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  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002-2005 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <lustre_lite.h>
  32 #include <lustre_ha.h>
  33 #include <lustre_dlm.h>
  34 #include <lprocfs_status.h>
  35 #include <lustre_disk.h>
  36 #include <lustre_param.h>
  37 #include <lustre_log.h>
  38 #include "llite_internal.h"
  39
  40 cfs_mem_cache_t *ll_file_data_slab;
  41
  42 LIST_HEAD(ll_super_blocks);
  43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52
  53 static struct ll_sb_info *ll_init_sbi(void)
  54 {
  55         struct ll_sb_info *sbi = NULL;
  56         class_uuid_t uuid;
  57         int i;
  58         ENTRY;
  59
  60         OBD_ALLOC(sbi, sizeof(*sbi));
  61         if (!sbi)
  62                 RETURN(NULL);
  63
  64         spin_lock_init(&sbi->ll_lock);
  65         spin_lock_init(&sbi->ll_lco.lco_lock);
  66         spin_lock_init(&sbi->ll_pp_extent_lock);
  67         spin_lock_init(&sbi->ll_process_lock);
  68         sbi->ll_rw_stats_on = 0;
  69         INIT_LIST_HEAD(&sbi->ll_pglist);
  70         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
  71                 sbi->ll_async_page_max = num_physpages / 2;
  72         else
  73                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  74         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  75                                            SBI_DEFAULT_READAHEAD_MAX);
  76         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  77                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  78
  79         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  80         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
  81
  82         ll_generate_random_uuid(uuid);
  83         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  84         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  85
  86         spin_lock(&ll_sb_lock);
  87         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  88         spin_unlock(&ll_sb_lock);
  89
  90 #ifdef ENABLE_LLITE_CHECKSUM
  91         sbi->ll_flags |= LL_SBI_CHECKSUM;
  92 #endif
  93
  94 #ifdef HAVE_EXPORT___IGET
  95         INIT_LIST_HEAD(&sbi->ll_deathrow);
  96         spin_lock_init(&sbi->ll_deathrow_lock);
  97 #endif
  98         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
  99                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 100                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 101         }
 102
 103         RETURN(sbi);
 104 }
 105
 106 void ll_free_sbi(struct super_block *sb)
 107 {
 108         struct ll_sb_info *sbi = ll_s2sbi(sb);
 109         ENTRY;
 110
 111         if (sbi != NULL) {
 112                 spin_lock(&ll_sb_lock);
 113                 list_del(&sbi->ll_list);
 114                 spin_unlock(&ll_sb_lock);
 115                 OBD_FREE(sbi, sizeof(*sbi));
 116         }
 117         EXIT;
 118 }
 119
 120 static struct dentry_operations ll_d_root_ops = {
 121 #ifdef LUSTRE_KERNEL_VERSION
 122         .d_compare = ll_dcompare,
 123 #endif
 124 };
 125
 126 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 127  * us to make MDS RPCs with large enough reply buffers to hold the
 128  * maximum-sized (= maximum striped) EA and cookie without having to
 129  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 130 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 131 {
 132         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 133         __u32 valsize = sizeof(struct lov_desc);
 134         int rc, easize, def_easize, cookiesize;
 135         struct lov_desc desc;
 136         __u32 stripes;
 137         ENTRY;
 138
 139         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 140                           &valsize, &desc);
 141         if (rc)
 142                 RETURN(rc);
 143
 144         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 145         lsm.lsm_stripe_count = stripes;
 146         easize = obd_size_diskmd(dt_exp, &lsm);
 147
 148         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 149         def_easize = obd_size_diskmd(dt_exp, &lsm);
 150
 151         cookiesize = stripes * sizeof(struct llog_cookie);
 152
 153         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 154                easize, cookiesize);
 155
 156         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 157         RETURN(rc);
 158 }
 159
 160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 161                                     uid_t nllu, gid_t nllg)
 162 {
 163         struct inode *root = 0;
 164         struct ll_sb_info *sbi = ll_s2sbi(sb);
 165         struct obd_device *obd;
 166         struct lu_fid rootfid;
 167         struct obd_capa *oc = NULL;
 168         struct obd_statfs osfs;
 169         struct ptlrpc_request *request = NULL;
 170         struct lustre_handle dt_conn = {0, };
 171         struct lustre_handle md_conn = {0, };
 172         struct obd_connect_data *data = NULL;
 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/lustre");
 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 |
 198                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET;
 201 #ifdef HAVE_LRU_RESIZE_SUPPORT
 202         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 203 #endif
 204 #ifdef CONFIG_FS_POSIX_ACL
 205         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 206 #endif
 207         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 208         data->ocd_version = LUSTRE_VERSION_CODE;
 209
 210         if (sb->s_flags & MS_RDONLY)
 211                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 212         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 213                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 214
 215 #ifdef HAVE_MS_FLOCK_LOCK
 216         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 217         sb->s_flags |= MS_FLOCK_LOCK;
 218 #endif
 219
 220         if (sbi->ll_flags & LL_SBI_FLOCK)
 221                 sbi->ll_fop = &ll_file_operations_flock;
 222         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 223                 sbi->ll_fop = &ll_file_operations;
 224         else
 225                 sbi->ll_fop = &ll_file_operations_noflock;
 226
 227         /* real client */
 228         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 229         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 230                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 231                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 232         } else {
 233                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 234                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 235         }
 236         data->ocd_nllu = nllu;
 237         data->ocd_nllg = nllg;
 238
 239         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
 240         if (err == -EBUSY) {
 241                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 242                                    "recovery, of which this client is not a "
 243                                    "part. Please wait for recovery to complete,"
 244                                    " abort, or time out.\n", md);
 245                 GOTO(out, err);
 246         } else if (err) {
 247                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 248                 GOTO(out, err);
 249         }
 250         sbi->ll_md_exp = class_conn2export(&md_conn);
 251
 252         err = obd_fid_init(sbi->ll_md_exp);
 253         if (err) {
 254                 CERROR("Can't init metadata layer FID infrastructure, "
 255                        "rc %d\n", err);
 256                 GOTO(out_md, err);
 257         }
 258
 259         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
 260         if (err)
 261                 GOTO(out_md_fid, err);
 262
 263         size = sizeof(*data);
 264         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 265                            KEY_CONN_DATA,  &size, data);
 266         if (err) {
 267                 CERROR("Get connect data failed: %d \n", err);
 268                 GOTO(out_md, err);
 269         }
 270
 271         LASSERT(osfs.os_bsize);
 272         sb->s_blocksize = osfs.os_bsize;
 273         sb->s_blocksize_bits = log2(osfs.os_bsize);
 274         sb->s_magic = LL_SUPER_MAGIC;
 275
 276         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 277          *         retval = in_file->f_op->sendfile(...);
 278          *         if (*ppos > max)
 279          *                 retval = -EOVERFLOW;
 280          *
 281          * it will check if *ppos is greater than max. However, max equals to
 282          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 283          * has been defined as a signed long long ineger in linux kernel. */
 284 #if BITS_PER_LONG == 64
 285         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 286 #else
 287         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 288 #endif
 289         sbi->ll_namelen = osfs.os_namelen;
 290         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 291
 292         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 293             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 294                 LCONSOLE_INFO("Disabling user_xattr feature because "
 295                               "it is not supported on the server\n");
 296                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 297         }
 298
 299         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 300 #ifdef MS_POSIXACL
 301                 sb->s_flags |= MS_POSIXACL;
 302 #endif
 303                 sbi->ll_flags |= LL_SBI_ACL;
 304         } else {
 305                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 306 #ifdef MS_POSIXACL
 307                 sb->s_flags &= ~MS_POSIXACL;
 308 #endif
 309                 sbi->ll_flags &= ~LL_SBI_ACL;
 310         }
 311
 312         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 313                 sbi->ll_flags |= LL_SBI_JOIN;
 314
 315         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 316                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 317                         /* sometimes local client claims to be remote, but mdt
 318                          * will disagree when client gss not applied. */
 319                         LCONSOLE_INFO("client claims to be remote, but server "
 320                                       "rejected, forced to be local.\n");
 321                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 322                 }
 323         } else {
 324                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 325                         /* with gss applied, remote client can not claim to be
 326                          * local, so mdt maybe force client to be remote. */
 327                         LCONSOLE_INFO("client claims to be local, but server "
 328                                       "rejected, forced to be remote.\n");
 329                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 330                 }
 331         }
 332
 333         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 334                 LCONSOLE_INFO("client enabled MDS capability!\n");
 335                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 339                 LCONSOLE_INFO("client enabled OSS capability!\n");
 340                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 341         }
 342
 343         sbi->ll_sdev_orig = sb->s_dev;
 344 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 345         /* We set sb->s_dev equal on all lustre clients in order to support
 346          * NFS export clustering.  NFSD requires that the FSID be the same
 347          * on all clients. */
 348         /* s_dev is also used in lt_compare() to compare two fs, but that is
 349          * only a node-local comparison. */
 350
 351         /* XXX: this will not work with LMV */
 352         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 353                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 354 #endif
 355
 356         obd = class_name2obd(dt);
 357         if (!obd) {
 358                 CERROR("DT %s: not setup or attached\n", dt);
 359                 GOTO(out_md_fid, err = -ENODEV);
 360         }
 361
 362         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 363                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 364                                   OBD_CONNECT_CANCELSET;
 365         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 366                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 367
 368 #ifdef HAVE_LRU_RESIZE_SUPPORT
 369         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 370 #endif
 371         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 372                "ocd_grant: %d\n", data->ocd_connect_flags,
 373                data->ocd_version, data->ocd_grant);
 374
 375         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 376         obd->obd_upcall.onu_upcall = ll_ocd_update;
 377         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 378
 379         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
 380         if (err == -EBUSY) {
 381                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 382                                    "recovery, of which this client is not a "
 383                                    "part.  Please wait for recovery to "
 384                                    "complete, abort, or time out.\n", dt);
 385                 GOTO(out_md_fid, err);
 386         } else if (err) {
 387                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 388                 GOTO(out_md_fid, err);
 389         }
 390
 391         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 392
 393         err = obd_fid_init(sbi->ll_dt_exp);
 394         if (err) {
 395                 CERROR("Can't init data layer FID infrastructure, "
 396                        "rc %d\n", err);
 397                 GOTO(out_dt, err);
 398         }
 399
 400         spin_lock(&sbi->ll_lco.lco_lock);
 401         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 402         spin_unlock(&sbi->ll_lco.lco_lock);
 403
 404         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 405
 406         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 407                                   0, NULL, NULL, NULL);
 408         if (err < 0) {
 409                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 410                                    "filesystem. There must be at least one "
 411                                    "active OST for a client to start.\n");
 412                 GOTO(out_dt_fid, err);
 413         }
 414
 415         if (!ll_async_page_slab) {
 416                 ll_async_page_slab_size =
 417                         size_round(sizeof(struct ll_async_page)) + err;
 418                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 419                                                         ll_async_page_slab_size,
 420                                                            0, 0);
 421                 if (!ll_async_page_slab)
 422                         GOTO(out_dt_fid, err = -ENOMEM);
 423         }
 424
 425         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 426         if (err) {
 427                 CERROR("cannot mds_connect: rc = %d\n", err);
 428                 GOTO(out_dt_fid, err);
 429         }
 430         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 431         sbi->ll_root_fid = rootfid;
 432
 433         sb->s_op = &lustre_super_operations;
 434 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 435         sb->s_export_op = &lustre_export_operations;
 436 #endif
 437
 438         /* make root inode
 439          * XXX: move this to after cbd setup? */
 440         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 441         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 442                 valid |= OBD_MD_FLRMTPERM;
 443         else if (sbi->ll_flags & LL_SBI_ACL)
 444                 valid |= OBD_MD_FLACL;
 445
 446         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 447         if (oc)
 448                 free_capa(oc);
 449         if (err) {
 450                 CERROR("md_getattr failed for root: rc = %d\n", err);
 451                 GOTO(out_dt_fid, err);
 452         }
 453         memset(&lmd, 0, sizeof(lmd));
 454         err = md_get_lustre_md(sbi->ll_md_exp, request,
 455                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
 456                                &lmd);
 457         if (err) {
 458                 CERROR("failed to understand root inode md: rc = %d\n", err);
 459                 ptlrpc_req_finished (request);
 460                 GOTO(out_dt_fid, err);
 461         }
 462
 463         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 464         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 465         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 466         ptlrpc_req_finished(request);
 467
 468         if (root == NULL || is_bad_inode(root)) {
 469                 if (lmd.lsm)
 470                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 471 #ifdef CONFIG_FS_POSIX_ACL
 472                 if (lmd.posix_acl) {
 473                         posix_acl_release(lmd.posix_acl);
 474                         lmd.posix_acl = NULL;
 475                 }
 476 #endif
 477                 CERROR("lustre_lite: bad iget4 for root\n");
 478                 GOTO(out_root, err = -EBADF);
 479         }
 480
 481         err = ll_close_thread_start(&sbi->ll_lcq);
 482         if (err) {
 483                 CERROR("cannot start close thread: rc %d\n", err);
 484                 GOTO(out_root, err);
 485         }
 486
 487         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 488         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 489                                  sizeof(checksum), &checksum, NULL);
 490
 491         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 492            backing dev info assigned to inode mapping is used for
 493            determining maximal readahead. */
 494 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 495     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 496         /* bug 2805 - set VM readahead to zero */
 497         vm_max_readahead = vm_min_readahead = 0;
 498 #endif
 499
 500         sb->s_root = d_alloc_root(root);
 501         if (data != NULL)
 502                 OBD_FREE(data, sizeof(*data));
 503         sb->s_root->d_op = &ll_d_root_ops;
 504         RETURN(err);
 505 out_root:
 506         if (root)
 507                 iput(root);
 508 out_dt_fid:
 509         obd_fid_fini(sbi->ll_dt_exp);
 510 out_dt:
 511         obd_disconnect(sbi->ll_dt_exp);
 512         sbi->ll_dt_exp = NULL;
 513 out_md_fid:
 514         obd_fid_fini(sbi->ll_md_exp);
 515 out_md:
 516         obd_disconnect(sbi->ll_md_exp);
 517         sbi->ll_md_exp = NULL;
 518 out:
 519         if (data != NULL)
 520                 OBD_FREE_PTR(data);
 521         lprocfs_unregister_mountpoint(sbi);
 522         return err;
 523 }
 524
 525 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 526 {
 527         int size, rc;
 528
 529         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 530         size = sizeof(int);
 531         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 532                           &size, lmmsize);
 533         if (rc)
 534                 CERROR("Get max mdsize error rc %d \n", rc);
 535
 536         RETURN(rc);
 537 }
 538
 539 void ll_dump_inode(struct inode *inode)
 540 {
 541         struct list_head *tmp;
 542         int dentry_count = 0;
 543
 544         LASSERT(inode != NULL);
 545
 546         list_for_each(tmp, &inode->i_dentry)
 547                 dentry_count++;
 548
 549         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 550                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 551                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 552 }
 553
 554 void lustre_dump_dentry(struct dentry *dentry, int recur)
 555 {
 556         struct list_head *tmp;
 557         int subdirs = 0;
 558
 559         LASSERT(dentry != NULL);
 560
 561         list_for_each(tmp, &dentry->d_subdirs)
 562                 subdirs++;
 563
 564         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 565                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 566                dentry->d_name.len, dentry->d_name.name,
 567                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 568                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 569                dentry->d_flags, dentry->d_fsdata, subdirs);
 570         if (dentry->d_inode != NULL)
 571                 ll_dump_inode(dentry->d_inode);
 572
 573         if (recur == 0)
 574                 return;
 575
 576         list_for_each(tmp, &dentry->d_subdirs) {
 577                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 578                 lustre_dump_dentry(d, recur - 1);
 579         }
 580 }
 581
 582 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 583 void lustre_throw_orphan_dentries(struct super_block *sb)
 584 {
 585         struct dentry *dentry, *next;
 586         struct ll_sb_info *sbi = ll_s2sbi(sb);
 587
 588         /* Do this to get rid of orphaned dentries. That is not really trw. */
 589         list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
 590                                  d_hash) {
 591                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 592                       "before and after shrink_dcache_parent\n",
 593                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 594                 lustre_dump_dentry(dentry, 1);
 595                 shrink_dcache_parent(dentry);
 596                 lustre_dump_dentry(dentry, 1);
 597         }
 598 }
 599 #else
 600 #define lustre_throw_orphan_dentries(sb)
 601 #endif
 602
 603 #ifdef HAVE_EXPORT___IGET
 604 static void prune_dir_dentries(struct inode *inode)
 605 {
 606         struct dentry *dentry, *prev = NULL;
 607
 608         /* due to lustre specific logic, a directory
 609          * can have few dentries - a bug from VFS POV */
 610 restart:
 611         spin_lock(&dcache_lock);
 612         if (!list_empty(&inode->i_dentry)) {
 613                 dentry = list_entry(inode->i_dentry.prev,
 614                                     struct dentry, d_alias);
 615                 /* in order to prevent infinite loops we
 616                  * break if previous dentry is busy */
 617                 if (dentry != prev) {
 618                         prev = dentry;
 619                         dget_locked(dentry);
 620                         spin_unlock(&dcache_lock);
 621
 622                         /* try to kill all child dentries */
 623                         lock_dentry(dentry);
 624                         shrink_dcache_parent(dentry);
 625                         unlock_dentry(dentry);
 626                         dput(dentry);
 627
 628                         /* now try to get rid of current dentry */
 629                         d_prune_aliases(inode);
 630                         goto restart;
 631                 }
 632         }
 633         spin_unlock(&dcache_lock);
 634 }
 635
 636 static void prune_deathrow_one(struct ll_inode_info *lli)
 637 {
 638         struct inode *inode = ll_info2i(lli);
 639
 640         /* first, try to drop any dentries - they hold a ref on the inode */
 641         if (S_ISDIR(inode->i_mode))
 642                 prune_dir_dentries(inode);
 643         else
 644                 d_prune_aliases(inode);
 645
 646
 647         /* if somebody still uses it, leave it */
 648         LASSERT(atomic_read(&inode->i_count) > 0);
 649         if (atomic_read(&inode->i_count) > 1)
 650                 goto out;
 651
 652         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 653                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 654
 655         /* seems nobody uses it anymore */
 656         inode->i_nlink = 0;
 657
 658 out:
 659         iput(inode);
 660         return;
 661 }
 662
 663 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 664 {
 665         struct ll_inode_info *lli;
 666         int empty;
 667
 668         do {
 669                 if (need_resched() && try)
 670                         break;
 671
 672                 if (try) {
 673                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 674                                 break;
 675                 } else {
 676                         spin_lock(&sbi->ll_deathrow_lock);
 677                 }
 678
 679                 empty = 1;
 680                 lli = NULL;
 681                 if (!list_empty(&sbi->ll_deathrow)) {
 682                         lli = list_entry(sbi->ll_deathrow.next,
 683                                          struct ll_inode_info,
 684                                          lli_dead_list);
 685                         list_del_init(&lli->lli_dead_list);
 686                         if (!list_empty(&sbi->ll_deathrow))
 687                                 empty = 0;
 688                 }
 689                 spin_unlock(&sbi->ll_deathrow_lock);
 690
 691                 if (lli)
 692                         prune_deathrow_one(lli);
 693
 694         } while (empty == 0);
 695 }
 696 #else /* !HAVE_EXPORT___IGET */
 697 #define prune_deathrow(sbi, try) do {} while (0)
 698 #endif /* HAVE_EXPORT___IGET */
 699
 700 void client_common_put_super(struct super_block *sb)
 701 {
 702         struct ll_sb_info *sbi = ll_s2sbi(sb);
 703         ENTRY;
 704
 705         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 706
 707         ll_close_thread_shutdown(sbi->ll_lcq);
 708
 709         /* destroy inodes in deathrow */
 710         prune_deathrow(sbi, 0);
 711
 712         list_del(&sbi->ll_conn_chain);
 713
 714         obd_fid_fini(sbi->ll_dt_exp);
 715         obd_disconnect(sbi->ll_dt_exp);
 716         sbi->ll_dt_exp = NULL;
 717
 718         lprocfs_unregister_mountpoint(sbi);
 719
 720         obd_fid_fini(sbi->ll_md_exp);
 721         obd_disconnect(sbi->ll_md_exp);
 722         sbi->ll_md_exp = NULL;
 723
 724         lustre_throw_orphan_dentries(sb);
 725
 726         EXIT;
 727 }
 728
 729 void ll_kill_super(struct super_block *sb)
 730 {
 731         struct ll_sb_info *sbi;
 732
 733         ENTRY;
 734
 735         /* not init sb ?*/
 736         if (!(sb->s_flags & MS_ACTIVE))
 737                 return;
 738
 739         sbi = ll_s2sbi(sb);
 740         /* we need restore s_dev from changed for clustred NFS before put_super
 741          * because new kernels have cached s_dev and change sb->s_dev in
 742          * put_super not affected real removing devices */
 743         if (sbi)
 744                 sb->s_dev = sbi->ll_sdev_orig;
 745         EXIT;
 746 }
 747
 748 char *ll_read_opt(const char *opt, char *data)
 749 {
 750         char *value;
 751         char *retval;
 752         ENTRY;
 753
 754         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 755         if (strncmp(opt, data, strlen(opt)))
 756                 RETURN(NULL);
 757         if ((value = strchr(data, '=')) == NULL)
 758                 RETURN(NULL);
 759
 760         value++;
 761         OBD_ALLOC(retval, strlen(value) + 1);
 762         if (!retval) {
 763                 CERROR("out of memory!\n");
 764                 RETURN(NULL);
 765         }
 766
 767         memcpy(retval, value, strlen(value)+1);
 768         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 769         RETURN(retval);
 770 }
 771
 772 static inline int ll_set_opt(const char *opt, char *data, int fl)
 773 {
 774         if (strncmp(opt, data, strlen(opt)) != 0)
 775                 return(0);
 776         else
 777                 return(fl);
 778 }
 779
 780 /* non-client-specific mount options are parsed in lmd_parse */
 781 static int ll_options(char *options, int *flags)
 782 {
 783         int tmp;
 784         char *s1 = options, *s2;
 785         ENTRY;
 786
 787         if (!options)
 788                 RETURN(0);
 789
 790         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 791
 792         while (*s1) {
 793                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 794                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 795                 if (tmp) {
 796                         *flags |= tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 805                 if (tmp) {
 806                         *flags |= tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 810                 if (tmp) {
 811                         *flags &= ~tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 815                 if (tmp) {
 816                         *flags |= tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 820                 if (tmp) {
 821                         *flags &= ~tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 825                 if (tmp) {
 826                         /* Ignore deprecated mount option.  The client will
 827                          * always try to mount with ACL support, whether this
 828                          * is used depends on whether server supports it. */
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 832                 if (tmp) {
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840
 841                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 842                 if (tmp) {
 843                         *flags |= tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 847                 if (tmp) {
 848                         *flags &= ~tmp;
 849                         goto next;
 850                 }
 851
 852                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 853                                    s1);
 854                 RETURN(-EINVAL);
 855
 856 next:
 857                 /* Find next opt */
 858                 s2 = strchr(s1, ',');
 859                 if (s2 == NULL)
 860                         break;
 861                 s1 = s2 + 1;
 862         }
 863         RETURN(0);
 864 }
 865
 866 void ll_lli_init(struct ll_inode_info *lli)
 867 {
 868         lli->lli_inode_magic = LLI_INODE_MAGIC;
 869         sema_init(&lli->lli_size_sem, 1);
 870         sema_init(&lli->lli_write_sem, 1);
 871         lli->lli_flags = 0;
 872         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 873         spin_lock_init(&lli->lli_lock);
 874         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 875         INIT_LIST_HEAD(&lli->lli_close_list);
 876         lli->lli_inode_magic = LLI_INODE_MAGIC;
 877         sema_init(&lli->lli_och_sem, 1);
 878         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 879         lli->lli_mds_exec_och = NULL;
 880         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 881         lli->lli_open_fd_exec_count = 0;
 882         INIT_LIST_HEAD(&lli->lli_dead_list);
 883         lli->lli_remote_perms = NULL;
 884         lli->lli_rmtperm_utime = 0;
 885         sema_init(&lli->lli_rmtperm_sem, 1);
 886         INIT_LIST_HEAD(&lli->lli_oss_capas);
 887 }
 888
 889 int ll_fill_super(struct super_block *sb)
 890 {
 891         struct lustre_profile *lprof;
 892         struct lustre_sb_info *lsi = s2lsi(sb);
 893         struct ll_sb_info *sbi;
 894         char  *dt = NULL, *md = NULL;
 895         char  *profilenm = get_profile_name(sb);
 896         struct config_llog_instance cfg = {0, };
 897         char   ll_instance[sizeof(sb) * 2 + 1];
 898         int    err;
 899         ENTRY;
 900
 901         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 902
 903         cfs_module_get();
 904
 905         sb->s_type->fs_flags |= FS_ODD_RENAME;
 906         /* client additional sb info */
 907         lsi->lsi_llsbi = sbi = ll_init_sbi();
 908         if (!sbi) {
 909                 cfs_module_put();
 910                 RETURN(-ENOMEM);
 911         }
 912
 913         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 914         if (err)
 915                 GOTO(out_free, err);
 916
 917         /* Generate a string unique to this super, in case some joker tries
 918            to mount the same fs at two mount points.
 919            Use the address of the super itself.*/
 920         sprintf(ll_instance, "%p", sb);
 921         cfg.cfg_instance = ll_instance;
 922         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 923
 924         /* set up client obds */
 925         err = lustre_process_log(sb, profilenm, &cfg);
 926         if (err < 0) {
 927                 CERROR("Unable to process log: %d\n", err);
 928                 GOTO(out_free, err);
 929         }
 930
 931         lprof = class_get_profile(profilenm);
 932         if (lprof == NULL) {
 933                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 934                                    " read from the MGS.  Does that filesystem "
 935                                    "exist?\n", profilenm);
 936                 GOTO(out_free, err = -EINVAL);
 937         }
 938         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 939                lprof->lp_md, lprof->lp_dt);
 940
 941         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 942                   strlen(ll_instance) + 2);
 943         if (!dt)
 944                 GOTO(out_free, err = -ENOMEM);
 945         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 946
 947         OBD_ALLOC(md, strlen(lprof->lp_md) +
 948                   strlen(ll_instance) + 2);
 949         if (!md)
 950                 GOTO(out_free, err = -ENOMEM);
 951         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 952
 953         /* connections, registrations, sb setup */
 954         err = client_common_fill_super(sb, md, dt,
 955                                        lsi->lsi_lmd->lmd_nllu,
 956                                        lsi->lsi_lmd->lmd_nllg);
 957
 958 out_free:
 959         if (md)
 960                 OBD_FREE(md, strlen(md) + 1);
 961         if (dt)
 962                 OBD_FREE(dt, strlen(dt) + 1);
 963         if (err)
 964                 ll_put_super(sb);
 965         else
 966                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 967
 968         RETURN(err);
 969 } /* ll_fill_super */
 970
 971
 972 void ll_put_super(struct super_block *sb)
 973 {
 974         struct config_llog_instance cfg;
 975         char   ll_instance[sizeof(sb) * 2 + 1];
 976         struct obd_device *obd;
 977         struct lustre_sb_info *lsi = s2lsi(sb);
 978         struct ll_sb_info *sbi = ll_s2sbi(sb);
 979         char *profilenm = get_profile_name(sb);
 980         int force = 1, next;
 981         ENTRY;
 982
 983         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 984
 985         ll_print_capa_stat(sbi);
 986
 987         sprintf(ll_instance, "%p", sb);
 988         cfg.cfg_instance = ll_instance;
 989         lustre_end_log(sb, NULL, &cfg);
 990
 991         if (sbi->ll_md_exp) {
 992                 obd = class_exp2obd(sbi->ll_md_exp);
 993                 if (obd)
 994                         force = obd->obd_force;
 995         }
 996
 997         /* We need to set force before the lov_disconnect in
 998            lustre_common_put_super, since l_d cleans up osc's as well. */
 999         if (force) {
1000                 next = 0;
1001                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1002                                                      &next)) != NULL) {
1003                         obd->obd_force = force;
1004                 }
1005         }
1006
1007         if (sbi->ll_lcq) {
1008                 /* Only if client_common_fill_super succeeded */
1009                 client_common_put_super(sb);
1010         }
1011         next = 0;
1012         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1013                 class_manual_cleanup(obd);
1014         }
1015
1016         if (profilenm)
1017                 class_del_profile(profilenm);
1018
1019         ll_free_sbi(sb);
1020         lsi->lsi_llsbi = NULL;
1021
1022         lustre_common_put_super(sb);
1023
1024         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1025
1026         cfs_module_put();
1027
1028         EXIT;
1029 } /* client_put_super */
1030
1031 #ifdef HAVE_REGISTER_CACHE
1032 #include <linux/cache_def.h>
1033 #ifdef HAVE_CACHE_RETURN_INT
1034 static int
1035 #else
1036 static void
1037 #endif
1038 ll_shrink_cache(int priority, unsigned int gfp_mask)
1039 {
1040         struct ll_sb_info *sbi;
1041         int count = 0;
1042
1043         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1044                 count += llap_shrink_cache(sbi, priority);
1045
1046 #ifdef HAVE_CACHE_RETURN_INT
1047         return count;
1048 #endif
1049 }
1050
1051 struct cache_definition ll_cache_definition = {
1052         .name = "llap_cache",
1053         .shrink = ll_shrink_cache
1054 };
1055 #endif /* HAVE_REGISTER_CACHE */
1056
1057 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1058 {
1059         struct inode *inode = NULL;
1060         /* NOTE: we depend on atomic igrab() -bzzz */
1061         lock_res_and_lock(lock);
1062         if (lock->l_ast_data) {
1063                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1064                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1065                         inode = igrab(lock->l_ast_data);
1066                 } else {
1067                         inode = lock->l_ast_data;
1068                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1069                                                 D_INFO : D_WARNING,
1070                                         lock, __FILE__, __func__, __LINE__,
1071                                         "l_ast_data %p is bogus: magic %08x",
1072                                         lock->l_ast_data, lli->lli_inode_magic);
1073                         inode = NULL;
1074                 }
1075         }
1076         unlock_res_and_lock(lock);
1077         return inode;
1078 }
1079
1080 static int null_if_equal(struct ldlm_lock *lock, void *data)
1081 {
1082         if (data == lock->l_ast_data) {
1083                 lock->l_ast_data = NULL;
1084
1085                 if (lock->l_req_mode != lock->l_granted_mode)
1086                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1087         }
1088
1089         return LDLM_ITER_CONTINUE;
1090 }
1091
1092 void ll_clear_inode(struct inode *inode)
1093 {
1094         struct ll_inode_info *lli = ll_i2info(inode);
1095         struct ll_sb_info *sbi = ll_i2sbi(inode);
1096         ENTRY;
1097
1098         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1099                inode->i_generation, inode);
1100
1101         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1102         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1103                          null_if_equal, inode);
1104
1105         LASSERT(!lli->lli_open_fd_write_count);
1106         LASSERT(!lli->lli_open_fd_read_count);
1107         LASSERT(!lli->lli_open_fd_exec_count);
1108
1109         if (lli->lli_mds_write_och)
1110                 ll_md_real_close(inode, FMODE_WRITE);
1111         if (lli->lli_mds_exec_och)
1112                 ll_md_real_close(inode, FMODE_EXEC);
1113         if (lli->lli_mds_read_och)
1114                 ll_md_real_close(inode, FMODE_READ);
1115
1116         if (lli->lli_smd) {
1117                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1118                                   null_if_equal, inode);
1119
1120                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1121                 lli->lli_smd = NULL;
1122         }
1123
1124         if (lli->lli_symlink_name) {
1125                 OBD_FREE(lli->lli_symlink_name,
1126                          strlen(lli->lli_symlink_name) + 1);
1127                 lli->lli_symlink_name = NULL;
1128         }
1129
1130         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1131                 LASSERT(lli->lli_posix_acl == NULL);
1132                 if (lli->lli_remote_perms) {
1133                         free_rmtperm_hash(lli->lli_remote_perms);
1134                         lli->lli_remote_perms = NULL;
1135                 }
1136         }
1137 #ifdef CONFIG_FS_POSIX_ACL
1138         else if (lli->lli_posix_acl) {
1139                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1140                 LASSERT(lli->lli_remote_perms == NULL);
1141                 posix_acl_release(lli->lli_posix_acl);
1142                 lli->lli_posix_acl = NULL;
1143         }
1144 #endif
1145         lli->lli_inode_magic = LLI_INODE_DEAD;
1146
1147 #ifdef HAVE_EXPORT___IGET
1148         spin_lock(&sbi->ll_deathrow_lock);
1149         list_del_init(&lli->lli_dead_list);
1150         spin_unlock(&sbi->ll_deathrow_lock);
1151 #endif
1152         ll_clear_inode_capas(inode);
1153
1154         EXIT;
1155 }
1156
1157 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1158                   struct md_open_data **mod)
1159 {
1160         struct lustre_md md;
1161         struct ll_sb_info *sbi = ll_i2sbi(inode);
1162         struct ptlrpc_request *request = NULL;
1163         int rc;
1164         ENTRY;
1165
1166         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1167                                      LUSTRE_OPC_ANY, NULL);
1168         if (IS_ERR(op_data))
1169                 RETURN(PTR_ERR(op_data));
1170
1171         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1172                         &request, mod);
1173         if (rc) {
1174                 ptlrpc_req_finished(request);
1175                 if (rc == -ENOENT) {
1176                         inode->i_nlink = 0;
1177                         /* Unlinked special device node? Or just a race?
1178                          * Pretend we done everything. */
1179                         if (!S_ISREG(inode->i_mode) &&
1180                             !S_ISDIR(inode->i_mode))
1181                                 rc = inode_setattr(inode, &op_data->op_attr);
1182                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1183                         CERROR("md_setattr fails: rc = %d\n", rc);
1184                 }
1185                 RETURN(rc);
1186         }
1187
1188         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1189                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1190         if (rc) {
1191                 ptlrpc_req_finished(request);
1192                 RETURN(rc);
1193         }
1194
1195         /* We call inode_setattr to adjust timestamps.
1196          * If there is at least some data in file, we cleared ATTR_SIZE
1197          * above to avoid invoking vmtruncate, otherwise it is important
1198          * to call vmtruncate in inode_setattr to update inode->i_size
1199          * (bug 6196) */
1200         rc = inode_setattr(inode, &op_data->op_attr);
1201
1202         /* Extract epoch data if obtained. */
1203         op_data->op_handle = md.body->handle;
1204         op_data->op_ioepoch = md.body->ioepoch;
1205
1206         ll_update_inode(inode, &md);
1207         ptlrpc_req_finished(request);
1208
1209         RETURN(rc);
1210 }
1211
1212 /* Close IO epoch and send Size-on-MDS attribute update. */
1213 static int ll_setattr_done_writing(struct inode *inode,
1214                                    struct md_op_data *op_data,
1215                                    struct md_open_data *mod)
1216 {
1217         struct ll_inode_info *lli = ll_i2info(inode);
1218         int rc = 0;
1219         ENTRY;
1220
1221         LASSERT(op_data != NULL);
1222         if (!S_ISREG(inode->i_mode))
1223                 RETURN(0);
1224
1225         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1226                op_data->op_ioepoch, PFID(&lli->lli_fid));
1227
1228         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1229         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1230         if (rc == -EAGAIN) {
1231                 /* MDS has instructed us to obtain Size-on-MDS attribute
1232                  * from OSTs and send setattr to back to MDS. */
1233                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1234                                          op_data->op_ioepoch);
1235         } else if (rc) {
1236                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1237                        inode->i_ino, rc);
1238         }
1239         RETURN(rc);
1240 }
1241
1242 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1243  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1244  * keep these values until such a time that objects are allocated for it.
1245  * We do the MDS operations first, as it is checking permissions for us.
1246  * We don't to the MDS RPC if there is nothing that we want to store there,
1247  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1248  * going to do an RPC anyways.
1249  *
1250  * If we are doing a truncate, we will send the mtime and ctime updates
1251  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1252  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1253  * at the same time.
1254  */
1255 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1256 {
1257         struct ll_inode_info *lli = ll_i2info(inode);
1258         struct lov_stripe_md *lsm = lli->lli_smd;
1259         struct ll_sb_info *sbi = ll_i2sbi(inode);
1260         struct md_op_data *op_data = NULL;
1261         struct md_open_data *mod = NULL;
1262         int ia_valid = attr->ia_valid;
1263         int rc = 0, rc1 = 0;
1264         ENTRY;
1265
1266         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1267                attr->ia_valid);
1268         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1269
1270         if (ia_valid & ATTR_SIZE) {
1271                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1272                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1273                                attr->ia_size, ll_file_maxbytes(inode));
1274                         RETURN(-EFBIG);
1275                 }
1276
1277                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1278         }
1279
1280         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1281         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1282                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1283                         RETURN(-EPERM);
1284         }
1285
1286         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1287         if (attr->ia_valid & ATTR_CTIME) {
1288                 attr->ia_ctime = CURRENT_TIME;
1289                 attr->ia_valid |= ATTR_CTIME_SET;
1290         }
1291         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1292                 attr->ia_atime = CURRENT_TIME;
1293                 attr->ia_valid |= ATTR_ATIME_SET;
1294         }
1295         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1296                 attr->ia_mtime = CURRENT_TIME;
1297                 attr->ia_valid |= ATTR_MTIME_SET;
1298         }
1299         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1300                 /* To avoid stale mtime on mds, obtain it from ost and send
1301                    to mds. */
1302                 rc = ll_glimpse_size(inode, 0);
1303                 if (rc)
1304                         RETURN(rc);
1305
1306                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1307                 attr->ia_mtime = inode->i_mtime;
1308         }
1309
1310         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1311                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1312                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1313                        CURRENT_SECONDS);
1314
1315         /* NB: ATTR_SIZE will only be set after this point if the size
1316          * resides on the MDS, ie, this file has no objects. */
1317         if (lsm)
1318                 attr->ia_valid &= ~ATTR_SIZE;
1319
1320         /* We always do an MDS RPC, even if we're only changing the size;
1321          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1322
1323         OBD_ALLOC_PTR(op_data);
1324         if (op_data == NULL)
1325                 RETURN(-ENOMEM);
1326
1327         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1328
1329         /* Open epoch for truncate. */
1330         if (ia_valid & ATTR_SIZE)
1331                 op_data->op_flags = MF_EPOCH_OPEN;
1332
1333         rc = ll_md_setattr(inode, op_data, &mod);
1334         if (rc)
1335                 GOTO(out, rc);
1336
1337         if (op_data->op_ioepoch)
1338                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1339                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1340
1341         if (!lsm || !S_ISREG(inode->i_mode)) {
1342                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1343                 GOTO(out, rc = 0);
1344         }
1345
1346         /* We really need to get our PW lock before we change inode->i_size.
1347          * If we don't we can race with other i_size updaters on our node, like
1348          * ll_file_read.  We can also race with i_size propogation to other
1349          * nodes through dirtying and writeback of final cached pages.  This
1350          * last one is especially bad for racing o_append users on other
1351          * nodes. */
1352         if (ia_valid & ATTR_SIZE) {
1353                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1354                                                            OBD_OBJECT_EOF } };
1355                 struct lustre_handle lockh = { 0 };
1356                 int err, ast_flags = 0;
1357                 /* XXX when we fix the AST intents to pass the discard-range
1358                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1359                  * XXX here. */
1360                 if (attr->ia_size == 0)
1361                         ast_flags = LDLM_AST_DISCARD_DATA;
1362
1363                 UNLOCK_INODE_MUTEX(inode);
1364                 UP_WRITE_I_ALLOC_SEM(inode);
1365                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1366                                     ast_flags);
1367 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1368                 DOWN_WRITE_I_ALLOC_SEM(inode);
1369                 LOCK_INODE_MUTEX(inode);
1370 #else
1371                 LOCK_INODE_MUTEX(inode);
1372                 DOWN_WRITE_I_ALLOC_SEM(inode);
1373 #endif
1374                 if (rc != 0)
1375                         GOTO(out, rc);
1376
1377                 /* Only ll_inode_size_lock is taken at this level.
1378                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1379                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1380                  * ll_truncate dropped ll_inode_size_lock() */
1381                 ll_inode_size_lock(inode, 0);
1382                 rc = vmtruncate(inode, attr->ia_size);
1383                 if (rc != 0) {
1384                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1385                         ll_inode_size_unlock(inode, 0);
1386                 }
1387
1388                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1389                 if (err) {
1390                         CERROR("ll_extent_unlock failed: %d\n", err);
1391                         if (!rc)
1392                                 rc = err;
1393                 }
1394         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1395                 obd_flag flags;
1396                 struct obd_info oinfo = { { { 0 } } };
1397                 struct obdo *oa;
1398
1399                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1400                        inode->i_ino, LTIME_S(attr->ia_mtime));
1401
1402                 OBDO_ALLOC(oa);
1403                 if (oa) {
1404                         oa->o_id = lsm->lsm_object_id;
1405                         oa->o_gr = lsm->lsm_object_gr;
1406                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1407
1408                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1409                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1410                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1411                                 OBD_MD_FLGROUP;
1412
1413                         obdo_from_inode(oa, inode, flags);
1414
1415                         oinfo.oi_oa = oa;
1416                         oinfo.oi_md = lsm;
1417                         oinfo.oi_capa = ll_mdscapa_get(inode);
1418
1419                         /* XXX: this looks unnecessary now. */
1420                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1421                         capa_put(oinfo.oi_capa);
1422                         if (rc)
1423                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1424                         OBDO_FREE(oa);
1425                 } else {
1426                         rc = -ENOMEM;
1427                 }
1428         }
1429         EXIT;
1430 out:
1431         if (op_data) {
1432                 if (op_data->op_ioepoch)
1433                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1434                 ll_finish_md_op_data(op_data);
1435         }
1436         return rc ? rc : rc1;
1437 }
1438
1439 int ll_setattr(struct dentry *de, struct iattr *attr)
1440 {
1441         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1442             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1443                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1444
1445         return ll_setattr_raw(de->d_inode, attr);
1446 }
1447
1448 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1449                        __u64 max_age)
1450 {
1451         struct ll_sb_info *sbi = ll_s2sbi(sb);
1452         struct obd_statfs obd_osfs;
1453         int rc;
1454         ENTRY;
1455
1456         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1457         if (rc) {
1458                 CERROR("md_statfs fails: rc = %d\n", rc);
1459                 RETURN(rc);
1460         }
1461
1462         osfs->os_type = sb->s_magic;
1463
1464         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1465                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1466
1467         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1468                               &obd_osfs, max_age);
1469         if (rc) {
1470                 CERROR("obd_statfs fails: rc = %d\n", rc);
1471                 RETURN(rc);
1472         }
1473
1474         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1475                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1476                obd_osfs.os_files);
1477
1478         osfs->os_bsize = obd_osfs.os_bsize;
1479         osfs->os_blocks = obd_osfs.os_blocks;
1480         osfs->os_bfree = obd_osfs.os_bfree;
1481         osfs->os_bavail = obd_osfs.os_bavail;
1482
1483         /* If we don't have as many objects free on the OST as inodes
1484          * on the MDS, we reduce the total number of inodes to
1485          * compensate, so that the "inodes in use" number is correct.
1486          */
1487         if (obd_osfs.os_ffree < osfs->os_ffree) {
1488                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1489                         obd_osfs.os_ffree;
1490                 osfs->os_ffree = obd_osfs.os_ffree;
1491         }
1492
1493         RETURN(rc);
1494 }
1495 #ifndef HAVE_STATFS_DENTRY_PARAM
1496 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1497 {
1498 #else
1499 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1500 {
1501         struct super_block *sb = de->d_sb;
1502 #endif
1503         struct obd_statfs osfs;
1504         int rc;
1505
1506         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1507         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1508
1509         /* For now we will always get up-to-date statfs values, but in the
1510          * future we may allow some amount of caching on the client (e.g.
1511          * from QOS or lprocfs updates). */
1512         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1513         if (rc)
1514                 return rc;
1515
1516         statfs_unpack(sfs, &osfs);
1517
1518         /* We need to downshift for all 32-bit kernels, because we can't
1519          * tell if the kernel is being called via sys_statfs64() or not.
1520          * Stop before overflowing f_bsize - in which case it is better
1521          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1522         if (sizeof(long) < 8) {
1523                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1524                         sfs->f_bsize <<= 1;
1525
1526                         osfs.os_blocks >>= 1;
1527                         osfs.os_bfree >>= 1;
1528                         osfs.os_bavail >>= 1;
1529                 }
1530         }
1531
1532         sfs->f_blocks = osfs.os_blocks;
1533         sfs->f_bfree = osfs.os_bfree;
1534         sfs->f_bavail = osfs.os_bavail;
1535
1536         return 0;
1537 }
1538
1539 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1540 {
1541         struct ll_inode_info *lli;
1542         struct lov_stripe_md *lsm;
1543
1544         lli = ll_i2info(inode);
1545         LASSERT(lli->lli_size_sem_owner != current);
1546         down(&lli->lli_size_sem);
1547         LASSERT(lli->lli_size_sem_owner == NULL);
1548         lli->lli_size_sem_owner = current;
1549         lsm = lli->lli_smd;
1550         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1551                  lsm, lock_lsm);
1552         if (lock_lsm)
1553                 lov_stripe_lock(lsm);
1554 }
1555
1556 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1557 {
1558         struct ll_inode_info *lli;
1559         struct lov_stripe_md *lsm;
1560
1561         lli = ll_i2info(inode);
1562         lsm = lli->lli_smd;
1563         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1564                  lsm, unlock_lsm);
1565         if (unlock_lsm)
1566                 lov_stripe_unlock(lsm);
1567         LASSERT(lli->lli_size_sem_owner == current);
1568         lli->lli_size_sem_owner = NULL;
1569         up(&lli->lli_size_sem);
1570 }
1571
1572 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1573 {
1574         struct ll_inode_info *lli = ll_i2info(inode);
1575
1576         dump_lsm(D_INODE, lsm);
1577         dump_lsm(D_INODE, lli->lli_smd);
1578         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1579                  "lsm must be joined lsm %p\n", lsm);
1580         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1581         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1582                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1583         lli->lli_smd = lsm;
1584         lli->lli_maxbytes = lsm->lsm_maxbytes;
1585         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1586                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1587 }
1588
1589 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1590 {
1591         struct ll_inode_info *lli = ll_i2info(inode);
1592         struct mdt_body *body = md->body;
1593         struct lov_stripe_md *lsm = md->lsm;
1594         struct ll_sb_info *sbi = ll_i2sbi(inode);
1595
1596         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1597         if (lsm != NULL) {
1598                 if (lli->lli_smd == NULL) {
1599                         if (lsm->lsm_magic != LOV_MAGIC &&
1600                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1601                                 dump_lsm(D_ERROR, lsm);
1602                                 LBUG();
1603                         }
1604                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1605                                lsm, inode->i_ino, inode->i_generation, inode);
1606                         /* ll_inode_size_lock() requires it is only called
1607                          * with lli_smd != NULL or lock_lsm == 0 or we can
1608                          * race between lock/unlock.  bug 9547 */
1609                         lli->lli_smd = lsm;
1610                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1611                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1612                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1613                 } else {
1614                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1615                              lli->lli_smd->lsm_stripe_count ==
1616                                         lsm->lsm_stripe_count) {
1617                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1618                                         CERROR("lsm mismatch for inode %ld\n",
1619                                                 inode->i_ino);
1620                                         CERROR("lli_smd:\n");
1621                                         dump_lsm(D_ERROR, lli->lli_smd);
1622                                         CERROR("lsm:\n");
1623                                         dump_lsm(D_ERROR, lsm);
1624                                         LBUG();
1625                                 }
1626                         } else
1627                                 ll_replace_lsm(inode, lsm);
1628                 }
1629                 if (lli->lli_smd != lsm)
1630                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1631         }
1632
1633         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1634                 if (body->valid & OBD_MD_FLRMTPERM)
1635                         ll_update_remote_perm(inode, md->remote_perm);
1636         }
1637 #ifdef CONFIG_FS_POSIX_ACL
1638         else if (body->valid & OBD_MD_FLACL) {
1639                 spin_lock(&lli->lli_lock);
1640                 if (lli->lli_posix_acl)
1641                         posix_acl_release(lli->lli_posix_acl);
1642                 lli->lli_posix_acl = md->posix_acl;
1643                 spin_unlock(&lli->lli_lock);
1644         }
1645 #endif
1646         if (body->valid & OBD_MD_FLATIME &&
1647             body->atime > LTIME_S(inode->i_atime))
1648                 LTIME_S(inode->i_atime) = body->atime;
1649
1650         /* mtime is always updated with ctime, but can be set in past.
1651            As write and utime(2) may happen within 1 second, and utime's
1652            mtime has a priority over write's one, so take mtime from mds
1653            for the same ctimes. */
1654         if (body->valid & OBD_MD_FLCTIME &&
1655             body->ctime >= LTIME_S(inode->i_ctime)) {
1656                 LTIME_S(inode->i_ctime) = body->ctime;
1657                 if (body->valid & OBD_MD_FLMTIME) {
1658                         CDEBUG(D_INODE, "setting ino %lu mtime "
1659                                "from %lu to "LPU64"\n", inode->i_ino,
1660                                LTIME_S(inode->i_mtime), body->mtime);
1661                         LTIME_S(inode->i_mtime) = body->mtime;
1662                 }
1663         }
1664         if (body->valid & OBD_MD_FLMODE)
1665                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1666         if (body->valid & OBD_MD_FLTYPE)
1667                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1668         if (S_ISREG(inode->i_mode)) {
1669                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1670         } else {
1671                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1672         }
1673 #ifdef HAVE_INODE_BLKSIZE
1674         inode->i_blksize = 1<<inode->i_blkbits;
1675 #endif
1676         if (body->valid & OBD_MD_FLUID)
1677                 inode->i_uid = body->uid;
1678         if (body->valid & OBD_MD_FLGID)
1679                 inode->i_gid = body->gid;
1680         if (body->valid & OBD_MD_FLFLAGS)
1681                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1682         if (body->valid & OBD_MD_FLNLINK)
1683                 inode->i_nlink = body->nlink;
1684         if (body->valid & OBD_MD_FLRDEV)
1685 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1686                 inode->i_rdev = body->rdev;
1687 #else
1688                 inode->i_rdev = old_decode_dev(body->rdev);
1689 #endif
1690         if (body->valid & OBD_MD_FLSIZE) {
1691                 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1692                         if (lli->lli_flags & (LLIF_DONE_WRITING |
1693                                               LLIF_EPOCH_PENDING |
1694                                               LLIF_SOM_DIRTY))
1695                           CWARN("ino %lu flags %lu still has size authority!"
1696                                 "do not trust the size got from MDS\n",
1697                                 inode->i_ino, lli->lli_flags);
1698                         else {
1699                                 i_size_write(inode, body->size);
1700                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1701                         }
1702                 } else {
1703                         i_size_write(inode, body->size);
1704                 }
1705
1706                 if (body->valid & OBD_MD_FLBLOCKS)
1707                         inode->i_blocks = body->blocks;
1708         }
1709
1710         if (body->valid & OBD_MD_FLID) {
1711                 /* FID shouldn't be changed! */
1712                 if (fid_is_sane(&lli->lli_fid)) {
1713                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1714                                  "Trying to change FID "DFID
1715                                  " to the "DFID", inode %lu/%u(%p)\n",
1716                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1717                                  inode->i_ino, inode->i_generation, inode);
1718                 } else
1719                         lli->lli_fid = body->fid1;
1720         }
1721
1722         LASSERT(fid_seq(&lli->lli_fid) != 0);
1723
1724         if (body->valid & OBD_MD_FLMDSCAPA) {
1725                 LASSERT(md->mds_capa);
1726                 ll_add_capa(inode, md->mds_capa);
1727         }
1728         if (body->valid & OBD_MD_FLOSSCAPA) {
1729                 LASSERT(md->oss_capa);
1730                 ll_add_capa(inode, md->oss_capa);
1731         }
1732 }
1733
1734 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1735 static struct backing_dev_info ll_backing_dev_info = {
1736         .ra_pages       = 0,    /* No readahead */
1737 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1738         .capabilities   = 0,    /* Does contribute to dirty memory */
1739 #else
1740         .memory_backed  = 0,    /* Does contribute to dirty memory */
1741 #endif
1742 };
1743 #endif
1744
1745 void ll_read_inode2(struct inode *inode, void *opaque)
1746 {
1747         struct lustre_md *md = opaque;
1748         struct ll_inode_info *lli = ll_i2info(inode);
1749         ENTRY;
1750
1751         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1752                inode->i_ino, inode->i_generation, inode);
1753
1754         ll_lli_init(lli);
1755
1756         LASSERT(!lli->lli_smd);
1757
1758         /* Core attributes from the MDS first.  This is a new inode, and
1759          * the VFS doesn't zero times in the core inode so we have to do
1760          * it ourselves.  They will be overwritten by either MDS or OST
1761          * attributes - we just need to make sure they aren't newer. */
1762         LTIME_S(inode->i_mtime) = 0;
1763         LTIME_S(inode->i_atime) = 0;
1764         LTIME_S(inode->i_ctime) = 0;
1765         inode->i_rdev = 0;
1766         ll_update_inode(inode, md);
1767
1768         /* OIDEBUG(inode); */
1769
1770         if (S_ISREG(inode->i_mode)) {
1771                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1772                 inode->i_op = &ll_file_inode_operations;
1773                 inode->i_fop = sbi->ll_fop;
1774                 inode->i_mapping->a_ops = &ll_aops;
1775                 EXIT;
1776         } else if (S_ISDIR(inode->i_mode)) {
1777                 inode->i_op = &ll_dir_inode_operations;
1778                 inode->i_fop = &ll_dir_operations;
1779                 inode->i_mapping->a_ops = &ll_dir_aops;
1780                 EXIT;
1781         } else if (S_ISLNK(inode->i_mode)) {
1782                 inode->i_op = &ll_fast_symlink_inode_operations;
1783                 EXIT;
1784         } else {
1785                 inode->i_op = &ll_special_inode_operations;
1786
1787 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1788                 init_special_inode(inode, inode->i_mode,
1789                                    kdev_t_to_nr(inode->i_rdev));
1790
1791                 /* initializing backing dev info. */
1792                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1793 #else
1794                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1795 #endif
1796                 EXIT;
1797         }
1798 }
1799
1800 void ll_delete_inode(struct inode *inode)
1801 {
1802         struct ll_sb_info *sbi = ll_i2sbi(inode);
1803         int rc;
1804         ENTRY;
1805
1806         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1807         if (rc) {
1808                 CERROR("fid_delete() failed, rc %d\n", rc);
1809         }
1810         clear_inode(inode);
1811
1812         EXIT;
1813 }
1814
1815 int ll_iocontrol(struct inode *inode, struct file *file,
1816                  unsigned int cmd, unsigned long arg)
1817 {
1818         struct ll_sb_info *sbi = ll_i2sbi(inode);
1819         struct ptlrpc_request *req = NULL;
1820         int rc, flags = 0;
1821         ENTRY;
1822
1823         switch(cmd) {
1824         case EXT3_IOC_GETFLAGS: {
1825                 struct mdt_body *body;
1826                 struct obd_capa *oc;
1827
1828                 oc = ll_mdscapa_get(inode);
1829                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1830                                 OBD_MD_FLFLAGS, 0, &req);
1831                 capa_put(oc);
1832                 if (rc) {
1833                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1834                         RETURN(-abs(rc));
1835                 }
1836
1837                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1838                                       sizeof(*body));
1839
1840                 /*Now the ext3 will be packed directly back to client,
1841                  *no need convert here*/
1842                 flags = body->flags;
1843
1844                 ptlrpc_req_finished (req);
1845
1846                 RETURN(put_user(flags, (int *)arg));
1847         }
1848         case EXT3_IOC_SETFLAGS: {
1849                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1850                 struct obd_info oinfo = { { { 0 } } };
1851                 struct md_op_data *op_data;
1852
1853                 if (get_user(flags, (int *)arg))
1854                         RETURN(-EFAULT);
1855
1856                 oinfo.oi_md = lsm;
1857                 OBDO_ALLOC(oinfo.oi_oa);
1858                 if (!oinfo.oi_oa)
1859                         RETURN(-ENOMEM);
1860
1861                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1862                                              LUSTRE_OPC_ANY, NULL);
1863                 if (IS_ERR(op_data))
1864                         RETURN(PTR_ERR(op_data));
1865
1866                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1867                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1868                 rc = md_setattr(sbi->ll_md_exp, op_data,
1869                                 NULL, 0, NULL, 0, &req, NULL);
1870                 ll_finish_md_op_data(op_data);
1871                 ptlrpc_req_finished(req);
1872                 if (rc || lsm == NULL) {
1873                         OBDO_FREE(oinfo.oi_oa);
1874                         RETURN(rc);
1875                 }
1876
1877                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1878                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1879                 oinfo.oi_oa->o_flags = flags;
1880                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1881                                        OBD_MD_FLGROUP;
1882                 oinfo.oi_capa = ll_mdscapa_get(inode);
1883
1884                 obdo_from_inode(oinfo.oi_oa, inode,
1885                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1886                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1887                 capa_put(oinfo.oi_capa);
1888                 OBDO_FREE(oinfo.oi_oa);
1889                 if (rc) {
1890                         if (rc != -EPERM && rc != -EACCES)
1891                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1892                         RETURN(rc);
1893                 }
1894
1895                 inode->i_flags = ll_ext_to_inode_flags(flags |
1896                                                        MDS_BFLAG_EXT_FLAGS);
1897                 RETURN(0);
1898         }
1899         default:
1900                 RETURN(-ENOSYS);
1901         }
1902
1903         RETURN(0);
1904 }
1905
1906 int ll_flush_ctx(struct inode *inode)
1907 {
1908         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1909
1910         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1911
1912         obd_set_info_async(sbi->ll_md_exp,
1913                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1914                            0, NULL, NULL);
1915         obd_set_info_async(sbi->ll_dt_exp,
1916                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1917                            0, NULL, NULL);
1918         return 0;
1919 }
1920
1921 /* umount -f client means force down, don't save state */
1922 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1923 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1924 {
1925         struct super_block *sb = vfsmnt->mnt_sb;
1926 #else
1927 void ll_umount_begin(struct super_block *sb)
1928 {
1929 #endif
1930         struct lustre_sb_info *lsi = s2lsi(sb);
1931         struct ll_sb_info *sbi = ll_s2sbi(sb);
1932         struct obd_device *obd;
1933         struct obd_ioctl_data ioc_data = { 0 };
1934         ENTRY;
1935
1936 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1937         if (!(flags & MNT_FORCE)) {
1938                 EXIT;
1939                 return;
1940         }
1941 #endif
1942
1943         /* Tell the MGC we got umount -f */
1944         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1945
1946         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1947                sb->s_count, atomic_read(&sb->s_active));
1948
1949         obd = class_exp2obd(sbi->ll_md_exp);
1950         if (obd == NULL) {
1951                 CERROR("Invalid MDC connection handle "LPX64"\n",
1952                        sbi->ll_md_exp->exp_handle.h_cookie);
1953                 EXIT;
1954                 return;
1955         }
1956         obd->obd_force = 1;
1957         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1958                       &ioc_data, NULL);
1959
1960         obd = class_exp2obd(sbi->ll_dt_exp);
1961         if (obd == NULL) {
1962                 CERROR("Invalid LOV connection handle "LPX64"\n",
1963                        sbi->ll_dt_exp->exp_handle.h_cookie);
1964                 EXIT;
1965                 return;
1966         }
1967
1968         obd->obd_force = 1;
1969         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1970                       &ioc_data, NULL);
1971
1972         /* Really, we'd like to wait until there are no requests outstanding,
1973          * and then continue.  For now, we just invalidate the requests,
1974          * schedule, and hope.
1975          */
1976         schedule();
1977
1978         EXIT;
1979 }
1980
1981 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1982 {
1983         struct ll_sb_info *sbi = ll_s2sbi(sb);
1984         int err;
1985         __u32 read_only;
1986
1987         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1988                 read_only = *flags & MS_RDONLY;
1989                 err = obd_set_info_async(sbi->ll_md_exp,
1990                                          sizeof(KEY_READ_ONLY) - 1,
1991                                          KEY_READ_ONLY, sizeof(read_only),
1992                                          &read_only, NULL);
1993                 if (err) {
1994                         CERROR("Failed to change the read-only flag during "
1995                                "remount: %d\n", err);
1996                         return err;
1997                 }
1998
1999                 if (read_only)
2000                         sb->s_flags |= MS_RDONLY;
2001                 else
2002                         sb->s_flags &= ~MS_RDONLY;
2003         }
2004         return 0;
2005 }
2006
2007 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2008                   int offset, struct super_block *sb)
2009 {
2010         struct ll_sb_info *sbi = NULL;
2011         struct lustre_md md;
2012         int rc = 0;
2013         ENTRY;
2014
2015         LASSERT(*inode || sb);
2016         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2017         prune_deathrow(sbi, 1);
2018         memset(&md, 0, sizeof(struct lustre_md));
2019
2020         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2021                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2022         if (rc)
2023                 RETURN(rc);
2024
2025         if (*inode) {
2026                 ll_update_inode(*inode, &md);
2027         } else {
2028                 LASSERT(sb != NULL);
2029
2030                 /*
2031                  * At this point server returns to client's same fid as client
2032                  * generated for creating. So using ->fid1 is okay here.
2033                  */
2034                 LASSERT(fid_is_sane(&md.body->fid1));
2035
2036                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2037                 if (*inode == NULL || is_bad_inode(*inode)) {
2038                         if (md.lsm)
2039                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2040 #ifdef CONFIG_FS_POSIX_ACL
2041                         if (md.posix_acl) {
2042                                 posix_acl_release(md.posix_acl);
2043                                 md.posix_acl = NULL;
2044                         }
2045 #endif
2046                         rc = -ENOMEM;
2047                         CERROR("new_inode -fatal: rc %d\n", rc);
2048                         GOTO(out, rc);
2049                 }
2050         }
2051
2052         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2053                          ll_i2info(*inode)->lli_smd);
2054 out:
2055         md_free_lustre_md(sbi->ll_md_exp, &md);
2056         RETURN(rc);
2057 }
2058
2059 char *llap_origins[] = {
2060         [LLAP_ORIGIN_UNKNOWN] = "--",
2061         [LLAP_ORIGIN_READPAGE] = "rp",
2062         [LLAP_ORIGIN_READAHEAD] = "ra",
2063         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2064         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2065 };
2066
2067 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2068                                              struct list_head *list)
2069 {
2070         struct ll_async_page *llap;
2071         struct list_head *pos;
2072
2073         list_for_each(pos, list) {
2074                 if (pos == &sbi->ll_pglist)
2075                         return NULL;
2076                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2077                 if (llap->llap_page == NULL)
2078                         continue;
2079                 return llap;
2080         }
2081         LBUG();
2082         return NULL;
2083 }
2084
2085 int ll_obd_statfs(struct inode *inode, void *arg)
2086 {
2087         struct ll_sb_info *sbi = NULL;
2088         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2089         struct lov_obd *lov = NULL;
2090         struct obd_statfs stat_buf = {0};
2091         char *buf = NULL;
2092         struct obd_ioctl_data *data = NULL;
2093         __u32 type, index;
2094         int len = 0, rc;
2095
2096         if (!inode || !(sbi = ll_i2sbi(inode)))
2097                 GOTO(out_statfs, rc = -EINVAL);
2098
2099         rc = obd_ioctl_getdata(&buf, &len, arg);
2100         if (rc)
2101                 GOTO(out_statfs, rc);
2102
2103         data = (void*)buf;
2104         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2105             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2106                 GOTO(out_statfs, rc = -EINVAL);
2107
2108         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2109         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2110
2111         if (type == LL_STATFS_MDC) {
2112                 if (index > 0)
2113                         GOTO(out_statfs, rc = -ENODEV);
2114                 client_obd = class_exp2obd(sbi->ll_md_exp);
2115         } else if (type == LL_STATFS_LOV) {
2116                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2117                 lov = &lov_obd->u.lov;
2118
2119                 if ((index >= lov->desc.ld_tgt_count))
2120                         GOTO(out_statfs, rc = -ENODEV);
2121                 if (!lov->lov_tgts[index])
2122                         /* Try again with the next index */
2123                         GOTO(out_statfs, rc = -EAGAIN);
2124
2125                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2126                 if (!lov->lov_tgts[index]->ltd_active)
2127                         GOTO(out_uuid, rc = -ENODATA);
2128         }
2129
2130         if (!client_obd)
2131                 GOTO(out_statfs, rc = -EINVAL);
2132
2133         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2134         if (rc)
2135                 GOTO(out_statfs, rc);
2136
2137         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2138                 GOTO(out_statfs, rc = -EFAULT);
2139
2140 out_uuid:
2141         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2142                          data->ioc_plen2))
2143                 rc = -EFAULT;
2144
2145 out_statfs:
2146         if (buf)
2147                 obd_ioctl_freedata(buf, len);
2148         return rc;
2149 }
2150
2151 int ll_process_config(struct lustre_cfg *lcfg)
2152 {
2153         char *ptr;
2154         void *sb;
2155         struct lprocfs_static_vars lvars;
2156         unsigned long x;
2157         int rc = 0;
2158
2159         lprocfs_init_vars(llite, &lvars);
2160
2161         /* The instance name contains the sb: lustre-client-aacfe000 */
2162         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2163         if (!ptr || !*(++ptr))
2164                 return -EINVAL;
2165         if (sscanf(ptr, "%lx", &x) != 1)
2166                 return -EINVAL;
2167         sb = (void *)x;
2168         /* This better be a real Lustre superblock! */
2169         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2170
2171         /* Note we have not called client_common_fill_super yet, so
2172            proc fns must be able to handle that! */
2173         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2174                                       lcfg, sb);
2175         return(rc);
2176 }
2177
2178 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2179 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2180                                        struct inode *i1, struct inode *i2,
2181                                        const char *name, int namelen,
2182                                        int mode, __u32 opc, void *data)
2183 {
2184         LASSERT(i1 != NULL);
2185
2186         if (namelen > ll_i2sbi(i1)->ll_namelen)
2187                 return ERR_PTR(-ENAMETOOLONG);
2188
2189         if (op_data == NULL)
2190                 OBD_ALLOC_PTR(op_data);
2191
2192         if (op_data == NULL)
2193                 return ERR_PTR(-ENOMEM);
2194
2195         ll_i2gids(op_data->op_suppgids, i1, i2);
2196         op_data->op_fid1 = *ll_inode2fid(i1);
2197         op_data->op_capa1 = ll_mdscapa_get(i1);
2198
2199         if (i2) {
2200                 op_data->op_fid2 = *ll_inode2fid(i2);
2201                 op_data->op_capa2 = ll_mdscapa_get(i2);
2202         } else {
2203                 fid_zero(&op_data->op_fid2);
2204         }
2205
2206         op_data->op_name = name;
2207         op_data->op_namelen = namelen;
2208         op_data->op_mode = mode;
2209         op_data->op_mod_time = CURRENT_SECONDS;
2210         op_data->op_fsuid = current->fsuid;
2211         op_data->op_fsgid = current->fsgid;
2212         op_data->op_cap = current->cap_effective;
2213         op_data->op_bias = MDS_CHECK_SPLIT;
2214         op_data->op_opc = opc;
2215         op_data->op_mds = 0;
2216         op_data->op_data = data;
2217
2218         return op_data;
2219 }
2220
2221 void ll_finish_md_op_data(struct md_op_data *op_data)
2222 {
2223         capa_put(op_data->op_capa1);
2224         capa_put(op_data->op_capa2);
2225         OBD_FREE_PTR(op_data);
2226 }
2227
2228 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2229 {
2230         struct ll_sb_info *sbi = ll_i2sbi(inode);
2231         struct ptlrpc_request *req = NULL;
2232         struct mdt_body *body;
2233         char *cmd, *buf;
2234         struct obd_capa *oc;
2235         int rc, buflen;
2236         ENTRY;
2237
2238         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2239                 RETURN(-EBADE);
2240
2241         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2242
2243         OBD_ALLOC(cmd, ioc->cmd_len);
2244         if (!cmd)
2245                 RETURN(-ENOMEM);
2246         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2247                 GOTO(out, rc = -EFAULT);
2248
2249         oc = ll_mdscapa_get(inode);
2250         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2251                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2252                          ioc->cmd_len, ioc->res_len, 0, &req);
2253         capa_put(oc);
2254         if (rc < 0) {
2255                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2256                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2257                 GOTO(out, rc);
2258         }
2259
2260         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2261         LASSERT(body);
2262
2263         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2264         LASSERT(buflen <= ioc->res_len);
2265         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2266         LASSERT(buf);
2267         if (copy_to_user(ioc->res, buf, buflen))
2268                 GOTO(out, rc = -EFAULT);
2269         EXIT;
2270 out:
2271         if (req)
2272                 ptlrpc_req_finished(req);
2273         OBD_FREE(cmd, ioc->cmd_len);
2274         return rc;
2275 }
2276
2277 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2278 {
2279         struct ll_sb_info *sbi = ll_i2sbi(inode);
2280         struct ptlrpc_request *req = NULL;
2281         char *cmd, *buf;
2282         struct obd_capa *oc;
2283         int buflen, rc;
2284         ENTRY;
2285
2286         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2287                 RETURN(-EBADE);
2288
2289         if (!(sbi->ll_flags & LL_SBI_ACL))
2290                 RETURN(-EOPNOTSUPP);
2291
2292         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2293
2294         OBD_ALLOC(cmd, ioc->cmd_len);
2295         if (!cmd)
2296                 RETURN(-ENOMEM);
2297         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2298                 GOTO(out, rc = -EFAULT);
2299
2300         oc = ll_mdscapa_get(inode);
2301         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2302                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2303                          ioc->cmd_len, ioc->res_len, 0, &req);
2304         capa_put(oc);
2305         if (rc) {
2306                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2307                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2308                 GOTO(out, rc);
2309         }
2310
2311         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2312         LASSERT(buflen <= ioc->res_len);
2313         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2314         LASSERT(buf);
2315         if (copy_to_user(ioc->res, buf, buflen))
2316                 GOTO(out, rc = -EFAULT);
2317         EXIT;
2318 out:
2319         if (req)
2320                 ptlrpc_req_finished(req);
2321         OBD_FREE(cmd, ioc->cmd_len);
2322         return rc;
2323 }