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         .d_compare = ll_dcompare,
 158         .d_revalidate = ll_revalidate_nd,
 159 };
 160
 161 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 162 {
 163         struct inode *root = 0;
 164         struct ll_sb_info *sbi = ll_s2sbi(sb);
 165         struct obd_device *obd;
 166         struct obd_capa *oc = NULL;
 167         struct obd_statfs osfs;
 168         struct ptlrpc_request *request = NULL;
 169         struct obd_connect_data *data = NULL;
 170         struct obd_uuid *uuid;
 171         struct lustre_md lmd;
 172         obd_valid valid;
 173         int size, err, checksum;
 174         ENTRY;
 175
 176         obd = class_name2obd(md);
 177         if (!obd) {
 178                 CERROR("MD %s: not setup or attached\n", md);
 179                 RETURN(-EINVAL);
 180         }
 181
 182         OBD_ALLOC_PTR(data);
 183         if (data == NULL)
 184                 RETURN(-ENOMEM);
 185
 186         if (proc_lustre_fs_root) {
 187                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 188                                                   dt, md);
 189                 if (err < 0)
 190                         CERROR("could not register mount in /proc/fs/lustre\n");
 191         }
 192
 193         /* indicate the features supported by this client */
 194         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 195                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 196                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 197                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 198                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 199                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 200                                   OBD_CONNECT_VBR      | OBD_CONNECT_SOM;
 201
 202 #ifdef HAVE_LRU_RESIZE_SUPPORT
 203         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 204                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 205 #endif
 206 #ifdef CONFIG_FS_POSIX_ACL
 207         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 208 #endif
 209         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 210         data->ocd_version = LUSTRE_VERSION_CODE;
 211
 212         if (sb->s_flags & MS_RDONLY)
 213                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 214         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 215                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 216
 217 #ifdef HAVE_MS_FLOCK_LOCK
 218         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 219         sb->s_flags |= MS_FLOCK_LOCK;
 220 #endif
 221
 222         if (sbi->ll_flags & LL_SBI_FLOCK)
 223                 sbi->ll_fop = &ll_file_operations_flock;
 224         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 225                 sbi->ll_fop = &ll_file_operations;
 226         else
 227                 sbi->ll_fop = &ll_file_operations_noflock;
 228
 229         /* real client */
 230         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 231         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 232                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 233
 234         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 235         if (err == -EBUSY) {
 236                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 237                                    "recovery, of which this client is not a "
 238                                    "part. Please wait for recovery to complete,"
 239                                    " abort, or time out.\n", md);
 240                 GOTO(out, err);
 241         } else if (err) {
 242                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 243                 GOTO(out, err);
 244         }
 245
 246         err = obd_fid_init(sbi->ll_md_exp);
 247         if (err) {
 248                 CERROR("Can't init metadata layer FID infrastructure, "
 249                        "rc %d\n", err);
 250                 GOTO(out_md, err);
 251         }
 252
 253         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 254         if (err)
 255                 GOTO(out_md_fid, err);
 256
 257         size = sizeof(*data);
 258         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 259                            KEY_CONN_DATA,  &size, data, NULL);
 260         if (err) {
 261                 CERROR("Get connect data failed: %d \n", err);
 262                 GOTO(out_md, err);
 263         }
 264
 265         LASSERT(osfs.os_bsize);
 266         sb->s_blocksize = osfs.os_bsize;
 267         sb->s_blocksize_bits = log2(osfs.os_bsize);
 268         sb->s_magic = LL_SUPER_MAGIC;
 269
 270         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 271          *         retval = in_file->f_op->sendfile(...);
 272          *         if (*ppos > max)
 273          *                 retval = -EOVERFLOW;
 274          *
 275          * it will check if *ppos is greater than max. However, max equals to
 276          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 277          * has been defined as a signed long long ineger in linux kernel. */
 278 #if BITS_PER_LONG == 64
 279         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 280 #else
 281         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 282 #endif
 283         sbi->ll_namelen = osfs.os_namelen;
 284         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 285
 286         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 287             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 288                 LCONSOLE_INFO("Disabling user_xattr feature because "
 289                               "it is not supported on the server\n");
 290                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 291         }
 292
 293         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 294 #ifdef MS_POSIXACL
 295                 sb->s_flags |= MS_POSIXACL;
 296 #endif
 297                 sbi->ll_flags |= LL_SBI_ACL;
 298         } else {
 299                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 300 #ifdef MS_POSIXACL
 301                 sb->s_flags &= ~MS_POSIXACL;
 302 #endif
 303                 sbi->ll_flags &= ~LL_SBI_ACL;
 304         }
 305
 306         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 307                 sbi->ll_flags |= LL_SBI_JOIN;
 308
 309         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 310                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 311                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 312                         LCONSOLE_INFO("client is set as remote by default.\n");
 313                 }
 314         } else {
 315                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 316                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 317                         LCONSOLE_INFO("client claims to be remote, but server "
 318                                       "rejected, forced to be local.\n");
 319                 }
 320         }
 321
 322         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 323                 LCONSOLE_INFO("client enabled MDS capability!\n");
 324                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 325         }
 326
 327         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 328                 LCONSOLE_INFO("client enabled OSS capability!\n");
 329                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 330         }
 331
 332         obd = class_name2obd(dt);
 333         if (!obd) {
 334                 CERROR("DT %s: not setup or attached\n", dt);
 335                 GOTO(out_md_fid, err = -ENODEV);
 336         }
 337
 338         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 339                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 340                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 341                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 342                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 343                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 344                                   OBD_CONNECT_SOM;
 345
 346         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 347                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 348                  * disabled by default, because it can still be enabled on the
 349                  * fly via /proc. As a consequence, we still need to come to an
 350                  * agreement on the supported algorithms at connect time */
 351                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 352
 353                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 354                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 355                 else
 356                         /* send the list of supported checksum types */
 357                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 358         }
 359
 360 #ifdef HAVE_LRU_RESIZE_SUPPORT
 361         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 362 #endif
 363         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 364                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 365
 366         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 367                "ocd_grant: %d\n", data->ocd_connect_flags,
 368                data->ocd_version, data->ocd_grant);
 369
 370         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 371         obd->obd_upcall.onu_upcall = cl_ocd_update;
 372         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 373
 374         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 375         if (err == -EBUSY) {
 376                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 377                                    "recovery, of which this client is not a "
 378                                    "part.  Please wait for recovery to "
 379                                    "complete, abort, or time out.\n", dt);
 380                 GOTO(out_md_fid, err);
 381         } else if (err) {
 382                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 383                 GOTO(out_md_fid, err);
 384         }
 385
 386         err = obd_fid_init(sbi->ll_dt_exp);
 387         if (err) {
 388                 CERROR("Can't init data layer FID infrastructure, "
 389                        "rc %d\n", err);
 390                 GOTO(out_dt, err);
 391         }
 392
 393         spin_lock(&sbi->ll_lco.lco_lock);
 394         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 395         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 396         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 397         spin_unlock(&sbi->ll_lco.lco_lock);
 398
 399         fid_zero(&sbi->ll_root_fid);
 400         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 401         if (err) {
 402                 CERROR("cannot mds_connect: rc = %d\n", err);
 403                 GOTO(out_lock_cn_cb, err);
 404         }
 405         if (!fid_is_sane(&sbi->ll_root_fid)) {
 406                 CERROR("Invalid root fid during mount\n");
 407                 GOTO(out_lock_cn_cb, err = -EINVAL);
 408         }
 409         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 410
 411         sb->s_op = &lustre_super_operations;
 412         sb->s_export_op = &lustre_export_operations;
 413
 414         /* make root inode
 415          * XXX: move this to after cbd setup? */
 416         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 417         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 418                 valid |= OBD_MD_FLRMTPERM;
 419         else if (sbi->ll_flags & LL_SBI_ACL)
 420                 valid |= OBD_MD_FLACL;
 421
 422         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 423                          &request);
 424         if (oc)
 425                 capa_put(oc);
 426         if (err) {
 427                 CERROR("md_getattr failed for root: rc = %d\n", err);
 428                 GOTO(out_lock_cn_cb, err);
 429         }
 430         memset(&lmd, 0, sizeof(lmd));
 431         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 432                                sbi->ll_md_exp, &lmd);
 433         if (err) {
 434                 CERROR("failed to understand root inode md: rc = %d\n", err);
 435                 ptlrpc_req_finished (request);
 436                 GOTO(out_lock_cn_cb, err);
 437         }
 438
 439         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 440         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 441         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 442         ptlrpc_req_finished(request);
 443
 444         if (root == NULL || IS_ERR(root)) {
 445                 if (lmd.lsm)
 446                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 447 #ifdef CONFIG_FS_POSIX_ACL
 448                 if (lmd.posix_acl) {
 449                         posix_acl_release(lmd.posix_acl);
 450                         lmd.posix_acl = NULL;
 451                 }
 452 #endif
 453                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 454                 root = NULL;
 455                 CERROR("lustre_lite: bad iget4 for root\n");
 456                 GOTO(out_root, err);
 457         }
 458
 459         err = ll_close_thread_start(&sbi->ll_lcq);
 460         if (err) {
 461                 CERROR("cannot start close thread: rc %d\n", err);
 462                 GOTO(out_root, err);
 463         }
 464
 465 #ifdef CONFIG_FS_POSIX_ACL
 466         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 467                 rct_init(&sbi->ll_rct);
 468                 et_init(&sbi->ll_et);
 469         }
 470 #endif
 471
 472         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 473         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 474                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 475                                  NULL);
 476         cl_sb_init(sb);
 477
 478         sb->s_root = d_alloc_root(root);
 479         if (data != NULL)
 480                 OBD_FREE(data, sizeof(*data));
 481
 482         sb->s_root->d_op = &ll_d_root_ops;
 483
 484         sbi->ll_sdev_orig = sb->s_dev;
 485
 486         /* We set sb->s_dev equal on all lustre clients in order to support
 487          * NFS export clustering.  NFSD requires that the FSID be the same
 488          * on all clients. */
 489         /* s_dev is also used in lt_compare() to compare two fs, but that is
 490          * only a node-local comparison. */
 491         uuid = obd_get_uuid(sbi->ll_md_exp);
 492         if (uuid != NULL)
 493                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 494
 495         RETURN(err);
 496 out_root:
 497         if (root)
 498                 iput(root);
 499 out_lock_cn_cb:
 500         obd_fid_fini(sbi->ll_dt_exp);
 501 out_dt:
 502         obd_disconnect(sbi->ll_dt_exp);
 503         sbi->ll_dt_exp = NULL;
 504 out_md_fid:
 505         obd_fid_fini(sbi->ll_md_exp);
 506 out_md:
 507         obd_disconnect(sbi->ll_md_exp);
 508         sbi->ll_md_exp = NULL;
 509 out:
 510         if (data != NULL)
 511                 OBD_FREE_PTR(data);
 512         lprocfs_unregister_mountpoint(sbi);
 513         return err;
 514 }
 515
 516 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 517 {
 518         int size, rc;
 519
 520         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 521         size = sizeof(int);
 522         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 523                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 524         if (rc)
 525                 CERROR("Get max mdsize error rc %d \n", rc);
 526
 527         RETURN(rc);
 528 }
 529
 530 void ll_dump_inode(struct inode *inode)
 531 {
 532         struct list_head *tmp;
 533         int dentry_count = 0;
 534
 535         LASSERT(inode != NULL);
 536
 537         list_for_each(tmp, &inode->i_dentry)
 538                 dentry_count++;
 539
 540         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 541                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 542                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 543 }
 544
 545 void lustre_dump_dentry(struct dentry *dentry, int recur)
 546 {
 547         struct list_head *tmp;
 548         int subdirs = 0;
 549
 550         LASSERT(dentry != NULL);
 551
 552         list_for_each(tmp, &dentry->d_subdirs)
 553                 subdirs++;
 554
 555         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 556                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 557                dentry->d_name.len, dentry->d_name.name,
 558                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 559                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 560                dentry->d_flags, dentry->d_fsdata, subdirs);
 561         if (dentry->d_inode != NULL)
 562                 ll_dump_inode(dentry->d_inode);
 563
 564         if (recur == 0)
 565                 return;
 566
 567         list_for_each(tmp, &dentry->d_subdirs) {
 568                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 569                 lustre_dump_dentry(d, recur - 1);
 570         }
 571 }
 572
 573 #ifdef HAVE_EXPORT___IGET
 574 static void prune_dir_dentries(struct inode *inode)
 575 {
 576         struct dentry *dentry, *prev = NULL;
 577
 578         /* due to lustre specific logic, a directory
 579          * can have few dentries - a bug from VFS POV */
 580 restart:
 581         spin_lock(&dcache_lock);
 582         if (!list_empty(&inode->i_dentry)) {
 583                 dentry = list_entry(inode->i_dentry.prev,
 584                                     struct dentry, d_alias);
 585                 /* in order to prevent infinite loops we
 586                  * break if previous dentry is busy */
 587                 if (dentry != prev) {
 588                         prev = dentry;
 589                         dget_locked(dentry);
 590                         spin_unlock(&dcache_lock);
 591
 592                         /* try to kill all child dentries */
 593                         lock_dentry(dentry);
 594                         shrink_dcache_parent(dentry);
 595                         unlock_dentry(dentry);
 596                         dput(dentry);
 597
 598                         /* now try to get rid of current dentry */
 599                         d_prune_aliases(inode);
 600                         goto restart;
 601                 }
 602         }
 603         spin_unlock(&dcache_lock);
 604 }
 605
 606 static void prune_deathrow_one(struct ll_inode_info *lli)
 607 {
 608         struct inode *inode = ll_info2i(lli);
 609
 610         /* first, try to drop any dentries - they hold a ref on the inode */
 611         if (S_ISDIR(inode->i_mode))
 612                 prune_dir_dentries(inode);
 613         else
 614                 d_prune_aliases(inode);
 615
 616
 617         /* if somebody still uses it, leave it */
 618         LASSERT(atomic_read(&inode->i_count) > 0);
 619         if (atomic_read(&inode->i_count) > 1)
 620                 goto out;
 621
 622         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 623                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 624
 625         /* seems nobody uses it anymore */
 626         inode->i_nlink = 0;
 627
 628 out:
 629         iput(inode);
 630         return;
 631 }
 632
 633 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 634 {
 635         struct ll_inode_info *lli;
 636         int empty;
 637
 638         do {
 639                 if (need_resched() && try)
 640                         break;
 641
 642                 if (try) {
 643                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 644                                 break;
 645                 } else {
 646                         spin_lock(&sbi->ll_deathrow_lock);
 647                 }
 648
 649                 empty = 1;
 650                 lli = NULL;
 651                 if (!list_empty(&sbi->ll_deathrow)) {
 652                         lli = list_entry(sbi->ll_deathrow.next,
 653                                          struct ll_inode_info,
 654                                          lli_dead_list);
 655                         list_del_init(&lli->lli_dead_list);
 656                         if (!list_empty(&sbi->ll_deathrow))
 657                                 empty = 0;
 658                 }
 659                 spin_unlock(&sbi->ll_deathrow_lock);
 660
 661                 if (lli)
 662                         prune_deathrow_one(lli);
 663
 664         } while (empty == 0);
 665 }
 666 #else /* !HAVE_EXPORT___IGET */
 667 #define prune_deathrow(sbi, try) do {} while (0)
 668 #endif /* HAVE_EXPORT___IGET */
 669
 670 void client_common_put_super(struct super_block *sb)
 671 {
 672         struct ll_sb_info *sbi = ll_s2sbi(sb);
 673         ENTRY;
 674
 675 #ifdef CONFIG_FS_POSIX_ACL
 676         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 677                 et_fini(&sbi->ll_et);
 678                 rct_fini(&sbi->ll_rct);
 679         }
 680 #endif
 681
 682         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 683
 684         ll_close_thread_shutdown(sbi->ll_lcq);
 685
 686         cl_sb_fini(sb);
 687
 688         /* destroy inodes in deathrow */
 689         prune_deathrow(sbi, 0);
 690
 691         list_del(&sbi->ll_conn_chain);
 692
 693         obd_fid_fini(sbi->ll_dt_exp);
 694         obd_disconnect(sbi->ll_dt_exp);
 695         sbi->ll_dt_exp = NULL;
 696
 697         lprocfs_unregister_mountpoint(sbi);
 698
 699         obd_fid_fini(sbi->ll_md_exp);
 700         obd_disconnect(sbi->ll_md_exp);
 701         sbi->ll_md_exp = NULL;
 702
 703         EXIT;
 704 }
 705
 706 void ll_kill_super(struct super_block *sb)
 707 {
 708         struct ll_sb_info *sbi;
 709
 710         ENTRY;
 711
 712         /* not init sb ?*/
 713         if (!(sb->s_flags & MS_ACTIVE))
 714                 return;
 715
 716         sbi = ll_s2sbi(sb);
 717         /* we need restore s_dev from changed for clustred NFS before put_super
 718          * because new kernels have cached s_dev and change sb->s_dev in
 719          * put_super not affected real removing devices */
 720         if (sbi)
 721                 sb->s_dev = sbi->ll_sdev_orig;
 722         EXIT;
 723 }
 724
 725 char *ll_read_opt(const char *opt, char *data)
 726 {
 727         char *value;
 728         char *retval;
 729         ENTRY;
 730
 731         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 732         if (strncmp(opt, data, strlen(opt)))
 733                 RETURN(NULL);
 734         if ((value = strchr(data, '=')) == NULL)
 735                 RETURN(NULL);
 736
 737         value++;
 738         OBD_ALLOC(retval, strlen(value) + 1);
 739         if (!retval) {
 740                 CERROR("out of memory!\n");
 741                 RETURN(NULL);
 742         }
 743
 744         memcpy(retval, value, strlen(value)+1);
 745         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 746         RETURN(retval);
 747 }
 748
 749 static inline int ll_set_opt(const char *opt, char *data, int fl)
 750 {
 751         if (strncmp(opt, data, strlen(opt)) != 0)
 752                 return(0);
 753         else
 754                 return(fl);
 755 }
 756
 757 /* non-client-specific mount options are parsed in lmd_parse */
 758 static int ll_options(char *options, int *flags)
 759 {
 760         int tmp;
 761         char *s1 = options, *s2;
 762         ENTRY;
 763
 764         if (!options)
 765                 RETURN(0);
 766
 767         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 768
 769         while (*s1) {
 770                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 771                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 787                 if (tmp) {
 788                         *flags &= ~tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 797                 if (tmp) {
 798                         *flags &= ~tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         /* Ignore deprecated mount option.  The client will
 804                          * always try to mount with ACL support, whether this
 805                          * is used depends on whether server supports it. */
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 809                 if (tmp) {
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817
 818                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848
 849                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 850                                    s1);
 851                 RETURN(-EINVAL);
 852
 853 next:
 854                 /* Find next opt */
 855                 s2 = strchr(s1, ',');
 856                 if (s2 == NULL)
 857                         break;
 858                 s1 = s2 + 1;
 859         }
 860         RETURN(0);
 861 }
 862
 863 void ll_lli_init(struct ll_inode_info *lli)
 864 {
 865         lli->lli_inode_magic = LLI_INODE_MAGIC;
 866         sema_init(&lli->lli_size_sem, 1);
 867         sema_init(&lli->lli_write_sem, 1);
 868         lli->lli_flags = 0;
 869         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 870         spin_lock_init(&lli->lli_lock);
 871         INIT_LIST_HEAD(&lli->lli_close_list);
 872         lli->lli_inode_magic = LLI_INODE_MAGIC;
 873         sema_init(&lli->lli_och_sem, 1);
 874         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 875         lli->lli_mds_exec_och = NULL;
 876         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 877         lli->lli_open_fd_exec_count = 0;
 878         INIT_LIST_HEAD(&lli->lli_dead_list);
 879         lli->lli_remote_perms = NULL;
 880         lli->lli_rmtperm_utime = 0;
 881         sema_init(&lli->lli_rmtperm_sem, 1);
 882         INIT_LIST_HEAD(&lli->lli_oss_capas);
 883 }
 884
 885 int ll_fill_super(struct super_block *sb)
 886 {
 887         struct lustre_profile *lprof;
 888         struct lustre_sb_info *lsi = s2lsi(sb);
 889         struct ll_sb_info *sbi;
 890         char  *dt = NULL, *md = NULL;
 891         char  *profilenm = get_profile_name(sb);
 892         struct config_llog_instance cfg = {0, };
 893         char   ll_instance[sizeof(sb) * 2 + 1];
 894         int    err;
 895         ENTRY;
 896
 897         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 898
 899         cfs_module_get();
 900
 901         /* client additional sb info */
 902         lsi->lsi_llsbi = sbi = ll_init_sbi();
 903         if (!sbi) {
 904                 cfs_module_put();
 905                 RETURN(-ENOMEM);
 906         }
 907
 908         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 909         if (err)
 910                 GOTO(out_free, err);
 911
 912         /* Generate a string unique to this super, in case some joker tries
 913            to mount the same fs at two mount points.
 914            Use the address of the super itself.*/
 915         sprintf(ll_instance, "%p", sb);
 916         cfg.cfg_instance = ll_instance;
 917         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 918
 919         /* set up client obds */
 920         err = lustre_process_log(sb, profilenm, &cfg);
 921         if (err < 0) {
 922                 CERROR("Unable to process log: %d\n", err);
 923                 GOTO(out_free, err);
 924         }
 925
 926         lprof = class_get_profile(profilenm);
 927         if (lprof == NULL) {
 928                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 929                                    " read from the MGS.  Does that filesystem "
 930                                    "exist?\n", profilenm);
 931                 GOTO(out_free, err = -EINVAL);
 932         }
 933         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 934                lprof->lp_md, lprof->lp_dt);
 935
 936         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 937                   strlen(ll_instance) + 2);
 938         if (!dt)
 939                 GOTO(out_free, err = -ENOMEM);
 940         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 941
 942         OBD_ALLOC(md, strlen(lprof->lp_md) +
 943                   strlen(ll_instance) + 2);
 944         if (!md)
 945                 GOTO(out_free, err = -ENOMEM);
 946         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 947
 948         /* connections, registrations, sb setup */
 949         err = client_common_fill_super(sb, md, dt);
 950
 951 out_free:
 952         if (md)
 953                 OBD_FREE(md, strlen(md) + 1);
 954         if (dt)
 955                 OBD_FREE(dt, strlen(dt) + 1);
 956         if (err)
 957                 ll_put_super(sb);
 958         else
 959                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 960
 961         RETURN(err);
 962 } /* ll_fill_super */
 963
 964
 965 void lu_context_keys_dump(void);
 966
 967 void ll_put_super(struct super_block *sb)
 968 {
 969         struct config_llog_instance cfg;
 970         char   ll_instance[sizeof(sb) * 2 + 1];
 971         struct obd_device *obd;
 972         struct lustre_sb_info *lsi = s2lsi(sb);
 973         struct ll_sb_info *sbi = ll_s2sbi(sb);
 974         char *profilenm = get_profile_name(sb);
 975         int force = 1, next;
 976         ENTRY;
 977
 978         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 979
 980         ll_print_capa_stat(sbi);
 981
 982         sprintf(ll_instance, "%p", sb);
 983         cfg.cfg_instance = ll_instance;
 984         lustre_end_log(sb, NULL, &cfg);
 985
 986         if (sbi->ll_md_exp) {
 987                 obd = class_exp2obd(sbi->ll_md_exp);
 988                 if (obd)
 989                         force = obd->obd_force;
 990         }
 991
 992         /* We need to set force before the lov_disconnect in
 993            lustre_common_put_super, since l_d cleans up osc's as well. */
 994         if (force) {
 995                 next = 0;
 996                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 997                                                      &next)) != NULL) {
 998                         obd->obd_force = force;
 999                 }
1000         }
1001
1002         if (sbi->ll_lcq) {
1003                 /* Only if client_common_fill_super succeeded */
1004                 client_common_put_super(sb);
1005         }
1006
1007         next = 0;
1008         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1009                 class_manual_cleanup(obd);
1010         }
1011
1012         if (profilenm)
1013                 class_del_profile(profilenm);
1014
1015         ll_free_sbi(sb);
1016         lsi->lsi_llsbi = NULL;
1017
1018         lustre_common_put_super(sb);
1019
1020         cl_env_cache_purge(~0);
1021
1022         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1023
1024         cfs_module_put();
1025
1026         EXIT;
1027 } /* client_put_super */
1028
1029 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1030 {
1031         struct inode *inode = NULL;
1032         /* NOTE: we depend on atomic igrab() -bzzz */
1033         lock_res_and_lock(lock);
1034         if (lock->l_ast_data) {
1035                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1036                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1037                         inode = igrab(lock->l_ast_data);
1038                 } else {
1039                         inode = lock->l_ast_data;
1040                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1041                                                 D_INFO : D_WARNING,
1042                                         lock, __FILE__, __func__, __LINE__,
1043                                         "l_ast_data %p is bogus: magic %08x",
1044                                         lock->l_ast_data, lli->lli_inode_magic);
1045                         inode = NULL;
1046                 }
1047         }
1048         unlock_res_and_lock(lock);
1049         return inode;
1050 }
1051
1052 static int null_if_equal(struct ldlm_lock *lock, void *data)
1053 {
1054         if (data == lock->l_ast_data) {
1055                 lock->l_ast_data = NULL;
1056
1057                 if (lock->l_req_mode != lock->l_granted_mode)
1058                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1059         }
1060
1061         return LDLM_ITER_CONTINUE;
1062 }
1063
1064 void ll_clear_inode(struct inode *inode)
1065 {
1066         struct ll_inode_info *lli = ll_i2info(inode);
1067         struct ll_sb_info *sbi = ll_i2sbi(inode);
1068         ENTRY;
1069
1070         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1071                inode->i_generation, inode);
1072
1073         if (S_ISDIR(inode->i_mode)) {
1074                 /* these should have been cleared in ll_file_release */
1075                 LASSERT(lli->lli_sai == NULL);
1076                 LASSERT(lli->lli_opendir_key == NULL);
1077                 LASSERT(lli->lli_opendir_pid == 0);
1078         }
1079
1080         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1081         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1082                          null_if_equal, inode);
1083
1084         LASSERT(!lli->lli_open_fd_write_count);
1085         LASSERT(!lli->lli_open_fd_read_count);
1086         LASSERT(!lli->lli_open_fd_exec_count);
1087
1088         if (lli->lli_mds_write_och)
1089                 ll_md_real_close(inode, FMODE_WRITE);
1090         if (lli->lli_mds_exec_och)
1091                 ll_md_real_close(inode, FMODE_EXEC);
1092         if (lli->lli_mds_read_och)
1093                 ll_md_real_close(inode, FMODE_READ);
1094
1095         if (lli->lli_symlink_name) {
1096                 OBD_FREE(lli->lli_symlink_name,
1097                          strlen(lli->lli_symlink_name) + 1);
1098                 lli->lli_symlink_name = NULL;
1099         }
1100
1101         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1102                 LASSERT(lli->lli_posix_acl == NULL);
1103                 if (lli->lli_remote_perms) {
1104                         free_rmtperm_hash(lli->lli_remote_perms);
1105                         lli->lli_remote_perms = NULL;
1106                 }
1107         }
1108 #ifdef CONFIG_FS_POSIX_ACL
1109         else if (lli->lli_posix_acl) {
1110                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1111                 LASSERT(lli->lli_remote_perms == NULL);
1112                 posix_acl_release(lli->lli_posix_acl);
1113                 lli->lli_posix_acl = NULL;
1114         }
1115 #endif
1116         lli->lli_inode_magic = LLI_INODE_DEAD;
1117
1118 #ifdef HAVE_EXPORT___IGET
1119         spin_lock(&sbi->ll_deathrow_lock);
1120         list_del_init(&lli->lli_dead_list);
1121         spin_unlock(&sbi->ll_deathrow_lock);
1122 #endif
1123         ll_clear_inode_capas(inode);
1124         /*
1125          * XXX This has to be done before lsm is freed below, because
1126          * cl_object still uses inode lsm.
1127          */
1128         cl_inode_fini(inode);
1129
1130         if (lli->lli_smd) {
1131                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1132                 lli->lli_smd = NULL;
1133         }
1134
1135
1136         EXIT;
1137 }
1138
1139 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1140                   struct md_open_data **mod)
1141 {
1142         struct lustre_md md;
1143         struct ll_sb_info *sbi = ll_i2sbi(inode);
1144         struct ptlrpc_request *request = NULL;
1145         int rc;
1146         ENTRY;
1147
1148         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1149                                      LUSTRE_OPC_ANY, NULL);
1150         if (IS_ERR(op_data))
1151                 RETURN(PTR_ERR(op_data));
1152
1153         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1154                         &request, mod);
1155         if (rc) {
1156                 ptlrpc_req_finished(request);
1157                 if (rc == -ENOENT) {
1158                         inode->i_nlink = 0;
1159                         /* Unlinked special device node? Or just a race?
1160                          * Pretend we done everything. */
1161                         if (!S_ISREG(inode->i_mode) &&
1162                             !S_ISDIR(inode->i_mode))
1163                                 rc = inode_setattr(inode, &op_data->op_attr);
1164                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1165                         CERROR("md_setattr fails: rc = %d\n", rc);
1166                 }
1167                 RETURN(rc);
1168         }
1169
1170         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1171                               sbi->ll_md_exp, &md);
1172         if (rc) {
1173                 ptlrpc_req_finished(request);
1174                 RETURN(rc);
1175         }
1176
1177         /* We call inode_setattr to adjust timestamps.
1178          * If there is at least some data in file, we cleared ATTR_SIZE
1179          * above to avoid invoking vmtruncate, otherwise it is important
1180          * to call vmtruncate in inode_setattr to update inode->i_size
1181          * (bug 6196) */
1182         rc = inode_setattr(inode, &op_data->op_attr);
1183
1184         /* Extract epoch data if obtained. */
1185         op_data->op_handle = md.body->handle;
1186         op_data->op_ioepoch = md.body->ioepoch;
1187
1188         ll_update_inode(inode, &md);
1189         ptlrpc_req_finished(request);
1190
1191         RETURN(rc);
1192 }
1193
1194 /* Close IO epoch and send Size-on-MDS attribute update. */
1195 static int ll_setattr_done_writing(struct inode *inode,
1196                                    struct md_op_data *op_data,
1197                                    struct md_open_data *mod)
1198 {
1199         struct ll_inode_info *lli = ll_i2info(inode);
1200         int rc = 0;
1201         ENTRY;
1202
1203         LASSERT(op_data != NULL);
1204         if (!S_ISREG(inode->i_mode))
1205                 RETURN(0);
1206
1207         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1208                op_data->op_ioepoch, PFID(&lli->lli_fid));
1209
1210         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1211         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1212         if (rc == -EAGAIN) {
1213                 /* MDS has instructed us to obtain Size-on-MDS attribute
1214                  * from OSTs and send setattr to back to MDS. */
1215                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1216                                          op_data->op_ioepoch);
1217         } else if (rc) {
1218                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1219                        inode->i_ino, rc);
1220         }
1221         RETURN(rc);
1222 }
1223
1224 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1225 {
1226         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1227         int rc;
1228
1229         rc = cl_setattr_do_truncate(inode, size, capa);
1230         ll_truncate_free_capa(capa);
1231         return rc;
1232 }
1233
1234 static int ll_setattr_ost(struct inode *inode)
1235 {
1236         struct obd_capa *capa = ll_mdscapa_get(inode);
1237         int rc;
1238
1239         rc = cl_setattr_ost(inode, capa);
1240         capa_put(capa);
1241
1242         return rc;
1243 }
1244
1245 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1246  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1247  * keep these values until such a time that objects are allocated for it.
1248  * We do the MDS operations first, as it is checking permissions for us.
1249  * We don't to the MDS RPC if there is nothing that we want to store there,
1250  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1251  * going to do an RPC anyways.
1252  *
1253  * If we are doing a truncate, we will send the mtime and ctime updates
1254  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1255  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1256  * at the same time.
1257  */
1258 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1259 {
1260         struct ll_inode_info *lli = ll_i2info(inode);
1261         struct lov_stripe_md *lsm = lli->lli_smd;
1262         struct md_op_data *op_data = NULL;
1263         struct md_open_data *mod = NULL;
1264         int ia_valid = attr->ia_valid;
1265         int rc = 0, rc1 = 0;
1266         ENTRY;
1267
1268         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1269                attr->ia_valid);
1270         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1271
1272         if (ia_valid & ATTR_SIZE) {
1273                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1274                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1275                                attr->ia_size, ll_file_maxbytes(inode));
1276                         RETURN(-EFBIG);
1277                 }
1278
1279                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1280         }
1281
1282         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1283         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1284                 if (current->fsuid != inode->i_uid &&
1285                     !cfs_capable(CFS_CAP_FOWNER))
1286                         RETURN(-EPERM);
1287         }
1288
1289         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1290         if (attr->ia_valid & ATTR_CTIME) {
1291                 attr->ia_ctime = CURRENT_TIME;
1292                 attr->ia_valid |= ATTR_CTIME_SET;
1293         }
1294         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1295                 attr->ia_atime = CURRENT_TIME;
1296                 attr->ia_valid |= ATTR_ATIME_SET;
1297         }
1298         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1299                 attr->ia_mtime = CURRENT_TIME;
1300                 attr->ia_valid |= ATTR_MTIME_SET;
1301         }
1302         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1303                 /* To avoid stale mtime on mds, obtain it from ost and send
1304                    to mds. */
1305                 rc = cl_glimpse_size(inode);
1306                 if (rc)
1307                         RETURN(rc);
1308
1309                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1310                 attr->ia_mtime = inode->i_mtime;
1311         }
1312
1313         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1314                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1315                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1316                        cfs_time_current_sec());
1317
1318         /* NB: ATTR_SIZE will only be set after this point if the size
1319          * resides on the MDS, ie, this file has no objects. */
1320         if (lsm)
1321                 attr->ia_valid &= ~ATTR_SIZE;
1322
1323         /* We always do an MDS RPC, even if we're only changing the size;
1324          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1325
1326         OBD_ALLOC_PTR(op_data);
1327         if (op_data == NULL)
1328                 RETURN(-ENOMEM);
1329
1330         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1331
1332         /* Open epoch for truncate. */
1333         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1334             (ia_valid & ATTR_SIZE))
1335                 op_data->op_flags = MF_EPOCH_OPEN;
1336
1337         rc = ll_md_setattr(inode, op_data, &mod);
1338         if (rc)
1339                 GOTO(out, rc);
1340
1341         ll_ioepoch_open(lli, op_data->op_ioepoch);
1342         if (!lsm || !S_ISREG(inode->i_mode)) {
1343                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1344                 GOTO(out, rc = 0);
1345         }
1346
1347         if (ia_valid & ATTR_SIZE)
1348                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1349         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1350                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1351                        inode->i_ino, LTIME_S(attr->ia_mtime));
1352                 rc = ll_setattr_ost(inode);
1353         }
1354         EXIT;
1355 out:
1356         if (op_data) {
1357                 if (op_data->op_ioepoch)
1358                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1359                 ll_finish_md_op_data(op_data);
1360         }
1361         return rc ? rc : rc1;
1362 }
1363
1364 int ll_setattr(struct dentry *de, struct iattr *attr)
1365 {
1366         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1367             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1368                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1369
1370         if ((de->d_inode->i_mode & S_ISUID) &&
1371             !(attr->ia_mode & S_ISUID) &&
1372             !(attr->ia_valid & ATTR_KILL_SUID))
1373                 attr->ia_valid |= ATTR_KILL_SUID;
1374
1375         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1376             !(attr->ia_mode & S_ISGID) &&
1377             !(attr->ia_valid & ATTR_KILL_SGID))
1378                 attr->ia_valid |= ATTR_KILL_SGID;
1379
1380         return ll_setattr_raw(de->d_inode, attr);
1381 }
1382
1383 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1384                        __u64 max_age, __u32 flags)
1385 {
1386         struct ll_sb_info *sbi = ll_s2sbi(sb);
1387         struct obd_statfs obd_osfs;
1388         int rc;
1389         ENTRY;
1390
1391         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1392         if (rc) {
1393                 CERROR("md_statfs fails: rc = %d\n", rc);
1394                 RETURN(rc);
1395         }
1396
1397         osfs->os_type = sb->s_magic;
1398
1399         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1400                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1401
1402         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1403                 flags |= OBD_STATFS_NODELAY;
1404
1405         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1406                               &obd_osfs, max_age, flags);
1407         if (rc) {
1408                 CERROR("obd_statfs fails: rc = %d\n", rc);
1409                 RETURN(rc);
1410         }
1411
1412         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1413                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1414                obd_osfs.os_files);
1415
1416         osfs->os_bsize = obd_osfs.os_bsize;
1417         osfs->os_blocks = obd_osfs.os_blocks;
1418         osfs->os_bfree = obd_osfs.os_bfree;
1419         osfs->os_bavail = obd_osfs.os_bavail;
1420
1421         /* If we don't have as many objects free on the OST as inodes
1422          * on the MDS, we reduce the total number of inodes to
1423          * compensate, so that the "inodes in use" number is correct.
1424          */
1425         if (obd_osfs.os_ffree < osfs->os_ffree) {
1426                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1427                         obd_osfs.os_ffree;
1428                 osfs->os_ffree = obd_osfs.os_ffree;
1429         }
1430
1431         RETURN(rc);
1432 }
1433 #ifndef HAVE_STATFS_DENTRY_PARAM
1434 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1435 {
1436 #else
1437 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1438 {
1439         struct super_block *sb = de->d_sb;
1440 #endif
1441         struct obd_statfs osfs;
1442         int rc;
1443
1444         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1445         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1446
1447         /* For now we will always get up-to-date statfs values, but in the
1448          * future we may allow some amount of caching on the client (e.g.
1449          * from QOS or lprocfs updates). */
1450         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1451         if (rc)
1452                 return rc;
1453
1454         statfs_unpack(sfs, &osfs);
1455
1456         /* We need to downshift for all 32-bit kernels, because we can't
1457          * tell if the kernel is being called via sys_statfs64() or not.
1458          * Stop before overflowing f_bsize - in which case it is better
1459          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1460         if (sizeof(long) < 8) {
1461                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1462                         sfs->f_bsize <<= 1;
1463
1464                         osfs.os_blocks >>= 1;
1465                         osfs.os_bfree >>= 1;
1466                         osfs.os_bavail >>= 1;
1467                 }
1468         }
1469
1470         sfs->f_blocks = osfs.os_blocks;
1471         sfs->f_bfree = osfs.os_bfree;
1472         sfs->f_bavail = osfs.os_bavail;
1473
1474         return 0;
1475 }
1476
1477 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1478 {
1479         struct ll_inode_info *lli;
1480         struct lov_stripe_md *lsm;
1481
1482         lli = ll_i2info(inode);
1483         LASSERT(lli->lli_size_sem_owner != current);
1484         down(&lli->lli_size_sem);
1485         LASSERT(lli->lli_size_sem_owner == NULL);
1486         lli->lli_size_sem_owner = current;
1487         lsm = lli->lli_smd;
1488         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1489                  lsm, lock_lsm);
1490         if (lock_lsm)
1491                 lov_stripe_lock(lsm);
1492 }
1493
1494 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1495 {
1496         struct ll_inode_info *lli;
1497         struct lov_stripe_md *lsm;
1498
1499         lli = ll_i2info(inode);
1500         lsm = lli->lli_smd;
1501         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1502                  lsm, unlock_lsm);
1503         if (unlock_lsm)
1504                 lov_stripe_unlock(lsm);
1505         LASSERT(lli->lli_size_sem_owner == current);
1506         lli->lli_size_sem_owner = NULL;
1507         up(&lli->lli_size_sem);
1508 }
1509
1510 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1511 {
1512         struct ll_inode_info *lli = ll_i2info(inode);
1513
1514         dump_lsm(D_INODE, lsm);
1515         dump_lsm(D_INODE, lli->lli_smd);
1516         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1517                  "lsm must be joined lsm %p\n", lsm);
1518         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1519         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1520                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1521         lli->lli_smd = lsm;
1522         lli->lli_maxbytes = lsm->lsm_maxbytes;
1523         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1524                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1525 }
1526
1527 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1528 {
1529         struct ll_inode_info *lli = ll_i2info(inode);
1530         struct mdt_body *body = md->body;
1531         struct lov_stripe_md *lsm = md->lsm;
1532         struct ll_sb_info *sbi = ll_i2sbi(inode);
1533
1534         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1535         if (lsm != NULL) {
1536                 if (lli->lli_smd == NULL) {
1537                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1538                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1539                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1540                                 dump_lsm(D_ERROR, lsm);
1541                                 LBUG();
1542                         }
1543                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1544                                lsm, inode->i_ino, inode->i_generation, inode);
1545                         cl_inode_init(inode, md);
1546                         /* ll_inode_size_lock() requires it is only
1547                          * called with lli_smd != NULL or lock_lsm == 0
1548                          *  or we can race between lock/unlock.
1549                          *  bug 9547 */
1550                         lli->lli_smd = lsm;
1551                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1552                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1553                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1554                 } else {
1555                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1556                              lli->lli_smd->lsm_stripe_count ==
1557                                         lsm->lsm_stripe_count) {
1558                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1559                                         CERROR("lsm mismatch for inode %ld\n",
1560                                                 inode->i_ino);
1561                                         CERROR("lli_smd:\n");
1562                                         dump_lsm(D_ERROR, lli->lli_smd);
1563                                         CERROR("lsm:\n");
1564                                         dump_lsm(D_ERROR, lsm);
1565                                         LBUG();
1566                                 }
1567                         } else {
1568                                 cl_inode_init(inode, md);
1569                                 ll_replace_lsm(inode, lsm);
1570                         }
1571                 }
1572                 if (lli->lli_smd != lsm)
1573                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1574         }
1575
1576         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1577                 if (body->valid & OBD_MD_FLRMTPERM)
1578                         ll_update_remote_perm(inode, md->remote_perm);
1579         }
1580 #ifdef CONFIG_FS_POSIX_ACL
1581         else if (body->valid & OBD_MD_FLACL) {
1582                 spin_lock(&lli->lli_lock);
1583                 if (lli->lli_posix_acl)
1584                         posix_acl_release(lli->lli_posix_acl);
1585                 lli->lli_posix_acl = md->posix_acl;
1586                 spin_unlock(&lli->lli_lock);
1587         }
1588 #endif
1589         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1590         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1591
1592         if (body->valid & OBD_MD_FLATIME &&
1593             body->atime > LTIME_S(inode->i_atime))
1594                 LTIME_S(inode->i_atime) = body->atime;
1595
1596         /* mtime is always updated with ctime, but can be set in past.
1597            As write and utime(2) may happen within 1 second, and utime's
1598            mtime has a priority over write's one, so take mtime from mds
1599            for the same ctimes. */
1600         if (body->valid & OBD_MD_FLCTIME &&
1601             body->ctime >= LTIME_S(inode->i_ctime)) {
1602                 LTIME_S(inode->i_ctime) = body->ctime;
1603                 if (body->valid & OBD_MD_FLMTIME) {
1604                         CDEBUG(D_INODE, "setting ino %lu mtime "
1605                                "from %lu to "LPU64"\n", inode->i_ino,
1606                                LTIME_S(inode->i_mtime), body->mtime);
1607                         LTIME_S(inode->i_mtime) = body->mtime;
1608                 }
1609         }
1610         if (body->valid & OBD_MD_FLMODE)
1611                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1612         if (body->valid & OBD_MD_FLTYPE)
1613                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1614         LASSERT(inode->i_mode != 0);
1615         if (S_ISREG(inode->i_mode)) {
1616                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1617         } else {
1618                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1619         }
1620 #ifdef HAVE_INODE_BLKSIZE
1621         inode->i_blksize = 1<<inode->i_blkbits;
1622 #endif
1623         if (body->valid & OBD_MD_FLUID)
1624                 inode->i_uid = body->uid;
1625         if (body->valid & OBD_MD_FLGID)
1626                 inode->i_gid = body->gid;
1627         if (body->valid & OBD_MD_FLFLAGS)
1628                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1629         if (body->valid & OBD_MD_FLNLINK)
1630                 inode->i_nlink = body->nlink;
1631         if (body->valid & OBD_MD_FLRDEV)
1632                 inode->i_rdev = old_decode_dev(body->rdev);
1633
1634         if (body->valid & OBD_MD_FLID) {
1635                 /* FID shouldn't be changed! */
1636                 if (fid_is_sane(&lli->lli_fid)) {
1637                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1638                                  "Trying to change FID "DFID
1639                                  " to the "DFID", inode %lu/%u(%p)\n",
1640                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1641                                  inode->i_ino, inode->i_generation, inode);
1642                 } else
1643                         lli->lli_fid = body->fid1;
1644         }
1645
1646         LASSERT(fid_seq(&lli->lli_fid) != 0);
1647
1648         if (body->valid & OBD_MD_FLSIZE) {
1649                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1650                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1651                         struct lustre_handle lockh;
1652                         ldlm_mode_t mode;
1653
1654                         /* As it is possible a blocking ast has been processed
1655                          * by this time, we need to check there is an UPDATE
1656                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1657                          * it. */
1658                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1659                                                &lockh);
1660                         if (mode) {
1661                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1662                                                       LLIF_EPOCH_PENDING |
1663                                                       LLIF_SOM_DIRTY)) {
1664                                         CERROR("ino %lu flags %lu still has "
1665                                                "size authority! do not trust "
1666                                                "the size got from MDS\n",
1667                                                inode->i_ino, lli->lli_flags);
1668                                 } else {
1669                                         /* Use old size assignment to avoid
1670                                          * deadlock bz14138 & bz14326 */
1671                                         inode->i_size = body->size;
1672                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1673                                 }
1674                                 ldlm_lock_decref(&lockh, mode);
1675                         }
1676                 } else {
1677                         /* Use old size assignment to avoid
1678                          * deadlock bz14138 & bz14326 */
1679                         inode->i_size = body->size;
1680                 }
1681
1682                 if (body->valid & OBD_MD_FLBLOCKS)
1683                         inode->i_blocks = body->blocks;
1684         }
1685
1686         if (body->valid & OBD_MD_FLMDSCAPA) {
1687                 LASSERT(md->mds_capa);
1688                 ll_add_capa(inode, md->mds_capa);
1689         }
1690         if (body->valid & OBD_MD_FLOSSCAPA) {
1691                 LASSERT(md->oss_capa);
1692                 ll_add_capa(inode, md->oss_capa);
1693         }
1694 }
1695
1696 static struct backing_dev_info ll_backing_dev_info = {
1697         .ra_pages       = 0,    /* No readahead */
1698 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1699         .capabilities   = 0,    /* Does contribute to dirty memory */
1700 #else
1701         .memory_backed  = 0,    /* Does contribute to dirty memory */
1702 #endif
1703 };
1704
1705 void ll_read_inode2(struct inode *inode, void *opaque)
1706 {
1707         struct lustre_md *md = opaque;
1708         struct ll_inode_info *lli = ll_i2info(inode);
1709         ENTRY;
1710
1711         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1712                inode->i_ino, inode->i_generation, inode);
1713
1714         ll_lli_init(lli);
1715
1716         LASSERT(!lli->lli_smd);
1717
1718         /* Core attributes from the MDS first.  This is a new inode, and
1719          * the VFS doesn't zero times in the core inode so we have to do
1720          * it ourselves.  They will be overwritten by either MDS or OST
1721          * attributes - we just need to make sure they aren't newer. */
1722         LTIME_S(inode->i_mtime) = 0;
1723         LTIME_S(inode->i_atime) = 0;
1724         LTIME_S(inode->i_ctime) = 0;
1725         inode->i_rdev = 0;
1726         ll_update_inode(inode, md);
1727
1728         /* OIDEBUG(inode); */
1729
1730         if (S_ISREG(inode->i_mode)) {
1731                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1732                 inode->i_op = &ll_file_inode_operations;
1733                 inode->i_fop = sbi->ll_fop;
1734                 inode->i_mapping->a_ops = &ll_aops;
1735                 EXIT;
1736         } else if (S_ISDIR(inode->i_mode)) {
1737                 inode->i_op = &ll_dir_inode_operations;
1738                 inode->i_fop = &ll_dir_operations;
1739                 inode->i_mapping->a_ops = &ll_dir_aops;
1740                 EXIT;
1741         } else if (S_ISLNK(inode->i_mode)) {
1742                 inode->i_op = &ll_fast_symlink_inode_operations;
1743                 EXIT;
1744         } else {
1745                 inode->i_op = &ll_special_inode_operations;
1746
1747                 init_special_inode(inode, inode->i_mode,
1748                                    kdev_t_to_nr(inode->i_rdev));
1749
1750                 /* initializing backing dev info. */
1751                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1752
1753                 EXIT;
1754         }
1755 }
1756
1757 void ll_delete_inode(struct inode *inode)
1758 {
1759         struct ll_sb_info *sbi = ll_i2sbi(inode);
1760         int rc;
1761         ENTRY;
1762
1763         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1764         if (rc) {
1765                 CERROR("fid_delete() failed, rc %d\n", rc);
1766         }
1767         truncate_inode_pages(&inode->i_data, 0);
1768         clear_inode(inode);
1769
1770         EXIT;
1771 }
1772
1773 int ll_iocontrol(struct inode *inode, struct file *file,
1774                  unsigned int cmd, unsigned long arg)
1775 {
1776         struct ll_sb_info *sbi = ll_i2sbi(inode);
1777         struct ptlrpc_request *req = NULL;
1778         int rc, flags = 0;
1779         ENTRY;
1780
1781         switch(cmd) {
1782         case EXT3_IOC_GETFLAGS: {
1783                 struct mdt_body *body;
1784                 struct obd_capa *oc;
1785
1786                 oc = ll_mdscapa_get(inode);
1787                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1788                                 OBD_MD_FLFLAGS, 0, &req);
1789                 capa_put(oc);
1790                 if (rc) {
1791                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1792                         RETURN(-abs(rc));
1793                 }
1794
1795                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1796
1797                 flags = body->flags;
1798
1799                 ptlrpc_req_finished(req);
1800
1801                 RETURN(put_user(flags, (int *)arg));
1802         }
1803         case EXT3_IOC_SETFLAGS: {
1804                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1805                 struct obd_info oinfo = { { { 0 } } };
1806                 struct md_op_data *op_data;
1807
1808                 if (get_user(flags, (int *)arg))
1809                         RETURN(-EFAULT);
1810
1811                 oinfo.oi_md = lsm;
1812                 OBDO_ALLOC(oinfo.oi_oa);
1813                 if (!oinfo.oi_oa)
1814                         RETURN(-ENOMEM);
1815
1816                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1817                                              LUSTRE_OPC_ANY, NULL);
1818                 if (IS_ERR(op_data))
1819                         RETURN(PTR_ERR(op_data));
1820
1821                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1822                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1823                 rc = md_setattr(sbi->ll_md_exp, op_data,
1824                                 NULL, 0, NULL, 0, &req, NULL);
1825                 ll_finish_md_op_data(op_data);
1826                 ptlrpc_req_finished(req);
1827                 if (rc || lsm == NULL) {
1828                         OBDO_FREE(oinfo.oi_oa);
1829                         RETURN(rc);
1830                 }
1831
1832                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1833                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1834                 oinfo.oi_oa->o_flags = flags;
1835                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1836                                        OBD_MD_FLGROUP;
1837                 oinfo.oi_capa = ll_mdscapa_get(inode);
1838
1839                 obdo_from_inode(oinfo.oi_oa, inode,
1840                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1841                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1842                 capa_put(oinfo.oi_capa);
1843                 OBDO_FREE(oinfo.oi_oa);
1844                 if (rc) {
1845                         if (rc != -EPERM && rc != -EACCES)
1846                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1847                         RETURN(rc);
1848                 }
1849
1850                 inode->i_flags = ll_ext_to_inode_flags(flags |
1851                                                        MDS_BFLAG_EXT_FLAGS);
1852                 RETURN(0);
1853         }
1854         default:
1855                 RETURN(-ENOSYS);
1856         }
1857
1858         RETURN(0);
1859 }
1860
1861 int ll_flush_ctx(struct inode *inode)
1862 {
1863         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1864
1865         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1866
1867         obd_set_info_async(sbi->ll_md_exp,
1868                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1869                            0, NULL, NULL);
1870         obd_set_info_async(sbi->ll_dt_exp,
1871                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1872                            0, NULL, NULL);
1873         return 0;
1874 }
1875
1876 /* umount -f client means force down, don't save state */
1877 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1878 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1879 {
1880         struct super_block *sb = vfsmnt->mnt_sb;
1881 #else
1882 void ll_umount_begin(struct super_block *sb)
1883 {
1884 #endif
1885         struct lustre_sb_info *lsi = s2lsi(sb);
1886         struct ll_sb_info *sbi = ll_s2sbi(sb);
1887         struct obd_device *obd;
1888         struct obd_ioctl_data ioc_data = { 0 };
1889         ENTRY;
1890
1891 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1892         if (!(flags & MNT_FORCE)) {
1893                 EXIT;
1894                 return;
1895         }
1896 #endif
1897
1898         /* Tell the MGC we got umount -f */
1899         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1900
1901         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1902                sb->s_count, atomic_read(&sb->s_active));
1903
1904         obd = class_exp2obd(sbi->ll_md_exp);
1905         if (obd == NULL) {
1906                 CERROR("Invalid MDC connection handle "LPX64"\n",
1907                        sbi->ll_md_exp->exp_handle.h_cookie);
1908                 EXIT;
1909                 return;
1910         }
1911         obd->obd_force = 1;
1912         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1913                       &ioc_data, NULL);
1914
1915         obd = class_exp2obd(sbi->ll_dt_exp);
1916         if (obd == NULL) {
1917                 CERROR("Invalid LOV connection handle "LPX64"\n",
1918                        sbi->ll_dt_exp->exp_handle.h_cookie);
1919                 EXIT;
1920                 return;
1921         }
1922
1923         obd->obd_force = 1;
1924         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1925                       &ioc_data, NULL);
1926
1927         /* Really, we'd like to wait until there are no requests outstanding,
1928          * and then continue.  For now, we just invalidate the requests,
1929          * schedule() and sleep one second if needed, and hope.
1930          */
1931         schedule();
1932 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1933         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1934                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1935                                      cfs_time_seconds(1));
1936                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1937                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1938                                       "may try to umount it a bit later\n",
1939                                       atomic_read(&vfsmnt->mnt_count));
1940         }
1941 #endif
1942
1943         EXIT;
1944 }
1945
1946 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1947 {
1948         struct ll_sb_info *sbi = ll_s2sbi(sb);
1949         int err;
1950         __u32 read_only;
1951
1952         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1953                 read_only = *flags & MS_RDONLY;
1954                 err = obd_set_info_async(sbi->ll_md_exp,
1955                                          sizeof(KEY_READ_ONLY),
1956                                          KEY_READ_ONLY, sizeof(read_only),
1957                                          &read_only, NULL);
1958                 if (err) {
1959                         CERROR("Failed to change the read-only flag during "
1960                                "remount: %d\n", err);
1961                         return err;
1962                 }
1963
1964                 if (read_only)
1965                         sb->s_flags |= MS_RDONLY;
1966                 else
1967                         sb->s_flags &= ~MS_RDONLY;
1968         }
1969         return 0;
1970 }
1971
1972 int ll_prep_inode(struct inode **inode,
1973                   struct ptlrpc_request *req,
1974                   struct super_block *sb)
1975 {
1976         struct ll_sb_info *sbi = NULL;
1977         struct lustre_md md;
1978         int rc;
1979         ENTRY;
1980
1981         LASSERT(*inode || sb);
1982         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1983         prune_deathrow(sbi, 1);
1984         memset(&md, 0, sizeof(struct lustre_md));
1985
1986         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1987                               sbi->ll_md_exp, &md);
1988         if (rc)
1989                 RETURN(rc);
1990
1991         if (*inode) {
1992                 ll_update_inode(*inode, &md);
1993         } else {
1994                 LASSERT(sb != NULL);
1995
1996                 /*
1997                  * At this point server returns to client's same fid as client
1998                  * generated for creating. So using ->fid1 is okay here.
1999                  */
2000                 LASSERT(fid_is_sane(&md.body->fid1));
2001
2002                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2003                 if (*inode == NULL || IS_ERR(*inode)) {
2004                         if (md.lsm)
2005                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2006 #ifdef CONFIG_FS_POSIX_ACL
2007                         if (md.posix_acl) {
2008                                 posix_acl_release(md.posix_acl);
2009                                 md.posix_acl = NULL;
2010                         }
2011 #endif
2012                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2013                         *inode = NULL;
2014                         CERROR("new_inode -fatal: rc %d\n", rc);
2015                         GOTO(out, rc);
2016                 }
2017         }
2018
2019         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2020                          ll_i2info(*inode)->lli_smd);
2021 out:
2022         md_free_lustre_md(sbi->ll_md_exp, &md);
2023         RETURN(rc);
2024 }
2025
2026 int ll_obd_statfs(struct inode *inode, void *arg)
2027 {
2028         struct ll_sb_info *sbi = NULL;
2029         struct obd_export *exp;
2030         char *buf = NULL;
2031         struct obd_ioctl_data *data = NULL;
2032         __u32 type;
2033         int len = 0, rc;
2034
2035         if (!inode || !(sbi = ll_i2sbi(inode)))
2036                 GOTO(out_statfs, rc = -EINVAL);
2037
2038         rc = obd_ioctl_getdata(&buf, &len, arg);
2039         if (rc)
2040                 GOTO(out_statfs, rc);
2041
2042         data = (void*)buf;
2043         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2044             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2045                 GOTO(out_statfs, rc = -EINVAL);
2046
2047         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2048         if (type == LL_STATFS_MDC)
2049                 exp = sbi->ll_md_exp;
2050         else if (type == LL_STATFS_LOV)
2051                 exp = sbi->ll_dt_exp;
2052         else
2053                 GOTO(out_statfs, rc = -ENODEV);
2054
2055         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2056         if (rc)
2057                 GOTO(out_statfs, rc);
2058 out_statfs:
2059         if (buf)
2060                 obd_ioctl_freedata(buf, len);
2061         return rc;
2062 }
2063
2064 int ll_process_config(struct lustre_cfg *lcfg)
2065 {
2066         char *ptr;
2067         void *sb;
2068         struct lprocfs_static_vars lvars;
2069         unsigned long x;
2070         int rc = 0;
2071
2072         lprocfs_llite_init_vars(&lvars);
2073
2074         /* The instance name contains the sb: lustre-client-aacfe000 */
2075         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2076         if (!ptr || !*(++ptr))
2077                 return -EINVAL;
2078         if (sscanf(ptr, "%lx", &x) != 1)
2079                 return -EINVAL;
2080         sb = (void *)x;
2081         /* This better be a real Lustre superblock! */
2082         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2083
2084         /* Note we have not called client_common_fill_super yet, so
2085            proc fns must be able to handle that! */
2086         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2087                                       lcfg, sb);
2088         if (rc > 0)
2089                 rc = 0;
2090         return(rc);
2091 }
2092
2093 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2094 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2095                                        struct inode *i1, struct inode *i2,
2096                                        const char *name, int namelen,
2097                                        int mode, __u32 opc, void *data)
2098 {
2099         LASSERT(i1 != NULL);
2100
2101         if (namelen > ll_i2sbi(i1)->ll_namelen)
2102                 return ERR_PTR(-ENAMETOOLONG);
2103
2104         if (op_data == NULL)
2105                 OBD_ALLOC_PTR(op_data);
2106
2107         if (op_data == NULL)
2108                 return ERR_PTR(-ENOMEM);
2109
2110         ll_i2gids(op_data->op_suppgids, i1, i2);
2111         op_data->op_fid1 = *ll_inode2fid(i1);
2112         op_data->op_capa1 = ll_mdscapa_get(i1);
2113
2114         if (i2) {
2115                 op_data->op_fid2 = *ll_inode2fid(i2);
2116                 op_data->op_capa2 = ll_mdscapa_get(i2);
2117         } else {
2118                 fid_zero(&op_data->op_fid2);
2119                 op_data->op_capa2 = NULL;
2120         }
2121
2122         op_data->op_name = name;
2123         op_data->op_namelen = namelen;
2124         op_data->op_mode = mode;
2125         op_data->op_mod_time = cfs_time_current_sec();
2126         op_data->op_fsuid = current->fsuid;
2127         op_data->op_fsgid = current->fsgid;
2128         op_data->op_cap = cfs_curproc_cap_pack();
2129         op_data->op_bias = MDS_CHECK_SPLIT;
2130         op_data->op_opc = opc;
2131         op_data->op_mds = 0;
2132         op_data->op_data = data;
2133
2134         return op_data;
2135 }
2136
2137 void ll_finish_md_op_data(struct md_op_data *op_data)
2138 {
2139         capa_put(op_data->op_capa1);
2140         capa_put(op_data->op_capa2);
2141         OBD_FREE_PTR(op_data);
2142 }
2143
2144 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2145 {
2146         struct ll_sb_info *sbi;
2147
2148         LASSERT((seq != NULL) && (vfs != NULL));
2149         sbi = ll_s2sbi(vfs->mnt_sb);
2150
2151         if (sbi->ll_flags & LL_SBI_NOLCK)
2152                 seq_puts(seq, ",nolock");
2153
2154         if (sbi->ll_flags & LL_SBI_FLOCK)
2155                 seq_puts(seq, ",flock");
2156
2157         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2158                 seq_puts(seq, ",localflock");
2159
2160         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2161                 seq_puts(seq, ",user_xattr");
2162
2163         if (sbi->ll_flags & LL_SBI_ACL)
2164                 seq_puts(seq, ",acl");
2165
2166         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2167                 seq_puts(seq, ",lazystatfs");
2168
2169         RETURN(0);
2170 }