lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  97                 lru_page_max = pages / 2;
  98         } else {
  99                 lru_page_max = (pages / 4) * 3;
 100         }
 101
 102         /* initialize ll_cache data */
 103         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 104         sbi->ll_cache.ccc_lru_max = lru_page_max;
 105         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 106         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 107         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 108
 109         cfs_atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 110         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 111
 112         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 113                                            SBI_DEFAULT_READAHEAD_MAX);
 114         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 115         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 116                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 117         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 118         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 119
 120         ll_generate_random_uuid(uuid);
 121         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 122         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 123
 124         spin_lock(&ll_sb_lock);
 125         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 126         spin_unlock(&ll_sb_lock);
 127
 128         sbi->ll_flags |= LL_SBI_VERBOSE;
 129 #ifdef ENABLE_CHECKSUM
 130         sbi->ll_flags |= LL_SBI_CHECKSUM;
 131 #endif
 132
 133 #ifdef HAVE_LRU_RESIZE_SUPPORT
 134         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 135 #endif
 136
 137         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 138                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 139                                pp_r_hist.oh_lock);
 140                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 141                                pp_w_hist.oh_lock);
 142         }
 143
 144         /* metadata statahead is enabled by default */
 145         sbi->ll_sa_max = LL_SA_RPC_DEF;
 146         cfs_atomic_set(&sbi->ll_sa_total, 0);
 147         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 148         cfs_atomic_set(&sbi->ll_agl_total, 0);
 149         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 150
 151         RETURN(sbi);
 152 }
 153
 154 void ll_free_sbi(struct super_block *sb)
 155 {
 156         struct ll_sb_info *sbi = ll_s2sbi(sb);
 157         ENTRY;
 158
 159         if (sbi != NULL) {
 160                 spin_lock(&ll_sb_lock);
 161                 cfs_list_del(&sbi->ll_list);
 162                 spin_unlock(&ll_sb_lock);
 163                 OBD_FREE(sbi, sizeof(*sbi));
 164         }
 165         EXIT;
 166 }
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs *osfs = NULL;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         OBD_ALLOC_PTR(osfs);
 196         if (osfs == NULL) {
 197                 OBD_FREE_PTR(data);
 198                 RETURN(-ENOMEM);
 199         }
 200
 201         /* indicate the features supported by this client */
 202         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 203                                   OBD_CONNECT_ATTRFID  |
 204                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 205                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 206                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 207                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 208                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 209                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 210                                   OBD_CONNECT_EINPROGRESS |
 211                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 212                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 213                                   OBD_CONNECT_MAX_EASIZE |
 214                                   OBD_CONNECT_FLOCK_DEAD;
 215
 216         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 217                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 218
 219 #ifdef HAVE_LRU_RESIZE_SUPPORT
 220         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 221                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 222 #endif
 223 #ifdef CONFIG_FS_POSIX_ACL
 224         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 225 #endif
 226
 227         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 228                 /* flag mdc connection as lightweight, only used for test
 229                  * purpose, use with care */
 230                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 231
 232         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 233         data->ocd_version = LUSTRE_VERSION_CODE;
 234
 235         if (sb->s_flags & MS_RDONLY)
 236                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 237         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 238                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 239
 240 #ifdef HAVE_MS_FLOCK_LOCK
 241         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 242         sb->s_flags |= MS_FLOCK_LOCK;
 243 #endif
 244 #ifdef MS_HAS_NEW_AOPS
 245         sb->s_flags |= MS_HAS_NEW_AOPS;
 246 #endif
 247
 248         if (sbi->ll_flags & LL_SBI_FLOCK)
 249                 sbi->ll_fop = &ll_file_operations_flock;
 250         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 251                 sbi->ll_fop = &ll_file_operations;
 252         else
 253                 sbi->ll_fop = &ll_file_operations_noflock;
 254
 255         /* real client */
 256         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 257         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 258                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 259
 260         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 261
 262         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 263         if (err == -EBUSY) {
 264                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 265                                    "recovery, of which this client is not a "
 266                                    "part. Please wait for recovery to complete,"
 267                                    " abort, or time out.\n", md);
 268                 GOTO(out, err);
 269         } else if (err) {
 270                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 271                 GOTO(out, err);
 272         }
 273
 274         sbi->ll_md_exp->exp_connect_data = *data;
 275
 276         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 277                            LUSTRE_SEQ_METADATA);
 278         if (err) {
 279                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 280                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 281                 GOTO(out_md, err);
 282         }
 283
 284         /* For mount, we only need fs info from MDT0, and also in DNE, it
 285          * can make sure the client can be mounted as long as MDT0 is
 286          * avaible */
 287         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 288                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 289                         OBD_STATFS_FOR_MDT0);
 290         if (err)
 291                 GOTO(out_md_fid, err);
 292
 293         /* This needs to be after statfs to ensure connect has finished.
 294          * Note that "data" does NOT contain the valid connect reply.
 295          * If connecting to a 1.8 server there will be no LMV device, so
 296          * we can access the MDC export directly and exp_connect_flags will
 297          * be non-zero, but if accessing an upgraded 2.1 server it will
 298          * have the correct flags filled in.
 299          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 300         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 301         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 302             valid != CLIENT_CONNECT_MDT_REQD) {
 303                 char *buf;
 304
 305                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 306                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 307                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 308                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 309                                    "feature(s) needed for correct operation "
 310                                    "of this client (%s). Please upgrade "
 311                                    "server or downgrade client.\n",
 312                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 313                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 314                 GOTO(out_md_fid, err = -EPROTO);
 315         }
 316
 317         size = sizeof(*data);
 318         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 319                            KEY_CONN_DATA,  &size, data, NULL);
 320         if (err) {
 321                 CERROR("%s: Get connect data failed: rc = %d\n",
 322                        sbi->ll_md_exp->exp_obd->obd_name, err);
 323                 GOTO(out_md_fid, err);
 324         }
 325
 326         LASSERT(osfs->os_bsize);
 327         sb->s_blocksize = osfs->os_bsize;
 328         sb->s_blocksize_bits = log2(osfs->os_bsize);
 329         sb->s_magic = LL_SUPER_MAGIC;
 330         sb->s_maxbytes = MAX_LFS_FILESIZE;
 331         sbi->ll_namelen = osfs->os_namelen;
 332         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 333
 334         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 335             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 336                 LCONSOLE_INFO("Disabling user_xattr feature because "
 337                               "it is not supported on the server\n");
 338                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 339         }
 340
 341         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 342 #ifdef MS_POSIXACL
 343                 sb->s_flags |= MS_POSIXACL;
 344 #endif
 345                 sbi->ll_flags |= LL_SBI_ACL;
 346         } else {
 347                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 348 #ifdef MS_POSIXACL
 349                 sb->s_flags &= ~MS_POSIXACL;
 350 #endif
 351                 sbi->ll_flags &= ~LL_SBI_ACL;
 352         }
 353
 354         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 355                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 356                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 357                         LCONSOLE_INFO("client is set as remote by default.\n");
 358                 }
 359         } else {
 360                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 361                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client claims to be remote, but server "
 363                                       "rejected, forced to be local.\n");
 364                 }
 365         }
 366
 367         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 368                 LCONSOLE_INFO("client enabled MDS capability!\n");
 369                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 370         }
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 373                 LCONSOLE_INFO("client enabled OSS capability!\n");
 374                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 378                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 379
 380         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 381                 sbi->ll_md_brw_size = data->ocd_brw_size;
 382         else
 383                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 384
 385         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 386                 LCONSOLE_INFO("Layout lock feature supported.\n");
 387                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 388         }
 389
 390         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 391                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 392                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 393                                       "unknown maximum xattr size.\n", dt);
 394                 } else {
 395                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 396                         sbi->ll_xattr_cache_enabled = 1;
 397                 }
 398         }
 399
 400         obd = class_name2obd(dt);
 401         if (!obd) {
 402                 CERROR("DT %s: not setup or attached\n", dt);
 403                 GOTO(out_md_fid, err = -ENODEV);
 404         }
 405
 406         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 407                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 408                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 409                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 410                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 411                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 412                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 413                                   OBD_CONNECT_MAXBYTES |
 414                                   OBD_CONNECT_EINPROGRESS |
 415                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 416                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 417
 418         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 419                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 420
 421         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 422                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 423                  * disabled by default, because it can still be enabled on the
 424                  * fly via /proc. As a consequence, we still need to come to an
 425                  * agreement on the supported algorithms at connect time */
 426                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 427
 428                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 429                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 430                 else
 431                         data->ocd_cksum_types = cksum_types_supported_client();
 432         }
 433
 434 #ifdef HAVE_LRU_RESIZE_SUPPORT
 435         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 436 #endif
 437         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 438                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 439
 440         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 441                "ocd_grant: %d\n", data->ocd_connect_flags,
 442                data->ocd_version, data->ocd_grant);
 443
 444         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 445         obd->obd_upcall.onu_upcall = cl_ocd_update;
 446
 447         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 448
 449         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 450                           NULL);
 451         if (err == -EBUSY) {
 452                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 453                                    "recovery, of which this client is not a "
 454                                    "part.  Please wait for recovery to "
 455                                    "complete, abort, or time out.\n", dt);
 456                 GOTO(out_md, err);
 457         } else if (err) {
 458                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 459                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 460                 GOTO(out_md, err);
 461         }
 462
 463         sbi->ll_dt_exp->exp_connect_data = *data;
 464
 465         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 466                            LUSTRE_SEQ_METADATA);
 467         if (err) {
 468                 CERROR("%s: Can't init data layer FID infrastructure, "
 469                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 470                 GOTO(out_dt, err);
 471         }
 472
 473         mutex_lock(&sbi->ll_lco.lco_lock);
 474         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 475         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 476         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 477         mutex_unlock(&sbi->ll_lco.lco_lock);
 478
 479         fid_zero(&sbi->ll_root_fid);
 480         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 481         if (err) {
 482                 CERROR("cannot mds_connect: rc = %d\n", err);
 483                 GOTO(out_lock_cn_cb, err);
 484         }
 485         if (!fid_is_sane(&sbi->ll_root_fid)) {
 486                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 487                        sbi->ll_md_exp->exp_obd->obd_name,
 488                        PFID(&sbi->ll_root_fid));
 489                 GOTO(out_lock_cn_cb, err = -EINVAL);
 490         }
 491         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 492
 493         sb->s_op = &lustre_super_operations;
 494 #if THREAD_SIZE >= 8192 /*b=17630*/
 495         sb->s_export_op = &lustre_export_operations;
 496 #endif
 497
 498         /* make root inode
 499          * XXX: move this to after cbd setup? */
 500         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 501         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 502                 valid |= OBD_MD_FLRMTPERM;
 503         else if (sbi->ll_flags & LL_SBI_ACL)
 504                 valid |= OBD_MD_FLACL;
 505
 506         OBD_ALLOC_PTR(op_data);
 507         if (op_data == NULL)
 508                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 509
 510         op_data->op_fid1 = sbi->ll_root_fid;
 511         op_data->op_mode = 0;
 512         op_data->op_capa1 = oc;
 513         op_data->op_valid = valid;
 514
 515         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 516         if (oc)
 517                 capa_put(oc);
 518         OBD_FREE_PTR(op_data);
 519         if (err) {
 520                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 521                        sbi->ll_md_exp->exp_obd->obd_name, err);
 522                 GOTO(out_lock_cn_cb, err);
 523         }
 524
 525         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 526                                sbi->ll_md_exp, &lmd);
 527         if (err) {
 528                 CERROR("failed to understand root inode md: rc = %d\n", err);
 529                 ptlrpc_req_finished(request);
 530                 GOTO(out_lock_cn_cb, err);
 531         }
 532
 533         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 534         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 535                                             sbi->ll_flags & LL_SBI_32BIT_API),
 536                        &lmd);
 537         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 538         ptlrpc_req_finished(request);
 539
 540         if (root == NULL || IS_ERR(root)) {
 541                 if (lmd.lsm)
 542                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 543 #ifdef CONFIG_FS_POSIX_ACL
 544                 if (lmd.posix_acl) {
 545                         posix_acl_release(lmd.posix_acl);
 546                         lmd.posix_acl = NULL;
 547                 }
 548 #endif
 549                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 550                 root = NULL;
 551                 CERROR("lustre_lite: bad iget4 for root\n");
 552                 GOTO(out_root, err);
 553         }
 554
 555         err = ll_close_thread_start(&sbi->ll_lcq);
 556         if (err) {
 557                 CERROR("cannot start close thread: rc %d\n", err);
 558                 GOTO(out_root, err);
 559         }
 560
 561 #ifdef CONFIG_FS_POSIX_ACL
 562         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 563                 rct_init(&sbi->ll_rct);
 564                 et_init(&sbi->ll_et);
 565         }
 566 #endif
 567
 568         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 570                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 571                                  NULL);
 572         cl_sb_init(sb);
 573
 574         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 575                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 576                                  &sbi->ll_cache, NULL);
 577
 578         sb->s_root = d_make_root(root);
 579         if (sb->s_root == NULL) {
 580                 CERROR("%s: can't make root dentry\n",
 581                         ll_get_fsname(sb, NULL, 0));
 582                 GOTO(out_root, err = -ENOMEM);
 583         }
 584 #ifdef HAVE_DCACHE_LOCK
 585         sb->s_root->d_op = &ll_d_ops;
 586 #endif
 587
 588         sbi->ll_sdev_orig = sb->s_dev;
 589
 590         /* We set sb->s_dev equal on all lustre clients in order to support
 591          * NFS export clustering.  NFSD requires that the FSID be the same
 592          * on all clients. */
 593         /* s_dev is also used in lt_compare() to compare two fs, but that is
 594          * only a node-local comparison. */
 595         uuid = obd_get_uuid(sbi->ll_md_exp);
 596         if (uuid != NULL)
 597                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 598
 599         if (data != NULL)
 600                 OBD_FREE_PTR(data);
 601         if (osfs != NULL)
 602                 OBD_FREE_PTR(osfs);
 603         if (proc_lustre_fs_root != NULL) {
 604                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 605                                                   dt, md);
 606                 if (err < 0) {
 607                         CERROR("%s: could not register mount in lprocfs: "
 608                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 609                         err = 0;
 610                 }
 611         }
 612
 613         RETURN(err);
 614 out_root:
 615         if (root)
 616                 iput(root);
 617 out_lock_cn_cb:
 618         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 619 out_dt:
 620         obd_disconnect(sbi->ll_dt_exp);
 621         sbi->ll_dt_exp = NULL;
 622         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 623         obd_zombie_barrier();
 624 out_md_fid:
 625         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 626 out_md:
 627         obd_disconnect(sbi->ll_md_exp);
 628         sbi->ll_md_exp = NULL;
 629 out:
 630         if (data != NULL)
 631                 OBD_FREE_PTR(data);
 632         if (osfs != NULL)
 633                 OBD_FREE_PTR(osfs);
 634         return err;
 635 }
 636
 637 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 638 {
 639         int size, rc;
 640
 641         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 642         size = sizeof(int);
 643         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 644                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 645         if (rc)
 646                 CERROR("Get max mdsize error rc %d \n", rc);
 647
 648         RETURN(rc);
 649 }
 650
 651 void ll_dump_inode(struct inode *inode)
 652 {
 653         struct ll_d_hlist_node *tmp;
 654         int dentry_count = 0;
 655
 656         LASSERT(inode != NULL);
 657
 658         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 659                 dentry_count++;
 660
 661         CERROR("%s: inode %p dump: dev=%s fid="DFID
 662                " mode=%o count=%u, %d dentries\n",
 663                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 664                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 665                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 666 }
 667
 668 void lustre_dump_dentry(struct dentry *dentry, int recur)
 669 {
 670         struct list_head *tmp;
 671         int subdirs = 0;
 672
 673         LASSERT(dentry != NULL);
 674
 675         list_for_each(tmp, &dentry->d_subdirs)
 676                 subdirs++;
 677
 678         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 679                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 680                dentry->d_name.len, dentry->d_name.name,
 681                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 682                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 683                dentry->d_flags, dentry->d_fsdata, subdirs);
 684         if (dentry->d_inode != NULL)
 685                 ll_dump_inode(dentry->d_inode);
 686
 687         if (recur == 0)
 688                 return;
 689
 690         list_for_each(tmp, &dentry->d_subdirs) {
 691                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 692                 lustre_dump_dentry(d, recur - 1);
 693         }
 694 }
 695
 696 void client_common_put_super(struct super_block *sb)
 697 {
 698         struct ll_sb_info *sbi = ll_s2sbi(sb);
 699         ENTRY;
 700
 701 #ifdef CONFIG_FS_POSIX_ACL
 702         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 703                 et_fini(&sbi->ll_et);
 704                 rct_fini(&sbi->ll_rct);
 705         }
 706 #endif
 707
 708         ll_close_thread_shutdown(sbi->ll_lcq);
 709
 710         cl_sb_fini(sb);
 711
 712         cfs_list_del(&sbi->ll_conn_chain);
 713
 714         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 715         obd_disconnect(sbi->ll_dt_exp);
 716         sbi->ll_dt_exp = NULL;
 717         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 718          * see LU-2543. */
 719         obd_zombie_barrier();
 720
 721         lprocfs_unregister_mountpoint(sbi);
 722
 723         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 724         obd_disconnect(sbi->ll_md_exp);
 725         sbi->ll_md_exp = NULL;
 726
 727         EXIT;
 728 }
 729
 730 void ll_kill_super(struct super_block *sb)
 731 {
 732         struct ll_sb_info *sbi;
 733
 734         ENTRY;
 735
 736         /* not init sb ?*/
 737         if (!(sb->s_flags & MS_ACTIVE))
 738                 return;
 739
 740         sbi = ll_s2sbi(sb);
 741         /* we need restore s_dev from changed for clustred NFS before put_super
 742          * because new kernels have cached s_dev and change sb->s_dev in
 743          * put_super not affected real removing devices */
 744         if (sbi) {
 745                 sb->s_dev = sbi->ll_sdev_orig;
 746                 sbi->ll_umounting = 1;
 747         }
 748         EXIT;
 749 }
 750
 751 char *ll_read_opt(const char *opt, char *data)
 752 {
 753         char *value;
 754         char *retval;
 755         ENTRY;
 756
 757         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 758         if (strncmp(opt, data, strlen(opt)))
 759                 RETURN(NULL);
 760         if ((value = strchr(data, '=')) == NULL)
 761                 RETURN(NULL);
 762
 763         value++;
 764         OBD_ALLOC(retval, strlen(value) + 1);
 765         if (!retval) {
 766                 CERROR("out of memory!\n");
 767                 RETURN(NULL);
 768         }
 769
 770         memcpy(retval, value, strlen(value)+1);
 771         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 772         RETURN(retval);
 773 }
 774
 775 static inline int ll_set_opt(const char *opt, char *data, int fl)
 776 {
 777         if (strncmp(opt, data, strlen(opt)) != 0)
 778                 return(0);
 779         else
 780                 return(fl);
 781 }
 782
 783 /* non-client-specific mount options are parsed in lmd_parse */
 784 static int ll_options(char *options, int *flags)
 785 {
 786         int tmp;
 787         char *s1 = options, *s2;
 788         ENTRY;
 789
 790         if (!options)
 791                 RETURN(0);
 792
 793         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 794
 795         while (*s1) {
 796                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 797                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 813                 if (tmp) {
 814                         *flags &= ~tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 823                 if (tmp) {
 824                         *flags &= ~tmp;
 825                         goto next;
 826                 }
 827 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 51, 0)
 828                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 829                 if (tmp) {
 830                         /* Ignore deprecated mount option.  The client will
 831                          * always try to mount with ACL support, whether this
 832                          * is used depends on whether server supports it. */
 833                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 834                                                   "mount option 'acl'.\n");
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 838                 if (tmp) {
 839                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 840                                                   "mount option 'noacl'.\n");
 841                         goto next;
 842                 }
 843 #else
 844 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 845 #endif
 846                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 852                 if (tmp) {
 853                         *flags |= tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 857                 if (tmp) {
 858                         *flags &= ~tmp;
 859                         goto next;
 860                 }
 861
 862                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 863                 if (tmp) {
 864                         *flags |= tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 868                 if (tmp) {
 869                         *flags &= ~tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 873                 if (tmp) {
 874                         *flags |= tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 878                 if (tmp) {
 879                         *flags &= ~tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 888                 if (tmp) {
 889                         *flags &= ~tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 893                 if (tmp) {
 894                         *flags |= tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 903                 if (tmp) {
 904                         *flags |= tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 908                 if (tmp) {
 909                         *flags &= ~tmp;
 910                         goto next;
 911                 }
 912                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 913                                    s1);
 914                 RETURN(-EINVAL);
 915
 916 next:
 917                 /* Find next opt */
 918                 s2 = strchr(s1, ',');
 919                 if (s2 == NULL)
 920                         break;
 921                 s1 = s2 + 1;
 922         }
 923         RETURN(0);
 924 }
 925
 926 void ll_lli_init(struct ll_inode_info *lli)
 927 {
 928         lli->lli_inode_magic = LLI_INODE_MAGIC;
 929         lli->lli_flags = 0;
 930         lli->lli_ioepoch = 0;
 931         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 932         spin_lock_init(&lli->lli_lock);
 933         lli->lli_posix_acl = NULL;
 934         lli->lli_remote_perms = NULL;
 935         mutex_init(&lli->lli_rmtperm_mutex);
 936         /* Do not set lli_fid, it has been initialized already. */
 937         fid_zero(&lli->lli_pfid);
 938         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 939         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 940         cfs_atomic_set(&lli->lli_open_count, 0);
 941         lli->lli_mds_capa = NULL;
 942         lli->lli_rmtperm_time = 0;
 943         lli->lli_pending_och = NULL;
 944         lli->lli_mds_read_och = NULL;
 945         lli->lli_mds_write_och = NULL;
 946         lli->lli_mds_exec_och = NULL;
 947         lli->lli_open_fd_read_count = 0;
 948         lli->lli_open_fd_write_count = 0;
 949         lli->lli_open_fd_exec_count = 0;
 950         mutex_init(&lli->lli_och_mutex);
 951         spin_lock_init(&lli->lli_agl_lock);
 952         lli->lli_has_smd = false;
 953         lli->lli_layout_gen = LL_LAYOUT_GEN_NONE;
 954         lli->lli_clob = NULL;
 955
 956         init_rwsem(&lli->lli_xattrs_list_rwsem);
 957         mutex_init(&lli->lli_xattrs_enq_lock);
 958
 959         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 960         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 961                 mutex_init(&lli->lli_readdir_mutex);
 962                 lli->lli_opendir_key = NULL;
 963                 lli->lli_sai = NULL;
 964                 lli->lli_def_acl = NULL;
 965                 spin_lock_init(&lli->lli_sa_lock);
 966                 lli->lli_opendir_pid = 0;
 967         } else {
 968                 sema_init(&lli->lli_size_sem, 1);
 969                 lli->lli_size_sem_owner = NULL;
 970                 lli->lli_symlink_name = NULL;
 971                 init_rwsem(&lli->lli_trunc_sem);
 972                 mutex_init(&lli->lli_write_mutex);
 973                 init_rwsem(&lli->lli_glimpse_sem);
 974                 lli->lli_glimpse_time = 0;
 975                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 976                 lli->lli_agl_index = 0;
 977                 lli->lli_async_rc = 0;
 978                 lli->lli_volatile = false;
 979         }
 980         mutex_init(&lli->lli_layout_mutex);
 981 }
 982
 983 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 984 {
 985         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 986
 987         bdi->name = "lustre";
 988         return bdi_register(bdi, NULL, "lustre-%d",
 989                             atomic_inc_return(&ll_bdi_num));
 990 }
 991
 992 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 993 {
 994         struct lustre_profile *lprof = NULL;
 995         struct lustre_sb_info *lsi = s2lsi(sb);
 996         struct ll_sb_info *sbi;
 997         char  *dt = NULL, *md = NULL;
 998         char  *profilenm = get_profile_name(sb);
 999         struct config_llog_instance *cfg;
1000         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1001         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1002         int    err;
1003         ENTRY;
1004
1005         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1006
1007         OBD_ALLOC_PTR(cfg);
1008         if (cfg == NULL)
1009                 RETURN(-ENOMEM);
1010
1011         try_module_get(THIS_MODULE);
1012
1013         /* client additional sb info */
1014         lsi->lsi_llsbi = sbi = ll_init_sbi();
1015         if (!sbi) {
1016                 module_put(THIS_MODULE);
1017                 OBD_FREE_PTR(cfg);
1018                 RETURN(-ENOMEM);
1019         }
1020
1021         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1022         if (err)
1023                 GOTO(out_free, err);
1024
1025         err = bdi_init(&lsi->lsi_bdi);
1026         if (err)
1027                 GOTO(out_free, err);
1028         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1029         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1030         err = ll_bdi_register(&lsi->lsi_bdi);
1031         if (err)
1032                 GOTO(out_free, err);
1033
1034         sb->s_bdi = &lsi->lsi_bdi;
1035 #ifndef HAVE_DCACHE_LOCK
1036         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1037         sb->s_d_op = &ll_d_ops;
1038 #endif
1039
1040         /* Generate a string unique to this super, in case some joker tries
1041            to mount the same fs at two mount points.
1042            Use the address of the super itself.*/
1043         cfg->cfg_instance = sb;
1044         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1045         cfg->cfg_callback = class_config_llog_handler;
1046         /* set up client obds */
1047         err = lustre_process_log(sb, profilenm, cfg);
1048         if (err < 0) {
1049                 CERROR("Unable to process log: %d\n", err);
1050                 GOTO(out_free, err);
1051         }
1052
1053         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1054         lprof = class_get_profile(profilenm);
1055         if (lprof == NULL) {
1056                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1057                                    " read from the MGS.  Does that filesystem "
1058                                    "exist?\n", profilenm);
1059                 GOTO(out_free, err = -EINVAL);
1060         }
1061         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1062                lprof->lp_md, lprof->lp_dt);
1063
1064         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1065         if (!dt)
1066                 GOTO(out_free, err = -ENOMEM);
1067         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1068
1069         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1070         if (!md)
1071                 GOTO(out_free, err = -ENOMEM);
1072         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1073
1074         /* connections, registrations, sb setup */
1075         err = client_common_fill_super(sb, md, dt, mnt);
1076
1077 out_free:
1078         if (md)
1079                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1080         if (dt)
1081                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1082         if (err)
1083                 ll_put_super(sb);
1084         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1085                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1086
1087         OBD_FREE_PTR(cfg);
1088         RETURN(err);
1089 } /* ll_fill_super */
1090
1091
1092 void lu_context_keys_dump(void);
1093
1094 void ll_put_super(struct super_block *sb)
1095 {
1096         struct config_llog_instance cfg, params_cfg;
1097         struct obd_device *obd;
1098         struct lustre_sb_info *lsi = s2lsi(sb);
1099         struct ll_sb_info *sbi = ll_s2sbi(sb);
1100         char *profilenm = get_profile_name(sb);
1101         int ccc_count, next, force = 1, rc = 0;
1102         ENTRY;
1103
1104         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1105
1106         ll_print_capa_stat(sbi);
1107
1108         cfg.cfg_instance = sb;
1109         lustre_end_log(sb, profilenm, &cfg);
1110
1111         params_cfg.cfg_instance = sb;
1112         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1113
1114         if (sbi->ll_md_exp) {
1115                 obd = class_exp2obd(sbi->ll_md_exp);
1116                 if (obd)
1117                         force = obd->obd_force;
1118         }
1119
1120         /* Wait for unstable pages to be committed to stable storage */
1121         if (force == 0) {
1122                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1123                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1124                         cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1125                         &lwi);
1126         }
1127
1128         ccc_count = cfs_atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1129         if (force == 0 && rc != -EINTR)
1130                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1131
1132
1133         /* We need to set force before the lov_disconnect in
1134            lustre_common_put_super, since l_d cleans up osc's as well. */
1135         if (force) {
1136                 next = 0;
1137                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1138                                                      &next)) != NULL) {
1139                         obd->obd_force = force;
1140                 }
1141         }
1142
1143         if (sbi->ll_lcq) {
1144                 /* Only if client_common_fill_super succeeded */
1145                 client_common_put_super(sb);
1146         }
1147
1148         next = 0;
1149         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1150                 class_manual_cleanup(obd);
1151         }
1152
1153         if (sbi->ll_flags & LL_SBI_VERBOSE)
1154                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1155
1156         if (profilenm)
1157                 class_del_profile(profilenm);
1158
1159         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1160                 bdi_destroy(&lsi->lsi_bdi);
1161                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1162         }
1163
1164         ll_free_sbi(sb);
1165         lsi->lsi_llsbi = NULL;
1166
1167         lustre_common_put_super(sb);
1168
1169         cl_env_cache_purge(~0);
1170
1171         module_put(THIS_MODULE);
1172
1173         EXIT;
1174 } /* client_put_super */
1175
1176 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1177 {
1178         struct inode *inode = NULL;
1179
1180         /* NOTE: we depend on atomic igrab() -bzzz */
1181         lock_res_and_lock(lock);
1182         if (lock->l_resource->lr_lvb_inode) {
1183                 struct ll_inode_info * lli;
1184                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1185                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1186                         inode = igrab(lock->l_resource->lr_lvb_inode);
1187                 } else {
1188                         inode = lock->l_resource->lr_lvb_inode;
1189                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1190                                          D_WARNING, lock, "lr_lvb_inode %p is "
1191                                          "bogus: magic %08x",
1192                                          lock->l_resource->lr_lvb_inode,
1193                                          lli->lli_inode_magic);
1194                         inode = NULL;
1195                 }
1196         }
1197         unlock_res_and_lock(lock);
1198         return inode;
1199 }
1200
1201 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1202 {
1203         struct inode *inode = NULL;
1204         /* NOTE: we depend on atomic igrab() -bzzz */
1205         lock_res_and_lock(lock);
1206         if (lock->l_ast_data) {
1207                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1208                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1209                         inode = igrab(lock->l_ast_data);
1210                 } else {
1211                         inode = lock->l_ast_data;
1212                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1213                                          D_WARNING, lock, "l_ast_data %p is "
1214                                          "bogus: magic %08x", lock->l_ast_data,
1215                                          lli->lli_inode_magic);
1216                         inode = NULL;
1217                 }
1218         }
1219         unlock_res_and_lock(lock);
1220         return inode;
1221 }
1222
1223 void ll_clear_inode(struct inode *inode)
1224 {
1225         struct ll_inode_info *lli = ll_i2info(inode);
1226         struct ll_sb_info *sbi = ll_i2sbi(inode);
1227         ENTRY;
1228
1229         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1230                PFID(ll_inode2fid(inode)), inode);
1231
1232         if (S_ISDIR(inode->i_mode)) {
1233                 /* these should have been cleared in ll_file_release */
1234                 LASSERT(lli->lli_opendir_key == NULL);
1235                 LASSERT(lli->lli_sai == NULL);
1236                 LASSERT(lli->lli_opendir_pid == 0);
1237         }
1238
1239         spin_lock(&lli->lli_lock);
1240         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1241         spin_unlock(&lli->lli_lock);
1242         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1243
1244         LASSERT(!lli->lli_open_fd_write_count);
1245         LASSERT(!lli->lli_open_fd_read_count);
1246         LASSERT(!lli->lli_open_fd_exec_count);
1247
1248         if (lli->lli_mds_write_och)
1249                 ll_md_real_close(inode, FMODE_WRITE);
1250         if (lli->lli_mds_exec_och)
1251                 ll_md_real_close(inode, FMODE_EXEC);
1252         if (lli->lli_mds_read_och)
1253                 ll_md_real_close(inode, FMODE_READ);
1254
1255         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1256                 OBD_FREE(lli->lli_symlink_name,
1257                          strlen(lli->lli_symlink_name) + 1);
1258                 lli->lli_symlink_name = NULL;
1259         }
1260
1261         ll_xattr_cache_destroy(inode);
1262
1263         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1264                 LASSERT(lli->lli_posix_acl == NULL);
1265                 if (lli->lli_remote_perms) {
1266                         free_rmtperm_hash(lli->lli_remote_perms);
1267                         lli->lli_remote_perms = NULL;
1268                 }
1269         }
1270 #ifdef CONFIG_FS_POSIX_ACL
1271         else if (lli->lli_posix_acl) {
1272                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1273                 LASSERT(lli->lli_remote_perms == NULL);
1274                 posix_acl_release(lli->lli_posix_acl);
1275                 lli->lli_posix_acl = NULL;
1276         }
1277 #endif
1278         lli->lli_inode_magic = LLI_INODE_DEAD;
1279
1280         ll_clear_inode_capas(inode);
1281         if (!S_ISDIR(inode->i_mode))
1282                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1283
1284         /*
1285          * XXX This has to be done before lsm is freed below, because
1286          * cl_object still uses inode lsm.
1287          */
1288         cl_inode_fini(inode);
1289         lli->lli_has_smd = false;
1290
1291         EXIT;
1292 }
1293
1294 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1295                   struct md_open_data **mod)
1296 {
1297         struct lustre_md md;
1298         struct inode *inode = dentry->d_inode;
1299         struct ll_sb_info *sbi = ll_i2sbi(inode);
1300         struct ptlrpc_request *request = NULL;
1301         int rc, ia_valid;
1302         ENTRY;
1303
1304         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1305                                      LUSTRE_OPC_ANY, NULL);
1306         if (IS_ERR(op_data))
1307                 RETURN(PTR_ERR(op_data));
1308
1309         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1310                         &request, mod);
1311         if (rc) {
1312                 ptlrpc_req_finished(request);
1313                 if (rc == -ENOENT) {
1314                         clear_nlink(inode);
1315                         /* Unlinked special device node? Or just a race?
1316                          * Pretend we done everything. */
1317                         if (!S_ISREG(inode->i_mode) &&
1318                             !S_ISDIR(inode->i_mode)) {
1319                                 ia_valid = op_data->op_attr.ia_valid;
1320                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1321                                 rc = simple_setattr(dentry, &op_data->op_attr);
1322                                 op_data->op_attr.ia_valid = ia_valid;
1323                         }
1324                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1325                         CERROR("md_setattr fails: rc = %d\n", rc);
1326                 }
1327                 RETURN(rc);
1328         }
1329
1330         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1331                               sbi->ll_md_exp, &md);
1332         if (rc) {
1333                 ptlrpc_req_finished(request);
1334                 RETURN(rc);
1335         }
1336
1337         ia_valid = op_data->op_attr.ia_valid;
1338         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1339          * cache is not cleared yet. */
1340         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1341         rc = simple_setattr(dentry, &op_data->op_attr);
1342         op_data->op_attr.ia_valid = ia_valid;
1343
1344         /* Extract epoch data if obtained. */
1345         op_data->op_handle = md.body->handle;
1346         op_data->op_ioepoch = md.body->ioepoch;
1347
1348         ll_update_inode(inode, &md);
1349         ptlrpc_req_finished(request);
1350
1351         RETURN(rc);
1352 }
1353
1354 /* Close IO epoch and send Size-on-MDS attribute update. */
1355 static int ll_setattr_done_writing(struct inode *inode,
1356                                    struct md_op_data *op_data,
1357                                    struct md_open_data *mod)
1358 {
1359         struct ll_inode_info *lli = ll_i2info(inode);
1360         int rc = 0;
1361         ENTRY;
1362
1363         LASSERT(op_data != NULL);
1364         if (!S_ISREG(inode->i_mode))
1365                 RETURN(0);
1366
1367         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1368                op_data->op_ioepoch, PFID(&lli->lli_fid));
1369
1370         op_data->op_flags = MF_EPOCH_CLOSE;
1371         ll_done_writing_attr(inode, op_data);
1372         ll_pack_inode2opdata(inode, op_data, NULL);
1373
1374         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1375         if (rc == -EAGAIN) {
1376                 /* MDS has instructed us to obtain Size-on-MDS attribute
1377                  * from OSTs and send setattr to back to MDS. */
1378                 rc = ll_som_update(inode, op_data);
1379         } else if (rc) {
1380                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1381                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1382                        PFID(ll_inode2fid(inode)), rc);
1383         }
1384         RETURN(rc);
1385 }
1386
1387 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1388 {
1389         struct obd_capa *capa;
1390         int rc;
1391
1392         if (attr->ia_valid & ATTR_SIZE)
1393                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1394         else
1395                 capa = ll_mdscapa_get(inode);
1396
1397         rc = cl_setattr_ost(inode, attr, capa);
1398
1399         if (attr->ia_valid & ATTR_SIZE)
1400                 ll_truncate_free_capa(capa);
1401         else
1402                 capa_put(capa);
1403
1404         return rc;
1405 }
1406
1407 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1408  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1409  * keep these values until such a time that objects are allocated for it.
1410  * We do the MDS operations first, as it is checking permissions for us.
1411  * We don't to the MDS RPC if there is nothing that we want to store there,
1412  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1413  * going to do an RPC anyways.
1414  *
1415  * If we are doing a truncate, we will send the mtime and ctime updates
1416  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1417  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1418  * at the same time.
1419  *
1420  * In case of HSMimport, we only set attr on MDS.
1421  */
1422 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1423 {
1424         struct inode *inode = dentry->d_inode;
1425         struct ll_inode_info *lli = ll_i2info(inode);
1426         struct md_op_data *op_data = NULL;
1427         struct md_open_data *mod = NULL;
1428         bool file_is_released = false;
1429         int rc = 0, rc1 = 0;
1430         ENTRY;
1431
1432         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1433                "valid %x, hsm_import %d\n",
1434                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1435                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1436                hsm_import);
1437
1438         if (attr->ia_valid & ATTR_SIZE) {
1439                 /* Check new size against VFS/VM file size limit and rlimit */
1440                 rc = inode_newsize_ok(inode, attr->ia_size);
1441                 if (rc)
1442                         RETURN(rc);
1443
1444                 /* The maximum Lustre file size is variable, based on the
1445                  * OST maximum object size and number of stripes.  This
1446                  * needs another check in addition to the VFS check above. */
1447                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1448                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1449                                PFID(&lli->lli_fid), attr->ia_size,
1450                                ll_file_maxbytes(inode));
1451                         RETURN(-EFBIG);
1452                 }
1453
1454                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1455         }
1456
1457         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1458         if (attr->ia_valid & TIMES_SET_FLAGS) {
1459                 if (current_fsuid() != inode->i_uid &&
1460                     !cfs_capable(CFS_CAP_FOWNER))
1461                         RETURN(-EPERM);
1462         }
1463
1464         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1465         if (attr->ia_valid & ATTR_CTIME) {
1466                 attr->ia_ctime = CFS_CURRENT_TIME;
1467                 attr->ia_valid |= ATTR_CTIME_SET;
1468         }
1469         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1470             (attr->ia_valid & ATTR_ATIME)) {
1471                 attr->ia_atime = CFS_CURRENT_TIME;
1472                 attr->ia_valid |= ATTR_ATIME_SET;
1473         }
1474         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1475             (attr->ia_valid & ATTR_MTIME)) {
1476                 attr->ia_mtime = CFS_CURRENT_TIME;
1477                 attr->ia_valid |= ATTR_MTIME_SET;
1478         }
1479
1480         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1481                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1482                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1483                        cfs_time_current_sec());
1484
1485         /* We always do an MDS RPC, even if we're only changing the size;
1486          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1487
1488         OBD_ALLOC_PTR(op_data);
1489         if (op_data == NULL)
1490                 RETURN(-ENOMEM);
1491
1492         if (!S_ISDIR(inode->i_mode)) {
1493                 if (attr->ia_valid & ATTR_SIZE)
1494                         inode_dio_write_done(inode);
1495                 mutex_unlock(&inode->i_mutex);
1496                 down_write(&lli->lli_trunc_sem);
1497         }
1498
1499         /* truncate on a released file must failed with -ENODATA,
1500          * so size must not be set on MDS for released file
1501          * but other attributes must be set
1502          */
1503         if (S_ISREG(inode->i_mode)) {
1504                 struct lov_stripe_md *lsm;
1505                 __u32 gen;
1506
1507                 ll_layout_refresh(inode, &gen);
1508                 lsm = ccc_inode_lsm_get(inode);
1509                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1510                         file_is_released = true;
1511                 ccc_inode_lsm_put(inode, lsm);
1512
1513                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1514                         if (file_is_released) {
1515                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1516                                 if (rc < 0)
1517                                         GOTO(out, rc);
1518
1519                                 file_is_released = false;
1520                                 ll_layout_refresh(inode, &gen);
1521                         }
1522
1523                         /* If we are changing file size, file content is
1524                          * modified, flag it. */
1525                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1526                         spin_lock(&lli->lli_lock);
1527                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1528                         spin_unlock(&lli->lli_lock);
1529                         op_data->op_bias |= MDS_DATA_MODIFIED;
1530                 }
1531         }
1532
1533         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1534
1535         /* Open epoch for truncate. */
1536         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1537             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1538                 op_data->op_flags = MF_EPOCH_OPEN;
1539
1540         rc = ll_md_setattr(dentry, op_data, &mod);
1541         if (rc)
1542                 GOTO(out, rc);
1543
1544         /* RPC to MDT is sent, cancel data modification flag */
1545         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1546                 spin_lock(&lli->lli_lock);
1547                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1548                 spin_unlock(&lli->lli_lock);
1549         }
1550
1551         ll_ioepoch_open(lli, op_data->op_ioepoch);
1552         if (!S_ISREG(inode->i_mode) || file_is_released)
1553                 GOTO(out, rc = 0);
1554
1555         if (attr->ia_valid & (ATTR_SIZE |
1556                               ATTR_ATIME | ATTR_ATIME_SET |
1557                               ATTR_MTIME | ATTR_MTIME_SET)) {
1558                 /* For truncate and utimes sending attributes to OSTs, setting
1559                  * mtime/atime to the past will be performed under PW [0:EOF]
1560                  * extent lock (new_size:EOF for truncate).  It may seem
1561                  * excessive to send mtime/atime updates to OSTs when not
1562                  * setting times to past, but it is necessary due to possible
1563                  * time de-synchronization between MDT inode and OST objects */
1564                 rc = ll_setattr_ost(inode, attr);
1565         }
1566         EXIT;
1567 out:
1568         if (op_data) {
1569                 if (op_data->op_ioepoch) {
1570                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1571                         if (!rc)
1572                                 rc = rc1;
1573                 }
1574                 ll_finish_md_op_data(op_data);
1575         }
1576         if (!S_ISDIR(inode->i_mode)) {
1577                 up_write(&lli->lli_trunc_sem);
1578                 mutex_lock(&inode->i_mutex);
1579                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1580                         inode_dio_wait(inode);
1581         }
1582
1583         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1584                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1585
1586         return rc;
1587 }
1588
1589 int ll_setattr(struct dentry *de, struct iattr *attr)
1590 {
1591         int mode = de->d_inode->i_mode;
1592
1593         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1594                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1595                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1596
1597         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1598                                (ATTR_SIZE|ATTR_MODE)) &&
1599             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1600              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1601               !(attr->ia_mode & S_ISGID))))
1602                 attr->ia_valid |= ATTR_FORCE;
1603
1604         if ((mode & S_ISUID) &&
1605             !(attr->ia_mode & S_ISUID) &&
1606             !(attr->ia_valid & ATTR_KILL_SUID))
1607                 attr->ia_valid |= ATTR_KILL_SUID;
1608
1609         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1610             !(attr->ia_mode & S_ISGID) &&
1611             !(attr->ia_valid & ATTR_KILL_SGID))
1612                 attr->ia_valid |= ATTR_KILL_SGID;
1613
1614         return ll_setattr_raw(de, attr, false);
1615 }
1616
1617 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1618                        __u64 max_age, __u32 flags)
1619 {
1620         struct ll_sb_info *sbi = ll_s2sbi(sb);
1621         struct obd_statfs obd_osfs;
1622         int rc;
1623         ENTRY;
1624
1625         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1626         if (rc) {
1627                 CERROR("md_statfs fails: rc = %d\n", rc);
1628                 RETURN(rc);
1629         }
1630
1631         osfs->os_type = sb->s_magic;
1632
1633         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1634                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1635
1636         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1637                 flags |= OBD_STATFS_NODELAY;
1638
1639         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1640         if (rc) {
1641                 CERROR("obd_statfs fails: rc = %d\n", rc);
1642                 RETURN(rc);
1643         }
1644
1645         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1646                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1647                obd_osfs.os_files);
1648
1649         osfs->os_bsize = obd_osfs.os_bsize;
1650         osfs->os_blocks = obd_osfs.os_blocks;
1651         osfs->os_bfree = obd_osfs.os_bfree;
1652         osfs->os_bavail = obd_osfs.os_bavail;
1653
1654         /* If we don't have as many objects free on the OST as inodes
1655          * on the MDS, we reduce the total number of inodes to
1656          * compensate, so that the "inodes in use" number is correct.
1657          */
1658         if (obd_osfs.os_ffree < osfs->os_ffree) {
1659                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1660                         obd_osfs.os_ffree;
1661                 osfs->os_ffree = obd_osfs.os_ffree;
1662         }
1663
1664         RETURN(rc);
1665 }
1666 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1667 {
1668         struct super_block *sb = de->d_sb;
1669         struct obd_statfs osfs;
1670         __u64 fsid = huge_encode_dev(sb->s_dev);
1671         int rc;
1672
1673         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1674         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1675
1676         /* Some amount of caching on the client is allowed */
1677         rc = ll_statfs_internal(sb, &osfs,
1678                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1679                                 0);
1680         if (rc)
1681                 return rc;
1682
1683         statfs_unpack(sfs, &osfs);
1684
1685         /* We need to downshift for all 32-bit kernels, because we can't
1686          * tell if the kernel is being called via sys_statfs64() or not.
1687          * Stop before overflowing f_bsize - in which case it is better
1688          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1689         if (sizeof(long) < 8) {
1690                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1691                         sfs->f_bsize <<= 1;
1692
1693                         osfs.os_blocks >>= 1;
1694                         osfs.os_bfree >>= 1;
1695                         osfs.os_bavail >>= 1;
1696                 }
1697         }
1698
1699         sfs->f_blocks = osfs.os_blocks;
1700         sfs->f_bfree = osfs.os_bfree;
1701         sfs->f_bavail = osfs.os_bavail;
1702         sfs->f_fsid.val[0] = (__u32)fsid;
1703         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1704         return 0;
1705 }
1706
1707 void ll_inode_size_lock(struct inode *inode)
1708 {
1709         struct ll_inode_info *lli;
1710
1711         LASSERT(!S_ISDIR(inode->i_mode));
1712
1713         lli = ll_i2info(inode);
1714         LASSERT(lli->lli_size_sem_owner != current);
1715         down(&lli->lli_size_sem);
1716         LASSERT(lli->lli_size_sem_owner == NULL);
1717         lli->lli_size_sem_owner = current;
1718 }
1719
1720 void ll_inode_size_unlock(struct inode *inode)
1721 {
1722         struct ll_inode_info *lli;
1723
1724         lli = ll_i2info(inode);
1725         LASSERT(lli->lli_size_sem_owner == current);
1726         lli->lli_size_sem_owner = NULL;
1727         up(&lli->lli_size_sem);
1728 }
1729
1730 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1731 {
1732         struct ll_inode_info *lli = ll_i2info(inode);
1733         struct mdt_body *body = md->body;
1734         struct lov_stripe_md *lsm = md->lsm;
1735         struct ll_sb_info *sbi = ll_i2sbi(inode);
1736
1737         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1738         if (lsm != NULL) {
1739                 if (!lli->lli_has_smd &&
1740                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1741                         cl_file_inode_init(inode, md);
1742
1743                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1744                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1745                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1746         }
1747
1748         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1749                 if (body->valid & OBD_MD_FLRMTPERM)
1750                         ll_update_remote_perm(inode, md->remote_perm);
1751         }
1752 #ifdef CONFIG_FS_POSIX_ACL
1753         else if (body->valid & OBD_MD_FLACL) {
1754                 spin_lock(&lli->lli_lock);
1755                 if (lli->lli_posix_acl)
1756                         posix_acl_release(lli->lli_posix_acl);
1757                 lli->lli_posix_acl = md->posix_acl;
1758                 spin_unlock(&lli->lli_lock);
1759         }
1760 #endif
1761         inode->i_ino = cl_fid_build_ino(&body->fid1,
1762                                         sbi->ll_flags & LL_SBI_32BIT_API);
1763         inode->i_generation = cl_fid_build_gen(&body->fid1);
1764
1765         if (body->valid & OBD_MD_FLATIME) {
1766                 if (body->atime > LTIME_S(inode->i_atime))
1767                         LTIME_S(inode->i_atime) = body->atime;
1768                 lli->lli_lvb.lvb_atime = body->atime;
1769         }
1770         if (body->valid & OBD_MD_FLMTIME) {
1771                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1772                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1773                                "to "LPU64"\n", inode->i_ino,
1774                                LTIME_S(inode->i_mtime), body->mtime);
1775                         LTIME_S(inode->i_mtime) = body->mtime;
1776                 }
1777                 lli->lli_lvb.lvb_mtime = body->mtime;
1778         }
1779         if (body->valid & OBD_MD_FLCTIME) {
1780                 if (body->ctime > LTIME_S(inode->i_ctime))
1781                         LTIME_S(inode->i_ctime) = body->ctime;
1782                 lli->lli_lvb.lvb_ctime = body->ctime;
1783         }
1784         if (body->valid & OBD_MD_FLMODE)
1785                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1786         if (body->valid & OBD_MD_FLTYPE)
1787                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1788         LASSERT(inode->i_mode != 0);
1789         if (S_ISREG(inode->i_mode)) {
1790                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1791         } else {
1792                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1793         }
1794         if (body->valid & OBD_MD_FLUID)
1795                 inode->i_uid = body->uid;
1796         if (body->valid & OBD_MD_FLGID)
1797                 inode->i_gid = body->gid;
1798         if (body->valid & OBD_MD_FLFLAGS)
1799                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1800         if (body->valid & OBD_MD_FLNLINK)
1801                 set_nlink(inode, body->nlink);
1802         if (body->valid & OBD_MD_FLRDEV)
1803                 inode->i_rdev = old_decode_dev(body->rdev);
1804
1805         if (body->valid & OBD_MD_FLID) {
1806                 /* FID shouldn't be changed! */
1807                 if (fid_is_sane(&lli->lli_fid)) {
1808                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1809                                  "Trying to change FID "DFID
1810                                  " to the "DFID", inode "DFID"(%p)\n",
1811                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1812                                  PFID(ll_inode2fid(inode)), inode);
1813                 } else
1814                         lli->lli_fid = body->fid1;
1815         }
1816
1817         LASSERT(fid_seq(&lli->lli_fid) != 0);
1818
1819         if (body->valid & OBD_MD_FLSIZE) {
1820                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1821                     S_ISREG(inode->i_mode)) {
1822                         struct lustre_handle lockh;
1823                         ldlm_mode_t mode;
1824
1825                         /* As it is possible a blocking ast has been processed
1826                          * by this time, we need to check there is an UPDATE
1827                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1828                          * it. */
1829                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1830                                                &lockh, LDLM_FL_CBPENDING,
1831                                                LCK_CR | LCK_CW |
1832                                                LCK_PR | LCK_PW);
1833                         if (mode) {
1834                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1835                                                       LLIF_EPOCH_PENDING |
1836                                                       LLIF_SOM_DIRTY)) {
1837                                         CERROR("%s: inode "DFID" flags %u still"
1838                                                " has size authority! do not "
1839                                                "trust the size from MDS\n",
1840                                                sbi->ll_md_exp->exp_obd->obd_name,
1841                                                PFID(ll_inode2fid(inode)),
1842                                                lli->lli_flags);
1843                                 } else {
1844                                         /* Use old size assignment to avoid
1845                                          * deadlock bz14138 & bz14326 */
1846                                         i_size_write(inode, body->size);
1847                                         spin_lock(&lli->lli_lock);
1848                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1849                                         spin_unlock(&lli->lli_lock);
1850                                 }
1851                                 ldlm_lock_decref(&lockh, mode);
1852                         }
1853                 } else {
1854                         /* Use old size assignment to avoid
1855                          * deadlock bz14138 & bz14326 */
1856                         i_size_write(inode, body->size);
1857
1858                         CDEBUG(D_VFSTRACE,
1859                                "inode="DFID", updating i_size %llu\n",
1860                                PFID(ll_inode2fid(inode)),
1861                                (unsigned long long)body->size);
1862                 }
1863
1864                 if (body->valid & OBD_MD_FLBLOCKS)
1865                         inode->i_blocks = body->blocks;
1866         }
1867
1868         if (body->valid & OBD_MD_FLMDSCAPA) {
1869                 LASSERT(md->mds_capa);
1870                 ll_add_capa(inode, md->mds_capa);
1871         }
1872         if (body->valid & OBD_MD_FLOSSCAPA) {
1873                 LASSERT(md->oss_capa);
1874                 ll_add_capa(inode, md->oss_capa);
1875         }
1876
1877         if (body->valid & OBD_MD_TSTATE) {
1878                 if (body->t_state & MS_RESTORE)
1879                         lli->lli_flags |= LLIF_FILE_RESTORING;
1880         }
1881 }
1882
1883 void ll_read_inode2(struct inode *inode, void *opaque)
1884 {
1885         struct lustre_md *md = opaque;
1886         struct ll_inode_info *lli = ll_i2info(inode);
1887         ENTRY;
1888
1889         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1890                PFID(&lli->lli_fid), inode);
1891
1892         LASSERT(!lli->lli_has_smd);
1893
1894         /* Core attributes from the MDS first.  This is a new inode, and
1895          * the VFS doesn't zero times in the core inode so we have to do
1896          * it ourselves.  They will be overwritten by either MDS or OST
1897          * attributes - we just need to make sure they aren't newer. */
1898         LTIME_S(inode->i_mtime) = 0;
1899         LTIME_S(inode->i_atime) = 0;
1900         LTIME_S(inode->i_ctime) = 0;
1901         inode->i_rdev = 0;
1902         ll_update_inode(inode, md);
1903
1904         /* OIDEBUG(inode); */
1905
1906         /* initializing backing dev info. */
1907         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1908
1909
1910         if (S_ISREG(inode->i_mode)) {
1911                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1912                 inode->i_op = &ll_file_inode_operations;
1913                 inode->i_fop = sbi->ll_fop;
1914                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1915                 EXIT;
1916         } else if (S_ISDIR(inode->i_mode)) {
1917                 inode->i_op = &ll_dir_inode_operations;
1918                 inode->i_fop = &ll_dir_operations;
1919                 EXIT;
1920         } else if (S_ISLNK(inode->i_mode)) {
1921                 inode->i_op = &ll_fast_symlink_inode_operations;
1922                 EXIT;
1923         } else {
1924                 inode->i_op = &ll_special_inode_operations;
1925
1926                 init_special_inode(inode, inode->i_mode,
1927                                    inode->i_rdev);
1928
1929                 EXIT;
1930         }
1931 }
1932
1933 void ll_delete_inode(struct inode *inode)
1934 {
1935         struct cl_inode_info *lli = cl_i2info(inode);
1936         ENTRY;
1937
1938         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1939                 /* discard all dirty pages before truncating them, required by
1940                  * osc_extent implementation at LU-1030. */
1941                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1942                                    CL_FSYNC_DISCARD, 1);
1943
1944         truncate_inode_pages(&inode->i_data, 0);
1945
1946         /* Workaround for LU-118 */
1947         if (inode->i_data.nrpages) {
1948                 spin_lock_irq(&inode->i_data.tree_lock);
1949                 spin_unlock_irq(&inode->i_data.tree_lock);
1950                 LASSERTF(inode->i_data.nrpages == 0,
1951                          "inode="DFID"(%p) nrpages=%lu, see "
1952                          "http://jira.whamcloud.com/browse/LU-118\n",
1953                          PFID(ll_inode2fid(inode)), inode,
1954                          inode->i_data.nrpages);
1955         }
1956         /* Workaround end */
1957
1958 #ifdef HAVE_SBOPS_EVICT_INODE
1959         ll_clear_inode(inode);
1960 #endif
1961         clear_inode(inode);
1962
1963         EXIT;
1964 }
1965
1966 int ll_iocontrol(struct inode *inode, struct file *file,
1967                  unsigned int cmd, unsigned long arg)
1968 {
1969         struct ll_sb_info *sbi = ll_i2sbi(inode);
1970         struct ptlrpc_request *req = NULL;
1971         int rc, flags = 0;
1972         ENTRY;
1973
1974         switch(cmd) {
1975         case FSFILT_IOC_GETFLAGS: {
1976                 struct mdt_body *body;
1977                 struct md_op_data *op_data;
1978
1979                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1980                                              0, 0, LUSTRE_OPC_ANY,
1981                                              NULL);
1982                 if (IS_ERR(op_data))
1983                         RETURN(PTR_ERR(op_data));
1984
1985                 op_data->op_valid = OBD_MD_FLFLAGS;
1986                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1987                 ll_finish_md_op_data(op_data);
1988                 if (rc) {
1989                         CERROR("%s: failure inode "DFID": rc = %d\n",
1990                                sbi->ll_md_exp->exp_obd->obd_name,
1991                                PFID(ll_inode2fid(inode)), rc);
1992                         RETURN(-abs(rc));
1993                 }
1994
1995                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1996
1997                 flags = body->flags;
1998
1999                 ptlrpc_req_finished(req);
2000
2001                 RETURN(put_user(flags, (int *)arg));
2002         }
2003         case FSFILT_IOC_SETFLAGS: {
2004                 struct lov_stripe_md *lsm;
2005                 struct obd_info oinfo = { { { 0 } } };
2006                 struct md_op_data *op_data;
2007
2008                 if (get_user(flags, (int *)arg))
2009                         RETURN(-EFAULT);
2010
2011                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2012                                              LUSTRE_OPC_ANY, NULL);
2013                 if (IS_ERR(op_data))
2014                         RETURN(PTR_ERR(op_data));
2015
2016                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2017                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2018                 rc = md_setattr(sbi->ll_md_exp, op_data,
2019                                 NULL, 0, NULL, 0, &req, NULL);
2020                 ll_finish_md_op_data(op_data);
2021                 ptlrpc_req_finished(req);
2022                 if (rc)
2023                         RETURN(rc);
2024
2025                 inode->i_flags = ll_ext_to_inode_flags(flags);
2026
2027                 lsm = ccc_inode_lsm_get(inode);
2028                 if (!lsm_has_objects(lsm)) {
2029                         ccc_inode_lsm_put(inode, lsm);
2030                         RETURN(0);
2031                 }
2032
2033                 OBDO_ALLOC(oinfo.oi_oa);
2034                 if (!oinfo.oi_oa) {
2035                         ccc_inode_lsm_put(inode, lsm);
2036                         RETURN(-ENOMEM);
2037                 }
2038                 oinfo.oi_md = lsm;
2039                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2040                 oinfo.oi_oa->o_flags = flags;
2041                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2042                                        OBD_MD_FLGROUP;
2043                 oinfo.oi_capa = ll_mdscapa_get(inode);
2044                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2045                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2046                 capa_put(oinfo.oi_capa);
2047                 OBDO_FREE(oinfo.oi_oa);
2048                 ccc_inode_lsm_put(inode, lsm);
2049
2050                 if (rc && rc != -EPERM && rc != -EACCES)
2051                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2052
2053                 RETURN(rc);
2054         }
2055         default:
2056                 RETURN(-ENOSYS);
2057         }
2058
2059         RETURN(0);
2060 }
2061
2062 int ll_flush_ctx(struct inode *inode)
2063 {
2064         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2065
2066         CDEBUG(D_SEC, "flush context for user %d\n", current_uid());
2067
2068         obd_set_info_async(NULL, sbi->ll_md_exp,
2069                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2070                            0, NULL, NULL);
2071         obd_set_info_async(NULL, sbi->ll_dt_exp,
2072                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2073                            0, NULL, NULL);
2074         return 0;
2075 }
2076
2077 /* umount -f client means force down, don't save state */
2078 void ll_umount_begin(struct super_block *sb)
2079 {
2080         struct ll_sb_info *sbi = ll_s2sbi(sb);
2081         struct obd_device *obd;
2082         struct obd_ioctl_data *ioc_data;
2083         ENTRY;
2084
2085         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2086                sb->s_count, atomic_read(&sb->s_active));
2087
2088         obd = class_exp2obd(sbi->ll_md_exp);
2089         if (obd == NULL) {
2090                 CERROR("Invalid MDC connection handle "LPX64"\n",
2091                        sbi->ll_md_exp->exp_handle.h_cookie);
2092                 EXIT;
2093                 return;
2094         }
2095         obd->obd_force = 1;
2096
2097         obd = class_exp2obd(sbi->ll_dt_exp);
2098         if (obd == NULL) {
2099                 CERROR("Invalid LOV connection handle "LPX64"\n",
2100                        sbi->ll_dt_exp->exp_handle.h_cookie);
2101                 EXIT;
2102                 return;
2103         }
2104         obd->obd_force = 1;
2105
2106         OBD_ALLOC_PTR(ioc_data);
2107         if (ioc_data) {
2108                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2109                               sizeof *ioc_data, ioc_data, NULL);
2110
2111                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2112                               sizeof *ioc_data, ioc_data, NULL);
2113
2114                 OBD_FREE_PTR(ioc_data);
2115         }
2116
2117         /* Really, we'd like to wait until there are no requests outstanding,
2118          * and then continue.  For now, we just invalidate the requests,
2119          * schedule() and sleep one second if needed, and hope.
2120          */
2121         schedule();
2122         EXIT;
2123 }
2124
2125 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2126 {
2127         struct ll_sb_info *sbi = ll_s2sbi(sb);
2128         char *profilenm = get_profile_name(sb);
2129         int err;
2130         __u32 read_only;
2131
2132         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2133                 read_only = *flags & MS_RDONLY;
2134                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2135                                          sizeof(KEY_READ_ONLY),
2136                                          KEY_READ_ONLY, sizeof(read_only),
2137                                          &read_only, NULL);
2138                 if (err) {
2139                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2140                                       profilenm, read_only ?
2141                                       "read-only" : "read-write", err);
2142                         return err;
2143                 }
2144
2145                 if (read_only)
2146                         sb->s_flags |= MS_RDONLY;
2147                 else
2148                         sb->s_flags &= ~MS_RDONLY;
2149
2150                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2151                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2152                                       read_only ?  "read-only" : "read-write");
2153         }
2154         return 0;
2155 }
2156
2157 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2158                   struct super_block *sb, struct lookup_intent *it)
2159 {
2160         struct ll_sb_info *sbi = NULL;
2161         struct lustre_md md;
2162         int rc;
2163         ENTRY;
2164
2165         LASSERT(*inode || sb);
2166         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2167         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2168                               sbi->ll_md_exp, &md);
2169         if (rc)
2170                 RETURN(rc);
2171
2172         if (*inode) {
2173                 ll_update_inode(*inode, &md);
2174         } else {
2175                 LASSERT(sb != NULL);
2176
2177                 /*
2178                  * At this point server returns to client's same fid as client
2179                  * generated for creating. So using ->fid1 is okay here.
2180                  */
2181                 LASSERT(fid_is_sane(&md.body->fid1));
2182
2183                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2184                                              sbi->ll_flags & LL_SBI_32BIT_API),
2185                                  &md);
2186                 if (*inode == NULL || IS_ERR(*inode)) {
2187 #ifdef CONFIG_FS_POSIX_ACL
2188                         if (md.posix_acl) {
2189                                 posix_acl_release(md.posix_acl);
2190                                 md.posix_acl = NULL;
2191                         }
2192 #endif
2193                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2194                         *inode = NULL;
2195                         CERROR("new_inode -fatal: rc %d\n", rc);
2196                         GOTO(out, rc);
2197                 }
2198         }
2199
2200         /* Handling piggyback layout lock.
2201          * Layout lock can be piggybacked by getattr and open request.
2202          * The lsm can be applied to inode only if it comes with a layout lock
2203          * otherwise correct layout may be overwritten, for example:
2204          * 1. proc1: mdt returns a lsm but not granting layout
2205          * 2. layout was changed by another client
2206          * 3. proc2: refresh layout and layout lock granted
2207          * 4. proc1: to apply a stale layout */
2208         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2209                 struct lustre_handle lockh;
2210                 struct ldlm_lock *lock;
2211
2212                 lockh.cookie = it->d.lustre.it_lock_handle;
2213                 lock = ldlm_handle2lock(&lockh);
2214                 LASSERT(lock != NULL);
2215                 if (ldlm_has_layout(lock)) {
2216                         struct cl_object_conf conf;
2217
2218                         memset(&conf, 0, sizeof(conf));
2219                         conf.coc_opc = OBJECT_CONF_SET;
2220                         conf.coc_inode = *inode;
2221                         conf.coc_lock = lock;
2222                         conf.u.coc_md = &md;
2223                         (void)ll_layout_conf(*inode, &conf);
2224                 }
2225                 LDLM_LOCK_PUT(lock);
2226         }
2227
2228 out:
2229         if (md.lsm != NULL)
2230                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2231         md_free_lustre_md(sbi->ll_md_exp, &md);
2232         RETURN(rc);
2233 }
2234
2235 int ll_obd_statfs(struct inode *inode, void *arg)
2236 {
2237         struct ll_sb_info *sbi = NULL;
2238         struct obd_export *exp;
2239         char *buf = NULL;
2240         struct obd_ioctl_data *data = NULL;
2241         __u32 type;
2242         __u32 flags;
2243         int len = 0, rc;
2244
2245         if (!inode || !(sbi = ll_i2sbi(inode)))
2246                 GOTO(out_statfs, rc = -EINVAL);
2247
2248         rc = obd_ioctl_getdata(&buf, &len, arg);
2249         if (rc)
2250                 GOTO(out_statfs, rc);
2251
2252         data = (void*)buf;
2253         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2254             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2255                 GOTO(out_statfs, rc = -EINVAL);
2256
2257         if (data->ioc_inllen1 != sizeof(__u32) ||
2258             data->ioc_inllen2 != sizeof(__u32) ||
2259             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2260             data->ioc_plen2 != sizeof(struct obd_uuid))
2261                 GOTO(out_statfs, rc = -EINVAL);
2262
2263         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2264         if (type & LL_STATFS_LMV)
2265                 exp = sbi->ll_md_exp;
2266         else if (type & LL_STATFS_LOV)
2267                 exp = sbi->ll_dt_exp;
2268         else
2269                 GOTO(out_statfs, rc = -ENODEV);
2270
2271         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2272         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2273         if (rc)
2274                 GOTO(out_statfs, rc);
2275 out_statfs:
2276         if (buf)
2277                 obd_ioctl_freedata(buf, len);
2278         return rc;
2279 }
2280
2281 int ll_process_config(struct lustre_cfg *lcfg)
2282 {
2283         char *ptr;
2284         void *sb;
2285         struct lprocfs_static_vars lvars;
2286         unsigned long x;
2287         int rc = 0;
2288
2289         lprocfs_llite_init_vars(&lvars);
2290
2291         /* The instance name contains the sb: lustre-client-aacfe000 */
2292         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2293         if (!ptr || !*(++ptr))
2294                 return -EINVAL;
2295         if (sscanf(ptr, "%lx", &x) != 1)
2296                 return -EINVAL;
2297         sb = (void *)x;
2298         /* This better be a real Lustre superblock! */
2299         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2300
2301         /* Note we have not called client_common_fill_super yet, so
2302            proc fns must be able to handle that! */
2303         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2304                                       lcfg, sb);
2305         if (rc > 0)
2306                 rc = 0;
2307         return(rc);
2308 }
2309
2310 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2311 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2312                                        struct inode *i1, struct inode *i2,
2313                                        const char *name, int namelen,
2314                                        int mode, __u32 opc, void *data)
2315 {
2316         LASSERT(i1 != NULL);
2317
2318         if (namelen > ll_i2sbi(i1)->ll_namelen)
2319                 return ERR_PTR(-ENAMETOOLONG);
2320
2321         if (op_data == NULL)
2322                 OBD_ALLOC_PTR(op_data);
2323
2324         if (op_data == NULL)
2325                 return ERR_PTR(-ENOMEM);
2326
2327         ll_i2gids(op_data->op_suppgids, i1, i2);
2328         op_data->op_fid1 = *ll_inode2fid(i1);
2329         op_data->op_capa1 = ll_mdscapa_get(i1);
2330
2331         if (i2) {
2332                 op_data->op_fid2 = *ll_inode2fid(i2);
2333                 op_data->op_capa2 = ll_mdscapa_get(i2);
2334         } else {
2335                 fid_zero(&op_data->op_fid2);
2336                 op_data->op_capa2 = NULL;
2337         }
2338
2339         op_data->op_name = name;
2340         op_data->op_namelen = namelen;
2341         op_data->op_mode = mode;
2342         op_data->op_mod_time = cfs_time_current_sec();
2343         op_data->op_fsuid = current_fsuid();
2344         op_data->op_fsgid = current_fsgid();
2345         op_data->op_cap = cfs_curproc_cap_pack();
2346         op_data->op_bias = 0;
2347         op_data->op_cli_flags = 0;
2348         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2349              filename_is_volatile(name, namelen, NULL))
2350                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2351         op_data->op_opc = opc;
2352         op_data->op_mds = 0;
2353         op_data->op_data = data;
2354
2355         /* If the file is being opened after mknod() (normally due to NFS)
2356          * try to use the default stripe data from parent directory for
2357          * allocating OST objects.  Try to pass the parent FID to MDS. */
2358         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2359             !ll_i2info(i2)->lli_has_smd) {
2360                 struct ll_inode_info *lli = ll_i2info(i2);
2361
2362                 spin_lock(&lli->lli_lock);
2363                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2364                         op_data->op_fid1 = lli->lli_pfid;
2365                 spin_unlock(&lli->lli_lock);
2366                 /** We ignore parent's capability temporary. */
2367         }
2368
2369         /* When called by ll_setattr_raw, file is i1. */
2370         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2371                 op_data->op_bias |= MDS_DATA_MODIFIED;
2372
2373         return op_data;
2374 }
2375
2376 void ll_finish_md_op_data(struct md_op_data *op_data)
2377 {
2378         capa_put(op_data->op_capa1);
2379         capa_put(op_data->op_capa2);
2380         OBD_FREE_PTR(op_data);
2381 }
2382
2383 #ifdef HAVE_SUPEROPS_USE_DENTRY
2384 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2385 #else
2386 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2387 #endif
2388 {
2389         struct ll_sb_info *sbi;
2390
2391 #ifdef HAVE_SUPEROPS_USE_DENTRY
2392         LASSERT((seq != NULL) && (dentry != NULL));
2393         sbi = ll_s2sbi(dentry->d_sb);
2394 #else
2395         LASSERT((seq != NULL) && (vfs != NULL));
2396         sbi = ll_s2sbi(vfs->mnt_sb);
2397 #endif
2398
2399         if (sbi->ll_flags & LL_SBI_NOLCK)
2400                 seq_puts(seq, ",nolock");
2401
2402         if (sbi->ll_flags & LL_SBI_FLOCK)
2403                 seq_puts(seq, ",flock");
2404
2405         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2406                 seq_puts(seq, ",localflock");
2407
2408         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2409                 seq_puts(seq, ",user_xattr");
2410
2411         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2412                 seq_puts(seq, ",lazystatfs");
2413
2414         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2415                 seq_puts(seq, ",user_fid2path");
2416
2417         RETURN(0);
2418 }
2419
2420 /**
2421  * Get obd name by cmd, and copy out to user space
2422  */
2423 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2424 {
2425         struct ll_sb_info *sbi = ll_i2sbi(inode);
2426         struct obd_device *obd;
2427         ENTRY;
2428
2429         if (cmd == OBD_IOC_GETDTNAME)
2430                 obd = class_exp2obd(sbi->ll_dt_exp);
2431         else if (cmd == OBD_IOC_GETMDNAME)
2432                 obd = class_exp2obd(sbi->ll_md_exp);
2433         else
2434                 RETURN(-EINVAL);
2435
2436         if (!obd)
2437                 RETURN(-ENOENT);
2438
2439         if (copy_to_user((void *)arg, obd->obd_name,
2440                          strlen(obd->obd_name) + 1))
2441                 RETURN(-EFAULT);
2442
2443         RETURN(0);
2444 }
2445
2446 /**
2447  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2448  * fsname will be returned in this buffer; otherwise, a static buffer will be
2449  * used to store the fsname and returned to caller.
2450  */
2451 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2452 {
2453         static char fsname_static[MTI_NAME_MAXLEN];
2454         struct lustre_sb_info *lsi = s2lsi(sb);
2455         char *ptr;
2456         int len;
2457
2458         if (buf == NULL) {
2459                 /* this means the caller wants to use static buffer
2460                  * and it doesn't care about race. Usually this is
2461                  * in error reporting path */
2462                 buf = fsname_static;
2463                 buflen = sizeof(fsname_static);
2464         }
2465
2466         len = strlen(lsi->lsi_lmd->lmd_profile);
2467         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2468         if (ptr && (strcmp(ptr, "-client") == 0))
2469                 len -= 7;
2470
2471         if (unlikely(len >= buflen))
2472                 len = buflen - 1;
2473         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2474         buf[len] = '\0';
2475
2476         return buf;
2477 }
2478
2479 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2480 {
2481         char *path = NULL;
2482
2483         struct path p;
2484
2485         p.dentry = dentry;
2486         p.mnt = current->fs->root.mnt;
2487         path_get(&p);
2488         path = d_path(&p, buf, bufsize);
2489         path_put(&p);
2490         return path;
2491 }
2492
2493 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2494 {
2495         char *buf, *path = NULL;
2496         struct dentry *dentry = NULL;
2497         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2498
2499         /* this can be called inside spin lock so use GFP_ATOMIC. */
2500         buf = (char *)__get_free_page(GFP_ATOMIC);
2501         if (buf != NULL) {
2502                 dentry = d_find_alias(page->mapping->host);
2503                 if (dentry != NULL)
2504                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2505         }
2506
2507         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2508               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2509               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2510               PFID(&obj->cob_header.coh_lu.loh_fid),
2511               (path && !IS_ERR(path)) ? path : "", ioret);
2512
2513         if (dentry != NULL)
2514                 dput(dentry);
2515
2516         if (buf != NULL)
2517                 free_page((unsigned long)buf);
2518 }