lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/llite_lib.c
  40  *
  41  * Lustre Light Super operations
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/module.h>
  47 #include <linux/types.h>
  48 #include <linux/version.h>
  49 #include <linux/mm.h>
  50
  51 #include <lustre_lite.h>
  52 #include <lustre_ha.h>
  53 #include <lustre_dlm.h>
  54 #include <lprocfs_status.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_param.h>
  57 #include <lustre_log.h>
  58 #include <cl_object.h>
  59 #include <obd_cksum.h>
  60 #include "llite_internal.h"
  61
  62 cfs_mem_cache_t *ll_file_data_slab;
  63
  64 CFS_LIST_HEAD(ll_super_blocks);
  65 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  66
  67 #ifndef MS_HAS_NEW_AOPS
  68 extern struct address_space_operations ll_aops;
  69 extern struct address_space_operations ll_dir_aops;
  70 #else
  71 extern struct address_space_operations_ext ll_aops;
  72 extern struct address_space_operations_ext ll_dir_aops;
  73 #endif
  74
  75 #ifndef log2
  76 #define log2(n) cfs_ffz(~(n))
  77 #endif
  78
  79 static struct ll_sb_info *ll_init_sbi(void)
  80 {
  81         struct ll_sb_info *sbi = NULL;
  82         unsigned long pages;
  83         struct sysinfo si;
  84         class_uuid_t uuid;
  85         int i;
  86         ENTRY;
  87
  88         OBD_ALLOC(sbi, sizeof(*sbi));
  89         if (!sbi)
  90                 RETURN(NULL);
  91
  92         cfs_spin_lock_init(&sbi->ll_lock);
  93         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  94         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  95         cfs_spin_lock_init(&sbi->ll_process_lock);
  96         sbi->ll_rw_stats_on = 0;
  97
  98         si_meminfo(&si);
  99         pages = si.totalram - si.totalhigh;
 100         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 101 #ifdef HAVE_BGL_SUPPORT
 102                 sbi->ll_async_page_max = pages / 4;
 103 #else
 104                 sbi->ll_async_page_max = pages / 2;
 105 #endif
 106         } else {
 107                 sbi->ll_async_page_max = (pages / 4) * 3;
 108         }
 109
 110         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 111                                            SBI_DEFAULT_READAHEAD_MAX);
 112         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 113         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 114                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 115         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 116         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 117
 118         ll_generate_random_uuid(uuid);
 119         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 120         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 121
 122         cfs_spin_lock(&ll_sb_lock);
 123         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 124         cfs_spin_unlock(&ll_sb_lock);
 125
 126 #ifdef ENABLE_CHECKSUM
 127         sbi->ll_flags |= LL_SBI_CHECKSUM;
 128 #endif
 129
 130 #ifdef HAVE_LRU_RESIZE_SUPPORT
 131         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 132 #endif
 133
 134         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 135                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 136                                    pp_r_hist.oh_lock);
 137                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 138                                    pp_w_hist.oh_lock);
 139         }
 140
 141         /* metadata statahead is enabled by default */
 142         sbi->ll_sa_max = LL_SA_RPC_DEF;
 143         atomic_set(&sbi->ll_sa_total, 0);
 144         atomic_set(&sbi->ll_sa_wrong, 0);
 145
 146         RETURN(sbi);
 147 }
 148
 149 void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 cfs_spin_lock(&ll_sb_lock);
 156                 cfs_list_del(&sbi->ll_list);
 157                 cfs_spin_unlock(&ll_sb_lock);
 158                 OBD_FREE(sbi, sizeof(*sbi));
 159         }
 160         EXIT;
 161 }
 162
 163 static struct dentry_operations ll_d_root_ops = {
 164         .d_compare = ll_dcompare,
 165         .d_revalidate = ll_revalidate_nd,
 166 };
 167
 168 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 169                                     struct vfsmount *mnt)
 170 {
 171         struct inode *root = 0;
 172         struct ll_sb_info *sbi = ll_s2sbi(sb);
 173         struct obd_device *obd;
 174         struct obd_capa *oc = NULL;
 175         struct obd_statfs osfs;
 176         struct ptlrpc_request *request = NULL;
 177         struct obd_connect_data *data = NULL;
 178         struct obd_uuid *uuid;
 179         struct md_op_data *op_data;
 180         struct lustre_md lmd;
 181         obd_valid valid;
 182         int size, err, checksum;
 183         ENTRY;
 184
 185         obd = class_name2obd(md);
 186         if (!obd) {
 187                 CERROR("MD %s: not setup or attached\n", md);
 188                 RETURN(-EINVAL);
 189         }
 190
 191         OBD_ALLOC_PTR(data);
 192         if (data == NULL)
 193                 RETURN(-ENOMEM);
 194
 195         if (proc_lustre_fs_root) {
 196                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 197                                                   dt, md);
 198                 if (err < 0)
 199                         CERROR("could not register mount in /proc/fs/lustre\n");
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET| OBD_CONNECT_FID      |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH;
 211
 212         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 213                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 214
 215 #ifdef HAVE_LRU_RESIZE_SUPPORT
 216         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 217                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 218 #endif
 219 #ifdef CONFIG_FS_POSIX_ACL
 220         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 221 #endif
 222         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 223         data->ocd_version = LUSTRE_VERSION_CODE;
 224
 225         if (sb->s_flags & MS_RDONLY)
 226                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 227         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 228                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 229
 230 #ifdef HAVE_MS_FLOCK_LOCK
 231         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 232         sb->s_flags |= MS_FLOCK_LOCK;
 233 #endif
 234 #ifdef MS_HAS_NEW_AOPS
 235         sb->s_flags |= MS_HAS_NEW_AOPS;
 236 #endif
 237
 238         if (sbi->ll_flags & LL_SBI_FLOCK)
 239                 sbi->ll_fop = &ll_file_operations_flock;
 240         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 241                 sbi->ll_fop = &ll_file_operations;
 242         else
 243                 sbi->ll_fop = &ll_file_operations_noflock;
 244
 245         /* real client */
 246         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 247         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 248                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 249
 250         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 251
 252         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 253         if (err == -EBUSY) {
 254                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 255                                    "recovery, of which this client is not a "
 256                                    "part. Please wait for recovery to complete,"
 257                                    " abort, or time out.\n", md);
 258                 GOTO(out, err);
 259         } else if (err) {
 260                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 261                 GOTO(out, err);
 262         }
 263
 264         err = obd_fid_init(sbi->ll_md_exp);
 265         if (err) {
 266                 CERROR("Can't init metadata layer FID infrastructure, "
 267                        "rc %d\n", err);
 268                 GOTO(out_md, err);
 269         }
 270
 271         err = obd_statfs(obd, &osfs,
 272                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 273         if (err)
 274                 GOTO(out_md_fid, err);
 275
 276         size = sizeof(*data);
 277         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 278                            KEY_CONN_DATA,  &size, data, NULL);
 279         if (err) {
 280                 CERROR("Get connect data failed: %d \n", err);
 281                 GOTO(out_md, err);
 282         }
 283
 284         LASSERT(osfs.os_bsize);
 285         sb->s_blocksize = osfs.os_bsize;
 286         sb->s_blocksize_bits = log2(osfs.os_bsize);
 287         sb->s_magic = LL_SUPER_MAGIC;
 288
 289         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 290          *         retval = in_file->f_op->sendfile(...);
 291          *         if (*ppos > max)
 292          *                 retval = -EOVERFLOW;
 293          *
 294          * it will check if *ppos is greater than max. However, max equals to
 295          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 296          * has been defined as a signed long long integer in linux kernel. */
 297 #if BITS_PER_LONG == 64
 298         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 299 #else
 300         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 301 #endif
 302         sbi->ll_namelen = osfs.os_namelen;
 303         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 304
 305         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 306             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 307                 LCONSOLE_INFO("Disabling user_xattr feature because "
 308                               "it is not supported on the server\n");
 309                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 310         }
 311
 312         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 313 #ifdef MS_POSIXACL
 314                 sb->s_flags |= MS_POSIXACL;
 315 #endif
 316                 sbi->ll_flags |= LL_SBI_ACL;
 317         } else {
 318                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 319 #ifdef MS_POSIXACL
 320                 sb->s_flags &= ~MS_POSIXACL;
 321 #endif
 322                 sbi->ll_flags &= ~LL_SBI_ACL;
 323         }
 324
 325         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 326                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 327                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 328                         LCONSOLE_INFO("client is set as remote by default.\n");
 329                 }
 330         } else {
 331                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 332                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 333                         LCONSOLE_INFO("client claims to be remote, but server "
 334                                       "rejected, forced to be local.\n");
 335                 }
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 339                 LCONSOLE_INFO("client enabled MDS capability!\n");
 340                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 344                 LCONSOLE_INFO("client enabled OSS capability!\n");
 345                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 346         }
 347
 348         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 349                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 350
 351         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 352                 sbi->ll_md_brw_size = data->ocd_brw_size;
 353         else
 354                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 355
 356         obd = class_name2obd(dt);
 357         if (!obd) {
 358                 CERROR("DT %s: not setup or attached\n", dt);
 359                 GOTO(out_md_fid, err = -ENODEV);
 360         }
 361
 362         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 363                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 364                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 365                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 366                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 367                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 368                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH;
 369
 370         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 371                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 372
 373         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 374                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 375                  * disabled by default, because it can still be enabled on the
 376                  * fly via /proc. As a consequence, we still need to come to an
 377                  * agreement on the supported algorithms at connect time */
 378                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 379
 380                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 381                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 382                 else
 383                         /* send the list of supported checksum types */
 384                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 385         }
 386
 387 #ifdef HAVE_LRU_RESIZE_SUPPORT
 388         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 389 #endif
 390         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 391                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 392
 393         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 394                "ocd_grant: %d\n", data->ocd_connect_flags,
 395                data->ocd_version, data->ocd_grant);
 396
 397         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 398         obd->obd_upcall.onu_upcall = cl_ocd_update;
 399
 400         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 401
 402         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 403         if (err == -EBUSY) {
 404                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 405                                    "recovery, of which this client is not a "
 406                                    "part.  Please wait for recovery to "
 407                                    "complete, abort, or time out.\n", dt);
 408                 GOTO(out_md_fid, err);
 409         } else if (err) {
 410                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 411                 GOTO(out_md_fid, err);
 412         }
 413
 414         err = obd_fid_init(sbi->ll_dt_exp);
 415         if (err) {
 416                 CERROR("Can't init data layer FID infrastructure, "
 417                        "rc %d\n", err);
 418                 GOTO(out_dt, err);
 419         }
 420
 421         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 422         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 423         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 424         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 425         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 426
 427         fid_zero(&sbi->ll_root_fid);
 428         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 429         if (err) {
 430                 CERROR("cannot mds_connect: rc = %d\n", err);
 431                 GOTO(out_lock_cn_cb, err);
 432         }
 433         if (!fid_is_sane(&sbi->ll_root_fid)) {
 434                 CERROR("Invalid root fid during mount\n");
 435                 GOTO(out_lock_cn_cb, err = -EINVAL);
 436         }
 437         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 438
 439         sb->s_op = &lustre_super_operations;
 440 #if THREAD_SIZE >= 8192
 441         sb->s_export_op = &lustre_export_operations;
 442 #endif
 443
 444         /* make root inode
 445          * XXX: move this to after cbd setup? */
 446         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 447         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 448                 valid |= OBD_MD_FLRMTPERM;
 449         else if (sbi->ll_flags & LL_SBI_ACL)
 450                 valid |= OBD_MD_FLACL;
 451
 452         OBD_ALLOC_PTR(op_data);
 453         if (op_data == NULL)
 454                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 455
 456         op_data->op_fid1 = sbi->ll_root_fid;
 457         op_data->op_mode = 0;
 458         op_data->op_capa1 = oc;
 459         op_data->op_valid = valid;
 460
 461         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 462         if (oc)
 463                 capa_put(oc);
 464         OBD_FREE_PTR(op_data);
 465         if (err) {
 466                 CERROR("md_getattr failed for root: rc = %d\n", err);
 467                 GOTO(out_lock_cn_cb, err);
 468         }
 469         memset(&lmd, 0, sizeof(lmd));
 470         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 471                                sbi->ll_md_exp, &lmd);
 472         if (err) {
 473                 CERROR("failed to understand root inode md: rc = %d\n", err);
 474                 ptlrpc_req_finished (request);
 475                 GOTO(out_lock_cn_cb, err);
 476         }
 477
 478         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 479         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 480         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 481         ptlrpc_req_finished(request);
 482
 483         if (root == NULL || IS_ERR(root)) {
 484                 if (lmd.lsm)
 485                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 486 #ifdef CONFIG_FS_POSIX_ACL
 487                 if (lmd.posix_acl) {
 488                         posix_acl_release(lmd.posix_acl);
 489                         lmd.posix_acl = NULL;
 490                 }
 491 #endif
 492                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 493                 root = NULL;
 494                 CERROR("lustre_lite: bad iget4 for root\n");
 495                 GOTO(out_root, err);
 496         }
 497
 498         err = ll_close_thread_start(&sbi->ll_lcq);
 499         if (err) {
 500                 CERROR("cannot start close thread: rc %d\n", err);
 501                 GOTO(out_root, err);
 502         }
 503
 504 #ifdef CONFIG_FS_POSIX_ACL
 505         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 506                 rct_init(&sbi->ll_rct);
 507                 et_init(&sbi->ll_et);
 508         }
 509 #endif
 510
 511         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 512         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 513                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 514                                  NULL);
 515         cl_sb_init(sb);
 516
 517         sb->s_root = d_alloc_root(root);
 518         if (data != NULL)
 519                 OBD_FREE(data, sizeof(*data));
 520
 521         sb->s_root->d_op = &ll_d_root_ops;
 522
 523         sbi->ll_sdev_orig = sb->s_dev;
 524
 525         /* We set sb->s_dev equal on all lustre clients in order to support
 526          * NFS export clustering.  NFSD requires that the FSID be the same
 527          * on all clients. */
 528         /* s_dev is also used in lt_compare() to compare two fs, but that is
 529          * only a node-local comparison. */
 530         uuid = obd_get_uuid(sbi->ll_md_exp);
 531         if (uuid != NULL)
 532                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 533         sbi->ll_mnt = mnt;
 534
 535         RETURN(err);
 536 out_root:
 537         if (root)
 538                 iput(root);
 539 out_lock_cn_cb:
 540         obd_fid_fini(sbi->ll_dt_exp);
 541 out_dt:
 542         obd_disconnect(sbi->ll_dt_exp);
 543         sbi->ll_dt_exp = NULL;
 544 out_md_fid:
 545         obd_fid_fini(sbi->ll_md_exp);
 546 out_md:
 547         obd_disconnect(sbi->ll_md_exp);
 548         sbi->ll_md_exp = NULL;
 549 out:
 550         if (data != NULL)
 551                 OBD_FREE_PTR(data);
 552         lprocfs_unregister_mountpoint(sbi);
 553         return err;
 554 }
 555
 556 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 557 {
 558         int size, rc;
 559
 560         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 561         size = sizeof(int);
 562         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 563                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 564         if (rc)
 565                 CERROR("Get max mdsize error rc %d \n", rc);
 566
 567         RETURN(rc);
 568 }
 569
 570 void ll_dump_inode(struct inode *inode)
 571 {
 572         struct list_head *tmp;
 573         int dentry_count = 0;
 574
 575         LASSERT(inode != NULL);
 576
 577         list_for_each(tmp, &inode->i_dentry)
 578                 dentry_count++;
 579
 580         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 581                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 582                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 583 }
 584
 585 void lustre_dump_dentry(struct dentry *dentry, int recur)
 586 {
 587         struct list_head *tmp;
 588         int subdirs = 0;
 589
 590         LASSERT(dentry != NULL);
 591
 592         list_for_each(tmp, &dentry->d_subdirs)
 593                 subdirs++;
 594
 595         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 596                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 597                dentry->d_name.len, dentry->d_name.name,
 598                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 599                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 600                dentry->d_flags, dentry->d_fsdata, subdirs);
 601         if (dentry->d_inode != NULL)
 602                 ll_dump_inode(dentry->d_inode);
 603
 604         if (recur == 0)
 605                 return;
 606
 607        list_for_each(tmp, &dentry->d_subdirs) {
 608                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 609                 lustre_dump_dentry(d, recur - 1);
 610         }
 611 }
 612
 613 void client_common_put_super(struct super_block *sb)
 614 {
 615         struct ll_sb_info *sbi = ll_s2sbi(sb);
 616         ENTRY;
 617
 618 #ifdef CONFIG_FS_POSIX_ACL
 619         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 620                 et_fini(&sbi->ll_et);
 621                 rct_fini(&sbi->ll_rct);
 622         }
 623 #endif
 624
 625         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 626
 627         ll_close_thread_shutdown(sbi->ll_lcq);
 628
 629         cl_sb_fini(sb);
 630
 631         cfs_list_del(&sbi->ll_conn_chain);
 632
 633         obd_fid_fini(sbi->ll_dt_exp);
 634         obd_disconnect(sbi->ll_dt_exp);
 635         sbi->ll_dt_exp = NULL;
 636
 637         lprocfs_unregister_mountpoint(sbi);
 638
 639         obd_fid_fini(sbi->ll_md_exp);
 640         obd_disconnect(sbi->ll_md_exp);
 641         sbi->ll_md_exp = NULL;
 642
 643         EXIT;
 644 }
 645
 646 void ll_kill_super(struct super_block *sb)
 647 {
 648         struct ll_sb_info *sbi;
 649
 650         ENTRY;
 651
 652         /* not init sb ?*/
 653         if (!(sb->s_flags & MS_ACTIVE))
 654                 return;
 655
 656         sbi = ll_s2sbi(sb);
 657         /* we need restore s_dev from changed for clustred NFS before put_super
 658          * because new kernels have cached s_dev and change sb->s_dev in
 659          * put_super not affected real removing devices */
 660         if (sbi)
 661                 sb->s_dev = sbi->ll_sdev_orig;
 662         EXIT;
 663 }
 664
 665 char *ll_read_opt(const char *opt, char *data)
 666 {
 667         char *value;
 668         char *retval;
 669         ENTRY;
 670
 671         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 672         if (strncmp(opt, data, strlen(opt)))
 673                 RETURN(NULL);
 674         if ((value = strchr(data, '=')) == NULL)
 675                 RETURN(NULL);
 676
 677         value++;
 678         OBD_ALLOC(retval, strlen(value) + 1);
 679         if (!retval) {
 680                 CERROR("out of memory!\n");
 681                 RETURN(NULL);
 682         }
 683
 684         memcpy(retval, value, strlen(value)+1);
 685         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 686         RETURN(retval);
 687 }
 688
 689 static inline int ll_set_opt(const char *opt, char *data, int fl)
 690 {
 691         if (strncmp(opt, data, strlen(opt)) != 0)
 692                 return(0);
 693         else
 694                 return(fl);
 695 }
 696
 697 /* non-client-specific mount options are parsed in lmd_parse */
 698 static int ll_options(char *options, int *flags)
 699 {
 700         int tmp;
 701         char *s1 = options, *s2;
 702         ENTRY;
 703
 704         if (!options)
 705                 RETURN(0);
 706
 707         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 708
 709         while (*s1) {
 710                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 711                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 712                 if (tmp) {
 713                         *flags |= tmp;
 714                         goto next;
 715                 }
 716                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 717                 if (tmp) {
 718                         *flags |= tmp;
 719                         goto next;
 720                 }
 721                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 722                 if (tmp) {
 723                         *flags |= tmp;
 724                         goto next;
 725                 }
 726                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 727                 if (tmp) {
 728                         *flags &= ~tmp;
 729                         goto next;
 730                 }
 731                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 732                 if (tmp) {
 733                         *flags |= tmp;
 734                         goto next;
 735                 }
 736                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 737                 if (tmp) {
 738                         *flags &= ~tmp;
 739                         goto next;
 740                 }
 741 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 742                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 743                 if (tmp) {
 744                         /* Ignore deprecated mount option.  The client will
 745                          * always try to mount with ACL support, whether this
 746                          * is used depends on whether server supports it. */
 747                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 748                                                   "mount option 'acl'.\n");
 749                         goto next;
 750                 }
 751                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 752                 if (tmp) {
 753                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 754                                                   "mount option 'noacl'.\n");
 755                         goto next;
 756                 }
 757 #else
 758 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 759 #endif
 760                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 761                 if (tmp) {
 762                         *flags |= tmp;
 763                         goto next;
 764                 }
 765
 766                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 767                 if (tmp) {
 768                         *flags |= tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 772                 if (tmp) {
 773                         *flags &= ~tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 782                 if (tmp) {
 783                         *flags &= ~tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 787                 if (tmp) {
 788                         *flags |= tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 792                 if (tmp) {
 793                         *flags &= ~tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806
 807                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 808                                    s1);
 809                 RETURN(-EINVAL);
 810
 811 next:
 812                 /* Find next opt */
 813                 s2 = strchr(s1, ',');
 814                 if (s2 == NULL)
 815                         break;
 816                 s1 = s2 + 1;
 817         }
 818         RETURN(0);
 819 }
 820
 821 void ll_lli_init(struct ll_inode_info *lli)
 822 {
 823         lli->lli_inode_magic = LLI_INODE_MAGIC;
 824         cfs_sema_init(&lli->lli_size_sem, 1);
 825         cfs_sema_init(&lli->lli_write_sem, 1);
 826         cfs_init_rwsem(&lli->lli_trunc_sem);
 827         lli->lli_flags = 0;
 828         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 829         cfs_spin_lock_init(&lli->lli_lock);
 830         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 831         lli->lli_inode_magic = LLI_INODE_MAGIC;
 832         cfs_sema_init(&lli->lli_och_sem, 1);
 833         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 834         lli->lli_mds_exec_och = NULL;
 835         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 836         lli->lli_open_fd_exec_count = 0;
 837         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 838         lli->lli_remote_perms = NULL;
 839         lli->lli_rmtperm_utime = 0;
 840         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 841         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 842         cfs_spin_lock_init(&lli->lli_sa_lock);
 843         cfs_sema_init(&lli->lli_readdir_sem, 1);
 844         fid_zero(&lli->lli_pfid);
 845 }
 846
 847 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 848 {
 849 #ifdef HAVE_BDI_REGISTER
 850         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 851
 852         bdi->name = "lustre";
 853         return bdi_register(bdi, NULL, "lustre-%d",
 854                             atomic_inc_return(&ll_bdi_num));
 855 #else
 856         return 0;
 857 #endif
 858 }
 859
 860 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 861 {
 862         struct lustre_profile *lprof;
 863         struct lustre_sb_info *lsi = s2lsi(sb);
 864         struct ll_sb_info *sbi;
 865         char  *dt = NULL, *md = NULL;
 866         char  *profilenm = get_profile_name(sb);
 867         struct config_llog_instance cfg = {0, };
 868         char   ll_instance[sizeof(sb) * 2 + 1];
 869         int    err;
 870         ENTRY;
 871
 872         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 873
 874         cfs_module_get();
 875
 876         /* client additional sb info */
 877         lsi->lsi_llsbi = sbi = ll_init_sbi();
 878         if (!sbi) {
 879                 cfs_module_put(THIS_MODULE);
 880                 RETURN(-ENOMEM);
 881         }
 882
 883         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 884         if (err)
 885                 GOTO(out_free, err);
 886
 887         err = ll_bdi_init(&lsi->lsi_bdi);
 888         if (err)
 889                 GOTO(out_free, err);
 890         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 891         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 892         err = ll_bdi_register(&lsi->lsi_bdi);
 893         if (err)
 894                 GOTO(out_free, err);
 895
 896 #ifdef HAVE_SB_BDI
 897         sb->s_bdi = &lsi->lsi_bdi;
 898 #endif
 899
 900         /* Generate a string unique to this super, in case some joker tries
 901            to mount the same fs at two mount points.
 902            Use the address of the super itself.*/
 903         sprintf(ll_instance, "%p", sb);
 904         cfg.cfg_instance = ll_instance;
 905         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 906
 907         /* set up client obds */
 908         err = lustre_process_log(sb, profilenm, &cfg);
 909         if (err < 0) {
 910                 CERROR("Unable to process log: %d\n", err);
 911                 GOTO(out_free, err);
 912         }
 913
 914         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 915         lprof = class_get_profile(profilenm);
 916         if (lprof == NULL) {
 917                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 918                                    " read from the MGS.  Does that filesystem "
 919                                    "exist?\n", profilenm);
 920                 GOTO(out_free, err = -EINVAL);
 921         }
 922         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 923                lprof->lp_md, lprof->lp_dt);
 924
 925         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 926                   strlen(ll_instance) + 2);
 927         if (!dt)
 928                 GOTO(out_free, err = -ENOMEM);
 929         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 930
 931         OBD_ALLOC(md, strlen(lprof->lp_md) +
 932                   strlen(ll_instance) + 2);
 933         if (!md)
 934                 GOTO(out_free, err = -ENOMEM);
 935         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 936
 937         /* connections, registrations, sb setup */
 938         err = client_common_fill_super(sb, md, dt, mnt);
 939
 940 out_free:
 941         if (md)
 942                 OBD_FREE(md, strlen(md) + 1);
 943         if (dt)
 944                 OBD_FREE(dt, strlen(dt) + 1);
 945         if (err)
 946                 ll_put_super(sb);
 947         else
 948                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 949
 950         RETURN(err);
 951 } /* ll_fill_super */
 952
 953
 954 void lu_context_keys_dump(void);
 955
 956 void ll_put_super(struct super_block *sb)
 957 {
 958         struct config_llog_instance cfg;
 959         char   ll_instance[sizeof(sb) * 2 + 1];
 960         struct obd_device *obd;
 961         struct lustre_sb_info *lsi = s2lsi(sb);
 962         struct ll_sb_info *sbi = ll_s2sbi(sb);
 963         char *profilenm = get_profile_name(sb);
 964         int force = 1, next;
 965         ENTRY;
 966
 967         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 968
 969         ll_print_capa_stat(sbi);
 970
 971         sprintf(ll_instance, "%p", sb);
 972         cfg.cfg_instance = ll_instance;
 973         lustre_end_log(sb, NULL, &cfg);
 974
 975         if (sbi->ll_md_exp) {
 976                 obd = class_exp2obd(sbi->ll_md_exp);
 977                 if (obd)
 978                         force = obd->obd_force;
 979         }
 980
 981         /* We need to set force before the lov_disconnect in
 982            lustre_common_put_super, since l_d cleans up osc's as well. */
 983         if (force) {
 984                 next = 0;
 985                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 986                                                      &next)) != NULL) {
 987                         obd->obd_force = force;
 988                 }
 989         }
 990
 991         if (sbi->ll_lcq) {
 992                 /* Only if client_common_fill_super succeeded */
 993                 client_common_put_super(sb);
 994         }
 995
 996         next = 0;
 997         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 998                 class_manual_cleanup(obd);
 999         }
1000
1001         if (profilenm)
1002                 class_del_profile(profilenm);
1003
1004         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1005                 ll_bdi_destroy(&lsi->lsi_bdi);
1006                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1007         }
1008
1009         ll_free_sbi(sb);
1010         lsi->lsi_llsbi = NULL;
1011
1012         lustre_common_put_super(sb);
1013
1014         cl_env_cache_purge(~0);
1015
1016         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1017
1018         cfs_module_put(THIS_MODULE);
1019
1020         EXIT;
1021 } /* client_put_super */
1022
1023 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1024 {
1025         struct inode *inode = NULL;
1026         /* NOTE: we depend on atomic igrab() -bzzz */
1027         lock_res_and_lock(lock);
1028         if (lock->l_ast_data) {
1029                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1030                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1031                         inode = igrab(lock->l_ast_data);
1032                 } else {
1033                         inode = lock->l_ast_data;
1034                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1035                                          D_WARNING, lock, "l_ast_data %p is "
1036                                          "bogus: magic %08x", lock->l_ast_data,
1037                                          lli->lli_inode_magic);
1038                         inode = NULL;
1039                 }
1040         }
1041         unlock_res_and_lock(lock);
1042         return inode;
1043 }
1044
1045 static int null_if_equal(struct ldlm_lock *lock, void *data)
1046 {
1047         if (data == lock->l_ast_data) {
1048                 lock->l_ast_data = NULL;
1049
1050                 if (lock->l_req_mode != lock->l_granted_mode)
1051                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1052         }
1053
1054         return LDLM_ITER_CONTINUE;
1055 }
1056
1057 void ll_clear_inode(struct inode *inode)
1058 {
1059         struct ll_inode_info *lli = ll_i2info(inode);
1060         struct ll_sb_info *sbi = ll_i2sbi(inode);
1061         ENTRY;
1062
1063         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1064                inode->i_generation, inode);
1065
1066         if (S_ISDIR(inode->i_mode)) {
1067                 /* these should have been cleared in ll_file_release */
1068                 LASSERT(lli->lli_sai == NULL);
1069                 LASSERT(lli->lli_opendir_key == NULL);
1070                 LASSERT(lli->lli_opendir_pid == 0);
1071         }
1072
1073         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1074         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1075                          null_if_equal, inode);
1076
1077         LASSERT(!lli->lli_open_fd_write_count);
1078         LASSERT(!lli->lli_open_fd_read_count);
1079         LASSERT(!lli->lli_open_fd_exec_count);
1080
1081         if (lli->lli_mds_write_och)
1082                 ll_md_real_close(inode, FMODE_WRITE);
1083         if (lli->lli_mds_exec_och)
1084                 ll_md_real_close(inode, FMODE_EXEC);
1085         if (lli->lli_mds_read_och)
1086                 ll_md_real_close(inode, FMODE_READ);
1087
1088         if (lli->lli_symlink_name) {
1089                 OBD_FREE(lli->lli_symlink_name,
1090                          strlen(lli->lli_symlink_name) + 1);
1091                 lli->lli_symlink_name = NULL;
1092         }
1093
1094         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1095                 LASSERT(lli->lli_posix_acl == NULL);
1096                 if (lli->lli_remote_perms) {
1097                         free_rmtperm_hash(lli->lli_remote_perms);
1098                         lli->lli_remote_perms = NULL;
1099                 }
1100         }
1101 #ifdef CONFIG_FS_POSIX_ACL
1102         else if (lli->lli_posix_acl) {
1103                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1104                 LASSERT(lli->lli_remote_perms == NULL);
1105                 posix_acl_release(lli->lli_posix_acl);
1106                 lli->lli_posix_acl = NULL;
1107         }
1108 #endif
1109         lli->lli_inode_magic = LLI_INODE_DEAD;
1110
1111         ll_clear_inode_capas(inode);
1112         /*
1113          * XXX This has to be done before lsm is freed below, because
1114          * cl_object still uses inode lsm.
1115          */
1116         cl_inode_fini(inode);
1117
1118         if (lli->lli_smd) {
1119                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1120                 lli->lli_smd = NULL;
1121         }
1122
1123
1124         EXIT;
1125 }
1126
1127 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1128                   struct md_open_data **mod)
1129 {
1130         struct lustre_md md;
1131         struct ll_sb_info *sbi = ll_i2sbi(inode);
1132         struct ptlrpc_request *request = NULL;
1133         int rc;
1134         ENTRY;
1135
1136         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1137                                      LUSTRE_OPC_ANY, NULL);
1138         if (IS_ERR(op_data))
1139                 RETURN(PTR_ERR(op_data));
1140
1141         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1142                         &request, mod);
1143         if (rc) {
1144                 ptlrpc_req_finished(request);
1145                 if (rc == -ENOENT) {
1146                         inode->i_nlink = 0;
1147                         /* Unlinked special device node? Or just a race?
1148                          * Pretend we done everything. */
1149                         if (!S_ISREG(inode->i_mode) &&
1150                             !S_ISDIR(inode->i_mode))
1151                                 rc = inode_setattr(inode, &op_data->op_attr);
1152                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1153                         CERROR("md_setattr fails: rc = %d\n", rc);
1154                 }
1155                 RETURN(rc);
1156         }
1157
1158         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1159                               sbi->ll_md_exp, &md);
1160         if (rc) {
1161                 ptlrpc_req_finished(request);
1162                 RETURN(rc);
1163         }
1164
1165         /* We call inode_setattr to adjust timestamps.
1166          * If there is at least some data in file, we cleared ATTR_SIZE
1167          * above to avoid invoking vmtruncate, otherwise it is important
1168          * to call vmtruncate in inode_setattr to update inode->i_size
1169          * (bug 6196) */
1170         rc = inode_setattr(inode, &op_data->op_attr);
1171
1172         /* Extract epoch data if obtained. */
1173         op_data->op_handle = md.body->handle;
1174         op_data->op_ioepoch = md.body->ioepoch;
1175
1176         ll_update_inode(inode, &md);
1177         ptlrpc_req_finished(request);
1178
1179         RETURN(rc);
1180 }
1181
1182 /* Close IO epoch and send Size-on-MDS attribute update. */
1183 static int ll_setattr_done_writing(struct inode *inode,
1184                                    struct md_op_data *op_data,
1185                                    struct md_open_data *mod)
1186 {
1187         struct ll_inode_info *lli = ll_i2info(inode);
1188         int rc = 0;
1189         ENTRY;
1190
1191         LASSERT(op_data != NULL);
1192         if (!S_ISREG(inode->i_mode))
1193                 RETURN(0);
1194
1195         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1196                op_data->op_ioepoch, PFID(&lli->lli_fid));
1197
1198         op_data->op_flags = MF_EPOCH_CLOSE;
1199         ll_done_writing_attr(inode, op_data);
1200         ll_pack_inode2opdata(inode, op_data, NULL);
1201
1202         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1203         if (rc == -EAGAIN) {
1204                 /* MDS has instructed us to obtain Size-on-MDS attribute
1205                  * from OSTs and send setattr to back to MDS. */
1206                 rc = ll_som_update(inode, op_data);
1207         } else if (rc) {
1208                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1209                        inode->i_ino, rc);
1210         }
1211         RETURN(rc);
1212 }
1213
1214 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1215 {
1216         struct obd_capa *capa;
1217         int rc;
1218
1219         if (attr->ia_valid & ATTR_SIZE)
1220                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1221         else
1222                 capa = ll_mdscapa_get(inode);
1223
1224         rc = cl_setattr_ost(inode, attr, capa);
1225
1226         if (attr->ia_valid & ATTR_SIZE)
1227                 ll_truncate_free_capa(capa);
1228         else
1229                 capa_put(capa);
1230
1231         return rc;
1232 }
1233
1234 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1235  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1236  * keep these values until such a time that objects are allocated for it.
1237  * We do the MDS operations first, as it is checking permissions for us.
1238  * We don't to the MDS RPC if there is nothing that we want to store there,
1239  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1240  * going to do an RPC anyways.
1241  *
1242  * If we are doing a truncate, we will send the mtime and ctime updates
1243  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1244  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1245  * at the same time.
1246  */
1247 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1248 {
1249         struct ll_inode_info *lli = ll_i2info(inode);
1250         struct lov_stripe_md *lsm = lli->lli_smd;
1251         struct md_op_data *op_data = NULL;
1252         struct md_open_data *mod = NULL;
1253         int ia_valid = attr->ia_valid;
1254         int rc = 0, rc1 = 0;
1255         ENTRY;
1256
1257         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1258                attr->ia_valid);
1259         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1260
1261         if (ia_valid & ATTR_SIZE) {
1262                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1263                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1264                                attr->ia_size, ll_file_maxbytes(inode));
1265                         RETURN(-EFBIG);
1266                 }
1267
1268                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1269         }
1270
1271         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1272         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1273                 if (cfs_curproc_fsuid() != inode->i_uid &&
1274                     !cfs_capable(CFS_CAP_FOWNER))
1275                         RETURN(-EPERM);
1276         }
1277
1278         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1279         if (attr->ia_valid & ATTR_CTIME) {
1280                 attr->ia_ctime = CFS_CURRENT_TIME;
1281                 attr->ia_valid |= ATTR_CTIME_SET;
1282         }
1283         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1284                 attr->ia_atime = CFS_CURRENT_TIME;
1285                 attr->ia_valid |= ATTR_ATIME_SET;
1286         }
1287         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1288                 attr->ia_mtime = CFS_CURRENT_TIME;
1289                 attr->ia_valid |= ATTR_MTIME_SET;
1290         }
1291
1292         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1293                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1294                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1295                        cfs_time_current_sec());
1296
1297         /* NB: ATTR_SIZE will only be set after this point if the size
1298          * resides on the MDS, ie, this file has no objects. */
1299         if (lsm)
1300                 attr->ia_valid &= ~ATTR_SIZE;
1301
1302         /* We always do an MDS RPC, even if we're only changing the size;
1303          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1304
1305         OBD_ALLOC_PTR(op_data);
1306         if (op_data == NULL)
1307                 RETURN(-ENOMEM);
1308
1309         UNLOCK_INODE_MUTEX(inode);
1310         if (ia_valid & ATTR_SIZE)
1311                 UP_WRITE_I_ALLOC_SEM(inode);
1312         cfs_down_write(&lli->lli_trunc_sem);
1313         LOCK_INODE_MUTEX(inode);
1314         if (ia_valid & ATTR_SIZE)
1315                 DOWN_WRITE_I_ALLOC_SEM(inode);
1316
1317         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1318
1319         /* Open epoch for truncate. */
1320         if (exp_connect_som(ll_i2mdexp(inode)) &&
1321             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1322                 op_data->op_flags = MF_EPOCH_OPEN;
1323
1324         rc = ll_md_setattr(inode, op_data, &mod);
1325         if (rc)
1326                 GOTO(out, rc);
1327
1328         ll_ioepoch_open(lli, op_data->op_ioepoch);
1329         if (!lsm || !S_ISREG(inode->i_mode)) {
1330                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1331                 GOTO(out, rc = 0);
1332         }
1333
1334         if (ia_valid & ATTR_SIZE)
1335                 attr->ia_valid |= ATTR_SIZE;
1336         if (ia_valid & (ATTR_SIZE |
1337                         ATTR_ATIME | ATTR_ATIME_SET |
1338                         ATTR_MTIME | ATTR_MTIME_SET))
1339                 /* on truncate and utimes send attributes to osts, setting
1340                  * mtime/atime to past will be performed under PW 0:EOF extent
1341                  * lock (new_size:EOF for truncate)
1342                  * it may seem excessive to send mtime/atime updates to osts
1343                  * when not setting times to past, but it is necessary due to
1344                  * possible time de-synchronization */
1345                 rc = ll_setattr_ost(inode, attr);
1346         EXIT;
1347 out:
1348         if (op_data) {
1349                 if (op_data->op_ioepoch)
1350                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1351                 ll_finish_md_op_data(op_data);
1352         }
1353         cfs_up_write(&lli->lli_trunc_sem);
1354         return rc ? rc : rc1;
1355 }
1356
1357 int ll_setattr(struct dentry *de, struct iattr *attr)
1358 {
1359         int mode = de->d_inode->i_mode;
1360
1361         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1362                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1363                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1364
1365         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1366                                (ATTR_SIZE|ATTR_MODE)) &&
1367             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1368              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1369               !(attr->ia_mode & S_ISGID))))
1370                 attr->ia_valid |= ATTR_FORCE;
1371
1372         if ((mode & S_ISUID) &&
1373             !(attr->ia_mode & S_ISUID) &&
1374             !(attr->ia_valid & ATTR_KILL_SUID))
1375                 attr->ia_valid |= ATTR_KILL_SUID;
1376
1377         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1378             !(attr->ia_mode & S_ISGID) &&
1379             !(attr->ia_valid & ATTR_KILL_SGID))
1380                 attr->ia_valid |= ATTR_KILL_SGID;
1381
1382         return ll_setattr_raw(de->d_inode, attr);
1383 }
1384
1385 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1386                        __u64 max_age, __u32 flags)
1387 {
1388         struct ll_sb_info *sbi = ll_s2sbi(sb);
1389         struct obd_statfs obd_osfs;
1390         int rc;
1391         ENTRY;
1392
1393         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1394         if (rc) {
1395                 CERROR("md_statfs fails: rc = %d\n", rc);
1396                 RETURN(rc);
1397         }
1398
1399         osfs->os_type = sb->s_magic;
1400
1401         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1402                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1403
1404         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1405                 flags |= OBD_STATFS_NODELAY;
1406
1407         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1408                               &obd_osfs, max_age, flags);
1409         if (rc) {
1410                 CERROR("obd_statfs fails: rc = %d\n", rc);
1411                 RETURN(rc);
1412         }
1413
1414         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1415                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1416                obd_osfs.os_files);
1417
1418         osfs->os_bsize = obd_osfs.os_bsize;
1419         osfs->os_blocks = obd_osfs.os_blocks;
1420         osfs->os_bfree = obd_osfs.os_bfree;
1421         osfs->os_bavail = obd_osfs.os_bavail;
1422
1423         /* If we don't have as many objects free on the OST as inodes
1424          * on the MDS, we reduce the total number of inodes to
1425          * compensate, so that the "inodes in use" number is correct.
1426          */
1427         if (obd_osfs.os_ffree < osfs->os_ffree) {
1428                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1429                         obd_osfs.os_ffree;
1430                 osfs->os_ffree = obd_osfs.os_ffree;
1431         }
1432
1433         RETURN(rc);
1434 }
1435 #ifndef HAVE_STATFS_DENTRY_PARAM
1436 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1437 {
1438 #else
1439 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1440 {
1441         struct super_block *sb = de->d_sb;
1442 #endif
1443         struct obd_statfs osfs;
1444         int rc;
1445
1446         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1447         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1448
1449         /* Some amount of caching on the client is allowed */
1450         rc = ll_statfs_internal(sb, &osfs,
1451                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1452                                 0);
1453         if (rc)
1454                 return rc;
1455
1456         statfs_unpack(sfs, &osfs);
1457
1458         /* We need to downshift for all 32-bit kernels, because we can't
1459          * tell if the kernel is being called via sys_statfs64() or not.
1460          * Stop before overflowing f_bsize - in which case it is better
1461          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1462         if (sizeof(long) < 8) {
1463                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1464                         sfs->f_bsize <<= 1;
1465
1466                         osfs.os_blocks >>= 1;
1467                         osfs.os_bfree >>= 1;
1468                         osfs.os_bavail >>= 1;
1469                 }
1470         }
1471
1472         sfs->f_blocks = osfs.os_blocks;
1473         sfs->f_bfree = osfs.os_bfree;
1474         sfs->f_bavail = osfs.os_bavail;
1475
1476         return 0;
1477 }
1478
1479 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1480 {
1481         struct ll_inode_info *lli;
1482         struct lov_stripe_md *lsm;
1483
1484         lli = ll_i2info(inode);
1485         LASSERT(lli->lli_size_sem_owner != current);
1486         cfs_down(&lli->lli_size_sem);
1487         LASSERT(lli->lli_size_sem_owner == NULL);
1488         lli->lli_size_sem_owner = current;
1489         lsm = lli->lli_smd;
1490         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1491                  lsm, lock_lsm);
1492         if (lock_lsm)
1493                 lov_stripe_lock(lsm);
1494 }
1495
1496 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1497 {
1498         struct ll_inode_info *lli;
1499         struct lov_stripe_md *lsm;
1500
1501         lli = ll_i2info(inode);
1502         lsm = lli->lli_smd;
1503         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1504                  lsm, unlock_lsm);
1505         if (unlock_lsm)
1506                 lov_stripe_unlock(lsm);
1507         LASSERT(lli->lli_size_sem_owner == current);
1508         lli->lli_size_sem_owner = NULL;
1509         cfs_up(&lli->lli_size_sem);
1510 }
1511
1512 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1513 {
1514         struct ll_inode_info *lli = ll_i2info(inode);
1515         struct mdt_body *body = md->body;
1516         struct lov_stripe_md *lsm = md->lsm;
1517         struct ll_sb_info *sbi = ll_i2sbi(inode);
1518
1519         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1520         if (lsm != NULL) {
1521                 cfs_down(&lli->lli_och_sem);
1522                 if (lli->lli_smd == NULL) {
1523                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1524                             lsm->lsm_magic != LOV_MAGIC_V3) {
1525                                 dump_lsm(D_ERROR, lsm);
1526                                 LBUG();
1527                         }
1528                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1529                                lsm, inode->i_ino, inode->i_generation, inode);
1530                         /* cl_inode_init must go before lli_smd or a race is
1531                          * possible where client thinks the file has stripes,
1532                          * but lov raid0 is not setup yet and parallel e.g.
1533                          * glimpse would try to use uninitialized lov */
1534                         cl_inode_init(inode, md);
1535                         cfs_spin_lock(&lli->lli_lock);
1536                         lli->lli_smd = lsm;
1537                         cfs_spin_unlock(&lli->lli_lock);
1538                         cfs_up(&lli->lli_och_sem);
1539                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1540                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1541                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1542                 } else {
1543                         cfs_up(&lli->lli_och_sem);
1544                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1545                                 lli->lli_smd->lsm_stripe_count ==
1546                                 lsm->lsm_stripe_count);
1547                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1548                                 CERROR("lsm mismatch for inode %ld\n",
1549                                        inode->i_ino);
1550                                 CERROR("lli_smd:\n");
1551                                 dump_lsm(D_ERROR, lli->lli_smd);
1552                                 CERROR("lsm:\n");
1553                                 dump_lsm(D_ERROR, lsm);
1554                                 LBUG();
1555                         }
1556                 }
1557                 if (lli->lli_smd != lsm)
1558                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1559         }
1560
1561         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1562                 if (body->valid & OBD_MD_FLRMTPERM)
1563                         ll_update_remote_perm(inode, md->remote_perm);
1564         }
1565 #ifdef CONFIG_FS_POSIX_ACL
1566         else if (body->valid & OBD_MD_FLACL) {
1567                 cfs_spin_lock(&lli->lli_lock);
1568                 if (lli->lli_posix_acl)
1569                         posix_acl_release(lli->lli_posix_acl);
1570                 lli->lli_posix_acl = md->posix_acl;
1571                 cfs_spin_unlock(&lli->lli_lock);
1572         }
1573 #endif
1574         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1575         inode->i_generation = cl_fid_build_gen(&body->fid1);
1576
1577         if (body->valid & OBD_MD_FLATIME) {
1578                 if (body->atime > LTIME_S(inode->i_atime))
1579                         LTIME_S(inode->i_atime) = body->atime;
1580                 lli->lli_lvb.lvb_atime = body->atime;
1581         }
1582         if (body->valid & OBD_MD_FLMTIME) {
1583                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1584                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1585                                "to "LPU64"\n", inode->i_ino,
1586                                LTIME_S(inode->i_mtime), body->mtime);
1587                         LTIME_S(inode->i_mtime) = body->mtime;
1588                 }
1589                 lli->lli_lvb.lvb_mtime = body->mtime;
1590         }
1591         if (body->valid & OBD_MD_FLCTIME) {
1592                 if (body->ctime > LTIME_S(inode->i_ctime))
1593                         LTIME_S(inode->i_ctime) = body->ctime;
1594                 lli->lli_lvb.lvb_ctime = body->ctime;
1595         }
1596         if (body->valid & OBD_MD_FLMODE)
1597                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1598         if (body->valid & OBD_MD_FLTYPE)
1599                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1600         LASSERT(inode->i_mode != 0);
1601         if (S_ISREG(inode->i_mode)) {
1602                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1603         } else {
1604                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1605         }
1606 #ifdef HAVE_INODE_BLKSIZE
1607         inode->i_blksize = 1<<inode->i_blkbits;
1608 #endif
1609         if (body->valid & OBD_MD_FLUID)
1610                 inode->i_uid = body->uid;
1611         if (body->valid & OBD_MD_FLGID)
1612                 inode->i_gid = body->gid;
1613         if (body->valid & OBD_MD_FLFLAGS)
1614                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1615         if (body->valid & OBD_MD_FLNLINK)
1616                 inode->i_nlink = body->nlink;
1617         if (body->valid & OBD_MD_FLRDEV)
1618                 inode->i_rdev = old_decode_dev(body->rdev);
1619
1620         if (body->valid & OBD_MD_FLID) {
1621                 /* FID shouldn't be changed! */
1622                 if (fid_is_sane(&lli->lli_fid)) {
1623                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1624                                  "Trying to change FID "DFID
1625                                  " to the "DFID", inode %lu/%u(%p)\n",
1626                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1627                                  inode->i_ino, inode->i_generation, inode);
1628                 } else
1629                         lli->lli_fid = body->fid1;
1630         }
1631
1632         LASSERT(fid_seq(&lli->lli_fid) != 0);
1633
1634         if (body->valid & OBD_MD_FLSIZE) {
1635                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1636                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1637                         struct lustre_handle lockh;
1638                         ldlm_mode_t mode;
1639
1640                         /* As it is possible a blocking ast has been processed
1641                          * by this time, we need to check there is an UPDATE
1642                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1643                          * it. */
1644                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1645                                                &lockh);
1646                         if (mode) {
1647                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1648                                                       LLIF_EPOCH_PENDING |
1649                                                       LLIF_SOM_DIRTY)) {
1650                                         CERROR("ino %lu flags %lu still has "
1651                                                "size authority! do not trust "
1652                                                "the size got from MDS\n",
1653                                                inode->i_ino, lli->lli_flags);
1654                                 } else {
1655                                         /* Use old size assignment to avoid
1656                                          * deadlock bz14138 & bz14326 */
1657                                         inode->i_size = body->size;
1658                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1659                                 }
1660                                 ldlm_lock_decref(&lockh, mode);
1661                         }
1662                 } else {
1663                         /* Use old size assignment to avoid
1664                          * deadlock bz14138 & bz14326 */
1665                         inode->i_size = body->size;
1666
1667                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1668                                inode->i_ino, (unsigned long long)body->size);
1669                 }
1670
1671                 if (body->valid & OBD_MD_FLBLOCKS)
1672                         inode->i_blocks = body->blocks;
1673         }
1674
1675         if (body->valid & OBD_MD_FLMDSCAPA) {
1676                 LASSERT(md->mds_capa);
1677                 ll_add_capa(inode, md->mds_capa);
1678         }
1679         if (body->valid & OBD_MD_FLOSSCAPA) {
1680                 LASSERT(md->oss_capa);
1681                 ll_add_capa(inode, md->oss_capa);
1682         }
1683 }
1684
1685 void ll_read_inode2(struct inode *inode, void *opaque)
1686 {
1687         struct lustre_md *md = opaque;
1688         struct ll_inode_info *lli = ll_i2info(inode);
1689         ENTRY;
1690
1691         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1692                inode->i_ino, inode->i_generation, inode);
1693
1694         ll_lli_init(lli);
1695
1696         LASSERT(!lli->lli_smd);
1697
1698         /* Core attributes from the MDS first.  This is a new inode, and
1699          * the VFS doesn't zero times in the core inode so we have to do
1700          * it ourselves.  They will be overwritten by either MDS or OST
1701          * attributes - we just need to make sure they aren't newer. */
1702         LTIME_S(inode->i_mtime) = 0;
1703         LTIME_S(inode->i_atime) = 0;
1704         LTIME_S(inode->i_ctime) = 0;
1705         inode->i_rdev = 0;
1706         ll_update_inode(inode, md);
1707
1708         /* OIDEBUG(inode); */
1709
1710         /* initializing backing dev info. */
1711         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1712
1713
1714         if (S_ISREG(inode->i_mode)) {
1715                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1716                 inode->i_op = &ll_file_inode_operations;
1717                 inode->i_fop = sbi->ll_fop;
1718                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1719                 EXIT;
1720         } else if (S_ISDIR(inode->i_mode)) {
1721                 inode->i_op = &ll_dir_inode_operations;
1722                 inode->i_fop = &ll_dir_operations;
1723                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_dir_aops;
1724                 EXIT;
1725         } else if (S_ISLNK(inode->i_mode)) {
1726                 inode->i_op = &ll_fast_symlink_inode_operations;
1727                 EXIT;
1728         } else {
1729                 inode->i_op = &ll_special_inode_operations;
1730
1731                 init_special_inode(inode, inode->i_mode,
1732                                    kdev_t_to_nr(inode->i_rdev));
1733
1734                 EXIT;
1735         }
1736 }
1737
1738 void ll_delete_inode(struct inode *inode)
1739 {
1740         struct ll_sb_info *sbi = ll_i2sbi(inode);
1741         int rc;
1742         ENTRY;
1743
1744         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1745         if (rc)
1746                 CERROR("fid_delete() failed, rc %d\n", rc);
1747
1748         truncate_inode_pages(&inode->i_data, 0);
1749
1750         /* Workaround for LU-118 */
1751         if (inode->i_data.nrpages) {
1752                 TREE_READ_LOCK_IRQ(&inode->i_data);
1753                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1754                 LASSERTF(inode->i_data.nrpages == 0,
1755                          "inode=%lu/%u(%p) nrpages=%lu, see "
1756                          "http://jira.whamcloud.com/browse/LU-118\n",
1757                          inode->i_ino, inode->i_generation, inode,
1758                          inode->i_data.nrpages);
1759         }
1760         /* Workaround end */
1761
1762         clear_inode(inode);
1763
1764         EXIT;
1765 }
1766
1767 int ll_iocontrol(struct inode *inode, struct file *file,
1768                  unsigned int cmd, unsigned long arg)
1769 {
1770         struct ll_sb_info *sbi = ll_i2sbi(inode);
1771         struct ptlrpc_request *req = NULL;
1772         int rc, flags = 0;
1773         ENTRY;
1774
1775         switch(cmd) {
1776         case FSFILT_IOC_GETFLAGS: {
1777                 struct mdt_body *body;
1778                 struct md_op_data *op_data;
1779
1780                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1781                                              0, 0, LUSTRE_OPC_ANY,
1782                                              NULL);
1783                 if (op_data == NULL)
1784                         RETURN(-ENOMEM);
1785
1786                 op_data->op_valid = OBD_MD_FLFLAGS;
1787                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1788                 ll_finish_md_op_data(op_data);
1789                 if (rc) {
1790                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1791                         RETURN(-abs(rc));
1792                 }
1793
1794                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1795
1796                 flags = body->flags;
1797
1798                 ptlrpc_req_finished(req);
1799
1800                 RETURN(put_user(flags, (int *)arg));
1801         }
1802         case FSFILT_IOC_SETFLAGS: {
1803                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1804                 struct obd_info oinfo = { { { 0 } } };
1805                 struct md_op_data *op_data;
1806
1807                 if (get_user(flags, (int *)arg))
1808                         RETURN(-EFAULT);
1809
1810                 oinfo.oi_md = lsm;
1811                 OBDO_ALLOC(oinfo.oi_oa);
1812                 if (!oinfo.oi_oa)
1813                         RETURN(-ENOMEM);
1814
1815                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1816                                              LUSTRE_OPC_ANY, NULL);
1817                 if (IS_ERR(op_data))
1818                         RETURN(PTR_ERR(op_data));
1819
1820                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1821                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1822                 rc = md_setattr(sbi->ll_md_exp, op_data,
1823                                 NULL, 0, NULL, 0, &req, NULL);
1824                 ll_finish_md_op_data(op_data);
1825                 ptlrpc_req_finished(req);
1826                 if (rc) {
1827                         OBDO_FREE(oinfo.oi_oa);
1828                         RETURN(rc);
1829                 }
1830
1831                 if (lsm == NULL) {
1832                         OBDO_FREE(oinfo.oi_oa);
1833                         GOTO(update_cache, rc);
1834                 }
1835
1836                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1837                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1838                 oinfo.oi_oa->o_flags = flags;
1839                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1840                                        OBD_MD_FLGROUP;
1841                 oinfo.oi_capa = ll_mdscapa_get(inode);
1842                 obdo_from_inode(oinfo.oi_oa, inode,
1843                                 &ll_i2info(inode)->lli_fid, 0);
1844                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1845                 capa_put(oinfo.oi_capa);
1846                 OBDO_FREE(oinfo.oi_oa);
1847                 if (rc) {
1848                         if (rc != -EPERM && rc != -EACCES)
1849                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1850                         RETURN(rc);
1851                 }
1852
1853                 EXIT;
1854 update_cache:
1855                 inode->i_flags = ll_ext_to_inode_flags(flags);
1856                 return 0;
1857         }
1858         default:
1859                 RETURN(-ENOSYS);
1860         }
1861
1862         RETURN(0);
1863 }
1864
1865 int ll_flush_ctx(struct inode *inode)
1866 {
1867         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1868
1869         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1870
1871         obd_set_info_async(sbi->ll_md_exp,
1872                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1873                            0, NULL, NULL);
1874         obd_set_info_async(sbi->ll_dt_exp,
1875                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1876                            0, NULL, NULL);
1877         return 0;
1878 }
1879
1880 /* umount -f client means force down, don't save state */
1881 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1882 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1883 {
1884         struct super_block *sb = vfsmnt->mnt_sb;
1885 #else
1886 void ll_umount_begin(struct super_block *sb)
1887 {
1888 #endif
1889         struct lustre_sb_info *lsi = s2lsi(sb);
1890         struct ll_sb_info *sbi = ll_s2sbi(sb);
1891         struct obd_device *obd;
1892         struct obd_ioctl_data *ioc_data;
1893         ENTRY;
1894
1895 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1896         if (!(flags & MNT_FORCE)) {
1897                 EXIT;
1898                 return;
1899         }
1900 #endif
1901
1902         /* Tell the MGC we got umount -f */
1903         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1904
1905         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1906                sb->s_count, atomic_read(&sb->s_active));
1907
1908         obd = class_exp2obd(sbi->ll_md_exp);
1909         if (obd == NULL) {
1910                 CERROR("Invalid MDC connection handle "LPX64"\n",
1911                        sbi->ll_md_exp->exp_handle.h_cookie);
1912                 EXIT;
1913                 return;
1914         }
1915         obd->obd_force = 1;
1916
1917         obd = class_exp2obd(sbi->ll_dt_exp);
1918         if (obd == NULL) {
1919                 CERROR("Invalid LOV connection handle "LPX64"\n",
1920                        sbi->ll_dt_exp->exp_handle.h_cookie);
1921                 EXIT;
1922                 return;
1923         }
1924         obd->obd_force = 1;
1925
1926         OBD_ALLOC_PTR(ioc_data);
1927         if (ioc_data) {
1928                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1929                               sizeof ioc_data, ioc_data, NULL);
1930
1931                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1932                               sizeof ioc_data, ioc_data, NULL);
1933
1934                 OBD_FREE_PTR(ioc_data);
1935         }
1936
1937
1938         /* Really, we'd like to wait until there are no requests outstanding,
1939          * and then continue.  For now, we just invalidate the requests,
1940          * schedule() and sleep one second if needed, and hope.
1941          */
1942         cfs_schedule();
1943 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1944         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1945                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1946                                                    cfs_time_seconds(1));
1947                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1948                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1949                                       "may try to umount it a bit later\n",
1950                                       atomic_read(&vfsmnt->mnt_count));
1951         }
1952 #endif
1953
1954         EXIT;
1955 }
1956
1957 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1958 {
1959         struct ll_sb_info *sbi = ll_s2sbi(sb);
1960         int err;
1961         __u32 read_only;
1962
1963         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1964                 read_only = *flags & MS_RDONLY;
1965                 err = obd_set_info_async(sbi->ll_md_exp,
1966                                          sizeof(KEY_READ_ONLY),
1967                                          KEY_READ_ONLY, sizeof(read_only),
1968                                          &read_only, NULL);
1969                 if (err) {
1970                         CERROR("Failed to change the read-only flag during "
1971                                "remount: %d\n", err);
1972                         return err;
1973                 }
1974
1975                 if (read_only)
1976                         sb->s_flags |= MS_RDONLY;
1977                 else
1978                         sb->s_flags &= ~MS_RDONLY;
1979         }
1980         return 0;
1981 }
1982
1983 int ll_prep_inode(struct inode **inode,
1984                   struct ptlrpc_request *req,
1985                   struct super_block *sb)
1986 {
1987         struct ll_sb_info *sbi = NULL;
1988         struct lustre_md md;
1989         int rc;
1990         ENTRY;
1991
1992         LASSERT(*inode || sb);
1993         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1994         memset(&md, 0, sizeof(struct lustre_md));
1995
1996         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1997                               sbi->ll_md_exp, &md);
1998         if (rc)
1999                 RETURN(rc);
2000
2001         if (*inode) {
2002                 ll_update_inode(*inode, &md);
2003         } else {
2004                 LASSERT(sb != NULL);
2005
2006                 /*
2007                  * At this point server returns to client's same fid as client
2008                  * generated for creating. So using ->fid1 is okay here.
2009                  */
2010                 LASSERT(fid_is_sane(&md.body->fid1));
2011
2012                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2013                 if (*inode == NULL || IS_ERR(*inode)) {
2014                         if (md.lsm)
2015                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2016 #ifdef CONFIG_FS_POSIX_ACL
2017                         if (md.posix_acl) {
2018                                 posix_acl_release(md.posix_acl);
2019                                 md.posix_acl = NULL;
2020                         }
2021 #endif
2022                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2023                         *inode = NULL;
2024                         CERROR("new_inode -fatal: rc %d\n", rc);
2025                         GOTO(out, rc);
2026                 }
2027         }
2028
2029 out:
2030         md_free_lustre_md(sbi->ll_md_exp, &md);
2031         RETURN(rc);
2032 }
2033
2034 int ll_obd_statfs(struct inode *inode, void *arg)
2035 {
2036         struct ll_sb_info *sbi = NULL;
2037         struct obd_export *exp;
2038         char *buf = NULL;
2039         struct obd_ioctl_data *data = NULL;
2040         __u32 type;
2041         int len = 0, rc;
2042
2043         if (!inode || !(sbi = ll_i2sbi(inode)))
2044                 GOTO(out_statfs, rc = -EINVAL);
2045
2046         rc = obd_ioctl_getdata(&buf, &len, arg);
2047         if (rc)
2048                 GOTO(out_statfs, rc);
2049
2050         data = (void*)buf;
2051         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2052             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2053                 GOTO(out_statfs, rc = -EINVAL);
2054
2055         if (data->ioc_inllen1 != sizeof(__u32) ||
2056             data->ioc_inllen2 != sizeof(__u32) ||
2057             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2058             data->ioc_plen2 != sizeof(struct obd_uuid))
2059                 GOTO(out_statfs, rc = -EINVAL);
2060
2061         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2062         if (type == LL_STATFS_MDC)
2063                 exp = sbi->ll_md_exp;
2064         else if (type == LL_STATFS_LOV)
2065                 exp = sbi->ll_dt_exp;
2066         else
2067                 GOTO(out_statfs, rc = -ENODEV);
2068
2069         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2070         if (rc)
2071                 GOTO(out_statfs, rc);
2072 out_statfs:
2073         if (buf)
2074                 obd_ioctl_freedata(buf, len);
2075         return rc;
2076 }
2077
2078 int ll_process_config(struct lustre_cfg *lcfg)
2079 {
2080         char *ptr;
2081         void *sb;
2082         struct lprocfs_static_vars lvars;
2083         unsigned long x;
2084         int rc = 0;
2085
2086         lprocfs_llite_init_vars(&lvars);
2087
2088         /* The instance name contains the sb: lustre-client-aacfe000 */
2089         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2090         if (!ptr || !*(++ptr))
2091                 return -EINVAL;
2092         if (sscanf(ptr, "%lx", &x) != 1)
2093                 return -EINVAL;
2094         sb = (void *)x;
2095         /* This better be a real Lustre superblock! */
2096         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2097
2098         /* Note we have not called client_common_fill_super yet, so
2099            proc fns must be able to handle that! */
2100         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2101                                       lcfg, sb);
2102         if (rc > 0)
2103                 rc = 0;
2104         return(rc);
2105 }
2106
2107 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2108 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2109                                        struct inode *i1, struct inode *i2,
2110                                        const char *name, int namelen,
2111                                        int mode, __u32 opc, void *data)
2112 {
2113         LASSERT(i1 != NULL);
2114
2115         if (namelen > ll_i2sbi(i1)->ll_namelen)
2116                 return ERR_PTR(-ENAMETOOLONG);
2117
2118         if (op_data == NULL)
2119                 OBD_ALLOC_PTR(op_data);
2120
2121         if (op_data == NULL)
2122                 return ERR_PTR(-ENOMEM);
2123
2124         ll_i2gids(op_data->op_suppgids, i1, i2);
2125         op_data->op_fid1 = *ll_inode2fid(i1);
2126         op_data->op_capa1 = ll_mdscapa_get(i1);
2127
2128         if (i2) {
2129                 op_data->op_fid2 = *ll_inode2fid(i2);
2130                 op_data->op_capa2 = ll_mdscapa_get(i2);
2131         } else {
2132                 fid_zero(&op_data->op_fid2);
2133                 op_data->op_capa2 = NULL;
2134         }
2135
2136         op_data->op_name = name;
2137         op_data->op_namelen = namelen;
2138         op_data->op_mode = mode;
2139         op_data->op_mod_time = cfs_time_current_sec();
2140         op_data->op_fsuid = cfs_curproc_fsuid();
2141         op_data->op_fsgid = cfs_curproc_fsgid();
2142         op_data->op_cap = cfs_curproc_cap_pack();
2143         op_data->op_bias = MDS_CHECK_SPLIT;
2144         op_data->op_opc = opc;
2145         op_data->op_mds = 0;
2146         op_data->op_data = data;
2147
2148         /* If the file is being opened after mknod() (normally due to NFS)
2149          * try to use the default stripe data from parent directory for
2150          * allocating OST objects.  Try to pass the parent FID to MDS. */
2151         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2152             ll_i2info(i2)->lli_smd == NULL) {
2153                 struct ll_inode_info *lli = ll_i2info(i2);
2154
2155                 cfs_spin_lock(&lli->lli_lock);
2156                 if (likely(lli->lli_smd == NULL &&
2157                            !fid_is_zero(&lli->lli_pfid)))
2158                         op_data->op_fid1 = lli->lli_pfid;
2159                 cfs_spin_unlock(&lli->lli_lock);
2160                 /** We ignore parent's capability temporary. */
2161         }
2162
2163         return op_data;
2164 }
2165
2166 void ll_finish_md_op_data(struct md_op_data *op_data)
2167 {
2168         capa_put(op_data->op_capa1);
2169         capa_put(op_data->op_capa2);
2170         OBD_FREE_PTR(op_data);
2171 }
2172
2173 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2174 {
2175         struct ll_sb_info *sbi;
2176
2177         LASSERT((seq != NULL) && (vfs != NULL));
2178         sbi = ll_s2sbi(vfs->mnt_sb);
2179
2180         if (sbi->ll_flags & LL_SBI_NOLCK)
2181                 seq_puts(seq, ",nolock");
2182
2183         if (sbi->ll_flags & LL_SBI_FLOCK)
2184                 seq_puts(seq, ",flock");
2185
2186         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2187                 seq_puts(seq, ",localflock");
2188
2189         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2190                 seq_puts(seq, ",user_xattr");
2191
2192         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2193                 seq_puts(seq, ",lazystatfs");
2194
2195         RETURN(0);
2196 }