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 = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         spin_lock(&ll_sb_lock);
 115         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_LLITE_CHECKSUM
 119         sbi->ll_flags |= LL_SBI_CHECKSUM;
 120 #endif
 121
 122 #ifdef HAVE_LRU_RESIZE_SUPPORT
 123         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 124 #endif
 125
 126 #ifdef HAVE_EXPORT___IGET
 127         INIT_LIST_HEAD(&sbi->ll_deathrow);
 128         spin_lock_init(&sbi->ll_deathrow_lock);
 129 #endif
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137
 138         RETURN(sbi);
 139 }
 140
 141 void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144         ENTRY;
 145
 146         if (sbi != NULL) {
 147                 spin_lock(&ll_sb_lock);
 148                 list_del(&sbi->ll_list);
 149                 spin_unlock(&ll_sb_lock);
 150                 OBD_FREE(sbi, sizeof(*sbi));
 151         }
 152         EXIT;
 153 }
 154
 155 static struct dentry_operations ll_d_root_ops = {
 156 #ifdef DCACHE_LUSTRE_INVALID
 157         .d_compare = ll_dcompare,
 158 #endif
 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;
 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_GRANT_SHRINK;
 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         /* destroy inodes in deathrow */
 687         prune_deathrow(sbi, 0);
 688
 689         list_del(&sbi->ll_conn_chain);
 690
 691         obd_fid_fini(sbi->ll_dt_exp);
 692         obd_disconnect(sbi->ll_dt_exp);
 693         sbi->ll_dt_exp = NULL;
 694
 695         lprocfs_unregister_mountpoint(sbi);
 696
 697         obd_fid_fini(sbi->ll_md_exp);
 698         obd_disconnect(sbi->ll_md_exp);
 699         sbi->ll_md_exp = NULL;
 700
 701         EXIT;
 702 }
 703
 704 void ll_kill_super(struct super_block *sb)
 705 {
 706         struct ll_sb_info *sbi;
 707
 708         ENTRY;
 709
 710         /* not init sb ?*/
 711         if (!(sb->s_flags & MS_ACTIVE))
 712                 return;
 713
 714         sbi = ll_s2sbi(sb);
 715         /* we need restore s_dev from changed for clustred NFS before put_super
 716          * because new kernels have cached s_dev and change sb->s_dev in
 717          * put_super not affected real removing devices */
 718         if (sbi)
 719                 sb->s_dev = sbi->ll_sdev_orig;
 720         EXIT;
 721 }
 722
 723 char *ll_read_opt(const char *opt, char *data)
 724 {
 725         char *value;
 726         char *retval;
 727         ENTRY;
 728
 729         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 730         if (strncmp(opt, data, strlen(opt)))
 731                 RETURN(NULL);
 732         if ((value = strchr(data, '=')) == NULL)
 733                 RETURN(NULL);
 734
 735         value++;
 736         OBD_ALLOC(retval, strlen(value) + 1);
 737         if (!retval) {
 738                 CERROR("out of memory!\n");
 739                 RETURN(NULL);
 740         }
 741
 742         memcpy(retval, value, strlen(value)+1);
 743         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 744         RETURN(retval);
 745 }
 746
 747 static inline int ll_set_opt(const char *opt, char *data, int fl)
 748 {
 749         if (strncmp(opt, data, strlen(opt)) != 0)
 750                 return(0);
 751         else
 752                 return(fl);
 753 }
 754
 755 /* non-client-specific mount options are parsed in lmd_parse */
 756 static int ll_options(char *options, int *flags)
 757 {
 758         int tmp;
 759         char *s1 = options, *s2;
 760         ENTRY;
 761
 762         if (!options)
 763                 RETURN(0);
 764
 765         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 766
 767         while (*s1) {
 768                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 769                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 770                 if (tmp) {
 771                         *flags |= tmp;
 772                         goto next;
 773                 }
 774                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 775                 if (tmp) {
 776                         *flags |= tmp;
 777                         goto next;
 778                 }
 779                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 780                 if (tmp) {
 781                         *flags |= tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 785                 if (tmp) {
 786                         *flags &= ~tmp;
 787                         goto next;
 788                 }
 789                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 795                 if (tmp) {
 796                         *flags &= ~tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 800                 if (tmp) {
 801                         /* Ignore deprecated mount option.  The client will
 802                          * always try to mount with ACL support, whether this
 803                          * is used depends on whether server supports it. */
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 807                 if (tmp) {
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815
 816                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 817                 if (tmp) {
 818                         *flags |= tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 822                 if (tmp) {
 823                         *flags &= ~tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836
 837                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 838                                    s1);
 839                 RETURN(-EINVAL);
 840
 841 next:
 842                 /* Find next opt */
 843                 s2 = strchr(s1, ',');
 844                 if (s2 == NULL)
 845                         break;
 846                 s1 = s2 + 1;
 847         }
 848         RETURN(0);
 849 }
 850
 851 void ll_lli_init(struct ll_inode_info *lli)
 852 {
 853         lli->lli_inode_magic = LLI_INODE_MAGIC;
 854         sema_init(&lli->lli_size_sem, 1);
 855         sema_init(&lli->lli_write_sem, 1);
 856         lli->lli_flags = 0;
 857         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 858         spin_lock_init(&lli->lli_lock);
 859         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 860         INIT_LIST_HEAD(&lli->lli_close_list);
 861         lli->lli_inode_magic = LLI_INODE_MAGIC;
 862         sema_init(&lli->lli_och_sem, 1);
 863         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 864         lli->lli_mds_exec_och = NULL;
 865         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 866         lli->lli_open_fd_exec_count = 0;
 867         INIT_LIST_HEAD(&lli->lli_dead_list);
 868         lli->lli_remote_perms = NULL;
 869         lli->lli_rmtperm_utime = 0;
 870         sema_init(&lli->lli_rmtperm_sem, 1);
 871         INIT_LIST_HEAD(&lli->lli_oss_capas);
 872 }
 873
 874 int ll_fill_super(struct super_block *sb)
 875 {
 876         struct lustre_profile *lprof;
 877         struct lustre_sb_info *lsi = s2lsi(sb);
 878         struct ll_sb_info *sbi;
 879         char  *dt = NULL, *md = NULL;
 880         char  *profilenm = get_profile_name(sb);
 881         struct config_llog_instance cfg = {0, };
 882         char   ll_instance[sizeof(sb) * 2 + 1];
 883         int    err;
 884         ENTRY;
 885
 886         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 887
 888         cfs_module_get();
 889
 890         /* client additional sb info */
 891         lsi->lsi_llsbi = sbi = ll_init_sbi();
 892         if (!sbi) {
 893                 cfs_module_put();
 894                 RETURN(-ENOMEM);
 895         }
 896
 897         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 898         if (err)
 899                 GOTO(out_free, err);
 900
 901         /* Generate a string unique to this super, in case some joker tries
 902            to mount the same fs at two mount points.
 903            Use the address of the super itself.*/
 904         sprintf(ll_instance, "%p", sb);
 905         cfg.cfg_instance = ll_instance;
 906         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 907
 908         /* set up client obds */
 909         err = lustre_process_log(sb, profilenm, &cfg);
 910         if (err < 0) {
 911                 CERROR("Unable to process log: %d\n", err);
 912                 GOTO(out_free, err);
 913         }
 914
 915         lprof = class_get_profile(profilenm);
 916         if (lprof == NULL) {
 917                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 918                                    " read from the MGS.  Does that filesystem "
 919                                    "exist?\n", profilenm);
 920                 GOTO(out_free, err = -EINVAL);
 921         }
 922         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 923                lprof->lp_md, lprof->lp_dt);
 924
 925         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 926                   strlen(ll_instance) + 2);
 927         if (!dt)
 928                 GOTO(out_free, err = -ENOMEM);
 929         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 930
 931         OBD_ALLOC(md, strlen(lprof->lp_md) +
 932                   strlen(ll_instance) + 2);
 933         if (!md)
 934                 GOTO(out_free, err = -ENOMEM);
 935         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 936
 937         /* connections, registrations, sb setup */
 938         err = client_common_fill_super(sb, md, dt);
 939
 940 out_free:
 941         if (md)
 942                 OBD_FREE(md, strlen(md) + 1);
 943         if (dt)
 944                 OBD_FREE(dt, strlen(dt) + 1);
 945         if (err)
 946                 ll_put_super(sb);
 947         else
 948                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 949
 950         RETURN(err);
 951 } /* ll_fill_super */
 952
 953
 954 void lu_context_keys_dump(void);
 955
 956 void ll_put_super(struct super_block *sb)
 957 {
 958         struct config_llog_instance cfg;
 959         char   ll_instance[sizeof(sb) * 2 + 1];
 960         struct obd_device *obd;
 961         struct lustre_sb_info *lsi = s2lsi(sb);
 962         struct ll_sb_info *sbi = ll_s2sbi(sb);
 963         char *profilenm = get_profile_name(sb);
 964         int force = 1, next;
 965         ENTRY;
 966
 967         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 968
 969         ll_print_capa_stat(sbi);
 970
 971         sprintf(ll_instance, "%p", sb);
 972         cfg.cfg_instance = ll_instance;
 973         lustre_end_log(sb, NULL, &cfg);
 974
 975         if (sbi->ll_md_exp) {
 976                 obd = class_exp2obd(sbi->ll_md_exp);
 977                 if (obd)
 978                         force = obd->obd_force;
 979         }
 980
 981         /* We need to set force before the lov_disconnect in
 982            lustre_common_put_super, since l_d cleans up osc's as well. */
 983         if (force) {
 984                 next = 0;
 985                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 986                                                      &next)) != NULL) {
 987                         obd->obd_force = force;
 988                 }
 989         }
 990
 991         cl_sb_fini(sb);
 992
 993         if (sbi->ll_lcq) {
 994                 /* Only if client_common_fill_super succeeded */
 995                 client_common_put_super(sb);
 996         }
 997         next = 0;
 998         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 999                 class_manual_cleanup(obd);
1000         }
1001
1002         if (profilenm)
1003                 class_del_profile(profilenm);
1004
1005         ll_free_sbi(sb);
1006         lsi->lsi_llsbi = NULL;
1007
1008         lustre_common_put_super(sb);
1009
1010         cl_env_cache_purge(~0);
1011
1012         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1013
1014         cfs_module_put();
1015
1016         EXIT;
1017 } /* client_put_super */
1018
1019 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1020 {
1021         struct inode *inode = NULL;
1022         /* NOTE: we depend on atomic igrab() -bzzz */
1023         lock_res_and_lock(lock);
1024         if (lock->l_ast_data) {
1025                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1026                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1027                         inode = igrab(lock->l_ast_data);
1028                 } else {
1029                         inode = lock->l_ast_data;
1030                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1031                                                 D_INFO : D_WARNING,
1032                                         lock, __FILE__, __func__, __LINE__,
1033                                         "l_ast_data %p is bogus: magic %08x",
1034                                         lock->l_ast_data, lli->lli_inode_magic);
1035                         inode = NULL;
1036                 }
1037         }
1038         unlock_res_and_lock(lock);
1039         return inode;
1040 }
1041
1042 static int null_if_equal(struct ldlm_lock *lock, void *data)
1043 {
1044         if (data == lock->l_ast_data) {
1045                 lock->l_ast_data = NULL;
1046
1047                 if (lock->l_req_mode != lock->l_granted_mode)
1048                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1049         }
1050
1051         return LDLM_ITER_CONTINUE;
1052 }
1053
1054 void ll_clear_inode(struct inode *inode)
1055 {
1056         struct ll_inode_info *lli = ll_i2info(inode);
1057         struct ll_sb_info *sbi = ll_i2sbi(inode);
1058         ENTRY;
1059
1060         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1061                inode->i_generation, inode);
1062
1063         if (S_ISDIR(inode->i_mode)) {
1064                 /* these should have been cleared in ll_file_release */
1065                 LASSERT(lli->lli_sai == NULL);
1066                 LASSERT(lli->lli_opendir_key == NULL);
1067                 LASSERT(lli->lli_opendir_pid == 0);
1068         }
1069
1070         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1071         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1072                          null_if_equal, inode);
1073
1074         LASSERT(!lli->lli_open_fd_write_count);
1075         LASSERT(!lli->lli_open_fd_read_count);
1076         LASSERT(!lli->lli_open_fd_exec_count);
1077
1078         if (lli->lli_mds_write_och)
1079                 ll_md_real_close(inode, FMODE_WRITE);
1080         if (lli->lli_mds_exec_och)
1081                 ll_md_real_close(inode, FMODE_EXEC);
1082         if (lli->lli_mds_read_och)
1083                 ll_md_real_close(inode, FMODE_READ);
1084
1085         if (lli->lli_symlink_name) {
1086                 OBD_FREE(lli->lli_symlink_name,
1087                          strlen(lli->lli_symlink_name) + 1);
1088                 lli->lli_symlink_name = NULL;
1089         }
1090
1091         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1092                 LASSERT(lli->lli_posix_acl == NULL);
1093                 if (lli->lli_remote_perms) {
1094                         free_rmtperm_hash(lli->lli_remote_perms);
1095                         lli->lli_remote_perms = NULL;
1096                 }
1097         }
1098 #ifdef CONFIG_FS_POSIX_ACL
1099         else if (lli->lli_posix_acl) {
1100                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1101                 LASSERT(lli->lli_remote_perms == NULL);
1102                 posix_acl_release(lli->lli_posix_acl);
1103                 lli->lli_posix_acl = NULL;
1104         }
1105 #endif
1106         lli->lli_inode_magic = LLI_INODE_DEAD;
1107
1108 #ifdef HAVE_EXPORT___IGET
1109         spin_lock(&sbi->ll_deathrow_lock);
1110         list_del_init(&lli->lli_dead_list);
1111         spin_unlock(&sbi->ll_deathrow_lock);
1112 #endif
1113         ll_clear_inode_capas(inode);
1114         /*
1115          * XXX This has to be done before lsm is freed below, because
1116          * cl_object still uses inode lsm.
1117          */
1118         cl_inode_fini(inode);
1119
1120         if (lli->lli_smd) {
1121                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1122                 lli->lli_smd = NULL;
1123         }
1124
1125
1126         EXIT;
1127 }
1128
1129 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1130                   struct md_open_data **mod)
1131 {
1132         struct lustre_md md;
1133         struct ll_sb_info *sbi = ll_i2sbi(inode);
1134         struct ptlrpc_request *request = NULL;
1135         int rc;
1136         ENTRY;
1137
1138         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1139                                      LUSTRE_OPC_ANY, NULL);
1140         if (IS_ERR(op_data))
1141                 RETURN(PTR_ERR(op_data));
1142
1143         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1144                         &request, mod);
1145         if (rc) {
1146                 ptlrpc_req_finished(request);
1147                 if (rc == -ENOENT) {
1148                         inode->i_nlink = 0;
1149                         /* Unlinked special device node? Or just a race?
1150                          * Pretend we done everything. */
1151                         if (!S_ISREG(inode->i_mode) &&
1152                             !S_ISDIR(inode->i_mode))
1153                                 rc = inode_setattr(inode, &op_data->op_attr);
1154                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1155                         CERROR("md_setattr fails: rc = %d\n", rc);
1156                 }
1157                 RETURN(rc);
1158         }
1159
1160         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1161                               sbi->ll_md_exp, &md);
1162         if (rc) {
1163                 ptlrpc_req_finished(request);
1164                 RETURN(rc);
1165         }
1166
1167         /* We call inode_setattr to adjust timestamps.
1168          * If there is at least some data in file, we cleared ATTR_SIZE
1169          * above to avoid invoking vmtruncate, otherwise it is important
1170          * to call vmtruncate in inode_setattr to update inode->i_size
1171          * (bug 6196) */
1172         rc = inode_setattr(inode, &op_data->op_attr);
1173
1174         /* Extract epoch data if obtained. */
1175         op_data->op_handle = md.body->handle;
1176         op_data->op_ioepoch = md.body->ioepoch;
1177
1178         ll_update_inode(inode, &md);
1179         ptlrpc_req_finished(request);
1180
1181         RETURN(rc);
1182 }
1183
1184 /* Close IO epoch and send Size-on-MDS attribute update. */
1185 static int ll_setattr_done_writing(struct inode *inode,
1186                                    struct md_op_data *op_data,
1187                                    struct md_open_data *mod)
1188 {
1189         struct ll_inode_info *lli = ll_i2info(inode);
1190         int rc = 0;
1191         ENTRY;
1192
1193         LASSERT(op_data != NULL);
1194         if (!S_ISREG(inode->i_mode))
1195                 RETURN(0);
1196
1197         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1198                op_data->op_ioepoch, PFID(&lli->lli_fid));
1199
1200         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1201         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1202         if (rc == -EAGAIN) {
1203                 /* MDS has instructed us to obtain Size-on-MDS attribute
1204                  * from OSTs and send setattr to back to MDS. */
1205                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1206                                          op_data->op_ioepoch);
1207         } else if (rc) {
1208                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1209                        inode->i_ino, rc);
1210         }
1211         RETURN(rc);
1212 }
1213
1214 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1215 {
1216         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1217         int rc;
1218
1219         rc = cl_setattr_do_truncate(inode, size, capa);
1220         ll_truncate_free_capa(capa);
1221         return rc;
1222 }
1223
1224 static int ll_setattr_ost(struct inode *inode)
1225 {
1226         struct obd_capa *capa = ll_mdscapa_get(inode);
1227         int rc;
1228
1229         rc = cl_setattr_ost(inode, capa);
1230         capa_put(capa);
1231
1232         return rc;
1233 }
1234
1235 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1236  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1237  * keep these values until such a time that objects are allocated for it.
1238  * We do the MDS operations first, as it is checking permissions for us.
1239  * We don't to the MDS RPC if there is nothing that we want to store there,
1240  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1241  * going to do an RPC anyways.
1242  *
1243  * If we are doing a truncate, we will send the mtime and ctime updates
1244  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1245  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1246  * at the same time.
1247  */
1248 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1249 {
1250         struct ll_inode_info *lli = ll_i2info(inode);
1251         struct lov_stripe_md *lsm = lli->lli_smd;
1252         struct md_op_data *op_data = NULL;
1253         struct md_open_data *mod = NULL;
1254         int ia_valid = attr->ia_valid;
1255         int rc = 0, rc1 = 0;
1256         ENTRY;
1257
1258         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1259                attr->ia_valid);
1260         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1261
1262         if (ia_valid & ATTR_SIZE) {
1263                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1264                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1265                                attr->ia_size, ll_file_maxbytes(inode));
1266                         RETURN(-EFBIG);
1267                 }
1268
1269                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1270         }
1271
1272         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1273         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1274                 if (current->fsuid != inode->i_uid &&
1275                     !cfs_capable(CFS_CAP_FOWNER))
1276                         RETURN(-EPERM);
1277         }
1278
1279         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1280         if (attr->ia_valid & ATTR_CTIME) {
1281                 attr->ia_ctime = CURRENT_TIME;
1282                 attr->ia_valid |= ATTR_CTIME_SET;
1283         }
1284         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1285                 attr->ia_atime = CURRENT_TIME;
1286                 attr->ia_valid |= ATTR_ATIME_SET;
1287         }
1288         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1289                 attr->ia_mtime = CURRENT_TIME;
1290                 attr->ia_valid |= ATTR_MTIME_SET;
1291         }
1292         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1293                 /* To avoid stale mtime on mds, obtain it from ost and send
1294                    to mds. */
1295                 rc = cl_glimpse_size(inode);
1296                 if (rc)
1297                         RETURN(rc);
1298
1299                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1300                 attr->ia_mtime = inode->i_mtime;
1301         }
1302
1303         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1304                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1305                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1306                        cfs_time_current_sec());
1307
1308         /* NB: ATTR_SIZE will only be set after this point if the size
1309          * resides on the MDS, ie, this file has no objects. */
1310         if (lsm)
1311                 attr->ia_valid &= ~ATTR_SIZE;
1312
1313         /* We always do an MDS RPC, even if we're only changing the size;
1314          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1315
1316         OBD_ALLOC_PTR(op_data);
1317         if (op_data == NULL)
1318                 RETURN(-ENOMEM);
1319
1320         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1321
1322         /* Open epoch for truncate. */
1323         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1324             (ia_valid & ATTR_SIZE))
1325                 op_data->op_flags = MF_EPOCH_OPEN;
1326
1327         rc = ll_md_setattr(inode, op_data, &mod);
1328         if (rc)
1329                 GOTO(out, rc);
1330
1331         if (op_data->op_ioepoch)
1332                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1333                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1334
1335         if (!lsm || !S_ISREG(inode->i_mode)) {
1336                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1337                 GOTO(out, rc = 0);
1338         }
1339
1340         if (ia_valid & ATTR_SIZE)
1341                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1342         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1343                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1344                        inode->i_ino, LTIME_S(attr->ia_mtime));
1345                 rc = ll_setattr_ost(inode);
1346         }
1347         EXIT;
1348 out:
1349         if (op_data) {
1350                 if (op_data->op_ioepoch)
1351                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1352                 ll_finish_md_op_data(op_data);
1353         }
1354         return rc ? rc : rc1;
1355 }
1356
1357 int ll_setattr(struct dentry *de, struct iattr *attr)
1358 {
1359         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1360             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1361                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1362
1363         if ((de->d_inode->i_mode & S_ISUID) &&
1364             !(attr->ia_mode & S_ISUID) &&
1365             !(attr->ia_valid & ATTR_KILL_SUID))
1366                 attr->ia_valid |= ATTR_KILL_SUID;
1367
1368         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1369             !(attr->ia_mode & S_ISGID) &&
1370             !(attr->ia_valid & ATTR_KILL_SGID))
1371                 attr->ia_valid |= ATTR_KILL_SGID;
1372
1373         return ll_setattr_raw(de->d_inode, attr);
1374 }
1375
1376 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1377                        __u64 max_age, __u32 flags)
1378 {
1379         struct ll_sb_info *sbi = ll_s2sbi(sb);
1380         struct obd_statfs obd_osfs;
1381         int rc;
1382         ENTRY;
1383
1384         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1385         if (rc) {
1386                 CERROR("md_statfs fails: rc = %d\n", rc);
1387                 RETURN(rc);
1388         }
1389
1390         osfs->os_type = sb->s_magic;
1391
1392         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1393                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1394
1395         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1396                               &obd_osfs, max_age, flags);
1397         if (rc) {
1398                 CERROR("obd_statfs fails: rc = %d\n", rc);
1399                 RETURN(rc);
1400         }
1401
1402         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1403                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1404                obd_osfs.os_files);
1405
1406         osfs->os_bsize = obd_osfs.os_bsize;
1407         osfs->os_blocks = obd_osfs.os_blocks;
1408         osfs->os_bfree = obd_osfs.os_bfree;
1409         osfs->os_bavail = obd_osfs.os_bavail;
1410
1411         /* If we don't have as many objects free on the OST as inodes
1412          * on the MDS, we reduce the total number of inodes to
1413          * compensate, so that the "inodes in use" number is correct.
1414          */
1415         if (obd_osfs.os_ffree < osfs->os_ffree) {
1416                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1417                         obd_osfs.os_ffree;
1418                 osfs->os_ffree = obd_osfs.os_ffree;
1419         }
1420
1421         RETURN(rc);
1422 }
1423 #ifndef HAVE_STATFS_DENTRY_PARAM
1424 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1425 {
1426 #else
1427 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1428 {
1429         struct super_block *sb = de->d_sb;
1430 #endif
1431         struct obd_statfs osfs;
1432         int rc;
1433
1434         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1435         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1436
1437         /* For now we will always get up-to-date statfs values, but in the
1438          * future we may allow some amount of caching on the client (e.g.
1439          * from QOS or lprocfs updates). */
1440         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1441         if (rc)
1442                 return rc;
1443
1444         statfs_unpack(sfs, &osfs);
1445
1446         /* We need to downshift for all 32-bit kernels, because we can't
1447          * tell if the kernel is being called via sys_statfs64() or not.
1448          * Stop before overflowing f_bsize - in which case it is better
1449          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1450         if (sizeof(long) < 8) {
1451                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1452                         sfs->f_bsize <<= 1;
1453
1454                         osfs.os_blocks >>= 1;
1455                         osfs.os_bfree >>= 1;
1456                         osfs.os_bavail >>= 1;
1457                 }
1458         }
1459
1460         sfs->f_blocks = osfs.os_blocks;
1461         sfs->f_bfree = osfs.os_bfree;
1462         sfs->f_bavail = osfs.os_bavail;
1463
1464         return 0;
1465 }
1466
1467 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1468 {
1469         struct ll_inode_info *lli;
1470         struct lov_stripe_md *lsm;
1471
1472         lli = ll_i2info(inode);
1473         LASSERT(lli->lli_size_sem_owner != current);
1474         down(&lli->lli_size_sem);
1475         LASSERT(lli->lli_size_sem_owner == NULL);
1476         lli->lli_size_sem_owner = current;
1477         lsm = lli->lli_smd;
1478         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1479                  lsm, lock_lsm);
1480         if (lock_lsm)
1481                 lov_stripe_lock(lsm);
1482 }
1483
1484 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1485 {
1486         struct ll_inode_info *lli;
1487         struct lov_stripe_md *lsm;
1488
1489         lli = ll_i2info(inode);
1490         lsm = lli->lli_smd;
1491         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1492                  lsm, unlock_lsm);
1493         if (unlock_lsm)
1494                 lov_stripe_unlock(lsm);
1495         LASSERT(lli->lli_size_sem_owner == current);
1496         lli->lli_size_sem_owner = NULL;
1497         up(&lli->lli_size_sem);
1498 }
1499
1500 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1501 {
1502         struct ll_inode_info *lli = ll_i2info(inode);
1503
1504         dump_lsm(D_INODE, lsm);
1505         dump_lsm(D_INODE, lli->lli_smd);
1506         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1507                  "lsm must be joined lsm %p\n", lsm);
1508         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1509         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1510                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1511         lli->lli_smd = lsm;
1512         lli->lli_maxbytes = lsm->lsm_maxbytes;
1513         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1514                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1515 }
1516
1517 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1518 {
1519         struct ll_inode_info *lli = ll_i2info(inode);
1520         struct mdt_body *body = md->body;
1521         struct lov_stripe_md *lsm = md->lsm;
1522         struct ll_sb_info *sbi = ll_i2sbi(inode);
1523
1524         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1525         if (lsm != NULL) {
1526                 if (lli->lli_smd == NULL) {
1527                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1528                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1529                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1530                                 dump_lsm(D_ERROR, lsm);
1531                                 LBUG();
1532                         }
1533                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1534                                lsm, inode->i_ino, inode->i_generation, inode);
1535                         cl_inode_init(inode, md);
1536                         /* ll_inode_size_lock() requires it is only
1537                          * called with lli_smd != NULL or lock_lsm == 0
1538                          *  or we can race between lock/unlock.
1539                          *  bug 9547 */
1540                         lli->lli_smd = lsm;
1541                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1542                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1543                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1544                 } else {
1545                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1546                              lli->lli_smd->lsm_stripe_count ==
1547                                         lsm->lsm_stripe_count) {
1548                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1549                                         CERROR("lsm mismatch for inode %ld\n",
1550                                                 inode->i_ino);
1551                                         CERROR("lli_smd:\n");
1552                                         dump_lsm(D_ERROR, lli->lli_smd);
1553                                         CERROR("lsm:\n");
1554                                         dump_lsm(D_ERROR, lsm);
1555                                         LBUG();
1556                                 }
1557                         } else {
1558                                 cl_inode_init(inode, md);
1559                                 ll_replace_lsm(inode, lsm);
1560                         }
1561                 }
1562                 if (lli->lli_smd != lsm)
1563                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1564         }
1565
1566         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1567                 if (body->valid & OBD_MD_FLRMTPERM)
1568                         ll_update_remote_perm(inode, md->remote_perm);
1569         }
1570 #ifdef CONFIG_FS_POSIX_ACL
1571         else if (body->valid & OBD_MD_FLACL) {
1572                 spin_lock(&lli->lli_lock);
1573                 if (lli->lli_posix_acl)
1574                         posix_acl_release(lli->lli_posix_acl);
1575                 lli->lli_posix_acl = md->posix_acl;
1576                 spin_unlock(&lli->lli_lock);
1577         }
1578 #endif
1579         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1580         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1581
1582         if (body->valid & OBD_MD_FLATIME &&
1583             body->atime > LTIME_S(inode->i_atime))
1584                 LTIME_S(inode->i_atime) = body->atime;
1585
1586         /* mtime is always updated with ctime, but can be set in past.
1587            As write and utime(2) may happen within 1 second, and utime's
1588            mtime has a priority over write's one, so take mtime from mds
1589            for the same ctimes. */
1590         if (body->valid & OBD_MD_FLCTIME &&
1591             body->ctime >= LTIME_S(inode->i_ctime)) {
1592                 LTIME_S(inode->i_ctime) = body->ctime;
1593                 if (body->valid & OBD_MD_FLMTIME) {
1594                         CDEBUG(D_INODE, "setting ino %lu mtime "
1595                                "from %lu to "LPU64"\n", inode->i_ino,
1596                                LTIME_S(inode->i_mtime), body->mtime);
1597                         LTIME_S(inode->i_mtime) = body->mtime;
1598                 }
1599         }
1600         if (body->valid & OBD_MD_FLMODE)
1601                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1602         if (body->valid & OBD_MD_FLTYPE)
1603                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1604         LASSERT(inode->i_mode != 0);
1605         if (S_ISREG(inode->i_mode)) {
1606                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1607         } else {
1608                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1609         }
1610 #ifdef HAVE_INODE_BLKSIZE
1611         inode->i_blksize = 1<<inode->i_blkbits;
1612 #endif
1613         if (body->valid & OBD_MD_FLUID)
1614                 inode->i_uid = body->uid;
1615         if (body->valid & OBD_MD_FLGID)
1616                 inode->i_gid = body->gid;
1617         if (body->valid & OBD_MD_FLFLAGS)
1618                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1619         if (body->valid & OBD_MD_FLNLINK)
1620                 inode->i_nlink = body->nlink;
1621         if (body->valid & OBD_MD_FLRDEV)
1622                 inode->i_rdev = old_decode_dev(body->rdev);
1623
1624         if (body->valid & OBD_MD_FLID) {
1625                 /* FID shouldn't be changed! */
1626                 if (fid_is_sane(&lli->lli_fid)) {
1627                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1628                                  "Trying to change FID "DFID
1629                                  " to the "DFID", inode %lu/%u(%p)\n",
1630                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1631                                  inode->i_ino, inode->i_generation, inode);
1632                 } else
1633                         lli->lli_fid = body->fid1;
1634         }
1635
1636         LASSERT(fid_seq(&lli->lli_fid) != 0);
1637
1638         if (body->valid & OBD_MD_FLSIZE) {
1639                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1640                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1641                         struct lustre_handle lockh;
1642                         ldlm_mode_t mode;
1643
1644                         /* As it is possible a blocking ast has been processed
1645                          * by this time, we need to check there is an UPDATE
1646                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1647                          * it. */
1648                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1649                                                &lockh);
1650                         if (mode) {
1651                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1652                                                       LLIF_EPOCH_PENDING |
1653                                                       LLIF_SOM_DIRTY)) {
1654                                         CERROR("ino %lu flags %lu still has "
1655                                                "size authority! do not trust "
1656                                                "the size got from MDS\n",
1657                                                inode->i_ino, lli->lli_flags);
1658                                 } else {
1659                                         /* Use old size assignment to avoid
1660                                          * deadlock bz14138 & bz14326 */
1661                                         inode->i_size = body->size;
1662                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1663                                 }
1664                                 ldlm_lock_decref(&lockh, mode);
1665                         }
1666                 } else {
1667                         /* Use old size assignment to avoid
1668                          * deadlock bz14138 & bz14326 */
1669                         inode->i_size = body->size;
1670                 }
1671
1672                 if (body->valid & OBD_MD_FLBLOCKS)
1673                         inode->i_blocks = body->blocks;
1674         }
1675
1676         if (body->valid & OBD_MD_FLMDSCAPA) {
1677                 LASSERT(md->mds_capa);
1678                 ll_add_capa(inode, md->mds_capa);
1679         }
1680         if (body->valid & OBD_MD_FLOSSCAPA) {
1681                 LASSERT(md->oss_capa);
1682                 ll_add_capa(inode, md->oss_capa);
1683         }
1684 }
1685
1686 static struct backing_dev_info ll_backing_dev_info = {
1687         .ra_pages       = 0,    /* No readahead */
1688 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1689         .capabilities   = 0,    /* Does contribute to dirty memory */
1690 #else
1691         .memory_backed  = 0,    /* Does contribute to dirty memory */
1692 #endif
1693 };
1694
1695 void ll_read_inode2(struct inode *inode, void *opaque)
1696 {
1697         struct lustre_md *md = opaque;
1698         struct ll_inode_info *lli = ll_i2info(inode);
1699         ENTRY;
1700
1701         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1702                inode->i_ino, inode->i_generation, inode);
1703
1704         ll_lli_init(lli);
1705
1706         LASSERT(!lli->lli_smd);
1707
1708         /* Core attributes from the MDS first.  This is a new inode, and
1709          * the VFS doesn't zero times in the core inode so we have to do
1710          * it ourselves.  They will be overwritten by either MDS or OST
1711          * attributes - we just need to make sure they aren't newer. */
1712         LTIME_S(inode->i_mtime) = 0;
1713         LTIME_S(inode->i_atime) = 0;
1714         LTIME_S(inode->i_ctime) = 0;
1715         inode->i_rdev = 0;
1716         ll_update_inode(inode, md);
1717
1718         /* OIDEBUG(inode); */
1719
1720         if (S_ISREG(inode->i_mode)) {
1721                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1722                 inode->i_op = &ll_file_inode_operations;
1723                 inode->i_fop = sbi->ll_fop;
1724                 inode->i_mapping->a_ops = &ll_aops;
1725                 EXIT;
1726         } else if (S_ISDIR(inode->i_mode)) {
1727                 inode->i_op = &ll_dir_inode_operations;
1728                 inode->i_fop = &ll_dir_operations;
1729                 inode->i_mapping->a_ops = &ll_dir_aops;
1730                 EXIT;
1731         } else if (S_ISLNK(inode->i_mode)) {
1732                 inode->i_op = &ll_fast_symlink_inode_operations;
1733                 EXIT;
1734         } else {
1735                 inode->i_op = &ll_special_inode_operations;
1736
1737                 init_special_inode(inode, inode->i_mode,
1738                                    kdev_t_to_nr(inode->i_rdev));
1739
1740                 /* initializing backing dev info. */
1741                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1742
1743                 EXIT;
1744         }
1745 }
1746
1747 void ll_delete_inode(struct inode *inode)
1748 {
1749         struct ll_sb_info *sbi = ll_i2sbi(inode);
1750         int rc;
1751         ENTRY;
1752
1753         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1754         if (rc) {
1755                 CERROR("fid_delete() failed, rc %d\n", rc);
1756         }
1757         truncate_inode_pages(&inode->i_data, 0);
1758         clear_inode(inode);
1759
1760         EXIT;
1761 }
1762
1763 int ll_iocontrol(struct inode *inode, struct file *file,
1764                  unsigned int cmd, unsigned long arg)
1765 {
1766         struct ll_sb_info *sbi = ll_i2sbi(inode);
1767         struct ptlrpc_request *req = NULL;
1768         int rc, flags = 0;
1769         ENTRY;
1770
1771         switch(cmd) {
1772         case EXT3_IOC_GETFLAGS: {
1773                 struct mdt_body *body;
1774                 struct obd_capa *oc;
1775
1776                 oc = ll_mdscapa_get(inode);
1777                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1778                                 OBD_MD_FLFLAGS, 0, &req);
1779                 capa_put(oc);
1780                 if (rc) {
1781                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1782                         RETURN(-abs(rc));
1783                 }
1784
1785                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1786
1787                 flags = body->flags;
1788
1789                 ptlrpc_req_finished(req);
1790
1791                 RETURN(put_user(flags, (int *)arg));
1792         }
1793         case EXT3_IOC_SETFLAGS: {
1794                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1795                 struct obd_info oinfo = { { { 0 } } };
1796                 struct md_op_data *op_data;
1797
1798                 if (get_user(flags, (int *)arg))
1799                         RETURN(-EFAULT);
1800
1801                 oinfo.oi_md = lsm;
1802                 OBDO_ALLOC(oinfo.oi_oa);
1803                 if (!oinfo.oi_oa)
1804                         RETURN(-ENOMEM);
1805
1806                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1807                                              LUSTRE_OPC_ANY, NULL);
1808                 if (IS_ERR(op_data))
1809                         RETURN(PTR_ERR(op_data));
1810
1811                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1812                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1813                 rc = md_setattr(sbi->ll_md_exp, op_data,
1814                                 NULL, 0, NULL, 0, &req, NULL);
1815                 ll_finish_md_op_data(op_data);
1816                 ptlrpc_req_finished(req);
1817                 if (rc || lsm == NULL) {
1818                         OBDO_FREE(oinfo.oi_oa);
1819                         RETURN(rc);
1820                 }
1821
1822                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1823                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1824                 oinfo.oi_oa->o_flags = flags;
1825                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1826                                        OBD_MD_FLGROUP;
1827                 oinfo.oi_capa = ll_mdscapa_get(inode);
1828
1829                 obdo_from_inode(oinfo.oi_oa, inode,
1830                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1831                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1832                 capa_put(oinfo.oi_capa);
1833                 OBDO_FREE(oinfo.oi_oa);
1834                 if (rc) {
1835                         if (rc != -EPERM && rc != -EACCES)
1836                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1837                         RETURN(rc);
1838                 }
1839
1840                 inode->i_flags = ll_ext_to_inode_flags(flags |
1841                                                        MDS_BFLAG_EXT_FLAGS);
1842                 RETURN(0);
1843         }
1844         default:
1845                 RETURN(-ENOSYS);
1846         }
1847
1848         RETURN(0);
1849 }
1850
1851 int ll_flush_ctx(struct inode *inode)
1852 {
1853         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1854
1855         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1856
1857         obd_set_info_async(sbi->ll_md_exp,
1858                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1859                            0, NULL, NULL);
1860         obd_set_info_async(sbi->ll_dt_exp,
1861                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1862                            0, NULL, NULL);
1863         return 0;
1864 }
1865
1866 /* umount -f client means force down, don't save state */
1867 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1868 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1869 {
1870         struct super_block *sb = vfsmnt->mnt_sb;
1871 #else
1872 void ll_umount_begin(struct super_block *sb)
1873 {
1874 #endif
1875         struct lustre_sb_info *lsi = s2lsi(sb);
1876         struct ll_sb_info *sbi = ll_s2sbi(sb);
1877         struct obd_device *obd;
1878         struct obd_ioctl_data ioc_data = { 0 };
1879         ENTRY;
1880
1881 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1882         if (!(flags & MNT_FORCE)) {
1883                 EXIT;
1884                 return;
1885         }
1886 #endif
1887
1888         /* Tell the MGC we got umount -f */
1889         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1890
1891         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1892                sb->s_count, atomic_read(&sb->s_active));
1893
1894         obd = class_exp2obd(sbi->ll_md_exp);
1895         if (obd == NULL) {
1896                 CERROR("Invalid MDC connection handle "LPX64"\n",
1897                        sbi->ll_md_exp->exp_handle.h_cookie);
1898                 EXIT;
1899                 return;
1900         }
1901         obd->obd_force = 1;
1902         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1903                       &ioc_data, NULL);
1904
1905         obd = class_exp2obd(sbi->ll_dt_exp);
1906         if (obd == NULL) {
1907                 CERROR("Invalid LOV connection handle "LPX64"\n",
1908                        sbi->ll_dt_exp->exp_handle.h_cookie);
1909                 EXIT;
1910                 return;
1911         }
1912
1913         obd->obd_force = 1;
1914         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1915                       &ioc_data, NULL);
1916
1917         /* Really, we'd like to wait until there are no requests outstanding,
1918          * and then continue.  For now, we just invalidate the requests,
1919          * schedule() and sleep one second if needed, and hope.
1920          */
1921         schedule();
1922 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1923         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1924                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1925                                      cfs_time_seconds(1));
1926                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1927                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1928                                       "may try to umount it a bit later\n",
1929                                       atomic_read(&vfsmnt->mnt_count));
1930         }
1931 #endif
1932
1933         EXIT;
1934 }
1935
1936 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1937 {
1938         struct ll_sb_info *sbi = ll_s2sbi(sb);
1939         int err;
1940         __u32 read_only;
1941
1942         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1943                 read_only = *flags & MS_RDONLY;
1944                 err = obd_set_info_async(sbi->ll_md_exp,
1945                                          sizeof(KEY_READ_ONLY),
1946                                          KEY_READ_ONLY, sizeof(read_only),
1947                                          &read_only, NULL);
1948                 if (err) {
1949                         CERROR("Failed to change the read-only flag during "
1950                                "remount: %d\n", err);
1951                         return err;
1952                 }
1953
1954                 if (read_only)
1955                         sb->s_flags |= MS_RDONLY;
1956                 else
1957                         sb->s_flags &= ~MS_RDONLY;
1958         }
1959         return 0;
1960 }
1961
1962 int ll_prep_inode(struct inode **inode,
1963                   struct ptlrpc_request *req,
1964                   struct super_block *sb)
1965 {
1966         struct ll_sb_info *sbi = NULL;
1967         struct lustre_md md;
1968         int rc;
1969         ENTRY;
1970
1971         LASSERT(*inode || sb);
1972         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1973         prune_deathrow(sbi, 1);
1974         memset(&md, 0, sizeof(struct lustre_md));
1975
1976         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1977                               sbi->ll_md_exp, &md);
1978         if (rc)
1979                 RETURN(rc);
1980
1981         if (*inode) {
1982                 ll_update_inode(*inode, &md);
1983         } else {
1984                 LASSERT(sb != NULL);
1985
1986                 /*
1987                  * At this point server returns to client's same fid as client
1988                  * generated for creating. So using ->fid1 is okay here.
1989                  */
1990                 LASSERT(fid_is_sane(&md.body->fid1));
1991
1992                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
1993                 if (*inode == NULL || IS_ERR(*inode)) {
1994                         if (md.lsm)
1995                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1996 #ifdef CONFIG_FS_POSIX_ACL
1997                         if (md.posix_acl) {
1998                                 posix_acl_release(md.posix_acl);
1999                                 md.posix_acl = NULL;
2000                         }
2001 #endif
2002                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2003                         *inode = NULL;
2004                         CERROR("new_inode -fatal: rc %d\n", rc);
2005                         GOTO(out, rc);
2006                 }
2007         }
2008
2009         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2010                          ll_i2info(*inode)->lli_smd);
2011 out:
2012         md_free_lustre_md(sbi->ll_md_exp, &md);
2013         RETURN(rc);
2014 }
2015
2016 int ll_obd_statfs(struct inode *inode, void *arg)
2017 {
2018         struct ll_sb_info *sbi = NULL;
2019         struct obd_export *exp;
2020         char *buf = NULL;
2021         struct obd_ioctl_data *data = NULL;
2022         __u32 type;
2023         int len = 0, rc;
2024
2025         if (!inode || !(sbi = ll_i2sbi(inode)))
2026                 GOTO(out_statfs, rc = -EINVAL);
2027
2028         rc = obd_ioctl_getdata(&buf, &len, arg);
2029         if (rc)
2030                 GOTO(out_statfs, rc);
2031
2032         data = (void*)buf;
2033         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2034             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2035                 GOTO(out_statfs, rc = -EINVAL);
2036
2037         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2038         if (type == LL_STATFS_MDC)
2039                 exp = sbi->ll_md_exp;
2040         else if (type == LL_STATFS_LOV)
2041                 exp = sbi->ll_dt_exp;
2042         else
2043                 GOTO(out_statfs, rc = -ENODEV);
2044
2045         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2046         if (rc)
2047                 GOTO(out_statfs, rc);
2048 out_statfs:
2049         if (buf)
2050                 obd_ioctl_freedata(buf, len);
2051         return rc;
2052 }
2053
2054 int ll_process_config(struct lustre_cfg *lcfg)
2055 {
2056         char *ptr;
2057         void *sb;
2058         struct lprocfs_static_vars lvars;
2059         unsigned long x;
2060         int rc = 0;
2061
2062         lprocfs_llite_init_vars(&lvars);
2063
2064         /* The instance name contains the sb: lustre-client-aacfe000 */
2065         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2066         if (!ptr || !*(++ptr))
2067                 return -EINVAL;
2068         if (sscanf(ptr, "%lx", &x) != 1)
2069                 return -EINVAL;
2070         sb = (void *)x;
2071         /* This better be a real Lustre superblock! */
2072         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2073
2074         /* Note we have not called client_common_fill_super yet, so
2075            proc fns must be able to handle that! */
2076         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2077                                       lcfg, sb);
2078         if (rc > 0)
2079                 rc = 0;
2080         return(rc);
2081 }
2082
2083 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2084 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2085                                        struct inode *i1, struct inode *i2,
2086                                        const char *name, int namelen,
2087                                        int mode, __u32 opc, void *data)
2088 {
2089         LASSERT(i1 != NULL);
2090
2091         if (namelen > ll_i2sbi(i1)->ll_namelen)
2092                 return ERR_PTR(-ENAMETOOLONG);
2093
2094         if (op_data == NULL)
2095                 OBD_ALLOC_PTR(op_data);
2096
2097         if (op_data == NULL)
2098                 return ERR_PTR(-ENOMEM);
2099
2100         ll_i2gids(op_data->op_suppgids, i1, i2);
2101         op_data->op_fid1 = *ll_inode2fid(i1);
2102         op_data->op_capa1 = ll_mdscapa_get(i1);
2103
2104         if (i2) {
2105                 op_data->op_fid2 = *ll_inode2fid(i2);
2106                 op_data->op_capa2 = ll_mdscapa_get(i2);
2107         } else {
2108                 fid_zero(&op_data->op_fid2);
2109                 op_data->op_capa2 = NULL;
2110         }
2111
2112         op_data->op_name = name;
2113         op_data->op_namelen = namelen;
2114         op_data->op_mode = mode;
2115         op_data->op_mod_time = cfs_time_current_sec();
2116         op_data->op_fsuid = current->fsuid;
2117         op_data->op_fsgid = current->fsgid;
2118         op_data->op_cap = cfs_curproc_cap_pack();
2119         op_data->op_bias = MDS_CHECK_SPLIT;
2120         op_data->op_opc = opc;
2121         op_data->op_mds = 0;
2122         op_data->op_data = data;
2123
2124         return op_data;
2125 }
2126
2127 void ll_finish_md_op_data(struct md_op_data *op_data)
2128 {
2129         capa_put(op_data->op_capa1);
2130         capa_put(op_data->op_capa2);
2131         OBD_FREE_PTR(op_data);
2132 }
2133
2134 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2135 {
2136         struct ll_sb_info *sbi;
2137
2138         LASSERT((seq != NULL) && (vfs != NULL));
2139         sbi = ll_s2sbi(vfs->mnt_sb);
2140
2141         if (sbi->ll_flags & LL_SBI_NOLCK)
2142                 seq_puts(seq, ",nolock");
2143
2144         if (sbi->ll_flags & LL_SBI_FLOCK)
2145                 seq_puts(seq, ",flock");
2146
2147         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2148                 seq_puts(seq, ",localflock");
2149
2150         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2151                 seq_puts(seq, ",user_xattr");
2152
2153         if (sbi->ll_flags & LL_SBI_ACL)
2154                 seq_puts(seq, ",acl");
2155
2156         RETURN(0);
2157 }