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