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 (c) 2003, 2010, Oracle and/or its affiliates. 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/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 cfs_spinlock_t ll_sb_lock = CFS_SPIN_LOCK_UNLOCKED;
  63
  64 extern struct address_space_operations ll_aops;
  65 extern struct address_space_operations ll_dir_aops;
  66
  67 #ifndef log2
  68 #define log2(n) cfs_ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC(sbi, sizeof(*sbi));
  81         if (!sbi)
  82                 RETURN(NULL);
  83
  84         cfs_spin_lock_init(&sbi->ll_lock);
  85         cfs_init_mutex(&sbi->ll_lco.lco_lock);
  86         cfs_spin_lock_init(&sbi->ll_pp_extent_lock);
  87         cfs_spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  93 #ifdef HAVE_BGL_SUPPORT
  94                 sbi->ll_async_page_max = pages / 4;
  95 #else
  96                 sbi->ll_async_page_max = pages / 2;
  97 #endif
  98         } else {
  99                 sbi->ll_async_page_max = (pages / 4) * 3;
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         CFS_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         cfs_spin_lock(&ll_sb_lock);
 115         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         cfs_spin_unlock(&ll_sb_lock);
 117
 118 #ifdef ENABLE_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         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 127                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 128                                    pp_r_hist.oh_lock);
 129                 cfs_spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i]. \
 130                                    pp_w_hist.oh_lock);
 131         }
 132
 133         /* metadata statahead is enabled by default */
 134         sbi->ll_sa_max = LL_SA_RPC_DEF;
 135         atomic_set(&sbi->ll_sa_total, 0);
 136         atomic_set(&sbi->ll_sa_wrong, 0);
 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                 cfs_spin_lock(&ll_sb_lock);
 148                 cfs_list_del(&sbi->ll_list);
 149                 cfs_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         .d_compare = ll_dcompare,
 157         .d_revalidate = ll_revalidate_nd,
 158 };
 159
 160 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 161 {
 162         struct inode *root = 0;
 163         struct ll_sb_info *sbi = ll_s2sbi(sb);
 164         struct obd_device *obd;
 165         struct obd_capa *oc = NULL;
 166         struct obd_statfs osfs;
 167         struct ptlrpc_request *request = NULL;
 168         struct obd_connect_data *data = NULL;
 169         struct obd_uuid *uuid;
 170         struct md_op_data *op_data;
 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_FULL20;
 201
 202         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 203                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 204
 205 #ifdef HAVE_LRU_RESIZE_SUPPORT
 206         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 207                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 208 #endif
 209 #ifdef CONFIG_FS_POSIX_ACL
 210         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 211 #endif
 212         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 213         data->ocd_version = LUSTRE_VERSION_CODE;
 214
 215         if (sb->s_flags & MS_RDONLY)
 216                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 217         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 218                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 219
 220 #ifdef HAVE_MS_FLOCK_LOCK
 221         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 222         sb->s_flags |= MS_FLOCK_LOCK;
 223 #endif
 224
 225         if (sbi->ll_flags & LL_SBI_FLOCK)
 226                 sbi->ll_fop = &ll_file_operations_flock;
 227         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 228                 sbi->ll_fop = &ll_file_operations;
 229         else
 230                 sbi->ll_fop = &ll_file_operations_noflock;
 231
 232         /* real client */
 233         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 234         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 235                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 236
 237         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 238         if (err == -EBUSY) {
 239                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 240                                    "recovery, of which this client is not a "
 241                                    "part. Please wait for recovery to complete,"
 242                                    " abort, or time out.\n", md);
 243                 GOTO(out, err);
 244         } else if (err) {
 245                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 246                 GOTO(out, err);
 247         }
 248
 249         err = obd_fid_init(sbi->ll_md_exp);
 250         if (err) {
 251                 CERROR("Can't init metadata layer FID infrastructure, "
 252                        "rc %d\n", err);
 253                 GOTO(out_md, err);
 254         }
 255
 256         err = obd_statfs(obd, &osfs,
 257                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 258         if (err)
 259                 GOTO(out_md_fid, err);
 260
 261         size = sizeof(*data);
 262         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 263                            KEY_CONN_DATA,  &size, data, NULL);
 264         if (err) {
 265                 CERROR("Get connect data failed: %d \n", err);
 266                 GOTO(out_md, err);
 267         }
 268
 269         LASSERT(osfs.os_bsize);
 270         sb->s_blocksize = osfs.os_bsize;
 271         sb->s_blocksize_bits = log2(osfs.os_bsize);
 272         sb->s_magic = LL_SUPER_MAGIC;
 273
 274         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 275          *         retval = in_file->f_op->sendfile(...);
 276          *         if (*ppos > max)
 277          *                 retval = -EOVERFLOW;
 278          *
 279          * it will check if *ppos is greater than max. However, max equals to
 280          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 281          * has been defined as a signed long long integer in linux kernel. */
 282 #if BITS_PER_LONG == 64
 283         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 284 #else
 285         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 286 #endif
 287         sbi->ll_namelen = osfs.os_namelen;
 288         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 289
 290         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 291             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 292                 LCONSOLE_INFO("Disabling user_xattr feature because "
 293                               "it is not supported on the server\n");
 294                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 295         }
 296
 297         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 298 #ifdef MS_POSIXACL
 299                 sb->s_flags |= MS_POSIXACL;
 300 #endif
 301                 sbi->ll_flags |= LL_SBI_ACL;
 302         } else {
 303                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 304 #ifdef MS_POSIXACL
 305                 sb->s_flags &= ~MS_POSIXACL;
 306 #endif
 307                 sbi->ll_flags &= ~LL_SBI_ACL;
 308         }
 309
 310         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 311                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 312                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 313                         LCONSOLE_INFO("client is set as remote by default.\n");
 314                 }
 315         } else {
 316                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 317                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 318                         LCONSOLE_INFO("client claims to be remote, but server "
 319                                       "rejected, forced to be local.\n");
 320                 }
 321         }
 322
 323         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 324                 LCONSOLE_INFO("client enabled MDS capability!\n");
 325                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 326         }
 327
 328         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 329                 LCONSOLE_INFO("client enabled OSS capability!\n");
 330                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 331         }
 332
 333         obd = class_name2obd(dt);
 334         if (!obd) {
 335                 CERROR("DT %s: not setup or attached\n", dt);
 336                 GOTO(out_md_fid, err = -ENODEV);
 337         }
 338
 339         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 340                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 341                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 342                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 343                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 344                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 345                                   OBD_CONNECT_FULL20;
 346
 347         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 348                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 349
 350         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 351                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 352                  * disabled by default, because it can still be enabled on the
 353                  * fly via /proc. As a consequence, we still need to come to an
 354                  * agreement on the supported algorithms at connect time */
 355                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 356
 357                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 358                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 359                 else
 360                         /* send the list of supported checksum types */
 361                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 362         }
 363
 364 #ifdef HAVE_LRU_RESIZE_SUPPORT
 365         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 366 #endif
 367         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 368                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 369
 370         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 371                "ocd_grant: %d\n", data->ocd_connect_flags,
 372                data->ocd_version, data->ocd_grant);
 373
 374         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 375         obd->obd_upcall.onu_upcall = cl_ocd_update;
 376         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 377
 378         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 379         if (err == -EBUSY) {
 380                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 381                                    "recovery, of which this client is not a "
 382                                    "part.  Please wait for recovery to "
 383                                    "complete, abort, or time out.\n", dt);
 384                 GOTO(out_md_fid, err);
 385         } else if (err) {
 386                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 387                 GOTO(out_md_fid, err);
 388         }
 389
 390         err = obd_fid_init(sbi->ll_dt_exp);
 391         if (err) {
 392                 CERROR("Can't init data layer FID infrastructure, "
 393                        "rc %d\n", err);
 394                 GOTO(out_dt, err);
 395         }
 396
 397         cfs_mutex_down(&sbi->ll_lco.lco_lock);
 398         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 399         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 400         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 401         cfs_mutex_up(&sbi->ll_lco.lco_lock);
 402
 403         fid_zero(&sbi->ll_root_fid);
 404         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 405         if (err) {
 406                 CERROR("cannot mds_connect: rc = %d\n", err);
 407                 GOTO(out_lock_cn_cb, err);
 408         }
 409         if (!fid_is_sane(&sbi->ll_root_fid)) {
 410                 CERROR("Invalid root fid during mount\n");
 411                 GOTO(out_lock_cn_cb, err = -EINVAL);
 412         }
 413         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 414
 415         sb->s_op = &lustre_super_operations;
 416 #if THREAD_SIZE >= 8192
 417         sb->s_export_op = &lustre_export_operations;
 418 #endif
 419
 420         /* make root inode
 421          * XXX: move this to after cbd setup? */
 422         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 423         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 424                 valid |= OBD_MD_FLRMTPERM;
 425         else if (sbi->ll_flags & LL_SBI_ACL)
 426                 valid |= OBD_MD_FLACL;
 427
 428         OBD_ALLOC_PTR(op_data);
 429         if (op_data == NULL)
 430                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 431
 432         op_data->op_fid1 = sbi->ll_root_fid;
 433         op_data->op_mode = 0;
 434         op_data->op_capa1 = oc;
 435         op_data->op_valid = valid;
 436
 437         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 438         if (oc)
 439                 capa_put(oc);
 440         OBD_FREE_PTR(op_data);
 441         if (err) {
 442                 CERROR("md_getattr failed for root: rc = %d\n", err);
 443                 GOTO(out_lock_cn_cb, err);
 444         }
 445         memset(&lmd, 0, sizeof(lmd));
 446         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 447                                sbi->ll_md_exp, &lmd);
 448         if (err) {
 449                 CERROR("failed to understand root inode md: rc = %d\n", err);
 450                 ptlrpc_req_finished (request);
 451                 GOTO(out_lock_cn_cb, err);
 452         }
 453
 454         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 455         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid), &lmd);
 456         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 457         ptlrpc_req_finished(request);
 458
 459         if (root == NULL || IS_ERR(root)) {
 460                 if (lmd.lsm)
 461                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 462 #ifdef CONFIG_FS_POSIX_ACL
 463                 if (lmd.posix_acl) {
 464                         posix_acl_release(lmd.posix_acl);
 465                         lmd.posix_acl = NULL;
 466                 }
 467 #endif
 468                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 469                 root = NULL;
 470                 CERROR("lustre_lite: bad iget4 for root\n");
 471                 GOTO(out_root, err);
 472         }
 473
 474         err = ll_close_thread_start(&sbi->ll_lcq);
 475         if (err) {
 476                 CERROR("cannot start close thread: rc %d\n", err);
 477                 GOTO(out_root, err);
 478         }
 479
 480 #ifdef CONFIG_FS_POSIX_ACL
 481         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 482                 rct_init(&sbi->ll_rct);
 483                 et_init(&sbi->ll_et);
 484         }
 485 #endif
 486
 487         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 488         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 489                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 490                                  NULL);
 491         cl_sb_init(sb);
 492
 493         sb->s_root = d_alloc_root(root);
 494         if (data != NULL)
 495                 OBD_FREE(data, sizeof(*data));
 496
 497         sb->s_root->d_op = &ll_d_root_ops;
 498
 499         sbi->ll_sdev_orig = sb->s_dev;
 500
 501         /* We set sb->s_dev equal on all lustre clients in order to support
 502          * NFS export clustering.  NFSD requires that the FSID be the same
 503          * on all clients. */
 504         /* s_dev is also used in lt_compare() to compare two fs, but that is
 505          * only a node-local comparison. */
 506         uuid = obd_get_uuid(sbi->ll_md_exp);
 507         if (uuid != NULL)
 508                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 509
 510         RETURN(err);
 511 out_root:
 512         if (root)
 513                 iput(root);
 514 out_lock_cn_cb:
 515         obd_fid_fini(sbi->ll_dt_exp);
 516 out_dt:
 517         obd_disconnect(sbi->ll_dt_exp);
 518         sbi->ll_dt_exp = NULL;
 519 out_md_fid:
 520         obd_fid_fini(sbi->ll_md_exp);
 521 out_md:
 522         obd_disconnect(sbi->ll_md_exp);
 523         sbi->ll_md_exp = NULL;
 524 out:
 525         if (data != NULL)
 526                 OBD_FREE_PTR(data);
 527         lprocfs_unregister_mountpoint(sbi);
 528         return err;
 529 }
 530
 531 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 532 {
 533         int size, rc;
 534
 535         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 536         size = sizeof(int);
 537         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 538                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 539         if (rc)
 540                 CERROR("Get max mdsize error rc %d \n", rc);
 541
 542         RETURN(rc);
 543 }
 544
 545 void ll_dump_inode(struct inode *inode)
 546 {
 547         struct list_head *tmp;
 548         int dentry_count = 0;
 549
 550         LASSERT(inode != NULL);
 551
 552         list_for_each(tmp, &inode->i_dentry)
 553                 dentry_count++;
 554
 555         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 556                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 557                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 558 }
 559
 560 void lustre_dump_dentry(struct dentry *dentry, int recur)
 561 {
 562         struct list_head *tmp;
 563         int subdirs = 0;
 564
 565         LASSERT(dentry != NULL);
 566
 567         list_for_each(tmp, &dentry->d_subdirs)
 568                 subdirs++;
 569
 570         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 571                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 572                dentry->d_name.len, dentry->d_name.name,
 573                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 574                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 575                dentry->d_flags, dentry->d_fsdata, subdirs);
 576         if (dentry->d_inode != NULL)
 577                 ll_dump_inode(dentry->d_inode);
 578
 579         if (recur == 0)
 580                 return;
 581
 582        list_for_each(tmp, &dentry->d_subdirs) {
 583                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 584                 lustre_dump_dentry(d, recur - 1);
 585         }
 586 }
 587
 588 void client_common_put_super(struct super_block *sb)
 589 {
 590         struct ll_sb_info *sbi = ll_s2sbi(sb);
 591         ENTRY;
 592
 593 #ifdef CONFIG_FS_POSIX_ACL
 594         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 595                 et_fini(&sbi->ll_et);
 596                 rct_fini(&sbi->ll_rct);
 597         }
 598 #endif
 599
 600         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 601
 602         ll_close_thread_shutdown(sbi->ll_lcq);
 603
 604         cl_sb_fini(sb);
 605
 606         cfs_list_del(&sbi->ll_conn_chain);
 607
 608         obd_fid_fini(sbi->ll_dt_exp);
 609         obd_disconnect(sbi->ll_dt_exp);
 610         sbi->ll_dt_exp = NULL;
 611
 612         lprocfs_unregister_mountpoint(sbi);
 613
 614         obd_fid_fini(sbi->ll_md_exp);
 615         obd_disconnect(sbi->ll_md_exp);
 616         sbi->ll_md_exp = NULL;
 617
 618         EXIT;
 619 }
 620
 621 void ll_kill_super(struct super_block *sb)
 622 {
 623         struct ll_sb_info *sbi;
 624
 625         ENTRY;
 626
 627         /* not init sb ?*/
 628         if (!(sb->s_flags & MS_ACTIVE))
 629                 return;
 630
 631         sbi = ll_s2sbi(sb);
 632         /* we need restore s_dev from changed for clustred NFS before put_super
 633          * because new kernels have cached s_dev and change sb->s_dev in
 634          * put_super not affected real removing devices */
 635         if (sbi)
 636                 sb->s_dev = sbi->ll_sdev_orig;
 637         EXIT;
 638 }
 639
 640 char *ll_read_opt(const char *opt, char *data)
 641 {
 642         char *value;
 643         char *retval;
 644         ENTRY;
 645
 646         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 647         if (strncmp(opt, data, strlen(opt)))
 648                 RETURN(NULL);
 649         if ((value = strchr(data, '=')) == NULL)
 650                 RETURN(NULL);
 651
 652         value++;
 653         OBD_ALLOC(retval, strlen(value) + 1);
 654         if (!retval) {
 655                 CERROR("out of memory!\n");
 656                 RETURN(NULL);
 657         }
 658
 659         memcpy(retval, value, strlen(value)+1);
 660         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 661         RETURN(retval);
 662 }
 663
 664 static inline int ll_set_opt(const char *opt, char *data, int fl)
 665 {
 666         if (strncmp(opt, data, strlen(opt)) != 0)
 667                 return(0);
 668         else
 669                 return(fl);
 670 }
 671
 672 /* non-client-specific mount options are parsed in lmd_parse */
 673 static int ll_options(char *options, int *flags)
 674 {
 675         int tmp;
 676         char *s1 = options, *s2;
 677         ENTRY;
 678
 679         if (!options)
 680                 RETURN(0);
 681
 682         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 683
 684         while (*s1) {
 685                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 686                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 687                 if (tmp) {
 688                         *flags |= tmp;
 689                         goto next;
 690                 }
 691                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 692                 if (tmp) {
 693                         *flags |= tmp;
 694                         goto next;
 695                 }
 696                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 697                 if (tmp) {
 698                         *flags |= tmp;
 699                         goto next;
 700                 }
 701                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 702                 if (tmp) {
 703                         *flags &= ~tmp;
 704                         goto next;
 705                 }
 706                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 707                 if (tmp) {
 708                         *flags |= tmp;
 709                         goto next;
 710                 }
 711                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 712                 if (tmp) {
 713                         *flags &= ~tmp;
 714                         goto next;
 715                 }
 716 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,5,50,0)
 717                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 718                 if (tmp) {
 719                         /* Ignore deprecated mount option.  The client will
 720                          * always try to mount with ACL support, whether this
 721                          * is used depends on whether server supports it. */
 722                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 723                                                   "mount option 'acl'.\n");
 724                         goto next;
 725                 }
 726                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 727                 if (tmp) {
 728                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 729                                                   "mount option 'noacl'.\n");
 730                         goto next;
 731                 }
 732 #else
 733 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 734 #endif
 735                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 736                 if (tmp) {
 737                         *flags |= tmp;
 738                         goto next;
 739                 }
 740
 741                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 742                 if (tmp) {
 743                         *flags |= tmp;
 744                         goto next;
 745                 }
 746                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 747                 if (tmp) {
 748                         *flags &= ~tmp;
 749                         goto next;
 750                 }
 751                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 752                 if (tmp) {
 753                         *flags |= tmp;
 754                         goto next;
 755                 }
 756                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 757                 if (tmp) {
 758                         *flags &= ~tmp;
 759                         goto next;
 760                 }
 761                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 762                 if (tmp) {
 763                         *flags |= tmp;
 764                         goto next;
 765                 }
 766                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 767                 if (tmp) {
 768                         *flags &= ~tmp;
 769                         goto next;
 770                 }
 771                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781
 782                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 783                                    s1);
 784                 RETURN(-EINVAL);
 785
 786 next:
 787                 /* Find next opt */
 788                 s2 = strchr(s1, ',');
 789                 if (s2 == NULL)
 790                         break;
 791                 s1 = s2 + 1;
 792         }
 793         RETURN(0);
 794 }
 795
 796 void ll_lli_init(struct ll_inode_info *lli)
 797 {
 798         lli->lli_inode_magic = LLI_INODE_MAGIC;
 799         cfs_sema_init(&lli->lli_size_sem, 1);
 800         cfs_sema_init(&lli->lli_write_sem, 1);
 801         cfs_init_rwsem(&lli->lli_trunc_sem);
 802         lli->lli_flags = 0;
 803         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 804         cfs_spin_lock_init(&lli->lli_lock);
 805         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 806         lli->lli_inode_magic = LLI_INODE_MAGIC;
 807         cfs_sema_init(&lli->lli_och_sem, 1);
 808         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 809         lli->lli_mds_exec_och = NULL;
 810         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 811         lli->lli_open_fd_exec_count = 0;
 812         CFS_INIT_LIST_HEAD(&lli->lli_dead_list);
 813         lli->lli_remote_perms = NULL;
 814         lli->lli_rmtperm_utime = 0;
 815         cfs_sema_init(&lli->lli_rmtperm_sem, 1);
 816         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 817         cfs_spin_lock_init(&lli->lli_sa_lock);
 818 }
 819
 820 #ifdef HAVE_NEW_BACKING_DEV_INFO
 821 static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 822 #endif
 823
 824 int ll_fill_super(struct super_block *sb)
 825 {
 826         struct lustre_profile *lprof;
 827         struct lustre_sb_info *lsi = s2lsi(sb);
 828         struct ll_sb_info *sbi;
 829         char  *dt = NULL, *md = NULL;
 830         char  *profilenm = get_profile_name(sb);
 831         struct config_llog_instance cfg = {0, };
 832         char   ll_instance[sizeof(sb) * 2 + 1];
 833         int    err;
 834         ENTRY;
 835
 836         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 837
 838         cfs_module_get();
 839
 840         /* client additional sb info */
 841         lsi->lsi_llsbi = sbi = ll_init_sbi();
 842         if (!sbi) {
 843                 cfs_module_put(THIS_MODULE);
 844                 RETURN(-ENOMEM);
 845         }
 846
 847         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 848         if (err)
 849                 GOTO(out_free, err);
 850
 851         err = ll_bdi_init(&lsi->bdi);
 852         if (err)
 853                 GOTO(out_free, err);
 854
 855 #ifdef HAVE_NEW_BACKING_DEV_INFO
 856         lsi->bdi.name = "lustre";
 857         lsi->bdi.capabilities = BDI_CAP_MAP_COPY;
 858         err = bdi_register(&lsi->bdi, NULL, "lustre-%d",
 859                            atomic_inc_return(&ll_bdi_num));
 860         sb->s_bdi = &lsi->bdi;
 861 #endif
 862
 863         /* Generate a string unique to this super, in case some joker tries
 864            to mount the same fs at two mount points.
 865            Use the address of the super itself.*/
 866         sprintf(ll_instance, "%p", sb);
 867         cfg.cfg_instance = ll_instance;
 868         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 869
 870         /* set up client obds */
 871         err = lustre_process_log(sb, profilenm, &cfg);
 872         if (err < 0) {
 873                 CERROR("Unable to process log: %d\n", err);
 874                 GOTO(out_free, err);
 875         }
 876
 877         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 878         lprof = class_get_profile(profilenm);
 879         if (lprof == NULL) {
 880                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 881                                    " read from the MGS.  Does that filesystem "
 882                                    "exist?\n", profilenm);
 883                 GOTO(out_free, err = -EINVAL);
 884         }
 885         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 886                lprof->lp_md, lprof->lp_dt);
 887
 888         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 889                   strlen(ll_instance) + 2);
 890         if (!dt)
 891                 GOTO(out_free, err = -ENOMEM);
 892         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 893
 894         OBD_ALLOC(md, strlen(lprof->lp_md) +
 895                   strlen(ll_instance) + 2);
 896         if (!md)
 897                 GOTO(out_free, err = -ENOMEM);
 898         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 899
 900         /* connections, registrations, sb setup */
 901         err = client_common_fill_super(sb, md, dt);
 902
 903 out_free:
 904         if (md)
 905                 OBD_FREE(md, strlen(md) + 1);
 906         if (dt)
 907                 OBD_FREE(dt, strlen(dt) + 1);
 908         if (err)
 909                 ll_put_super(sb);
 910         else
 911                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 912
 913         RETURN(err);
 914 } /* ll_fill_super */
 915
 916
 917 void lu_context_keys_dump(void);
 918
 919 void ll_put_super(struct super_block *sb)
 920 {
 921         struct config_llog_instance cfg;
 922         char   ll_instance[sizeof(sb) * 2 + 1];
 923         struct obd_device *obd;
 924         struct lustre_sb_info *lsi = s2lsi(sb);
 925         struct ll_sb_info *sbi = ll_s2sbi(sb);
 926         char *profilenm = get_profile_name(sb);
 927         int force = 1, next;
 928         ENTRY;
 929
 930         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 931
 932         ll_print_capa_stat(sbi);
 933
 934         sprintf(ll_instance, "%p", sb);
 935         cfg.cfg_instance = ll_instance;
 936         lustre_end_log(sb, NULL, &cfg);
 937
 938         if (sbi->ll_md_exp) {
 939                 obd = class_exp2obd(sbi->ll_md_exp);
 940                 if (obd)
 941                         force = obd->obd_force;
 942         }
 943
 944         /* We need to set force before the lov_disconnect in
 945            lustre_common_put_super, since l_d cleans up osc's as well. */
 946         if (force) {
 947                 next = 0;
 948                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 949                                                      &next)) != NULL) {
 950                         obd->obd_force = force;
 951                 }
 952         }
 953
 954         if (sbi->ll_lcq) {
 955                 /* Only if client_common_fill_super succeeded */
 956                 client_common_put_super(sb);
 957         }
 958
 959         next = 0;
 960         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
 961                 class_manual_cleanup(obd);
 962         }
 963
 964         if (profilenm)
 965                 class_del_profile(profilenm);
 966
 967         if (ll_bdi_wb_cnt(lsi->bdi) > 0)
 968                 ll_bdi_destroy(&lsi->bdi);
 969
 970         ll_free_sbi(sb);
 971         lsi->lsi_llsbi = NULL;
 972
 973         lustre_common_put_super(sb);
 974
 975         cl_env_cache_purge(~0);
 976
 977         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
 978
 979         cfs_module_put(THIS_MODULE);
 980
 981         EXIT;
 982 } /* client_put_super */
 983
 984 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 985 {
 986         struct inode *inode = NULL;
 987         /* NOTE: we depend on atomic igrab() -bzzz */
 988         lock_res_and_lock(lock);
 989         if (lock->l_ast_data) {
 990                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 991                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 992                         inode = igrab(lock->l_ast_data);
 993                 } else {
 994                         inode = lock->l_ast_data;
 995                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
 996                                          D_WARNING, lock, "l_ast_data %p is "
 997                                          "bogus: magic %08x", lock->l_ast_data,
 998                                          lli->lli_inode_magic);
 999                         inode = NULL;
1000                 }
1001         }
1002         unlock_res_and_lock(lock);
1003         return inode;
1004 }
1005
1006 static int null_if_equal(struct ldlm_lock *lock, void *data)
1007 {
1008         if (data == lock->l_ast_data) {
1009                 lock->l_ast_data = NULL;
1010
1011                 if (lock->l_req_mode != lock->l_granted_mode)
1012                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1013         }
1014
1015         return LDLM_ITER_CONTINUE;
1016 }
1017
1018 void ll_clear_inode(struct inode *inode)
1019 {
1020         struct ll_inode_info *lli = ll_i2info(inode);
1021         struct ll_sb_info *sbi = ll_i2sbi(inode);
1022         ENTRY;
1023
1024         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1025                inode->i_generation, inode);
1026
1027         if (S_ISDIR(inode->i_mode)) {
1028                 /* these should have been cleared in ll_file_release */
1029                 LASSERT(lli->lli_sai == NULL);
1030                 LASSERT(lli->lli_opendir_key == NULL);
1031                 LASSERT(lli->lli_opendir_pid == 0);
1032         }
1033
1034         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1035         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1036                          null_if_equal, inode);
1037
1038         LASSERT(!lli->lli_open_fd_write_count);
1039         LASSERT(!lli->lli_open_fd_read_count);
1040         LASSERT(!lli->lli_open_fd_exec_count);
1041
1042         if (lli->lli_mds_write_och)
1043                 ll_md_real_close(inode, FMODE_WRITE);
1044         if (lli->lli_mds_exec_och)
1045                 ll_md_real_close(inode, FMODE_EXEC);
1046         if (lli->lli_mds_read_och)
1047                 ll_md_real_close(inode, FMODE_READ);
1048
1049         if (lli->lli_symlink_name) {
1050                 OBD_FREE(lli->lli_symlink_name,
1051                          strlen(lli->lli_symlink_name) + 1);
1052                 lli->lli_symlink_name = NULL;
1053         }
1054
1055         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1056                 LASSERT(lli->lli_posix_acl == NULL);
1057                 if (lli->lli_remote_perms) {
1058                         free_rmtperm_hash(lli->lli_remote_perms);
1059                         lli->lli_remote_perms = NULL;
1060                 }
1061         }
1062 #ifdef CONFIG_FS_POSIX_ACL
1063         else if (lli->lli_posix_acl) {
1064                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1065                 LASSERT(lli->lli_remote_perms == NULL);
1066                 posix_acl_release(lli->lli_posix_acl);
1067                 lli->lli_posix_acl = NULL;
1068         }
1069 #endif
1070         lli->lli_inode_magic = LLI_INODE_DEAD;
1071
1072         ll_clear_inode_capas(inode);
1073         /*
1074          * XXX This has to be done before lsm is freed below, because
1075          * cl_object still uses inode lsm.
1076          */
1077         cl_inode_fini(inode);
1078
1079         if (lli->lli_smd) {
1080                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1081                 lli->lli_smd = NULL;
1082         }
1083
1084
1085         EXIT;
1086 }
1087
1088 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1089                   struct md_open_data **mod)
1090 {
1091         struct lustre_md md;
1092         struct ll_sb_info *sbi = ll_i2sbi(inode);
1093         struct ptlrpc_request *request = NULL;
1094         int rc;
1095         ENTRY;
1096
1097         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1098                                      LUSTRE_OPC_ANY, NULL);
1099         if (IS_ERR(op_data))
1100                 RETURN(PTR_ERR(op_data));
1101
1102         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1103                         &request, mod);
1104         if (rc) {
1105                 ptlrpc_req_finished(request);
1106                 if (rc == -ENOENT) {
1107                         inode->i_nlink = 0;
1108                         /* Unlinked special device node? Or just a race?
1109                          * Pretend we done everything. */
1110                         if (!S_ISREG(inode->i_mode) &&
1111                             !S_ISDIR(inode->i_mode))
1112                                 rc = inode_setattr(inode, &op_data->op_attr);
1113                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1114                         CERROR("md_setattr fails: rc = %d\n", rc);
1115                 }
1116                 RETURN(rc);
1117         }
1118
1119         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1120                               sbi->ll_md_exp, &md);
1121         if (rc) {
1122                 ptlrpc_req_finished(request);
1123                 RETURN(rc);
1124         }
1125
1126         /* We call inode_setattr to adjust timestamps.
1127          * If there is at least some data in file, we cleared ATTR_SIZE
1128          * above to avoid invoking vmtruncate, otherwise it is important
1129          * to call vmtruncate in inode_setattr to update inode->i_size
1130          * (bug 6196) */
1131         rc = inode_setattr(inode, &op_data->op_attr);
1132
1133         /* Extract epoch data if obtained. */
1134         op_data->op_handle = md.body->handle;
1135         op_data->op_ioepoch = md.body->ioepoch;
1136
1137         ll_update_inode(inode, &md);
1138         ptlrpc_req_finished(request);
1139
1140         RETURN(rc);
1141 }
1142
1143 /* Close IO epoch and send Size-on-MDS attribute update. */
1144 static int ll_setattr_done_writing(struct inode *inode,
1145                                    struct md_op_data *op_data,
1146                                    struct md_open_data *mod)
1147 {
1148         struct ll_inode_info *lli = ll_i2info(inode);
1149         int rc = 0;
1150         ENTRY;
1151
1152         LASSERT(op_data != NULL);
1153         if (!S_ISREG(inode->i_mode))
1154                 RETURN(0);
1155
1156         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1157                op_data->op_ioepoch, PFID(&lli->lli_fid));
1158
1159         op_data->op_flags = MF_EPOCH_CLOSE;
1160         ll_done_writing_attr(inode, op_data);
1161         ll_pack_inode2opdata(inode, op_data, NULL);
1162
1163         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1164         if (rc == -EAGAIN) {
1165                 /* MDS has instructed us to obtain Size-on-MDS attribute
1166                  * from OSTs and send setattr to back to MDS. */
1167                 rc = ll_som_update(inode, op_data);
1168         } else if (rc) {
1169                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1170                        inode->i_ino, rc);
1171         }
1172         RETURN(rc);
1173 }
1174
1175 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1176 {
1177         struct obd_capa *capa;
1178         int rc;
1179
1180         if (attr->ia_valid & ATTR_SIZE)
1181                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1182         else
1183                 capa = ll_mdscapa_get(inode);
1184
1185         rc = cl_setattr_ost(inode, attr, capa);
1186
1187         if (attr->ia_valid & ATTR_SIZE)
1188                 ll_truncate_free_capa(capa);
1189         else
1190                 capa_put(capa);
1191
1192         return rc;
1193 }
1194
1195 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1196  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1197  * keep these values until such a time that objects are allocated for it.
1198  * We do the MDS operations first, as it is checking permissions for us.
1199  * We don't to the MDS RPC if there is nothing that we want to store there,
1200  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1201  * going to do an RPC anyways.
1202  *
1203  * If we are doing a truncate, we will send the mtime and ctime updates
1204  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1205  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1206  * at the same time.
1207  */
1208 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1209 {
1210         struct ll_inode_info *lli = ll_i2info(inode);
1211         struct lov_stripe_md *lsm = lli->lli_smd;
1212         struct md_op_data *op_data = NULL;
1213         struct md_open_data *mod = NULL;
1214         int ia_valid = attr->ia_valid;
1215         int rc = 0, rc1 = 0;
1216         ENTRY;
1217
1218         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1219                attr->ia_valid);
1220         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1221
1222         if (ia_valid & ATTR_SIZE) {
1223                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1224                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1225                                attr->ia_size, ll_file_maxbytes(inode));
1226                         RETURN(-EFBIG);
1227                 }
1228
1229                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1230         }
1231
1232         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1233         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1234                 if (cfs_curproc_fsuid() != inode->i_uid &&
1235                     !cfs_capable(CFS_CAP_FOWNER))
1236                         RETURN(-EPERM);
1237         }
1238
1239         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1240         if (attr->ia_valid & ATTR_CTIME) {
1241                 attr->ia_ctime = CFS_CURRENT_TIME;
1242                 attr->ia_valid |= ATTR_CTIME_SET;
1243         }
1244         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1245                 attr->ia_atime = CFS_CURRENT_TIME;
1246                 attr->ia_valid |= ATTR_ATIME_SET;
1247         }
1248         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1249                 attr->ia_mtime = CFS_CURRENT_TIME;
1250                 attr->ia_valid |= ATTR_MTIME_SET;
1251         }
1252
1253         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1254                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1255                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1256                        cfs_time_current_sec());
1257
1258         /* NB: ATTR_SIZE will only be set after this point if the size
1259          * resides on the MDS, ie, this file has no objects. */
1260         if (lsm)
1261                 attr->ia_valid &= ~ATTR_SIZE;
1262
1263         /* We always do an MDS RPC, even if we're only changing the size;
1264          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1265
1266         OBD_ALLOC_PTR(op_data);
1267         if (op_data == NULL)
1268                 RETURN(-ENOMEM);
1269
1270         UNLOCK_INODE_MUTEX(inode);
1271         if (ia_valid & ATTR_SIZE)
1272                 UP_WRITE_I_ALLOC_SEM(inode);
1273         cfs_down_write(&lli->lli_trunc_sem);
1274         LOCK_INODE_MUTEX(inode);
1275         if (ia_valid & ATTR_SIZE)
1276                 DOWN_WRITE_I_ALLOC_SEM(inode);
1277
1278         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1279
1280         /* Open epoch for truncate. */
1281         if (exp_connect_som(ll_i2mdexp(inode)) &&
1282             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1283                 op_data->op_flags = MF_EPOCH_OPEN;
1284
1285         rc = ll_md_setattr(inode, op_data, &mod);
1286         if (rc)
1287                 GOTO(out, rc);
1288
1289         ll_ioepoch_open(lli, op_data->op_ioepoch);
1290         if (!lsm || !S_ISREG(inode->i_mode)) {
1291                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1292                 GOTO(out, rc = 0);
1293         }
1294
1295         if (ia_valid & ATTR_SIZE)
1296                 attr->ia_valid |= ATTR_SIZE;
1297         if (ia_valid & (ATTR_SIZE |
1298                         ATTR_ATIME | ATTR_ATIME_SET |
1299                         ATTR_MTIME | ATTR_MTIME_SET))
1300                 /* on truncate and utimes send attributes to osts, setting
1301                  * mtime/atime to past will be performed under PW 0:EOF extent
1302                  * lock (new_size:EOF for truncate)
1303                  * it may seem excessive to send mtime/atime updates to osts
1304                  * when not setting times to past, but it is necessary due to
1305                  * possible time de-synchronization */
1306                 rc = ll_setattr_ost(inode, attr);
1307         EXIT;
1308 out:
1309         if (op_data) {
1310                 if (op_data->op_ioepoch)
1311                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1312                 ll_finish_md_op_data(op_data);
1313         }
1314         cfs_up_write(&lli->lli_trunc_sem);
1315         return rc ? rc : rc1;
1316 }
1317
1318 int ll_setattr(struct dentry *de, struct iattr *attr)
1319 {
1320         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1321             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1322                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1323
1324         if ((de->d_inode->i_mode & S_ISUID) &&
1325             !(attr->ia_mode & S_ISUID) &&
1326             !(attr->ia_valid & ATTR_KILL_SUID))
1327                 attr->ia_valid |= ATTR_KILL_SUID;
1328
1329         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1330             !(attr->ia_mode & S_ISGID) &&
1331             !(attr->ia_valid & ATTR_KILL_SGID))
1332                 attr->ia_valid |= ATTR_KILL_SGID;
1333
1334         return ll_setattr_raw(de->d_inode, attr);
1335 }
1336
1337 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1338                        __u64 max_age, __u32 flags)
1339 {
1340         struct ll_sb_info *sbi = ll_s2sbi(sb);
1341         struct obd_statfs obd_osfs;
1342         int rc;
1343         ENTRY;
1344
1345         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1346         if (rc) {
1347                 CERROR("md_statfs fails: rc = %d\n", rc);
1348                 RETURN(rc);
1349         }
1350
1351         osfs->os_type = sb->s_magic;
1352
1353         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1354                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1355
1356         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1357                 flags |= OBD_STATFS_NODELAY;
1358
1359         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1360                               &obd_osfs, max_age, flags);
1361         if (rc) {
1362                 CERROR("obd_statfs fails: rc = %d\n", rc);
1363                 RETURN(rc);
1364         }
1365
1366         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1367                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1368                obd_osfs.os_files);
1369
1370         osfs->os_bsize = obd_osfs.os_bsize;
1371         osfs->os_blocks = obd_osfs.os_blocks;
1372         osfs->os_bfree = obd_osfs.os_bfree;
1373         osfs->os_bavail = obd_osfs.os_bavail;
1374
1375         /* If we don't have as many objects free on the OST as inodes
1376          * on the MDS, we reduce the total number of inodes to
1377          * compensate, so that the "inodes in use" number is correct.
1378          */
1379         if (obd_osfs.os_ffree < osfs->os_ffree) {
1380                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1381                         obd_osfs.os_ffree;
1382                 osfs->os_ffree = obd_osfs.os_ffree;
1383         }
1384
1385         RETURN(rc);
1386 }
1387 #ifndef HAVE_STATFS_DENTRY_PARAM
1388 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1389 {
1390 #else
1391 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1392 {
1393         struct super_block *sb = de->d_sb;
1394 #endif
1395         struct obd_statfs osfs;
1396         int rc;
1397
1398         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1399         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1400
1401         /* Some amount of caching on the client is allowed */
1402         rc = ll_statfs_internal(sb, &osfs,
1403                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1404                                 0);
1405         if (rc)
1406                 return rc;
1407
1408         statfs_unpack(sfs, &osfs);
1409
1410         /* We need to downshift for all 32-bit kernels, because we can't
1411          * tell if the kernel is being called via sys_statfs64() or not.
1412          * Stop before overflowing f_bsize - in which case it is better
1413          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1414         if (sizeof(long) < 8) {
1415                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1416                         sfs->f_bsize <<= 1;
1417
1418                         osfs.os_blocks >>= 1;
1419                         osfs.os_bfree >>= 1;
1420                         osfs.os_bavail >>= 1;
1421                 }
1422         }
1423
1424         sfs->f_blocks = osfs.os_blocks;
1425         sfs->f_bfree = osfs.os_bfree;
1426         sfs->f_bavail = osfs.os_bavail;
1427
1428         return 0;
1429 }
1430
1431 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1432 {
1433         struct ll_inode_info *lli;
1434         struct lov_stripe_md *lsm;
1435
1436         lli = ll_i2info(inode);
1437         LASSERT(lli->lli_size_sem_owner != current);
1438         cfs_down(&lli->lli_size_sem);
1439         LASSERT(lli->lli_size_sem_owner == NULL);
1440         lli->lli_size_sem_owner = current;
1441         lsm = lli->lli_smd;
1442         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1443                  lsm, lock_lsm);
1444         if (lock_lsm)
1445                 lov_stripe_lock(lsm);
1446 }
1447
1448 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1449 {
1450         struct ll_inode_info *lli;
1451         struct lov_stripe_md *lsm;
1452
1453         lli = ll_i2info(inode);
1454         lsm = lli->lli_smd;
1455         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1456                  lsm, unlock_lsm);
1457         if (unlock_lsm)
1458                 lov_stripe_unlock(lsm);
1459         LASSERT(lli->lli_size_sem_owner == current);
1460         lli->lli_size_sem_owner = NULL;
1461         cfs_up(&lli->lli_size_sem);
1462 }
1463
1464 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1465 {
1466         struct ll_inode_info *lli = ll_i2info(inode);
1467         struct mdt_body *body = md->body;
1468         struct lov_stripe_md *lsm = md->lsm;
1469         struct ll_sb_info *sbi = ll_i2sbi(inode);
1470
1471         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1472         if (lsm != NULL) {
1473                 cfs_down(&lli->lli_och_sem);
1474                 if (lli->lli_smd == NULL) {
1475                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1476                             lsm->lsm_magic != LOV_MAGIC_V3) {
1477                                 dump_lsm(D_ERROR, lsm);
1478                                 LBUG();
1479                         }
1480                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1481                                lsm, inode->i_ino, inode->i_generation, inode);
1482                         /* cl_inode_init must go before lli_smd or a race is
1483                          * possible where client thinks the file has stripes,
1484                          * but lov raid0 is not setup yet and parallel e.g.
1485                          * glimpse would try to use uninitialized lov */
1486                         cl_inode_init(inode, md);
1487                         lli->lli_smd = lsm;
1488                         cfs_up(&lli->lli_och_sem);
1489                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1490                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1491                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1492                 } else {
1493                         cfs_up(&lli->lli_och_sem);
1494                         LASSERT(lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1495                                 lli->lli_smd->lsm_stripe_count ==
1496                                 lsm->lsm_stripe_count);
1497                         if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1498                                 CERROR("lsm mismatch for inode %ld\n",
1499                                        inode->i_ino);
1500                                 CERROR("lli_smd:\n");
1501                                 dump_lsm(D_ERROR, lli->lli_smd);
1502                                 CERROR("lsm:\n");
1503                                 dump_lsm(D_ERROR, lsm);
1504                                 LBUG();
1505                         }
1506                 }
1507                 if (lli->lli_smd != lsm)
1508                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1509         }
1510
1511         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1512                 if (body->valid & OBD_MD_FLRMTPERM)
1513                         ll_update_remote_perm(inode, md->remote_perm);
1514         }
1515 #ifdef CONFIG_FS_POSIX_ACL
1516         else if (body->valid & OBD_MD_FLACL) {
1517                 cfs_spin_lock(&lli->lli_lock);
1518                 if (lli->lli_posix_acl)
1519                         posix_acl_release(lli->lli_posix_acl);
1520                 lli->lli_posix_acl = md->posix_acl;
1521                 cfs_spin_unlock(&lli->lli_lock);
1522         }
1523 #endif
1524         inode->i_ino = cl_fid_build_ino(&body->fid1);
1525         inode->i_generation = cl_fid_build_gen(&body->fid1);
1526
1527         if (body->valid & OBD_MD_FLATIME) {
1528                 if (body->atime > LTIME_S(inode->i_atime))
1529                         LTIME_S(inode->i_atime) = body->atime;
1530                 lli->lli_lvb.lvb_atime = body->atime;
1531         }
1532         if (body->valid & OBD_MD_FLMTIME) {
1533                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1534                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1535                                "to "LPU64"\n", inode->i_ino,
1536                                LTIME_S(inode->i_mtime), body->mtime);
1537                         LTIME_S(inode->i_mtime) = body->mtime;
1538                 }
1539                 lli->lli_lvb.lvb_mtime = body->mtime;
1540         }
1541         if (body->valid & OBD_MD_FLCTIME) {
1542                 if (body->ctime > LTIME_S(inode->i_ctime))
1543                         LTIME_S(inode->i_ctime) = body->ctime;
1544                 lli->lli_lvb.lvb_ctime = body->ctime;
1545         }
1546         if (body->valid & OBD_MD_FLMODE)
1547                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1548         if (body->valid & OBD_MD_FLTYPE)
1549                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1550         LASSERT(inode->i_mode != 0);
1551         if (S_ISREG(inode->i_mode)) {
1552                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1553         } else {
1554                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1555         }
1556 #ifdef HAVE_INODE_BLKSIZE
1557         inode->i_blksize = 1<<inode->i_blkbits;
1558 #endif
1559         if (body->valid & OBD_MD_FLUID)
1560                 inode->i_uid = body->uid;
1561         if (body->valid & OBD_MD_FLGID)
1562                 inode->i_gid = body->gid;
1563         if (body->valid & OBD_MD_FLFLAGS)
1564                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1565         if (body->valid & OBD_MD_FLNLINK)
1566                 inode->i_nlink = body->nlink;
1567         if (body->valid & OBD_MD_FLRDEV)
1568                 inode->i_rdev = old_decode_dev(body->rdev);
1569
1570         if (body->valid & OBD_MD_FLID) {
1571                 /* FID shouldn't be changed! */
1572                 if (fid_is_sane(&lli->lli_fid)) {
1573                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1574                                  "Trying to change FID "DFID
1575                                  " to the "DFID", inode %lu/%u(%p)\n",
1576                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1577                                  inode->i_ino, inode->i_generation, inode);
1578                 } else
1579                         lli->lli_fid = body->fid1;
1580         }
1581
1582         LASSERT(fid_seq(&lli->lli_fid) != 0);
1583
1584         if (body->valid & OBD_MD_FLSIZE) {
1585                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1586                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1587                         struct lustre_handle lockh;
1588                         ldlm_mode_t mode;
1589
1590                         /* As it is possible a blocking ast has been processed
1591                          * by this time, we need to check there is an UPDATE
1592                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1593                          * it. */
1594                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1595                                                &lockh);
1596                         if (mode) {
1597                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1598                                                       LLIF_EPOCH_PENDING |
1599                                                       LLIF_SOM_DIRTY)) {
1600                                         CERROR("ino %lu flags %lu still has "
1601                                                "size authority! do not trust "
1602                                                "the size got from MDS\n",
1603                                                inode->i_ino, lli->lli_flags);
1604                                 } else {
1605                                         /* Use old size assignment to avoid
1606                                          * deadlock bz14138 & bz14326 */
1607                                         inode->i_size = body->size;
1608                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1609                                 }
1610                                 ldlm_lock_decref(&lockh, mode);
1611                         }
1612                 } else {
1613                         /* Use old size assignment to avoid
1614                          * deadlock bz14138 & bz14326 */
1615                         inode->i_size = body->size;
1616
1617                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1618                                inode->i_ino, (unsigned long long)body->size);
1619                 }
1620
1621                 if (body->valid & OBD_MD_FLBLOCKS)
1622                         inode->i_blocks = body->blocks;
1623         }
1624
1625         if (body->valid & OBD_MD_FLMDSCAPA) {
1626                 LASSERT(md->mds_capa);
1627                 ll_add_capa(inode, md->mds_capa);
1628         }
1629         if (body->valid & OBD_MD_FLOSSCAPA) {
1630                 LASSERT(md->oss_capa);
1631                 ll_add_capa(inode, md->oss_capa);
1632         }
1633 }
1634
1635 void ll_read_inode2(struct inode *inode, void *opaque)
1636 {
1637         struct lustre_md *md = opaque;
1638         struct ll_inode_info *lli = ll_i2info(inode);
1639         ENTRY;
1640
1641         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1642                inode->i_ino, inode->i_generation, inode);
1643
1644         ll_lli_init(lli);
1645
1646         LASSERT(!lli->lli_smd);
1647
1648         /* Core attributes from the MDS first.  This is a new inode, and
1649          * the VFS doesn't zero times in the core inode so we have to do
1650          * it ourselves.  They will be overwritten by either MDS or OST
1651          * attributes - we just need to make sure they aren't newer. */
1652         LTIME_S(inode->i_mtime) = 0;
1653         LTIME_S(inode->i_atime) = 0;
1654         LTIME_S(inode->i_ctime) = 0;
1655         inode->i_rdev = 0;
1656         ll_update_inode(inode, md);
1657
1658         /* OIDEBUG(inode); */
1659
1660         /* initializing backing dev info. */
1661         inode->i_mapping->backing_dev_info = &(s2lsi(inode->i_sb)->bdi);
1662
1663
1664         if (S_ISREG(inode->i_mode)) {
1665                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1666                 inode->i_op = &ll_file_inode_operations;
1667                 inode->i_fop = sbi->ll_fop;
1668                 inode->i_mapping->a_ops = &ll_aops;
1669                 EXIT;
1670         } else if (S_ISDIR(inode->i_mode)) {
1671                 inode->i_op = &ll_dir_inode_operations;
1672                 inode->i_fop = &ll_dir_operations;
1673                 inode->i_mapping->a_ops = &ll_dir_aops;
1674                 EXIT;
1675         } else if (S_ISLNK(inode->i_mode)) {
1676                 inode->i_op = &ll_fast_symlink_inode_operations;
1677                 EXIT;
1678         } else {
1679                 inode->i_op = &ll_special_inode_operations;
1680
1681                 init_special_inode(inode, inode->i_mode,
1682                                    kdev_t_to_nr(inode->i_rdev));
1683
1684                 EXIT;
1685         }
1686 }
1687
1688 void ll_delete_inode(struct inode *inode)
1689 {
1690         struct ll_sb_info *sbi = ll_i2sbi(inode);
1691         int rc;
1692         ENTRY;
1693
1694         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1695         if (rc) {
1696                 CERROR("fid_delete() failed, rc %d\n", rc);
1697         }
1698         truncate_inode_pages(&inode->i_data, 0);
1699         clear_inode(inode);
1700
1701         EXIT;
1702 }
1703
1704 int ll_iocontrol(struct inode *inode, struct file *file,
1705                  unsigned int cmd, unsigned long arg)
1706 {
1707         struct ll_sb_info *sbi = ll_i2sbi(inode);
1708         struct ptlrpc_request *req = NULL;
1709         int rc, flags = 0;
1710         ENTRY;
1711
1712         switch(cmd) {
1713         case FSFILT_IOC_GETFLAGS: {
1714                 struct mdt_body *body;
1715                 struct md_op_data *op_data;
1716
1717                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1718                                              0, 0, LUSTRE_OPC_ANY,
1719                                              NULL);
1720                 if (op_data == NULL)
1721                         RETURN(-ENOMEM);
1722
1723                 op_data->op_valid = OBD_MD_FLFLAGS;
1724                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1725                 ll_finish_md_op_data(op_data);
1726                 if (rc) {
1727                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1728                         RETURN(-abs(rc));
1729                 }
1730
1731                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1732
1733                 flags = body->flags;
1734
1735                 ptlrpc_req_finished(req);
1736
1737                 RETURN(put_user(flags, (int *)arg));
1738         }
1739         case FSFILT_IOC_SETFLAGS: {
1740                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1741                 struct obd_info oinfo = { { { 0 } } };
1742                 struct md_op_data *op_data;
1743
1744                 if (get_user(flags, (int *)arg))
1745                         RETURN(-EFAULT);
1746
1747                 oinfo.oi_md = lsm;
1748                 OBDO_ALLOC(oinfo.oi_oa);
1749                 if (!oinfo.oi_oa)
1750                         RETURN(-ENOMEM);
1751
1752                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1753                                              LUSTRE_OPC_ANY, NULL);
1754                 if (IS_ERR(op_data))
1755                         RETURN(PTR_ERR(op_data));
1756
1757                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1758                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1759                 rc = md_setattr(sbi->ll_md_exp, op_data,
1760                                 NULL, 0, NULL, 0, &req, NULL);
1761                 ll_finish_md_op_data(op_data);
1762                 ptlrpc_req_finished(req);
1763                 if (rc) {
1764                         OBDO_FREE(oinfo.oi_oa);
1765                         RETURN(rc);
1766                 }
1767
1768                 if (lsm == NULL) {
1769                         OBDO_FREE(oinfo.oi_oa);
1770                         GOTO(update_cache, rc);
1771                 }
1772
1773                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1774                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1775                 oinfo.oi_oa->o_flags = flags;
1776                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1777                                        OBD_MD_FLGROUP;
1778                 oinfo.oi_capa = ll_mdscapa_get(inode);
1779                 obdo_from_inode(oinfo.oi_oa, inode,
1780                                 &ll_i2info(inode)->lli_fid, 0);
1781                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1782                 capa_put(oinfo.oi_capa);
1783                 OBDO_FREE(oinfo.oi_oa);
1784                 if (rc) {
1785                         if (rc != -EPERM && rc != -EACCES)
1786                                 CERROR("osc_setattr_async fails: rc = %d\n",rc);
1787                         RETURN(rc);
1788                 }
1789
1790                 EXIT;
1791 update_cache:
1792                 inode->i_flags = ll_ext_to_inode_flags(flags);
1793                 return 0;
1794         }
1795         default:
1796                 RETURN(-ENOSYS);
1797         }
1798
1799         RETURN(0);
1800 }
1801
1802 int ll_flush_ctx(struct inode *inode)
1803 {
1804         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1805
1806         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
1807
1808         obd_set_info_async(sbi->ll_md_exp,
1809                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1810                            0, NULL, NULL);
1811         obd_set_info_async(sbi->ll_dt_exp,
1812                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1813                            0, NULL, NULL);
1814         return 0;
1815 }
1816
1817 /* umount -f client means force down, don't save state */
1818 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1819 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1820 {
1821         struct super_block *sb = vfsmnt->mnt_sb;
1822 #else
1823 void ll_umount_begin(struct super_block *sb)
1824 {
1825 #endif
1826         struct lustre_sb_info *lsi = s2lsi(sb);
1827         struct ll_sb_info *sbi = ll_s2sbi(sb);
1828         struct obd_device *obd;
1829         struct obd_ioctl_data *ioc_data;
1830         ENTRY;
1831
1832 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1833         if (!(flags & MNT_FORCE)) {
1834                 EXIT;
1835                 return;
1836         }
1837 #endif
1838
1839         /* Tell the MGC we got umount -f */
1840         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1841
1842         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1843                sb->s_count, atomic_read(&sb->s_active));
1844
1845         obd = class_exp2obd(sbi->ll_md_exp);
1846         if (obd == NULL) {
1847                 CERROR("Invalid MDC connection handle "LPX64"\n",
1848                        sbi->ll_md_exp->exp_handle.h_cookie);
1849                 EXIT;
1850                 return;
1851         }
1852         obd->obd_force = 1;
1853
1854         obd = class_exp2obd(sbi->ll_dt_exp);
1855         if (obd == NULL) {
1856                 CERROR("Invalid LOV connection handle "LPX64"\n",
1857                        sbi->ll_dt_exp->exp_handle.h_cookie);
1858                 EXIT;
1859                 return;
1860         }
1861         obd->obd_force = 1;
1862
1863         OBD_ALLOC_PTR(ioc_data);
1864         if (ioc_data) {
1865                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1866                               sizeof ioc_data, ioc_data, NULL);
1867
1868                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1869                               sizeof ioc_data, ioc_data, NULL);
1870
1871                 OBD_FREE_PTR(ioc_data);
1872         }
1873
1874
1875         /* Really, we'd like to wait until there are no requests outstanding,
1876          * and then continue.  For now, we just invalidate the requests,
1877          * schedule() and sleep one second if needed, and hope.
1878          */
1879         cfs_schedule();
1880 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1881         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1882                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
1883                                                    cfs_time_seconds(1));
1884                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1885                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1886                                       "may try to umount it a bit later\n",
1887                                       atomic_read(&vfsmnt->mnt_count));
1888         }
1889 #endif
1890
1891         EXIT;
1892 }
1893
1894 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1895 {
1896         struct ll_sb_info *sbi = ll_s2sbi(sb);
1897         int err;
1898         __u32 read_only;
1899
1900         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1901                 read_only = *flags & MS_RDONLY;
1902                 err = obd_set_info_async(sbi->ll_md_exp,
1903                                          sizeof(KEY_READ_ONLY),
1904                                          KEY_READ_ONLY, sizeof(read_only),
1905                                          &read_only, NULL);
1906                 if (err) {
1907                         CERROR("Failed to change the read-only flag during "
1908                                "remount: %d\n", err);
1909                         return err;
1910                 }
1911
1912                 if (read_only)
1913                         sb->s_flags |= MS_RDONLY;
1914                 else
1915                         sb->s_flags &= ~MS_RDONLY;
1916         }
1917         return 0;
1918 }
1919
1920 int ll_prep_inode(struct inode **inode,
1921                   struct ptlrpc_request *req,
1922                   struct super_block *sb)
1923 {
1924         struct ll_sb_info *sbi = NULL;
1925         struct lustre_md md;
1926         int rc;
1927         ENTRY;
1928
1929         LASSERT(*inode || sb);
1930         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1931         memset(&md, 0, sizeof(struct lustre_md));
1932
1933         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1934                               sbi->ll_md_exp, &md);
1935         if (rc)
1936                 RETURN(rc);
1937
1938         if (*inode) {
1939                 ll_update_inode(*inode, &md);
1940         } else {
1941                 LASSERT(sb != NULL);
1942
1943                 /*
1944                  * At this point server returns to client's same fid as client
1945                  * generated for creating. So using ->fid1 is okay here.
1946                  */
1947                 LASSERT(fid_is_sane(&md.body->fid1));
1948
1949                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1), &md);
1950                 if (*inode == NULL || IS_ERR(*inode)) {
1951                         if (md.lsm)
1952                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
1953 #ifdef CONFIG_FS_POSIX_ACL
1954                         if (md.posix_acl) {
1955                                 posix_acl_release(md.posix_acl);
1956                                 md.posix_acl = NULL;
1957                         }
1958 #endif
1959                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
1960                         *inode = NULL;
1961                         CERROR("new_inode -fatal: rc %d\n", rc);
1962                         GOTO(out, rc);
1963                 }
1964         }
1965
1966 out:
1967         md_free_lustre_md(sbi->ll_md_exp, &md);
1968         RETURN(rc);
1969 }
1970
1971 int ll_obd_statfs(struct inode *inode, void *arg)
1972 {
1973         struct ll_sb_info *sbi = NULL;
1974         struct obd_export *exp;
1975         char *buf = NULL;
1976         struct obd_ioctl_data *data = NULL;
1977         __u32 type;
1978         int len = 0, rc;
1979
1980         if (!inode || !(sbi = ll_i2sbi(inode)))
1981                 GOTO(out_statfs, rc = -EINVAL);
1982
1983         rc = obd_ioctl_getdata(&buf, &len, arg);
1984         if (rc)
1985                 GOTO(out_statfs, rc);
1986
1987         data = (void*)buf;
1988         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1989             !data->ioc_pbuf1 || !data->ioc_pbuf2)
1990                 GOTO(out_statfs, rc = -EINVAL);
1991
1992         if (data->ioc_inllen1 != sizeof(__u32) ||
1993             data->ioc_inllen2 != sizeof(__u32) ||
1994             data->ioc_plen1 != sizeof(struct obd_statfs) ||
1995             data->ioc_plen2 != sizeof(struct obd_uuid))
1996                 GOTO(out_statfs, rc = -EINVAL);
1997
1998         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
1999         if (type == LL_STATFS_MDC)
2000                 exp = sbi->ll_md_exp;
2001         else if (type == LL_STATFS_LOV)
2002                 exp = sbi->ll_dt_exp;
2003         else
2004                 GOTO(out_statfs, rc = -ENODEV);
2005
2006         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2007         if (rc)
2008                 GOTO(out_statfs, rc);
2009 out_statfs:
2010         if (buf)
2011                 obd_ioctl_freedata(buf, len);
2012         return rc;
2013 }
2014
2015 int ll_process_config(struct lustre_cfg *lcfg)
2016 {
2017         char *ptr;
2018         void *sb;
2019         struct lprocfs_static_vars lvars;
2020         unsigned long x;
2021         int rc = 0;
2022
2023         lprocfs_llite_init_vars(&lvars);
2024
2025         /* The instance name contains the sb: lustre-client-aacfe000 */
2026         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2027         if (!ptr || !*(++ptr))
2028                 return -EINVAL;
2029         if (sscanf(ptr, "%lx", &x) != 1)
2030                 return -EINVAL;
2031         sb = (void *)x;
2032         /* This better be a real Lustre superblock! */
2033         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2034
2035         /* Note we have not called client_common_fill_super yet, so
2036            proc fns must be able to handle that! */
2037         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2038                                       lcfg, sb);
2039         if (rc > 0)
2040                 rc = 0;
2041         return(rc);
2042 }
2043
2044 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2045 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2046                                        struct inode *i1, struct inode *i2,
2047                                        const char *name, int namelen,
2048                                        int mode, __u32 opc, void *data)
2049 {
2050         LASSERT(i1 != NULL);
2051
2052         if (namelen > ll_i2sbi(i1)->ll_namelen)
2053                 return ERR_PTR(-ENAMETOOLONG);
2054
2055         if (op_data == NULL)
2056                 OBD_ALLOC_PTR(op_data);
2057
2058         if (op_data == NULL)
2059                 return ERR_PTR(-ENOMEM);
2060
2061         ll_i2gids(op_data->op_suppgids, i1, i2);
2062         op_data->op_fid1 = *ll_inode2fid(i1);
2063         op_data->op_capa1 = ll_mdscapa_get(i1);
2064
2065         if (i2) {
2066                 op_data->op_fid2 = *ll_inode2fid(i2);
2067                 op_data->op_capa2 = ll_mdscapa_get(i2);
2068         } else {
2069                 fid_zero(&op_data->op_fid2);
2070                 op_data->op_capa2 = NULL;
2071         }
2072
2073         op_data->op_name = name;
2074         op_data->op_namelen = namelen;
2075         op_data->op_mode = mode;
2076         op_data->op_mod_time = cfs_time_current_sec();
2077         op_data->op_fsuid = cfs_curproc_fsuid();
2078         op_data->op_fsgid = cfs_curproc_fsgid();
2079         op_data->op_cap = cfs_curproc_cap_pack();
2080         op_data->op_bias = MDS_CHECK_SPLIT;
2081         op_data->op_opc = opc;
2082         op_data->op_mds = 0;
2083         op_data->op_data = data;
2084
2085         return op_data;
2086 }
2087
2088 void ll_finish_md_op_data(struct md_op_data *op_data)
2089 {
2090         capa_put(op_data->op_capa1);
2091         capa_put(op_data->op_capa2);
2092         OBD_FREE_PTR(op_data);
2093 }
2094
2095 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2096 {
2097         struct ll_sb_info *sbi;
2098
2099         LASSERT((seq != NULL) && (vfs != NULL));
2100         sbi = ll_s2sbi(vfs->mnt_sb);
2101
2102         if (sbi->ll_flags & LL_SBI_NOLCK)
2103                 seq_puts(seq, ",nolock");
2104
2105         if (sbi->ll_flags & LL_SBI_FLOCK)
2106                 seq_puts(seq, ",flock");
2107
2108         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2109                 seq_puts(seq, ",localflock");
2110
2111         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2112                 seq_puts(seq, ",user_xattr");
2113
2114         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2115                 seq_puts(seq, ",lazystatfs");
2116
2117         RETURN(0);
2118 }