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_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  85
  86         spin_lock(&ll_sb_lock);
  87         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  88         spin_unlock(&ll_sb_lock);
  89
  90 #ifdef ENABLE_LLITE_CHECKSUM
  91         sbi->ll_flags |= LL_SBI_CHECKSUM;
  92 #endif
  93
  94 #ifdef HAVE_LRU_RESIZE_SUPPORT
  95         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
  96 #endif
  97
  98 #ifdef HAVE_EXPORT___IGET
  99         INIT_LIST_HEAD(&sbi->ll_deathrow);
 100         spin_lock_init(&sbi->ll_deathrow_lock);
 101 #endif
 102         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 103                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 104                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 105         }
 106
 107         RETURN(sbi);
 108 }
 109
 110 void ll_free_sbi(struct super_block *sb)
 111 {
 112         struct ll_sb_info *sbi = ll_s2sbi(sb);
 113         ENTRY;
 114
 115         if (sbi != NULL) {
 116                 spin_lock(&ll_sb_lock);
 117                 list_del(&sbi->ll_list);
 118                 spin_unlock(&ll_sb_lock);
 119                 OBD_FREE(sbi, sizeof(*sbi));
 120         }
 121         EXIT;
 122 }
 123
 124 static struct dentry_operations ll_d_root_ops = {
 125 #ifdef DCACHE_LUSTRE_INVALID
 126         .d_compare = ll_dcompare,
 127 #endif
 128 };
 129
 130 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 131  * us to make MDS RPCs with large enough reply buffers to hold the
 132  * maximum-sized (= maximum striped) EA and cookie without having to
 133  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 134 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 135 {
 136         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 137         __u32 valsize = sizeof(struct lov_desc);
 138         int rc, easize, def_easize, cookiesize;
 139         struct lov_desc desc;
 140         __u32 stripes;
 141         ENTRY;
 142
 143         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 144                           &valsize, &desc);
 145         if (rc)
 146                 RETURN(rc);
 147
 148         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 149         lsm.lsm_stripe_count = stripes;
 150         easize = obd_size_diskmd(dt_exp, &lsm);
 151
 152         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 153         def_easize = obd_size_diskmd(dt_exp, &lsm);
 154
 155         cookiesize = stripes * sizeof(struct llog_cookie);
 156
 157         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 158                easize, cookiesize);
 159
 160         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 161         RETURN(rc);
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     uid_t nllu, gid_t nllg)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct lu_fid rootfid;
 171         struct obd_capa *oc = NULL;
 172         struct obd_statfs osfs;
 173         struct ptlrpc_request *request = NULL;
 174         struct lustre_handle dt_conn = {0, };
 175         struct lustre_handle md_conn = {0, };
 176         struct obd_connect_data *data = NULL;
 177         struct lustre_md lmd;
 178         obd_valid valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         if (proc_lustre_fs_root) {
 193                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 194                                                   dt, md);
 195                 if (err < 0)
 196                         CERROR("could not register mount in /proc/lustre");
 197         }
 198
 199         /* indicate the features supported by this client */
 200         data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
 201                                   OBD_CONNECT_JOIN |
 202                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
 203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 204                                   OBD_CONNECT_CANCELSET;
 205 #ifdef HAVE_LRU_RESIZE_SUPPORT
 206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208 #endif
 209 #ifdef CONFIG_FS_POSIX_ACL
 210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 211 #endif
 212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 213         data->ocd_version = LUSTRE_VERSION_CODE;
 214
 215         if (sb->s_flags & MS_RDONLY)
 216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 219
 220 #ifdef HAVE_MS_FLOCK_LOCK
 221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 222         sb->s_flags |= MS_FLOCK_LOCK;
 223 #endif
 224
 225         if (sbi->ll_flags & LL_SBI_FLOCK)
 226                 sbi->ll_fop = &ll_file_operations_flock;
 227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 228                 sbi->ll_fop = &ll_file_operations;
 229         else
 230                 sbi->ll_fop = &ll_file_operations_noflock;
 231
 232         /* real client */
 233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 235                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 236                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 237         } else {
 238                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 239                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 240         }
 241         data->ocd_nllu = nllu;
 242         data->ocd_nllg = nllg;
 243
 244         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
 245         if (err == -EBUSY) {
 246                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 247                                    "recovery, of which this client is not a "
 248                                    "part. Please wait for recovery to complete,"
 249                                    " abort, or time out.\n", md);
 250                 GOTO(out, err);
 251         } else if (err) {
 252                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 253                 GOTO(out, err);
 254         }
 255         sbi->ll_md_exp = class_conn2export(&md_conn);
 256
 257         err = obd_fid_init(sbi->ll_md_exp);
 258         if (err) {
 259                 CERROR("Can't init metadata layer FID infrastructure, "
 260                        "rc %d\n", err);
 261                 GOTO(out_md, err);
 262         }
 263
 264         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
 265         if (err)
 266                 GOTO(out_md_fid, err);
 267
 268         size = sizeof(*data);
 269         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 270                            KEY_CONN_DATA,  &size, data);
 271         if (err) {
 272                 CERROR("Get connect data failed: %d \n", err);
 273                 GOTO(out_md, err);
 274         }
 275
 276         LASSERT(osfs.os_bsize);
 277         sb->s_blocksize = osfs.os_bsize;
 278         sb->s_blocksize_bits = log2(osfs.os_bsize);
 279         sb->s_magic = LL_SUPER_MAGIC;
 280
 281         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 282          *         retval = in_file->f_op->sendfile(...);
 283          *         if (*ppos > max)
 284          *                 retval = -EOVERFLOW;
 285          *
 286          * it will check if *ppos is greater than max. However, max equals to
 287          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 288          * has been defined as a signed long long ineger in linux kernel. */
 289 #if BITS_PER_LONG == 64
 290         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 291 #else
 292         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 293 #endif
 294         sbi->ll_namelen = osfs.os_namelen;
 295         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 296
 297         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 298             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 299                 LCONSOLE_INFO("Disabling user_xattr feature because "
 300                               "it is not supported on the server\n");
 301                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 302         }
 303
 304         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 305 #ifdef MS_POSIXACL
 306                 sb->s_flags |= MS_POSIXACL;
 307 #endif
 308                 sbi->ll_flags |= LL_SBI_ACL;
 309         } else {
 310                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 311 #ifdef MS_POSIXACL
 312                 sb->s_flags &= ~MS_POSIXACL;
 313 #endif
 314                 sbi->ll_flags &= ~LL_SBI_ACL;
 315         }
 316
 317         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 318                 sbi->ll_flags |= LL_SBI_JOIN;
 319
 320         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 321                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 322                         /* sometimes local client claims to be remote, but mdt
 323                          * will disagree when client gss not applied. */
 324                         LCONSOLE_INFO("client claims to be remote, but server "
 325                                       "rejected, forced to be local.\n");
 326                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 327                 }
 328         } else {
 329                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 330                         /* with gss applied, remote client can not claim to be
 331                          * local, so mdt maybe force client to be remote. */
 332                         LCONSOLE_INFO("client claims to be local, but server "
 333                                       "rejected, forced to be remote.\n");
 334                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 335                 }
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 339                 LCONSOLE_INFO("client enabled MDS capability!\n");
 340                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 344                 LCONSOLE_INFO("client enabled OSS capability!\n");
 345                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 346         }
 347
 348         sbi->ll_sdev_orig = sb->s_dev;
 349 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 350         /* We set sb->s_dev equal on all lustre clients in order to support
 351          * NFS export clustering.  NFSD requires that the FSID be the same
 352          * on all clients. */
 353         /* s_dev is also used in lt_compare() to compare two fs, but that is
 354          * only a node-local comparison. */
 355
 356         /* XXX: this will not work with LMV */
 357         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 358                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 359 #endif
 360
 361         obd = class_name2obd(dt);
 362         if (!obd) {
 363                 CERROR("DT %s: not setup or attached\n", dt);
 364                 GOTO(out_md_fid, err = -ENODEV);
 365         }
 366
 367         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 368                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 369                                   OBD_CONNECT_CANCELSET;
 370         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 371                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 372
 373 #ifdef HAVE_LRU_RESIZE_SUPPORT
 374         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 375 #endif
 376         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 377                "ocd_grant: %d\n", data->ocd_connect_flags,
 378                data->ocd_version, data->ocd_grant);
 379
 380         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 381         obd->obd_upcall.onu_upcall = ll_ocd_update;
 382         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 383
 384         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
 385         if (err == -EBUSY) {
 386                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 387                                    "recovery, of which this client is not a "
 388                                    "part.  Please wait for recovery to "
 389                                    "complete, abort, or time out.\n", dt);
 390                 GOTO(out_md_fid, err);
 391         } else if (err) {
 392                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 393                 GOTO(out_md_fid, err);
 394         }
 395
 396         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 397
 398         err = obd_fid_init(sbi->ll_dt_exp);
 399         if (err) {
 400                 CERROR("Can't init data layer FID infrastructure, "
 401                        "rc %d\n", err);
 402                 GOTO(out_dt, err);
 403         }
 404
 405         spin_lock(&sbi->ll_lco.lco_lock);
 406         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 407         spin_unlock(&sbi->ll_lco.lco_lock);
 408
 409         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 410
 411         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 412                                   0, NULL, NULL, NULL);
 413         if (err < 0) {
 414                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 415                                    "filesystem. There must be at least one "
 416                                    "active OST for a client to start.\n");
 417                 GOTO(out_dt_fid, err);
 418         }
 419
 420         if (!ll_async_page_slab) {
 421                 ll_async_page_slab_size =
 422                         size_round(sizeof(struct ll_async_page)) + err;
 423                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 424                                                           ll_async_page_slab_size,
 425                                                           0, 0);
 426                 if (!ll_async_page_slab)
 427                         GOTO(out_dt_fid, err = -ENOMEM);
 428         }
 429
 430         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 431         if (err) {
 432                 CERROR("cannot mds_connect: rc = %d\n", err);
 433                 GOTO(out_dt_fid, err);
 434         }
 435         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 436         sbi->ll_root_fid = rootfid;
 437
 438         sb->s_op = &lustre_super_operations;
 439         sb->s_export_op = &lustre_export_operations;
 440
 441         /* make root inode
 442          * XXX: move this to after cbd setup? */
 443         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 444         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 445                 valid |= OBD_MD_FLRMTPERM;
 446         else if (sbi->ll_flags & LL_SBI_ACL)
 447                 valid |= OBD_MD_FLACL;
 448
 449         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 450         if (oc)
 451                 free_capa(oc);
 452         if (err) {
 453                 CERROR("md_getattr failed for root: rc = %d\n", err);
 454                 GOTO(out_dt_fid, err);
 455         }
 456         memset(&lmd, 0, sizeof(lmd));
 457         err = md_get_lustre_md(sbi->ll_md_exp, request,
 458                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
 459                                &lmd);
 460         if (err) {
 461                 CERROR("failed to understand root inode md: rc = %d\n", err);
 462                 ptlrpc_req_finished (request);
 463                 GOTO(out_dt_fid, err);
 464         }
 465
 466         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 467         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 468         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 469         ptlrpc_req_finished(request);
 470
 471         if (root == NULL || is_bad_inode(root)) {
 472                 if (lmd.lsm)
 473                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 474 #ifdef CONFIG_FS_POSIX_ACL
 475                 if (lmd.posix_acl) {
 476                         posix_acl_release(lmd.posix_acl);
 477                         lmd.posix_acl = NULL;
 478                 }
 479 #endif
 480                 CERROR("lustre_lite: bad iget4 for root\n");
 481                 GOTO(out_root, err = -EBADF);
 482         }
 483
 484         err = ll_close_thread_start(&sbi->ll_lcq);
 485         if (err) {
 486                 CERROR("cannot start close thread: rc %d\n", err);
 487                 GOTO(out_root, err);
 488         }
 489
 490         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 491         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 492                                  sizeof(checksum), &checksum, NULL);
 493
 494         sb->s_root = d_alloc_root(root);
 495         if (data != NULL)
 496                 OBD_FREE(data, sizeof(*data));
 497         sb->s_root->d_op = &ll_d_root_ops;
 498         RETURN(err);
 499 out_root:
 500         if (root)
 501                 iput(root);
 502 out_dt_fid:
 503         obd_fid_fini(sbi->ll_dt_exp);
 504 out_dt:
 505         obd_disconnect(sbi->ll_dt_exp);
 506         sbi->ll_dt_exp = NULL;
 507 out_md_fid:
 508         obd_fid_fini(sbi->ll_md_exp);
 509 out_md:
 510         obd_disconnect(sbi->ll_md_exp);
 511         sbi->ll_md_exp = NULL;
 512 out:
 513         if (data != NULL)
 514                 OBD_FREE_PTR(data);
 515         lprocfs_unregister_mountpoint(sbi);
 516         return err;
 517 }
 518
 519 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 520 {
 521         int size, rc;
 522
 523         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 524         size = sizeof(int);
 525         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 526                           &size, lmmsize);
 527         if (rc)
 528                 CERROR("Get max mdsize error rc %d \n", rc);
 529
 530         RETURN(rc);
 531 }
 532
 533 void ll_dump_inode(struct inode *inode)
 534 {
 535         struct list_head *tmp;
 536         int dentry_count = 0;
 537
 538         LASSERT(inode != NULL);
 539
 540         list_for_each(tmp, &inode->i_dentry)
 541                 dentry_count++;
 542
 543         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 544                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 545                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 546 }
 547
 548 void lustre_dump_dentry(struct dentry *dentry, int recur)
 549 {
 550         struct list_head *tmp;
 551         int subdirs = 0;
 552
 553         LASSERT(dentry != NULL);
 554
 555         list_for_each(tmp, &dentry->d_subdirs)
 556                 subdirs++;
 557
 558         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 559                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 560                dentry->d_name.len, dentry->d_name.name,
 561                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 562                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 563                dentry->d_flags, dentry->d_fsdata, subdirs);
 564         if (dentry->d_inode != NULL)
 565                 ll_dump_inode(dentry->d_inode);
 566
 567         if (recur == 0)
 568                 return;
 569
 570         list_for_each(tmp, &dentry->d_subdirs) {
 571                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 572                 lustre_dump_dentry(d, recur - 1);
 573         }
 574 }
 575
 576 #ifdef HAVE_EXPORT___IGET
 577 static void prune_dir_dentries(struct inode *inode)
 578 {
 579         struct dentry *dentry, *prev = NULL;
 580
 581         /* due to lustre specific logic, a directory
 582          * can have few dentries - a bug from VFS POV */
 583 restart:
 584         spin_lock(&dcache_lock);
 585         if (!list_empty(&inode->i_dentry)) {
 586                 dentry = list_entry(inode->i_dentry.prev,
 587                                     struct dentry, d_alias);
 588                 /* in order to prevent infinite loops we
 589                  * break if previous dentry is busy */
 590                 if (dentry != prev) {
 591                         prev = dentry;
 592                         dget_locked(dentry);
 593                         spin_unlock(&dcache_lock);
 594
 595                         /* try to kill all child dentries */
 596                         lock_dentry(dentry);
 597                         shrink_dcache_parent(dentry);
 598                         unlock_dentry(dentry);
 599                         dput(dentry);
 600
 601                         /* now try to get rid of current dentry */
 602                         d_prune_aliases(inode);
 603                         goto restart;
 604                 }
 605         }
 606         spin_unlock(&dcache_lock);
 607 }
 608
 609 static void prune_deathrow_one(struct ll_inode_info *lli)
 610 {
 611         struct inode *inode = ll_info2i(lli);
 612
 613         /* first, try to drop any dentries - they hold a ref on the inode */
 614         if (S_ISDIR(inode->i_mode))
 615                 prune_dir_dentries(inode);
 616         else
 617                 d_prune_aliases(inode);
 618
 619
 620         /* if somebody still uses it, leave it */
 621         LASSERT(atomic_read(&inode->i_count) > 0);
 622         if (atomic_read(&inode->i_count) > 1)
 623                 goto out;
 624
 625         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 626                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 627
 628         /* seems nobody uses it anymore */
 629         inode->i_nlink = 0;
 630
 631 out:
 632         iput(inode);
 633         return;
 634 }
 635
 636 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 637 {
 638         struct ll_inode_info *lli;
 639         int empty;
 640
 641         do {
 642                 if (need_resched() && try)
 643                         break;
 644
 645                 if (try) {
 646                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 647                                 break;
 648                 } else {
 649                         spin_lock(&sbi->ll_deathrow_lock);
 650                 }
 651
 652                 empty = 1;
 653                 lli = NULL;
 654                 if (!list_empty(&sbi->ll_deathrow)) {
 655                         lli = list_entry(sbi->ll_deathrow.next,
 656                                          struct ll_inode_info,
 657                                          lli_dead_list);
 658                         list_del_init(&lli->lli_dead_list);
 659                         if (!list_empty(&sbi->ll_deathrow))
 660                                 empty = 0;
 661                 }
 662                 spin_unlock(&sbi->ll_deathrow_lock);
 663
 664                 if (lli)
 665                         prune_deathrow_one(lli);
 666
 667         } while (empty == 0);
 668 }
 669 #else /* !HAVE_EXPORT___IGET */
 670 #define prune_deathrow(sbi, try) do {} while (0)
 671 #endif /* HAVE_EXPORT___IGET */
 672
 673 void client_common_put_super(struct super_block *sb)
 674 {
 675         struct ll_sb_info *sbi = ll_s2sbi(sb);
 676         ENTRY;
 677
 678         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 679
 680         ll_close_thread_shutdown(sbi->ll_lcq);
 681
 682         /* destroy inodes in deathrow */
 683         prune_deathrow(sbi, 0);
 684
 685         list_del(&sbi->ll_conn_chain);
 686
 687         obd_fid_fini(sbi->ll_dt_exp);
 688         obd_disconnect(sbi->ll_dt_exp);
 689         sbi->ll_dt_exp = NULL;
 690
 691         lprocfs_unregister_mountpoint(sbi);
 692
 693         obd_fid_fini(sbi->ll_md_exp);
 694         obd_disconnect(sbi->ll_md_exp);
 695         sbi->ll_md_exp = NULL;
 696
 697         EXIT;
 698 }
 699
 700 void ll_kill_super(struct super_block *sb)
 701 {
 702         struct ll_sb_info *sbi;
 703
 704         ENTRY;
 705
 706         /* not init sb ?*/
 707         if (!(sb->s_flags & MS_ACTIVE))
 708                 return;
 709
 710         sbi = ll_s2sbi(sb);
 711         /* we need restore s_dev from changed for clustred NFS before put_super
 712          * because new kernels have cached s_dev and change sb->s_dev in
 713          * put_super not affected real removing devices */
 714         if (sbi)
 715                 sb->s_dev = sbi->ll_sdev_orig;
 716         EXIT;
 717 }
 718
 719 char *ll_read_opt(const char *opt, char *data)
 720 {
 721         char *value;
 722         char *retval;
 723         ENTRY;
 724
 725         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 726         if (strncmp(opt, data, strlen(opt)))
 727                 RETURN(NULL);
 728         if ((value = strchr(data, '=')) == NULL)
 729                 RETURN(NULL);
 730
 731         value++;
 732         OBD_ALLOC(retval, strlen(value) + 1);
 733         if (!retval) {
 734                 CERROR("out of memory!\n");
 735                 RETURN(NULL);
 736         }
 737
 738         memcpy(retval, value, strlen(value)+1);
 739         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 740         RETURN(retval);
 741 }
 742
 743 static inline int ll_set_opt(const char *opt, char *data, int fl)
 744 {
 745         if (strncmp(opt, data, strlen(opt)) != 0)
 746                 return(0);
 747         else
 748                 return(fl);
 749 }
 750
 751 /* non-client-specific mount options are parsed in lmd_parse */
 752 static int ll_options(char *options, int *flags)
 753 {
 754         int tmp;
 755         char *s1 = options, *s2;
 756         ENTRY;
 757
 758         if (!options)
 759                 RETURN(0);
 760
 761         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 762
 763         while (*s1) {
 764                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 765                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 766                 if (tmp) {
 767                         *flags |= tmp;
 768                         goto next;
 769                 }
 770                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 771                 if (tmp) {
 772                         *flags |= tmp;
 773                         goto next;
 774                 }
 775                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 776                 if (tmp) {
 777                         *flags |= tmp;
 778                         goto next;
 779                 }
 780                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 781                 if (tmp) {
 782                         *flags &= ~tmp;
 783                         goto next;
 784                 }
 785                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 786                 if (tmp) {
 787                         *flags |= tmp;
 788                         goto next;
 789                 }
 790                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 791                 if (tmp) {
 792                         *flags &= ~tmp;
 793                         goto next;
 794                 }
 795                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 796                 if (tmp) {
 797                         /* Ignore deprecated mount option.  The client will
 798                          * always try to mount with ACL support, whether this
 799                          * is used depends on whether server supports it. */
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 803                 if (tmp) {
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811
 812                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 818                 if (tmp) {
 819                         *flags &= ~tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 828                 if (tmp) {
 829                         *flags &= ~tmp;
 830                         goto next;
 831                 }
 832
 833                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 834                                    s1);
 835                 RETURN(-EINVAL);
 836
 837 next:
 838                 /* Find next opt */
 839                 s2 = strchr(s1, ',');
 840                 if (s2 == NULL)
 841                         break;
 842                 s1 = s2 + 1;
 843         }
 844         RETURN(0);
 845 }
 846
 847 void ll_lli_init(struct ll_inode_info *lli)
 848 {
 849         lli->lli_inode_magic = LLI_INODE_MAGIC;
 850         sema_init(&lli->lli_size_sem, 1);
 851         sema_init(&lli->lli_write_sem, 1);
 852         lli->lli_flags = 0;
 853         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 854         spin_lock_init(&lli->lli_lock);
 855         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 856         INIT_LIST_HEAD(&lli->lli_close_list);
 857         lli->lli_inode_magic = LLI_INODE_MAGIC;
 858         sema_init(&lli->lli_och_sem, 1);
 859         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 860         lli->lli_mds_exec_och = NULL;
 861         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 862         lli->lli_open_fd_exec_count = 0;
 863         INIT_LIST_HEAD(&lli->lli_dead_list);
 864         lli->lli_remote_perms = NULL;
 865         lli->lli_rmtperm_utime = 0;
 866         sema_init(&lli->lli_rmtperm_sem, 1);
 867         INIT_LIST_HEAD(&lli->lli_oss_capas);
 868 }
 869
 870 int ll_fill_super(struct super_block *sb)
 871 {
 872         struct lustre_profile *lprof;
 873         struct lustre_sb_info *lsi = s2lsi(sb);
 874         struct ll_sb_info *sbi;
 875         char  *dt = NULL, *md = NULL;
 876         char  *profilenm = get_profile_name(sb);
 877         struct config_llog_instance cfg = {0, };
 878         char   ll_instance[sizeof(sb) * 2 + 1];
 879         int    err;
 880         ENTRY;
 881
 882         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 883
 884         cfs_module_get();
 885
 886         sb->s_type->fs_flags |= FS_ODD_RENAME;
 887         /* client additional sb info */
 888         lsi->lsi_llsbi = sbi = ll_init_sbi();
 889         if (!sbi) {
 890                 cfs_module_put();
 891                 RETURN(-ENOMEM);
 892         }
 893
 894         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 895         if (err)
 896                 GOTO(out_free, err);
 897
 898         /* Generate a string unique to this super, in case some joker tries
 899            to mount the same fs at two mount points.
 900            Use the address of the super itself.*/
 901         sprintf(ll_instance, "%p", sb);
 902         cfg.cfg_instance = ll_instance;
 903         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 904
 905         /* set up client obds */
 906         err = lustre_process_log(sb, profilenm, &cfg);
 907         if (err < 0) {
 908                 CERROR("Unable to process log: %d\n", err);
 909                 GOTO(out_free, err);
 910         }
 911
 912         lprof = class_get_profile(profilenm);
 913         if (lprof == NULL) {
 914                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 915                                    " read from the MGS.  Does that filesystem "
 916                                    "exist?\n", profilenm);
 917                 GOTO(out_free, err = -EINVAL);
 918         }
 919         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 920                lprof->lp_md, lprof->lp_dt);
 921
 922         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 923                   strlen(ll_instance) + 2);
 924         if (!dt)
 925                 GOTO(out_free, err = -ENOMEM);
 926         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 927
 928         OBD_ALLOC(md, strlen(lprof->lp_md) +
 929                   strlen(ll_instance) + 2);
 930         if (!md)
 931                 GOTO(out_free, err = -ENOMEM);
 932         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 933
 934         /* connections, registrations, sb setup */
 935         err = client_common_fill_super(sb, md, dt,
 936                                        lsi->lsi_lmd->lmd_nllu,
 937                                        lsi->lsi_lmd->lmd_nllg);
 938
 939 out_free:
 940         if (md)
 941                 OBD_FREE(md, strlen(md) + 1);
 942         if (dt)
 943                 OBD_FREE(dt, strlen(dt) + 1);
 944         if (err)
 945                 ll_put_super(sb);
 946         else
 947                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 948
 949         RETURN(err);
 950 } /* ll_fill_super */
 951
 952
 953 void ll_put_super(struct super_block *sb)
 954 {
 955         struct config_llog_instance cfg;
 956         char   ll_instance[sizeof(sb) * 2 + 1];
 957         struct obd_device *obd;
 958         struct lustre_sb_info *lsi = s2lsi(sb);
 959         struct ll_sb_info *sbi = ll_s2sbi(sb);
 960         char *profilenm = get_profile_name(sb);
 961         int force = 1, next;
 962         ENTRY;
 963
 964         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 965
 966         ll_print_capa_stat(sbi);
 967
 968         sprintf(ll_instance, "%p", sb);
 969         cfg.cfg_instance = ll_instance;
 970         lustre_end_log(sb, NULL, &cfg);
 971
 972         if (sbi->ll_md_exp) {
 973                 obd = class_exp2obd(sbi->ll_md_exp);
 974                 if (obd)
 975                         force = obd->obd_force;
 976         }
 977
 978         /* We need to set force before the lov_disconnect in
 979            lustre_common_put_super, since l_d cleans up osc's as well. */
 980         if (force) {
 981                 next = 0;
 982                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 983                                                      &next)) != NULL) {
 984                         obd->obd_force = force;
 985                 }
 986         }
 987
 988         if (sbi->ll_lcq) {
 989                 /* Only if client_common_fill_super succeeded */
 990                 client_common_put_super(sb);
 991         }
 992         next = 0;
 993         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 994                 class_manual_cleanup(obd);
 995         }
 996
 997         if (profilenm)
 998                 class_del_profile(profilenm);
 999
1000         ll_free_sbi(sb);
1001         lsi->lsi_llsbi = NULL;
1002
1003         lustre_common_put_super(sb);
1004
1005         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1006
1007         cfs_module_put();
1008
1009         EXIT;
1010 } /* client_put_super */
1011
1012 #ifdef HAVE_REGISTER_CACHE
1013 #include <linux/cache_def.h>
1014 #ifdef HAVE_CACHE_RETURN_INT
1015 static int
1016 #else
1017 static void
1018 #endif
1019 ll_shrink_cache(int priority, unsigned int gfp_mask)
1020 {
1021         struct ll_sb_info *sbi;
1022         int count = 0;
1023
1024         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1025                 count += llap_shrink_cache(sbi, priority);
1026
1027 #ifdef HAVE_CACHE_RETURN_INT
1028         return count;
1029 #endif
1030 }
1031
1032 struct cache_definition ll_cache_definition = {
1033         .name = "llap_cache",
1034         .shrink = ll_shrink_cache
1035 };
1036 #endif /* HAVE_REGISTER_CACHE */
1037
1038 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1039 {
1040         struct inode *inode = NULL;
1041         /* NOTE: we depend on atomic igrab() -bzzz */
1042         lock_res_and_lock(lock);
1043         if (lock->l_ast_data) {
1044                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1045                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1046                         inode = igrab(lock->l_ast_data);
1047                 } else {
1048                         inode = lock->l_ast_data;
1049                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1050                                                 D_INFO : D_WARNING,
1051                                         lock, __FILE__, __func__, __LINE__,
1052                                         "l_ast_data %p is bogus: magic %08x",
1053                                         lock->l_ast_data, lli->lli_inode_magic);
1054                         inode = NULL;
1055                 }
1056         }
1057         unlock_res_and_lock(lock);
1058         return inode;
1059 }
1060
1061 static int null_if_equal(struct ldlm_lock *lock, void *data)
1062 {
1063         if (data == lock->l_ast_data) {
1064                 lock->l_ast_data = NULL;
1065
1066                 if (lock->l_req_mode != lock->l_granted_mode)
1067                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1068         }
1069
1070         return LDLM_ITER_CONTINUE;
1071 }
1072
1073 void ll_clear_inode(struct inode *inode)
1074 {
1075         struct ll_inode_info *lli = ll_i2info(inode);
1076         struct ll_sb_info *sbi = ll_i2sbi(inode);
1077         ENTRY;
1078
1079         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1080                inode->i_generation, inode);
1081
1082         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1083         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1084                          null_if_equal, inode);
1085
1086         LASSERT(!lli->lli_open_fd_write_count);
1087         LASSERT(!lli->lli_open_fd_read_count);
1088         LASSERT(!lli->lli_open_fd_exec_count);
1089
1090         if (lli->lli_mds_write_och)
1091                 ll_md_real_close(inode, FMODE_WRITE);
1092         if (lli->lli_mds_exec_och)
1093                 ll_md_real_close(inode, FMODE_EXEC);
1094         if (lli->lli_mds_read_och)
1095                 ll_md_real_close(inode, FMODE_READ);
1096
1097         if (lli->lli_smd) {
1098                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1099                                   null_if_equal, inode);
1100
1101                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1102                 lli->lli_smd = NULL;
1103         }
1104
1105         if (lli->lli_symlink_name) {
1106                 OBD_FREE(lli->lli_symlink_name,
1107                          strlen(lli->lli_symlink_name) + 1);
1108                 lli->lli_symlink_name = NULL;
1109         }
1110
1111         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1112                 LASSERT(lli->lli_posix_acl == NULL);
1113                 if (lli->lli_remote_perms) {
1114                         free_rmtperm_hash(lli->lli_remote_perms);
1115                         lli->lli_remote_perms = NULL;
1116                 }
1117         }
1118 #ifdef CONFIG_FS_POSIX_ACL
1119         else if (lli->lli_posix_acl) {
1120                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1121                 LASSERT(lli->lli_remote_perms == NULL);
1122                 posix_acl_release(lli->lli_posix_acl);
1123                 lli->lli_posix_acl = NULL;
1124         }
1125 #endif
1126         lli->lli_inode_magic = LLI_INODE_DEAD;
1127
1128 #ifdef HAVE_EXPORT___IGET
1129         spin_lock(&sbi->ll_deathrow_lock);
1130         list_del_init(&lli->lli_dead_list);
1131         spin_unlock(&sbi->ll_deathrow_lock);
1132 #endif
1133         ll_clear_inode_capas(inode);
1134
1135         EXIT;
1136 }
1137
1138 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1139                   struct md_open_data **mod)
1140 {
1141         struct lustre_md md;
1142         struct ll_sb_info *sbi = ll_i2sbi(inode);
1143         struct ptlrpc_request *request = NULL;
1144         int rc;
1145         ENTRY;
1146
1147         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1148                                      LUSTRE_OPC_ANY, NULL);
1149         if (IS_ERR(op_data))
1150                 RETURN(PTR_ERR(op_data));
1151
1152         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1153                         &request, mod);
1154         if (rc) {
1155                 ptlrpc_req_finished(request);
1156                 if (rc == -ENOENT) {
1157                         inode->i_nlink = 0;
1158                         /* Unlinked special device node? Or just a race?
1159                          * Pretend we done everything. */
1160                         if (!S_ISREG(inode->i_mode) &&
1161                             !S_ISDIR(inode->i_mode))
1162                                 rc = inode_setattr(inode, &op_data->op_attr);
1163                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1164                         CERROR("md_setattr fails: rc = %d\n", rc);
1165                 }
1166                 RETURN(rc);
1167         }
1168
1169         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1170                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1171         if (rc) {
1172                 ptlrpc_req_finished(request);
1173                 RETURN(rc);
1174         }
1175
1176         /* We call inode_setattr to adjust timestamps.
1177          * If there is at least some data in file, we cleared ATTR_SIZE
1178          * above to avoid invoking vmtruncate, otherwise it is important
1179          * to call vmtruncate in inode_setattr to update inode->i_size
1180          * (bug 6196) */
1181         rc = inode_setattr(inode, &op_data->op_attr);
1182
1183         /* Extract epoch data if obtained. */
1184         op_data->op_handle = md.body->handle;
1185         op_data->op_ioepoch = md.body->ioepoch;
1186
1187         ll_update_inode(inode, &md);
1188         ptlrpc_req_finished(request);
1189
1190         RETURN(rc);
1191 }
1192
1193 /* Close IO epoch and send Size-on-MDS attribute update. */
1194 static int ll_setattr_done_writing(struct inode *inode,
1195                                    struct md_op_data *op_data,
1196                                    struct md_open_data *mod)
1197 {
1198         struct ll_inode_info *lli = ll_i2info(inode);
1199         int rc = 0;
1200         ENTRY;
1201
1202         LASSERT(op_data != NULL);
1203         if (!S_ISREG(inode->i_mode))
1204                 RETURN(0);
1205
1206         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1207                op_data->op_ioepoch, PFID(&lli->lli_fid));
1208
1209         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1210         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1211         if (rc == -EAGAIN) {
1212                 /* MDS has instructed us to obtain Size-on-MDS attribute
1213                  * from OSTs and send setattr to back to MDS. */
1214                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1215                                          op_data->op_ioepoch);
1216         } else if (rc) {
1217                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1218                        inode->i_ino, rc);
1219         }
1220         RETURN(rc);
1221 }
1222
1223 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1224  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1225  * keep these values until such a time that objects are allocated for it.
1226  * We do the MDS operations first, as it is checking permissions for us.
1227  * We don't to the MDS RPC if there is nothing that we want to store there,
1228  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1229  * going to do an RPC anyways.
1230  *
1231  * If we are doing a truncate, we will send the mtime and ctime updates
1232  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1233  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1234  * at the same time.
1235  */
1236 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1237 {
1238         struct ll_inode_info *lli = ll_i2info(inode);
1239         struct lov_stripe_md *lsm = lli->lli_smd;
1240         struct ll_sb_info *sbi = ll_i2sbi(inode);
1241         struct md_op_data *op_data = NULL;
1242         struct md_open_data *mod = NULL;
1243         int ia_valid = attr->ia_valid;
1244         int rc = 0, rc1 = 0;
1245         ENTRY;
1246
1247         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1248                attr->ia_valid);
1249         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1250
1251         if (ia_valid & ATTR_SIZE) {
1252                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1253                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1254                                attr->ia_size, ll_file_maxbytes(inode));
1255                         RETURN(-EFBIG);
1256                 }
1257
1258                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1259         }
1260
1261         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1262         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1263                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1264                         RETURN(-EPERM);
1265         }
1266
1267         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1268         if (attr->ia_valid & ATTR_CTIME) {
1269                 attr->ia_ctime = CURRENT_TIME;
1270                 attr->ia_valid |= ATTR_CTIME_SET;
1271         }
1272         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1273                 attr->ia_atime = CURRENT_TIME;
1274                 attr->ia_valid |= ATTR_ATIME_SET;
1275         }
1276         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1277                 attr->ia_mtime = CURRENT_TIME;
1278                 attr->ia_valid |= ATTR_MTIME_SET;
1279         }
1280         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1281                 /* To avoid stale mtime on mds, obtain it from ost and send
1282                    to mds. */
1283                 rc = ll_glimpse_size(inode, 0);
1284                 if (rc)
1285                         RETURN(rc);
1286
1287                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1288                 attr->ia_mtime = inode->i_mtime;
1289         }
1290
1291         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1292                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1293                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1294                        CURRENT_SECONDS);
1295
1296         /* NB: ATTR_SIZE will only be set after this point if the size
1297          * resides on the MDS, ie, this file has no objects. */
1298         if (lsm)
1299                 attr->ia_valid &= ~ATTR_SIZE;
1300
1301         /* We always do an MDS RPC, even if we're only changing the size;
1302          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1303
1304         OBD_ALLOC_PTR(op_data);
1305         if (op_data == NULL)
1306                 RETURN(-ENOMEM);
1307
1308         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1309
1310         /* Open epoch for truncate. */
1311         if (ia_valid & ATTR_SIZE)
1312                 op_data->op_flags = MF_EPOCH_OPEN;
1313
1314         rc = ll_md_setattr(inode, op_data, &mod);
1315         if (rc)
1316                 GOTO(out, rc);
1317
1318         if (op_data->op_ioepoch)
1319                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1320                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1321
1322         if (!lsm || !S_ISREG(inode->i_mode)) {
1323                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1324                 GOTO(out, rc = 0);
1325         }
1326
1327         /* We really need to get our PW lock before we change inode->i_size.
1328          * If we don't we can race with other i_size updaters on our node, like
1329          * ll_file_read.  We can also race with i_size propogation to other
1330          * nodes through dirtying and writeback of final cached pages.  This
1331          * last one is especially bad for racing o_append users on other
1332          * nodes. */
1333         if (ia_valid & ATTR_SIZE) {
1334                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1335                                                            OBD_OBJECT_EOF } };
1336                 struct lustre_handle lockh = { 0 };
1337                 int err, ast_flags = 0;
1338                 /* XXX when we fix the AST intents to pass the discard-range
1339                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1340                  * XXX here. */
1341                 if (attr->ia_size == 0)
1342                         ast_flags = LDLM_AST_DISCARD_DATA;
1343
1344                 UNLOCK_INODE_MUTEX(inode);
1345                 UP_WRITE_I_ALLOC_SEM(inode);
1346                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1347                                     ast_flags);
1348                 LOCK_INODE_MUTEX(inode);
1349                 DOWN_WRITE_I_ALLOC_SEM(inode);
1350
1351                 if (rc != 0)
1352                         GOTO(out, rc);
1353
1354                 /* Only ll_inode_size_lock is taken at this level.
1355                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1356                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1357                  * ll_truncate dropped ll_inode_size_lock() */
1358                 ll_inode_size_lock(inode, 0);
1359                 rc = vmtruncate(inode, attr->ia_size);
1360                 if (rc != 0) {
1361                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1362                         ll_inode_size_unlock(inode, 0);
1363                 }
1364
1365                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1366                 if (err) {
1367                         CERROR("ll_extent_unlock failed: %d\n", err);
1368                         if (!rc)
1369                                 rc = err;
1370                 }
1371         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1372                 obd_flag flags;
1373                 struct obd_info oinfo = { { { 0 } } };
1374                 struct obdo *oa;
1375
1376                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1377                        inode->i_ino, LTIME_S(attr->ia_mtime));
1378
1379                 OBDO_ALLOC(oa);
1380                 if (oa) {
1381                         oa->o_id = lsm->lsm_object_id;
1382                         oa->o_gr = lsm->lsm_object_gr;
1383                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1384
1385                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1386                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1387                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1388                                 OBD_MD_FLGROUP;
1389
1390                         obdo_from_inode(oa, inode, flags);
1391
1392                         oinfo.oi_oa = oa;
1393                         oinfo.oi_md = lsm;
1394                         oinfo.oi_capa = ll_mdscapa_get(inode);
1395
1396                         /* XXX: this looks unnecessary now. */
1397                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1398                         capa_put(oinfo.oi_capa);
1399                         if (rc)
1400                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1401                         OBDO_FREE(oa);
1402                 } else {
1403                         rc = -ENOMEM;
1404                 }
1405         }
1406         EXIT;
1407 out:
1408         if (op_data) {
1409                 if (op_data->op_ioepoch)
1410                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1411                 ll_finish_md_op_data(op_data);
1412         }
1413         return rc ? rc : rc1;
1414 }
1415
1416 int ll_setattr(struct dentry *de, struct iattr *attr)
1417 {
1418         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1419             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1420                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1421
1422         return ll_setattr_raw(de->d_inode, attr);
1423 }
1424
1425 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1426                        __u64 max_age)
1427 {
1428         struct ll_sb_info *sbi = ll_s2sbi(sb);
1429         struct obd_statfs obd_osfs;
1430         int rc;
1431         ENTRY;
1432
1433         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1434         if (rc) {
1435                 CERROR("md_statfs fails: rc = %d\n", rc);
1436                 RETURN(rc);
1437         }
1438
1439         osfs->os_type = sb->s_magic;
1440
1441         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1442                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1443
1444         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1445                               &obd_osfs, max_age, 0);
1446         if (rc) {
1447                 CERROR("obd_statfs fails: rc = %d\n", rc);
1448                 RETURN(rc);
1449         }
1450
1451         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1452                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1453                obd_osfs.os_files);
1454
1455         osfs->os_bsize = obd_osfs.os_bsize;
1456         osfs->os_blocks = obd_osfs.os_blocks;
1457         osfs->os_bfree = obd_osfs.os_bfree;
1458         osfs->os_bavail = obd_osfs.os_bavail;
1459
1460         /* If we don't have as many objects free on the OST as inodes
1461          * on the MDS, we reduce the total number of inodes to
1462          * compensate, so that the "inodes in use" number is correct.
1463          */
1464         if (obd_osfs.os_ffree < osfs->os_ffree) {
1465                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1466                         obd_osfs.os_ffree;
1467                 osfs->os_ffree = obd_osfs.os_ffree;
1468         }
1469
1470         RETURN(rc);
1471 }
1472 #ifndef HAVE_STATFS_DENTRY_PARAM
1473 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1474 {
1475 #else
1476 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1477 {
1478         struct super_block *sb = de->d_sb;
1479 #endif
1480         struct obd_statfs osfs;
1481         int rc;
1482
1483         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1484         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1485
1486         /* For now we will always get up-to-date statfs values, but in the
1487          * future we may allow some amount of caching on the client (e.g.
1488          * from QOS or lprocfs updates). */
1489         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1490         if (rc)
1491                 return rc;
1492
1493         statfs_unpack(sfs, &osfs);
1494
1495         /* We need to downshift for all 32-bit kernels, because we can't
1496          * tell if the kernel is being called via sys_statfs64() or not.
1497          * Stop before overflowing f_bsize - in which case it is better
1498          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1499         if (sizeof(long) < 8) {
1500                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1501                         sfs->f_bsize <<= 1;
1502
1503                         osfs.os_blocks >>= 1;
1504                         osfs.os_bfree >>= 1;
1505                         osfs.os_bavail >>= 1;
1506                 }
1507         }
1508
1509         sfs->f_blocks = osfs.os_blocks;
1510         sfs->f_bfree = osfs.os_bfree;
1511         sfs->f_bavail = osfs.os_bavail;
1512
1513         return 0;
1514 }
1515
1516 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1517 {
1518         struct ll_inode_info *lli;
1519         struct lov_stripe_md *lsm;
1520
1521         lli = ll_i2info(inode);
1522         LASSERT(lli->lli_size_sem_owner != current);
1523         down(&lli->lli_size_sem);
1524         LASSERT(lli->lli_size_sem_owner == NULL);
1525         lli->lli_size_sem_owner = current;
1526         lsm = lli->lli_smd;
1527         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1528                  lsm, lock_lsm);
1529         if (lock_lsm)
1530                 lov_stripe_lock(lsm);
1531 }
1532
1533 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1534 {
1535         struct ll_inode_info *lli;
1536         struct lov_stripe_md *lsm;
1537
1538         lli = ll_i2info(inode);
1539         lsm = lli->lli_smd;
1540         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1541                  lsm, unlock_lsm);
1542         if (unlock_lsm)
1543                 lov_stripe_unlock(lsm);
1544         LASSERT(lli->lli_size_sem_owner == current);
1545         lli->lli_size_sem_owner = NULL;
1546         up(&lli->lli_size_sem);
1547 }
1548
1549 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1550 {
1551         struct ll_inode_info *lli = ll_i2info(inode);
1552
1553         dump_lsm(D_INODE, lsm);
1554         dump_lsm(D_INODE, lli->lli_smd);
1555         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1556                  "lsm must be joined lsm %p\n", lsm);
1557         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1558         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1559                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1560         lli->lli_smd = lsm;
1561         lli->lli_maxbytes = lsm->lsm_maxbytes;
1562         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1563                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1564 }
1565
1566 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1567 {
1568         struct ll_inode_info *lli = ll_i2info(inode);
1569         struct mdt_body *body = md->body;
1570         struct lov_stripe_md *lsm = md->lsm;
1571         struct ll_sb_info *sbi = ll_i2sbi(inode);
1572
1573         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1574         if (lsm != NULL) {
1575                 if (lli->lli_smd == NULL) {
1576                         if (lsm->lsm_magic != LOV_MAGIC &&
1577                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1578                                 dump_lsm(D_ERROR, lsm);
1579                                 LBUG();
1580                         }
1581                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1582                                lsm, inode->i_ino, inode->i_generation, inode);
1583                         /* ll_inode_size_lock() requires it is only called
1584                          * with lli_smd != NULL or lock_lsm == 0 or we can
1585                          * race between lock/unlock.  bug 9547 */
1586                         lli->lli_smd = lsm;
1587                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1588                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1589                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1590                 } else {
1591                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1592                              lli->lli_smd->lsm_stripe_count ==
1593                                         lsm->lsm_stripe_count) {
1594                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1595                                         CERROR("lsm mismatch for inode %ld\n",
1596                                                 inode->i_ino);
1597                                         CERROR("lli_smd:\n");
1598                                         dump_lsm(D_ERROR, lli->lli_smd);
1599                                         CERROR("lsm:\n");
1600                                         dump_lsm(D_ERROR, lsm);
1601                                         LBUG();
1602                                 }
1603                         } else
1604                                 ll_replace_lsm(inode, lsm);
1605                 }
1606                 if (lli->lli_smd != lsm)
1607                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1608         }
1609
1610         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1611                 if (body->valid & OBD_MD_FLRMTPERM)
1612                         ll_update_remote_perm(inode, md->remote_perm);
1613         }
1614 #ifdef CONFIG_FS_POSIX_ACL
1615         else if (body->valid & OBD_MD_FLACL) {
1616                 spin_lock(&lli->lli_lock);
1617                 if (lli->lli_posix_acl)
1618                         posix_acl_release(lli->lli_posix_acl);
1619                 lli->lli_posix_acl = md->posix_acl;
1620                 spin_unlock(&lli->lli_lock);
1621         }
1622 #endif
1623         if (body->valid & OBD_MD_FLATIME &&
1624             body->atime > LTIME_S(inode->i_atime))
1625                 LTIME_S(inode->i_atime) = body->atime;
1626
1627         /* mtime is always updated with ctime, but can be set in past.
1628            As write and utime(2) may happen within 1 second, and utime's
1629            mtime has a priority over write's one, so take mtime from mds
1630            for the same ctimes. */
1631         if (body->valid & OBD_MD_FLCTIME &&
1632             body->ctime >= LTIME_S(inode->i_ctime)) {
1633                 LTIME_S(inode->i_ctime) = body->ctime;
1634                 if (body->valid & OBD_MD_FLMTIME) {
1635                         CDEBUG(D_INODE, "setting ino %lu mtime "
1636                                "from %lu to "LPU64"\n", inode->i_ino,
1637                                LTIME_S(inode->i_mtime), body->mtime);
1638                         LTIME_S(inode->i_mtime) = body->mtime;
1639                 }
1640         }
1641         if (body->valid & OBD_MD_FLMODE)
1642                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1643         if (body->valid & OBD_MD_FLTYPE)
1644                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1645         if (S_ISREG(inode->i_mode)) {
1646                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1647         } else {
1648                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1649         }
1650 #ifdef HAVE_INODE_BLKSIZE
1651         inode->i_blksize = 1<<inode->i_blkbits;
1652 #endif
1653         if (body->valid & OBD_MD_FLUID)
1654                 inode->i_uid = body->uid;
1655         if (body->valid & OBD_MD_FLGID)
1656                 inode->i_gid = body->gid;
1657         if (body->valid & OBD_MD_FLFLAGS)
1658                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1659         if (body->valid & OBD_MD_FLNLINK)
1660                 inode->i_nlink = body->nlink;
1661         if (body->valid & OBD_MD_FLRDEV)
1662                 inode->i_rdev = old_decode_dev(body->rdev);
1663
1664         if (body->valid & OBD_MD_FLID) {
1665                 /* FID shouldn't be changed! */
1666                 if (fid_is_sane(&lli->lli_fid)) {
1667                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1668                                  "Trying to change FID "DFID
1669                                  " to the "DFID", inode %lu/%u(%p)\n",
1670                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1671                                  inode->i_ino, inode->i_generation, inode);
1672                 } else
1673                         lli->lli_fid = body->fid1;
1674         }
1675
1676         LASSERT(fid_seq(&lli->lli_fid) != 0);
1677
1678         if (body->valid & OBD_MD_FLSIZE) {
1679                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1680                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1681                         struct lustre_handle lockh;
1682                         ldlm_mode_t mode;
1683
1684                         /* As it is possible a blocking ast has been processed
1685                          * by this time, we need to check there is an UPDATE
1686                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1687                          * it. */
1688                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1689                                                &lockh);
1690                         if (mode) {
1691                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1692                                                       LLIF_EPOCH_PENDING |
1693                                                       LLIF_SOM_DIRTY)) {
1694                                         CERROR("ino %lu flags %lu still has "
1695                                                "size authority! do not trust "
1696                                                "the size got from MDS\n",
1697                                                inode->i_ino, lli->lli_flags);
1698                                 } else {
1699                                         /* Use old size assignment to avoid
1700                                          * deadlock bz14138 & bz14326 */
1701                                         inode->i_size = body->size;
1702                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1703                                 }
1704                                 ldlm_lock_decref(&lockh, mode);
1705                         }
1706                 } else {
1707                         /* Use old size assignment to avoid
1708                          * deadlock bz14138 & bz14326 */
1709                         inode->i_size = body->size;
1710                 }
1711
1712                 if (body->valid & OBD_MD_FLBLOCKS)
1713                         inode->i_blocks = body->blocks;
1714         }
1715
1716         if (body->valid & OBD_MD_FLMDSCAPA) {
1717                 LASSERT(md->mds_capa);
1718                 ll_add_capa(inode, md->mds_capa);
1719         }
1720         if (body->valid & OBD_MD_FLOSSCAPA) {
1721                 LASSERT(md->oss_capa);
1722                 ll_add_capa(inode, md->oss_capa);
1723         }
1724 }
1725
1726 static struct backing_dev_info ll_backing_dev_info = {
1727         .ra_pages       = 0,    /* No readahead */
1728 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1729         .capabilities   = 0,    /* Does contribute to dirty memory */
1730 #else
1731         .memory_backed  = 0,    /* Does contribute to dirty memory */
1732 #endif
1733 };
1734
1735 void ll_read_inode2(struct inode *inode, void *opaque)
1736 {
1737         struct lustre_md *md = opaque;
1738         struct ll_inode_info *lli = ll_i2info(inode);
1739         ENTRY;
1740
1741         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1742                inode->i_ino, inode->i_generation, inode);
1743
1744         ll_lli_init(lli);
1745
1746         LASSERT(!lli->lli_smd);
1747
1748         /* Core attributes from the MDS first.  This is a new inode, and
1749          * the VFS doesn't zero times in the core inode so we have to do
1750          * it ourselves.  They will be overwritten by either MDS or OST
1751          * attributes - we just need to make sure they aren't newer. */
1752         LTIME_S(inode->i_mtime) = 0;
1753         LTIME_S(inode->i_atime) = 0;
1754         LTIME_S(inode->i_ctime) = 0;
1755         inode->i_rdev = 0;
1756         ll_update_inode(inode, md);
1757
1758         /* OIDEBUG(inode); */
1759
1760         if (S_ISREG(inode->i_mode)) {
1761                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1762                 inode->i_op = &ll_file_inode_operations;
1763                 inode->i_fop = sbi->ll_fop;
1764                 inode->i_mapping->a_ops = &ll_aops;
1765                 EXIT;
1766         } else if (S_ISDIR(inode->i_mode)) {
1767                 inode->i_op = &ll_dir_inode_operations;
1768                 inode->i_fop = &ll_dir_operations;
1769                 inode->i_mapping->a_ops = &ll_dir_aops;
1770                 EXIT;
1771         } else if (S_ISLNK(inode->i_mode)) {
1772                 inode->i_op = &ll_fast_symlink_inode_operations;
1773                 EXIT;
1774         } else {
1775                 inode->i_op = &ll_special_inode_operations;
1776
1777                 init_special_inode(inode, inode->i_mode,
1778                                    kdev_t_to_nr(inode->i_rdev));
1779
1780                 /* initializing backing dev info. */
1781                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1782
1783                 EXIT;
1784         }
1785 }
1786
1787 void ll_delete_inode(struct inode *inode)
1788 {
1789         struct ll_sb_info *sbi = ll_i2sbi(inode);
1790         int rc;
1791         ENTRY;
1792
1793         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1794         if (rc) {
1795                 CERROR("fid_delete() failed, rc %d\n", rc);
1796         }
1797         clear_inode(inode);
1798
1799         EXIT;
1800 }
1801
1802 int ll_iocontrol(struct inode *inode, struct file *file,
1803                  unsigned int cmd, unsigned long arg)
1804 {
1805         struct ll_sb_info *sbi = ll_i2sbi(inode);
1806         struct ptlrpc_request *req = NULL;
1807         int rc, flags = 0;
1808         ENTRY;
1809
1810         switch(cmd) {
1811         case EXT3_IOC_GETFLAGS: {
1812                 struct mdt_body *body;
1813                 struct obd_capa *oc;
1814
1815                 oc = ll_mdscapa_get(inode);
1816                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1817                                 OBD_MD_FLFLAGS, 0, &req);
1818                 capa_put(oc);
1819                 if (rc) {
1820                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1821                         RETURN(-abs(rc));
1822                 }
1823
1824                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1825                                       sizeof(*body));
1826                 flags = body->flags;
1827
1828                 ptlrpc_req_finished (req);
1829
1830                 RETURN(put_user(flags, (int *)arg));
1831         }
1832         case EXT3_IOC_SETFLAGS: {
1833                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1834                 struct obd_info oinfo = { { { 0 } } };
1835                 struct md_op_data *op_data;
1836
1837                 if (get_user(flags, (int *)arg))
1838                         RETURN(-EFAULT);
1839
1840                 oinfo.oi_md = lsm;
1841                 OBDO_ALLOC(oinfo.oi_oa);
1842                 if (!oinfo.oi_oa)
1843                         RETURN(-ENOMEM);
1844
1845                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1846                                              LUSTRE_OPC_ANY, NULL);
1847                 if (IS_ERR(op_data))
1848                         RETURN(PTR_ERR(op_data));
1849
1850                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1851                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1852                 rc = md_setattr(sbi->ll_md_exp, op_data,
1853                                 NULL, 0, NULL, 0, &req, NULL);
1854                 ll_finish_md_op_data(op_data);
1855                 ptlrpc_req_finished(req);
1856                 if (rc || lsm == NULL) {
1857                         OBDO_FREE(oinfo.oi_oa);
1858                         RETURN(rc);
1859                 }
1860
1861                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1862                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1863                 oinfo.oi_oa->o_flags = flags;
1864                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1865                                        OBD_MD_FLGROUP;
1866                 oinfo.oi_capa = ll_mdscapa_get(inode);
1867
1868                 obdo_from_inode(oinfo.oi_oa, inode,
1869                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1870                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1871                 capa_put(oinfo.oi_capa);
1872                 OBDO_FREE(oinfo.oi_oa);
1873                 if (rc) {
1874                         if (rc != -EPERM && rc != -EACCES)
1875                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1876                         RETURN(rc);
1877                 }
1878
1879                 inode->i_flags = ll_ext_to_inode_flags(flags |
1880                                                        MDS_BFLAG_EXT_FLAGS);
1881                 RETURN(0);
1882         }
1883         default:
1884                 RETURN(-ENOSYS);
1885         }
1886
1887         RETURN(0);
1888 }
1889
1890 int ll_flush_ctx(struct inode *inode)
1891 {
1892         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1893
1894         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1895
1896         obd_set_info_async(sbi->ll_md_exp,
1897                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1898                            0, NULL, NULL);
1899         obd_set_info_async(sbi->ll_dt_exp,
1900                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1901                            0, NULL, NULL);
1902         return 0;
1903 }
1904
1905 /* umount -f client means force down, don't save state */
1906 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1907 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1908 {
1909         struct super_block *sb = vfsmnt->mnt_sb;
1910 #else
1911 void ll_umount_begin(struct super_block *sb)
1912 {
1913 #endif
1914         struct lustre_sb_info *lsi = s2lsi(sb);
1915         struct ll_sb_info *sbi = ll_s2sbi(sb);
1916         struct obd_device *obd;
1917         struct obd_ioctl_data ioc_data = { 0 };
1918         ENTRY;
1919
1920 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1921         if (!(flags & MNT_FORCE)) {
1922                 EXIT;
1923                 return;
1924         }
1925 #endif
1926
1927         /* Tell the MGC we got umount -f */
1928         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1929
1930         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1931                sb->s_count, atomic_read(&sb->s_active));
1932
1933         obd = class_exp2obd(sbi->ll_md_exp);
1934         if (obd == NULL) {
1935                 CERROR("Invalid MDC connection handle "LPX64"\n",
1936                        sbi->ll_md_exp->exp_handle.h_cookie);
1937                 EXIT;
1938                 return;
1939         }
1940         obd->obd_force = 1;
1941         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1942                       &ioc_data, NULL);
1943
1944         obd = class_exp2obd(sbi->ll_dt_exp);
1945         if (obd == NULL) {
1946                 CERROR("Invalid LOV connection handle "LPX64"\n",
1947                        sbi->ll_dt_exp->exp_handle.h_cookie);
1948                 EXIT;
1949                 return;
1950         }
1951
1952         obd->obd_force = 1;
1953         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1954                       &ioc_data, NULL);
1955
1956         /* Really, we'd like to wait until there are no requests outstanding,
1957          * and then continue.  For now, we just invalidate the requests,
1958          * schedule, and hope.
1959          */
1960         schedule();
1961
1962         EXIT;
1963 }
1964
1965 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1966 {
1967         struct ll_sb_info *sbi = ll_s2sbi(sb);
1968         int err;
1969         __u32 read_only;
1970
1971         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1972                 read_only = *flags & MS_RDONLY;
1973                 err = obd_set_info_async(sbi->ll_md_exp,
1974                                          sizeof(KEY_READ_ONLY) - 1,
1975                                          KEY_READ_ONLY, sizeof(read_only),
1976                                          &read_only, NULL);
1977                 if (err) {
1978                         CERROR("Failed to change the read-only flag during "
1979                                "remount: %d\n", err);
1980                         return err;
1981                 }
1982
1983                 if (read_only)
1984                         sb->s_flags |= MS_RDONLY;
1985                 else
1986                         sb->s_flags &= ~MS_RDONLY;
1987         }
1988         return 0;
1989 }
1990
1991 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1992                   int offset, struct super_block *sb)
1993 {
1994         struct ll_sb_info *sbi = NULL;
1995         struct lustre_md md;
1996         int rc = 0;
1997         ENTRY;
1998
1999         LASSERT(*inode || sb);
2000         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2001         prune_deathrow(sbi, 1);
2002         memset(&md, 0, sizeof(struct lustre_md));
2003
2004         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2005                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2006         if (rc)
2007                 RETURN(rc);
2008
2009         if (*inode) {
2010                 ll_update_inode(*inode, &md);
2011         } else {
2012                 LASSERT(sb != NULL);
2013
2014                 /*
2015                  * At this point server returns to client's same fid as client
2016                  * generated for creating. So using ->fid1 is okay here.
2017                  */
2018                 LASSERT(fid_is_sane(&md.body->fid1));
2019
2020                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2021                 if (*inode == NULL || is_bad_inode(*inode)) {
2022                         if (md.lsm)
2023                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2024 #ifdef CONFIG_FS_POSIX_ACL
2025                         if (md.posix_acl) {
2026                                 posix_acl_release(md.posix_acl);
2027                                 md.posix_acl = NULL;
2028                         }
2029 #endif
2030                         rc = -ENOMEM;
2031                         CERROR("new_inode -fatal: rc %d\n", rc);
2032                         GOTO(out, rc);
2033                 }
2034         }
2035
2036         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2037                          ll_i2info(*inode)->lli_smd);
2038 out:
2039         md_free_lustre_md(sbi->ll_md_exp, &md);
2040         RETURN(rc);
2041 }
2042
2043 char *llap_origins[] = {
2044         [LLAP_ORIGIN_UNKNOWN] = "--",
2045         [LLAP_ORIGIN_READPAGE] = "rp",
2046         [LLAP_ORIGIN_READAHEAD] = "ra",
2047         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2048         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2049 };
2050
2051 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2052                                              struct list_head *list)
2053 {
2054         struct ll_async_page *llap;
2055         struct list_head *pos;
2056
2057         list_for_each(pos, list) {
2058                 if (pos == &sbi->ll_pglist)
2059                         return NULL;
2060                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2061                 if (llap->llap_page == NULL)
2062                         continue;
2063                 return llap;
2064         }
2065         LBUG();
2066         return NULL;
2067 }
2068
2069 int ll_obd_statfs(struct inode *inode, void *arg)
2070 {
2071         struct ll_sb_info *sbi = NULL;
2072         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2073         struct lov_obd *lov = NULL;
2074         struct obd_statfs stat_buf = {0};
2075         char *buf = NULL;
2076         struct obd_ioctl_data *data = NULL;
2077         __u32 type, index;
2078         int len = 0, rc;
2079
2080         if (!inode || !(sbi = ll_i2sbi(inode)))
2081                 GOTO(out_statfs, rc = -EINVAL);
2082
2083         rc = obd_ioctl_getdata(&buf, &len, arg);
2084         if (rc)
2085                 GOTO(out_statfs, rc);
2086
2087         data = (void*)buf;
2088         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2089             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2090                 GOTO(out_statfs, rc = -EINVAL);
2091
2092         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2093         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2094
2095         if (type == LL_STATFS_MDC) {
2096                 if (index > 0)
2097                         GOTO(out_statfs, rc = -ENODEV);
2098                 client_obd = class_exp2obd(sbi->ll_md_exp);
2099         } else if (type == LL_STATFS_LOV) {
2100                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2101                 lov = &lov_obd->u.lov;
2102
2103                 if ((index >= lov->desc.ld_tgt_count))
2104                         GOTO(out_statfs, rc = -ENODEV);
2105                 if (!lov->lov_tgts[index])
2106                         /* Try again with the next index */
2107                         GOTO(out_statfs, rc = -EAGAIN);
2108
2109                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2110                 if (!lov->lov_tgts[index]->ltd_active)
2111                         GOTO(out_uuid, rc = -ENODATA);
2112         }
2113
2114         if (!client_obd)
2115                 GOTO(out_statfs, rc = -EINVAL);
2116
2117         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2118         if (rc)
2119                 GOTO(out_statfs, rc);
2120
2121         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2122                 GOTO(out_statfs, rc = -EFAULT);
2123
2124 out_uuid:
2125         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2126                          data->ioc_plen2))
2127                 rc = -EFAULT;
2128
2129 out_statfs:
2130         if (buf)
2131                 obd_ioctl_freedata(buf, len);
2132         return rc;
2133 }
2134
2135 int ll_process_config(struct lustre_cfg *lcfg)
2136 {
2137         char *ptr;
2138         void *sb;
2139         struct lprocfs_static_vars lvars;
2140         unsigned long x;
2141         int rc = 0;
2142
2143         lprocfs_llite_init_vars(&lvars);
2144
2145         /* The instance name contains the sb: lustre-client-aacfe000 */
2146         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2147         if (!ptr || !*(++ptr))
2148                 return -EINVAL;
2149         if (sscanf(ptr, "%lx", &x) != 1)
2150                 return -EINVAL;
2151         sb = (void *)x;
2152         /* This better be a real Lustre superblock! */
2153         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2154
2155         /* Note we have not called client_common_fill_super yet, so
2156            proc fns must be able to handle that! */
2157         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2158                                       lcfg, sb);
2159         return(rc);
2160 }
2161
2162 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2163 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2164                                        struct inode *i1, struct inode *i2,
2165                                        const char *name, int namelen,
2166                                        int mode, __u32 opc, void *data)
2167 {
2168         LASSERT(i1 != NULL);
2169
2170         if (namelen > ll_i2sbi(i1)->ll_namelen)
2171                 return ERR_PTR(-ENAMETOOLONG);
2172
2173         if (op_data == NULL)
2174                 OBD_ALLOC_PTR(op_data);
2175
2176         if (op_data == NULL)
2177                 return ERR_PTR(-ENOMEM);
2178
2179         ll_i2gids(op_data->op_suppgids, i1, i2);
2180         op_data->op_fid1 = *ll_inode2fid(i1);
2181         op_data->op_capa1 = ll_mdscapa_get(i1);
2182
2183         if (i2) {
2184                 op_data->op_fid2 = *ll_inode2fid(i2);
2185                 op_data->op_capa2 = ll_mdscapa_get(i2);
2186         } else {
2187                 fid_zero(&op_data->op_fid2);
2188         }
2189
2190         op_data->op_name = name;
2191         op_data->op_namelen = namelen;
2192         op_data->op_mode = mode;
2193         op_data->op_mod_time = CURRENT_SECONDS;
2194         op_data->op_fsuid = current->fsuid;
2195         op_data->op_fsgid = current->fsgid;
2196         op_data->op_cap = current->cap_effective;
2197         op_data->op_bias = MDS_CHECK_SPLIT;
2198         op_data->op_opc = opc;
2199         op_data->op_mds = 0;
2200         op_data->op_data = data;
2201
2202         return op_data;
2203 }
2204
2205 void ll_finish_md_op_data(struct md_op_data *op_data)
2206 {
2207         capa_put(op_data->op_capa1);
2208         capa_put(op_data->op_capa2);
2209         OBD_FREE_PTR(op_data);
2210 }
2211
2212 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2213 {
2214         struct ll_sb_info *sbi = ll_i2sbi(inode);
2215         struct ptlrpc_request *req = NULL;
2216         struct mdt_body *body;
2217         char *cmd, *buf;
2218         struct obd_capa *oc;
2219         int rc, buflen;
2220         ENTRY;
2221
2222         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2223                 RETURN(-EBADE);
2224
2225         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2226
2227         OBD_ALLOC(cmd, ioc->cmd_len);
2228         if (!cmd)
2229                 RETURN(-ENOMEM);
2230         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2231                 GOTO(out, rc = -EFAULT);
2232
2233         oc = ll_mdscapa_get(inode);
2234         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2235                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2236                          ioc->cmd_len, ioc->res_len, 0, &req);
2237         capa_put(oc);
2238         if (rc < 0) {
2239                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2240                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2241                 GOTO(out, rc);
2242         }
2243
2244         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2245         LASSERT(body);
2246
2247         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2248         LASSERT(buflen <= ioc->res_len);
2249         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2250         LASSERT(buf);
2251         if (copy_to_user(ioc->res, buf, buflen))
2252                 GOTO(out, rc = -EFAULT);
2253         EXIT;
2254 out:
2255         if (req)
2256                 ptlrpc_req_finished(req);
2257         OBD_FREE(cmd, ioc->cmd_len);
2258         return rc;
2259 }
2260
2261 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2262 {
2263         struct ll_sb_info *sbi = ll_i2sbi(inode);
2264         struct ptlrpc_request *req = NULL;
2265         char *cmd, *buf;
2266         struct obd_capa *oc;
2267         int buflen, rc;
2268         ENTRY;
2269
2270         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2271                 RETURN(-EBADE);
2272
2273         if (!(sbi->ll_flags & LL_SBI_ACL))
2274                 RETURN(-EOPNOTSUPP);
2275
2276         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2277
2278         OBD_ALLOC(cmd, ioc->cmd_len);
2279         if (!cmd)
2280                 RETURN(-ENOMEM);
2281         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2282                 GOTO(out, rc = -EFAULT);
2283
2284         oc = ll_mdscapa_get(inode);
2285         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2286                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2287                          ioc->cmd_len, ioc->res_len, 0, &req);
2288         capa_put(oc);
2289         if (rc) {
2290                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2291                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2292                 GOTO(out, rc);
2293         }
2294
2295         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2296         LASSERT(buflen <= ioc->res_len);
2297         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2298         LASSERT(buf);
2299         if (copy_to_user(ioc->res, buf, buflen))
2300                 GOTO(out, rc = -EFAULT);
2301         EXIT;
2302 out:
2303         if (req)
2304                 ptlrpc_req_finished(req);
2305         OBD_FREE(cmd, ioc->cmd_len);
2306         return rc;
2307 }