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