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  2008 Sun Microsystems, Inc. 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/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 CFS_LIST_HEAD(ll_super_blocks);
  63 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) cfs_ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC(sbi, sizeof(*sbi));
  82         if (!sbi)
  83                 RETURN(NULL);
  84
  85         cfs_spin_lock_init(&sbi->ll_lock);
  86         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  87         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  88         cfs_spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         cfs_spin_lock(&ll_sb_lock);
 116         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         cfs_spin_unlock(&ll_sb_lock);
 118
 119 #ifdef ENABLE_LLITE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127 #ifdef HAVE_EXPORT___IGET
 128         CFS_INIT_LIST_HEAD(&sbi->ll_deathrow);
 129         cfs_spin_lock_init(&sbi->ll_deathrow_lock);
 130 #endif
 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
 141         RETURN(sbi);
 142 }
 143
 144 void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 cfs_spin_lock(&ll_sb_lock);
 151                 cfs_list_del(&sbi->ll_list);
 152                 cfs_spin_unlock(&ll_sb_lock);
 153                 OBD_FREE(sbi, sizeof(*sbi));
 154         }
 155         EXIT;
 156 }
 157
 158 static struct dentry_operations ll_d_root_ops = {
 159         .d_compare = ll_dcompare,
 160         .d_revalidate = ll_revalidate_nd,
 161 };
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 164 {
 165         struct inode *root = 0;
 166         struct ll_sb_info *sbi = ll_s2sbi(sb);
 167         struct obd_device *obd;
 168         struct obd_capa *oc = NULL;
 169         struct obd_statfs osfs;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct lustre_md lmd;
 174         obd_valid valid;
 175         int size, err, checksum;
 176         ENTRY;
 177
 178         obd = class_name2obd(md);
 179         if (!obd) {
 180                 CERROR("MD %s: not setup or attached\n", md);
 181                 RETURN(-EINVAL);
 182         }
 183
 184         OBD_ALLOC_PTR(data);
 185         if (data == NULL)
 186                 RETURN(-ENOMEM);
 187
 188         if (proc_lustre_fs_root) {
 189                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 190                                                   dt, md);
 191                 if (err < 0)
 192                         CERROR("could not register mount in /proc/fs/lustre\n");
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 199                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 200                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 201                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 202                                   OBD_CONNECT_VBR      | OBD_CONNECT_SOM;
 203
 204 #ifdef HAVE_LRU_RESIZE_SUPPORT
 205         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 206                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 207 #endif
 208 #ifdef CONFIG_FS_POSIX_ACL
 209         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 210 #endif
 211         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213
 214         if (sb->s_flags & MS_RDONLY)
 215                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 216         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 217                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 218
 219 #ifdef HAVE_MS_FLOCK_LOCK
 220         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 221         sb->s_flags |= MS_FLOCK_LOCK;
 222 #endif
 223
 224         if (sbi->ll_flags & LL_SBI_FLOCK)
 225                 sbi->ll_fop = &ll_file_operations_flock;
 226         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 227                 sbi->ll_fop = &ll_file_operations;
 228         else
 229                 sbi->ll_fop = &ll_file_operations_noflock;
 230
 231         /* real client */
 232         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 233         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 235
 236         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 237         if (err == -EBUSY) {
 238                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 239                                    "recovery, of which this client is not a "
 240                                    "part. Please wait for recovery to complete,"
 241                                    " abort, or time out.\n", md);
 242                 GOTO(out, err);
 243         } else if (err) {
 244                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 245                 GOTO(out, err);
 246         }
 247
 248         err = obd_fid_init(sbi->ll_md_exp);
 249         if (err) {
 250                 CERROR("Can't init metadata layer FID infrastructure, "
 251                        "rc %d\n", err);
 252                 GOTO(out_md, err);
 253         }
 254
 255         err = obd_statfs(obd, &osfs, cfs_time_current_64() - CFS_HZ, 0);
 256         if (err)
 257                 GOTO(out_md_fid, err);
 258
 259         size = sizeof(*data);
 260         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 261                            KEY_CONN_DATA,  &size, data, NULL);
 262         if (err) {
 263                 CERROR("Get connect data failed: %d \n", err);
 264                 GOTO(out_md, err);
 265         }
 266
 267         LASSERT(osfs.os_bsize);
 268         sb->s_blocksize = osfs.os_bsize;
 269         sb->s_blocksize_bits = log2(osfs.os_bsize);
 270         sb->s_magic = LL_SUPER_MAGIC;
 271
 272         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 273          *         retval = in_file->f_op->sendfile(...);
 274          *         if (*ppos > max)
 275          *                 retval = -EOVERFLOW;
 276          *
 277          * it will check if *ppos is greater than max. However, max equals to
 278          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 279          * has been defined as a signed long long integer in linux kernel. */
 280 #if BITS_PER_LONG == 64
 281         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 282 #else
 283         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 284 #endif
 285         sbi->ll_namelen = osfs.os_namelen;
 286         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 287
 288         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 289             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 290                 LCONSOLE_INFO("Disabling user_xattr feature because "
 291                               "it is not supported on the server\n");
 292                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 293         }
 294
 295         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 296 #ifdef MS_POSIXACL
 297                 sb->s_flags |= MS_POSIXACL;
 298 #endif
 299                 sbi->ll_flags |= LL_SBI_ACL;
 300         } else {
 301                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 302 #ifdef MS_POSIXACL
 303                 sb->s_flags &= ~MS_POSIXACL;
 304 #endif
 305                 sbi->ll_flags &= ~LL_SBI_ACL;
 306         }
 307
 308         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 309                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 310                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 311                         LCONSOLE_INFO("client is set as remote by default.\n");
 312                 }
 313         } else {
 314                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 315                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 316                         LCONSOLE_INFO("client claims to be remote, but server "
 317                                       "rejected, forced to be local.\n");
 318                 }
 319         }
 320
 321         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 322                 LCONSOLE_INFO("client enabled MDS capability!\n");
 323                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 324         }
 325
 326         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 327                 LCONSOLE_INFO("client enabled OSS capability!\n");
 328                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 329         }
 330
 331         obd = class_name2obd(dt);
 332         if (!obd) {
 333                 CERROR("DT %s: not setup or attached\n", dt);
 334                 GOTO(out_md_fid, err = -ENODEV);
 335         }
 336
 337         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 338                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 339                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 340                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 341                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 342                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 343                                   OBD_CONNECT_SOM;
 344
 345         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 346                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 347                  * disabled by default, because it can still be enabled on the
 348                  * fly via /proc. As a consequence, we still need to come to an
 349                  * agreement on the supported algorithms at connect time */
 350                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 351
 352                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 353                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 354                 else
 355                         /* send the list of supported checksum types */
 356                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 357         }
 358
 359 #ifdef HAVE_LRU_RESIZE_SUPPORT
 360         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 361 #endif
 362         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 363                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 364
 365         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 366                "ocd_grant: %d\n", data->ocd_connect_flags,
 367                data->ocd_version, data->ocd_grant);
 368
 369         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 370         obd->obd_upcall.onu_upcall = cl_ocd_update;
 371         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 372
 373         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 374         if (err == -EBUSY) {
 375                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 376                                    "recovery, of which this client is not a "
 377                                    "part.  Please wait for recovery to "
 378                                    "complete, abort, or time out.\n", dt);
 379                 GOTO(out_md_fid, err);
 380         } else if (err) {
 381                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 382                 GOTO(out_md_fid, err);
 383         }
 384
 385         err = obd_fid_init(sbi->ll_dt_exp);
 386         if (err) {
 387                 CERROR("Can't init data layer FID infrastructure, "
 388                        "rc %d\n", err);
 389                 GOTO(out_dt, err);
 390         }
 391
 392         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 393         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 394         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 395         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 396         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 397
 398         fid_zero(&sbi->ll_root_fid);
 399         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 400         if (err) {
 401                 CERROR("cannot mds_connect: rc = %d\n", err);
 402                 GOTO(out_lock_cn_cb, err);
 403         }
 404         if (!fid_is_sane(&sbi->ll_root_fid)) {
 405                 CERROR("Invalid root fid during mount\n");
 406                 GOTO(out_lock_cn_cb, err = -EINVAL);
 407         }
 408         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 409
 410         sb->s_op = &lustre_super_operations;
 411         sb->s_export_op = &lustre_export_operations;
 412
 413         /* make root inode
 414          * XXX: move this to after cbd setup? */
 415         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 416         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 417                 valid |= OBD_MD_FLRMTPERM;
 418         else if (sbi->ll_flags & LL_SBI_ACL)
 419                 valid |= OBD_MD_FLACL;
 420
 421         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 422                          &request);
 423         if (oc)
 424                 capa_put(oc);
 425         if (err) {
 426                 CERROR("md_getattr failed for root: rc = %d\n", err);
 427                 GOTO(out_lock_cn_cb, err);
 428         }
 429         memset(&lmd, 0, sizeof(lmd));
 430         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 431                                sbi->ll_md_exp, &lmd);
 432         if (err) {
 433                 CERROR("failed to understand root inode md: rc = %d\n", err);
 434                 ptlrpc_req_finished (request);
 435                 GOTO(out_lock_cn_cb, err);
 436         }
 437
 438         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 439         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
 440         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 441         ptlrpc_req_finished(request);
 442
 443         if (root == NULL || IS_ERR(root)) {
 444                 if (lmd.lsm)
 445                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 446 #ifdef CONFIG_FS_POSIX_ACL
 447                 if (lmd.posix_acl) {
 448                         posix_acl_release(lmd.posix_acl);
 449                         lmd.posix_acl = NULL;
 450                 }
 451 #endif
 452                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 453                 root = NULL;
 454                 CERROR("lustre_lite: bad iget4 for root\n");
 455                 GOTO(out_root, err);
 456         }
 457
 458         err = ll_close_thread_start(&sbi->ll_lcq);
 459         if (err) {
 460                 CERROR("cannot start close thread: rc %d\n", err);
 461                 GOTO(out_root, err);
 462         }
 463
 464 #ifdef CONFIG_FS_POSIX_ACL
 465         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 466                 rct_init(&sbi->ll_rct);
 467                 et_init(&sbi->ll_et);
 468         }
 469 #endif
 470
 471         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 472         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 473                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 474                                  NULL);
 475         cl_sb_init(sb);
 476
 477         sb->s_root = d_alloc_root(root);
 478         if (data != NULL)
 479                 OBD_FREE(data, sizeof(*data));
 480
 481         sb->s_root->d_op = &ll_d_root_ops;
 482
 483         sbi->ll_sdev_orig = sb->s_dev;
 484
 485         /* We set sb->s_dev equal on all lustre clients in order to support
 486          * NFS export clustering.  NFSD requires that the FSID be the same
 487          * on all clients. */
 488         /* s_dev is also used in lt_compare() to compare two fs, but that is
 489          * only a node-local comparison. */
 490         uuid = obd_get_uuid(sbi->ll_md_exp);
 491         if (uuid != NULL)
 492                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 493
 494         RETURN(err);
 495 out_root:
 496         if (root)
 497                 iput(root);
 498 out_lock_cn_cb:
 499         obd_fid_fini(sbi->ll_dt_exp);
 500 out_dt:
 501         obd_disconnect(sbi->ll_dt_exp);
 502         sbi->ll_dt_exp = NULL;
 503 out_md_fid:
 504         obd_fid_fini(sbi->ll_md_exp);
 505 out_md:
 506         obd_disconnect(sbi->ll_md_exp);
 507         sbi->ll_md_exp = NULL;
 508 out:
 509         if (data != NULL)
 510                 OBD_FREE_PTR(data);
 511         lprocfs_unregister_mountpoint(sbi);
 512         return err;
 513 }
 514
 515 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 516 {
 517         int size, rc;
 518
 519         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 520         size = sizeof(int);
 521         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 522                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 523         if (rc)
 524                 CERROR("Get max mdsize error rc %d \n", rc);
 525
 526         RETURN(rc);
 527 }
 528
 529 void ll_dump_inode(struct inode *inode)
 530 {
 531         struct list_head *tmp;
 532         int dentry_count = 0;
 533
 534         LASSERT(inode != NULL);
 535
 536         list_for_each(tmp, &inode->i_dentry)
 537                 dentry_count++;
 538
 539         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 540                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 541                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 542 }
 543
 544 void lustre_dump_dentry(struct dentry *dentry, int recur)
 545 {
 546         struct list_head *tmp;
 547         int subdirs = 0;
 548
 549         LASSERT(dentry != NULL);
 550
 551         list_for_each(tmp, &dentry->d_subdirs)
 552                 subdirs++;
 553
 554         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 555                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 556                dentry->d_name.len, dentry->d_name.name,
 557                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 558                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 559                dentry->d_flags, dentry->d_fsdata, subdirs);
 560         if (dentry->d_inode != NULL)
 561                 ll_dump_inode(dentry->d_inode);
 562
 563         if (recur == 0)
 564                 return;
 565
 566        list_for_each(tmp, &dentry->d_subdirs) {
 567                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 568                 lustre_dump_dentry(d, recur - 1);
 569         }
 570 }
 571
 572 #ifdef HAVE_EXPORT___IGET
 573 static void prune_dir_dentries(struct inode *inode)
 574 {
 575         struct dentry *dentry, *prev = NULL;
 576
 577         /* due to lustre specific logic, a directory
 578          * can have few dentries - a bug from VFS POV */
 579 restart:
 580         spin_lock(&dcache_lock);
 581         if (!list_empty(&inode->i_dentry)) {
 582                 dentry = list_entry(inode->i_dentry.prev,
 583                                     struct dentry, d_alias);
 584                 /* in order to prevent infinite loops we
 585                  * break if previous dentry is busy */
 586                 if (dentry != prev) {
 587                         prev = dentry;
 588                         dget_locked(dentry);
 589                         spin_unlock(&dcache_lock);
 590
 591                         /* try to kill all child dentries */
 592                         lock_dentry(dentry);
 593                         shrink_dcache_parent(dentry);
 594                         unlock_dentry(dentry);
 595                         dput(dentry);
 596
 597                         /* now try to get rid of current dentry */
 598                         d_prune_aliases(inode);
 599                         goto restart;
 600                 }
 601         }
 602         spin_unlock(&dcache_lock);
 603 }
 604
 605 static void prune_deathrow_one(struct ll_inode_info *lli)
 606 {
 607         struct inode *inode = ll_info2i(lli);
 608
 609         /* first, try to drop any dentries - they hold a ref on the inode */
 610         if (S_ISDIR(inode->i_mode))
 611                 prune_dir_dentries(inode);
 612         else
 613                 d_prune_aliases(inode);
 614
 615
 616         /* if somebody still uses it, leave it */
 617         LASSERT(atomic_read(&inode->i_count) > 0);
 618         if (atomic_read(&inode->i_count) > 1)
 619                 goto out;
 620
 621         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 622                inode->i_ino,inode->i_generation,
 623                atomic_read(&inode->i_count));
 624
 625         /* seems nobody uses it anymore */
 626         inode->i_nlink = 0;
 627
 628 out:
 629         iput(inode);
 630         return;
 631 }
 632
 633 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 634 {
 635         struct ll_inode_info *lli;
 636         int empty;
 637
 638         do {
 639                 if (need_resched() && try)
 640                         break;
 641
 642                 if (try) {
 643                         if (!cfs_spin_trylock(&sbi->ll_deathrow_lock))
 644                                 break;
 645                 } else {
 646                         cfs_spin_lock(&sbi->ll_deathrow_lock);
 647                 }
 648
 649                 empty = 1;
 650                 lli = NULL;
 651                 if (!cfs_list_empty(&sbi->ll_deathrow)) {
 652                         lli = cfs_list_entry(sbi->ll_deathrow.next,
 653                                              struct ll_inode_info,
 654                                              lli_dead_list);
 655                         cfs_list_del_init(&lli->lli_dead_list);
 656                         if (!cfs_list_empty(&sbi->ll_deathrow))
 657                                 empty = 0;
 658                 }
 659                 cfs_spin_unlock(&sbi->ll_deathrow_lock);
 660
 661                 if (lli)
 662                         prune_deathrow_one(lli);
 663
 664         } while (empty == 0);
 665 }
 666 #else /* !HAVE_EXPORT___IGET */
 667 #define prune_deathrow(sbi, try) do {} while (0)
 668 #endif /* HAVE_EXPORT___IGET */
 669
 670 void client_common_put_super(struct super_block *sb)
 671 {
 672         struct ll_sb_info *sbi = ll_s2sbi(sb);
 673         ENTRY;
 674
 675 #ifdef CONFIG_FS_POSIX_ACL
 676         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 677                 et_fini(&sbi->ll_et);
 678                 rct_fini(&sbi->ll_rct);
 679         }
 680 #endif
 681
 682         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 683
 684         ll_close_thread_shutdown(sbi->ll_lcq);
 685
 686         cl_sb_fini(sb);
 687
 688         /* destroy inodes in deathrow */
 689         prune_deathrow(sbi, 0);
 690
 691         cfs_list_del(&sbi->ll_conn_chain);
 692
 693         obd_fid_fini(sbi->ll_dt_exp);
 694         obd_disconnect(sbi->ll_dt_exp);
 695         sbi->ll_dt_exp = NULL;
 696
 697         lprocfs_unregister_mountpoint(sbi);
 698
 699         obd_fid_fini(sbi->ll_md_exp);
 700         obd_disconnect(sbi->ll_md_exp);
 701         sbi->ll_md_exp = NULL;
 702
 703         EXIT;
 704 }
 705
 706 void ll_kill_super(struct super_block *sb)
 707 {
 708         struct ll_sb_info *sbi;
 709
 710         ENTRY;
 711
 712         /* not init sb ?*/
 713         if (!(sb->s_flags & MS_ACTIVE))
 714                 return;
 715
 716         sbi = ll_s2sbi(sb);
 717         /* we need restore s_dev from changed for clustred NFS before put_super
 718          * because new kernels have cached s_dev and change sb->s_dev in
 719          * put_super not affected real removing devices */
 720         if (sbi)
 721                 sb->s_dev = sbi->ll_sdev_orig;
 722         EXIT;
 723 }
 724
 725 char *ll_read_opt(const char *opt, char *data)
 726 {
 727         char *value;
 728         char *retval;
 729         ENTRY;
 730
 731         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 732         if (strncmp(opt, data, strlen(opt)))
 733                 RETURN(NULL);
 734         if ((value = strchr(data, '=')) == NULL)
 735                 RETURN(NULL);
 736
 737         value++;
 738         OBD_ALLOC(retval, strlen(value) + 1);
 739         if (!retval) {
 740                 CERROR("out of memory!\n");
 741                 RETURN(NULL);
 742         }
 743
 744         memcpy(retval, value, strlen(value)+1);
 745         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 746         RETURN(retval);
 747 }
 748
 749 static inline int ll_set_opt(const char *opt, char *data, int fl)
 750 {
 751         if (strncmp(opt, data, strlen(opt)) != 0)
 752                 return(0);
 753         else
 754                 return(fl);
 755 }
 756
 757 /* non-client-specific mount options are parsed in lmd_parse */
 758 static int ll_options(char *options, int *flags)
 759 {
 760         int tmp;
 761         char *s1 = options, *s2;
 762         ENTRY;
 763
 764         if (!options)
 765                 RETURN(0);
 766
 767         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 768
 769         while (*s1) {
 770                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 771                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 787                 if (tmp) {
 788                         *flags &= ~tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 797                 if (tmp) {
 798                         *flags &= ~tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         /* Ignore deprecated mount option.  The client will
 804                          * always try to mount with ACL support, whether this
 805                          * is used depends on whether server supports it. */
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 809                 if (tmp) {
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817
 818                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848
 849                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 850                                    s1);
 851                 RETURN(-EINVAL);
 852
 853 next:
 854                 /* Find next opt */
 855                 s2 = strchr(s1, ',');
 856                 if (s2 == NULL)
 857                         break;
 858                 s1 = s2 + 1;
 859         }
 860         RETURN(0);
 861 }
 862
 863 void ll_lli_init(struct ll_inode_info *lli)
 864 {
 865         lli->lli_inode_magic = LLI_INODE_MAGIC;
 866         cfs_sema_init(&lli->lli_size_sem, 1);
 867         cfs_sema_init(&lli->lli_write_sem, 1);
 868         cfs_sema_init(&lli->lli_trunc_sem, 1);
 869         lli->lli_flags = 0;
 870         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 871         cfs_spin_lock_init(&lli->lli_lock);
 872         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 873         lli->lli_inode_magic = LLI_INODE_MAGIC;
 874         cfs_sema_init(&lli->lli_och_sem, 1);
 875         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 876         lli->lli_mds_exec_och = NULL;
 877         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 878         lli->lli_open_fd_exec_count = 0;
 879         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 880         lli->lli_remote_perms = NULL;
 881         lli->lli_rmtperm_utime = 0;
 882         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 883         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 884 }
 885
 886 int ll_fill_super(struct super_block *sb)
 887 {
 888         struct lustre_profile *lprof;
 889         struct lustre_sb_info *lsi = s2lsi(sb);
 890         struct ll_sb_info *sbi;
 891         char  *dt = NULL, *md = NULL;
 892         char  *profilenm = get_profile_name(sb);
 893         struct config_llog_instance cfg = {0, };
 894         char   ll_instance[sizeof(sb) * 2 + 1];
 895         int    err;
 896         ENTRY;
 897
 898         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 899
 900         cfs_module_get();
 901
 902         /* client additional sb info */
 903         lsi->lsi_llsbi = sbi = ll_init_sbi();
 904         if (!sbi) {
 905                 cfs_module_put(THIS_MODULE);
 906                 RETURN(-ENOMEM);
 907         }
 908
 909         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 910         if (err)
 911                 GOTO(out_free, err);
 912
 913         /* Generate a string unique to this super, in case some joker tries
 914            to mount the same fs at two mount points.
 915            Use the address of the super itself.*/
 916         sprintf(ll_instance, "%p", sb);
 917         cfg.cfg_instance = ll_instance;
 918         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 919
 920         /* set up client obds */
 921         err = lustre_process_log(sb, profilenm, &cfg);
 922         if (err < 0) {
 923                 CERROR("Unable to process log: %d\n", err);
 924                 GOTO(out_free, err);
 925         }
 926
 927         lprof = class_get_profile(profilenm);
 928         if (lprof == NULL) {
 929                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 930                                    " read from the MGS.  Does that filesystem "
 931                                    "exist?\n", profilenm);
 932                 GOTO(out_free, err = -EINVAL);
 933         }
 934         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 935                lprof->lp_md, lprof->lp_dt);
 936
 937         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 938                   strlen(ll_instance) + 2);
 939         if (!dt)
 940                 GOTO(out_free, err = -ENOMEM);
 941         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 942
 943         OBD_ALLOC(md, strlen(lprof->lp_md) +
 944                   strlen(ll_instance) + 2);
 945         if (!md)
 946                 GOTO(out_free, err = -ENOMEM);
 947         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 948
 949         /* connections, registrations, sb setup */
 950         err = client_common_fill_super(sb, md, dt);
 951
 952 out_free:
 953         if (md)
 954                 OBD_FREE(md, strlen(md) + 1);
 955         if (dt)
 956                 OBD_FREE(dt, strlen(dt) + 1);
 957         if (err)
 958                 ll_put_super(sb);
 959         else
 960                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 961
 962         RETURN(err);
 963 } /* ll_fill_super */
 964
 965
 966 void lu_context_keys_dump(void);
 967
 968 void ll_put_super(struct super_block *sb)
 969 {
 970         struct config_llog_instance cfg;
 971         char   ll_instance[sizeof(sb) * 2 + 1];
 972         struct obd_device *obd;
 973         struct lustre_sb_info *lsi = s2lsi(sb);
 974         struct ll_sb_info *sbi = ll_s2sbi(sb);
 975         char *profilenm = get_profile_name(sb);
 976         int force = 1, next;
 977         ENTRY;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 980
 981         ll_print_capa_stat(sbi);
 982
 983         sprintf(ll_instance, "%p", sb);
 984         cfg.cfg_instance = ll_instance;
 985         lustre_end_log(sb, NULL, &cfg);
 986
 987         if (sbi->ll_md_exp) {
 988                 obd = class_exp2obd(sbi->ll_md_exp);
 989                 if (obd)
 990                         force = obd->obd_force;
 991         }
 992
 993         /* We need to set force before the lov_disconnect in
 994            lustre_common_put_super, since l_d cleans up osc's as well. */
 995         if (force) {
 996                 next = 0;
 997                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 998                                                      &next)) != NULL) {
 999                         obd->obd_force = force;
1000                 }
1001         }
1002
1003         if (sbi->ll_lcq) {
1004                 /* Only if client_common_fill_super succeeded */
1005                 client_common_put_super(sb);
1006         }
1007
1008         next = 0;
1009         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1010                 class_manual_cleanup(obd);
1011         }
1012
1013         if (profilenm)
1014                 class_del_profile(profilenm);
1015
1016         ll_free_sbi(sb);
1017         lsi->lsi_llsbi = NULL;
1018
1019         lustre_common_put_super(sb);
1020
1021         cl_env_cache_purge(~0);
1022
1023         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1024
1025         cfs_module_put(THIS_MODULE);
1026
1027         EXIT;
1028 } /* client_put_super */
1029
1030 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1031 {
1032         struct inode *inode = NULL;
1033         /* NOTE: we depend on atomic igrab() -bzzz */
1034         lock_res_and_lock(lock);
1035         if (lock->l_ast_data) {
1036                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1037                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1038                         inode = igrab(lock->l_ast_data);
1039                 } else {
1040                         inode = lock->l_ast_data;
1041                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1042                                                 D_INFO : D_WARNING,
1043                                         lock, __FILE__, __func__, __LINE__,
1044                                         "l_ast_data %p is bogus: magic %08x",
1045                                         lock->l_ast_data, lli->lli_inode_magic);
1046                         inode = NULL;
1047                 }
1048         }
1049         unlock_res_and_lock(lock);
1050         return inode;
1051 }
1052
1053 static int null_if_equal(struct ldlm_lock *lock, void *data)
1054 {
1055         if (data == lock->l_ast_data) {
1056                 lock->l_ast_data = NULL;
1057
1058                 if (lock->l_req_mode != lock->l_granted_mode)
1059                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1060         }
1061
1062         return LDLM_ITER_CONTINUE;
1063 }
1064
1065 void ll_clear_inode(struct inode *inode)
1066 {
1067         struct ll_inode_info *lli = ll_i2info(inode);
1068         struct ll_sb_info *sbi = ll_i2sbi(inode);
1069         ENTRY;
1070
1071         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1072                inode->i_generation, inode);
1073
1074         if (S_ISDIR(inode->i_mode)) {
1075                 /* these should have been cleared in ll_file_release */
1076                 LASSERT(lli->lli_sai == NULL);
1077                 LASSERT(lli->lli_opendir_key == NULL);
1078                 LASSERT(lli->lli_opendir_pid == 0);
1079         }
1080
1081         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1082         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1083                          null_if_equal, inode);
1084
1085         LASSERT(!lli->lli_open_fd_write_count);
1086         LASSERT(!lli->lli_open_fd_read_count);
1087         LASSERT(!lli->lli_open_fd_exec_count);
1088
1089         if (lli->lli_mds_write_och)
1090                 ll_md_real_close(inode, FMODE_WRITE);
1091         if (lli->lli_mds_exec_och)
1092                 ll_md_real_close(inode, FMODE_EXEC);
1093         if (lli->lli_mds_read_och)
1094                 ll_md_real_close(inode, FMODE_READ);
1095
1096         if (lli->lli_symlink_name) {
1097                 OBD_FREE(lli->lli_symlink_name,
1098                          strlen(lli->lli_symlink_name) + 1);
1099                 lli->lli_symlink_name = NULL;
1100         }
1101
1102         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1103                 LASSERT(lli->lli_posix_acl == NULL);
1104                 if (lli->lli_remote_perms) {
1105                         free_rmtperm_hash(lli->lli_remote_perms);
1106                         lli->lli_remote_perms = NULL;
1107                 }
1108         }
1109 #ifdef CONFIG_FS_POSIX_ACL
1110         else if (lli->lli_posix_acl) {
1111                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1112                 LASSERT(lli->lli_remote_perms == NULL);
1113                 posix_acl_release(lli->lli_posix_acl);
1114                 lli->lli_posix_acl = NULL;
1115         }
1116 #endif
1117         lli->lli_inode_magic = LLI_INODE_DEAD;
1118
1119 #ifdef HAVE_EXPORT___IGET
1120         cfs_spin_lock(&sbi->ll_deathrow_lock);
1121         cfs_list_del_init(&lli->lli_dead_list);
1122         cfs_spin_unlock(&sbi->ll_deathrow_lock);
1123 #endif
1124         ll_clear_inode_capas(inode);
1125         /*
1126          * XXX This has to be done before lsm is freed below, because
1127          * cl_object still uses inode lsm.
1128          */
1129         cl_inode_fini(inode);
1130
1131         if (lli->lli_smd) {
1132                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1133                 lli->lli_smd = NULL;
1134         }
1135
1136
1137         EXIT;
1138 }
1139
1140 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1141                   struct md_open_data **mod)
1142 {
1143         struct lustre_md md;
1144         struct ll_sb_info *sbi = ll_i2sbi(inode);
1145         struct ptlrpc_request *request = NULL;
1146         int rc;
1147         ENTRY;
1148
1149         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1150                                      LUSTRE_OPC_ANY, NULL);
1151         if (IS_ERR(op_data))
1152                 RETURN(PTR_ERR(op_data));
1153
1154         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1155                         &request, mod);
1156         if (rc) {
1157                 ptlrpc_req_finished(request);
1158                 if (rc == -ENOENT) {
1159                         inode->i_nlink = 0;
1160                         /* Unlinked special device node? Or just a race?
1161                          * Pretend we done everything. */
1162                         if (!S_ISREG(inode->i_mode) &&
1163                             !S_ISDIR(inode->i_mode))
1164                                 rc = inode_setattr(inode, &op_data->op_attr);
1165                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1166                         CERROR("md_setattr fails: rc = %d\n", rc);
1167                 }
1168                 RETURN(rc);
1169         }
1170
1171         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1172                               sbi->ll_md_exp, &md);
1173         if (rc) {
1174                 ptlrpc_req_finished(request);
1175                 RETURN(rc);
1176         }
1177
1178         /* We call inode_setattr to adjust timestamps.
1179          * If there is at least some data in file, we cleared ATTR_SIZE
1180          * above to avoid invoking vmtruncate, otherwise it is important
1181          * to call vmtruncate in inode_setattr to update inode->i_size
1182          * (bug 6196) */
1183         rc = inode_setattr(inode, &op_data->op_attr);
1184
1185         /* Extract epoch data if obtained. */
1186         op_data->op_handle = md.body->handle;
1187         op_data->op_ioepoch = md.body->ioepoch;
1188
1189         ll_update_inode(inode, &md);
1190         ptlrpc_req_finished(request);
1191
1192         RETURN(rc);
1193 }
1194
1195 /* Close IO epoch and send Size-on-MDS attribute update. */
1196 static int ll_setattr_done_writing(struct inode *inode,
1197                                    struct md_op_data *op_data,
1198                                    struct md_open_data *mod)
1199 {
1200         struct ll_inode_info *lli = ll_i2info(inode);
1201         int rc = 0;
1202         ENTRY;
1203
1204         LASSERT(op_data != NULL);
1205         if (!S_ISREG(inode->i_mode))
1206                 RETURN(0);
1207
1208         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1209                op_data->op_ioepoch, PFID(&lli->lli_fid));
1210
1211         op_data->op_flags = MF_EPOCH_CLOSE;
1212         ll_done_writing_attr(inode, op_data);
1213         ll_pack_inode2opdata(inode, op_data, NULL);
1214
1215         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1216         if (rc == -EAGAIN) {
1217                 /* MDS has instructed us to obtain Size-on-MDS attribute
1218                  * from OSTs and send setattr to back to MDS. */
1219                 rc = ll_som_update(inode, op_data);
1220         } else if (rc) {
1221                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1222                        inode->i_ino, rc);
1223         }
1224         RETURN(rc);
1225 }
1226
1227 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1228 {
1229         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1230         int rc;
1231
1232         rc = cl_setattr_do_truncate(inode, size, capa);
1233         ll_truncate_free_capa(capa);
1234         return rc;
1235 }
1236
1237 static int ll_setattr_ost(struct inode *inode)
1238 {
1239         struct obd_capa *capa = ll_mdscapa_get(inode);
1240         int rc;
1241
1242         rc = cl_setattr_ost(inode, capa);
1243         capa_put(capa);
1244
1245         return rc;
1246 }
1247
1248 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1249  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1250  * keep these values until such a time that objects are allocated for it.
1251  * We do the MDS operations first, as it is checking permissions for us.
1252  * We don't to the MDS RPC if there is nothing that we want to store there,
1253  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1254  * going to do an RPC anyways.
1255  *
1256  * If we are doing a truncate, we will send the mtime and ctime updates
1257  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1258  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1259  * at the same time.
1260  */
1261 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1262 {
1263         struct ll_inode_info *lli = ll_i2info(inode);
1264         struct lov_stripe_md *lsm = lli->lli_smd;
1265         struct md_op_data *op_data = NULL;
1266         struct md_open_data *mod = NULL;
1267         int ia_valid = attr->ia_valid;
1268         int rc = 0, rc1 = 0;
1269         ENTRY;
1270
1271         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1272                attr->ia_valid);
1273         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1274
1275         if (ia_valid & ATTR_SIZE) {
1276                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1277                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1278                                attr->ia_size, ll_file_maxbytes(inode));
1279                         RETURN(-EFBIG);
1280                 }
1281
1282                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1283         }
1284
1285         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1286         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1287                 if (cfs_curproc_fsuid() != inode->i_uid &&
1288                     !cfs_capable(CFS_CAP_FOWNER))
1289                         RETURN(-EPERM);
1290         }
1291
1292         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1293         if (attr->ia_valid & ATTR_CTIME) {
1294                 attr->ia_ctime = CFS_CURRENT_TIME;
1295                 attr->ia_valid |= ATTR_CTIME_SET;
1296         }
1297         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1298                 attr->ia_atime = CFS_CURRENT_TIME;
1299                 attr->ia_valid |= ATTR_ATIME_SET;
1300         }
1301         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1302                 attr->ia_mtime = CFS_CURRENT_TIME;
1303                 attr->ia_valid |= ATTR_MTIME_SET;
1304         }
1305         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1306                 /* To avoid stale mtime on mds, obtain it from ost and send
1307                    to mds. */
1308                 rc = cl_glimpse_size(inode);
1309                 if (rc)
1310                         RETURN(rc);
1311
1312                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1313                 attr->ia_mtime = inode->i_mtime;
1314         }
1315
1316         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1317                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1318                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1319                        cfs_time_current_sec());
1320
1321         /* NB: ATTR_SIZE will only be set after this point if the size
1322          * resides on the MDS, ie, this file has no objects. */
1323         if (lsm)
1324                 attr->ia_valid &= ~ATTR_SIZE;
1325
1326         /* We always do an MDS RPC, even if we're only changing the size;
1327          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1328
1329         OBD_ALLOC_PTR(op_data);
1330         if (op_data == NULL)
1331                 RETURN(-ENOMEM);
1332
1333         UNLOCK_INODE_MUTEX(inode);
1334         if (ia_valid & ATTR_SIZE)
1335                 UP_WRITE_I_ALLOC_SEM(inode);
1336         cfs_down(&lli->lli_trunc_sem);
1337         LOCK_INODE_MUTEX(inode);
1338         if (ia_valid & ATTR_SIZE)
1339                 DOWN_WRITE_I_ALLOC_SEM(inode);
1340
1341         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1342
1343         /* Open epoch for truncate. */
1344         if (exp_connect_som(ll_i2mdexp(inode)) && (ia_valid & ATTR_SIZE))
1345                 op_data->op_flags = MF_EPOCH_OPEN;
1346
1347         rc = ll_md_setattr(inode, op_data, &mod);
1348         if (rc)
1349                 GOTO(out, rc);
1350
1351         ll_ioepoch_open(lli, op_data->op_ioepoch);
1352         if (!lsm || !S_ISREG(inode->i_mode)) {
1353                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1354                 GOTO(out, rc = 0);
1355         }
1356
1357         if (ia_valid & ATTR_SIZE)
1358                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1359         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1360                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1361                        inode->i_ino, LTIME_S(attr->ia_mtime));
1362                 rc = ll_setattr_ost(inode);
1363         }
1364         EXIT;
1365 out:
1366         if (op_data) {
1367                 if (op_data->op_ioepoch)
1368                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1369                 ll_finish_md_op_data(op_data);
1370         }
1371         cfs_up(&lli->lli_trunc_sem);
1372         return rc ? rc : rc1;
1373 }
1374
1375 int ll_setattr(struct dentry *de, struct iattr *attr)
1376 {
1377         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1378             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1379                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1380
1381         if ((de->d_inode->i_mode & S_ISUID) &&
1382             !(attr->ia_mode & S_ISUID) &&
1383             !(attr->ia_valid & ATTR_KILL_SUID))
1384                 attr->ia_valid |= ATTR_KILL_SUID;
1385
1386         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1387             !(attr->ia_mode & S_ISGID) &&
1388             !(attr->ia_valid & ATTR_KILL_SGID))
1389                 attr->ia_valid |= ATTR_KILL_SGID;
1390
1391         return ll_setattr_raw(de->d_inode, attr);
1392 }
1393
1394 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1395                        __u64 max_age, __u32 flags)
1396 {
1397         struct ll_sb_info *sbi = ll_s2sbi(sb);
1398         struct obd_statfs obd_osfs;
1399         int rc;
1400         ENTRY;
1401
1402         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1403         if (rc) {
1404                 CERROR("md_statfs fails: rc = %d\n", rc);
1405                 RETURN(rc);
1406         }
1407
1408         osfs->os_type = sb->s_magic;
1409
1410         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1411                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1412
1413         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1414                 flags |= OBD_STATFS_NODELAY;
1415
1416         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1417                               &obd_osfs, max_age, flags);
1418         if (rc) {
1419                 CERROR("obd_statfs fails: rc = %d\n", rc);
1420                 RETURN(rc);
1421         }
1422
1423         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1424                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1425                obd_osfs.os_files);
1426
1427         osfs->os_bsize = obd_osfs.os_bsize;
1428         osfs->os_blocks = obd_osfs.os_blocks;
1429         osfs->os_bfree = obd_osfs.os_bfree;
1430         osfs->os_bavail = obd_osfs.os_bavail;
1431
1432         /* If we don't have as many objects free on the OST as inodes
1433          * on the MDS, we reduce the total number of inodes to
1434          * compensate, so that the "inodes in use" number is correct.
1435          */
1436         if (obd_osfs.os_ffree < osfs->os_ffree) {
1437                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1438                         obd_osfs.os_ffree;
1439                 osfs->os_ffree = obd_osfs.os_ffree;
1440         }
1441
1442         RETURN(rc);
1443 }
1444 #ifndef HAVE_STATFS_DENTRY_PARAM
1445 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1446 {
1447 #else
1448 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1449 {
1450         struct super_block *sb = de->d_sb;
1451 #endif
1452         struct obd_statfs osfs;
1453         int rc;
1454
1455         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1456         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1457
1458         /* For now we will always get up-to-date statfs values, but in the
1459          * future we may allow some amount of caching on the client (e.g.
1460          * from QOS or lprocfs updates). */
1461         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1462         if (rc)
1463                 return rc;
1464
1465         statfs_unpack(sfs, &osfs);
1466
1467         /* We need to downshift for all 32-bit kernels, because we can't
1468          * tell if the kernel is being called via sys_statfs64() or not.
1469          * Stop before overflowing f_bsize - in which case it is better
1470          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1471         if (sizeof(long) < 8) {
1472                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1473                         sfs->f_bsize <<= 1;
1474
1475                         osfs.os_blocks >>= 1;
1476                         osfs.os_bfree >>= 1;
1477                         osfs.os_bavail >>= 1;
1478                 }
1479         }
1480
1481         sfs->f_blocks = osfs.os_blocks;
1482         sfs->f_bfree = osfs.os_bfree;
1483         sfs->f_bavail = osfs.os_bavail;
1484
1485         return 0;
1486 }
1487
1488 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1489 {
1490         struct ll_inode_info *lli;
1491         struct lov_stripe_md *lsm;
1492
1493         lli = ll_i2info(inode);
1494         LASSERT(lli->lli_size_sem_owner != current);
1495         cfs_down(&lli->lli_size_sem);
1496         LASSERT(lli->lli_size_sem_owner == NULL);
1497         lli->lli_size_sem_owner = current;
1498         lsm = lli->lli_smd;
1499         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1500                  lsm, lock_lsm);
1501         if (lock_lsm)
1502                 lov_stripe_lock(lsm);
1503 }
1504
1505 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1506 {
1507         struct ll_inode_info *lli;
1508         struct lov_stripe_md *lsm;
1509
1510         lli = ll_i2info(inode);
1511         lsm = lli->lli_smd;
1512         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1513                  lsm, unlock_lsm);
1514         if (unlock_lsm)
1515                 lov_stripe_unlock(lsm);
1516         LASSERT(lli->lli_size_sem_owner == current);
1517         lli->lli_size_sem_owner = NULL;
1518         cfs_up(&lli->lli_size_sem);
1519 }
1520
1521 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1522 {
1523         struct ll_inode_info *lli = ll_i2info(inode);
1524         struct mdt_body *body = md->body;
1525         struct lov_stripe_md *lsm = md->lsm;
1526         struct ll_sb_info *sbi = ll_i2sbi(inode);
1527
1528         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1529         if (lsm != NULL) {
1530                 if (lli->lli_smd == NULL) {
1531                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1532                             lsm->lsm_magic != LOV_MAGIC_V3) {
1533                                 dump_lsm(D_ERROR, lsm);
1534                                 LBUG();
1535                         }
1536                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1537                                lsm, inode->i_ino, inode->i_generation, inode);
1538                         cl_inode_init(inode, md);
1539                         /* ll_inode_size_lock() requires it is only
1540                          * called with lli_smd != NULL or lock_lsm == 0
1541                          *  or we can race between lock/unlock.
1542                          *  bug 9547 */
1543                         lli->lli_smd = lsm;
1544                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1545                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1546                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1547                 } else {
1548                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1549                                 lli->lli_smd->lsm_stripe_count ==
1550                                 lsm->lsm_stripe_count);
1551                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1552                                 CERROR("lsm mismatch for inode %ld\n",
1553                                        inode->i_ino);
1554                                 CERROR("lli_smd:\n");
1555                                 dump_lsm(D_ERROR, lli->lli_smd);
1556                                 CERROR("lsm:\n");
1557                                 dump_lsm(D_ERROR, lsm);
1558                                 LBUG();
1559                         }
1560                 }
1561                 if (lli->lli_smd != lsm)
1562                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1563         }
1564
1565         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1566                 if (body->valid & OBD_MD_FLRMTPERM)
1567                         ll_update_remote_perm(inode, md->remote_perm);
1568         }
1569 #ifdef CONFIG_FS_POSIX_ACL
1570         else if (body->valid & OBD_MD_FLACL) {
1571                 cfs_spin_lock(&lli->lli_lock);
1572                 if (lli->lli_posix_acl)
1573                         posix_acl_release(lli->lli_posix_acl);
1574                 lli->lli_posix_acl = md->posix_acl;
1575                 cfs_spin_unlock(&lli->lli_lock);
1576         }
1577 #endif
1578         inode->i_ino = cl_fid_build_ino(&body->fid1);
1579         inode->i_generation = cl_fid_build_gen(&body->fid1);
1580
1581         if (body->valid & OBD_MD_FLATIME &&
1582             body->atime > LTIME_S(inode->i_atime))
1583                 LTIME_S(inode->i_atime) = body->atime;
1584
1585         /* mtime is always updated with ctime, but can be set in past.
1586            As write and utime(2) may happen within 1 second, and utime's
1587            mtime has a priority over write's one, so take mtime from mds
1588            for the same ctimes. */
1589         if (body->valid & OBD_MD_FLCTIME &&
1590             body->ctime >= LTIME_S(inode->i_ctime)) {
1591                 LTIME_S(inode->i_ctime) = body->ctime;
1592                 if (body->valid & OBD_MD_FLMTIME) {
1593                         CDEBUG(D_INODE, "setting ino %lu mtime "
1594                                "from %lu to "LPU64"\n", inode->i_ino,
1595                                LTIME_S(inode->i_mtime), body->mtime);
1596                         LTIME_S(inode->i_mtime) = body->mtime;
1597                 }
1598         }
1599         if (body->valid & OBD_MD_FLMODE)
1600                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1601         if (body->valid & OBD_MD_FLTYPE)
1602                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1603         LASSERT(inode->i_mode != 0);
1604         if (S_ISREG(inode->i_mode)) {
1605                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1606         } else {
1607                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1608         }
1609 #ifdef HAVE_INODE_BLKSIZE
1610         inode->i_blksize = 1<<inode->i_blkbits;
1611 #endif
1612         if (body->valid & OBD_MD_FLUID)
1613                 inode->i_uid = body->uid;
1614         if (body->valid & OBD_MD_FLGID)
1615                 inode->i_gid = body->gid;
1616         if (body->valid & OBD_MD_FLFLAGS)
1617                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1618         if (body->valid & OBD_MD_FLNLINK)
1619                 inode->i_nlink = body->nlink;
1620         if (body->valid & OBD_MD_FLRDEV)
1621                 inode->i_rdev = old_decode_dev(body->rdev);
1622
1623         if (body->valid & OBD_MD_FLID) {
1624                 /* FID shouldn't be changed! */
1625                 if (fid_is_sane(&lli->lli_fid)) {
1626                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1627                                  "Trying to change FID "DFID
1628                                  " to the "DFID", inode %lu/%u(%p)\n",
1629                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1630                                  inode->i_ino, inode->i_generation, inode);
1631                 } else
1632                         lli->lli_fid = body->fid1;
1633         }
1634
1635         LASSERT(fid_seq(&lli->lli_fid) != 0);
1636
1637         if (body->valid & OBD_MD_FLSIZE) {
1638                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1639                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1640                         struct lustre_handle lockh;
1641                         ldlm_mode_t mode;
1642
1643                         /* As it is possible a blocking ast has been processed
1644                          * by this time, we need to check there is an UPDATE
1645                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1646                          * it. */
1647                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1648                                                &lockh);
1649                         if (mode) {
1650                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1651                                                       LLIF_EPOCH_PENDING |
1652                                                       LLIF_SOM_DIRTY)) {
1653                                         CERROR("ino %lu flags %lu still has "
1654                                                "size authority! do not trust "
1655                                                "the size got from MDS\n",
1656                                                inode->i_ino, lli->lli_flags);
1657                                 } else {
1658                                         /* Use old size assignment to avoid
1659                                          * deadlock bz14138 & bz14326 */
1660                                         inode->i_size = body->size;
1661                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1662                                 }
1663                                 ldlm_lock_decref(&lockh, mode);
1664                         }
1665                 } else {
1666                         /* Use old size assignment to avoid
1667                          * deadlock bz14138 & bz14326 */
1668                         inode->i_size = 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 static struct backing_dev_info ll_backing_dev_info = {
1686         .ra_pages       = 0,    /* No readahead */
1687 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1688         .capabilities   = 0,    /* Does contribute to dirty memory */
1689 #else
1690         .memory_backed  = 0,    /* Does contribute to dirty memory */
1691 #endif
1692 };
1693
1694 void ll_read_inode2(struct inode *inode, void *opaque)
1695 {
1696         struct lustre_md *md = opaque;
1697         struct ll_inode_info *lli = ll_i2info(inode);
1698         ENTRY;
1699
1700         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1701                inode->i_ino, inode->i_generation, inode);
1702
1703         ll_lli_init(lli);
1704
1705         LASSERT(!lli->lli_smd);
1706
1707         /* Core attributes from the MDS first.  This is a new inode, and
1708          * the VFS doesn't zero times in the core inode so we have to do
1709          * it ourselves.  They will be overwritten by either MDS or OST
1710          * attributes - we just need to make sure they aren't newer. */
1711         LTIME_S(inode->i_mtime) = 0;
1712         LTIME_S(inode->i_atime) = 0;
1713         LTIME_S(inode->i_ctime) = 0;
1714         inode->i_rdev = 0;
1715         ll_update_inode(inode, md);
1716
1717         /* OIDEBUG(inode); */
1718
1719         if (S_ISREG(inode->i_mode)) {
1720                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1721                 inode->i_op = &ll_file_inode_operations;
1722                 inode->i_fop = sbi->ll_fop;
1723                 inode->i_mapping->a_ops = &ll_aops;
1724                 EXIT;
1725         } else if (S_ISDIR(inode->i_mode)) {
1726                 inode->i_op = &ll_dir_inode_operations;
1727                 inode->i_fop = &ll_dir_operations;
1728                 inode->i_mapping->a_ops = &ll_dir_aops;
1729                 EXIT;
1730         } else if (S_ISLNK(inode->i_mode)) {
1731                 inode->i_op = &ll_fast_symlink_inode_operations;
1732                 EXIT;
1733         } else {
1734                 inode->i_op = &ll_special_inode_operations;
1735
1736                 init_special_inode(inode, inode->i_mode,
1737                                    kdev_t_to_nr(inode->i_rdev));
1738
1739                 /* initializing backing dev info. */
1740                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1741
1742                 EXIT;
1743         }
1744 }
1745
1746 void ll_delete_inode(struct inode *inode)
1747 {
1748         struct ll_sb_info *sbi = ll_i2sbi(inode);
1749         int rc;
1750         ENTRY;
1751
1752         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1753         if (rc) {
1754                 CERROR("fid_delete() failed, rc %d\n", rc);
1755         }
1756         truncate_inode_pages(&inode->i_data, 0);
1757         clear_inode(inode);
1758
1759         EXIT;
1760 }
1761
1762 int ll_iocontrol(struct inode *inode, struct file *file,
1763                  unsigned int cmd, unsigned long arg)
1764 {
1765         struct ll_sb_info *sbi = ll_i2sbi(inode);
1766         struct ptlrpc_request *req = NULL;
1767         int rc, flags = 0;
1768         ENTRY;
1769
1770         switch(cmd) {
1771         case FSFILT_IOC_GETFLAGS: {
1772                 struct mdt_body *body;
1773                 struct obd_capa *oc;
1774
1775                 oc = ll_mdscapa_get(inode);
1776                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1777                                 OBD_MD_FLFLAGS, 0, &req);
1778                 capa_put(oc);
1779                 if (rc) {
1780                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1781                         RETURN(-abs(rc));
1782                 }
1783
1784                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1785
1786                 flags = body->flags;
1787
1788                 ptlrpc_req_finished(req);
1789
1790                 RETURN(put_user(flags, (int *)arg));
1791         }
1792         case FSFILT_IOC_SETFLAGS: {
1793                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1794                 struct obd_info oinfo = { { { 0 } } };
1795                 struct md_op_data *op_data;
1796
1797                 if (get_user(flags, (int *)arg))
1798                         RETURN(-EFAULT);
1799
1800                 oinfo.oi_md = lsm;
1801                 OBDO_ALLOC(oinfo.oi_oa);
1802                 if (!oinfo.oi_oa)
1803                         RETURN(-ENOMEM);
1804
1805                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1806                                              LUSTRE_OPC_ANY, NULL);
1807                 if (IS_ERR(op_data))
1808                         RETURN(PTR_ERR(op_data));
1809
1810                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1811                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1812                 rc = md_setattr(sbi->ll_md_exp, op_data,
1813                                 NULL, 0, NULL, 0, &req, NULL);
1814                 ll_finish_md_op_data(op_data);
1815                 ptlrpc_req_finished(req);
1816                 if (rc) {
1817                         OBDO_FREE(oinfo.oi_oa);
1818                         RETURN(rc);
1819                 }
1820
1821                 if (lsm == NULL) {
1822                         OBDO_FREE(oinfo.oi_oa);
1823                         GOTO(update_cache, rc);
1824                 }
1825
1826                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1827                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1828                 oinfo.oi_oa->o_flags = flags;
1829                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1830                                        OBD_MD_FLGROUP;
1831                 oinfo.oi_capa = ll_mdscapa_get(inode);
1832
1833                 obdo_from_inode(oinfo.oi_oa, inode,
1834                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1835                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1836                 capa_put(oinfo.oi_capa);
1837                 OBDO_FREE(oinfo.oi_oa);
1838                 if (rc) {
1839                         if (rc != -EPERM && rc != -EACCES)
1840                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1841                         RETURN(rc);
1842                 }
1843
1844                 EXIT;
1845 update_cache:
1846                 inode->i_flags = ll_ext_to_inode_flags(flags |
1847                                                        MDS_BFLAG_EXT_FLAGS);
1848                 return 0;
1849         }
1850         default:
1851                 RETURN(-ENOSYS);
1852         }
1853
1854         RETURN(0);
1855 }
1856
1857 int ll_flush_ctx(struct inode *inode)
1858 {
1859         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1860
1861         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1862
1863         obd_set_info_async(sbi->ll_md_exp,
1864                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1865                            0, NULL, NULL);
1866         obd_set_info_async(sbi->ll_dt_exp,
1867                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1868                            0, NULL, NULL);
1869         return 0;
1870 }
1871
1872 /* umount -f client means force down, don't save state */
1873 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1874 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1875 {
1876         struct super_block *sb = vfsmnt->mnt_sb;
1877 #else
1878 void ll_umount_begin(struct super_block *sb)
1879 {
1880 #endif
1881         struct lustre_sb_info *lsi = s2lsi(sb);
1882         struct ll_sb_info *sbi = ll_s2sbi(sb);
1883         struct obd_device *obd;
1884         struct obd_ioctl_data *ioc_data;
1885         ENTRY;
1886
1887 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1888         if (!(flags & MNT_FORCE)) {
1889                 EXIT;
1890                 return;
1891         }
1892 #endif
1893
1894         /* Tell the MGC we got umount -f */
1895         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1896
1897         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1898                sb->s_count, atomic_read(&sb->s_active));
1899
1900         obd = class_exp2obd(sbi->ll_md_exp);
1901         if (obd == NULL) {
1902                 CERROR("Invalid MDC connection handle "LPX64"\n",
1903                        sbi->ll_md_exp->exp_handle.h_cookie);
1904                 EXIT;
1905                 return;
1906         }
1907         obd->obd_force = 1;
1908
1909         obd = class_exp2obd(sbi->ll_dt_exp);
1910         if (obd == NULL) {
1911                 CERROR("Invalid LOV connection handle "LPX64"\n",
1912                        sbi->ll_dt_exp->exp_handle.h_cookie);
1913                 EXIT;
1914                 return;
1915         }
1916         obd->obd_force = 1;
1917
1918         OBD_ALLOC_PTR(ioc_data);
1919         if (ioc_data) {
1920                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1921                               sizeof ioc_data, ioc_data, NULL);
1922
1923                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1924                               sizeof ioc_data, ioc_data, NULL);
1925
1926                 OBD_FREE_PTR(ioc_data);
1927         }
1928
1929
1930         /* Really, we'd like to wait until there are no requests outstanding,
1931          * and then continue.  For now, we just invalidate the requests,
1932          * schedule() and sleep one second if needed, and hope.
1933          */
1934         cfs_schedule();
1935 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1936         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1937                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1938                                                    cfs_time_seconds(1));
1939                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1940                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1941                                       "may try to umount it a bit later\n",
1942                                       atomic_read(&vfsmnt->mnt_count));
1943         }
1944 #endif
1945
1946         EXIT;
1947 }
1948
1949 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1950 {
1951         struct ll_sb_info *sbi = ll_s2sbi(sb);
1952         int err;
1953         __u32 read_only;
1954
1955         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1956                 read_only = *flags & MS_RDONLY;
1957                 err = obd_set_info_async(sbi->ll_md_exp,
1958                                          sizeof(KEY_READ_ONLY),
1959                                          KEY_READ_ONLY, sizeof(read_only),
1960                                          &read_only, NULL);
1961                 if (err) {
1962                         CERROR("Failed to change the read-only flag during "
1963                                "remount: %d\n", err);
1964                         return err;
1965                 }
1966
1967                 if (read_only)
1968                         sb->s_flags |= MS_RDONLY;
1969                 else
1970                         sb->s_flags &= ~MS_RDONLY;
1971         }
1972         return 0;
1973 }
1974
1975 int ll_prep_inode(struct inode **inode,
1976                   struct ptlrpc_request *req,
1977                   struct super_block *sb)
1978 {
1979         struct ll_sb_info *sbi = NULL;
1980         struct lustre_md md;
1981         int rc;
1982         ENTRY;
1983
1984         LASSERT(*inode || sb);
1985         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1986         prune_deathrow(sbi, 1);
1987         memset(&md, 0, sizeof(struct lustre_md));
1988
1989         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1990                               sbi->ll_md_exp, &md);
1991         if (rc)
1992                 RETURN(rc);
1993
1994         if (*inode) {
1995                 ll_update_inode(*inode, &md);
1996         } else {
1997                 LASSERT(sb != NULL);
1998
1999                 /*
2000                  * At this point server returns to client's same fid as client
2001                  * generated for creating. So using ->fid1 is okay here.
2002                  */
2003                 LASSERT(fid_is_sane(&md.body->fid1));
2004
2005                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
2006                 if (*inode == NULL || IS_ERR(*inode)) {
2007                         if (md.lsm)
2008                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2009 #ifdef CONFIG_FS_POSIX_ACL
2010                         if (md.posix_acl) {
2011                                 posix_acl_release(md.posix_acl);
2012                                 md.posix_acl = NULL;
2013                         }
2014 #endif
2015                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2016                         *inode = NULL;
2017                         CERROR("new_inode -fatal: rc %d\n", rc);
2018                         GOTO(out, rc);
2019                 }
2020         }
2021
2022 out:
2023         md_free_lustre_md(sbi->ll_md_exp, &md);
2024         RETURN(rc);
2025 }
2026
2027 int ll_obd_statfs(struct inode *inode, void *arg)
2028 {
2029         struct ll_sb_info *sbi = NULL;
2030         struct obd_export *exp;
2031         char *buf = NULL;
2032         struct obd_ioctl_data *data = NULL;
2033         __u32 type;
2034         int len = 0, rc;
2035
2036         if (!inode || !(sbi = ll_i2sbi(inode)))
2037                 GOTO(out_statfs, rc = -EINVAL);
2038
2039         rc = obd_ioctl_getdata(&buf, &len, arg);
2040         if (rc)
2041                 GOTO(out_statfs, rc);
2042
2043         data = (void*)buf;
2044         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2045             !data->ioc_pbuf1 || !data->ioc_pbuf2)
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         return op_data;
2136 }
2137
2138 void ll_finish_md_op_data(struct md_op_data *op_data)
2139 {
2140         capa_put(op_data->op_capa1);
2141         capa_put(op_data->op_capa2);
2142         OBD_FREE_PTR(op_data);
2143 }
2144
2145 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2146 {
2147         struct ll_sb_info *sbi;
2148
2149         LASSERT((seq != NULL) && (vfs != NULL));
2150         sbi = ll_s2sbi(vfs->mnt_sb);
2151
2152         if (sbi->ll_flags & LL_SBI_NOLCK)
2153                 seq_puts(seq, ",nolock");
2154
2155         if (sbi->ll_flags & LL_SBI_FLOCK)
2156                 seq_puts(seq, ",flock");
2157
2158         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2159                 seq_puts(seq, ",localflock");
2160
2161         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2162                 seq_puts(seq, ",user_xattr");
2163
2164         if (sbi->ll_flags & LL_SBI_ACL)
2165                 seq_puts(seq, ",acl");
2166
2167         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2168                 seq_puts(seq, ",lazystatfs");
2169
2170         RETURN(0);
2171 }