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         init_mutex(&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         mutex_down(&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         mutex_up(&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         struct dentry *dentry;
 534         int dentry_count = 0;
 535
 536         LASSERT(inode != NULL);
 537         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u state %lx\n",
 538                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 539                inode->i_mode, atomic_read(&inode->i_count), inode->i_state);
 540
 541         list_for_each(tmp, &inode->i_dentry) {
 542                 dentry = list_entry(tmp, struct dentry, d_alias);
 543                 CERROR("Alias[%d] dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u, "
 544                        "flags=0x%x, fsdata=%p\n", dentry_count, dentry,
 545                        dentry->d_name.len, dentry->d_name.name,
 546                        dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 547                        dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 548                        dentry->d_flags, dentry->d_fsdata);
 549
 550                 dentry_count++;
 551         }
 552
 553 }
 554
 555 void lustre_dump_dentry(struct dentry *dentry, int recur)
 556 {
 557         struct list_head *tmp;
 558         int subdirs = 0;
 559
 560         LASSERT(dentry != NULL);
 561
 562         list_for_each(tmp, &dentry->d_subdirs)
 563                 subdirs++;
 564
 565         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 566                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 567                dentry->d_name.len, dentry->d_name.name,
 568                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 569                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 570                dentry->d_flags, dentry->d_fsdata, subdirs);
 571         if (dentry->d_inode != NULL)
 572                 ll_dump_inode(dentry->d_inode);
 573
 574         if (recur == 0)
 575                 return;
 576
 577         list_for_each(tmp, &dentry->d_subdirs) {
 578                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 579                 lustre_dump_dentry(d, recur - 1);
 580         }
 581 }
 582
 583 #ifdef HAVE_EXPORT___IGET
 584 static void prune_dir_dentries(struct inode *inode)
 585 {
 586         struct dentry *dentry, *prev = NULL;
 587
 588         /* due to lustre specific logic, a directory
 589          * can have few dentries - a bug from VFS POV */
 590 restart:
 591         spin_lock(&dcache_lock);
 592         if (!list_empty(&inode->i_dentry)) {
 593                 dentry = list_entry(inode->i_dentry.prev,
 594                                     struct dentry, d_alias);
 595                 /* in order to prevent infinite loops we
 596                  * break if previous dentry is busy */
 597                 if (dentry != prev) {
 598                         prev = dentry;
 599                         dget_locked(dentry);
 600                         spin_unlock(&dcache_lock);
 601
 602                         /* try to kill all child dentries */
 603                         lock_dentry(dentry);
 604                         shrink_dcache_parent(dentry);
 605                         unlock_dentry(dentry);
 606                         dput(dentry);
 607
 608                         /* now try to get rid of current dentry */
 609                         d_prune_aliases(inode);
 610                         goto restart;
 611                 }
 612         }
 613         spin_unlock(&dcache_lock);
 614 }
 615
 616 static void prune_deathrow_one(struct ll_inode_info *lli)
 617 {
 618         struct inode *inode = ll_info2i(lli);
 619
 620         /* first, try to drop any dentries - they hold a ref on the inode */
 621         if (S_ISDIR(inode->i_mode))
 622                 prune_dir_dentries(inode);
 623         else
 624                 d_prune_aliases(inode);
 625
 626
 627         /* if somebody still uses it, leave it */
 628         LASSERT(atomic_read(&inode->i_count) > 0);
 629         if (atomic_read(&inode->i_count) > 1)
 630                 goto out;
 631
 632         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 633                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 634
 635         /* seems nobody uses it anymore */
 636         inode->i_nlink = 0;
 637
 638 out:
 639         iput(inode);
 640         return;
 641 }
 642
 643 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 644 {
 645         struct ll_inode_info *lli;
 646         int empty;
 647
 648         do {
 649                 if (need_resched() && try)
 650                         break;
 651
 652                 if (try) {
 653                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 654                                 break;
 655                 } else {
 656                         spin_lock(&sbi->ll_deathrow_lock);
 657                 }
 658
 659                 empty = 1;
 660                 lli = NULL;
 661                 if (!list_empty(&sbi->ll_deathrow)) {
 662                         lli = list_entry(sbi->ll_deathrow.next,
 663                                          struct ll_inode_info,
 664                                          lli_dead_list);
 665                         list_del_init(&lli->lli_dead_list);
 666                         if (!list_empty(&sbi->ll_deathrow))
 667                                 empty = 0;
 668                 }
 669                 spin_unlock(&sbi->ll_deathrow_lock);
 670
 671                 if (lli)
 672                         prune_deathrow_one(lli);
 673
 674         } while (empty == 0);
 675 }
 676 #else /* !HAVE_EXPORT___IGET */
 677 #define prune_deathrow(sbi, try) do {} while (0)
 678 #endif /* HAVE_EXPORT___IGET */
 679
 680 void client_common_put_super(struct super_block *sb)
 681 {
 682         struct ll_sb_info *sbi = ll_s2sbi(sb);
 683         ENTRY;
 684
 685 #ifdef CONFIG_FS_POSIX_ACL
 686         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 687                 et_fini(&sbi->ll_et);
 688                 rct_fini(&sbi->ll_rct);
 689         }
 690 #endif
 691
 692         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 693
 694         ll_close_thread_shutdown(sbi->ll_lcq);
 695
 696         cl_sb_fini(sb);
 697
 698         /* destroy inodes in deathrow */
 699         prune_deathrow(sbi, 0);
 700
 701         list_del(&sbi->ll_conn_chain);
 702
 703         obd_fid_fini(sbi->ll_dt_exp);
 704         obd_disconnect(sbi->ll_dt_exp);
 705         sbi->ll_dt_exp = NULL;
 706
 707         lprocfs_unregister_mountpoint(sbi);
 708
 709         obd_fid_fini(sbi->ll_md_exp);
 710         obd_disconnect(sbi->ll_md_exp);
 711         sbi->ll_md_exp = NULL;
 712
 713         EXIT;
 714 }
 715
 716 void ll_kill_super(struct super_block *sb)
 717 {
 718         struct ll_sb_info *sbi;
 719
 720         ENTRY;
 721
 722         /* not init sb ?*/
 723         if (!(sb->s_flags & MS_ACTIVE))
 724                 return;
 725
 726         sbi = ll_s2sbi(sb);
 727         /* we need restore s_dev from changed for clustred NFS before put_super
 728          * because new kernels have cached s_dev and change sb->s_dev in
 729          * put_super not affected real removing devices */
 730         if (sbi)
 731                 sb->s_dev = sbi->ll_sdev_orig;
 732         EXIT;
 733 }
 734
 735 char *ll_read_opt(const char *opt, char *data)
 736 {
 737         char *value;
 738         char *retval;
 739         ENTRY;
 740
 741         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 742         if (strncmp(opt, data, strlen(opt)))
 743                 RETURN(NULL);
 744         if ((value = strchr(data, '=')) == NULL)
 745                 RETURN(NULL);
 746
 747         value++;
 748         OBD_ALLOC(retval, strlen(value) + 1);
 749         if (!retval) {
 750                 CERROR("out of memory!\n");
 751                 RETURN(NULL);
 752         }
 753
 754         memcpy(retval, value, strlen(value)+1);
 755         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 756         RETURN(retval);
 757 }
 758
 759 static inline int ll_set_opt(const char *opt, char *data, int fl)
 760 {
 761         if (strncmp(opt, data, strlen(opt)) != 0)
 762                 return(0);
 763         else
 764                 return(fl);
 765 }
 766
 767 /* non-client-specific mount options are parsed in lmd_parse */
 768 static int ll_options(char *options, int *flags)
 769 {
 770         int tmp;
 771         char *s1 = options, *s2;
 772         ENTRY;
 773
 774         if (!options)
 775                 RETURN(0);
 776
 777         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 778
 779         while (*s1) {
 780                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 781                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 787                 if (tmp) {
 788                         *flags |= tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 797                 if (tmp) {
 798                         *flags &= ~tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 802                 if (tmp) {
 803                         *flags |= tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 807                 if (tmp) {
 808                         *flags &= ~tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 812                 if (tmp) {
 813                         /* Ignore deprecated mount option.  The client will
 814                          * always try to mount with ACL support, whether this
 815                          * is used depends on whether server supports it. */
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 819                 if (tmp) {
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827
 828                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 849                 if (tmp) {
 850                         *flags |= tmp;
 851                         goto next;
 852                 }
 853                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 854                 if (tmp) {
 855                         *flags &= ~tmp;
 856                         goto next;
 857                 }
 858
 859                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 860                                    s1);
 861                 RETURN(-EINVAL);
 862
 863 next:
 864                 /* Find next opt */
 865                 s2 = strchr(s1, ',');
 866                 if (s2 == NULL)
 867                         break;
 868                 s1 = s2 + 1;
 869         }
 870         RETURN(0);
 871 }
 872
 873 void ll_lli_init(struct ll_inode_info *lli)
 874 {
 875         lli->lli_inode_magic = LLI_INODE_MAGIC;
 876         sema_init(&lli->lli_size_sem, 1);
 877         sema_init(&lli->lli_write_sem, 1);
 878         lli->lli_flags = 0;
 879         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 880         spin_lock_init(&lli->lli_lock);
 881         INIT_LIST_HEAD(&lli->lli_close_list);
 882         lli->lli_inode_magic = LLI_INODE_MAGIC;
 883         sema_init(&lli->lli_och_sem, 1);
 884         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 885         lli->lli_mds_exec_och = NULL;
 886         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 887         lli->lli_open_fd_exec_count = 0;
 888         INIT_LIST_HEAD(&lli->lli_dead_list);
 889         lli->lli_remote_perms = NULL;
 890         lli->lli_rmtperm_utime = 0;
 891         sema_init(&lli->lli_rmtperm_sem, 1);
 892         INIT_LIST_HEAD(&lli->lli_oss_capas);
 893 }
 894
 895 int ll_fill_super(struct super_block *sb)
 896 {
 897         struct lustre_profile *lprof;
 898         struct lustre_sb_info *lsi = s2lsi(sb);
 899         struct ll_sb_info *sbi;
 900         char  *dt = NULL, *md = NULL;
 901         char  *profilenm = get_profile_name(sb);
 902         struct config_llog_instance cfg = {0, };
 903         char   ll_instance[sizeof(sb) * 2 + 1];
 904         int    err;
 905         ENTRY;
 906
 907         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 908
 909         cfs_module_get();
 910
 911         /* client additional sb info */
 912         lsi->lsi_llsbi = sbi = ll_init_sbi();
 913         if (!sbi) {
 914                 cfs_module_put();
 915                 RETURN(-ENOMEM);
 916         }
 917
 918         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 919         if (err)
 920                 GOTO(out_free, err);
 921
 922         /* Generate a string unique to this super, in case some joker tries
 923            to mount the same fs at two mount points.
 924            Use the address of the super itself.*/
 925         sprintf(ll_instance, "%p", sb);
 926         cfg.cfg_instance = ll_instance;
 927         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 928
 929         /* set up client obds */
 930         err = lustre_process_log(sb, profilenm, &cfg);
 931         if (err < 0) {
 932                 CERROR("Unable to process log: %d\n", err);
 933                 GOTO(out_free, err);
 934         }
 935
 936         lprof = class_get_profile(profilenm);
 937         if (lprof == NULL) {
 938                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 939                                    " read from the MGS.  Does that filesystem "
 940                                    "exist?\n", profilenm);
 941                 GOTO(out_free, err = -EINVAL);
 942         }
 943         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 944                lprof->lp_md, lprof->lp_dt);
 945
 946         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 947                   strlen(ll_instance) + 2);
 948         if (!dt)
 949                 GOTO(out_free, err = -ENOMEM);
 950         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 951
 952         OBD_ALLOC(md, strlen(lprof->lp_md) +
 953                   strlen(ll_instance) + 2);
 954         if (!md)
 955                 GOTO(out_free, err = -ENOMEM);
 956         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 957
 958         /* connections, registrations, sb setup */
 959         err = client_common_fill_super(sb, md, dt);
 960
 961 out_free:
 962         if (md)
 963                 OBD_FREE(md, strlen(md) + 1);
 964         if (dt)
 965                 OBD_FREE(dt, strlen(dt) + 1);
 966         if (err)
 967                 ll_put_super(sb);
 968         else
 969                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 970
 971         RETURN(err);
 972 } /* ll_fill_super */
 973
 974
 975 void lu_context_keys_dump(void);
 976
 977 void ll_put_super(struct super_block *sb)
 978 {
 979         struct config_llog_instance cfg;
 980         char   ll_instance[sizeof(sb) * 2 + 1];
 981         struct obd_device *obd;
 982         struct lustre_sb_info *lsi = s2lsi(sb);
 983         struct ll_sb_info *sbi = ll_s2sbi(sb);
 984         char *profilenm = get_profile_name(sb);
 985         int force = 1, next;
 986         ENTRY;
 987
 988         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 989
 990         ll_print_capa_stat(sbi);
 991
 992         sprintf(ll_instance, "%p", sb);
 993         cfg.cfg_instance = ll_instance;
 994         lustre_end_log(sb, NULL, &cfg);
 995
 996         if (sbi->ll_md_exp) {
 997                 obd = class_exp2obd(sbi->ll_md_exp);
 998                 if (obd)
 999                         force = obd->obd_force;
1000         }
1001
1002         /* We need to set force before the lov_disconnect in
1003            lustre_common_put_super, since l_d cleans up osc's as well. */
1004         if (force) {
1005                 next = 0;
1006                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1007                                                      &next)) != NULL) {
1008                         obd->obd_force = force;
1009                 }
1010         }
1011
1012         if (sbi->ll_lcq) {
1013                 /* Only if client_common_fill_super succeeded */
1014                 client_common_put_super(sb);
1015         }
1016
1017         next = 0;
1018         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1019                 class_manual_cleanup(obd);
1020         }
1021
1022         if (profilenm)
1023                 class_del_profile(profilenm);
1024
1025         ll_free_sbi(sb);
1026         lsi->lsi_llsbi = NULL;
1027
1028         lustre_common_put_super(sb);
1029
1030         cl_env_cache_purge(~0);
1031
1032         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1033
1034         cfs_module_put();
1035
1036         EXIT;
1037 } /* client_put_super */
1038
1039 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1040 {
1041         struct inode *inode = NULL;
1042         /* NOTE: we depend on atomic igrab() -bzzz */
1043         lock_res_and_lock(lock);
1044         if (lock->l_ast_data) {
1045                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1046                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1047                         inode = igrab(lock->l_ast_data);
1048                 } else {
1049                         inode = lock->l_ast_data;
1050                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1051                                                 D_INFO : D_WARNING,
1052                                         lock, __FILE__, __func__, __LINE__,
1053                                         "l_ast_data %p is bogus: magic %08x",
1054                                         lock->l_ast_data, lli->lli_inode_magic);
1055                         inode = NULL;
1056                 }
1057         }
1058         unlock_res_and_lock(lock);
1059         return inode;
1060 }
1061
1062 static int null_if_equal(struct ldlm_lock *lock, void *data)
1063 {
1064         if (data == lock->l_ast_data) {
1065                 lock->l_ast_data = NULL;
1066
1067                 if (lock->l_req_mode != lock->l_granted_mode)
1068                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1069         }
1070
1071         return LDLM_ITER_CONTINUE;
1072 }
1073
1074 void ll_clear_inode(struct inode *inode)
1075 {
1076         struct ll_inode_info *lli = ll_i2info(inode);
1077         struct ll_sb_info *sbi = ll_i2sbi(inode);
1078         ENTRY;
1079
1080         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1081                inode->i_generation, inode);
1082
1083         if (S_ISDIR(inode->i_mode)) {
1084                 /* these should have been cleared in ll_file_release */
1085                 LASSERT(lli->lli_sai == NULL);
1086                 LASSERT(lli->lli_opendir_key == NULL);
1087                 LASSERT(lli->lli_opendir_pid == 0);
1088         }
1089
1090         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1091         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1092                          null_if_equal, inode);
1093
1094         LASSERT(!lli->lli_open_fd_write_count);
1095         LASSERT(!lli->lli_open_fd_read_count);
1096         LASSERT(!lli->lli_open_fd_exec_count);
1097
1098         if (lli->lli_mds_write_och)
1099                 ll_md_real_close(inode, FMODE_WRITE);
1100         if (lli->lli_mds_exec_och)
1101                 ll_md_real_close(inode, FMODE_EXEC);
1102         if (lli->lli_mds_read_och)
1103                 ll_md_real_close(inode, FMODE_READ);
1104
1105         if (lli->lli_symlink_name) {
1106                 OBD_FREE(lli->lli_symlink_name,
1107                          strlen(lli->lli_symlink_name) + 1);
1108                 lli->lli_symlink_name = NULL;
1109         }
1110
1111         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1112                 LASSERT(lli->lli_posix_acl == NULL);
1113                 if (lli->lli_remote_perms) {
1114                         free_rmtperm_hash(lli->lli_remote_perms);
1115                         lli->lli_remote_perms = NULL;
1116                 }
1117         }
1118 #ifdef CONFIG_FS_POSIX_ACL
1119         else if (lli->lli_posix_acl) {
1120                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1121                 LASSERT(lli->lli_remote_perms == NULL);
1122                 posix_acl_release(lli->lli_posix_acl);
1123                 lli->lli_posix_acl = NULL;
1124         }
1125 #endif
1126         lli->lli_inode_magic = LLI_INODE_DEAD;
1127
1128 #ifdef HAVE_EXPORT___IGET
1129         spin_lock(&sbi->ll_deathrow_lock);
1130         list_del_init(&lli->lli_dead_list);
1131         spin_unlock(&sbi->ll_deathrow_lock);
1132 #endif
1133         ll_clear_inode_capas(inode);
1134         /*
1135          * XXX This has to be done before lsm is freed below, because
1136          * cl_object still uses inode lsm.
1137          */
1138         cl_inode_fini(inode);
1139
1140         if (lli->lli_smd) {
1141                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1142                 lli->lli_smd = NULL;
1143         }
1144
1145
1146         EXIT;
1147 }
1148
1149 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1150                   struct md_open_data **mod)
1151 {
1152         struct lustre_md md;
1153         struct ll_sb_info *sbi = ll_i2sbi(inode);
1154         struct ptlrpc_request *request = NULL;
1155         int rc;
1156         ENTRY;
1157
1158         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1159                                      LUSTRE_OPC_ANY, NULL);
1160         if (IS_ERR(op_data))
1161                 RETURN(PTR_ERR(op_data));
1162
1163         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1164                         &request, mod);
1165         if (rc) {
1166                 ptlrpc_req_finished(request);
1167                 if (rc == -ENOENT) {
1168                         inode->i_nlink = 0;
1169                         /* Unlinked special device node? Or just a race?
1170                          * Pretend we done everything. */
1171                         if (!S_ISREG(inode->i_mode) &&
1172                             !S_ISDIR(inode->i_mode))
1173                                 rc = inode_setattr(inode, &op_data->op_attr);
1174                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1175                         CERROR("md_setattr fails: rc = %d\n", rc);
1176                 }
1177                 RETURN(rc);
1178         }
1179
1180         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1181                               sbi->ll_md_exp, &md);
1182         if (rc) {
1183                 ptlrpc_req_finished(request);
1184                 RETURN(rc);
1185         }
1186
1187         /* We call inode_setattr to adjust timestamps.
1188          * If there is at least some data in file, we cleared ATTR_SIZE
1189          * above to avoid invoking vmtruncate, otherwise it is important
1190          * to call vmtruncate in inode_setattr to update inode->i_size
1191          * (bug 6196) */
1192         rc = inode_setattr(inode, &op_data->op_attr);
1193
1194         /* Extract epoch data if obtained. */
1195         op_data->op_handle = md.body->handle;
1196         op_data->op_ioepoch = md.body->ioepoch;
1197
1198         ll_update_inode(inode, &md);
1199         ptlrpc_req_finished(request);
1200
1201         RETURN(rc);
1202 }
1203
1204 /* Close IO epoch and send Size-on-MDS attribute update. */
1205 static int ll_setattr_done_writing(struct inode *inode,
1206                                    struct md_op_data *op_data,
1207                                    struct md_open_data *mod)
1208 {
1209         struct ll_inode_info *lli = ll_i2info(inode);
1210         int rc = 0;
1211         ENTRY;
1212
1213         LASSERT(op_data != NULL);
1214         if (!S_ISREG(inode->i_mode))
1215                 RETURN(0);
1216
1217         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1218                op_data->op_ioepoch, PFID(&lli->lli_fid));
1219
1220         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1221         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1222         if (rc == -EAGAIN) {
1223                 /* MDS has instructed us to obtain Size-on-MDS attribute
1224                  * from OSTs and send setattr to back to MDS. */
1225                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1226                                          op_data->op_ioepoch);
1227         } else if (rc) {
1228                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1229                        inode->i_ino, rc);
1230         }
1231         RETURN(rc);
1232 }
1233
1234 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1235 {
1236         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1237         int rc;
1238
1239         rc = cl_setattr_do_truncate(inode, size, capa);
1240         ll_truncate_free_capa(capa);
1241         return rc;
1242 }
1243
1244 static int ll_setattr_ost(struct inode *inode)
1245 {
1246         struct obd_capa *capa = ll_mdscapa_get(inode);
1247         int rc;
1248
1249         rc = cl_setattr_ost(inode, capa);
1250         capa_put(capa);
1251
1252         return rc;
1253 }
1254
1255 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1256  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1257  * keep these values until such a time that objects are allocated for it.
1258  * We do the MDS operations first, as it is checking permissions for us.
1259  * We don't to the MDS RPC if there is nothing that we want to store there,
1260  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1261  * going to do an RPC anyways.
1262  *
1263  * If we are doing a truncate, we will send the mtime and ctime updates
1264  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1265  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1266  * at the same time.
1267  */
1268 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1269 {
1270         struct ll_inode_info *lli = ll_i2info(inode);
1271         struct lov_stripe_md *lsm = lli->lli_smd;
1272         struct md_op_data *op_data = NULL;
1273         struct md_open_data *mod = NULL;
1274         int ia_valid = attr->ia_valid;
1275         int rc = 0, rc1 = 0;
1276         ENTRY;
1277
1278         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1279                attr->ia_valid);
1280         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1281
1282         if (ia_valid & ATTR_SIZE) {
1283                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1284                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1285                                attr->ia_size, ll_file_maxbytes(inode));
1286                         RETURN(-EFBIG);
1287                 }
1288
1289                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1290         }
1291
1292         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1293         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1294                 if (current->fsuid != inode->i_uid &&
1295                     !cfs_capable(CFS_CAP_FOWNER))
1296                         RETURN(-EPERM);
1297         }
1298
1299         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1300         if (attr->ia_valid & ATTR_CTIME) {
1301                 attr->ia_ctime = CURRENT_TIME;
1302                 attr->ia_valid |= ATTR_CTIME_SET;
1303         }
1304         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1305                 attr->ia_atime = CURRENT_TIME;
1306                 attr->ia_valid |= ATTR_ATIME_SET;
1307         }
1308         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1309                 attr->ia_mtime = CURRENT_TIME;
1310                 attr->ia_valid |= ATTR_MTIME_SET;
1311         }
1312         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1313                 /* To avoid stale mtime on mds, obtain it from ost and send
1314                    to mds. */
1315                 rc = cl_glimpse_size(inode);
1316                 if (rc)
1317                         RETURN(rc);
1318
1319                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1320                 attr->ia_mtime = inode->i_mtime;
1321         }
1322
1323         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1324                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1325                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1326                        cfs_time_current_sec());
1327
1328         /* NB: ATTR_SIZE will only be set after this point if the size
1329          * resides on the MDS, ie, this file has no objects. */
1330         if (lsm)
1331                 attr->ia_valid &= ~ATTR_SIZE;
1332
1333         /* We always do an MDS RPC, even if we're only changing the size;
1334          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1335
1336         OBD_ALLOC_PTR(op_data);
1337         if (op_data == NULL)
1338                 RETURN(-ENOMEM);
1339
1340         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1341
1342         /* Open epoch for truncate. */
1343         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1344             (ia_valid & ATTR_SIZE))
1345                 op_data->op_flags = MF_EPOCH_OPEN;
1346
1347         rc = ll_md_setattr(inode, op_data, &mod);
1348         if (rc)
1349                 GOTO(out, rc);
1350
1351         ll_ioepoch_open(lli, op_data->op_ioepoch);
1352         if (!lsm || !S_ISREG(inode->i_mode)) {
1353                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1354                 GOTO(out, rc = 0);
1355         }
1356
1357         if (ia_valid & ATTR_SIZE)
1358                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1359         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1360                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1361                        inode->i_ino, LTIME_S(attr->ia_mtime));
1362                 rc = ll_setattr_ost(inode);
1363         }
1364         EXIT;
1365 out:
1366         if (op_data) {
1367                 if (op_data->op_ioepoch)
1368                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1369                 ll_finish_md_op_data(op_data);
1370         }
1371         return rc ? rc : rc1;
1372 }
1373
1374 int ll_setattr(struct dentry *de, struct iattr *attr)
1375 {
1376         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1377             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1378                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1379
1380         if ((de->d_inode->i_mode & S_ISUID) &&
1381             !(attr->ia_mode & S_ISUID) &&
1382             !(attr->ia_valid & ATTR_KILL_SUID))
1383                 attr->ia_valid |= ATTR_KILL_SUID;
1384
1385         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1386             !(attr->ia_mode & S_ISGID) &&
1387             !(attr->ia_valid & ATTR_KILL_SGID))
1388                 attr->ia_valid |= ATTR_KILL_SGID;
1389
1390         return ll_setattr_raw(de->d_inode, attr);
1391 }
1392
1393 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1394                        __u64 max_age, __u32 flags)
1395 {
1396         struct ll_sb_info *sbi = ll_s2sbi(sb);
1397         struct obd_statfs obd_osfs;
1398         int rc;
1399         ENTRY;
1400
1401         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1402         if (rc) {
1403                 CERROR("md_statfs fails: rc = %d\n", rc);
1404                 RETURN(rc);
1405         }
1406
1407         osfs->os_type = sb->s_magic;
1408
1409         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1410                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1411
1412         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1413                 flags |= OBD_STATFS_NODELAY;
1414
1415         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1416                               &obd_osfs, max_age, flags);
1417         if (rc) {
1418                 CERROR("obd_statfs fails: rc = %d\n", rc);
1419                 RETURN(rc);
1420         }
1421
1422         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1423                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1424                obd_osfs.os_files);
1425
1426         osfs->os_bsize = obd_osfs.os_bsize;
1427         osfs->os_blocks = obd_osfs.os_blocks;
1428         osfs->os_bfree = obd_osfs.os_bfree;
1429         osfs->os_bavail = obd_osfs.os_bavail;
1430
1431         /* If we don't have as many objects free on the OST as inodes
1432          * on the MDS, we reduce the total number of inodes to
1433          * compensate, so that the "inodes in use" number is correct.
1434          */
1435         if (obd_osfs.os_ffree < osfs->os_ffree) {
1436                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1437                         obd_osfs.os_ffree;
1438                 osfs->os_ffree = obd_osfs.os_ffree;
1439         }
1440
1441         RETURN(rc);
1442 }
1443 #ifndef HAVE_STATFS_DENTRY_PARAM
1444 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1445 {
1446 #else
1447 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1448 {
1449         struct super_block *sb = de->d_sb;
1450 #endif
1451         struct obd_statfs osfs;
1452         int rc;
1453
1454         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1455         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1456
1457         /* For now we will always get up-to-date statfs values, but in the
1458          * future we may allow some amount of caching on the client (e.g.
1459          * from QOS or lprocfs updates). */
1460         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1461         if (rc)
1462                 return rc;
1463
1464         statfs_unpack(sfs, &osfs);
1465
1466         /* We need to downshift for all 32-bit kernels, because we can't
1467          * tell if the kernel is being called via sys_statfs64() or not.
1468          * Stop before overflowing f_bsize - in which case it is better
1469          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1470         if (sizeof(long) < 8) {
1471                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1472                         sfs->f_bsize <<= 1;
1473
1474                         osfs.os_blocks >>= 1;
1475                         osfs.os_bfree >>= 1;
1476                         osfs.os_bavail >>= 1;
1477                 }
1478         }
1479
1480         sfs->f_blocks = osfs.os_blocks;
1481         sfs->f_bfree = osfs.os_bfree;
1482         sfs->f_bavail = osfs.os_bavail;
1483
1484         return 0;
1485 }
1486
1487 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1488 {
1489         struct ll_inode_info *lli;
1490         struct lov_stripe_md *lsm;
1491
1492         lli = ll_i2info(inode);
1493         LASSERT(lli->lli_size_sem_owner != current);
1494         down(&lli->lli_size_sem);
1495         LASSERT(lli->lli_size_sem_owner == NULL);
1496         lli->lli_size_sem_owner = current;
1497         lsm = lli->lli_smd;
1498         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1499                  lsm, lock_lsm);
1500         if (lock_lsm)
1501                 lov_stripe_lock(lsm);
1502 }
1503
1504 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1505 {
1506         struct ll_inode_info *lli;
1507         struct lov_stripe_md *lsm;
1508
1509         lli = ll_i2info(inode);
1510         lsm = lli->lli_smd;
1511         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1512                  lsm, unlock_lsm);
1513         if (unlock_lsm)
1514                 lov_stripe_unlock(lsm);
1515         LASSERT(lli->lli_size_sem_owner == current);
1516         lli->lli_size_sem_owner = NULL;
1517         up(&lli->lli_size_sem);
1518 }
1519
1520 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1521 {
1522         struct ll_inode_info *lli = ll_i2info(inode);
1523
1524         dump_lsm(D_INODE, lsm);
1525         dump_lsm(D_INODE, lli->lli_smd);
1526         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1527                  "lsm must be joined lsm %p\n", lsm);
1528         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1529         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1530                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1531         lli->lli_smd = lsm;
1532         lli->lli_maxbytes = lsm->lsm_maxbytes;
1533         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1534                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1535 }
1536
1537 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1538 {
1539         struct ll_inode_info *lli = ll_i2info(inode);
1540         struct mdt_body *body = md->body;
1541         struct lov_stripe_md *lsm = md->lsm;
1542         struct ll_sb_info *sbi = ll_i2sbi(inode);
1543
1544         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1545         if (lsm != NULL) {
1546                 if (lli->lli_smd == NULL) {
1547                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1548                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1549                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1550                                 dump_lsm(D_ERROR, lsm);
1551                                 LBUG();
1552                         }
1553                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1554                                lsm, inode->i_ino, inode->i_generation, inode);
1555                         cl_inode_init(inode, md);
1556                         /* ll_inode_size_lock() requires it is only
1557                          * called with lli_smd != NULL or lock_lsm == 0
1558                          *  or we can race between lock/unlock.
1559                          *  bug 9547 */
1560                         lli->lli_smd = lsm;
1561                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1562                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1563                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1564                 } else {
1565                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1566                              lli->lli_smd->lsm_stripe_count ==
1567                                         lsm->lsm_stripe_count) {
1568                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1569                                         CERROR("lsm mismatch for inode %ld\n",
1570                                                 inode->i_ino);
1571                                         CERROR("lli_smd:\n");
1572                                         dump_lsm(D_ERROR, lli->lli_smd);
1573                                         CERROR("lsm:\n");
1574                                         dump_lsm(D_ERROR, lsm);
1575                                         LBUG();
1576                                 }
1577                         } else {
1578                                 cl_inode_init(inode, md);
1579                                 ll_replace_lsm(inode, lsm);
1580                         }
1581                 }
1582                 if (lli->lli_smd != lsm)
1583                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1584         }
1585
1586         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1587                 if (body->valid & OBD_MD_FLRMTPERM)
1588                         ll_update_remote_perm(inode, md->remote_perm);
1589         }
1590 #ifdef CONFIG_FS_POSIX_ACL
1591         else if (body->valid & OBD_MD_FLACL) {
1592                 spin_lock(&lli->lli_lock);
1593                 if (lli->lli_posix_acl)
1594                         posix_acl_release(lli->lli_posix_acl);
1595                 lli->lli_posix_acl = md->posix_acl;
1596                 spin_unlock(&lli->lli_lock);
1597         }
1598 #endif
1599         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1600         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1601
1602         if (body->valid & OBD_MD_FLATIME &&
1603             body->atime > LTIME_S(inode->i_atime))
1604                 LTIME_S(inode->i_atime) = body->atime;
1605
1606         /* mtime is always updated with ctime, but can be set in past.
1607            As write and utime(2) may happen within 1 second, and utime's
1608            mtime has a priority over write's one, so take mtime from mds
1609            for the same ctimes. */
1610         if (body->valid & OBD_MD_FLCTIME &&
1611             body->ctime >= LTIME_S(inode->i_ctime)) {
1612                 LTIME_S(inode->i_ctime) = body->ctime;
1613                 if (body->valid & OBD_MD_FLMTIME) {
1614                         CDEBUG(D_INODE, "setting ino %lu mtime "
1615                                "from %lu to "LPU64"\n", inode->i_ino,
1616                                LTIME_S(inode->i_mtime), body->mtime);
1617                         LTIME_S(inode->i_mtime) = body->mtime;
1618                 }
1619         }
1620         if (body->valid & OBD_MD_FLMODE)
1621                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1622         if (body->valid & OBD_MD_FLTYPE)
1623                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1624         LASSERT(inode->i_mode != 0);
1625         if (S_ISREG(inode->i_mode)) {
1626                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1627         } else {
1628                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1629         }
1630 #ifdef HAVE_INODE_BLKSIZE
1631         inode->i_blksize = 1<<inode->i_blkbits;
1632 #endif
1633         if (body->valid & OBD_MD_FLUID)
1634                 inode->i_uid = body->uid;
1635         if (body->valid & OBD_MD_FLGID)
1636                 inode->i_gid = body->gid;
1637         if (body->valid & OBD_MD_FLFLAGS)
1638                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1639         if (body->valid & OBD_MD_FLNLINK)
1640                 inode->i_nlink = body->nlink;
1641         if (body->valid & OBD_MD_FLRDEV)
1642                 inode->i_rdev = old_decode_dev(body->rdev);
1643
1644         if (body->valid & OBD_MD_FLID) {
1645                 /* FID shouldn't be changed! */
1646                 if (fid_is_sane(&lli->lli_fid)) {
1647                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1648                                  "Trying to change FID "DFID
1649                                  " to the "DFID", inode %lu/%u(%p)\n",
1650                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1651                                  inode->i_ino, inode->i_generation, inode);
1652                 } else
1653                         lli->lli_fid = body->fid1;
1654         }
1655
1656         LASSERT(fid_seq(&lli->lli_fid) != 0);
1657
1658         if (body->valid & OBD_MD_FLSIZE) {
1659                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1660                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1661                         struct lustre_handle lockh;
1662                         ldlm_mode_t mode;
1663
1664                         /* As it is possible a blocking ast has been processed
1665                          * by this time, we need to check there is an UPDATE
1666                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1667                          * it. */
1668                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1669                                                &lockh);
1670                         if (mode) {
1671                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1672                                                       LLIF_EPOCH_PENDING |
1673                                                       LLIF_SOM_DIRTY)) {
1674                                         CERROR("ino %lu flags %lu still has "
1675                                                "size authority! do not trust "
1676                                                "the size got from MDS\n",
1677                                                inode->i_ino, lli->lli_flags);
1678                                 } else {
1679                                         /* Use old size assignment to avoid
1680                                          * deadlock bz14138 & bz14326 */
1681                                         inode->i_size = body->size;
1682                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1683                                 }
1684                                 ldlm_lock_decref(&lockh, mode);
1685                         }
1686                 } else {
1687                         /* Use old size assignment to avoid
1688                          * deadlock bz14138 & bz14326 */
1689                         inode->i_size = body->size;
1690                 }
1691
1692                 if (body->valid & OBD_MD_FLBLOCKS)
1693                         inode->i_blocks = body->blocks;
1694         }
1695
1696         if (body->valid & OBD_MD_FLMDSCAPA) {
1697                 LASSERT(md->mds_capa);
1698                 ll_add_capa(inode, md->mds_capa);
1699         }
1700         if (body->valid & OBD_MD_FLOSSCAPA) {
1701                 LASSERT(md->oss_capa);
1702                 ll_add_capa(inode, md->oss_capa);
1703         }
1704 }
1705
1706 static struct backing_dev_info ll_backing_dev_info = {
1707         .ra_pages       = 0,    /* No readahead */
1708 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1709         .capabilities   = 0,    /* Does contribute to dirty memory */
1710 #else
1711         .memory_backed  = 0,    /* Does contribute to dirty memory */
1712 #endif
1713 };
1714
1715 void ll_read_inode2(struct inode *inode, void *opaque)
1716 {
1717         struct lustre_md *md = opaque;
1718         struct ll_inode_info *lli = ll_i2info(inode);
1719         ENTRY;
1720
1721         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1722                inode->i_ino, inode->i_generation, inode);
1723
1724         ll_lli_init(lli);
1725
1726         LASSERT(!lli->lli_smd);
1727
1728         /* Core attributes from the MDS first.  This is a new inode, and
1729          * the VFS doesn't zero times in the core inode so we have to do
1730          * it ourselves.  They will be overwritten by either MDS or OST
1731          * attributes - we just need to make sure they aren't newer. */
1732         LTIME_S(inode->i_mtime) = 0;
1733         LTIME_S(inode->i_atime) = 0;
1734         LTIME_S(inode->i_ctime) = 0;
1735         inode->i_rdev = 0;
1736         ll_update_inode(inode, md);
1737
1738         /* OIDEBUG(inode); */
1739
1740         if (S_ISREG(inode->i_mode)) {
1741                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1742                 inode->i_op = &ll_file_inode_operations;
1743                 inode->i_fop = sbi->ll_fop;
1744                 inode->i_mapping->a_ops = &ll_aops;
1745                 EXIT;
1746         } else if (S_ISDIR(inode->i_mode)) {
1747                 inode->i_op = &ll_dir_inode_operations;
1748                 inode->i_fop = &ll_dir_operations;
1749                 inode->i_mapping->a_ops = &ll_dir_aops;
1750                 EXIT;
1751         } else if (S_ISLNK(inode->i_mode)) {
1752                 inode->i_op = &ll_fast_symlink_inode_operations;
1753                 EXIT;
1754         } else {
1755                 inode->i_op = &ll_special_inode_operations;
1756
1757                 init_special_inode(inode, inode->i_mode,
1758                                    kdev_t_to_nr(inode->i_rdev));
1759
1760                 /* initializing backing dev info. */
1761                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1762
1763                 EXIT;
1764         }
1765 }
1766
1767 void ll_delete_inode(struct inode *inode)
1768 {
1769         struct ll_sb_info *sbi = ll_i2sbi(inode);
1770         int rc;
1771         ENTRY;
1772
1773         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1774         if (rc) {
1775                 CERROR("fid_delete() failed, rc %d\n", rc);
1776         }
1777         truncate_inode_pages(&inode->i_data, 0);
1778         clear_inode(inode);
1779
1780         EXIT;
1781 }
1782
1783 int ll_iocontrol(struct inode *inode, struct file *file,
1784                  unsigned int cmd, unsigned long arg)
1785 {
1786         struct ll_sb_info *sbi = ll_i2sbi(inode);
1787         struct ptlrpc_request *req = NULL;
1788         int rc, flags = 0;
1789         ENTRY;
1790
1791         switch(cmd) {
1792         case FSFILT_IOC_GETFLAGS: {
1793                 struct mdt_body *body;
1794                 struct obd_capa *oc;
1795
1796                 oc = ll_mdscapa_get(inode);
1797                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1798                                 OBD_MD_FLFLAGS, 0, &req);
1799                 capa_put(oc);
1800                 if (rc) {
1801                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1802                         RETURN(-abs(rc));
1803                 }
1804
1805                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1806
1807                 flags = body->flags;
1808
1809                 ptlrpc_req_finished(req);
1810
1811                 RETURN(put_user(flags, (int *)arg));
1812         }
1813         case FSFILT_IOC_SETFLAGS: {
1814                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1815                 struct obd_info oinfo = { { { 0 } } };
1816                 struct md_op_data *op_data;
1817
1818                 if (get_user(flags, (int *)arg))
1819                         RETURN(-EFAULT);
1820
1821                 oinfo.oi_md = lsm;
1822                 OBDO_ALLOC(oinfo.oi_oa);
1823                 if (!oinfo.oi_oa)
1824                         RETURN(-ENOMEM);
1825
1826                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1827                                              LUSTRE_OPC_ANY, NULL);
1828                 if (IS_ERR(op_data))
1829                         RETURN(PTR_ERR(op_data));
1830
1831                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1832                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1833                 rc = md_setattr(sbi->ll_md_exp, op_data,
1834                                 NULL, 0, NULL, 0, &req, NULL);
1835                 ll_finish_md_op_data(op_data);
1836                 ptlrpc_req_finished(req);
1837                 if (rc) {
1838                         OBDO_FREE(oinfo.oi_oa);
1839                         RETURN(rc);
1840                 }
1841
1842                 if (lsm == NULL) {
1843                         OBDO_FREE(oinfo.oi_oa);
1844                         GOTO(update_cache, rc);
1845                 }
1846
1847                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1848                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1849                 oinfo.oi_oa->o_flags = flags;
1850                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1851                                        OBD_MD_FLGROUP;
1852                 oinfo.oi_capa = ll_mdscapa_get(inode);
1853
1854                 obdo_from_inode(oinfo.oi_oa, inode,
1855                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1856                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1857                 capa_put(oinfo.oi_capa);
1858                 OBDO_FREE(oinfo.oi_oa);
1859                 if (rc) {
1860                         if (rc != -EPERM && rc != -EACCES)
1861                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1862                         RETURN(rc);
1863                 }
1864
1865                 EXIT;
1866 update_cache:
1867                 inode->i_flags = ll_ext_to_inode_flags(flags |
1868                                                        MDS_BFLAG_EXT_FLAGS);
1869                 return 0;
1870         }
1871         default:
1872                 RETURN(-ENOSYS);
1873         }
1874
1875         RETURN(0);
1876 }
1877
1878 int ll_flush_ctx(struct inode *inode)
1879 {
1880         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1881
1882         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1883
1884         obd_set_info_async(sbi->ll_md_exp,
1885                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1886                            0, NULL, NULL);
1887         obd_set_info_async(sbi->ll_dt_exp,
1888                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1889                            0, NULL, NULL);
1890         return 0;
1891 }
1892
1893 /* umount -f client means force down, don't save state */
1894 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1895 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1896 {
1897         struct super_block *sb = vfsmnt->mnt_sb;
1898 #else
1899 void ll_umount_begin(struct super_block *sb)
1900 {
1901 #endif
1902         struct lustre_sb_info *lsi = s2lsi(sb);
1903         struct ll_sb_info *sbi = ll_s2sbi(sb);
1904         struct obd_device *obd;
1905         struct obd_ioctl_data ioc_data = { 0 };
1906         ENTRY;
1907
1908 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1909         if (!(flags & MNT_FORCE)) {
1910                 EXIT;
1911                 return;
1912         }
1913 #endif
1914
1915         /* Tell the MGC we got umount -f */
1916         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1917
1918         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1919                sb->s_count, atomic_read(&sb->s_active));
1920
1921         obd = class_exp2obd(sbi->ll_md_exp);
1922         if (obd == NULL) {
1923                 CERROR("Invalid MDC connection handle "LPX64"\n",
1924                        sbi->ll_md_exp->exp_handle.h_cookie);
1925                 EXIT;
1926                 return;
1927         }
1928         obd->obd_force = 1;
1929         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1930                       &ioc_data, NULL);
1931
1932         obd = class_exp2obd(sbi->ll_dt_exp);
1933         if (obd == NULL) {
1934                 CERROR("Invalid LOV connection handle "LPX64"\n",
1935                        sbi->ll_dt_exp->exp_handle.h_cookie);
1936                 EXIT;
1937                 return;
1938         }
1939
1940         obd->obd_force = 1;
1941         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1942                       &ioc_data, NULL);
1943
1944         /* Really, we'd like to wait until there are no requests outstanding,
1945          * and then continue.  For now, we just invalidate the requests,
1946          * schedule() and sleep one second if needed, and hope.
1947          */
1948         schedule();
1949 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1950         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1951                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1952                                      cfs_time_seconds(1));
1953                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1954                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1955                                       "may try to umount it a bit later\n",
1956                                       atomic_read(&vfsmnt->mnt_count));
1957         }
1958 #endif
1959
1960         EXIT;
1961 }
1962
1963 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1964 {
1965         struct ll_sb_info *sbi = ll_s2sbi(sb);
1966         int err;
1967         __u32 read_only;
1968
1969         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1970                 read_only = *flags & MS_RDONLY;
1971                 err = obd_set_info_async(sbi->ll_md_exp,
1972                                          sizeof(KEY_READ_ONLY),
1973                                          KEY_READ_ONLY, sizeof(read_only),
1974                                          &read_only, NULL);
1975                 if (err) {
1976                         CERROR("Failed to change the read-only flag during "
1977                                "remount: %d\n", err);
1978                         return err;
1979                 }
1980
1981                 if (read_only)
1982                         sb->s_flags |= MS_RDONLY;
1983                 else
1984                         sb->s_flags &= ~MS_RDONLY;
1985         }
1986         return 0;
1987 }
1988
1989 int ll_prep_inode(struct inode **inode,
1990                   struct ptlrpc_request *req,
1991                   struct super_block *sb)
1992 {
1993         struct ll_sb_info *sbi = NULL;
1994         struct lustre_md md;
1995         int rc;
1996         ENTRY;
1997
1998         LASSERT(*inode || sb);
1999         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2000         prune_deathrow(sbi, 1);
2001         memset(&md, 0, sizeof(struct lustre_md));
2002
2003         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2004                               sbi->ll_md_exp, &md);
2005         if (rc)
2006                 RETURN(rc);
2007
2008         if (*inode) {
2009                 ll_update_inode(*inode, &md);
2010         } else {
2011                 LASSERT(sb != NULL);
2012
2013                 /*
2014                  * At this point server returns to client's same fid as client
2015                  * generated for creating. So using ->fid1 is okay here.
2016                  */
2017                 LASSERT(fid_is_sane(&md.body->fid1));
2018
2019                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2020                 if (*inode == NULL || IS_ERR(*inode)) {
2021                         if (md.lsm)
2022                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2023 #ifdef CONFIG_FS_POSIX_ACL
2024                         if (md.posix_acl) {
2025                                 posix_acl_release(md.posix_acl);
2026                                 md.posix_acl = NULL;
2027                         }
2028 #endif
2029                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2030                         *inode = NULL;
2031                         CERROR("new_inode -fatal: rc %d\n", rc);
2032                         GOTO(out, rc);
2033                 }
2034         }
2035
2036         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2037                          ll_i2info(*inode)->lli_smd);
2038 out:
2039         md_free_lustre_md(sbi->ll_md_exp, &md);
2040         RETURN(rc);
2041 }
2042
2043 int ll_obd_statfs(struct inode *inode, void *arg)
2044 {
2045         struct ll_sb_info *sbi = NULL;
2046         struct obd_export *exp;
2047         char *buf = NULL;
2048         struct obd_ioctl_data *data = NULL;
2049         __u32 type;
2050         int len = 0, rc;
2051
2052         if (!inode || !(sbi = ll_i2sbi(inode)))
2053                 GOTO(out_statfs, rc = -EINVAL);
2054
2055         rc = obd_ioctl_getdata(&buf, &len, arg);
2056         if (rc)
2057                 GOTO(out_statfs, rc);
2058
2059         data = (void*)buf;
2060         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2061             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2062                 GOTO(out_statfs, rc = -EINVAL);
2063
2064         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2065         if (type == LL_STATFS_MDC)
2066                 exp = sbi->ll_md_exp;
2067         else if (type == LL_STATFS_LOV)
2068                 exp = sbi->ll_dt_exp;
2069         else
2070                 GOTO(out_statfs, rc = -ENODEV);
2071
2072         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2073         if (rc)
2074                 GOTO(out_statfs, rc);
2075 out_statfs:
2076         if (buf)
2077                 obd_ioctl_freedata(buf, len);
2078         return rc;
2079 }
2080
2081 int ll_process_config(struct lustre_cfg *lcfg)
2082 {
2083         char *ptr;
2084         void *sb;
2085         struct lprocfs_static_vars lvars;
2086         unsigned long x;
2087         int rc = 0;
2088
2089         lprocfs_llite_init_vars(&lvars);
2090
2091         /* The instance name contains the sb: lustre-client-aacfe000 */
2092         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2093         if (!ptr || !*(++ptr))
2094                 return -EINVAL;
2095         if (sscanf(ptr, "%lx", &x) != 1)
2096                 return -EINVAL;
2097         sb = (void *)x;
2098         /* This better be a real Lustre superblock! */
2099         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2100
2101         /* Note we have not called client_common_fill_super yet, so
2102            proc fns must be able to handle that! */
2103         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2104                                       lcfg, sb);
2105         if (rc > 0)
2106                 rc = 0;
2107         return(rc);
2108 }
2109
2110 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2111 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2112                                        struct inode *i1, struct inode *i2,
2113                                        const char *name, int namelen,
2114                                        int mode, __u32 opc, void *data)
2115 {
2116         LASSERT(i1 != NULL);
2117
2118         if (namelen > ll_i2sbi(i1)->ll_namelen)
2119                 return ERR_PTR(-ENAMETOOLONG);
2120
2121         if (op_data == NULL)
2122                 OBD_ALLOC_PTR(op_data);
2123
2124         if (op_data == NULL)
2125                 return ERR_PTR(-ENOMEM);
2126
2127         ll_i2gids(op_data->op_suppgids, i1, i2);
2128         op_data->op_fid1 = *ll_inode2fid(i1);
2129         op_data->op_capa1 = ll_mdscapa_get(i1);
2130
2131         if (i2) {
2132                 op_data->op_fid2 = *ll_inode2fid(i2);
2133                 op_data->op_capa2 = ll_mdscapa_get(i2);
2134         } else {
2135                 fid_zero(&op_data->op_fid2);
2136                 op_data->op_capa2 = NULL;
2137         }
2138
2139         op_data->op_name = name;
2140         op_data->op_namelen = namelen;
2141         op_data->op_mode = mode;
2142         op_data->op_mod_time = cfs_time_current_sec();
2143         op_data->op_fsuid = current->fsuid;
2144         op_data->op_fsgid = current->fsgid;
2145         op_data->op_cap = cfs_curproc_cap_pack();
2146         op_data->op_bias = MDS_CHECK_SPLIT;
2147         op_data->op_opc = opc;
2148         op_data->op_mds = 0;
2149         op_data->op_data = data;
2150
2151         return op_data;
2152 }
2153
2154 void ll_finish_md_op_data(struct md_op_data *op_data)
2155 {
2156         capa_put(op_data->op_capa1);
2157         capa_put(op_data->op_capa2);
2158         OBD_FREE_PTR(op_data);
2159 }
2160
2161 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2162 {
2163         struct ll_sb_info *sbi;
2164
2165         LASSERT((seq != NULL) && (vfs != NULL));
2166         sbi = ll_s2sbi(vfs->mnt_sb);
2167
2168         if (sbi->ll_flags & LL_SBI_NOLCK)
2169                 seq_puts(seq, ",nolock");
2170
2171         if (sbi->ll_flags & LL_SBI_FLOCK)
2172                 seq_puts(seq, ",flock");
2173
2174         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2175                 seq_puts(seq, ",localflock");
2176
2177         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2178                 seq_puts(seq, ",user_xattr");
2179
2180         if (sbi->ll_flags & LL_SBI_ACL)
2181                 seq_puts(seq, ",acl");
2182
2183         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2184                 seq_puts(seq, ",lazystatfs");
2185
2186         RETURN(0);
2187 }