lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #ifdef HAVE_UIDGID_HEADER
  50 # include <linux/uidgid.h>
  51 #endif
  52
  53 #include <lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC_PTR(sbi);
  81         if (sbi == NULL)
  82                 RETURN(NULL);
  83
  84         spin_lock_init(&sbi->ll_lock);
  85         mutex_init(&sbi->ll_lco.lco_lock);
  86         spin_lock_init(&sbi->ll_pp_extent_lock);
  87         spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         lru_page_max = pages / 2;
  93
  94         /* initialize ll_cache data */
  95         sbi->ll_cache = cl_cache_init(lru_page_max);
  96         if (sbi->ll_cache == NULL) {
  97                 OBD_FREE(sbi, sizeof(*sbi));
  98                 RETURN(NULL);
  99         }
 100
 101         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 102                                            SBI_DEFAULT_READAHEAD_MAX);
 103         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 104         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 105                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 106         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 107         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 108
 109         ll_generate_random_uuid(uuid);
 110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 112
 113         sbi->ll_flags |= LL_SBI_VERBOSE;
 114 #ifdef ENABLE_CHECKSUM
 115         sbi->ll_flags |= LL_SBI_CHECKSUM;
 116 #endif
 117
 118 #ifdef HAVE_LRU_RESIZE_SUPPORT
 119         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 120 #endif
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136
 137         /* root squash */
 138         sbi->ll_squash.rsi_uid = 0;
 139         sbi->ll_squash.rsi_gid = 0;
 140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 141         init_rwsem(&sbi->ll_squash.rsi_sem);
 142
 143         RETURN(sbi);
 144 }
 145
 146 static void ll_free_sbi(struct super_block *sb)
 147 {
 148         struct ll_sb_info *sbi = ll_s2sbi(sb);
 149         ENTRY;
 150
 151         if (sbi != NULL) {
 152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 154                 if (sbi->ll_cache != NULL) {
 155                         cl_cache_decref(sbi->ll_cache);
 156                         sbi->ll_cache = NULL;
 157                 }
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 164                                     struct vfsmount *mnt)
 165 {
 166         struct inode *root = NULL;
 167         struct ll_sb_info *sbi = ll_s2sbi(sb);
 168         struct obd_device *obd;
 169         struct obd_statfs *osfs = NULL;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct md_op_data *op_data;
 174         struct lustre_md lmd;
 175         u64 valid;
 176         int size, err, checksum;
 177         ENTRY;
 178
 179         obd = class_name2obd(md);
 180         if (!obd) {
 181                 CERROR("MD %s: not setup or attached\n", md);
 182                 RETURN(-EINVAL);
 183         }
 184
 185         OBD_ALLOC_PTR(data);
 186         if (data == NULL)
 187                 RETURN(-ENOMEM);
 188
 189         OBD_ALLOC_PTR(osfs);
 190         if (osfs == NULL) {
 191                 OBD_FREE_PTR(data);
 192                 RETURN(-ENOMEM);
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 204                                   OBD_CONNECT_EINPROGRESS |
 205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 207                                   OBD_CONNECT_MAX_EASIZE |
 208                                   OBD_CONNECT_FLOCK_DEAD |
 209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 210                                   OBD_CONNECT_OPEN_BY_FID |
 211                                   OBD_CONNECT_DIR_STRIPE |
 212                                   OBD_CONNECT_BULK_MBITS;
 213
 214 #ifdef HAVE_LRU_RESIZE_SUPPORT
 215         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 216                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 217 #endif
 218 #ifdef CONFIG_FS_POSIX_ACL
 219         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 220 #endif
 221
 222         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 223                 /* flag mdc connection as lightweight, only used for test
 224                  * purpose, use with care */
 225                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 226
 227         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 228         data->ocd_version = LUSTRE_VERSION_CODE;
 229
 230         if (sb->s_flags & MS_RDONLY)
 231                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 232         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 233                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 234
 235 #ifdef HAVE_MS_FLOCK_LOCK
 236         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 237         sb->s_flags |= MS_FLOCK_LOCK;
 238 #endif
 239 #ifdef MS_HAS_NEW_AOPS
 240         sb->s_flags |= MS_HAS_NEW_AOPS;
 241 #endif
 242
 243         if (sbi->ll_flags & LL_SBI_FLOCK)
 244                 sbi->ll_fop = &ll_file_operations_flock;
 245         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 246                 sbi->ll_fop = &ll_file_operations;
 247         else
 248                 sbi->ll_fop = &ll_file_operations_noflock;
 249
 250         /* real client */
 251         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 252         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 253                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 254
 255         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 256
 257         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 258         if (err == -EBUSY) {
 259                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 260                                    "recovery, of which this client is not a "
 261                                    "part. Please wait for recovery to complete,"
 262                                    " abort, or time out.\n", md);
 263                 GOTO(out, err);
 264         } else if (err) {
 265                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 266                 GOTO(out, err);
 267         }
 268
 269         sbi->ll_md_exp->exp_connect_data = *data;
 270
 271         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 272                            LUSTRE_SEQ_METADATA);
 273         if (err) {
 274                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 275                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 276                 GOTO(out_md, err);
 277         }
 278
 279         /* For mount, we only need fs info from MDT0, and also in DNE, it
 280          * can make sure the client can be mounted as long as MDT0 is
 281          * avaible */
 282         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 283                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 284                         OBD_STATFS_FOR_MDT0);
 285         if (err)
 286                 GOTO(out_md_fid, err);
 287
 288         /* This needs to be after statfs to ensure connect has finished.
 289          * Note that "data" does NOT contain the valid connect reply.
 290          * If connecting to a 1.8 server there will be no LMV device, so
 291          * we can access the MDC export directly and exp_connect_flags will
 292          * be non-zero, but if accessing an upgraded 2.1 server it will
 293          * have the correct flags filled in.
 294          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 295         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 296         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 297             valid != CLIENT_CONNECT_MDT_REQD) {
 298                 char *buf;
 299
 300                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 301                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 302                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 303                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 304                                    "feature(s) needed for correct operation "
 305                                    "of this client (%s). Please upgrade "
 306                                    "server or downgrade client.\n",
 307                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 308                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 309                 GOTO(out_md_fid, err = -EPROTO);
 310         }
 311
 312         size = sizeof(*data);
 313         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 314                            KEY_CONN_DATA,  &size, data);
 315         if (err) {
 316                 CERROR("%s: Get connect data failed: rc = %d\n",
 317                        sbi->ll_md_exp->exp_obd->obd_name, err);
 318                 GOTO(out_md_fid, err);
 319         }
 320
 321         LASSERT(osfs->os_bsize);
 322         sb->s_blocksize = osfs->os_bsize;
 323         sb->s_blocksize_bits = log2(osfs->os_bsize);
 324         sb->s_magic = LL_SUPER_MAGIC;
 325         sb->s_maxbytes = MAX_LFS_FILESIZE;
 326         sbi->ll_namelen = osfs->os_namelen;
 327
 328         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 329             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 330                 LCONSOLE_INFO("Disabling user_xattr feature because "
 331                               "it is not supported on the server\n");
 332                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 333         }
 334
 335         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 336 #ifdef MS_POSIXACL
 337                 sb->s_flags |= MS_POSIXACL;
 338 #endif
 339                 sbi->ll_flags |= LL_SBI_ACL;
 340         } else {
 341                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 342 #ifdef MS_POSIXACL
 343                 sb->s_flags &= ~MS_POSIXACL;
 344 #endif
 345                 sbi->ll_flags &= ~LL_SBI_ACL;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 349                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 350                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 351                         LCONSOLE_INFO("client is set as remote by default.\n");
 352                 }
 353         } else {
 354                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 355                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 356                         LCONSOLE_INFO("client claims to be remote, but server "
 357                                       "rejected, forced to be local.\n");
 358                 }
 359         }
 360
 361         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 362                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 365                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 366         else
 367                 sbi->ll_md_brw_pages = 1;
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 370                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 371
 372         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 373                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 374                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 375                                       "unknown maximum xattr size.\n", dt);
 376                 } else {
 377                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 378                         sbi->ll_xattr_cache_enabled = 1;
 379                 }
 380         }
 381
 382         obd = class_name2obd(dt);
 383         if (!obd) {
 384                 CERROR("DT %s: not setup or attached\n", dt);
 385                 GOTO(out_md_fid, err = -ENODEV);
 386         }
 387
 388         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 389                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 390                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 391                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 392                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 393                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 394                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 395                                   OBD_CONNECT_MAXBYTES |
 396                                   OBD_CONNECT_EINPROGRESS |
 397                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 398                                   OBD_CONNECT_LAYOUTLOCK |
 399                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 400                                   OBD_CONNECT_BULK_MBITS;
 401
 402         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 403                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 404                  * disabled by default, because it can still be enabled on the
 405                  * fly via /proc. As a consequence, we still need to come to an
 406                  * agreement on the supported algorithms at connect time */
 407                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 408
 409                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 410                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 411                 else
 412                         data->ocd_cksum_types = cksum_types_supported_client();
 413         }
 414
 415 #ifdef HAVE_LRU_RESIZE_SUPPORT
 416         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 417 #endif
 418         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 419                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 420
 421         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 422                "ocd_grant: %d\n", data->ocd_connect_flags,
 423                data->ocd_version, data->ocd_grant);
 424
 425         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 426         obd->obd_upcall.onu_upcall = cl_ocd_update;
 427
 428         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 429
 430         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 431                           NULL);
 432         if (err == -EBUSY) {
 433                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 434                                    "recovery, of which this client is not a "
 435                                    "part.  Please wait for recovery to "
 436                                    "complete, abort, or time out.\n", dt);
 437                 GOTO(out_md, err);
 438         } else if (err) {
 439                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 440                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 441                 GOTO(out_md, err);
 442         }
 443
 444         sbi->ll_dt_exp->exp_connect_data = *data;
 445
 446         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 447                            LUSTRE_SEQ_METADATA);
 448         if (err) {
 449                 CERROR("%s: Can't init data layer FID infrastructure, "
 450                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 451                 GOTO(out_dt, err);
 452         }
 453
 454         mutex_lock(&sbi->ll_lco.lco_lock);
 455         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 456         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 457         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 458         mutex_unlock(&sbi->ll_lco.lco_lock);
 459
 460         fid_zero(&sbi->ll_root_fid);
 461         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
 462         if (err) {
 463                 CERROR("cannot mds_connect: rc = %d\n", err);
 464                 GOTO(out_lock_cn_cb, err);
 465         }
 466         if (!fid_is_sane(&sbi->ll_root_fid)) {
 467                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 468                        sbi->ll_md_exp->exp_obd->obd_name,
 469                        PFID(&sbi->ll_root_fid));
 470                 GOTO(out_lock_cn_cb, err = -EINVAL);
 471         }
 472         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 473
 474         sb->s_op = &lustre_super_operations;
 475 #if THREAD_SIZE >= 8192 /*b=17630*/
 476         sb->s_export_op = &lustre_export_operations;
 477 #endif
 478
 479         /* make root inode
 480          * XXX: move this to after cbd setup? */
 481         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 482         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 483                 valid |= OBD_MD_FLRMTPERM;
 484         else if (sbi->ll_flags & LL_SBI_ACL)
 485                 valid |= OBD_MD_FLACL;
 486
 487         OBD_ALLOC_PTR(op_data);
 488         if (op_data == NULL)
 489                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 490
 491         op_data->op_fid1 = sbi->ll_root_fid;
 492         op_data->op_mode = 0;
 493         op_data->op_valid = valid;
 494
 495         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 496
 497         OBD_FREE_PTR(op_data);
 498         if (err) {
 499                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 500                        sbi->ll_md_exp->exp_obd->obd_name, err);
 501                 GOTO(out_lock_cn_cb, err);
 502         }
 503
 504         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 505                                sbi->ll_md_exp, &lmd);
 506         if (err) {
 507                 CERROR("failed to understand root inode md: rc = %d\n", err);
 508                 ptlrpc_req_finished(request);
 509                 GOTO(out_lock_cn_cb, err);
 510         }
 511
 512         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 513         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 514                                             sbi->ll_flags & LL_SBI_32BIT_API),
 515                        &lmd);
 516         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 517         ptlrpc_req_finished(request);
 518
 519         if (IS_ERR(root)) {
 520 #ifdef CONFIG_FS_POSIX_ACL
 521                 if (lmd.posix_acl) {
 522                         posix_acl_release(lmd.posix_acl);
 523                         lmd.posix_acl = NULL;
 524                 }
 525 #endif
 526                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 527                 root = NULL;
 528                 CERROR("lustre_lite: bad iget4 for root\n");
 529                 GOTO(out_root, err);
 530         }
 531
 532 #ifdef CONFIG_FS_POSIX_ACL
 533         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 534                 rct_init(&sbi->ll_rct);
 535                 et_init(&sbi->ll_et);
 536         }
 537 #endif
 538
 539         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 540         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 541                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 542                                  NULL);
 543         cl_sb_init(sb);
 544
 545         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 546                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 547                                  sbi->ll_cache, NULL);
 548
 549         sb->s_root = d_make_root(root);
 550         if (sb->s_root == NULL) {
 551                 CERROR("%s: can't make root dentry\n",
 552                         ll_get_fsname(sb, NULL, 0));
 553                 GOTO(out_root, err = -ENOMEM);
 554         }
 555 #ifdef HAVE_DCACHE_LOCK
 556         sb->s_root->d_op = &ll_d_ops;
 557 #endif
 558
 559         sbi->ll_sdev_orig = sb->s_dev;
 560
 561         /* We set sb->s_dev equal on all lustre clients in order to support
 562          * NFS export clustering.  NFSD requires that the FSID be the same
 563          * on all clients. */
 564         /* s_dev is also used in lt_compare() to compare two fs, but that is
 565          * only a node-local comparison. */
 566         uuid = obd_get_uuid(sbi->ll_md_exp);
 567         if (uuid != NULL)
 568                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 569
 570         if (data != NULL)
 571                 OBD_FREE_PTR(data);
 572         if (osfs != NULL)
 573                 OBD_FREE_PTR(osfs);
 574         if (proc_lustre_fs_root != NULL) {
 575                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 576                                                   dt, md);
 577                 if (err < 0) {
 578                         CERROR("%s: could not register mount in lprocfs: "
 579                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 580                         err = 0;
 581                 }
 582         }
 583
 584         RETURN(err);
 585 out_root:
 586         if (root)
 587                 iput(root);
 588 out_lock_cn_cb:
 589         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 590 out_dt:
 591         obd_disconnect(sbi->ll_dt_exp);
 592         sbi->ll_dt_exp = NULL;
 593 out_md_fid:
 594         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 595 out_md:
 596         obd_disconnect(sbi->ll_md_exp);
 597         sbi->ll_md_exp = NULL;
 598 out:
 599         if (data != NULL)
 600                 OBD_FREE_PTR(data);
 601         if (osfs != NULL)
 602                 OBD_FREE_PTR(osfs);
 603         return err;
 604 }
 605
 606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 607 {
 608         int size, rc;
 609
 610         size = sizeof(*lmmsize);
 611         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 612                           KEY_MAX_EASIZE, &size, lmmsize);
 613         if (rc != 0) {
 614                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 615                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 616                 RETURN(rc);
 617         }
 618
 619         size = sizeof(int);
 620         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 621                           KEY_MAX_EASIZE, &size, lmmsize);
 622         if (rc)
 623                 CERROR("Get max mdsize error rc %d\n", rc);
 624
 625         RETURN(rc);
 626 }
 627
 628 /**
 629  * Get the value of the default_easize parameter.
 630  *
 631  * \see client_obd::cl_default_mds_easize
 632  *
 633  * \param[in] sbi       superblock info for this filesystem
 634  * \param[out] lmmsize  pointer to storage location for value
 635  *
 636  * \retval 0            on success
 637  * \retval negative     negated errno on failure
 638  */
 639 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 640 {
 641         int size, rc;
 642
 643         size = sizeof(int);
 644         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 645                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 646         if (rc)
 647                 CERROR("Get default mdsize error rc %d\n", rc);
 648
 649         RETURN(rc);
 650 }
 651
 652 /**
 653  * Set the default_easize parameter to the given value.
 654  *
 655  * \see client_obd::cl_default_mds_easize
 656  *
 657  * \param[in] sbi       superblock info for this filesystem
 658  * \param[in] lmmsize   the size to set
 659  *
 660  * \retval 0            on success
 661  * \retval negative     negated errno on failure
 662  */
 663 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 664 {
 665         int rc;
 666
 667         if (lmmsize < sizeof(struct lov_mds_md) ||
 668             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 669                 return -EINVAL;
 670
 671         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 672                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 673                                 sizeof(int), &lmmsize, NULL);
 674
 675         RETURN(rc);
 676 }
 677
 678 static void ll_dump_inode(struct inode *inode)
 679 {
 680         struct ll_d_hlist_node *tmp;
 681         int dentry_count = 0;
 682
 683         LASSERT(inode != NULL);
 684
 685         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 686                 dentry_count++;
 687
 688         CERROR("%s: inode %p dump: dev=%s fid="DFID
 689                " mode=%o count=%u, %d dentries\n",
 690                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 691                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 692                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 693 }
 694
 695 void lustre_dump_dentry(struct dentry *dentry, int recur)
 696 {
 697         struct list_head *tmp;
 698         int subdirs = 0;
 699
 700         LASSERT(dentry != NULL);
 701
 702         list_for_each(tmp, &dentry->d_subdirs)
 703                 subdirs++;
 704
 705         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 706                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 707                dentry->d_name.len, dentry->d_name.name,
 708                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 709                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 710                dentry->d_flags, dentry->d_fsdata, subdirs);
 711         if (dentry->d_inode != NULL)
 712                 ll_dump_inode(dentry->d_inode);
 713
 714         if (recur == 0)
 715                 return;
 716
 717         list_for_each(tmp, &dentry->d_subdirs) {
 718                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 719                 lustre_dump_dentry(d, recur - 1);
 720         }
 721 }
 722
 723 static void client_common_put_super(struct super_block *sb)
 724 {
 725         struct ll_sb_info *sbi = ll_s2sbi(sb);
 726         ENTRY;
 727
 728 #ifdef CONFIG_FS_POSIX_ACL
 729         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 730                 et_fini(&sbi->ll_et);
 731                 rct_fini(&sbi->ll_rct);
 732         }
 733 #endif
 734
 735         cl_sb_fini(sb);
 736
 737         list_del(&sbi->ll_conn_chain);
 738
 739         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 740         obd_disconnect(sbi->ll_dt_exp);
 741         sbi->ll_dt_exp = NULL;
 742
 743         lprocfs_unregister_mountpoint(sbi);
 744
 745         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 746         obd_disconnect(sbi->ll_md_exp);
 747         sbi->ll_md_exp = NULL;
 748
 749         EXIT;
 750 }
 751
 752 void ll_kill_super(struct super_block *sb)
 753 {
 754         struct ll_sb_info *sbi;
 755         ENTRY;
 756
 757         /* not init sb ?*/
 758         if (!(sb->s_flags & MS_ACTIVE))
 759                 return;
 760
 761         sbi = ll_s2sbi(sb);
 762         /* we need restore s_dev from changed for clustred NFS before put_super
 763          * because new kernels have cached s_dev and change sb->s_dev in
 764          * put_super not affected real removing devices */
 765         if (sbi) {
 766                 sb->s_dev = sbi->ll_sdev_orig;
 767                 sbi->ll_umounting = 1;
 768
 769                 /* wait running statahead threads to quit */
 770                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 771                         set_current_state(TASK_UNINTERRUPTIBLE);
 772                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 773                 }
 774         }
 775
 776         EXIT;
 777 }
 778
 779 static inline int ll_set_opt(const char *opt, char *data, int fl)
 780 {
 781         if (strncmp(opt, data, strlen(opt)) != 0)
 782                 return(0);
 783         else
 784                 return(fl);
 785 }
 786
 787 /* non-client-specific mount options are parsed in lmd_parse */
 788 static int ll_options(char *options, int *flags)
 789 {
 790         int tmp;
 791         char *s1 = options, *s2;
 792         ENTRY;
 793
 794         if (!options)
 795                 RETURN(0);
 796
 797         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 798
 799         while (*s1) {
 800                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 801                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 812                 if (tmp) {
 813                         *flags |= tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 817                 if (tmp) {
 818                         *flags &= ~tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 827                 if (tmp) {
 828                         *flags &= ~tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 832                 if (tmp) {
 833                         *flags |= tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846
 847                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 853                 if (tmp) {
 854                         *flags &= ~tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 888                 if (tmp) {
 889                         *flags &= ~tmp;
 890                         goto next;
 891                 }
 892                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 893                                    s1);
 894                 RETURN(-EINVAL);
 895
 896 next:
 897                 /* Find next opt */
 898                 s2 = strchr(s1, ',');
 899                 if (s2 == NULL)
 900                         break;
 901                 s1 = s2 + 1;
 902         }
 903         RETURN(0);
 904 }
 905
 906 void ll_lli_init(struct ll_inode_info *lli)
 907 {
 908         lli->lli_inode_magic = LLI_INODE_MAGIC;
 909         lli->lli_flags = 0;
 910         spin_lock_init(&lli->lli_lock);
 911         lli->lli_posix_acl = NULL;
 912         lli->lli_remote_perms = NULL;
 913         mutex_init(&lli->lli_rmtperm_mutex);
 914         /* Do not set lli_fid, it has been initialized already. */
 915         fid_zero(&lli->lli_pfid);
 916         atomic_set(&lli->lli_open_count, 0);
 917         lli->lli_rmtperm_time = 0;
 918         lli->lli_mds_read_och = NULL;
 919         lli->lli_mds_write_och = NULL;
 920         lli->lli_mds_exec_och = NULL;
 921         lli->lli_open_fd_read_count = 0;
 922         lli->lli_open_fd_write_count = 0;
 923         lli->lli_open_fd_exec_count = 0;
 924         mutex_init(&lli->lli_och_mutex);
 925         spin_lock_init(&lli->lli_agl_lock);
 926         spin_lock_init(&lli->lli_layout_lock);
 927         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 928         lli->lli_clob = NULL;
 929
 930         init_rwsem(&lli->lli_xattrs_list_rwsem);
 931         mutex_init(&lli->lli_xattrs_enq_lock);
 932
 933         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 934         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 935                 mutex_init(&lli->lli_readdir_mutex);
 936                 lli->lli_opendir_key = NULL;
 937                 lli->lli_sai = NULL;
 938                 spin_lock_init(&lli->lli_sa_lock);
 939                 lli->lli_opendir_pid = 0;
 940                 lli->lli_sa_enabled = 0;
 941                 lli->lli_def_stripe_offset = -1;
 942         } else {
 943                 mutex_init(&lli->lli_size_mutex);
 944                 lli->lli_symlink_name = NULL;
 945                 init_rwsem(&lli->lli_trunc_sem);
 946                 range_lock_tree_init(&lli->lli_write_tree);
 947                 init_rwsem(&lli->lli_glimpse_sem);
 948                 lli->lli_glimpse_time = 0;
 949                 INIT_LIST_HEAD(&lli->lli_agl_list);
 950                 lli->lli_agl_index = 0;
 951                 lli->lli_async_rc = 0;
 952         }
 953         mutex_init(&lli->lli_layout_mutex);
 954 }
 955
 956 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 957 {
 958         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 959
 960         bdi->name = "lustre";
 961         return bdi_register(bdi, NULL, "lustre-%d",
 962                             atomic_inc_return(&ll_bdi_num));
 963 }
 964
 965 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 966 {
 967         struct lustre_profile *lprof = NULL;
 968         struct lustre_sb_info *lsi = s2lsi(sb);
 969         struct ll_sb_info *sbi;
 970         char  *dt = NULL, *md = NULL;
 971         char  *profilenm = get_profile_name(sb);
 972         struct config_llog_instance *cfg;
 973         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 974         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 975         int    err;
 976         ENTRY;
 977
 978         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 979
 980         OBD_ALLOC_PTR(cfg);
 981         if (cfg == NULL)
 982                 RETURN(-ENOMEM);
 983
 984         try_module_get(THIS_MODULE);
 985
 986         /* client additional sb info */
 987         lsi->lsi_llsbi = sbi = ll_init_sbi();
 988         if (!sbi) {
 989                 module_put(THIS_MODULE);
 990                 OBD_FREE_PTR(cfg);
 991                 RETURN(-ENOMEM);
 992         }
 993
 994         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 995         if (err)
 996                 GOTO(out_free, err);
 997
 998         err = bdi_init(&lsi->lsi_bdi);
 999         if (err)
1000                 GOTO(out_free, err);
1001         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1002         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1003         err = ll_bdi_register(&lsi->lsi_bdi);
1004         if (err)
1005                 GOTO(out_free, err);
1006
1007         sb->s_bdi = &lsi->lsi_bdi;
1008 #ifndef HAVE_DCACHE_LOCK
1009         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1010         sb->s_d_op = &ll_d_ops;
1011 #endif
1012
1013         /* Generate a string unique to this super, in case some joker tries
1014            to mount the same fs at two mount points.
1015            Use the address of the super itself.*/
1016         cfg->cfg_instance = sb;
1017         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1018         cfg->cfg_callback = class_config_llog_handler;
1019         /* set up client obds */
1020         err = lustre_process_log(sb, profilenm, cfg);
1021         if (err < 0)
1022                 GOTO(out_free, err);
1023
1024         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1025         lprof = class_get_profile(profilenm);
1026         if (lprof == NULL) {
1027                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1028                                    " read from the MGS.  Does that filesystem "
1029                                    "exist?\n", profilenm);
1030                 GOTO(out_free, err = -EINVAL);
1031         }
1032         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1033                lprof->lp_md, lprof->lp_dt);
1034
1035         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1036         if (!dt)
1037                 GOTO(out_free, err = -ENOMEM);
1038         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1039
1040         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1041         if (!md)
1042                 GOTO(out_free, err = -ENOMEM);
1043         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1044
1045         /* connections, registrations, sb setup */
1046         err = client_common_fill_super(sb, md, dt, mnt);
1047         if (err < 0)
1048                 GOTO(out_free, err);
1049
1050         sbi->ll_client_common_fill_super_succeeded = 1;
1051
1052 out_free:
1053         if (md)
1054                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1055         if (dt)
1056                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1057         if (err)
1058                 ll_put_super(sb);
1059         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1060                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1061
1062         OBD_FREE_PTR(cfg);
1063         RETURN(err);
1064 } /* ll_fill_super */
1065
1066 void ll_put_super(struct super_block *sb)
1067 {
1068         struct config_llog_instance cfg, params_cfg;
1069         struct obd_device *obd;
1070         struct lustre_sb_info *lsi = s2lsi(sb);
1071         struct ll_sb_info *sbi = ll_s2sbi(sb);
1072         char *profilenm = get_profile_name(sb);
1073         long ccc_count;
1074         int next, force = 1, rc = 0;
1075         ENTRY;
1076
1077         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1078
1079         cfg.cfg_instance = sb;
1080         lustre_end_log(sb, profilenm, &cfg);
1081
1082         params_cfg.cfg_instance = sb;
1083         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1084
1085         if (sbi->ll_md_exp) {
1086                 obd = class_exp2obd(sbi->ll_md_exp);
1087                 if (obd)
1088                         force = obd->obd_force;
1089         }
1090
1091         /* Wait for unstable pages to be committed to stable storage */
1092         if (force == 0) {
1093                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1094                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1095                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1096                         &lwi);
1097         }
1098
1099         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1100         if (force == 0 && rc != -EINTR)
1101                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1102
1103
1104         /* We need to set force before the lov_disconnect in
1105            lustre_common_put_super, since l_d cleans up osc's as well. */
1106         if (force) {
1107                 next = 0;
1108                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1109                                                      &next)) != NULL) {
1110                         obd->obd_force = force;
1111                 }
1112         }
1113
1114         if (sbi->ll_client_common_fill_super_succeeded) {
1115                 /* Only if client_common_fill_super succeeded */
1116                 client_common_put_super(sb);
1117         }
1118
1119         next = 0;
1120         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1121                 class_manual_cleanup(obd);
1122         }
1123
1124         if (sbi->ll_flags & LL_SBI_VERBOSE)
1125                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1126
1127         if (profilenm)
1128                 class_del_profile(profilenm);
1129
1130         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1131                 bdi_destroy(&lsi->lsi_bdi);
1132                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1133         }
1134
1135         ll_free_sbi(sb);
1136         lsi->lsi_llsbi = NULL;
1137
1138         lustre_common_put_super(sb);
1139
1140         cl_env_cache_purge(~0);
1141
1142         module_put(THIS_MODULE);
1143
1144         EXIT;
1145 } /* client_put_super */
1146
1147 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1148 {
1149         struct inode *inode = NULL;
1150
1151         /* NOTE: we depend on atomic igrab() -bzzz */
1152         lock_res_and_lock(lock);
1153         if (lock->l_resource->lr_lvb_inode) {
1154                 struct ll_inode_info * lli;
1155                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1156                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1157                         inode = igrab(lock->l_resource->lr_lvb_inode);
1158                 } else {
1159                         inode = lock->l_resource->lr_lvb_inode;
1160                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1161                                          D_WARNING, lock, "lr_lvb_inode %p is "
1162                                          "bogus: magic %08x",
1163                                          lock->l_resource->lr_lvb_inode,
1164                                          lli->lli_inode_magic);
1165                         inode = NULL;
1166                 }
1167         }
1168         unlock_res_and_lock(lock);
1169         return inode;
1170 }
1171
1172 static void ll_dir_clear_lsm_md(struct inode *inode)
1173 {
1174         struct ll_inode_info *lli = ll_i2info(inode);
1175
1176         LASSERT(S_ISDIR(inode->i_mode));
1177
1178         if (lli->lli_lsm_md != NULL) {
1179                 lmv_free_memmd(lli->lli_lsm_md);
1180                 lli->lli_lsm_md = NULL;
1181         }
1182 }
1183
1184 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1185                                       const struct lu_fid *fid,
1186                                       struct lustre_md *md)
1187 {
1188         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1189         struct mdt_body         *body = md->body;
1190         struct inode            *inode;
1191         ino_t                   ino;
1192         ENTRY;
1193
1194         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1195         inode = iget_locked(sb, ino);
1196         if (inode == NULL) {
1197                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1198                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1199                 RETURN(ERR_PTR(-ENOENT));
1200         }
1201
1202         if (inode->i_state & I_NEW) {
1203                 struct ll_inode_info *lli = ll_i2info(inode);
1204                 struct lmv_stripe_md *lsm = md->lmv;
1205
1206                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1207                                 (body->mbo_mode & S_IFMT);
1208                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1209                          PFID(fid));
1210
1211                 LTIME_S(inode->i_mtime) = 0;
1212                 LTIME_S(inode->i_atime) = 0;
1213                 LTIME_S(inode->i_ctime) = 0;
1214                 inode->i_rdev = 0;
1215
1216                 /* initializing backing dev info. */
1217                 inode->i_mapping->backing_dev_info =
1218                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1219                 inode->i_op = &ll_dir_inode_operations;
1220                 inode->i_fop = &ll_dir_operations;
1221                 lli->lli_fid = *fid;
1222                 ll_lli_init(lli);
1223
1224                 LASSERT(lsm != NULL);
1225                 /* master object FID */
1226                 lli->lli_pfid = body->mbo_fid1;
1227                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1228                        lli, PFID(fid), PFID(&lli->lli_pfid));
1229                 unlock_new_inode(inode);
1230         }
1231
1232         RETURN(inode);
1233 }
1234
1235 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1236 {
1237         struct lu_fid *fid;
1238         struct lmv_stripe_md *lsm = md->lmv;
1239         int i;
1240
1241         LASSERT(lsm != NULL);
1242         /* XXX sigh, this lsm_root initialization should be in
1243          * LMV layer, but it needs ll_iget right now, so we
1244          * put this here right now. */
1245         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1246                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1247                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1248                 /* Unfortunately ll_iget will call ll_update_inode,
1249                  * where the initialization of slave inode is slightly
1250                  * different, so it reset lsm_md to NULL to avoid
1251                  * initializing lsm for slave inode. */
1252                 /* For migrating inode, master stripe and master object will
1253                  * be same, so we only need assign this inode */
1254                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1255                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1256                 else
1257                         lsm->lsm_md_oinfo[i].lmo_root =
1258                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1259
1260                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1261                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1262
1263                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1264                         return rc;
1265                 }
1266         }
1267
1268         return 0;
1269 }
1270
1271 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1272                                 const struct lmv_stripe_md *lsm_md2)
1273 {
1274         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1275                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1276                lsm_md1->lsm_md_master_mdt_index ==
1277                                         lsm_md2->lsm_md_master_mdt_index &&
1278                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1279                lsm_md1->lsm_md_layout_version ==
1280                                         lsm_md2->lsm_md_layout_version &&
1281                strcmp(lsm_md1->lsm_md_pool_name,
1282                       lsm_md2->lsm_md_pool_name) == 0;
1283 }
1284
1285 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1286 {
1287         struct ll_inode_info *lli = ll_i2info(inode);
1288         struct lmv_stripe_md *lsm = md->lmv;
1289         int     rc;
1290         ENTRY;
1291
1292         LASSERT(S_ISDIR(inode->i_mode));
1293         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1294                PFID(ll_inode2fid(inode)));
1295
1296         /* no striped information from request. */
1297         if (lsm == NULL) {
1298                 if (lli->lli_lsm_md == NULL) {
1299                         RETURN(0);
1300                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1301                                                 LMV_HASH_FLAG_MIGRATION) {
1302                         /* migration is done, the temporay MIGRATE layout has
1303                          * been removed */
1304                         CDEBUG(D_INODE, DFID" finish migration.\n",
1305                                PFID(ll_inode2fid(inode)));
1306                         lmv_free_memmd(lli->lli_lsm_md);
1307                         lli->lli_lsm_md = NULL;
1308                         RETURN(0);
1309                 } else {
1310                         /* The lustre_md from req does not include stripeEA,
1311                          * see ll_md_setattr */
1312                         RETURN(0);
1313                 }
1314         }
1315
1316         /* set the directory layout */
1317         if (lli->lli_lsm_md == NULL) {
1318                 struct cl_attr  *attr;
1319
1320                 rc = ll_init_lsm_md(inode, md);
1321                 if (rc != 0)
1322                         RETURN(rc);
1323
1324                 lli->lli_lsm_md = lsm;
1325
1326                 OBD_ALLOC_PTR(attr);
1327                 if (attr == NULL)
1328                         RETURN(-ENOMEM);
1329
1330                 /* validate the lsm */
1331                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1332                                    ll_md_blocking_ast);
1333                 if (rc != 0) {
1334                         OBD_FREE_PTR(attr);
1335                         RETURN(rc);
1336                 }
1337
1338                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1339                         md->body->mbo_nlink = attr->cat_nlink;
1340                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1341                         md->body->mbo_size = attr->cat_size;
1342                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1343                         md->body->mbo_atime = attr->cat_atime;
1344                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1345                         md->body->mbo_ctime = attr->cat_ctime;
1346                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1347                         md->body->mbo_mtime = attr->cat_mtime;
1348
1349                 OBD_FREE_PTR(attr);
1350
1351
1352                         /* set lsm_md to NULL, so the following free lustre_md
1353                  * will not free this lsm */
1354                 md->lmv = NULL;
1355                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1356                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1357                 RETURN(0);
1358         }
1359
1360         /* Compare the old and new stripe information */
1361         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1362                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1363                 int                     idx;
1364
1365                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1366                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1367                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1368                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1369                        inode, lsm, old_lsm,
1370                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1371                        lsm->lsm_md_stripe_count,
1372                        old_lsm->lsm_md_stripe_count,
1373                        lsm->lsm_md_master_mdt_index,
1374                        old_lsm->lsm_md_master_mdt_index,
1375                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1376                        lsm->lsm_md_layout_version,
1377                        old_lsm->lsm_md_layout_version,
1378                        lsm->lsm_md_pool_name,
1379                        old_lsm->lsm_md_pool_name);
1380
1381                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1382                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1383                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1385                 }
1386
1387                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1388                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1389                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1390                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1391                 }
1392
1393                 RETURN(-EIO);
1394         }
1395
1396         RETURN(0);
1397 }
1398
1399 void ll_clear_inode(struct inode *inode)
1400 {
1401         struct ll_inode_info *lli = ll_i2info(inode);
1402         struct ll_sb_info *sbi = ll_i2sbi(inode);
1403         ENTRY;
1404
1405         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1406                PFID(ll_inode2fid(inode)), inode);
1407
1408         if (S_ISDIR(inode->i_mode)) {
1409                 /* these should have been cleared in ll_file_release */
1410                 LASSERT(lli->lli_opendir_key == NULL);
1411                 LASSERT(lli->lli_sai == NULL);
1412                 LASSERT(lli->lli_opendir_pid == 0);
1413         }
1414
1415         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1416
1417         LASSERT(!lli->lli_open_fd_write_count);
1418         LASSERT(!lli->lli_open_fd_read_count);
1419         LASSERT(!lli->lli_open_fd_exec_count);
1420
1421         if (lli->lli_mds_write_och)
1422                 ll_md_real_close(inode, FMODE_WRITE);
1423         if (lli->lli_mds_exec_och)
1424                 ll_md_real_close(inode, FMODE_EXEC);
1425         if (lli->lli_mds_read_och)
1426                 ll_md_real_close(inode, FMODE_READ);
1427
1428         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429                 OBD_FREE(lli->lli_symlink_name,
1430                          strlen(lli->lli_symlink_name) + 1);
1431                 lli->lli_symlink_name = NULL;
1432         }
1433
1434         ll_xattr_cache_destroy(inode);
1435
1436         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1437                 LASSERT(lli->lli_posix_acl == NULL);
1438                 if (lli->lli_remote_perms) {
1439                         free_rmtperm_hash(lli->lli_remote_perms);
1440                         lli->lli_remote_perms = NULL;
1441                 }
1442         }
1443 #ifdef CONFIG_FS_POSIX_ACL
1444         else if (lli->lli_posix_acl) {
1445                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1446                 LASSERT(lli->lli_remote_perms == NULL);
1447                 posix_acl_release(lli->lli_posix_acl);
1448                 lli->lli_posix_acl = NULL;
1449         }
1450 #endif
1451         lli->lli_inode_magic = LLI_INODE_DEAD;
1452
1453         if (S_ISDIR(inode->i_mode))
1454                 ll_dir_clear_lsm_md(inode);
1455         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1456                 LASSERT(list_empty(&lli->lli_agl_list));
1457
1458         /*
1459          * XXX This has to be done before lsm is freed below, because
1460          * cl_object still uses inode lsm.
1461          */
1462         cl_inode_fini(inode);
1463
1464         EXIT;
1465 }
1466
1467 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1468 {
1469         struct lustre_md md;
1470         struct inode *inode = dentry->d_inode;
1471         struct ll_sb_info *sbi = ll_i2sbi(inode);
1472         struct ptlrpc_request *request = NULL;
1473         int rc, ia_valid;
1474         ENTRY;
1475
1476         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1477                                      LUSTRE_OPC_ANY, NULL);
1478         if (IS_ERR(op_data))
1479                 RETURN(PTR_ERR(op_data));
1480
1481         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1482         if (rc) {
1483                 ptlrpc_req_finished(request);
1484                 if (rc == -ENOENT) {
1485                         clear_nlink(inode);
1486                         /* Unlinked special device node? Or just a race?
1487                          * Pretend we done everything. */
1488                         if (!S_ISREG(inode->i_mode) &&
1489                             !S_ISDIR(inode->i_mode)) {
1490                                 ia_valid = op_data->op_attr.ia_valid;
1491                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1492                                 rc = simple_setattr(dentry, &op_data->op_attr);
1493                                 op_data->op_attr.ia_valid = ia_valid;
1494                         }
1495                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1496                         CERROR("md_setattr fails: rc = %d\n", rc);
1497                 }
1498                 RETURN(rc);
1499         }
1500
1501         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1502                               sbi->ll_md_exp, &md);
1503         if (rc) {
1504                 ptlrpc_req_finished(request);
1505                 RETURN(rc);
1506         }
1507
1508         ia_valid = op_data->op_attr.ia_valid;
1509         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1510          * cache is not cleared yet. */
1511         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1512         if (S_ISREG(inode->i_mode))
1513                 mutex_lock(&inode->i_mutex);
1514         rc = simple_setattr(dentry, &op_data->op_attr);
1515         if (S_ISREG(inode->i_mode))
1516                 mutex_unlock(&inode->i_mutex);
1517         op_data->op_attr.ia_valid = ia_valid;
1518
1519         rc = ll_update_inode(inode, &md);
1520         ptlrpc_req_finished(request);
1521
1522         RETURN(rc);
1523 }
1524
1525 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1526  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1527  * keep these values until such a time that objects are allocated for it.
1528  * We do the MDS operations first, as it is checking permissions for us.
1529  * We don't to the MDS RPC if there is nothing that we want to store there,
1530  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1531  * going to do an RPC anyways.
1532  *
1533  * If we are doing a truncate, we will send the mtime and ctime updates
1534  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1535  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1536  * at the same time.
1537  *
1538  * In case of HSMimport, we only set attr on MDS.
1539  */
1540 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1541 {
1542         struct inode *inode = dentry->d_inode;
1543         struct ll_inode_info *lli = ll_i2info(inode);
1544         struct md_op_data *op_data = NULL;
1545         int rc = 0;
1546         ENTRY;
1547
1548         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1549                "valid %x, hsm_import %d\n",
1550                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1551                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1552                hsm_import);
1553
1554         if (attr->ia_valid & ATTR_SIZE) {
1555                 /* Check new size against VFS/VM file size limit and rlimit */
1556                 rc = inode_newsize_ok(inode, attr->ia_size);
1557                 if (rc)
1558                         RETURN(rc);
1559
1560                 /* The maximum Lustre file size is variable, based on the
1561                  * OST maximum object size and number of stripes.  This
1562                  * needs another check in addition to the VFS check above. */
1563                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1564                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1565                                PFID(&lli->lli_fid), attr->ia_size,
1566                                ll_file_maxbytes(inode));
1567                         RETURN(-EFBIG);
1568                 }
1569
1570                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1571         }
1572
1573         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1574         if (attr->ia_valid & TIMES_SET_FLAGS) {
1575                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1576                     !cfs_capable(CFS_CAP_FOWNER))
1577                         RETURN(-EPERM);
1578         }
1579
1580         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1581         if (attr->ia_valid & ATTR_CTIME) {
1582                 attr->ia_ctime = CFS_CURRENT_TIME;
1583                 attr->ia_valid |= ATTR_CTIME_SET;
1584         }
1585         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1586             (attr->ia_valid & ATTR_ATIME)) {
1587                 attr->ia_atime = CFS_CURRENT_TIME;
1588                 attr->ia_valid |= ATTR_ATIME_SET;
1589         }
1590         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1591             (attr->ia_valid & ATTR_MTIME)) {
1592                 attr->ia_mtime = CFS_CURRENT_TIME;
1593                 attr->ia_valid |= ATTR_MTIME_SET;
1594         }
1595
1596         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1597                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1598                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1599                        cfs_time_current_sec());
1600
1601         if (S_ISREG(inode->i_mode)) {
1602                 if (attr->ia_valid & ATTR_SIZE)
1603                         inode_dio_write_done(inode);
1604                 mutex_unlock(&inode->i_mutex);
1605         }
1606
1607         /* We always do an MDS RPC, even if we're only changing the size;
1608          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1609
1610         OBD_ALLOC_PTR(op_data);
1611         if (op_data == NULL)
1612                 GOTO(out, rc = -ENOMEM);
1613
1614         op_data->op_attr = *attr;
1615
1616         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1617                 /* If we are changing file size, file content is
1618                  * modified, flag it. */
1619                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1620                 op_data->op_bias |= MDS_DATA_MODIFIED;
1621         }
1622
1623         rc = ll_md_setattr(dentry, op_data);
1624         if (rc)
1625                 GOTO(out, rc);
1626
1627         if (!S_ISREG(inode->i_mode) || hsm_import)
1628                 GOTO(out, rc = 0);
1629
1630         if (attr->ia_valid & (ATTR_SIZE |
1631                               ATTR_ATIME | ATTR_ATIME_SET |
1632                               ATTR_MTIME | ATTR_MTIME_SET)) {
1633                 /* For truncate and utimes sending attributes to OSTs, setting
1634                  * mtime/atime to the past will be performed under PW [0:EOF]
1635                  * extent lock (new_size:EOF for truncate).  It may seem
1636                  * excessive to send mtime/atime updates to OSTs when not
1637                  * setting times to past, but it is necessary due to possible
1638                  * time de-synchronization between MDT inode and OST objects */
1639                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1640         }
1641
1642         /* If the file was restored, it needs to set dirty flag.
1643          *
1644          * We've already sent MDS_DATA_MODIFIED flag in
1645          * ll_md_setattr() for truncate. However, the MDT refuses to
1646          * set the HS_DIRTY flag on released files, so we have to set
1647          * it again if the file has been restored. Please check how
1648          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1649          *
1650          * Please notice that if the file is not released, the previous
1651          * MDS_DATA_MODIFIED has taken effect and usually
1652          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1653          * This way we can save an RPC for common open + trunc
1654          * operation. */
1655         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1656                 struct hsm_state_set hss = {
1657                         .hss_valid = HSS_SETMASK,
1658                         .hss_setmask = HS_DIRTY,
1659                 };
1660                 int rc2;
1661
1662                 rc2 = ll_hsm_state_set(inode, &hss);
1663                 if (rc2 < 0)
1664                         CERROR(DFID "HSM set dirty failed: rc2 = %d\n",
1665                                PFID(ll_inode2fid(inode)), rc2);
1666         }
1667
1668         EXIT;
1669 out:
1670         if (op_data != NULL)
1671                 ll_finish_md_op_data(op_data);
1672
1673         if (S_ISREG(inode->i_mode)) {
1674                 mutex_lock(&inode->i_mutex);
1675                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1676                         inode_dio_wait(inode);
1677         }
1678
1679         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1680                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1681
1682         return rc;
1683 }
1684
1685 int ll_setattr(struct dentry *de, struct iattr *attr)
1686 {
1687         int mode = de->d_inode->i_mode;
1688
1689         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1690                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1691                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1692
1693         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1694                                (ATTR_SIZE|ATTR_MODE)) &&
1695             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1696              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1697               !(attr->ia_mode & S_ISGID))))
1698                 attr->ia_valid |= ATTR_FORCE;
1699
1700         if ((attr->ia_valid & ATTR_MODE) &&
1701             (mode & S_ISUID) &&
1702             !(attr->ia_mode & S_ISUID) &&
1703             !(attr->ia_valid & ATTR_KILL_SUID))
1704                 attr->ia_valid |= ATTR_KILL_SUID;
1705
1706         if ((attr->ia_valid & ATTR_MODE) &&
1707             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1708             !(attr->ia_mode & S_ISGID) &&
1709             !(attr->ia_valid & ATTR_KILL_SGID))
1710                 attr->ia_valid |= ATTR_KILL_SGID;
1711
1712         return ll_setattr_raw(de, attr, false);
1713 }
1714
1715 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1716                        __u64 max_age, __u32 flags)
1717 {
1718         struct ll_sb_info *sbi = ll_s2sbi(sb);
1719         struct obd_statfs obd_osfs;
1720         int rc;
1721         ENTRY;
1722
1723         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1724         if (rc) {
1725                 CERROR("md_statfs fails: rc = %d\n", rc);
1726                 RETURN(rc);
1727         }
1728
1729         osfs->os_type = sb->s_magic;
1730
1731         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1732                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1733
1734         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1735                 flags |= OBD_STATFS_NODELAY;
1736
1737         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1738         if (rc) {
1739                 CERROR("obd_statfs fails: rc = %d\n", rc);
1740                 RETURN(rc);
1741         }
1742
1743         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1744                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1745                obd_osfs.os_files);
1746
1747         osfs->os_bsize = obd_osfs.os_bsize;
1748         osfs->os_blocks = obd_osfs.os_blocks;
1749         osfs->os_bfree = obd_osfs.os_bfree;
1750         osfs->os_bavail = obd_osfs.os_bavail;
1751
1752         /* If we don't have as many objects free on the OST as inodes
1753          * on the MDS, we reduce the total number of inodes to
1754          * compensate, so that the "inodes in use" number is correct.
1755          */
1756         if (obd_osfs.os_ffree < osfs->os_ffree) {
1757                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1758                         obd_osfs.os_ffree;
1759                 osfs->os_ffree = obd_osfs.os_ffree;
1760         }
1761
1762         RETURN(rc);
1763 }
1764 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1765 {
1766         struct super_block *sb = de->d_sb;
1767         struct obd_statfs osfs;
1768         __u64 fsid = huge_encode_dev(sb->s_dev);
1769         int rc;
1770
1771         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1772         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1773
1774         /* Some amount of caching on the client is allowed */
1775         rc = ll_statfs_internal(sb, &osfs,
1776                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1777                                 0);
1778         if (rc)
1779                 return rc;
1780
1781         statfs_unpack(sfs, &osfs);
1782
1783         /* We need to downshift for all 32-bit kernels, because we can't
1784          * tell if the kernel is being called via sys_statfs64() or not.
1785          * Stop before overflowing f_bsize - in which case it is better
1786          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1787         if (sizeof(long) < 8) {
1788                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1789                         sfs->f_bsize <<= 1;
1790
1791                         osfs.os_blocks >>= 1;
1792                         osfs.os_bfree >>= 1;
1793                         osfs.os_bavail >>= 1;
1794                 }
1795         }
1796
1797         sfs->f_blocks = osfs.os_blocks;
1798         sfs->f_bfree = osfs.os_bfree;
1799         sfs->f_bavail = osfs.os_bavail;
1800         sfs->f_fsid.val[0] = (__u32)fsid;
1801         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1802         return 0;
1803 }
1804
1805 void ll_inode_size_lock(struct inode *inode)
1806 {
1807         struct ll_inode_info *lli;
1808
1809         LASSERT(!S_ISDIR(inode->i_mode));
1810
1811         lli = ll_i2info(inode);
1812         mutex_lock(&lli->lli_size_mutex);
1813 }
1814
1815 void ll_inode_size_unlock(struct inode *inode)
1816 {
1817         struct ll_inode_info *lli;
1818
1819         lli = ll_i2info(inode);
1820         mutex_unlock(&lli->lli_size_mutex);
1821 }
1822
1823 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1824 {
1825         struct ll_inode_info *lli = ll_i2info(inode);
1826         struct mdt_body *body = md->body;
1827         struct ll_sb_info *sbi = ll_i2sbi(inode);
1828
1829         if (body->mbo_valid & OBD_MD_FLEASIZE)
1830                 cl_file_inode_init(inode, md);
1831
1832         if (S_ISDIR(inode->i_mode)) {
1833                 int     rc;
1834
1835                 rc = ll_update_lsm_md(inode, md);
1836                 if (rc != 0)
1837                         return rc;
1838         }
1839
1840         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1841                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1842                         ll_update_remote_perm(inode, md->remote_perm);
1843         }
1844 #ifdef CONFIG_FS_POSIX_ACL
1845         else if (body->mbo_valid & OBD_MD_FLACL) {
1846                 spin_lock(&lli->lli_lock);
1847                 if (lli->lli_posix_acl)
1848                         posix_acl_release(lli->lli_posix_acl);
1849                 lli->lli_posix_acl = md->posix_acl;
1850                 spin_unlock(&lli->lli_lock);
1851         }
1852 #endif
1853         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1854                                         sbi->ll_flags & LL_SBI_32BIT_API);
1855         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1856
1857         if (body->mbo_valid & OBD_MD_FLATIME) {
1858                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1859                         LTIME_S(inode->i_atime) = body->mbo_atime;
1860                 lli->lli_atime = body->mbo_atime;
1861         }
1862
1863         if (body->mbo_valid & OBD_MD_FLMTIME) {
1864                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1865                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1866                                "to "LPU64"\n", inode->i_ino,
1867                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1868                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1869                 }
1870                 lli->lli_mtime = body->mbo_mtime;
1871         }
1872
1873         if (body->mbo_valid & OBD_MD_FLCTIME) {
1874                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1875                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1876                 lli->lli_ctime = body->mbo_ctime;
1877         }
1878
1879         if (body->mbo_valid & OBD_MD_FLMODE)
1880                 inode->i_mode = (inode->i_mode & S_IFMT) |
1881                                 (body->mbo_mode & ~S_IFMT);
1882
1883         if (body->mbo_valid & OBD_MD_FLTYPE)
1884                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1885                                 (body->mbo_mode & S_IFMT);
1886
1887         LASSERT(inode->i_mode != 0);
1888         if (S_ISREG(inode->i_mode))
1889                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1890                                        LL_MAX_BLKSIZE_BITS);
1891         else
1892                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1893
1894         if (body->mbo_valid & OBD_MD_FLUID)
1895                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1896         if (body->mbo_valid & OBD_MD_FLGID)
1897                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1898         if (body->mbo_valid & OBD_MD_FLFLAGS)
1899                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1900         if (body->mbo_valid & OBD_MD_FLNLINK)
1901                 set_nlink(inode, body->mbo_nlink);
1902         if (body->mbo_valid & OBD_MD_FLRDEV)
1903                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1904
1905         if (body->mbo_valid & OBD_MD_FLID) {
1906                 /* FID shouldn't be changed! */
1907                 if (fid_is_sane(&lli->lli_fid)) {
1908                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1909                                  "Trying to change FID "DFID
1910                                  " to the "DFID", inode "DFID"(%p)\n",
1911                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1912                                  PFID(ll_inode2fid(inode)), inode);
1913                 } else {
1914                         lli->lli_fid = body->mbo_fid1;
1915                 }
1916         }
1917
1918         LASSERT(fid_seq(&lli->lli_fid) != 0);
1919
1920         if (body->mbo_valid & OBD_MD_FLSIZE) {
1921                 i_size_write(inode, body->mbo_size);
1922
1923                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1924                        PFID(ll_inode2fid(inode)),
1925                        (unsigned long long)body->mbo_size);
1926
1927                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1928                         inode->i_blocks = body->mbo_blocks;
1929         }
1930
1931         if (body->mbo_valid & OBD_MD_TSTATE) {
1932                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1933                  * clear it when done to ensure to start again
1934                  * glimpsing updated attrs
1935                  */
1936                 if (body->mbo_t_state & MS_RESTORE)
1937                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1938                 else
1939                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1940         }
1941
1942         return 0;
1943 }
1944
1945 int ll_read_inode2(struct inode *inode, void *opaque)
1946 {
1947         struct lustre_md *md = opaque;
1948         struct ll_inode_info *lli = ll_i2info(inode);
1949         int     rc;
1950         ENTRY;
1951
1952         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1953                PFID(&lli->lli_fid), inode);
1954
1955         /* Core attributes from the MDS first.  This is a new inode, and
1956          * the VFS doesn't zero times in the core inode so we have to do
1957          * it ourselves.  They will be overwritten by either MDS or OST
1958          * attributes - we just need to make sure they aren't newer. */
1959         LTIME_S(inode->i_mtime) = 0;
1960         LTIME_S(inode->i_atime) = 0;
1961         LTIME_S(inode->i_ctime) = 0;
1962         inode->i_rdev = 0;
1963         rc = ll_update_inode(inode, md);
1964         if (rc != 0)
1965                 RETURN(rc);
1966
1967         /* OIDEBUG(inode); */
1968
1969         /* initializing backing dev info. */
1970         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1971
1972
1973         if (S_ISREG(inode->i_mode)) {
1974                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1975                 inode->i_op = &ll_file_inode_operations;
1976                 inode->i_fop = sbi->ll_fop;
1977                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1978                 EXIT;
1979         } else if (S_ISDIR(inode->i_mode)) {
1980                 inode->i_op = &ll_dir_inode_operations;
1981                 inode->i_fop = &ll_dir_operations;
1982                 EXIT;
1983         } else if (S_ISLNK(inode->i_mode)) {
1984                 inode->i_op = &ll_fast_symlink_inode_operations;
1985                 EXIT;
1986         } else {
1987                 inode->i_op = &ll_special_inode_operations;
1988
1989                 init_special_inode(inode, inode->i_mode,
1990                                    inode->i_rdev);
1991
1992                 EXIT;
1993         }
1994
1995         return 0;
1996 }
1997
1998 void ll_delete_inode(struct inode *inode)
1999 {
2000         struct ll_inode_info *lli = ll_i2info(inode);
2001         ENTRY;
2002
2003         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2004                 /* It is last chance to write out dirty pages,
2005                  * otherwise we may lose data while umount */
2006                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2007
2008         truncate_inode_pages_final(&inode->i_data);
2009
2010         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2011                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2012                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2013
2014 #ifdef HAVE_SBOPS_EVICT_INODE
2015         ll_clear_inode(inode);
2016 #endif
2017         clear_inode(inode);
2018
2019         EXIT;
2020 }
2021
2022 int ll_iocontrol(struct inode *inode, struct file *file,
2023                  unsigned int cmd, unsigned long arg)
2024 {
2025         struct ll_sb_info *sbi = ll_i2sbi(inode);
2026         struct ptlrpc_request *req = NULL;
2027         int rc, flags = 0;
2028         ENTRY;
2029
2030         switch(cmd) {
2031         case FSFILT_IOC_GETFLAGS: {
2032                 struct mdt_body *body;
2033                 struct md_op_data *op_data;
2034
2035                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2036                                              0, 0, LUSTRE_OPC_ANY,
2037                                              NULL);
2038                 if (IS_ERR(op_data))
2039                         RETURN(PTR_ERR(op_data));
2040
2041                 op_data->op_valid = OBD_MD_FLFLAGS;
2042                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2043                 ll_finish_md_op_data(op_data);
2044                 if (rc) {
2045                         CERROR("%s: failure inode "DFID": rc = %d\n",
2046                                sbi->ll_md_exp->exp_obd->obd_name,
2047                                PFID(ll_inode2fid(inode)), rc);
2048                         RETURN(-abs(rc));
2049                 }
2050
2051                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2052
2053                 flags = body->mbo_flags;
2054
2055                 ptlrpc_req_finished(req);
2056
2057                 RETURN(put_user(flags, (int __user *)arg));
2058         }
2059         case FSFILT_IOC_SETFLAGS: {
2060                 struct iattr *attr;
2061                 struct md_op_data *op_data;
2062                 struct cl_object *obj;
2063
2064                 if (get_user(flags, (int __user *)arg))
2065                         RETURN(-EFAULT);
2066
2067                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2068                                              LUSTRE_OPC_ANY, NULL);
2069                 if (IS_ERR(op_data))
2070                         RETURN(PTR_ERR(op_data));
2071
2072                 op_data->op_attr_flags = flags;
2073                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2074                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2075                 ll_finish_md_op_data(op_data);
2076                 ptlrpc_req_finished(req);
2077                 if (rc)
2078                         RETURN(rc);
2079
2080                 inode->i_flags = ll_ext_to_inode_flags(flags);
2081
2082                 obj = ll_i2info(inode)->lli_clob;
2083                 if (obj == NULL)
2084                         RETURN(0);
2085
2086                 OBD_ALLOC_PTR(attr);
2087                 if (attr == NULL)
2088                         RETURN(-ENOMEM);
2089
2090                 attr->ia_valid = ATTR_ATTR_FLAG;
2091                 rc = cl_setattr_ost(obj, attr, flags);
2092
2093                 OBD_FREE_PTR(attr);
2094                 RETURN(rc);
2095         }
2096         default:
2097                 RETURN(-ENOSYS);
2098         }
2099
2100         RETURN(0);
2101 }
2102
2103 int ll_flush_ctx(struct inode *inode)
2104 {
2105         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2106
2107         CDEBUG(D_SEC, "flush context for user %d\n",
2108                from_kuid(&init_user_ns, current_uid()));
2109
2110         obd_set_info_async(NULL, sbi->ll_md_exp,
2111                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2112                            0, NULL, NULL);
2113         obd_set_info_async(NULL, sbi->ll_dt_exp,
2114                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2115                            0, NULL, NULL);
2116         return 0;
2117 }
2118
2119 /* umount -f client means force down, don't save state */
2120 void ll_umount_begin(struct super_block *sb)
2121 {
2122         struct ll_sb_info *sbi = ll_s2sbi(sb);
2123         struct obd_device *obd;
2124         struct obd_ioctl_data *ioc_data;
2125         ENTRY;
2126
2127         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2128                sb->s_count, atomic_read(&sb->s_active));
2129
2130         obd = class_exp2obd(sbi->ll_md_exp);
2131         if (obd == NULL) {
2132                 CERROR("Invalid MDC connection handle "LPX64"\n",
2133                        sbi->ll_md_exp->exp_handle.h_cookie);
2134                 EXIT;
2135                 return;
2136         }
2137         obd->obd_force = 1;
2138
2139         obd = class_exp2obd(sbi->ll_dt_exp);
2140         if (obd == NULL) {
2141                 CERROR("Invalid LOV connection handle "LPX64"\n",
2142                        sbi->ll_dt_exp->exp_handle.h_cookie);
2143                 EXIT;
2144                 return;
2145         }
2146         obd->obd_force = 1;
2147
2148         OBD_ALLOC_PTR(ioc_data);
2149         if (ioc_data) {
2150                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2151                               sizeof *ioc_data, ioc_data, NULL);
2152
2153                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2154                               sizeof *ioc_data, ioc_data, NULL);
2155
2156                 OBD_FREE_PTR(ioc_data);
2157         }
2158
2159         /* Really, we'd like to wait until there are no requests outstanding,
2160          * and then continue.  For now, we just invalidate the requests,
2161          * schedule() and sleep one second if needed, and hope.
2162          */
2163         schedule();
2164         EXIT;
2165 }
2166
2167 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2168 {
2169         struct ll_sb_info *sbi = ll_s2sbi(sb);
2170         char *profilenm = get_profile_name(sb);
2171         int err;
2172         __u32 read_only;
2173
2174         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2175                 read_only = *flags & MS_RDONLY;
2176                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2177                                          sizeof(KEY_READ_ONLY),
2178                                          KEY_READ_ONLY, sizeof(read_only),
2179                                          &read_only, NULL);
2180                 if (err) {
2181                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2182                                       profilenm, read_only ?
2183                                       "read-only" : "read-write", err);
2184                         return err;
2185                 }
2186
2187                 if (read_only)
2188                         sb->s_flags |= MS_RDONLY;
2189                 else
2190                         sb->s_flags &= ~MS_RDONLY;
2191
2192                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2193                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2194                                       read_only ?  "read-only" : "read-write");
2195         }
2196         return 0;
2197 }
2198
2199 /**
2200  * Cleanup the open handle that is cached on MDT-side.
2201  *
2202  * For open case, the client side open handling thread may hit error
2203  * after the MDT grant the open. Under such case, the client should
2204  * send close RPC to the MDT as cleanup; otherwise, the open handle
2205  * on the MDT will be leaked there until the client umount or evicted.
2206  *
2207  * In further, if someone unlinked the file, because the open handle
2208  * holds the reference on such file/object, then it will block the
2209  * subsequent threads that want to locate such object via FID.
2210  *
2211  * \param[in] sb        super block for this file-system
2212  * \param[in] open_req  pointer to the original open request
2213  */
2214 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2215 {
2216         struct mdt_body                 *body;
2217         struct md_op_data               *op_data;
2218         struct ptlrpc_request           *close_req = NULL;
2219         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2220         ENTRY;
2221
2222         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2223         OBD_ALLOC_PTR(op_data);
2224         if (op_data == NULL) {
2225                 CWARN("%s: cannot allocate op_data to release open handle for "
2226                       DFID"\n",
2227                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2228
2229                 RETURN_EXIT;
2230         }
2231
2232         op_data->op_fid1 = body->mbo_fid1;
2233         op_data->op_handle = body->mbo_handle;
2234         op_data->op_mod_time = cfs_time_current_sec();
2235         md_close(exp, op_data, NULL, &close_req);
2236         ptlrpc_req_finished(close_req);
2237         ll_finish_md_op_data(op_data);
2238
2239         EXIT;
2240 }
2241
2242 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2243                   struct super_block *sb, struct lookup_intent *it)
2244 {
2245         struct ll_sb_info *sbi = NULL;
2246         struct lustre_md md = { NULL };
2247         int rc;
2248         ENTRY;
2249
2250         LASSERT(*inode || sb);
2251         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2252         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2253                               sbi->ll_md_exp, &md);
2254         if (rc != 0)
2255                 GOTO(cleanup, rc);
2256
2257         if (*inode) {
2258                 rc = ll_update_inode(*inode, &md);
2259                 if (rc != 0)
2260                         GOTO(out, rc);
2261         } else {
2262                 LASSERT(sb != NULL);
2263
2264                 /*
2265                  * At this point server returns to client's same fid as client
2266                  * generated for creating. So using ->fid1 is okay here.
2267                  */
2268                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2269
2270                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2271                                              sbi->ll_flags & LL_SBI_32BIT_API),
2272                                  &md);
2273                 if (IS_ERR(*inode)) {
2274 #ifdef CONFIG_FS_POSIX_ACL
2275                         if (md.posix_acl) {
2276                                 posix_acl_release(md.posix_acl);
2277                                 md.posix_acl = NULL;
2278                         }
2279 #endif
2280                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2281                         *inode = NULL;
2282                         CERROR("new_inode -fatal: rc %d\n", rc);
2283                         GOTO(out, rc);
2284                 }
2285         }
2286
2287         /* Handling piggyback layout lock.
2288          * Layout lock can be piggybacked by getattr and open request.
2289          * The lsm can be applied to inode only if it comes with a layout lock
2290          * otherwise correct layout may be overwritten, for example:
2291          * 1. proc1: mdt returns a lsm but not granting layout
2292          * 2. layout was changed by another client
2293          * 3. proc2: refresh layout and layout lock granted
2294          * 4. proc1: to apply a stale layout */
2295         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2296                 struct lustre_handle lockh;
2297                 struct ldlm_lock *lock;
2298
2299                 lockh.cookie = it->d.lustre.it_lock_handle;
2300                 lock = ldlm_handle2lock(&lockh);
2301                 LASSERT(lock != NULL);
2302                 if (ldlm_has_layout(lock)) {
2303                         struct cl_object_conf conf;
2304
2305                         memset(&conf, 0, sizeof(conf));
2306                         conf.coc_opc = OBJECT_CONF_SET;
2307                         conf.coc_inode = *inode;
2308                         conf.coc_lock = lock;
2309                         conf.u.coc_layout = md.layout;
2310                         (void)ll_layout_conf(*inode, &conf);
2311                 }
2312                 LDLM_LOCK_PUT(lock);
2313         }
2314
2315         GOTO(out, rc = 0);
2316
2317 out:
2318         md_free_lustre_md(sbi->ll_md_exp, &md);
2319
2320 cleanup:
2321         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2322                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2323
2324         return rc;
2325 }
2326
2327 int ll_obd_statfs(struct inode *inode, void __user *arg)
2328 {
2329         struct ll_sb_info *sbi = NULL;
2330         struct obd_export *exp;
2331         char *buf = NULL;
2332         struct obd_ioctl_data *data = NULL;
2333         __u32 type;
2334         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2335         int len = 0, rc;
2336
2337         if (!inode || !(sbi = ll_i2sbi(inode)))
2338                 GOTO(out_statfs, rc = -EINVAL);
2339
2340         rc = obd_ioctl_getdata(&buf, &len, arg);
2341         if (rc)
2342                 GOTO(out_statfs, rc);
2343
2344         data = (void*)buf;
2345         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2346             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2347                 GOTO(out_statfs, rc = -EINVAL);
2348
2349         if (data->ioc_inllen1 != sizeof(__u32) ||
2350             data->ioc_inllen2 != sizeof(__u32) ||
2351             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2352             data->ioc_plen2 != sizeof(struct obd_uuid))
2353                 GOTO(out_statfs, rc = -EINVAL);
2354
2355         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2356         if (type & LL_STATFS_LMV)
2357                 exp = sbi->ll_md_exp;
2358         else if (type & LL_STATFS_LOV)
2359                 exp = sbi->ll_dt_exp;
2360         else
2361                 GOTO(out_statfs, rc = -ENODEV);
2362
2363         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2364         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2365         if (rc)
2366                 GOTO(out_statfs, rc);
2367 out_statfs:
2368         if (buf)
2369                 obd_ioctl_freedata(buf, len);
2370         return rc;
2371 }
2372
2373 int ll_process_config(struct lustre_cfg *lcfg)
2374 {
2375         struct super_block *sb;
2376         unsigned long x;
2377         int rc = 0;
2378         char *ptr;
2379
2380         /* The instance name contains the sb: lustre-client-aacfe000 */
2381         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2382         if (!ptr || !*(++ptr))
2383                 return -EINVAL;
2384         if (sscanf(ptr, "%lx", &x) != 1)
2385                 return -EINVAL;
2386         sb = (struct super_block *)x;
2387         /* This better be a real Lustre superblock! */
2388         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2389
2390         /* Note we have not called client_common_fill_super yet, so
2391            proc fns must be able to handle that! */
2392         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2393                                       lcfg, sb);
2394         if (rc > 0)
2395                 rc = 0;
2396         return rc;
2397 }
2398
2399 /* this function prepares md_op_data hint for passing it down to MD stack. */
2400 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2401                                       struct inode *i1, struct inode *i2,
2402                                       const char *name, size_t namelen,
2403                                       __u32 mode, __u32 opc, void *data)
2404 {
2405         LASSERT(i1 != NULL);
2406
2407         if (name == NULL) {
2408                 /* Do not reuse namelen for something else. */
2409                 if (namelen != 0)
2410                         return ERR_PTR(-EINVAL);
2411         } else {
2412                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2413                         return ERR_PTR(-ENAMETOOLONG);
2414
2415                 if (!lu_name_is_valid_2(name, namelen))
2416                         return ERR_PTR(-EINVAL);
2417         }
2418
2419         if (op_data == NULL)
2420                 OBD_ALLOC_PTR(op_data);
2421
2422         if (op_data == NULL)
2423                 return ERR_PTR(-ENOMEM);
2424
2425         ll_i2gids(op_data->op_suppgids, i1, i2);
2426         op_data->op_fid1 = *ll_inode2fid(i1);
2427         op_data->op_default_stripe_offset = -1;
2428         if (S_ISDIR(i1->i_mode)) {
2429                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2430                 if (opc == LUSTRE_OPC_MKDIR)
2431                         op_data->op_default_stripe_offset =
2432                                    ll_i2info(i1)->lli_def_stripe_offset;
2433         }
2434
2435         if (i2) {
2436                 op_data->op_fid2 = *ll_inode2fid(i2);
2437                 if (S_ISDIR(i2->i_mode))
2438                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2439         } else {
2440                 fid_zero(&op_data->op_fid2);
2441         }
2442
2443         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2444                 op_data->op_cli_flags |= CLI_HASH64;
2445
2446         if (ll_need_32bit_api(ll_i2sbi(i1)))
2447                 op_data->op_cli_flags |= CLI_API32;
2448
2449         op_data->op_name = name;
2450         op_data->op_namelen = namelen;
2451         op_data->op_mode = mode;
2452         op_data->op_mod_time = cfs_time_current_sec();
2453         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2454         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2455         op_data->op_cap = cfs_curproc_cap_pack();
2456         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2457              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2458                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2459         } else {
2460                 op_data->op_mds = 0;
2461         }
2462         op_data->op_data = data;
2463
2464         return op_data;
2465 }
2466
2467 void ll_finish_md_op_data(struct md_op_data *op_data)
2468 {
2469         OBD_FREE_PTR(op_data);
2470 }
2471
2472 #ifdef HAVE_SUPEROPS_USE_DENTRY
2473 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2474 #else
2475 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2476 #endif
2477 {
2478         struct ll_sb_info *sbi;
2479
2480 #ifdef HAVE_SUPEROPS_USE_DENTRY
2481         LASSERT((seq != NULL) && (dentry != NULL));
2482         sbi = ll_s2sbi(dentry->d_sb);
2483 #else
2484         LASSERT((seq != NULL) && (vfs != NULL));
2485         sbi = ll_s2sbi(vfs->mnt_sb);
2486 #endif
2487
2488         if (sbi->ll_flags & LL_SBI_NOLCK)
2489                 seq_puts(seq, ",nolock");
2490
2491         if (sbi->ll_flags & LL_SBI_FLOCK)
2492                 seq_puts(seq, ",flock");
2493
2494         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2495                 seq_puts(seq, ",localflock");
2496
2497         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2498                 seq_puts(seq, ",user_xattr");
2499
2500         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2501                 seq_puts(seq, ",lazystatfs");
2502
2503         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2504                 seq_puts(seq, ",user_fid2path");
2505
2506         RETURN(0);
2507 }
2508
2509 /**
2510  * Get obd name by cmd, and copy out to user space
2511  */
2512 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2513 {
2514         struct ll_sb_info *sbi = ll_i2sbi(inode);
2515         struct obd_device *obd;
2516         ENTRY;
2517
2518         if (cmd == OBD_IOC_GETDTNAME)
2519                 obd = class_exp2obd(sbi->ll_dt_exp);
2520         else if (cmd == OBD_IOC_GETMDNAME)
2521                 obd = class_exp2obd(sbi->ll_md_exp);
2522         else
2523                 RETURN(-EINVAL);
2524
2525         if (!obd)
2526                 RETURN(-ENOENT);
2527
2528         if (copy_to_user((void __user *)arg, obd->obd_name,
2529                          strlen(obd->obd_name) + 1))
2530                 RETURN(-EFAULT);
2531
2532         RETURN(0);
2533 }
2534
2535 /**
2536  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2537  * fsname will be returned in this buffer; otherwise, a static buffer will be
2538  * used to store the fsname and returned to caller.
2539  */
2540 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2541 {
2542         static char fsname_static[MTI_NAME_MAXLEN];
2543         struct lustre_sb_info *lsi = s2lsi(sb);
2544         char *ptr;
2545         int len;
2546
2547         if (buf == NULL) {
2548                 /* this means the caller wants to use static buffer
2549                  * and it doesn't care about race. Usually this is
2550                  * in error reporting path */
2551                 buf = fsname_static;
2552                 buflen = sizeof(fsname_static);
2553         }
2554
2555         len = strlen(lsi->lsi_lmd->lmd_profile);
2556         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2557         if (ptr && (strcmp(ptr, "-client") == 0))
2558                 len -= 7;
2559
2560         if (unlikely(len >= buflen))
2561                 len = buflen - 1;
2562         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2563         buf[len] = '\0';
2564
2565         return buf;
2566 }
2567
2568 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2569 {
2570         char *path = NULL;
2571
2572         struct path p;
2573
2574         p.dentry = dentry;
2575         p.mnt = current->fs->root.mnt;
2576         path_get(&p);
2577         path = d_path(&p, buf, bufsize);
2578         path_put(&p);
2579         return path;
2580 }
2581
2582 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2583 {
2584         char *buf, *path = NULL;
2585         struct dentry *dentry = NULL;
2586         struct inode *inode = page->mapping->host;
2587
2588         /* this can be called inside spin lock so use GFP_ATOMIC. */
2589         buf = (char *)__get_free_page(GFP_ATOMIC);
2590         if (buf != NULL) {
2591                 dentry = d_find_alias(page->mapping->host);
2592                 if (dentry != NULL)
2593                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2594         }
2595
2596         CDEBUG(D_WARNING,
2597                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2598                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2599                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2600                PFID(ll_inode2fid(inode)),
2601                (path && !IS_ERR(path)) ? path : "", ioret);
2602
2603         if (dentry != NULL)
2604                 dput(dentry);
2605
2606         if (buf != NULL)
2607                 free_page((unsigned long)buf);
2608 }
2609
2610 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2611                         struct lov_user_md **kbuf)
2612 {
2613         struct lov_user_md      lum;
2614         ssize_t                 lum_size;
2615         ENTRY;
2616
2617         if (copy_from_user(&lum, md, sizeof(lum)))
2618                 RETURN(-EFAULT);
2619
2620         lum_size = ll_lov_user_md_size(&lum);
2621         if (lum_size < 0)
2622                 RETURN(lum_size);
2623
2624         OBD_ALLOC(*kbuf, lum_size);
2625         if (*kbuf == NULL)
2626                 RETURN(-ENOMEM);
2627
2628         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2629                 OBD_FREE(*kbuf, lum_size);
2630                 RETURN(-EFAULT);
2631         }
2632
2633         RETURN(lum_size);
2634 }
2635
2636 /*
2637  * Compute llite root squash state after a change of root squash
2638  * configuration setting or add/remove of a lnet nid
2639  */
2640 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2641 {
2642         struct root_squash_info *squash = &sbi->ll_squash;
2643         int i;
2644         bool matched;
2645         lnet_process_id_t id;
2646
2647         /* Update norootsquash flag */
2648         down_write(&squash->rsi_sem);
2649         if (list_empty(&squash->rsi_nosquash_nids))
2650                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2651         else {
2652                 /* Do not apply root squash as soon as one of our NIDs is
2653                  * in the nosquash_nids list */
2654                 matched = false;
2655                 i = 0;
2656                 while (LNetGetId(i++, &id) != -ENOENT) {
2657                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2658                                 continue;
2659                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2660                                 matched = true;
2661                                 break;
2662                         }
2663                 }
2664                 if (matched)
2665                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2666                 else
2667                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2668         }
2669         up_write(&squash->rsi_sem);
2670 }
2671
2672 /**
2673  * Parse linkea content to extract information about a given hardlink
2674  *
2675  * \param[in]   ldata      - Initialized linkea data
2676  * \param[in]   linkno     - Link identifier
2677  * \param[out]  parent_fid - The entry's parent FID
2678  * \param[out]  ln         - Entry name destination buffer
2679  *
2680  * \retval 0 on success
2681  * \retval Appropriate negative error code on failure
2682  */
2683 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2684                             struct lu_fid *parent_fid, struct lu_name *ln)
2685 {
2686         unsigned int    idx;
2687         int             rc;
2688         ENTRY;
2689
2690         rc = linkea_init(ldata);
2691         if (rc < 0)
2692                 RETURN(rc);
2693
2694         if (linkno >= ldata->ld_leh->leh_reccount)
2695                 /* beyond last link */
2696                 RETURN(-ENODATA);
2697
2698         linkea_first_entry(ldata);
2699         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2700                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2701                                     parent_fid);
2702                 if (idx == linkno)
2703                         break;
2704
2705                 linkea_next_entry(ldata);
2706         }
2707
2708         if (idx < linkno)
2709                 RETURN(-ENODATA);
2710
2711         RETURN(0);
2712 }
2713
2714 /**
2715  * Get parent FID and name of an identified link. Operation is performed for
2716  * a given link number, letting the caller iterate over linkno to list one or
2717  * all links of an entry.
2718  *
2719  * \param[in]     file - File descriptor against which to perform the operation
2720  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2721  *                       on and the available size. It is eventually filled with
2722  *                       the requested information or left untouched on error
2723  *
2724  * \retval - 0 on success
2725  * \retval - Appropriate negative error code on failure
2726  */
2727 int ll_getparent(struct file *file, struct getparent __user *arg)
2728 {
2729         struct dentry           *dentry = file->f_path.dentry;
2730         struct inode            *inode = dentry->d_inode;
2731         struct linkea_data      *ldata;
2732         struct lu_buf            buf = LU_BUF_NULL;
2733         struct lu_name           ln;
2734         struct lu_fid            parent_fid;
2735         __u32                    linkno;
2736         __u32                    name_size;
2737         int                      rc;
2738
2739         ENTRY;
2740
2741         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2742             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2743                 RETURN(-EPERM);
2744
2745         if (get_user(name_size, &arg->gp_name_size))
2746                 RETURN(-EFAULT);
2747
2748         if (get_user(linkno, &arg->gp_linkno))
2749                 RETURN(-EFAULT);
2750
2751         if (name_size > PATH_MAX)
2752                 RETURN(-EINVAL);
2753
2754         OBD_ALLOC(ldata, sizeof(*ldata));
2755         if (ldata == NULL)
2756                 RETURN(-ENOMEM);
2757
2758         rc = linkea_data_new(ldata, &buf);
2759         if (rc < 0)
2760                 GOTO(ldata_free, rc);
2761
2762         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2763         if (rc < 0)
2764                 GOTO(lb_free, rc);
2765
2766         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2767         if (rc < 0)
2768                 GOTO(lb_free, rc);
2769
2770         if (ln.ln_namelen >= name_size)
2771                 GOTO(lb_free, rc = -EOVERFLOW);
2772
2773         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2774                 GOTO(lb_free, rc = -EFAULT);
2775
2776         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2777                 GOTO(lb_free, rc = -EFAULT);
2778
2779         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2780                 GOTO(lb_free, rc = -EFAULT);
2781
2782 lb_free:
2783         lu_buf_free(&buf);
2784 ldata_free:
2785         OBD_FREE(ldata, sizeof(*ldata));
2786
2787         RETURN(rc);
2788 }