lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002-2005 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <lustre_lite.h>
  32 #include <lustre_ha.h>
  33 #include <lustre_dlm.h>
  34 #include <lprocfs_status.h>
  35 #include <lustre_disk.h>
  36 #include <lustre_param.h>
  37 #include <lustre_log.h>
  38 #include "llite_internal.h"
  39
  40 cfs_mem_cache_t *ll_file_data_slab;
  41
  42 LIST_HEAD(ll_super_blocks);
  43 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52
  53 static struct ll_sb_info *ll_init_sbi(void)
  54 {
  55         struct ll_sb_info *sbi = NULL;
  56         class_uuid_t uuid;
  57         int i;
  58         ENTRY;
  59
  60         OBD_ALLOC(sbi, sizeof(*sbi));
  61         if (!sbi)
  62                 RETURN(NULL);
  63
  64         spin_lock_init(&sbi->ll_lock);
  65         spin_lock_init(&sbi->ll_lco.lco_lock);
  66         spin_lock_init(&sbi->ll_pp_extent_lock);
  67         spin_lock_init(&sbi->ll_process_lock);
  68         sbi->ll_rw_stats_on = 0;
  69         INIT_LIST_HEAD(&sbi->ll_pglist);
  70         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
  71                 sbi->ll_async_page_max = num_physpages / 2;
  72         else
  73                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  74         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  75                                            SBI_DEFAULT_READAHEAD_MAX);
  76         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  77                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  78
  79         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  80         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
  81
  82         ll_generate_random_uuid(uuid);
  83         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  84         CDEBUG(D_HA, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  85
  86         spin_lock(&ll_sb_lock);
  87         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  88         spin_unlock(&ll_sb_lock);
  89
  90 #ifdef ENABLE_LLITE_CHECKSUM
  91         sbi->ll_flags |= LL_SBI_CHECKSUM;
  92 #endif
  93
  94 #ifdef HAVE_LRU_RESIZE_SUPPORT
  95         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
  96 #endif
  97
  98 #ifdef HAVE_EXPORT___IGET
  99         INIT_LIST_HEAD(&sbi->ll_deathrow);
 100         spin_lock_init(&sbi->ll_deathrow_lock);
 101 #endif
 102         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 103                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 104                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 105         }
 106
 107         RETURN(sbi);
 108 }
 109
 110 void ll_free_sbi(struct super_block *sb)
 111 {
 112         struct ll_sb_info *sbi = ll_s2sbi(sb);
 113         ENTRY;
 114
 115         if (sbi != NULL) {
 116                 spin_lock(&ll_sb_lock);
 117                 list_del(&sbi->ll_list);
 118                 spin_unlock(&ll_sb_lock);
 119                 OBD_FREE(sbi, sizeof(*sbi));
 120         }
 121         EXIT;
 122 }
 123
 124 static struct dentry_operations ll_d_root_ops = {
 125 #ifdef LUSTRE_KERNEL_VERSION
 126         .d_compare = ll_dcompare,
 127 #endif
 128 };
 129
 130 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 131  * us to make MDS RPCs with large enough reply buffers to hold the
 132  * maximum-sized (= maximum striped) EA and cookie without having to
 133  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 134 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 135 {
 136         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 137         __u32 valsize = sizeof(struct lov_desc);
 138         int rc, easize, def_easize, cookiesize;
 139         struct lov_desc desc;
 140         __u32 stripes;
 141         ENTRY;
 142
 143         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 144                           &valsize, &desc);
 145         if (rc)
 146                 RETURN(rc);
 147
 148         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 149         lsm.lsm_stripe_count = stripes;
 150         easize = obd_size_diskmd(dt_exp, &lsm);
 151
 152         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 153         def_easize = obd_size_diskmd(dt_exp, &lsm);
 154
 155         cookiesize = stripes * sizeof(struct llog_cookie);
 156
 157         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 158                easize, cookiesize);
 159
 160         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 161         RETURN(rc);
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     uid_t nllu, gid_t nllg)
 166 {
 167         struct inode *root = 0;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct lu_fid rootfid;
 171         struct obd_capa *oc = NULL;
 172         struct obd_statfs osfs;
 173         struct ptlrpc_request *request = NULL;
 174         struct lustre_handle dt_conn = {0, };
 175         struct lustre_handle md_conn = {0, };
 176         struct obd_connect_data *data = NULL;
 177         struct lustre_md lmd;
 178         obd_valid valid;
 179         int size, err, checksum;
 180         ENTRY;
 181
 182         obd = class_name2obd(md);
 183         if (!obd) {
 184                 CERROR("MD %s: not setup or attached\n", md);
 185                 RETURN(-EINVAL);
 186         }
 187
 188         OBD_ALLOC_PTR(data);
 189         if (data == NULL)
 190                 RETURN(-ENOMEM);
 191
 192         if (proc_lustre_fs_root) {
 193                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 194                                                   dt, md);
 195                 if (err < 0)
 196                         CERROR("could not register mount in /proc/lustre");
 197         }
 198
 199         /* indicate the features supported by this client */
 200         data->ocd_connect_flags = OBD_CONNECT_IBITS | OBD_CONNECT_NODEVOH |
 201                                   OBD_CONNECT_JOIN |
 202                                   OBD_CONNECT_ATTRFID | OBD_CONNECT_VERSION |
 203                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 204                                   OBD_CONNECT_CANCELSET;
 205 #ifdef HAVE_LRU_RESIZE_SUPPORT
 206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208 #endif
 209 #ifdef CONFIG_FS_POSIX_ACL
 210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 211 #endif
 212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 213         data->ocd_version = LUSTRE_VERSION_CODE;
 214
 215         if (sb->s_flags & MS_RDONLY)
 216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 219
 220 #ifdef HAVE_MS_FLOCK_LOCK
 221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 222         sb->s_flags |= MS_FLOCK_LOCK;
 223 #endif
 224
 225         if (sbi->ll_flags & LL_SBI_FLOCK)
 226                 sbi->ll_fop = &ll_file_operations_flock;
 227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 228                 sbi->ll_fop = &ll_file_operations;
 229         else
 230                 sbi->ll_fop = &ll_file_operations_noflock;
 231
 232         /* real client */
 233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 235                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 236                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 237         } else {
 238                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 239                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 240         }
 241         data->ocd_nllu = nllu;
 242         data->ocd_nllg = nllg;
 243
 244         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data);
 245         if (err == -EBUSY) {
 246                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 247                                    "recovery, of which this client is not a "
 248                                    "part. Please wait for recovery to complete,"
 249                                    " abort, or time out.\n", md);
 250                 GOTO(out, err);
 251         } else if (err) {
 252                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 253                 GOTO(out, err);
 254         }
 255         sbi->ll_md_exp = class_conn2export(&md_conn);
 256
 257         err = obd_fid_init(sbi->ll_md_exp);
 258         if (err) {
 259                 CERROR("Can't init metadata layer FID infrastructure, "
 260                        "rc %d\n", err);
 261                 GOTO(out_md, err);
 262         }
 263
 264         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ);
 265         if (err)
 266                 GOTO(out_md_fid, err);
 267
 268         size = sizeof(*data);
 269         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 270                            KEY_CONN_DATA,  &size, data);
 271         if (err) {
 272                 CERROR("Get connect data failed: %d \n", err);
 273                 GOTO(out_md, err);
 274         }
 275
 276         LASSERT(osfs.os_bsize);
 277         sb->s_blocksize = osfs.os_bsize;
 278         sb->s_blocksize_bits = log2(osfs.os_bsize);
 279         sb->s_magic = LL_SUPER_MAGIC;
 280
 281         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 282          *         retval = in_file->f_op->sendfile(...);
 283          *         if (*ppos > max)
 284          *                 retval = -EOVERFLOW;
 285          *
 286          * it will check if *ppos is greater than max. However, max equals to
 287          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 288          * has been defined as a signed long long ineger in linux kernel. */
 289 #if BITS_PER_LONG == 64
 290         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 291 #else
 292         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 293 #endif
 294         sbi->ll_namelen = osfs.os_namelen;
 295         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 296
 297         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 298             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 299                 LCONSOLE_INFO("Disabling user_xattr feature because "
 300                               "it is not supported on the server\n");
 301                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 302         }
 303
 304         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 305 #ifdef MS_POSIXACL
 306                 sb->s_flags |= MS_POSIXACL;
 307 #endif
 308                 sbi->ll_flags |= LL_SBI_ACL;
 309         } else {
 310                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 311 #ifdef MS_POSIXACL
 312                 sb->s_flags &= ~MS_POSIXACL;
 313 #endif
 314                 sbi->ll_flags &= ~LL_SBI_ACL;
 315         }
 316
 317         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 318                 sbi->ll_flags |= LL_SBI_JOIN;
 319
 320         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 321                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 322                         /* sometimes local client claims to be remote, but mdt
 323                          * will disagree when client gss not applied. */
 324                         LCONSOLE_INFO("client claims to be remote, but server "
 325                                       "rejected, forced to be local.\n");
 326                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 327                 }
 328         } else {
 329                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 330                         /* with gss applied, remote client can not claim to be
 331                          * local, so mdt maybe force client to be remote. */
 332                         LCONSOLE_INFO("client claims to be local, but server "
 333                                       "rejected, forced to be remote.\n");
 334                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 335                 }
 336         }
 337
 338         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 339                 LCONSOLE_INFO("client enabled MDS capability!\n");
 340                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 344                 LCONSOLE_INFO("client enabled OSS capability!\n");
 345                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 346         }
 347
 348         sbi->ll_sdev_orig = sb->s_dev;
 349 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 350         /* We set sb->s_dev equal on all lustre clients in order to support
 351          * NFS export clustering.  NFSD requires that the FSID be the same
 352          * on all clients. */
 353         /* s_dev is also used in lt_compare() to compare two fs, but that is
 354          * only a node-local comparison. */
 355
 356         /* XXX: this will not work with LMV */
 357         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 358                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 359 #endif
 360
 361         obd = class_name2obd(dt);
 362         if (!obd) {
 363                 CERROR("DT %s: not setup or attached\n", dt);
 364                 GOTO(out_md_fid, err = -ENODEV);
 365         }
 366
 367         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 368                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 369                                   OBD_CONNECT_CANCELSET;
 370         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 371                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 372
 373 #ifdef HAVE_LRU_RESIZE_SUPPORT
 374         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 375 #endif
 376         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 377                "ocd_grant: %d\n", data->ocd_connect_flags,
 378                data->ocd_version, data->ocd_grant);
 379
 380         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 381         obd->obd_upcall.onu_upcall = ll_ocd_update;
 382         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 383
 384         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data);
 385         if (err == -EBUSY) {
 386                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 387                                    "recovery, of which this client is not a "
 388                                    "part.  Please wait for recovery to "
 389                                    "complete, abort, or time out.\n", dt);
 390                 GOTO(out_md_fid, err);
 391         } else if (err) {
 392                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 393                 GOTO(out_md_fid, err);
 394         }
 395
 396         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 397
 398         err = obd_fid_init(sbi->ll_dt_exp);
 399         if (err) {
 400                 CERROR("Can't init data layer FID infrastructure, "
 401                        "rc %d\n", err);
 402                 GOTO(out_dt, err);
 403         }
 404
 405         spin_lock(&sbi->ll_lco.lco_lock);
 406         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 407         spin_unlock(&sbi->ll_lco.lco_lock);
 408
 409         ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);
 410
 411         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 412                                   0, NULL, NULL, NULL);
 413         if (err < 0) {
 414                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 415                                    "filesystem. There must be at least one "
 416                                    "active OST for a client to start.\n");
 417                 GOTO(out_dt_fid, err);
 418         }
 419
 420         if (!ll_async_page_slab) {
 421                 ll_async_page_slab_size =
 422                         size_round(sizeof(struct ll_async_page)) + err;
 423                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 424                                                         ll_async_page_slab_size,
 425                                                            0, 0);
 426                 if (!ll_async_page_slab)
 427                         GOTO(out_dt_fid, err = -ENOMEM);
 428         }
 429
 430         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 431         if (err) {
 432                 CERROR("cannot mds_connect: rc = %d\n", err);
 433                 GOTO(out_dt_fid, err);
 434         }
 435         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 436         sbi->ll_root_fid = rootfid;
 437
 438         sb->s_op = &lustre_super_operations;
 439 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 440         sb->s_export_op = &lustre_export_operations;
 441 #endif
 442
 443         /* make root inode
 444          * XXX: move this to after cbd setup? */
 445         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 446         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 447                 valid |= OBD_MD_FLRMTPERM;
 448         else if (sbi->ll_flags & LL_SBI_ACL)
 449                 valid |= OBD_MD_FLACL;
 450
 451         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 452         if (oc)
 453                 free_capa(oc);
 454         if (err) {
 455                 CERROR("md_getattr failed for root: rc = %d\n", err);
 456                 GOTO(out_dt_fid, err);
 457         }
 458         memset(&lmd, 0, sizeof(lmd));
 459         err = md_get_lustre_md(sbi->ll_md_exp, request,
 460                                REPLY_REC_OFF, sbi->ll_dt_exp, sbi->ll_md_exp,
 461                                &lmd);
 462         if (err) {
 463                 CERROR("failed to understand root inode md: rc = %d\n", err);
 464                 ptlrpc_req_finished (request);
 465                 GOTO(out_dt_fid, err);
 466         }
 467
 468         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 469         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 470         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 471         ptlrpc_req_finished(request);
 472
 473         if (root == NULL || is_bad_inode(root)) {
 474                 if (lmd.lsm)
 475                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 476 #ifdef CONFIG_FS_POSIX_ACL
 477                 if (lmd.posix_acl) {
 478                         posix_acl_release(lmd.posix_acl);
 479                         lmd.posix_acl = NULL;
 480                 }
 481 #endif
 482                 CERROR("lustre_lite: bad iget4 for root\n");
 483                 GOTO(out_root, err = -EBADF);
 484         }
 485
 486         err = ll_close_thread_start(&sbi->ll_lcq);
 487         if (err) {
 488                 CERROR("cannot start close thread: rc %d\n", err);
 489                 GOTO(out_root, err);
 490         }
 491
 492         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 493         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 494                                  sizeof(checksum), &checksum, NULL);
 495
 496         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 497            backing dev info assigned to inode mapping is used for
 498            determining maximal readahead. */
 499 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 500     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 501         /* bug 2805 - set VM readahead to zero */
 502         vm_max_readahead = vm_min_readahead = 0;
 503 #endif
 504
 505         sb->s_root = d_alloc_root(root);
 506         if (data != NULL)
 507                 OBD_FREE(data, sizeof(*data));
 508         sb->s_root->d_op = &ll_d_root_ops;
 509         RETURN(err);
 510 out_root:
 511         if (root)
 512                 iput(root);
 513 out_dt_fid:
 514         obd_fid_fini(sbi->ll_dt_exp);
 515 out_dt:
 516         obd_disconnect(sbi->ll_dt_exp);
 517         sbi->ll_dt_exp = NULL;
 518 out_md_fid:
 519         obd_fid_fini(sbi->ll_md_exp);
 520 out_md:
 521         obd_disconnect(sbi->ll_md_exp);
 522         sbi->ll_md_exp = NULL;
 523 out:
 524         if (data != NULL)
 525                 OBD_FREE_PTR(data);
 526         lprocfs_unregister_mountpoint(sbi);
 527         return err;
 528 }
 529
 530 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 531 {
 532         int size, rc;
 533
 534         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 535         size = sizeof(int);
 536         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 537                           &size, lmmsize);
 538         if (rc)
 539                 CERROR("Get max mdsize error rc %d \n", rc);
 540
 541         RETURN(rc);
 542 }
 543
 544 void ll_dump_inode(struct inode *inode)
 545 {
 546         struct list_head *tmp;
 547         int dentry_count = 0;
 548
 549         LASSERT(inode != NULL);
 550
 551         list_for_each(tmp, &inode->i_dentry)
 552                 dentry_count++;
 553
 554         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 555                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 556                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 557 }
 558
 559 void lustre_dump_dentry(struct dentry *dentry, int recur)
 560 {
 561         struct list_head *tmp;
 562         int subdirs = 0;
 563
 564         LASSERT(dentry != NULL);
 565
 566         list_for_each(tmp, &dentry->d_subdirs)
 567                 subdirs++;
 568
 569         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 570                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 571                dentry->d_name.len, dentry->d_name.name,
 572                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 573                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 574                dentry->d_flags, dentry->d_fsdata, subdirs);
 575         if (dentry->d_inode != NULL)
 576                 ll_dump_inode(dentry->d_inode);
 577
 578         if (recur == 0)
 579                 return;
 580
 581         list_for_each(tmp, &dentry->d_subdirs) {
 582                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 583                 lustre_dump_dentry(d, recur - 1);
 584         }
 585 }
 586
 587 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 588 void lustre_throw_orphan_dentries(struct super_block *sb)
 589 {
 590         struct dentry *dentry, *next;
 591         struct ll_sb_info *sbi = ll_s2sbi(sb);
 592
 593         /* Do this to get rid of orphaned dentries. That is not really trw. */
 594         list_for_each_entry_safe(dentry, next, &sbi->ll_orphan_dentry_list,
 595                                  d_hash) {
 596                 CWARN("found orphan dentry %.*s (%p->%p) at unmount, dumping "
 597                       "before and after shrink_dcache_parent\n",
 598                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 599                 lustre_dump_dentry(dentry, 1);
 600                 shrink_dcache_parent(dentry);
 601                 lustre_dump_dentry(dentry, 1);
 602         }
 603 }
 604 #else
 605 #define lustre_throw_orphan_dentries(sb)
 606 #endif
 607
 608 #ifdef HAVE_EXPORT___IGET
 609 static void prune_dir_dentries(struct inode *inode)
 610 {
 611         struct dentry *dentry, *prev = NULL;
 612
 613         /* due to lustre specific logic, a directory
 614          * can have few dentries - a bug from VFS POV */
 615 restart:
 616         spin_lock(&dcache_lock);
 617         if (!list_empty(&inode->i_dentry)) {
 618                 dentry = list_entry(inode->i_dentry.prev,
 619                                     struct dentry, d_alias);
 620                 /* in order to prevent infinite loops we
 621                  * break if previous dentry is busy */
 622                 if (dentry != prev) {
 623                         prev = dentry;
 624                         dget_locked(dentry);
 625                         spin_unlock(&dcache_lock);
 626
 627                         /* try to kill all child dentries */
 628                         lock_dentry(dentry);
 629                         shrink_dcache_parent(dentry);
 630                         unlock_dentry(dentry);
 631                         dput(dentry);
 632
 633                         /* now try to get rid of current dentry */
 634                         d_prune_aliases(inode);
 635                         goto restart;
 636                 }
 637         }
 638         spin_unlock(&dcache_lock);
 639 }
 640
 641 static void prune_deathrow_one(struct ll_inode_info *lli)
 642 {
 643         struct inode *inode = ll_info2i(lli);
 644
 645         /* first, try to drop any dentries - they hold a ref on the inode */
 646         if (S_ISDIR(inode->i_mode))
 647                 prune_dir_dentries(inode);
 648         else
 649                 d_prune_aliases(inode);
 650
 651
 652         /* if somebody still uses it, leave it */
 653         LASSERT(atomic_read(&inode->i_count) > 0);
 654         if (atomic_read(&inode->i_count) > 1)
 655                 goto out;
 656
 657         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 658                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 659
 660         /* seems nobody uses it anymore */
 661         inode->i_nlink = 0;
 662
 663 out:
 664         iput(inode);
 665         return;
 666 }
 667
 668 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 669 {
 670         struct ll_inode_info *lli;
 671         int empty;
 672
 673         do {
 674                 if (need_resched() && try)
 675                         break;
 676
 677                 if (try) {
 678                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 679                                 break;
 680                 } else {
 681                         spin_lock(&sbi->ll_deathrow_lock);
 682                 }
 683
 684                 empty = 1;
 685                 lli = NULL;
 686                 if (!list_empty(&sbi->ll_deathrow)) {
 687                         lli = list_entry(sbi->ll_deathrow.next,
 688                                          struct ll_inode_info,
 689                                          lli_dead_list);
 690                         list_del_init(&lli->lli_dead_list);
 691                         if (!list_empty(&sbi->ll_deathrow))
 692                                 empty = 0;
 693                 }
 694                 spin_unlock(&sbi->ll_deathrow_lock);
 695
 696                 if (lli)
 697                         prune_deathrow_one(lli);
 698
 699         } while (empty == 0);
 700 }
 701 #else /* !HAVE_EXPORT___IGET */
 702 #define prune_deathrow(sbi, try) do {} while (0)
 703 #endif /* HAVE_EXPORT___IGET */
 704
 705 void client_common_put_super(struct super_block *sb)
 706 {
 707         struct ll_sb_info *sbi = ll_s2sbi(sb);
 708         ENTRY;
 709
 710         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 711
 712         ll_close_thread_shutdown(sbi->ll_lcq);
 713
 714         /* destroy inodes in deathrow */
 715         prune_deathrow(sbi, 0);
 716
 717         list_del(&sbi->ll_conn_chain);
 718
 719         obd_fid_fini(sbi->ll_dt_exp);
 720         obd_disconnect(sbi->ll_dt_exp);
 721         sbi->ll_dt_exp = NULL;
 722
 723         lprocfs_unregister_mountpoint(sbi);
 724
 725         obd_fid_fini(sbi->ll_md_exp);
 726         obd_disconnect(sbi->ll_md_exp);
 727         sbi->ll_md_exp = NULL;
 728
 729         lustre_throw_orphan_dentries(sb);
 730
 731         EXIT;
 732 }
 733
 734 void ll_kill_super(struct super_block *sb)
 735 {
 736         struct ll_sb_info *sbi;
 737
 738         ENTRY;
 739
 740         /* not init sb ?*/
 741         if (!(sb->s_flags & MS_ACTIVE))
 742                 return;
 743
 744         sbi = ll_s2sbi(sb);
 745         /* we need restore s_dev from changed for clustred NFS before put_super
 746          * because new kernels have cached s_dev and change sb->s_dev in
 747          * put_super not affected real removing devices */
 748         if (sbi)
 749                 sb->s_dev = sbi->ll_sdev_orig;
 750         EXIT;
 751 }
 752
 753 char *ll_read_opt(const char *opt, char *data)
 754 {
 755         char *value;
 756         char *retval;
 757         ENTRY;
 758
 759         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 760         if (strncmp(opt, data, strlen(opt)))
 761                 RETURN(NULL);
 762         if ((value = strchr(data, '=')) == NULL)
 763                 RETURN(NULL);
 764
 765         value++;
 766         OBD_ALLOC(retval, strlen(value) + 1);
 767         if (!retval) {
 768                 CERROR("out of memory!\n");
 769                 RETURN(NULL);
 770         }
 771
 772         memcpy(retval, value, strlen(value)+1);
 773         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 774         RETURN(retval);
 775 }
 776
 777 static inline int ll_set_opt(const char *opt, char *data, int fl)
 778 {
 779         if (strncmp(opt, data, strlen(opt)) != 0)
 780                 return(0);
 781         else
 782                 return(fl);
 783 }
 784
 785 /* non-client-specific mount options are parsed in lmd_parse */
 786 static int ll_options(char *options, int *flags)
 787 {
 788         int tmp;
 789         char *s1 = options, *s2;
 790         ENTRY;
 791
 792         if (!options)
 793                 RETURN(0);
 794
 795         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 796
 797         while (*s1) {
 798                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 799                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 805                 if (tmp) {
 806                         *flags |= tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 810                 if (tmp) {
 811                         *flags |= tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 815                 if (tmp) {
 816                         *flags &= ~tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 825                 if (tmp) {
 826                         *flags &= ~tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 830                 if (tmp) {
 831                         /* Ignore deprecated mount option.  The client will
 832                          * always try to mount with ACL support, whether this
 833                          * is used depends on whether server supports it. */
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 837                 if (tmp) {
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845
 846                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 862                 if (tmp) {
 863                         *flags &= ~tmp;
 864                         goto next;
 865                 }
 866
 867                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 868                                    s1);
 869                 RETURN(-EINVAL);
 870
 871 next:
 872                 /* Find next opt */
 873                 s2 = strchr(s1, ',');
 874                 if (s2 == NULL)
 875                         break;
 876                 s1 = s2 + 1;
 877         }
 878         RETURN(0);
 879 }
 880
 881 void ll_lli_init(struct ll_inode_info *lli)
 882 {
 883         lli->lli_inode_magic = LLI_INODE_MAGIC;
 884         sema_init(&lli->lli_size_sem, 1);
 885         sema_init(&lli->lli_write_sem, 1);
 886         lli->lli_flags = 0;
 887         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 888         spin_lock_init(&lli->lli_lock);
 889         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 890         INIT_LIST_HEAD(&lli->lli_close_list);
 891         lli->lli_inode_magic = LLI_INODE_MAGIC;
 892         sema_init(&lli->lli_och_sem, 1);
 893         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 894         lli->lli_mds_exec_och = NULL;
 895         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 896         lli->lli_open_fd_exec_count = 0;
 897         INIT_LIST_HEAD(&lli->lli_dead_list);
 898         lli->lli_remote_perms = NULL;
 899         lli->lli_rmtperm_utime = 0;
 900         sema_init(&lli->lli_rmtperm_sem, 1);
 901         INIT_LIST_HEAD(&lli->lli_oss_capas);
 902 }
 903
 904 int ll_fill_super(struct super_block *sb)
 905 {
 906         struct lustre_profile *lprof;
 907         struct lustre_sb_info *lsi = s2lsi(sb);
 908         struct ll_sb_info *sbi;
 909         char  *dt = NULL, *md = NULL;
 910         char  *profilenm = get_profile_name(sb);
 911         struct config_llog_instance cfg = {0, };
 912         char   ll_instance[sizeof(sb) * 2 + 1];
 913         int    err;
 914         ENTRY;
 915
 916         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 917
 918         cfs_module_get();
 919
 920         sb->s_type->fs_flags |= FS_ODD_RENAME;
 921         /* client additional sb info */
 922         lsi->lsi_llsbi = sbi = ll_init_sbi();
 923         if (!sbi) {
 924                 cfs_module_put();
 925                 RETURN(-ENOMEM);
 926         }
 927
 928         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 929         if (err)
 930                 GOTO(out_free, err);
 931
 932         /* Generate a string unique to this super, in case some joker tries
 933            to mount the same fs at two mount points.
 934            Use the address of the super itself.*/
 935         sprintf(ll_instance, "%p", sb);
 936         cfg.cfg_instance = ll_instance;
 937         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 938
 939         /* set up client obds */
 940         err = lustre_process_log(sb, profilenm, &cfg);
 941         if (err < 0) {
 942                 CERROR("Unable to process log: %d\n", err);
 943                 GOTO(out_free, err);
 944         }
 945
 946         lprof = class_get_profile(profilenm);
 947         if (lprof == NULL) {
 948                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 949                                    " read from the MGS.  Does that filesystem "
 950                                    "exist?\n", profilenm);
 951                 GOTO(out_free, err = -EINVAL);
 952         }
 953         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 954                lprof->lp_md, lprof->lp_dt);
 955
 956         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 957                   strlen(ll_instance) + 2);
 958         if (!dt)
 959                 GOTO(out_free, err = -ENOMEM);
 960         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 961
 962         OBD_ALLOC(md, strlen(lprof->lp_md) +
 963                   strlen(ll_instance) + 2);
 964         if (!md)
 965                 GOTO(out_free, err = -ENOMEM);
 966         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 967
 968         /* connections, registrations, sb setup */
 969         err = client_common_fill_super(sb, md, dt,
 970                                        lsi->lsi_lmd->lmd_nllu,
 971                                        lsi->lsi_lmd->lmd_nllg);
 972
 973 out_free:
 974         if (md)
 975                 OBD_FREE(md, strlen(md) + 1);
 976         if (dt)
 977                 OBD_FREE(dt, strlen(dt) + 1);
 978         if (err)
 979                 ll_put_super(sb);
 980         else
 981                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 982
 983         RETURN(err);
 984 } /* ll_fill_super */
 985
 986
 987 void ll_put_super(struct super_block *sb)
 988 {
 989         struct config_llog_instance cfg;
 990         char   ll_instance[sizeof(sb) * 2 + 1];
 991         struct obd_device *obd;
 992         struct lustre_sb_info *lsi = s2lsi(sb);
 993         struct ll_sb_info *sbi = ll_s2sbi(sb);
 994         char *profilenm = get_profile_name(sb);
 995         int force = 1, next;
 996         ENTRY;
 997
 998         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 999
1000         ll_print_capa_stat(sbi);
1001
1002         sprintf(ll_instance, "%p", sb);
1003         cfg.cfg_instance = ll_instance;
1004         lustre_end_log(sb, NULL, &cfg);
1005
1006         if (sbi->ll_md_exp) {
1007                 obd = class_exp2obd(sbi->ll_md_exp);
1008                 if (obd)
1009                         force = obd->obd_force;
1010         }
1011
1012         /* We need to set force before the lov_disconnect in
1013            lustre_common_put_super, since l_d cleans up osc's as well. */
1014         if (force) {
1015                 next = 0;
1016                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1017                                                      &next)) != NULL) {
1018                         obd->obd_force = force;
1019                 }
1020         }
1021
1022         if (sbi->ll_lcq) {
1023                 /* Only if client_common_fill_super succeeded */
1024                 client_common_put_super(sb);
1025         }
1026         next = 0;
1027         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1028                 class_manual_cleanup(obd);
1029         }
1030
1031         if (profilenm)
1032                 class_del_profile(profilenm);
1033
1034         ll_free_sbi(sb);
1035         lsi->lsi_llsbi = NULL;
1036
1037         lustre_common_put_super(sb);
1038
1039         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1040
1041         cfs_module_put();
1042
1043         EXIT;
1044 } /* client_put_super */
1045
1046 #ifdef HAVE_REGISTER_CACHE
1047 #include <linux/cache_def.h>
1048 #ifdef HAVE_CACHE_RETURN_INT
1049 static int
1050 #else
1051 static void
1052 #endif
1053 ll_shrink_cache(int priority, unsigned int gfp_mask)
1054 {
1055         struct ll_sb_info *sbi;
1056         int count = 0;
1057
1058         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1059                 count += llap_shrink_cache(sbi, priority);
1060
1061 #ifdef HAVE_CACHE_RETURN_INT
1062         return count;
1063 #endif
1064 }
1065
1066 struct cache_definition ll_cache_definition = {
1067         .name = "llap_cache",
1068         .shrink = ll_shrink_cache
1069 };
1070 #endif /* HAVE_REGISTER_CACHE */
1071
1072 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1073 {
1074         struct inode *inode = NULL;
1075         /* NOTE: we depend on atomic igrab() -bzzz */
1076         lock_res_and_lock(lock);
1077         if (lock->l_ast_data) {
1078                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1079                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1080                         inode = igrab(lock->l_ast_data);
1081                 } else {
1082                         inode = lock->l_ast_data;
1083                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1084                                                 D_INFO : D_WARNING,
1085                                         lock, __FILE__, __func__, __LINE__,
1086                                         "l_ast_data %p is bogus: magic %08x",
1087                                         lock->l_ast_data, lli->lli_inode_magic);
1088                         inode = NULL;
1089                 }
1090         }
1091         unlock_res_and_lock(lock);
1092         return inode;
1093 }
1094
1095 static int null_if_equal(struct ldlm_lock *lock, void *data)
1096 {
1097         if (data == lock->l_ast_data) {
1098                 lock->l_ast_data = NULL;
1099
1100                 if (lock->l_req_mode != lock->l_granted_mode)
1101                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1102         }
1103
1104         return LDLM_ITER_CONTINUE;
1105 }
1106
1107 void ll_clear_inode(struct inode *inode)
1108 {
1109         struct ll_inode_info *lli = ll_i2info(inode);
1110         struct ll_sb_info *sbi = ll_i2sbi(inode);
1111         ENTRY;
1112
1113         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1114                inode->i_generation, inode);
1115
1116         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1117         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1118                          null_if_equal, inode);
1119
1120         LASSERT(!lli->lli_open_fd_write_count);
1121         LASSERT(!lli->lli_open_fd_read_count);
1122         LASSERT(!lli->lli_open_fd_exec_count);
1123
1124         if (lli->lli_mds_write_och)
1125                 ll_md_real_close(inode, FMODE_WRITE);
1126         if (lli->lli_mds_exec_och)
1127                 ll_md_real_close(inode, FMODE_EXEC);
1128         if (lli->lli_mds_read_och)
1129                 ll_md_real_close(inode, FMODE_READ);
1130
1131         if (lli->lli_smd) {
1132                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1133                                   null_if_equal, inode);
1134
1135                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1136                 lli->lli_smd = NULL;
1137         }
1138
1139         if (lli->lli_symlink_name) {
1140                 OBD_FREE(lli->lli_symlink_name,
1141                          strlen(lli->lli_symlink_name) + 1);
1142                 lli->lli_symlink_name = NULL;
1143         }
1144
1145         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1146                 LASSERT(lli->lli_posix_acl == NULL);
1147                 if (lli->lli_remote_perms) {
1148                         free_rmtperm_hash(lli->lli_remote_perms);
1149                         lli->lli_remote_perms = NULL;
1150                 }
1151         }
1152 #ifdef CONFIG_FS_POSIX_ACL
1153         else if (lli->lli_posix_acl) {
1154                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1155                 LASSERT(lli->lli_remote_perms == NULL);
1156                 posix_acl_release(lli->lli_posix_acl);
1157                 lli->lli_posix_acl = NULL;
1158         }
1159 #endif
1160         lli->lli_inode_magic = LLI_INODE_DEAD;
1161
1162 #ifdef HAVE_EXPORT___IGET
1163         spin_lock(&sbi->ll_deathrow_lock);
1164         list_del_init(&lli->lli_dead_list);
1165         spin_unlock(&sbi->ll_deathrow_lock);
1166 #endif
1167         ll_clear_inode_capas(inode);
1168
1169         EXIT;
1170 }
1171
1172 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1173                   struct md_open_data **mod)
1174 {
1175         struct lustre_md md;
1176         struct ll_sb_info *sbi = ll_i2sbi(inode);
1177         struct ptlrpc_request *request = NULL;
1178         int rc;
1179         ENTRY;
1180
1181         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1182                                      LUSTRE_OPC_ANY, NULL);
1183         if (IS_ERR(op_data))
1184                 RETURN(PTR_ERR(op_data));
1185
1186         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1187                         &request, mod);
1188         if (rc) {
1189                 ptlrpc_req_finished(request);
1190                 if (rc == -ENOENT) {
1191                         inode->i_nlink = 0;
1192                         /* Unlinked special device node? Or just a race?
1193                          * Pretend we done everything. */
1194                         if (!S_ISREG(inode->i_mode) &&
1195                             !S_ISDIR(inode->i_mode))
1196                                 rc = inode_setattr(inode, &op_data->op_attr);
1197                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1198                         CERROR("md_setattr fails: rc = %d\n", rc);
1199                 }
1200                 RETURN(rc);
1201         }
1202
1203         rc = md_get_lustre_md(sbi->ll_md_exp, request, REPLY_REC_OFF,
1204                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
1205         if (rc) {
1206                 ptlrpc_req_finished(request);
1207                 RETURN(rc);
1208         }
1209
1210         /* We call inode_setattr to adjust timestamps.
1211          * If there is at least some data in file, we cleared ATTR_SIZE
1212          * above to avoid invoking vmtruncate, otherwise it is important
1213          * to call vmtruncate in inode_setattr to update inode->i_size
1214          * (bug 6196) */
1215         rc = inode_setattr(inode, &op_data->op_attr);
1216
1217         /* Extract epoch data if obtained. */
1218         op_data->op_handle = md.body->handle;
1219         op_data->op_ioepoch = md.body->ioepoch;
1220
1221         ll_update_inode(inode, &md);
1222         ptlrpc_req_finished(request);
1223
1224         RETURN(rc);
1225 }
1226
1227 /* Close IO epoch and send Size-on-MDS attribute update. */
1228 static int ll_setattr_done_writing(struct inode *inode,
1229                                    struct md_op_data *op_data,
1230                                    struct md_open_data *mod)
1231 {
1232         struct ll_inode_info *lli = ll_i2info(inode);
1233         int rc = 0;
1234         ENTRY;
1235
1236         LASSERT(op_data != NULL);
1237         if (!S_ISREG(inode->i_mode))
1238                 RETURN(0);
1239
1240         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1241                op_data->op_ioepoch, PFID(&lli->lli_fid));
1242
1243         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1244         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1245         if (rc == -EAGAIN) {
1246                 /* MDS has instructed us to obtain Size-on-MDS attribute
1247                  * from OSTs and send setattr to back to MDS. */
1248                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1249                                          op_data->op_ioepoch);
1250         } else if (rc) {
1251                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1252                        inode->i_ino, rc);
1253         }
1254         RETURN(rc);
1255 }
1256
1257 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1258  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1259  * keep these values until such a time that objects are allocated for it.
1260  * We do the MDS operations first, as it is checking permissions for us.
1261  * We don't to the MDS RPC if there is nothing that we want to store there,
1262  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1263  * going to do an RPC anyways.
1264  *
1265  * If we are doing a truncate, we will send the mtime and ctime updates
1266  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1267  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1268  * at the same time.
1269  */
1270 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1271 {
1272         struct ll_inode_info *lli = ll_i2info(inode);
1273         struct lov_stripe_md *lsm = lli->lli_smd;
1274         struct ll_sb_info *sbi = ll_i2sbi(inode);
1275         struct md_op_data *op_data = NULL;
1276         struct md_open_data *mod = NULL;
1277         int ia_valid = attr->ia_valid;
1278         int rc = 0, rc1 = 0;
1279         ENTRY;
1280
1281         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1282                attr->ia_valid);
1283         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1284
1285         if (ia_valid & ATTR_SIZE) {
1286                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1287                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1288                                attr->ia_size, ll_file_maxbytes(inode));
1289                         RETURN(-EFBIG);
1290                 }
1291
1292                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1293         }
1294
1295         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1296         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1297                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1298                         RETURN(-EPERM);
1299         }
1300
1301         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1302         if (attr->ia_valid & ATTR_CTIME) {
1303                 attr->ia_ctime = CURRENT_TIME;
1304                 attr->ia_valid |= ATTR_CTIME_SET;
1305         }
1306         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1307                 attr->ia_atime = CURRENT_TIME;
1308                 attr->ia_valid |= ATTR_ATIME_SET;
1309         }
1310         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1311                 attr->ia_mtime = CURRENT_TIME;
1312                 attr->ia_valid |= ATTR_MTIME_SET;
1313         }
1314         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1315                 /* To avoid stale mtime on mds, obtain it from ost and send
1316                    to mds. */
1317                 rc = ll_glimpse_size(inode, 0);
1318                 if (rc)
1319                         RETURN(rc);
1320
1321                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1322                 attr->ia_mtime = inode->i_mtime;
1323         }
1324
1325         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1326                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1327                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1328                        CURRENT_SECONDS);
1329
1330         /* NB: ATTR_SIZE will only be set after this point if the size
1331          * resides on the MDS, ie, this file has no objects. */
1332         if (lsm)
1333                 attr->ia_valid &= ~ATTR_SIZE;
1334
1335         /* We always do an MDS RPC, even if we're only changing the size;
1336          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1337
1338         OBD_ALLOC_PTR(op_data);
1339         if (op_data == NULL)
1340                 RETURN(-ENOMEM);
1341
1342         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1343
1344         /* Open epoch for truncate. */
1345         if (ia_valid & ATTR_SIZE)
1346                 op_data->op_flags = MF_EPOCH_OPEN;
1347
1348         rc = ll_md_setattr(inode, op_data, &mod);
1349         if (rc)
1350                 GOTO(out, rc);
1351
1352         if (op_data->op_ioepoch)
1353                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1354                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1355
1356         if (!lsm || !S_ISREG(inode->i_mode)) {
1357                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1358                 GOTO(out, rc = 0);
1359         }
1360
1361         /* We really need to get our PW lock before we change inode->i_size.
1362          * If we don't we can race with other i_size updaters on our node, like
1363          * ll_file_read.  We can also race with i_size propogation to other
1364          * nodes through dirtying and writeback of final cached pages.  This
1365          * last one is especially bad for racing o_append users on other
1366          * nodes. */
1367         if (ia_valid & ATTR_SIZE) {
1368                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1369                                                            OBD_OBJECT_EOF } };
1370                 struct lustre_handle lockh = { 0 };
1371                 int err, ast_flags = 0;
1372                 /* XXX when we fix the AST intents to pass the discard-range
1373                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1374                  * XXX here. */
1375                 if (attr->ia_size == 0)
1376                         ast_flags = LDLM_AST_DISCARD_DATA;
1377
1378                 UNLOCK_INODE_MUTEX(inode);
1379                 UP_WRITE_I_ALLOC_SEM(inode);
1380                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1381                                     ast_flags);
1382 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1383                 DOWN_WRITE_I_ALLOC_SEM(inode);
1384                 LOCK_INODE_MUTEX(inode);
1385 #else
1386                 LOCK_INODE_MUTEX(inode);
1387                 DOWN_WRITE_I_ALLOC_SEM(inode);
1388 #endif
1389                 if (rc != 0)
1390                         GOTO(out, rc);
1391
1392                 /* Only ll_inode_size_lock is taken at this level.
1393                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1394                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1395                  * ll_truncate dropped ll_inode_size_lock() */
1396                 ll_inode_size_lock(inode, 0);
1397                 rc = vmtruncate(inode, attr->ia_size);
1398                 if (rc != 0) {
1399                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1400                         ll_inode_size_unlock(inode, 0);
1401                 }
1402
1403                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1404                 if (err) {
1405                         CERROR("ll_extent_unlock failed: %d\n", err);
1406                         if (!rc)
1407                                 rc = err;
1408                 }
1409         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1410                 obd_flag flags;
1411                 struct obd_info oinfo = { { { 0 } } };
1412                 struct obdo *oa;
1413
1414                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1415                        inode->i_ino, LTIME_S(attr->ia_mtime));
1416
1417                 OBDO_ALLOC(oa);
1418                 if (oa) {
1419                         oa->o_id = lsm->lsm_object_id;
1420                         oa->o_gr = lsm->lsm_object_gr;
1421                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1422
1423                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1424                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1425                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1426                                 OBD_MD_FLGROUP;
1427
1428                         obdo_from_inode(oa, inode, flags);
1429
1430                         oinfo.oi_oa = oa;
1431                         oinfo.oi_md = lsm;
1432                         oinfo.oi_capa = ll_mdscapa_get(inode);
1433
1434                         /* XXX: this looks unnecessary now. */
1435                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1436                         capa_put(oinfo.oi_capa);
1437                         if (rc)
1438                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1439                         OBDO_FREE(oa);
1440                 } else {
1441                         rc = -ENOMEM;
1442                 }
1443         }
1444         EXIT;
1445 out:
1446         if (op_data) {
1447                 if (op_data->op_ioepoch)
1448                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1449                 ll_finish_md_op_data(op_data);
1450         }
1451         return rc ? rc : rc1;
1452 }
1453
1454 int ll_setattr(struct dentry *de, struct iattr *attr)
1455 {
1456         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1457             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1458                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1459
1460         return ll_setattr_raw(de->d_inode, attr);
1461 }
1462
1463 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1464                        __u64 max_age)
1465 {
1466         struct ll_sb_info *sbi = ll_s2sbi(sb);
1467         struct obd_statfs obd_osfs;
1468         int rc;
1469         ENTRY;
1470
1471         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1472         if (rc) {
1473                 CERROR("md_statfs fails: rc = %d\n", rc);
1474                 RETURN(rc);
1475         }
1476
1477         osfs->os_type = sb->s_magic;
1478
1479         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1480                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1481
1482         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1483                               &obd_osfs, max_age);
1484         if (rc) {
1485                 CERROR("obd_statfs fails: rc = %d\n", rc);
1486                 RETURN(rc);
1487         }
1488
1489         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1490                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1491                obd_osfs.os_files);
1492
1493         osfs->os_bsize = obd_osfs.os_bsize;
1494         osfs->os_blocks = obd_osfs.os_blocks;
1495         osfs->os_bfree = obd_osfs.os_bfree;
1496         osfs->os_bavail = obd_osfs.os_bavail;
1497
1498         /* If we don't have as many objects free on the OST as inodes
1499          * on the MDS, we reduce the total number of inodes to
1500          * compensate, so that the "inodes in use" number is correct.
1501          */
1502         if (obd_osfs.os_ffree < osfs->os_ffree) {
1503                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1504                         obd_osfs.os_ffree;
1505                 osfs->os_ffree = obd_osfs.os_ffree;
1506         }
1507
1508         RETURN(rc);
1509 }
1510 #ifndef HAVE_STATFS_DENTRY_PARAM
1511 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1512 {
1513 #else
1514 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1515 {
1516         struct super_block *sb = de->d_sb;
1517 #endif
1518         struct obd_statfs osfs;
1519         int rc;
1520
1521         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1522         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1523
1524         /* For now we will always get up-to-date statfs values, but in the
1525          * future we may allow some amount of caching on the client (e.g.
1526          * from QOS or lprocfs updates). */
1527         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1);
1528         if (rc)
1529                 return rc;
1530
1531         statfs_unpack(sfs, &osfs);
1532
1533         /* We need to downshift for all 32-bit kernels, because we can't
1534          * tell if the kernel is being called via sys_statfs64() or not.
1535          * Stop before overflowing f_bsize - in which case it is better
1536          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1537         if (sizeof(long) < 8) {
1538                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1539                         sfs->f_bsize <<= 1;
1540
1541                         osfs.os_blocks >>= 1;
1542                         osfs.os_bfree >>= 1;
1543                         osfs.os_bavail >>= 1;
1544                 }
1545         }
1546
1547         sfs->f_blocks = osfs.os_blocks;
1548         sfs->f_bfree = osfs.os_bfree;
1549         sfs->f_bavail = osfs.os_bavail;
1550
1551         return 0;
1552 }
1553
1554 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1555 {
1556         struct ll_inode_info *lli;
1557         struct lov_stripe_md *lsm;
1558
1559         lli = ll_i2info(inode);
1560         LASSERT(lli->lli_size_sem_owner != current);
1561         down(&lli->lli_size_sem);
1562         LASSERT(lli->lli_size_sem_owner == NULL);
1563         lli->lli_size_sem_owner = current;
1564         lsm = lli->lli_smd;
1565         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1566                  lsm, lock_lsm);
1567         if (lock_lsm)
1568                 lov_stripe_lock(lsm);
1569 }
1570
1571 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1572 {
1573         struct ll_inode_info *lli;
1574         struct lov_stripe_md *lsm;
1575
1576         lli = ll_i2info(inode);
1577         lsm = lli->lli_smd;
1578         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1579                  lsm, unlock_lsm);
1580         if (unlock_lsm)
1581                 lov_stripe_unlock(lsm);
1582         LASSERT(lli->lli_size_sem_owner == current);
1583         lli->lli_size_sem_owner = NULL;
1584         up(&lli->lli_size_sem);
1585 }
1586
1587 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1588 {
1589         struct ll_inode_info *lli = ll_i2info(inode);
1590
1591         dump_lsm(D_INODE, lsm);
1592         dump_lsm(D_INODE, lli->lli_smd);
1593         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1594                  "lsm must be joined lsm %p\n", lsm);
1595         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1596         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1597                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1598         lli->lli_smd = lsm;
1599         lli->lli_maxbytes = lsm->lsm_maxbytes;
1600         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1601                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1602 }
1603
1604 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1605 {
1606         struct ll_inode_info *lli = ll_i2info(inode);
1607         struct mdt_body *body = md->body;
1608         struct lov_stripe_md *lsm = md->lsm;
1609         struct ll_sb_info *sbi = ll_i2sbi(inode);
1610
1611         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1612         if (lsm != NULL) {
1613                 if (lli->lli_smd == NULL) {
1614                         if (lsm->lsm_magic != LOV_MAGIC &&
1615                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1616                                 dump_lsm(D_ERROR, lsm);
1617                                 LBUG();
1618                         }
1619                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1620                                lsm, inode->i_ino, inode->i_generation, inode);
1621                         /* ll_inode_size_lock() requires it is only called
1622                          * with lli_smd != NULL or lock_lsm == 0 or we can
1623                          * race between lock/unlock.  bug 9547 */
1624                         lli->lli_smd = lsm;
1625                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1626                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1627                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1628                 } else {
1629                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1630                              lli->lli_smd->lsm_stripe_count ==
1631                                         lsm->lsm_stripe_count) {
1632                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1633                                         CERROR("lsm mismatch for inode %ld\n",
1634                                                 inode->i_ino);
1635                                         CERROR("lli_smd:\n");
1636                                         dump_lsm(D_ERROR, lli->lli_smd);
1637                                         CERROR("lsm:\n");
1638                                         dump_lsm(D_ERROR, lsm);
1639                                         LBUG();
1640                                 }
1641                         } else
1642                                 ll_replace_lsm(inode, lsm);
1643                 }
1644                 if (lli->lli_smd != lsm)
1645                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1646         }
1647
1648         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1649                 if (body->valid & OBD_MD_FLRMTPERM)
1650                         ll_update_remote_perm(inode, md->remote_perm);
1651         }
1652 #ifdef CONFIG_FS_POSIX_ACL
1653         else if (body->valid & OBD_MD_FLACL) {
1654                 spin_lock(&lli->lli_lock);
1655                 if (lli->lli_posix_acl)
1656                         posix_acl_release(lli->lli_posix_acl);
1657                 lli->lli_posix_acl = md->posix_acl;
1658                 spin_unlock(&lli->lli_lock);
1659         }
1660 #endif
1661         if (body->valid & OBD_MD_FLATIME &&
1662             body->atime > LTIME_S(inode->i_atime))
1663                 LTIME_S(inode->i_atime) = body->atime;
1664
1665         /* mtime is always updated with ctime, but can be set in past.
1666            As write and utime(2) may happen within 1 second, and utime's
1667            mtime has a priority over write's one, so take mtime from mds
1668            for the same ctimes. */
1669         if (body->valid & OBD_MD_FLCTIME &&
1670             body->ctime >= LTIME_S(inode->i_ctime)) {
1671                 LTIME_S(inode->i_ctime) = body->ctime;
1672                 if (body->valid & OBD_MD_FLMTIME) {
1673                         CDEBUG(D_INODE, "setting ino %lu mtime "
1674                                "from %lu to "LPU64"\n", inode->i_ino,
1675                                LTIME_S(inode->i_mtime), body->mtime);
1676                         LTIME_S(inode->i_mtime) = body->mtime;
1677                 }
1678         }
1679         if (body->valid & OBD_MD_FLMODE)
1680                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1681         if (body->valid & OBD_MD_FLTYPE)
1682                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1683         if (S_ISREG(inode->i_mode)) {
1684                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1685         } else {
1686                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1687         }
1688 #ifdef HAVE_INODE_BLKSIZE
1689         inode->i_blksize = 1<<inode->i_blkbits;
1690 #endif
1691         if (body->valid & OBD_MD_FLUID)
1692                 inode->i_uid = body->uid;
1693         if (body->valid & OBD_MD_FLGID)
1694                 inode->i_gid = body->gid;
1695         if (body->valid & OBD_MD_FLFLAGS)
1696                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1697         if (body->valid & OBD_MD_FLNLINK)
1698                 inode->i_nlink = body->nlink;
1699         if (body->valid & OBD_MD_FLRDEV)
1700 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1701                 inode->i_rdev = body->rdev;
1702 #else
1703                 inode->i_rdev = old_decode_dev(body->rdev);
1704 #endif
1705         if (body->valid & OBD_MD_FLSIZE) {
1706                 if (ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) {
1707                         if (lli->lli_flags & (LLIF_DONE_WRITING |
1708                                               LLIF_EPOCH_PENDING |
1709                                               LLIF_SOM_DIRTY))
1710                           CWARN("ino %lu flags %lu still has size authority!"
1711                                 "do not trust the size got from MDS\n",
1712                                 inode->i_ino, lli->lli_flags);
1713                         else {
1714                                 i_size_write(inode, body->size);
1715                                 lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1716                         }
1717                 } else {
1718                         i_size_write(inode, body->size);
1719                 }
1720
1721                 if (body->valid & OBD_MD_FLBLOCKS)
1722                         inode->i_blocks = body->blocks;
1723         }
1724
1725         if (body->valid & OBD_MD_FLID) {
1726                 /* FID shouldn't be changed! */
1727                 if (fid_is_sane(&lli->lli_fid)) {
1728                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1729                                  "Trying to change FID "DFID
1730                                  " to the "DFID", inode %lu/%u(%p)\n",
1731                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1732                                  inode->i_ino, inode->i_generation, inode);
1733                 } else
1734                         lli->lli_fid = body->fid1;
1735         }
1736
1737         LASSERT(fid_seq(&lli->lli_fid) != 0);
1738
1739         if (body->valid & OBD_MD_FLMDSCAPA) {
1740                 LASSERT(md->mds_capa);
1741                 ll_add_capa(inode, md->mds_capa);
1742         }
1743         if (body->valid & OBD_MD_FLOSSCAPA) {
1744                 LASSERT(md->oss_capa);
1745                 ll_add_capa(inode, md->oss_capa);
1746         }
1747 }
1748
1749 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1750 static struct backing_dev_info ll_backing_dev_info = {
1751         .ra_pages       = 0,    /* No readahead */
1752 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1753         .capabilities   = 0,    /* Does contribute to dirty memory */
1754 #else
1755         .memory_backed  = 0,    /* Does contribute to dirty memory */
1756 #endif
1757 };
1758 #endif
1759
1760 void ll_read_inode2(struct inode *inode, void *opaque)
1761 {
1762         struct lustre_md *md = opaque;
1763         struct ll_inode_info *lli = ll_i2info(inode);
1764         ENTRY;
1765
1766         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1767                inode->i_ino, inode->i_generation, inode);
1768
1769         ll_lli_init(lli);
1770
1771         LASSERT(!lli->lli_smd);
1772
1773         /* Core attributes from the MDS first.  This is a new inode, and
1774          * the VFS doesn't zero times in the core inode so we have to do
1775          * it ourselves.  They will be overwritten by either MDS or OST
1776          * attributes - we just need to make sure they aren't newer. */
1777         LTIME_S(inode->i_mtime) = 0;
1778         LTIME_S(inode->i_atime) = 0;
1779         LTIME_S(inode->i_ctime) = 0;
1780         inode->i_rdev = 0;
1781         ll_update_inode(inode, md);
1782
1783         /* OIDEBUG(inode); */
1784
1785         if (S_ISREG(inode->i_mode)) {
1786                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1787                 inode->i_op = &ll_file_inode_operations;
1788                 inode->i_fop = sbi->ll_fop;
1789                 inode->i_mapping->a_ops = &ll_aops;
1790                 EXIT;
1791         } else if (S_ISDIR(inode->i_mode)) {
1792                 inode->i_op = &ll_dir_inode_operations;
1793                 inode->i_fop = &ll_dir_operations;
1794                 inode->i_mapping->a_ops = &ll_dir_aops;
1795                 EXIT;
1796         } else if (S_ISLNK(inode->i_mode)) {
1797                 inode->i_op = &ll_fast_symlink_inode_operations;
1798                 EXIT;
1799         } else {
1800                 inode->i_op = &ll_special_inode_operations;
1801
1802 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1803                 init_special_inode(inode, inode->i_mode,
1804                                    kdev_t_to_nr(inode->i_rdev));
1805
1806                 /* initializing backing dev info. */
1807                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1808 #else
1809                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1810 #endif
1811                 EXIT;
1812         }
1813 }
1814
1815 void ll_delete_inode(struct inode *inode)
1816 {
1817         struct ll_sb_info *sbi = ll_i2sbi(inode);
1818         int rc;
1819         ENTRY;
1820
1821         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1822         if (rc) {
1823                 CERROR("fid_delete() failed, rc %d\n", rc);
1824         }
1825         clear_inode(inode);
1826
1827         EXIT;
1828 }
1829
1830 int ll_iocontrol(struct inode *inode, struct file *file,
1831                  unsigned int cmd, unsigned long arg)
1832 {
1833         struct ll_sb_info *sbi = ll_i2sbi(inode);
1834         struct ptlrpc_request *req = NULL;
1835         int rc, flags = 0;
1836         ENTRY;
1837
1838         switch(cmd) {
1839         case EXT3_IOC_GETFLAGS: {
1840                 struct mdt_body *body;
1841                 struct obd_capa *oc;
1842
1843                 oc = ll_mdscapa_get(inode);
1844                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1845                                 OBD_MD_FLFLAGS, 0, &req);
1846                 capa_put(oc);
1847                 if (rc) {
1848                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1849                         RETURN(-abs(rc));
1850                 }
1851
1852                 body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF,
1853                                       sizeof(*body));
1854
1855                 /*Now the ext3 will be packed directly back to client,
1856                  *no need convert here*/
1857                 flags = body->flags;
1858
1859                 ptlrpc_req_finished (req);
1860
1861                 RETURN(put_user(flags, (int *)arg));
1862         }
1863         case EXT3_IOC_SETFLAGS: {
1864                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1865                 struct obd_info oinfo = { { { 0 } } };
1866                 struct md_op_data *op_data;
1867
1868                 if (get_user(flags, (int *)arg))
1869                         RETURN(-EFAULT);
1870
1871                 oinfo.oi_md = lsm;
1872                 OBDO_ALLOC(oinfo.oi_oa);
1873                 if (!oinfo.oi_oa)
1874                         RETURN(-ENOMEM);
1875
1876                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1877                                              LUSTRE_OPC_ANY, NULL);
1878                 if (IS_ERR(op_data))
1879                         RETURN(PTR_ERR(op_data));
1880
1881                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1882                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1883                 rc = md_setattr(sbi->ll_md_exp, op_data,
1884                                 NULL, 0, NULL, 0, &req, NULL);
1885                 ll_finish_md_op_data(op_data);
1886                 ptlrpc_req_finished(req);
1887                 if (rc || lsm == NULL) {
1888                         OBDO_FREE(oinfo.oi_oa);
1889                         RETURN(rc);
1890                 }
1891
1892                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1893                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1894                 oinfo.oi_oa->o_flags = flags;
1895                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1896                                        OBD_MD_FLGROUP;
1897                 oinfo.oi_capa = ll_mdscapa_get(inode);
1898
1899                 obdo_from_inode(oinfo.oi_oa, inode,
1900                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1901                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1902                 capa_put(oinfo.oi_capa);
1903                 OBDO_FREE(oinfo.oi_oa);
1904                 if (rc) {
1905                         if (rc != -EPERM && rc != -EACCES)
1906                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1907                         RETURN(rc);
1908                 }
1909
1910                 inode->i_flags = ll_ext_to_inode_flags(flags |
1911                                                        MDS_BFLAG_EXT_FLAGS);
1912                 RETURN(0);
1913         }
1914         default:
1915                 RETURN(-ENOSYS);
1916         }
1917
1918         RETURN(0);
1919 }
1920
1921 int ll_flush_ctx(struct inode *inode)
1922 {
1923         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1924
1925         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1926
1927         obd_set_info_async(sbi->ll_md_exp,
1928                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1929                            0, NULL, NULL);
1930         obd_set_info_async(sbi->ll_dt_exp,
1931                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
1932                            0, NULL, NULL);
1933         return 0;
1934 }
1935
1936 /* umount -f client means force down, don't save state */
1937 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1938 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1939 {
1940         struct super_block *sb = vfsmnt->mnt_sb;
1941 #else
1942 void ll_umount_begin(struct super_block *sb)
1943 {
1944 #endif
1945         struct lustre_sb_info *lsi = s2lsi(sb);
1946         struct ll_sb_info *sbi = ll_s2sbi(sb);
1947         struct obd_device *obd;
1948         struct obd_ioctl_data ioc_data = { 0 };
1949         ENTRY;
1950
1951 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1952         if (!(flags & MNT_FORCE)) {
1953                 EXIT;
1954                 return;
1955         }
1956 #endif
1957
1958         /* Tell the MGC we got umount -f */
1959         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1960
1961         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1962                sb->s_count, atomic_read(&sb->s_active));
1963
1964         obd = class_exp2obd(sbi->ll_md_exp);
1965         if (obd == NULL) {
1966                 CERROR("Invalid MDC connection handle "LPX64"\n",
1967                        sbi->ll_md_exp->exp_handle.h_cookie);
1968                 EXIT;
1969                 return;
1970         }
1971         obd->obd_force = 1;
1972         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1973                       &ioc_data, NULL);
1974
1975         obd = class_exp2obd(sbi->ll_dt_exp);
1976         if (obd == NULL) {
1977                 CERROR("Invalid LOV connection handle "LPX64"\n",
1978                        sbi->ll_dt_exp->exp_handle.h_cookie);
1979                 EXIT;
1980                 return;
1981         }
1982
1983         obd->obd_force = 1;
1984         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1985                       &ioc_data, NULL);
1986
1987         /* Really, we'd like to wait until there are no requests outstanding,
1988          * and then continue.  For now, we just invalidate the requests,
1989          * schedule, and hope.
1990          */
1991         schedule();
1992
1993         EXIT;
1994 }
1995
1996 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1997 {
1998         struct ll_sb_info *sbi = ll_s2sbi(sb);
1999         int err;
2000         __u32 read_only;
2001
2002         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2003                 read_only = *flags & MS_RDONLY;
2004                 err = obd_set_info_async(sbi->ll_md_exp,
2005                                          sizeof(KEY_READ_ONLY) - 1,
2006                                          KEY_READ_ONLY, sizeof(read_only),
2007                                          &read_only, NULL);
2008                 if (err) {
2009                         CERROR("Failed to change the read-only flag during "
2010                                "remount: %d\n", err);
2011                         return err;
2012                 }
2013
2014                 if (read_only)
2015                         sb->s_flags |= MS_RDONLY;
2016                 else
2017                         sb->s_flags &= ~MS_RDONLY;
2018         }
2019         return 0;
2020 }
2021
2022 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2023                   int offset, struct super_block *sb)
2024 {
2025         struct ll_sb_info *sbi = NULL;
2026         struct lustre_md md;
2027         int rc = 0;
2028         ENTRY;
2029
2030         LASSERT(*inode || sb);
2031         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2032         prune_deathrow(sbi, 1);
2033         memset(&md, 0, sizeof(struct lustre_md));
2034
2035         rc = md_get_lustre_md(sbi->ll_md_exp, req, offset,
2036                               sbi->ll_dt_exp, sbi->ll_md_exp, &md);
2037         if (rc)
2038                 RETURN(rc);
2039
2040         if (*inode) {
2041                 ll_update_inode(*inode, &md);
2042         } else {
2043                 LASSERT(sb != NULL);
2044
2045                 /*
2046                  * At this point server returns to client's same fid as client
2047                  * generated for creating. So using ->fid1 is okay here.
2048                  */
2049                 LASSERT(fid_is_sane(&md.body->fid1));
2050
2051                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2052                 if (*inode == NULL || is_bad_inode(*inode)) {
2053                         if (md.lsm)
2054                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2055 #ifdef CONFIG_FS_POSIX_ACL
2056                         if (md.posix_acl) {
2057                                 posix_acl_release(md.posix_acl);
2058                                 md.posix_acl = NULL;
2059                         }
2060 #endif
2061                         rc = -ENOMEM;
2062                         CERROR("new_inode -fatal: rc %d\n", rc);
2063                         GOTO(out, rc);
2064                 }
2065         }
2066
2067         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2068                          ll_i2info(*inode)->lli_smd);
2069 out:
2070         md_free_lustre_md(sbi->ll_md_exp, &md);
2071         RETURN(rc);
2072 }
2073
2074 char *llap_origins[] = {
2075         [LLAP_ORIGIN_UNKNOWN] = "--",
2076         [LLAP_ORIGIN_READPAGE] = "rp",
2077         [LLAP_ORIGIN_READAHEAD] = "ra",
2078         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2079         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2080 };
2081
2082 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2083                                              struct list_head *list)
2084 {
2085         struct ll_async_page *llap;
2086         struct list_head *pos;
2087
2088         list_for_each(pos, list) {
2089                 if (pos == &sbi->ll_pglist)
2090                         return NULL;
2091                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2092                 if (llap->llap_page == NULL)
2093                         continue;
2094                 return llap;
2095         }
2096         LBUG();
2097         return NULL;
2098 }
2099
2100 int ll_obd_statfs(struct inode *inode, void *arg)
2101 {
2102         struct ll_sb_info *sbi = NULL;
2103         struct obd_device *client_obd = NULL, *lov_obd = NULL;
2104         struct lov_obd *lov = NULL;
2105         struct obd_statfs stat_buf = {0};
2106         char *buf = NULL;
2107         struct obd_ioctl_data *data = NULL;
2108         __u32 type, index;
2109         int len = 0, rc;
2110
2111         if (!inode || !(sbi = ll_i2sbi(inode)))
2112                 GOTO(out_statfs, rc = -EINVAL);
2113
2114         rc = obd_ioctl_getdata(&buf, &len, arg);
2115         if (rc)
2116                 GOTO(out_statfs, rc);
2117
2118         data = (void*)buf;
2119         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2120             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2121                 GOTO(out_statfs, rc = -EINVAL);
2122
2123         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2124         memcpy(&index, data->ioc_inlbuf2, sizeof(__u32));
2125
2126         if (type == LL_STATFS_MDC) {
2127                 if (index > 0)
2128                         GOTO(out_statfs, rc = -ENODEV);
2129                 client_obd = class_exp2obd(sbi->ll_md_exp);
2130         } else if (type == LL_STATFS_LOV) {
2131                 lov_obd = class_exp2obd(sbi->ll_dt_exp);
2132                 lov = &lov_obd->u.lov;
2133
2134                 if ((index >= lov->desc.ld_tgt_count))
2135                         GOTO(out_statfs, rc = -ENODEV);
2136                 if (!lov->lov_tgts[index])
2137                         /* Try again with the next index */
2138                         GOTO(out_statfs, rc = -EAGAIN);
2139
2140                 client_obd = class_exp2obd(lov->lov_tgts[index]->ltd_exp);
2141                 if (!lov->lov_tgts[index]->ltd_active)
2142                         GOTO(out_uuid, rc = -ENODATA);
2143         }
2144
2145         if (!client_obd)
2146                 GOTO(out_statfs, rc = -EINVAL);
2147
2148         rc = obd_statfs(client_obd, &stat_buf, cfs_time_current_64() - 1);
2149         if (rc)
2150                 GOTO(out_statfs, rc);
2151
2152         if (copy_to_user(data->ioc_pbuf1, &stat_buf, data->ioc_plen1))
2153                 GOTO(out_statfs, rc = -EFAULT);
2154
2155 out_uuid:
2156         if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(client_obd),
2157                          data->ioc_plen2))
2158                 rc = -EFAULT;
2159
2160 out_statfs:
2161         if (buf)
2162                 obd_ioctl_freedata(buf, len);
2163         return rc;
2164 }
2165
2166 int ll_process_config(struct lustre_cfg *lcfg)
2167 {
2168         char *ptr;
2169         void *sb;
2170         struct lprocfs_static_vars lvars;
2171         unsigned long x;
2172         int rc = 0;
2173
2174         lprocfs_init_vars(llite, &lvars);
2175
2176         /* The instance name contains the sb: lustre-client-aacfe000 */
2177         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2178         if (!ptr || !*(++ptr))
2179                 return -EINVAL;
2180         if (sscanf(ptr, "%lx", &x) != 1)
2181                 return -EINVAL;
2182         sb = (void *)x;
2183         /* This better be a real Lustre superblock! */
2184         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2185
2186         /* Note we have not called client_common_fill_super yet, so
2187            proc fns must be able to handle that! */
2188         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2189                                       lcfg, sb);
2190         return(rc);
2191 }
2192
2193 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2194 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2195                                        struct inode *i1, struct inode *i2,
2196                                        const char *name, int namelen,
2197                                        int mode, __u32 opc, void *data)
2198 {
2199         LASSERT(i1 != NULL);
2200
2201         if (namelen > ll_i2sbi(i1)->ll_namelen)
2202                 return ERR_PTR(-ENAMETOOLONG);
2203
2204         if (op_data == NULL)
2205                 OBD_ALLOC_PTR(op_data);
2206
2207         if (op_data == NULL)
2208                 return ERR_PTR(-ENOMEM);
2209
2210         ll_i2gids(op_data->op_suppgids, i1, i2);
2211         op_data->op_fid1 = *ll_inode2fid(i1);
2212         op_data->op_capa1 = ll_mdscapa_get(i1);
2213
2214         if (i2) {
2215                 op_data->op_fid2 = *ll_inode2fid(i2);
2216                 op_data->op_capa2 = ll_mdscapa_get(i2);
2217         } else {
2218                 fid_zero(&op_data->op_fid2);
2219         }
2220
2221         op_data->op_name = name;
2222         op_data->op_namelen = namelen;
2223         op_data->op_mode = mode;
2224         op_data->op_mod_time = CURRENT_SECONDS;
2225         op_data->op_fsuid = current->fsuid;
2226         op_data->op_fsgid = current->fsgid;
2227         op_data->op_cap = current->cap_effective;
2228         op_data->op_bias = MDS_CHECK_SPLIT;
2229         op_data->op_opc = opc;
2230         op_data->op_mds = 0;
2231         op_data->op_data = data;
2232
2233         return op_data;
2234 }
2235
2236 void ll_finish_md_op_data(struct md_op_data *op_data)
2237 {
2238         capa_put(op_data->op_capa1);
2239         capa_put(op_data->op_capa2);
2240         OBD_FREE_PTR(op_data);
2241 }
2242
2243 int ll_ioctl_getfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2244 {
2245         struct ll_sb_info *sbi = ll_i2sbi(inode);
2246         struct ptlrpc_request *req = NULL;
2247         struct mdt_body *body;
2248         char *cmd, *buf;
2249         struct obd_capa *oc;
2250         int rc, buflen;
2251         ENTRY;
2252
2253         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2254                 RETURN(-EBADE);
2255
2256         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2257
2258         OBD_ALLOC(cmd, ioc->cmd_len);
2259         if (!cmd)
2260                 RETURN(-ENOMEM);
2261         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2262                 GOTO(out, rc = -EFAULT);
2263
2264         oc = ll_mdscapa_get(inode);
2265         rc = md_getxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2266                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2267                          ioc->cmd_len, ioc->res_len, 0, &req);
2268         capa_put(oc);
2269         if (rc < 0) {
2270                 CERROR("mdc_getxattr %s [%s] failed: %d\n",
2271                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2272                 GOTO(out, rc);
2273         }
2274
2275         body = lustre_msg_buf(req->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
2276         LASSERT(body);
2277
2278         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2279         LASSERT(buflen <= ioc->res_len);
2280         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF + 1, ioc->res_len);
2281         LASSERT(buf);
2282         if (copy_to_user(ioc->res, buf, buflen))
2283                 GOTO(out, rc = -EFAULT);
2284         EXIT;
2285 out:
2286         if (req)
2287                 ptlrpc_req_finished(req);
2288         OBD_FREE(cmd, ioc->cmd_len);
2289         return rc;
2290 }
2291
2292 int ll_ioctl_setfacl(struct inode *inode, struct rmtacl_ioctl_data *ioc)
2293 {
2294         struct ll_sb_info *sbi = ll_i2sbi(inode);
2295         struct ptlrpc_request *req = NULL;
2296         char *cmd, *buf;
2297         struct obd_capa *oc;
2298         int buflen, rc;
2299         ENTRY;
2300
2301         if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT))
2302                 RETURN(-EBADE);
2303
2304         if (!(sbi->ll_flags & LL_SBI_ACL))
2305                 RETURN(-EOPNOTSUPP);
2306
2307         LASSERT(ioc->cmd && ioc->cmd_len && ioc->res && ioc->res_len);
2308
2309         OBD_ALLOC(cmd, ioc->cmd_len);
2310         if (!cmd)
2311                 RETURN(-ENOMEM);
2312         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
2313                 GOTO(out, rc = -EFAULT);
2314
2315         oc = ll_mdscapa_get(inode);
2316         rc = md_setxattr(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode), oc,
2317                          OBD_MD_FLXATTR, XATTR_NAME_LUSTRE_ACL, cmd,
2318                          ioc->cmd_len, ioc->res_len, 0, &req);
2319         capa_put(oc);
2320         if (rc) {
2321                 CERROR("mdc_setxattr %s [%s] failed: %d\n",
2322                        XATTR_NAME_LUSTRE_ACL, cmd, rc);
2323                 GOTO(out, rc);
2324         }
2325
2326         buflen = lustre_msg_buflen(req->rq_repmsg, REPLY_REC_OFF);
2327         LASSERT(buflen <= ioc->res_len);
2328         buf = lustre_msg_string(req->rq_repmsg, REPLY_REC_OFF, ioc->res_len);
2329         LASSERT(buf);
2330         if (copy_to_user(ioc->res, buf, buflen))
2331                 GOTO(out, rc = -EFAULT);
2332         EXIT;
2333 out:
2334         if (req)
2335                 ptlrpc_req_finished(req);
2336         OBD_FREE(cmd, ioc->cmd_len);
2337         return rc;
2338 }