lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2016, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/time.h>
  42 #include <linux/types.h>
  43 #include <linux/version.h>
  44 #include <linux/mm.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <linux/security.h>
  50
  51 #include <uapi/linux/lustre/lustre_ioctl.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <uapi/linux/lustre/lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 struct kmem_cache *ll_file_data_slab;
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC_PTR(sbi);
  79         if (sbi == NULL)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         sbi->ll_cache = cl_cache_init(lru_page_max);
  94         if (sbi->ll_cache == NULL) {
  95                 OBD_FREE(sbi, sizeof(*sbi));
  96                 RETURN(NULL);
  97         }
  98
  99         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 100                                            SBI_DEFAULT_READAHEAD_MAX);
 101         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 102         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 103
 104         ll_generate_random_uuid(uuid);
 105         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 106         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 107
 108         sbi->ll_flags |= LL_SBI_VERBOSE;
 109 #ifdef ENABLE_CHECKSUM
 110         sbi->ll_flags |= LL_SBI_CHECKSUM;
 111 #endif
 112
 113 #ifdef HAVE_LRU_RESIZE_SUPPORT
 114         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 115 #endif
 116         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 117
 118         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 119                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 120                                pp_r_hist.oh_lock);
 121                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 122                                pp_w_hist.oh_lock);
 123         }
 124
 125         /* metadata statahead is enabled by default */
 126         sbi->ll_sa_max = LL_SA_RPC_DEF;
 127         atomic_set(&sbi->ll_sa_total, 0);
 128         atomic_set(&sbi->ll_sa_wrong, 0);
 129         atomic_set(&sbi->ll_sa_running, 0);
 130         atomic_set(&sbi->ll_agl_total, 0);
 131         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 132         sbi->ll_flags |= LL_SBI_FAST_READ;
 133
 134         /* root squash */
 135         sbi->ll_squash.rsi_uid = 0;
 136         sbi->ll_squash.rsi_gid = 0;
 137         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 138         init_rwsem(&sbi->ll_squash.rsi_sem);
 139
 140         RETURN(sbi);
 141 }
 142
 143 static void ll_free_sbi(struct super_block *sb)
 144 {
 145         struct ll_sb_info *sbi = ll_s2sbi(sb);
 146         ENTRY;
 147
 148         if (sbi != NULL) {
 149                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 150                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 151                 if (sbi->ll_cache != NULL) {
 152                         cl_cache_decref(sbi->ll_cache);
 153                         sbi->ll_cache = NULL;
 154                 }
 155                 OBD_FREE(sbi, sizeof(*sbi));
 156         }
 157         EXIT;
 158 }
 159
 160 static inline int obd_connect_has_secctx(struct obd_connect_data *data)
 161 {
 162         return data->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
 163                data->ocd_connect_flags2 & OBD_CONNECT2_FILE_SECCTX;
 164 }
 165
 166 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 167                                     struct vfsmount *mnt)
 168 {
 169         struct inode *root = NULL;
 170         struct ll_sb_info *sbi = ll_s2sbi(sb);
 171         struct obd_device *obd;
 172         struct obd_statfs *osfs = NULL;
 173         struct ptlrpc_request *request = NULL;
 174         struct obd_connect_data *data = NULL;
 175         struct obd_uuid *uuid;
 176         struct md_op_data *op_data;
 177         struct lustre_md lmd;
 178         u64 valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         OBD_ALLOC_PTR(osfs);
 193         if (osfs == NULL) {
 194                 OBD_FREE_PTR(data);
 195                 RETURN(-ENOMEM);
 196         }
 197
 198         /* pass client page size via ocd_grant_blkbits, the server should report
 199          * back its backend blocksize for grant calculation purpose */
 200         data->ocd_grant_blkbits = PAGE_SHIFT;
 201
 202         /* indicate MDT features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 207                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 208                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 209                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 210                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 211                                   OBD_CONNECT_64BITHASH |
 212                                   OBD_CONNECT_EINPROGRESS |
 213                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 214                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 215                                   OBD_CONNECT_MAX_EASIZE |
 216                                   OBD_CONNECT_FLOCK_DEAD |
 217                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 218                                   OBD_CONNECT_OPEN_BY_FID |
 219                                   OBD_CONNECT_DIR_STRIPE |
 220                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 221                                   OBD_CONNECT_SUBTREE |
 222                                   OBD_CONNECT_MULTIMODRPCS |
 223                                   OBD_CONNECT_GRANT_PARAM | OBD_CONNECT_FLAGS2;
 224
 225         data->ocd_connect_flags2 = 0;
 226
 227 #ifdef HAVE_LRU_RESIZE_SUPPORT
 228         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 229                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 230 #endif
 231 #ifdef CONFIG_FS_POSIX_ACL
 232         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 233                                    OBD_CONNECT_LARGE_ACL;
 234 #endif
 235
 236         data->ocd_cksum_types = cksum_types_supported_client();
 237
 238         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 239                 /* flag mdc connection as lightweight, only used for test
 240                  * purpose, use with care */
 241                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 242
 243         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 244         data->ocd_version = LUSTRE_VERSION_CODE;
 245
 246         if (sb->s_flags & MS_RDONLY)
 247                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 248         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 249                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 250
 251 #ifdef MS_NOSEC
 252         /* Setting this indicates we correctly support S_NOSEC (See kernel
 253          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 254          */
 255         sb->s_flags |= MS_NOSEC;
 256 #endif
 257
 258         if (sbi->ll_flags & LL_SBI_FLOCK)
 259                 sbi->ll_fop = &ll_file_operations_flock;
 260         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 261                 sbi->ll_fop = &ll_file_operations;
 262         else
 263                 sbi->ll_fop = &ll_file_operations_noflock;
 264
 265         /* always ping even if server suppress_pings */
 266         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 267                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 268
 269 #if defined(HAVE_SECURITY_DENTRY_INIT_SECURITY) && defined(CONFIG_SECURITY)
 270         data->ocd_connect_flags2 |= OBD_CONNECT2_FILE_SECCTX;
 271 #endif /* HAVE_SECURITY_DENTRY_INIT_SECURITY */
 272
 273         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 274
 275         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 276         if (err == -EBUSY) {
 277                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 278                                    "recovery, of which this client is not a "
 279                                    "part. Please wait for recovery to complete,"
 280                                    " abort, or time out.\n", md);
 281                 GOTO(out, err);
 282         } else if (err) {
 283                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 284                 GOTO(out, err);
 285         }
 286
 287         sbi->ll_md_exp->exp_connect_data = *data;
 288
 289         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 290                            LUSTRE_SEQ_METADATA);
 291         if (err) {
 292                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 293                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 294                 GOTO(out_md, err);
 295         }
 296
 297         /* For mount, we only need fs info from MDT0, and also in DNE, it
 298          * can make sure the client can be mounted as long as MDT0 is
 299          * avaible */
 300         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 301                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 302                         OBD_STATFS_FOR_MDT0);
 303         if (err)
 304                 GOTO(out_md_fid, err);
 305
 306         /* This needs to be after statfs to ensure connect has finished.
 307          * Note that "data" does NOT contain the valid connect reply.
 308          * If connecting to a 1.8 server there will be no LMV device, so
 309          * we can access the MDC export directly and exp_connect_flags will
 310          * be non-zero, but if accessing an upgraded 2.1 server it will
 311          * have the correct flags filled in.
 312          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 313         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 314         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 315             valid != CLIENT_CONNECT_MDT_REQD) {
 316                 char *buf;
 317
 318                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 319                 obd_connect_flags2str(buf, PAGE_SIZE,
 320                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 321                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 322                                    "feature(s) needed for correct operation "
 323                                    "of this client (%s). Please upgrade "
 324                                    "server or downgrade client.\n",
 325                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 326                 OBD_FREE(buf, PAGE_SIZE);
 327                 GOTO(out_md_fid, err = -EPROTO);
 328         }
 329
 330         size = sizeof(*data);
 331         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 332                            KEY_CONN_DATA,  &size, data);
 333         if (err) {
 334                 CERROR("%s: Get connect data failed: rc = %d\n",
 335                        sbi->ll_md_exp->exp_obd->obd_name, err);
 336                 GOTO(out_md_fid, err);
 337         }
 338
 339         LASSERT(osfs->os_bsize);
 340         sb->s_blocksize = osfs->os_bsize;
 341         sb->s_blocksize_bits = log2(osfs->os_bsize);
 342         sb->s_magic = LL_SUPER_MAGIC;
 343         sb->s_maxbytes = MAX_LFS_FILESIZE;
 344         sbi->ll_namelen = osfs->os_namelen;
 345         sbi->ll_mnt.mnt = current->fs->root.mnt;
 346
 347         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 348             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 349                 LCONSOLE_INFO("Disabling user_xattr feature because "
 350                               "it is not supported on the server\n");
 351                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 352         }
 353
 354         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 355 #ifdef MS_POSIXACL
 356                 sb->s_flags |= MS_POSIXACL;
 357 #endif
 358                 sbi->ll_flags |= LL_SBI_ACL;
 359         } else {
 360                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 361 #ifdef MS_POSIXACL
 362                 sb->s_flags &= ~MS_POSIXACL;
 363 #endif
 364                 sbi->ll_flags &= ~LL_SBI_ACL;
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 368                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 369
 370         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 371                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 372
 373         if (obd_connect_has_secctx(data))
 374                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 375
 376         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 377                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 378                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 379                                       "unknown maximum xattr size.\n", dt);
 380                 } else if (!sbi->ll_xattr_cache_set) {
 381                         /* If xattr_cache is already set (no matter 0 or 1)
 382                          * during processing llog, it won't be enabled here. */
 383                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 384                         sbi->ll_xattr_cache_enabled = 1;
 385                 }
 386         }
 387
 388         obd = class_name2obd(dt);
 389         if (!obd) {
 390                 CERROR("DT %s: not setup or attached\n", dt);
 391                 GOTO(out_md_fid, err = -ENODEV);
 392         }
 393
 394         /* pass client page size via ocd_grant_blkbits, the server should report
 395          * back its backend blocksize for grant calculation purpose */
 396         data->ocd_grant_blkbits = PAGE_SHIFT;
 397
 398         /* indicate OST features supported by this client */
 399         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 400                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 401                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 402                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 403                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 404                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 405                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 406                                   OBD_CONNECT_EINPROGRESS |
 407                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 408                                   OBD_CONNECT_LAYOUTLOCK |
 409                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 410                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 411                                   OBD_CONNECT_FLAGS2;
 412
 413 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 414  * can interoperate with an older version of lockahead which was released prior
 415  * to landing in master. This support will be dropped when 2.13 development
 416  * starts.  At the point, we should not just drop the connect flag (below), we
 417  * should also remove the support in the code.
 418  *
 419  * Removing it means a few things:
 420  * 1. Remove this section here
 421  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 422  * 3. Remove function exp_connect_lockahead_old
 423  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 424  * */
 425 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 426         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 427 #endif
 428
 429         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 430
 431         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 432                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 433
 434         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 435          * disabled by default, because it can still be enabled on the
 436          * fly via /proc. As a consequence, we still need to come to an
 437          * agreement on the supported algorithms at connect time */
 438         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 439
 440         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 441                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 442         else
 443                 data->ocd_cksum_types = cksum_types_supported_client();
 444
 445 #ifdef HAVE_LRU_RESIZE_SUPPORT
 446         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 447 #endif
 448         /* always ping even if server suppress_pings */
 449         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 450                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 451
 452         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 453                "ocd_grant: %d\n", data->ocd_connect_flags,
 454                data->ocd_version, data->ocd_grant);
 455
 456         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 457         obd->obd_upcall.onu_upcall = cl_ocd_update;
 458
 459         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 460
 461         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 462                           NULL);
 463         if (err == -EBUSY) {
 464                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 465                                    "recovery, of which this client is not a "
 466                                    "part.  Please wait for recovery to "
 467                                    "complete, abort, or time out.\n", dt);
 468                 GOTO(out_md, err);
 469         } else if (err) {
 470                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 471                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 472                 GOTO(out_md, err);
 473         }
 474
 475         sbi->ll_dt_exp->exp_connect_data = *data;
 476
 477         /* Don't change value if it was specified in the config log */
 478         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1)
 479                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 480                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 481                               (data->ocd_brw_size >> PAGE_SHIFT));
 482
 483         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 484                            LUSTRE_SEQ_METADATA);
 485         if (err) {
 486                 CERROR("%s: Can't init data layer FID infrastructure, "
 487                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 488                 GOTO(out_dt, err);
 489         }
 490
 491         mutex_lock(&sbi->ll_lco.lco_lock);
 492         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 493         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 494         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 495         mutex_unlock(&sbi->ll_lco.lco_lock);
 496
 497         fid_zero(&sbi->ll_root_fid);
 498         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 499                            &sbi->ll_root_fid);
 500         if (err) {
 501                 CERROR("cannot mds_connect: rc = %d\n", err);
 502                 GOTO(out_lock_cn_cb, err);
 503         }
 504         if (!fid_is_sane(&sbi->ll_root_fid)) {
 505                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 506                        sbi->ll_md_exp->exp_obd->obd_name,
 507                        PFID(&sbi->ll_root_fid));
 508                 GOTO(out_lock_cn_cb, err = -EINVAL);
 509         }
 510         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 511
 512         sb->s_op = &lustre_super_operations;
 513 #ifdef HAVE_XATTR_HANDLER_FLAGS
 514         sb->s_xattr = ll_xattr_handlers;
 515 #endif
 516 #if THREAD_SIZE >= 8192 /*b=17630*/
 517         sb->s_export_op = &lustre_export_operations;
 518 #endif
 519
 520         /* make root inode
 521          * XXX: move this to after cbd setup? */
 522         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 523         if (sbi->ll_flags & LL_SBI_ACL)
 524                 valid |= OBD_MD_FLACL;
 525
 526         OBD_ALLOC_PTR(op_data);
 527         if (op_data == NULL)
 528                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 529
 530         op_data->op_fid1 = sbi->ll_root_fid;
 531         op_data->op_mode = 0;
 532         op_data->op_valid = valid;
 533
 534         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 535
 536         OBD_FREE_PTR(op_data);
 537         if (err) {
 538                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 539                        sbi->ll_md_exp->exp_obd->obd_name, err);
 540                 GOTO(out_lock_cn_cb, err);
 541         }
 542
 543         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 544                                sbi->ll_md_exp, &lmd);
 545         if (err) {
 546                 CERROR("failed to understand root inode md: rc = %d\n", err);
 547                 ptlrpc_req_finished(request);
 548                 GOTO(out_lock_cn_cb, err);
 549         }
 550
 551         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 552         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 553                                             sbi->ll_flags & LL_SBI_32BIT_API),
 554                        &lmd);
 555         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 556         ptlrpc_req_finished(request);
 557
 558         if (IS_ERR(root)) {
 559 #ifdef CONFIG_FS_POSIX_ACL
 560                 if (lmd.posix_acl) {
 561                         posix_acl_release(lmd.posix_acl);
 562                         lmd.posix_acl = NULL;
 563                 }
 564 #endif
 565                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 566                 root = NULL;
 567                 CERROR("lustre_lite: bad iget4 for root\n");
 568                 GOTO(out_root, err);
 569         }
 570
 571         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 572         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 573                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 574                                  NULL);
 575         if (err) {
 576                 CERROR("%s: Set checksum failed: rc = %d\n",
 577                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 578                 GOTO(out_root, err);
 579         }
 580         cl_sb_init(sb);
 581
 582         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 583                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 584                                  sbi->ll_cache, NULL);
 585         if (err) {
 586                 CERROR("%s: Set cache_set failed: rc = %d\n",
 587                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 588                 GOTO(out_root, err);
 589         }
 590
 591         sb->s_root = d_make_root(root);
 592         if (sb->s_root == NULL) {
 593                 CERROR("%s: can't make root dentry\n",
 594                         ll_get_fsname(sb, NULL, 0));
 595                 GOTO(out_root, err = -ENOMEM);
 596         }
 597 #ifdef HAVE_DCACHE_LOCK
 598         sb->s_root->d_op = &ll_d_ops;
 599 #endif
 600
 601         sbi->ll_sdev_orig = sb->s_dev;
 602
 603         /* We set sb->s_dev equal on all lustre clients in order to support
 604          * NFS export clustering.  NFSD requires that the FSID be the same
 605          * on all clients. */
 606         /* s_dev is also used in lt_compare() to compare two fs, but that is
 607          * only a node-local comparison. */
 608         uuid = obd_get_uuid(sbi->ll_md_exp);
 609         if (uuid != NULL)
 610                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 611
 612         if (data != NULL)
 613                 OBD_FREE_PTR(data);
 614         if (osfs != NULL)
 615                 OBD_FREE_PTR(osfs);
 616
 617         if (sbi->ll_proc_root != NULL) {
 618                 err = lprocfs_ll_register_obd(sb, dt);
 619                 if (err < 0) {
 620                         CERROR("%s: could not register %s in llite: rc = %d\n",
 621                                dt, ll_get_fsname(sb, NULL, 0), err);
 622                         err = 0;
 623                 }
 624                 err = lprocfs_ll_register_obd(sb, md);
 625                 if (err < 0) {
 626                         CERROR("%s: could not register %s in llite: rc = %d\n",
 627                                md, ll_get_fsname(sb, NULL, 0), err);
 628                         err = 0;
 629                 }
 630         }
 631
 632         RETURN(err);
 633 out_root:
 634         if (root)
 635                 iput(root);
 636 out_lock_cn_cb:
 637         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 638 out_dt:
 639         obd_disconnect(sbi->ll_dt_exp);
 640         sbi->ll_dt_exp = NULL;
 641 out_md_fid:
 642         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 643 out_md:
 644         obd_disconnect(sbi->ll_md_exp);
 645         sbi->ll_md_exp = NULL;
 646 out:
 647         if (data != NULL)
 648                 OBD_FREE_PTR(data);
 649         if (osfs != NULL)
 650                 OBD_FREE_PTR(osfs);
 651         return err;
 652 }
 653
 654 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 655 {
 656         int size, rc;
 657
 658         size = sizeof(*lmmsize);
 659         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 660                           KEY_MAX_EASIZE, &size, lmmsize);
 661         if (rc != 0) {
 662                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 663                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 664                 RETURN(rc);
 665         }
 666
 667         size = sizeof(int);
 668         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 669                           KEY_MAX_EASIZE, &size, lmmsize);
 670         if (rc)
 671                 CERROR("Get max mdsize error rc %d\n", rc);
 672
 673         RETURN(rc);
 674 }
 675
 676 /**
 677  * Get the value of the default_easize parameter.
 678  *
 679  * \see client_obd::cl_default_mds_easize
 680  *
 681  * \param[in] sbi       superblock info for this filesystem
 682  * \param[out] lmmsize  pointer to storage location for value
 683  *
 684  * \retval 0            on success
 685  * \retval negative     negated errno on failure
 686  */
 687 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 688 {
 689         int size, rc;
 690
 691         size = sizeof(int);
 692         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 693                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 694         if (rc)
 695                 CERROR("Get default mdsize error rc %d\n", rc);
 696
 697         RETURN(rc);
 698 }
 699
 700 /**
 701  * Set the default_easize parameter to the given value.
 702  *
 703  * \see client_obd::cl_default_mds_easize
 704  *
 705  * \param[in] sbi       superblock info for this filesystem
 706  * \param[in] lmmsize   the size to set
 707  *
 708  * \retval 0            on success
 709  * \retval negative     negated errno on failure
 710  */
 711 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 712 {
 713         int rc;
 714
 715         if (lmmsize < sizeof(struct lov_mds_md) ||
 716             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 717                 return -EINVAL;
 718
 719         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 720                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 721                                 sizeof(int), &lmmsize, NULL);
 722
 723         RETURN(rc);
 724 }
 725
 726 static void client_common_put_super(struct super_block *sb)
 727 {
 728         struct ll_sb_info *sbi = ll_s2sbi(sb);
 729         ENTRY;
 730
 731         cl_sb_fini(sb);
 732
 733         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 734         obd_disconnect(sbi->ll_dt_exp);
 735         sbi->ll_dt_exp = NULL;
 736
 737         lprocfs_ll_unregister_mountpoint(sbi);
 738
 739         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 740         obd_disconnect(sbi->ll_md_exp);
 741         sbi->ll_md_exp = NULL;
 742
 743         EXIT;
 744 }
 745
 746 void ll_kill_super(struct super_block *sb)
 747 {
 748         struct ll_sb_info *sbi;
 749         ENTRY;
 750
 751         /* not init sb ?*/
 752         if (!(sb->s_flags & MS_ACTIVE))
 753                 return;
 754
 755         sbi = ll_s2sbi(sb);
 756         /* we need restore s_dev from changed for clustred NFS before put_super
 757          * because new kernels have cached s_dev and change sb->s_dev in
 758          * put_super not affected real removing devices */
 759         if (sbi) {
 760                 sb->s_dev = sbi->ll_sdev_orig;
 761                 sbi->ll_umounting = 1;
 762
 763                 /* wait running statahead threads to quit */
 764                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 765                         set_current_state(TASK_UNINTERRUPTIBLE);
 766                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 767                 }
 768         }
 769
 770         EXIT;
 771 }
 772
 773 static inline int ll_set_opt(const char *opt, char *data, int fl)
 774 {
 775         if (strncmp(opt, data, strlen(opt)) != 0)
 776                 return(0);
 777         else
 778                 return(fl);
 779 }
 780
 781 /* non-client-specific mount options are parsed in lmd_parse */
 782 static int ll_options(char *options, int *flags)
 783 {
 784         int tmp;
 785         char *s1 = options, *s2;
 786         ENTRY;
 787
 788         if (!options)
 789                 RETURN(0);
 790
 791         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 792
 793         while (*s1) {
 794                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 795                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 796                 if (tmp) {
 797                         *flags |= tmp;
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 806                 if (tmp) {
 807                         *flags |= tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 811                 if (tmp) {
 812                         *flags &= ~tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 821                 if (tmp) {
 822                         *flags &= ~tmp;
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("context", s1, 1);
 826                 if (tmp)
 827                         goto next;
 828                 tmp = ll_set_opt("fscontext", s1, 1);
 829                 if (tmp)
 830                         goto next;
 831                 tmp = ll_set_opt("defcontext", s1, 1);
 832                 if (tmp)
 833                         goto next;
 834                 tmp = ll_set_opt("rootcontext", s1, 1);
 835                 if (tmp)
 836                         goto next;
 837                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847
 848                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 859                 if (tmp) {
 860                         *flags |= tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 864                 if (tmp) {
 865                         *flags &= ~tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 884                 if (tmp) {
 885                         *flags |= tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 889                 if (tmp) {
 890                         *flags &= ~tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 894                 if (tmp) {
 895                         *flags |= tmp;
 896                         goto next;
 897                 }
 898                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 899                                    s1);
 900                 RETURN(-EINVAL);
 901
 902 next:
 903                 /* Find next opt */
 904                 s2 = strchr(s1, ',');
 905                 if (s2 == NULL)
 906                         break;
 907                 s1 = s2 + 1;
 908         }
 909         RETURN(0);
 910 }
 911
 912 void ll_lli_init(struct ll_inode_info *lli)
 913 {
 914         lli->lli_inode_magic = LLI_INODE_MAGIC;
 915         lli->lli_flags = 0;
 916         spin_lock_init(&lli->lli_lock);
 917         lli->lli_posix_acl = NULL;
 918         /* Do not set lli_fid, it has been initialized already. */
 919         fid_zero(&lli->lli_pfid);
 920         lli->lli_mds_read_och = NULL;
 921         lli->lli_mds_write_och = NULL;
 922         lli->lli_mds_exec_och = NULL;
 923         lli->lli_open_fd_read_count = 0;
 924         lli->lli_open_fd_write_count = 0;
 925         lli->lli_open_fd_exec_count = 0;
 926         mutex_init(&lli->lli_och_mutex);
 927         spin_lock_init(&lli->lli_agl_lock);
 928         spin_lock_init(&lli->lli_layout_lock);
 929         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 930         lli->lli_clob = NULL;
 931
 932         init_rwsem(&lli->lli_xattrs_list_rwsem);
 933         mutex_init(&lli->lli_xattrs_enq_lock);
 934
 935         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 936         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 937                 mutex_init(&lli->lli_readdir_mutex);
 938                 lli->lli_opendir_key = NULL;
 939                 lli->lli_sai = NULL;
 940                 spin_lock_init(&lli->lli_sa_lock);
 941                 lli->lli_opendir_pid = 0;
 942                 lli->lli_sa_enabled = 0;
 943                 lli->lli_def_stripe_offset = -1;
 944         } else {
 945                 mutex_init(&lli->lli_size_mutex);
 946                 lli->lli_symlink_name = NULL;
 947                 init_rwsem(&lli->lli_trunc_sem);
 948                 range_lock_tree_init(&lli->lli_write_tree);
 949                 init_rwsem(&lli->lli_glimpse_sem);
 950                 lli->lli_glimpse_time = 0;
 951                 INIT_LIST_HEAD(&lli->lli_agl_list);
 952                 lli->lli_agl_index = 0;
 953                 lli->lli_async_rc = 0;
 954         }
 955         mutex_init(&lli->lli_layout_mutex);
 956         memset(lli->lli_jobid, 0, LUSTRE_JOBID_SIZE);
 957 }
 958
 959 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 960
 961 #define LSI_BDI_INITIALIZED     0x00400000
 962
 963 #ifndef HAVE_BDI_CAP_MAP_COPY
 964 # define BDI_CAP_MAP_COPY       0
 965 #endif
 966
 967 #define MAX_STRING_SIZE 128
 968
 969 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 970 {
 971         struct  lustre_sb_info *lsi = s2lsi(sb);
 972         char buf[MAX_STRING_SIZE];
 973         va_list args;
 974         int err;
 975
 976         err = bdi_init(&lsi->lsi_bdi);
 977         if (err)
 978                 return err;
 979
 980         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 981         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 982         lsi->lsi_bdi.name = "lustre";
 983         va_start(args, fmt);
 984         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
 985         va_end(args);
 986         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
 987         va_end(args);
 988         if (!err)
 989                 sb->s_bdi = &lsi->lsi_bdi;
 990
 991         return err;
 992 }
 993 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
 994
 995 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 996 {
 997         struct  lustre_profile *lprof = NULL;
 998         struct  lustre_sb_info *lsi = s2lsi(sb);
 999         struct  ll_sb_info *sbi;
1000         char    *dt = NULL, *md = NULL;
1001         char    *profilenm = get_profile_name(sb);
1002         struct config_llog_instance *cfg;
1003         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1004         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1005         int     md_len = 0;
1006         int     dt_len = 0;
1007         int     err;
1008         ENTRY;
1009
1010         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1011
1012         OBD_ALLOC_PTR(cfg);
1013         if (cfg == NULL)
1014                 RETURN(-ENOMEM);
1015
1016         try_module_get(THIS_MODULE);
1017
1018         /* client additional sb info */
1019         lsi->lsi_llsbi = sbi = ll_init_sbi();
1020         if (!sbi) {
1021                 module_put(THIS_MODULE);
1022                 OBD_FREE_PTR(cfg);
1023                 RETURN(-ENOMEM);
1024         }
1025
1026         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1027         if (err)
1028                 GOTO(out_free, err);
1029
1030         err = super_setup_bdi_name(sb, "lustre-%p", sb);
1031         if (err)
1032                 GOTO(out_free, err);
1033
1034 #ifndef HAVE_DCACHE_LOCK
1035         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1036         sb->s_d_op = &ll_d_ops;
1037 #endif
1038
1039         /* Call lprocfs_ll_register_mountpoint() before lustre_process_log()
1040          * so that "llite.*.*" params can be processed correctly. */
1041         if (proc_lustre_fs_root != NULL) {
1042                 err = lprocfs_ll_register_mountpoint(proc_lustre_fs_root, sb);
1043                 if (err < 0) {
1044                         CERROR("%s: could not register mountpoint in llite: "
1045                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
1046                         err = 0;
1047                 }
1048         }
1049
1050         /* Generate a string unique to this super, in case some joker tries
1051            to mount the same fs at two mount points.
1052            Use the address of the super itself.*/
1053         cfg->cfg_instance = sb;
1054         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1055         cfg->cfg_callback = class_config_llog_handler;
1056         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1057         /* set up client obds */
1058         err = lustre_process_log(sb, profilenm, cfg);
1059         if (err < 0)
1060                 GOTO(out_proc, err);
1061
1062         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1063         lprof = class_get_profile(profilenm);
1064         if (lprof == NULL) {
1065                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1066                                    " read from the MGS.  Does that filesystem "
1067                                    "exist?\n", profilenm);
1068                 GOTO(out_proc, err = -EINVAL);
1069         }
1070         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1071                lprof->lp_md, lprof->lp_dt);
1072
1073         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1074         OBD_ALLOC(dt, dt_len);
1075         if (!dt)
1076                 GOTO(out_proc, err = -ENOMEM);
1077         snprintf(dt, dt_len - 1, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1078
1079         md_len = strlen(lprof->lp_md) + instlen + 2;
1080         OBD_ALLOC(md, md_len);
1081         if (!md)
1082                 GOTO(out_proc, err = -ENOMEM);
1083         snprintf(md, md_len - 1, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1084
1085         /* connections, registrations, sb setup */
1086         err = client_common_fill_super(sb, md, dt, mnt);
1087         if (err < 0)
1088                 GOTO(out_proc, err);
1089
1090         sbi->ll_client_common_fill_super_succeeded = 1;
1091
1092 out_proc:
1093         if (err < 0)
1094                 lprocfs_ll_unregister_mountpoint(sbi);
1095 out_free:
1096         if (md)
1097                 OBD_FREE(md, md_len);
1098         if (dt)
1099                 OBD_FREE(dt, dt_len);
1100         if (lprof != NULL)
1101                 class_put_profile(lprof);
1102         if (err)
1103                 ll_put_super(sb);
1104         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1105                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1106
1107         OBD_FREE_PTR(cfg);
1108         RETURN(err);
1109 } /* ll_fill_super */
1110
1111 void ll_put_super(struct super_block *sb)
1112 {
1113         struct config_llog_instance cfg, params_cfg;
1114         struct obd_device *obd;
1115         struct lustre_sb_info *lsi = s2lsi(sb);
1116         struct ll_sb_info *sbi = ll_s2sbi(sb);
1117         char *profilenm = get_profile_name(sb);
1118         long ccc_count;
1119         int next, force = 1, rc = 0;
1120         ENTRY;
1121
1122         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1123
1124         cfg.cfg_instance = sb;
1125         lustre_end_log(sb, profilenm, &cfg);
1126
1127         params_cfg.cfg_instance = sb;
1128         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1129
1130         if (sbi->ll_md_exp) {
1131                 obd = class_exp2obd(sbi->ll_md_exp);
1132                 if (obd)
1133                         force = obd->obd_force;
1134         }
1135
1136         /* Wait for unstable pages to be committed to stable storage */
1137         if (force == 0) {
1138                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1139                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1140                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1141                         &lwi);
1142         }
1143
1144         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1145         if (force == 0 && rc != -EINTR)
1146                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1147
1148
1149         /* We need to set force before the lov_disconnect in
1150            lustre_common_put_super, since l_d cleans up osc's as well. */
1151         if (force) {
1152                 next = 0;
1153                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1154                                                      &next)) != NULL) {
1155                         obd->obd_force = force;
1156                 }
1157         }
1158
1159         if (sbi->ll_client_common_fill_super_succeeded) {
1160                 /* Only if client_common_fill_super succeeded */
1161                 client_common_put_super(sb);
1162         }
1163
1164         next = 0;
1165         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1166                 class_manual_cleanup(obd);
1167         }
1168
1169         if (sbi->ll_flags & LL_SBI_VERBOSE)
1170                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1171
1172         if (profilenm)
1173                 class_del_profile(profilenm);
1174
1175 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1176         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1177                 bdi_destroy(&lsi->lsi_bdi);
1178                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1179         }
1180 #endif
1181
1182         ll_free_sbi(sb);
1183         lsi->lsi_llsbi = NULL;
1184
1185         lustre_common_put_super(sb);
1186
1187         cl_env_cache_purge(~0);
1188
1189         module_put(THIS_MODULE);
1190
1191         EXIT;
1192 } /* client_put_super */
1193
1194 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1195 {
1196         struct inode *inode = NULL;
1197
1198         /* NOTE: we depend on atomic igrab() -bzzz */
1199         lock_res_and_lock(lock);
1200         if (lock->l_resource->lr_lvb_inode) {
1201                 struct ll_inode_info * lli;
1202                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1203                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1204                         inode = igrab(lock->l_resource->lr_lvb_inode);
1205                 } else {
1206                         inode = lock->l_resource->lr_lvb_inode;
1207                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1208                                          D_WARNING, lock, "lr_lvb_inode %p is "
1209                                          "bogus: magic %08x",
1210                                          lock->l_resource->lr_lvb_inode,
1211                                          lli->lli_inode_magic);
1212                         inode = NULL;
1213                 }
1214         }
1215         unlock_res_and_lock(lock);
1216         return inode;
1217 }
1218
1219 void ll_dir_clear_lsm_md(struct inode *inode)
1220 {
1221         struct ll_inode_info *lli = ll_i2info(inode);
1222
1223         LASSERT(S_ISDIR(inode->i_mode));
1224
1225         if (lli->lli_lsm_md != NULL) {
1226                 lmv_free_memmd(lli->lli_lsm_md);
1227                 lli->lli_lsm_md = NULL;
1228         }
1229 }
1230
1231 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1232                                       const struct lu_fid *fid,
1233                                       struct lustre_md *md)
1234 {
1235         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1236         struct mdt_body         *body = md->body;
1237         struct inode            *inode;
1238         ino_t                   ino;
1239         ENTRY;
1240
1241         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1242         inode = iget_locked(sb, ino);
1243         if (inode == NULL) {
1244                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1245                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1246                 RETURN(ERR_PTR(-ENOENT));
1247         }
1248
1249         if (inode->i_state & I_NEW) {
1250                 struct ll_inode_info *lli = ll_i2info(inode);
1251                 struct lmv_stripe_md *lsm = md->lmv;
1252
1253                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1254                                 (body->mbo_mode & S_IFMT);
1255                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1256                          PFID(fid));
1257
1258                 LTIME_S(inode->i_mtime) = 0;
1259                 LTIME_S(inode->i_atime) = 0;
1260                 LTIME_S(inode->i_ctime) = 0;
1261                 inode->i_rdev = 0;
1262
1263 #ifdef HAVE_BACKING_DEV_INFO
1264                 /* initializing backing dev info. */
1265                 inode->i_mapping->backing_dev_info =
1266                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1267 #endif
1268                 inode->i_op = &ll_dir_inode_operations;
1269                 inode->i_fop = &ll_dir_operations;
1270                 lli->lli_fid = *fid;
1271                 ll_lli_init(lli);
1272
1273                 LASSERT(lsm != NULL);
1274                 /* master object FID */
1275                 lli->lli_pfid = body->mbo_fid1;
1276                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1277                        lli, PFID(fid), PFID(&lli->lli_pfid));
1278                 unlock_new_inode(inode);
1279         }
1280
1281         RETURN(inode);
1282 }
1283
1284 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1285 {
1286         struct lu_fid *fid;
1287         struct lmv_stripe_md *lsm = md->lmv;
1288         int i;
1289
1290         LASSERT(lsm != NULL);
1291         /* XXX sigh, this lsm_root initialization should be in
1292          * LMV layer, but it needs ll_iget right now, so we
1293          * put this here right now. */
1294         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1295                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1296                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1297                 /* Unfortunately ll_iget will call ll_update_inode,
1298                  * where the initialization of slave inode is slightly
1299                  * different, so it reset lsm_md to NULL to avoid
1300                  * initializing lsm for slave inode. */
1301                 /* For migrating inode, master stripe and master object will
1302                  * be same, so we only need assign this inode */
1303                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1304                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1305                 else
1306                         lsm->lsm_md_oinfo[i].lmo_root =
1307                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1308
1309                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1310                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1311
1312                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1313                         return rc;
1314                 }
1315         }
1316
1317         return 0;
1318 }
1319
1320 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1321                                 const struct lmv_stripe_md *lsm_md2)
1322 {
1323         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1324                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1325                lsm_md1->lsm_md_master_mdt_index ==
1326                                         lsm_md2->lsm_md_master_mdt_index &&
1327                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1328                lsm_md1->lsm_md_layout_version ==
1329                                         lsm_md2->lsm_md_layout_version &&
1330                strcmp(lsm_md1->lsm_md_pool_name,
1331                       lsm_md2->lsm_md_pool_name) == 0;
1332 }
1333
1334 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1335 {
1336         struct ll_inode_info *lli = ll_i2info(inode);
1337         struct lmv_stripe_md *lsm = md->lmv;
1338         int     rc;
1339         ENTRY;
1340
1341         LASSERT(S_ISDIR(inode->i_mode));
1342         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1343                PFID(ll_inode2fid(inode)));
1344
1345         /* no striped information from request. */
1346         if (lsm == NULL) {
1347                 if (lli->lli_lsm_md == NULL) {
1348                         RETURN(0);
1349                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1350                                                 LMV_HASH_FLAG_MIGRATION) {
1351                         /* migration is done, the temporay MIGRATE layout has
1352                          * been removed */
1353                         CDEBUG(D_INODE, DFID" finish migration.\n",
1354                                PFID(ll_inode2fid(inode)));
1355                         lmv_free_memmd(lli->lli_lsm_md);
1356                         lli->lli_lsm_md = NULL;
1357                         RETURN(0);
1358                 } else {
1359                         /* The lustre_md from req does not include stripeEA,
1360                          * see ll_md_setattr */
1361                         RETURN(0);
1362                 }
1363         }
1364
1365         /* set the directory layout */
1366         if (lli->lli_lsm_md == NULL) {
1367                 struct cl_attr  *attr;
1368
1369                 rc = ll_init_lsm_md(inode, md);
1370                 if (rc != 0)
1371                         RETURN(rc);
1372
1373                 /* set md->lmv to NULL, so the following free lustre_md
1374                  * will not free this lsm */
1375                 md->lmv = NULL;
1376                 lli->lli_lsm_md = lsm;
1377
1378                 OBD_ALLOC_PTR(attr);
1379                 if (attr == NULL)
1380                         RETURN(-ENOMEM);
1381
1382                 /* validate the lsm */
1383                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1384                                    ll_md_blocking_ast);
1385                 if (rc != 0) {
1386                         OBD_FREE_PTR(attr);
1387                         RETURN(rc);
1388                 }
1389
1390                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1391                         md->body->mbo_nlink = attr->cat_nlink;
1392                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1393                         md->body->mbo_size = attr->cat_size;
1394                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1395                         md->body->mbo_atime = attr->cat_atime;
1396                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1397                         md->body->mbo_ctime = attr->cat_ctime;
1398                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1399                         md->body->mbo_mtime = attr->cat_mtime;
1400
1401                 OBD_FREE_PTR(attr);
1402
1403                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1404                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1405                 RETURN(0);
1406         }
1407
1408         /* Compare the old and new stripe information */
1409         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1410                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1411                 int                     idx;
1412
1413                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1414                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1415                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1416                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1417                        inode, lsm, old_lsm,
1418                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1419                        lsm->lsm_md_stripe_count,
1420                        old_lsm->lsm_md_stripe_count,
1421                        lsm->lsm_md_master_mdt_index,
1422                        old_lsm->lsm_md_master_mdt_index,
1423                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1424                        lsm->lsm_md_layout_version,
1425                        old_lsm->lsm_md_layout_version,
1426                        lsm->lsm_md_pool_name,
1427                        old_lsm->lsm_md_pool_name);
1428
1429                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1430                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1431                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1432                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1433                 }
1434
1435                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1436                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1437                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1438                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1439                 }
1440
1441                 RETURN(-EIO);
1442         }
1443
1444         RETURN(0);
1445 }
1446
1447 void ll_clear_inode(struct inode *inode)
1448 {
1449         struct ll_inode_info *lli = ll_i2info(inode);
1450         struct ll_sb_info *sbi = ll_i2sbi(inode);
1451         ENTRY;
1452
1453         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1454                PFID(ll_inode2fid(inode)), inode);
1455
1456         if (S_ISDIR(inode->i_mode)) {
1457                 /* these should have been cleared in ll_file_release */
1458                 LASSERT(lli->lli_opendir_key == NULL);
1459                 LASSERT(lli->lli_sai == NULL);
1460                 LASSERT(lli->lli_opendir_pid == 0);
1461         }
1462
1463         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1464
1465         LASSERT(!lli->lli_open_fd_write_count);
1466         LASSERT(!lli->lli_open_fd_read_count);
1467         LASSERT(!lli->lli_open_fd_exec_count);
1468
1469         if (lli->lli_mds_write_och)
1470                 ll_md_real_close(inode, FMODE_WRITE);
1471         if (lli->lli_mds_exec_och)
1472                 ll_md_real_close(inode, FMODE_EXEC);
1473         if (lli->lli_mds_read_och)
1474                 ll_md_real_close(inode, FMODE_READ);
1475
1476         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1477                 OBD_FREE(lli->lli_symlink_name,
1478                          strlen(lli->lli_symlink_name) + 1);
1479                 lli->lli_symlink_name = NULL;
1480         }
1481
1482         ll_xattr_cache_destroy(inode);
1483
1484 #ifdef CONFIG_FS_POSIX_ACL
1485         forget_all_cached_acls(inode);
1486         if (lli->lli_posix_acl) {
1487                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1488                 posix_acl_release(lli->lli_posix_acl);
1489                 lli->lli_posix_acl = NULL;
1490         }
1491 #endif
1492         lli->lli_inode_magic = LLI_INODE_DEAD;
1493
1494         if (S_ISDIR(inode->i_mode))
1495                 ll_dir_clear_lsm_md(inode);
1496         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1497                 LASSERT(list_empty(&lli->lli_agl_list));
1498
1499         /*
1500          * XXX This has to be done before lsm is freed below, because
1501          * cl_object still uses inode lsm.
1502          */
1503         cl_inode_fini(inode);
1504
1505         EXIT;
1506 }
1507
1508 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1509 {
1510         struct lustre_md md;
1511         struct inode *inode = dentry->d_inode;
1512         struct ll_sb_info *sbi = ll_i2sbi(inode);
1513         struct ptlrpc_request *request = NULL;
1514         int rc, ia_valid;
1515         ENTRY;
1516
1517         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1518                                      LUSTRE_OPC_ANY, NULL);
1519         if (IS_ERR(op_data))
1520                 RETURN(PTR_ERR(op_data));
1521
1522         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1523         if (rc) {
1524                 ptlrpc_req_finished(request);
1525                 if (rc == -ENOENT) {
1526                         clear_nlink(inode);
1527                         /* Unlinked special device node? Or just a race?
1528                          * Pretend we done everything. */
1529                         if (!S_ISREG(inode->i_mode) &&
1530                             !S_ISDIR(inode->i_mode)) {
1531                                 ia_valid = op_data->op_attr.ia_valid;
1532                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1533                                 rc = simple_setattr(dentry, &op_data->op_attr);
1534                                 op_data->op_attr.ia_valid = ia_valid;
1535                         }
1536                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1537                         CERROR("md_setattr fails: rc = %d\n", rc);
1538                 }
1539                 RETURN(rc);
1540         }
1541
1542         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1543                               sbi->ll_md_exp, &md);
1544         if (rc) {
1545                 ptlrpc_req_finished(request);
1546                 RETURN(rc);
1547         }
1548
1549         ia_valid = op_data->op_attr.ia_valid;
1550         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1551          * cache is not cleared yet. */
1552         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1553         if (S_ISREG(inode->i_mode))
1554                 inode_lock(inode);
1555         rc = simple_setattr(dentry, &op_data->op_attr);
1556         if (S_ISREG(inode->i_mode))
1557                 inode_unlock(inode);
1558         op_data->op_attr.ia_valid = ia_valid;
1559
1560         rc = ll_update_inode(inode, &md);
1561         ptlrpc_req_finished(request);
1562
1563         RETURN(rc);
1564 }
1565
1566 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1567  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1568  * keep these values until such a time that objects are allocated for it.
1569  * We do the MDS operations first, as it is checking permissions for us.
1570  * We don't to the MDS RPC if there is nothing that we want to store there,
1571  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1572  * going to do an RPC anyways.
1573  *
1574  * If we are doing a truncate, we will send the mtime and ctime updates
1575  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1576  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1577  * at the same time.
1578  *
1579  * In case of HSMimport, we only set attr on MDS.
1580  */
1581 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1582 {
1583         struct inode *inode = dentry->d_inode;
1584         struct ll_inode_info *lli = ll_i2info(inode);
1585         struct md_op_data *op_data = NULL;
1586         int rc = 0;
1587         ENTRY;
1588
1589         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1590                "valid %x, hsm_import %d\n",
1591                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1592                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1593                hsm_import);
1594
1595         if (attr->ia_valid & ATTR_SIZE) {
1596                 /* Check new size against VFS/VM file size limit and rlimit */
1597                 rc = inode_newsize_ok(inode, attr->ia_size);
1598                 if (rc)
1599                         RETURN(rc);
1600
1601                 /* The maximum Lustre file size is variable, based on the
1602                  * OST maximum object size and number of stripes.  This
1603                  * needs another check in addition to the VFS check above. */
1604                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1605                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1606                                PFID(&lli->lli_fid), attr->ia_size,
1607                                ll_file_maxbytes(inode));
1608                         RETURN(-EFBIG);
1609                 }
1610
1611                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1612         }
1613
1614         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1615         if (attr->ia_valid & TIMES_SET_FLAGS) {
1616                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1617                     !cfs_capable(CFS_CAP_FOWNER))
1618                         RETURN(-EPERM);
1619         }
1620
1621         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1622         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1623             (attr->ia_valid & ATTR_CTIME)) {
1624                 attr->ia_ctime = current_time(inode);
1625                 attr->ia_valid |= ATTR_CTIME_SET;
1626         }
1627         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1628             (attr->ia_valid & ATTR_ATIME)) {
1629                 attr->ia_atime = current_time(inode);
1630                 attr->ia_valid |= ATTR_ATIME_SET;
1631         }
1632         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1633             (attr->ia_valid & ATTR_MTIME)) {
1634                 attr->ia_mtime = current_time(inode);
1635                 attr->ia_valid |= ATTR_MTIME_SET;
1636         }
1637
1638         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1639                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1640                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1641                        (s64)ktime_get_real_seconds());
1642
1643         if (S_ISREG(inode->i_mode)) {
1644                 if (attr->ia_valid & ATTR_SIZE)
1645                         inode_dio_write_done(inode);
1646                 inode_unlock(inode);
1647         }
1648
1649         /* We always do an MDS RPC, even if we're only changing the size;
1650          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1651
1652         OBD_ALLOC_PTR(op_data);
1653         if (op_data == NULL)
1654                 GOTO(out, rc = -ENOMEM);
1655
1656         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1657                 /* If we are changing file size, file content is
1658                  * modified, flag it. */
1659                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1660                 op_data->op_bias |= MDS_DATA_MODIFIED;
1661                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1662         }
1663
1664         op_data->op_attr = *attr;
1665
1666         rc = ll_md_setattr(dentry, op_data);
1667         if (rc)
1668                 GOTO(out, rc);
1669
1670         if (!S_ISREG(inode->i_mode) || hsm_import)
1671                 GOTO(out, rc = 0);
1672
1673         if (attr->ia_valid & (ATTR_SIZE |
1674                               ATTR_ATIME | ATTR_ATIME_SET |
1675                               ATTR_MTIME | ATTR_MTIME_SET |
1676                               ATTR_CTIME | ATTR_CTIME_SET)) {
1677                 /* For truncate and utimes sending attributes to OSTs, setting
1678                  * mtime/atime to the past will be performed under PW [0:EOF]
1679                  * extent lock (new_size:EOF for truncate).  It may seem
1680                  * excessive to send mtime/atime updates to OSTs when not
1681                  * setting times to past, but it is necessary due to possible
1682                  * time de-synchronization between MDT inode and OST objects */
1683                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1684         }
1685
1686         /* If the file was restored, it needs to set dirty flag.
1687          *
1688          * We've already sent MDS_DATA_MODIFIED flag in
1689          * ll_md_setattr() for truncate. However, the MDT refuses to
1690          * set the HS_DIRTY flag on released files, so we have to set
1691          * it again if the file has been restored. Please check how
1692          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1693          *
1694          * Please notice that if the file is not released, the previous
1695          * MDS_DATA_MODIFIED has taken effect and usually
1696          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1697          * This way we can save an RPC for common open + trunc
1698          * operation. */
1699         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1700                 struct hsm_state_set hss = {
1701                         .hss_valid = HSS_SETMASK,
1702                         .hss_setmask = HS_DIRTY,
1703                 };
1704                 int rc2;
1705
1706                 rc2 = ll_hsm_state_set(inode, &hss);
1707                 /* truncate and write can happen at the same time, so that
1708                  * the file can be set modified even though the file is not
1709                  * restored from released state, and ll_hsm_state_set() is
1710                  * not applicable for the file, and rc2 < 0 is normal in this
1711                  * case. */
1712                 if (rc2 < 0)
1713                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1714                                PFID(ll_inode2fid(inode)), rc2);
1715         }
1716
1717         EXIT;
1718 out:
1719         if (op_data != NULL)
1720                 ll_finish_md_op_data(op_data);
1721
1722         if (S_ISREG(inode->i_mode)) {
1723                 inode_lock(inode);
1724                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1725                         inode_dio_wait(inode);
1726                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1727                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1728                  * inode flags, so there is a gap where S_NOSEC is not set.
1729                  * This can cause a writer to take the i_mutex unnecessarily,
1730                  * but this is safe to do and should be rare. */
1731                 inode_has_no_xattr(inode);
1732         }
1733
1734         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1735                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1736
1737         return rc;
1738 }
1739
1740 int ll_setattr(struct dentry *de, struct iattr *attr)
1741 {
1742         int mode = de->d_inode->i_mode;
1743
1744         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1745                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1746                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1747
1748         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1749                                (ATTR_SIZE|ATTR_MODE)) &&
1750             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1751              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1752               !(attr->ia_mode & S_ISGID))))
1753                 attr->ia_valid |= ATTR_FORCE;
1754
1755         if ((attr->ia_valid & ATTR_MODE) &&
1756             (mode & S_ISUID) &&
1757             !(attr->ia_mode & S_ISUID) &&
1758             !(attr->ia_valid & ATTR_KILL_SUID))
1759                 attr->ia_valid |= ATTR_KILL_SUID;
1760
1761         if ((attr->ia_valid & ATTR_MODE) &&
1762             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1763             !(attr->ia_mode & S_ISGID) &&
1764             !(attr->ia_valid & ATTR_KILL_SGID))
1765                 attr->ia_valid |= ATTR_KILL_SGID;
1766
1767         /* avoid polluted from ATTR_TIMES_SET,
1768          * projid is not expected to be set here */
1769         attr->ia_valid &= ~MDS_ATTR_PROJID;
1770
1771         return ll_setattr_raw(de, attr, false);
1772 }
1773
1774 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1775                        __u64 max_age, __u32 flags)
1776 {
1777         struct ll_sb_info *sbi = ll_s2sbi(sb);
1778         struct obd_statfs obd_osfs;
1779         int rc;
1780         ENTRY;
1781
1782         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1783         if (rc) {
1784                 CERROR("md_statfs fails: rc = %d\n", rc);
1785                 RETURN(rc);
1786         }
1787
1788         osfs->os_type = sb->s_magic;
1789
1790         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1791                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1792
1793         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1794                 flags |= OBD_STATFS_NODELAY;
1795
1796         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1797         if (rc) {
1798                 CERROR("obd_statfs fails: rc = %d\n", rc);
1799                 RETURN(rc);
1800         }
1801
1802         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1803                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1804                obd_osfs.os_files);
1805
1806         osfs->os_bsize = obd_osfs.os_bsize;
1807         osfs->os_blocks = obd_osfs.os_blocks;
1808         osfs->os_bfree = obd_osfs.os_bfree;
1809         osfs->os_bavail = obd_osfs.os_bavail;
1810
1811         /* If we don't have as many objects free on the OST as inodes
1812          * on the MDS, we reduce the total number of inodes to
1813          * compensate, so that the "inodes in use" number is correct.
1814          */
1815         if (obd_osfs.os_ffree < osfs->os_ffree) {
1816                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1817                         obd_osfs.os_ffree;
1818                 osfs->os_ffree = obd_osfs.os_ffree;
1819         }
1820
1821         RETURN(rc);
1822 }
1823 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1824 {
1825         struct super_block *sb = de->d_sb;
1826         struct obd_statfs osfs;
1827         __u64 fsid = huge_encode_dev(sb->s_dev);
1828         int rc;
1829
1830         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1831         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1832
1833         /* Some amount of caching on the client is allowed */
1834         rc = ll_statfs_internal(sb, &osfs,
1835                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1836                                 0);
1837         if (rc)
1838                 return rc;
1839
1840         statfs_unpack(sfs, &osfs);
1841
1842         /* We need to downshift for all 32-bit kernels, because we can't
1843          * tell if the kernel is being called via sys_statfs64() or not.
1844          * Stop before overflowing f_bsize - in which case it is better
1845          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1846         if (sizeof(long) < 8) {
1847                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1848                         sfs->f_bsize <<= 1;
1849
1850                         osfs.os_blocks >>= 1;
1851                         osfs.os_bfree >>= 1;
1852                         osfs.os_bavail >>= 1;
1853                 }
1854         }
1855
1856         sfs->f_blocks = osfs.os_blocks;
1857         sfs->f_bfree = osfs.os_bfree;
1858         sfs->f_bavail = osfs.os_bavail;
1859         sfs->f_fsid.val[0] = (__u32)fsid;
1860         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1861         return 0;
1862 }
1863
1864 void ll_inode_size_lock(struct inode *inode)
1865 {
1866         struct ll_inode_info *lli;
1867
1868         LASSERT(!S_ISDIR(inode->i_mode));
1869
1870         lli = ll_i2info(inode);
1871         mutex_lock(&lli->lli_size_mutex);
1872 }
1873
1874 void ll_inode_size_unlock(struct inode *inode)
1875 {
1876         struct ll_inode_info *lli;
1877
1878         lli = ll_i2info(inode);
1879         mutex_unlock(&lli->lli_size_mutex);
1880 }
1881
1882 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1883 {
1884         struct ll_inode_info *lli = ll_i2info(inode);
1885         struct mdt_body *body = md->body;
1886         struct ll_sb_info *sbi = ll_i2sbi(inode);
1887         int rc = 0;
1888
1889         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1890                 rc = cl_file_inode_init(inode, md);
1891                 if (rc)
1892                         return rc;
1893         }
1894
1895         if (S_ISDIR(inode->i_mode)) {
1896                 rc = ll_update_lsm_md(inode, md);
1897                 if (rc != 0)
1898                         return rc;
1899         }
1900
1901 #ifdef CONFIG_FS_POSIX_ACL
1902         if (body->mbo_valid & OBD_MD_FLACL) {
1903                 spin_lock(&lli->lli_lock);
1904                 if (lli->lli_posix_acl)
1905                         posix_acl_release(lli->lli_posix_acl);
1906                 lli->lli_posix_acl = md->posix_acl;
1907                 spin_unlock(&lli->lli_lock);
1908         }
1909 #endif
1910         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1911                                         sbi->ll_flags & LL_SBI_32BIT_API);
1912         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1913
1914         if (body->mbo_valid & OBD_MD_FLATIME) {
1915                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1916                         LTIME_S(inode->i_atime) = body->mbo_atime;
1917                 lli->lli_atime = body->mbo_atime;
1918         }
1919
1920         if (body->mbo_valid & OBD_MD_FLMTIME) {
1921                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1922                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1923                                "to %llu\n", inode->i_ino,
1924                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1925                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1926                 }
1927                 lli->lli_mtime = body->mbo_mtime;
1928         }
1929
1930         if (body->mbo_valid & OBD_MD_FLCTIME) {
1931                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1932                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1933                 lli->lli_ctime = body->mbo_ctime;
1934         }
1935
1936         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1937         if (body->mbo_valid & OBD_MD_FLFLAGS)
1938                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1939         if (body->mbo_valid & OBD_MD_FLMODE)
1940                 inode->i_mode = (inode->i_mode & S_IFMT) |
1941                                 (body->mbo_mode & ~S_IFMT);
1942
1943         if (body->mbo_valid & OBD_MD_FLTYPE)
1944                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1945                                 (body->mbo_mode & S_IFMT);
1946
1947         LASSERT(inode->i_mode != 0);
1948         if (S_ISREG(inode->i_mode))
1949                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1950                                        LL_MAX_BLKSIZE_BITS);
1951         else
1952                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1953
1954         if (body->mbo_valid & OBD_MD_FLUID)
1955                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1956         if (body->mbo_valid & OBD_MD_FLGID)
1957                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1958         if (body->mbo_valid & OBD_MD_FLPROJID)
1959                 lli->lli_projid = body->mbo_projid;
1960         if (body->mbo_valid & OBD_MD_FLNLINK)
1961                 set_nlink(inode, body->mbo_nlink);
1962         if (body->mbo_valid & OBD_MD_FLRDEV)
1963                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1964
1965         if (body->mbo_valid & OBD_MD_FLID) {
1966                 /* FID shouldn't be changed! */
1967                 if (fid_is_sane(&lli->lli_fid)) {
1968                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1969                                  "Trying to change FID "DFID
1970                                  " to the "DFID", inode "DFID"(%p)\n",
1971                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1972                                  PFID(ll_inode2fid(inode)), inode);
1973                 } else {
1974                         lli->lli_fid = body->mbo_fid1;
1975                 }
1976         }
1977
1978         LASSERT(fid_seq(&lli->lli_fid) != 0);
1979
1980         if (body->mbo_valid & OBD_MD_FLSIZE) {
1981                 i_size_write(inode, body->mbo_size);
1982
1983                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1984                        PFID(ll_inode2fid(inode)),
1985                        (unsigned long long)body->mbo_size);
1986
1987                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1988                         inode->i_blocks = body->mbo_blocks;
1989         }
1990
1991         if (body->mbo_valid & OBD_MD_TSTATE) {
1992                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1993                  * clear it when done to ensure to start again
1994                  * glimpsing updated attrs
1995                  */
1996                 if (body->mbo_t_state & MS_RESTORE)
1997                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1998                 else
1999                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2000         }
2001
2002         return 0;
2003 }
2004
2005 int ll_read_inode2(struct inode *inode, void *opaque)
2006 {
2007         struct lustre_md *md = opaque;
2008         struct ll_inode_info *lli = ll_i2info(inode);
2009         int     rc;
2010         ENTRY;
2011
2012         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2013                PFID(&lli->lli_fid), inode);
2014
2015         /* Core attributes from the MDS first.  This is a new inode, and
2016          * the VFS doesn't zero times in the core inode so we have to do
2017          * it ourselves.  They will be overwritten by either MDS or OST
2018          * attributes - we just need to make sure they aren't newer. */
2019         LTIME_S(inode->i_mtime) = 0;
2020         LTIME_S(inode->i_atime) = 0;
2021         LTIME_S(inode->i_ctime) = 0;
2022         inode->i_rdev = 0;
2023         rc = ll_update_inode(inode, md);
2024         if (rc != 0)
2025                 RETURN(rc);
2026
2027         /* OIDEBUG(inode); */
2028
2029 #ifdef HAVE_BACKING_DEV_INFO
2030         /* initializing backing dev info. */
2031         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2032 #endif
2033         if (S_ISREG(inode->i_mode)) {
2034                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2035                 inode->i_op = &ll_file_inode_operations;
2036                 inode->i_fop = sbi->ll_fop;
2037                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2038                 EXIT;
2039         } else if (S_ISDIR(inode->i_mode)) {
2040                 inode->i_op = &ll_dir_inode_operations;
2041                 inode->i_fop = &ll_dir_operations;
2042                 EXIT;
2043         } else if (S_ISLNK(inode->i_mode)) {
2044                 inode->i_op = &ll_fast_symlink_inode_operations;
2045                 EXIT;
2046         } else {
2047                 inode->i_op = &ll_special_inode_operations;
2048
2049                 init_special_inode(inode, inode->i_mode,
2050                                    inode->i_rdev);
2051
2052                 EXIT;
2053         }
2054
2055         return 0;
2056 }
2057
2058 void ll_delete_inode(struct inode *inode)
2059 {
2060         struct ll_inode_info *lli = ll_i2info(inode);
2061         ENTRY;
2062
2063         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2064                 /* It is last chance to write out dirty pages,
2065                  * otherwise we may lose data while umount */
2066                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2067
2068         truncate_inode_pages_final(&inode->i_data);
2069
2070         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2071                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2072                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2073
2074 #ifdef HAVE_SBOPS_EVICT_INODE
2075         ll_clear_inode(inode);
2076 #endif
2077         clear_inode(inode);
2078
2079         EXIT;
2080 }
2081
2082 int ll_iocontrol(struct inode *inode, struct file *file,
2083                  unsigned int cmd, unsigned long arg)
2084 {
2085         struct ll_sb_info *sbi = ll_i2sbi(inode);
2086         struct ptlrpc_request *req = NULL;
2087         int rc, flags = 0;
2088         ENTRY;
2089
2090         switch(cmd) {
2091         case FSFILT_IOC_GETFLAGS: {
2092                 struct mdt_body *body;
2093                 struct md_op_data *op_data;
2094
2095                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2096                                              0, 0, LUSTRE_OPC_ANY,
2097                                              NULL);
2098                 if (IS_ERR(op_data))
2099                         RETURN(PTR_ERR(op_data));
2100
2101                 op_data->op_valid = OBD_MD_FLFLAGS;
2102                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2103                 ll_finish_md_op_data(op_data);
2104                 if (rc) {
2105                         CERROR("%s: failure inode "DFID": rc = %d\n",
2106                                sbi->ll_md_exp->exp_obd->obd_name,
2107                                PFID(ll_inode2fid(inode)), rc);
2108                         RETURN(-abs(rc));
2109                 }
2110
2111                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2112
2113                 flags = body->mbo_flags;
2114
2115                 ptlrpc_req_finished(req);
2116
2117                 RETURN(put_user(flags, (int __user *)arg));
2118         }
2119         case FSFILT_IOC_SETFLAGS: {
2120                 struct iattr *attr;
2121                 struct md_op_data *op_data;
2122                 struct cl_object *obj;
2123
2124                 if (get_user(flags, (int __user *)arg))
2125                         RETURN(-EFAULT);
2126
2127                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2128                                              LUSTRE_OPC_ANY, NULL);
2129                 if (IS_ERR(op_data))
2130                         RETURN(PTR_ERR(op_data));
2131
2132                 op_data->op_attr_flags = flags;
2133                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2134                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2135                 ll_finish_md_op_data(op_data);
2136                 ptlrpc_req_finished(req);
2137                 if (rc)
2138                         RETURN(rc);
2139
2140                 inode->i_flags = ll_ext_to_inode_flags(flags);
2141
2142                 obj = ll_i2info(inode)->lli_clob;
2143                 if (obj == NULL)
2144                         RETURN(0);
2145
2146                 OBD_ALLOC_PTR(attr);
2147                 if (attr == NULL)
2148                         RETURN(-ENOMEM);
2149
2150                 attr->ia_valid = ATTR_ATTR_FLAG;
2151                 rc = cl_setattr_ost(obj, attr, flags);
2152
2153                 OBD_FREE_PTR(attr);
2154                 RETURN(rc);
2155         }
2156         default:
2157                 RETURN(-ENOSYS);
2158         }
2159
2160         RETURN(0);
2161 }
2162
2163 int ll_flush_ctx(struct inode *inode)
2164 {
2165         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2166
2167         CDEBUG(D_SEC, "flush context for user %d\n",
2168                from_kuid(&init_user_ns, current_uid()));
2169
2170         obd_set_info_async(NULL, sbi->ll_md_exp,
2171                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2172                            0, NULL, NULL);
2173         obd_set_info_async(NULL, sbi->ll_dt_exp,
2174                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2175                            0, NULL, NULL);
2176         return 0;
2177 }
2178
2179 /* umount -f client means force down, don't save state */
2180 void ll_umount_begin(struct super_block *sb)
2181 {
2182         struct ll_sb_info *sbi = ll_s2sbi(sb);
2183         struct obd_device *obd;
2184         struct obd_ioctl_data *ioc_data;
2185         struct l_wait_info lwi;
2186         wait_queue_head_t waitq;
2187         ENTRY;
2188
2189         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2190                sb->s_count, atomic_read(&sb->s_active));
2191
2192         obd = class_exp2obd(sbi->ll_md_exp);
2193         if (obd == NULL) {
2194                 CERROR("Invalid MDC connection handle %#llx\n",
2195                        sbi->ll_md_exp->exp_handle.h_cookie);
2196                 EXIT;
2197                 return;
2198         }
2199         obd->obd_force = 1;
2200
2201         obd = class_exp2obd(sbi->ll_dt_exp);
2202         if (obd == NULL) {
2203                 CERROR("Invalid LOV connection handle %#llx\n",
2204                        sbi->ll_dt_exp->exp_handle.h_cookie);
2205                 EXIT;
2206                 return;
2207         }
2208         obd->obd_force = 1;
2209
2210         OBD_ALLOC_PTR(ioc_data);
2211         if (ioc_data) {
2212                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2213                               sizeof *ioc_data, ioc_data, NULL);
2214
2215                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2216                               sizeof *ioc_data, ioc_data, NULL);
2217
2218                 OBD_FREE_PTR(ioc_data);
2219         }
2220
2221         /* Really, we'd like to wait until there are no requests outstanding,
2222          * and then continue.  For now, we just periodically checking for vfs
2223          * to decrement mnt_cnt and hope to finish it within 10sec.
2224          */
2225         init_waitqueue_head(&waitq);
2226         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2227                                    cfs_time_seconds(1), NULL, NULL);
2228         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2229
2230         EXIT;
2231 }
2232
2233 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2234 {
2235         struct ll_sb_info *sbi = ll_s2sbi(sb);
2236         char *profilenm = get_profile_name(sb);
2237         int err;
2238         __u32 read_only;
2239
2240         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2241                 read_only = *flags & MS_RDONLY;
2242                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2243                                          sizeof(KEY_READ_ONLY),
2244                                          KEY_READ_ONLY, sizeof(read_only),
2245                                          &read_only, NULL);
2246                 if (err) {
2247                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2248                                       profilenm, read_only ?
2249                                       "read-only" : "read-write", err);
2250                         return err;
2251                 }
2252
2253                 if (read_only)
2254                         sb->s_flags |= MS_RDONLY;
2255                 else
2256                         sb->s_flags &= ~MS_RDONLY;
2257
2258                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2259                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2260                                       read_only ?  "read-only" : "read-write");
2261         }
2262         return 0;
2263 }
2264
2265 /**
2266  * Cleanup the open handle that is cached on MDT-side.
2267  *
2268  * For open case, the client side open handling thread may hit error
2269  * after the MDT grant the open. Under such case, the client should
2270  * send close RPC to the MDT as cleanup; otherwise, the open handle
2271  * on the MDT will be leaked there until the client umount or evicted.
2272  *
2273  * In further, if someone unlinked the file, because the open handle
2274  * holds the reference on such file/object, then it will block the
2275  * subsequent threads that want to locate such object via FID.
2276  *
2277  * \param[in] sb        super block for this file-system
2278  * \param[in] open_req  pointer to the original open request
2279  */
2280 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2281 {
2282         struct mdt_body                 *body;
2283         struct md_op_data               *op_data;
2284         struct ptlrpc_request           *close_req = NULL;
2285         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2286         ENTRY;
2287
2288         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2289         OBD_ALLOC_PTR(op_data);
2290         if (op_data == NULL) {
2291                 CWARN("%s: cannot allocate op_data to release open handle for "
2292                       DFID"\n",
2293                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2294
2295                 RETURN_EXIT;
2296         }
2297
2298         op_data->op_fid1 = body->mbo_fid1;
2299         op_data->op_handle = body->mbo_handle;
2300         op_data->op_mod_time = ktime_get_real_seconds();
2301         md_close(exp, op_data, NULL, &close_req);
2302         ptlrpc_req_finished(close_req);
2303         ll_finish_md_op_data(op_data);
2304
2305         EXIT;
2306 }
2307
2308 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2309                   struct super_block *sb, struct lookup_intent *it)
2310 {
2311         struct ll_sb_info *sbi = NULL;
2312         struct lustre_md md = { NULL };
2313         int rc;
2314         ENTRY;
2315
2316         LASSERT(*inode || sb);
2317         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2318         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2319                               sbi->ll_md_exp, &md);
2320         if (rc != 0)
2321                 GOTO(cleanup, rc);
2322
2323         if (*inode) {
2324                 rc = ll_update_inode(*inode, &md);
2325                 if (rc != 0)
2326                         GOTO(out, rc);
2327         } else {
2328                 LASSERT(sb != NULL);
2329
2330                 /*
2331                  * At this point server returns to client's same fid as client
2332                  * generated for creating. So using ->fid1 is okay here.
2333                  */
2334                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2335                         CERROR("%s: Fid is insane "DFID"\n",
2336                                 ll_get_fsname(sb, NULL, 0),
2337                                 PFID(&md.body->mbo_fid1));
2338                         GOTO(out, rc = -EINVAL);
2339                 }
2340
2341                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2342                                              sbi->ll_flags & LL_SBI_32BIT_API),
2343                                  &md);
2344                 if (IS_ERR(*inode)) {
2345 #ifdef CONFIG_FS_POSIX_ACL
2346                         if (md.posix_acl) {
2347                                 posix_acl_release(md.posix_acl);
2348                                 md.posix_acl = NULL;
2349                         }
2350 #endif
2351                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2352                         *inode = NULL;
2353                         CERROR("new_inode -fatal: rc %d\n", rc);
2354                         GOTO(out, rc);
2355                 }
2356         }
2357
2358         /* Handling piggyback layout lock.
2359          * Layout lock can be piggybacked by getattr and open request.
2360          * The lsm can be applied to inode only if it comes with a layout lock
2361          * otherwise correct layout may be overwritten, for example:
2362          * 1. proc1: mdt returns a lsm but not granting layout
2363          * 2. layout was changed by another client
2364          * 3. proc2: refresh layout and layout lock granted
2365          * 4. proc1: to apply a stale layout */
2366         if (it != NULL && it->it_lock_mode != 0) {
2367                 struct lustre_handle lockh;
2368                 struct ldlm_lock *lock;
2369
2370                 lockh.cookie = it->it_lock_handle;
2371                 lock = ldlm_handle2lock(&lockh);
2372                 LASSERT(lock != NULL);
2373                 if (ldlm_has_layout(lock)) {
2374                         struct cl_object_conf conf;
2375
2376                         memset(&conf, 0, sizeof(conf));
2377                         conf.coc_opc = OBJECT_CONF_SET;
2378                         conf.coc_inode = *inode;
2379                         conf.coc_lock = lock;
2380                         conf.u.coc_layout = md.layout;
2381                         (void)ll_layout_conf(*inode, &conf);
2382                 }
2383                 LDLM_LOCK_PUT(lock);
2384         }
2385
2386         GOTO(out, rc = 0);
2387
2388 out:
2389         md_free_lustre_md(sbi->ll_md_exp, &md);
2390
2391 cleanup:
2392         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2393                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2394
2395         return rc;
2396 }
2397
2398 int ll_obd_statfs(struct inode *inode, void __user *arg)
2399 {
2400         struct ll_sb_info *sbi = NULL;
2401         struct obd_export *exp;
2402         char *buf = NULL;
2403         struct obd_ioctl_data *data = NULL;
2404         __u32 type;
2405         int len = 0, rc;
2406
2407         if (!inode || !(sbi = ll_i2sbi(inode)))
2408                 GOTO(out_statfs, rc = -EINVAL);
2409
2410         rc = obd_ioctl_getdata(&buf, &len, arg);
2411         if (rc)
2412                 GOTO(out_statfs, rc);
2413
2414         data = (void*)buf;
2415         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2416             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2417                 GOTO(out_statfs, rc = -EINVAL);
2418
2419         if (data->ioc_inllen1 != sizeof(__u32) ||
2420             data->ioc_inllen2 != sizeof(__u32) ||
2421             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2422             data->ioc_plen2 != sizeof(struct obd_uuid))
2423                 GOTO(out_statfs, rc = -EINVAL);
2424
2425         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2426         if (type & LL_STATFS_LMV)
2427                 exp = sbi->ll_md_exp;
2428         else if (type & LL_STATFS_LOV)
2429                 exp = sbi->ll_dt_exp;
2430         else
2431                 GOTO(out_statfs, rc = -ENODEV);
2432
2433         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2434         if (rc)
2435                 GOTO(out_statfs, rc);
2436 out_statfs:
2437         OBD_FREE_LARGE(buf, len);
2438         return rc;
2439 }
2440
2441 int ll_process_config(struct lustre_cfg *lcfg)
2442 {
2443         struct super_block *sb;
2444         unsigned long x;
2445         int rc = 0;
2446         char *ptr;
2447
2448         /* The instance name contains the sb: lustre-client-aacfe000 */
2449         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2450         if (!ptr || !*(++ptr))
2451                 return -EINVAL;
2452         if (sscanf(ptr, "%lx", &x) != 1)
2453                 return -EINVAL;
2454         sb = (struct super_block *)x;
2455         /* This better be a real Lustre superblock! */
2456         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2457
2458         /* Note we have not called client_common_fill_super yet, so
2459            proc fns must be able to handle that! */
2460         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2461                                       lcfg, sb);
2462         if (rc > 0)
2463                 rc = 0;
2464         return rc;
2465 }
2466
2467 /* this function prepares md_op_data hint for passing it down to MD stack. */
2468 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2469                                       struct inode *i1, struct inode *i2,
2470                                       const char *name, size_t namelen,
2471                                       __u32 mode, __u32 opc, void *data)
2472 {
2473         LASSERT(i1 != NULL);
2474
2475         if (name == NULL) {
2476                 /* Do not reuse namelen for something else. */
2477                 if (namelen != 0)
2478                         return ERR_PTR(-EINVAL);
2479         } else {
2480                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2481                         return ERR_PTR(-ENAMETOOLONG);
2482
2483                 if (!lu_name_is_valid_2(name, namelen))
2484                         return ERR_PTR(-EINVAL);
2485         }
2486
2487         if (op_data == NULL)
2488                 OBD_ALLOC_PTR(op_data);
2489
2490         if (op_data == NULL)
2491                 return ERR_PTR(-ENOMEM);
2492
2493         ll_i2gids(op_data->op_suppgids, i1, i2);
2494         op_data->op_fid1 = *ll_inode2fid(i1);
2495         op_data->op_default_stripe_offset = -1;
2496         if (S_ISDIR(i1->i_mode)) {
2497                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2498                 if (opc == LUSTRE_OPC_MKDIR)
2499                         op_data->op_default_stripe_offset =
2500                                    ll_i2info(i1)->lli_def_stripe_offset;
2501         }
2502
2503         if (i2) {
2504                 op_data->op_fid2 = *ll_inode2fid(i2);
2505                 if (S_ISDIR(i2->i_mode))
2506                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2507         } else {
2508                 fid_zero(&op_data->op_fid2);
2509         }
2510
2511         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2512                 op_data->op_cli_flags |= CLI_HASH64;
2513
2514         if (ll_need_32bit_api(ll_i2sbi(i1)))
2515                 op_data->op_cli_flags |= CLI_API32;
2516
2517         op_data->op_name = name;
2518         op_data->op_namelen = namelen;
2519         op_data->op_mode = mode;
2520         op_data->op_mod_time = cfs_time_current_sec();
2521         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2522         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2523         op_data->op_cap = cfs_curproc_cap_pack();
2524         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2525              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2526                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2527         } else {
2528                 op_data->op_mds = 0;
2529         }
2530         op_data->op_data = data;
2531
2532         return op_data;
2533 }
2534
2535 void ll_finish_md_op_data(struct md_op_data *op_data)
2536 {
2537         security_release_secctx(op_data->op_file_secctx,
2538                                 op_data->op_file_secctx_size);
2539         OBD_FREE_PTR(op_data);
2540 }
2541
2542 #ifdef HAVE_SUPEROPS_USE_DENTRY
2543 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2544 #else
2545 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2546 #endif
2547 {
2548         struct ll_sb_info *sbi;
2549
2550 #ifdef HAVE_SUPEROPS_USE_DENTRY
2551         LASSERT((seq != NULL) && (dentry != NULL));
2552         sbi = ll_s2sbi(dentry->d_sb);
2553 #else
2554         LASSERT((seq != NULL) && (vfs != NULL));
2555         sbi = ll_s2sbi(vfs->mnt_sb);
2556 #endif
2557
2558         if (sbi->ll_flags & LL_SBI_NOLCK)
2559                 seq_puts(seq, ",nolock");
2560
2561         if (sbi->ll_flags & LL_SBI_FLOCK)
2562                 seq_puts(seq, ",flock");
2563
2564         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2565                 seq_puts(seq, ",localflock");
2566
2567         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2568                 seq_puts(seq, ",user_xattr");
2569
2570         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2571                 seq_puts(seq, ",lazystatfs");
2572
2573         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2574                 seq_puts(seq, ",user_fid2path");
2575
2576         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2577                 seq_puts(seq, ",always_ping");
2578
2579         RETURN(0);
2580 }
2581
2582 /**
2583  * Get obd name by cmd, and copy out to user space
2584  */
2585 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2586 {
2587         struct ll_sb_info *sbi = ll_i2sbi(inode);
2588         struct obd_device *obd;
2589         ENTRY;
2590
2591         if (cmd == OBD_IOC_GETDTNAME)
2592                 obd = class_exp2obd(sbi->ll_dt_exp);
2593         else if (cmd == OBD_IOC_GETMDNAME)
2594                 obd = class_exp2obd(sbi->ll_md_exp);
2595         else
2596                 RETURN(-EINVAL);
2597
2598         if (!obd)
2599                 RETURN(-ENOENT);
2600
2601         if (copy_to_user((void __user *)arg, obd->obd_name,
2602                          strlen(obd->obd_name) + 1))
2603                 RETURN(-EFAULT);
2604
2605         RETURN(0);
2606 }
2607
2608 /**
2609  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2610  * fsname will be returned in this buffer; otherwise, a static buffer will be
2611  * used to store the fsname and returned to caller.
2612  */
2613 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2614 {
2615         static char fsname_static[MTI_NAME_MAXLEN];
2616         struct lustre_sb_info *lsi = s2lsi(sb);
2617         char *ptr;
2618         int len;
2619
2620         if (buf == NULL) {
2621                 /* this means the caller wants to use static buffer
2622                  * and it doesn't care about race. Usually this is
2623                  * in error reporting path */
2624                 buf = fsname_static;
2625                 buflen = sizeof(fsname_static);
2626         }
2627
2628         len = strlen(lsi->lsi_lmd->lmd_profile);
2629         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2630         if (ptr && (strcmp(ptr, "-client") == 0))
2631                 len -= 7;
2632
2633         if (unlikely(len >= buflen))
2634                 len = buflen - 1;
2635         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2636         buf[len] = '\0';
2637
2638         return buf;
2639 }
2640
2641 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2642 {
2643         char *path = NULL;
2644
2645         struct path p;
2646
2647         p.dentry = dentry;
2648         p.mnt = current->fs->root.mnt;
2649         path_get(&p);
2650         path = d_path(&p, buf, bufsize);
2651         path_put(&p);
2652         return path;
2653 }
2654
2655 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2656 {
2657         char *buf, *path = NULL;
2658         struct dentry *dentry = NULL;
2659         struct inode *inode = page->mapping->host;
2660
2661         /* this can be called inside spin lock so use GFP_ATOMIC. */
2662         buf = (char *)__get_free_page(GFP_ATOMIC);
2663         if (buf != NULL) {
2664                 dentry = d_find_alias(page->mapping->host);
2665                 if (dentry != NULL)
2666                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2667         }
2668
2669         CDEBUG(D_WARNING,
2670                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2671                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2672                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2673                PFID(ll_inode2fid(inode)),
2674                (path && !IS_ERR(path)) ? path : "", ioret);
2675
2676         if (dentry != NULL)
2677                 dput(dentry);
2678
2679         if (buf != NULL)
2680                 free_page((unsigned long)buf);
2681 }
2682
2683 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2684                         struct lov_user_md **kbuf)
2685 {
2686         struct lov_user_md      lum;
2687         ssize_t                 lum_size;
2688         ENTRY;
2689
2690         if (copy_from_user(&lum, md, sizeof(lum)))
2691                 RETURN(-EFAULT);
2692
2693         lum_size = ll_lov_user_md_size(&lum);
2694         if (lum_size < 0)
2695                 RETURN(lum_size);
2696
2697         OBD_ALLOC(*kbuf, lum_size);
2698         if (*kbuf == NULL)
2699                 RETURN(-ENOMEM);
2700
2701         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2702                 OBD_FREE(*kbuf, lum_size);
2703                 RETURN(-EFAULT);
2704         }
2705
2706         RETURN(lum_size);
2707 }
2708
2709 /*
2710  * Compute llite root squash state after a change of root squash
2711  * configuration setting or add/remove of a lnet nid
2712  */
2713 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2714 {
2715         struct root_squash_info *squash = &sbi->ll_squash;
2716         int i;
2717         bool matched;
2718         struct lnet_process_id id;
2719
2720         /* Update norootsquash flag */
2721         down_write(&squash->rsi_sem);
2722         if (list_empty(&squash->rsi_nosquash_nids))
2723                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2724         else {
2725                 /* Do not apply root squash as soon as one of our NIDs is
2726                  * in the nosquash_nids list */
2727                 matched = false;
2728                 i = 0;
2729                 while (LNetGetId(i++, &id) != -ENOENT) {
2730                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2731                                 continue;
2732                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2733                                 matched = true;
2734                                 break;
2735                         }
2736                 }
2737                 if (matched)
2738                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2739                 else
2740                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2741         }
2742         up_write(&squash->rsi_sem);
2743 }
2744
2745 /**
2746  * Parse linkea content to extract information about a given hardlink
2747  *
2748  * \param[in]   ldata      - Initialized linkea data
2749  * \param[in]   linkno     - Link identifier
2750  * \param[out]  parent_fid - The entry's parent FID
2751  * \param[out]  ln         - Entry name destination buffer
2752  *
2753  * \retval 0 on success
2754  * \retval Appropriate negative error code on failure
2755  */
2756 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2757                             struct lu_fid *parent_fid, struct lu_name *ln)
2758 {
2759         unsigned int    idx;
2760         int             rc;
2761         ENTRY;
2762
2763         rc = linkea_init_with_rec(ldata);
2764         if (rc < 0)
2765                 RETURN(rc);
2766
2767         if (linkno >= ldata->ld_leh->leh_reccount)
2768                 /* beyond last link */
2769                 RETURN(-ENODATA);
2770
2771         linkea_first_entry(ldata);
2772         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2773                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2774                                     parent_fid);
2775                 if (idx == linkno)
2776                         break;
2777
2778                 linkea_next_entry(ldata);
2779         }
2780
2781         if (idx < linkno)
2782                 RETURN(-ENODATA);
2783
2784         RETURN(0);
2785 }
2786
2787 /**
2788  * Get parent FID and name of an identified link. Operation is performed for
2789  * a given link number, letting the caller iterate over linkno to list one or
2790  * all links of an entry.
2791  *
2792  * \param[in]     file - File descriptor against which to perform the operation
2793  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2794  *                       on and the available size. It is eventually filled with
2795  *                       the requested information or left untouched on error
2796  *
2797  * \retval - 0 on success
2798  * \retval - Appropriate negative error code on failure
2799  */
2800 int ll_getparent(struct file *file, struct getparent __user *arg)
2801 {
2802         struct inode            *inode = file_inode(file);
2803         struct linkea_data      *ldata;
2804         struct lu_buf            buf = LU_BUF_NULL;
2805         struct lu_name           ln;
2806         struct lu_fid            parent_fid;
2807         __u32                    linkno;
2808         __u32                    name_size;
2809         int                      rc;
2810
2811         ENTRY;
2812
2813         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2814             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2815                 RETURN(-EPERM);
2816
2817         if (get_user(name_size, &arg->gp_name_size))
2818                 RETURN(-EFAULT);
2819
2820         if (get_user(linkno, &arg->gp_linkno))
2821                 RETURN(-EFAULT);
2822
2823         if (name_size > PATH_MAX)
2824                 RETURN(-EINVAL);
2825
2826         OBD_ALLOC(ldata, sizeof(*ldata));
2827         if (ldata == NULL)
2828                 RETURN(-ENOMEM);
2829
2830         rc = linkea_data_new(ldata, &buf);
2831         if (rc < 0)
2832                 GOTO(ldata_free, rc);
2833
2834 #ifdef HAVE_XATTR_HANDLER_FLAGS
2835         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2836                            buf.lb_len, OBD_MD_FLXATTR);
2837 #else
2838         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2839                          buf.lb_len);
2840 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2841         if (rc < 0)
2842                 GOTO(lb_free, rc);
2843
2844         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2845         if (rc < 0)
2846                 GOTO(lb_free, rc);
2847
2848         if (ln.ln_namelen >= name_size)
2849                 GOTO(lb_free, rc = -EOVERFLOW);
2850
2851         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2852                 GOTO(lb_free, rc = -EFAULT);
2853
2854         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2855                 GOTO(lb_free, rc = -EFAULT);
2856
2857         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2858                 GOTO(lb_free, rc = -EFAULT);
2859
2860 lb_free:
2861         lu_buf_free(&buf);
2862 ldata_free:
2863         OBD_FREE(ldata, sizeof(*ldata));
2864
2865         RETURN(rc);
2866 }