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, 2013, 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/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_ioctl.h>
  49 #include <lustre_lite.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_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_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_agl_total, 0);
 142         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 spin_lock(&ll_sb_lock);
 160                 list_del(&sbi->ll_list);
 161                 spin_unlock(&ll_sb_lock);
 162                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 163                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 170                                     struct vfsmount *mnt)
 171 {
 172         struct inode *root = 0;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct obd_capa *oc = NULL;
 176         struct obd_statfs *osfs = NULL;
 177         struct ptlrpc_request *request = NULL;
 178         struct obd_connect_data *data = NULL;
 179         struct obd_uuid *uuid;
 180         struct md_op_data *op_data;
 181         struct lustre_md lmd;
 182         obd_valid valid;
 183         int size, err, checksum;
 184         ENTRY;
 185
 186         obd = class_name2obd(md);
 187         if (!obd) {
 188                 CERROR("MD %s: not setup or attached\n", md);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         OBD_ALLOC_PTR(data);
 193         if (data == NULL)
 194                 RETURN(-ENOMEM);
 195
 196         OBD_ALLOC_PTR(osfs);
 197         if (osfs == NULL) {
 198                 OBD_FREE_PTR(data);
 199                 RETURN(-ENOMEM);
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 211                                   OBD_CONNECT_EINPROGRESS |
 212                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 213                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 214                                   OBD_CONNECT_MAX_EASIZE |
 215                                   OBD_CONNECT_FLOCK_DEAD |
 216                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK;
 217
 218         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 219                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 220
 221 #ifdef HAVE_LRU_RESIZE_SUPPORT
 222         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 223                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 224 #endif
 225 #ifdef CONFIG_FS_POSIX_ACL
 226         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 227 #endif
 228
 229         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 230                 /* flag mdc connection as lightweight, only used for test
 231                  * purpose, use with care */
 232                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 233
 234         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 235         data->ocd_version = LUSTRE_VERSION_CODE;
 236
 237         if (sb->s_flags & MS_RDONLY)
 238                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 239         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 240                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 241
 242 #ifdef HAVE_MS_FLOCK_LOCK
 243         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 244         sb->s_flags |= MS_FLOCK_LOCK;
 245 #endif
 246 #ifdef MS_HAS_NEW_AOPS
 247         sb->s_flags |= MS_HAS_NEW_AOPS;
 248 #endif
 249
 250         if (sbi->ll_flags & LL_SBI_FLOCK)
 251                 sbi->ll_fop = &ll_file_operations_flock;
 252         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 253                 sbi->ll_fop = &ll_file_operations;
 254         else
 255                 sbi->ll_fop = &ll_file_operations_noflock;
 256
 257         /* real client */
 258         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 259         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 260                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 261
 262         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 263
 264         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 265         if (err == -EBUSY) {
 266                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 267                                    "recovery, of which this client is not a "
 268                                    "part. Please wait for recovery to complete,"
 269                                    " abort, or time out.\n", md);
 270                 GOTO(out, err);
 271         } else if (err) {
 272                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 273                 GOTO(out, err);
 274         }
 275
 276         sbi->ll_md_exp->exp_connect_data = *data;
 277
 278         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 279                            LUSTRE_SEQ_METADATA);
 280         if (err) {
 281                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 282                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 283                 GOTO(out_md, err);
 284         }
 285
 286         /* For mount, we only need fs info from MDT0, and also in DNE, it
 287          * can make sure the client can be mounted as long as MDT0 is
 288          * avaible */
 289         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 290                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 291                         OBD_STATFS_FOR_MDT0);
 292         if (err)
 293                 GOTO(out_md_fid, err);
 294
 295         /* This needs to be after statfs to ensure connect has finished.
 296          * Note that "data" does NOT contain the valid connect reply.
 297          * If connecting to a 1.8 server there will be no LMV device, so
 298          * we can access the MDC export directly and exp_connect_flags will
 299          * be non-zero, but if accessing an upgraded 2.1 server it will
 300          * have the correct flags filled in.
 301          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 302         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 303         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 304             valid != CLIENT_CONNECT_MDT_REQD) {
 305                 char *buf;
 306
 307                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 308                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 309                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 310                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 311                                    "feature(s) needed for correct operation "
 312                                    "of this client (%s). Please upgrade "
 313                                    "server or downgrade client.\n",
 314                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 315                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 316                 GOTO(out_md_fid, err = -EPROTO);
 317         }
 318
 319         size = sizeof(*data);
 320         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 321                            KEY_CONN_DATA,  &size, data, NULL);
 322         if (err) {
 323                 CERROR("%s: Get connect data failed: rc = %d\n",
 324                        sbi->ll_md_exp->exp_obd->obd_name, err);
 325                 GOTO(out_md_fid, err);
 326         }
 327
 328         LASSERT(osfs->os_bsize);
 329         sb->s_blocksize = osfs->os_bsize;
 330         sb->s_blocksize_bits = log2(osfs->os_bsize);
 331         sb->s_magic = LL_SUPER_MAGIC;
 332         sb->s_maxbytes = MAX_LFS_FILESIZE;
 333         sbi->ll_namelen = osfs->os_namelen;
 334         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 335
 336         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 337             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 338                 LCONSOLE_INFO("Disabling user_xattr feature because "
 339                               "it is not supported on the server\n");
 340                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 344 #ifdef MS_POSIXACL
 345                 sb->s_flags |= MS_POSIXACL;
 346 #endif
 347                 sbi->ll_flags |= LL_SBI_ACL;
 348         } else {
 349                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 350 #ifdef MS_POSIXACL
 351                 sb->s_flags &= ~MS_POSIXACL;
 352 #endif
 353                 sbi->ll_flags &= ~LL_SBI_ACL;
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 357                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 358                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client is set as remote by default.\n");
 360                 }
 361         } else {
 362                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 363                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client claims to be remote, but server "
 365                                       "rejected, forced to be local.\n");
 366                 }
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 370                 LCONSOLE_INFO("client enabled MDS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 375                 LCONSOLE_INFO("client enabled OSS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 380                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 383                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 384         else
 385                 sbi->ll_md_brw_pages = 1;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 388                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 389
 390         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 391                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 392                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 393                                       "unknown maximum xattr size.\n", dt);
 394                 } else {
 395                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 396                         sbi->ll_xattr_cache_enabled = 1;
 397                 }
 398         }
 399
 400         obd = class_name2obd(dt);
 401         if (!obd) {
 402                 CERROR("DT %s: not setup or attached\n", dt);
 403                 GOTO(out_md_fid, err = -ENODEV);
 404         }
 405
 406         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 407                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 408                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 409                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 410                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 411                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 412                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 413                                   OBD_CONNECT_MAXBYTES |
 414                                   OBD_CONNECT_EINPROGRESS |
 415                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 416                                   OBD_CONNECT_LAYOUTLOCK |
 417                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 418
 419         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 420                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 421
 422         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 423                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 424                  * disabled by default, because it can still be enabled on the
 425                  * fly via /proc. As a consequence, we still need to come to an
 426                  * agreement on the supported algorithms at connect time */
 427                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 428
 429                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 430                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 431                 else
 432                         data->ocd_cksum_types = cksum_types_supported_client();
 433         }
 434
 435 #ifdef HAVE_LRU_RESIZE_SUPPORT
 436         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 437 #endif
 438         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 439                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 440
 441         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 442                "ocd_grant: %d\n", data->ocd_connect_flags,
 443                data->ocd_version, data->ocd_grant);
 444
 445         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 446         obd->obd_upcall.onu_upcall = cl_ocd_update;
 447
 448         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 449
 450         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 451                           NULL);
 452         if (err == -EBUSY) {
 453                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 454                                    "recovery, of which this client is not a "
 455                                    "part.  Please wait for recovery to "
 456                                    "complete, abort, or time out.\n", dt);
 457                 GOTO(out_md, err);
 458         } else if (err) {
 459                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 460                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 461                 GOTO(out_md, err);
 462         }
 463
 464         sbi->ll_dt_exp->exp_connect_data = *data;
 465
 466         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 467                            LUSTRE_SEQ_METADATA);
 468         if (err) {
 469                 CERROR("%s: Can't init data layer FID infrastructure, "
 470                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 471                 GOTO(out_dt, err);
 472         }
 473
 474         mutex_lock(&sbi->ll_lco.lco_lock);
 475         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 476         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 477         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 478         mutex_unlock(&sbi->ll_lco.lco_lock);
 479
 480         fid_zero(&sbi->ll_root_fid);
 481         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 482         if (err) {
 483                 CERROR("cannot mds_connect: rc = %d\n", err);
 484                 GOTO(out_lock_cn_cb, err);
 485         }
 486         if (!fid_is_sane(&sbi->ll_root_fid)) {
 487                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 488                        sbi->ll_md_exp->exp_obd->obd_name,
 489                        PFID(&sbi->ll_root_fid));
 490                 GOTO(out_lock_cn_cb, err = -EINVAL);
 491         }
 492         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 493
 494         sb->s_op = &lustre_super_operations;
 495 #if THREAD_SIZE >= 8192 /*b=17630*/
 496         sb->s_export_op = &lustre_export_operations;
 497 #endif
 498
 499         /* make root inode
 500          * XXX: move this to after cbd setup? */
 501         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 502                 OBD_MD_FLMODEASIZE;
 503         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 504                 valid |= OBD_MD_FLRMTPERM;
 505         else if (sbi->ll_flags & LL_SBI_ACL)
 506                 valid |= OBD_MD_FLACL;
 507
 508         OBD_ALLOC_PTR(op_data);
 509         if (op_data == NULL)
 510                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 511
 512         op_data->op_fid1 = sbi->ll_root_fid;
 513         op_data->op_mode = 0;
 514         op_data->op_capa1 = oc;
 515         op_data->op_valid = valid;
 516
 517         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 518         if (oc)
 519                 capa_put(oc);
 520         OBD_FREE_PTR(op_data);
 521         if (err) {
 522                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 523                        sbi->ll_md_exp->exp_obd->obd_name, err);
 524                 GOTO(out_lock_cn_cb, err);
 525         }
 526
 527         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 528                                sbi->ll_md_exp, &lmd);
 529         if (err) {
 530                 CERROR("failed to understand root inode md: rc = %d\n", err);
 531                 ptlrpc_req_finished(request);
 532                 GOTO(out_lock_cn_cb, err);
 533         }
 534
 535         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 536         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 537                                             sbi->ll_flags & LL_SBI_32BIT_API),
 538                        &lmd);
 539         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 540         ptlrpc_req_finished(request);
 541
 542         if (IS_ERR(root)) {
 543                 if (lmd.lsm)
 544                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 545 #ifdef CONFIG_FS_POSIX_ACL
 546                 if (lmd.posix_acl) {
 547                         posix_acl_release(lmd.posix_acl);
 548                         lmd.posix_acl = NULL;
 549                 }
 550 #endif
 551                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 552                 root = NULL;
 553                 CERROR("lustre_lite: bad iget4 for root\n");
 554                 GOTO(out_root, err);
 555         }
 556
 557         err = ll_close_thread_start(&sbi->ll_lcq);
 558         if (err) {
 559                 CERROR("cannot start close thread: rc %d\n", err);
 560                 GOTO(out_root, err);
 561         }
 562
 563 #ifdef CONFIG_FS_POSIX_ACL
 564         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 565                 rct_init(&sbi->ll_rct);
 566                 et_init(&sbi->ll_et);
 567         }
 568 #endif
 569
 570         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 571         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 572                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 573                                  NULL);
 574         cl_sb_init(sb);
 575
 576         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 577                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 578                                  &sbi->ll_cache, NULL);
 579
 580         sb->s_root = d_make_root(root);
 581         if (sb->s_root == NULL) {
 582                 CERROR("%s: can't make root dentry\n",
 583                         ll_get_fsname(sb, NULL, 0));
 584                 GOTO(out_root, err = -ENOMEM);
 585         }
 586 #ifdef HAVE_DCACHE_LOCK
 587         sb->s_root->d_op = &ll_d_ops;
 588 #endif
 589
 590         sbi->ll_sdev_orig = sb->s_dev;
 591
 592         /* We set sb->s_dev equal on all lustre clients in order to support
 593          * NFS export clustering.  NFSD requires that the FSID be the same
 594          * on all clients. */
 595         /* s_dev is also used in lt_compare() to compare two fs, but that is
 596          * only a node-local comparison. */
 597         uuid = obd_get_uuid(sbi->ll_md_exp);
 598         if (uuid != NULL)
 599                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 600
 601         if (data != NULL)
 602                 OBD_FREE_PTR(data);
 603         if (osfs != NULL)
 604                 OBD_FREE_PTR(osfs);
 605         if (proc_lustre_fs_root != NULL) {
 606                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 607                                                   dt, md);
 608                 if (err < 0) {
 609                         CERROR("%s: could not register mount in lprocfs: "
 610                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 611                         err = 0;
 612                 }
 613         }
 614
 615         RETURN(err);
 616 out_root:
 617         if (root)
 618                 iput(root);
 619 out_lock_cn_cb:
 620         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 621 out_dt:
 622         obd_disconnect(sbi->ll_dt_exp);
 623         sbi->ll_dt_exp = NULL;
 624         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 625         obd_zombie_barrier();
 626 out_md_fid:
 627         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 628 out_md:
 629         obd_disconnect(sbi->ll_md_exp);
 630         sbi->ll_md_exp = NULL;
 631 out:
 632         if (data != NULL)
 633                 OBD_FREE_PTR(data);
 634         if (osfs != NULL)
 635                 OBD_FREE_PTR(osfs);
 636         return err;
 637 }
 638
 639 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 640 {
 641         int size, rc;
 642
 643         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 644         size = sizeof(int);
 645         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 646                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 647         if (rc)
 648                 CERROR("Get max mdsize error rc %d\n", rc);
 649
 650         RETURN(rc);
 651 }
 652
 653 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 654 {
 655         int size, rc;
 656
 657         size = sizeof(int);
 658         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 659                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 660         if (rc)
 661                 CERROR("Get default mdsize error rc %d\n", rc);
 662
 663         RETURN(rc);
 664 }
 665
 666 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 667 {
 668         int size, rc;
 669
 670         size = sizeof(int);
 671         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 672                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 673         if (rc)
 674                 CERROR("Get max cookiesize error rc %d\n", rc);
 675
 676         RETURN(rc);
 677 }
 678
 679 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 680 {
 681         int size, rc;
 682
 683         size = sizeof(int);
 684         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 685                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 686         if (rc)
 687                 CERROR("Get default cookiesize error rc %d\n", rc);
 688
 689         RETURN(rc);
 690 }
 691
 692 static void ll_dump_inode(struct inode *inode)
 693 {
 694         struct ll_d_hlist_node *tmp;
 695         int dentry_count = 0;
 696
 697         LASSERT(inode != NULL);
 698
 699         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 700                 dentry_count++;
 701
 702         CERROR("%s: inode %p dump: dev=%s fid="DFID
 703                " mode=%o count=%u, %d dentries\n",
 704                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 705                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 706                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 707 }
 708
 709 void lustre_dump_dentry(struct dentry *dentry, int recur)
 710 {
 711         struct list_head *tmp;
 712         int subdirs = 0;
 713
 714         LASSERT(dentry != NULL);
 715
 716         list_for_each(tmp, &dentry->d_subdirs)
 717                 subdirs++;
 718
 719         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 720                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 721                dentry->d_name.len, dentry->d_name.name,
 722                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 723                dentry->d_parent, dentry->d_inode, d_count(dentry),
 724                dentry->d_flags, dentry->d_fsdata, subdirs);
 725         if (dentry->d_inode != NULL)
 726                 ll_dump_inode(dentry->d_inode);
 727
 728         if (recur == 0)
 729                 return;
 730
 731         list_for_each(tmp, &dentry->d_subdirs) {
 732                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 733                 lustre_dump_dentry(d, recur - 1);
 734         }
 735 }
 736
 737 static void client_common_put_super(struct super_block *sb)
 738 {
 739         struct ll_sb_info *sbi = ll_s2sbi(sb);
 740         ENTRY;
 741
 742 #ifdef CONFIG_FS_POSIX_ACL
 743         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 744                 et_fini(&sbi->ll_et);
 745                 rct_fini(&sbi->ll_rct);
 746         }
 747 #endif
 748
 749         ll_close_thread_shutdown(sbi->ll_lcq);
 750
 751         cl_sb_fini(sb);
 752
 753         list_del(&sbi->ll_conn_chain);
 754
 755         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 756         obd_disconnect(sbi->ll_dt_exp);
 757         sbi->ll_dt_exp = NULL;
 758         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 759          * see LU-2543. */
 760         obd_zombie_barrier();
 761
 762         lprocfs_unregister_mountpoint(sbi);
 763
 764         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 765         obd_disconnect(sbi->ll_md_exp);
 766         sbi->ll_md_exp = NULL;
 767
 768         EXIT;
 769 }
 770
 771 void ll_kill_super(struct super_block *sb)
 772 {
 773         struct ll_sb_info *sbi;
 774
 775         ENTRY;
 776
 777         /* not init sb ?*/
 778         if (!(sb->s_flags & MS_ACTIVE))
 779                 return;
 780
 781         sbi = ll_s2sbi(sb);
 782         /* we need restore s_dev from changed for clustred NFS before put_super
 783          * because new kernels have cached s_dev and change sb->s_dev in
 784          * put_super not affected real removing devices */
 785         if (sbi) {
 786                 sb->s_dev = sbi->ll_sdev_orig;
 787                 sbi->ll_umounting = 1;
 788         }
 789         EXIT;
 790 }
 791
 792 static inline int ll_set_opt(const char *opt, char *data, int fl)
 793 {
 794         if (strncmp(opt, data, strlen(opt)) != 0)
 795                 return(0);
 796         else
 797                 return(fl);
 798 }
 799
 800 /* non-client-specific mount options are parsed in lmd_parse */
 801 static int ll_options(char *options, int *flags)
 802 {
 803         int tmp;
 804         char *s1 = options, *s2;
 805         ENTRY;
 806
 807         if (!options)
 808                 RETURN(0);
 809
 810         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 811
 812         while (*s1) {
 813                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 814                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 830                 if (tmp) {
 831                         *flags &= ~tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 835                 if (tmp) {
 836                         *flags |= tmp;
 837                         goto next;
 838                 }
 839                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 840                 if (tmp) {
 841                         *flags &= ~tmp;
 842                         goto next;
 843                 }
 844 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 845                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 846                 if (tmp) {
 847                         /* Ignore deprecated mount option.  The client will
 848                          * always try to mount with ACL support, whether this
 849                          * is used depends on whether server supports it. */
 850                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 851                                                   "mount option 'acl'.\n");
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 855                 if (tmp) {
 856                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 857                                                   "mount option 'noacl'.\n");
 858                         goto next;
 859                 }
 860 #else
 861 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 862 #endif
 863                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878
 879                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 880                 if (tmp) {
 881                         *flags |= tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 885                 if (tmp) {
 886                         *flags &= ~tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 895                 if (tmp) {
 896                         *flags &= ~tmp;
 897                         goto next;
 898                 }
 899                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 900                 if (tmp) {
 901                         *flags |= tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 905                 if (tmp) {
 906                         *flags &= ~tmp;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 910                 if (tmp) {
 911                         *flags |= tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 915                 if (tmp) {
 916                         *flags |= tmp;
 917                         goto next;
 918                 }
 919                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 920                 if (tmp) {
 921                         *flags |= tmp;
 922                         goto next;
 923                 }
 924                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 925                 if (tmp) {
 926                         *flags &= ~tmp;
 927                         goto next;
 928                 }
 929                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 930                                    s1);
 931                 RETURN(-EINVAL);
 932
 933 next:
 934                 /* Find next opt */
 935                 s2 = strchr(s1, ',');
 936                 if (s2 == NULL)
 937                         break;
 938                 s1 = s2 + 1;
 939         }
 940         RETURN(0);
 941 }
 942
 943 void ll_lli_init(struct ll_inode_info *lli)
 944 {
 945         lli->lli_inode_magic = LLI_INODE_MAGIC;
 946         lli->lli_flags = 0;
 947         lli->lli_ioepoch = 0;
 948         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 949         spin_lock_init(&lli->lli_lock);
 950         lli->lli_posix_acl = NULL;
 951         lli->lli_remote_perms = NULL;
 952         mutex_init(&lli->lli_rmtperm_mutex);
 953         /* Do not set lli_fid, it has been initialized already. */
 954         fid_zero(&lli->lli_pfid);
 955         INIT_LIST_HEAD(&lli->lli_close_list);
 956         INIT_LIST_HEAD(&lli->lli_oss_capas);
 957         atomic_set(&lli->lli_open_count, 0);
 958         lli->lli_mds_capa = NULL;
 959         lli->lli_rmtperm_time = 0;
 960         lli->lli_pending_och = NULL;
 961         lli->lli_mds_read_och = NULL;
 962         lli->lli_mds_write_och = NULL;
 963         lli->lli_mds_exec_och = NULL;
 964         lli->lli_open_fd_read_count = 0;
 965         lli->lli_open_fd_write_count = 0;
 966         lli->lli_open_fd_exec_count = 0;
 967         mutex_init(&lli->lli_och_mutex);
 968         spin_lock_init(&lli->lli_agl_lock);
 969         lli->lli_has_smd = false;
 970         spin_lock_init(&lli->lli_layout_lock);
 971         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 972         lli->lli_clob = NULL;
 973
 974         init_rwsem(&lli->lli_xattrs_list_rwsem);
 975         mutex_init(&lli->lli_xattrs_enq_lock);
 976
 977         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 978         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 979                 mutex_init(&lli->lli_readdir_mutex);
 980                 lli->lli_opendir_key = NULL;
 981                 lli->lli_sai = NULL;
 982                 spin_lock_init(&lli->lli_sa_lock);
 983                 lli->lli_opendir_pid = 0;
 984         } else {
 985                 mutex_init(&lli->lli_size_mutex);
 986                 lli->lli_symlink_name = NULL;
 987                 init_rwsem(&lli->lli_trunc_sem);
 988                 mutex_init(&lli->lli_write_mutex);
 989                 init_rwsem(&lli->lli_glimpse_sem);
 990                 lli->lli_glimpse_time = 0;
 991                 INIT_LIST_HEAD(&lli->lli_agl_list);
 992                 lli->lli_agl_index = 0;
 993                 lli->lli_async_rc = 0;
 994         }
 995         mutex_init(&lli->lli_layout_mutex);
 996 }
 997
 998 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 999 {
1000         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1001
1002         bdi->name = "lustre";
1003         return bdi_register(bdi, NULL, "lustre-%d",
1004                             atomic_inc_return(&ll_bdi_num));
1005 }
1006
1007 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1008 {
1009         struct lustre_profile *lprof = NULL;
1010         struct lustre_sb_info *lsi = s2lsi(sb);
1011         struct ll_sb_info *sbi;
1012         char  *dt = NULL, *md = NULL;
1013         char  *profilenm = get_profile_name(sb);
1014         struct config_llog_instance *cfg;
1015         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1016         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1017         int    err;
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1021
1022         OBD_ALLOC_PTR(cfg);
1023         if (cfg == NULL)
1024                 RETURN(-ENOMEM);
1025
1026         try_module_get(THIS_MODULE);
1027
1028         /* client additional sb info */
1029         lsi->lsi_llsbi = sbi = ll_init_sbi();
1030         if (!sbi) {
1031                 module_put(THIS_MODULE);
1032                 OBD_FREE_PTR(cfg);
1033                 RETURN(-ENOMEM);
1034         }
1035
1036         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1037         if (err)
1038                 GOTO(out_free, err);
1039
1040         err = bdi_init(&lsi->lsi_bdi);
1041         if (err)
1042                 GOTO(out_free, err);
1043         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1044         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1045         err = ll_bdi_register(&lsi->lsi_bdi);
1046         if (err)
1047                 GOTO(out_free, err);
1048
1049         sb->s_bdi = &lsi->lsi_bdi;
1050 #ifndef HAVE_DCACHE_LOCK
1051         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1052         sb->s_d_op = &ll_d_ops;
1053 #endif
1054
1055         /* Generate a string unique to this super, in case some joker tries
1056            to mount the same fs at two mount points.
1057            Use the address of the super itself.*/
1058         cfg->cfg_instance = sb;
1059         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1060         cfg->cfg_callback = class_config_llog_handler;
1061         /* set up client obds */
1062         err = lustre_process_log(sb, profilenm, cfg);
1063         if (err < 0)
1064                 GOTO(out_free, err);
1065
1066         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1067         lprof = class_get_profile(profilenm);
1068         if (lprof == NULL) {
1069                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1070                                    " read from the MGS.  Does that filesystem "
1071                                    "exist?\n", profilenm);
1072                 GOTO(out_free, err = -EINVAL);
1073         }
1074         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1075                lprof->lp_md, lprof->lp_dt);
1076
1077         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1078         if (!dt)
1079                 GOTO(out_free, err = -ENOMEM);
1080         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1081
1082         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1083         if (!md)
1084                 GOTO(out_free, err = -ENOMEM);
1085         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1086
1087         /* connections, registrations, sb setup */
1088         err = client_common_fill_super(sb, md, dt, mnt);
1089
1090 out_free:
1091         if (md)
1092                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1093         if (dt)
1094                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1095         if (err)
1096                 ll_put_super(sb);
1097         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1098                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1099
1100         OBD_FREE_PTR(cfg);
1101         RETURN(err);
1102 } /* ll_fill_super */
1103
1104 void ll_put_super(struct super_block *sb)
1105 {
1106         struct config_llog_instance cfg, params_cfg;
1107         struct obd_device *obd;
1108         struct lustre_sb_info *lsi = s2lsi(sb);
1109         struct ll_sb_info *sbi = ll_s2sbi(sb);
1110         char *profilenm = get_profile_name(sb);
1111         int ccc_count, next, force = 1, rc = 0;
1112         ENTRY;
1113
1114         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1115
1116         ll_print_capa_stat(sbi);
1117
1118         cfg.cfg_instance = sb;
1119         lustre_end_log(sb, profilenm, &cfg);
1120
1121         params_cfg.cfg_instance = sb;
1122         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1123
1124         if (sbi->ll_md_exp) {
1125                 obd = class_exp2obd(sbi->ll_md_exp);
1126                 if (obd)
1127                         force = obd->obd_force;
1128         }
1129
1130         /* Wait for unstable pages to be committed to stable storage */
1131         if (force == 0) {
1132                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1133                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1134                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1135                         &lwi);
1136         }
1137
1138         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1139         if (force == 0 && rc != -EINTR)
1140                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1141
1142
1143         /* We need to set force before the lov_disconnect in
1144            lustre_common_put_super, since l_d cleans up osc's as well. */
1145         if (force) {
1146                 next = 0;
1147                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1148                                                      &next)) != NULL) {
1149                         obd->obd_force = force;
1150                 }
1151         }
1152
1153         if (sbi->ll_lcq) {
1154                 /* Only if client_common_fill_super succeeded */
1155                 client_common_put_super(sb);
1156         }
1157
1158         next = 0;
1159         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1160                 class_manual_cleanup(obd);
1161         }
1162
1163         if (sbi->ll_flags & LL_SBI_VERBOSE)
1164                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1165
1166         if (profilenm)
1167                 class_del_profile(profilenm);
1168
1169         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1170                 bdi_destroy(&lsi->lsi_bdi);
1171                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1172         }
1173
1174         ll_free_sbi(sb);
1175         lsi->lsi_llsbi = NULL;
1176
1177         lustre_common_put_super(sb);
1178
1179         cl_env_cache_purge(~0);
1180
1181         module_put(THIS_MODULE);
1182
1183         EXIT;
1184 } /* client_put_super */
1185
1186 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1187 {
1188         struct inode *inode = NULL;
1189
1190         /* NOTE: we depend on atomic igrab() -bzzz */
1191         lock_res_and_lock(lock);
1192         if (lock->l_resource->lr_lvb_inode) {
1193                 struct ll_inode_info * lli;
1194                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1195                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1196                         inode = igrab(lock->l_resource->lr_lvb_inode);
1197                 } else {
1198                         inode = lock->l_resource->lr_lvb_inode;
1199                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1200                                          D_WARNING, lock, "lr_lvb_inode %p is "
1201                                          "bogus: magic %08x",
1202                                          lock->l_resource->lr_lvb_inode,
1203                                          lli->lli_inode_magic);
1204                         inode = NULL;
1205                 }
1206         }
1207         unlock_res_and_lock(lock);
1208         return inode;
1209 }
1210
1211 static void ll_dir_clear_lsm_md(struct inode *inode)
1212 {
1213         struct ll_inode_info *lli = ll_i2info(inode);
1214
1215         LASSERT(S_ISDIR(inode->i_mode));
1216
1217         if (lli->lli_lsm_md != NULL) {
1218                 lmv_free_memmd(lli->lli_lsm_md);
1219                 lli->lli_lsm_md = NULL;
1220         }
1221 }
1222
1223 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1224                                       const struct lu_fid *fid,
1225                                       struct lustre_md *md)
1226 {
1227         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1228         struct mdt_body         *body = md->body;
1229         struct inode            *inode;
1230         ino_t                   ino;
1231         ENTRY;
1232
1233         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1234         inode = iget_locked(sb, ino);
1235         if (inode == NULL) {
1236                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1237                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1238                 RETURN(ERR_PTR(-ENOENT));
1239         }
1240
1241         if (inode->i_state & I_NEW) {
1242                 struct ll_inode_info *lli = ll_i2info(inode);
1243                 struct lmv_stripe_md *lsm = md->lmv;
1244
1245                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1246                                 (body->mbo_mode & S_IFMT);
1247                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1248                          PFID(fid));
1249
1250                 LTIME_S(inode->i_mtime) = 0;
1251                 LTIME_S(inode->i_atime) = 0;
1252                 LTIME_S(inode->i_ctime) = 0;
1253                 inode->i_rdev = 0;
1254
1255                 /* initializing backing dev info. */
1256                 inode->i_mapping->backing_dev_info =
1257                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1258                 inode->i_op = &ll_dir_inode_operations;
1259                 inode->i_fop = &ll_dir_operations;
1260                 lli->lli_fid = *fid;
1261                 ll_lli_init(lli);
1262
1263                 LASSERT(lsm != NULL);
1264                 /* master object FID */
1265                 lli->lli_pfid = body->mbo_fid1;
1266                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1267                        lli, PFID(fid), PFID(&lli->lli_pfid));
1268                 unlock_new_inode(inode);
1269         }
1270
1271         RETURN(inode);
1272 }
1273
1274 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1275 {
1276         struct lu_fid *fid;
1277         struct lmv_stripe_md *lsm = md->lmv;
1278         int i;
1279
1280         LASSERT(lsm != NULL);
1281         /* XXX sigh, this lsm_root initialization should be in
1282          * LMV layer, but it needs ll_iget right now, so we
1283          * put this here right now. */
1284         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1285                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1286                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1287                 /* Unfortunately ll_iget will call ll_update_inode,
1288                  * where the initialization of slave inode is slightly
1289                  * different, so it reset lsm_md to NULL to avoid
1290                  * initializing lsm for slave inode. */
1291                 /* For migrating inode, master stripe and master object will
1292                  * be same, so we only need assign this inode */
1293                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1294                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1295                 else
1296                         lsm->lsm_md_oinfo[i].lmo_root =
1297                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1298
1299                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1300                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1301
1302                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1303                         return rc;
1304                 }
1305         }
1306
1307         /* Here is where the lsm is being initialized(fill lmo_info) after
1308          * client retrieve MD stripe information from MDT. */
1309         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1310                                 ll_md_blocking_ast);
1311 }
1312
1313 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1314                                 const struct lmv_stripe_md *lsm_md2)
1315 {
1316         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1317                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1318                lsm_md1->lsm_md_master_mdt_index ==
1319                                         lsm_md2->lsm_md_master_mdt_index &&
1320                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1321                lsm_md1->lsm_md_layout_version ==
1322                                         lsm_md2->lsm_md_layout_version &&
1323                strcmp(lsm_md1->lsm_md_pool_name,
1324                       lsm_md2->lsm_md_pool_name) == 0;
1325 }
1326
1327 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1328 {
1329         struct ll_inode_info *lli = ll_i2info(inode);
1330         struct lmv_stripe_md *lsm = md->lmv;
1331         int     rc;
1332         ENTRY;
1333
1334         LASSERT(S_ISDIR(inode->i_mode));
1335         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1336                PFID(ll_inode2fid(inode)));
1337
1338         /* no striped information from request. */
1339         if (lsm == NULL) {
1340                 if (lli->lli_lsm_md == NULL) {
1341                         RETURN(0);
1342                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1343                                                 LMV_HASH_FLAG_MIGRATION) {
1344                         /* migration is done, the temporay MIGRATE layout has
1345                          * been removed */
1346                         CDEBUG(D_INODE, DFID" finish migration.\n",
1347                                PFID(ll_inode2fid(inode)));
1348                         lmv_free_memmd(lli->lli_lsm_md);
1349                         lli->lli_lsm_md = NULL;
1350                         RETURN(0);
1351                 } else {
1352                         /* The lustre_md from req does not include stripeEA,
1353                          * see ll_md_setattr */
1354                         RETURN(0);
1355                 }
1356         }
1357
1358         /* set the directory layout */
1359         if (lli->lli_lsm_md == NULL) {
1360
1361                 rc = ll_init_lsm_md(inode, md);
1362                 if (rc != 0)
1363                         RETURN(rc);
1364
1365                 lli->lli_lsm_md = lsm;
1366                 /* set lsm_md to NULL, so the following free lustre_md
1367                  * will not free this lsm */
1368                 md->lmv = NULL;
1369                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1370                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1371                 RETURN(0);
1372         }
1373
1374         /* Compare the old and new stripe information */
1375         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1376                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1377                 int                     idx;
1378
1379                 CERROR("%s: lmv layout mismatch "DFID"(%p)/"DFID"(%p)"
1380                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1381                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1382                        ll_get_fsname(inode->i_sb, NULL, 0),
1383                        PFID(&lsm->lsm_md_master_fid), lsm,
1384                        PFID(&old_lsm->lsm_md_master_fid), old_lsm,
1385                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1386                        lsm->lsm_md_stripe_count,
1387                        old_lsm->lsm_md_stripe_count,
1388                        lsm->lsm_md_master_mdt_index,
1389                        old_lsm->lsm_md_master_mdt_index,
1390                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1391                        lsm->lsm_md_layout_version,
1392                        old_lsm->lsm_md_layout_version,
1393                        lsm->lsm_md_pool_name,
1394                        old_lsm->lsm_md_pool_name);
1395
1396                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1397                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1398                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1399                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1400                 }
1401
1402                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1403                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1404                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1405                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1406                 }
1407
1408                 RETURN(-EIO);
1409         }
1410
1411         rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1412                               md->body, ll_md_blocking_ast);
1413
1414         RETURN(rc);
1415 }
1416
1417 void ll_clear_inode(struct inode *inode)
1418 {
1419         struct ll_inode_info *lli = ll_i2info(inode);
1420         struct ll_sb_info *sbi = ll_i2sbi(inode);
1421         ENTRY;
1422
1423         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1424                PFID(ll_inode2fid(inode)), inode);
1425
1426         if (S_ISDIR(inode->i_mode)) {
1427                 /* these should have been cleared in ll_file_release */
1428                 LASSERT(lli->lli_opendir_key == NULL);
1429                 LASSERT(lli->lli_sai == NULL);
1430                 LASSERT(lli->lli_opendir_pid == 0);
1431         }
1432
1433         spin_lock(&lli->lli_lock);
1434         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1435         spin_unlock(&lli->lli_lock);
1436         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1437
1438         LASSERT(!lli->lli_open_fd_write_count);
1439         LASSERT(!lli->lli_open_fd_read_count);
1440         LASSERT(!lli->lli_open_fd_exec_count);
1441
1442         if (lli->lli_mds_write_och)
1443                 ll_md_real_close(inode, FMODE_WRITE);
1444         if (lli->lli_mds_exec_och)
1445                 ll_md_real_close(inode, FMODE_EXEC);
1446         if (lli->lli_mds_read_och)
1447                 ll_md_real_close(inode, FMODE_READ);
1448
1449         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1450                 OBD_FREE(lli->lli_symlink_name,
1451                          strlen(lli->lli_symlink_name) + 1);
1452                 lli->lli_symlink_name = NULL;
1453         }
1454
1455         ll_xattr_cache_destroy(inode);
1456
1457         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1458                 LASSERT(lli->lli_posix_acl == NULL);
1459                 if (lli->lli_remote_perms) {
1460                         free_rmtperm_hash(lli->lli_remote_perms);
1461                         lli->lli_remote_perms = NULL;
1462                 }
1463         }
1464 #ifdef CONFIG_FS_POSIX_ACL
1465         else if (lli->lli_posix_acl) {
1466                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1467                 LASSERT(lli->lli_remote_perms == NULL);
1468                 posix_acl_release(lli->lli_posix_acl);
1469                 lli->lli_posix_acl = NULL;
1470         }
1471 #endif
1472         lli->lli_inode_magic = LLI_INODE_DEAD;
1473
1474         ll_clear_inode_capas(inode);
1475         if (S_ISDIR(inode->i_mode))
1476                 ll_dir_clear_lsm_md(inode);
1477         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1478                 LASSERT(list_empty(&lli->lli_agl_list));
1479
1480         /*
1481          * XXX This has to be done before lsm is freed below, because
1482          * cl_object still uses inode lsm.
1483          */
1484         cl_inode_fini(inode);
1485         lli->lli_has_smd = false;
1486
1487         EXIT;
1488 }
1489
1490 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1491                   struct md_open_data **mod)
1492 {
1493         struct lustre_md md;
1494         struct inode *inode = dentry->d_inode;
1495         struct ll_sb_info *sbi = ll_i2sbi(inode);
1496         struct ptlrpc_request *request = NULL;
1497         int rc, ia_valid;
1498         ENTRY;
1499
1500         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1501                                      LUSTRE_OPC_ANY, NULL);
1502         if (IS_ERR(op_data))
1503                 RETURN(PTR_ERR(op_data));
1504
1505         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1506                         &request, mod);
1507         if (rc) {
1508                 ptlrpc_req_finished(request);
1509                 if (rc == -ENOENT) {
1510                         clear_nlink(inode);
1511                         /* Unlinked special device node? Or just a race?
1512                          * Pretend we done everything. */
1513                         if (!S_ISREG(inode->i_mode) &&
1514                             !S_ISDIR(inode->i_mode)) {
1515                                 ia_valid = op_data->op_attr.ia_valid;
1516                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1517                                 rc = simple_setattr(dentry, &op_data->op_attr);
1518                                 op_data->op_attr.ia_valid = ia_valid;
1519                         }
1520                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1521                         CERROR("md_setattr fails: rc = %d\n", rc);
1522                 }
1523                 RETURN(rc);
1524         }
1525
1526         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1527                               sbi->ll_md_exp, &md);
1528         if (rc) {
1529                 ptlrpc_req_finished(request);
1530                 RETURN(rc);
1531         }
1532
1533         ia_valid = op_data->op_attr.ia_valid;
1534         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1535          * cache is not cleared yet. */
1536         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1537         rc = simple_setattr(dentry, &op_data->op_attr);
1538         op_data->op_attr.ia_valid = ia_valid;
1539
1540         /* Extract epoch data if obtained. */
1541         op_data->op_handle = md.body->mbo_handle;
1542         op_data->op_ioepoch = md.body->mbo_ioepoch;
1543
1544         rc = ll_update_inode(inode, &md);
1545         ptlrpc_req_finished(request);
1546
1547         RETURN(rc);
1548 }
1549
1550 /* Close IO epoch and send Size-on-MDS attribute update. */
1551 static int ll_setattr_done_writing(struct inode *inode,
1552                                    struct md_op_data *op_data,
1553                                    struct md_open_data *mod)
1554 {
1555         struct ll_inode_info *lli = ll_i2info(inode);
1556         int rc = 0;
1557         ENTRY;
1558
1559         LASSERT(op_data != NULL);
1560         if (!S_ISREG(inode->i_mode))
1561                 RETURN(0);
1562
1563         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1564                op_data->op_ioepoch, PFID(&lli->lli_fid));
1565
1566         op_data->op_flags = MF_EPOCH_CLOSE;
1567         ll_done_writing_attr(inode, op_data);
1568         ll_pack_inode2opdata(inode, op_data, NULL);
1569
1570         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1571         if (rc == -EAGAIN) {
1572                 /* MDS has instructed us to obtain Size-on-MDS attribute
1573                  * from OSTs and send setattr to back to MDS. */
1574                 rc = ll_som_update(inode, op_data);
1575         } else if (rc) {
1576                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1577                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1578                        PFID(ll_inode2fid(inode)), rc);
1579         }
1580         RETURN(rc);
1581 }
1582
1583 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1584 {
1585         struct obd_capa *capa;
1586         int rc;
1587
1588         if (attr->ia_valid & ATTR_SIZE)
1589                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1590         else
1591                 capa = ll_mdscapa_get(inode);
1592
1593         rc = cl_setattr_ost(inode, attr, capa);
1594
1595         if (attr->ia_valid & ATTR_SIZE)
1596                 ll_truncate_free_capa(capa);
1597         else
1598                 capa_put(capa);
1599
1600         return rc;
1601 }
1602
1603 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1604  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1605  * keep these values until such a time that objects are allocated for it.
1606  * We do the MDS operations first, as it is checking permissions for us.
1607  * We don't to the MDS RPC if there is nothing that we want to store there,
1608  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1609  * going to do an RPC anyways.
1610  *
1611  * If we are doing a truncate, we will send the mtime and ctime updates
1612  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1613  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1614  * at the same time.
1615  *
1616  * In case of HSMimport, we only set attr on MDS.
1617  */
1618 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1619 {
1620         struct inode *inode = dentry->d_inode;
1621         struct ll_inode_info *lli = ll_i2info(inode);
1622         struct md_op_data *op_data = NULL;
1623         struct md_open_data *mod = NULL;
1624         bool file_is_released = false;
1625         int rc = 0, rc1 = 0;
1626         ENTRY;
1627
1628         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1629                "valid %x, hsm_import %d\n",
1630                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1631                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1632                hsm_import);
1633
1634         if (attr->ia_valid & ATTR_SIZE) {
1635                 /* Check new size against VFS/VM file size limit and rlimit */
1636                 rc = inode_newsize_ok(inode, attr->ia_size);
1637                 if (rc)
1638                         RETURN(rc);
1639
1640                 /* The maximum Lustre file size is variable, based on the
1641                  * OST maximum object size and number of stripes.  This
1642                  * needs another check in addition to the VFS check above. */
1643                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1644                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1645                                PFID(&lli->lli_fid), attr->ia_size,
1646                                ll_file_maxbytes(inode));
1647                         RETURN(-EFBIG);
1648                 }
1649
1650                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1651         }
1652
1653         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1654         if (attr->ia_valid & TIMES_SET_FLAGS) {
1655                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1656                     !cfs_capable(CFS_CAP_FOWNER))
1657                         RETURN(-EPERM);
1658         }
1659
1660         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1661         if (attr->ia_valid & ATTR_CTIME) {
1662                 attr->ia_ctime = CFS_CURRENT_TIME;
1663                 attr->ia_valid |= ATTR_CTIME_SET;
1664         }
1665         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1666             (attr->ia_valid & ATTR_ATIME)) {
1667                 attr->ia_atime = CFS_CURRENT_TIME;
1668                 attr->ia_valid |= ATTR_ATIME_SET;
1669         }
1670         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1671             (attr->ia_valid & ATTR_MTIME)) {
1672                 attr->ia_mtime = CFS_CURRENT_TIME;
1673                 attr->ia_valid |= ATTR_MTIME_SET;
1674         }
1675
1676         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1677                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1678                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1679                        cfs_time_current_sec());
1680
1681         /* We always do an MDS RPC, even if we're only changing the size;
1682          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1683
1684         OBD_ALLOC_PTR(op_data);
1685         if (op_data == NULL)
1686                 RETURN(-ENOMEM);
1687
1688         if (!S_ISDIR(inode->i_mode)) {
1689                 if (attr->ia_valid & ATTR_SIZE)
1690                         inode_dio_write_done(inode);
1691                 mutex_unlock(&inode->i_mutex);
1692         }
1693
1694         /* truncate on a released file must failed with -ENODATA,
1695          * so size must not be set on MDS for released file
1696          * but other attributes must be set
1697          */
1698         if (S_ISREG(inode->i_mode)) {
1699                 struct lov_stripe_md *lsm;
1700                 __u32 gen;
1701
1702                 ll_layout_refresh(inode, &gen);
1703                 lsm = ccc_inode_lsm_get(inode);
1704                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1705                         file_is_released = true;
1706                 ccc_inode_lsm_put(inode, lsm);
1707
1708                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1709                         if (file_is_released) {
1710                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1711                                 if (rc < 0)
1712                                         GOTO(out, rc);
1713
1714                                 file_is_released = false;
1715                                 ll_layout_refresh(inode, &gen);
1716                         }
1717
1718                         /* If we are changing file size, file content is
1719                          * modified, flag it. */
1720                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1721                         spin_lock(&lli->lli_lock);
1722                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1723                         spin_unlock(&lli->lli_lock);
1724                         op_data->op_bias |= MDS_DATA_MODIFIED;
1725                 }
1726         }
1727
1728         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1729
1730         /* Open epoch for truncate. */
1731         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1732             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1733                 op_data->op_flags = MF_EPOCH_OPEN;
1734
1735         rc = ll_md_setattr(dentry, op_data, &mod);
1736         if (rc)
1737                 GOTO(out, rc);
1738
1739         /* RPC to MDT is sent, cancel data modification flag */
1740         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1741                 spin_lock(&lli->lli_lock);
1742                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1743                 spin_unlock(&lli->lli_lock);
1744         }
1745
1746         ll_ioepoch_open(lli, op_data->op_ioepoch);
1747         if (!S_ISREG(inode->i_mode) || file_is_released)
1748                 GOTO(out, rc = 0);
1749
1750         if (attr->ia_valid & (ATTR_SIZE |
1751                               ATTR_ATIME | ATTR_ATIME_SET |
1752                               ATTR_MTIME | ATTR_MTIME_SET)) {
1753                 /* For truncate and utimes sending attributes to OSTs, setting
1754                  * mtime/atime to the past will be performed under PW [0:EOF]
1755                  * extent lock (new_size:EOF for truncate).  It may seem
1756                  * excessive to send mtime/atime updates to OSTs when not
1757                  * setting times to past, but it is necessary due to possible
1758                  * time de-synchronization between MDT inode and OST objects */
1759                 if (attr->ia_valid & ATTR_SIZE)
1760                         down_write(&lli->lli_trunc_sem);
1761                 rc = ll_setattr_ost(inode, attr);
1762                 if (attr->ia_valid & ATTR_SIZE)
1763                         up_write(&lli->lli_trunc_sem);
1764         }
1765         EXIT;
1766 out:
1767         if (op_data) {
1768                 if (op_data->op_ioepoch) {
1769                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1770                         if (!rc)
1771                                 rc = rc1;
1772                 }
1773                 ll_finish_md_op_data(op_data);
1774         }
1775         if (!S_ISDIR(inode->i_mode)) {
1776                 mutex_lock(&inode->i_mutex);
1777                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1778                         inode_dio_wait(inode);
1779         }
1780
1781         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1782                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1783
1784         return rc;
1785 }
1786
1787 int ll_setattr(struct dentry *de, struct iattr *attr)
1788 {
1789         int mode = de->d_inode->i_mode;
1790
1791         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1792                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1793                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1794
1795         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1796                                (ATTR_SIZE|ATTR_MODE)) &&
1797             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1798              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1799               !(attr->ia_mode & S_ISGID))))
1800                 attr->ia_valid |= ATTR_FORCE;
1801
1802         if ((attr->ia_valid & ATTR_MODE) &&
1803             (mode & S_ISUID) &&
1804             !(attr->ia_mode & S_ISUID) &&
1805             !(attr->ia_valid & ATTR_KILL_SUID))
1806                 attr->ia_valid |= ATTR_KILL_SUID;
1807
1808         if ((attr->ia_valid & ATTR_MODE) &&
1809             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1810             !(attr->ia_mode & S_ISGID) &&
1811             !(attr->ia_valid & ATTR_KILL_SGID))
1812                 attr->ia_valid |= ATTR_KILL_SGID;
1813
1814         return ll_setattr_raw(de, attr, false);
1815 }
1816
1817 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1818                        __u64 max_age, __u32 flags)
1819 {
1820         struct ll_sb_info *sbi = ll_s2sbi(sb);
1821         struct obd_statfs obd_osfs;
1822         int rc;
1823         ENTRY;
1824
1825         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1826         if (rc) {
1827                 CERROR("md_statfs fails: rc = %d\n", rc);
1828                 RETURN(rc);
1829         }
1830
1831         osfs->os_type = sb->s_magic;
1832
1833         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1834                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1835
1836         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1837                 flags |= OBD_STATFS_NODELAY;
1838
1839         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1840         if (rc) {
1841                 CERROR("obd_statfs fails: rc = %d\n", rc);
1842                 RETURN(rc);
1843         }
1844
1845         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1846                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1847                obd_osfs.os_files);
1848
1849         osfs->os_bsize = obd_osfs.os_bsize;
1850         osfs->os_blocks = obd_osfs.os_blocks;
1851         osfs->os_bfree = obd_osfs.os_bfree;
1852         osfs->os_bavail = obd_osfs.os_bavail;
1853
1854         /* If we don't have as many objects free on the OST as inodes
1855          * on the MDS, we reduce the total number of inodes to
1856          * compensate, so that the "inodes in use" number is correct.
1857          */
1858         if (obd_osfs.os_ffree < osfs->os_ffree) {
1859                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1860                         obd_osfs.os_ffree;
1861                 osfs->os_ffree = obd_osfs.os_ffree;
1862         }
1863
1864         RETURN(rc);
1865 }
1866 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1867 {
1868         struct super_block *sb = de->d_sb;
1869         struct obd_statfs osfs;
1870         __u64 fsid = huge_encode_dev(sb->s_dev);
1871         int rc;
1872
1873         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1874         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1875
1876         /* Some amount of caching on the client is allowed */
1877         rc = ll_statfs_internal(sb, &osfs,
1878                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1879                                 0);
1880         if (rc)
1881                 return rc;
1882
1883         statfs_unpack(sfs, &osfs);
1884
1885         /* We need to downshift for all 32-bit kernels, because we can't
1886          * tell if the kernel is being called via sys_statfs64() or not.
1887          * Stop before overflowing f_bsize - in which case it is better
1888          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1889         if (sizeof(long) < 8) {
1890                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1891                         sfs->f_bsize <<= 1;
1892
1893                         osfs.os_blocks >>= 1;
1894                         osfs.os_bfree >>= 1;
1895                         osfs.os_bavail >>= 1;
1896                 }
1897         }
1898
1899         sfs->f_blocks = osfs.os_blocks;
1900         sfs->f_bfree = osfs.os_bfree;
1901         sfs->f_bavail = osfs.os_bavail;
1902         sfs->f_fsid.val[0] = (__u32)fsid;
1903         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1904         return 0;
1905 }
1906
1907 void ll_inode_size_lock(struct inode *inode)
1908 {
1909         struct ll_inode_info *lli;
1910
1911         LASSERT(!S_ISDIR(inode->i_mode));
1912
1913         lli = ll_i2info(inode);
1914         mutex_lock(&lli->lli_size_mutex);
1915 }
1916
1917 void ll_inode_size_unlock(struct inode *inode)
1918 {
1919         struct ll_inode_info *lli;
1920
1921         lli = ll_i2info(inode);
1922         mutex_unlock(&lli->lli_size_mutex);
1923 }
1924
1925 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1926 {
1927         struct ll_inode_info *lli = ll_i2info(inode);
1928         struct mdt_body *body = md->body;
1929         struct lov_stripe_md *lsm = md->lsm;
1930         struct ll_sb_info *sbi = ll_i2sbi(inode);
1931
1932         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1933         if (lsm != NULL) {
1934                 if (!lli->lli_has_smd &&
1935                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1936                         cl_file_inode_init(inode, md);
1937
1938                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1939                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1940                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1941         }
1942
1943         if (S_ISDIR(inode->i_mode)) {
1944                 int     rc;
1945
1946                 rc = ll_update_lsm_md(inode, md);
1947                 if (rc != 0)
1948                         return rc;
1949         }
1950
1951         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1952                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1953                         ll_update_remote_perm(inode, md->remote_perm);
1954         }
1955 #ifdef CONFIG_FS_POSIX_ACL
1956         else if (body->mbo_valid & OBD_MD_FLACL) {
1957                 spin_lock(&lli->lli_lock);
1958                 if (lli->lli_posix_acl)
1959                         posix_acl_release(lli->lli_posix_acl);
1960                 lli->lli_posix_acl = md->posix_acl;
1961                 spin_unlock(&lli->lli_lock);
1962         }
1963 #endif
1964         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1965                                         sbi->ll_flags & LL_SBI_32BIT_API);
1966         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1967
1968         if (body->mbo_valid & OBD_MD_FLATIME) {
1969                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1970                         LTIME_S(inode->i_atime) = body->mbo_atime;
1971                 lli->lli_lvb.lvb_atime = body->mbo_atime;
1972         }
1973
1974         if (body->mbo_valid & OBD_MD_FLMTIME) {
1975                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1976                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1977                                "to "LPU64"\n", inode->i_ino,
1978                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1979                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1980                 }
1981                 lli->lli_lvb.lvb_mtime = body->mbo_mtime;
1982         }
1983
1984         if (body->mbo_valid & OBD_MD_FLCTIME) {
1985                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1986                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1987                 lli->lli_lvb.lvb_ctime = body->mbo_ctime;
1988         }
1989
1990         if (body->mbo_valid & OBD_MD_FLMODE)
1991                 inode->i_mode = (inode->i_mode & S_IFMT) |
1992                                 (body->mbo_mode & ~S_IFMT);
1993
1994         if (body->mbo_valid & OBD_MD_FLTYPE)
1995                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1996                                 (body->mbo_mode & S_IFMT);
1997
1998         LASSERT(inode->i_mode != 0);
1999         if (S_ISREG(inode->i_mode))
2000                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2001                                        LL_MAX_BLKSIZE_BITS);
2002         else
2003                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2004
2005         if (body->mbo_valid & OBD_MD_FLUID)
2006                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2007         if (body->mbo_valid & OBD_MD_FLGID)
2008                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2009         if (body->mbo_valid & OBD_MD_FLFLAGS)
2010                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
2011         if (body->mbo_valid & OBD_MD_FLNLINK)
2012                 set_nlink(inode, body->mbo_nlink);
2013         if (body->mbo_valid & OBD_MD_FLRDEV)
2014                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2015
2016         if (body->mbo_valid & OBD_MD_FLID) {
2017                 /* FID shouldn't be changed! */
2018                 if (fid_is_sane(&lli->lli_fid)) {
2019                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2020                                  "Trying to change FID "DFID
2021                                  " to the "DFID", inode "DFID"(%p)\n",
2022                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2023                                  PFID(ll_inode2fid(inode)), inode);
2024                 } else {
2025                         lli->lli_fid = body->mbo_fid1;
2026                 }
2027         }
2028
2029         LASSERT(fid_seq(&lli->lli_fid) != 0);
2030
2031         if (body->mbo_valid & OBD_MD_FLSIZE) {
2032                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2033                     S_ISREG(inode->i_mode)) {
2034                         struct lustre_handle lockh;
2035                         ldlm_mode_t mode;
2036
2037                         /* As it is possible a blocking ast has been processed
2038                          * by this time, we need to check there is an UPDATE
2039                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2040                          * it. */
2041                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2042                                                &lockh, LDLM_FL_CBPENDING,
2043                                                LCK_CR | LCK_CW |
2044                                                LCK_PR | LCK_PW);
2045                         if (mode) {
2046                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2047                                                       LLIF_EPOCH_PENDING |
2048                                                       LLIF_SOM_DIRTY)) {
2049                                         CERROR("%s: inode "DFID" flags %u still"
2050                                                " has size authority! do not "
2051                                                "trust the size from MDS\n",
2052                                                sbi->ll_md_exp->exp_obd->obd_name,
2053                                                PFID(ll_inode2fid(inode)),
2054                                                lli->lli_flags);
2055                                 } else {
2056                                         /* Use old size assignment to avoid
2057                                          * deadlock bz14138 & bz14326 */
2058                                         i_size_write(inode, body->mbo_size);
2059                                         spin_lock(&lli->lli_lock);
2060                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2061                                         spin_unlock(&lli->lli_lock);
2062                                 }
2063                                 ldlm_lock_decref(&lockh, mode);
2064                         }
2065                 } else {
2066                         /* Use old size assignment to avoid
2067                          * deadlock bz14138 & bz14326 */
2068                         i_size_write(inode, body->mbo_size);
2069
2070                         CDEBUG(D_VFSTRACE,
2071                                "inode="DFID", updating i_size %llu\n",
2072                                PFID(ll_inode2fid(inode)),
2073                                (unsigned long long)body->mbo_size);
2074                 }
2075
2076                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2077                         inode->i_blocks = body->mbo_blocks;
2078         }
2079
2080         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2081                 LASSERT(md->mds_capa);
2082                 ll_add_capa(inode, md->mds_capa);
2083         }
2084
2085         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2086                 LASSERT(md->oss_capa);
2087                 ll_add_capa(inode, md->oss_capa);
2088         }
2089
2090         if (body->mbo_valid & OBD_MD_TSTATE) {
2091                 if (body->mbo_t_state & MS_RESTORE)
2092                         lli->lli_flags |= LLIF_FILE_RESTORING;
2093         }
2094
2095         return 0;
2096 }
2097
2098 int ll_read_inode2(struct inode *inode, void *opaque)
2099 {
2100         struct lustre_md *md = opaque;
2101         struct ll_inode_info *lli = ll_i2info(inode);
2102         int     rc;
2103         ENTRY;
2104
2105         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2106                PFID(&lli->lli_fid), inode);
2107
2108         LASSERT(!lli->lli_has_smd);
2109
2110         /* Core attributes from the MDS first.  This is a new inode, and
2111          * the VFS doesn't zero times in the core inode so we have to do
2112          * it ourselves.  They will be overwritten by either MDS or OST
2113          * attributes - we just need to make sure they aren't newer. */
2114         LTIME_S(inode->i_mtime) = 0;
2115         LTIME_S(inode->i_atime) = 0;
2116         LTIME_S(inode->i_ctime) = 0;
2117         inode->i_rdev = 0;
2118         rc = ll_update_inode(inode, md);
2119         if (rc != 0)
2120                 RETURN(rc);
2121
2122         /* OIDEBUG(inode); */
2123
2124         /* initializing backing dev info. */
2125         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2126
2127
2128         if (S_ISREG(inode->i_mode)) {
2129                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2130                 inode->i_op = &ll_file_inode_operations;
2131                 inode->i_fop = sbi->ll_fop;
2132                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2133                 EXIT;
2134         } else if (S_ISDIR(inode->i_mode)) {
2135                 inode->i_op = &ll_dir_inode_operations;
2136                 inode->i_fop = &ll_dir_operations;
2137                 EXIT;
2138         } else if (S_ISLNK(inode->i_mode)) {
2139                 inode->i_op = &ll_fast_symlink_inode_operations;
2140                 EXIT;
2141         } else {
2142                 inode->i_op = &ll_special_inode_operations;
2143
2144                 init_special_inode(inode, inode->i_mode,
2145                                    inode->i_rdev);
2146
2147                 EXIT;
2148         }
2149
2150         return 0;
2151 }
2152
2153 void ll_delete_inode(struct inode *inode)
2154 {
2155         struct cl_inode_info *lli = cl_i2info(inode);
2156         ENTRY;
2157
2158         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2159                 /* discard all dirty pages before truncating them, required by
2160                  * osc_extent implementation at LU-1030. */
2161                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2162                                    CL_FSYNC_DISCARD, 1);
2163
2164         truncate_inode_pages(&inode->i_data, 0);
2165
2166         /* Workaround for LU-118 */
2167         if (inode->i_data.nrpages) {
2168                 spin_lock_irq(&inode->i_data.tree_lock);
2169                 spin_unlock_irq(&inode->i_data.tree_lock);
2170                 LASSERTF(inode->i_data.nrpages == 0,
2171                          "inode="DFID"(%p) nrpages=%lu, see "
2172                          "http://jira.whamcloud.com/browse/LU-118\n",
2173                          PFID(ll_inode2fid(inode)), inode,
2174                          inode->i_data.nrpages);
2175         }
2176         /* Workaround end */
2177
2178 #ifdef HAVE_SBOPS_EVICT_INODE
2179         ll_clear_inode(inode);
2180 #endif
2181         clear_inode(inode);
2182
2183         EXIT;
2184 }
2185
2186 int ll_iocontrol(struct inode *inode, struct file *file,
2187                  unsigned int cmd, unsigned long arg)
2188 {
2189         struct ll_sb_info *sbi = ll_i2sbi(inode);
2190         struct ptlrpc_request *req = NULL;
2191         int rc, flags = 0;
2192         ENTRY;
2193
2194         switch(cmd) {
2195         case FSFILT_IOC_GETFLAGS: {
2196                 struct mdt_body *body;
2197                 struct md_op_data *op_data;
2198
2199                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2200                                              0, 0, LUSTRE_OPC_ANY,
2201                                              NULL);
2202                 if (IS_ERR(op_data))
2203                         RETURN(PTR_ERR(op_data));
2204
2205                 op_data->op_valid = OBD_MD_FLFLAGS;
2206                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2207                 ll_finish_md_op_data(op_data);
2208                 if (rc) {
2209                         CERROR("%s: failure inode "DFID": rc = %d\n",
2210                                sbi->ll_md_exp->exp_obd->obd_name,
2211                                PFID(ll_inode2fid(inode)), rc);
2212                         RETURN(-abs(rc));
2213                 }
2214
2215                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2216
2217                 flags = body->mbo_flags;
2218
2219                 ptlrpc_req_finished(req);
2220
2221                 RETURN(put_user(flags, (int *)arg));
2222         }
2223         case FSFILT_IOC_SETFLAGS: {
2224                 struct lov_stripe_md *lsm;
2225                 struct obd_info oinfo = { { { 0 } } };
2226                 struct md_op_data *op_data;
2227
2228                 if (get_user(flags, (int *)arg))
2229                         RETURN(-EFAULT);
2230
2231                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2232                                              LUSTRE_OPC_ANY, NULL);
2233                 if (IS_ERR(op_data))
2234                         RETURN(PTR_ERR(op_data));
2235
2236                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2237                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2238                 rc = md_setattr(sbi->ll_md_exp, op_data,
2239                                 NULL, 0, NULL, 0, &req, NULL);
2240                 ll_finish_md_op_data(op_data);
2241                 ptlrpc_req_finished(req);
2242                 if (rc)
2243                         RETURN(rc);
2244
2245                 inode->i_flags = ll_ext_to_inode_flags(flags);
2246
2247                 lsm = ccc_inode_lsm_get(inode);
2248                 if (!lsm_has_objects(lsm)) {
2249                         ccc_inode_lsm_put(inode, lsm);
2250                         RETURN(0);
2251                 }
2252
2253                 OBDO_ALLOC(oinfo.oi_oa);
2254                 if (!oinfo.oi_oa) {
2255                         ccc_inode_lsm_put(inode, lsm);
2256                         RETURN(-ENOMEM);
2257                 }
2258                 oinfo.oi_md = lsm;
2259                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2260                 oinfo.oi_oa->o_flags = flags;
2261                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2262                                        OBD_MD_FLGROUP;
2263                 oinfo.oi_capa = ll_mdscapa_get(inode);
2264                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2265                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2266                 capa_put(oinfo.oi_capa);
2267                 OBDO_FREE(oinfo.oi_oa);
2268                 ccc_inode_lsm_put(inode, lsm);
2269
2270                 if (rc && rc != -EPERM && rc != -EACCES)
2271                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2272
2273                 RETURN(rc);
2274         }
2275         default:
2276                 RETURN(-ENOSYS);
2277         }
2278
2279         RETURN(0);
2280 }
2281
2282 int ll_flush_ctx(struct inode *inode)
2283 {
2284         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2285
2286         CDEBUG(D_SEC, "flush context for user %d\n",
2287                from_kuid(&init_user_ns, current_uid()));
2288
2289         obd_set_info_async(NULL, sbi->ll_md_exp,
2290                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2291                            0, NULL, NULL);
2292         obd_set_info_async(NULL, sbi->ll_dt_exp,
2293                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2294                            0, NULL, NULL);
2295         return 0;
2296 }
2297
2298 /* umount -f client means force down, don't save state */
2299 void ll_umount_begin(struct super_block *sb)
2300 {
2301         struct ll_sb_info *sbi = ll_s2sbi(sb);
2302         struct obd_device *obd;
2303         struct obd_ioctl_data *ioc_data;
2304         ENTRY;
2305
2306         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2307                sb->s_count, atomic_read(&sb->s_active));
2308
2309         obd = class_exp2obd(sbi->ll_md_exp);
2310         if (obd == NULL) {
2311                 CERROR("Invalid MDC connection handle "LPX64"\n",
2312                        sbi->ll_md_exp->exp_handle.h_cookie);
2313                 EXIT;
2314                 return;
2315         }
2316         obd->obd_force = 1;
2317
2318         obd = class_exp2obd(sbi->ll_dt_exp);
2319         if (obd == NULL) {
2320                 CERROR("Invalid LOV connection handle "LPX64"\n",
2321                        sbi->ll_dt_exp->exp_handle.h_cookie);
2322                 EXIT;
2323                 return;
2324         }
2325         obd->obd_force = 1;
2326
2327         OBD_ALLOC_PTR(ioc_data);
2328         if (ioc_data) {
2329                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2330                               sizeof *ioc_data, ioc_data, NULL);
2331
2332                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2333                               sizeof *ioc_data, ioc_data, NULL);
2334
2335                 OBD_FREE_PTR(ioc_data);
2336         }
2337
2338         /* Really, we'd like to wait until there are no requests outstanding,
2339          * and then continue.  For now, we just invalidate the requests,
2340          * schedule() and sleep one second if needed, and hope.
2341          */
2342         schedule();
2343         EXIT;
2344 }
2345
2346 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2347 {
2348         struct ll_sb_info *sbi = ll_s2sbi(sb);
2349         char *profilenm = get_profile_name(sb);
2350         int err;
2351         __u32 read_only;
2352
2353         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2354                 read_only = *flags & MS_RDONLY;
2355                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2356                                          sizeof(KEY_READ_ONLY),
2357                                          KEY_READ_ONLY, sizeof(read_only),
2358                                          &read_only, NULL);
2359                 if (err) {
2360                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2361                                       profilenm, read_only ?
2362                                       "read-only" : "read-write", err);
2363                         return err;
2364                 }
2365
2366                 if (read_only)
2367                         sb->s_flags |= MS_RDONLY;
2368                 else
2369                         sb->s_flags &= ~MS_RDONLY;
2370
2371                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2372                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2373                                       read_only ?  "read-only" : "read-write");
2374         }
2375         return 0;
2376 }
2377
2378 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2379                   struct super_block *sb, struct lookup_intent *it)
2380 {
2381         struct ll_sb_info *sbi = NULL;
2382         struct lustre_md md = { 0 };
2383         int rc;
2384         ENTRY;
2385
2386         LASSERT(*inode || sb);
2387         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2388         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2389                               sbi->ll_md_exp, &md);
2390         if (rc)
2391                 RETURN(rc);
2392
2393         if (*inode) {
2394                 rc = ll_update_inode(*inode, &md);
2395                 if (rc != 0)
2396                         GOTO(out, rc);
2397         } else {
2398                 LASSERT(sb != NULL);
2399
2400                 /*
2401                  * At this point server returns to client's same fid as client
2402                  * generated for creating. So using ->fid1 is okay here.
2403                  */
2404                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2405
2406                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2407                                              sbi->ll_flags & LL_SBI_32BIT_API),
2408                                  &md);
2409                 if (IS_ERR(*inode)) {
2410 #ifdef CONFIG_FS_POSIX_ACL
2411                         if (md.posix_acl) {
2412                                 posix_acl_release(md.posix_acl);
2413                                 md.posix_acl = NULL;
2414                         }
2415 #endif
2416                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2417                         *inode = NULL;
2418                         CERROR("new_inode -fatal: rc %d\n", rc);
2419                         GOTO(out, rc);
2420                 }
2421         }
2422
2423         /* Handling piggyback layout lock.
2424          * Layout lock can be piggybacked by getattr and open request.
2425          * The lsm can be applied to inode only if it comes with a layout lock
2426          * otherwise correct layout may be overwritten, for example:
2427          * 1. proc1: mdt returns a lsm but not granting layout
2428          * 2. layout was changed by another client
2429          * 3. proc2: refresh layout and layout lock granted
2430          * 4. proc1: to apply a stale layout */
2431         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2432                 struct lustre_handle lockh;
2433                 struct ldlm_lock *lock;
2434
2435                 lockh.cookie = it->d.lustre.it_lock_handle;
2436                 lock = ldlm_handle2lock(&lockh);
2437                 LASSERT(lock != NULL);
2438                 if (ldlm_has_layout(lock)) {
2439                         struct cl_object_conf conf;
2440
2441                         memset(&conf, 0, sizeof(conf));
2442                         conf.coc_opc = OBJECT_CONF_SET;
2443                         conf.coc_inode = *inode;
2444                         conf.coc_lock = lock;
2445                         conf.u.coc_md = &md;
2446                         (void)ll_layout_conf(*inode, &conf);
2447                 }
2448                 LDLM_LOCK_PUT(lock);
2449         }
2450
2451 out:
2452         if (md.lsm != NULL)
2453                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2454         md_free_lustre_md(sbi->ll_md_exp, &md);
2455         RETURN(rc);
2456 }
2457
2458 int ll_obd_statfs(struct inode *inode, void *arg)
2459 {
2460         struct ll_sb_info *sbi = NULL;
2461         struct obd_export *exp;
2462         char *buf = NULL;
2463         struct obd_ioctl_data *data = NULL;
2464         __u32 type;
2465         __u32 flags;
2466         int len = 0, rc;
2467
2468         if (!inode || !(sbi = ll_i2sbi(inode)))
2469                 GOTO(out_statfs, rc = -EINVAL);
2470
2471         rc = obd_ioctl_getdata(&buf, &len, arg);
2472         if (rc)
2473                 GOTO(out_statfs, rc);
2474
2475         data = (void*)buf;
2476         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2477             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2478                 GOTO(out_statfs, rc = -EINVAL);
2479
2480         if (data->ioc_inllen1 != sizeof(__u32) ||
2481             data->ioc_inllen2 != sizeof(__u32) ||
2482             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2483             data->ioc_plen2 != sizeof(struct obd_uuid))
2484                 GOTO(out_statfs, rc = -EINVAL);
2485
2486         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2487         if (type & LL_STATFS_LMV)
2488                 exp = sbi->ll_md_exp;
2489         else if (type & LL_STATFS_LOV)
2490                 exp = sbi->ll_dt_exp;
2491         else
2492                 GOTO(out_statfs, rc = -ENODEV);
2493
2494         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2495         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2496         if (rc)
2497                 GOTO(out_statfs, rc);
2498 out_statfs:
2499         if (buf)
2500                 obd_ioctl_freedata(buf, len);
2501         return rc;
2502 }
2503
2504 int ll_process_config(struct lustre_cfg *lcfg)
2505 {
2506         struct super_block *sb;
2507         unsigned long x;
2508         int rc = 0;
2509         char *ptr;
2510
2511         /* The instance name contains the sb: lustre-client-aacfe000 */
2512         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2513         if (!ptr || !*(++ptr))
2514                 return -EINVAL;
2515         if (sscanf(ptr, "%lx", &x) != 1)
2516                 return -EINVAL;
2517         sb = (struct super_block *)x;
2518         /* This better be a real Lustre superblock! */
2519         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2520
2521         /* Note we have not called client_common_fill_super yet, so
2522            proc fns must be able to handle that! */
2523         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2524                                           lcfg, sb);
2525         if (rc > 0)
2526                 rc = 0;
2527         return rc;
2528 }
2529
2530 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2531 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2532                                        struct inode *i1, struct inode *i2,
2533                                        const char *name, int namelen,
2534                                        int mode, __u32 opc, void *data)
2535 {
2536         LASSERT(i1 != NULL);
2537
2538         if (name == NULL) {
2539                 /* Do not reuse namelen for something else. */
2540                 if (namelen != 0)
2541                         return ERR_PTR(-EINVAL);
2542         } else {
2543                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2544                         return ERR_PTR(-ENAMETOOLONG);
2545
2546                 if (!lu_name_is_valid_2(name, namelen))
2547                         return ERR_PTR(-EINVAL);
2548         }
2549
2550         if (op_data == NULL)
2551                 OBD_ALLOC_PTR(op_data);
2552
2553         if (op_data == NULL)
2554                 return ERR_PTR(-ENOMEM);
2555
2556         ll_i2gids(op_data->op_suppgids, i1, i2);
2557         op_data->op_fid1 = *ll_inode2fid(i1);
2558         op_data->op_capa1 = ll_mdscapa_get(i1);
2559         if (S_ISDIR(i1->i_mode))
2560                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2561
2562         if (i2) {
2563                 op_data->op_fid2 = *ll_inode2fid(i2);
2564                 op_data->op_capa2 = ll_mdscapa_get(i2);
2565                 if (S_ISDIR(i2->i_mode))
2566                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2567         } else {
2568                 fid_zero(&op_data->op_fid2);
2569                 op_data->op_capa2 = NULL;
2570         }
2571
2572         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2573                 op_data->op_cli_flags |= CLI_HASH64;
2574
2575         if (ll_need_32bit_api(ll_i2sbi(i1)))
2576                 op_data->op_cli_flags |= CLI_API32;
2577
2578         op_data->op_name = name;
2579         op_data->op_namelen = namelen;
2580         op_data->op_mode = mode;
2581         op_data->op_mod_time = cfs_time_current_sec();
2582         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2583         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2584         op_data->op_cap = cfs_curproc_cap_pack();
2585         op_data->op_bias = 0;
2586         op_data->op_cli_flags = 0;
2587         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2588              filename_is_volatile(name, namelen, NULL))
2589                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2590         op_data->op_mds = 0;
2591         op_data->op_data = data;
2592
2593         /* If the file is being opened after mknod() (normally due to NFS)
2594          * try to use the default stripe data from parent directory for
2595          * allocating OST objects.  Try to pass the parent FID to MDS. */
2596         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2597             !ll_i2info(i2)->lli_has_smd) {
2598                 struct ll_inode_info *lli = ll_i2info(i2);
2599
2600                 spin_lock(&lli->lli_lock);
2601                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2602                         op_data->op_fid1 = lli->lli_pfid;
2603                 spin_unlock(&lli->lli_lock);
2604                 /** We ignore parent's capability temporary. */
2605         }
2606
2607         /* When called by ll_setattr_raw, file is i1. */
2608         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2609                 op_data->op_bias |= MDS_DATA_MODIFIED;
2610
2611         return op_data;
2612 }
2613
2614 void ll_finish_md_op_data(struct md_op_data *op_data)
2615 {
2616         capa_put(op_data->op_capa1);
2617         capa_put(op_data->op_capa2);
2618         OBD_FREE_PTR(op_data);
2619 }
2620
2621 #ifdef HAVE_SUPEROPS_USE_DENTRY
2622 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2623 #else
2624 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2625 #endif
2626 {
2627         struct ll_sb_info *sbi;
2628
2629 #ifdef HAVE_SUPEROPS_USE_DENTRY
2630         LASSERT((seq != NULL) && (dentry != NULL));
2631         sbi = ll_s2sbi(dentry->d_sb);
2632 #else
2633         LASSERT((seq != NULL) && (vfs != NULL));
2634         sbi = ll_s2sbi(vfs->mnt_sb);
2635 #endif
2636
2637         if (sbi->ll_flags & LL_SBI_NOLCK)
2638                 seq_puts(seq, ",nolock");
2639
2640         if (sbi->ll_flags & LL_SBI_FLOCK)
2641                 seq_puts(seq, ",flock");
2642
2643         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2644                 seq_puts(seq, ",localflock");
2645
2646         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2647                 seq_puts(seq, ",user_xattr");
2648
2649         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2650                 seq_puts(seq, ",lazystatfs");
2651
2652         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2653                 seq_puts(seq, ",user_fid2path");
2654
2655         RETURN(0);
2656 }
2657
2658 /**
2659  * Get obd name by cmd, and copy out to user space
2660  */
2661 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2662 {
2663         struct ll_sb_info *sbi = ll_i2sbi(inode);
2664         struct obd_device *obd;
2665         ENTRY;
2666
2667         if (cmd == OBD_IOC_GETDTNAME)
2668                 obd = class_exp2obd(sbi->ll_dt_exp);
2669         else if (cmd == OBD_IOC_GETMDNAME)
2670                 obd = class_exp2obd(sbi->ll_md_exp);
2671         else
2672                 RETURN(-EINVAL);
2673
2674         if (!obd)
2675                 RETURN(-ENOENT);
2676
2677         if (copy_to_user((void *)arg, obd->obd_name,
2678                          strlen(obd->obd_name) + 1))
2679                 RETURN(-EFAULT);
2680
2681         RETURN(0);
2682 }
2683
2684 /**
2685  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2686  * fsname will be returned in this buffer; otherwise, a static buffer will be
2687  * used to store the fsname and returned to caller.
2688  */
2689 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2690 {
2691         static char fsname_static[MTI_NAME_MAXLEN];
2692         struct lustre_sb_info *lsi = s2lsi(sb);
2693         char *ptr;
2694         int len;
2695
2696         if (buf == NULL) {
2697                 /* this means the caller wants to use static buffer
2698                  * and it doesn't care about race. Usually this is
2699                  * in error reporting path */
2700                 buf = fsname_static;
2701                 buflen = sizeof(fsname_static);
2702         }
2703
2704         len = strlen(lsi->lsi_lmd->lmd_profile);
2705         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2706         if (ptr && (strcmp(ptr, "-client") == 0))
2707                 len -= 7;
2708
2709         if (unlikely(len >= buflen))
2710                 len = buflen - 1;
2711         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2712         buf[len] = '\0';
2713
2714         return buf;
2715 }
2716
2717 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2718 {
2719         char *path = NULL;
2720
2721         struct path p;
2722
2723         p.dentry = dentry;
2724         p.mnt = current->fs->root.mnt;
2725         path_get(&p);
2726         path = d_path(&p, buf, bufsize);
2727         path_put(&p);
2728         return path;
2729 }
2730
2731 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2732 {
2733         char *buf, *path = NULL;
2734         struct dentry *dentry = NULL;
2735         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2736
2737         /* this can be called inside spin lock so use GFP_ATOMIC. */
2738         buf = (char *)__get_free_page(GFP_ATOMIC);
2739         if (buf != NULL) {
2740                 dentry = d_find_alias(page->mapping->host);
2741                 if (dentry != NULL)
2742                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2743         }
2744
2745         CDEBUG(D_WARNING,
2746                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2747                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2748                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2749                PFID(&obj->cob_header.coh_lu.loh_fid),
2750                (path && !IS_ERR(path)) ? path : "", ioret);
2751
2752         if (dentry != NULL)
2753                 dput(dentry);
2754
2755         if (buf != NULL)
2756                 free_page((unsigned long)buf);
2757 }
2758
2759 /*
2760  * Compute llite root squash state after a change of root squash
2761  * configuration setting or add/remove of a lnet nid
2762  */
2763 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2764 {
2765         struct root_squash_info *squash = &sbi->ll_squash;
2766         int i;
2767         bool matched;
2768         lnet_process_id_t id;
2769
2770         /* Update norootsquash flag */
2771         down_write(&squash->rsi_sem);
2772         if (list_empty(&squash->rsi_nosquash_nids))
2773                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2774         else {
2775                 /* Do not apply root squash as soon as one of our NIDs is
2776                  * in the nosquash_nids list */
2777                 matched = false;
2778                 i = 0;
2779                 while (LNetGetId(i++, &id) != -ENOENT) {
2780                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2781                                 continue;
2782                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2783                                 matched = true;
2784                                 break;
2785                         }
2786                 }
2787                 if (matched)
2788                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2789                 else
2790                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2791         }
2792         up_write(&squash->rsi_sem);
2793 }
2794
2795