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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC(sbi, sizeof(*sbi));
  82         if (!sbi)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         spin_lock_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119 #ifdef ENABLE_LLITE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127 #ifdef HAVE_EXPORT___IGET
 128         INIT_LIST_HEAD(&sbi->ll_deathrow);
 129         spin_lock_init(&sbi->ll_deathrow_lock);
 130 #endif
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 133                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 134         }
 135
 136         /* metadata statahead is enabled by default */
 137         sbi->ll_sa_max = LL_SA_RPC_DEF;
 138
 139         RETURN(sbi);
 140 }
 141
 142 void ll_free_sbi(struct super_block *sb)
 143 {
 144         struct ll_sb_info *sbi = ll_s2sbi(sb);
 145         ENTRY;
 146
 147         if (sbi != NULL) {
 148                 spin_lock(&ll_sb_lock);
 149                 list_del(&sbi->ll_list);
 150                 spin_unlock(&ll_sb_lock);
 151                 OBD_FREE(sbi, sizeof(*sbi));
 152         }
 153         EXIT;
 154 }
 155
 156 static struct dentry_operations ll_d_root_ops = {
 157 #ifdef DCACHE_LUSTRE_INVALID
 158         .d_compare = ll_dcompare,
 159 #endif
 160         .d_revalidate = ll_revalidate_nd,
 161 };
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 164 {
 165         struct inode *root = 0;
 166         struct ll_sb_info *sbi = ll_s2sbi(sb);
 167         struct obd_device *obd;
 168         struct obd_capa *oc = NULL;
 169         struct obd_statfs osfs;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct lustre_md lmd;
 174         obd_valid valid;
 175         int size, err, checksum;
 176         ENTRY;
 177
 178         obd = class_name2obd(md);
 179         if (!obd) {
 180                 CERROR("MD %s: not setup or attached\n", md);
 181                 RETURN(-EINVAL);
 182         }
 183
 184         OBD_ALLOC_PTR(data);
 185         if (data == NULL)
 186                 RETURN(-ENOMEM);
 187
 188         if (proc_lustre_fs_root) {
 189                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 190                                                   dt, md);
 191                 if (err < 0)
 192                         CERROR("could not register mount in /proc/fs/lustre\n");
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 199                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 200                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 201                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 202                                   OBD_CONNECT_VBR      | OBD_CONNECT_SOM;
 203
 204 #ifdef HAVE_LRU_RESIZE_SUPPORT
 205         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 206                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 207 #endif
 208 #ifdef CONFIG_FS_POSIX_ACL
 209         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 210 #endif
 211         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213
 214         if (sb->s_flags & MS_RDONLY)
 215                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 216         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 217                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 218
 219 #ifdef HAVE_MS_FLOCK_LOCK
 220         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 221         sb->s_flags |= MS_FLOCK_LOCK;
 222 #endif
 223
 224         if (sbi->ll_flags & LL_SBI_FLOCK)
 225                 sbi->ll_fop = &ll_file_operations_flock;
 226         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 227                 sbi->ll_fop = &ll_file_operations;
 228         else
 229                 sbi->ll_fop = &ll_file_operations_noflock;
 230
 231         /* real client */
 232         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 233         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 235
 236         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 237         if (err == -EBUSY) {
 238                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 239                                    "recovery, of which this client is not a "
 240                                    "part. Please wait for recovery to complete,"
 241                                    " abort, or time out.\n", md);
 242                 GOTO(out, err);
 243         } else if (err) {
 244                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 245                 GOTO(out, err);
 246         }
 247
 248         err = obd_fid_init(sbi->ll_md_exp);
 249         if (err) {
 250                 CERROR("Can't init metadata layer FID infrastructure, "
 251                        "rc %d\n", err);
 252                 GOTO(out_md, err);
 253         }
 254
 255         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 256         if (err)
 257                 GOTO(out_md_fid, err);
 258
 259         size = sizeof(*data);
 260         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 261                            KEY_CONN_DATA,  &size, data, NULL);
 262         if (err) {
 263                 CERROR("Get connect data failed: %d \n", err);
 264                 GOTO(out_md, err);
 265         }
 266
 267         LASSERT(osfs.os_bsize);
 268         sb->s_blocksize = osfs.os_bsize;
 269         sb->s_blocksize_bits = log2(osfs.os_bsize);
 270         sb->s_magic = LL_SUPER_MAGIC;
 271
 272         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 273          *         retval = in_file->f_op->sendfile(...);
 274          *         if (*ppos > max)
 275          *                 retval = -EOVERFLOW;
 276          *
 277          * it will check if *ppos is greater than max. However, max equals to
 278          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 279          * has been defined as a signed long long ineger in linux kernel. */
 280 #if BITS_PER_LONG == 64
 281         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 282 #else
 283         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 284 #endif
 285         sbi->ll_namelen = osfs.os_namelen;
 286         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 287
 288         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 289             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 290                 LCONSOLE_INFO("Disabling user_xattr feature because "
 291                               "it is not supported on the server\n");
 292                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 293         }
 294
 295         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 296 #ifdef MS_POSIXACL
 297                 sb->s_flags |= MS_POSIXACL;
 298 #endif
 299                 sbi->ll_flags |= LL_SBI_ACL;
 300         } else {
 301                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 302 #ifdef MS_POSIXACL
 303                 sb->s_flags &= ~MS_POSIXACL;
 304 #endif
 305                 sbi->ll_flags &= ~LL_SBI_ACL;
 306         }
 307
 308         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 309                 sbi->ll_flags |= LL_SBI_JOIN;
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 312                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 313                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 314                         LCONSOLE_INFO("client is set as remote by default.\n");
 315                 }
 316         } else {
 317                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 318                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 319                         LCONSOLE_INFO("client claims to be remote, but server "
 320                                       "rejected, forced to be local.\n");
 321                 }
 322         }
 323
 324         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 325                 LCONSOLE_INFO("client enabled MDS capability!\n");
 326                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 327         }
 328
 329         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 330                 LCONSOLE_INFO("client enabled OSS capability!\n");
 331                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 332         }
 333
 334         obd = class_name2obd(dt);
 335         if (!obd) {
 336                 CERROR("DT %s: not setup or attached\n", dt);
 337                 GOTO(out_md_fid, err = -ENODEV);
 338         }
 339
 340         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 341                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 342                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 343                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 344                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 345                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 346                                   OBD_CONNECT_SOM;
 347
 348         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 349                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 350                  * disabled by default, because it can still be enabled on the
 351                  * fly via /proc. As a consequence, we still need to come to an
 352                  * agreement on the supported algorithms at connect time */
 353                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 354
 355                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 356                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 357                 else
 358                         /* send the list of supported checksum types */
 359                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 360         }
 361
 362 #ifdef HAVE_LRU_RESIZE_SUPPORT
 363         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 364 #endif
 365         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 366                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 367
 368         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 369                "ocd_grant: %d\n", data->ocd_connect_flags,
 370                data->ocd_version, data->ocd_grant);
 371
 372         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 373         obd->obd_upcall.onu_upcall = cl_ocd_update;
 374         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 375
 376         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 377         if (err == -EBUSY) {
 378                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 379                                    "recovery, of which this client is not a "
 380                                    "part.  Please wait for recovery to "
 381                                    "complete, abort, or time out.\n", dt);
 382                 GOTO(out_md_fid, err);
 383         } else if (err) {
 384                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 385                 GOTO(out_md_fid, err);
 386         }
 387
 388         err = obd_fid_init(sbi->ll_dt_exp);
 389         if (err) {
 390                 CERROR("Can't init data layer FID infrastructure, "
 391                        "rc %d\n", err);
 392                 GOTO(out_dt, err);
 393         }
 394
 395         spin_lock(&sbi->ll_lco.lco_lock);
 396         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 397         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 398         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 399         spin_unlock(&sbi->ll_lco.lco_lock);
 400
 401         fid_zero(&sbi->ll_root_fid);
 402         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 403         if (err) {
 404                 CERROR("cannot mds_connect: rc = %d\n", err);
 405                 GOTO(out_lock_cn_cb, err);
 406         }
 407         if (!fid_is_sane(&sbi->ll_root_fid)) {
 408                 CERROR("Invalid root fid during mount\n");
 409                 GOTO(out_lock_cn_cb, err = -EINVAL);
 410         }
 411         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 412
 413         sb->s_op = &lustre_super_operations;
 414         sb->s_export_op = &lustre_export_operations;
 415
 416         /* make root inode
 417          * XXX: move this to after cbd setup? */
 418         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 419         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 420                 valid |= OBD_MD_FLRMTPERM;
 421         else if (sbi->ll_flags & LL_SBI_ACL)
 422                 valid |= OBD_MD_FLACL;
 423
 424         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 425                          &request);
 426         if (oc)
 427                 capa_put(oc);
 428         if (err) {
 429                 CERROR("md_getattr failed for root: rc = %d\n", err);
 430                 GOTO(out_lock_cn_cb, err);
 431         }
 432         memset(&lmd, 0, sizeof(lmd));
 433         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 434                                sbi->ll_md_exp, &lmd);
 435         if (err) {
 436                 CERROR("failed to understand root inode md: rc = %d\n", err);
 437                 ptlrpc_req_finished (request);
 438                 GOTO(out_lock_cn_cb, err);
 439         }
 440
 441         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 442         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 443         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 444         ptlrpc_req_finished(request);
 445
 446         if (root == NULL || IS_ERR(root)) {
 447                 if (lmd.lsm)
 448                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 449 #ifdef CONFIG_FS_POSIX_ACL
 450                 if (lmd.posix_acl) {
 451                         posix_acl_release(lmd.posix_acl);
 452                         lmd.posix_acl = NULL;
 453                 }
 454 #endif
 455                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 456                 root = NULL;
 457                 CERROR("lustre_lite: bad iget4 for root\n");
 458                 GOTO(out_root, err);
 459         }
 460
 461         err = ll_close_thread_start(&sbi->ll_lcq);
 462         if (err) {
 463                 CERROR("cannot start close thread: rc %d\n", err);
 464                 GOTO(out_root, err);
 465         }
 466
 467 #ifdef CONFIG_FS_POSIX_ACL
 468         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 469                 rct_init(&sbi->ll_rct);
 470                 et_init(&sbi->ll_et);
 471         }
 472 #endif
 473
 474         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 475         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 476                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 477                                  NULL);
 478         cl_sb_init(sb);
 479
 480         sb->s_root = d_alloc_root(root);
 481         if (data != NULL)
 482                 OBD_FREE(data, sizeof(*data));
 483
 484         sb->s_root->d_op = &ll_d_root_ops;
 485
 486         sbi->ll_sdev_orig = sb->s_dev;
 487
 488         /* We set sb->s_dev equal on all lustre clients in order to support
 489          * NFS export clustering.  NFSD requires that the FSID be the same
 490          * on all clients. */
 491         /* s_dev is also used in lt_compare() to compare two fs, but that is
 492          * only a node-local comparison. */
 493         uuid = obd_get_uuid(sbi->ll_md_exp);
 494         if (uuid != NULL)
 495                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 496
 497         RETURN(err);
 498 out_root:
 499         if (root)
 500                 iput(root);
 501 out_lock_cn_cb:
 502         obd_fid_fini(sbi->ll_dt_exp);
 503 out_dt:
 504         obd_disconnect(sbi->ll_dt_exp);
 505         sbi->ll_dt_exp = NULL;
 506 out_md_fid:
 507         obd_fid_fini(sbi->ll_md_exp);
 508 out_md:
 509         obd_disconnect(sbi->ll_md_exp);
 510         sbi->ll_md_exp = NULL;
 511 out:
 512         if (data != NULL)
 513                 OBD_FREE_PTR(data);
 514         lprocfs_unregister_mountpoint(sbi);
 515         return err;
 516 }
 517
 518 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 519 {
 520         int size, rc;
 521
 522         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 523         size = sizeof(int);
 524         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 525                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 526         if (rc)
 527                 CERROR("Get max mdsize error rc %d \n", rc);
 528
 529         RETURN(rc);
 530 }
 531
 532 void ll_dump_inode(struct inode *inode)
 533 {
 534         struct list_head *tmp;
 535         struct dentry *dentry;
 536         int dentry_count = 0;
 537
 538         LASSERT(inode != NULL);
 539         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u state %lx\n",
 540                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 541                inode->i_mode, atomic_read(&inode->i_count), inode->i_state);
 542
 543         list_for_each(tmp, &inode->i_dentry) {
 544                 dentry = list_entry(tmp, struct dentry, d_alias);
 545                 CERROR("Alias[%d] dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u, "
 546                        "flags=0x%x, fsdata=%p\n", dentry_count, dentry,
 547                        dentry->d_name.len, dentry->d_name.name,
 548                        dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 549                        dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 550                        dentry->d_flags, dentry->d_fsdata);
 551
 552                 dentry_count++;
 553         }
 554
 555 }
 556
 557 void lustre_dump_dentry(struct dentry *dentry, int recur)
 558 {
 559         struct list_head *tmp;
 560         int subdirs = 0;
 561
 562         LASSERT(dentry != NULL);
 563
 564         list_for_each(tmp, &dentry->d_subdirs)
 565                 subdirs++;
 566
 567         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 568                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 569                dentry->d_name.len, dentry->d_name.name,
 570                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 571                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 572                dentry->d_flags, dentry->d_fsdata, subdirs);
 573         if (dentry->d_inode != NULL)
 574                 ll_dump_inode(dentry->d_inode);
 575
 576         if (recur == 0)
 577                 return;
 578
 579         list_for_each(tmp, &dentry->d_subdirs) {
 580                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 581                 lustre_dump_dentry(d, recur - 1);
 582         }
 583 }
 584
 585 #ifdef HAVE_EXPORT___IGET
 586 static void prune_dir_dentries(struct inode *inode)
 587 {
 588         struct dentry *dentry, *prev = NULL;
 589
 590         /* due to lustre specific logic, a directory
 591          * can have few dentries - a bug from VFS POV */
 592 restart:
 593         spin_lock(&dcache_lock);
 594         if (!list_empty(&inode->i_dentry)) {
 595                 dentry = list_entry(inode->i_dentry.prev,
 596                                     struct dentry, d_alias);
 597                 /* in order to prevent infinite loops we
 598                  * break if previous dentry is busy */
 599                 if (dentry != prev) {
 600                         prev = dentry;
 601                         dget_locked(dentry);
 602                         spin_unlock(&dcache_lock);
 603
 604                         /* try to kill all child dentries */
 605                         lock_dentry(dentry);
 606                         shrink_dcache_parent(dentry);
 607                         unlock_dentry(dentry);
 608                         dput(dentry);
 609
 610                         /* now try to get rid of current dentry */
 611                         d_prune_aliases(inode);
 612                         goto restart;
 613                 }
 614         }
 615         spin_unlock(&dcache_lock);
 616 }
 617
 618 static void prune_deathrow_one(struct ll_inode_info *lli)
 619 {
 620         struct inode *inode = ll_info2i(lli);
 621
 622         /* first, try to drop any dentries - they hold a ref on the inode */
 623         if (S_ISDIR(inode->i_mode))
 624                 prune_dir_dentries(inode);
 625         else
 626                 d_prune_aliases(inode);
 627
 628
 629         /* if somebody still uses it, leave it */
 630         LASSERT(atomic_read(&inode->i_count) > 0);
 631         if (atomic_read(&inode->i_count) > 1)
 632                 goto out;
 633
 634         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 635                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 636
 637         /* seems nobody uses it anymore */
 638         inode->i_nlink = 0;
 639
 640 out:
 641         iput(inode);
 642         return;
 643 }
 644
 645 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 646 {
 647         struct ll_inode_info *lli;
 648         int empty;
 649
 650         do {
 651                 if (need_resched() && try)
 652                         break;
 653
 654                 if (try) {
 655                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 656                                 break;
 657                 } else {
 658                         spin_lock(&sbi->ll_deathrow_lock);
 659                 }
 660
 661                 empty = 1;
 662                 lli = NULL;
 663                 if (!list_empty(&sbi->ll_deathrow)) {
 664                         lli = list_entry(sbi->ll_deathrow.next,
 665                                          struct ll_inode_info,
 666                                          lli_dead_list);
 667                         list_del_init(&lli->lli_dead_list);
 668                         if (!list_empty(&sbi->ll_deathrow))
 669                                 empty = 0;
 670                 }
 671                 spin_unlock(&sbi->ll_deathrow_lock);
 672
 673                 if (lli)
 674                         prune_deathrow_one(lli);
 675
 676         } while (empty == 0);
 677 }
 678 #else /* !HAVE_EXPORT___IGET */
 679 #define prune_deathrow(sbi, try) do {} while (0)
 680 #endif /* HAVE_EXPORT___IGET */
 681
 682 void client_common_put_super(struct super_block *sb)
 683 {
 684         struct ll_sb_info *sbi = ll_s2sbi(sb);
 685         ENTRY;
 686
 687 #ifdef CONFIG_FS_POSIX_ACL
 688         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 689                 et_fini(&sbi->ll_et);
 690                 rct_fini(&sbi->ll_rct);
 691         }
 692 #endif
 693
 694         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 695
 696         ll_close_thread_shutdown(sbi->ll_lcq);
 697
 698         cl_sb_fini(sb);
 699
 700         /* destroy inodes in deathrow */
 701         prune_deathrow(sbi, 0);
 702
 703         list_del(&sbi->ll_conn_chain);
 704
 705         obd_fid_fini(sbi->ll_dt_exp);
 706         obd_disconnect(sbi->ll_dt_exp);
 707         sbi->ll_dt_exp = NULL;
 708
 709         lprocfs_unregister_mountpoint(sbi);
 710
 711         obd_fid_fini(sbi->ll_md_exp);
 712         obd_disconnect(sbi->ll_md_exp);
 713         sbi->ll_md_exp = NULL;
 714
 715         EXIT;
 716 }
 717
 718 void ll_kill_super(struct super_block *sb)
 719 {
 720         struct ll_sb_info *sbi;
 721
 722         ENTRY;
 723
 724         /* not init sb ?*/
 725         if (!(sb->s_flags & MS_ACTIVE))
 726                 return;
 727
 728         sbi = ll_s2sbi(sb);
 729         /* we need restore s_dev from changed for clustred NFS before put_super
 730          * because new kernels have cached s_dev and change sb->s_dev in
 731          * put_super not affected real removing devices */
 732         if (sbi)
 733                 sb->s_dev = sbi->ll_sdev_orig;
 734         EXIT;
 735 }
 736
 737 char *ll_read_opt(const char *opt, char *data)
 738 {
 739         char *value;
 740         char *retval;
 741         ENTRY;
 742
 743         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 744         if (strncmp(opt, data, strlen(opt)))
 745                 RETURN(NULL);
 746         if ((value = strchr(data, '=')) == NULL)
 747                 RETURN(NULL);
 748
 749         value++;
 750         OBD_ALLOC(retval, strlen(value) + 1);
 751         if (!retval) {
 752                 CERROR("out of memory!\n");
 753                 RETURN(NULL);
 754         }
 755
 756         memcpy(retval, value, strlen(value)+1);
 757         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 758         RETURN(retval);
 759 }
 760
 761 static inline int ll_set_opt(const char *opt, char *data, int fl)
 762 {
 763         if (strncmp(opt, data, strlen(opt)) != 0)
 764                 return(0);
 765         else
 766                 return(fl);
 767 }
 768
 769 /* non-client-specific mount options are parsed in lmd_parse */
 770 static int ll_options(char *options, int *flags)
 771 {
 772         int tmp;
 773         char *s1 = options, *s2;
 774         ENTRY;
 775
 776         if (!options)
 777                 RETURN(0);
 778
 779         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 780
 781         while (*s1) {
 782                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 783                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 784                 if (tmp) {
 785                         *flags |= tmp;
 786                         goto next;
 787                 }
 788                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 789                 if (tmp) {
 790                         *flags |= tmp;
 791                         goto next;
 792                 }
 793                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 794                 if (tmp) {
 795                         *flags |= tmp;
 796                         goto next;
 797                 }
 798                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 799                 if (tmp) {
 800                         *flags &= ~tmp;
 801                         goto next;
 802                 }
 803                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 804                 if (tmp) {
 805                         *flags |= tmp;
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 809                 if (tmp) {
 810                         *flags &= ~tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 814                 if (tmp) {
 815                         /* Ignore deprecated mount option.  The client will
 816                          * always try to mount with ACL support, whether this
 817                          * is used depends on whether server supports it. */
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 821                 if (tmp) {
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829
 830                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 831                 if (tmp) {
 832                         *flags |= tmp;
 833                         goto next;
 834                 }
 835                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags &= ~tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 841                 if (tmp) {
 842                         *flags |= tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 846                 if (tmp) {
 847                         *flags &= ~tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 856                 if (tmp) {
 857                         *flags &= ~tmp;
 858                         goto next;
 859                 }
 860
 861                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 862                                    s1);
 863                 RETURN(-EINVAL);
 864
 865 next:
 866                 /* Find next opt */
 867                 s2 = strchr(s1, ',');
 868                 if (s2 == NULL)
 869                         break;
 870                 s1 = s2 + 1;
 871         }
 872         RETURN(0);
 873 }
 874
 875 void ll_lli_init(struct ll_inode_info *lli)
 876 {
 877         lli->lli_inode_magic = LLI_INODE_MAGIC;
 878         sema_init(&lli->lli_size_sem, 1);
 879         sema_init(&lli->lli_write_sem, 1);
 880         lli->lli_flags = 0;
 881         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 882         spin_lock_init(&lli->lli_lock);
 883         INIT_LIST_HEAD(&lli->lli_close_list);
 884         lli->lli_inode_magic = LLI_INODE_MAGIC;
 885         sema_init(&lli->lli_och_sem, 1);
 886         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 887         lli->lli_mds_exec_och = NULL;
 888         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 889         lli->lli_open_fd_exec_count = 0;
 890         INIT_LIST_HEAD(&lli->lli_dead_list);
 891         lli->lli_remote_perms = NULL;
 892         lli->lli_rmtperm_utime = 0;
 893         sema_init(&lli->lli_rmtperm_sem, 1);
 894         INIT_LIST_HEAD(&lli->lli_oss_capas);
 895 }
 896
 897 int ll_fill_super(struct super_block *sb)
 898 {
 899         struct lustre_profile *lprof;
 900         struct lustre_sb_info *lsi = s2lsi(sb);
 901         struct ll_sb_info *sbi;
 902         char  *dt = NULL, *md = NULL;
 903         char  *profilenm = get_profile_name(sb);
 904         struct config_llog_instance cfg = {0, };
 905         char   ll_instance[sizeof(sb) * 2 + 1];
 906         int    err;
 907         ENTRY;
 908
 909         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 910
 911         cfs_module_get();
 912
 913         /* client additional sb info */
 914         lsi->lsi_llsbi = sbi = ll_init_sbi();
 915         if (!sbi) {
 916                 cfs_module_put();
 917                 RETURN(-ENOMEM);
 918         }
 919
 920         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 921         if (err)
 922                 GOTO(out_free, err);
 923
 924         /* Generate a string unique to this super, in case some joker tries
 925            to mount the same fs at two mount points.
 926            Use the address of the super itself.*/
 927         sprintf(ll_instance, "%p", sb);
 928         cfg.cfg_instance = ll_instance;
 929         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 930
 931         /* set up client obds */
 932         err = lustre_process_log(sb, profilenm, &cfg);
 933         if (err < 0) {
 934                 CERROR("Unable to process log: %d\n", err);
 935                 GOTO(out_free, err);
 936         }
 937
 938         lprof = class_get_profile(profilenm);
 939         if (lprof == NULL) {
 940                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 941                                    " read from the MGS.  Does that filesystem "
 942                                    "exist?\n", profilenm);
 943                 GOTO(out_free, err = -EINVAL);
 944         }
 945         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 946                lprof->lp_md, lprof->lp_dt);
 947
 948         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 949                   strlen(ll_instance) + 2);
 950         if (!dt)
 951                 GOTO(out_free, err = -ENOMEM);
 952         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 953
 954         OBD_ALLOC(md, strlen(lprof->lp_md) +
 955                   strlen(ll_instance) + 2);
 956         if (!md)
 957                 GOTO(out_free, err = -ENOMEM);
 958         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 959
 960         /* connections, registrations, sb setup */
 961         err = client_common_fill_super(sb, md, dt);
 962
 963 out_free:
 964         if (md)
 965                 OBD_FREE(md, strlen(md) + 1);
 966         if (dt)
 967                 OBD_FREE(dt, strlen(dt) + 1);
 968         if (err)
 969                 ll_put_super(sb);
 970         else
 971                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 972
 973         RETURN(err);
 974 } /* ll_fill_super */
 975
 976
 977 void lu_context_keys_dump(void);
 978
 979 void ll_put_super(struct super_block *sb)
 980 {
 981         struct config_llog_instance cfg;
 982         char   ll_instance[sizeof(sb) * 2 + 1];
 983         struct obd_device *obd;
 984         struct lustre_sb_info *lsi = s2lsi(sb);
 985         struct ll_sb_info *sbi = ll_s2sbi(sb);
 986         char *profilenm = get_profile_name(sb);
 987         int force = 1, next;
 988         ENTRY;
 989
 990         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 991
 992         ll_print_capa_stat(sbi);
 993
 994         sprintf(ll_instance, "%p", sb);
 995         cfg.cfg_instance = ll_instance;
 996         lustre_end_log(sb, NULL, &cfg);
 997
 998         if (sbi->ll_md_exp) {
 999                 obd = class_exp2obd(sbi->ll_md_exp);
1000                 if (obd)
1001                         force = obd->obd_force;
1002         }
1003
1004         /* We need to set force before the lov_disconnect in
1005            lustre_common_put_super, since l_d cleans up osc's as well. */
1006         if (force) {
1007                 next = 0;
1008                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1009                                                      &next)) != NULL) {
1010                         obd->obd_force = force;
1011                 }
1012         }
1013
1014         if (sbi->ll_lcq) {
1015                 /* Only if client_common_fill_super succeeded */
1016                 client_common_put_super(sb);
1017         }
1018
1019         next = 0;
1020         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1021                 class_manual_cleanup(obd);
1022         }
1023
1024         if (profilenm)
1025                 class_del_profile(profilenm);
1026
1027         ll_free_sbi(sb);
1028         lsi->lsi_llsbi = NULL;
1029
1030         lustre_common_put_super(sb);
1031
1032         cl_env_cache_purge(~0);
1033
1034         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1035
1036         cfs_module_put();
1037
1038         EXIT;
1039 } /* client_put_super */
1040
1041 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1042 {
1043         struct inode *inode = NULL;
1044         /* NOTE: we depend on atomic igrab() -bzzz */
1045         lock_res_and_lock(lock);
1046         if (lock->l_ast_data) {
1047                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1048                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1049                         inode = igrab(lock->l_ast_data);
1050                 } else {
1051                         inode = lock->l_ast_data;
1052                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1053                                                 D_INFO : D_WARNING,
1054                                         lock, __FILE__, __func__, __LINE__,
1055                                         "l_ast_data %p is bogus: magic %08x",
1056                                         lock->l_ast_data, lli->lli_inode_magic);
1057                         inode = NULL;
1058                 }
1059         }
1060         unlock_res_and_lock(lock);
1061         return inode;
1062 }
1063
1064 static int null_if_equal(struct ldlm_lock *lock, void *data)
1065 {
1066         if (data == lock->l_ast_data) {
1067                 lock->l_ast_data = NULL;
1068
1069                 if (lock->l_req_mode != lock->l_granted_mode)
1070                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1071         }
1072
1073         return LDLM_ITER_CONTINUE;
1074 }
1075
1076 void ll_clear_inode(struct inode *inode)
1077 {
1078         struct ll_inode_info *lli = ll_i2info(inode);
1079         struct ll_sb_info *sbi = ll_i2sbi(inode);
1080         ENTRY;
1081
1082         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1083                inode->i_generation, inode);
1084
1085         if (S_ISDIR(inode->i_mode)) {
1086                 /* these should have been cleared in ll_file_release */
1087                 LASSERT(lli->lli_sai == NULL);
1088                 LASSERT(lli->lli_opendir_key == NULL);
1089                 LASSERT(lli->lli_opendir_pid == 0);
1090         }
1091
1092         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1093         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1094                          null_if_equal, inode);
1095
1096         LASSERT(!lli->lli_open_fd_write_count);
1097         LASSERT(!lli->lli_open_fd_read_count);
1098         LASSERT(!lli->lli_open_fd_exec_count);
1099
1100         if (lli->lli_mds_write_och)
1101                 ll_md_real_close(inode, FMODE_WRITE);
1102         if (lli->lli_mds_exec_och)
1103                 ll_md_real_close(inode, FMODE_EXEC);
1104         if (lli->lli_mds_read_och)
1105                 ll_md_real_close(inode, FMODE_READ);
1106
1107         if (lli->lli_symlink_name) {
1108                 OBD_FREE(lli->lli_symlink_name,
1109                          strlen(lli->lli_symlink_name) + 1);
1110                 lli->lli_symlink_name = NULL;
1111         }
1112
1113         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1114                 LASSERT(lli->lli_posix_acl == NULL);
1115                 if (lli->lli_remote_perms) {
1116                         free_rmtperm_hash(lli->lli_remote_perms);
1117                         lli->lli_remote_perms = NULL;
1118                 }
1119         }
1120 #ifdef CONFIG_FS_POSIX_ACL
1121         else if (lli->lli_posix_acl) {
1122                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1123                 LASSERT(lli->lli_remote_perms == NULL);
1124                 posix_acl_release(lli->lli_posix_acl);
1125                 lli->lli_posix_acl = NULL;
1126         }
1127 #endif
1128         lli->lli_inode_magic = LLI_INODE_DEAD;
1129
1130 #ifdef HAVE_EXPORT___IGET
1131         spin_lock(&sbi->ll_deathrow_lock);
1132         list_del_init(&lli->lli_dead_list);
1133         spin_unlock(&sbi->ll_deathrow_lock);
1134 #endif
1135         ll_clear_inode_capas(inode);
1136         /*
1137          * XXX This has to be done before lsm is freed below, because
1138          * cl_object still uses inode lsm.
1139          */
1140         cl_inode_fini(inode);
1141
1142         if (lli->lli_smd) {
1143                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1144                 lli->lli_smd = NULL;
1145         }
1146
1147
1148         EXIT;
1149 }
1150
1151 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1152                   struct md_open_data **mod)
1153 {
1154         struct lustre_md md;
1155         struct ll_sb_info *sbi = ll_i2sbi(inode);
1156         struct ptlrpc_request *request = NULL;
1157         int rc;
1158         ENTRY;
1159
1160         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1161                                      LUSTRE_OPC_ANY, NULL);
1162         if (IS_ERR(op_data))
1163                 RETURN(PTR_ERR(op_data));
1164
1165         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1166                         &request, mod);
1167         if (rc) {
1168                 ptlrpc_req_finished(request);
1169                 if (rc == -ENOENT) {
1170                         inode->i_nlink = 0;
1171                         /* Unlinked special device node? Or just a race?
1172                          * Pretend we done everything. */
1173                         if (!S_ISREG(inode->i_mode) &&
1174                             !S_ISDIR(inode->i_mode))
1175                                 rc = inode_setattr(inode, &op_data->op_attr);
1176                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1177                         CERROR("md_setattr fails: rc = %d\n", rc);
1178                 }
1179                 RETURN(rc);
1180         }
1181
1182         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1183                               sbi->ll_md_exp, &md);
1184         if (rc) {
1185                 ptlrpc_req_finished(request);
1186                 RETURN(rc);
1187         }
1188
1189         /* We call inode_setattr to adjust timestamps.
1190          * If there is at least some data in file, we cleared ATTR_SIZE
1191          * above to avoid invoking vmtruncate, otherwise it is important
1192          * to call vmtruncate in inode_setattr to update inode->i_size
1193          * (bug 6196) */
1194         rc = inode_setattr(inode, &op_data->op_attr);
1195
1196         /* Extract epoch data if obtained. */
1197         op_data->op_handle = md.body->handle;
1198         op_data->op_ioepoch = md.body->ioepoch;
1199
1200         ll_update_inode(inode, &md);
1201         ptlrpc_req_finished(request);
1202
1203         RETURN(rc);
1204 }
1205
1206 /* Close IO epoch and send Size-on-MDS attribute update. */
1207 static int ll_setattr_done_writing(struct inode *inode,
1208                                    struct md_op_data *op_data,
1209                                    struct md_open_data *mod)
1210 {
1211         struct ll_inode_info *lli = ll_i2info(inode);
1212         int rc = 0;
1213         ENTRY;
1214
1215         LASSERT(op_data != NULL);
1216         if (!S_ISREG(inode->i_mode))
1217                 RETURN(0);
1218
1219         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1220                op_data->op_ioepoch, PFID(&lli->lli_fid));
1221
1222         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1223         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1224         if (rc == -EAGAIN) {
1225                 /* MDS has instructed us to obtain Size-on-MDS attribute
1226                  * from OSTs and send setattr to back to MDS. */
1227                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1228                                          op_data->op_ioepoch);
1229         } else if (rc) {
1230                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1231                        inode->i_ino, rc);
1232         }
1233         RETURN(rc);
1234 }
1235
1236 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1237 {
1238         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1239         int rc;
1240
1241         rc = cl_setattr_do_truncate(inode, size, capa);
1242         ll_truncate_free_capa(capa);
1243         return rc;
1244 }
1245
1246 static int ll_setattr_ost(struct inode *inode)
1247 {
1248         struct obd_capa *capa = ll_mdscapa_get(inode);
1249         int rc;
1250
1251         rc = cl_setattr_ost(inode, capa);
1252         capa_put(capa);
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 md_op_data *op_data = NULL;
1275         struct md_open_data *mod = NULL;
1276         int ia_valid = attr->ia_valid;
1277         int rc = 0, rc1 = 0;
1278         ENTRY;
1279
1280         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1281                attr->ia_valid);
1282         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1283
1284         if (ia_valid & ATTR_SIZE) {
1285                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1286                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1287                                attr->ia_size, ll_file_maxbytes(inode));
1288                         RETURN(-EFBIG);
1289                 }
1290
1291                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1292         }
1293
1294         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1295         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1296                 if (current->fsuid != inode->i_uid &&
1297                     !cfs_capable(CFS_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 = cl_glimpse_size(inode);
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                        cfs_time_current_sec());
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 ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1346             (ia_valid & ATTR_SIZE))
1347                 op_data->op_flags = MF_EPOCH_OPEN;
1348
1349         rc = ll_md_setattr(inode, op_data, &mod);
1350         if (rc)
1351                 GOTO(out, rc);
1352
1353         ll_ioepoch_open(lli, op_data->op_ioepoch);
1354         if (!lsm || !S_ISREG(inode->i_mode)) {
1355                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1356                 GOTO(out, rc = 0);
1357         }
1358
1359         if (ia_valid & ATTR_SIZE)
1360                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1361         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1362                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1363                        inode->i_ino, LTIME_S(attr->ia_mtime));
1364                 rc = ll_setattr_ost(inode);
1365         }
1366         EXIT;
1367 out:
1368         if (op_data) {
1369                 if (op_data->op_ioepoch)
1370                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1371                 ll_finish_md_op_data(op_data);
1372         }
1373         return rc ? rc : rc1;
1374 }
1375
1376 int ll_setattr(struct dentry *de, struct iattr *attr)
1377 {
1378         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1379             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1380                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1381
1382         if ((de->d_inode->i_mode & S_ISUID) &&
1383             !(attr->ia_mode & S_ISUID) &&
1384             !(attr->ia_valid & ATTR_KILL_SUID))
1385                 attr->ia_valid |= ATTR_KILL_SUID;
1386
1387         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1388             !(attr->ia_mode & S_ISGID) &&
1389             !(attr->ia_valid & ATTR_KILL_SGID))
1390                 attr->ia_valid |= ATTR_KILL_SGID;
1391
1392         return ll_setattr_raw(de->d_inode, attr);
1393 }
1394
1395 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1396                        __u64 max_age, __u32 flags)
1397 {
1398         struct ll_sb_info *sbi = ll_s2sbi(sb);
1399         struct obd_statfs obd_osfs;
1400         int rc;
1401         ENTRY;
1402
1403         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1404         if (rc) {
1405                 CERROR("md_statfs fails: rc = %d\n", rc);
1406                 RETURN(rc);
1407         }
1408
1409         osfs->os_type = sb->s_magic;
1410
1411         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1412                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1413
1414         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1415                 flags |= OBD_STATFS_NODELAY;
1416
1417         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1418                               &obd_osfs, max_age, flags);
1419         if (rc) {
1420                 CERROR("obd_statfs fails: rc = %d\n", rc);
1421                 RETURN(rc);
1422         }
1423
1424         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1425                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1426                obd_osfs.os_files);
1427
1428         osfs->os_bsize = obd_osfs.os_bsize;
1429         osfs->os_blocks = obd_osfs.os_blocks;
1430         osfs->os_bfree = obd_osfs.os_bfree;
1431         osfs->os_bavail = obd_osfs.os_bavail;
1432
1433         /* If we don't have as many objects free on the OST as inodes
1434          * on the MDS, we reduce the total number of inodes to
1435          * compensate, so that the "inodes in use" number is correct.
1436          */
1437         if (obd_osfs.os_ffree < osfs->os_ffree) {
1438                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1439                         obd_osfs.os_ffree;
1440                 osfs->os_ffree = obd_osfs.os_ffree;
1441         }
1442
1443         RETURN(rc);
1444 }
1445 #ifndef HAVE_STATFS_DENTRY_PARAM
1446 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1447 {
1448 #else
1449 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1450 {
1451         struct super_block *sb = de->d_sb;
1452 #endif
1453         struct obd_statfs osfs;
1454         int rc;
1455
1456         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1457         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1458
1459         /* For now we will always get up-to-date statfs values, but in the
1460          * future we may allow some amount of caching on the client (e.g.
1461          * from QOS or lprocfs updates). */
1462         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1463         if (rc)
1464                 return rc;
1465
1466         statfs_unpack(sfs, &osfs);
1467
1468         /* We need to downshift for all 32-bit kernels, because we can't
1469          * tell if the kernel is being called via sys_statfs64() or not.
1470          * Stop before overflowing f_bsize - in which case it is better
1471          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1472         if (sizeof(long) < 8) {
1473                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1474                         sfs->f_bsize <<= 1;
1475
1476                         osfs.os_blocks >>= 1;
1477                         osfs.os_bfree >>= 1;
1478                         osfs.os_bavail >>= 1;
1479                 }
1480         }
1481
1482         sfs->f_blocks = osfs.os_blocks;
1483         sfs->f_bfree = osfs.os_bfree;
1484         sfs->f_bavail = osfs.os_bavail;
1485
1486         return 0;
1487 }
1488
1489 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1490 {
1491         struct ll_inode_info *lli;
1492         struct lov_stripe_md *lsm;
1493
1494         lli = ll_i2info(inode);
1495         LASSERT(lli->lli_size_sem_owner != current);
1496         down(&lli->lli_size_sem);
1497         LASSERT(lli->lli_size_sem_owner == NULL);
1498         lli->lli_size_sem_owner = current;
1499         lsm = lli->lli_smd;
1500         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1501                  lsm, lock_lsm);
1502         if (lock_lsm)
1503                 lov_stripe_lock(lsm);
1504 }
1505
1506 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1507 {
1508         struct ll_inode_info *lli;
1509         struct lov_stripe_md *lsm;
1510
1511         lli = ll_i2info(inode);
1512         lsm = lli->lli_smd;
1513         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1514                  lsm, unlock_lsm);
1515         if (unlock_lsm)
1516                 lov_stripe_unlock(lsm);
1517         LASSERT(lli->lli_size_sem_owner == current);
1518         lli->lli_size_sem_owner = NULL;
1519         up(&lli->lli_size_sem);
1520 }
1521
1522 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1523 {
1524         struct ll_inode_info *lli = ll_i2info(inode);
1525
1526         dump_lsm(D_INODE, lsm);
1527         dump_lsm(D_INODE, lli->lli_smd);
1528         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1529                  "lsm must be joined lsm %p\n", lsm);
1530         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1531         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1532                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1533         lli->lli_smd = lsm;
1534         lli->lli_maxbytes = lsm->lsm_maxbytes;
1535         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1536                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1537 }
1538
1539 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1540 {
1541         struct ll_inode_info *lli = ll_i2info(inode);
1542         struct mdt_body *body = md->body;
1543         struct lov_stripe_md *lsm = md->lsm;
1544         struct ll_sb_info *sbi = ll_i2sbi(inode);
1545
1546         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1547         if (lsm != NULL) {
1548                 if (lli->lli_smd == NULL) {
1549                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1550                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1551                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1552                                 dump_lsm(D_ERROR, lsm);
1553                                 LBUG();
1554                         }
1555                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1556                                lsm, inode->i_ino, inode->i_generation, inode);
1557                         cl_inode_init(inode, md);
1558                         /* ll_inode_size_lock() requires it is only
1559                          * called with lli_smd != NULL or lock_lsm == 0
1560                          *  or we can race between lock/unlock.
1561                          *  bug 9547 */
1562                         lli->lli_smd = lsm;
1563                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1564                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1565                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1566                 } else {
1567                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1568                              lli->lli_smd->lsm_stripe_count ==
1569                                         lsm->lsm_stripe_count) {
1570                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1571                                         CERROR("lsm mismatch for inode %ld\n",
1572                                                 inode->i_ino);
1573                                         CERROR("lli_smd:\n");
1574                                         dump_lsm(D_ERROR, lli->lli_smd);
1575                                         CERROR("lsm:\n");
1576                                         dump_lsm(D_ERROR, lsm);
1577                                         LBUG();
1578                                 }
1579                         } else {
1580                                 cl_inode_init(inode, md);
1581                                 ll_replace_lsm(inode, lsm);
1582                         }
1583                 }
1584                 if (lli->lli_smd != lsm)
1585                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1586         }
1587
1588         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1589                 if (body->valid & OBD_MD_FLRMTPERM)
1590                         ll_update_remote_perm(inode, md->remote_perm);
1591         }
1592 #ifdef CONFIG_FS_POSIX_ACL
1593         else if (body->valid & OBD_MD_FLACL) {
1594                 spin_lock(&lli->lli_lock);
1595                 if (lli->lli_posix_acl)
1596                         posix_acl_release(lli->lli_posix_acl);
1597                 lli->lli_posix_acl = md->posix_acl;
1598                 spin_unlock(&lli->lli_lock);
1599         }
1600 #endif
1601         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1602         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1603
1604         if (body->valid & OBD_MD_FLATIME &&
1605             body->atime > LTIME_S(inode->i_atime))
1606                 LTIME_S(inode->i_atime) = body->atime;
1607
1608         /* mtime is always updated with ctime, but can be set in past.
1609            As write and utime(2) may happen within 1 second, and utime's
1610            mtime has a priority over write's one, so take mtime from mds
1611            for the same ctimes. */
1612         if (body->valid & OBD_MD_FLCTIME &&
1613             body->ctime >= LTIME_S(inode->i_ctime)) {
1614                 LTIME_S(inode->i_ctime) = body->ctime;
1615                 if (body->valid & OBD_MD_FLMTIME) {
1616                         CDEBUG(D_INODE, "setting ino %lu mtime "
1617                                "from %lu to "LPU64"\n", inode->i_ino,
1618                                LTIME_S(inode->i_mtime), body->mtime);
1619                         LTIME_S(inode->i_mtime) = body->mtime;
1620                 }
1621         }
1622         if (body->valid & OBD_MD_FLMODE)
1623                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1624         if (body->valid & OBD_MD_FLTYPE)
1625                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1626         LASSERT(inode->i_mode != 0);
1627         if (S_ISREG(inode->i_mode)) {
1628                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1629         } else {
1630                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1631         }
1632 #ifdef HAVE_INODE_BLKSIZE
1633         inode->i_blksize = 1<<inode->i_blkbits;
1634 #endif
1635         if (body->valid & OBD_MD_FLUID)
1636                 inode->i_uid = body->uid;
1637         if (body->valid & OBD_MD_FLGID)
1638                 inode->i_gid = body->gid;
1639         if (body->valid & OBD_MD_FLFLAGS)
1640                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1641         if (body->valid & OBD_MD_FLNLINK)
1642                 inode->i_nlink = body->nlink;
1643         if (body->valid & OBD_MD_FLRDEV)
1644                 inode->i_rdev = old_decode_dev(body->rdev);
1645
1646         if (body->valid & OBD_MD_FLID) {
1647                 /* FID shouldn't be changed! */
1648                 if (fid_is_sane(&lli->lli_fid)) {
1649                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1650                                  "Trying to change FID "DFID
1651                                  " to the "DFID", inode %lu/%u(%p)\n",
1652                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1653                                  inode->i_ino, inode->i_generation, inode);
1654                 } else
1655                         lli->lli_fid = body->fid1;
1656         }
1657
1658         LASSERT(fid_seq(&lli->lli_fid) != 0);
1659
1660         if (body->valid & OBD_MD_FLSIZE) {
1661                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1662                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1663                         struct lustre_handle lockh;
1664                         ldlm_mode_t mode;
1665
1666                         /* As it is possible a blocking ast has been processed
1667                          * by this time, we need to check there is an UPDATE
1668                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1669                          * it. */
1670                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1671                                                &lockh);
1672                         if (mode) {
1673                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1674                                                       LLIF_EPOCH_PENDING |
1675                                                       LLIF_SOM_DIRTY)) {
1676                                         CERROR("ino %lu flags %lu still has "
1677                                                "size authority! do not trust "
1678                                                "the size got from MDS\n",
1679                                                inode->i_ino, lli->lli_flags);
1680                                 } else {
1681                                         /* Use old size assignment to avoid
1682                                          * deadlock bz14138 & bz14326 */
1683                                         inode->i_size = body->size;
1684                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1685                                 }
1686                                 ldlm_lock_decref(&lockh, mode);
1687                         }
1688                 } else {
1689                         /* Use old size assignment to avoid
1690                          * deadlock bz14138 & bz14326 */
1691                         inode->i_size = body->size;
1692                 }
1693
1694                 if (body->valid & OBD_MD_FLBLOCKS)
1695                         inode->i_blocks = body->blocks;
1696         }
1697
1698         if (body->valid & OBD_MD_FLMDSCAPA) {
1699                 LASSERT(md->mds_capa);
1700                 ll_add_capa(inode, md->mds_capa);
1701         }
1702         if (body->valid & OBD_MD_FLOSSCAPA) {
1703                 LASSERT(md->oss_capa);
1704                 ll_add_capa(inode, md->oss_capa);
1705         }
1706 }
1707
1708 static struct backing_dev_info ll_backing_dev_info = {
1709         .ra_pages       = 0,    /* No readahead */
1710 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1711         .capabilities   = 0,    /* Does contribute to dirty memory */
1712 #else
1713         .memory_backed  = 0,    /* Does contribute to dirty memory */
1714 #endif
1715 };
1716
1717 void ll_read_inode2(struct inode *inode, void *opaque)
1718 {
1719         struct lustre_md *md = opaque;
1720         struct ll_inode_info *lli = ll_i2info(inode);
1721         ENTRY;
1722
1723         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1724                inode->i_ino, inode->i_generation, inode);
1725
1726         ll_lli_init(lli);
1727
1728         LASSERT(!lli->lli_smd);
1729
1730         /* Core attributes from the MDS first.  This is a new inode, and
1731          * the VFS doesn't zero times in the core inode so we have to do
1732          * it ourselves.  They will be overwritten by either MDS or OST
1733          * attributes - we just need to make sure they aren't newer. */
1734         LTIME_S(inode->i_mtime) = 0;
1735         LTIME_S(inode->i_atime) = 0;
1736         LTIME_S(inode->i_ctime) = 0;
1737         inode->i_rdev = 0;
1738         ll_update_inode(inode, md);
1739
1740         /* OIDEBUG(inode); */
1741
1742         if (S_ISREG(inode->i_mode)) {
1743                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1744                 inode->i_op = &ll_file_inode_operations;
1745                 inode->i_fop = sbi->ll_fop;
1746                 inode->i_mapping->a_ops = &ll_aops;
1747                 EXIT;
1748         } else if (S_ISDIR(inode->i_mode)) {
1749                 inode->i_op = &ll_dir_inode_operations;
1750                 inode->i_fop = &ll_dir_operations;
1751                 inode->i_mapping->a_ops = &ll_dir_aops;
1752                 EXIT;
1753         } else if (S_ISLNK(inode->i_mode)) {
1754                 inode->i_op = &ll_fast_symlink_inode_operations;
1755                 EXIT;
1756         } else {
1757                 inode->i_op = &ll_special_inode_operations;
1758
1759                 init_special_inode(inode, inode->i_mode,
1760                                    kdev_t_to_nr(inode->i_rdev));
1761
1762                 /* initializing backing dev info. */
1763                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1764
1765                 EXIT;
1766         }
1767 }
1768
1769 void ll_delete_inode(struct inode *inode)
1770 {
1771         struct ll_sb_info *sbi = ll_i2sbi(inode);
1772         int rc;
1773         ENTRY;
1774
1775         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1776         if (rc) {
1777                 CERROR("fid_delete() failed, rc %d\n", rc);
1778         }
1779         truncate_inode_pages(&inode->i_data, 0);
1780         clear_inode(inode);
1781
1782         EXIT;
1783 }
1784
1785 int ll_iocontrol(struct inode *inode, struct file *file,
1786                  unsigned int cmd, unsigned long arg)
1787 {
1788         struct ll_sb_info *sbi = ll_i2sbi(inode);
1789         struct ptlrpc_request *req = NULL;
1790         int rc, flags = 0;
1791         ENTRY;
1792
1793         switch(cmd) {
1794         case EXT3_IOC_GETFLAGS: {
1795                 struct mdt_body *body;
1796                 struct obd_capa *oc;
1797
1798                 oc = ll_mdscapa_get(inode);
1799                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1800                                 OBD_MD_FLFLAGS, 0, &req);
1801                 capa_put(oc);
1802                 if (rc) {
1803                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1804                         RETURN(-abs(rc));
1805                 }
1806
1807                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1808
1809                 flags = body->flags;
1810
1811                 ptlrpc_req_finished(req);
1812
1813                 RETURN(put_user(flags, (int *)arg));
1814         }
1815         case EXT3_IOC_SETFLAGS: {
1816                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1817                 struct obd_info oinfo = { { { 0 } } };
1818                 struct md_op_data *op_data;
1819
1820                 if (get_user(flags, (int *)arg))
1821                         RETURN(-EFAULT);
1822
1823                 oinfo.oi_md = lsm;
1824                 OBDO_ALLOC(oinfo.oi_oa);
1825                 if (!oinfo.oi_oa)
1826                         RETURN(-ENOMEM);
1827
1828                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1829                                              LUSTRE_OPC_ANY, NULL);
1830                 if (IS_ERR(op_data))
1831                         RETURN(PTR_ERR(op_data));
1832
1833                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1834                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1835                 rc = md_setattr(sbi->ll_md_exp, op_data,
1836                                 NULL, 0, NULL, 0, &req, NULL);
1837                 ll_finish_md_op_data(op_data);
1838                 ptlrpc_req_finished(req);
1839                 if (rc || lsm == NULL) {
1840                         OBDO_FREE(oinfo.oi_oa);
1841                         RETURN(rc);
1842                 }
1843
1844                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1845                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1846                 oinfo.oi_oa->o_flags = flags;
1847                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1848                                        OBD_MD_FLGROUP;
1849                 oinfo.oi_capa = ll_mdscapa_get(inode);
1850
1851                 obdo_from_inode(oinfo.oi_oa, inode,
1852                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1853                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1854                 capa_put(oinfo.oi_capa);
1855                 OBDO_FREE(oinfo.oi_oa);
1856                 if (rc) {
1857                         if (rc != -EPERM && rc != -EACCES)
1858                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1859                         RETURN(rc);
1860                 }
1861
1862                 inode->i_flags = ll_ext_to_inode_flags(flags |
1863                                                        MDS_BFLAG_EXT_FLAGS);
1864                 RETURN(0);
1865         }
1866         default:
1867                 RETURN(-ENOSYS);
1868         }
1869
1870         RETURN(0);
1871 }
1872
1873 int ll_flush_ctx(struct inode *inode)
1874 {
1875         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1876
1877         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1878
1879         obd_set_info_async(sbi->ll_md_exp,
1880                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1881                            0, NULL, NULL);
1882         obd_set_info_async(sbi->ll_dt_exp,
1883                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1884                            0, NULL, NULL);
1885         return 0;
1886 }
1887
1888 /* umount -f client means force down, don't save state */
1889 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1890 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1891 {
1892         struct super_block *sb = vfsmnt->mnt_sb;
1893 #else
1894 void ll_umount_begin(struct super_block *sb)
1895 {
1896 #endif
1897         struct lustre_sb_info *lsi = s2lsi(sb);
1898         struct ll_sb_info *sbi = ll_s2sbi(sb);
1899         struct obd_device *obd;
1900         struct obd_ioctl_data ioc_data = { 0 };
1901         ENTRY;
1902
1903 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1904         if (!(flags & MNT_FORCE)) {
1905                 EXIT;
1906                 return;
1907         }
1908 #endif
1909
1910         /* Tell the MGC we got umount -f */
1911         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1912
1913         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1914                sb->s_count, atomic_read(&sb->s_active));
1915
1916         obd = class_exp2obd(sbi->ll_md_exp);
1917         if (obd == NULL) {
1918                 CERROR("Invalid MDC connection handle "LPX64"\n",
1919                        sbi->ll_md_exp->exp_handle.h_cookie);
1920                 EXIT;
1921                 return;
1922         }
1923         obd->obd_force = 1;
1924         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1925                       &ioc_data, NULL);
1926
1927         obd = class_exp2obd(sbi->ll_dt_exp);
1928         if (obd == NULL) {
1929                 CERROR("Invalid LOV connection handle "LPX64"\n",
1930                        sbi->ll_dt_exp->exp_handle.h_cookie);
1931                 EXIT;
1932                 return;
1933         }
1934
1935         obd->obd_force = 1;
1936         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1937                       &ioc_data, NULL);
1938
1939         /* Really, we'd like to wait until there are no requests outstanding,
1940          * and then continue.  For now, we just invalidate the requests,
1941          * schedule() and sleep one second if needed, and hope.
1942          */
1943         schedule();
1944 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1945         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1946                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1947                                      cfs_time_seconds(1));
1948                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1949                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1950                                       "may try to umount it a bit later\n",
1951                                       atomic_read(&vfsmnt->mnt_count));
1952         }
1953 #endif
1954
1955         EXIT;
1956 }
1957
1958 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1959 {
1960         struct ll_sb_info *sbi = ll_s2sbi(sb);
1961         int err;
1962         __u32 read_only;
1963
1964         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1965                 read_only = *flags & MS_RDONLY;
1966                 err = obd_set_info_async(sbi->ll_md_exp,
1967                                          sizeof(KEY_READ_ONLY),
1968                                          KEY_READ_ONLY, sizeof(read_only),
1969                                          &read_only, NULL);
1970                 if (err) {
1971                         CERROR("Failed to change the read-only flag during "
1972                                "remount: %d\n", err);
1973                         return err;
1974                 }
1975
1976                 if (read_only)
1977                         sb->s_flags |= MS_RDONLY;
1978                 else
1979                         sb->s_flags &= ~MS_RDONLY;
1980         }
1981         return 0;
1982 }
1983
1984 int ll_prep_inode(struct inode **inode,
1985                   struct ptlrpc_request *req,
1986                   struct super_block *sb)
1987 {
1988         struct ll_sb_info *sbi = NULL;
1989         struct lustre_md md;
1990         int rc;
1991         ENTRY;
1992
1993         LASSERT(*inode || sb);
1994         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1995         prune_deathrow(sbi, 1);
1996         memset(&md, 0, sizeof(struct lustre_md));
1997
1998         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1999                               sbi->ll_md_exp, &md);
2000         if (rc)
2001                 RETURN(rc);
2002
2003         if (*inode) {
2004                 ll_update_inode(*inode, &md);
2005         } else {
2006                 LASSERT(sb != NULL);
2007
2008                 /*
2009                  * At this point server returns to client's same fid as client
2010                  * generated for creating. So using ->fid1 is okay here.
2011                  */
2012                 LASSERT(fid_is_sane(&md.body->fid1));
2013
2014                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2015                 if (*inode == NULL || IS_ERR(*inode)) {
2016                         if (md.lsm)
2017                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2018 #ifdef CONFIG_FS_POSIX_ACL
2019                         if (md.posix_acl) {
2020                                 posix_acl_release(md.posix_acl);
2021                                 md.posix_acl = NULL;
2022                         }
2023 #endif
2024                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2025                         *inode = NULL;
2026                         CERROR("new_inode -fatal: rc %d\n", rc);
2027                         GOTO(out, rc);
2028                 }
2029         }
2030
2031         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2032                          ll_i2info(*inode)->lli_smd);
2033 out:
2034         md_free_lustre_md(sbi->ll_md_exp, &md);
2035         RETURN(rc);
2036 }
2037
2038 int ll_obd_statfs(struct inode *inode, void *arg)
2039 {
2040         struct ll_sb_info *sbi = NULL;
2041         struct obd_export *exp;
2042         char *buf = NULL;
2043         struct obd_ioctl_data *data = NULL;
2044         __u32 type;
2045         int len = 0, rc;
2046
2047         if (!inode || !(sbi = ll_i2sbi(inode)))
2048                 GOTO(out_statfs, rc = -EINVAL);
2049
2050         rc = obd_ioctl_getdata(&buf, &len, arg);
2051         if (rc)
2052                 GOTO(out_statfs, rc);
2053
2054         data = (void*)buf;
2055         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2056             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2057                 GOTO(out_statfs, rc = -EINVAL);
2058
2059         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2060         if (type == LL_STATFS_MDC)
2061                 exp = sbi->ll_md_exp;
2062         else if (type == LL_STATFS_LOV)
2063                 exp = sbi->ll_dt_exp;
2064         else
2065                 GOTO(out_statfs, rc = -ENODEV);
2066
2067         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2068         if (rc)
2069                 GOTO(out_statfs, rc);
2070 out_statfs:
2071         if (buf)
2072                 obd_ioctl_freedata(buf, len);
2073         return rc;
2074 }
2075
2076 int ll_process_config(struct lustre_cfg *lcfg)
2077 {
2078         char *ptr;
2079         void *sb;
2080         struct lprocfs_static_vars lvars;
2081         unsigned long x;
2082         int rc = 0;
2083
2084         lprocfs_llite_init_vars(&lvars);
2085
2086         /* The instance name contains the sb: lustre-client-aacfe000 */
2087         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2088         if (!ptr || !*(++ptr))
2089                 return -EINVAL;
2090         if (sscanf(ptr, "%lx", &x) != 1)
2091                 return -EINVAL;
2092         sb = (void *)x;
2093         /* This better be a real Lustre superblock! */
2094         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2095
2096         /* Note we have not called client_common_fill_super yet, so
2097            proc fns must be able to handle that! */
2098         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2099                                       lcfg, sb);
2100         if (rc > 0)
2101                 rc = 0;
2102         return(rc);
2103 }
2104
2105 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2106 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2107                                        struct inode *i1, struct inode *i2,
2108                                        const char *name, int namelen,
2109                                        int mode, __u32 opc, void *data)
2110 {
2111         LASSERT(i1 != NULL);
2112
2113         if (namelen > ll_i2sbi(i1)->ll_namelen)
2114                 return ERR_PTR(-ENAMETOOLONG);
2115
2116         if (op_data == NULL)
2117                 OBD_ALLOC_PTR(op_data);
2118
2119         if (op_data == NULL)
2120                 return ERR_PTR(-ENOMEM);
2121
2122         ll_i2gids(op_data->op_suppgids, i1, i2);
2123         op_data->op_fid1 = *ll_inode2fid(i1);
2124         op_data->op_capa1 = ll_mdscapa_get(i1);
2125
2126         if (i2) {
2127                 op_data->op_fid2 = *ll_inode2fid(i2);
2128                 op_data->op_capa2 = ll_mdscapa_get(i2);
2129         } else {
2130                 fid_zero(&op_data->op_fid2);
2131                 op_data->op_capa2 = NULL;
2132         }
2133
2134         op_data->op_name = name;
2135         op_data->op_namelen = namelen;
2136         op_data->op_mode = mode;
2137         op_data->op_mod_time = cfs_time_current_sec();
2138         op_data->op_fsuid = current->fsuid;
2139         op_data->op_fsgid = current->fsgid;
2140         op_data->op_cap = cfs_curproc_cap_pack();
2141         op_data->op_bias = MDS_CHECK_SPLIT;
2142         op_data->op_opc = opc;
2143         op_data->op_mds = 0;
2144         op_data->op_data = data;
2145
2146         return op_data;
2147 }
2148
2149 void ll_finish_md_op_data(struct md_op_data *op_data)
2150 {
2151         capa_put(op_data->op_capa1);
2152         capa_put(op_data->op_capa2);
2153         OBD_FREE_PTR(op_data);
2154 }
2155
2156 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2157 {
2158         struct ll_sb_info *sbi;
2159
2160         LASSERT((seq != NULL) && (vfs != NULL));
2161         sbi = ll_s2sbi(vfs->mnt_sb);
2162
2163         if (sbi->ll_flags & LL_SBI_NOLCK)
2164                 seq_puts(seq, ",nolock");
2165
2166         if (sbi->ll_flags & LL_SBI_FLOCK)
2167                 seq_puts(seq, ",flock");
2168
2169         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2170                 seq_puts(seq, ",localflock");
2171
2172         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2173                 seq_puts(seq, ",user_xattr");
2174
2175         if (sbi->ll_flags & LL_SBI_ACL)
2176                 seq_puts(seq, ",acl");
2177
2178         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2179                 seq_puts(seq, ",lazystatfs");
2180
2181         RETURN(0);
2182 }