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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/time.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49 #include <linux/user_namespace.h>
  50 #ifdef HAVE_UIDGID_HEADER
  51 # include <linux/uidgid.h>
  52 #endif
  53
  54 #include <lustre_ioctl.h>
  55 #include <lustre_ha.h>
  56 #include <lustre_dlm.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_disk.h>
  59 #include <lustre_param.h>
  60 #include <lustre_log.h>
  61 #include <cl_object.h>
  62 #include <obd_cksum.h>
  63 #include "llite_internal.h"
  64
  65 struct kmem_cache *ll_file_data_slab;
  66
  67 #ifndef log2
  68 #define log2(n) ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         unsigned long lru_page_max;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         mutex_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         lru_page_max = pages / 2;
  94
  95         /* initialize ll_cache data */
  96         sbi->ll_cache = cl_cache_init(lru_page_max);
  97         if (sbi->ll_cache == NULL) {
  98                 OBD_FREE(sbi, sizeof(*sbi));
  99                 RETURN(NULL);
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         sbi->ll_flags |= LL_SBI_VERBOSE;
 115 #ifdef ENABLE_CHECKSUM
 116         sbi->ll_flags |= LL_SBI_CHECKSUM;
 117 #endif
 118
 119 #ifdef HAVE_LRU_RESIZE_SUPPORT
 120         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 121 #endif
 122
 123         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 124                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 125                                pp_r_hist.oh_lock);
 126                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 127                                pp_w_hist.oh_lock);
 128         }
 129
 130         /* metadata statahead is enabled by default */
 131         sbi->ll_sa_max = LL_SA_RPC_DEF;
 132         atomic_set(&sbi->ll_sa_total, 0);
 133         atomic_set(&sbi->ll_sa_wrong, 0);
 134         atomic_set(&sbi->ll_sa_running, 0);
 135         atomic_set(&sbi->ll_agl_total, 0);
 136         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 137
 138         /* root squash */
 139         sbi->ll_squash.rsi_uid = 0;
 140         sbi->ll_squash.rsi_gid = 0;
 141         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 142         init_rwsem(&sbi->ll_squash.rsi_sem);
 143
 144         RETURN(sbi);
 145 }
 146
 147 static void ll_free_sbi(struct super_block *sb)
 148 {
 149         struct ll_sb_info *sbi = ll_s2sbi(sb);
 150         ENTRY;
 151
 152         if (sbi != NULL) {
 153                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 154                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 155                 if (sbi->ll_cache != NULL) {
 156                         cl_cache_decref(sbi->ll_cache);
 157                         sbi->ll_cache = NULL;
 158                 }
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = NULL;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178         ENTRY;
 179
 180         obd = class_name2obd(md);
 181         if (!obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* indicate the features supported by this client */
 197         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 198                                   OBD_CONNECT_ATTRFID  |
 199                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 200                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 201                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 202                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 203                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 204                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 205                                   OBD_CONNECT_EINPROGRESS |
 206                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 207                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 208                                   OBD_CONNECT_MAX_EASIZE |
 209                                   OBD_CONNECT_FLOCK_DEAD |
 210                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 211                                   OBD_CONNECT_OPEN_BY_FID |
 212                                   OBD_CONNECT_DIR_STRIPE |
 213                                   OBD_CONNECT_BULK_MBITS |
 214                                   OBD_CONNECT_SUBTREE;
 215
 216 #ifdef HAVE_LRU_RESIZE_SUPPORT
 217         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 218                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 219 #endif
 220 #ifdef CONFIG_FS_POSIX_ACL
 221         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 222 #endif
 223
 224         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 225                 /* flag mdc connection as lightweight, only used for test
 226                  * purpose, use with care */
 227                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 228
 229         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 230         data->ocd_version = LUSTRE_VERSION_CODE;
 231
 232         if (sb->s_flags & MS_RDONLY)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 234         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 235                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 236
 237         if (sbi->ll_flags & LL_SBI_FLOCK)
 238                 sbi->ll_fop = &ll_file_operations_flock;
 239         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 240                 sbi->ll_fop = &ll_file_operations;
 241         else
 242                 sbi->ll_fop = &ll_file_operations_noflock;
 243
 244         /* real client */
 245         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 246         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 247                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 248
 249         /* always ping even if server suppress_pings */
 250         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 251                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 252
 253         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 254
 255         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 256         if (err == -EBUSY) {
 257                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 258                                    "recovery, of which this client is not a "
 259                                    "part. Please wait for recovery to complete,"
 260                                    " abort, or time out.\n", md);
 261                 GOTO(out, err);
 262         } else if (err) {
 263                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 264                 GOTO(out, err);
 265         }
 266
 267         sbi->ll_md_exp->exp_connect_data = *data;
 268
 269         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 270                            LUSTRE_SEQ_METADATA);
 271         if (err) {
 272                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 273                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 274                 GOTO(out_md, err);
 275         }
 276
 277         /* For mount, we only need fs info from MDT0, and also in DNE, it
 278          * can make sure the client can be mounted as long as MDT0 is
 279          * avaible */
 280         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 281                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 282                         OBD_STATFS_FOR_MDT0);
 283         if (err)
 284                 GOTO(out_md_fid, err);
 285
 286         /* This needs to be after statfs to ensure connect has finished.
 287          * Note that "data" does NOT contain the valid connect reply.
 288          * If connecting to a 1.8 server there will be no LMV device, so
 289          * we can access the MDC export directly and exp_connect_flags will
 290          * be non-zero, but if accessing an upgraded 2.1 server it will
 291          * have the correct flags filled in.
 292          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 293         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 294         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 295             valid != CLIENT_CONNECT_MDT_REQD) {
 296                 char *buf;
 297
 298                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 299                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 300                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 301                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 302                                    "feature(s) needed for correct operation "
 303                                    "of this client (%s). Please upgrade "
 304                                    "server or downgrade client.\n",
 305                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 306                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 307                 GOTO(out_md_fid, err = -EPROTO);
 308         }
 309
 310         size = sizeof(*data);
 311         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 312                            KEY_CONN_DATA,  &size, data);
 313         if (err) {
 314                 CERROR("%s: Get connect data failed: rc = %d\n",
 315                        sbi->ll_md_exp->exp_obd->obd_name, err);
 316                 GOTO(out_md_fid, err);
 317         }
 318
 319         LASSERT(osfs->os_bsize);
 320         sb->s_blocksize = osfs->os_bsize;
 321         sb->s_blocksize_bits = log2(osfs->os_bsize);
 322         sb->s_magic = LL_SUPER_MAGIC;
 323         sb->s_maxbytes = MAX_LFS_FILESIZE;
 324         sbi->ll_namelen = osfs->os_namelen;
 325
 326         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 327             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 328                 LCONSOLE_INFO("Disabling user_xattr feature because "
 329                               "it is not supported on the server\n");
 330                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 331         }
 332
 333         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 334 #ifdef MS_POSIXACL
 335                 sb->s_flags |= MS_POSIXACL;
 336 #endif
 337                 sbi->ll_flags |= LL_SBI_ACL;
 338         } else {
 339                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 340 #ifdef MS_POSIXACL
 341                 sb->s_flags &= ~MS_POSIXACL;
 342 #endif
 343                 sbi->ll_flags &= ~LL_SBI_ACL;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 347                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 348                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 349                         LCONSOLE_INFO("client is set as remote by default.\n");
 350                 }
 351         } else {
 352                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 353                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client claims to be remote, but server "
 355                                       "rejected, forced to be local.\n");
 356                 }
 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_BRW_SIZE)
 363                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 364         else
 365                 sbi->ll_md_brw_pages = 1;
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 368                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 369
 370         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 371                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 372                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 373                                       "unknown maximum xattr size.\n", dt);
 374                 } else {
 375                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 376                         sbi->ll_xattr_cache_enabled = 1;
 377                 }
 378         }
 379
 380         obd = class_name2obd(dt);
 381         if (!obd) {
 382                 CERROR("DT %s: not setup or attached\n", dt);
 383                 GOTO(out_md_fid, err = -ENODEV);
 384         }
 385
 386         /* pass client page size via ocd_grant_blkbits, the server should report
 387          * back its backend blocksize for grant calculation purpose */
 388         data->ocd_grant_blkbits = PAGE_SHIFT;
 389
 390         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 391                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 392                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 393                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 394                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 395                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 396                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 397                                   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
 404         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 405                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 406
 407         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 408                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 409                  * disabled by default, because it can still be enabled on the
 410                  * fly via /proc. As a consequence, we still need to come to an
 411                  * agreement on the supported algorithms at connect time */
 412                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 413
 414                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 415                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 416                 else
 417                         data->ocd_cksum_types = cksum_types_supported_client();
 418         }
 419
 420 #ifdef HAVE_LRU_RESIZE_SUPPORT
 421         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 422 #endif
 423         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 424                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 425
 426         /* always ping even if server suppress_pings */
 427         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 428                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 429
 430         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 431                "ocd_grant: %d\n", data->ocd_connect_flags,
 432                data->ocd_version, data->ocd_grant);
 433
 434         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 435         obd->obd_upcall.onu_upcall = cl_ocd_update;
 436
 437         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 438
 439         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 440                           NULL);
 441         if (err == -EBUSY) {
 442                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 443                                    "recovery, of which this client is not a "
 444                                    "part.  Please wait for recovery to "
 445                                    "complete, abort, or time out.\n", dt);
 446                 GOTO(out_md, err);
 447         } else if (err) {
 448                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 449                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 450                 GOTO(out_md, err);
 451         }
 452
 453         sbi->ll_dt_exp->exp_connect_data = *data;
 454
 455         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 456                            LUSTRE_SEQ_METADATA);
 457         if (err) {
 458                 CERROR("%s: Can't init data layer FID infrastructure, "
 459                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 460                 GOTO(out_dt, err);
 461         }
 462
 463         mutex_lock(&sbi->ll_lco.lco_lock);
 464         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 465         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 466         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 467         mutex_unlock(&sbi->ll_lco.lco_lock);
 468
 469         fid_zero(&sbi->ll_root_fid);
 470         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 471                            &sbi->ll_root_fid);
 472         if (err) {
 473                 CERROR("cannot mds_connect: rc = %d\n", err);
 474                 GOTO(out_lock_cn_cb, err);
 475         }
 476         if (!fid_is_sane(&sbi->ll_root_fid)) {
 477                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 478                        sbi->ll_md_exp->exp_obd->obd_name,
 479                        PFID(&sbi->ll_root_fid));
 480                 GOTO(out_lock_cn_cb, err = -EINVAL);
 481         }
 482         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 483
 484         sb->s_op = &lustre_super_operations;
 485 #if THREAD_SIZE >= 8192 /*b=17630*/
 486         sb->s_export_op = &lustre_export_operations;
 487 #endif
 488
 489         /* make root inode
 490          * XXX: move this to after cbd setup? */
 491         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 492         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 493                 valid |= OBD_MD_FLRMTPERM;
 494         else if (sbi->ll_flags & LL_SBI_ACL)
 495                 valid |= OBD_MD_FLACL;
 496
 497         OBD_ALLOC_PTR(op_data);
 498         if (op_data == NULL)
 499                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 500
 501         op_data->op_fid1 = sbi->ll_root_fid;
 502         op_data->op_mode = 0;
 503         op_data->op_valid = valid;
 504
 505         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 506
 507         OBD_FREE_PTR(op_data);
 508         if (err) {
 509                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 510                        sbi->ll_md_exp->exp_obd->obd_name, err);
 511                 GOTO(out_lock_cn_cb, err);
 512         }
 513
 514         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 515                                sbi->ll_md_exp, &lmd);
 516         if (err) {
 517                 CERROR("failed to understand root inode md: rc = %d\n", err);
 518                 ptlrpc_req_finished(request);
 519                 GOTO(out_lock_cn_cb, err);
 520         }
 521
 522         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 523         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 524                                             sbi->ll_flags & LL_SBI_32BIT_API),
 525                        &lmd);
 526         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 527         ptlrpc_req_finished(request);
 528
 529         if (IS_ERR(root)) {
 530 #ifdef CONFIG_FS_POSIX_ACL
 531                 if (lmd.posix_acl) {
 532                         posix_acl_release(lmd.posix_acl);
 533                         lmd.posix_acl = NULL;
 534                 }
 535 #endif
 536                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 537                 root = NULL;
 538                 CERROR("lustre_lite: bad iget4 for root\n");
 539                 GOTO(out_root, err);
 540         }
 541
 542 #ifdef CONFIG_FS_POSIX_ACL
 543         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 544                 rct_init(&sbi->ll_rct);
 545                 et_init(&sbi->ll_et);
 546         }
 547 #endif
 548
 549         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 550         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 551                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 552                                  NULL);
 553         cl_sb_init(sb);
 554
 555         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 556                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 557                                  sbi->ll_cache, NULL);
 558
 559         sb->s_root = d_make_root(root);
 560         if (sb->s_root == NULL) {
 561                 CERROR("%s: can't make root dentry\n",
 562                         ll_get_fsname(sb, NULL, 0));
 563                 GOTO(out_root, err = -ENOMEM);
 564         }
 565 #ifdef HAVE_DCACHE_LOCK
 566         sb->s_root->d_op = &ll_d_ops;
 567 #endif
 568
 569         sbi->ll_sdev_orig = sb->s_dev;
 570
 571         /* We set sb->s_dev equal on all lustre clients in order to support
 572          * NFS export clustering.  NFSD requires that the FSID be the same
 573          * on all clients. */
 574         /* s_dev is also used in lt_compare() to compare two fs, but that is
 575          * only a node-local comparison. */
 576         uuid = obd_get_uuid(sbi->ll_md_exp);
 577         if (uuid != NULL)
 578                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 579
 580         if (data != NULL)
 581                 OBD_FREE_PTR(data);
 582         if (osfs != NULL)
 583                 OBD_FREE_PTR(osfs);
 584         if (proc_lustre_fs_root != NULL) {
 585                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 586                                                   dt, md);
 587                 if (err < 0) {
 588                         CERROR("%s: could not register mount in lprocfs: "
 589                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 590                         err = 0;
 591                 }
 592         }
 593
 594         RETURN(err);
 595 out_root:
 596         if (root)
 597                 iput(root);
 598 out_lock_cn_cb:
 599         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 600 out_dt:
 601         obd_disconnect(sbi->ll_dt_exp);
 602         sbi->ll_dt_exp = NULL;
 603 out_md_fid:
 604         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 605 out_md:
 606         obd_disconnect(sbi->ll_md_exp);
 607         sbi->ll_md_exp = NULL;
 608 out:
 609         if (data != NULL)
 610                 OBD_FREE_PTR(data);
 611         if (osfs != NULL)
 612                 OBD_FREE_PTR(osfs);
 613         return err;
 614 }
 615
 616 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 617 {
 618         int size, rc;
 619
 620         size = sizeof(*lmmsize);
 621         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 622                           KEY_MAX_EASIZE, &size, lmmsize);
 623         if (rc != 0) {
 624                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 625                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 626                 RETURN(rc);
 627         }
 628
 629         size = sizeof(int);
 630         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 631                           KEY_MAX_EASIZE, &size, lmmsize);
 632         if (rc)
 633                 CERROR("Get max mdsize error rc %d\n", rc);
 634
 635         RETURN(rc);
 636 }
 637
 638 /**
 639  * Get the value of the default_easize parameter.
 640  *
 641  * \see client_obd::cl_default_mds_easize
 642  *
 643  * \param[in] sbi       superblock info for this filesystem
 644  * \param[out] lmmsize  pointer to storage location for value
 645  *
 646  * \retval 0            on success
 647  * \retval negative     negated errno on failure
 648  */
 649 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 650 {
 651         int size, rc;
 652
 653         size = sizeof(int);
 654         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 655                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 656         if (rc)
 657                 CERROR("Get default mdsize error rc %d\n", rc);
 658
 659         RETURN(rc);
 660 }
 661
 662 /**
 663  * Set the default_easize parameter to the given value.
 664  *
 665  * \see client_obd::cl_default_mds_easize
 666  *
 667  * \param[in] sbi       superblock info for this filesystem
 668  * \param[in] lmmsize   the size to set
 669  *
 670  * \retval 0            on success
 671  * \retval negative     negated errno on failure
 672  */
 673 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 674 {
 675         int rc;
 676
 677         if (lmmsize < sizeof(struct lov_mds_md) ||
 678             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 679                 return -EINVAL;
 680
 681         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 682                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 683                                 sizeof(int), &lmmsize, NULL);
 684
 685         RETURN(rc);
 686 }
 687
 688 static void ll_dump_inode(struct inode *inode)
 689 {
 690         struct ll_d_hlist_node *tmp;
 691         int dentry_count = 0;
 692
 693         LASSERT(inode != NULL);
 694
 695         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 696                 dentry_count++;
 697
 698         CERROR("%s: inode %p dump: dev=%s fid="DFID
 699                " mode=%o count=%u, %d dentries\n",
 700                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 701                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 702                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 703 }
 704
 705 void lustre_dump_dentry(struct dentry *dentry, int recur)
 706 {
 707         struct list_head *tmp;
 708         int subdirs = 0;
 709
 710         LASSERT(dentry != NULL);
 711
 712         list_for_each(tmp, &dentry->d_subdirs)
 713                 subdirs++;
 714
 715         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 716                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 717                dentry->d_name.len, dentry->d_name.name,
 718                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 719                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 720                dentry->d_flags, dentry->d_fsdata, subdirs);
 721         if (dentry->d_inode != NULL)
 722                 ll_dump_inode(dentry->d_inode);
 723
 724         if (recur == 0)
 725                 return;
 726
 727         list_for_each(tmp, &dentry->d_subdirs) {
 728                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 729                 lustre_dump_dentry(d, recur - 1);
 730         }
 731 }
 732
 733 static void client_common_put_super(struct super_block *sb)
 734 {
 735         struct ll_sb_info *sbi = ll_s2sbi(sb);
 736         ENTRY;
 737
 738 #ifdef CONFIG_FS_POSIX_ACL
 739         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 740                 et_fini(&sbi->ll_et);
 741                 rct_fini(&sbi->ll_rct);
 742         }
 743 #endif
 744
 745         cl_sb_fini(sb);
 746
 747         list_del(&sbi->ll_conn_chain);
 748
 749         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 750         obd_disconnect(sbi->ll_dt_exp);
 751         sbi->ll_dt_exp = NULL;
 752
 753         lprocfs_unregister_mountpoint(sbi);
 754
 755         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 756         obd_disconnect(sbi->ll_md_exp);
 757         sbi->ll_md_exp = NULL;
 758
 759         EXIT;
 760 }
 761
 762 void ll_kill_super(struct super_block *sb)
 763 {
 764         struct ll_sb_info *sbi;
 765         ENTRY;
 766
 767         /* not init sb ?*/
 768         if (!(sb->s_flags & MS_ACTIVE))
 769                 return;
 770
 771         sbi = ll_s2sbi(sb);
 772         /* we need restore s_dev from changed for clustred NFS before put_super
 773          * because new kernels have cached s_dev and change sb->s_dev in
 774          * put_super not affected real removing devices */
 775         if (sbi) {
 776                 sb->s_dev = sbi->ll_sdev_orig;
 777                 sbi->ll_umounting = 1;
 778
 779                 /* wait running statahead threads to quit */
 780                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 781                         set_current_state(TASK_UNINTERRUPTIBLE);
 782                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 783                 }
 784         }
 785
 786         EXIT;
 787 }
 788
 789 static inline int ll_set_opt(const char *opt, char *data, int fl)
 790 {
 791         if (strncmp(opt, data, strlen(opt)) != 0)
 792                 return(0);
 793         else
 794                 return(fl);
 795 }
 796
 797 /* non-client-specific mount options are parsed in lmd_parse */
 798 static int ll_options(char *options, int *flags)
 799 {
 800         int tmp;
 801         char *s1 = options, *s2;
 802         ENTRY;
 803
 804         if (!options)
 805                 RETURN(0);
 806
 807         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 808
 809         while (*s1) {
 810                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 811                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 817                 if (tmp) {
 818                         *flags |= tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 827                 if (tmp) {
 828                         *flags &= ~tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 832                 if (tmp) {
 833                         *flags |= tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 837                 if (tmp) {
 838                         *flags &= ~tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("context", s1, 1);
 842                 if (tmp)
 843                         goto next;
 844                 tmp = ll_set_opt("fscontext", s1, 1);
 845                 if (tmp)
 846                         goto next;
 847                 tmp = ll_set_opt("defcontext", s1, 1);
 848                 if (tmp)
 849                         goto next;
 850                 tmp = ll_set_opt("rootcontext", s1, 1);
 851                 if (tmp)
 852                         goto next;
 853                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 854                 if (tmp) {
 855                         *flags |= tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 859                 if (tmp) {
 860                         *flags |= tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 864                 if (tmp) {
 865                         *flags &= ~tmp;
 866                         goto next;
 867                 }
 868
 869                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 875                 if (tmp) {
 876                         *flags &= ~tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 895                 if (tmp) {
 896                         *flags &= ~tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 900                 if (tmp) {
 901                         *flags |= tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 905                 if (tmp) {
 906                         *flags |= tmp;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 910                 if (tmp) {
 911                         *flags &= ~tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 915                 if (tmp) {
 916                         *flags |= tmp;
 917                         goto next;
 918                 }
 919                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 920                                    s1);
 921                 RETURN(-EINVAL);
 922
 923 next:
 924                 /* Find next opt */
 925                 s2 = strchr(s1, ',');
 926                 if (s2 == NULL)
 927                         break;
 928                 s1 = s2 + 1;
 929         }
 930         RETURN(0);
 931 }
 932
 933 void ll_lli_init(struct ll_inode_info *lli)
 934 {
 935         lli->lli_inode_magic = LLI_INODE_MAGIC;
 936         lli->lli_flags = 0;
 937         spin_lock_init(&lli->lli_lock);
 938         lli->lli_posix_acl = NULL;
 939         lli->lli_remote_perms = NULL;
 940         mutex_init(&lli->lli_rmtperm_mutex);
 941         /* Do not set lli_fid, it has been initialized already. */
 942         fid_zero(&lli->lli_pfid);
 943         atomic_set(&lli->lli_open_count, 0);
 944         lli->lli_rmtperm_time = 0;
 945         lli->lli_mds_read_och = NULL;
 946         lli->lli_mds_write_och = NULL;
 947         lli->lli_mds_exec_och = NULL;
 948         lli->lli_open_fd_read_count = 0;
 949         lli->lli_open_fd_write_count = 0;
 950         lli->lli_open_fd_exec_count = 0;
 951         mutex_init(&lli->lli_och_mutex);
 952         spin_lock_init(&lli->lli_agl_lock);
 953         spin_lock_init(&lli->lli_layout_lock);
 954         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 955         lli->lli_clob = NULL;
 956
 957         init_rwsem(&lli->lli_xattrs_list_rwsem);
 958         mutex_init(&lli->lli_xattrs_enq_lock);
 959
 960         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 961         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 962                 mutex_init(&lli->lli_readdir_mutex);
 963                 lli->lli_opendir_key = NULL;
 964                 lli->lli_sai = NULL;
 965                 spin_lock_init(&lli->lli_sa_lock);
 966                 lli->lli_opendir_pid = 0;
 967                 lli->lli_sa_enabled = 0;
 968                 lli->lli_def_stripe_offset = -1;
 969         } else {
 970                 mutex_init(&lli->lli_size_mutex);
 971                 lli->lli_symlink_name = NULL;
 972                 init_rwsem(&lli->lli_trunc_sem);
 973                 range_lock_tree_init(&lli->lli_write_tree);
 974                 init_rwsem(&lli->lli_glimpse_sem);
 975                 lli->lli_glimpse_time = 0;
 976                 INIT_LIST_HEAD(&lli->lli_agl_list);
 977                 lli->lli_agl_index = 0;
 978                 lli->lli_async_rc = 0;
 979         }
 980         mutex_init(&lli->lli_layout_mutex);
 981 }
 982
 983 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 984 {
 985         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 986
 987         bdi->name = "lustre";
 988         return bdi_register(bdi, NULL, "lustre-%d",
 989                             atomic_inc_return(&ll_bdi_num));
 990 }
 991
 992 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 993 {
 994         struct lustre_profile *lprof = NULL;
 995         struct lustre_sb_info *lsi = s2lsi(sb);
 996         struct ll_sb_info *sbi;
 997         char  *dt = NULL, *md = NULL;
 998         char  *profilenm = get_profile_name(sb);
 999         struct config_llog_instance *cfg;
1000         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1001         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1002         int    err;
1003         ENTRY;
1004
1005         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1006
1007         OBD_ALLOC_PTR(cfg);
1008         if (cfg == NULL)
1009                 RETURN(-ENOMEM);
1010
1011         try_module_get(THIS_MODULE);
1012
1013         /* client additional sb info */
1014         lsi->lsi_llsbi = sbi = ll_init_sbi();
1015         if (!sbi) {
1016                 module_put(THIS_MODULE);
1017                 OBD_FREE_PTR(cfg);
1018                 RETURN(-ENOMEM);
1019         }
1020
1021         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1022         if (err)
1023                 GOTO(out_free, err);
1024
1025         err = bdi_init(&lsi->lsi_bdi);
1026         if (err)
1027                 GOTO(out_free, err);
1028         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1029 #ifdef HAVE_BDI_CAP_MAP_COPY
1030         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1031 #else
1032         lsi->lsi_bdi.capabilities = 0;
1033 #endif
1034         err = ll_bdi_register(&lsi->lsi_bdi);
1035         if (err)
1036                 GOTO(out_free, err);
1037
1038         sb->s_bdi = &lsi->lsi_bdi;
1039 #ifndef HAVE_DCACHE_LOCK
1040         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1041         sb->s_d_op = &ll_d_ops;
1042 #endif
1043
1044         /* Generate a string unique to this super, in case some joker tries
1045            to mount the same fs at two mount points.
1046            Use the address of the super itself.*/
1047         cfg->cfg_instance = sb;
1048         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1049         cfg->cfg_callback = class_config_llog_handler;
1050         /* set up client obds */
1051         err = lustre_process_log(sb, profilenm, cfg);
1052         if (err < 0)
1053                 GOTO(out_free, err);
1054
1055         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1056         lprof = class_get_profile(profilenm);
1057         if (lprof == NULL) {
1058                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1059                                    " read from the MGS.  Does that filesystem "
1060                                    "exist?\n", profilenm);
1061                 GOTO(out_free, err = -EINVAL);
1062         }
1063         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1064                lprof->lp_md, lprof->lp_dt);
1065
1066         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1067         if (!dt)
1068                 GOTO(out_free, err = -ENOMEM);
1069         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1070
1071         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1072         if (!md)
1073                 GOTO(out_free, err = -ENOMEM);
1074         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1075
1076         /* connections, registrations, sb setup */
1077         err = client_common_fill_super(sb, md, dt, mnt);
1078         if (err < 0)
1079                 GOTO(out_free, err);
1080
1081         sbi->ll_client_common_fill_super_succeeded = 1;
1082
1083 out_free:
1084         if (md)
1085                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1086         if (dt)
1087                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1088         if (lprof != NULL)
1089                 class_put_profile(lprof);
1090         if (err)
1091                 ll_put_super(sb);
1092         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1093                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1094
1095         OBD_FREE_PTR(cfg);
1096         RETURN(err);
1097 } /* ll_fill_super */
1098
1099 void ll_put_super(struct super_block *sb)
1100 {
1101         struct config_llog_instance cfg, params_cfg;
1102         struct obd_device *obd;
1103         struct lustre_sb_info *lsi = s2lsi(sb);
1104         struct ll_sb_info *sbi = ll_s2sbi(sb);
1105         char *profilenm = get_profile_name(sb);
1106         long ccc_count;
1107         int next, force = 1, rc = 0;
1108         ENTRY;
1109
1110         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1111
1112         cfg.cfg_instance = sb;
1113         lustre_end_log(sb, profilenm, &cfg);
1114
1115         params_cfg.cfg_instance = sb;
1116         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1117
1118         if (sbi->ll_md_exp) {
1119                 obd = class_exp2obd(sbi->ll_md_exp);
1120                 if (obd)
1121                         force = obd->obd_force;
1122         }
1123
1124         /* Wait for unstable pages to be committed to stable storage */
1125         if (force == 0) {
1126                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1127                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1128                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1129                         &lwi);
1130         }
1131
1132         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1133         if (force == 0 && rc != -EINTR)
1134                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1135
1136
1137         /* We need to set force before the lov_disconnect in
1138            lustre_common_put_super, since l_d cleans up osc's as well. */
1139         if (force) {
1140                 next = 0;
1141                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1142                                                      &next)) != NULL) {
1143                         obd->obd_force = force;
1144                 }
1145         }
1146
1147         if (sbi->ll_client_common_fill_super_succeeded) {
1148                 /* Only if client_common_fill_super succeeded */
1149                 client_common_put_super(sb);
1150         }
1151
1152         next = 0;
1153         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1154                 class_manual_cleanup(obd);
1155         }
1156
1157         if (sbi->ll_flags & LL_SBI_VERBOSE)
1158                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1159
1160         if (profilenm)
1161                 class_del_profile(profilenm);
1162
1163         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1164                 bdi_destroy(&lsi->lsi_bdi);
1165                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1166         }
1167
1168         ll_free_sbi(sb);
1169         lsi->lsi_llsbi = NULL;
1170
1171         lustre_common_put_super(sb);
1172
1173         cl_env_cache_purge(~0);
1174
1175         module_put(THIS_MODULE);
1176
1177         EXIT;
1178 } /* client_put_super */
1179
1180 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1181 {
1182         struct inode *inode = NULL;
1183
1184         /* NOTE: we depend on atomic igrab() -bzzz */
1185         lock_res_and_lock(lock);
1186         if (lock->l_resource->lr_lvb_inode) {
1187                 struct ll_inode_info * lli;
1188                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1189                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1190                         inode = igrab(lock->l_resource->lr_lvb_inode);
1191                 } else {
1192                         inode = lock->l_resource->lr_lvb_inode;
1193                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1194                                          D_WARNING, lock, "lr_lvb_inode %p is "
1195                                          "bogus: magic %08x",
1196                                          lock->l_resource->lr_lvb_inode,
1197                                          lli->lli_inode_magic);
1198                         inode = NULL;
1199                 }
1200         }
1201         unlock_res_and_lock(lock);
1202         return inode;
1203 }
1204
1205 void ll_dir_clear_lsm_md(struct inode *inode)
1206 {
1207         struct ll_inode_info *lli = ll_i2info(inode);
1208
1209         LASSERT(S_ISDIR(inode->i_mode));
1210
1211         if (lli->lli_lsm_md != NULL) {
1212                 lmv_free_memmd(lli->lli_lsm_md);
1213                 lli->lli_lsm_md = NULL;
1214         }
1215 }
1216
1217 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1218                                       const struct lu_fid *fid,
1219                                       struct lustre_md *md)
1220 {
1221         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1222         struct mdt_body         *body = md->body;
1223         struct inode            *inode;
1224         ino_t                   ino;
1225         ENTRY;
1226
1227         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1228         inode = iget_locked(sb, ino);
1229         if (inode == NULL) {
1230                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1231                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1232                 RETURN(ERR_PTR(-ENOENT));
1233         }
1234
1235         if (inode->i_state & I_NEW) {
1236                 struct ll_inode_info *lli = ll_i2info(inode);
1237                 struct lmv_stripe_md *lsm = md->lmv;
1238
1239                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1240                                 (body->mbo_mode & S_IFMT);
1241                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1242                          PFID(fid));
1243
1244                 LTIME_S(inode->i_mtime) = 0;
1245                 LTIME_S(inode->i_atime) = 0;
1246                 LTIME_S(inode->i_ctime) = 0;
1247                 inode->i_rdev = 0;
1248
1249 #ifdef HAVE_BACKING_DEV_INFO
1250                 /* initializing backing dev info. */
1251                 inode->i_mapping->backing_dev_info =
1252                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1253 #endif
1254                 inode->i_op = &ll_dir_inode_operations;
1255                 inode->i_fop = &ll_dir_operations;
1256                 lli->lli_fid = *fid;
1257                 ll_lli_init(lli);
1258
1259                 LASSERT(lsm != NULL);
1260                 /* master object FID */
1261                 lli->lli_pfid = body->mbo_fid1;
1262                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1263                        lli, PFID(fid), PFID(&lli->lli_pfid));
1264                 unlock_new_inode(inode);
1265         }
1266
1267         RETURN(inode);
1268 }
1269
1270 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1271 {
1272         struct lu_fid *fid;
1273         struct lmv_stripe_md *lsm = md->lmv;
1274         int i;
1275
1276         LASSERT(lsm != NULL);
1277         /* XXX sigh, this lsm_root initialization should be in
1278          * LMV layer, but it needs ll_iget right now, so we
1279          * put this here right now. */
1280         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1281                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1282                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1283                 /* Unfortunately ll_iget will call ll_update_inode,
1284                  * where the initialization of slave inode is slightly
1285                  * different, so it reset lsm_md to NULL to avoid
1286                  * initializing lsm for slave inode. */
1287                 /* For migrating inode, master stripe and master object will
1288                  * be same, so we only need assign this inode */
1289                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1290                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1291                 else
1292                         lsm->lsm_md_oinfo[i].lmo_root =
1293                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1294
1295                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1296                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1297
1298                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1299                         return rc;
1300                 }
1301         }
1302
1303         return 0;
1304 }
1305
1306 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1307                                 const struct lmv_stripe_md *lsm_md2)
1308 {
1309         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1310                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1311                lsm_md1->lsm_md_master_mdt_index ==
1312                                         lsm_md2->lsm_md_master_mdt_index &&
1313                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1314                lsm_md1->lsm_md_layout_version ==
1315                                         lsm_md2->lsm_md_layout_version &&
1316                strcmp(lsm_md1->lsm_md_pool_name,
1317                       lsm_md2->lsm_md_pool_name) == 0;
1318 }
1319
1320 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1321 {
1322         struct ll_inode_info *lli = ll_i2info(inode);
1323         struct lmv_stripe_md *lsm = md->lmv;
1324         int     rc;
1325         ENTRY;
1326
1327         LASSERT(S_ISDIR(inode->i_mode));
1328         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1329                PFID(ll_inode2fid(inode)));
1330
1331         /* no striped information from request. */
1332         if (lsm == NULL) {
1333                 if (lli->lli_lsm_md == NULL) {
1334                         RETURN(0);
1335                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1336                                                 LMV_HASH_FLAG_MIGRATION) {
1337                         /* migration is done, the temporay MIGRATE layout has
1338                          * been removed */
1339                         CDEBUG(D_INODE, DFID" finish migration.\n",
1340                                PFID(ll_inode2fid(inode)));
1341                         lmv_free_memmd(lli->lli_lsm_md);
1342                         lli->lli_lsm_md = NULL;
1343                         RETURN(0);
1344                 } else {
1345                         /* The lustre_md from req does not include stripeEA,
1346                          * see ll_md_setattr */
1347                         RETURN(0);
1348                 }
1349         }
1350
1351         /* set the directory layout */
1352         if (lli->lli_lsm_md == NULL) {
1353                 struct cl_attr  *attr;
1354
1355                 rc = ll_init_lsm_md(inode, md);
1356                 if (rc != 0)
1357                         RETURN(rc);
1358
1359                 /* set md->lmv to NULL, so the following free lustre_md
1360                  * will not free this lsm */
1361                 md->lmv = NULL;
1362                 lli->lli_lsm_md = lsm;
1363
1364                 OBD_ALLOC_PTR(attr);
1365                 if (attr == NULL)
1366                         RETURN(-ENOMEM);
1367
1368                 /* validate the lsm */
1369                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1370                                    ll_md_blocking_ast);
1371                 if (rc != 0) {
1372                         OBD_FREE_PTR(attr);
1373                         RETURN(rc);
1374                 }
1375
1376                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1377                         md->body->mbo_nlink = attr->cat_nlink;
1378                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1379                         md->body->mbo_size = attr->cat_size;
1380                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1381                         md->body->mbo_atime = attr->cat_atime;
1382                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1383                         md->body->mbo_ctime = attr->cat_ctime;
1384                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1385                         md->body->mbo_mtime = attr->cat_mtime;
1386
1387                 OBD_FREE_PTR(attr);
1388
1389                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1390                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1391                 RETURN(0);
1392         }
1393
1394         /* Compare the old and new stripe information */
1395         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1396                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1397                 int                     idx;
1398
1399                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1400                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1401                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1402                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1403                        inode, lsm, old_lsm,
1404                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1405                        lsm->lsm_md_stripe_count,
1406                        old_lsm->lsm_md_stripe_count,
1407                        lsm->lsm_md_master_mdt_index,
1408                        old_lsm->lsm_md_master_mdt_index,
1409                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1410                        lsm->lsm_md_layout_version,
1411                        old_lsm->lsm_md_layout_version,
1412                        lsm->lsm_md_pool_name,
1413                        old_lsm->lsm_md_pool_name);
1414
1415                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1416                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1417                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1418                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1419                 }
1420
1421                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1422                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1423                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1424                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1425                 }
1426
1427                 RETURN(-EIO);
1428         }
1429
1430         RETURN(0);
1431 }
1432
1433 void ll_clear_inode(struct inode *inode)
1434 {
1435         struct ll_inode_info *lli = ll_i2info(inode);
1436         struct ll_sb_info *sbi = ll_i2sbi(inode);
1437         ENTRY;
1438
1439         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1440                PFID(ll_inode2fid(inode)), inode);
1441
1442         if (S_ISDIR(inode->i_mode)) {
1443                 /* these should have been cleared in ll_file_release */
1444                 LASSERT(lli->lli_opendir_key == NULL);
1445                 LASSERT(lli->lli_sai == NULL);
1446                 LASSERT(lli->lli_opendir_pid == 0);
1447         }
1448
1449         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1450
1451         LASSERT(!lli->lli_open_fd_write_count);
1452         LASSERT(!lli->lli_open_fd_read_count);
1453         LASSERT(!lli->lli_open_fd_exec_count);
1454
1455         if (lli->lli_mds_write_och)
1456                 ll_md_real_close(inode, FMODE_WRITE);
1457         if (lli->lli_mds_exec_och)
1458                 ll_md_real_close(inode, FMODE_EXEC);
1459         if (lli->lli_mds_read_och)
1460                 ll_md_real_close(inode, FMODE_READ);
1461
1462         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1463                 OBD_FREE(lli->lli_symlink_name,
1464                          strlen(lli->lli_symlink_name) + 1);
1465                 lli->lli_symlink_name = NULL;
1466         }
1467
1468         ll_xattr_cache_destroy(inode);
1469
1470         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1471                 LASSERT(lli->lli_posix_acl == NULL);
1472                 if (lli->lli_remote_perms) {
1473                         free_rmtperm_hash(lli->lli_remote_perms);
1474                         lli->lli_remote_perms = NULL;
1475                 }
1476         }
1477 #ifdef CONFIG_FS_POSIX_ACL
1478         else if (lli->lli_posix_acl) {
1479                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1480                 LASSERT(lli->lli_remote_perms == NULL);
1481                 posix_acl_release(lli->lli_posix_acl);
1482                 lli->lli_posix_acl = NULL;
1483         }
1484 #endif
1485         lli->lli_inode_magic = LLI_INODE_DEAD;
1486
1487         if (S_ISDIR(inode->i_mode))
1488                 ll_dir_clear_lsm_md(inode);
1489         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1490                 LASSERT(list_empty(&lli->lli_agl_list));
1491
1492         /*
1493          * XXX This has to be done before lsm is freed below, because
1494          * cl_object still uses inode lsm.
1495          */
1496         cl_inode_fini(inode);
1497
1498         EXIT;
1499 }
1500
1501 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1502 {
1503         struct lustre_md md;
1504         struct inode *inode = dentry->d_inode;
1505         struct ll_sb_info *sbi = ll_i2sbi(inode);
1506         struct ptlrpc_request *request = NULL;
1507         int rc, ia_valid;
1508         ENTRY;
1509
1510         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1511                                      LUSTRE_OPC_ANY, NULL);
1512         if (IS_ERR(op_data))
1513                 RETURN(PTR_ERR(op_data));
1514
1515         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1516         if (rc) {
1517                 ptlrpc_req_finished(request);
1518                 if (rc == -ENOENT) {
1519                         clear_nlink(inode);
1520                         /* Unlinked special device node? Or just a race?
1521                          * Pretend we done everything. */
1522                         if (!S_ISREG(inode->i_mode) &&
1523                             !S_ISDIR(inode->i_mode)) {
1524                                 ia_valid = op_data->op_attr.ia_valid;
1525                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1526                                 rc = simple_setattr(dentry, &op_data->op_attr);
1527                                 op_data->op_attr.ia_valid = ia_valid;
1528                         }
1529                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1530                         CERROR("md_setattr fails: rc = %d\n", rc);
1531                 }
1532                 RETURN(rc);
1533         }
1534
1535         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1536                               sbi->ll_md_exp, &md);
1537         if (rc) {
1538                 ptlrpc_req_finished(request);
1539                 RETURN(rc);
1540         }
1541
1542         ia_valid = op_data->op_attr.ia_valid;
1543         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1544          * cache is not cleared yet. */
1545         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1546         if (S_ISREG(inode->i_mode))
1547                 mutex_lock(&inode->i_mutex);
1548         rc = simple_setattr(dentry, &op_data->op_attr);
1549         if (S_ISREG(inode->i_mode))
1550                 mutex_unlock(&inode->i_mutex);
1551         op_data->op_attr.ia_valid = ia_valid;
1552
1553         rc = ll_update_inode(inode, &md);
1554         ptlrpc_req_finished(request);
1555
1556         RETURN(rc);
1557 }
1558
1559 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1560  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1561  * keep these values until such a time that objects are allocated for it.
1562  * We do the MDS operations first, as it is checking permissions for us.
1563  * We don't to the MDS RPC if there is nothing that we want to store there,
1564  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1565  * going to do an RPC anyways.
1566  *
1567  * If we are doing a truncate, we will send the mtime and ctime updates
1568  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1569  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1570  * at the same time.
1571  *
1572  * In case of HSMimport, we only set attr on MDS.
1573  */
1574 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1575 {
1576         struct inode *inode = dentry->d_inode;
1577         struct ll_inode_info *lli = ll_i2info(inode);
1578         struct md_op_data *op_data = NULL;
1579         int rc = 0;
1580         ENTRY;
1581
1582         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1583                "valid %x, hsm_import %d\n",
1584                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1585                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1586                hsm_import);
1587
1588         if (attr->ia_valid & ATTR_SIZE) {
1589                 /* Check new size against VFS/VM file size limit and rlimit */
1590                 rc = inode_newsize_ok(inode, attr->ia_size);
1591                 if (rc)
1592                         RETURN(rc);
1593
1594                 /* The maximum Lustre file size is variable, based on the
1595                  * OST maximum object size and number of stripes.  This
1596                  * needs another check in addition to the VFS check above. */
1597                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1598                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1599                                PFID(&lli->lli_fid), attr->ia_size,
1600                                ll_file_maxbytes(inode));
1601                         RETURN(-EFBIG);
1602                 }
1603
1604                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1605         }
1606
1607         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1608         if (attr->ia_valid & TIMES_SET_FLAGS) {
1609                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1610                     !cfs_capable(CFS_CAP_FOWNER))
1611                         RETURN(-EPERM);
1612         }
1613
1614         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1615         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1616             (attr->ia_valid & ATTR_CTIME)) {
1617                 attr->ia_ctime = CURRENT_TIME;
1618                 attr->ia_valid |= ATTR_CTIME_SET;
1619         }
1620         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1621             (attr->ia_valid & ATTR_ATIME)) {
1622                 attr->ia_atime = CURRENT_TIME;
1623                 attr->ia_valid |= ATTR_ATIME_SET;
1624         }
1625         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1626             (attr->ia_valid & ATTR_MTIME)) {
1627                 attr->ia_mtime = CURRENT_TIME;
1628                 attr->ia_valid |= ATTR_MTIME_SET;
1629         }
1630
1631         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1632                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1633                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1634                        cfs_time_current_sec());
1635
1636         if (S_ISREG(inode->i_mode)) {
1637                 if (attr->ia_valid & ATTR_SIZE)
1638                         inode_dio_write_done(inode);
1639                 mutex_unlock(&inode->i_mutex);
1640         }
1641
1642         /* We always do an MDS RPC, even if we're only changing the size;
1643          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1644
1645         OBD_ALLOC_PTR(op_data);
1646         if (op_data == NULL)
1647                 GOTO(out, rc = -ENOMEM);
1648
1649         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1650                 /* If we are changing file size, file content is
1651                  * modified, flag it. */
1652                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1653                 op_data->op_bias |= MDS_DATA_MODIFIED;
1654                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1655         }
1656
1657         op_data->op_attr = *attr;
1658
1659         rc = ll_md_setattr(dentry, op_data);
1660         if (rc)
1661                 GOTO(out, rc);
1662
1663         if (!S_ISREG(inode->i_mode) || hsm_import)
1664                 GOTO(out, rc = 0);
1665
1666         if (attr->ia_valid & (ATTR_SIZE |
1667                               ATTR_ATIME | ATTR_ATIME_SET |
1668                               ATTR_MTIME | ATTR_MTIME_SET |
1669                               ATTR_CTIME | ATTR_CTIME_SET)) {
1670                 /* For truncate and utimes sending attributes to OSTs, setting
1671                  * mtime/atime to the past will be performed under PW [0:EOF]
1672                  * extent lock (new_size:EOF for truncate).  It may seem
1673                  * excessive to send mtime/atime updates to OSTs when not
1674                  * setting times to past, but it is necessary due to possible
1675                  * time de-synchronization between MDT inode and OST objects */
1676                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1677         }
1678
1679         /* If the file was restored, it needs to set dirty flag.
1680          *
1681          * We've already sent MDS_DATA_MODIFIED flag in
1682          * ll_md_setattr() for truncate. However, the MDT refuses to
1683          * set the HS_DIRTY flag on released files, so we have to set
1684          * it again if the file has been restored. Please check how
1685          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1686          *
1687          * Please notice that if the file is not released, the previous
1688          * MDS_DATA_MODIFIED has taken effect and usually
1689          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1690          * This way we can save an RPC for common open + trunc
1691          * operation. */
1692         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1693                 struct hsm_state_set hss = {
1694                         .hss_valid = HSS_SETMASK,
1695                         .hss_setmask = HS_DIRTY,
1696                 };
1697                 int rc2;
1698
1699                 rc2 = ll_hsm_state_set(inode, &hss);
1700                 /* truncate and write can happen at the same time, so that
1701                  * the file can be set modified even though the file is not
1702                  * restored from released state, and ll_hsm_state_set() is
1703                  * not applicable for the file, and rc2 < 0 is normal in this
1704                  * case. */
1705                 if (rc2 < 0)
1706                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1707                                PFID(ll_inode2fid(inode)), rc2);
1708         }
1709
1710         EXIT;
1711 out:
1712         if (op_data != NULL)
1713                 ll_finish_md_op_data(op_data);
1714
1715         if (S_ISREG(inode->i_mode)) {
1716                 mutex_lock(&inode->i_mutex);
1717                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1718                         inode_dio_wait(inode);
1719         }
1720
1721         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1722                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1723
1724         return rc;
1725 }
1726
1727 int ll_setattr(struct dentry *de, struct iattr *attr)
1728 {
1729         int mode = de->d_inode->i_mode;
1730
1731         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1732                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1733                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1734
1735         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1736                                (ATTR_SIZE|ATTR_MODE)) &&
1737             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1738              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1739               !(attr->ia_mode & S_ISGID))))
1740                 attr->ia_valid |= ATTR_FORCE;
1741
1742         if ((attr->ia_valid & ATTR_MODE) &&
1743             (mode & S_ISUID) &&
1744             !(attr->ia_mode & S_ISUID) &&
1745             !(attr->ia_valid & ATTR_KILL_SUID))
1746                 attr->ia_valid |= ATTR_KILL_SUID;
1747
1748         if ((attr->ia_valid & ATTR_MODE) &&
1749             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1750             !(attr->ia_mode & S_ISGID) &&
1751             !(attr->ia_valid & ATTR_KILL_SGID))
1752                 attr->ia_valid |= ATTR_KILL_SGID;
1753
1754         return ll_setattr_raw(de, attr, false);
1755 }
1756
1757 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1758                        __u64 max_age, __u32 flags)
1759 {
1760         struct ll_sb_info *sbi = ll_s2sbi(sb);
1761         struct obd_statfs obd_osfs;
1762         int rc;
1763         ENTRY;
1764
1765         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1766         if (rc) {
1767                 CERROR("md_statfs fails: rc = %d\n", rc);
1768                 RETURN(rc);
1769         }
1770
1771         osfs->os_type = sb->s_magic;
1772
1773         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1774                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1775
1776         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1777                 flags |= OBD_STATFS_NODELAY;
1778
1779         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1780         if (rc) {
1781                 CERROR("obd_statfs fails: rc = %d\n", rc);
1782                 RETURN(rc);
1783         }
1784
1785         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1786                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1787                obd_osfs.os_files);
1788
1789         osfs->os_bsize = obd_osfs.os_bsize;
1790         osfs->os_blocks = obd_osfs.os_blocks;
1791         osfs->os_bfree = obd_osfs.os_bfree;
1792         osfs->os_bavail = obd_osfs.os_bavail;
1793
1794         /* If we don't have as many objects free on the OST as inodes
1795          * on the MDS, we reduce the total number of inodes to
1796          * compensate, so that the "inodes in use" number is correct.
1797          */
1798         if (obd_osfs.os_ffree < osfs->os_ffree) {
1799                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1800                         obd_osfs.os_ffree;
1801                 osfs->os_ffree = obd_osfs.os_ffree;
1802         }
1803
1804         RETURN(rc);
1805 }
1806 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1807 {
1808         struct super_block *sb = de->d_sb;
1809         struct obd_statfs osfs;
1810         __u64 fsid = huge_encode_dev(sb->s_dev);
1811         int rc;
1812
1813         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1814         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1815
1816         /* Some amount of caching on the client is allowed */
1817         rc = ll_statfs_internal(sb, &osfs,
1818                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1819                                 0);
1820         if (rc)
1821                 return rc;
1822
1823         statfs_unpack(sfs, &osfs);
1824
1825         /* We need to downshift for all 32-bit kernels, because we can't
1826          * tell if the kernel is being called via sys_statfs64() or not.
1827          * Stop before overflowing f_bsize - in which case it is better
1828          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1829         if (sizeof(long) < 8) {
1830                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1831                         sfs->f_bsize <<= 1;
1832
1833                         osfs.os_blocks >>= 1;
1834                         osfs.os_bfree >>= 1;
1835                         osfs.os_bavail >>= 1;
1836                 }
1837         }
1838
1839         sfs->f_blocks = osfs.os_blocks;
1840         sfs->f_bfree = osfs.os_bfree;
1841         sfs->f_bavail = osfs.os_bavail;
1842         sfs->f_fsid.val[0] = (__u32)fsid;
1843         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1844         return 0;
1845 }
1846
1847 void ll_inode_size_lock(struct inode *inode)
1848 {
1849         struct ll_inode_info *lli;
1850
1851         LASSERT(!S_ISDIR(inode->i_mode));
1852
1853         lli = ll_i2info(inode);
1854         mutex_lock(&lli->lli_size_mutex);
1855 }
1856
1857 void ll_inode_size_unlock(struct inode *inode)
1858 {
1859         struct ll_inode_info *lli;
1860
1861         lli = ll_i2info(inode);
1862         mutex_unlock(&lli->lli_size_mutex);
1863 }
1864
1865 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1866 {
1867         struct ll_inode_info *lli = ll_i2info(inode);
1868         struct mdt_body *body = md->body;
1869         struct ll_sb_info *sbi = ll_i2sbi(inode);
1870
1871         if (body->mbo_valid & OBD_MD_FLEASIZE)
1872                 cl_file_inode_init(inode, md);
1873
1874         if (S_ISDIR(inode->i_mode)) {
1875                 int     rc;
1876
1877                 rc = ll_update_lsm_md(inode, md);
1878                 if (rc != 0)
1879                         return rc;
1880         }
1881
1882         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1883                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1884                         ll_update_remote_perm(inode, md->remote_perm);
1885         }
1886 #ifdef CONFIG_FS_POSIX_ACL
1887         else if (body->mbo_valid & OBD_MD_FLACL) {
1888                 spin_lock(&lli->lli_lock);
1889                 if (lli->lli_posix_acl)
1890                         posix_acl_release(lli->lli_posix_acl);
1891                 lli->lli_posix_acl = md->posix_acl;
1892                 spin_unlock(&lli->lli_lock);
1893         }
1894 #endif
1895         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1896                                         sbi->ll_flags & LL_SBI_32BIT_API);
1897         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1898
1899         if (body->mbo_valid & OBD_MD_FLATIME) {
1900                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1901                         LTIME_S(inode->i_atime) = body->mbo_atime;
1902                 lli->lli_atime = body->mbo_atime;
1903         }
1904
1905         if (body->mbo_valid & OBD_MD_FLMTIME) {
1906                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1907                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1908                                "to "LPU64"\n", inode->i_ino,
1909                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1910                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1911                 }
1912                 lli->lli_mtime = body->mbo_mtime;
1913         }
1914
1915         if (body->mbo_valid & OBD_MD_FLCTIME) {
1916                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1917                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1918                 lli->lli_ctime = body->mbo_ctime;
1919         }
1920
1921         if (body->mbo_valid & OBD_MD_FLMODE)
1922                 inode->i_mode = (inode->i_mode & S_IFMT) |
1923                                 (body->mbo_mode & ~S_IFMT);
1924
1925         if (body->mbo_valid & OBD_MD_FLTYPE)
1926                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1927                                 (body->mbo_mode & S_IFMT);
1928
1929         LASSERT(inode->i_mode != 0);
1930         if (S_ISREG(inode->i_mode))
1931                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1932                                        LL_MAX_BLKSIZE_BITS);
1933         else
1934                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1935
1936         if (body->mbo_valid & OBD_MD_FLUID)
1937                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1938         if (body->mbo_valid & OBD_MD_FLGID)
1939                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1940         if (body->mbo_valid & OBD_MD_FLFLAGS)
1941                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1942         if (body->mbo_valid & OBD_MD_FLNLINK)
1943                 set_nlink(inode, body->mbo_nlink);
1944         if (body->mbo_valid & OBD_MD_FLRDEV)
1945                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1946
1947         if (body->mbo_valid & OBD_MD_FLID) {
1948                 /* FID shouldn't be changed! */
1949                 if (fid_is_sane(&lli->lli_fid)) {
1950                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1951                                  "Trying to change FID "DFID
1952                                  " to the "DFID", inode "DFID"(%p)\n",
1953                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1954                                  PFID(ll_inode2fid(inode)), inode);
1955                 } else {
1956                         lli->lli_fid = body->mbo_fid1;
1957                 }
1958         }
1959
1960         LASSERT(fid_seq(&lli->lli_fid) != 0);
1961
1962         if (body->mbo_valid & OBD_MD_FLSIZE) {
1963                 i_size_write(inode, body->mbo_size);
1964
1965                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1966                        PFID(ll_inode2fid(inode)),
1967                        (unsigned long long)body->mbo_size);
1968
1969                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1970                         inode->i_blocks = body->mbo_blocks;
1971         }
1972
1973         if (body->mbo_valid & OBD_MD_TSTATE) {
1974                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1975                  * clear it when done to ensure to start again
1976                  * glimpsing updated attrs
1977                  */
1978                 if (body->mbo_t_state & MS_RESTORE)
1979                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1980                 else
1981                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1982         }
1983
1984         return 0;
1985 }
1986
1987 int ll_read_inode2(struct inode *inode, void *opaque)
1988 {
1989         struct lustre_md *md = opaque;
1990         struct ll_inode_info *lli = ll_i2info(inode);
1991         int     rc;
1992         ENTRY;
1993
1994         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1995                PFID(&lli->lli_fid), inode);
1996
1997         /* Core attributes from the MDS first.  This is a new inode, and
1998          * the VFS doesn't zero times in the core inode so we have to do
1999          * it ourselves.  They will be overwritten by either MDS or OST
2000          * attributes - we just need to make sure they aren't newer. */
2001         LTIME_S(inode->i_mtime) = 0;
2002         LTIME_S(inode->i_atime) = 0;
2003         LTIME_S(inode->i_ctime) = 0;
2004         inode->i_rdev = 0;
2005         rc = ll_update_inode(inode, md);
2006         if (rc != 0)
2007                 RETURN(rc);
2008
2009         /* OIDEBUG(inode); */
2010
2011 #ifdef HAVE_BACKING_DEV_INFO
2012         /* initializing backing dev info. */
2013         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2014 #endif
2015         if (S_ISREG(inode->i_mode)) {
2016                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2017                 inode->i_op = &ll_file_inode_operations;
2018                 inode->i_fop = sbi->ll_fop;
2019                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2020                 EXIT;
2021         } else if (S_ISDIR(inode->i_mode)) {
2022                 inode->i_op = &ll_dir_inode_operations;
2023                 inode->i_fop = &ll_dir_operations;
2024                 EXIT;
2025         } else if (S_ISLNK(inode->i_mode)) {
2026                 inode->i_op = &ll_fast_symlink_inode_operations;
2027                 EXIT;
2028         } else {
2029                 inode->i_op = &ll_special_inode_operations;
2030
2031                 init_special_inode(inode, inode->i_mode,
2032                                    inode->i_rdev);
2033
2034                 EXIT;
2035         }
2036
2037         return 0;
2038 }
2039
2040 void ll_delete_inode(struct inode *inode)
2041 {
2042         struct ll_inode_info *lli = ll_i2info(inode);
2043         ENTRY;
2044
2045         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2046                 /* It is last chance to write out dirty pages,
2047                  * otherwise we may lose data while umount */
2048                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2049
2050         truncate_inode_pages_final(&inode->i_data);
2051
2052         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2053                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2054                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2055
2056 #ifdef HAVE_SBOPS_EVICT_INODE
2057         ll_clear_inode(inode);
2058 #endif
2059         clear_inode(inode);
2060
2061         EXIT;
2062 }
2063
2064 int ll_iocontrol(struct inode *inode, struct file *file,
2065                  unsigned int cmd, unsigned long arg)
2066 {
2067         struct ll_sb_info *sbi = ll_i2sbi(inode);
2068         struct ptlrpc_request *req = NULL;
2069         int rc, flags = 0;
2070         ENTRY;
2071
2072         switch(cmd) {
2073         case FSFILT_IOC_GETFLAGS: {
2074                 struct mdt_body *body;
2075                 struct md_op_data *op_data;
2076
2077                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2078                                              0, 0, LUSTRE_OPC_ANY,
2079                                              NULL);
2080                 if (IS_ERR(op_data))
2081                         RETURN(PTR_ERR(op_data));
2082
2083                 op_data->op_valid = OBD_MD_FLFLAGS;
2084                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2085                 ll_finish_md_op_data(op_data);
2086                 if (rc) {
2087                         CERROR("%s: failure inode "DFID": rc = %d\n",
2088                                sbi->ll_md_exp->exp_obd->obd_name,
2089                                PFID(ll_inode2fid(inode)), rc);
2090                         RETURN(-abs(rc));
2091                 }
2092
2093                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2094
2095                 flags = body->mbo_flags;
2096
2097                 ptlrpc_req_finished(req);
2098
2099                 RETURN(put_user(flags, (int __user *)arg));
2100         }
2101         case FSFILT_IOC_SETFLAGS: {
2102                 struct iattr *attr;
2103                 struct md_op_data *op_data;
2104                 struct cl_object *obj;
2105
2106                 if (get_user(flags, (int __user *)arg))
2107                         RETURN(-EFAULT);
2108
2109                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2110                                              LUSTRE_OPC_ANY, NULL);
2111                 if (IS_ERR(op_data))
2112                         RETURN(PTR_ERR(op_data));
2113
2114                 op_data->op_attr_flags = flags;
2115                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2116                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2117                 ll_finish_md_op_data(op_data);
2118                 ptlrpc_req_finished(req);
2119                 if (rc)
2120                         RETURN(rc);
2121
2122                 inode->i_flags = ll_ext_to_inode_flags(flags);
2123
2124                 obj = ll_i2info(inode)->lli_clob;
2125                 if (obj == NULL)
2126                         RETURN(0);
2127
2128                 OBD_ALLOC_PTR(attr);
2129                 if (attr == NULL)
2130                         RETURN(-ENOMEM);
2131
2132                 attr->ia_valid = ATTR_ATTR_FLAG;
2133                 rc = cl_setattr_ost(obj, attr, flags);
2134
2135                 OBD_FREE_PTR(attr);
2136                 RETURN(rc);
2137         }
2138         default:
2139                 RETURN(-ENOSYS);
2140         }
2141
2142         RETURN(0);
2143 }
2144
2145 int ll_flush_ctx(struct inode *inode)
2146 {
2147         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2148
2149         CDEBUG(D_SEC, "flush context for user %d\n",
2150                from_kuid(&init_user_ns, current_uid()));
2151
2152         obd_set_info_async(NULL, sbi->ll_md_exp,
2153                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2154                            0, NULL, NULL);
2155         obd_set_info_async(NULL, sbi->ll_dt_exp,
2156                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2157                            0, NULL, NULL);
2158         return 0;
2159 }
2160
2161 /* umount -f client means force down, don't save state */
2162 void ll_umount_begin(struct super_block *sb)
2163 {
2164         struct ll_sb_info *sbi = ll_s2sbi(sb);
2165         struct obd_device *obd;
2166         struct obd_ioctl_data *ioc_data;
2167         ENTRY;
2168
2169         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2170                sb->s_count, atomic_read(&sb->s_active));
2171
2172         obd = class_exp2obd(sbi->ll_md_exp);
2173         if (obd == NULL) {
2174                 CERROR("Invalid MDC connection handle "LPX64"\n",
2175                        sbi->ll_md_exp->exp_handle.h_cookie);
2176                 EXIT;
2177                 return;
2178         }
2179         obd->obd_force = 1;
2180
2181         obd = class_exp2obd(sbi->ll_dt_exp);
2182         if (obd == NULL) {
2183                 CERROR("Invalid LOV connection handle "LPX64"\n",
2184                        sbi->ll_dt_exp->exp_handle.h_cookie);
2185                 EXIT;
2186                 return;
2187         }
2188         obd->obd_force = 1;
2189
2190         OBD_ALLOC_PTR(ioc_data);
2191         if (ioc_data) {
2192                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2193                               sizeof *ioc_data, ioc_data, NULL);
2194
2195                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2196                               sizeof *ioc_data, ioc_data, NULL);
2197
2198                 OBD_FREE_PTR(ioc_data);
2199         }
2200
2201         /* Really, we'd like to wait until there are no requests outstanding,
2202          * and then continue.  For now, we just invalidate the requests,
2203          * schedule() and sleep one second if needed, and hope.
2204          */
2205         schedule();
2206         EXIT;
2207 }
2208
2209 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2210 {
2211         struct ll_sb_info *sbi = ll_s2sbi(sb);
2212         char *profilenm = get_profile_name(sb);
2213         int err;
2214         __u32 read_only;
2215
2216         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2217                 read_only = *flags & MS_RDONLY;
2218                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2219                                          sizeof(KEY_READ_ONLY),
2220                                          KEY_READ_ONLY, sizeof(read_only),
2221                                          &read_only, NULL);
2222                 if (err) {
2223                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2224                                       profilenm, read_only ?
2225                                       "read-only" : "read-write", err);
2226                         return err;
2227                 }
2228
2229                 if (read_only)
2230                         sb->s_flags |= MS_RDONLY;
2231                 else
2232                         sb->s_flags &= ~MS_RDONLY;
2233
2234                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2235                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2236                                       read_only ?  "read-only" : "read-write");
2237         }
2238         return 0;
2239 }
2240
2241 /**
2242  * Cleanup the open handle that is cached on MDT-side.
2243  *
2244  * For open case, the client side open handling thread may hit error
2245  * after the MDT grant the open. Under such case, the client should
2246  * send close RPC to the MDT as cleanup; otherwise, the open handle
2247  * on the MDT will be leaked there until the client umount or evicted.
2248  *
2249  * In further, if someone unlinked the file, because the open handle
2250  * holds the reference on such file/object, then it will block the
2251  * subsequent threads that want to locate such object via FID.
2252  *
2253  * \param[in] sb        super block for this file-system
2254  * \param[in] open_req  pointer to the original open request
2255  */
2256 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2257 {
2258         struct mdt_body                 *body;
2259         struct md_op_data               *op_data;
2260         struct ptlrpc_request           *close_req = NULL;
2261         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2262         ENTRY;
2263
2264         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2265         OBD_ALLOC_PTR(op_data);
2266         if (op_data == NULL) {
2267                 CWARN("%s: cannot allocate op_data to release open handle for "
2268                       DFID"\n",
2269                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2270
2271                 RETURN_EXIT;
2272         }
2273
2274         op_data->op_fid1 = body->mbo_fid1;
2275         op_data->op_handle = body->mbo_handle;
2276         op_data->op_mod_time = cfs_time_current_sec();
2277         md_close(exp, op_data, NULL, &close_req);
2278         ptlrpc_req_finished(close_req);
2279         ll_finish_md_op_data(op_data);
2280
2281         EXIT;
2282 }
2283
2284 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2285                   struct super_block *sb, struct lookup_intent *it)
2286 {
2287         struct ll_sb_info *sbi = NULL;
2288         struct lustre_md md = { NULL };
2289         int rc;
2290         ENTRY;
2291
2292         LASSERT(*inode || sb);
2293         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2294         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2295                               sbi->ll_md_exp, &md);
2296         if (rc != 0)
2297                 GOTO(cleanup, rc);
2298
2299         if (*inode) {
2300                 rc = ll_update_inode(*inode, &md);
2301                 if (rc != 0)
2302                         GOTO(out, rc);
2303         } else {
2304                 LASSERT(sb != NULL);
2305
2306                 /*
2307                  * At this point server returns to client's same fid as client
2308                  * generated for creating. So using ->fid1 is okay here.
2309                  */
2310                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2311                         CERROR("%s: Fid is insane "DFID"\n",
2312                                 ll_get_fsname(sb, NULL, 0),
2313                                 PFID(&md.body->mbo_fid1));
2314                         GOTO(out, rc = -EINVAL);
2315                 }
2316
2317                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2318                                              sbi->ll_flags & LL_SBI_32BIT_API),
2319                                  &md);
2320                 if (IS_ERR(*inode)) {
2321 #ifdef CONFIG_FS_POSIX_ACL
2322                         if (md.posix_acl) {
2323                                 posix_acl_release(md.posix_acl);
2324                                 md.posix_acl = NULL;
2325                         }
2326 #endif
2327                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2328                         *inode = NULL;
2329                         CERROR("new_inode -fatal: rc %d\n", rc);
2330                         GOTO(out, rc);
2331                 }
2332         }
2333
2334         /* Handling piggyback layout lock.
2335          * Layout lock can be piggybacked by getattr and open request.
2336          * The lsm can be applied to inode only if it comes with a layout lock
2337          * otherwise correct layout may be overwritten, for example:
2338          * 1. proc1: mdt returns a lsm but not granting layout
2339          * 2. layout was changed by another client
2340          * 3. proc2: refresh layout and layout lock granted
2341          * 4. proc1: to apply a stale layout */
2342         if (it != NULL && it->it_lock_mode != 0) {
2343                 struct lustre_handle lockh;
2344                 struct ldlm_lock *lock;
2345
2346                 lockh.cookie = it->it_lock_handle;
2347                 lock = ldlm_handle2lock(&lockh);
2348                 LASSERT(lock != NULL);
2349                 if (ldlm_has_layout(lock)) {
2350                         struct cl_object_conf conf;
2351
2352                         memset(&conf, 0, sizeof(conf));
2353                         conf.coc_opc = OBJECT_CONF_SET;
2354                         conf.coc_inode = *inode;
2355                         conf.coc_lock = lock;
2356                         conf.u.coc_layout = md.layout;
2357                         (void)ll_layout_conf(*inode, &conf);
2358                 }
2359                 LDLM_LOCK_PUT(lock);
2360         }
2361
2362         GOTO(out, rc = 0);
2363
2364 out:
2365         md_free_lustre_md(sbi->ll_md_exp, &md);
2366
2367 cleanup:
2368         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2369                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2370
2371         return rc;
2372 }
2373
2374 int ll_obd_statfs(struct inode *inode, void __user *arg)
2375 {
2376         struct ll_sb_info *sbi = NULL;
2377         struct obd_export *exp;
2378         char *buf = NULL;
2379         struct obd_ioctl_data *data = NULL;
2380         __u32 type;
2381         int len = 0, rc;
2382
2383         if (!inode || !(sbi = ll_i2sbi(inode)))
2384                 GOTO(out_statfs, rc = -EINVAL);
2385
2386         rc = obd_ioctl_getdata(&buf, &len, arg);
2387         if (rc)
2388                 GOTO(out_statfs, rc);
2389
2390         data = (void*)buf;
2391         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2392             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2393                 GOTO(out_statfs, rc = -EINVAL);
2394
2395         if (data->ioc_inllen1 != sizeof(__u32) ||
2396             data->ioc_inllen2 != sizeof(__u32) ||
2397             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2398             data->ioc_plen2 != sizeof(struct obd_uuid))
2399                 GOTO(out_statfs, rc = -EINVAL);
2400
2401         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2402         if (type & LL_STATFS_LMV)
2403                 exp = sbi->ll_md_exp;
2404         else if (type & LL_STATFS_LOV)
2405                 exp = sbi->ll_dt_exp;
2406         else
2407                 GOTO(out_statfs, rc = -ENODEV);
2408
2409         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2410         if (rc)
2411                 GOTO(out_statfs, rc);
2412 out_statfs:
2413         if (buf)
2414                 obd_ioctl_freedata(buf, len);
2415         return rc;
2416 }
2417
2418 int ll_process_config(struct lustre_cfg *lcfg)
2419 {
2420         struct super_block *sb;
2421         unsigned long x;
2422         int rc = 0;
2423         char *ptr;
2424
2425         /* The instance name contains the sb: lustre-client-aacfe000 */
2426         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2427         if (!ptr || !*(++ptr))
2428                 return -EINVAL;
2429         if (sscanf(ptr, "%lx", &x) != 1)
2430                 return -EINVAL;
2431         sb = (struct super_block *)x;
2432         /* This better be a real Lustre superblock! */
2433         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2434
2435         /* Note we have not called client_common_fill_super yet, so
2436            proc fns must be able to handle that! */
2437         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2438                                       lcfg, sb);
2439         if (rc > 0)
2440                 rc = 0;
2441         return rc;
2442 }
2443
2444 /* this function prepares md_op_data hint for passing it down to MD stack. */
2445 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2446                                       struct inode *i1, struct inode *i2,
2447                                       const char *name, size_t namelen,
2448                                       __u32 mode, __u32 opc, void *data)
2449 {
2450         LASSERT(i1 != NULL);
2451
2452         if (name == NULL) {
2453                 /* Do not reuse namelen for something else. */
2454                 if (namelen != 0)
2455                         return ERR_PTR(-EINVAL);
2456         } else {
2457                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2458                         return ERR_PTR(-ENAMETOOLONG);
2459
2460                 if (!lu_name_is_valid_2(name, namelen))
2461                         return ERR_PTR(-EINVAL);
2462         }
2463
2464         if (op_data == NULL)
2465                 OBD_ALLOC_PTR(op_data);
2466
2467         if (op_data == NULL)
2468                 return ERR_PTR(-ENOMEM);
2469
2470         ll_i2gids(op_data->op_suppgids, i1, i2);
2471         op_data->op_fid1 = *ll_inode2fid(i1);
2472         op_data->op_default_stripe_offset = -1;
2473         if (S_ISDIR(i1->i_mode)) {
2474                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2475                 if (opc == LUSTRE_OPC_MKDIR)
2476                         op_data->op_default_stripe_offset =
2477                                    ll_i2info(i1)->lli_def_stripe_offset;
2478         }
2479
2480         if (i2) {
2481                 op_data->op_fid2 = *ll_inode2fid(i2);
2482                 if (S_ISDIR(i2->i_mode))
2483                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2484         } else {
2485                 fid_zero(&op_data->op_fid2);
2486         }
2487
2488         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2489                 op_data->op_cli_flags |= CLI_HASH64;
2490
2491         if (ll_need_32bit_api(ll_i2sbi(i1)))
2492                 op_data->op_cli_flags |= CLI_API32;
2493
2494         op_data->op_name = name;
2495         op_data->op_namelen = namelen;
2496         op_data->op_mode = mode;
2497         op_data->op_mod_time = cfs_time_current_sec();
2498         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2499         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2500         op_data->op_cap = cfs_curproc_cap_pack();
2501         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2502              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2503                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2504         } else {
2505                 op_data->op_mds = 0;
2506         }
2507         op_data->op_data = data;
2508
2509         return op_data;
2510 }
2511
2512 void ll_finish_md_op_data(struct md_op_data *op_data)
2513 {
2514         OBD_FREE_PTR(op_data);
2515 }
2516
2517 #ifdef HAVE_SUPEROPS_USE_DENTRY
2518 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2519 #else
2520 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2521 #endif
2522 {
2523         struct ll_sb_info *sbi;
2524
2525 #ifdef HAVE_SUPEROPS_USE_DENTRY
2526         LASSERT((seq != NULL) && (dentry != NULL));
2527         sbi = ll_s2sbi(dentry->d_sb);
2528 #else
2529         LASSERT((seq != NULL) && (vfs != NULL));
2530         sbi = ll_s2sbi(vfs->mnt_sb);
2531 #endif
2532
2533         if (sbi->ll_flags & LL_SBI_NOLCK)
2534                 seq_puts(seq, ",nolock");
2535
2536         if (sbi->ll_flags & LL_SBI_FLOCK)
2537                 seq_puts(seq, ",flock");
2538
2539         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2540                 seq_puts(seq, ",localflock");
2541
2542         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2543                 seq_puts(seq, ",user_xattr");
2544
2545         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2546                 seq_puts(seq, ",lazystatfs");
2547
2548         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2549                 seq_puts(seq, ",user_fid2path");
2550
2551         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2552                 seq_puts(seq, ",always_ping");
2553
2554         RETURN(0);
2555 }
2556
2557 /**
2558  * Get obd name by cmd, and copy out to user space
2559  */
2560 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2561 {
2562         struct ll_sb_info *sbi = ll_i2sbi(inode);
2563         struct obd_device *obd;
2564         ENTRY;
2565
2566         if (cmd == OBD_IOC_GETDTNAME)
2567                 obd = class_exp2obd(sbi->ll_dt_exp);
2568         else if (cmd == OBD_IOC_GETMDNAME)
2569                 obd = class_exp2obd(sbi->ll_md_exp);
2570         else
2571                 RETURN(-EINVAL);
2572
2573         if (!obd)
2574                 RETURN(-ENOENT);
2575
2576         if (copy_to_user((void __user *)arg, obd->obd_name,
2577                          strlen(obd->obd_name) + 1))
2578                 RETURN(-EFAULT);
2579
2580         RETURN(0);
2581 }
2582
2583 /**
2584  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2585  * fsname will be returned in this buffer; otherwise, a static buffer will be
2586  * used to store the fsname and returned to caller.
2587  */
2588 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2589 {
2590         static char fsname_static[MTI_NAME_MAXLEN];
2591         struct lustre_sb_info *lsi = s2lsi(sb);
2592         char *ptr;
2593         int len;
2594
2595         if (buf == NULL) {
2596                 /* this means the caller wants to use static buffer
2597                  * and it doesn't care about race. Usually this is
2598                  * in error reporting path */
2599                 buf = fsname_static;
2600                 buflen = sizeof(fsname_static);
2601         }
2602
2603         len = strlen(lsi->lsi_lmd->lmd_profile);
2604         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2605         if (ptr && (strcmp(ptr, "-client") == 0))
2606                 len -= 7;
2607
2608         if (unlikely(len >= buflen))
2609                 len = buflen - 1;
2610         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2611         buf[len] = '\0';
2612
2613         return buf;
2614 }
2615
2616 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2617 {
2618         char *path = NULL;
2619
2620         struct path p;
2621
2622         p.dentry = dentry;
2623         p.mnt = current->fs->root.mnt;
2624         path_get(&p);
2625         path = d_path(&p, buf, bufsize);
2626         path_put(&p);
2627         return path;
2628 }
2629
2630 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2631 {
2632         char *buf, *path = NULL;
2633         struct dentry *dentry = NULL;
2634         struct inode *inode = page->mapping->host;
2635
2636         /* this can be called inside spin lock so use GFP_ATOMIC. */
2637         buf = (char *)__get_free_page(GFP_ATOMIC);
2638         if (buf != NULL) {
2639                 dentry = d_find_alias(page->mapping->host);
2640                 if (dentry != NULL)
2641                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2642         }
2643
2644         CDEBUG(D_WARNING,
2645                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2646                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2647                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2648                PFID(ll_inode2fid(inode)),
2649                (path && !IS_ERR(path)) ? path : "", ioret);
2650
2651         if (dentry != NULL)
2652                 dput(dentry);
2653
2654         if (buf != NULL)
2655                 free_page((unsigned long)buf);
2656 }
2657
2658 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2659                         struct lov_user_md **kbuf)
2660 {
2661         struct lov_user_md      lum;
2662         ssize_t                 lum_size;
2663         ENTRY;
2664
2665         if (copy_from_user(&lum, md, sizeof(lum)))
2666                 RETURN(-EFAULT);
2667
2668         lum_size = ll_lov_user_md_size(&lum);
2669         if (lum_size < 0)
2670                 RETURN(lum_size);
2671
2672         OBD_ALLOC(*kbuf, lum_size);
2673         if (*kbuf == NULL)
2674                 RETURN(-ENOMEM);
2675
2676         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2677                 OBD_FREE(*kbuf, lum_size);
2678                 RETURN(-EFAULT);
2679         }
2680
2681         RETURN(lum_size);
2682 }
2683
2684 /*
2685  * Compute llite root squash state after a change of root squash
2686  * configuration setting or add/remove of a lnet nid
2687  */
2688 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2689 {
2690         struct root_squash_info *squash = &sbi->ll_squash;
2691         int i;
2692         bool matched;
2693         lnet_process_id_t id;
2694
2695         /* Update norootsquash flag */
2696         down_write(&squash->rsi_sem);
2697         if (list_empty(&squash->rsi_nosquash_nids))
2698                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2699         else {
2700                 /* Do not apply root squash as soon as one of our NIDs is
2701                  * in the nosquash_nids list */
2702                 matched = false;
2703                 i = 0;
2704                 while (LNetGetId(i++, &id) != -ENOENT) {
2705                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2706                                 continue;
2707                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2708                                 matched = true;
2709                                 break;
2710                         }
2711                 }
2712                 if (matched)
2713                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2714                 else
2715                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2716         }
2717         up_write(&squash->rsi_sem);
2718 }
2719
2720 /**
2721  * Parse linkea content to extract information about a given hardlink
2722  *
2723  * \param[in]   ldata      - Initialized linkea data
2724  * \param[in]   linkno     - Link identifier
2725  * \param[out]  parent_fid - The entry's parent FID
2726  * \param[out]  ln         - Entry name destination buffer
2727  *
2728  * \retval 0 on success
2729  * \retval Appropriate negative error code on failure
2730  */
2731 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2732                             struct lu_fid *parent_fid, struct lu_name *ln)
2733 {
2734         unsigned int    idx;
2735         int             rc;
2736         ENTRY;
2737
2738         rc = linkea_init(ldata);
2739         if (rc < 0)
2740                 RETURN(rc);
2741
2742         if (linkno >= ldata->ld_leh->leh_reccount)
2743                 /* beyond last link */
2744                 RETURN(-ENODATA);
2745
2746         linkea_first_entry(ldata);
2747         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2748                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2749                                     parent_fid);
2750                 if (idx == linkno)
2751                         break;
2752
2753                 linkea_next_entry(ldata);
2754         }
2755
2756         if (idx < linkno)
2757                 RETURN(-ENODATA);
2758
2759         RETURN(0);
2760 }
2761
2762 /**
2763  * Get parent FID and name of an identified link. Operation is performed for
2764  * a given link number, letting the caller iterate over linkno to list one or
2765  * all links of an entry.
2766  *
2767  * \param[in]     file - File descriptor against which to perform the operation
2768  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2769  *                       on and the available size. It is eventually filled with
2770  *                       the requested information or left untouched on error
2771  *
2772  * \retval - 0 on success
2773  * \retval - Appropriate negative error code on failure
2774  */
2775 int ll_getparent(struct file *file, struct getparent __user *arg)
2776 {
2777         struct dentry           *dentry = file->f_path.dentry;
2778         struct inode            *inode = dentry->d_inode;
2779         struct linkea_data      *ldata;
2780         struct lu_buf            buf = LU_BUF_NULL;
2781         struct lu_name           ln;
2782         struct lu_fid            parent_fid;
2783         __u32                    linkno;
2784         __u32                    name_size;
2785         int                      rc;
2786
2787         ENTRY;
2788
2789         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2790             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2791                 RETURN(-EPERM);
2792
2793         if (get_user(name_size, &arg->gp_name_size))
2794                 RETURN(-EFAULT);
2795
2796         if (get_user(linkno, &arg->gp_linkno))
2797                 RETURN(-EFAULT);
2798
2799         if (name_size > PATH_MAX)
2800                 RETURN(-EINVAL);
2801
2802         OBD_ALLOC(ldata, sizeof(*ldata));
2803         if (ldata == NULL)
2804                 RETURN(-ENOMEM);
2805
2806         rc = linkea_data_new(ldata, &buf);
2807         if (rc < 0)
2808                 GOTO(ldata_free, rc);
2809
2810         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2811         if (rc < 0)
2812                 GOTO(lb_free, rc);
2813
2814         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2815         if (rc < 0)
2816                 GOTO(lb_free, rc);
2817
2818         if (ln.ln_namelen >= name_size)
2819                 GOTO(lb_free, rc = -EOVERFLOW);
2820
2821         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2822                 GOTO(lb_free, rc = -EFAULT);
2823
2824         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2825                 GOTO(lb_free, rc = -EFAULT);
2826
2827         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2828                 GOTO(lb_free, rc = -EFAULT);
2829
2830 lb_free:
2831         lu_buf_free(&buf);
2832 ldata_free:
2833         OBD_FREE(ldata, sizeof(*ldata));
2834
2835         RETURN(rc);
2836 }