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 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = 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) 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         spin_lock_init(&sbi->ll_lock);
  86         spin_lock_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         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 = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         spin_lock(&ll_sb_lock);
 115         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_LLITE_CHECKSUM
 119         sbi->ll_flags |= LL_SBI_CHECKSUM;
 120 #endif
 121
 122 #ifdef HAVE_LRU_RESIZE_SUPPORT
 123         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 124 #endif
 125
 126 #ifdef HAVE_EXPORT___IGET
 127         INIT_LIST_HEAD(&sbi->ll_deathrow);
 128         spin_lock_init(&sbi->ll_deathrow_lock);
 129 #endif
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137
 138         RETURN(sbi);
 139 }
 140
 141 void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144         ENTRY;
 145
 146         if (sbi != NULL) {
 147                 spin_lock(&ll_sb_lock);
 148                 list_del(&sbi->ll_list);
 149                 spin_unlock(&ll_sb_lock);
 150                 OBD_FREE(sbi, sizeof(*sbi));
 151         }
 152         EXIT;
 153 }
 154
 155 static struct dentry_operations ll_d_root_ops = {
 156 #ifdef DCACHE_LUSTRE_INVALID
 157         .d_compare = ll_dcompare,
 158 #endif
 159 };
 160
 161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 162 {
 163         struct inode *root = 0;
 164         struct ll_sb_info *sbi = ll_s2sbi(sb);
 165         struct obd_device *obd;
 166         struct obd_capa *oc = NULL;
 167         struct obd_statfs osfs;
 168         struct ptlrpc_request *request = NULL;
 169         struct lustre_handle dt_conn = {0, };
 170         struct lustre_handle md_conn = {0, };
 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
 203 #ifdef HAVE_LRU_RESIZE_SUPPORT
 204         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 205                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 206 #endif
 207 #ifdef CONFIG_FS_POSIX_ACL
 208         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 209 #endif
 210         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 211         data->ocd_version = LUSTRE_VERSION_CODE;
 212
 213         if (sb->s_flags & MS_RDONLY)
 214                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 215         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 216                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 217
 218 #ifdef HAVE_MS_FLOCK_LOCK
 219         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 220         sb->s_flags |= MS_FLOCK_LOCK;
 221 #endif
 222
 223         if (sbi->ll_flags & LL_SBI_FLOCK)
 224                 sbi->ll_fop = &ll_file_operations_flock;
 225         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 226                 sbi->ll_fop = &ll_file_operations;
 227         else
 228                 sbi->ll_fop = &ll_file_operations_noflock;
 229
 230         /* real client */
 231         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 232         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 233                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 234
 235         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 236         if (err == -EBUSY) {
 237                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 238                                    "recovery, of which this client is not a "
 239                                    "part. Please wait for recovery to complete,"
 240                                    " abort, or time out.\n", md);
 241                 GOTO(out, err);
 242         } else if (err) {
 243                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 244                 GOTO(out, err);
 245         }
 246         sbi->ll_md_exp = class_conn2export(&md_conn);
 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() - 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 ineger 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_JOIN)
 309                 sbi->ll_flags |= LL_SBI_JOIN;
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 312                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 313                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 314                         LCONSOLE_INFO("client is set as remote by default.\n");
 315                 }
 316         } else {
 317                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 318                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 319                         LCONSOLE_INFO("client claims to be remote, but server "
 320                                       "rejected, forced to be local.\n");
 321                 }
 322         }
 323
 324         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 325                 LCONSOLE_INFO("client enabled MDS capability!\n");
 326                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 327         }
 328
 329         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 330                 LCONSOLE_INFO("client enabled OSS capability!\n");
 331                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 332         }
 333
 334         obd = class_name2obd(dt);
 335         if (!obd) {
 336                 CERROR("DT %s: not setup or attached\n", dt);
 337                 GOTO(out_md_fid, err = -ENODEV);
 338         }
 339
 340         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 341                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 342                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 343                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 344                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 345                                   OBD_CONNECT_OSS_CAPA;
 346
 347         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 348                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 349                  * disabled by default, because it can still be enabled on the
 350                  * fly via /proc. As a consequence, we still need to come to an
 351                  * agreement on the supported algorithms at connect time */
 352                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 353
 354                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 355                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 356                 else
 357                         /* send the list of supported checksum types */
 358                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 359         }
 360
 361 #ifdef HAVE_LRU_RESIZE_SUPPORT
 362         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 363 #endif
 364         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 365                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 366
 367         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 368                "ocd_grant: %d\n", data->ocd_connect_flags,
 369                data->ocd_version, data->ocd_grant);
 370
 371         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 372         obd->obd_upcall.onu_upcall = cl_ocd_update;
 373         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 374
 375         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 376         if (err == -EBUSY) {
 377                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 378                                    "recovery, of which this client is not a "
 379                                    "part.  Please wait for recovery to "
 380                                    "complete, abort, or time out.\n", dt);
 381                 GOTO(out_md_fid, err);
 382         } else if (err) {
 383                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 384                 GOTO(out_md_fid, err);
 385         }
 386
 387         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 388
 389         err = obd_fid_init(sbi->ll_dt_exp);
 390         if (err) {
 391                 CERROR("Can't init data layer FID infrastructure, "
 392                        "rc %d\n", err);
 393                 GOTO(out_dt, err);
 394         }
 395
 396         spin_lock(&sbi->ll_lco.lco_lock);
 397         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 398         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 399         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 400         spin_unlock(&sbi->ll_lco.lco_lock);
 401
 402         fid_zero(&sbi->ll_root_fid);
 403         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 404         if (err) {
 405                 CERROR("cannot mds_connect: rc = %d\n", err);
 406                 GOTO(out_lock_cn_cb, err);
 407         }
 408         if (!fid_is_sane(&sbi->ll_root_fid)) {
 409                 CERROR("Invalid root fid during mount\n");
 410                 GOTO(out_lock_cn_cb, err = -EINVAL);
 411         }
 412         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 413
 414         sb->s_op = &lustre_super_operations;
 415         sb->s_export_op = &lustre_export_operations;
 416
 417         /* make root inode
 418          * XXX: move this to after cbd setup? */
 419         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 420         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 421                 valid |= OBD_MD_FLRMTPERM;
 422         else if (sbi->ll_flags & LL_SBI_ACL)
 423                 valid |= OBD_MD_FLACL;
 424
 425         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 426                          &request);
 427         if (oc)
 428                 capa_put(oc);
 429         if (err) {
 430                 CERROR("md_getattr failed for root: rc = %d\n", err);
 431                 GOTO(out_lock_cn_cb, err);
 432         }
 433         memset(&lmd, 0, sizeof(lmd));
 434         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 435                                sbi->ll_md_exp, &lmd);
 436         if (err) {
 437                 CERROR("failed to understand root inode md: rc = %d\n", err);
 438                 ptlrpc_req_finished (request);
 439                 GOTO(out_lock_cn_cb, err);
 440         }
 441
 442         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 443         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 444         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 445         ptlrpc_req_finished(request);
 446
 447         if (root == NULL || IS_ERR(root)) {
 448                 if (lmd.lsm)
 449                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 450 #ifdef CONFIG_FS_POSIX_ACL
 451                 if (lmd.posix_acl) {
 452                         posix_acl_release(lmd.posix_acl);
 453                         lmd.posix_acl = NULL;
 454                 }
 455 #endif
 456                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 457                 root = NULL;
 458                 CERROR("lustre_lite: bad iget4 for root\n");
 459                 GOTO(out_root, err);
 460         }
 461
 462         err = ll_close_thread_start(&sbi->ll_lcq);
 463         if (err) {
 464                 CERROR("cannot start close thread: rc %d\n", err);
 465                 GOTO(out_root, err);
 466         }
 467
 468 #ifdef CONFIG_FS_POSIX_ACL
 469         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 470                 rct_init(&sbi->ll_rct);
 471                 et_init(&sbi->ll_et);
 472         }
 473 #endif
 474
 475         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 476         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 477                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 478                                  NULL);
 479         cl_sb_init(sb);
 480
 481         sb->s_root = d_alloc_root(root);
 482         if (data != NULL)
 483                 OBD_FREE(data, sizeof(*data));
 484
 485         sb->s_root->d_op = &ll_d_root_ops;
 486
 487         sbi->ll_sdev_orig = sb->s_dev;
 488
 489         /* We set sb->s_dev equal on all lustre clients in order to support
 490          * NFS export clustering.  NFSD requires that the FSID be the same
 491          * on all clients. */
 492         /* s_dev is also used in lt_compare() to compare two fs, but that is
 493          * only a node-local comparison. */
 494         uuid = obd_get_uuid(sbi->ll_md_exp);
 495         if (uuid != NULL)
 496                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 497
 498         RETURN(err);
 499 out_root:
 500         if (root)
 501                 iput(root);
 502 out_lock_cn_cb:
 503         obd_fid_fini(sbi->ll_dt_exp);
 504 out_dt:
 505         obd_disconnect(sbi->ll_dt_exp);
 506         sbi->ll_dt_exp = NULL;
 507 out_md_fid:
 508         obd_fid_fini(sbi->ll_md_exp);
 509 out_md:
 510         obd_disconnect(sbi->ll_md_exp);
 511         sbi->ll_md_exp = NULL;
 512 out:
 513         if (data != NULL)
 514                 OBD_FREE_PTR(data);
 515         lprocfs_unregister_mountpoint(sbi);
 516         return err;
 517 }
 518
 519 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 520 {
 521         int size, rc;
 522
 523         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 524         size = sizeof(int);
 525         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 526                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 527         if (rc)
 528                 CERROR("Get max mdsize error rc %d \n", rc);
 529
 530         RETURN(rc);
 531 }
 532
 533 void ll_dump_inode(struct inode *inode)
 534 {
 535         struct list_head *tmp;
 536         int dentry_count = 0;
 537
 538         LASSERT(inode != NULL);
 539
 540         list_for_each(tmp, &inode->i_dentry)
 541                 dentry_count++;
 542
 543         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 544                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 545                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 546 }
 547
 548 void lustre_dump_dentry(struct dentry *dentry, int recur)
 549 {
 550         struct list_head *tmp;
 551         int subdirs = 0;
 552
 553         LASSERT(dentry != NULL);
 554
 555         list_for_each(tmp, &dentry->d_subdirs)
 556                 subdirs++;
 557
 558         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 559                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 560                dentry->d_name.len, dentry->d_name.name,
 561                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 562                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 563                dentry->d_flags, dentry->d_fsdata, subdirs);
 564         if (dentry->d_inode != NULL)
 565                 ll_dump_inode(dentry->d_inode);
 566
 567         if (recur == 0)
 568                 return;
 569
 570         list_for_each(tmp, &dentry->d_subdirs) {
 571                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 572                 lustre_dump_dentry(d, recur - 1);
 573         }
 574 }
 575
 576 #ifdef HAVE_EXPORT___IGET
 577 static void prune_dir_dentries(struct inode *inode)
 578 {
 579         struct dentry *dentry, *prev = NULL;
 580
 581         /* due to lustre specific logic, a directory
 582          * can have few dentries - a bug from VFS POV */
 583 restart:
 584         spin_lock(&dcache_lock);
 585         if (!list_empty(&inode->i_dentry)) {
 586                 dentry = list_entry(inode->i_dentry.prev,
 587                                     struct dentry, d_alias);
 588                 /* in order to prevent infinite loops we
 589                  * break if previous dentry is busy */
 590                 if (dentry != prev) {
 591                         prev = dentry;
 592                         dget_locked(dentry);
 593                         spin_unlock(&dcache_lock);
 594
 595                         /* try to kill all child dentries */
 596                         lock_dentry(dentry);
 597                         shrink_dcache_parent(dentry);
 598                         unlock_dentry(dentry);
 599                         dput(dentry);
 600
 601                         /* now try to get rid of current dentry */
 602                         d_prune_aliases(inode);
 603                         goto restart;
 604                 }
 605         }
 606         spin_unlock(&dcache_lock);
 607 }
 608
 609 static void prune_deathrow_one(struct ll_inode_info *lli)
 610 {
 611         struct inode *inode = ll_info2i(lli);
 612
 613         /* first, try to drop any dentries - they hold a ref on the inode */
 614         if (S_ISDIR(inode->i_mode))
 615                 prune_dir_dentries(inode);
 616         else
 617                 d_prune_aliases(inode);
 618
 619
 620         /* if somebody still uses it, leave it */
 621         LASSERT(atomic_read(&inode->i_count) > 0);
 622         if (atomic_read(&inode->i_count) > 1)
 623                 goto out;
 624
 625         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 626                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 627
 628         /* seems nobody uses it anymore */
 629         inode->i_nlink = 0;
 630
 631 out:
 632         iput(inode);
 633         return;
 634 }
 635
 636 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 637 {
 638         struct ll_inode_info *lli;
 639         int empty;
 640
 641         do {
 642                 if (need_resched() && try)
 643                         break;
 644
 645                 if (try) {
 646                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 647                                 break;
 648                 } else {
 649                         spin_lock(&sbi->ll_deathrow_lock);
 650                 }
 651
 652                 empty = 1;
 653                 lli = NULL;
 654                 if (!list_empty(&sbi->ll_deathrow)) {
 655                         lli = list_entry(sbi->ll_deathrow.next,
 656                                          struct ll_inode_info,
 657                                          lli_dead_list);
 658                         list_del_init(&lli->lli_dead_list);
 659                         if (!list_empty(&sbi->ll_deathrow))
 660                                 empty = 0;
 661                 }
 662                 spin_unlock(&sbi->ll_deathrow_lock);
 663
 664                 if (lli)
 665                         prune_deathrow_one(lli);
 666
 667         } while (empty == 0);
 668 }
 669 #else /* !HAVE_EXPORT___IGET */
 670 #define prune_deathrow(sbi, try) do {} while (0)
 671 #endif /* HAVE_EXPORT___IGET */
 672
 673 void client_common_put_super(struct super_block *sb)
 674 {
 675         struct ll_sb_info *sbi = ll_s2sbi(sb);
 676         ENTRY;
 677
 678 #ifdef CONFIG_FS_POSIX_ACL
 679         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 680                 et_fini(&sbi->ll_et);
 681                 rct_fini(&sbi->ll_rct);
 682         }
 683 #endif
 684
 685         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 686
 687         ll_close_thread_shutdown(sbi->ll_lcq);
 688
 689         /* destroy inodes in deathrow */
 690         prune_deathrow(sbi, 0);
 691
 692         list_del(&sbi->ll_conn_chain);
 693
 694         obd_fid_fini(sbi->ll_dt_exp);
 695         obd_disconnect(sbi->ll_dt_exp);
 696         sbi->ll_dt_exp = NULL;
 697
 698         lprocfs_unregister_mountpoint(sbi);
 699
 700         obd_fid_fini(sbi->ll_md_exp);
 701         obd_disconnect(sbi->ll_md_exp);
 702         sbi->ll_md_exp = NULL;
 703
 704         EXIT;
 705 }
 706
 707 void ll_kill_super(struct super_block *sb)
 708 {
 709         struct ll_sb_info *sbi;
 710
 711         ENTRY;
 712
 713         /* not init sb ?*/
 714         if (!(sb->s_flags & MS_ACTIVE))
 715                 return;
 716
 717         sbi = ll_s2sbi(sb);
 718         /* we need restore s_dev from changed for clustred NFS before put_super
 719          * because new kernels have cached s_dev and change sb->s_dev in
 720          * put_super not affected real removing devices */
 721         if (sbi)
 722                 sb->s_dev = sbi->ll_sdev_orig;
 723         EXIT;
 724 }
 725
 726 char *ll_read_opt(const char *opt, char *data)
 727 {
 728         char *value;
 729         char *retval;
 730         ENTRY;
 731
 732         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 733         if (strncmp(opt, data, strlen(opt)))
 734                 RETURN(NULL);
 735         if ((value = strchr(data, '=')) == NULL)
 736                 RETURN(NULL);
 737
 738         value++;
 739         OBD_ALLOC(retval, strlen(value) + 1);
 740         if (!retval) {
 741                 CERROR("out of memory!\n");
 742                 RETURN(NULL);
 743         }
 744
 745         memcpy(retval, value, strlen(value)+1);
 746         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 747         RETURN(retval);
 748 }
 749
 750 static inline int ll_set_opt(const char *opt, char *data, int fl)
 751 {
 752         if (strncmp(opt, data, strlen(opt)) != 0)
 753                 return(0);
 754         else
 755                 return(fl);
 756 }
 757
 758 /* non-client-specific mount options are parsed in lmd_parse */
 759 static int ll_options(char *options, int *flags)
 760 {
 761         int tmp;
 762         char *s1 = options, *s2;
 763         ENTRY;
 764
 765         if (!options)
 766                 RETURN(0);
 767
 768         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 769
 770         while (*s1) {
 771                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 772                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 773                 if (tmp) {
 774                         *flags |= tmp;
 775                         goto next;
 776                 }
 777                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 778                 if (tmp) {
 779                         *flags |= tmp;
 780                         goto next;
 781                 }
 782                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 783                 if (tmp) {
 784                         *flags |= tmp;
 785                         goto next;
 786                 }
 787                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 788                 if (tmp) {
 789                         *flags &= ~tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 793                 if (tmp) {
 794                         *flags |= tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 798                 if (tmp) {
 799                         *flags &= ~tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 803                 if (tmp) {
 804                         /* Ignore deprecated mount option.  The client will
 805                          * always try to mount with ACL support, whether this
 806                          * is used depends on whether server supports it. */
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 810                 if (tmp) {
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818
 819                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 825                 if (tmp) {
 826                         *flags &= ~tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 835                 if (tmp) {
 836                         *flags &= ~tmp;
 837                         goto next;
 838                 }
 839
 840                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 841                                    s1);
 842                 RETURN(-EINVAL);
 843
 844 next:
 845                 /* Find next opt */
 846                 s2 = strchr(s1, ',');
 847                 if (s2 == NULL)
 848                         break;
 849                 s1 = s2 + 1;
 850         }
 851         RETURN(0);
 852 }
 853
 854 void ll_lli_init(struct ll_inode_info *lli)
 855 {
 856         lli->lli_inode_magic = LLI_INODE_MAGIC;
 857         sema_init(&lli->lli_size_sem, 1);
 858         sema_init(&lli->lli_write_sem, 1);
 859         lli->lli_flags = 0;
 860         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 861         spin_lock_init(&lli->lli_lock);
 862         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 863         INIT_LIST_HEAD(&lli->lli_close_list);
 864         lli->lli_inode_magic = LLI_INODE_MAGIC;
 865         sema_init(&lli->lli_och_sem, 1);
 866         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 867         lli->lli_mds_exec_och = NULL;
 868         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 869         lli->lli_open_fd_exec_count = 0;
 870         INIT_LIST_HEAD(&lli->lli_dead_list);
 871         lli->lli_remote_perms = NULL;
 872         lli->lli_rmtperm_utime = 0;
 873         sema_init(&lli->lli_rmtperm_sem, 1);
 874         INIT_LIST_HEAD(&lli->lli_oss_capas);
 875 }
 876
 877 int ll_fill_super(struct super_block *sb)
 878 {
 879         struct lustre_profile *lprof;
 880         struct lustre_sb_info *lsi = s2lsi(sb);
 881         struct ll_sb_info *sbi;
 882         char  *dt = NULL, *md = NULL;
 883         char  *profilenm = get_profile_name(sb);
 884         struct config_llog_instance cfg = {0, };
 885         char   ll_instance[sizeof(sb) * 2 + 1];
 886         int    err;
 887         ENTRY;
 888
 889         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 890
 891         cfs_module_get();
 892
 893         /* client additional sb info */
 894         lsi->lsi_llsbi = sbi = ll_init_sbi();
 895         if (!sbi) {
 896                 cfs_module_put();
 897                 RETURN(-ENOMEM);
 898         }
 899
 900         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 901         if (err)
 902                 GOTO(out_free, err);
 903
 904         /* Generate a string unique to this super, in case some joker tries
 905            to mount the same fs at two mount points.
 906            Use the address of the super itself.*/
 907         sprintf(ll_instance, "%p", sb);
 908         cfg.cfg_instance = ll_instance;
 909         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 910
 911         /* set up client obds */
 912         err = lustre_process_log(sb, profilenm, &cfg);
 913         if (err < 0) {
 914                 CERROR("Unable to process log: %d\n", err);
 915                 GOTO(out_free, err);
 916         }
 917
 918         lprof = class_get_profile(profilenm);
 919         if (lprof == NULL) {
 920                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 921                                    " read from the MGS.  Does that filesystem "
 922                                    "exist?\n", profilenm);
 923                 GOTO(out_free, err = -EINVAL);
 924         }
 925         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 926                lprof->lp_md, lprof->lp_dt);
 927
 928         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 929                   strlen(ll_instance) + 2);
 930         if (!dt)
 931                 GOTO(out_free, err = -ENOMEM);
 932         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 933
 934         OBD_ALLOC(md, strlen(lprof->lp_md) +
 935                   strlen(ll_instance) + 2);
 936         if (!md)
 937                 GOTO(out_free, err = -ENOMEM);
 938         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 939
 940         /* connections, registrations, sb setup */
 941         err = client_common_fill_super(sb, md, dt);
 942
 943 out_free:
 944         if (md)
 945                 OBD_FREE(md, strlen(md) + 1);
 946         if (dt)
 947                 OBD_FREE(dt, strlen(dt) + 1);
 948         if (err)
 949                 ll_put_super(sb);
 950         else
 951                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 952
 953         RETURN(err);
 954 } /* ll_fill_super */
 955
 956
 957 void lu_context_keys_dump(void);
 958
 959 void ll_put_super(struct super_block *sb)
 960 {
 961         struct config_llog_instance cfg;
 962         char   ll_instance[sizeof(sb) * 2 + 1];
 963         struct obd_device *obd;
 964         struct lustre_sb_info *lsi = s2lsi(sb);
 965         struct ll_sb_info *sbi = ll_s2sbi(sb);
 966         char *profilenm = get_profile_name(sb);
 967         int force = 1, next;
 968         ENTRY;
 969
 970         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 971
 972         ll_print_capa_stat(sbi);
 973
 974         sprintf(ll_instance, "%p", sb);
 975         cfg.cfg_instance = ll_instance;
 976         lustre_end_log(sb, NULL, &cfg);
 977
 978         if (sbi->ll_md_exp) {
 979                 obd = class_exp2obd(sbi->ll_md_exp);
 980                 if (obd)
 981                         force = obd->obd_force;
 982         }
 983
 984         /* We need to set force before the lov_disconnect in
 985            lustre_common_put_super, since l_d cleans up osc's as well. */
 986         if (force) {
 987                 next = 0;
 988                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 989                                                      &next)) != NULL) {
 990                         obd->obd_force = force;
 991                 }
 992         }
 993
 994         cl_sb_fini(sb);
 995
 996         if (sbi->ll_lcq) {
 997                 /* Only if client_common_fill_super succeeded */
 998                 client_common_put_super(sb);
 999         }
1000         next = 0;
1001         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1002                 class_manual_cleanup(obd);
1003         }
1004
1005         if (profilenm)
1006                 class_del_profile(profilenm);
1007
1008         ll_free_sbi(sb);
1009         lsi->lsi_llsbi = NULL;
1010
1011         lustre_common_put_super(sb);
1012
1013         cl_env_cache_purge(~0);
1014         lu_context_keys_dump();
1015
1016         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1017
1018         cfs_module_put();
1019
1020         EXIT;
1021 } /* client_put_super */
1022
1023 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1024 {
1025         struct inode *inode = NULL;
1026         /* NOTE: we depend on atomic igrab() -bzzz */
1027         lock_res_and_lock(lock);
1028         if (lock->l_ast_data) {
1029                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1030                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1031                         inode = igrab(lock->l_ast_data);
1032                 } else {
1033                         inode = lock->l_ast_data;
1034                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1035                                                 D_INFO : D_WARNING,
1036                                         lock, __FILE__, __func__, __LINE__,
1037                                         "l_ast_data %p is bogus: magic %08x",
1038                                         lock->l_ast_data, lli->lli_inode_magic);
1039                         inode = NULL;
1040                 }
1041         }
1042         unlock_res_and_lock(lock);
1043         return inode;
1044 }
1045
1046 static int null_if_equal(struct ldlm_lock *lock, void *data)
1047 {
1048         if (data == lock->l_ast_data) {
1049                 lock->l_ast_data = NULL;
1050
1051                 if (lock->l_req_mode != lock->l_granted_mode)
1052                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1053         }
1054
1055         return LDLM_ITER_CONTINUE;
1056 }
1057
1058 void ll_clear_inode(struct inode *inode)
1059 {
1060         struct ll_inode_info *lli = ll_i2info(inode);
1061         struct ll_sb_info *sbi = ll_i2sbi(inode);
1062         ENTRY;
1063
1064         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1065                inode->i_generation, inode);
1066
1067         if (S_ISDIR(inode->i_mode)) {
1068                 /* these should have been cleared in ll_file_release */
1069                 LASSERT(lli->lli_sai == NULL);
1070                 LASSERT(lli->lli_opendir_key == NULL);
1071                 LASSERT(lli->lli_opendir_pid == 0);
1072         }
1073
1074         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1075         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1076                          null_if_equal, inode);
1077
1078         LASSERT(!lli->lli_open_fd_write_count);
1079         LASSERT(!lli->lli_open_fd_read_count);
1080         LASSERT(!lli->lli_open_fd_exec_count);
1081
1082         if (lli->lli_mds_write_och)
1083                 ll_md_real_close(inode, FMODE_WRITE);
1084         if (lli->lli_mds_exec_och)
1085                 ll_md_real_close(inode, FMODE_EXEC);
1086         if (lli->lli_mds_read_och)
1087                 ll_md_real_close(inode, FMODE_READ);
1088
1089         if (lli->lli_symlink_name) {
1090                 OBD_FREE(lli->lli_symlink_name,
1091                          strlen(lli->lli_symlink_name) + 1);
1092                 lli->lli_symlink_name = NULL;
1093         }
1094
1095         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1096                 LASSERT(lli->lli_posix_acl == NULL);
1097                 if (lli->lli_remote_perms) {
1098                         free_rmtperm_hash(lli->lli_remote_perms);
1099                         lli->lli_remote_perms = NULL;
1100                 }
1101         }
1102 #ifdef CONFIG_FS_POSIX_ACL
1103         else if (lli->lli_posix_acl) {
1104                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1105                 LASSERT(lli->lli_remote_perms == NULL);
1106                 posix_acl_release(lli->lli_posix_acl);
1107                 lli->lli_posix_acl = NULL;
1108         }
1109 #endif
1110         lli->lli_inode_magic = LLI_INODE_DEAD;
1111
1112 #ifdef HAVE_EXPORT___IGET
1113         spin_lock(&sbi->ll_deathrow_lock);
1114         list_del_init(&lli->lli_dead_list);
1115         spin_unlock(&sbi->ll_deathrow_lock);
1116 #endif
1117         ll_clear_inode_capas(inode);
1118         /*
1119          * XXX This has to be done before lsm is freed below, because
1120          * cl_object still uses inode lsm.
1121          */
1122         cl_inode_fini(inode);
1123
1124         if (lli->lli_smd) {
1125                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1126                 lli->lli_smd = NULL;
1127         }
1128
1129
1130         EXIT;
1131 }
1132
1133 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1134                   struct md_open_data **mod)
1135 {
1136         struct lustre_md md;
1137         struct ll_sb_info *sbi = ll_i2sbi(inode);
1138         struct ptlrpc_request *request = NULL;
1139         int rc;
1140         ENTRY;
1141
1142         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1143                                      LUSTRE_OPC_ANY, NULL);
1144         if (IS_ERR(op_data))
1145                 RETURN(PTR_ERR(op_data));
1146
1147         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1148                         &request, mod);
1149         if (rc) {
1150                 ptlrpc_req_finished(request);
1151                 if (rc == -ENOENT) {
1152                         inode->i_nlink = 0;
1153                         /* Unlinked special device node? Or just a race?
1154                          * Pretend we done everything. */
1155                         if (!S_ISREG(inode->i_mode) &&
1156                             !S_ISDIR(inode->i_mode))
1157                                 rc = inode_setattr(inode, &op_data->op_attr);
1158                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1159                         CERROR("md_setattr fails: rc = %d\n", rc);
1160                 }
1161                 RETURN(rc);
1162         }
1163
1164         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1165                               sbi->ll_md_exp, &md);
1166         if (rc) {
1167                 ptlrpc_req_finished(request);
1168                 RETURN(rc);
1169         }
1170
1171         /* We call inode_setattr to adjust timestamps.
1172          * If there is at least some data in file, we cleared ATTR_SIZE
1173          * above to avoid invoking vmtruncate, otherwise it is important
1174          * to call vmtruncate in inode_setattr to update inode->i_size
1175          * (bug 6196) */
1176         rc = inode_setattr(inode, &op_data->op_attr);
1177
1178         /* Extract epoch data if obtained. */
1179         op_data->op_handle = md.body->handle;
1180         op_data->op_ioepoch = md.body->ioepoch;
1181
1182         ll_update_inode(inode, &md);
1183         ptlrpc_req_finished(request);
1184
1185         RETURN(rc);
1186 }
1187
1188 /* Close IO epoch and send Size-on-MDS attribute update. */
1189 static int ll_setattr_done_writing(struct inode *inode,
1190                                    struct md_op_data *op_data,
1191                                    struct md_open_data *mod)
1192 {
1193         struct ll_inode_info *lli = ll_i2info(inode);
1194         int rc = 0;
1195         ENTRY;
1196
1197         LASSERT(op_data != NULL);
1198         if (!S_ISREG(inode->i_mode))
1199                 RETURN(0);
1200
1201         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1202                op_data->op_ioepoch, PFID(&lli->lli_fid));
1203
1204         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1205         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1206         if (rc == -EAGAIN) {
1207                 /* MDS has instructed us to obtain Size-on-MDS attribute
1208                  * from OSTs and send setattr to back to MDS. */
1209                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1210                                          op_data->op_ioepoch);
1211         } else if (rc) {
1212                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1213                        inode->i_ino, rc);
1214         }
1215         RETURN(rc);
1216 }
1217
1218 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1219 {
1220         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1221         int rc;
1222
1223         rc = cl_setattr_do_truncate(inode, size, capa);
1224         ll_truncate_free_capa(capa);
1225         return rc;
1226 }
1227
1228 static int ll_setattr_ost(struct inode *inode)
1229 {
1230         struct obd_capa *capa = ll_mdscapa_get(inode);
1231         int rc;
1232
1233         rc = cl_setattr_ost(inode, capa);
1234         capa_put(capa);
1235
1236         return rc;
1237 }
1238
1239 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1240  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1241  * keep these values until such a time that objects are allocated for it.
1242  * We do the MDS operations first, as it is checking permissions for us.
1243  * We don't to the MDS RPC if there is nothing that we want to store there,
1244  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1245  * going to do an RPC anyways.
1246  *
1247  * If we are doing a truncate, we will send the mtime and ctime updates
1248  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1249  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1250  * at the same time.
1251  */
1252 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1253 {
1254         struct ll_inode_info *lli = ll_i2info(inode);
1255         struct lov_stripe_md *lsm = lli->lli_smd;
1256         struct md_op_data *op_data = NULL;
1257         struct md_open_data *mod = NULL;
1258         int ia_valid = attr->ia_valid;
1259         int rc = 0, rc1 = 0;
1260         ENTRY;
1261
1262         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1263                attr->ia_valid);
1264         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1265
1266         if (ia_valid & ATTR_SIZE) {
1267                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1268                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1269                                attr->ia_size, ll_file_maxbytes(inode));
1270                         RETURN(-EFBIG);
1271                 }
1272
1273                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1274         }
1275
1276         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1277         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1278                 if (current->fsuid != inode->i_uid &&
1279                     !cfs_capable(CFS_CAP_FOWNER))
1280                         RETURN(-EPERM);
1281         }
1282
1283         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1284         if (attr->ia_valid & ATTR_CTIME) {
1285                 attr->ia_ctime = CURRENT_TIME;
1286                 attr->ia_valid |= ATTR_CTIME_SET;
1287         }
1288         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1289                 attr->ia_atime = CURRENT_TIME;
1290                 attr->ia_valid |= ATTR_ATIME_SET;
1291         }
1292         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1293                 attr->ia_mtime = CURRENT_TIME;
1294                 attr->ia_valid |= ATTR_MTIME_SET;
1295         }
1296         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1297                 /* To avoid stale mtime on mds, obtain it from ost and send
1298                    to mds. */
1299                 rc = cl_glimpse_size(inode);
1300                 if (rc)
1301                         RETURN(rc);
1302
1303                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1304                 attr->ia_mtime = inode->i_mtime;
1305         }
1306
1307         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1308                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1309                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1310                        cfs_time_current_sec());
1311
1312         /* NB: ATTR_SIZE will only be set after this point if the size
1313          * resides on the MDS, ie, this file has no objects. */
1314         if (lsm)
1315                 attr->ia_valid &= ~ATTR_SIZE;
1316
1317         /* We always do an MDS RPC, even if we're only changing the size;
1318          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1319
1320         OBD_ALLOC_PTR(op_data);
1321         if (op_data == NULL)
1322                 RETURN(-ENOMEM);
1323
1324         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1325
1326         /* Open epoch for truncate. */
1327         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1328             (ia_valid & ATTR_SIZE))
1329                 op_data->op_flags = MF_EPOCH_OPEN;
1330
1331         rc = ll_md_setattr(inode, op_data, &mod);
1332         if (rc)
1333                 GOTO(out, rc);
1334
1335         if (op_data->op_ioepoch)
1336                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1337                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1338
1339         if (!lsm || !S_ISREG(inode->i_mode)) {
1340                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1341                 GOTO(out, rc = 0);
1342         }
1343
1344         if (ia_valid & ATTR_SIZE)
1345                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1346         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1347                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1348                        inode->i_ino, LTIME_S(attr->ia_mtime));
1349                 rc = ll_setattr_ost(inode);
1350         }
1351         EXIT;
1352 out:
1353         if (op_data) {
1354                 if (op_data->op_ioepoch)
1355                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1356                 ll_finish_md_op_data(op_data);
1357         }
1358         return rc ? rc : rc1;
1359 }
1360
1361 int ll_setattr(struct dentry *de, struct iattr *attr)
1362 {
1363         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1364             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1365                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1366
1367         if ((de->d_inode->i_mode & S_ISUID) &&
1368             !(attr->ia_mode & S_ISUID) &&
1369             !(attr->ia_valid & ATTR_KILL_SUID))
1370                 attr->ia_valid |= ATTR_KILL_SUID;
1371
1372         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1373             !(attr->ia_mode & S_ISGID) &&
1374             !(attr->ia_valid & ATTR_KILL_SGID))
1375                 attr->ia_valid |= ATTR_KILL_SGID;
1376
1377         return ll_setattr_raw(de->d_inode, attr);
1378 }
1379
1380 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1381                        __u64 max_age, __u32 flags)
1382 {
1383         struct ll_sb_info *sbi = ll_s2sbi(sb);
1384         struct obd_statfs obd_osfs;
1385         int rc;
1386         ENTRY;
1387
1388         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1389         if (rc) {
1390                 CERROR("md_statfs fails: rc = %d\n", rc);
1391                 RETURN(rc);
1392         }
1393
1394         osfs->os_type = sb->s_magic;
1395
1396         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1397                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1398
1399         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1400                               &obd_osfs, max_age, flags);
1401         if (rc) {
1402                 CERROR("obd_statfs fails: rc = %d\n", rc);
1403                 RETURN(rc);
1404         }
1405
1406         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1407                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1408                obd_osfs.os_files);
1409
1410         osfs->os_bsize = obd_osfs.os_bsize;
1411         osfs->os_blocks = obd_osfs.os_blocks;
1412         osfs->os_bfree = obd_osfs.os_bfree;
1413         osfs->os_bavail = obd_osfs.os_bavail;
1414
1415         /* If we don't have as many objects free on the OST as inodes
1416          * on the MDS, we reduce the total number of inodes to
1417          * compensate, so that the "inodes in use" number is correct.
1418          */
1419         if (obd_osfs.os_ffree < osfs->os_ffree) {
1420                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1421                         obd_osfs.os_ffree;
1422                 osfs->os_ffree = obd_osfs.os_ffree;
1423         }
1424
1425         RETURN(rc);
1426 }
1427 #ifndef HAVE_STATFS_DENTRY_PARAM
1428 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1429 {
1430 #else
1431 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1432 {
1433         struct super_block *sb = de->d_sb;
1434 #endif
1435         struct obd_statfs osfs;
1436         int rc;
1437
1438         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1439         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1440
1441         /* For now we will always get up-to-date statfs values, but in the
1442          * future we may allow some amount of caching on the client (e.g.
1443          * from QOS or lprocfs updates). */
1444         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1445         if (rc)
1446                 return rc;
1447
1448         statfs_unpack(sfs, &osfs);
1449
1450         /* We need to downshift for all 32-bit kernels, because we can't
1451          * tell if the kernel is being called via sys_statfs64() or not.
1452          * Stop before overflowing f_bsize - in which case it is better
1453          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1454         if (sizeof(long) < 8) {
1455                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1456                         sfs->f_bsize <<= 1;
1457
1458                         osfs.os_blocks >>= 1;
1459                         osfs.os_bfree >>= 1;
1460                         osfs.os_bavail >>= 1;
1461                 }
1462         }
1463
1464         sfs->f_blocks = osfs.os_blocks;
1465         sfs->f_bfree = osfs.os_bfree;
1466         sfs->f_bavail = osfs.os_bavail;
1467
1468         return 0;
1469 }
1470
1471 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1472 {
1473         struct ll_inode_info *lli;
1474         struct lov_stripe_md *lsm;
1475
1476         lli = ll_i2info(inode);
1477         LASSERT(lli->lli_size_sem_owner != current);
1478         down(&lli->lli_size_sem);
1479         LASSERT(lli->lli_size_sem_owner == NULL);
1480         lli->lli_size_sem_owner = current;
1481         lsm = lli->lli_smd;
1482         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1483                  lsm, lock_lsm);
1484         if (lock_lsm)
1485                 lov_stripe_lock(lsm);
1486 }
1487
1488 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1489 {
1490         struct ll_inode_info *lli;
1491         struct lov_stripe_md *lsm;
1492
1493         lli = ll_i2info(inode);
1494         lsm = lli->lli_smd;
1495         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1496                  lsm, unlock_lsm);
1497         if (unlock_lsm)
1498                 lov_stripe_unlock(lsm);
1499         LASSERT(lli->lli_size_sem_owner == current);
1500         lli->lli_size_sem_owner = NULL;
1501         up(&lli->lli_size_sem);
1502 }
1503
1504 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1505 {
1506         struct ll_inode_info *lli = ll_i2info(inode);
1507
1508         dump_lsm(D_INODE, lsm);
1509         dump_lsm(D_INODE, lli->lli_smd);
1510         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1511                  "lsm must be joined lsm %p\n", lsm);
1512         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1513         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1514                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1515         lli->lli_smd = lsm;
1516         lli->lli_maxbytes = lsm->lsm_maxbytes;
1517         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1518                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
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                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1534                                 dump_lsm(D_ERROR, lsm);
1535                                 LBUG();
1536                         }
1537                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1538                                lsm, inode->i_ino, inode->i_generation, inode);
1539                         cl_inode_init(inode, md);
1540                         /* ll_inode_size_lock() requires it is only
1541                          * called with lli_smd != NULL or lock_lsm == 0
1542                          *  or we can race between lock/unlock.
1543                          *  bug 9547 */
1544                         lli->lli_smd = lsm;
1545                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1546                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1547                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1548                 } else {
1549                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1550                              lli->lli_smd->lsm_stripe_count ==
1551                                         lsm->lsm_stripe_count) {
1552                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1553                                         CERROR("lsm mismatch for inode %ld\n",
1554                                                 inode->i_ino);
1555                                         CERROR("lli_smd:\n");
1556                                         dump_lsm(D_ERROR, lli->lli_smd);
1557                                         CERROR("lsm:\n");
1558                                         dump_lsm(D_ERROR, lsm);
1559                                         LBUG();
1560                                 }
1561                         } else {
1562                                 cl_inode_init(inode, md);
1563                                 ll_replace_lsm(inode, lsm);
1564                         }
1565                 }
1566                 if (lli->lli_smd != lsm)
1567                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1568         }
1569
1570         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1571                 if (body->valid & OBD_MD_FLRMTPERM)
1572                         ll_update_remote_perm(inode, md->remote_perm);
1573         }
1574 #ifdef CONFIG_FS_POSIX_ACL
1575         else if (body->valid & OBD_MD_FLACL) {
1576                 spin_lock(&lli->lli_lock);
1577                 if (lli->lli_posix_acl)
1578                         posix_acl_release(lli->lli_posix_acl);
1579                 lli->lli_posix_acl = md->posix_acl;
1580                 spin_unlock(&lli->lli_lock);
1581         }
1582 #endif
1583         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1584         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1585
1586         if (body->valid & OBD_MD_FLATIME &&
1587             body->atime > LTIME_S(inode->i_atime))
1588                 LTIME_S(inode->i_atime) = body->atime;
1589
1590         /* mtime is always updated with ctime, but can be set in past.
1591            As write and utime(2) may happen within 1 second, and utime's
1592            mtime has a priority over write's one, so take mtime from mds
1593            for the same ctimes. */
1594         if (body->valid & OBD_MD_FLCTIME &&
1595             body->ctime >= LTIME_S(inode->i_ctime)) {
1596                 LTIME_S(inode->i_ctime) = body->ctime;
1597                 if (body->valid & OBD_MD_FLMTIME) {
1598                         CDEBUG(D_INODE, "setting ino %lu mtime "
1599                                "from %lu to "LPU64"\n", inode->i_ino,
1600                                LTIME_S(inode->i_mtime), body->mtime);
1601                         LTIME_S(inode->i_mtime) = body->mtime;
1602                 }
1603         }
1604         if (body->valid & OBD_MD_FLMODE)
1605                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1606         if (body->valid & OBD_MD_FLTYPE)
1607                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1608         LASSERT(inode->i_mode != 0);
1609         if (S_ISREG(inode->i_mode)) {
1610                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1611         } else {
1612                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1613         }
1614 #ifdef HAVE_INODE_BLKSIZE
1615         inode->i_blksize = 1<<inode->i_blkbits;
1616 #endif
1617         if (body->valid & OBD_MD_FLUID)
1618                 inode->i_uid = body->uid;
1619         if (body->valid & OBD_MD_FLGID)
1620                 inode->i_gid = body->gid;
1621         if (body->valid & OBD_MD_FLFLAGS)
1622                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1623         if (body->valid & OBD_MD_FLNLINK)
1624                 inode->i_nlink = body->nlink;
1625         if (body->valid & OBD_MD_FLRDEV)
1626                 inode->i_rdev = old_decode_dev(body->rdev);
1627
1628         if (body->valid & OBD_MD_FLID) {
1629                 /* FID shouldn't be changed! */
1630                 if (fid_is_sane(&lli->lli_fid)) {
1631                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1632                                  "Trying to change FID "DFID
1633                                  " to the "DFID", inode %lu/%u(%p)\n",
1634                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1635                                  inode->i_ino, inode->i_generation, inode);
1636                 } else
1637                         lli->lli_fid = body->fid1;
1638         }
1639
1640         LASSERT(fid_seq(&lli->lli_fid) != 0);
1641
1642         if (body->valid & OBD_MD_FLSIZE) {
1643                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1644                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1645                         struct lustre_handle lockh;
1646                         ldlm_mode_t mode;
1647
1648                         /* As it is possible a blocking ast has been processed
1649                          * by this time, we need to check there is an UPDATE
1650                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1651                          * it. */
1652                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1653                                                &lockh);
1654                         if (mode) {
1655                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1656                                                       LLIF_EPOCH_PENDING |
1657                                                       LLIF_SOM_DIRTY)) {
1658                                         CERROR("ino %lu flags %lu still has "
1659                                                "size authority! do not trust "
1660                                                "the size got from MDS\n",
1661                                                inode->i_ino, lli->lli_flags);
1662                                 } else {
1663                                         /* Use old size assignment to avoid
1664                                          * deadlock bz14138 & bz14326 */
1665                                         inode->i_size = body->size;
1666                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1667                                 }
1668                                 ldlm_lock_decref(&lockh, mode);
1669                         }
1670                 } else {
1671                         /* Use old size assignment to avoid
1672                          * deadlock bz14138 & bz14326 */
1673                         inode->i_size = body->size;
1674                 }
1675
1676                 if (body->valid & OBD_MD_FLBLOCKS)
1677                         inode->i_blocks = body->blocks;
1678         }
1679
1680         if (body->valid & OBD_MD_FLMDSCAPA) {
1681                 LASSERT(md->mds_capa);
1682                 ll_add_capa(inode, md->mds_capa);
1683         }
1684         if (body->valid & OBD_MD_FLOSSCAPA) {
1685                 LASSERT(md->oss_capa);
1686                 ll_add_capa(inode, md->oss_capa);
1687         }
1688 }
1689
1690 static struct backing_dev_info ll_backing_dev_info = {
1691         .ra_pages       = 0,    /* No readahead */
1692 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1693         .capabilities   = 0,    /* Does contribute to dirty memory */
1694 #else
1695         .memory_backed  = 0,    /* Does contribute to dirty memory */
1696 #endif
1697 };
1698
1699 void ll_read_inode2(struct inode *inode, void *opaque)
1700 {
1701         struct lustre_md *md = opaque;
1702         struct ll_inode_info *lli = ll_i2info(inode);
1703         ENTRY;
1704
1705         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1706                inode->i_ino, inode->i_generation, inode);
1707
1708         ll_lli_init(lli);
1709
1710         LASSERT(!lli->lli_smd);
1711
1712         /* Core attributes from the MDS first.  This is a new inode, and
1713          * the VFS doesn't zero times in the core inode so we have to do
1714          * it ourselves.  They will be overwritten by either MDS or OST
1715          * attributes - we just need to make sure they aren't newer. */
1716         LTIME_S(inode->i_mtime) = 0;
1717         LTIME_S(inode->i_atime) = 0;
1718         LTIME_S(inode->i_ctime) = 0;
1719         inode->i_rdev = 0;
1720         ll_update_inode(inode, md);
1721
1722         /* OIDEBUG(inode); */
1723
1724         if (S_ISREG(inode->i_mode)) {
1725                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1726                 inode->i_op = &ll_file_inode_operations;
1727                 inode->i_fop = sbi->ll_fop;
1728                 inode->i_mapping->a_ops = &ll_aops;
1729                 EXIT;
1730         } else if (S_ISDIR(inode->i_mode)) {
1731                 inode->i_op = &ll_dir_inode_operations;
1732                 inode->i_fop = &ll_dir_operations;
1733                 inode->i_mapping->a_ops = &ll_dir_aops;
1734                 EXIT;
1735         } else if (S_ISLNK(inode->i_mode)) {
1736                 inode->i_op = &ll_fast_symlink_inode_operations;
1737                 EXIT;
1738         } else {
1739                 inode->i_op = &ll_special_inode_operations;
1740
1741                 init_special_inode(inode, inode->i_mode,
1742                                    kdev_t_to_nr(inode->i_rdev));
1743
1744                 /* initializing backing dev info. */
1745                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1746
1747                 EXIT;
1748         }
1749 }
1750
1751 void ll_delete_inode(struct inode *inode)
1752 {
1753         struct ll_sb_info *sbi = ll_i2sbi(inode);
1754         int rc;
1755         ENTRY;
1756
1757         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1758         if (rc) {
1759                 CERROR("fid_delete() failed, rc %d\n", rc);
1760         }
1761         truncate_inode_pages(&inode->i_data, 0);
1762         clear_inode(inode);
1763
1764         EXIT;
1765 }
1766
1767 int ll_iocontrol(struct inode *inode, struct file *file,
1768                  unsigned int cmd, unsigned long arg)
1769 {
1770         struct ll_sb_info *sbi = ll_i2sbi(inode);
1771         struct ptlrpc_request *req = NULL;
1772         int rc, flags = 0;
1773         ENTRY;
1774
1775         switch(cmd) {
1776         case EXT3_IOC_GETFLAGS: {
1777                 struct mdt_body *body;
1778                 struct obd_capa *oc;
1779
1780                 oc = ll_mdscapa_get(inode);
1781                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1782                                 OBD_MD_FLFLAGS, 0, &req);
1783                 capa_put(oc);
1784                 if (rc) {
1785                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1786                         RETURN(-abs(rc));
1787                 }
1788
1789                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1790
1791                 flags = body->flags;
1792
1793                 ptlrpc_req_finished(req);
1794
1795                 RETURN(put_user(flags, (int *)arg));
1796         }
1797         case EXT3_IOC_SETFLAGS: {
1798                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1799                 struct obd_info oinfo = { { { 0 } } };
1800                 struct md_op_data *op_data;
1801
1802                 if (get_user(flags, (int *)arg))
1803                         RETURN(-EFAULT);
1804
1805                 oinfo.oi_md = lsm;
1806                 OBDO_ALLOC(oinfo.oi_oa);
1807                 if (!oinfo.oi_oa)
1808                         RETURN(-ENOMEM);
1809
1810                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1811                                              LUSTRE_OPC_ANY, NULL);
1812                 if (IS_ERR(op_data))
1813                         RETURN(PTR_ERR(op_data));
1814
1815                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1816                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1817                 rc = md_setattr(sbi->ll_md_exp, op_data,
1818                                 NULL, 0, NULL, 0, &req, NULL);
1819                 ll_finish_md_op_data(op_data);
1820                 ptlrpc_req_finished(req);
1821                 if (rc || lsm == NULL) {
1822                         OBDO_FREE(oinfo.oi_oa);
1823                         RETURN(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("md_setattr_async fails: rc = %d\n", rc);
1841                         RETURN(rc);
1842                 }
1843
1844                 inode->i_flags = ll_ext_to_inode_flags(flags |
1845                                                        MDS_BFLAG_EXT_FLAGS);
1846                 RETURN(0);
1847         }
1848         default:
1849                 RETURN(-ENOSYS);
1850         }
1851
1852         RETURN(0);
1853 }
1854
1855 int ll_flush_ctx(struct inode *inode)
1856 {
1857         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1858
1859         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1860
1861         obd_set_info_async(sbi->ll_md_exp,
1862                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1863                            0, NULL, NULL);
1864         obd_set_info_async(sbi->ll_dt_exp,
1865                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1866                            0, NULL, NULL);
1867         return 0;
1868 }
1869
1870 /* umount -f client means force down, don't save state */
1871 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1872 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1873 {
1874         struct super_block *sb = vfsmnt->mnt_sb;
1875 #else
1876 void ll_umount_begin(struct super_block *sb)
1877 {
1878 #endif
1879         struct lustre_sb_info *lsi = s2lsi(sb);
1880         struct ll_sb_info *sbi = ll_s2sbi(sb);
1881         struct obd_device *obd;
1882         struct obd_ioctl_data ioc_data = { 0 };
1883         ENTRY;
1884
1885 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1886         if (!(flags & MNT_FORCE)) {
1887                 EXIT;
1888                 return;
1889         }
1890 #endif
1891
1892         /* Tell the MGC we got umount -f */
1893         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1894
1895         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1896                sb->s_count, atomic_read(&sb->s_active));
1897
1898         obd = class_exp2obd(sbi->ll_md_exp);
1899         if (obd == NULL) {
1900                 CERROR("Invalid MDC connection handle "LPX64"\n",
1901                        sbi->ll_md_exp->exp_handle.h_cookie);
1902                 EXIT;
1903                 return;
1904         }
1905         obd->obd_force = 1;
1906         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1907                       &ioc_data, NULL);
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
1917         obd->obd_force = 1;
1918         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1919                       &ioc_data, NULL);
1920
1921         /* Really, we'd like to wait until there are no requests outstanding,
1922          * and then continue.  For now, we just invalidate the requests,
1923          * schedule() and sleep one second if needed, and hope.
1924          */
1925         schedule();
1926 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1927         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1928                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1929                                      cfs_time_seconds(1));
1930                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1931                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1932                                       "may try to umount it a bit later\n",
1933                                       atomic_read(&vfsmnt->mnt_count));
1934         }
1935 #endif
1936
1937         EXIT;
1938 }
1939
1940 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1941 {
1942         struct ll_sb_info *sbi = ll_s2sbi(sb);
1943         int err;
1944         __u32 read_only;
1945
1946         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1947                 read_only = *flags & MS_RDONLY;
1948                 err = obd_set_info_async(sbi->ll_md_exp,
1949                                          sizeof(KEY_READ_ONLY),
1950                                          KEY_READ_ONLY, sizeof(read_only),
1951                                          &read_only, NULL);
1952                 if (err) {
1953                         CERROR("Failed to change the read-only flag during "
1954                                "remount: %d\n", err);
1955                         return err;
1956                 }
1957
1958                 if (read_only)
1959                         sb->s_flags |= MS_RDONLY;
1960                 else
1961                         sb->s_flags &= ~MS_RDONLY;
1962         }
1963         return 0;
1964 }
1965
1966 int ll_prep_inode(struct inode **inode,
1967                   struct ptlrpc_request *req,
1968                   struct super_block *sb)
1969 {
1970         struct ll_sb_info *sbi = NULL;
1971         struct lustre_md md;
1972         int rc;
1973         ENTRY;
1974
1975         LASSERT(*inode || sb);
1976         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1977         prune_deathrow(sbi, 1);
1978         memset(&md, 0, sizeof(struct lustre_md));
1979
1980         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1981                               sbi->ll_md_exp, &md);
1982         if (rc)
1983                 RETURN(rc);
1984
1985         if (*inode) {
1986                 ll_update_inode(*inode, &md);
1987         } else {
1988                 LASSERT(sb != NULL);
1989
1990                 /*
1991                  * At this point server returns to client's same fid as client
1992                  * generated for creating. So using ->fid1 is okay here.
1993                  */
1994                 LASSERT(fid_is_sane(&md.body->fid1));
1995
1996                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
1997                 if (*inode == NULL || IS_ERR(*inode)) {
1998                         if (md.lsm)
1999                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2000 #ifdef CONFIG_FS_POSIX_ACL
2001                         if (md.posix_acl) {
2002                                 posix_acl_release(md.posix_acl);
2003                                 md.posix_acl = NULL;
2004                         }
2005 #endif
2006                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2007                         *inode = NULL;
2008                         CERROR("new_inode -fatal: rc %d\n", rc);
2009                         GOTO(out, rc);
2010                 }
2011         }
2012
2013         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2014                          ll_i2info(*inode)->lli_smd);
2015 out:
2016         md_free_lustre_md(sbi->ll_md_exp, &md);
2017         RETURN(rc);
2018 }
2019
2020 int ll_obd_statfs(struct inode *inode, void *arg)
2021 {
2022         struct ll_sb_info *sbi = NULL;
2023         struct obd_export *exp;
2024         char *buf = NULL;
2025         struct obd_ioctl_data *data = NULL;
2026         __u32 type;
2027         int len = 0, rc;
2028
2029         if (!inode || !(sbi = ll_i2sbi(inode)))
2030                 GOTO(out_statfs, rc = -EINVAL);
2031
2032         rc = obd_ioctl_getdata(&buf, &len, arg);
2033         if (rc)
2034                 GOTO(out_statfs, rc);
2035
2036         data = (void*)buf;
2037         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2038             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2039                 GOTO(out_statfs, rc = -EINVAL);
2040
2041         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2042         if (type == LL_STATFS_MDC)
2043                 exp = sbi->ll_md_exp;
2044         else if (type == LL_STATFS_LOV)
2045                 exp = sbi->ll_dt_exp;
2046         else
2047                 GOTO(out_statfs, rc = -ENODEV);
2048
2049         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2050         if (rc)
2051                 GOTO(out_statfs, rc);
2052 out_statfs:
2053         if (buf)
2054                 obd_ioctl_freedata(buf, len);
2055         return rc;
2056 }
2057
2058 int ll_process_config(struct lustre_cfg *lcfg)
2059 {
2060         char *ptr;
2061         void *sb;
2062         struct lprocfs_static_vars lvars;
2063         unsigned long x;
2064         int rc = 0;
2065
2066         lprocfs_llite_init_vars(&lvars);
2067
2068         /* The instance name contains the sb: lustre-client-aacfe000 */
2069         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2070         if (!ptr || !*(++ptr))
2071                 return -EINVAL;
2072         if (sscanf(ptr, "%lx", &x) != 1)
2073                 return -EINVAL;
2074         sb = (void *)x;
2075         /* This better be a real Lustre superblock! */
2076         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2077
2078         /* Note we have not called client_common_fill_super yet, so
2079            proc fns must be able to handle that! */
2080         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2081                                       lcfg, sb);
2082         if (rc > 0)
2083                 rc = 0;
2084         return(rc);
2085 }
2086
2087 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2088 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2089                                        struct inode *i1, struct inode *i2,
2090                                        const char *name, int namelen,
2091                                        int mode, __u32 opc, void *data)
2092 {
2093         LASSERT(i1 != NULL);
2094
2095         if (namelen > ll_i2sbi(i1)->ll_namelen)
2096                 return ERR_PTR(-ENAMETOOLONG);
2097
2098         if (op_data == NULL)
2099                 OBD_ALLOC_PTR(op_data);
2100
2101         if (op_data == NULL)
2102                 return ERR_PTR(-ENOMEM);
2103
2104         ll_i2gids(op_data->op_suppgids, i1, i2);
2105         op_data->op_fid1 = *ll_inode2fid(i1);
2106         op_data->op_capa1 = ll_mdscapa_get(i1);
2107
2108         if (i2) {
2109                 op_data->op_fid2 = *ll_inode2fid(i2);
2110                 op_data->op_capa2 = ll_mdscapa_get(i2);
2111         } else {
2112                 fid_zero(&op_data->op_fid2);
2113                 op_data->op_capa2 = NULL;
2114         }
2115
2116         op_data->op_name = name;
2117         op_data->op_namelen = namelen;
2118         op_data->op_mode = mode;
2119         op_data->op_mod_time = cfs_time_current_sec();
2120         op_data->op_fsuid = current->fsuid;
2121         op_data->op_fsgid = current->fsgid;
2122         op_data->op_cap = cfs_curproc_cap_pack();
2123         op_data->op_bias = MDS_CHECK_SPLIT;
2124         op_data->op_opc = opc;
2125         op_data->op_mds = 0;
2126         op_data->op_data = data;
2127
2128         return op_data;
2129 }
2130
2131 void ll_finish_md_op_data(struct md_op_data *op_data)
2132 {
2133         capa_put(op_data->op_capa1);
2134         capa_put(op_data->op_capa2);
2135         OBD_FREE_PTR(op_data);
2136 }
2137
2138 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2139 {
2140         struct ll_sb_info *sbi;
2141
2142         LASSERT((seq != NULL) && (vfs != NULL));
2143         sbi = ll_s2sbi(vfs->mnt_sb);
2144
2145         if (sbi->ll_flags & LL_SBI_NOLCK)
2146                 seq_puts(seq, ",nolock");
2147
2148         if (sbi->ll_flags & LL_SBI_FLOCK)
2149                 seq_puts(seq, ",flock");
2150
2151         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2152                 seq_puts(seq, ",localflock");
2153
2154         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2155                 seq_puts(seq, ",user_xattr");
2156
2157         if (sbi->ll_flags & LL_SBI_ACL)
2158                 seq_puts(seq, ",acl");
2159
2160         RETURN(0);
2161 }