lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002-2005 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <lustre_lite.h>
  32 #include <lustre_ha.h>
  33 #include <lustre_dlm.h>
  34 #include <lprocfs_status.h>
  35 #include <lustre_disk.h>
  36 #include <lustre_param.h>
  37 #include <lustre_log.h>
  38 #include "llite_internal.h"
  39
  40 cfs_mem_cache_t *ll_file_data_slab;
  41
  42 LIST_HEAD(ll_super_blocks);
  43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52
  53 static struct ll_sb_info *ll_init_sbi(void)
  54 {
  55         struct ll_sb_info *sbi = NULL;
  56         class_uuid_t uuid;
  57         int i;
  58         ENTRY;
  59
  60         OBD_ALLOC(sbi, sizeof(*sbi));
  61         if (!sbi)
  62                 RETURN(NULL);
  63
  64         spin_lock_init(&sbi->ll_lock);
  65         spin_lock_init(&sbi->ll_lco.lco_lock);
  66         spin_lock_init(&sbi->ll_pp_extent_lock);
  67         spin_lock_init(&sbi->ll_process_lock);
  68         sbi->ll_rw_stats_on = 0;
  69         INIT_LIST_HEAD(&sbi->ll_pglist);
  70         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
  71                 sbi->ll_async_page_max = num_physpages / 2;
  72         else
  73                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  74         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  75                                            SBI_DEFAULT_READAHEAD_MAX);
  76         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  77                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  78
  79         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  80         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
  81
  82         ll_generate_random_uuid(uuid);
  83         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  84         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  85
  86         spin_lock(&ll_sb_lock);
  87         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  88         spin_unlock(&ll_sb_lock);
  89
  90 #ifdef ENABLE_LLITE_CHECKSUM
  91         sbi->ll_flags |= LL_SBI_CHECKSUM;
  92 #endif
  93
  94 #ifdef HAVE_EXPORT___IGET
  95         INIT_LIST_HEAD(&sbi->ll_deathrow);
  96         spin_lock_init(&sbi->ll_deathrow_lock);
  97 #endif
  98         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
  99                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 100                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 101         }
 102
 103         RETURN(sbi);
 104 }
 105
 106 void ll_free_sbi(struct super_block *sb)
 107 {
 108         struct ll_sb_info *sbi = ll_s2sbi(sb);
 109         ENTRY;
 110
 111         if (sbi != NULL) {
 112                 spin_lock(&ll_sb_lock);
 113                 list_del(&sbi->ll_list);
 114                 spin_unlock(&ll_sb_lock);
 115                 OBD_FREE(sbi, sizeof(*sbi));
 116         }
 117         EXIT;
 118 }
 119
 120 static struct dentry_operations ll_d_root_ops = {
 121 #ifdef LUSTRE_KERNEL_VERSION
 122         .d_compare = ll_dcompare,
 123 #endif
 124 };
 125
 126 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 127  * us to make MDS RPCs with large enough reply buffers to hold the
 128  * maximum-sized (= maximum striped) EA and cookie without having to
 129  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 130 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 131 {
 132         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 133         __u32 valsize = sizeof(struct lov_desc);
 134         int rc, easize, def_easize, cookiesize;
 135         struct lov_desc desc;
 136         __u32 stripes;
 137         ENTRY;
 138
 139         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 140                           &valsize, &desc);
 141         if (rc)
 142                 RETURN(rc);
 143
 144         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 145         lsm.lsm_stripe_count = stripes;
 146         easize = obd_size_diskmd(dt_exp, &lsm);
 147
 148         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 149         def_easize = obd_size_diskmd(dt_exp, &lsm);
 150
 151         cookiesize = stripes * sizeof(struct llog_cookie);
 152
 153         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 154                easize, cookiesize);
 155
 156         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 157         RETURN(rc);
 158 }
 159
 160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 161                                     uid_t nllu, gid_t nllg)
 162 {
 163         struct inode *root = 0;
 164         struct ll_sb_info *sbi = ll_s2sbi(sb);
 165         struct obd_device *obd;
 166         struct lu_fid rootfid;
 167         struct obd_capa *oc = NULL;
 168         struct obd_statfs osfs;
 169         struct ptlrpc_request *request = NULL;
 170         struct lustre_handle dt_conn = {0, };
 171         struct lustre_handle md_conn = {0, };
 172         struct obd_connect_data *data = NULL;
 173         struct lustre_md lmd;
 174         obd_valid valid;
 175         int size, err, checksum;
 176         ENTRY;
 177
 178         obd = class_name2obd(md);
 179         if (!obd) {
 180                 CERROR("MD %s: not setup or attached\n", md);
 181                 RETURN(-EINVAL);
 182         }
 183
 184         OBD_ALLOC_PTR(data);
 185         if (data == NULL)
 186                 RETURN(-ENOMEM);
 187
 188         if (proc_lustre_fs_root) {
 189                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 190                                                   dt, md);
 191                 if (err < 0)
 192                         CERROR("could not register mount in /proc/lustre");
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
 197                                   OBD_CONNECT_JOIN |
 198                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
 199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 200                                   OBD_CONNECT_CANCELSET;
 201 #ifdef CONFIG_FS_POSIX_ACL
 202         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 203 #endif
 204         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 205         data->ocd_version = LUSTRE_VERSION_CODE;
 206
 207         if (sb->s_flags & MS_RDONLY)
 208                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 209         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 210                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 211
 212 #ifdef HAVE_MS_FLOCK_LOCK
 213         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 214         sb->s_flags |= MS_FLOCK_LOCK;
 215 #endif
 216
 217         if (sbi->ll_flags & LL_SBI_FLOCK)
 218                 sbi->ll_fop = &ll_file_operations_flock;
 219         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 220                 sbi->ll_fop = &ll_file_operations;
 221         else
 222                 sbi->ll_fop = &ll_file_operations_noflock;
 223
 224         /* real client */
 225         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 226         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 227                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 228                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 229         } else {
 230                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 231                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 232         }
 233         data->ocd_nllu = nllu;
 234         data->ocd_nllg = nllg;
 235
 236         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
 237         if (err == -EBUSY) {
 238                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 239                                    "recovery, of which this client is not a "
 240                                    "part. Please wait for recovery to complete,"
 241                                    " abort, or time out.\n", md);
 242                 GOTO(out, err);
 243         } else if (err) {
 244                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 245                 GOTO(out, err);
 246         }
 247         sbi->ll_md_exp = class_conn2export(&md_conn);
 248
 249         err = obd_fid_init(sbi->ll_md_exp);
 250         if (err) {
 251                 CERROR("Can't init metadata layer FID infrastructure, "
 252                        "rc %d\n", err);
 253                 GOTO(out_md, err);
 254         }
 255
 256         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
 257         if (err)
 258                 GOTO(out_md_fid, err);
 259
 260         size = sizeof(*data);
 261         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 262                            KEY_CONN_DATA,  &size, data);
 263         if (err) {
 264                 CERROR("Get connect data failed: %d \n", err);
 265                 GOTO(out_md, err);
 266         }
 267
 268         LASSERT(osfs.os_bsize);
 269         sb->s_blocksize = osfs.os_bsize;
 270         sb->s_blocksize_bits = log2(osfs.os_bsize);
 271         sb->s_magic = LL_SUPER_MAGIC;
 272
 273         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 274          *         retval = in_file->f_op->sendfile(...);
 275          *         if (*ppos > max)
 276          *                 retval = -EOVERFLOW;
 277          *
 278          * it will check if *ppos is greater than max. However, max equals to
 279          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 280          * has been defined as a signed long long ineger in linux kernel. */
 281 #if BITS_PER_LONG == 64
 282         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 283 #else
 284         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 285 #endif
 286         sbi->ll_namelen = osfs.os_namelen;
 287         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 288
 289         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 290             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 291                 LCONSOLE_INFO("Disabling user_xattr feature because "
 292                               "it is not supported on the server\n");
 293                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 294         }
 295
 296         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 297 #ifdef MS_POSIXACL
 298                 sb->s_flags |= MS_POSIXACL;
 299 #endif
 300                 sbi->ll_flags |= LL_SBI_ACL;
 301         } else {
 302                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 303 #ifdef MS_POSIXACL
 304                 sb->s_flags &= ~MS_POSIXACL;
 305 #endif
 306                 sbi->ll_flags &= ~LL_SBI_ACL;
 307         }
 308
 309         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 310                 sbi->ll_flags |= LL_SBI_JOIN;
 311
 312         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 313                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 314                         /* sometimes local client claims to be remote, but mdt
 315                          * will disagree when client gss not applied. */
 316                         LCONSOLE_INFO("client claims to be remote, but server "
 317                                       "rejected, forced to be local.\n");
 318                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 319                 }
 320         } else {
 321                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 322                         /* with gss applied, remote client can not claim to be
 323                          * local, so mdt maybe force client to be remote. */
 324                         LCONSOLE_INFO("client claims to be local, but server "
 325                                       "rejected, forced to be remote.\n");
 326                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 327                 }
 328         }
 329
 330         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 331                 LCONSOLE_INFO("client enabled MDS capability!\n");
 332                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 333         }
 334
 335         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 336                 LCONSOLE_INFO("client enabled OSS capability!\n");
 337                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 338         }
 339
 340         sbi->ll_sdev_orig = sb->s_dev;
 341 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 342         /* We set sb->s_dev equal on all lustre clients in order to support
 343          * NFS export clustering.  NFSD requires that the FSID be the same
 344          * on all clients. */
 345         /* s_dev is also used in lt_compare() to compare two fs, but that is
 346          * only a node-local comparison. */
 347
 348         /* XXX: this will not work with LMV */
 349         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 350                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 351 #endif
 352
 353         obd = class_name2obd(dt);
 354         if (!obd) {
 355                 CERROR("DT %s: not setup or attached\n", dt);
 356                 GOTO(out_md_fid, err = -ENODEV);
 357         }
 358
 359         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 360                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 361                                   OBD_CONNECT_CANCELSET;
 362         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 363                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 364
 365         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 366                "ocd_grant: %d\n", data->ocd_connect_flags,
 367                data->ocd_version, data->ocd_grant);
 368
 369         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 370         obd->obd_upcall.onu_upcall = ll_ocd_update;
 371         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 372
 373         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
 374         if (err == -EBUSY) {
 375                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 376                                    "recovery, of which this client is not a "
 377                                    "part.  Please wait for recovery to "
 378                                    "complete, abort, or time out.\n", dt);
 379                 GOTO(out_md_fid, err);
 380         } else if (err) {
 381                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 382                 GOTO(out_md_fid, err);
 383         }
 384
 385         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 386
 387         err = obd_fid_init(sbi->ll_dt_exp);
 388         if (err) {
 389                 CERROR("Can't init data layer FID infrastructure, "
 390                        "rc %d\n", err);
 391                 GOTO(out_dt, err);
 392         }
 393
 394         spin_lock(&sbi->ll_lco.lco_lock);
 395         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 396         spin_unlock(&sbi->ll_lco.lco_lock);
 397
 398         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 399
 400         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 401                                   0, NULL, NULL, NULL);
 402         if (err < 0) {
 403                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 404                                    "filesystem. There must be at least one "
 405                                    "active OST for a client to start.\n");
 406                 GOTO(out_dt_fid, err);
 407         }
 408
 409         if (!ll_async_page_slab) {
 410                 ll_async_page_slab_size =
 411                         size_round(sizeof(struct ll_async_page)) + err;
 412                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 413                                                         ll_async_page_slab_size,
 414                                                            0, 0);
 415                 if (!ll_async_page_slab)
 416                         GOTO(out_dt_fid, err = -ENOMEM);
 417         }
 418
 419         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 420         if (err) {
 421                 CERROR("cannot mds_connect: rc = %d\n", err);
 422                 GOTO(out_dt_fid, err);
 423         }
 424         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 425         sbi->ll_root_fid = rootfid;
 426
 427         sb->s_op = &lustre_super_operations;
 428 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 429         sb->s_export_op = &lustre_export_operations;
 430 #endif
 431
 432         /* make root inode
 433          * XXX: move this to after cbd setup? */
 434         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 435         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 436                 valid |= OBD_MD_FLRMTPERM;
 437         else if (sbi->ll_flags & LL_SBI_ACL)
 438                 valid |= OBD_MD_FLACL;
 439
 440         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 441         if (oc)
 442                 free_capa(oc);
 443         if (err) {
 444                 CERROR("md_getattr failed for root: rc = %d\n", err);
 445                 GOTO(out_dt_fid, err);
 446         }
 447         memset(&lmd, 0, sizeof(lmd));
 448         err = md_get_lustre_md(sbi->ll_md_exp, request,
 449                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
 450                                &lmd);
 451         if (err) {
 452                 CERROR("failed to understand root inode md: rc = %d\n", err);
 453                 ptlrpc_req_finished (request);
 454                 GOTO(out_dt_fid, err);
 455         }
 456
 457         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 458         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 459         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 460         ptlrpc_req_finished(request);
 461
 462         if (root == NULL || is_bad_inode(root)) {
 463                 if (lmd.lsm)
 464                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 465 #ifdef CONFIG_FS_POSIX_ACL
 466                 if (lmd.posix_acl) {
 467                         posix_acl_release(lmd.posix_acl);
 468                         lmd.posix_acl = NULL;
 469                 }
 470 #endif
 471                 CERROR("lustre_lite: bad iget4 for root\n");
 472                 GOTO(out_root, err = -EBADF);
 473         }
 474
 475         err = ll_close_thread_start(&sbi->ll_lcq);
 476         if (err) {
 477                 CERROR("cannot start close thread: rc %d\n", err);
 478                 GOTO(out_root, err);
 479         }
 480
 481         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 482         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 483                                  sizeof(checksum), &checksum, NULL);
 484
 485         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 486            backing dev info assigned to inode mapping is used for
 487            determining maximal readahead. */
 488 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 489     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 490         /* bug 2805 - set VM readahead to zero */
 491         vm_max_readahead = vm_min_readahead = 0;
 492 #endif
 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 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 577 void lustre_throw_orphan_dentries(struct super_block *sb)
 578 {
 579         struct dentry *dentry, *next;
 580         struct ll_sb_info *sbi = ll_s2sbi(sb);
 581
 582         /* Do this to get rid of orphaned dentries. That is not really trw. */
 583         list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
 584                                  d_hash) {
 585                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 586                       "before and after shrink_dcache_parent\n",
 587                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 588                 lustre_dump_dentry(dentry, 1);
 589                 shrink_dcache_parent(dentry);
 590                 lustre_dump_dentry(dentry, 1);
 591         }
 592 }
 593 #else
 594 #define lustre_throw_orphan_dentries(sb)
 595 #endif
 596
 597 #ifdef HAVE_EXPORT___IGET
 598 static void prune_dir_dentries(struct inode *inode)
 599 {
 600         struct dentry *dentry, *prev = NULL;
 601
 602         /* due to lustre specific logic, a directory
 603          * can have few dentries - a bug from VFS POV */
 604 restart:
 605         spin_lock(&dcache_lock);
 606         if (!list_empty(&inode->i_dentry)) {
 607                 dentry = list_entry(inode->i_dentry.prev,
 608                                     struct dentry, d_alias);
 609                 /* in order to prevent infinite loops we
 610                  * break if previous dentry is busy */
 611                 if (dentry != prev) {
 612                         prev = dentry;
 613                         dget_locked(dentry);
 614                         spin_unlock(&dcache_lock);
 615
 616                         /* try to kill all child dentries */
 617                         lock_dentry(dentry);
 618                         shrink_dcache_parent(dentry);
 619                         unlock_dentry(dentry);
 620                         dput(dentry);
 621
 622                         /* now try to get rid of current dentry */
 623                         d_prune_aliases(inode);
 624                         goto restart;
 625                 }
 626         }
 627         spin_unlock(&dcache_lock);
 628 }
 629
 630 static void prune_deathrow_one(struct ll_inode_info *lli)
 631 {
 632         struct inode *inode = ll_info2i(lli);
 633
 634         /* first, try to drop any dentries - they hold a ref on the inode */
 635         if (S_ISDIR(inode->i_mode))
 636                 prune_dir_dentries(inode);
 637         else
 638                 d_prune_aliases(inode);
 639
 640
 641         /* if somebody still uses it, leave it */
 642         LASSERT(atomic_read(&inode->i_count) > 0);
 643         if (atomic_read(&inode->i_count) > 1)
 644                 goto out;
 645
 646         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 647                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 648
 649         /* seems nobody uses it anymore */
 650         inode->i_nlink = 0;
 651
 652 out:
 653         iput(inode);
 654         return;
 655 }
 656
 657 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 658 {
 659         struct ll_inode_info *lli;
 660         int empty;
 661
 662         do {
 663                 if (need_resched() && try)
 664                         break;
 665
 666                 if (try) {
 667                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 668                                 break;
 669                 } else {
 670                         spin_lock(&sbi->ll_deathrow_lock);
 671                 }
 672
 673                 empty = 1;
 674                 lli = NULL;
 675                 if (!list_empty(&sbi->ll_deathrow)) {
 676                         lli = list_entry(sbi->ll_deathrow.next,
 677                                          struct ll_inode_info,
 678                                          lli_dead_list);
 679                         list_del_init(&lli->lli_dead_list);
 680                         if (!list_empty(&sbi->ll_deathrow))
 681                                 empty = 0;
 682                 }
 683                 spin_unlock(&sbi->ll_deathrow_lock);
 684
 685                 if (lli)
 686                         prune_deathrow_one(lli);
 687
 688         } while (empty == 0);
 689 }
 690 #else /* !HAVE_EXPORT___IGET */
 691 #define prune_deathrow(sbi, try) do {} while (0)
 692 #endif /* HAVE_EXPORT___IGET */
 693
 694 void client_common_put_super(struct super_block *sb)
 695 {
 696         struct ll_sb_info *sbi = ll_s2sbi(sb);
 697         ENTRY;
 698
 699         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 700
 701         ll_close_thread_shutdown(sbi->ll_lcq);
 702
 703         /* destroy inodes in deathrow */
 704         prune_deathrow(sbi, 0);
 705
 706         list_del(&sbi->ll_conn_chain);
 707
 708         obd_fid_fini(sbi->ll_dt_exp);
 709         obd_disconnect(sbi->ll_dt_exp);
 710         sbi->ll_dt_exp = NULL;
 711
 712         lprocfs_unregister_mountpoint(sbi);
 713
 714         obd_fid_fini(sbi->ll_md_exp);
 715         obd_disconnect(sbi->ll_md_exp);
 716         sbi->ll_md_exp = NULL;
 717
 718         lustre_throw_orphan_dentries(sb);
 719
 720         EXIT;
 721 }
 722
 723 void ll_kill_super(struct super_block *sb)
 724 {
 725         struct ll_sb_info *sbi;
 726
 727         ENTRY;
 728
 729         /* not init sb ?*/
 730         if (!(sb->s_flags & MS_ACTIVE))
 731                 return;
 732
 733         sbi = ll_s2sbi(sb);
 734         /* we need restore s_dev from changed for clustred NFS before put_super
 735          * because new kernels have cached s_dev and change sb->s_dev in
 736          * put_super not affected real removing devices */
 737         if (sbi)
 738                 sb->s_dev = sbi->ll_sdev_orig;
 739         EXIT;
 740 }
 741
 742 char *ll_read_opt(const char *opt, char *data)
 743 {
 744         char *value;
 745         char *retval;
 746         ENTRY;
 747
 748         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 749         if (strncmp(opt, data, strlen(opt)))
 750                 RETURN(NULL);
 751         if ((value = strchr(data, '=')) == NULL)
 752                 RETURN(NULL);
 753
 754         value++;
 755         OBD_ALLOC(retval, strlen(value) + 1);
 756         if (!retval) {
 757                 CERROR("out of memory!\n");
 758                 RETURN(NULL);
 759         }
 760
 761         memcpy(retval, value, strlen(value)+1);
 762         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 763         RETURN(retval);
 764 }
 765
 766 static inline int ll_set_opt(const char *opt, char *data, int fl)
 767 {
 768         if (strncmp(opt, data, strlen(opt)) != 0)
 769                 return(0);
 770         else
 771                 return(fl);
 772 }
 773
 774 /* non-client-specific mount options are parsed in lmd_parse */
 775 static int ll_options(char *options, int *flags)
 776 {
 777         int tmp;
 778         char *s1 = options, *s2;
 779         ENTRY;
 780
 781         if (!options)
 782                 RETURN(0);
 783
 784         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 785
 786         while (*s1) {
 787                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 788                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 789                 if (tmp) {
 790                         *flags |= tmp;
 791                         goto next;
 792                 }
 793                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 794                 if (tmp) {
 795                         *flags |= tmp;
 796                         goto next;
 797                 }
 798                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 799                 if (tmp) {
 800                         *flags |= tmp;
 801                         goto next;
 802                 }
 803                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 804                 if (tmp) {
 805                         *flags &= ~tmp;
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 809                 if (tmp) {
 810                         *flags |= tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 814                 if (tmp) {
 815                         *flags &= ~tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 819                 if (tmp) {
 820                         /* Ignore deprecated mount option.  The client will
 821                          * always try to mount with ACL support, whether this
 822                          * is used depends on whether server supports it. */
 823                         goto next;
 824                 }
 825                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 826                 if (tmp) {
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834
 835                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 841                 if (tmp) {
 842                         *flags &= ~tmp;
 843                         goto next;
 844                 }
 845
 846                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 847                                    s1);
 848                 RETURN(-EINVAL);
 849
 850 next:
 851                 /* Find next opt */
 852                 s2 = strchr(s1, ',');
 853                 if (s2 == NULL)
 854                         break;
 855                 s1 = s2 + 1;
 856         }
 857         RETURN(0);
 858 }
 859
 860 void ll_lli_init(struct ll_inode_info *lli)
 861 {
 862         lli->lli_inode_magic = LLI_INODE_MAGIC;
 863         sema_init(&lli->lli_size_sem, 1);
 864         sema_init(&lli->lli_write_sem, 1);
 865         lli->lli_flags = 0;
 866         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 867         spin_lock_init(&lli->lli_lock);
 868         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 869         INIT_LIST_HEAD(&lli->lli_close_list);
 870         lli->lli_inode_magic = LLI_INODE_MAGIC;
 871         sema_init(&lli->lli_och_sem, 1);
 872         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 873         lli->lli_mds_exec_och = NULL;
 874         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 875         lli->lli_open_fd_exec_count = 0;
 876         INIT_LIST_HEAD(&lli->lli_dead_list);
 877         lli->lli_remote_perms = NULL;
 878         lli->lli_rmtperm_utime = 0;
 879         sema_init(&lli->lli_rmtperm_sem, 1);
 880         INIT_LIST_HEAD(&lli->lli_oss_capas);
 881 }
 882
 883 int ll_fill_super(struct super_block *sb)
 884 {
 885         struct lustre_profile *lprof;
 886         struct lustre_sb_info *lsi = s2lsi(sb);
 887         struct ll_sb_info *sbi;
 888         char  *dt = NULL, *md = NULL;
 889         char  *profilenm = get_profile_name(sb);
 890         struct config_llog_instance cfg = {0, };
 891         char   ll_instance[sizeof(sb) * 2 + 1];
 892         int    err;
 893         ENTRY;
 894
 895         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 896
 897         cfs_module_get();
 898
 899         sb->s_type->fs_flags |= FS_ODD_RENAME;
 900         /* client additional sb info */
 901         lsi->lsi_llsbi = sbi = ll_init_sbi();
 902         if (!sbi) {
 903                 cfs_module_put();
 904                 RETURN(-ENOMEM);
 905         }
 906
 907         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 908         if (err)
 909                 GOTO(out_free, err);
 910
 911         /* Generate a string unique to this super, in case some joker tries
 912            to mount the same fs at two mount points.
 913            Use the address of the super itself.*/
 914         sprintf(ll_instance, "%p", sb);
 915         cfg.cfg_instance = ll_instance;
 916         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 917
 918         /* set up client obds */
 919         err = lustre_process_log(sb, profilenm, &cfg);
 920         if (err < 0) {
 921                 CERROR("Unable to process log: %d\n", err);
 922                 GOTO(out_free, err);
 923         }
 924
 925         lprof = class_get_profile(profilenm);
 926         if (lprof == NULL) {
 927                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 928                                    " read from the MGS.  Does that filesystem "
 929                                    "exist?\n", profilenm);
 930                 GOTO(out_free, err = -EINVAL);
 931         }
 932         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 933                lprof->lp_md, lprof->lp_dt);
 934
 935         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 936                   strlen(ll_instance) + 2);
 937         if (!dt)
 938                 GOTO(out_free, err = -ENOMEM);
 939         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 940
 941         OBD_ALLOC(md, strlen(lprof->lp_md) +
 942                   strlen(ll_instance) + 2);
 943         if (!md)
 944                 GOTO(out_free, err = -ENOMEM);
 945         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 946
 947         /* connections, registrations, sb setup */
 948         err = client_common_fill_super(sb, md, dt,
 949                                        lsi->lsi_lmd->lmd_nllu,
 950                                        lsi->lsi_lmd->lmd_nllg);
 951
 952 out_free:
 953         if (md)
 954                 OBD_FREE(md, strlen(md) + 1);
 955         if (dt)
 956                 OBD_FREE(dt, strlen(dt) + 1);
 957         if (err)
 958                 ll_put_super(sb);
 959         else
 960                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 961
 962         RETURN(err);
 963 } /* ll_fill_super */
 964
 965
 966 void ll_put_super(struct super_block *sb)
 967 {
 968         struct config_llog_instance cfg;
 969         char   ll_instance[sizeof(sb) * 2 + 1];
 970         struct obd_device *obd;
 971         struct lustre_sb_info *lsi = s2lsi(sb);
 972         struct ll_sb_info *sbi = ll_s2sbi(sb);
 973         char *profilenm = get_profile_name(sb);
 974         int force = 1, next;
 975         ENTRY;
 976
 977         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 978
 979         ll_print_capa_stat(sbi);
 980
 981         sprintf(ll_instance, "%p", sb);
 982         cfg.cfg_instance = ll_instance;
 983         lustre_end_log(sb, NULL, &cfg);
 984
 985         if (sbi->ll_md_exp) {
 986                 obd = class_exp2obd(sbi->ll_md_exp);
 987                 if (obd)
 988                         force = obd->obd_force;
 989         }
 990
 991         /* We need to set force before the lov_disconnect in
 992            lustre_common_put_super, since l_d cleans up osc's as well. */
 993         if (force) {
 994                 next = 0;
 995                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 996                                                      &next)) != NULL) {
 997                         obd->obd_force = force;
 998                 }
 999         }
1000
1001         if (sbi->ll_lcq) {
1002                 /* Only if client_common_fill_super succeeded */
1003                 client_common_put_super(sb);
1004         }
1005         next = 0;
1006         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1007                 class_manual_cleanup(obd);
1008         }
1009
1010         if (profilenm)
1011                 class_del_profile(profilenm);
1012
1013         ll_free_sbi(sb);
1014         lsi->lsi_llsbi = NULL;
1015
1016         lustre_common_put_super(sb);
1017
1018         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1019
1020         cfs_module_put();
1021
1022         EXIT;
1023 } /* client_put_super */
1024
1025 #ifdef HAVE_REGISTER_CACHE
1026 #include <linux/cache_def.h>
1027 #ifdef HAVE_CACHE_RETURN_INT
1028 static int
1029 #else
1030 static void
1031 #endif
1032 ll_shrink_cache(int priority, unsigned int gfp_mask)
1033 {
1034         struct ll_sb_info *sbi;
1035         int count = 0;
1036
1037         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1038                 count += llap_shrink_cache(sbi, priority);
1039
1040 #ifdef HAVE_CACHE_RETURN_INT
1041         return count;
1042 #endif
1043 }
1044
1045 struct cache_definition ll_cache_definition = {
1046         .name = "llap_cache",
1047         .shrink = ll_shrink_cache
1048 };
1049 #endif /* HAVE_REGISTER_CACHE */
1050
1051 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1052 {
1053         struct inode *inode = NULL;
1054         /* NOTE: we depend on atomic igrab() -bzzz */
1055         lock_res_and_lock(lock);
1056         if (lock->l_ast_data) {
1057                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1058                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1059                         inode = igrab(lock->l_ast_data);
1060                 } else {
1061                         inode = lock->l_ast_data;
1062                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1063                                                 D_INFO : D_WARNING,
1064                                         lock, __FILE__, __func__, __LINE__,
1065                                         "l_ast_data %p is bogus: magic %08x",
1066                                         lock->l_ast_data, lli->lli_inode_magic);
1067                         inode = NULL;
1068                 }
1069         }
1070         unlock_res_and_lock(lock);
1071         return inode;
1072 }
1073
1074 static int null_if_equal(struct ldlm_lock *lock, void *data)
1075 {
1076         if (data == lock->l_ast_data) {
1077                 lock->l_ast_data = NULL;
1078
1079                 if (lock->l_req_mode != lock->l_granted_mode)
1080                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1081         }
1082
1083         return LDLM_ITER_CONTINUE;
1084 }
1085
1086 void ll_clear_inode(struct inode *inode)
1087 {
1088         struct ll_inode_info *lli = ll_i2info(inode);
1089         struct ll_sb_info *sbi = ll_i2sbi(inode);
1090         ENTRY;
1091
1092         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1093                inode->i_generation, inode);
1094
1095         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1096         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1097                          null_if_equal, inode);
1098
1099         LASSERT(!lli->lli_open_fd_write_count);
1100         LASSERT(!lli->lli_open_fd_read_count);
1101         LASSERT(!lli->lli_open_fd_exec_count);
1102
1103         if (lli->lli_mds_write_och)
1104                 ll_md_real_close(inode, FMODE_WRITE);
1105         if (lli->lli_mds_exec_och)
1106                 ll_md_real_close(inode, FMODE_EXEC);
1107         if (lli->lli_mds_read_och)
1108                 ll_md_real_close(inode, FMODE_READ);
1109
1110         if (lli->lli_smd) {
1111                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1112                                   null_if_equal, inode);
1113
1114                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1115                 lli->lli_smd = NULL;
1116         }
1117
1118         if (lli->lli_symlink_name) {
1119                 OBD_FREE(lli->lli_symlink_name,
1120                          strlen(lli->lli_symlink_name) + 1);
1121                 lli->lli_symlink_name = NULL;
1122         }
1123
1124         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1125                 LASSERT(lli->lli_posix_acl == NULL);
1126                 if (lli->lli_remote_perms) {
1127                         free_rmtperm_hash(lli->lli_remote_perms);
1128                         lli->lli_remote_perms = NULL;
1129                 }
1130         }
1131 #ifdef CONFIG_FS_POSIX_ACL
1132         else if (lli->lli_posix_acl) {
1133                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1134                 LASSERT(lli->lli_remote_perms == NULL);
1135                 posix_acl_release(lli->lli_posix_acl);
1136                 lli->lli_posix_acl = NULL;
1137         }
1138 #endif
1139         lli->lli_inode_magic = LLI_INODE_DEAD;
1140
1141 #ifdef HAVE_EXPORT___IGET
1142         spin_lock(&sbi->ll_deathrow_lock);
1143         list_del_init(&lli->lli_dead_list);
1144         spin_unlock(&sbi->ll_deathrow_lock);
1145 #endif
1146         ll_clear_inode_capas(inode);
1147
1148         EXIT;
1149 }
1150
1151 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1152                   struct md_open_data **mod)
1153 {
1154         struct lustre_md md;
1155         struct ll_sb_info *sbi = ll_i2sbi(inode);
1156         struct ptlrpc_request *request = NULL;
1157         int rc;
1158         ENTRY;
1159
1160         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1161                                      LUSTRE_OPC_ANY, NULL);
1162         if (IS_ERR(op_data))
1163                 RETURN(PTR_ERR(op_data));
1164
1165         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1166                         &request, mod);
1167         if (rc) {
1168                 ptlrpc_req_finished(request);
1169                 if (rc == -ENOENT) {
1170                         inode->i_nlink = 0;
1171                         /* Unlinked special device node? Or just a race?
1172                          * Pretend we done everything. */
1173                         if (!S_ISREG(inode->i_mode) &&
1174                             !S_ISDIR(inode->i_mode))
1175                                 rc = inode_setattr(inode, &op_data->op_attr);
1176                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1177                         CERROR("md_setattr fails: rc = %d\n", rc);
1178                 }
1179                 RETURN(rc);
1180         }
1181
1182         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1183                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1184         if (rc) {
1185                 ptlrpc_req_finished(request);
1186                 RETURN(rc);
1187         }
1188
1189         /* We call inode_setattr to adjust timestamps.
1190          * If there is at least some data in file, we cleared ATTR_SIZE
1191          * above to avoid invoking vmtruncate, otherwise it is important
1192          * to call vmtruncate in inode_setattr to update inode->i_size
1193          * (bug 6196) */
1194         rc = inode_setattr(inode, &op_data->op_attr);
1195
1196         /* Extract epoch data if obtained. */
1197         op_data->op_handle = md.body->handle;
1198         op_data->op_ioepoch = md.body->ioepoch;
1199
1200         ll_update_inode(inode, &md);
1201         ptlrpc_req_finished(request);
1202
1203         RETURN(rc);
1204 }
1205
1206 /* Close IO epoch and send Size-on-MDS attribute update. */
1207 static int ll_setattr_done_writing(struct inode *inode,
1208                                    struct md_op_data *op_data,
1209                                    struct md_open_data *mod)
1210 {
1211         struct ll_inode_info *lli = ll_i2info(inode);
1212         int rc = 0;
1213         ENTRY;
1214
1215         LASSERT(op_data != NULL);
1216         if (!S_ISREG(inode->i_mode))
1217                 RETURN(0);
1218
1219         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1220                op_data->op_ioepoch, PFID(&lli->lli_fid));
1221
1222         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1223         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1224         if (rc == -EAGAIN) {
1225                 /* MDS has instructed us to obtain Size-on-MDS attribute
1226                  * from OSTs and send setattr to back to MDS. */
1227                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1228                                          op_data->op_ioepoch);
1229         } else if (rc) {
1230                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1231                        inode->i_ino, rc);
1232         }
1233         RETURN(rc);
1234 }
1235
1236 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1237  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1238  * keep these values until such a time that objects are allocated for it.
1239  * We do the MDS operations first, as it is checking permissions for us.
1240  * We don't to the MDS RPC if there is nothing that we want to store there,
1241  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1242  * going to do an RPC anyways.
1243  *
1244  * If we are doing a truncate, we will send the mtime and ctime updates
1245  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1246  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1247  * at the same time.
1248  */
1249 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1250 {
1251         struct ll_inode_info *lli = ll_i2info(inode);
1252         struct lov_stripe_md *lsm = lli->lli_smd;
1253         struct ll_sb_info *sbi = ll_i2sbi(inode);
1254         struct md_op_data *op_data = NULL;
1255         struct md_open_data *mod = NULL;
1256         int ia_valid = attr->ia_valid;
1257         int rc = 0, rc1 = 0;
1258         ENTRY;
1259
1260         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1261                attr->ia_valid);
1262         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1263
1264         if (ia_valid & ATTR_SIZE) {
1265                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1266                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1267                                attr->ia_size, ll_file_maxbytes(inode));
1268                         RETURN(-EFBIG);
1269                 }
1270
1271                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1272         }
1273
1274         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1275         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1276                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1277                         RETURN(-EPERM);
1278         }
1279
1280         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1281         if (attr->ia_valid & ATTR_CTIME) {
1282                 attr->ia_ctime = CURRENT_TIME;
1283                 attr->ia_valid |= ATTR_CTIME_SET;
1284         }
1285         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1286                 attr->ia_atime = CURRENT_TIME;
1287                 attr->ia_valid |= ATTR_ATIME_SET;
1288         }
1289         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1290                 attr->ia_mtime = CURRENT_TIME;
1291                 attr->ia_valid |= ATTR_MTIME_SET;
1292         }
1293         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1294                 /* To avoid stale mtime on mds, obtain it from ost and send
1295                    to mds. */
1296                 rc = ll_glimpse_size(inode, 0);
1297                 if (rc)
1298                         RETURN(rc);
1299
1300                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1301                 attr->ia_mtime = inode->i_mtime;
1302         }
1303
1304         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1305                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1306                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1307                        CURRENT_SECONDS);
1308
1309         /* NB: ATTR_SIZE will only be set after this point if the size
1310          * resides on the MDS, ie, this file has no objects. */
1311         if (lsm)
1312                 attr->ia_valid &= ~ATTR_SIZE;
1313
1314         /* We always do an MDS RPC, even if we're only changing the size;
1315          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1316
1317         OBD_ALLOC_PTR(op_data);
1318         if (op_data == NULL)
1319                 RETURN(-ENOMEM);
1320
1321         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1322
1323         /* Open epoch for truncate. */
1324         if (ia_valid & ATTR_SIZE)
1325                 op_data->op_flags = MF_EPOCH_OPEN;
1326
1327         rc = ll_md_setattr(inode, op_data, &mod);
1328         if (rc)
1329                 GOTO(out, rc);
1330
1331         if (op_data->op_ioepoch)
1332                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1333                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1334
1335         if (!lsm || !S_ISREG(inode->i_mode)) {
1336                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1337                 GOTO(out, rc = 0);
1338         }
1339
1340         /* We really need to get our PW lock before we change inode->i_size.
1341          * If we don't we can race with other i_size updaters on our node, like
1342          * ll_file_read.  We can also race with i_size propogation to other
1343          * nodes through dirtying and writeback of final cached pages.  This
1344          * last one is especially bad for racing o_append users on other
1345          * nodes. */
1346         if (ia_valid & ATTR_SIZE) {
1347                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1348                                                            OBD_OBJECT_EOF } };
1349                 struct lustre_handle lockh = { 0 };
1350                 int err, ast_flags = 0;
1351                 /* XXX when we fix the AST intents to pass the discard-range
1352                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1353                  * XXX here. */
1354                 if (attr->ia_size == 0)
1355                         ast_flags = LDLM_AST_DISCARD_DATA;
1356
1357                 UNLOCK_INODE_MUTEX(inode);
1358                 UP_WRITE_I_ALLOC_SEM(inode);
1359                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1360                                     ast_flags);
1361 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1362                 DOWN_WRITE_I_ALLOC_SEM(inode);
1363                 LOCK_INODE_MUTEX(inode);
1364 #else
1365                 LOCK_INODE_MUTEX(inode);
1366                 DOWN_WRITE_I_ALLOC_SEM(inode);
1367 #endif
1368                 if (rc != 0)
1369                         GOTO(out, rc);
1370
1371                 /* Only ll_inode_size_lock is taken at this level.
1372                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1373                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1374                  * ll_truncate dropped ll_inode_size_lock() */
1375                 ll_inode_size_lock(inode, 0);
1376                 rc = vmtruncate(inode, attr->ia_size);
1377                 if (rc != 0) {
1378                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1379                         ll_inode_size_unlock(inode, 0);
1380                 }
1381
1382                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1383                 if (err) {
1384                         CERROR("ll_extent_unlock failed: %d\n", err);
1385                         if (!rc)
1386                                 rc = err;
1387                 }
1388         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1389                 obd_flag flags;
1390                 struct obd_info oinfo = { { { 0 } } };
1391                 struct obdo *oa;
1392
1393                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1394                        inode->i_ino, LTIME_S(attr->ia_mtime));
1395
1396                 OBDO_ALLOC(oa);
1397                 if (oa) {
1398                         oa->o_id = lsm->lsm_object_id;
1399                         oa->o_gr = lsm->lsm_object_gr;
1400                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1401
1402                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1403                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1404                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1405                                 OBD_MD_FLGROUP;
1406
1407                         obdo_from_inode(oa, inode, flags);
1408
1409                         oinfo.oi_oa = oa;
1410                         oinfo.oi_md = lsm;
1411                         oinfo.oi_capa = ll_mdscapa_get(inode);
1412
1413                         /* XXX: this looks unnecessary now. */
1414                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1415                         capa_put(oinfo.oi_capa);
1416                         if (rc)
1417                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1418                         OBDO_FREE(oa);
1419                 } else {
1420                         rc = -ENOMEM;
1421                 }
1422         }
1423         EXIT;
1424 out:
1425         if (op_data) {
1426                 if (op_data->op_ioepoch)
1427                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1428                 ll_finish_md_op_data(op_data);
1429         }
1430         return rc ? rc : rc1;
1431 }
1432
1433 int ll_setattr(struct dentry *de, struct iattr *attr)
1434 {
1435         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1436             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1437                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1438
1439         return ll_setattr_raw(de->d_inode, attr);
1440 }
1441
1442 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1443                        __u64 max_age)
1444 {
1445         struct ll_sb_info *sbi = ll_s2sbi(sb);
1446         struct obd_statfs obd_osfs;
1447         int rc;
1448         ENTRY;
1449
1450         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1451         if (rc) {
1452                 CERROR("md_statfs fails: rc = %d\n", rc);
1453                 RETURN(rc);
1454         }
1455
1456         osfs->os_type = sb->s_magic;
1457
1458         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1459                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1460
1461         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1462                               &obd_osfs, max_age);
1463         if (rc) {
1464                 CERROR("obd_statfs fails: rc = %d\n", rc);
1465                 RETURN(rc);
1466         }
1467
1468         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1469                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1470                obd_osfs.os_files);
1471
1472         osfs->os_bsize = obd_osfs.os_bsize;
1473         osfs->os_blocks = obd_osfs.os_blocks;
1474         osfs->os_bfree = obd_osfs.os_bfree;
1475         osfs->os_bavail = obd_osfs.os_bavail;
1476
1477         /* If we don't have as many objects free on the OST as inodes
1478          * on the MDS, we reduce the total number of inodes to
1479          * compensate, so that the "inodes in use" number is correct.
1480          */
1481         if (obd_osfs.os_ffree < osfs->os_ffree) {
1482                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1483                         obd_osfs.os_ffree;
1484                 osfs->os_ffree = obd_osfs.os_ffree;
1485         }
1486
1487         RETURN(rc);
1488 }
1489 #ifndef HAVE_STATFS_DENTRY_PARAM
1490 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1491 {
1492 #else
1493 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1494 {
1495         struct super_block *sb = de->d_sb;
1496 #endif
1497         struct obd_statfs osfs;
1498         int rc;
1499
1500         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1501         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1502
1503         /* For now we will always get up-to-date statfs values, but in the
1504          * future we may allow some amount of caching on the client (e.g.
1505          * from QOS or lprocfs updates). */
1506         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1507         if (rc)
1508                 return rc;
1509
1510         statfs_unpack(sfs, &osfs);
1511
1512         /* We need to downshift for all 32-bit kernels, because we can't
1513          * tell if the kernel is being called via sys_statfs64() or not.
1514          * Stop before overflowing f_bsize - in which case it is better
1515          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1516         if (sizeof(long) < 8) {
1517                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1518                         sfs->f_bsize <<= 1;
1519
1520                         osfs.os_blocks >>= 1;
1521                         osfs.os_bfree >>= 1;
1522                         osfs.os_bavail >>= 1;
1523                 }
1524         }
1525
1526         sfs->f_blocks = osfs.os_blocks;
1527         sfs->f_bfree = osfs.os_bfree;
1528         sfs->f_bavail = osfs.os_bavail;
1529
1530         return 0;
1531 }
1532
1533 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1534 {
1535         struct ll_inode_info *lli;
1536         struct lov_stripe_md *lsm;
1537
1538         lli = ll_i2info(inode);
1539         LASSERT(lli->lli_size_sem_owner != current);
1540         down(&lli->lli_size_sem);
1541         LASSERT(lli->lli_size_sem_owner == NULL);
1542         lli->lli_size_sem_owner = current;
1543         lsm = lli->lli_smd;
1544         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1545                  lsm, lock_lsm);
1546         if (lock_lsm)
1547                 lov_stripe_lock(lsm);
1548 }
1549
1550 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1551 {
1552         struct ll_inode_info *lli;
1553         struct lov_stripe_md *lsm;
1554
1555         lli = ll_i2info(inode);
1556         lsm = lli->lli_smd;
1557         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1558                  lsm, unlock_lsm);
1559         if (unlock_lsm)
1560                 lov_stripe_unlock(lsm);
1561         LASSERT(lli->lli_size_sem_owner == current);
1562         lli->lli_size_sem_owner = NULL;
1563         up(&lli->lli_size_sem);
1564 }
1565
1566 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1567 {
1568         struct ll_inode_info *lli = ll_i2info(inode);
1569
1570         dump_lsm(D_INODE, lsm);
1571         dump_lsm(D_INODE, lli->lli_smd);
1572         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1573                  "lsm must be joined lsm %p\n", lsm);
1574         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1575         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1576                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1577         lli->lli_smd = lsm;
1578         lli->lli_maxbytes = lsm->lsm_maxbytes;
1579         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1580                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1581 }
1582
1583 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1584 {
1585         struct ll_inode_info *lli = ll_i2info(inode);
1586         struct mdt_body *body = md->body;
1587         struct lov_stripe_md *lsm = md->lsm;
1588         struct ll_sb_info *sbi = ll_i2sbi(inode);
1589
1590         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1591         if (lsm != NULL) {
1592                 if (lli->lli_smd == NULL) {
1593                         if (lsm->lsm_magic != LOV_MAGIC &&
1594                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1595                                 dump_lsm(D_ERROR, lsm);
1596                                 LBUG();
1597                         }
1598                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1599                                lsm, inode->i_ino, inode->i_generation, inode);
1600                         /* ll_inode_size_lock() requires it is only called
1601                          * with lli_smd != NULL or lock_lsm == 0 or we can
1602                          * race between lock/unlock.  bug 9547 */
1603                         lli->lli_smd = lsm;
1604                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1605                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1606                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1607                 } else {
1608                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1609                              lli->lli_smd->lsm_stripe_count ==
1610                                         lsm->lsm_stripe_count) {
1611                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1612                                         CERROR("lsm mismatch for inode %ld\n",
1613                                                 inode->i_ino);
1614                                         CERROR("lli_smd:\n");
1615                                         dump_lsm(D_ERROR, lli->lli_smd);
1616                                         CERROR("lsm:\n");
1617                                         dump_lsm(D_ERROR, lsm);
1618                                         LBUG();
1619                                 }
1620                         } else
1621                                 ll_replace_lsm(inode, lsm);
1622                 }
1623                 if (lli->lli_smd != lsm)
1624                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1625         }
1626
1627         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1628                 if (body->valid & OBD_MD_FLRMTPERM)
1629                         ll_update_remote_perm(inode, md->remote_perm);
1630         }
1631 #ifdef CONFIG_FS_POSIX_ACL
1632         else if (body->valid & OBD_MD_FLACL) {
1633                 spin_lock(&lli->lli_lock);
1634                 if (lli->lli_posix_acl)
1635                         posix_acl_release(lli->lli_posix_acl);
1636                 lli->lli_posix_acl = md->posix_acl;
1637                 spin_unlock(&lli->lli_lock);
1638         }
1639 #endif
1640         if (body->valid & OBD_MD_FLATIME &&
1641             body->atime > LTIME_S(inode->i_atime))
1642                 LTIME_S(inode->i_atime) = body->atime;
1643
1644         /* mtime is always updated with ctime, but can be set in past.
1645            As write and utime(2) may happen within 1 second, and utime's
1646            mtime has a priority over write's one, so take mtime from mds
1647            for the same ctimes. */
1648         if (body->valid & OBD_MD_FLCTIME &&
1649             body->ctime >= LTIME_S(inode->i_ctime)) {
1650                 LTIME_S(inode->i_ctime) = body->ctime;
1651                 if (body->valid & OBD_MD_FLMTIME) {
1652                         CDEBUG(D_INODE, "setting ino %lu mtime "
1653                                "from %lu to "LPU64"\n", inode->i_ino,
1654                                LTIME_S(inode->i_mtime), body->mtime);
1655                         LTIME_S(inode->i_mtime) = body->mtime;
1656                 }
1657         }
1658         if (body->valid & OBD_MD_FLMODE)
1659                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1660         if (body->valid & OBD_MD_FLTYPE)
1661                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1662         if (S_ISREG(inode->i_mode)) {
1663                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1664         } else {
1665                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1666         }
1667 #ifdef HAVE_INODE_BLKSIZE
1668         inode->i_blksize = 1<<inode->i_blkbits;
1669 #endif
1670         if (body->valid & OBD_MD_FLUID)
1671                 inode->i_uid = body->uid;
1672         if (body->valid & OBD_MD_FLGID)
1673                 inode->i_gid = body->gid;
1674         if (body->valid & OBD_MD_FLFLAGS)
1675                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1676         if (body->valid & OBD_MD_FLNLINK)
1677                 inode->i_nlink = body->nlink;
1678         if (body->valid & OBD_MD_FLRDEV)
1679 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1680                 inode->i_rdev = body->rdev;
1681 #else
1682                 inode->i_rdev = old_decode_dev(body->rdev);
1683 #endif
1684         if (body->valid & OBD_MD_FLSIZE) {
1685                 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1686                         if (lli->lli_flags & (LLIF_DONE_WRITING |
1687                                               LLIF_EPOCH_PENDING |
1688                                               LLIF_SOM_DIRTY))
1689                           CWARN("ino %lu flags %lu still has size authority!"
1690                                 "do not trust the size got from MDS\n",
1691                                 inode->i_ino, lli->lli_flags);
1692                         else {
1693                                 i_size_write(inode, body->size);
1694                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1695                         }
1696                 } else {
1697                         i_size_write(inode, body->size);
1698                 }
1699
1700                 if (body->valid & OBD_MD_FLBLOCKS)
1701                         inode->i_blocks = body->blocks;
1702         }
1703
1704         if (body->valid & OBD_MD_FLID) {
1705                 /* FID shouldn't be changed! */
1706                 if (fid_is_sane(&lli->lli_fid)) {
1707                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1708                                  "Trying to change FID "DFID
1709                                  " to the "DFID", inode %lu/%u(%p)\n",
1710                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1711                                  inode->i_ino, inode->i_generation, inode);
1712                 } else
1713                         lli->lli_fid = body->fid1;
1714         }
1715
1716         LASSERT(fid_seq(&lli->lli_fid) != 0);
1717
1718         if (body->valid & OBD_MD_FLMDSCAPA) {
1719                 LASSERT(md->mds_capa);
1720                 ll_add_capa(inode, md->mds_capa);
1721         }
1722         if (body->valid & OBD_MD_FLOSSCAPA) {
1723                 LASSERT(md->oss_capa);
1724                 ll_add_capa(inode, md->oss_capa);
1725         }
1726 }
1727
1728 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1729 static struct backing_dev_info ll_backing_dev_info = {
1730         .ra_pages       = 0,    /* No readahead */
1731 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1732         .capabilities   = 0,    /* Does contribute to dirty memory */
1733 #else
1734         .memory_backed  = 0,    /* Does contribute to dirty memory */
1735 #endif
1736 };
1737 #endif
1738
1739 void ll_read_inode2(struct inode *inode, void *opaque)
1740 {
1741         struct lustre_md *md = opaque;
1742         struct ll_inode_info *lli = ll_i2info(inode);
1743         ENTRY;
1744
1745         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1746                inode->i_ino, inode->i_generation, inode);
1747
1748         ll_lli_init(lli);
1749
1750         LASSERT(!lli->lli_smd);
1751
1752         /* Core attributes from the MDS first.  This is a new inode, and
1753          * the VFS doesn't zero times in the core inode so we have to do
1754          * it ourselves.  They will be overwritten by either MDS or OST
1755          * attributes - we just need to make sure they aren't newer. */
1756         LTIME_S(inode->i_mtime) = 0;
1757         LTIME_S(inode->i_atime) = 0;
1758         LTIME_S(inode->i_ctime) = 0;
1759         inode->i_rdev = 0;
1760         ll_update_inode(inode, md);
1761
1762         /* OIDEBUG(inode); */
1763
1764         if (S_ISREG(inode->i_mode)) {
1765                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1766                 inode->i_op = &ll_file_inode_operations;
1767                 inode->i_fop = sbi->ll_fop;
1768                 inode->i_mapping->a_ops = &ll_aops;
1769                 EXIT;
1770         } else if (S_ISDIR(inode->i_mode)) {
1771                 inode->i_op = &ll_dir_inode_operations;
1772                 inode->i_fop = &ll_dir_operations;
1773                 inode->i_mapping->a_ops = &ll_dir_aops;
1774                 EXIT;
1775         } else if (S_ISLNK(inode->i_mode)) {
1776                 inode->i_op = &ll_fast_symlink_inode_operations;
1777                 EXIT;
1778         } else {
1779                 inode->i_op = &ll_special_inode_operations;
1780
1781 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1782                 init_special_inode(inode, inode->i_mode,
1783                                    kdev_t_to_nr(inode->i_rdev));
1784
1785                 /* initializing backing dev info. */
1786                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1787 #else
1788                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1789 #endif
1790                 EXIT;
1791         }
1792 }
1793
1794 void ll_delete_inode(struct inode *inode)
1795 {
1796         struct ll_sb_info *sbi = ll_i2sbi(inode);
1797         int rc;
1798         ENTRY;
1799
1800         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1801         if (rc) {
1802                 CERROR("fid_delete() failed, rc %d\n", rc);
1803         }
1804         clear_inode(inode);
1805
1806         EXIT;
1807 }
1808
1809 int ll_iocontrol(struct inode *inode, struct file *file,
1810                  unsigned int cmd, unsigned long arg)
1811 {
1812         struct ll_sb_info *sbi = ll_i2sbi(inode);
1813         struct ptlrpc_request *req = NULL;
1814         int rc, flags = 0;
1815         ENTRY;
1816
1817         switch(cmd) {
1818         case EXT3_IOC_GETFLAGS: {
1819                 struct mdt_body *body;
1820                 struct obd_capa *oc;
1821
1822                 oc = ll_mdscapa_get(inode);
1823                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1824                                 OBD_MD_FLFLAGS, 0, &req);
1825                 capa_put(oc);
1826                 if (rc) {
1827                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1828                         RETURN(-abs(rc));
1829                 }
1830
1831                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1832                                       sizeof(*body));
1833
1834                 /*Now the ext3 will be packed directly back to client,
1835                  *no need convert here*/
1836                 flags = body->flags;
1837
1838                 ptlrpc_req_finished (req);
1839
1840                 RETURN(put_user(flags, (int *)arg));
1841         }
1842         case EXT3_IOC_SETFLAGS: {
1843                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1844                 struct obd_info oinfo = { { { 0 } } };
1845                 struct md_op_data *op_data;
1846
1847                 if (get_user(flags, (int *)arg))
1848                         RETURN(-EFAULT);
1849
1850                 oinfo.oi_md = lsm;
1851                 OBDO_ALLOC(oinfo.oi_oa);
1852                 if (!oinfo.oi_oa)
1853                         RETURN(-ENOMEM);
1854
1855                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1856                                              LUSTRE_OPC_ANY, NULL);
1857                 if (IS_ERR(op_data))
1858                         RETURN(PTR_ERR(op_data));
1859
1860                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1861                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1862                 rc = md_setattr(sbi->ll_md_exp, op_data,
1863                                 NULL, 0, NULL, 0, &req, NULL);
1864                 ll_finish_md_op_data(op_data);
1865                 ptlrpc_req_finished(req);
1866                 if (rc || lsm == NULL) {
1867                         OBDO_FREE(oinfo.oi_oa);
1868                         RETURN(rc);
1869                 }
1870
1871                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1872                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1873                 oinfo.oi_oa->o_flags = flags;
1874                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1875                                        OBD_MD_FLGROUP;
1876                 oinfo.oi_capa = ll_mdscapa_get(inode);
1877
1878                 obdo_from_inode(oinfo.oi_oa, inode,
1879                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1880                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1881                 capa_put(oinfo.oi_capa);
1882                 OBDO_FREE(oinfo.oi_oa);
1883                 if (rc) {
1884                         if (rc != -EPERM && rc != -EACCES)
1885                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1886                         RETURN(rc);
1887                 }
1888
1889                 inode->i_flags = ll_ext_to_inode_flags(flags |
1890                                                        MDS_BFLAG_EXT_FLAGS);
1891                 RETURN(0);
1892         }
1893         default:
1894                 RETURN(-ENOSYS);
1895         }
1896
1897         RETURN(0);
1898 }
1899
1900 int ll_flush_ctx(struct inode *inode)
1901 {
1902         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1903
1904         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1905
1906         obd_set_info_async(sbi->ll_md_exp,
1907                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1908                            0, NULL, NULL);
1909         obd_set_info_async(sbi->ll_dt_exp,
1910                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1911                            0, NULL, NULL);
1912         return 0;
1913 }
1914
1915 /* umount -f client means force down, don't save state */
1916 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1917 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1918 {
1919         struct super_block *sb = vfsmnt->mnt_sb;
1920 #else
1921 void ll_umount_begin(struct super_block *sb)
1922 {
1923 #endif
1924         struct lustre_sb_info *lsi = s2lsi(sb);
1925         struct ll_sb_info *sbi = ll_s2sbi(sb);
1926         struct obd_device *obd;
1927         struct obd_ioctl_data ioc_data = { 0 };
1928         ENTRY;
1929
1930 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1931         if (!(flags & MNT_FORCE)) {
1932                 EXIT;
1933                 return;
1934         }
1935 #endif
1936
1937         /* Tell the MGC we got umount -f */
1938         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1939
1940         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1941                sb->s_count, atomic_read(&sb->s_active));
1942
1943         obd = class_exp2obd(sbi->ll_md_exp);
1944         if (obd == NULL) {
1945                 CERROR("Invalid MDC connection handle "LPX64"\n",
1946                        sbi->ll_md_exp->exp_handle.h_cookie);
1947                 EXIT;
1948                 return;
1949         }
1950         obd->obd_force = 1;
1951         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1952                       &ioc_data, NULL);
1953
1954         obd = class_exp2obd(sbi->ll_dt_exp);
1955         if (obd == NULL) {
1956                 CERROR("Invalid LOV connection handle "LPX64"\n",
1957                        sbi->ll_dt_exp->exp_handle.h_cookie);
1958                 EXIT;
1959                 return;
1960         }
1961
1962         obd->obd_force = 1;
1963         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1964                       &ioc_data, NULL);
1965
1966         /* Really, we'd like to wait until there are no requests outstanding,
1967          * and then continue.  For now, we just invalidate the requests,
1968          * schedule, and hope.
1969          */
1970         schedule();
1971
1972         EXIT;
1973 }
1974
1975 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1976 {
1977         struct ll_sb_info *sbi = ll_s2sbi(sb);
1978         int err;
1979         __u32 read_only;
1980
1981         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1982                 read_only = *flags & MS_RDONLY;
1983                 err = obd_set_info_async(sbi->ll_md_exp,
1984                                          sizeof(KEY_READ_ONLY) - 1,
1985                                          KEY_READ_ONLY, sizeof(read_only),
1986                                          &read_only, NULL);
1987                 if (err) {
1988                         CERROR("Failed to change the read-only flag during "
1989                                "remount: %d\n", err);
1990                         return err;
1991                 }
1992
1993                 if (read_only)
1994                         sb->s_flags |= MS_RDONLY;
1995                 else
1996                         sb->s_flags &= ~MS_RDONLY;
1997         }
1998         return 0;
1999 }
2000
2001 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2002                   int offset, struct super_block *sb)
2003 {
2004         struct ll_sb_info *sbi = NULL;
2005         struct lustre_md md;
2006         int rc = 0;
2007         ENTRY;
2008
2009         LASSERT(*inode || sb);
2010         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2011         prune_deathrow(sbi, 1);
2012         memset(&md, 0, sizeof(struct lustre_md));
2013
2014         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2015                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2016         if (rc)
2017                 RETURN(rc);
2018
2019         if (*inode) {
2020                 ll_update_inode(*inode, &md);
2021         } else {
2022                 LASSERT(sb != NULL);
2023
2024                 /*
2025                  * At this point server returns to client's same fid as client
2026                  * generated for creating. So using ->fid1 is okay here.
2027                  */
2028                 LASSERT(fid_is_sane(&md.body->fid1));
2029
2030                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2031                 if (*inode == NULL || is_bad_inode(*inode)) {
2032                         if (md.lsm)
2033                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2034 #ifdef CONFIG_FS_POSIX_ACL
2035                         if (md.posix_acl) {
2036                                 posix_acl_release(md.posix_acl);
2037                                 md.posix_acl = NULL;
2038                         }
2039 #endif
2040                         rc = -ENOMEM;
2041                         CERROR("new_inode -fatal: rc %d\n", rc);
2042                         GOTO(out, rc);
2043                 }
2044         }
2045
2046         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2047                          ll_i2info(*inode)->lli_smd);
2048 out:
2049         md_free_lustre_md(sbi->ll_md_exp, &md);
2050         RETURN(rc);
2051 }
2052
2053 char *llap_origins[] = {
2054         [LLAP_ORIGIN_UNKNOWN] = "--",
2055         [LLAP_ORIGIN_READPAGE] = "rp",
2056         [LLAP_ORIGIN_READAHEAD] = "ra",
2057         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2058         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2059 };
2060
2061 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2062                                              struct list_head *list)
2063 {
2064         struct ll_async_page *llap;
2065         struct list_head *pos;
2066
2067         list_for_each(pos, list) {
2068                 if (pos == &sbi->ll_pglist)
2069                         return NULL;
2070                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2071                 if (llap->llap_page == NULL)
2072                         continue;
2073                 return llap;
2074         }
2075         LBUG();
2076         return NULL;
2077 }
2078
2079 int ll_obd_statfs(struct inode *inode, void *arg)
2080 {
2081         struct ll_sb_info *sbi = NULL;
2082         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2083         struct lov_obd *lov = NULL;
2084         struct obd_statfs stat_buf = {0};
2085         char *buf = NULL;
2086         struct obd_ioctl_data *data = NULL;
2087         __u32 type, index;
2088         int len = 0, rc;
2089
2090         if (!inode || !(sbi = ll_i2sbi(inode)))
2091                 GOTO(out_statfs, rc = -EINVAL);
2092
2093         rc = obd_ioctl_getdata(&buf, &len, arg);
2094         if (rc)
2095                 GOTO(out_statfs, rc);
2096
2097         data = (void*)buf;
2098         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2099             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2100                 GOTO(out_statfs, rc = -EINVAL);
2101
2102         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2103         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2104
2105         if (type == LL_STATFS_MDC) {
2106                 if (index > 0)
2107                         GOTO(out_statfs, rc = -ENODEV);
2108                 client_obd = class_exp2obd(sbi->ll_md_exp);
2109         } else if (type == LL_STATFS_LOV) {
2110                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2111                 lov = &lov_obd->u.lov;
2112
2113                 if ((index >= lov->desc.ld_tgt_count))
2114                         GOTO(out_statfs, rc = -ENODEV);
2115                 if (!lov->lov_tgts[index])
2116                         /* Try again with the next index */
2117                         GOTO(out_statfs, rc = -EAGAIN);
2118
2119                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2120                 if (!lov->lov_tgts[index]->ltd_active)
2121                         GOTO(out_uuid, rc = -ENODATA);
2122         }
2123
2124         if (!client_obd)
2125                 GOTO(out_statfs, rc = -EINVAL);
2126
2127         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2128         if (rc)
2129                 GOTO(out_statfs, rc);
2130
2131         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2132                 GOTO(out_statfs, rc = -EFAULT);
2133
2134 out_uuid:
2135         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2136                          data->ioc_plen2))
2137                 rc = -EFAULT;
2138
2139 out_statfs:
2140         if (buf)
2141                 obd_ioctl_freedata(buf, len);
2142         return rc;
2143 }
2144
2145 int ll_process_config(struct lustre_cfg *lcfg)
2146 {
2147         char *ptr;
2148         void *sb;
2149         struct lprocfs_static_vars lvars;
2150         unsigned long x;
2151         int rc = 0;
2152
2153         lprocfs_init_vars(llite, &lvars);
2154
2155         /* The instance name contains the sb: lustre-client-aacfe000 */
2156         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2157         if (!ptr || !*(++ptr))
2158                 return -EINVAL;
2159         if (sscanf(ptr, "%lx", &x) != 1)
2160                 return -EINVAL;
2161         sb = (void *)x;
2162         /* This better be a real Lustre superblock! */
2163         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2164
2165         /* Note we have not called client_common_fill_super yet, so
2166            proc fns must be able to handle that! */
2167         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2168                                       lcfg, sb);
2169         return(rc);
2170 }
2171
2172 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2173 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2174                                        struct inode *i1, struct inode *i2,
2175                                        const char *name, int namelen,
2176                                        int mode, __u32 opc, void *data)
2177 {
2178         LASSERT(i1 != NULL);
2179
2180         if (namelen > ll_i2sbi(i1)->ll_namelen)
2181                 return ERR_PTR(-ENAMETOOLONG);
2182
2183         if (op_data == NULL)
2184                 OBD_ALLOC_PTR(op_data);
2185
2186         if (op_data == NULL)
2187                 return ERR_PTR(-ENOMEM);
2188
2189         ll_i2gids(op_data->op_suppgids, i1, i2);
2190         op_data->op_fid1 = *ll_inode2fid(i1);
2191         op_data->op_capa1 = ll_mdscapa_get(i1);
2192
2193         if (i2) {
2194                 op_data->op_fid2 = *ll_inode2fid(i2);
2195                 op_data->op_capa2 = ll_mdscapa_get(i2);
2196         } else {
2197                 fid_zero(&op_data->op_fid2);
2198         }
2199
2200         op_data->op_name = name;
2201         op_data->op_namelen = namelen;
2202         op_data->op_mode = mode;
2203         op_data->op_mod_time = CURRENT_SECONDS;
2204         op_data->op_fsuid = current->fsuid;
2205         op_data->op_fsgid = current->fsgid;
2206         op_data->op_cap = current->cap_effective;
2207         op_data->op_bias = MDS_CHECK_SPLIT;
2208         op_data->op_opc = opc;
2209         op_data->op_mds = 0;
2210         op_data->op_data = data;
2211
2212         return op_data;
2213 }
2214
2215 void ll_finish_md_op_data(struct md_op_data *op_data)
2216 {
2217         capa_put(op_data->op_capa1);
2218         capa_put(op_data->op_capa2);
2219         OBD_FREE_PTR(op_data);
2220 }
2221
2222 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2223 {
2224         struct ll_sb_info *sbi = ll_i2sbi(inode);
2225         struct ptlrpc_request *req = NULL;
2226         struct mdt_body *body;
2227         char *cmd, *buf;
2228         struct obd_capa *oc;
2229         int rc, buflen;
2230         ENTRY;
2231
2232         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2233                 RETURN(-EBADE);
2234
2235         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2236
2237         OBD_ALLOC(cmd, ioc->cmd_len);
2238         if (!cmd)
2239                 RETURN(-ENOMEM);
2240         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2241                 GOTO(out, rc = -EFAULT);
2242
2243         oc = ll_mdscapa_get(inode);
2244         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2245                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2246                          ioc->cmd_len, ioc->res_len, 0, &req);
2247         capa_put(oc);
2248         if (rc < 0) {
2249                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2250                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2251                 GOTO(out, rc);
2252         }
2253
2254         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2255         LASSERT(body);
2256
2257         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2258         LASSERT(buflen <= ioc->res_len);
2259         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2260         LASSERT(buf);
2261         if (copy_to_user(ioc->res, buf, buflen))
2262                 GOTO(out, rc = -EFAULT);
2263         EXIT;
2264 out:
2265         if (req)
2266                 ptlrpc_req_finished(req);
2267         OBD_FREE(cmd, ioc->cmd_len);
2268         return rc;
2269 }
2270
2271 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2272 {
2273         struct ll_sb_info *sbi = ll_i2sbi(inode);
2274         struct ptlrpc_request *req = NULL;
2275         char *cmd, *buf;
2276         struct obd_capa *oc;
2277         int buflen, rc;
2278         ENTRY;
2279
2280         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2281                 RETURN(-EBADE);
2282
2283         if (!(sbi->ll_flags & LL_SBI_ACL))
2284                 RETURN(-EOPNOTSUPP);
2285
2286         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2287
2288         OBD_ALLOC(cmd, ioc->cmd_len);
2289         if (!cmd)
2290                 RETURN(-ENOMEM);
2291         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2292                 GOTO(out, rc = -EFAULT);
2293
2294         oc = ll_mdscapa_get(inode);
2295         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2296                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2297                          ioc->cmd_len, ioc->res_len, 0, &req);
2298         capa_put(oc);
2299         if (rc) {
2300                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2301                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2302                 GOTO(out, rc);
2303         }
2304
2305         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2306         LASSERT(buflen <= ioc->res_len);
2307         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2308         LASSERT(buf);
2309         if (copy_to_user(ioc->res, buf, buflen))
2310                 GOTO(out, rc = -EFAULT);
2311         EXIT;
2312 out:
2313         if (req)
2314                 ptlrpc_req_finished(req);
2315         OBD_FREE(cmd, ioc->cmd_len);
2316         return rc;
2317 }