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