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, 2014, 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/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_ioctl.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         atomic_set(&sbi->ll_cache.ccc_users, 0);
  95         sbi->ll_cache.ccc_lru_max = lru_page_max;
  96         atomic_long_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  97         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  98         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
  99
 100         /* turn unstable check off by default as it impacts performance */
 101         sbi->ll_cache.ccc_unstable_check = 0;
 102         atomic_long_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 103         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 104
 105         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 106                                            SBI_DEFAULT_READAHEAD_MAX);
 107         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 108         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 109                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 110         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 111         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 112
 113         ll_generate_random_uuid(uuid);
 114         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 115         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 116
 117         spin_lock(&ll_sb_lock);
 118         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 119         spin_unlock(&ll_sb_lock);
 120
 121         sbi->ll_flags |= LL_SBI_VERBOSE;
 122 #ifdef ENABLE_CHECKSUM
 123         sbi->ll_flags |= LL_SBI_CHECKSUM;
 124 #endif
 125
 126 #ifdef HAVE_LRU_RESIZE_SUPPORT
 127         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 128 #endif
 129
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 132                                pp_r_hist.oh_lock);
 133                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 134                                pp_w_hist.oh_lock);
 135         }
 136
 137         /* metadata statahead is enabled by default */
 138         sbi->ll_sa_max = LL_SA_RPC_DEF;
 139         atomic_set(&sbi->ll_sa_total, 0);
 140         atomic_set(&sbi->ll_sa_wrong, 0);
 141         atomic_set(&sbi->ll_sa_running, 0);
 142         atomic_set(&sbi->ll_agl_total, 0);
 143         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 144
 145         /* root squash */
 146         sbi->ll_squash.rsi_uid = 0;
 147         sbi->ll_squash.rsi_gid = 0;
 148         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 149         init_rwsem(&sbi->ll_squash.rsi_sem);
 150
 151         RETURN(sbi);
 152 }
 153
 154 static void ll_free_sbi(struct super_block *sb)
 155 {
 156         struct ll_sb_info *sbi = ll_s2sbi(sb);
 157         ENTRY;
 158
 159         if (sbi != NULL) {
 160                 spin_lock(&ll_sb_lock);
 161                 list_del(&sbi->ll_list);
 162                 spin_unlock(&ll_sb_lock);
 163                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 164                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 165                 OBD_FREE(sbi, sizeof(*sbi));
 166         }
 167         EXIT;
 168 }
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = NULL;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         u64 valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         /* indicate the features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 208                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 209                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 210                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 211                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 212                                   OBD_CONNECT_EINPROGRESS |
 213                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 214                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 215                                   OBD_CONNECT_MAX_EASIZE |
 216                                   OBD_CONNECT_FLOCK_DEAD |
 217                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 218                                   OBD_CONNECT_OPEN_BY_FID |
 219                                   OBD_CONNECT_DIR_STRIPE;
 220
 221         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 222                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 230 #endif
 231
 232         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 233                 /* flag mdc connection as lightweight, only used for test
 234                  * purpose, use with care */
 235                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 236
 237         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 238         data->ocd_version = LUSTRE_VERSION_CODE;
 239
 240         if (sb->s_flags & MS_RDONLY)
 241                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 242         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 243                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 244
 245 #ifdef HAVE_MS_FLOCK_LOCK
 246         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 247         sb->s_flags |= MS_FLOCK_LOCK;
 248 #endif
 249 #ifdef MS_HAS_NEW_AOPS
 250         sb->s_flags |= MS_HAS_NEW_AOPS;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* real client */
 261         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 262         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 263                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data, NULL);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 338
 339         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 340             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 341                 LCONSOLE_INFO("Disabling user_xattr feature because "
 342                               "it is not supported on the server\n");
 343                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags |= MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags |= LL_SBI_ACL;
 351         } else {
 352                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 353 #ifdef MS_POSIXACL
 354                 sb->s_flags &= ~MS_POSIXACL;
 355 #endif
 356                 sbi->ll_flags &= ~LL_SBI_ACL;
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 360                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 361                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client is set as remote by default.\n");
 363                 }
 364         } else {
 365                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 366                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 367                         LCONSOLE_INFO("client claims to be remote, but server "
 368                                       "rejected, forced to be local.\n");
 369                 }
 370         }
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 373                 LCONSOLE_INFO("client enabled MDS capability!\n");
 374                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 378                 LCONSOLE_INFO("client enabled OSS capability!\n");
 379                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 380         }
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 383                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 384
 385         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 386                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 387         else
 388                 sbi->ll_md_brw_pages = 1;
 389
 390         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 391                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 392
 393         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 394                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 395                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 396                                       "unknown maximum xattr size.\n", dt);
 397                 } else {
 398                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 399                         sbi->ll_xattr_cache_enabled = 1;
 400                 }
 401         }
 402
 403         obd = class_name2obd(dt);
 404         if (!obd) {
 405                 CERROR("DT %s: not setup or attached\n", dt);
 406                 GOTO(out_md_fid, err = -ENODEV);
 407         }
 408
 409         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 410                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 411                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 412                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 413                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 414                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 415                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 416                                   OBD_CONNECT_MAXBYTES |
 417                                   OBD_CONNECT_EINPROGRESS |
 418                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 419                                   OBD_CONNECT_LAYOUTLOCK |
 420                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 421
 422         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 423                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 424
 425         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 426                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 427                  * disabled by default, because it can still be enabled on the
 428                  * fly via /proc. As a consequence, we still need to come to an
 429                  * agreement on the supported algorithms at connect time */
 430                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 431
 432                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 433                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 434                 else
 435                         data->ocd_cksum_types = cksum_types_supported_client();
 436         }
 437
 438 #ifdef HAVE_LRU_RESIZE_SUPPORT
 439         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 440 #endif
 441         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 442                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 443
 444         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 445                "ocd_grant: %d\n", data->ocd_connect_flags,
 446                data->ocd_version, data->ocd_grant);
 447
 448         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 449         obd->obd_upcall.onu_upcall = cl_ocd_update;
 450
 451         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 452
 453         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 454                           NULL);
 455         if (err == -EBUSY) {
 456                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 457                                    "recovery, of which this client is not a "
 458                                    "part.  Please wait for recovery to "
 459                                    "complete, abort, or time out.\n", dt);
 460                 GOTO(out_md, err);
 461         } else if (err) {
 462                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 463                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 464                 GOTO(out_md, err);
 465         }
 466
 467         sbi->ll_dt_exp->exp_connect_data = *data;
 468
 469         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 470                            LUSTRE_SEQ_METADATA);
 471         if (err) {
 472                 CERROR("%s: Can't init data layer FID infrastructure, "
 473                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 474                 GOTO(out_dt, err);
 475         }
 476
 477         mutex_lock(&sbi->ll_lco.lco_lock);
 478         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 479         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 480         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 481         mutex_unlock(&sbi->ll_lco.lco_lock);
 482
 483         fid_zero(&sbi->ll_root_fid);
 484         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 485         if (err) {
 486                 CERROR("cannot mds_connect: rc = %d\n", err);
 487                 GOTO(out_lock_cn_cb, err);
 488         }
 489         if (!fid_is_sane(&sbi->ll_root_fid)) {
 490                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 491                        sbi->ll_md_exp->exp_obd->obd_name,
 492                        PFID(&sbi->ll_root_fid));
 493                 GOTO(out_lock_cn_cb, err = -EINVAL);
 494         }
 495         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 496
 497         sb->s_op = &lustre_super_operations;
 498 #if THREAD_SIZE >= 8192 /*b=17630*/
 499         sb->s_export_op = &lustre_export_operations;
 500 #endif
 501
 502         /* make root inode
 503          * XXX: move this to after cbd setup? */
 504         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 505                 OBD_MD_FLMODEASIZE;
 506         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 507                 valid |= OBD_MD_FLRMTPERM;
 508         else if (sbi->ll_flags & LL_SBI_ACL)
 509                 valid |= OBD_MD_FLACL;
 510
 511         OBD_ALLOC_PTR(op_data);
 512         if (op_data == NULL)
 513                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 514
 515         op_data->op_fid1 = sbi->ll_root_fid;
 516         op_data->op_mode = 0;
 517         op_data->op_capa1 = oc;
 518         op_data->op_valid = valid;
 519
 520         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 521         if (oc)
 522                 capa_put(oc);
 523         OBD_FREE_PTR(op_data);
 524         if (err) {
 525                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 526                        sbi->ll_md_exp->exp_obd->obd_name, err);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 531                                sbi->ll_md_exp, &lmd);
 532         if (err) {
 533                 CERROR("failed to understand root inode md: rc = %d\n", err);
 534                 ptlrpc_req_finished(request);
 535                 GOTO(out_lock_cn_cb, err);
 536         }
 537
 538         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 539         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 540                                             sbi->ll_flags & LL_SBI_32BIT_API),
 541                        &lmd);
 542         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 543         ptlrpc_req_finished(request);
 544
 545         if (IS_ERR(root)) {
 546                 if (lmd.lsm)
 547                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 548 #ifdef CONFIG_FS_POSIX_ACL
 549                 if (lmd.posix_acl) {
 550                         posix_acl_release(lmd.posix_acl);
 551                         lmd.posix_acl = NULL;
 552                 }
 553 #endif
 554                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 555                 root = NULL;
 556                 CERROR("lustre_lite: bad iget4 for root\n");
 557                 GOTO(out_root, err);
 558         }
 559
 560         err = ll_close_thread_start(&sbi->ll_lcq);
 561         if (err) {
 562                 CERROR("cannot start close thread: rc %d\n", err);
 563                 GOTO(out_root, err);
 564         }
 565
 566 #ifdef CONFIG_FS_POSIX_ACL
 567         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 568                 rct_init(&sbi->ll_rct);
 569                 et_init(&sbi->ll_et);
 570         }
 571 #endif
 572
 573         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 574         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 575                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 576                                  NULL);
 577         cl_sb_init(sb);
 578
 579         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 580                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 581                                  &sbi->ll_cache, NULL);
 582
 583         sb->s_root = d_make_root(root);
 584         if (sb->s_root == NULL) {
 585                 CERROR("%s: can't make root dentry\n",
 586                         ll_get_fsname(sb, NULL, 0));
 587                 GOTO(out_root, err = -ENOMEM);
 588         }
 589 #ifdef HAVE_DCACHE_LOCK
 590         sb->s_root->d_op = &ll_d_ops;
 591 #endif
 592
 593         sbi->ll_sdev_orig = sb->s_dev;
 594
 595         /* We set sb->s_dev equal on all lustre clients in order to support
 596          * NFS export clustering.  NFSD requires that the FSID be the same
 597          * on all clients. */
 598         /* s_dev is also used in lt_compare() to compare two fs, but that is
 599          * only a node-local comparison. */
 600         uuid = obd_get_uuid(sbi->ll_md_exp);
 601         if (uuid != NULL)
 602                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 603
 604         if (data != NULL)
 605                 OBD_FREE_PTR(data);
 606         if (osfs != NULL)
 607                 OBD_FREE_PTR(osfs);
 608         if (proc_lustre_fs_root != NULL) {
 609                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 610                                                   dt, md);
 611                 if (err < 0) {
 612                         CERROR("%s: could not register mount in lprocfs: "
 613                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 614                         err = 0;
 615                 }
 616         }
 617
 618         RETURN(err);
 619 out_root:
 620         if (root)
 621                 iput(root);
 622 out_lock_cn_cb:
 623         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 624 out_dt:
 625         obd_disconnect(sbi->ll_dt_exp);
 626         sbi->ll_dt_exp = NULL;
 627         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 628         obd_zombie_barrier();
 629 out_md_fid:
 630         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 631 out_md:
 632         obd_disconnect(sbi->ll_md_exp);
 633         sbi->ll_md_exp = NULL;
 634 out:
 635         if (data != NULL)
 636                 OBD_FREE_PTR(data);
 637         if (osfs != NULL)
 638                 OBD_FREE_PTR(osfs);
 639         return err;
 640 }
 641
 642 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 643 {
 644         int size, rc;
 645
 646         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 647         size = sizeof(int);
 648         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 649                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 650         if (rc)
 651                 CERROR("Get max mdsize error rc %d\n", rc);
 652
 653         RETURN(rc);
 654 }
 655
 656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 657 {
 658         int size, rc;
 659
 660         size = sizeof(int);
 661         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 662                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 663         if (rc)
 664                 CERROR("Get default mdsize error rc %d\n", rc);
 665
 666         RETURN(rc);
 667 }
 668
 669 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 670 {
 671         int size, rc;
 672
 673         size = sizeof(int);
 674         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 675                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 676         if (rc)
 677                 CERROR("Get max cookiesize error rc %d\n", rc);
 678
 679         RETURN(rc);
 680 }
 681
 682 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 683 {
 684         int size, rc;
 685
 686         size = sizeof(int);
 687         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 688                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 689         if (rc)
 690                 CERROR("Get default cookiesize error rc %d\n", rc);
 691
 692         RETURN(rc);
 693 }
 694
 695 static void ll_dump_inode(struct inode *inode)
 696 {
 697         struct ll_d_hlist_node *tmp;
 698         int dentry_count = 0;
 699
 700         LASSERT(inode != NULL);
 701
 702         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 703                 dentry_count++;
 704
 705         CERROR("%s: inode %p dump: dev=%s fid="DFID
 706                " mode=%o count=%u, %d dentries\n",
 707                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 708                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 709                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 710 }
 711
 712 void lustre_dump_dentry(struct dentry *dentry, int recur)
 713 {
 714         struct list_head *tmp;
 715         int subdirs = 0;
 716
 717         LASSERT(dentry != NULL);
 718
 719         list_for_each(tmp, &dentry->d_subdirs)
 720                 subdirs++;
 721
 722         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 723                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 724                dentry->d_name.len, dentry->d_name.name,
 725                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 726                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 727                dentry->d_flags, dentry->d_fsdata, subdirs);
 728         if (dentry->d_inode != NULL)
 729                 ll_dump_inode(dentry->d_inode);
 730
 731         if (recur == 0)
 732                 return;
 733
 734         list_for_each(tmp, &dentry->d_subdirs) {
 735                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 736                 lustre_dump_dentry(d, recur - 1);
 737         }
 738 }
 739
 740 static void client_common_put_super(struct super_block *sb)
 741 {
 742         struct ll_sb_info *sbi = ll_s2sbi(sb);
 743         ENTRY;
 744
 745 #ifdef CONFIG_FS_POSIX_ACL
 746         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 747                 et_fini(&sbi->ll_et);
 748                 rct_fini(&sbi->ll_rct);
 749         }
 750 #endif
 751
 752         ll_close_thread_shutdown(sbi->ll_lcq);
 753
 754         cl_sb_fini(sb);
 755
 756         list_del(&sbi->ll_conn_chain);
 757
 758         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 759         obd_disconnect(sbi->ll_dt_exp);
 760         sbi->ll_dt_exp = NULL;
 761         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 762          * see LU-2543. */
 763         obd_zombie_barrier();
 764
 765         lprocfs_unregister_mountpoint(sbi);
 766
 767         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 768         obd_disconnect(sbi->ll_md_exp);
 769         sbi->ll_md_exp = NULL;
 770
 771         EXIT;
 772 }
 773
 774 void ll_kill_super(struct super_block *sb)
 775 {
 776         struct ll_sb_info *sbi;
 777         ENTRY;
 778
 779         /* not init sb ?*/
 780         if (!(sb->s_flags & MS_ACTIVE))
 781                 return;
 782
 783         sbi = ll_s2sbi(sb);
 784         /* we need restore s_dev from changed for clustred NFS before put_super
 785          * because new kernels have cached s_dev and change sb->s_dev in
 786          * put_super not affected real removing devices */
 787         if (sbi) {
 788                 sb->s_dev = sbi->ll_sdev_orig;
 789                 sbi->ll_umounting = 1;
 790
 791                 /* wait running statahead threads to quit */
 792                 while (atomic_read(&sbi->ll_sa_running) > 0)
 793                         schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
 794                                 msecs_to_jiffies(MSEC_PER_SEC >> 3));
 795         }
 796
 797         EXIT;
 798 }
 799
 800 static inline int ll_set_opt(const char *opt, char *data, int fl)
 801 {
 802         if (strncmp(opt, data, strlen(opt)) != 0)
 803                 return(0);
 804         else
 805                 return(fl);
 806 }
 807
 808 /* non-client-specific mount options are parsed in lmd_parse */
 809 static int ll_options(char *options, int *flags)
 810 {
 811         int tmp;
 812         char *s1 = options, *s2;
 813         ENTRY;
 814
 815         if (!options)
 816                 RETURN(0);
 817
 818         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 819
 820         while (*s1) {
 821                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 822                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 828                 if (tmp) {
 829                         *flags |= tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 833                 if (tmp) {
 834                         *flags |= tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 838                 if (tmp) {
 839                         *flags &= ~tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867
 868                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 894                 if (tmp) {
 895                         *flags &= ~tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 904                 if (tmp) {
 905                         *flags |= tmp;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 909                 if (tmp) {
 910                         *flags |= tmp;
 911                         goto next;
 912                 }
 913                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 914                 if (tmp) {
 915                         *flags &= ~tmp;
 916                         goto next;
 917                 }
 918                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 919                                    s1);
 920                 RETURN(-EINVAL);
 921
 922 next:
 923                 /* Find next opt */
 924                 s2 = strchr(s1, ',');
 925                 if (s2 == NULL)
 926                         break;
 927                 s1 = s2 + 1;
 928         }
 929         RETURN(0);
 930 }
 931
 932 void ll_lli_init(struct ll_inode_info *lli)
 933 {
 934         lli->lli_inode_magic = LLI_INODE_MAGIC;
 935         lli->lli_flags = 0;
 936         lli->lli_ioepoch = 0;
 937         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 938         spin_lock_init(&lli->lli_lock);
 939         lli->lli_posix_acl = NULL;
 940         lli->lli_remote_perms = NULL;
 941         mutex_init(&lli->lli_rmtperm_mutex);
 942         /* Do not set lli_fid, it has been initialized already. */
 943         fid_zero(&lli->lli_pfid);
 944         INIT_LIST_HEAD(&lli->lli_close_list);
 945         INIT_LIST_HEAD(&lli->lli_oss_capas);
 946         atomic_set(&lli->lli_open_count, 0);
 947         lli->lli_mds_capa = NULL;
 948         lli->lli_rmtperm_time = 0;
 949         lli->lli_pending_och = NULL;
 950         lli->lli_mds_read_och = NULL;
 951         lli->lli_mds_write_och = NULL;
 952         lli->lli_mds_exec_och = NULL;
 953         lli->lli_open_fd_read_count = 0;
 954         lli->lli_open_fd_write_count = 0;
 955         lli->lli_open_fd_exec_count = 0;
 956         mutex_init(&lli->lli_och_mutex);
 957         spin_lock_init(&lli->lli_agl_lock);
 958         lli->lli_has_smd = false;
 959         spin_lock_init(&lli->lli_layout_lock);
 960         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 961         lli->lli_clob = NULL;
 962
 963         init_rwsem(&lli->lli_xattrs_list_rwsem);
 964         mutex_init(&lli->lli_xattrs_enq_lock);
 965
 966         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 967         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 968                 mutex_init(&lli->lli_readdir_mutex);
 969                 lli->lli_opendir_key = NULL;
 970                 lli->lli_sai = NULL;
 971                 spin_lock_init(&lli->lli_sa_lock);
 972                 lli->lli_opendir_pid = 0;
 973                 lli->lli_sa_enabled = 0;
 974         } else {
 975                 mutex_init(&lli->lli_size_mutex);
 976                 lli->lli_symlink_name = NULL;
 977                 init_rwsem(&lli->lli_trunc_sem);
 978                 range_lock_tree_init(&lli->lli_write_tree);
 979                 init_rwsem(&lli->lli_glimpse_sem);
 980                 lli->lli_glimpse_time = 0;
 981                 INIT_LIST_HEAD(&lli->lli_agl_list);
 982                 lli->lli_agl_index = 0;
 983                 lli->lli_async_rc = 0;
 984         }
 985         mutex_init(&lli->lli_layout_mutex);
 986 }
 987
 988 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 989 {
 990         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 991
 992         bdi->name = "lustre";
 993         return bdi_register(bdi, NULL, "lustre-%d",
 994                             atomic_inc_return(&ll_bdi_num));
 995 }
 996
 997 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 998 {
 999         struct lustre_profile *lprof = NULL;
1000         struct lustre_sb_info *lsi = s2lsi(sb);
1001         struct ll_sb_info *sbi;
1002         char  *dt = NULL, *md = NULL;
1003         char  *profilenm = get_profile_name(sb);
1004         struct config_llog_instance *cfg;
1005         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1006         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1007         int    err;
1008         ENTRY;
1009
1010         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1011
1012         OBD_ALLOC_PTR(cfg);
1013         if (cfg == NULL)
1014                 RETURN(-ENOMEM);
1015
1016         try_module_get(THIS_MODULE);
1017
1018         /* client additional sb info */
1019         lsi->lsi_llsbi = sbi = ll_init_sbi();
1020         if (!sbi) {
1021                 module_put(THIS_MODULE);
1022                 OBD_FREE_PTR(cfg);
1023                 RETURN(-ENOMEM);
1024         }
1025
1026         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1027         if (err)
1028                 GOTO(out_free, err);
1029
1030         err = bdi_init(&lsi->lsi_bdi);
1031         if (err)
1032                 GOTO(out_free, err);
1033         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1034         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1035         err = ll_bdi_register(&lsi->lsi_bdi);
1036         if (err)
1037                 GOTO(out_free, err);
1038
1039         sb->s_bdi = &lsi->lsi_bdi;
1040 #ifndef HAVE_DCACHE_LOCK
1041         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1042         sb->s_d_op = &ll_d_ops;
1043 #endif
1044
1045         /* Generate a string unique to this super, in case some joker tries
1046            to mount the same fs at two mount points.
1047            Use the address of the super itself.*/
1048         cfg->cfg_instance = sb;
1049         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1050         cfg->cfg_callback = class_config_llog_handler;
1051         /* set up client obds */
1052         err = lustre_process_log(sb, profilenm, cfg);
1053         if (err < 0)
1054                 GOTO(out_free, err);
1055
1056         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1057         lprof = class_get_profile(profilenm);
1058         if (lprof == NULL) {
1059                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1060                                    " read from the MGS.  Does that filesystem "
1061                                    "exist?\n", profilenm);
1062                 GOTO(out_free, err = -EINVAL);
1063         }
1064         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1065                lprof->lp_md, lprof->lp_dt);
1066
1067         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1068         if (!dt)
1069                 GOTO(out_free, err = -ENOMEM);
1070         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1071
1072         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1073         if (!md)
1074                 GOTO(out_free, err = -ENOMEM);
1075         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1076
1077         /* connections, registrations, sb setup */
1078         err = client_common_fill_super(sb, md, dt, mnt);
1079
1080 out_free:
1081         if (md)
1082                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1083         if (dt)
1084                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1085         if (err)
1086                 ll_put_super(sb);
1087         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1088                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1089
1090         OBD_FREE_PTR(cfg);
1091         RETURN(err);
1092 } /* ll_fill_super */
1093
1094 void ll_put_super(struct super_block *sb)
1095 {
1096         struct config_llog_instance cfg, params_cfg;
1097         struct obd_device *obd;
1098         struct lustre_sb_info *lsi = s2lsi(sb);
1099         struct ll_sb_info *sbi = ll_s2sbi(sb);
1100         char *profilenm = get_profile_name(sb);
1101         long ccc_count;
1102         int next, force = 1, rc = 0;
1103         ENTRY;
1104
1105         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1106
1107         ll_print_capa_stat(sbi);
1108
1109         cfg.cfg_instance = sb;
1110         lustre_end_log(sb, profilenm, &cfg);
1111
1112         params_cfg.cfg_instance = sb;
1113         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1114
1115         if (sbi->ll_md_exp) {
1116                 obd = class_exp2obd(sbi->ll_md_exp);
1117                 if (obd)
1118                         force = obd->obd_force;
1119         }
1120
1121         /* Wait for unstable pages to be committed to stable storage */
1122         if (force == 0) {
1123                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1124                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1125                         atomic_long_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1126                         &lwi);
1127         }
1128
1129         ccc_count = atomic_long_read(&sbi->ll_cache.ccc_unstable_nr);
1130         if (force == 0 && rc != -EINTR)
1131                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1132
1133
1134         /* We need to set force before the lov_disconnect in
1135            lustre_common_put_super, since l_d cleans up osc's as well. */
1136         if (force) {
1137                 next = 0;
1138                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1139                                                      &next)) != NULL) {
1140                         obd->obd_force = force;
1141                 }
1142         }
1143
1144         if (sbi->ll_lcq) {
1145                 /* Only if client_common_fill_super succeeded */
1146                 client_common_put_super(sb);
1147         }
1148
1149         next = 0;
1150         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1151                 class_manual_cleanup(obd);
1152         }
1153
1154         if (sbi->ll_flags & LL_SBI_VERBOSE)
1155                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1156
1157         if (profilenm)
1158                 class_del_profile(profilenm);
1159
1160         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1161                 bdi_destroy(&lsi->lsi_bdi);
1162                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1163         }
1164
1165         ll_free_sbi(sb);
1166         lsi->lsi_llsbi = NULL;
1167
1168         lustre_common_put_super(sb);
1169
1170         cl_env_cache_purge(~0);
1171
1172         module_put(THIS_MODULE);
1173
1174         EXIT;
1175 } /* client_put_super */
1176
1177 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1178 {
1179         struct inode *inode = NULL;
1180
1181         /* NOTE: we depend on atomic igrab() -bzzz */
1182         lock_res_and_lock(lock);
1183         if (lock->l_resource->lr_lvb_inode) {
1184                 struct ll_inode_info * lli;
1185                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1186                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1187                         inode = igrab(lock->l_resource->lr_lvb_inode);
1188                 } else {
1189                         inode = lock->l_resource->lr_lvb_inode;
1190                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1191                                          D_WARNING, lock, "lr_lvb_inode %p is "
1192                                          "bogus: magic %08x",
1193                                          lock->l_resource->lr_lvb_inode,
1194                                          lli->lli_inode_magic);
1195                         inode = NULL;
1196                 }
1197         }
1198         unlock_res_and_lock(lock);
1199         return inode;
1200 }
1201
1202 static void ll_dir_clear_lsm_md(struct inode *inode)
1203 {
1204         struct ll_inode_info *lli = ll_i2info(inode);
1205
1206         LASSERT(S_ISDIR(inode->i_mode));
1207
1208         if (lli->lli_lsm_md != NULL) {
1209                 lmv_free_memmd(lli->lli_lsm_md);
1210                 lli->lli_lsm_md = NULL;
1211         }
1212 }
1213
1214 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1215                                       const struct lu_fid *fid,
1216                                       struct lustre_md *md)
1217 {
1218         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1219         struct mdt_body         *body = md->body;
1220         struct inode            *inode;
1221         ino_t                   ino;
1222         ENTRY;
1223
1224         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1225         inode = iget_locked(sb, ino);
1226         if (inode == NULL) {
1227                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1228                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1229                 RETURN(ERR_PTR(-ENOENT));
1230         }
1231
1232         if (inode->i_state & I_NEW) {
1233                 struct ll_inode_info *lli = ll_i2info(inode);
1234                 struct lmv_stripe_md *lsm = md->lmv;
1235
1236                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1237                                 (body->mbo_mode & S_IFMT);
1238                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1239                          PFID(fid));
1240
1241                 LTIME_S(inode->i_mtime) = 0;
1242                 LTIME_S(inode->i_atime) = 0;
1243                 LTIME_S(inode->i_ctime) = 0;
1244                 inode->i_rdev = 0;
1245
1246                 /* initializing backing dev info. */
1247                 inode->i_mapping->backing_dev_info =
1248                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1249                 inode->i_op = &ll_dir_inode_operations;
1250                 inode->i_fop = &ll_dir_operations;
1251                 lli->lli_fid = *fid;
1252                 ll_lli_init(lli);
1253
1254                 LASSERT(lsm != NULL);
1255                 /* master object FID */
1256                 lli->lli_pfid = body->mbo_fid1;
1257                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1258                        lli, PFID(fid), PFID(&lli->lli_pfid));
1259                 unlock_new_inode(inode);
1260         }
1261
1262         RETURN(inode);
1263 }
1264
1265 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1266 {
1267         struct lu_fid *fid;
1268         struct lmv_stripe_md *lsm = md->lmv;
1269         int i;
1270
1271         LASSERT(lsm != NULL);
1272         /* XXX sigh, this lsm_root initialization should be in
1273          * LMV layer, but it needs ll_iget right now, so we
1274          * put this here right now. */
1275         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1276                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1277                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1278                 /* Unfortunately ll_iget will call ll_update_inode,
1279                  * where the initialization of slave inode is slightly
1280                  * different, so it reset lsm_md to NULL to avoid
1281                  * initializing lsm for slave inode. */
1282                 /* For migrating inode, master stripe and master object will
1283                  * be same, so we only need assign this inode */
1284                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1285                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1286                 else
1287                         lsm->lsm_md_oinfo[i].lmo_root =
1288                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1289
1290                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1291                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1292
1293                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1294                         return rc;
1295                 }
1296         }
1297
1298         return 0;
1299 }
1300
1301 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1302                                 const struct lmv_stripe_md *lsm_md2)
1303 {
1304         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1305                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1306                lsm_md1->lsm_md_master_mdt_index ==
1307                                         lsm_md2->lsm_md_master_mdt_index &&
1308                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1309                lsm_md1->lsm_md_layout_version ==
1310                                         lsm_md2->lsm_md_layout_version &&
1311                strcmp(lsm_md1->lsm_md_pool_name,
1312                       lsm_md2->lsm_md_pool_name) == 0;
1313 }
1314
1315 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1316 {
1317         struct ll_inode_info *lli = ll_i2info(inode);
1318         struct lmv_stripe_md *lsm = md->lmv;
1319         int     rc;
1320         ENTRY;
1321
1322         LASSERT(S_ISDIR(inode->i_mode));
1323         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1324                PFID(ll_inode2fid(inode)));
1325
1326         /* no striped information from request. */
1327         if (lsm == NULL) {
1328                 if (lli->lli_lsm_md == NULL) {
1329                         RETURN(0);
1330                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1331                                                 LMV_HASH_FLAG_MIGRATION) {
1332                         /* migration is done, the temporay MIGRATE layout has
1333                          * been removed */
1334                         CDEBUG(D_INODE, DFID" finish migration.\n",
1335                                PFID(ll_inode2fid(inode)));
1336                         lmv_free_memmd(lli->lli_lsm_md);
1337                         lli->lli_lsm_md = NULL;
1338                         RETURN(0);
1339                 } else {
1340                         /* The lustre_md from req does not include stripeEA,
1341                          * see ll_md_setattr */
1342                         RETURN(0);
1343                 }
1344         }
1345
1346         /* set the directory layout */
1347         if (lli->lli_lsm_md == NULL) {
1348
1349                 rc = ll_init_lsm_md(inode, md);
1350                 if (rc != 0)
1351                         RETURN(rc);
1352
1353                 lli->lli_lsm_md = lsm;
1354                 /* set lsm_md to NULL, so the following free lustre_md
1355                  * will not free this lsm */
1356                 md->lmv = NULL;
1357                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1358                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1359                 RETURN(0);
1360         }
1361
1362         /* Compare the old and new stripe information */
1363         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1364                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1365                 int                     idx;
1366
1367                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1368                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1369                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1370                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1371                        inode, lsm, old_lsm,
1372                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1373                        lsm->lsm_md_stripe_count,
1374                        old_lsm->lsm_md_stripe_count,
1375                        lsm->lsm_md_master_mdt_index,
1376                        old_lsm->lsm_md_master_mdt_index,
1377                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1378                        lsm->lsm_md_layout_version,
1379                        old_lsm->lsm_md_layout_version,
1380                        lsm->lsm_md_pool_name,
1381                        old_lsm->lsm_md_pool_name);
1382
1383                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1384                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1385                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1386                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1387                 }
1388
1389                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1390                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1391                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1392                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1393                 }
1394
1395                 RETURN(-EIO);
1396         }
1397
1398         RETURN(0);
1399 }
1400
1401 void ll_clear_inode(struct inode *inode)
1402 {
1403         struct ll_inode_info *lli = ll_i2info(inode);
1404         struct ll_sb_info *sbi = ll_i2sbi(inode);
1405         ENTRY;
1406
1407         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1408                PFID(ll_inode2fid(inode)), inode);
1409
1410         if (S_ISDIR(inode->i_mode)) {
1411                 /* these should have been cleared in ll_file_release */
1412                 LASSERT(lli->lli_opendir_key == NULL);
1413                 LASSERT(lli->lli_sai == NULL);
1414                 LASSERT(lli->lli_opendir_pid == 0);
1415         }
1416
1417         spin_lock(&lli->lli_lock);
1418         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1419         spin_unlock(&lli->lli_lock);
1420         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1421
1422         LASSERT(!lli->lli_open_fd_write_count);
1423         LASSERT(!lli->lli_open_fd_read_count);
1424         LASSERT(!lli->lli_open_fd_exec_count);
1425
1426         if (lli->lli_mds_write_och)
1427                 ll_md_real_close(inode, FMODE_WRITE);
1428         if (lli->lli_mds_exec_och)
1429                 ll_md_real_close(inode, FMODE_EXEC);
1430         if (lli->lli_mds_read_och)
1431                 ll_md_real_close(inode, FMODE_READ);
1432
1433         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1434                 OBD_FREE(lli->lli_symlink_name,
1435                          strlen(lli->lli_symlink_name) + 1);
1436                 lli->lli_symlink_name = NULL;
1437         }
1438
1439         ll_xattr_cache_destroy(inode);
1440
1441         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1442                 LASSERT(lli->lli_posix_acl == NULL);
1443                 if (lli->lli_remote_perms) {
1444                         free_rmtperm_hash(lli->lli_remote_perms);
1445                         lli->lli_remote_perms = NULL;
1446                 }
1447         }
1448 #ifdef CONFIG_FS_POSIX_ACL
1449         else if (lli->lli_posix_acl) {
1450                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1451                 LASSERT(lli->lli_remote_perms == NULL);
1452                 posix_acl_release(lli->lli_posix_acl);
1453                 lli->lli_posix_acl = NULL;
1454         }
1455 #endif
1456         lli->lli_inode_magic = LLI_INODE_DEAD;
1457
1458         ll_clear_inode_capas(inode);
1459         if (S_ISDIR(inode->i_mode))
1460                 ll_dir_clear_lsm_md(inode);
1461         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1462                 LASSERT(list_empty(&lli->lli_agl_list));
1463
1464         /*
1465          * XXX This has to be done before lsm is freed below, because
1466          * cl_object still uses inode lsm.
1467          */
1468         cl_inode_fini(inode);
1469         lli->lli_has_smd = false;
1470
1471         EXIT;
1472 }
1473
1474 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1475                          struct md_open_data **mod)
1476 {
1477         struct lustre_md md;
1478         struct inode *inode = dentry->d_inode;
1479         struct ll_sb_info *sbi = ll_i2sbi(inode);
1480         struct ptlrpc_request *request = NULL;
1481         int rc, ia_valid;
1482         ENTRY;
1483
1484         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1485                                      LUSTRE_OPC_ANY, NULL);
1486         if (IS_ERR(op_data))
1487                 RETURN(PTR_ERR(op_data));
1488
1489         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1490                         &request, mod);
1491         if (rc) {
1492                 ptlrpc_req_finished(request);
1493                 if (rc == -ENOENT) {
1494                         clear_nlink(inode);
1495                         /* Unlinked special device node? Or just a race?
1496                          * Pretend we done everything. */
1497                         if (!S_ISREG(inode->i_mode) &&
1498                             !S_ISDIR(inode->i_mode)) {
1499                                 ia_valid = op_data->op_attr.ia_valid;
1500                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1501                                 rc = simple_setattr(dentry, &op_data->op_attr);
1502                                 op_data->op_attr.ia_valid = ia_valid;
1503                         }
1504                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1505                         CERROR("md_setattr fails: rc = %d\n", rc);
1506                 }
1507                 RETURN(rc);
1508         }
1509
1510         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1511                               sbi->ll_md_exp, &md);
1512         if (rc) {
1513                 ptlrpc_req_finished(request);
1514                 RETURN(rc);
1515         }
1516
1517         ia_valid = op_data->op_attr.ia_valid;
1518         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1519          * cache is not cleared yet. */
1520         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1521         rc = simple_setattr(dentry, &op_data->op_attr);
1522         op_data->op_attr.ia_valid = ia_valid;
1523
1524         /* Extract epoch data if obtained. */
1525         op_data->op_handle = md.body->mbo_handle;
1526         op_data->op_ioepoch = md.body->mbo_ioepoch;
1527
1528         rc = ll_update_inode(inode, &md);
1529         ptlrpc_req_finished(request);
1530
1531         RETURN(rc);
1532 }
1533
1534 /* Close IO epoch and send Size-on-MDS attribute update. */
1535 static int ll_setattr_done_writing(struct inode *inode,
1536                                    struct md_op_data *op_data,
1537                                    struct md_open_data *mod)
1538 {
1539         struct ll_inode_info *lli = ll_i2info(inode);
1540         int rc = 0;
1541         ENTRY;
1542
1543         LASSERT(op_data != NULL);
1544         if (!S_ISREG(inode->i_mode))
1545                 RETURN(0);
1546
1547         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1548                op_data->op_ioepoch, PFID(&lli->lli_fid));
1549
1550         op_data->op_flags = MF_EPOCH_CLOSE;
1551         ll_done_writing_attr(inode, op_data);
1552         ll_pack_inode2opdata(inode, op_data, NULL);
1553
1554         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1555         if (rc == -EAGAIN) {
1556                 /* MDS has instructed us to obtain Size-on-MDS attribute
1557                  * from OSTs and send setattr to back to MDS. */
1558                 rc = ll_som_update(inode, op_data);
1559         } else if (rc) {
1560                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1561                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1562                        PFID(ll_inode2fid(inode)), rc);
1563         }
1564         RETURN(rc);
1565 }
1566
1567 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1568 {
1569         struct obd_capa *capa;
1570         int rc;
1571
1572         if (attr->ia_valid & ATTR_SIZE)
1573                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1574         else
1575                 capa = ll_mdscapa_get(inode);
1576
1577         rc = cl_setattr_ost(inode, attr, capa);
1578
1579         if (attr->ia_valid & ATTR_SIZE)
1580                 ll_truncate_free_capa(capa);
1581         else
1582                 capa_put(capa);
1583
1584         return rc;
1585 }
1586
1587 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1588  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1589  * keep these values until such a time that objects are allocated for it.
1590  * We do the MDS operations first, as it is checking permissions for us.
1591  * We don't to the MDS RPC if there is nothing that we want to store there,
1592  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1593  * going to do an RPC anyways.
1594  *
1595  * If we are doing a truncate, we will send the mtime and ctime updates
1596  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1597  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1598  * at the same time.
1599  *
1600  * In case of HSMimport, we only set attr on MDS.
1601  */
1602 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1603 {
1604         struct inode *inode = dentry->d_inode;
1605         struct ll_inode_info *lli = ll_i2info(inode);
1606         struct md_op_data *op_data = NULL;
1607         struct md_open_data *mod = NULL;
1608         bool file_is_released = false;
1609         int rc = 0, rc1 = 0;
1610         ENTRY;
1611
1612         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1613                "valid %x, hsm_import %d\n",
1614                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1615                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1616                hsm_import);
1617
1618         if (attr->ia_valid & ATTR_SIZE) {
1619                 /* Check new size against VFS/VM file size limit and rlimit */
1620                 rc = inode_newsize_ok(inode, attr->ia_size);
1621                 if (rc)
1622                         RETURN(rc);
1623
1624                 /* The maximum Lustre file size is variable, based on the
1625                  * OST maximum object size and number of stripes.  This
1626                  * needs another check in addition to the VFS check above. */
1627                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1628                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1629                                PFID(&lli->lli_fid), attr->ia_size,
1630                                ll_file_maxbytes(inode));
1631                         RETURN(-EFBIG);
1632                 }
1633
1634                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1635         }
1636
1637         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1638         if (attr->ia_valid & TIMES_SET_FLAGS) {
1639                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1640                     !cfs_capable(CFS_CAP_FOWNER))
1641                         RETURN(-EPERM);
1642         }
1643
1644         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1645         if (attr->ia_valid & ATTR_CTIME) {
1646                 attr->ia_ctime = CFS_CURRENT_TIME;
1647                 attr->ia_valid |= ATTR_CTIME_SET;
1648         }
1649         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1650             (attr->ia_valid & ATTR_ATIME)) {
1651                 attr->ia_atime = CFS_CURRENT_TIME;
1652                 attr->ia_valid |= ATTR_ATIME_SET;
1653         }
1654         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1655             (attr->ia_valid & ATTR_MTIME)) {
1656                 attr->ia_mtime = CFS_CURRENT_TIME;
1657                 attr->ia_valid |= ATTR_MTIME_SET;
1658         }
1659
1660         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1661                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1662                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1663                        cfs_time_current_sec());
1664
1665         /* We always do an MDS RPC, even if we're only changing the size;
1666          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1667
1668         OBD_ALLOC_PTR(op_data);
1669         if (op_data == NULL)
1670                 RETURN(-ENOMEM);
1671
1672         if (!S_ISDIR(inode->i_mode)) {
1673                 if (attr->ia_valid & ATTR_SIZE)
1674                         inode_dio_write_done(inode);
1675                 mutex_unlock(&inode->i_mutex);
1676         }
1677
1678         /* truncate on a released file must failed with -ENODATA,
1679          * so size must not be set on MDS for released file
1680          * but other attributes must be set
1681          */
1682         if (S_ISREG(inode->i_mode)) {
1683                 struct lov_stripe_md *lsm;
1684                 __u32 gen;
1685
1686                 ll_layout_refresh(inode, &gen);
1687                 lsm = ccc_inode_lsm_get(inode);
1688                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1689                         file_is_released = true;
1690                 ccc_inode_lsm_put(inode, lsm);
1691
1692                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1693                         if (file_is_released) {
1694                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1695                                 if (rc < 0)
1696                                         GOTO(out, rc);
1697
1698                                 file_is_released = false;
1699                                 ll_layout_refresh(inode, &gen);
1700                         }
1701
1702                         /* If we are changing file size, file content is
1703                          * modified, flag it. */
1704                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1705                         spin_lock(&lli->lli_lock);
1706                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1707                         spin_unlock(&lli->lli_lock);
1708                         op_data->op_bias |= MDS_DATA_MODIFIED;
1709                 }
1710         }
1711
1712         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1713
1714         /* Open epoch for truncate. */
1715         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1716             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1717                 op_data->op_flags = MF_EPOCH_OPEN;
1718
1719         rc = ll_md_setattr(dentry, op_data, &mod);
1720         if (rc)
1721                 GOTO(out, rc);
1722
1723         /* RPC to MDT is sent, cancel data modification flag */
1724         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1725                 spin_lock(&lli->lli_lock);
1726                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1727                 spin_unlock(&lli->lli_lock);
1728         }
1729
1730         ll_ioepoch_open(lli, op_data->op_ioepoch);
1731         if (!S_ISREG(inode->i_mode) || file_is_released)
1732                 GOTO(out, rc = 0);
1733
1734         if (attr->ia_valid & (ATTR_SIZE |
1735                               ATTR_ATIME | ATTR_ATIME_SET |
1736                               ATTR_MTIME | ATTR_MTIME_SET)) {
1737                 /* For truncate and utimes sending attributes to OSTs, setting
1738                  * mtime/atime to the past will be performed under PW [0:EOF]
1739                  * extent lock (new_size:EOF for truncate).  It may seem
1740                  * excessive to send mtime/atime updates to OSTs when not
1741                  * setting times to past, but it is necessary due to possible
1742                  * time de-synchronization between MDT inode and OST objects */
1743                 if (attr->ia_valid & ATTR_SIZE)
1744                         down_write(&lli->lli_trunc_sem);
1745                 rc = ll_setattr_ost(inode, attr);
1746                 if (attr->ia_valid & ATTR_SIZE)
1747                         up_write(&lli->lli_trunc_sem);
1748         }
1749         EXIT;
1750 out:
1751         if (op_data) {
1752                 if (op_data->op_ioepoch) {
1753                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1754                         if (!rc)
1755                                 rc = rc1;
1756                 }
1757                 ll_finish_md_op_data(op_data);
1758         }
1759         if (!S_ISDIR(inode->i_mode)) {
1760                 mutex_lock(&inode->i_mutex);
1761                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1762                         inode_dio_wait(inode);
1763         }
1764
1765         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1766                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1767
1768         return rc;
1769 }
1770
1771 int ll_setattr(struct dentry *de, struct iattr *attr)
1772 {
1773         int mode = de->d_inode->i_mode;
1774
1775         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1776                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1777                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1778
1779         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1780                                (ATTR_SIZE|ATTR_MODE)) &&
1781             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1782              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1783               !(attr->ia_mode & S_ISGID))))
1784                 attr->ia_valid |= ATTR_FORCE;
1785
1786         if ((attr->ia_valid & ATTR_MODE) &&
1787             (mode & S_ISUID) &&
1788             !(attr->ia_mode & S_ISUID) &&
1789             !(attr->ia_valid & ATTR_KILL_SUID))
1790                 attr->ia_valid |= ATTR_KILL_SUID;
1791
1792         if ((attr->ia_valid & ATTR_MODE) &&
1793             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1794             !(attr->ia_mode & S_ISGID) &&
1795             !(attr->ia_valid & ATTR_KILL_SGID))
1796                 attr->ia_valid |= ATTR_KILL_SGID;
1797
1798         return ll_setattr_raw(de, attr, false);
1799 }
1800
1801 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1802                        __u64 max_age, __u32 flags)
1803 {
1804         struct ll_sb_info *sbi = ll_s2sbi(sb);
1805         struct obd_statfs obd_osfs;
1806         int rc;
1807         ENTRY;
1808
1809         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1810         if (rc) {
1811                 CERROR("md_statfs fails: rc = %d\n", rc);
1812                 RETURN(rc);
1813         }
1814
1815         osfs->os_type = sb->s_magic;
1816
1817         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1818                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1819
1820         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1821                 flags |= OBD_STATFS_NODELAY;
1822
1823         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1824         if (rc) {
1825                 CERROR("obd_statfs fails: rc = %d\n", rc);
1826                 RETURN(rc);
1827         }
1828
1829         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1830                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1831                obd_osfs.os_files);
1832
1833         osfs->os_bsize = obd_osfs.os_bsize;
1834         osfs->os_blocks = obd_osfs.os_blocks;
1835         osfs->os_bfree = obd_osfs.os_bfree;
1836         osfs->os_bavail = obd_osfs.os_bavail;
1837
1838         /* If we don't have as many objects free on the OST as inodes
1839          * on the MDS, we reduce the total number of inodes to
1840          * compensate, so that the "inodes in use" number is correct.
1841          */
1842         if (obd_osfs.os_ffree < osfs->os_ffree) {
1843                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1844                         obd_osfs.os_ffree;
1845                 osfs->os_ffree = obd_osfs.os_ffree;
1846         }
1847
1848         RETURN(rc);
1849 }
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1851 {
1852         struct super_block *sb = de->d_sb;
1853         struct obd_statfs osfs;
1854         __u64 fsid = huge_encode_dev(sb->s_dev);
1855         int rc;
1856
1857         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1858         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1859
1860         /* Some amount of caching on the client is allowed */
1861         rc = ll_statfs_internal(sb, &osfs,
1862                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1863                                 0);
1864         if (rc)
1865                 return rc;
1866
1867         statfs_unpack(sfs, &osfs);
1868
1869         /* We need to downshift for all 32-bit kernels, because we can't
1870          * tell if the kernel is being called via sys_statfs64() or not.
1871          * Stop before overflowing f_bsize - in which case it is better
1872          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1873         if (sizeof(long) < 8) {
1874                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1875                         sfs->f_bsize <<= 1;
1876
1877                         osfs.os_blocks >>= 1;
1878                         osfs.os_bfree >>= 1;
1879                         osfs.os_bavail >>= 1;
1880                 }
1881         }
1882
1883         sfs->f_blocks = osfs.os_blocks;
1884         sfs->f_bfree = osfs.os_bfree;
1885         sfs->f_bavail = osfs.os_bavail;
1886         sfs->f_fsid.val[0] = (__u32)fsid;
1887         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1888         return 0;
1889 }
1890
1891 void ll_inode_size_lock(struct inode *inode)
1892 {
1893         struct ll_inode_info *lli;
1894
1895         LASSERT(!S_ISDIR(inode->i_mode));
1896
1897         lli = ll_i2info(inode);
1898         mutex_lock(&lli->lli_size_mutex);
1899 }
1900
1901 void ll_inode_size_unlock(struct inode *inode)
1902 {
1903         struct ll_inode_info *lli;
1904
1905         lli = ll_i2info(inode);
1906         mutex_unlock(&lli->lli_size_mutex);
1907 }
1908
1909 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1910 {
1911         struct ll_inode_info *lli = ll_i2info(inode);
1912         struct mdt_body *body = md->body;
1913         struct lov_stripe_md *lsm = md->lsm;
1914         struct ll_sb_info *sbi = ll_i2sbi(inode);
1915
1916         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1917         if (lsm != NULL) {
1918                 if (!lli->lli_has_smd &&
1919                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1920                         cl_file_inode_init(inode, md);
1921
1922                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1923                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1924                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1925         }
1926
1927         if (S_ISDIR(inode->i_mode)) {
1928                 int     rc;
1929
1930                 rc = ll_update_lsm_md(inode, md);
1931                 if (rc != 0)
1932                         return rc;
1933         }
1934
1935         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1936                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1937                         ll_update_remote_perm(inode, md->remote_perm);
1938         }
1939 #ifdef CONFIG_FS_POSIX_ACL
1940         else if (body->mbo_valid & OBD_MD_FLACL) {
1941                 spin_lock(&lli->lli_lock);
1942                 if (lli->lli_posix_acl)
1943                         posix_acl_release(lli->lli_posix_acl);
1944                 lli->lli_posix_acl = md->posix_acl;
1945                 spin_unlock(&lli->lli_lock);
1946         }
1947 #endif
1948         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1949                                         sbi->ll_flags & LL_SBI_32BIT_API);
1950         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1951
1952         if (body->mbo_valid & OBD_MD_FLATIME) {
1953                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1954                         LTIME_S(inode->i_atime) = body->mbo_atime;
1955                 lli->lli_atime = body->mbo_atime;
1956         }
1957
1958         if (body->mbo_valid & OBD_MD_FLMTIME) {
1959                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1960                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1961                                "to "LPU64"\n", inode->i_ino,
1962                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1963                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1964                 }
1965                 lli->lli_mtime = body->mbo_mtime;
1966         }
1967
1968         if (body->mbo_valid & OBD_MD_FLCTIME) {
1969                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1970                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1971                 lli->lli_ctime = body->mbo_ctime;
1972         }
1973
1974         if (body->mbo_valid & OBD_MD_FLMODE)
1975                 inode->i_mode = (inode->i_mode & S_IFMT) |
1976                                 (body->mbo_mode & ~S_IFMT);
1977
1978         if (body->mbo_valid & OBD_MD_FLTYPE)
1979                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1980                                 (body->mbo_mode & S_IFMT);
1981
1982         LASSERT(inode->i_mode != 0);
1983         if (S_ISREG(inode->i_mode))
1984                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1985                                        LL_MAX_BLKSIZE_BITS);
1986         else
1987                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1988
1989         if (body->mbo_valid & OBD_MD_FLUID)
1990                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1991         if (body->mbo_valid & OBD_MD_FLGID)
1992                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1993         if (body->mbo_valid & OBD_MD_FLFLAGS)
1994                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1995         if (body->mbo_valid & OBD_MD_FLNLINK)
1996                 set_nlink(inode, body->mbo_nlink);
1997         if (body->mbo_valid & OBD_MD_FLRDEV)
1998                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1999
2000         if (body->mbo_valid & OBD_MD_FLID) {
2001                 /* FID shouldn't be changed! */
2002                 if (fid_is_sane(&lli->lli_fid)) {
2003                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2004                                  "Trying to change FID "DFID
2005                                  " to the "DFID", inode "DFID"(%p)\n",
2006                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2007                                  PFID(ll_inode2fid(inode)), inode);
2008                 } else {
2009                         lli->lli_fid = body->mbo_fid1;
2010                 }
2011         }
2012
2013         LASSERT(fid_seq(&lli->lli_fid) != 0);
2014
2015         if (body->mbo_valid & OBD_MD_FLSIZE) {
2016                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2017                     S_ISREG(inode->i_mode)) {
2018                         struct lustre_handle lockh;
2019                         ldlm_mode_t mode;
2020
2021                         /* As it is possible a blocking ast has been processed
2022                          * by this time, we need to check there is an UPDATE
2023                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2024                          * it. */
2025                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2026                                                &lockh, LDLM_FL_CBPENDING,
2027                                                LCK_CR | LCK_CW |
2028                                                LCK_PR | LCK_PW);
2029                         if (mode) {
2030                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2031                                                       LLIF_EPOCH_PENDING |
2032                                                       LLIF_SOM_DIRTY)) {
2033                                         CERROR("%s: inode "DFID" flags %u still"
2034                                                " has size authority! do not "
2035                                                "trust the size from MDS\n",
2036                                                sbi->ll_md_exp->exp_obd->obd_name,
2037                                                PFID(ll_inode2fid(inode)),
2038                                                lli->lli_flags);
2039                                 } else {
2040                                         /* Use old size assignment to avoid
2041                                          * deadlock bz14138 & bz14326 */
2042                                         i_size_write(inode, body->mbo_size);
2043                                         spin_lock(&lli->lli_lock);
2044                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2045                                         spin_unlock(&lli->lli_lock);
2046                                 }
2047                                 ldlm_lock_decref(&lockh, mode);
2048                         }
2049                 } else {
2050                         /* Use old size assignment to avoid
2051                          * deadlock bz14138 & bz14326 */
2052                         i_size_write(inode, body->mbo_size);
2053
2054                         CDEBUG(D_VFSTRACE,
2055                                "inode="DFID", updating i_size %llu\n",
2056                                PFID(ll_inode2fid(inode)),
2057                                (unsigned long long)body->mbo_size);
2058                 }
2059
2060                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2061                         inode->i_blocks = body->mbo_blocks;
2062         }
2063
2064         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2065                 LASSERT(md->mds_capa);
2066                 ll_add_capa(inode, md->mds_capa);
2067         }
2068
2069         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2070                 LASSERT(md->oss_capa);
2071                 ll_add_capa(inode, md->oss_capa);
2072         }
2073
2074         if (body->mbo_valid & OBD_MD_TSTATE) {
2075                 if (body->mbo_t_state & MS_RESTORE)
2076                         lli->lli_flags |= LLIF_FILE_RESTORING;
2077         }
2078
2079         return 0;
2080 }
2081
2082 int ll_read_inode2(struct inode *inode, void *opaque)
2083 {
2084         struct lustre_md *md = opaque;
2085         struct ll_inode_info *lli = ll_i2info(inode);
2086         int     rc;
2087         ENTRY;
2088
2089         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2090                PFID(&lli->lli_fid), inode);
2091
2092         LASSERT(!lli->lli_has_smd);
2093
2094         /* Core attributes from the MDS first.  This is a new inode, and
2095          * the VFS doesn't zero times in the core inode so we have to do
2096          * it ourselves.  They will be overwritten by either MDS or OST
2097          * attributes - we just need to make sure they aren't newer. */
2098         LTIME_S(inode->i_mtime) = 0;
2099         LTIME_S(inode->i_atime) = 0;
2100         LTIME_S(inode->i_ctime) = 0;
2101         inode->i_rdev = 0;
2102         rc = ll_update_inode(inode, md);
2103         if (rc != 0)
2104                 RETURN(rc);
2105
2106         /* OIDEBUG(inode); */
2107
2108         /* initializing backing dev info. */
2109         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2110
2111
2112         if (S_ISREG(inode->i_mode)) {
2113                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2114                 inode->i_op = &ll_file_inode_operations;
2115                 inode->i_fop = sbi->ll_fop;
2116                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2117                 EXIT;
2118         } else if (S_ISDIR(inode->i_mode)) {
2119                 inode->i_op = &ll_dir_inode_operations;
2120                 inode->i_fop = &ll_dir_operations;
2121                 EXIT;
2122         } else if (S_ISLNK(inode->i_mode)) {
2123                 inode->i_op = &ll_fast_symlink_inode_operations;
2124                 EXIT;
2125         } else {
2126                 inode->i_op = &ll_special_inode_operations;
2127
2128                 init_special_inode(inode, inode->i_mode,
2129                                    inode->i_rdev);
2130
2131                 EXIT;
2132         }
2133
2134         return 0;
2135 }
2136
2137 void ll_delete_inode(struct inode *inode)
2138 {
2139         struct ll_inode_info *lli = ll_i2info(inode);
2140         ENTRY;
2141
2142         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2143                 /* It is last chance to write out dirty pages,
2144                  * otherwise we may lose data while umount */
2145                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2146
2147         truncate_inode_pages_final(&inode->i_data);
2148
2149         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2150                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2151                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2152
2153 #ifdef HAVE_SBOPS_EVICT_INODE
2154         ll_clear_inode(inode);
2155 #endif
2156         clear_inode(inode);
2157
2158         EXIT;
2159 }
2160
2161 int ll_iocontrol(struct inode *inode, struct file *file,
2162                  unsigned int cmd, unsigned long arg)
2163 {
2164         struct ll_sb_info *sbi = ll_i2sbi(inode);
2165         struct ptlrpc_request *req = NULL;
2166         int rc, flags = 0;
2167         ENTRY;
2168
2169         switch(cmd) {
2170         case FSFILT_IOC_GETFLAGS: {
2171                 struct mdt_body *body;
2172                 struct md_op_data *op_data;
2173
2174                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2175                                              0, 0, LUSTRE_OPC_ANY,
2176                                              NULL);
2177                 if (IS_ERR(op_data))
2178                         RETURN(PTR_ERR(op_data));
2179
2180                 op_data->op_valid = OBD_MD_FLFLAGS;
2181                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2182                 ll_finish_md_op_data(op_data);
2183                 if (rc) {
2184                         CERROR("%s: failure inode "DFID": rc = %d\n",
2185                                sbi->ll_md_exp->exp_obd->obd_name,
2186                                PFID(ll_inode2fid(inode)), rc);
2187                         RETURN(-abs(rc));
2188                 }
2189
2190                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2191
2192                 flags = body->mbo_flags;
2193
2194                 ptlrpc_req_finished(req);
2195
2196                 RETURN(put_user(flags, (int __user *)arg));
2197         }
2198         case FSFILT_IOC_SETFLAGS: {
2199                 struct lov_stripe_md *lsm;
2200                 struct obd_info oinfo = { { { 0 } } };
2201                 struct md_op_data *op_data;
2202
2203                 if (get_user(flags, (int __user *)arg))
2204                         RETURN(-EFAULT);
2205
2206                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2207                                              LUSTRE_OPC_ANY, NULL);
2208                 if (IS_ERR(op_data))
2209                         RETURN(PTR_ERR(op_data));
2210
2211                 op_data->op_attr_flags = flags;
2212                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2213                 rc = md_setattr(sbi->ll_md_exp, op_data,
2214                                 NULL, 0, NULL, 0, &req, NULL);
2215                 ll_finish_md_op_data(op_data);
2216                 ptlrpc_req_finished(req);
2217                 if (rc)
2218                         RETURN(rc);
2219
2220                 inode->i_flags = ll_ext_to_inode_flags(flags);
2221
2222                 lsm = ccc_inode_lsm_get(inode);
2223                 if (!lsm_has_objects(lsm)) {
2224                         ccc_inode_lsm_put(inode, lsm);
2225                         RETURN(0);
2226                 }
2227
2228                 OBDO_ALLOC(oinfo.oi_oa);
2229                 if (!oinfo.oi_oa) {
2230                         ccc_inode_lsm_put(inode, lsm);
2231                         RETURN(-ENOMEM);
2232                 }
2233                 oinfo.oi_md = lsm;
2234                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2235                 oinfo.oi_oa->o_flags = flags;
2236                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2237                                        OBD_MD_FLGROUP;
2238                 oinfo.oi_capa = ll_mdscapa_get(inode);
2239                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2240                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2241                 capa_put(oinfo.oi_capa);
2242                 OBDO_FREE(oinfo.oi_oa);
2243                 ccc_inode_lsm_put(inode, lsm);
2244
2245                 if (rc && rc != -EPERM && rc != -EACCES)
2246                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2247
2248                 RETURN(rc);
2249         }
2250         default:
2251                 RETURN(-ENOSYS);
2252         }
2253
2254         RETURN(0);
2255 }
2256
2257 int ll_flush_ctx(struct inode *inode)
2258 {
2259         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2260
2261         CDEBUG(D_SEC, "flush context for user %d\n",
2262                from_kuid(&init_user_ns, current_uid()));
2263
2264         obd_set_info_async(NULL, sbi->ll_md_exp,
2265                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2266                            0, NULL, NULL);
2267         obd_set_info_async(NULL, sbi->ll_dt_exp,
2268                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2269                            0, NULL, NULL);
2270         return 0;
2271 }
2272
2273 /* umount -f client means force down, don't save state */
2274 void ll_umount_begin(struct super_block *sb)
2275 {
2276         struct ll_sb_info *sbi = ll_s2sbi(sb);
2277         struct obd_device *obd;
2278         struct obd_ioctl_data *ioc_data;
2279         ENTRY;
2280
2281         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2282                sb->s_count, atomic_read(&sb->s_active));
2283
2284         obd = class_exp2obd(sbi->ll_md_exp);
2285         if (obd == NULL) {
2286                 CERROR("Invalid MDC connection handle "LPX64"\n",
2287                        sbi->ll_md_exp->exp_handle.h_cookie);
2288                 EXIT;
2289                 return;
2290         }
2291         obd->obd_force = 1;
2292
2293         obd = class_exp2obd(sbi->ll_dt_exp);
2294         if (obd == NULL) {
2295                 CERROR("Invalid LOV connection handle "LPX64"\n",
2296                        sbi->ll_dt_exp->exp_handle.h_cookie);
2297                 EXIT;
2298                 return;
2299         }
2300         obd->obd_force = 1;
2301
2302         OBD_ALLOC_PTR(ioc_data);
2303         if (ioc_data) {
2304                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2305                               sizeof *ioc_data, ioc_data, NULL);
2306
2307                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2308                               sizeof *ioc_data, ioc_data, NULL);
2309
2310                 OBD_FREE_PTR(ioc_data);
2311         }
2312
2313         /* Really, we'd like to wait until there are no requests outstanding,
2314          * and then continue.  For now, we just invalidate the requests,
2315          * schedule() and sleep one second if needed, and hope.
2316          */
2317         schedule();
2318         EXIT;
2319 }
2320
2321 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2322 {
2323         struct ll_sb_info *sbi = ll_s2sbi(sb);
2324         char *profilenm = get_profile_name(sb);
2325         int err;
2326         __u32 read_only;
2327
2328         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2329                 read_only = *flags & MS_RDONLY;
2330                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2331                                          sizeof(KEY_READ_ONLY),
2332                                          KEY_READ_ONLY, sizeof(read_only),
2333                                          &read_only, NULL);
2334                 if (err) {
2335                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2336                                       profilenm, read_only ?
2337                                       "read-only" : "read-write", err);
2338                         return err;
2339                 }
2340
2341                 if (read_only)
2342                         sb->s_flags |= MS_RDONLY;
2343                 else
2344                         sb->s_flags &= ~MS_RDONLY;
2345
2346                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2347                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2348                                       read_only ?  "read-only" : "read-write");
2349         }
2350         return 0;
2351 }
2352
2353 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2354                   struct super_block *sb, struct lookup_intent *it)
2355 {
2356         struct ll_sb_info *sbi = NULL;
2357         struct lustre_md md = { NULL };
2358         int rc;
2359         ENTRY;
2360
2361         LASSERT(*inode || sb);
2362         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2363         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2364                               sbi->ll_md_exp, &md);
2365         if (rc)
2366                 RETURN(rc);
2367
2368         if (*inode) {
2369                 rc = ll_update_inode(*inode, &md);
2370                 if (rc != 0)
2371                         GOTO(out, rc);
2372         } else {
2373                 LASSERT(sb != NULL);
2374
2375                 /*
2376                  * At this point server returns to client's same fid as client
2377                  * generated for creating. So using ->fid1 is okay here.
2378                  */
2379                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2380
2381                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2382                                              sbi->ll_flags & LL_SBI_32BIT_API),
2383                                  &md);
2384                 if (IS_ERR(*inode)) {
2385 #ifdef CONFIG_FS_POSIX_ACL
2386                         if (md.posix_acl) {
2387                                 posix_acl_release(md.posix_acl);
2388                                 md.posix_acl = NULL;
2389                         }
2390 #endif
2391                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2392                         *inode = NULL;
2393                         CERROR("new_inode -fatal: rc %d\n", rc);
2394                         GOTO(out, rc);
2395                 }
2396         }
2397
2398         /* Handling piggyback layout lock.
2399          * Layout lock can be piggybacked by getattr and open request.
2400          * The lsm can be applied to inode only if it comes with a layout lock
2401          * otherwise correct layout may be overwritten, for example:
2402          * 1. proc1: mdt returns a lsm but not granting layout
2403          * 2. layout was changed by another client
2404          * 3. proc2: refresh layout and layout lock granted
2405          * 4. proc1: to apply a stale layout */
2406         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2407                 struct lustre_handle lockh;
2408                 struct ldlm_lock *lock;
2409
2410                 lockh.cookie = it->d.lustre.it_lock_handle;
2411                 lock = ldlm_handle2lock(&lockh);
2412                 LASSERT(lock != NULL);
2413                 if (ldlm_has_layout(lock)) {
2414                         struct cl_object_conf conf;
2415
2416                         memset(&conf, 0, sizeof(conf));
2417                         conf.coc_opc = OBJECT_CONF_SET;
2418                         conf.coc_inode = *inode;
2419                         conf.coc_lock = lock;
2420                         conf.u.coc_md = &md;
2421                         (void)ll_layout_conf(*inode, &conf);
2422                 }
2423                 LDLM_LOCK_PUT(lock);
2424         }
2425
2426 out:
2427         if (md.lsm != NULL)
2428                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2429         md_free_lustre_md(sbi->ll_md_exp, &md);
2430         RETURN(rc);
2431 }
2432
2433 int ll_obd_statfs(struct inode *inode, void __user *arg)
2434 {
2435         struct ll_sb_info *sbi = NULL;
2436         struct obd_export *exp;
2437         char *buf = NULL;
2438         struct obd_ioctl_data *data = NULL;
2439         __u32 type;
2440         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2441         int len = 0, rc;
2442
2443         if (!inode || !(sbi = ll_i2sbi(inode)))
2444                 GOTO(out_statfs, rc = -EINVAL);
2445
2446         rc = obd_ioctl_getdata(&buf, &len, arg);
2447         if (rc)
2448                 GOTO(out_statfs, rc);
2449
2450         data = (void*)buf;
2451         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2452             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2453                 GOTO(out_statfs, rc = -EINVAL);
2454
2455         if (data->ioc_inllen1 != sizeof(__u32) ||
2456             data->ioc_inllen2 != sizeof(__u32) ||
2457             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2458             data->ioc_plen2 != sizeof(struct obd_uuid))
2459                 GOTO(out_statfs, rc = -EINVAL);
2460
2461         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2462         if (type & LL_STATFS_LMV)
2463                 exp = sbi->ll_md_exp;
2464         else if (type & LL_STATFS_LOV)
2465                 exp = sbi->ll_dt_exp;
2466         else
2467                 GOTO(out_statfs, rc = -ENODEV);
2468
2469         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2470         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2471         if (rc)
2472                 GOTO(out_statfs, rc);
2473 out_statfs:
2474         if (buf)
2475                 obd_ioctl_freedata(buf, len);
2476         return rc;
2477 }
2478
2479 int ll_process_config(struct lustre_cfg *lcfg)
2480 {
2481         struct super_block *sb;
2482         unsigned long x;
2483         int rc = 0;
2484         char *ptr;
2485
2486         /* The instance name contains the sb: lustre-client-aacfe000 */
2487         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2488         if (!ptr || !*(++ptr))
2489                 return -EINVAL;
2490         if (sscanf(ptr, "%lx", &x) != 1)
2491                 return -EINVAL;
2492         sb = (struct super_block *)x;
2493         /* This better be a real Lustre superblock! */
2494         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2495
2496         /* Note we have not called client_common_fill_super yet, so
2497            proc fns must be able to handle that! */
2498         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2499                                       lcfg, sb);
2500         if (rc > 0)
2501                 rc = 0;
2502         return rc;
2503 }
2504
2505 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2506 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2507                                        struct inode *i1, struct inode *i2,
2508                                        const char *name, size_t namelen,
2509                                        __u32 mode, __u32 opc, void *data)
2510 {
2511         LASSERT(i1 != NULL);
2512
2513         if (name == NULL) {
2514                 /* Do not reuse namelen for something else. */
2515                 if (namelen != 0)
2516                         return ERR_PTR(-EINVAL);
2517         } else {
2518                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2519                         return ERR_PTR(-ENAMETOOLONG);
2520
2521                 if (!lu_name_is_valid_2(name, namelen))
2522                         return ERR_PTR(-EINVAL);
2523         }
2524
2525         if (op_data == NULL)
2526                 OBD_ALLOC_PTR(op_data);
2527
2528         if (op_data == NULL)
2529                 return ERR_PTR(-ENOMEM);
2530
2531         ll_i2gids(op_data->op_suppgids, i1, i2);
2532         op_data->op_fid1 = *ll_inode2fid(i1);
2533         op_data->op_capa1 = ll_mdscapa_get(i1);
2534         if (S_ISDIR(i1->i_mode))
2535                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2536
2537         if (i2) {
2538                 op_data->op_fid2 = *ll_inode2fid(i2);
2539                 op_data->op_capa2 = ll_mdscapa_get(i2);
2540                 if (S_ISDIR(i2->i_mode))
2541                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2542         } else {
2543                 fid_zero(&op_data->op_fid2);
2544                 op_data->op_capa2 = NULL;
2545         }
2546
2547         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2548                 op_data->op_cli_flags |= CLI_HASH64;
2549
2550         if (ll_need_32bit_api(ll_i2sbi(i1)))
2551                 op_data->op_cli_flags |= CLI_API32;
2552
2553         op_data->op_name = name;
2554         op_data->op_namelen = namelen;
2555         op_data->op_mode = mode;
2556         op_data->op_mod_time = cfs_time_current_sec();
2557         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2558         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2559         op_data->op_cap = cfs_curproc_cap_pack();
2560         op_data->op_bias = 0;
2561         op_data->op_cli_flags = 0;
2562         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2563              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2564                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2565         } else {
2566                 op_data->op_mds = 0;
2567         }
2568         op_data->op_data = data;
2569
2570         /* When called by ll_setattr_raw, file is i1. */
2571         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2572                 op_data->op_bias |= MDS_DATA_MODIFIED;
2573
2574         return op_data;
2575 }
2576
2577 void ll_finish_md_op_data(struct md_op_data *op_data)
2578 {
2579         capa_put(op_data->op_capa1);
2580         capa_put(op_data->op_capa2);
2581         OBD_FREE_PTR(op_data);
2582 }
2583
2584 #ifdef HAVE_SUPEROPS_USE_DENTRY
2585 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2586 #else
2587 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2588 #endif
2589 {
2590         struct ll_sb_info *sbi;
2591
2592 #ifdef HAVE_SUPEROPS_USE_DENTRY
2593         LASSERT((seq != NULL) && (dentry != NULL));
2594         sbi = ll_s2sbi(dentry->d_sb);
2595 #else
2596         LASSERT((seq != NULL) && (vfs != NULL));
2597         sbi = ll_s2sbi(vfs->mnt_sb);
2598 #endif
2599
2600         if (sbi->ll_flags & LL_SBI_NOLCK)
2601                 seq_puts(seq, ",nolock");
2602
2603         if (sbi->ll_flags & LL_SBI_FLOCK)
2604                 seq_puts(seq, ",flock");
2605
2606         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2607                 seq_puts(seq, ",localflock");
2608
2609         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2610                 seq_puts(seq, ",user_xattr");
2611
2612         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2613                 seq_puts(seq, ",lazystatfs");
2614
2615         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2616                 seq_puts(seq, ",user_fid2path");
2617
2618         RETURN(0);
2619 }
2620
2621 /**
2622  * Get obd name by cmd, and copy out to user space
2623  */
2624 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2625 {
2626         struct ll_sb_info *sbi = ll_i2sbi(inode);
2627         struct obd_device *obd;
2628         ENTRY;
2629
2630         if (cmd == OBD_IOC_GETDTNAME)
2631                 obd = class_exp2obd(sbi->ll_dt_exp);
2632         else if (cmd == OBD_IOC_GETMDNAME)
2633                 obd = class_exp2obd(sbi->ll_md_exp);
2634         else
2635                 RETURN(-EINVAL);
2636
2637         if (!obd)
2638                 RETURN(-ENOENT);
2639
2640         if (copy_to_user((void __user *)arg, obd->obd_name,
2641                          strlen(obd->obd_name) + 1))
2642                 RETURN(-EFAULT);
2643
2644         RETURN(0);
2645 }
2646
2647 /**
2648  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2649  * fsname will be returned in this buffer; otherwise, a static buffer will be
2650  * used to store the fsname and returned to caller.
2651  */
2652 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2653 {
2654         static char fsname_static[MTI_NAME_MAXLEN];
2655         struct lustre_sb_info *lsi = s2lsi(sb);
2656         char *ptr;
2657         int len;
2658
2659         if (buf == NULL) {
2660                 /* this means the caller wants to use static buffer
2661                  * and it doesn't care about race. Usually this is
2662                  * in error reporting path */
2663                 buf = fsname_static;
2664                 buflen = sizeof(fsname_static);
2665         }
2666
2667         len = strlen(lsi->lsi_lmd->lmd_profile);
2668         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2669         if (ptr && (strcmp(ptr, "-client") == 0))
2670                 len -= 7;
2671
2672         if (unlikely(len >= buflen))
2673                 len = buflen - 1;
2674         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2675         buf[len] = '\0';
2676
2677         return buf;
2678 }
2679
2680 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2681 {
2682         char *path = NULL;
2683
2684         struct path p;
2685
2686         p.dentry = dentry;
2687         p.mnt = current->fs->root.mnt;
2688         path_get(&p);
2689         path = d_path(&p, buf, bufsize);
2690         path_put(&p);
2691         return path;
2692 }
2693
2694 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2695 {
2696         char *buf, *path = NULL;
2697         struct dentry *dentry = NULL;
2698         struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2699
2700         /* this can be called inside spin lock so use GFP_ATOMIC. */
2701         buf = (char *)__get_free_page(GFP_ATOMIC);
2702         if (buf != NULL) {
2703                 dentry = d_find_alias(page->mapping->host);
2704                 if (dentry != NULL)
2705                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2706         }
2707
2708         CDEBUG(D_WARNING,
2709                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2710                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2711                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2712                PFID(&obj->vob_header.coh_lu.loh_fid),
2713                (path && !IS_ERR(path)) ? path : "", ioret);
2714
2715         if (dentry != NULL)
2716                 dput(dentry);
2717
2718         if (buf != NULL)
2719                 free_page((unsigned long)buf);
2720 }
2721
2722 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2723                         struct lov_user_md **kbuf)
2724 {
2725         struct lov_user_md      lum;
2726         ssize_t                 lum_size;
2727         ENTRY;
2728
2729         if (copy_from_user(&lum, md, sizeof(lum)))
2730                 RETURN(-EFAULT);
2731
2732         lum_size = ll_lov_user_md_size(&lum);
2733         if (lum_size < 0)
2734                 RETURN(lum_size);
2735
2736         OBD_ALLOC(*kbuf, lum_size);
2737         if (*kbuf == NULL)
2738                 RETURN(-ENOMEM);
2739
2740         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2741                 OBD_FREE(*kbuf, lum_size);
2742                 RETURN(-EFAULT);
2743         }
2744
2745         RETURN(lum_size);
2746 }
2747
2748 /*
2749  * Compute llite root squash state after a change of root squash
2750  * configuration setting or add/remove of a lnet nid
2751  */
2752 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2753 {
2754         struct root_squash_info *squash = &sbi->ll_squash;
2755         int i;
2756         bool matched;
2757         lnet_process_id_t id;
2758
2759         /* Update norootsquash flag */
2760         down_write(&squash->rsi_sem);
2761         if (list_empty(&squash->rsi_nosquash_nids))
2762                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2763         else {
2764                 /* Do not apply root squash as soon as one of our NIDs is
2765                  * in the nosquash_nids list */
2766                 matched = false;
2767                 i = 0;
2768                 while (LNetGetId(i++, &id) != -ENOENT) {
2769                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2770                                 continue;
2771                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2772                                 matched = true;
2773                                 break;
2774                         }
2775                 }
2776                 if (matched)
2777                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2778                 else
2779                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2780         }
2781         up_write(&squash->rsi_sem);
2782 }
2783
2784 /**
2785  * Parse linkea content to extract information about a given hardlink
2786  *
2787  * \param[in]   ldata      - Initialized linkea data
2788  * \param[in]   linkno     - Link identifier
2789  * \param[out]  parent_fid - The entry's parent FID
2790  * \param[out]  ln         - Entry name destination buffer
2791  *
2792  * \retval 0 on success
2793  * \retval Appropriate negative error code on failure
2794  */
2795 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2796                             struct lu_fid *parent_fid, struct lu_name *ln)
2797 {
2798         unsigned int    idx;
2799         int             rc;
2800         ENTRY;
2801
2802         rc = linkea_init(ldata);
2803         if (rc < 0)
2804                 RETURN(rc);
2805
2806         if (linkno >= ldata->ld_leh->leh_reccount)
2807                 /* beyond last link */
2808                 RETURN(-ENODATA);
2809
2810         linkea_first_entry(ldata);
2811         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2812                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2813                                     parent_fid);
2814                 if (idx == linkno)
2815                         break;
2816
2817                 linkea_next_entry(ldata);
2818         }
2819
2820         if (idx < linkno)
2821                 RETURN(-ENODATA);
2822
2823         RETURN(0);
2824 }
2825
2826 /**
2827  * Get parent FID and name of an identified link. Operation is performed for
2828  * a given link number, letting the caller iterate over linkno to list one or
2829  * all links of an entry.
2830  *
2831  * \param[in]     file - File descriptor against which to perform the operation
2832  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2833  *                       on and the available size. It is eventually filled with
2834  *                       the requested information or left untouched on error
2835  *
2836  * \retval - 0 on success
2837  * \retval - Appropriate negative error code on failure
2838  */
2839 int ll_getparent(struct file *file, struct getparent __user *arg)
2840 {
2841         struct dentry           *dentry = file->f_dentry;
2842         struct inode            *inode = file->f_dentry->d_inode;
2843         struct linkea_data      *ldata;
2844         struct lu_buf            buf = LU_BUF_NULL;
2845         struct lu_name           ln;
2846         struct lu_fid            parent_fid;
2847         __u32                    linkno;
2848         __u32                    name_size;
2849         int                      rc;
2850
2851         ENTRY;
2852
2853         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2854             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2855                 RETURN(-EPERM);
2856
2857         if (get_user(name_size, &arg->gp_name_size))
2858                 RETURN(-EFAULT);
2859
2860         if (get_user(linkno, &arg->gp_linkno))
2861                 RETURN(-EFAULT);
2862
2863         if (name_size > PATH_MAX)
2864                 RETURN(-EINVAL);
2865
2866         OBD_ALLOC(ldata, sizeof(*ldata));
2867         if (ldata == NULL)
2868                 RETURN(-ENOMEM);
2869
2870         rc = linkea_data_new(ldata, &buf);
2871         if (rc < 0)
2872                 GOTO(ldata_free, rc);
2873
2874         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2875         if (rc < 0)
2876                 GOTO(lb_free, rc);
2877
2878         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2879         if (rc < 0)
2880                 GOTO(lb_free, rc);
2881
2882         if (ln.ln_namelen >= name_size)
2883                 GOTO(lb_free, rc = -EOVERFLOW);
2884
2885         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2886                 GOTO(lb_free, rc = -EFAULT);
2887
2888         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2889                 GOTO(lb_free, rc = -EFAULT);
2890
2891         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2892                 GOTO(lb_free, rc = -EFAULT);
2893
2894 lb_free:
2895         lu_buf_free(&buf);
2896 ldata_free:
2897         OBD_FREE(ldata, sizeof(*ldata));
2898
2899         RETURN(rc);
2900 }