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 HAVE_EXPORT___IGET
  91         INIT_LIST_HEAD(&sbi->ll_deathrow);
  92         spin_lock_init(&sbi->ll_deathrow_lock);
  93 #endif
  94         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
  95                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
  96                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
  97         }
  98
  99         RETURN(sbi);
 100 }
 101
 102 void ll_free_sbi(struct super_block *sb)
 103 {
 104         struct ll_sb_info *sbi = ll_s2sbi(sb);
 105         ENTRY;
 106
 107         if (sbi != NULL) {
 108                 spin_lock(&ll_sb_lock);
 109                 list_del(&sbi->ll_list);
 110                 spin_unlock(&ll_sb_lock);
 111                 OBD_FREE(sbi, sizeof(*sbi));
 112         }
 113         EXIT;
 114 }
 115
 116 static struct dentry_operations ll_d_root_ops = {
 117 #ifdef LUSTRE_KERNEL_VERSION
 118         .d_compare = ll_dcompare,
 119 #endif
 120 };
 121
 122 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 123  * us to make MDS RPCs with large enough reply buffers to hold the
 124  * maximum-sized (= maximum striped) EA and cookie without having to
 125  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 126 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 127 {
 128         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 129         __u32 valsize = sizeof(struct lov_desc);
 130         int rc, easize, def_easize, cookiesize;
 131         struct lov_desc desc;
 132         __u32 stripes;
 133         ENTRY;
 134
 135         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 136                           &valsize, &desc);
 137         if (rc)
 138                 RETURN(rc);
 139
 140         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 141         lsm.lsm_stripe_count = stripes;
 142         easize = obd_size_diskmd(dt_exp, &lsm);
 143
 144         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 145         def_easize = obd_size_diskmd(dt_exp, &lsm);
 146
 147         cookiesize = stripes * sizeof(struct llog_cookie);
 148
 149         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 150                easize, cookiesize);
 151
 152         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 153         RETURN(rc);
 154 }
 155
 156 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 157                                     uid_t nllu, gid_t nllg)
 158 {
 159         struct inode *root = 0;
 160         struct ll_sb_info *sbi = ll_s2sbi(sb);
 161         struct obd_device *obd;
 162         struct lu_fid rootfid;
 163         struct obd_capa *oc = NULL;
 164         struct obd_statfs osfs;
 165         struct ptlrpc_request *request = NULL;
 166         struct lustre_handle dt_conn = {0, };
 167         struct lustre_handle md_conn = {0, };
 168         struct obd_connect_data *data = NULL;
 169         struct lustre_md lmd;
 170         obd_valid valid;
 171         int size, err;
 172         ENTRY;
 173
 174         obd = class_name2obd(md);
 175         if (!obd) {
 176                 CERROR("MD %s: not setup or attached\n", md);
 177                 RETURN(-EINVAL);
 178         }
 179
 180         OBD_ALLOC_PTR(data);
 181         if (data == NULL)
 182                 RETURN(-ENOMEM);
 183
 184         if (proc_lustre_fs_root) {
 185                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 186                                                   dt, md);
 187                 if (err < 0)
 188                         CERROR("could not register mount in /proc/lustre");
 189         }
 190
 191         /* indicate the features supported by this client */
 192         data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
 193                                   OBD_CONNECT_JOIN |
 194                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
 195                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 196                                   OBD_CONNECT_CANCELSET;
 197 #ifdef CONFIG_FS_POSIX_ACL
 198         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 199 #endif
 200         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 201         data->ocd_version = LUSTRE_VERSION_CODE;
 202
 203         if (sb->s_flags & MS_RDONLY)
 204                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 205         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 206                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 207
 208 #ifdef HAVE_MS_FLOCK_LOCK
 209         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 210         sb->s_flags |= MS_FLOCK_LOCK;
 211 #endif
 212
 213         if (sbi->ll_flags & LL_SBI_FLOCK)
 214                 sbi->ll_fop = &ll_file_operations_flock;
 215         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 216                 sbi->ll_fop = &ll_file_operations;
 217         else
 218                 sbi->ll_fop = &ll_file_operations_noflock;
 219
 220         /* real client */
 221         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 222         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 223                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 224                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 225         } else {
 226                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 227                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 228         }
 229         data->ocd_nllu = nllu;
 230         data->ocd_nllg = nllg;
 231
 232         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
 233         if (err == -EBUSY) {
 234                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 235                                    "recovery, of which this client is not a "
 236                                    "part. Please wait for recovery to complete,"
 237                                    " abort, or time out.\n", md);
 238                 GOTO(out, err);
 239         } else if (err) {
 240                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 241                 GOTO(out, err);
 242         }
 243         sbi->ll_md_exp = class_conn2export(&md_conn);
 244
 245         err = obd_fid_init(sbi->ll_md_exp);
 246         if (err) {
 247                 CERROR("Can't init metadata layer FID infrastructure, "
 248                        "rc %d\n", err);
 249                 GOTO(out_md, err);
 250         }
 251
 252         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
 253         if (err)
 254                 GOTO(out_md_fid, err);
 255
 256         size = sizeof(*data);
 257         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 258                            KEY_CONN_DATA,  &size, data);
 259         if (err) {
 260                 CERROR("Get connect data failed: %d \n", err);
 261                 GOTO(out_md, err);
 262         }
 263
 264         LASSERT(osfs.os_bsize);
 265         sb->s_blocksize = osfs.os_bsize;
 266         sb->s_blocksize_bits = log2(osfs.os_bsize);
 267         sb->s_magic = LL_SUPER_MAGIC;
 268
 269         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 270          *         retval = in_file->f_op->sendfile(...);
 271          *         if (*ppos > max)
 272          *                 retval = -EOVERFLOW;
 273          *
 274          * it will check if *ppos is greater than max. However, max equals to
 275          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 276          * has been defined as a signed long long ineger in linux kernel. */
 277 #if BITS_PER_LONG == 64
 278         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 279 #else
 280         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 281 #endif
 282         sbi->ll_namelen = osfs.os_namelen;
 283         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 284
 285         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 286             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 287                 LCONSOLE_INFO("Disabling user_xattr feature because "
 288                               "it is not supported on the server\n");
 289                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 290         }
 291
 292         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 293 #ifdef MS_POSIXACL
 294                 sb->s_flags |= MS_POSIXACL;
 295 #endif
 296                 sbi->ll_flags |= LL_SBI_ACL;
 297         } else {
 298                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 299 #ifdef MS_POSIXACL
 300                 sb->s_flags &= ~MS_POSIXACL;
 301 #endif
 302                 sbi->ll_flags &= ~LL_SBI_ACL;
 303         }
 304
 305         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 306                 sbi->ll_flags |= LL_SBI_JOIN;
 307
 308         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 309                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 310                         /* sometimes local client claims to be remote, but mdt
 311                          * will disagree when client gss not applied. */
 312                         LCONSOLE_INFO("client claims to be remote, but server "
 313                                       "rejected, forced to be local.\n");
 314                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 315                 }
 316         } else {
 317                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 318                         /* with gss applied, remote client can not claim to be
 319                          * local, so mdt maybe force client to be remote. */
 320                         LCONSOLE_INFO("client claims to be local, but server "
 321                                       "rejected, forced to be remote.\n");
 322                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 323                 }
 324         }
 325
 326         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 327                 LCONSOLE_INFO("client enabled MDS capability!\n");
 328                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 332                 LCONSOLE_INFO("client enabled OSS capability!\n");
 333                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 334         }
 335
 336 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 337         /* We set sb->s_dev equal on all lustre clients in order to support
 338          * NFS export clustering.  NFSD requires that the FSID be the same
 339          * on all clients. */
 340         /* s_dev is also used in lt_compare() to compare two fs, but that is
 341          * only a node-local comparison. */
 342
 343         /* XXX: this will not work with LMV */
 344         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 345                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 346 #endif
 347
 348         obd = class_name2obd(dt);
 349         if (!obd) {
 350                 CERROR("DT %s: not setup or attached\n", dt);
 351                 GOTO(out_md_fid, err = -ENODEV);
 352         }
 353
 354         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 355                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 356                                   OBD_CONNECT_CANCELSET;
 357         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 358                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 359
 360         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 361                "ocd_grant: %d\n", data->ocd_connect_flags,
 362                data->ocd_version, data->ocd_grant);
 363
 364         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 365         obd->obd_upcall.onu_upcall = ll_ocd_update;
 366         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 367
 368         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
 369         if (err == -EBUSY) {
 370                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 371                                    "recovery, of which this client is not a "
 372                                    "part.  Please wait for recovery to "
 373                                    "complete, abort, or time out.\n", dt);
 374                 GOTO(out_md_fid, err);
 375         } else if (err) {
 376                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 377                 GOTO(out_md_fid, err);
 378         }
 379
 380         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 381
 382         err = obd_fid_init(sbi->ll_dt_exp);
 383         if (err) {
 384                 CERROR("Can't init data layer FID infrastructure, "
 385                        "rc %d\n", err);
 386                 GOTO(out_dt, err);
 387         }
 388
 389         spin_lock(&sbi->ll_lco.lco_lock);
 390         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 391         spin_unlock(&sbi->ll_lco.lco_lock);
 392
 393         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 394
 395         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 396                                   0, NULL, NULL, NULL);
 397         if (err < 0) {
 398                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 399                                    "filesystem. There must be at least one "
 400                                    "active OST for a client to start.\n");
 401                 GOTO(out_dt_fid, err);
 402         }
 403
 404         if (!ll_async_page_slab) {
 405                 ll_async_page_slab_size =
 406                         size_round(sizeof(struct ll_async_page)) + err;
 407                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 408                                                         ll_async_page_slab_size,
 409                                                            0, 0);
 410                 if (!ll_async_page_slab)
 411                         GOTO(out_dt_fid, err = -ENOMEM);
 412         }
 413
 414         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 415         if (err) {
 416                 CERROR("cannot mds_connect: rc = %d\n", err);
 417                 GOTO(out_dt_fid, err);
 418         }
 419         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 420         sbi->ll_root_fid = rootfid;
 421
 422         sb->s_op = &lustre_super_operations;
 423 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 424         sb->s_export_op = &lustre_export_operations;
 425 #endif
 426
 427         /* make root inode
 428          * XXX: move this to after cbd setup? */
 429         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 430         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 431                 valid |= OBD_MD_FLRMTPERM;
 432         else if (sbi->ll_flags & LL_SBI_ACL)
 433                 valid |= OBD_MD_FLACL;
 434
 435         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 436         if (oc)
 437                 free_capa(oc);
 438         if (err) {
 439                 CERROR("md_getattr failed for root: rc = %d\n", err);
 440                 GOTO(out_dt_fid, err);
 441         }
 442         memset(&lmd, 0, sizeof(lmd));
 443         err = md_get_lustre_md(sbi->ll_md_exp, request,
 444                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
 445                                &lmd);
 446         if (err) {
 447                 CERROR("failed to understand root inode md: rc = %d\n", err);
 448                 ptlrpc_req_finished (request);
 449                 GOTO(out_dt_fid, err);
 450         }
 451
 452         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 453         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 454         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 455         ptlrpc_req_finished(request);
 456
 457         if (root == NULL || is_bad_inode(root)) {
 458                 if (lmd.lsm)
 459                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 460 #ifdef CONFIG_FS_POSIX_ACL
 461                 if (lmd.posix_acl) {
 462                         posix_acl_release(lmd.posix_acl);
 463                         lmd.posix_acl = NULL;
 464                 }
 465 #endif
 466                 CERROR("lustre_lite: bad iget4 for root\n");
 467                 GOTO(out_root, err = -EBADF);
 468         }
 469
 470         err = ll_close_thread_start(&sbi->ll_lcq);
 471         if (err) {
 472                 CERROR("cannot start close thread: rc %d\n", err);
 473                 GOTO(out_root, err);
 474         }
 475
 476         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 477            backing dev info assigned to inode mapping is used for
 478            determining maximal readahead. */
 479 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 480     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 481         /* bug 2805 - set VM readahead to zero */
 482         vm_max_readahead = vm_min_readahead = 0;
 483 #endif
 484
 485         sb->s_root = d_alloc_root(root);
 486         if (data != NULL)
 487                 OBD_FREE(data, sizeof(*data));
 488         sb->s_root->d_op = &ll_d_root_ops;
 489         RETURN(err);
 490 out_root:
 491         if (root)
 492                 iput(root);
 493 out_dt_fid:
 494         obd_fid_fini(sbi->ll_dt_exp);
 495 out_dt:
 496         obd_disconnect(sbi->ll_dt_exp);
 497         sbi->ll_dt_exp = NULL;
 498 out_md_fid:
 499         obd_fid_fini(sbi->ll_md_exp);
 500 out_md:
 501         obd_disconnect(sbi->ll_md_exp);
 502         sbi->ll_md_exp = NULL;
 503 out:
 504         if (data != NULL)
 505                 OBD_FREE_PTR(data);
 506         lprocfs_unregister_mountpoint(sbi);
 507         return err;
 508 }
 509
 510 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 511 {
 512         int size, rc;
 513
 514         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 515         size = sizeof(int);
 516         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 517                           &size, lmmsize);
 518         if (rc)
 519                 CERROR("Get max mdsize error rc %d \n", rc);
 520
 521         RETURN(rc);
 522 }
 523
 524 void ll_dump_inode(struct inode *inode)
 525 {
 526         struct list_head *tmp;
 527         int dentry_count = 0;
 528
 529         LASSERT(inode != NULL);
 530
 531         list_for_each(tmp, &inode->i_dentry)
 532                 dentry_count++;
 533
 534         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 535                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 536                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 537 }
 538
 539 void lustre_dump_dentry(struct dentry *dentry, int recur)
 540 {
 541         struct list_head *tmp;
 542         int subdirs = 0;
 543
 544         LASSERT(dentry != NULL);
 545
 546         list_for_each(tmp, &dentry->d_subdirs)
 547                 subdirs++;
 548
 549         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 550                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 551                dentry->d_name.len, dentry->d_name.name,
 552                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 553                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 554                dentry->d_flags, dentry->d_fsdata, subdirs);
 555         if (dentry->d_inode != NULL)
 556                 ll_dump_inode(dentry->d_inode);
 557
 558         if (recur == 0)
 559                 return;
 560
 561         list_for_each(tmp, &dentry->d_subdirs) {
 562                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 563                 lustre_dump_dentry(d, recur - 1);
 564         }
 565 }
 566
 567 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 568 void lustre_throw_orphan_dentries(struct super_block *sb)
 569 {
 570         struct dentry *dentry, *next;
 571         struct ll_sb_info *sbi = ll_s2sbi(sb);
 572
 573         /* Do this to get rid of orphaned dentries. That is not really trw. */
 574         list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
 575                                  d_hash) {
 576                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 577                       "before and after shrink_dcache_parent\n",
 578                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 579                 lustre_dump_dentry(dentry, 1);
 580                 shrink_dcache_parent(dentry);
 581                 lustre_dump_dentry(dentry, 1);
 582         }
 583 }
 584 #else
 585 #define lustre_throw_orphan_dentries(sb)
 586 #endif
 587
 588 #ifdef HAVE_EXPORT___IGET
 589 static void prune_dir_dentries(struct inode *inode)
 590 {
 591         struct dentry *dentry, *prev = NULL;
 592
 593         /* due to lustre specific logic, a directory
 594          * can have few dentries - a bug from VFS POV */
 595 restart:
 596         spin_lock(&dcache_lock);
 597         if (!list_empty(&inode->i_dentry)) {
 598                 dentry = list_entry(inode->i_dentry.prev,
 599                                     struct dentry, d_alias);
 600                 /* in order to prevent infinite loops we
 601                  * break if previous dentry is busy */
 602                 if (dentry != prev) {
 603                         prev = dentry;
 604                         dget_locked(dentry);
 605                         spin_unlock(&dcache_lock);
 606
 607                         /* try to kill all child dentries */
 608                         lock_dentry(dentry);
 609                         shrink_dcache_parent(dentry);
 610                         unlock_dentry(dentry);
 611                         dput(dentry);
 612
 613                         /* now try to get rid of current dentry */
 614                         d_prune_aliases(inode);
 615                         goto restart;
 616                 }
 617         }
 618         spin_unlock(&dcache_lock);
 619 }
 620
 621 static void prune_deathrow_one(struct ll_inode_info *lli)
 622 {
 623         struct inode *inode = ll_info2i(lli);
 624
 625         /* first, try to drop any dentries - they hold a ref on the inode */
 626         if (S_ISDIR(inode->i_mode))
 627                 prune_dir_dentries(inode);
 628         else
 629                 d_prune_aliases(inode);
 630
 631
 632         /* if somebody still uses it, leave it */
 633         LASSERT(atomic_read(&inode->i_count) > 0);
 634         if (atomic_read(&inode->i_count) > 1)
 635                 goto out;
 636
 637         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 638                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 639
 640         /* seems nobody uses it anymore */
 641         inode->i_nlink = 0;
 642
 643 out:
 644         iput(inode);
 645         return;
 646 }
 647
 648 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 649 {
 650         struct ll_inode_info *lli;
 651         int empty;
 652
 653         do {
 654                 if (need_resched() && try)
 655                         break;
 656
 657                 if (try) {
 658                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 659                                 break;
 660                 } else {
 661                         spin_lock(&sbi->ll_deathrow_lock);
 662                 }
 663
 664                 empty = 1;
 665                 lli = NULL;
 666                 if (!list_empty(&sbi->ll_deathrow)) {
 667                         lli = list_entry(sbi->ll_deathrow.next,
 668                                          struct ll_inode_info,
 669                                          lli_dead_list);
 670                         list_del_init(&lli->lli_dead_list);
 671                         if (!list_empty(&sbi->ll_deathrow))
 672                                 empty = 0;
 673                 }
 674                 spin_unlock(&sbi->ll_deathrow_lock);
 675
 676                 if (lli)
 677                         prune_deathrow_one(lli);
 678
 679         } while (empty == 0);
 680 }
 681 #else /* !HAVE_EXPORT___IGET */
 682 #define prune_deathrow(sbi, try) do {} while (0)
 683 #endif /* HAVE_EXPORT___IGET */
 684
 685 void client_common_put_super(struct super_block *sb)
 686 {
 687         struct ll_sb_info *sbi = ll_s2sbi(sb);
 688         ENTRY;
 689
 690         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 691
 692         ll_close_thread_shutdown(sbi->ll_lcq);
 693
 694         /* destroy inodes in deathrow */
 695         prune_deathrow(sbi, 0);
 696
 697         list_del(&sbi->ll_conn_chain);
 698
 699         obd_fid_fini(sbi->ll_dt_exp);
 700         obd_disconnect(sbi->ll_dt_exp);
 701         sbi->ll_dt_exp = NULL;
 702
 703         lprocfs_unregister_mountpoint(sbi);
 704
 705         obd_fid_fini(sbi->ll_md_exp);
 706         obd_disconnect(sbi->ll_md_exp);
 707         sbi->ll_md_exp = NULL;
 708
 709         lustre_throw_orphan_dentries(sb);
 710         EXIT;
 711 }
 712
 713 char *ll_read_opt(const char *opt, char *data)
 714 {
 715         char *value;
 716         char *retval;
 717         ENTRY;
 718
 719         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 720         if (strncmp(opt, data, strlen(opt)))
 721                 RETURN(NULL);
 722         if ((value = strchr(data, '=')) == NULL)
 723                 RETURN(NULL);
 724
 725         value++;
 726         OBD_ALLOC(retval, strlen(value) + 1);
 727         if (!retval) {
 728                 CERROR("out of memory!\n");
 729                 RETURN(NULL);
 730         }
 731
 732         memcpy(retval, value, strlen(value)+1);
 733         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 734         RETURN(retval);
 735 }
 736
 737 static inline int ll_set_opt(const char *opt, char *data, int fl)
 738 {
 739         if (strncmp(opt, data, strlen(opt)) != 0)
 740                 return(0);
 741         else
 742                 return(fl);
 743 }
 744
 745 /* non-client-specific mount options are parsed in lmd_parse */
 746 static int ll_options(char *options, int *flags)
 747 {
 748         int tmp;
 749         char *s1 = options, *s2;
 750         ENTRY;
 751
 752         if (!options)
 753                 RETURN(0);
 754
 755         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 756
 757         while (*s1) {
 758                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 759                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 760                 if (tmp) {
 761                         *flags |= tmp;
 762                         goto next;
 763                 }
 764                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 765                 if (tmp) {
 766                         *flags |= tmp;
 767                         goto next;
 768                 }
 769                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 770                 if (tmp) {
 771                         *flags |= tmp;
 772                         goto next;
 773                 }
 774                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 775                 if (tmp) {
 776                         *flags &= ~tmp;
 777                         goto next;
 778                 }
 779                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 780                 if (tmp) {
 781                         *flags |= tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 785                 if (tmp) {
 786                         *flags &= ~tmp;
 787                         goto next;
 788                 }
 789                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 790                 if (tmp) {
 791                         /* Ignore deprecated mount option.  The client will
 792                          * always try to mount with ACL support, whether this
 793                          * is used depends on whether server supports it. */
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 797                 if (tmp) {
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805
 806                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 807                                    s1);
 808                 RETURN(-EINVAL);
 809
 810 next:
 811                 /* Find next opt */
 812                 s2 = strchr(s1, ',');
 813                 if (s2 == NULL)
 814                         break;
 815                 s1 = s2 + 1;
 816         }
 817         RETURN(0);
 818 }
 819
 820 void ll_lli_init(struct ll_inode_info *lli)
 821 {
 822         lli->lli_inode_magic = LLI_INODE_MAGIC;
 823         sema_init(&lli->lli_size_sem, 1);
 824         sema_init(&lli->lli_write_sem, 1);
 825         lli->lli_flags = 0;
 826         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 827         spin_lock_init(&lli->lli_lock);
 828         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 829         INIT_LIST_HEAD(&lli->lli_close_list);
 830         lli->lli_inode_magic = LLI_INODE_MAGIC;
 831         sema_init(&lli->lli_och_sem, 1);
 832         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 833         lli->lli_mds_exec_och = NULL;
 834         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 835         lli->lli_open_fd_exec_count = 0;
 836         INIT_LIST_HEAD(&lli->lli_dead_list);
 837         lli->lli_remote_perms = NULL;
 838         lli->lli_rmtperm_utime = 0;
 839         sema_init(&lli->lli_rmtperm_sem, 1);
 840         INIT_LIST_HEAD(&lli->lli_oss_capas);
 841 }
 842
 843 int ll_fill_super(struct super_block *sb)
 844 {
 845         struct lustre_profile *lprof;
 846         struct lustre_sb_info *lsi = s2lsi(sb);
 847         struct ll_sb_info *sbi;
 848         char  *dt = NULL, *md = NULL;
 849         char  *profilenm = get_profile_name(sb);
 850         struct config_llog_instance cfg = {0, };
 851         char   ll_instance[sizeof(sb) * 2 + 1];
 852         int    err;
 853         ENTRY;
 854
 855         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 856
 857         cfs_module_get();
 858
 859         sb->s_type->fs_flags |= FS_ODD_RENAME;
 860         /* client additional sb info */
 861         lsi->lsi_llsbi = sbi = ll_init_sbi();
 862         if (!sbi) {
 863                 cfs_module_put();
 864                 RETURN(-ENOMEM);
 865         }
 866
 867         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 868         if (err)
 869                 GOTO(out_free, err);
 870
 871         /* Generate a string unique to this super, in case some joker tries
 872            to mount the same fs at two mount points.
 873            Use the address of the super itself.*/
 874         sprintf(ll_instance, "%p", sb);
 875         cfg.cfg_instance = ll_instance;
 876         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 877
 878         /* set up client obds */
 879         err = lustre_process_log(sb, profilenm, &cfg);
 880         if (err < 0) {
 881                 CERROR("Unable to process log: %d\n", err);
 882                 GOTO(out_free, err);
 883         }
 884
 885         lprof = class_get_profile(profilenm);
 886         if (lprof == NULL) {
 887                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 888                                    " read from the MGS.  Does that filesystem "
 889                                    "exist?\n", profilenm);
 890                 GOTO(out_free, err = -EINVAL);
 891         }
 892         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 893                lprof->lp_md, lprof->lp_dt);
 894
 895         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 896                   strlen(ll_instance) + 2);
 897         if (!dt)
 898                 GOTO(out_free, err = -ENOMEM);
 899         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 900
 901         OBD_ALLOC(md, strlen(lprof->lp_md) +
 902                   strlen(ll_instance) + 2);
 903         if (!md)
 904                 GOTO(out_free, err = -ENOMEM);
 905         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 906
 907         /* connections, registrations, sb setup */
 908         err = client_common_fill_super(sb, md, dt,
 909                                        lsi->lsi_lmd->lmd_nllu,
 910                                        lsi->lsi_lmd->lmd_nllg);
 911
 912 out_free:
 913         if (md)
 914                 OBD_FREE(md, strlen(md) + 1);
 915         if (dt)
 916                 OBD_FREE(dt, strlen(dt) + 1);
 917         if (err)
 918                 ll_put_super(sb);
 919         else
 920                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 921
 922         RETURN(err);
 923 } /* ll_fill_super */
 924
 925
 926 void ll_put_super(struct super_block *sb)
 927 {
 928         struct config_llog_instance cfg;
 929         char   ll_instance[sizeof(sb) * 2 + 1];
 930         struct obd_device *obd;
 931         struct lustre_sb_info *lsi = s2lsi(sb);
 932         struct ll_sb_info *sbi = ll_s2sbi(sb);
 933         char *profilenm = get_profile_name(sb);
 934         int force = 1, next;
 935         ENTRY;
 936
 937         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 938
 939         ll_print_capa_stat(sbi);
 940
 941         sprintf(ll_instance, "%p", sb);
 942         cfg.cfg_instance = ll_instance;
 943         lustre_end_log(sb, NULL, &cfg);
 944
 945         if (sbi->ll_md_exp) {
 946                 obd = class_exp2obd(sbi->ll_md_exp);
 947                 if (obd)
 948                         force = obd->obd_force;
 949         }
 950
 951         /* We need to set force before the lov_disconnect in
 952            lustre_common_put_super, since l_d cleans up osc's as well. */
 953         if (force) {
 954                 next = 0;
 955                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 956                                                      &next)) != NULL) {
 957                         obd->obd_force = force;
 958                 }
 959         }
 960
 961         if (sbi->ll_lcq) {
 962                 /* Only if client_common_fill_super succeeded */
 963                 client_common_put_super(sb);
 964         }
 965         next = 0;
 966         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 967                 class_manual_cleanup(obd);
 968         }
 969
 970         if (profilenm)
 971                 class_del_profile(profilenm);
 972
 973         ll_free_sbi(sb);
 974         lsi->lsi_llsbi = NULL;
 975
 976         lustre_common_put_super(sb);
 977
 978         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
 979
 980         cfs_module_put();
 981
 982         EXIT;
 983 } /* client_put_super */
 984
 985 #ifdef HAVE_REGISTER_CACHE
 986 #include <linux/cache_def.h>
 987 #ifdef HAVE_CACHE_RETURN_INT
 988 static int
 989 #else
 990 static void
 991 #endif
 992 ll_shrink_cache(int priority, unsigned int gfp_mask)
 993 {
 994         struct ll_sb_info *sbi;
 995         int count = 0;
 996
 997         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
 998                 count += llap_shrink_cache(sbi, priority);
 999
1000 #ifdef HAVE_CACHE_RETURN_INT
1001         return count;
1002 #endif
1003 }
1004
1005 struct cache_definition ll_cache_definition = {
1006         .name = "llap_cache",
1007         .shrink = ll_shrink_cache
1008 };
1009 #endif /* HAVE_REGISTER_CACHE */
1010
1011 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1012 {
1013         struct inode *inode = NULL;
1014         /* NOTE: we depend on atomic igrab() -bzzz */
1015         lock_res_and_lock(lock);
1016         if (lock->l_ast_data) {
1017                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1018                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1019                         inode = igrab(lock->l_ast_data);
1020                 } else {
1021                         inode = lock->l_ast_data;
1022                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1023                                                 D_INFO : D_WARNING,
1024                                         lock, __FILE__, __func__, __LINE__,
1025                                         "l_ast_data %p is bogus: magic %08x",
1026                                         lock->l_ast_data, lli->lli_inode_magic);
1027                         inode = NULL;
1028                 }
1029         }
1030         unlock_res_and_lock(lock);
1031         return inode;
1032 }
1033
1034 static int null_if_equal(struct ldlm_lock *lock, void *data)
1035 {
1036         if (data == lock->l_ast_data) {
1037                 lock->l_ast_data = NULL;
1038
1039                 if (lock->l_req_mode != lock->l_granted_mode)
1040                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1041         }
1042
1043         return LDLM_ITER_CONTINUE;
1044 }
1045
1046 void ll_clear_inode(struct inode *inode)
1047 {
1048         struct ll_inode_info *lli = ll_i2info(inode);
1049         struct ll_sb_info *sbi = ll_i2sbi(inode);
1050         ENTRY;
1051
1052         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1053                inode->i_generation, inode);
1054
1055         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1056         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1057                          null_if_equal, inode);
1058
1059         LASSERT(!lli->lli_open_fd_write_count);
1060         LASSERT(!lli->lli_open_fd_read_count);
1061         LASSERT(!lli->lli_open_fd_exec_count);
1062
1063         if (lli->lli_mds_write_och)
1064                 ll_md_real_close(inode, FMODE_WRITE);
1065         if (lli->lli_mds_exec_och)
1066                 ll_md_real_close(inode, FMODE_EXEC);
1067         if (lli->lli_mds_read_och)
1068                 ll_md_real_close(inode, FMODE_READ);
1069
1070         if (lli->lli_smd) {
1071                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1072                                   null_if_equal, inode);
1073
1074                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1075                 lli->lli_smd = NULL;
1076         }
1077
1078         if (lli->lli_symlink_name) {
1079                 OBD_FREE(lli->lli_symlink_name,
1080                          strlen(lli->lli_symlink_name) + 1);
1081                 lli->lli_symlink_name = NULL;
1082         }
1083
1084         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1085                 LASSERT(lli->lli_posix_acl == NULL);
1086                 if (lli->lli_remote_perms) {
1087                         free_rmtperm_hash(lli->lli_remote_perms);
1088                         lli->lli_remote_perms = NULL;
1089                 }
1090         }
1091 #ifdef CONFIG_FS_POSIX_ACL
1092         else if (lli->lli_posix_acl) {
1093                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1094                 LASSERT(lli->lli_remote_perms == NULL);
1095                 posix_acl_release(lli->lli_posix_acl);
1096                 lli->lli_posix_acl = NULL;
1097         }
1098 #endif
1099         lli->lli_inode_magic = LLI_INODE_DEAD;
1100
1101 #ifdef HAVE_EXPORT___IGET
1102         spin_lock(&sbi->ll_deathrow_lock);
1103         list_del_init(&lli->lli_dead_list);
1104         spin_unlock(&sbi->ll_deathrow_lock);
1105 #endif
1106         ll_clear_inode_capas(inode);
1107
1108         EXIT;
1109 }
1110
1111 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data)
1112 {
1113         struct lustre_md md;
1114         struct ll_sb_info *sbi = ll_i2sbi(inode);
1115         struct ptlrpc_request *request = NULL;
1116         int rc;
1117         ENTRY;
1118
1119         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1120                                      LUSTRE_OPC_ANY, NULL);
1121         if (IS_ERR(op_data))
1122                 RETURN(PTR_ERR(op_data));
1123
1124         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1125                         &request);
1126         if (rc) {
1127                 ptlrpc_req_finished(request);
1128                 if (rc == -ENOENT) {
1129                         inode->i_nlink = 0;
1130                         /* Unlinked special device node? Or just a race?
1131                          * Pretend we done everything. */
1132                         if (!S_ISREG(inode->i_mode) &&
1133                             !S_ISDIR(inode->i_mode))
1134                                 rc = inode_setattr(inode, &op_data->op_attr);
1135                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1136                         CERROR("md_setattr fails: rc = %d\n", rc);
1137                 }
1138                 RETURN(rc);
1139         }
1140
1141         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1142                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1143         if (rc) {
1144                 ptlrpc_req_finished(request);
1145                 RETURN(rc);
1146         }
1147
1148         /* We call inode_setattr to adjust timestamps.
1149          * If there is at least some data in file, we cleared ATTR_SIZE
1150          * above to avoid invoking vmtruncate, otherwise it is important
1151          * to call vmtruncate in inode_setattr to update inode->i_size
1152          * (bug 6196) */
1153         rc = inode_setattr(inode, &op_data->op_attr);
1154
1155         /* Extract epoch data if obtained. */
1156         memcpy(&op_data->op_handle, &md.body->handle, sizeof(op_data->op_handle));
1157         op_data->op_ioepoch = md.body->ioepoch;
1158
1159         ll_update_inode(inode, &md);
1160         ptlrpc_req_finished(request);
1161
1162         RETURN(rc);
1163 }
1164
1165 /* Close IO epoch and send Size-on-MDS attribute update. */
1166 static int ll_setattr_done_writing(struct inode *inode,
1167                                    struct md_op_data *op_data)
1168 {
1169         struct ll_inode_info *lli = ll_i2info(inode);
1170         int rc = 0;
1171         ENTRY;
1172
1173         LASSERT(op_data != NULL);
1174         if (!S_ISREG(inode->i_mode))
1175                 RETURN(0);
1176
1177         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1178                op_data->op_ioepoch, PFID(&lli->lli_fid));
1179
1180         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1181         /* XXX: pass och here for the recovery purpose. */
1182         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
1183         if (rc == -EAGAIN) {
1184                 /* MDS has instructed us to obtain Size-on-MDS attribute
1185                  * from OSTs and send setattr to back to MDS. */
1186                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1187                                          op_data->op_ioepoch);
1188         } else if (rc) {
1189                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1190                        inode->i_ino, rc);
1191         }
1192         RETURN(rc);
1193 }
1194
1195 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1196  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1197  * keep these values until such a time that objects are allocated for it.
1198  * We do the MDS operations first, as it is checking permissions for us.
1199  * We don't to the MDS RPC if there is nothing that we want to store there,
1200  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1201  * going to do an RPC anyways.
1202  *
1203  * If we are doing a truncate, we will send the mtime and ctime updates
1204  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1205  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1206  * at the same time.
1207  */
1208 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1209 {
1210         struct ll_inode_info *lli = ll_i2info(inode);
1211         struct lov_stripe_md *lsm = lli->lli_smd;
1212         struct ll_sb_info *sbi = ll_i2sbi(inode);
1213         struct md_op_data *op_data = NULL;
1214         int ia_valid = attr->ia_valid;
1215         int rc = 0, rc1 = 0;
1216         ENTRY;
1217
1218         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1219                attr->ia_valid);
1220         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1221
1222         if (ia_valid & ATTR_SIZE) {
1223                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1224                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1225                                attr->ia_size, ll_file_maxbytes(inode));
1226                         RETURN(-EFBIG);
1227                 }
1228
1229                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1230         }
1231
1232         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1233         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1234                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1235                         RETURN(-EPERM);
1236         }
1237
1238         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1239         if (attr->ia_valid & ATTR_CTIME) {
1240                 attr->ia_ctime = CURRENT_TIME;
1241                 attr->ia_valid |= ATTR_CTIME_SET;
1242         }
1243         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1244                 attr->ia_atime = CURRENT_TIME;
1245                 attr->ia_valid |= ATTR_ATIME_SET;
1246         }
1247         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1248                 attr->ia_mtime = CURRENT_TIME;
1249                 attr->ia_valid |= ATTR_MTIME_SET;
1250         }
1251         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1252                 /* To avoid stale mtime on mds, obtain it from ost and send
1253                    to mds. */
1254                 rc = ll_glimpse_size(inode, 0);
1255                 if (rc)
1256                         RETURN(rc);
1257
1258                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1259                 attr->ia_mtime = inode->i_mtime;
1260         }
1261
1262         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1263                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1264                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1265                        CURRENT_SECONDS);
1266
1267         /* NB: ATTR_SIZE will only be set after this point if the size
1268          * resides on the MDS, ie, this file has no objects. */
1269         if (lsm)
1270                 attr->ia_valid &= ~ATTR_SIZE;
1271
1272         /* We always do an MDS RPC, even if we're only changing the size;
1273          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1274
1275         OBD_ALLOC_PTR(op_data);
1276         if (op_data == NULL)
1277                 RETURN(-ENOMEM);
1278
1279         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1280
1281         /* Open epoch for truncate. */
1282         if (ia_valid & ATTR_SIZE)
1283                 op_data->op_flags = MF_EPOCH_OPEN;
1284
1285         rc = ll_md_setattr(inode, op_data);
1286         if (rc)
1287                 GOTO(out, rc);
1288
1289         if (op_data->op_ioepoch)
1290                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1291                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1292
1293         if (!lsm || !S_ISREG(inode->i_mode)) {
1294                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1295                         GOTO(out, rc = 0);
1296         }
1297
1298         /* We really need to get our PW lock before we change inode->i_size.
1299          * If we don't we can race with other i_size updaters on our node, like
1300          * ll_file_read.  We can also race with i_size propogation to other
1301          * nodes through dirtying and writeback of final cached pages.  This
1302          * last one is especially bad for racing o_append users on other
1303          * nodes. */
1304         if (ia_valid & ATTR_SIZE) {
1305                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1306                                                            OBD_OBJECT_EOF } };
1307                 struct lustre_handle lockh = { 0 };
1308                 int err, ast_flags = 0;
1309                 /* XXX when we fix the AST intents to pass the discard-range
1310                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1311                  * XXX here. */
1312                 if (attr->ia_size == 0)
1313                         ast_flags = LDLM_AST_DISCARD_DATA;
1314
1315                 UNLOCK_INODE_MUTEX(inode);
1316                 UP_WRITE_I_ALLOC_SEM(inode);
1317                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1318                                     ast_flags);
1319 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1320                 DOWN_WRITE_I_ALLOC_SEM(inode);
1321                 LOCK_INODE_MUTEX(inode);
1322 #else
1323                 LOCK_INODE_MUTEX(inode);
1324                 DOWN_WRITE_I_ALLOC_SEM(inode);
1325 #endif
1326                 if (rc != 0)
1327                         GOTO(out, rc);
1328
1329                 /* Only ll_inode_size_lock is taken at this level.
1330                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1331                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1332                  * ll_truncate dropped ll_inode_size_lock() */
1333                 ll_inode_size_lock(inode, 0);
1334                 rc = vmtruncate(inode, attr->ia_size);
1335                 if (rc != 0) {
1336                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1337                         ll_inode_size_unlock(inode, 0);
1338                 }
1339
1340                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1341                 if (err) {
1342                         CERROR("ll_extent_unlock failed: %d\n", err);
1343                         if (!rc)
1344                                 rc = err;
1345                 }
1346         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1347                 obd_flag flags;
1348                 struct obd_info oinfo = { { { 0 } } };
1349                 struct obdo *oa;
1350
1351                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1352                        inode->i_ino, LTIME_S(attr->ia_mtime));
1353
1354                 OBDO_ALLOC(oa);
1355                 if (oa) {
1356                         oa->o_id = lsm->lsm_object_id;
1357                         oa->o_gr = lsm->lsm_object_gr;
1358                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1359
1360                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1361                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1362                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1363                                 OBD_MD_FLGROUP;
1364
1365                         obdo_from_inode(oa, inode, flags);
1366
1367                         oinfo.oi_oa = oa;
1368                         oinfo.oi_md = lsm;
1369                         oinfo.oi_capa = ll_mdscapa_get(inode);
1370
1371                         /* XXX: this looks unnecessary now. */
1372                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1373                         capa_put(oinfo.oi_capa);
1374                         if (rc)
1375                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1376                         OBDO_FREE(oa);
1377                 } else {
1378                         rc = -ENOMEM;
1379                 }
1380         }
1381         EXIT;
1382 out:
1383         if (op_data) {
1384                 if (op_data->op_ioepoch) {
1385                         rc1 = ll_setattr_done_writing(inode, op_data);
1386                 }
1387                 ll_finish_md_op_data(op_data);
1388         }
1389         return rc ? rc : rc1;
1390 }
1391
1392 int ll_setattr(struct dentry *de, struct iattr *attr)
1393 {
1394         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1395             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1396                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1397
1398         return ll_setattr_raw(de->d_inode, attr);
1399 }
1400
1401 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1402                        __u64 max_age)
1403 {
1404         struct ll_sb_info *sbi = ll_s2sbi(sb);
1405         struct obd_statfs obd_osfs;
1406         int rc;
1407         ENTRY;
1408
1409         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1410         if (rc) {
1411                 CERROR("md_statfs fails: rc = %d\n", rc);
1412                 RETURN(rc);
1413         }
1414
1415         osfs->os_type = sb->s_magic;
1416
1417         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1418                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1419
1420         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1421                               &obd_osfs, max_age);
1422         if (rc) {
1423                 CERROR("obd_statfs fails: rc = %d\n", rc);
1424                 RETURN(rc);
1425         }
1426
1427         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1428                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1429                obd_osfs.os_files);
1430
1431         osfs->os_blocks = obd_osfs.os_blocks;
1432         osfs->os_bfree = obd_osfs.os_bfree;
1433         osfs->os_bavail = obd_osfs.os_bavail;
1434
1435         /* If we don't have as many objects free on the OST as inodes
1436          * on the MDS, we reduce the total number of inodes to
1437          * compensate, so that the "inodes in use" number is correct.
1438          */
1439         if (obd_osfs.os_ffree < osfs->os_ffree) {
1440                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1441                         obd_osfs.os_ffree;
1442                 osfs->os_ffree = obd_osfs.os_ffree;
1443         }
1444
1445         RETURN(rc);
1446 }
1447 #ifndef HAVE_STATFS_DENTRY_PARAM
1448 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1449 {
1450 #else
1451 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1452 {
1453         struct super_block *sb = de->d_sb;
1454 #endif
1455         struct obd_statfs osfs;
1456         int rc;
1457
1458         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1459         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1460
1461         /* For now we will always get up-to-date statfs values, but in the
1462          * future we may allow some amount of caching on the client (e.g.
1463          * from QOS or lprocfs updates). */
1464         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1465         if (rc)
1466                 return rc;
1467
1468         statfs_unpack(sfs, &osfs);
1469
1470         /* We need to downshift for all 32-bit kernels, because we can't
1471          * tell if the kernel is being called via sys_statfs64() or not.
1472          * Stop before overflowing f_bsize - in which case it is better
1473          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1474         if (sizeof(long) < 8) {
1475                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1476                         sfs->f_bsize <<= 1;
1477
1478                         osfs.os_blocks >>= 1;
1479                         osfs.os_bfree >>= 1;
1480                         osfs.os_bavail >>= 1;
1481                 }
1482         }
1483
1484         sfs->f_blocks = osfs.os_blocks;
1485         sfs->f_bfree = osfs.os_bfree;
1486         sfs->f_bavail = osfs.os_bavail;
1487
1488         return 0;
1489 }
1490
1491 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1492 {
1493         struct ll_inode_info *lli;
1494         struct lov_stripe_md *lsm;
1495
1496         lli = ll_i2info(inode);
1497         LASSERT(lli->lli_size_sem_owner != current);
1498         down(&lli->lli_size_sem);
1499         LASSERT(lli->lli_size_sem_owner == NULL);
1500         lli->lli_size_sem_owner = current;
1501         lsm = lli->lli_smd;
1502         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1503                  lsm, lock_lsm);
1504         if (lock_lsm)
1505                 lov_stripe_lock(lsm);
1506 }
1507
1508 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1509 {
1510         struct ll_inode_info *lli;
1511         struct lov_stripe_md *lsm;
1512
1513         lli = ll_i2info(inode);
1514         lsm = lli->lli_smd;
1515         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1516                  lsm, unlock_lsm);
1517         if (unlock_lsm)
1518                 lov_stripe_unlock(lsm);
1519         LASSERT(lli->lli_size_sem_owner == current);
1520         lli->lli_size_sem_owner = NULL;
1521         up(&lli->lli_size_sem);
1522 }
1523
1524 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1525 {
1526         struct ll_inode_info *lli = ll_i2info(inode);
1527
1528         dump_lsm(D_INODE, lsm);
1529         dump_lsm(D_INODE, lli->lli_smd);
1530         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1531                  "lsm must be joined lsm %p\n", lsm);
1532         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1533         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1534                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1535         lli->lli_smd = lsm;
1536         lli->lli_maxbytes = lsm->lsm_maxbytes;
1537         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1538                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1539 }
1540
1541 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1542 {
1543         struct ll_inode_info *lli = ll_i2info(inode);
1544         struct mdt_body *body = md->body;
1545         struct lov_stripe_md *lsm = md->lsm;
1546         struct ll_sb_info *sbi = ll_i2sbi(inode);
1547
1548         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1549         if (lsm != NULL) {
1550                 if (lli->lli_smd == NULL) {
1551                         if (lsm->lsm_magic != LOV_MAGIC &&
1552                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1553                                 dump_lsm(D_ERROR, lsm);
1554                                 LBUG();
1555                         }
1556                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1557                                lsm, inode->i_ino, inode->i_generation, inode);
1558                         /* ll_inode_size_lock() requires it is only called
1559                          * with lli_smd != NULL or lock_lsm == 0 or we can
1560                          * race between lock/unlock.  bug 9547 */
1561                         lli->lli_smd = lsm;
1562                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1563                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1564                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1565                 } else {
1566                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1567                              lli->lli_smd->lsm_stripe_count ==
1568                                         lsm->lsm_stripe_count) {
1569                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1570                                         CERROR("lsm mismatch for inode %ld\n",
1571                                                 inode->i_ino);
1572                                         CERROR("lli_smd:\n");
1573                                         dump_lsm(D_ERROR, lli->lli_smd);
1574                                         CERROR("lsm:\n");
1575                                         dump_lsm(D_ERROR, lsm);
1576                                         LBUG();
1577                                 }
1578                         } else
1579                                 ll_replace_lsm(inode, lsm);
1580                 }
1581                 if (lli->lli_smd != lsm)
1582                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1583         }
1584
1585         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1586                 if (body->valid & OBD_MD_FLRMTPERM)
1587                         ll_update_remote_perm(inode, md->remote_perm);
1588         }
1589 #ifdef CONFIG_FS_POSIX_ACL
1590         else if (body->valid & OBD_MD_FLACL) {
1591                 spin_lock(&lli->lli_lock);
1592                 if (lli->lli_posix_acl)
1593                         posix_acl_release(lli->lli_posix_acl);
1594                 lli->lli_posix_acl = md->posix_acl;
1595                 spin_unlock(&lli->lli_lock);
1596         }
1597 #endif
1598         if (body->valid & OBD_MD_FLATIME &&
1599             body->atime > LTIME_S(inode->i_atime))
1600                 LTIME_S(inode->i_atime) = body->atime;
1601
1602         /* mtime is always updated with ctime, but can be set in past.
1603            As write and utime(2) may happen within 1 second, and utime's
1604            mtime has a priority over write's one, so take mtime from mds
1605            for the same ctimes. */
1606         if (body->valid & OBD_MD_FLCTIME &&
1607             body->ctime >= LTIME_S(inode->i_ctime)) {
1608                 LTIME_S(inode->i_ctime) = body->ctime;
1609                 if (body->valid & OBD_MD_FLMTIME) {
1610                         CDEBUG(D_INODE, "setting ino %lu mtime "
1611                                "from %lu to "LPU64"\n", inode->i_ino,
1612                                LTIME_S(inode->i_mtime), body->mtime);
1613                         LTIME_S(inode->i_mtime) = body->mtime;
1614                 }
1615         }
1616         if (body->valid & OBD_MD_FLMODE)
1617                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1618         if (body->valid & OBD_MD_FLTYPE)
1619                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1620         if (S_ISREG(inode->i_mode)) {
1621                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1622         } else {
1623                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1624         }
1625 #ifdef HAVE_INODE_BLKSIZE
1626         inode->i_blksize = 1<<inode->i_blkbits;
1627 #endif
1628         if (body->valid & OBD_MD_FLUID)
1629                 inode->i_uid = body->uid;
1630         if (body->valid & OBD_MD_FLGID)
1631                 inode->i_gid = body->gid;
1632         if (body->valid & OBD_MD_FLFLAGS)
1633                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1634         if (body->valid & OBD_MD_FLNLINK)
1635                 inode->i_nlink = body->nlink;
1636         if (body->valid & OBD_MD_FLRDEV)
1637 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1638                 inode->i_rdev = body->rdev;
1639 #else
1640                 inode->i_rdev = old_decode_dev(body->rdev);
1641 #endif
1642         if (body->valid & OBD_MD_FLSIZE) {
1643                 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1644                         if (lli->lli_flags & (LLIF_DONE_WRITING |
1645                                               LLIF_EPOCH_PENDING |
1646                                               LLIF_SOM_DIRTY))
1647                           CWARN("ino %lu flags %lu still has size authority!"
1648                                 "do not trust the size got from MDS\n",
1649                                 inode->i_ino, lli->lli_flags);
1650                         else {
1651                                 inode->i_size = body->size;
1652                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1653                         }
1654                 } else {
1655                         inode->i_size = body->size;
1656                 }
1657
1658                 if (body->valid & OBD_MD_FLBLOCKS)
1659                         inode->i_blocks = body->blocks;
1660         }
1661
1662         if (body->valid & OBD_MD_FLID) {
1663                 /* FID shouldn't be changed! */
1664                 if (fid_is_sane(&lli->lli_fid)) {
1665                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1666                                  "Trying to change FID "DFID
1667                                  " to the "DFID", inode %lu/%u(%p)\n",
1668                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1669                                  inode->i_ino, inode->i_generation, inode);
1670                 } else
1671                         lli->lli_fid = body->fid1;
1672         }
1673
1674         LASSERT(fid_seq(&lli->lli_fid) != 0);
1675
1676         if (body->valid & OBD_MD_FLMDSCAPA) {
1677                 LASSERT(md->mds_capa);
1678                 ll_add_capa(inode, md->mds_capa);
1679         }
1680         if (body->valid & OBD_MD_FLOSSCAPA) {
1681                 LASSERT(md->oss_capa);
1682                 ll_add_capa(inode, md->oss_capa);
1683         }
1684 }
1685
1686 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1687 static struct backing_dev_info ll_backing_dev_info = {
1688         .ra_pages       = 0,    /* No readahead */
1689 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1690         .capabilities   = 0,    /* Does contribute to dirty memory */
1691 #else
1692         .memory_backed  = 0,    /* Does contribute to dirty memory */
1693 #endif
1694 };
1695 #endif
1696
1697 void ll_read_inode2(struct inode *inode, void *opaque)
1698 {
1699         struct lustre_md *md = opaque;
1700         struct ll_inode_info *lli = ll_i2info(inode);
1701         ENTRY;
1702
1703         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1704                inode->i_ino, inode->i_generation, inode);
1705
1706         ll_lli_init(lli);
1707
1708         LASSERT(!lli->lli_smd);
1709
1710         /* Core attributes from the MDS first.  This is a new inode, and
1711          * the VFS doesn't zero times in the core inode so we have to do
1712          * it ourselves.  They will be overwritten by either MDS or OST
1713          * attributes - we just need to make sure they aren't newer. */
1714         LTIME_S(inode->i_mtime) = 0;
1715         LTIME_S(inode->i_atime) = 0;
1716         LTIME_S(inode->i_ctime) = 0;
1717         inode->i_rdev = 0;
1718         ll_update_inode(inode, md);
1719
1720         /* OIDEBUG(inode); */
1721
1722         if (S_ISREG(inode->i_mode)) {
1723                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1724                 inode->i_op = &ll_file_inode_operations;
1725                 inode->i_fop = sbi->ll_fop;
1726                 inode->i_mapping->a_ops = &ll_aops;
1727                 EXIT;
1728         } else if (S_ISDIR(inode->i_mode)) {
1729                 inode->i_op = &ll_dir_inode_operations;
1730                 inode->i_fop = &ll_dir_operations;
1731                 inode->i_mapping->a_ops = &ll_dir_aops;
1732                 EXIT;
1733         } else if (S_ISLNK(inode->i_mode)) {
1734                 inode->i_op = &ll_fast_symlink_inode_operations;
1735                 EXIT;
1736         } else {
1737                 inode->i_op = &ll_special_inode_operations;
1738
1739 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1740                 init_special_inode(inode, inode->i_mode,
1741                                    kdev_t_to_nr(inode->i_rdev));
1742
1743                 /* initializing backing dev info. */
1744                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1745 #else
1746                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1747 #endif
1748                 EXIT;
1749         }
1750 }
1751
1752 void ll_delete_inode(struct inode *inode)
1753 {
1754         struct ll_sb_info *sbi = ll_i2sbi(inode);
1755         int rc;
1756         ENTRY;
1757
1758         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1759         if (rc) {
1760                 CERROR("fid_delete() failed, rc %d\n", rc);
1761         }
1762         clear_inode(inode);
1763
1764         EXIT;
1765 }
1766
1767 int ll_iocontrol(struct inode *inode, struct file *file,
1768                  unsigned int cmd, unsigned long arg)
1769 {
1770         struct ll_sb_info *sbi = ll_i2sbi(inode);
1771         struct ptlrpc_request *req = NULL;
1772         int rc, flags = 0;
1773         ENTRY;
1774
1775         switch(cmd) {
1776         case EXT3_IOC_GETFLAGS: {
1777                 struct mdt_body *body;
1778                 struct obd_capa *oc;
1779
1780                 oc = ll_mdscapa_get(inode);
1781                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1782                                 OBD_MD_FLFLAGS, 0, &req);
1783                 capa_put(oc);
1784                 if (rc) {
1785                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1786                         RETURN(-abs(rc));
1787                 }
1788
1789                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1790                                       sizeof(*body));
1791
1792                 /*Now the ext3 will be packed directly back to client,
1793                  *no need convert here*/
1794                 flags = body->flags;
1795
1796                 ptlrpc_req_finished (req);
1797
1798                 RETURN(put_user(flags, (int *)arg));
1799         }
1800         case EXT3_IOC_SETFLAGS: {
1801                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1802                 struct obd_info oinfo = { { { 0 } } };
1803                 struct md_op_data *op_data;
1804
1805                 if (get_user(flags, (int *)arg))
1806                         RETURN(-EFAULT);
1807
1808                 oinfo.oi_md = lsm;
1809                 OBDO_ALLOC(oinfo.oi_oa);
1810                 if (!oinfo.oi_oa)
1811                         RETURN(-ENOMEM);
1812
1813                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1814                                              LUSTRE_OPC_ANY, NULL);
1815                 if (IS_ERR(op_data))
1816                         RETURN(PTR_ERR(op_data));
1817
1818                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1819                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1820                 rc = md_setattr(sbi->ll_md_exp, op_data,
1821                                 NULL, 0, NULL, 0, &req);
1822                 ll_finish_md_op_data(op_data);
1823                 ptlrpc_req_finished(req);
1824                 if (rc || lsm == NULL) {
1825                         OBDO_FREE(oinfo.oi_oa);
1826                         RETURN(rc);
1827                 }
1828
1829                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1830                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1831                 oinfo.oi_oa->o_flags = flags;
1832                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1833                                        OBD_MD_FLGROUP;
1834                 oinfo.oi_capa = ll_mdscapa_get(inode);
1835
1836                 obdo_from_inode(oinfo.oi_oa, inode,
1837                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1838                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1839                 capa_put(oinfo.oi_capa);
1840                 OBDO_FREE(oinfo.oi_oa);
1841                 if (rc) {
1842                         if (rc != -EPERM && rc != -EACCES)
1843                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1844                         RETURN(rc);
1845                 }
1846
1847                 inode->i_flags = ll_ext_to_inode_flags(flags |
1848                                                        MDS_BFLAG_EXT_FLAGS);
1849                 RETURN(0);
1850         }
1851         default:
1852                 RETURN(-ENOSYS);
1853         }
1854
1855         RETURN(0);
1856 }
1857
1858 int ll_flush_ctx(struct inode *inode)
1859 {
1860         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1861
1862         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1863
1864         obd_set_info_async(sbi->ll_md_exp,
1865                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1866                            0, NULL, NULL);
1867         obd_set_info_async(sbi->ll_dt_exp,
1868                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1869                            0, NULL, NULL);
1870         return 0;
1871 }
1872
1873 /* umount -f client means force down, don't save state */
1874 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1875 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1876 {
1877         struct super_block *sb = vfsmnt->mnt_sb;
1878 #else
1879 void ll_umount_begin(struct super_block *sb)
1880 {
1881 #endif
1882         struct lustre_sb_info *lsi = s2lsi(sb);
1883         struct ll_sb_info *sbi = ll_s2sbi(sb);
1884         struct obd_device *obd;
1885         struct obd_ioctl_data ioc_data = { 0 };
1886         ENTRY;
1887
1888 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1889         if (!(flags & MNT_FORCE)) {
1890                 EXIT;
1891                 return;
1892         }
1893 #endif
1894
1895         /* Tell the MGC we got umount -f */
1896         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1897
1898         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1899                sb->s_count, atomic_read(&sb->s_active));
1900
1901         obd = class_exp2obd(sbi->ll_md_exp);
1902         if (obd == NULL) {
1903                 CERROR("Invalid MDC connection handle "LPX64"\n",
1904                        sbi->ll_md_exp->exp_handle.h_cookie);
1905                 EXIT;
1906                 return;
1907         }
1908         obd->obd_force = 1;
1909         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1910                       &ioc_data, NULL);
1911
1912         obd = class_exp2obd(sbi->ll_dt_exp);
1913         if (obd == NULL) {
1914                 CERROR("Invalid LOV connection handle "LPX64"\n",
1915                        sbi->ll_dt_exp->exp_handle.h_cookie);
1916                 EXIT;
1917                 return;
1918         }
1919
1920         obd->obd_force = 1;
1921         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1922                       &ioc_data, NULL);
1923
1924         /* Really, we'd like to wait until there are no requests outstanding,
1925          * and then continue.  For now, we just invalidate the requests,
1926          * schedule, and hope.
1927          */
1928         schedule();
1929
1930         EXIT;
1931 }
1932
1933 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1934 {
1935         struct ll_sb_info *sbi = ll_s2sbi(sb);
1936         int err;
1937         __u32 read_only;
1938
1939         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1940                 read_only = *flags & MS_RDONLY;
1941                 err = obd_set_info_async(sbi->ll_md_exp,
1942                                          sizeof(KEY_READ_ONLY) - 1,
1943                                          KEY_READ_ONLY, sizeof(read_only),
1944                                          &read_only, NULL);
1945                 if (err) {
1946                         CERROR("Failed to change the read-only flag during "
1947                                "remount: %d\n", err);
1948                         return err;
1949                 }
1950
1951                 if (read_only)
1952                         sb->s_flags |= MS_RDONLY;
1953                 else
1954                         sb->s_flags &= ~MS_RDONLY;
1955         }
1956         return 0;
1957 }
1958
1959 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
1960                   int offset, struct super_block *sb)
1961 {
1962         struct ll_sb_info *sbi = NULL;
1963         struct lustre_md md;
1964         int rc = 0;
1965         ENTRY;
1966
1967         LASSERT(*inode || sb);
1968         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1969         prune_deathrow(sbi, 1);
1970         memset(&md, 0, sizeof(struct lustre_md));
1971
1972         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
1973                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1974         if (rc)
1975                 RETURN(rc);
1976
1977         if (*inode) {
1978                 ll_update_inode(*inode, &md);
1979         } else {
1980                 LASSERT(sb != NULL);
1981
1982                 /*
1983                  * At this point server returns to client's same fid as client
1984                  * generated for creating. So using ->fid1 is okay here.
1985                  */
1986                 LASSERT(fid_is_sane(&md.body->fid1));
1987
1988                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
1989                 if (*inode == NULL || is_bad_inode(*inode)) {
1990                         if (md.lsm)
1991                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1992 #ifdef CONFIG_FS_POSIX_ACL
1993                         if (md.posix_acl) {
1994                                 posix_acl_release(md.posix_acl);
1995                                 md.posix_acl = NULL;
1996                         }
1997 #endif
1998                         rc = -ENOMEM;
1999                         CERROR("new_inode -fatal: rc %d\n", rc);
2000                         GOTO(out, rc);
2001                 }
2002         }
2003
2004         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2005                          ll_i2info(*inode)->lli_smd);
2006 out:
2007         md_free_lustre_md(sbi->ll_md_exp, &md);
2008         RETURN(rc);
2009 }
2010
2011 char *llap_origins[] = {
2012         [LLAP_ORIGIN_UNKNOWN] = "--",
2013         [LLAP_ORIGIN_READPAGE] = "rp",
2014         [LLAP_ORIGIN_READAHEAD] = "ra",
2015         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2016         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2017 };
2018
2019 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2020                                              struct list_head *list)
2021 {
2022         struct ll_async_page *llap;
2023         struct list_head *pos;
2024
2025         list_for_each(pos, list) {
2026                 if (pos == &sbi->ll_pglist)
2027                         return NULL;
2028                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2029                 if (llap->llap_page == NULL)
2030                         continue;
2031                 return llap;
2032         }
2033         LBUG();
2034         return NULL;
2035 }
2036
2037 int ll_obd_statfs(struct inode *inode, void *arg)
2038 {
2039         struct ll_sb_info *sbi = NULL;
2040         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2041         struct lov_obd *lov = NULL;
2042         struct obd_statfs stat_buf = {0};
2043         char *buf = NULL;
2044         struct obd_ioctl_data *data = NULL;
2045         __u32 type, index;
2046         int len = 0, rc;
2047
2048         if (!inode || !(sbi = ll_i2sbi(inode)))
2049                 GOTO(out_statfs, rc = -EINVAL);
2050
2051         rc = obd_ioctl_getdata(&buf, &len, arg);
2052         if (rc)
2053                 GOTO(out_statfs, rc);
2054
2055         data = (void*)buf;
2056         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2057             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2058                 GOTO(out_statfs, rc = -EINVAL);
2059
2060         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2061         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2062
2063         if (type == LL_STATFS_MDC) {
2064                 if (index > 0)
2065                         GOTO(out_statfs, rc = -ENODEV);
2066                 client_obd = class_exp2obd(sbi->ll_md_exp);
2067         } else if (type == LL_STATFS_LOV) {
2068                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2069                 lov = &lov_obd->u.lov;
2070
2071                 if ((index >= lov->desc.ld_tgt_count))
2072                         GOTO(out_statfs, rc = -ENODEV);
2073                 if (!lov->lov_tgts[index])
2074                         /* Try again with the next index */
2075                         GOTO(out_statfs, rc = -EAGAIN);
2076
2077                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2078                 if (!lov->lov_tgts[index]->ltd_active)
2079                         GOTO(out_uuid, rc = -ENODATA);
2080         }
2081
2082         if (!client_obd)
2083                 GOTO(out_statfs, rc = -EINVAL);
2084
2085         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2086         if (rc)
2087                 GOTO(out_statfs, rc);
2088
2089         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2090                 GOTO(out_statfs, rc = -EFAULT);
2091
2092 out_uuid:
2093         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2094                          data->ioc_plen2))
2095                 rc = -EFAULT;
2096
2097 out_statfs:
2098         if (buf)
2099                 obd_ioctl_freedata(buf, len);
2100         return rc;
2101 }
2102
2103 int ll_process_config(struct lustre_cfg *lcfg)
2104 {
2105         char *ptr;
2106         void *sb;
2107         struct lprocfs_static_vars lvars;
2108         unsigned long x;
2109         int rc = 0;
2110
2111         lprocfs_init_vars(llite, &lvars);
2112
2113         /* The instance name contains the sb: lustre-client-aacfe000 */
2114         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2115         if (!ptr || !*(++ptr))
2116                 return -EINVAL;
2117         if (sscanf(ptr, "%lx", &x) != 1)
2118                 return -EINVAL;
2119         sb = (void *)x;
2120         /* This better be a real Lustre superblock! */
2121         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2122
2123         /* Note we have not called client_common_fill_super yet, so
2124            proc fns must be able to handle that! */
2125         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2126                                       lcfg, sb);
2127         return(rc);
2128 }
2129
2130 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2131 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2132                                        struct inode *i1, struct inode *i2,
2133                                        const char *name, int namelen,
2134                                        int mode, __u32 opc, void *data)
2135 {
2136         LASSERT(i1 != NULL);
2137
2138         if (namelen > ll_i2sbi(i1)->ll_namelen)
2139                 return ERR_PTR(-ENAMETOOLONG);
2140
2141         if (op_data == NULL)
2142                 OBD_ALLOC_PTR(op_data);
2143
2144         if (op_data == NULL)
2145                 return ERR_PTR(-ENOMEM);
2146
2147         ll_i2gids(op_data->op_suppgids, i1, i2);
2148         op_data->op_fid1 = *ll_inode2fid(i1);
2149         op_data->op_capa1 = ll_mdscapa_get(i1);
2150
2151         if (i2) {
2152                 op_data->op_fid2 = *ll_inode2fid(i2);
2153                 op_data->op_capa2 = ll_mdscapa_get(i2);
2154         } else {
2155                 fid_zero(&op_data->op_fid2);
2156         }
2157
2158         op_data->op_name = name;
2159         op_data->op_namelen = namelen;
2160         op_data->op_mode = mode;
2161         op_data->op_mod_time = CURRENT_SECONDS;
2162         op_data->op_fsuid = current->fsuid;
2163         op_data->op_fsgid = current->fsgid;
2164         op_data->op_cap = current->cap_effective;
2165         op_data->op_bias = MDS_CHECK_SPLIT;
2166         op_data->op_opc = opc;
2167         op_data->op_mds = 0;
2168         op_data->op_data = data;
2169
2170         return op_data;
2171 }
2172
2173 void ll_finish_md_op_data(struct md_op_data *op_data)
2174 {
2175         capa_put(op_data->op_capa1);
2176         capa_put(op_data->op_capa2);
2177         OBD_FREE_PTR(op_data);
2178 }
2179
2180 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2181 {
2182         struct ll_sb_info *sbi = ll_i2sbi(inode);
2183         struct ptlrpc_request *req = NULL;
2184         struct mdt_body *body;
2185         char *cmd, *buf;
2186         struct obd_capa *oc;
2187         int rc, buflen;
2188         ENTRY;
2189
2190         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2191                 RETURN(-EBADE);
2192
2193         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2194
2195         OBD_ALLOC(cmd, ioc->cmd_len);
2196         if (!cmd)
2197                 RETURN(-ENOMEM);
2198         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2199                 GOTO(out, rc = -EFAULT);
2200
2201         oc = ll_mdscapa_get(inode);
2202         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2203                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2204                          ioc->cmd_len, ioc->res_len, 0, &req);
2205         capa_put(oc);
2206         if (rc < 0) {
2207                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2208                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2209                 GOTO(out, rc);
2210         }
2211
2212         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2213         LASSERT(body);
2214
2215         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2216         LASSERT(buflen <= ioc->res_len);
2217         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2218         LASSERT(buf);
2219         if (copy_to_user(ioc->res, buf, buflen))
2220                 GOTO(out, rc = -EFAULT);
2221         EXIT;
2222 out:
2223         if (req)
2224                 ptlrpc_req_finished(req);
2225         OBD_FREE(cmd, ioc->cmd_len);
2226         return rc;
2227 }
2228
2229 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2230 {
2231         struct ll_sb_info *sbi = ll_i2sbi(inode);
2232         struct ptlrpc_request *req = NULL;
2233         char *cmd, *buf;
2234         struct obd_capa *oc;
2235         int buflen, rc;
2236         ENTRY;
2237
2238         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2239                 RETURN(-EBADE);
2240
2241         if (!(sbi->ll_flags & LL_SBI_ACL))
2242                 RETURN(-EOPNOTSUPP);
2243
2244         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2245
2246         OBD_ALLOC(cmd, ioc->cmd_len);
2247         if (!cmd)
2248                 RETURN(-ENOMEM);
2249         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2250                 GOTO(out, rc = -EFAULT);
2251
2252         oc = ll_mdscapa_get(inode);
2253         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2254                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2255                          ioc->cmd_len, ioc->res_len, 0, &req);
2256         capa_put(oc);
2257         if (rc) {
2258                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2259                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2260                 GOTO(out, rc);
2261         }
2262
2263         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2264         LASSERT(buflen <= ioc->res_len);
2265         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2266         LASSERT(buf);
2267         if (copy_to_user(ioc->res, buf, buflen))
2268                 GOTO(out, rc = -EFAULT);
2269         EXIT;
2270 out:
2271         if (req)
2272                 ptlrpc_req_finished(req);
2273         OBD_FREE(cmd, ioc->cmd_len);
2274         return rc;
2275 }