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, 2012, Whamcloud, Inc.
  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 cfs_mem_cache_t *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) cfs_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         cfs_spin_lock_init(&sbi->ll_lock);
  89         cfs_mutex_init(&sbi->ll_lco.lco_lock);
  90         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  91         cfs_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 - CFS_PAGE_SHIFT) < 512) {
  97 #ifdef HAVE_BGL_SUPPORT
  98                 lru_page_max = pages / 4;
  99 #else
 100                 lru_page_max = pages / 2;
 101 #endif
 102         } else {
 103                 lru_page_max = (pages / 4) * 3;
 104         }
 105
 106         /* initialize lru data */
 107         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 108         sbi->ll_cache.ccc_lru_max = lru_page_max;
 109         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 110         cfs_spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 111         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 112
 113         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 114                                            SBI_DEFAULT_READAHEAD_MAX);
 115         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 117                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 118         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 119         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 120
 121         ll_generate_random_uuid(uuid);
 122         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 123         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 124
 125         cfs_spin_lock(&ll_sb_lock);
 126         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 127         cfs_spin_unlock(&ll_sb_lock);
 128
 129         sbi->ll_flags |= LL_SBI_VERBOSE;
 130 #ifdef ENABLE_CHECKSUM
 131         sbi->ll_flags |= LL_SBI_CHECKSUM;
 132 #endif
 133
 134 #ifdef HAVE_LRU_RESIZE_SUPPORT
 135         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 136 #endif
 137
 138         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 139                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 140                                    pp_r_hist.oh_lock);
 141                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 142                                    pp_w_hist.oh_lock);
 143         }
 144
 145         /* metadata statahead is enabled by default */
 146         sbi->ll_sa_max = LL_SA_RPC_DEF;
 147         cfs_atomic_set(&sbi->ll_sa_total, 0);
 148         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 149         cfs_atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151
 152         RETURN(sbi);
 153 }
 154
 155 void ll_free_sbi(struct super_block *sb)
 156 {
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         ENTRY;
 159
 160         if (sbi != NULL) {
 161                 cfs_spin_lock(&ll_sb_lock);
 162                 cfs_list_del(&sbi->ll_list);
 163                 cfs_spin_unlock(&ll_sb_lock);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static struct dentry_operations ll_d_root_ops = {
 170         .d_compare = ll_dcompare,
 171         .d_revalidate = ll_revalidate_nd,
 172 };
 173
 174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 175                                     struct vfsmount *mnt)
 176 {
 177         struct inode *root = 0;
 178         struct ll_sb_info *sbi = ll_s2sbi(sb);
 179         struct obd_device *obd;
 180         struct obd_capa *oc = NULL;
 181         struct obd_statfs *osfs = NULL;
 182         struct ptlrpc_request *request = NULL;
 183         struct obd_connect_data *data = NULL;
 184         struct obd_uuid *uuid;
 185         struct md_op_data *op_data;
 186         struct lustre_md lmd;
 187         obd_valid valid;
 188         int size, err, checksum;
 189         ENTRY;
 190
 191         obd = class_name2obd(md);
 192         if (!obd) {
 193                 CERROR("MD %s: not setup or attached\n", md);
 194                 RETURN(-EINVAL);
 195         }
 196
 197         OBD_ALLOC_PTR(data);
 198         if (data == NULL)
 199                 RETURN(-ENOMEM);
 200
 201         OBD_ALLOC_PTR(osfs);
 202         if (osfs == NULL) {
 203                 OBD_FREE_PTR(data);
 204                 RETURN(-ENOMEM);
 205         }
 206
 207         if (proc_lustre_fs_root) {
 208                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 209                                                   dt, md);
 210                 if (err < 0)
 211                         CERROR("could not register mount in /proc/fs/lustre\n");
 212         }
 213
 214         /* indicate the features supported by this client */
 215         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 216                                   OBD_CONNECT_ATTRFID  |
 217                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 218                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 219                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 220                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 221                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 222                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 223                                   OBD_CONNECT_EINPROGRESS |
 224                                   OBD_CONNECT_JOBSTATS;
 225
 226         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 227                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 228
 229 #ifdef HAVE_LRU_RESIZE_SUPPORT
 230         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 231                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 232 #endif
 233 #ifdef CONFIG_FS_POSIX_ACL
 234         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 235 #endif
 236
 237         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 238                 /* flag mdc connection as lightweight, only used for test
 239                  * purpose, use with care */
 240                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 241
 242         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 243         data->ocd_version = LUSTRE_VERSION_CODE;
 244
 245         if (sb->s_flags & MS_RDONLY)
 246                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 247         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 248                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 249
 250 #ifdef HAVE_MS_FLOCK_LOCK
 251         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 252         sb->s_flags |= MS_FLOCK_LOCK;
 253 #endif
 254 #ifdef MS_HAS_NEW_AOPS
 255         sb->s_flags |= MS_HAS_NEW_AOPS;
 256 #endif
 257
 258         if (sbi->ll_flags & LL_SBI_FLOCK)
 259                 sbi->ll_fop = &ll_file_operations_flock;
 260         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 261                 sbi->ll_fop = &ll_file_operations;
 262         else
 263                 sbi->ll_fop = &ll_file_operations_noflock;
 264
 265         /* real client */
 266         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 267         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 268                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 269
 270         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 271
 272         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 273         if (err == -EBUSY) {
 274                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 275                                    "recovery, of which this client is not a "
 276                                    "part. Please wait for recovery to complete,"
 277                                    " abort, or time out.\n", md);
 278                 GOTO(out, err);
 279         } else if (err) {
 280                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 281                 GOTO(out, err);
 282         }
 283
 284         err = obd_fid_init(sbi->ll_md_exp);
 285         if (err) {
 286                 CERROR("Can't init metadata layer FID infrastructure, "
 287                        "rc %d\n", err);
 288                 GOTO(out_md, err);
 289         }
 290
 291         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 292                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 293         if (err)
 294                 GOTO(out_md_fid, err);
 295
 296         /* This needs to be after statfs to ensure connect has finished.
 297          * Note that "data" does NOT contain the valid connect reply.
 298          * If connecting to a 1.8 server there will be no LMV device, so
 299          * we can access the MDC export directly and exp_connect_flags will
 300          * be non-zero, but if accessing an upgraded 2.1 server it will
 301          * have the correct flags filled in.
 302          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 303         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 304         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 305             valid != CLIENT_CONNECT_MDT_REQD) {
 306                 char *buf;
 307
 308                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 309                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 310                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 311                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 312                                    "feature(s) needed for correct operation "
 313                                    "of this client (%s). Please upgrade "
 314                                    "server or downgrade client.\n",
 315                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 316                 OBD_FREE(buf, CFS_PAGE_SIZE);
 317                 GOTO(out_md, err = -EPROTO);
 318         }
 319
 320         size = sizeof(*data);
 321         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 322                            KEY_CONN_DATA,  &size, data, NULL);
 323         if (err) {
 324                 CERROR("Get connect data failed: %d \n", err);
 325                 GOTO(out_md, err);
 326         }
 327
 328         LASSERT(osfs->os_bsize);
 329         sb->s_blocksize = osfs->os_bsize;
 330         sb->s_blocksize_bits = log2(osfs->os_bsize);
 331         sb->s_magic = LL_SUPER_MAGIC;
 332         sb->s_maxbytes = MAX_LFS_FILESIZE;
 333         sbi->ll_namelen = osfs->os_namelen;
 334         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 335
 336         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 337             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 338                 LCONSOLE_INFO("Disabling user_xattr feature because "
 339                               "it is not supported on the server\n");
 340                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 344 #ifdef MS_POSIXACL
 345                 sb->s_flags |= MS_POSIXACL;
 346 #endif
 347                 sbi->ll_flags |= LL_SBI_ACL;
 348         } else {
 349                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 350 #ifdef MS_POSIXACL
 351                 sb->s_flags &= ~MS_POSIXACL;
 352 #endif
 353                 sbi->ll_flags &= ~LL_SBI_ACL;
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 357                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 358                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client is set as remote by default.\n");
 360                 }
 361         } else {
 362                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 363                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client claims to be remote, but server "
 365                                       "rejected, forced to be local.\n");
 366                 }
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 370                 LCONSOLE_INFO("client enabled MDS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 375                 LCONSOLE_INFO("client enabled OSS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 380                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 383                 sbi->ll_md_brw_size = data->ocd_brw_size;
 384         else
 385                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 388                 LCONSOLE_INFO("Layout lock feature supported.\n");
 389                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 390         }
 391
 392         obd = class_name2obd(dt);
 393         if (!obd) {
 394                 CERROR("DT %s: not setup or attached\n", dt);
 395                 GOTO(out_md_fid, err = -ENODEV);
 396         }
 397
 398         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 399                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 400                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 401                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 402                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 403                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 404                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 405                                   OBD_CONNECT_MAXBYTES |
 406                                   OBD_CONNECT_EINPROGRESS |
 407                                   OBD_CONNECT_JOBSTATS;
 408
 409         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 410                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 411
 412         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 413                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 414                  * disabled by default, because it can still be enabled on the
 415                  * fly via /proc. As a consequence, we still need to come to an
 416                  * agreement on the supported algorithms at connect time */
 417                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 418
 419                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 420                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 421                 else
 422                         data->ocd_cksum_types = cksum_types_supported_client();
 423         }
 424
 425 #ifdef HAVE_LRU_RESIZE_SUPPORT
 426         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 427 #endif
 428         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 429                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 430
 431         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 432                "ocd_grant: %d\n", data->ocd_connect_flags,
 433                data->ocd_version, data->ocd_grant);
 434
 435         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 436         obd->obd_upcall.onu_upcall = cl_ocd_update;
 437
 438         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 439
 440         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 441         if (err == -EBUSY) {
 442                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 443                                    "recovery, of which this client is not a "
 444                                    "part.  Please wait for recovery to "
 445                                    "complete, abort, or time out.\n", dt);
 446                 GOTO(out_md_fid, err);
 447         } else if (err) {
 448                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 449                 GOTO(out_md_fid, err);
 450         }
 451
 452         err = obd_fid_init(sbi->ll_dt_exp);
 453         if (err) {
 454                 CERROR("Can't init data layer FID infrastructure, "
 455                        "rc %d\n", err);
 456                 GOTO(out_dt, err);
 457         }
 458
 459         cfs_mutex_lock(&sbi->ll_lco.lco_lock);
 460         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 461         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 462         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 463         cfs_mutex_unlock(&sbi->ll_lco.lco_lock);
 464
 465         fid_zero(&sbi->ll_root_fid);
 466         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 467         if (err) {
 468                 CERROR("cannot mds_connect: rc = %d\n", err);
 469                 GOTO(out_lock_cn_cb, err);
 470         }
 471         if (!fid_is_sane(&sbi->ll_root_fid)) {
 472                 CERROR("Invalid root fid during mount\n");
 473                 GOTO(out_lock_cn_cb, err = -EINVAL);
 474         }
 475         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 476
 477         sb->s_op = &lustre_super_operations;
 478 #if THREAD_SIZE >= 8192 /*b=17630*/
 479         sb->s_export_op = &lustre_export_operations;
 480 #endif
 481
 482         /* make root inode
 483          * XXX: move this to after cbd setup? */
 484         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 485         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 486                 valid |= OBD_MD_FLRMTPERM;
 487         else if (sbi->ll_flags & LL_SBI_ACL)
 488                 valid |= OBD_MD_FLACL;
 489
 490         OBD_ALLOC_PTR(op_data);
 491         if (op_data == NULL)
 492                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 493
 494         op_data->op_fid1 = sbi->ll_root_fid;
 495         op_data->op_mode = 0;
 496         op_data->op_capa1 = oc;
 497         op_data->op_valid = valid;
 498
 499         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 500         if (oc)
 501                 capa_put(oc);
 502         OBD_FREE_PTR(op_data);
 503         if (err) {
 504                 CERROR("md_getattr failed for root: rc = %d\n", err);
 505                 GOTO(out_lock_cn_cb, err);
 506         }
 507         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 508                                sbi->ll_md_exp, &lmd);
 509         if (err) {
 510                 CERROR("failed to understand root inode md: rc = %d\n", err);
 511                 ptlrpc_req_finished (request);
 512                 GOTO(out_lock_cn_cb, err);
 513         }
 514
 515         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 516         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 517         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 518         ptlrpc_req_finished(request);
 519
 520         if (root == NULL || IS_ERR(root)) {
 521                 if (lmd.lsm)
 522                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 523 #ifdef CONFIG_FS_POSIX_ACL
 524                 if (lmd.posix_acl) {
 525                         posix_acl_release(lmd.posix_acl);
 526                         lmd.posix_acl = NULL;
 527                 }
 528 #endif
 529                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 530                 root = NULL;
 531                 CERROR("lustre_lite: bad iget4 for root\n");
 532                 GOTO(out_root, err);
 533         }
 534
 535         err = ll_close_thread_start(&sbi->ll_lcq);
 536         if (err) {
 537                 CERROR("cannot start close thread: rc %d\n", err);
 538                 GOTO(out_root, err);
 539         }
 540
 541 #ifdef CONFIG_FS_POSIX_ACL
 542         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 543                 rct_init(&sbi->ll_rct);
 544                 et_init(&sbi->ll_et);
 545         }
 546 #endif
 547
 548         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 549         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 550                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 551                                  NULL);
 552         cl_sb_init(sb);
 553
 554         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 555                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 556                                  &sbi->ll_cache, NULL);
 557
 558         sb->s_root = d_alloc_root(root);
 559 #ifdef HAVE_DCACHE_LOCK
 560         sb->s_root->d_op = &ll_d_root_ops;
 561 #else
 562         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 563         d_set_d_op(sb->s_root, &ll_d_root_ops);
 564         sb->s_d_op = &ll_d_ops;
 565 #endif
 566
 567         sbi->ll_sdev_orig = sb->s_dev;
 568
 569         /* We set sb->s_dev equal on all lustre clients in order to support
 570          * NFS export clustering.  NFSD requires that the FSID be the same
 571          * on all clients. */
 572         /* s_dev is also used in lt_compare() to compare two fs, but that is
 573          * only a node-local comparison. */
 574         uuid = obd_get_uuid(sbi->ll_md_exp);
 575         if (uuid != NULL)
 576                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 577
 578         if (data != NULL)
 579                 OBD_FREE_PTR(data);
 580         if (osfs != NULL)
 581                 OBD_FREE_PTR(osfs);
 582
 583         RETURN(err);
 584 out_root:
 585         if (root)
 586                 iput(root);
 587 out_lock_cn_cb:
 588         obd_fid_fini(sbi->ll_dt_exp);
 589 out_dt:
 590         obd_disconnect(sbi->ll_dt_exp);
 591         sbi->ll_dt_exp = NULL;
 592 out_md_fid:
 593         obd_fid_fini(sbi->ll_md_exp);
 594 out_md:
 595         obd_disconnect(sbi->ll_md_exp);
 596         sbi->ll_md_exp = NULL;
 597 out:
 598         if (data != NULL)
 599                 OBD_FREE_PTR(data);
 600         if (osfs != NULL)
 601                 OBD_FREE_PTR(osfs);
 602         lprocfs_unregister_mountpoint(sbi);
 603         return err;
 604 }
 605
 606 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 607 {
 608         int size, rc;
 609
 610         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 611         size = sizeof(int);
 612         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 613                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 614         if (rc)
 615                 CERROR("Get max mdsize error rc %d \n", rc);
 616
 617         RETURN(rc);
 618 }
 619
 620 void ll_dump_inode(struct inode *inode)
 621 {
 622         struct list_head *tmp;
 623         int dentry_count = 0;
 624
 625         LASSERT(inode != NULL);
 626
 627         list_for_each(tmp, &inode->i_dentry)
 628                 dentry_count++;
 629
 630         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 631                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 632                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 633 }
 634
 635 void lustre_dump_dentry(struct dentry *dentry, int recur)
 636 {
 637         struct list_head *tmp;
 638         int subdirs = 0;
 639
 640         LASSERT(dentry != NULL);
 641
 642         list_for_each(tmp, &dentry->d_subdirs)
 643                 subdirs++;
 644
 645         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 646                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 647                dentry->d_name.len, dentry->d_name.name,
 648                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 649                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 650                dentry->d_flags, dentry->d_fsdata, subdirs);
 651         if (dentry->d_inode != NULL)
 652                 ll_dump_inode(dentry->d_inode);
 653
 654         if (recur == 0)
 655                 return;
 656
 657         list_for_each(tmp, &dentry->d_subdirs) {
 658                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 659                 lustre_dump_dentry(d, recur - 1);
 660         }
 661 }
 662
 663 void client_common_put_super(struct super_block *sb)
 664 {
 665         struct ll_sb_info *sbi = ll_s2sbi(sb);
 666         ENTRY;
 667
 668 #ifdef CONFIG_FS_POSIX_ACL
 669         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 670                 et_fini(&sbi->ll_et);
 671                 rct_fini(&sbi->ll_rct);
 672         }
 673 #endif
 674
 675         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 676
 677         ll_close_thread_shutdown(sbi->ll_lcq);
 678
 679         cl_sb_fini(sb);
 680
 681         cfs_list_del(&sbi->ll_conn_chain);
 682
 683         obd_fid_fini(sbi->ll_dt_exp);
 684         obd_disconnect(sbi->ll_dt_exp);
 685         sbi->ll_dt_exp = NULL;
 686
 687         lprocfs_unregister_mountpoint(sbi);
 688
 689         obd_fid_fini(sbi->ll_md_exp);
 690         obd_disconnect(sbi->ll_md_exp);
 691         sbi->ll_md_exp = NULL;
 692
 693         EXIT;
 694 }
 695
 696 void ll_kill_super(struct super_block *sb)
 697 {
 698         struct ll_sb_info *sbi;
 699
 700         ENTRY;
 701
 702         /* not init sb ?*/
 703         if (!(sb->s_flags & MS_ACTIVE))
 704                 return;
 705
 706         sbi = ll_s2sbi(sb);
 707         /* we need restore s_dev from changed for clustred NFS before put_super
 708          * because new kernels have cached s_dev and change sb->s_dev in
 709          * put_super not affected real removing devices */
 710         if (sbi)
 711                 sb->s_dev = sbi->ll_sdev_orig;
 712         EXIT;
 713 }
 714
 715 char *ll_read_opt(const char *opt, char *data)
 716 {
 717         char *value;
 718         char *retval;
 719         ENTRY;
 720
 721         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 722         if (strncmp(opt, data, strlen(opt)))
 723                 RETURN(NULL);
 724         if ((value = strchr(data, '=')) == NULL)
 725                 RETURN(NULL);
 726
 727         value++;
 728         OBD_ALLOC(retval, strlen(value) + 1);
 729         if (!retval) {
 730                 CERROR("out of memory!\n");
 731                 RETURN(NULL);
 732         }
 733
 734         memcpy(retval, value, strlen(value)+1);
 735         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 736         RETURN(retval);
 737 }
 738
 739 static inline int ll_set_opt(const char *opt, char *data, int fl)
 740 {
 741         if (strncmp(opt, data, strlen(opt)) != 0)
 742                 return(0);
 743         else
 744                 return(fl);
 745 }
 746
 747 /* non-client-specific mount options are parsed in lmd_parse */
 748 static int ll_options(char *options, int *flags)
 749 {
 750         int tmp;
 751         char *s1 = options, *s2;
 752         ENTRY;
 753
 754         if (!options)
 755                 RETURN(0);
 756
 757         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 758
 759         while (*s1) {
 760                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 761                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 767                 if (tmp) {
 768                         *flags |= tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 777                 if (tmp) {
 778                         *flags &= ~tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 787                 if (tmp) {
 788                         *flags &= ~tmp;
 789                         goto next;
 790                 }
 791 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 792                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 793                 if (tmp) {
 794                         /* Ignore deprecated mount option.  The client will
 795                          * always try to mount with ACL support, whether this
 796                          * is used depends on whether server supports it. */
 797                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 798                                                   "mount option 'acl'.\n");
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 804                                                   "mount option 'noacl'.\n");
 805                         goto next;
 806                 }
 807 #else
 808 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 809 #endif
 810                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 821                 if (tmp) {
 822                         *flags &= ~tmp;
 823                         goto next;
 824                 }
 825
 826                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 877                                    s1);
 878                 RETURN(-EINVAL);
 879
 880 next:
 881                 /* Find next opt */
 882                 s2 = strchr(s1, ',');
 883                 if (s2 == NULL)
 884                         break;
 885                 s1 = s2 + 1;
 886         }
 887         RETURN(0);
 888 }
 889
 890 void ll_lli_init(struct ll_inode_info *lli)
 891 {
 892         lli->lli_inode_magic = LLI_INODE_MAGIC;
 893         lli->lli_flags = 0;
 894         lli->lli_ioepoch = 0;
 895         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 896         cfs_spin_lock_init(&lli->lli_lock);
 897         lli->lli_posix_acl = NULL;
 898         lli->lli_remote_perms = NULL;
 899         cfs_mutex_init(&lli->lli_rmtperm_mutex);
 900         /* Do not set lli_fid, it has been initialized already. */
 901         fid_zero(&lli->lli_pfid);
 902         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 903         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 904         cfs_atomic_set(&lli->lli_open_count, 0);
 905         lli->lli_mds_capa = NULL;
 906         lli->lli_rmtperm_time = 0;
 907         lli->lli_pending_och = NULL;
 908         lli->lli_mds_read_och = NULL;
 909         lli->lli_mds_write_och = NULL;
 910         lli->lli_mds_exec_och = NULL;
 911         lli->lli_open_fd_read_count = 0;
 912         lli->lli_open_fd_write_count = 0;
 913         lli->lli_open_fd_exec_count = 0;
 914         cfs_mutex_init(&lli->lli_och_mutex);
 915         cfs_spin_lock_init(&lli->lli_agl_lock);
 916         lli->lli_has_smd = false;
 917         lli->lli_clob = NULL;
 918
 919         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 920         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 921                 cfs_mutex_init(&lli->lli_readdir_mutex);
 922                 lli->lli_opendir_key = NULL;
 923                 lli->lli_sai = NULL;
 924                 lli->lli_def_acl = NULL;
 925                 cfs_spin_lock_init(&lli->lli_sa_lock);
 926                 lli->lli_opendir_pid = 0;
 927         } else {
 928                 cfs_sema_init(&lli->lli_size_sem, 1);
 929                 lli->lli_size_sem_owner = NULL;
 930                 lli->lli_symlink_name = NULL;
 931                 cfs_init_rwsem(&lli->lli_trunc_sem);
 932                 cfs_mutex_init(&lli->lli_write_mutex);
 933                 cfs_init_rwsem(&lli->lli_glimpse_sem);
 934                 lli->lli_glimpse_time = 0;
 935                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 936                 lli->lli_agl_index = 0;
 937                 lli->lli_async_rc = 0;
 938         }
 939         cfs_mutex_init(&lli->lli_layout_mutex);
 940 }
 941
 942 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 943 {
 944 #ifdef HAVE_BDI_REGISTER
 945         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 946
 947 #ifdef HAVE_BDI_NAME
 948         bdi->name = "lustre";
 949 #endif
 950         return bdi_register(bdi, NULL, "lustre-%d",
 951                             atomic_inc_return(&ll_bdi_num));
 952 #else
 953         return 0;
 954 #endif
 955 }
 956
 957 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 958 {
 959         struct lustre_profile *lprof = NULL;
 960         struct lustre_sb_info *lsi = s2lsi(sb);
 961         struct ll_sb_info *sbi;
 962         char  *dt = NULL, *md = NULL;
 963         char  *profilenm = get_profile_name(sb);
 964         struct config_llog_instance *cfg;
 965         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 966         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 967         int    err;
 968         ENTRY;
 969
 970         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 971
 972         OBD_ALLOC_PTR(cfg);
 973         if (cfg == NULL)
 974                 RETURN(-ENOMEM);
 975
 976         cfs_module_get();
 977
 978         /* client additional sb info */
 979         lsi->lsi_llsbi = sbi = ll_init_sbi();
 980         if (!sbi) {
 981                 cfs_module_put(THIS_MODULE);
 982                 OBD_FREE_PTR(cfg);
 983                 RETURN(-ENOMEM);
 984         }
 985
 986         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 987         if (err)
 988                 GOTO(out_free, err);
 989
 990         err = bdi_init(&lsi->lsi_bdi);
 991         if (err)
 992                 GOTO(out_free, err);
 993         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 994         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 995         err = ll_bdi_register(&lsi->lsi_bdi);
 996         if (err)
 997                 GOTO(out_free, err);
 998
 999 #ifdef HAVE_SB_BDI
1000         sb->s_bdi = &lsi->lsi_bdi;
1001 #endif
1002
1003         /* Generate a string unique to this super, in case some joker tries
1004            to mount the same fs at two mount points.
1005            Use the address of the super itself.*/
1006         cfg->cfg_instance = sb;
1007         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1008         cfg->cfg_callback = class_config_llog_handler;
1009         /* set up client obds */
1010         err = lustre_process_log(sb, profilenm, cfg);
1011         if (err < 0) {
1012                 CERROR("Unable to process log: %d\n", err);
1013                 GOTO(out_free, err);
1014         }
1015
1016         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1017         lprof = class_get_profile(profilenm);
1018         if (lprof == NULL) {
1019                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1020                                    " read from the MGS.  Does that filesystem "
1021                                    "exist?\n", profilenm);
1022                 GOTO(out_free, err = -EINVAL);
1023         }
1024         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1025                lprof->lp_md, lprof->lp_dt);
1026
1027         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1028         if (!dt)
1029                 GOTO(out_free, err = -ENOMEM);
1030         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1031
1032         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1033         if (!md)
1034                 GOTO(out_free, err = -ENOMEM);
1035         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1036
1037         /* connections, registrations, sb setup */
1038         err = client_common_fill_super(sb, md, dt, mnt);
1039
1040 out_free:
1041         if (md)
1042                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1043         if (dt)
1044                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1045         if (err)
1046                 ll_put_super(sb);
1047         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1048                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1049
1050         OBD_FREE_PTR(cfg);
1051         RETURN(err);
1052 } /* ll_fill_super */
1053
1054
1055 void lu_context_keys_dump(void);
1056
1057 void ll_put_super(struct super_block *sb)
1058 {
1059         struct config_llog_instance cfg;
1060         struct obd_device *obd;
1061         struct lustre_sb_info *lsi = s2lsi(sb);
1062         struct ll_sb_info *sbi = ll_s2sbi(sb);
1063         char *profilenm = get_profile_name(sb);
1064         int force = 1, next;
1065         ENTRY;
1066
1067         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1068
1069         ll_print_capa_stat(sbi);
1070
1071         cfg.cfg_instance = sb;
1072         lustre_end_log(sb, profilenm, &cfg);
1073
1074         if (sbi->ll_md_exp) {
1075                 obd = class_exp2obd(sbi->ll_md_exp);
1076                 if (obd)
1077                         force = obd->obd_force;
1078         }
1079
1080         /* We need to set force before the lov_disconnect in
1081            lustre_common_put_super, since l_d cleans up osc's as well. */
1082         if (force) {
1083                 next = 0;
1084                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1085                                                      &next)) != NULL) {
1086                         obd->obd_force = force;
1087                 }
1088         }
1089
1090         if (sbi->ll_lcq) {
1091                 /* Only if client_common_fill_super succeeded */
1092                 client_common_put_super(sb);
1093         }
1094
1095         next = 0;
1096         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1097                 class_manual_cleanup(obd);
1098         }
1099
1100         if (sbi->ll_flags & LL_SBI_VERBOSE)
1101                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1102
1103         if (profilenm)
1104                 class_del_profile(profilenm);
1105
1106         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1107                 bdi_destroy(&lsi->lsi_bdi);
1108                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1109         }
1110
1111         ll_free_sbi(sb);
1112         lsi->lsi_llsbi = NULL;
1113
1114         lustre_common_put_super(sb);
1115
1116         cl_env_cache_purge(~0);
1117
1118         cfs_module_put(THIS_MODULE);
1119
1120         EXIT;
1121 } /* client_put_super */
1122
1123 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1124 {
1125         struct inode *inode = NULL;
1126         /* NOTE: we depend on atomic igrab() -bzzz */
1127         lock_res_and_lock(lock);
1128         if (lock->l_ast_data) {
1129                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1130                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1131                         inode = igrab(lock->l_ast_data);
1132                 } else {
1133                         inode = lock->l_ast_data;
1134                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1135                                          D_WARNING, lock, "l_ast_data %p is "
1136                                          "bogus: magic %08x", lock->l_ast_data,
1137                                          lli->lli_inode_magic);
1138                         inode = NULL;
1139                 }
1140         }
1141         unlock_res_and_lock(lock);
1142         return inode;
1143 }
1144
1145 static int null_if_equal(struct ldlm_lock *lock, void *data)
1146 {
1147         if (data == lock->l_ast_data) {
1148                 lock->l_ast_data = NULL;
1149
1150                 if (lock->l_req_mode != lock->l_granted_mode)
1151                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1152         }
1153
1154         return LDLM_ITER_CONTINUE;
1155 }
1156
1157 void ll_clear_inode(struct inode *inode)
1158 {
1159         struct ll_inode_info *lli = ll_i2info(inode);
1160         struct ll_sb_info *sbi = ll_i2sbi(inode);
1161         ENTRY;
1162
1163         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1164                inode->i_generation, inode);
1165
1166         if (S_ISDIR(inode->i_mode)) {
1167                 /* these should have been cleared in ll_file_release */
1168                 LASSERT(lli->lli_opendir_key == NULL);
1169                 LASSERT(lli->lli_sai == NULL);
1170                 LASSERT(lli->lli_opendir_pid == 0);
1171         }
1172
1173         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1174         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1175                          null_if_equal, inode);
1176
1177         LASSERT(!lli->lli_open_fd_write_count);
1178         LASSERT(!lli->lli_open_fd_read_count);
1179         LASSERT(!lli->lli_open_fd_exec_count);
1180
1181         if (lli->lli_mds_write_och)
1182                 ll_md_real_close(inode, FMODE_WRITE);
1183         if (lli->lli_mds_exec_och)
1184                 ll_md_real_close(inode, FMODE_EXEC);
1185         if (lli->lli_mds_read_och)
1186                 ll_md_real_close(inode, FMODE_READ);
1187
1188         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1189                 OBD_FREE(lli->lli_symlink_name,
1190                          strlen(lli->lli_symlink_name) + 1);
1191                 lli->lli_symlink_name = NULL;
1192         }
1193
1194         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1195                 LASSERT(lli->lli_posix_acl == NULL);
1196                 if (lli->lli_remote_perms) {
1197                         free_rmtperm_hash(lli->lli_remote_perms);
1198                         lli->lli_remote_perms = NULL;
1199                 }
1200         }
1201 #ifdef CONFIG_FS_POSIX_ACL
1202         else if (lli->lli_posix_acl) {
1203                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1204                 LASSERT(lli->lli_remote_perms == NULL);
1205                 posix_acl_release(lli->lli_posix_acl);
1206                 lli->lli_posix_acl = NULL;
1207         }
1208 #endif
1209         lli->lli_inode_magic = LLI_INODE_DEAD;
1210
1211         ll_clear_inode_capas(inode);
1212         if (!S_ISDIR(inode->i_mode))
1213                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1214
1215         /*
1216          * XXX This has to be done before lsm is freed below, because
1217          * cl_object still uses inode lsm.
1218          */
1219         cl_inode_fini(inode);
1220         lli->lli_has_smd = false;
1221
1222         EXIT;
1223 }
1224
1225 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1226                   struct md_open_data **mod)
1227 {
1228         struct lustre_md md;
1229         struct inode *inode = dentry->d_inode;
1230         struct ll_sb_info *sbi = ll_i2sbi(inode);
1231         struct ptlrpc_request *request = NULL;
1232         int rc, ia_valid;
1233         ENTRY;
1234
1235         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1236                                      LUSTRE_OPC_ANY, NULL);
1237         if (IS_ERR(op_data))
1238                 RETURN(PTR_ERR(op_data));
1239
1240         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1241                         &request, mod);
1242         if (rc) {
1243                 ptlrpc_req_finished(request);
1244                 if (rc == -ENOENT) {
1245                         clear_nlink(inode);
1246                         /* Unlinked special device node? Or just a race?
1247                          * Pretend we done everything. */
1248                         if (!S_ISREG(inode->i_mode) &&
1249                             !S_ISDIR(inode->i_mode)) {
1250                                 ia_valid = op_data->op_attr.ia_valid;
1251                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1252                                 rc = simple_setattr(dentry, &op_data->op_attr);
1253                                 op_data->op_attr.ia_valid = ia_valid;
1254                         }
1255                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1256                         CERROR("md_setattr fails: rc = %d\n", rc);
1257                 }
1258                 RETURN(rc);
1259         }
1260
1261         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1262                               sbi->ll_md_exp, &md);
1263         if (rc) {
1264                 ptlrpc_req_finished(request);
1265                 RETURN(rc);
1266         }
1267
1268         /* We want to adjust timestamps.
1269          * If there is at least some data in file, we cleared ATTR_SIZE
1270          * to avoid update size, otherwise it is important to do.(SOM case)
1271          * (bug 6196) */
1272         ia_valid = op_data->op_attr.ia_valid;
1273         /* Since we set ATTR_*_SET flags above, and already done permission
1274          * check, So don't let inode_change_ok() check it again.  */
1275         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1276         rc = simple_setattr(dentry, &op_data->op_attr);
1277         op_data->op_attr.ia_valid = ia_valid;
1278
1279         /* Extract epoch data if obtained. */
1280         op_data->op_handle = md.body->handle;
1281         op_data->op_ioepoch = md.body->ioepoch;
1282
1283         ll_update_inode(inode, &md);
1284         ptlrpc_req_finished(request);
1285
1286         RETURN(rc);
1287 }
1288
1289 /* Close IO epoch and send Size-on-MDS attribute update. */
1290 static int ll_setattr_done_writing(struct inode *inode,
1291                                    struct md_op_data *op_data,
1292                                    struct md_open_data *mod)
1293 {
1294         struct ll_inode_info *lli = ll_i2info(inode);
1295         int rc = 0;
1296         ENTRY;
1297
1298         LASSERT(op_data != NULL);
1299         if (!S_ISREG(inode->i_mode))
1300                 RETURN(0);
1301
1302         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1303                op_data->op_ioepoch, PFID(&lli->lli_fid));
1304
1305         op_data->op_flags = MF_EPOCH_CLOSE;
1306         ll_done_writing_attr(inode, op_data);
1307         ll_pack_inode2opdata(inode, op_data, NULL);
1308
1309         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1310         if (rc == -EAGAIN) {
1311                 /* MDS has instructed us to obtain Size-on-MDS attribute
1312                  * from OSTs and send setattr to back to MDS. */
1313                 rc = ll_som_update(inode, op_data);
1314         } else if (rc) {
1315                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1316                        inode->i_ino, rc);
1317         }
1318         RETURN(rc);
1319 }
1320
1321 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1322 {
1323         struct obd_capa *capa;
1324         int rc;
1325
1326         if (attr->ia_valid & ATTR_SIZE)
1327                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1328         else
1329                 capa = ll_mdscapa_get(inode);
1330
1331         rc = cl_setattr_ost(inode, attr, capa);
1332
1333         if (attr->ia_valid & ATTR_SIZE)
1334                 ll_truncate_free_capa(capa);
1335         else
1336                 capa_put(capa);
1337
1338         return rc;
1339 }
1340
1341 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1342 /**
1343  * inode_newsize_ok - may this inode be truncated to a given size
1344  * @inode:      the inode to be truncated
1345  * @offset:     the new size to assign to the inode
1346  * @Returns:    0 on success, -ve errno on failure
1347  *
1348  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1349  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1350  * when necessary. Caller must not proceed with inode size change if failure is
1351  * returned. @inode must be a file (not directory), with appropriate
1352  * permissions to allow truncate (inode_newsize_ok does NOT check these
1353  * conditions).
1354  *
1355  * inode_newsize_ok must be called with i_mutex held.
1356  */
1357 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1358 {
1359         if (inode->i_size < offset) {
1360                 unsigned long limit;
1361
1362                 limit = rlimit(RLIMIT_FSIZE);
1363                 if (limit != RLIM_INFINITY && offset > limit)
1364                         goto out_sig;
1365                 if (offset > inode->i_sb->s_maxbytes)
1366                         goto out_big;
1367         } else {
1368                 /*
1369                  * truncation of in-use swapfiles is disallowed - it would
1370                  * cause subsequent swapout to scribble on the now-freed
1371                  * blocks.
1372                  */
1373                 if (IS_SWAPFILE(inode))
1374                         return -ETXTBSY;
1375         }
1376
1377         return 0;
1378 out_sig:
1379         send_sig(SIGXFSZ, current, 0);
1380 out_big:
1381         return -EFBIG;
1382 }
1383 #endif
1384
1385 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1386  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1387  * keep these values until such a time that objects are allocated for it.
1388  * We do the MDS operations first, as it is checking permissions for us.
1389  * We don't to the MDS RPC if there is nothing that we want to store there,
1390  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1391  * going to do an RPC anyways.
1392  *
1393  * If we are doing a truncate, we will send the mtime and ctime updates
1394  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1395  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1396  * at the same time.
1397  */
1398 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1399 {
1400         struct lov_stripe_md *lsm;
1401         struct inode *inode = dentry->d_inode;
1402         struct ll_inode_info *lli = ll_i2info(inode);
1403         struct md_op_data *op_data = NULL;
1404         struct md_open_data *mod = NULL;
1405         int ia_valid = attr->ia_valid;
1406         int rc = 0, rc1 = 0;
1407         ENTRY;
1408
1409         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu (%p) to %llu, valid %x\n",
1410                 inode->i_ino,
1411                 inode,i_size_read(inode),
1412                 attr->ia_valid);
1413
1414         if (ia_valid & ATTR_SIZE) {
1415                 /* Check new size against VFS/VM file size limit and rlimit */
1416                 rc = inode_newsize_ok(inode, attr->ia_size);
1417                 if (rc)
1418                         RETURN(rc);
1419
1420                 /* The maximum Lustre file size is variable, based on the
1421                  * OST maximum object size and number of stripes.  This
1422                  * needs another check in addition to the VFS check above. */
1423                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1424                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1425                                PFID(&lli->lli_fid), attr->ia_size,
1426                                ll_file_maxbytes(inode));
1427                         RETURN(-EFBIG);
1428                 }
1429
1430                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1431         }
1432
1433         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1434         if (ia_valid & TIMES_SET_FLAGS) {
1435                 if (cfs_curproc_fsuid() != inode->i_uid &&
1436                     !cfs_capable(CFS_CAP_FOWNER))
1437                         RETURN(-EPERM);
1438         }
1439
1440         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1441         if (attr->ia_valid & ATTR_CTIME) {
1442                 attr->ia_ctime = CFS_CURRENT_TIME;
1443                 attr->ia_valid |= ATTR_CTIME_SET;
1444         }
1445         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1446                 attr->ia_atime = CFS_CURRENT_TIME;
1447                 attr->ia_valid |= ATTR_ATIME_SET;
1448         }
1449         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1450                 attr->ia_mtime = CFS_CURRENT_TIME;
1451                 attr->ia_valid |= ATTR_MTIME_SET;
1452         }
1453
1454         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1455                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1456                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1457                        cfs_time_current_sec());
1458
1459         /* We always do an MDS RPC, even if we're only changing the size;
1460          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1461
1462         OBD_ALLOC_PTR(op_data);
1463         if (op_data == NULL)
1464                 RETURN(-ENOMEM);
1465
1466         if (!S_ISDIR(inode->i_mode)) {
1467                 if (ia_valid & ATTR_SIZE)
1468                         inode_dio_write_done(inode);
1469                 mutex_unlock(&inode->i_mutex);
1470                 cfs_down_write(&lli->lli_trunc_sem);
1471                 mutex_lock(&inode->i_mutex);
1472                 if (ia_valid & ATTR_SIZE)
1473                         inode_dio_wait(inode);
1474         }
1475
1476         /* We need a steady stripe configuration for setattr to avoid
1477          * confusion. */
1478         lsm = ccc_inode_lsm_get(inode);
1479
1480         /* NB: ATTR_SIZE will only be set after this point if the size
1481          * resides on the MDS, ie, this file has no objects. */
1482         if (lsm != NULL)
1483                 attr->ia_valid &= ~ATTR_SIZE;
1484
1485         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1486
1487         /* Open epoch for truncate. */
1488         if (exp_connect_som(ll_i2mdexp(inode)) &&
1489             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1490                 op_data->op_flags = MF_EPOCH_OPEN;
1491
1492         rc = ll_md_setattr(dentry, op_data, &mod);
1493         if (rc)
1494                 GOTO(out, rc);
1495
1496         ll_ioepoch_open(lli, op_data->op_ioepoch);
1497         if (lsm == NULL || !S_ISREG(inode->i_mode)) {
1498                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1499                 GOTO(out, rc = 0);
1500         }
1501
1502         if (ia_valid & ATTR_SIZE)
1503                 attr->ia_valid |= ATTR_SIZE;
1504         if (ia_valid & (ATTR_SIZE |
1505                         ATTR_ATIME | ATTR_ATIME_SET |
1506                         ATTR_MTIME | ATTR_MTIME_SET))
1507                 /* For truncate and utimes sending attributes to OSTs, setting
1508                  * mtime/atime to the past will be performed under PW [0:EOF]
1509                  * extent lock (new_size:EOF for truncate).  It may seem
1510                  * excessive to send mtime/atime updates to OSTs when not
1511                  * setting times to past, but it is necessary due to possible
1512                  * time de-synchronization between MDT inode and OST objects */
1513                 rc = ll_setattr_ost(inode, attr);
1514         EXIT;
1515 out:
1516         ccc_inode_lsm_put(inode, lsm);
1517         if (op_data) {
1518                 if (op_data->op_ioepoch) {
1519                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1520                         if (!rc)
1521                                 rc = rc1;
1522                 }
1523                 ll_finish_md_op_data(op_data);
1524         }
1525         if (!S_ISDIR(inode->i_mode))
1526                 cfs_up_write(&lli->lli_trunc_sem);
1527
1528         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1529                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1530
1531         return rc;
1532 }
1533
1534 int ll_setattr(struct dentry *de, struct iattr *attr)
1535 {
1536         int mode = de->d_inode->i_mode;
1537
1538         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1539                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1540                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1541
1542         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1543                                (ATTR_SIZE|ATTR_MODE)) &&
1544             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1545              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1546               !(attr->ia_mode & S_ISGID))))
1547                 attr->ia_valid |= ATTR_FORCE;
1548
1549         if ((mode & S_ISUID) &&
1550             !(attr->ia_mode & S_ISUID) &&
1551             !(attr->ia_valid & ATTR_KILL_SUID))
1552                 attr->ia_valid |= ATTR_KILL_SUID;
1553
1554         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1555             !(attr->ia_mode & S_ISGID) &&
1556             !(attr->ia_valid & ATTR_KILL_SGID))
1557                 attr->ia_valid |= ATTR_KILL_SGID;
1558
1559         return ll_setattr_raw(de, attr);
1560 }
1561
1562 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1563                        __u64 max_age, __u32 flags)
1564 {
1565         struct ll_sb_info *sbi = ll_s2sbi(sb);
1566         struct obd_statfs obd_osfs;
1567         int rc;
1568         ENTRY;
1569
1570         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1571         if (rc) {
1572                 CERROR("md_statfs fails: rc = %d\n", rc);
1573                 RETURN(rc);
1574         }
1575
1576         osfs->os_type = sb->s_magic;
1577
1578         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1579                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1580
1581         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1582                 flags |= OBD_STATFS_NODELAY;
1583
1584         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1585         if (rc) {
1586                 CERROR("obd_statfs fails: rc = %d\n", rc);
1587                 RETURN(rc);
1588         }
1589
1590         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1591                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1592                obd_osfs.os_files);
1593
1594         osfs->os_bsize = obd_osfs.os_bsize;
1595         osfs->os_blocks = obd_osfs.os_blocks;
1596         osfs->os_bfree = obd_osfs.os_bfree;
1597         osfs->os_bavail = obd_osfs.os_bavail;
1598
1599         /* If we don't have as many objects free on the OST as inodes
1600          * on the MDS, we reduce the total number of inodes to
1601          * compensate, so that the "inodes in use" number is correct.
1602          */
1603         if (obd_osfs.os_ffree < osfs->os_ffree) {
1604                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1605                         obd_osfs.os_ffree;
1606                 osfs->os_ffree = obd_osfs.os_ffree;
1607         }
1608
1609         RETURN(rc);
1610 }
1611 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1612 {
1613         struct super_block *sb = de->d_sb;
1614         struct obd_statfs osfs;
1615         int rc;
1616
1617         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1618         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1619
1620         /* Some amount of caching on the client is allowed */
1621         rc = ll_statfs_internal(sb, &osfs,
1622                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1623                                 0);
1624         if (rc)
1625                 return rc;
1626
1627         statfs_unpack(sfs, &osfs);
1628
1629         /* We need to downshift for all 32-bit kernels, because we can't
1630          * tell if the kernel is being called via sys_statfs64() or not.
1631          * Stop before overflowing f_bsize - in which case it is better
1632          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1633         if (sizeof(long) < 8) {
1634                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1635                         sfs->f_bsize <<= 1;
1636
1637                         osfs.os_blocks >>= 1;
1638                         osfs.os_bfree >>= 1;
1639                         osfs.os_bavail >>= 1;
1640                 }
1641         }
1642
1643         sfs->f_blocks = osfs.os_blocks;
1644         sfs->f_bfree = osfs.os_bfree;
1645         sfs->f_bavail = osfs.os_bavail;
1646
1647         return 0;
1648 }
1649
1650 void ll_inode_size_lock(struct inode *inode)
1651 {
1652         struct ll_inode_info *lli;
1653
1654         LASSERT(!S_ISDIR(inode->i_mode));
1655
1656         lli = ll_i2info(inode);
1657         LASSERT(lli->lli_size_sem_owner != current);
1658         cfs_down(&lli->lli_size_sem);
1659         LASSERT(lli->lli_size_sem_owner == NULL);
1660         lli->lli_size_sem_owner = current;
1661 }
1662
1663 void ll_inode_size_unlock(struct inode *inode)
1664 {
1665         struct ll_inode_info *lli;
1666
1667         lli = ll_i2info(inode);
1668         LASSERT(lli->lli_size_sem_owner == current);
1669         lli->lli_size_sem_owner = NULL;
1670         cfs_up(&lli->lli_size_sem);
1671 }
1672
1673 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1674 {
1675         struct ll_inode_info *lli = ll_i2info(inode);
1676         struct mdt_body *body = md->body;
1677         struct lov_stripe_md *lsm = md->lsm;
1678         struct ll_sb_info *sbi = ll_i2sbi(inode);
1679
1680         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1681         if (lsm != NULL) {
1682                 LASSERT(S_ISREG(inode->i_mode));
1683                 cfs_mutex_lock(&lli->lli_och_mutex);
1684                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1685                                 lsm, inode->i_ino, inode->i_generation, inode);
1686                 /* cl_file_inode_init must go before lli_has_smd or a race
1687                  * is possible where client thinks the file has stripes,
1688                  * but lov raid0 is not setup yet and parallel e.g.
1689                  * glimpse would try to use uninitialized lov */
1690                 if (cl_file_inode_init(inode, md) == 0)
1691                         lli->lli_has_smd = true;
1692                 cfs_mutex_unlock(&lli->lli_och_mutex);
1693
1694                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1695                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1696                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1697                 if (md->lsm != NULL)
1698                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1699         }
1700
1701         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1702                 if (body->valid & OBD_MD_FLRMTPERM)
1703                         ll_update_remote_perm(inode, md->remote_perm);
1704         }
1705 #ifdef CONFIG_FS_POSIX_ACL
1706         else if (body->valid & OBD_MD_FLACL) {
1707                 cfs_spin_lock(&lli->lli_lock);
1708                 if (lli->lli_posix_acl)
1709                         posix_acl_release(lli->lli_posix_acl);
1710                 lli->lli_posix_acl = md->posix_acl;
1711                 cfs_spin_unlock(&lli->lli_lock);
1712         }
1713 #endif
1714         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1715         inode->i_generation = cl_fid_build_gen(&body->fid1);
1716
1717         if (body->valid & OBD_MD_FLATIME) {
1718                 if (body->atime > LTIME_S(inode->i_atime))
1719                         LTIME_S(inode->i_atime) = body->atime;
1720                 lli->lli_lvb.lvb_atime = body->atime;
1721         }
1722         if (body->valid & OBD_MD_FLMTIME) {
1723                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1724                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1725                                "to "LPU64"\n", inode->i_ino,
1726                                LTIME_S(inode->i_mtime), body->mtime);
1727                         LTIME_S(inode->i_mtime) = body->mtime;
1728                 }
1729                 lli->lli_lvb.lvb_mtime = body->mtime;
1730         }
1731         if (body->valid & OBD_MD_FLCTIME) {
1732                 if (body->ctime > LTIME_S(inode->i_ctime))
1733                         LTIME_S(inode->i_ctime) = body->ctime;
1734                 lli->lli_lvb.lvb_ctime = body->ctime;
1735         }
1736         if (body->valid & OBD_MD_FLMODE)
1737                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1738         if (body->valid & OBD_MD_FLTYPE)
1739                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1740         LASSERT(inode->i_mode != 0);
1741         if (S_ISREG(inode->i_mode)) {
1742                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1743         } else {
1744                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1745         }
1746         if (body->valid & OBD_MD_FLUID)
1747                 inode->i_uid = body->uid;
1748         if (body->valid & OBD_MD_FLGID)
1749                 inode->i_gid = body->gid;
1750         if (body->valid & OBD_MD_FLFLAGS)
1751                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1752         if (body->valid & OBD_MD_FLNLINK)
1753                 set_nlink(inode, body->nlink);
1754         if (body->valid & OBD_MD_FLRDEV)
1755                 inode->i_rdev = old_decode_dev(body->rdev);
1756
1757         if (body->valid & OBD_MD_FLID) {
1758                 /* FID shouldn't be changed! */
1759                 if (fid_is_sane(&lli->lli_fid)) {
1760                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1761                                  "Trying to change FID "DFID
1762                                  " to the "DFID", inode %lu/%u(%p)\n",
1763                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1764                                  inode->i_ino, inode->i_generation, inode);
1765                 } else
1766                         lli->lli_fid = body->fid1;
1767         }
1768
1769         LASSERT(fid_seq(&lli->lli_fid) != 0);
1770
1771         if (body->valid & OBD_MD_FLSIZE) {
1772                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1773                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1774                         struct lustre_handle lockh;
1775                         ldlm_mode_t mode;
1776
1777                         /* As it is possible a blocking ast has been processed
1778                          * by this time, we need to check there is an UPDATE
1779                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1780                          * it. */
1781                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1782                                                &lockh);
1783                         if (mode) {
1784                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1785                                                       LLIF_EPOCH_PENDING |
1786                                                       LLIF_SOM_DIRTY)) {
1787                                         CERROR("ino %lu flags %u still has "
1788                                                "size authority! do not trust "
1789                                                "the size got from MDS\n",
1790                                                inode->i_ino, lli->lli_flags);
1791                                 } else {
1792                                         /* Use old size assignment to avoid
1793                                          * deadlock bz14138 & bz14326 */
1794                                         i_size_write(inode, body->size);
1795                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1796                                 }
1797                                 ldlm_lock_decref(&lockh, mode);
1798                         }
1799                 } else {
1800                         /* Use old size assignment to avoid
1801                          * deadlock bz14138 & bz14326 */
1802                         i_size_write(inode, body->size);
1803
1804                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1805                                inode->i_ino, (unsigned long long)body->size);
1806                 }
1807
1808                 if (body->valid & OBD_MD_FLBLOCKS)
1809                         inode->i_blocks = body->blocks;
1810         }
1811
1812         if (body->valid & OBD_MD_FLMDSCAPA) {
1813                 LASSERT(md->mds_capa);
1814                 ll_add_capa(inode, md->mds_capa);
1815         }
1816         if (body->valid & OBD_MD_FLOSSCAPA) {
1817                 LASSERT(md->oss_capa);
1818                 ll_add_capa(inode, md->oss_capa);
1819         }
1820 }
1821
1822 void ll_read_inode2(struct inode *inode, void *opaque)
1823 {
1824         struct lustre_md *md = opaque;
1825         struct ll_inode_info *lli = ll_i2info(inode);
1826         ENTRY;
1827
1828         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1829                PFID(&lli->lli_fid), inode);
1830
1831         LASSERT(!lli->lli_has_smd);
1832
1833         /* Core attributes from the MDS first.  This is a new inode, and
1834          * the VFS doesn't zero times in the core inode so we have to do
1835          * it ourselves.  They will be overwritten by either MDS or OST
1836          * attributes - we just need to make sure they aren't newer. */
1837         LTIME_S(inode->i_mtime) = 0;
1838         LTIME_S(inode->i_atime) = 0;
1839         LTIME_S(inode->i_ctime) = 0;
1840         inode->i_rdev = 0;
1841         ll_update_inode(inode, md);
1842
1843         /* OIDEBUG(inode); */
1844
1845         /* initializing backing dev info. */
1846         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1847
1848
1849         if (S_ISREG(inode->i_mode)) {
1850                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1851                 inode->i_op = &ll_file_inode_operations;
1852                 inode->i_fop = sbi->ll_fop;
1853                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1854                 EXIT;
1855         } else if (S_ISDIR(inode->i_mode)) {
1856                 inode->i_op = &ll_dir_inode_operations;
1857                 inode->i_fop = &ll_dir_operations;
1858                 EXIT;
1859         } else if (S_ISLNK(inode->i_mode)) {
1860                 inode->i_op = &ll_fast_symlink_inode_operations;
1861                 EXIT;
1862         } else {
1863                 inode->i_op = &ll_special_inode_operations;
1864
1865                 init_special_inode(inode, inode->i_mode,
1866                                    inode->i_rdev);
1867
1868                 EXIT;
1869         }
1870 }
1871
1872 void ll_delete_inode(struct inode *inode)
1873 {
1874         struct cl_inode_info *lli = cl_i2info(inode);
1875         struct ll_sb_info    *sbi = ll_i2sbi(inode);
1876         int rc;
1877         ENTRY;
1878
1879         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1880         if (rc)
1881                 CERROR("fid_delete() failed, rc %d\n", rc);
1882
1883         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1884                 /* discard all dirty pages before truncating them, required by
1885                  * osc_extent implementation at LU-1030. */
1886                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1887
1888         truncate_inode_pages(&inode->i_data, 0);
1889
1890         /* Workaround for LU-118 */
1891         if (inode->i_data.nrpages) {
1892                 TREE_READ_LOCK_IRQ(&inode->i_data);
1893                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1894                 LASSERTF(inode->i_data.nrpages == 0,
1895                          "inode=%lu/%u(%p) nrpages=%lu, see "
1896                          "http://jira.whamcloud.com/browse/LU-118\n",
1897                          inode->i_ino, inode->i_generation, inode,
1898                          inode->i_data.nrpages);
1899         }
1900         /* Workaround end */
1901
1902 #ifdef HAVE_SBOPS_EVICT_INODE
1903         ll_clear_inode(inode);
1904         end_writeback(inode);
1905 #else
1906         clear_inode(inode);
1907 #endif
1908
1909         EXIT;
1910 }
1911
1912 int ll_iocontrol(struct inode *inode, struct file *file,
1913                  unsigned int cmd, unsigned long arg)
1914 {
1915         struct ll_sb_info *sbi = ll_i2sbi(inode);
1916         struct ptlrpc_request *req = NULL;
1917         int rc, flags = 0;
1918         ENTRY;
1919
1920         switch(cmd) {
1921         case FSFILT_IOC_GETFLAGS: {
1922                 struct mdt_body *body;
1923                 struct md_op_data *op_data;
1924
1925                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1926                                              0, 0, LUSTRE_OPC_ANY,
1927                                              NULL);
1928                 if (IS_ERR(op_data))
1929                         RETURN(PTR_ERR(op_data));
1930
1931                 op_data->op_valid = OBD_MD_FLFLAGS;
1932                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1933                 ll_finish_md_op_data(op_data);
1934                 if (rc) {
1935                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1936                         RETURN(-abs(rc));
1937                 }
1938
1939                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1940
1941                 flags = body->flags;
1942
1943                 ptlrpc_req_finished(req);
1944
1945                 RETURN(put_user(flags, (int *)arg));
1946         }
1947         case FSFILT_IOC_SETFLAGS: {
1948                 struct lov_stripe_md *lsm;
1949                 struct obd_info oinfo = { { { 0 } } };
1950                 struct md_op_data *op_data;
1951
1952                 if (get_user(flags, (int *)arg))
1953                         RETURN(-EFAULT);
1954
1955                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1956                                              LUSTRE_OPC_ANY, NULL);
1957                 if (IS_ERR(op_data))
1958                         RETURN(PTR_ERR(op_data));
1959
1960                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1961                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1962                 rc = md_setattr(sbi->ll_md_exp, op_data,
1963                                 NULL, 0, NULL, 0, &req, NULL);
1964                 ll_finish_md_op_data(op_data);
1965                 ptlrpc_req_finished(req);
1966                 if (rc)
1967                         RETURN(rc);
1968
1969                 inode->i_flags = ll_ext_to_inode_flags(flags);
1970
1971                 lsm = ccc_inode_lsm_get(inode);
1972                 if (lsm == NULL)
1973                         RETURN(0);
1974
1975                 OBDO_ALLOC(oinfo.oi_oa);
1976                 if (!oinfo.oi_oa) {
1977                         ccc_inode_lsm_put(inode, lsm);
1978                         RETURN(-ENOMEM);
1979                 }
1980                 oinfo.oi_md = lsm;
1981                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1982                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1983                 oinfo.oi_oa->o_flags = flags;
1984                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1985                                        OBD_MD_FLGROUP;
1986                 oinfo.oi_capa = ll_mdscapa_get(inode);
1987                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1988                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1989                 capa_put(oinfo.oi_capa);
1990                 OBDO_FREE(oinfo.oi_oa);
1991                 ccc_inode_lsm_put(inode, lsm);
1992
1993                 if (rc && rc != -EPERM && rc != -EACCES)
1994                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1995
1996                 RETURN(rc);
1997         }
1998         default:
1999                 RETURN(-ENOSYS);
2000         }
2001
2002         RETURN(0);
2003 }
2004
2005 int ll_flush_ctx(struct inode *inode)
2006 {
2007         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2008
2009         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2010
2011         obd_set_info_async(NULL, sbi->ll_md_exp,
2012                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2013                            0, NULL, NULL);
2014         obd_set_info_async(NULL, sbi->ll_dt_exp,
2015                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2016                            0, NULL, NULL);
2017         return 0;
2018 }
2019
2020 /* umount -f client means force down, don't save state */
2021 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2022 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2023 {
2024         struct super_block *sb = vfsmnt->mnt_sb;
2025 #else
2026 void ll_umount_begin(struct super_block *sb)
2027 {
2028 #endif
2029         struct ll_sb_info *sbi = ll_s2sbi(sb);
2030         struct obd_device *obd;
2031         struct obd_ioctl_data *ioc_data;
2032         ENTRY;
2033
2034 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2035         if (!(flags & MNT_FORCE)) {
2036                 EXIT;
2037                 return;
2038         }
2039 #endif
2040
2041         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2042                sb->s_count, atomic_read(&sb->s_active));
2043
2044         obd = class_exp2obd(sbi->ll_md_exp);
2045         if (obd == NULL) {
2046                 CERROR("Invalid MDC connection handle "LPX64"\n",
2047                        sbi->ll_md_exp->exp_handle.h_cookie);
2048                 EXIT;
2049                 return;
2050         }
2051         obd->obd_force = 1;
2052
2053         obd = class_exp2obd(sbi->ll_dt_exp);
2054         if (obd == NULL) {
2055                 CERROR("Invalid LOV connection handle "LPX64"\n",
2056                        sbi->ll_dt_exp->exp_handle.h_cookie);
2057                 EXIT;
2058                 return;
2059         }
2060         obd->obd_force = 1;
2061
2062         OBD_ALLOC_PTR(ioc_data);
2063         if (ioc_data) {
2064                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2065                               sizeof ioc_data, ioc_data, NULL);
2066
2067                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2068                               sizeof ioc_data, ioc_data, NULL);
2069
2070                 OBD_FREE_PTR(ioc_data);
2071         }
2072
2073
2074         /* Really, we'd like to wait until there are no requests outstanding,
2075          * and then continue.  For now, we just invalidate the requests,
2076          * schedule() and sleep one second if needed, and hope.
2077          */
2078         cfs_schedule();
2079 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2080         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2081                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2082                                                    cfs_time_seconds(1));
2083                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2084                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2085                                       "may try to umount it a bit later\n",
2086                                       atomic_read(&vfsmnt->mnt_count));
2087         }
2088 #endif
2089
2090         EXIT;
2091 }
2092
2093 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2094 {
2095         struct ll_sb_info *sbi = ll_s2sbi(sb);
2096         char *profilenm = get_profile_name(sb);
2097         int err;
2098         __u32 read_only;
2099
2100         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2101                 read_only = *flags & MS_RDONLY;
2102                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2103                                          sizeof(KEY_READ_ONLY),
2104                                          KEY_READ_ONLY, sizeof(read_only),
2105                                          &read_only, NULL);
2106                 if (err) {
2107                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2108                                       profilenm, read_only ?
2109                                       "read-only" : "read-write", err);
2110                         return err;
2111                 }
2112
2113                 if (read_only)
2114                         sb->s_flags |= MS_RDONLY;
2115                 else
2116                         sb->s_flags &= ~MS_RDONLY;
2117
2118                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2119                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2120                                       read_only ?  "read-only" : "read-write");
2121         }
2122         return 0;
2123 }
2124
2125 int ll_prep_inode(struct inode **inode,
2126                   struct ptlrpc_request *req,
2127                   struct super_block *sb)
2128 {
2129         struct ll_sb_info *sbi = NULL;
2130         struct lustre_md md;
2131         __u64 ibits;
2132         int rc;
2133         ENTRY;
2134
2135         LASSERT(*inode || sb);
2136         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2137         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2138                               sbi->ll_md_exp, &md);
2139         if (rc)
2140                 RETURN(rc);
2141
2142         if (*inode) {
2143                 ll_update_inode(*inode, &md);
2144         } else {
2145                 LASSERT(sb != NULL);
2146
2147                 /*
2148                  * At this point server returns to client's same fid as client
2149                  * generated for creating. So using ->fid1 is okay here.
2150                  */
2151                 LASSERT(fid_is_sane(&md.body->fid1));
2152
2153                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2154                 if (*inode == NULL || IS_ERR(*inode)) {
2155                         if (md.lsm)
2156                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2157 #ifdef CONFIG_FS_POSIX_ACL
2158                         if (md.posix_acl) {
2159                                 posix_acl_release(md.posix_acl);
2160                                 md.posix_acl = NULL;
2161                         }
2162 #endif
2163                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2164                         *inode = NULL;
2165                         CERROR("new_inode -fatal: rc %d\n", rc);
2166                         GOTO(out, rc);
2167                 }
2168         }
2169
2170         /* sanity check for LAYOUT lock. */
2171         ibits = MDS_INODELOCK_LAYOUT;
2172         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2173             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2174                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2175                         ll_get_fsname(sb, NULL, 0),
2176                         PFID(ll_inode2fid(*inode)), *inode);
2177         }
2178
2179 out:
2180         md_free_lustre_md(sbi->ll_md_exp, &md);
2181         RETURN(rc);
2182 }
2183
2184 int ll_obd_statfs(struct inode *inode, void *arg)
2185 {
2186         struct ll_sb_info *sbi = NULL;
2187         struct obd_export *exp;
2188         char *buf = NULL;
2189         struct obd_ioctl_data *data = NULL;
2190         __u32 type;
2191         __u32 flags;
2192         int len = 0, rc;
2193
2194         if (!inode || !(sbi = ll_i2sbi(inode)))
2195                 GOTO(out_statfs, rc = -EINVAL);
2196
2197         rc = obd_ioctl_getdata(&buf, &len, arg);
2198         if (rc)
2199                 GOTO(out_statfs, rc);
2200
2201         data = (void*)buf;
2202         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2203             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2204                 GOTO(out_statfs, rc = -EINVAL);
2205
2206         if (data->ioc_inllen1 != sizeof(__u32) ||
2207             data->ioc_inllen2 != sizeof(__u32) ||
2208             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2209             data->ioc_plen2 != sizeof(struct obd_uuid))
2210                 GOTO(out_statfs, rc = -EINVAL);
2211
2212         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2213         if (type & LL_STATFS_LMV)
2214                 exp = sbi->ll_md_exp;
2215         else if (type & LL_STATFS_LOV)
2216                 exp = sbi->ll_dt_exp;
2217         else
2218                 GOTO(out_statfs, rc = -ENODEV);
2219
2220         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2221         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2222         if (rc)
2223                 GOTO(out_statfs, rc);
2224 out_statfs:
2225         if (buf)
2226                 obd_ioctl_freedata(buf, len);
2227         return rc;
2228 }
2229
2230 int ll_process_config(struct lustre_cfg *lcfg)
2231 {
2232         char *ptr;
2233         void *sb;
2234         struct lprocfs_static_vars lvars;
2235         unsigned long x;
2236         int rc = 0;
2237
2238         lprocfs_llite_init_vars(&lvars);
2239
2240         /* The instance name contains the sb: lustre-client-aacfe000 */
2241         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2242         if (!ptr || !*(++ptr))
2243                 return -EINVAL;
2244         if (sscanf(ptr, "%lx", &x) != 1)
2245                 return -EINVAL;
2246         sb = (void *)x;
2247         /* This better be a real Lustre superblock! */
2248         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2249
2250         /* Note we have not called client_common_fill_super yet, so
2251            proc fns must be able to handle that! */
2252         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2253                                       lcfg, sb);
2254         if (rc > 0)
2255                 rc = 0;
2256         return(rc);
2257 }
2258
2259 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2260 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2261                                        struct inode *i1, struct inode *i2,
2262                                        const char *name, int namelen,
2263                                        int mode, __u32 opc, void *data)
2264 {
2265         LASSERT(i1 != NULL);
2266
2267         if (namelen > ll_i2sbi(i1)->ll_namelen)
2268                 return ERR_PTR(-ENAMETOOLONG);
2269
2270         if (op_data == NULL)
2271                 OBD_ALLOC_PTR(op_data);
2272
2273         if (op_data == NULL)
2274                 return ERR_PTR(-ENOMEM);
2275
2276         ll_i2gids(op_data->op_suppgids, i1, i2);
2277         op_data->op_fid1 = *ll_inode2fid(i1);
2278         op_data->op_capa1 = ll_mdscapa_get(i1);
2279
2280         if (i2) {
2281                 op_data->op_fid2 = *ll_inode2fid(i2);
2282                 op_data->op_capa2 = ll_mdscapa_get(i2);
2283         } else {
2284                 fid_zero(&op_data->op_fid2);
2285                 op_data->op_capa2 = NULL;
2286         }
2287
2288         op_data->op_name = name;
2289         op_data->op_namelen = namelen;
2290         op_data->op_mode = mode;
2291         op_data->op_mod_time = cfs_time_current_sec();
2292         op_data->op_fsuid = cfs_curproc_fsuid();
2293         op_data->op_fsgid = cfs_curproc_fsgid();
2294         op_data->op_cap = cfs_curproc_cap_pack();
2295         op_data->op_bias = MDS_CHECK_SPLIT;
2296         op_data->op_opc = opc;
2297         op_data->op_mds = 0;
2298         op_data->op_data = data;
2299
2300         /* If the file is being opened after mknod() (normally due to NFS)
2301          * try to use the default stripe data from parent directory for
2302          * allocating OST objects.  Try to pass the parent FID to MDS. */
2303         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2304             !ll_i2info(i2)->lli_has_smd) {
2305                 struct ll_inode_info *lli = ll_i2info(i2);
2306
2307                 cfs_spin_lock(&lli->lli_lock);
2308                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2309                         op_data->op_fid1 = lli->lli_pfid;
2310                 cfs_spin_unlock(&lli->lli_lock);
2311                 /** We ignore parent's capability temporary. */
2312         }
2313
2314         return op_data;
2315 }
2316
2317 void ll_finish_md_op_data(struct md_op_data *op_data)
2318 {
2319         capa_put(op_data->op_capa1);
2320         capa_put(op_data->op_capa2);
2321         OBD_FREE_PTR(op_data);
2322 }
2323
2324 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2325 {
2326         struct ll_sb_info *sbi;
2327
2328         LASSERT((seq != NULL) && (vfs != NULL));
2329         sbi = ll_s2sbi(vfs->mnt_sb);
2330
2331         if (sbi->ll_flags & LL_SBI_NOLCK)
2332                 seq_puts(seq, ",nolock");
2333
2334         if (sbi->ll_flags & LL_SBI_FLOCK)
2335                 seq_puts(seq, ",flock");
2336
2337         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2338                 seq_puts(seq, ",localflock");
2339
2340         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2341                 seq_puts(seq, ",user_xattr");
2342
2343         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2344                 seq_puts(seq, ",lazystatfs");
2345
2346         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2347                 seq_puts(seq, ",user_fid2path");
2348
2349         RETURN(0);
2350 }
2351
2352 /**
2353  * Get obd name by cmd, and copy out to user space
2354  */
2355 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2356 {
2357         struct ll_sb_info *sbi = ll_i2sbi(inode);
2358         struct obd_device *obd;
2359         ENTRY;
2360
2361         if (cmd == OBD_IOC_GETDTNAME)
2362                 obd = class_exp2obd(sbi->ll_dt_exp);
2363         else if (cmd == OBD_IOC_GETMDNAME)
2364                 obd = class_exp2obd(sbi->ll_md_exp);
2365         else
2366                 RETURN(-EINVAL);
2367
2368         if (!obd)
2369                 RETURN(-ENOENT);
2370
2371         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2372                              strlen(obd->obd_name) + 1))
2373                 RETURN(-EFAULT);
2374
2375         RETURN(0);
2376 }
2377
2378 /**
2379  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2380  * fsname will be returned in this buffer; otherwise, a static buffer will be
2381  * used to store the fsname and returned to caller.
2382  */
2383 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2384 {
2385         static char fsname_static[MTI_NAME_MAXLEN];
2386         struct lustre_sb_info *lsi = s2lsi(sb);
2387         char *ptr;
2388         int len;
2389
2390         if (buf == NULL) {
2391                 /* this means the caller wants to use static buffer
2392                  * and it doesn't care about race. Usually this is
2393                  * in error reporting path */
2394                 buf = fsname_static;
2395                 buflen = sizeof(fsname_static);
2396         }
2397
2398         len = strlen(lsi->lsi_lmd->lmd_profile);
2399         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2400         if (ptr && (strcmp(ptr, "-client") == 0))
2401                 len -= 7;
2402
2403         if (unlikely(len >= buflen))
2404                 len = buflen - 1;
2405         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2406         buf[len] = '\0';
2407
2408         return buf;
2409 }
2410
2411 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2412 {
2413         char *path = NULL;
2414
2415 #ifdef HAVE_FS_STRUCT_USE_PATH
2416         struct path p;
2417
2418         p.dentry = dentry;
2419         p.mnt = current->fs->root.mnt;
2420         path_get(&p);
2421         path = d_path(&p, buf, bufsize);
2422         path_put(&p);
2423 #else
2424         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2425 #endif
2426
2427         return path;
2428 }
2429
2430 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2431 {
2432         char *buf, *path = NULL;
2433         struct dentry *dentry = NULL;
2434         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2435
2436         buf = (char *)__get_free_page(GFP_KERNEL);
2437         if (buf != NULL) {
2438                 dentry = d_find_alias(page->mapping->host);
2439                 if (dentry != NULL)
2440                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2441         }
2442
2443         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2444               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2445               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2446               PFID(&obj->cob_header.coh_lu.loh_fid),
2447               (path && !IS_ERR(path)) ? path : "", ioret);
2448
2449         if (dentry != NULL)
2450                 dput(dentry);
2451
2452         if (buf != NULL)
2453                 free_page((unsigned long)buf);
2454 }