif
2613 {
2614         struct ll_sb_info *sbi;
2615
2616 #ifdef HAVE_SUPEROPS_USE_DENTRY
2617         LASSERT((seq != NULL) && (dentry != NULL));
2618         sbi = ll_s2sbi(dentry->d_sb);
2619 #else
2620         LASSERT((seq != NULL) && (vfs != NULL));
2621         sbi = ll_s2sbi(vfs->mnt_sb);
2622 #endif
2623
2624         if (sbi->ll_flags & LL_SBI_NOLCK)
2625                 seq_puts(seq, ",nolock");
2626
2627         if (sbi->ll_flags & LL_SBI_FLOCK)
2628                 seq_puts(seq, ",flock");
2629
2630         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2631                 seq_puts(seq, ",localflock");
2632
2633         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2634                 seq_puts(seq, ",user_xattr");
2635
2636         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2637                 seq_puts(seq, ",lazystatfs");
2638
2639         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2640                 seq_puts(seq, ",user_fid2path");
2641
2642         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2643                 seq_puts(seq, ",always_ping");
2644
2645         RETURN(0);
2646 }
2647
2648 /**
2649  * Get obd name by cmd, and copy out to user space
2650  */
2651 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2652 {
2653         struct ll_sb_info *sbi = ll_i2sbi(inode);
2654         struct obd_device *obd;
2655         ENTRY;
2656
2657         if (cmd == OBD_IOC_GETDTNAME)
2658                 obd = class_exp2obd(sbi->ll_dt_exp);
2659         else if (cmd == OBD_IOC_GETMDNAME)
2660                 obd = class_exp2obd(sbi->ll_md_exp);
2661         else
2662                 RETURN(-EINVAL);
2663
2664         if (!obd)
2665                 RETURN(-ENOENT);
2666
2667         if (copy_to_user((void __user *)arg, obd->obd_name,
2668                          strlen(obd->obd_name) + 1))
2669                 RETURN(-EFAULT);
2670
2671         RETURN(0);
2672 }
2673
2674 /**
2675  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2676  * fsname will be returned in this buffer; otherwise, a static buffer will be
2677  * used to store the fsname and returned to caller.
2678  */
2679 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2680 {
2681         static char fsname_static[MTI_NAME_MAXLEN];
2682         struct lustre_sb_info *lsi = s2lsi(sb);
2683         char *ptr;
2684         int len;
2685
2686         if (buf == NULL) {
2687                 /* this means the caller wants to use static buffer
2688                  * and it doesn't care about race. Usually this is
2689                  * in error reporting path */
2690                 buf = fsname_static;
2691                 buflen = sizeof(fsname_static);
2692         }
2693
2694         len = strlen(lsi->lsi_lmd->lmd_profile);
2695         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2696         if (ptr && (strcmp(ptr, "-client") == 0))
2697                 len -= 7;
2698
2699         if (unlikely(len >= buflen))
2700                 len = buflen - 1;
2701         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2702         buf[len] = '\0';
2703
2704         return buf;
2705 }
2706
2707 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2708 {
2709         char *path = NULL;
2710
2711         struct path p;
2712
2713         p.dentry = dentry;
2714         p.mnt = current->fs->root.mnt;
2715         path_get(&p);
2716         path = d_path(&p, buf, bufsize);
2717         path_put(&p);
2718         return path;
2719 }
2720
2721 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2722 {
2723         char *buf, *path = NULL;
2724         struct dentry *dentry = NULL;
2725         struct inode *inode = page->mapping->host;
2726
2727         /* this can be called inside spin lock so use GFP_ATOMIC. */
2728         buf = (char *)__get_free_page(GFP_ATOMIC);
2729         if (buf != NULL) {
2730                 dentry = d_find_alias(page->mapping->host);
2731                 if (dentry != NULL)
2732                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2733         }
2734
2735         CDEBUG(D_WARNING,
2736                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2737                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2738                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2739                PFID(ll_inode2fid(inode)),
2740                (path && !IS_ERR(path)) ? path : "", ioret);
2741
2742         if (dentry != NULL)
2743                 dput(dentry);
2744
2745         if (buf != NULL)
2746                 free_page((unsigned long)buf);
2747 }
2748
2749 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2750                         struct lov_user_md **kbuf)
2751 {
2752         struct lov_user_md      lum;
2753         ssize_t                 lum_size;
2754         ENTRY;
2755
2756         if (copy_from_user(&lum, md, sizeof(lum)))
2757                 RETURN(-EFAULT);
2758
2759         lum_size = ll_lov_user_md_size(&lum);
2760         if (lum_size < 0)
2761                 RETURN(lum_size);
2762
2763         OBD_ALLOC(*kbuf, lum_size);
2764         if (*kbuf == NULL)
2765                 RETURN(-ENOMEM);
2766
2767         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2768                 OBD_FREE(*kbuf, lum_size);
2769                 RETURN(-EFAULT);
2770         }
2771
2772         RETURN(lum_size);
2773 }
2774
2775 /*
2776  * Compute llite root squash state after a change of root squash
2777  * configuration setting or add/remove of a lnet nid
2778  */
2779 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2780 {
2781         struct root_squash_info *squash = &sbi->ll_squash;
2782         int i;
2783         bool matched;
2784         struct lnet_process_id id;
2785
2786         /* Update norootsquash flag */
2787         down_write(&squash->rsi_sem);
2788         if (list_empty(&squash->rsi_nosquash_nids))
2789                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2790         else {
2791                 /* Do not apply root squash as soon as one of our NIDs is
2792                  * in the nosquash_nids list */
2793                 matched = false;
2794                 i = 0;
2795                 while (LNetGetId(i++, &id) != -ENOENT) {
2796                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2797                                 continue;
2798                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2799                                 matched = true;
2800                                 break;
2801                         }
2802                 }
2803                 if (matched)
2804                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2805                 else
2806                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2807         }
2808         up_write(&squash->rsi_sem);
2809 }
2810
2811 /**
2812  * Parse linkea content to extract information about a given hardlink
2813  *
2814  * \param[in]   ldata      - Initialized linkea data
2815  * \param[in]   linkno     - Link identifier
2816  * \param[out]  parent_fid - The entry's parent FID
2817  * \param[out]  ln         - Entry name destination buffer
2818  *
2819  * \retval 0 on success
2820  * \retval Appropriate negative error code on failure
2821  */
2822 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2823                             struct lu_fid *parent_fid, struct lu_name *ln)
2824 {
2825         unsigned int    idx;
2826         int             rc;
2827         ENTRY;
2828
2829         rc = linkea_init_with_rec(ldata);
2830         if (rc < 0)
2831                 RETURN(rc);
2832
2833         if (linkno >= ldata->ld_leh->leh_reccount)
2834                 /* beyond last link */
2835                 RETURN(-ENODATA);
2836
2837         linkea_first_entry(ldata);
2838         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2839                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2840                                     parent_fid);
2841                 if (idx == linkno)
2842                         break;
2843
2844                 linkea_next_entry(ldata);
2845         }
2846
2847         if (idx < linkno)
2848                 RETURN(-ENODATA);
2849
2850         RETURN(0);
2851 }
2852
2853 /**
2854  * Get parent FID and name of an identified link. Operation is performed for
2855  * a given link number, letting the caller iterate over linkno to list one or
2856  * all links of an entry.
2857  *
2858  * \param[in]     file - File descriptor against which to perform the operation
2859  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2860  *                       on and the available size. It is eventually filled with
2861  *                       the requested information or left untouched on error
2862  *
2863  * \retval - 0 on success
2864  * \retval - Appropriate negative error code on failure
2865  */
2866 int ll_getparent(struct file *file, struct getparent __user *arg)
2867 {
2868         struct inode            *inode = file_inode(file);
2869         struct linkea_data      *ldata;
2870         struct lu_buf            buf = LU_BUF_NULL;
2871         struct lu_name           ln;
2872         struct lu_fid            parent_fid;
2873         __u32                    linkno;
2874         __u32                    name_size;
2875         int                      rc;
2876
2877         ENTRY;
2878
2879         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2880             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2881                 RETURN(-EPERM);
2882
2883         if (get_user(name_size, &arg->gp_name_size))
2884                 RETURN(-EFAULT);
2885
2886         if (get_user(linkno, &arg->gp_linkno))
2887                 RETURN(-EFAULT);
2888
2889         if (name_size > PATH_MAX)
2890                 RETURN(-EINVAL);
2891
2892         OBD_ALLOC(ldata, sizeof(*ldata));
2893         if (ldata == NULL)
2894                 RETURN(-ENOMEM);
2895
2896         rc = linkea_data_new(ldata, &buf);
2897         if (rc < 0)
2898                 GOTO(ldata_free, rc);
2899
2900 #ifdef HAVE_XATTR_HANDLER_FLAGS
2901         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2902                            buf.lb_len, OBD_MD_FLXATTR);
2903 #else
2904         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2905                          buf.lb_len);
2906 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2907         if (rc < 0)
2908                 GOTO(lb_free, rc);
2909
2910         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2911         if (rc < 0)
2912                 GOTO(lb_free, rc);
2913
2914         if (ln.ln_namelen >= name_size)
2915                 GOTO(lb_free, rc = -EOVERFLOW);
2916
2917         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2918                 GOTO(lb_free, rc = -EFAULT);
2919
2920         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2921                 GOTO(lb_free, rc = -EFAULT);
2922
2923         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2924                 GOTO(lb_free, rc = -EFAULT);
2925
2926 lb_free:
2927         lu_buf_free(&buf);
2928 ldata_free:
2929         OBD_FREE(ldata, sizeof(*ldata));
2930
2931         RETURN(rc);
2932 }