lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static struct ll_sb_info *ll_init_sbi(void)
  73 {
  74         struct ll_sb_info *sbi = NULL;
  75         unsigned long pages;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC(sbi, sizeof(*sbi));
  82         if (!sbi)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         spin_lock_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  94 #ifdef HAVE_BGL_SUPPORT
  95                 sbi->ll_async_page_max = pages / 4;
  96 #else
  97                 sbi->ll_async_page_max = pages / 2;
  98 #endif
  99         } else {
 100                 sbi->ll_async_page_max = (pages / 4) * 3;
 101         }
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119 #ifdef ENABLE_LLITE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127 #ifdef HAVE_EXPORT___IGET
 128         INIT_LIST_HEAD(&sbi->ll_deathrow);
 129         spin_lock_init(&sbi->ll_deathrow_lock);
 130 #endif
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 133                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 134         }
 135
 136         /* metadata statahead is enabled by default */
 137         sbi->ll_sa_max = LL_SA_RPC_DEF;
 138
 139         RETURN(sbi);
 140 }
 141
 142 void ll_free_sbi(struct super_block *sb)
 143 {
 144         struct ll_sb_info *sbi = ll_s2sbi(sb);
 145         ENTRY;
 146
 147         if (sbi != NULL) {
 148                 spin_lock(&ll_sb_lock);
 149                 list_del(&sbi->ll_list);
 150                 spin_unlock(&ll_sb_lock);
 151                 OBD_FREE(sbi, sizeof(*sbi));
 152         }
 153         EXIT;
 154 }
 155
 156 static struct dentry_operations ll_d_root_ops = {
 157 #ifdef DCACHE_LUSTRE_INVALID
 158         .d_compare = ll_dcompare,
 159 #endif
 160         .d_revalidate = ll_revalidate_nd,
 161 };
 162
 163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 164 {
 165         struct inode *root = 0;
 166         struct ll_sb_info *sbi = ll_s2sbi(sb);
 167         struct obd_device *obd;
 168         struct obd_capa *oc = NULL;
 169         struct obd_statfs osfs;
 170         struct ptlrpc_request *request = NULL;
 171         struct obd_connect_data *data = NULL;
 172         struct obd_uuid *uuid;
 173         struct lustre_md lmd;
 174         obd_valid valid;
 175         int size, err, checksum;
 176         ENTRY;
 177
 178         obd = class_name2obd(md);
 179         if (!obd) {
 180                 CERROR("MD %s: not setup or attached\n", md);
 181                 RETURN(-EINVAL);
 182         }
 183
 184         OBD_ALLOC_PTR(data);
 185         if (data == NULL)
 186                 RETURN(-ENOMEM);
 187
 188         if (proc_lustre_fs_root) {
 189                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 190                                                   dt, md);
 191                 if (err < 0)
 192                         CERROR("could not register mount in /proc/fs/lustre\n");
 193         }
 194
 195         /* indicate the features supported by this client */
 196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 197                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 199                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 200                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 201                                   OBD_CONNECT_LOV_V3 | OBD_CONNECT_RMT_CLIENT |
 202                                   OBD_CONNECT_VBR;
 203
 204 #ifdef HAVE_LRU_RESIZE_SUPPORT
 205         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 206                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 207 #endif
 208 #ifdef CONFIG_FS_POSIX_ACL
 209         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 210 #endif
 211         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213
 214         if (sb->s_flags & MS_RDONLY)
 215                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 216         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 217                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 218
 219 #ifdef HAVE_MS_FLOCK_LOCK
 220         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 221         sb->s_flags |= MS_FLOCK_LOCK;
 222 #endif
 223
 224         if (sbi->ll_flags & LL_SBI_FLOCK)
 225                 sbi->ll_fop = &ll_file_operations_flock;
 226         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 227                 sbi->ll_fop = &ll_file_operations;
 228         else
 229                 sbi->ll_fop = &ll_file_operations_noflock;
 230
 231         /* real client */
 232         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 233         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 234                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 235
 236         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 237         if (err == -EBUSY) {
 238                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 239                                    "recovery, of which this client is not a "
 240                                    "part. Please wait for recovery to complete,"
 241                                    " abort, or time out.\n", md);
 242                 GOTO(out, err);
 243         } else if (err) {
 244                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 245                 GOTO(out, err);
 246         }
 247
 248         err = obd_fid_init(sbi->ll_md_exp);
 249         if (err) {
 250                 CERROR("Can't init metadata layer FID infrastructure, "
 251                        "rc %d\n", err);
 252                 GOTO(out_md, err);
 253         }
 254
 255         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 256         if (err)
 257                 GOTO(out_md_fid, err);
 258
 259         size = sizeof(*data);
 260         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 261                            KEY_CONN_DATA,  &size, data, NULL);
 262         if (err) {
 263                 CERROR("Get connect data failed: %d \n", err);
 264                 GOTO(out_md, err);
 265         }
 266
 267         LASSERT(osfs.os_bsize);
 268         sb->s_blocksize = osfs.os_bsize;
 269         sb->s_blocksize_bits = log2(osfs.os_bsize);
 270         sb->s_magic = LL_SUPER_MAGIC;
 271
 272         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 273          *         retval = in_file->f_op->sendfile(...);
 274          *         if (*ppos > max)
 275          *                 retval = -EOVERFLOW;
 276          *
 277          * it will check if *ppos is greater than max. However, max equals to
 278          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 279          * has been defined as a signed long long ineger in linux kernel. */
 280 #if BITS_PER_LONG == 64
 281         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 282 #else
 283         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 284 #endif
 285         sbi->ll_namelen = osfs.os_namelen;
 286         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 287
 288         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 289             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 290                 LCONSOLE_INFO("Disabling user_xattr feature because "
 291                               "it is not supported on the server\n");
 292                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 293         }
 294
 295         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 296 #ifdef MS_POSIXACL
 297                 sb->s_flags |= MS_POSIXACL;
 298 #endif
 299                 sbi->ll_flags |= LL_SBI_ACL;
 300         } else {
 301                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 302 #ifdef MS_POSIXACL
 303                 sb->s_flags &= ~MS_POSIXACL;
 304 #endif
 305                 sbi->ll_flags &= ~LL_SBI_ACL;
 306         }
 307
 308         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 309                 sbi->ll_flags |= LL_SBI_JOIN;
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 312                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 313                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 314                         LCONSOLE_INFO("client is set as remote by default.\n");
 315                 }
 316         } else {
 317                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 318                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 319                         LCONSOLE_INFO("client claims to be remote, but server "
 320                                       "rejected, forced to be local.\n");
 321                 }
 322         }
 323
 324         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 325                 LCONSOLE_INFO("client enabled MDS capability!\n");
 326                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 327         }
 328
 329         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 330                 LCONSOLE_INFO("client enabled OSS capability!\n");
 331                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 332         }
 333
 334         obd = class_name2obd(dt);
 335         if (!obd) {
 336                 CERROR("DT %s: not setup or attached\n", dt);
 337                 GOTO(out_md_fid, err = -ENODEV);
 338         }
 339
 340         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 341                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 342                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 343                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 344                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 345                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 346                                   OBD_CONNECT_GRANT_SHRINK;
 347
 348         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 349                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 350                  * disabled by default, because it can still be enabled on the
 351                  * fly via /proc. As a consequence, we still need to come to an
 352                  * agreement on the supported algorithms at connect time */
 353                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 354
 355                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 356                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 357                 else
 358                         /* send the list of supported checksum types */
 359                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 360         }
 361
 362 #ifdef HAVE_LRU_RESIZE_SUPPORT
 363         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 364 #endif
 365         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 366                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 367
 368         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 369                "ocd_grant: %d\n", data->ocd_connect_flags,
 370                data->ocd_version, data->ocd_grant);
 371
 372         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 373         obd->obd_upcall.onu_upcall = cl_ocd_update;
 374         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 375
 376         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 377         if (err == -EBUSY) {
 378                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 379                                    "recovery, of which this client is not a "
 380                                    "part.  Please wait for recovery to "
 381                                    "complete, abort, or time out.\n", dt);
 382                 GOTO(out_md_fid, err);
 383         } else if (err) {
 384                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 385                 GOTO(out_md_fid, err);
 386         }
 387
 388         err = obd_fid_init(sbi->ll_dt_exp);
 389         if (err) {
 390                 CERROR("Can't init data layer FID infrastructure, "
 391                        "rc %d\n", err);
 392                 GOTO(out_dt, err);
 393         }
 394
 395         spin_lock(&sbi->ll_lco.lco_lock);
 396         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 397         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 398         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 399         spin_unlock(&sbi->ll_lco.lco_lock);
 400
 401         fid_zero(&sbi->ll_root_fid);
 402         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 403         if (err) {
 404                 CERROR("cannot mds_connect: rc = %d\n", err);
 405                 GOTO(out_lock_cn_cb, err);
 406         }
 407         if (!fid_is_sane(&sbi->ll_root_fid)) {
 408                 CERROR("Invalid root fid during mount\n");
 409                 GOTO(out_lock_cn_cb, err = -EINVAL);
 410         }
 411         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 412
 413         sb->s_op = &lustre_super_operations;
 414         sb->s_export_op = &lustre_export_operations;
 415
 416         /* make root inode
 417          * XXX: move this to after cbd setup? */
 418         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 419         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 420                 valid |= OBD_MD_FLRMTPERM;
 421         else if (sbi->ll_flags & LL_SBI_ACL)
 422                 valid |= OBD_MD_FLACL;
 423
 424         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 425                          &request);
 426         if (oc)
 427                 capa_put(oc);
 428         if (err) {
 429                 CERROR("md_getattr failed for root: rc = %d\n", err);
 430                 GOTO(out_lock_cn_cb, err);
 431         }
 432         memset(&lmd, 0, sizeof(lmd));
 433         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 434                                sbi->ll_md_exp, &lmd);
 435         if (err) {
 436                 CERROR("failed to understand root inode md: rc = %d\n", err);
 437                 ptlrpc_req_finished (request);
 438                 GOTO(out_lock_cn_cb, err);
 439         }
 440
 441         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 442         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 443         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 444         ptlrpc_req_finished(request);
 445
 446         if (root == NULL || IS_ERR(root)) {
 447                 if (lmd.lsm)
 448                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 449 #ifdef CONFIG_FS_POSIX_ACL
 450                 if (lmd.posix_acl) {
 451                         posix_acl_release(lmd.posix_acl);
 452                         lmd.posix_acl = NULL;
 453                 }
 454 #endif
 455                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 456                 root = NULL;
 457                 CERROR("lustre_lite: bad iget4 for root\n");
 458                 GOTO(out_root, err);
 459         }
 460
 461         err = ll_close_thread_start(&sbi->ll_lcq);
 462         if (err) {
 463                 CERROR("cannot start close thread: rc %d\n", err);
 464                 GOTO(out_root, err);
 465         }
 466
 467 #ifdef CONFIG_FS_POSIX_ACL
 468         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 469                 rct_init(&sbi->ll_rct);
 470                 et_init(&sbi->ll_et);
 471         }
 472 #endif
 473
 474         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 475         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 476                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 477                                  NULL);
 478         cl_sb_init(sb);
 479
 480         sb->s_root = d_alloc_root(root);
 481         if (data != NULL)
 482                 OBD_FREE(data, sizeof(*data));
 483
 484         sb->s_root->d_op = &ll_d_root_ops;
 485
 486         sbi->ll_sdev_orig = sb->s_dev;
 487
 488         /* We set sb->s_dev equal on all lustre clients in order to support
 489          * NFS export clustering.  NFSD requires that the FSID be the same
 490          * on all clients. */
 491         /* s_dev is also used in lt_compare() to compare two fs, but that is
 492          * only a node-local comparison. */
 493         uuid = obd_get_uuid(sbi->ll_md_exp);
 494         if (uuid != NULL)
 495                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 496
 497         RETURN(err);
 498 out_root:
 499         if (root)
 500                 iput(root);
 501 out_lock_cn_cb:
 502         obd_fid_fini(sbi->ll_dt_exp);
 503 out_dt:
 504         obd_disconnect(sbi->ll_dt_exp);
 505         sbi->ll_dt_exp = NULL;
 506 out_md_fid:
 507         obd_fid_fini(sbi->ll_md_exp);
 508 out_md:
 509         obd_disconnect(sbi->ll_md_exp);
 510         sbi->ll_md_exp = NULL;
 511 out:
 512         if (data != NULL)
 513                 OBD_FREE_PTR(data);
 514         lprocfs_unregister_mountpoint(sbi);
 515         return err;
 516 }
 517
 518 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 519 {
 520         int size, rc;
 521
 522         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 523         size = sizeof(int);
 524         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 525                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 526         if (rc)
 527                 CERROR("Get max mdsize error rc %d \n", rc);
 528
 529         RETURN(rc);
 530 }
 531
 532 void ll_dump_inode(struct inode *inode)
 533 {
 534         struct list_head *tmp;
 535         int dentry_count = 0;
 536
 537         LASSERT(inode != NULL);
 538
 539         list_for_each(tmp, &inode->i_dentry)
 540                 dentry_count++;
 541
 542         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 543                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 544                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 545 }
 546
 547 void lustre_dump_dentry(struct dentry *dentry, int recur)
 548 {
 549         struct list_head *tmp;
 550         int subdirs = 0;
 551
 552         LASSERT(dentry != NULL);
 553
 554         list_for_each(tmp, &dentry->d_subdirs)
 555                 subdirs++;
 556
 557         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 558                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 559                dentry->d_name.len, dentry->d_name.name,
 560                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 561                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 562                dentry->d_flags, dentry->d_fsdata, subdirs);
 563         if (dentry->d_inode != NULL)
 564                 ll_dump_inode(dentry->d_inode);
 565
 566         if (recur == 0)
 567                 return;
 568
 569         list_for_each(tmp, &dentry->d_subdirs) {
 570                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 571                 lustre_dump_dentry(d, recur - 1);
 572         }
 573 }
 574
 575 #ifdef HAVE_EXPORT___IGET
 576 static void prune_dir_dentries(struct inode *inode)
 577 {
 578         struct dentry *dentry, *prev = NULL;
 579
 580         /* due to lustre specific logic, a directory
 581          * can have few dentries - a bug from VFS POV */
 582 restart:
 583         spin_lock(&dcache_lock);
 584         if (!list_empty(&inode->i_dentry)) {
 585                 dentry = list_entry(inode->i_dentry.prev,
 586                                     struct dentry, d_alias);
 587                 /* in order to prevent infinite loops we
 588                  * break if previous dentry is busy */
 589                 if (dentry != prev) {
 590                         prev = dentry;
 591                         dget_locked(dentry);
 592                         spin_unlock(&dcache_lock);
 593
 594                         /* try to kill all child dentries */
 595                         lock_dentry(dentry);
 596                         shrink_dcache_parent(dentry);
 597                         unlock_dentry(dentry);
 598                         dput(dentry);
 599
 600                         /* now try to get rid of current dentry */
 601                         d_prune_aliases(inode);
 602                         goto restart;
 603                 }
 604         }
 605         spin_unlock(&dcache_lock);
 606 }
 607
 608 static void prune_deathrow_one(struct ll_inode_info *lli)
 609 {
 610         struct inode *inode = ll_info2i(lli);
 611
 612         /* first, try to drop any dentries - they hold a ref on the inode */
 613         if (S_ISDIR(inode->i_mode))
 614                 prune_dir_dentries(inode);
 615         else
 616                 d_prune_aliases(inode);
 617
 618
 619         /* if somebody still uses it, leave it */
 620         LASSERT(atomic_read(&inode->i_count) > 0);
 621         if (atomic_read(&inode->i_count) > 1)
 622                 goto out;
 623
 624         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 625                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 626
 627         /* seems nobody uses it anymore */
 628         inode->i_nlink = 0;
 629
 630 out:
 631         iput(inode);
 632         return;
 633 }
 634
 635 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 636 {
 637         struct ll_inode_info *lli;
 638         int empty;
 639
 640         do {
 641                 if (need_resched() && try)
 642                         break;
 643
 644                 if (try) {
 645                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 646                                 break;
 647                 } else {
 648                         spin_lock(&sbi->ll_deathrow_lock);
 649                 }
 650
 651                 empty = 1;
 652                 lli = NULL;
 653                 if (!list_empty(&sbi->ll_deathrow)) {
 654                         lli = list_entry(sbi->ll_deathrow.next,
 655                                          struct ll_inode_info,
 656                                          lli_dead_list);
 657                         list_del_init(&lli->lli_dead_list);
 658                         if (!list_empty(&sbi->ll_deathrow))
 659                                 empty = 0;
 660                 }
 661                 spin_unlock(&sbi->ll_deathrow_lock);
 662
 663                 if (lli)
 664                         prune_deathrow_one(lli);
 665
 666         } while (empty == 0);
 667 }
 668 #else /* !HAVE_EXPORT___IGET */
 669 #define prune_deathrow(sbi, try) do {} while (0)
 670 #endif /* HAVE_EXPORT___IGET */
 671
 672 void client_common_put_super(struct super_block *sb)
 673 {
 674         struct ll_sb_info *sbi = ll_s2sbi(sb);
 675         ENTRY;
 676
 677 #ifdef CONFIG_FS_POSIX_ACL
 678         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 679                 et_fini(&sbi->ll_et);
 680                 rct_fini(&sbi->ll_rct);
 681         }
 682 #endif
 683
 684         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 685
 686         ll_close_thread_shutdown(sbi->ll_lcq);
 687
 688         /* destroy inodes in deathrow */
 689         prune_deathrow(sbi, 0);
 690
 691         list_del(&sbi->ll_conn_chain);
 692
 693         obd_fid_fini(sbi->ll_dt_exp);
 694         obd_disconnect(sbi->ll_dt_exp);
 695         sbi->ll_dt_exp = NULL;
 696
 697         lprocfs_unregister_mountpoint(sbi);
 698
 699         obd_fid_fini(sbi->ll_md_exp);
 700         obd_disconnect(sbi->ll_md_exp);
 701         sbi->ll_md_exp = NULL;
 702
 703         EXIT;
 704 }
 705
 706 void ll_kill_super(struct super_block *sb)
 707 {
 708         struct ll_sb_info *sbi;
 709
 710         ENTRY;
 711
 712         /* not init sb ?*/
 713         if (!(sb->s_flags & MS_ACTIVE))
 714                 return;
 715
 716         sbi = ll_s2sbi(sb);
 717         /* we need restore s_dev from changed for clustred NFS before put_super
 718          * because new kernels have cached s_dev and change sb->s_dev in
 719          * put_super not affected real removing devices */
 720         if (sbi)
 721                 sb->s_dev = sbi->ll_sdev_orig;
 722         EXIT;
 723 }
 724
 725 char *ll_read_opt(const char *opt, char *data)
 726 {
 727         char *value;
 728         char *retval;
 729         ENTRY;
 730
 731         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 732         if (strncmp(opt, data, strlen(opt)))
 733                 RETURN(NULL);
 734         if ((value = strchr(data, '=')) == NULL)
 735                 RETURN(NULL);
 736
 737         value++;
 738         OBD_ALLOC(retval, strlen(value) + 1);
 739         if (!retval) {
 740                 CERROR("out of memory!\n");
 741                 RETURN(NULL);
 742         }
 743
 744         memcpy(retval, value, strlen(value)+1);
 745         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 746         RETURN(retval);
 747 }
 748
 749 static inline int ll_set_opt(const char *opt, char *data, int fl)
 750 {
 751         if (strncmp(opt, data, strlen(opt)) != 0)
 752                 return(0);
 753         else
 754                 return(fl);
 755 }
 756
 757 /* non-client-specific mount options are parsed in lmd_parse */
 758 static int ll_options(char *options, int *flags)
 759 {
 760         int tmp;
 761         char *s1 = options, *s2;
 762         ENTRY;
 763
 764         if (!options)
 765                 RETURN(0);
 766
 767         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 768
 769         while (*s1) {
 770                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 771                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 772                 if (tmp) {
 773                         *flags |= tmp;
 774                         goto next;
 775                 }
 776                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 777                 if (tmp) {
 778                         *flags |= tmp;
 779                         goto next;
 780                 }
 781                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 782                 if (tmp) {
 783                         *flags |= tmp;
 784                         goto next;
 785                 }
 786                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 787                 if (tmp) {
 788                         *flags &= ~tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 797                 if (tmp) {
 798                         *flags &= ~tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         /* Ignore deprecated mount option.  The client will
 804                          * always try to mount with ACL support, whether this
 805                          * is used depends on whether server supports it. */
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 809                 if (tmp) {
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817
 818                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 839                 if (tmp) {
 840                         *flags |= tmp;
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 844                 if (tmp) {
 845                         *flags &= ~tmp;
 846                         goto next;
 847                 }
 848
 849                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 850                                    s1);
 851                 RETURN(-EINVAL);
 852
 853 next:
 854                 /* Find next opt */
 855                 s2 = strchr(s1, ',');
 856                 if (s2 == NULL)
 857                         break;
 858                 s1 = s2 + 1;
 859         }
 860         RETURN(0);
 861 }
 862
 863 void ll_lli_init(struct ll_inode_info *lli)
 864 {
 865         lli->lli_inode_magic = LLI_INODE_MAGIC;
 866         sema_init(&lli->lli_size_sem, 1);
 867         sema_init(&lli->lli_write_sem, 1);
 868         lli->lli_flags = 0;
 869         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 870         spin_lock_init(&lli->lli_lock);
 871         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 872         INIT_LIST_HEAD(&lli->lli_close_list);
 873         lli->lli_inode_magic = LLI_INODE_MAGIC;
 874         sema_init(&lli->lli_och_sem, 1);
 875         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 876         lli->lli_mds_exec_och = NULL;
 877         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 878         lli->lli_open_fd_exec_count = 0;
 879         INIT_LIST_HEAD(&lli->lli_dead_list);
 880         lli->lli_remote_perms = NULL;
 881         lli->lli_rmtperm_utime = 0;
 882         sema_init(&lli->lli_rmtperm_sem, 1);
 883         INIT_LIST_HEAD(&lli->lli_oss_capas);
 884 }
 885
 886 int ll_fill_super(struct super_block *sb)
 887 {
 888         struct lustre_profile *lprof;
 889         struct lustre_sb_info *lsi = s2lsi(sb);
 890         struct ll_sb_info *sbi;
 891         char  *dt = NULL, *md = NULL;
 892         char  *profilenm = get_profile_name(sb);
 893         struct config_llog_instance cfg = {0, };
 894         char   ll_instance[sizeof(sb) * 2 + 1];
 895         int    err;
 896         ENTRY;
 897
 898         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 899
 900         cfs_module_get();
 901
 902         /* client additional sb info */
 903         lsi->lsi_llsbi = sbi = ll_init_sbi();
 904         if (!sbi) {
 905                 cfs_module_put();
 906                 RETURN(-ENOMEM);
 907         }
 908
 909         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 910         if (err)
 911                 GOTO(out_free, err);
 912
 913         /* Generate a string unique to this super, in case some joker tries
 914            to mount the same fs at two mount points.
 915            Use the address of the super itself.*/
 916         sprintf(ll_instance, "%p", sb);
 917         cfg.cfg_instance = ll_instance;
 918         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 919
 920         /* set up client obds */
 921         err = lustre_process_log(sb, profilenm, &cfg);
 922         if (err < 0) {
 923                 CERROR("Unable to process log: %d\n", err);
 924                 GOTO(out_free, err);
 925         }
 926
 927         lprof = class_get_profile(profilenm);
 928         if (lprof == NULL) {
 929                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 930                                    " read from the MGS.  Does that filesystem "
 931                                    "exist?\n", profilenm);
 932                 GOTO(out_free, err = -EINVAL);
 933         }
 934         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 935                lprof->lp_md, lprof->lp_dt);
 936
 937         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 938                   strlen(ll_instance) + 2);
 939         if (!dt)
 940                 GOTO(out_free, err = -ENOMEM);
 941         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 942
 943         OBD_ALLOC(md, strlen(lprof->lp_md) +
 944                   strlen(ll_instance) + 2);
 945         if (!md)
 946                 GOTO(out_free, err = -ENOMEM);
 947         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 948
 949         /* connections, registrations, sb setup */
 950         err = client_common_fill_super(sb, md, dt);
 951
 952 out_free:
 953         if (md)
 954                 OBD_FREE(md, strlen(md) + 1);
 955         if (dt)
 956                 OBD_FREE(dt, strlen(dt) + 1);
 957         if (err)
 958                 ll_put_super(sb);
 959         else
 960                 LCONSOLE_WARN("Client %s has started\n", profilenm);
 961
 962         RETURN(err);
 963 } /* ll_fill_super */
 964
 965
 966 void lu_context_keys_dump(void);
 967
 968 void ll_put_super(struct super_block *sb)
 969 {
 970         struct config_llog_instance cfg;
 971         char   ll_instance[sizeof(sb) * 2 + 1];
 972         struct obd_device *obd;
 973         struct lustre_sb_info *lsi = s2lsi(sb);
 974         struct ll_sb_info *sbi = ll_s2sbi(sb);
 975         char *profilenm = get_profile_name(sb);
 976         int force = 1, next;
 977         ENTRY;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 980
 981         ll_print_capa_stat(sbi);
 982
 983         sprintf(ll_instance, "%p", sb);
 984         cfg.cfg_instance = ll_instance;
 985         lustre_end_log(sb, NULL, &cfg);
 986
 987         if (sbi->ll_md_exp) {
 988                 obd = class_exp2obd(sbi->ll_md_exp);
 989                 if (obd)
 990                         force = obd->obd_force;
 991         }
 992
 993         /* We need to set force before the lov_disconnect in
 994            lustre_common_put_super, since l_d cleans up osc's as well. */
 995         if (force) {
 996                 next = 0;
 997                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
 998                                                      &next)) != NULL) {
 999                         obd->obd_force = force;
1000                 }
1001         }
1002
1003         cl_sb_fini(sb);
1004
1005         if (sbi->ll_lcq) {
1006                 /* Only if client_common_fill_super succeeded */
1007                 client_common_put_super(sb);
1008         }
1009         next = 0;
1010         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1011                 class_manual_cleanup(obd);
1012         }
1013
1014         if (profilenm)
1015                 class_del_profile(profilenm);
1016
1017         ll_free_sbi(sb);
1018         lsi->lsi_llsbi = NULL;
1019
1020         lustre_common_put_super(sb);
1021
1022         cl_env_cache_purge(~0);
1023
1024         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1025
1026         cfs_module_put();
1027
1028         EXIT;
1029 } /* client_put_super */
1030
1031 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1032 {
1033         struct inode *inode = NULL;
1034         /* NOTE: we depend on atomic igrab() -bzzz */
1035         lock_res_and_lock(lock);
1036         if (lock->l_ast_data) {
1037                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1038                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1039                         inode = igrab(lock->l_ast_data);
1040                 } else {
1041                         inode = lock->l_ast_data;
1042                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1043                                                 D_INFO : D_WARNING,
1044                                         lock, __FILE__, __func__, __LINE__,
1045                                         "l_ast_data %p is bogus: magic %08x",
1046                                         lock->l_ast_data, lli->lli_inode_magic);
1047                         inode = NULL;
1048                 }
1049         }
1050         unlock_res_and_lock(lock);
1051         return inode;
1052 }
1053
1054 static int null_if_equal(struct ldlm_lock *lock, void *data)
1055 {
1056         if (data == lock->l_ast_data) {
1057                 lock->l_ast_data = NULL;
1058
1059                 if (lock->l_req_mode != lock->l_granted_mode)
1060                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1061         }
1062
1063         return LDLM_ITER_CONTINUE;
1064 }
1065
1066 void ll_clear_inode(struct inode *inode)
1067 {
1068         struct ll_inode_info *lli = ll_i2info(inode);
1069         struct ll_sb_info *sbi = ll_i2sbi(inode);
1070         ENTRY;
1071
1072         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1073                inode->i_generation, inode);
1074
1075         if (S_ISDIR(inode->i_mode)) {
1076                 /* these should have been cleared in ll_file_release */
1077                 LASSERT(lli->lli_sai == NULL);
1078                 LASSERT(lli->lli_opendir_key == NULL);
1079                 LASSERT(lli->lli_opendir_pid == 0);
1080         }
1081
1082         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1083         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1084                          null_if_equal, inode);
1085
1086         LASSERT(!lli->lli_open_fd_write_count);
1087         LASSERT(!lli->lli_open_fd_read_count);
1088         LASSERT(!lli->lli_open_fd_exec_count);
1089
1090         if (lli->lli_mds_write_och)
1091                 ll_md_real_close(inode, FMODE_WRITE);
1092         if (lli->lli_mds_exec_och)
1093                 ll_md_real_close(inode, FMODE_EXEC);
1094         if (lli->lli_mds_read_och)
1095                 ll_md_real_close(inode, FMODE_READ);
1096
1097         if (lli->lli_symlink_name) {
1098                 OBD_FREE(lli->lli_symlink_name,
1099                          strlen(lli->lli_symlink_name) + 1);
1100                 lli->lli_symlink_name = NULL;
1101         }
1102
1103         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1104                 LASSERT(lli->lli_posix_acl == NULL);
1105                 if (lli->lli_remote_perms) {
1106                         free_rmtperm_hash(lli->lli_remote_perms);
1107                         lli->lli_remote_perms = NULL;
1108                 }
1109         }
1110 #ifdef CONFIG_FS_POSIX_ACL
1111         else if (lli->lli_posix_acl) {
1112                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1113                 LASSERT(lli->lli_remote_perms == NULL);
1114                 posix_acl_release(lli->lli_posix_acl);
1115                 lli->lli_posix_acl = NULL;
1116         }
1117 #endif
1118         lli->lli_inode_magic = LLI_INODE_DEAD;
1119
1120 #ifdef HAVE_EXPORT___IGET
1121         spin_lock(&sbi->ll_deathrow_lock);
1122         list_del_init(&lli->lli_dead_list);
1123         spin_unlock(&sbi->ll_deathrow_lock);
1124 #endif
1125         ll_clear_inode_capas(inode);
1126         /*
1127          * XXX This has to be done before lsm is freed below, because
1128          * cl_object still uses inode lsm.
1129          */
1130         cl_inode_fini(inode);
1131
1132         if (lli->lli_smd) {
1133                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1134                 lli->lli_smd = NULL;
1135         }
1136
1137
1138         EXIT;
1139 }
1140
1141 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1142                   struct md_open_data **mod)
1143 {
1144         struct lustre_md md;
1145         struct ll_sb_info *sbi = ll_i2sbi(inode);
1146         struct ptlrpc_request *request = NULL;
1147         int rc;
1148         ENTRY;
1149
1150         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1151                                      LUSTRE_OPC_ANY, NULL);
1152         if (IS_ERR(op_data))
1153                 RETURN(PTR_ERR(op_data));
1154
1155         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1156                         &request, mod);
1157         if (rc) {
1158                 ptlrpc_req_finished(request);
1159                 if (rc == -ENOENT) {
1160                         inode->i_nlink = 0;
1161                         /* Unlinked special device node? Or just a race?
1162                          * Pretend we done everything. */
1163                         if (!S_ISREG(inode->i_mode) &&
1164                             !S_ISDIR(inode->i_mode))
1165                                 rc = inode_setattr(inode, &op_data->op_attr);
1166                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1167                         CERROR("md_setattr fails: rc = %d\n", rc);
1168                 }
1169                 RETURN(rc);
1170         }
1171
1172         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1173                               sbi->ll_md_exp, &md);
1174         if (rc) {
1175                 ptlrpc_req_finished(request);
1176                 RETURN(rc);
1177         }
1178
1179         /* We call inode_setattr to adjust timestamps.
1180          * If there is at least some data in file, we cleared ATTR_SIZE
1181          * above to avoid invoking vmtruncate, otherwise it is important
1182          * to call vmtruncate in inode_setattr to update inode->i_size
1183          * (bug 6196) */
1184         rc = inode_setattr(inode, &op_data->op_attr);
1185
1186         /* Extract epoch data if obtained. */
1187         op_data->op_handle = md.body->handle;
1188         op_data->op_ioepoch = md.body->ioepoch;
1189
1190         ll_update_inode(inode, &md);
1191         ptlrpc_req_finished(request);
1192
1193         RETURN(rc);
1194 }
1195
1196 /* Close IO epoch and send Size-on-MDS attribute update. */
1197 static int ll_setattr_done_writing(struct inode *inode,
1198                                    struct md_op_data *op_data,
1199                                    struct md_open_data *mod)
1200 {
1201         struct ll_inode_info *lli = ll_i2info(inode);
1202         int rc = 0;
1203         ENTRY;
1204
1205         LASSERT(op_data != NULL);
1206         if (!S_ISREG(inode->i_mode))
1207                 RETURN(0);
1208
1209         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1210                op_data->op_ioepoch, PFID(&lli->lli_fid));
1211
1212         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1213         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1214         if (rc == -EAGAIN) {
1215                 /* MDS has instructed us to obtain Size-on-MDS attribute
1216                  * from OSTs and send setattr to back to MDS. */
1217                 rc = ll_sizeonmds_update(inode, &op_data->op_handle,
1218                                          op_data->op_ioepoch);
1219         } else if (rc) {
1220                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1221                        inode->i_ino, rc);
1222         }
1223         RETURN(rc);
1224 }
1225
1226 static int ll_setattr_do_truncate(struct inode *inode, loff_t size)
1227 {
1228         struct obd_capa *capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1229         int rc;
1230
1231         rc = cl_setattr_do_truncate(inode, size, capa);
1232         ll_truncate_free_capa(capa);
1233         return rc;
1234 }
1235
1236 static int ll_setattr_ost(struct inode *inode)
1237 {
1238         struct obd_capa *capa = ll_mdscapa_get(inode);
1239         int rc;
1240
1241         rc = cl_setattr_ost(inode, capa);
1242         capa_put(capa);
1243
1244         return rc;
1245 }
1246
1247 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1248  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1249  * keep these values until such a time that objects are allocated for it.
1250  * We do the MDS operations first, as it is checking permissions for us.
1251  * We don't to the MDS RPC if there is nothing that we want to store there,
1252  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1253  * going to do an RPC anyways.
1254  *
1255  * If we are doing a truncate, we will send the mtime and ctime updates
1256  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1257  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1258  * at the same time.
1259  */
1260 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1261 {
1262         struct ll_inode_info *lli = ll_i2info(inode);
1263         struct lov_stripe_md *lsm = lli->lli_smd;
1264         struct md_op_data *op_data = NULL;
1265         struct md_open_data *mod = NULL;
1266         int ia_valid = attr->ia_valid;
1267         int rc = 0, rc1 = 0;
1268         ENTRY;
1269
1270         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1271                attr->ia_valid);
1272         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1273
1274         if (ia_valid & ATTR_SIZE) {
1275                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1276                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1277                                attr->ia_size, ll_file_maxbytes(inode));
1278                         RETURN(-EFBIG);
1279                 }
1280
1281                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1282         }
1283
1284         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1285         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1286                 if (current->fsuid != inode->i_uid &&
1287                     !cfs_capable(CFS_CAP_FOWNER))
1288                         RETURN(-EPERM);
1289         }
1290
1291         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1292         if (attr->ia_valid & ATTR_CTIME) {
1293                 attr->ia_ctime = CURRENT_TIME;
1294                 attr->ia_valid |= ATTR_CTIME_SET;
1295         }
1296         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1297                 attr->ia_atime = CURRENT_TIME;
1298                 attr->ia_valid |= ATTR_ATIME_SET;
1299         }
1300         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1301                 attr->ia_mtime = CURRENT_TIME;
1302                 attr->ia_valid |= ATTR_MTIME_SET;
1303         }
1304         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1305                 /* To avoid stale mtime on mds, obtain it from ost and send
1306                    to mds. */
1307                 rc = cl_glimpse_size(inode);
1308                 if (rc)
1309                         RETURN(rc);
1310
1311                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1312                 attr->ia_mtime = inode->i_mtime;
1313         }
1314
1315         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1316                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1317                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1318                        cfs_time_current_sec());
1319
1320         /* NB: ATTR_SIZE will only be set after this point if the size
1321          * resides on the MDS, ie, this file has no objects. */
1322         if (lsm)
1323                 attr->ia_valid &= ~ATTR_SIZE;
1324
1325         /* We always do an MDS RPC, even if we're only changing the size;
1326          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1327
1328         OBD_ALLOC_PTR(op_data);
1329         if (op_data == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1333
1334         /* Open epoch for truncate. */
1335         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1336             (ia_valid & ATTR_SIZE))
1337                 op_data->op_flags = MF_EPOCH_OPEN;
1338
1339         rc = ll_md_setattr(inode, op_data, &mod);
1340         if (rc)
1341                 GOTO(out, rc);
1342
1343         if (op_data->op_ioepoch)
1344                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1345                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1346
1347         if (!lsm || !S_ISREG(inode->i_mode)) {
1348                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1349                 GOTO(out, rc = 0);
1350         }
1351
1352         if (ia_valid & ATTR_SIZE)
1353                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1354         else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1355                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1356                        inode->i_ino, LTIME_S(attr->ia_mtime));
1357                 rc = ll_setattr_ost(inode);
1358         }
1359         EXIT;
1360 out:
1361         if (op_data) {
1362                 if (op_data->op_ioepoch)
1363                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1364                 ll_finish_md_op_data(op_data);
1365         }
1366         return rc ? rc : rc1;
1367 }
1368
1369 int ll_setattr(struct dentry *de, struct iattr *attr)
1370 {
1371         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1372             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1373                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1374
1375         if ((de->d_inode->i_mode & S_ISUID) &&
1376             !(attr->ia_mode & S_ISUID) &&
1377             !(attr->ia_valid & ATTR_KILL_SUID))
1378                 attr->ia_valid |= ATTR_KILL_SUID;
1379
1380         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1381             !(attr->ia_mode & S_ISGID) &&
1382             !(attr->ia_valid & ATTR_KILL_SGID))
1383                 attr->ia_valid |= ATTR_KILL_SGID;
1384
1385         return ll_setattr_raw(de->d_inode, attr);
1386 }
1387
1388 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1389                        __u64 max_age, __u32 flags)
1390 {
1391         struct ll_sb_info *sbi = ll_s2sbi(sb);
1392         struct obd_statfs obd_osfs;
1393         int rc;
1394         ENTRY;
1395
1396         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1397         if (rc) {
1398                 CERROR("md_statfs fails: rc = %d\n", rc);
1399                 RETURN(rc);
1400         }
1401
1402         osfs->os_type = sb->s_magic;
1403
1404         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1405                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1406
1407         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1408                 flags |= OBD_STATFS_NODELAY;
1409
1410         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1411                               &obd_osfs, max_age, flags);
1412         if (rc) {
1413                 CERROR("obd_statfs fails: rc = %d\n", rc);
1414                 RETURN(rc);
1415         }
1416
1417         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1418                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1419                obd_osfs.os_files);
1420
1421         osfs->os_bsize = obd_osfs.os_bsize;
1422         osfs->os_blocks = obd_osfs.os_blocks;
1423         osfs->os_bfree = obd_osfs.os_bfree;
1424         osfs->os_bavail = obd_osfs.os_bavail;
1425
1426         /* If we don't have as many objects free on the OST as inodes
1427          * on the MDS, we reduce the total number of inodes to
1428          * compensate, so that the "inodes in use" number is correct.
1429          */
1430         if (obd_osfs.os_ffree < osfs->os_ffree) {
1431                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1432                         obd_osfs.os_ffree;
1433                 osfs->os_ffree = obd_osfs.os_ffree;
1434         }
1435
1436         RETURN(rc);
1437 }
1438 #ifndef HAVE_STATFS_DENTRY_PARAM
1439 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1440 {
1441 #else
1442 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1443 {
1444         struct super_block *sb = de->d_sb;
1445 #endif
1446         struct obd_statfs osfs;
1447         int rc;
1448
1449         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1450         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1451
1452         /* For now we will always get up-to-date statfs values, but in the
1453          * future we may allow some amount of caching on the client (e.g.
1454          * from QOS or lprocfs updates). */
1455         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1456         if (rc)
1457                 return rc;
1458
1459         statfs_unpack(sfs, &osfs);
1460
1461         /* We need to downshift for all 32-bit kernels, because we can't
1462          * tell if the kernel is being called via sys_statfs64() or not.
1463          * Stop before overflowing f_bsize - in which case it is better
1464          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1465         if (sizeof(long) < 8) {
1466                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1467                         sfs->f_bsize <<= 1;
1468
1469                         osfs.os_blocks >>= 1;
1470                         osfs.os_bfree >>= 1;
1471                         osfs.os_bavail >>= 1;
1472                 }
1473         }
1474
1475         sfs->f_blocks = osfs.os_blocks;
1476         sfs->f_bfree = osfs.os_bfree;
1477         sfs->f_bavail = osfs.os_bavail;
1478
1479         return 0;
1480 }
1481
1482 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1483 {
1484         struct ll_inode_info *lli;
1485         struct lov_stripe_md *lsm;
1486
1487         lli = ll_i2info(inode);
1488         LASSERT(lli->lli_size_sem_owner != current);
1489         down(&lli->lli_size_sem);
1490         LASSERT(lli->lli_size_sem_owner == NULL);
1491         lli->lli_size_sem_owner = current;
1492         lsm = lli->lli_smd;
1493         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1494                  lsm, lock_lsm);
1495         if (lock_lsm)
1496                 lov_stripe_lock(lsm);
1497 }
1498
1499 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1500 {
1501         struct ll_inode_info *lli;
1502         struct lov_stripe_md *lsm;
1503
1504         lli = ll_i2info(inode);
1505         lsm = lli->lli_smd;
1506         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1507                  lsm, unlock_lsm);
1508         if (unlock_lsm)
1509                 lov_stripe_unlock(lsm);
1510         LASSERT(lli->lli_size_sem_owner == current);
1511         lli->lli_size_sem_owner = NULL;
1512         up(&lli->lli_size_sem);
1513 }
1514
1515 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1516 {
1517         struct ll_inode_info *lli = ll_i2info(inode);
1518
1519         dump_lsm(D_INODE, lsm);
1520         dump_lsm(D_INODE, lli->lli_smd);
1521         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1522                  "lsm must be joined lsm %p\n", lsm);
1523         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1524         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1525                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1526         lli->lli_smd = lsm;
1527         lli->lli_maxbytes = lsm->lsm_maxbytes;
1528         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1529                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1530 }
1531
1532 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1533 {
1534         struct ll_inode_info *lli = ll_i2info(inode);
1535         struct mdt_body *body = md->body;
1536         struct lov_stripe_md *lsm = md->lsm;
1537         struct ll_sb_info *sbi = ll_i2sbi(inode);
1538
1539         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1540         if (lsm != NULL) {
1541                 if (lli->lli_smd == NULL) {
1542                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1543                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1544                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1545                                 dump_lsm(D_ERROR, lsm);
1546                                 LBUG();
1547                         }
1548                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1549                                lsm, inode->i_ino, inode->i_generation, inode);
1550                         cl_inode_init(inode, md);
1551                         /* ll_inode_size_lock() requires it is only
1552                          * called with lli_smd != NULL or lock_lsm == 0
1553                          *  or we can race between lock/unlock.
1554                          *  bug 9547 */
1555                         lli->lli_smd = lsm;
1556                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1557                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1558                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1559                 } else {
1560                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1561                              lli->lli_smd->lsm_stripe_count ==
1562                                         lsm->lsm_stripe_count) {
1563                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1564                                         CERROR("lsm mismatch for inode %ld\n",
1565                                                 inode->i_ino);
1566                                         CERROR("lli_smd:\n");
1567                                         dump_lsm(D_ERROR, lli->lli_smd);
1568                                         CERROR("lsm:\n");
1569                                         dump_lsm(D_ERROR, lsm);
1570                                         LBUG();
1571                                 }
1572                         } else {
1573                                 cl_inode_init(inode, md);
1574                                 ll_replace_lsm(inode, lsm);
1575                         }
1576                 }
1577                 if (lli->lli_smd != lsm)
1578                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1579         }
1580
1581         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1582                 if (body->valid & OBD_MD_FLRMTPERM)
1583                         ll_update_remote_perm(inode, md->remote_perm);
1584         }
1585 #ifdef CONFIG_FS_POSIX_ACL
1586         else if (body->valid & OBD_MD_FLACL) {
1587                 spin_lock(&lli->lli_lock);
1588                 if (lli->lli_posix_acl)
1589                         posix_acl_release(lli->lli_posix_acl);
1590                 lli->lli_posix_acl = md->posix_acl;
1591                 spin_unlock(&lli->lli_lock);
1592         }
1593 #endif
1594         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1595         inode->i_generation = ll_fid_build_gen(sbi, &body->fid1);
1596
1597         if (body->valid & OBD_MD_FLATIME &&
1598             body->atime > LTIME_S(inode->i_atime))
1599                 LTIME_S(inode->i_atime) = body->atime;
1600
1601         /* mtime is always updated with ctime, but can be set in past.
1602            As write and utime(2) may happen within 1 second, and utime's
1603            mtime has a priority over write's one, so take mtime from mds
1604            for the same ctimes. */
1605         if (body->valid & OBD_MD_FLCTIME &&
1606             body->ctime >= LTIME_S(inode->i_ctime)) {
1607                 LTIME_S(inode->i_ctime) = body->ctime;
1608                 if (body->valid & OBD_MD_FLMTIME) {
1609                         CDEBUG(D_INODE, "setting ino %lu mtime "
1610                                "from %lu to "LPU64"\n", inode->i_ino,
1611                                LTIME_S(inode->i_mtime), body->mtime);
1612                         LTIME_S(inode->i_mtime) = body->mtime;
1613                 }
1614         }
1615         if (body->valid & OBD_MD_FLMODE)
1616                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1617         if (body->valid & OBD_MD_FLTYPE)
1618                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1619         LASSERT(inode->i_mode != 0);
1620         if (S_ISREG(inode->i_mode)) {
1621                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1622         } else {
1623                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1624         }
1625 #ifdef HAVE_INODE_BLKSIZE
1626         inode->i_blksize = 1<<inode->i_blkbits;
1627 #endif
1628         if (body->valid & OBD_MD_FLUID)
1629                 inode->i_uid = body->uid;
1630         if (body->valid & OBD_MD_FLGID)
1631                 inode->i_gid = body->gid;
1632         if (body->valid & OBD_MD_FLFLAGS)
1633                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1634         if (body->valid & OBD_MD_FLNLINK)
1635                 inode->i_nlink = body->nlink;
1636         if (body->valid & OBD_MD_FLRDEV)
1637                 inode->i_rdev = old_decode_dev(body->rdev);
1638
1639         if (body->valid & OBD_MD_FLID) {
1640                 /* FID shouldn't be changed! */
1641                 if (fid_is_sane(&lli->lli_fid)) {
1642                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1643                                  "Trying to change FID "DFID
1644                                  " to the "DFID", inode %lu/%u(%p)\n",
1645                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1646                                  inode->i_ino, inode->i_generation, inode);
1647                 } else
1648                         lli->lli_fid = body->fid1;
1649         }
1650
1651         LASSERT(fid_seq(&lli->lli_fid) != 0);
1652
1653         if (body->valid & OBD_MD_FLSIZE) {
1654                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1655                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1656                         struct lustre_handle lockh;
1657                         ldlm_mode_t mode;
1658
1659                         /* As it is possible a blocking ast has been processed
1660                          * by this time, we need to check there is an UPDATE
1661                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1662                          * it. */
1663                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1664                                                &lockh);
1665                         if (mode) {
1666                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1667                                                       LLIF_EPOCH_PENDING |
1668                                                       LLIF_SOM_DIRTY)) {
1669                                         CERROR("ino %lu flags %lu still has "
1670                                                "size authority! do not trust "
1671                                                "the size got from MDS\n",
1672                                                inode->i_ino, lli->lli_flags);
1673                                 } else {
1674                                         /* Use old size assignment to avoid
1675                                          * deadlock bz14138 & bz14326 */
1676                                         inode->i_size = body->size;
1677                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1678                                 }
1679                                 ldlm_lock_decref(&lockh, mode);
1680                         }
1681                 } else {
1682                         /* Use old size assignment to avoid
1683                          * deadlock bz14138 & bz14326 */
1684                         inode->i_size = body->size;
1685                 }
1686
1687                 if (body->valid & OBD_MD_FLBLOCKS)
1688                         inode->i_blocks = body->blocks;
1689         }
1690
1691         if (body->valid & OBD_MD_FLMDSCAPA) {
1692                 LASSERT(md->mds_capa);
1693                 ll_add_capa(inode, md->mds_capa);
1694         }
1695         if (body->valid & OBD_MD_FLOSSCAPA) {
1696                 LASSERT(md->oss_capa);
1697                 ll_add_capa(inode, md->oss_capa);
1698         }
1699 }
1700
1701 static struct backing_dev_info ll_backing_dev_info = {
1702         .ra_pages       = 0,    /* No readahead */
1703 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1704         .capabilities   = 0,    /* Does contribute to dirty memory */
1705 #else
1706         .memory_backed  = 0,    /* Does contribute to dirty memory */
1707 #endif
1708 };
1709
1710 void ll_read_inode2(struct inode *inode, void *opaque)
1711 {
1712         struct lustre_md *md = opaque;
1713         struct ll_inode_info *lli = ll_i2info(inode);
1714         ENTRY;
1715
1716         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1717                inode->i_ino, inode->i_generation, inode);
1718
1719         ll_lli_init(lli);
1720
1721         LASSERT(!lli->lli_smd);
1722
1723         /* Core attributes from the MDS first.  This is a new inode, and
1724          * the VFS doesn't zero times in the core inode so we have to do
1725          * it ourselves.  They will be overwritten by either MDS or OST
1726          * attributes - we just need to make sure they aren't newer. */
1727         LTIME_S(inode->i_mtime) = 0;
1728         LTIME_S(inode->i_atime) = 0;
1729         LTIME_S(inode->i_ctime) = 0;
1730         inode->i_rdev = 0;
1731         ll_update_inode(inode, md);
1732
1733         /* OIDEBUG(inode); */
1734
1735         if (S_ISREG(inode->i_mode)) {
1736                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1737                 inode->i_op = &ll_file_inode_operations;
1738                 inode->i_fop = sbi->ll_fop;
1739                 inode->i_mapping->a_ops = &ll_aops;
1740                 EXIT;
1741         } else if (S_ISDIR(inode->i_mode)) {
1742                 inode->i_op = &ll_dir_inode_operations;
1743                 inode->i_fop = &ll_dir_operations;
1744                 inode->i_mapping->a_ops = &ll_dir_aops;
1745                 EXIT;
1746         } else if (S_ISLNK(inode->i_mode)) {
1747                 inode->i_op = &ll_fast_symlink_inode_operations;
1748                 EXIT;
1749         } else {
1750                 inode->i_op = &ll_special_inode_operations;
1751
1752                 init_special_inode(inode, inode->i_mode,
1753                                    kdev_t_to_nr(inode->i_rdev));
1754
1755                 /* initializing backing dev info. */
1756                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1757
1758                 EXIT;
1759         }
1760 }
1761
1762 void ll_delete_inode(struct inode *inode)
1763 {
1764         struct ll_sb_info *sbi = ll_i2sbi(inode);
1765         int rc;
1766         ENTRY;
1767
1768         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1769         if (rc) {
1770                 CERROR("fid_delete() failed, rc %d\n", rc);
1771         }
1772         truncate_inode_pages(&inode->i_data, 0);
1773         clear_inode(inode);
1774
1775         EXIT;
1776 }
1777
1778 int ll_iocontrol(struct inode *inode, struct file *file,
1779                  unsigned int cmd, unsigned long arg)
1780 {
1781         struct ll_sb_info *sbi = ll_i2sbi(inode);
1782         struct ptlrpc_request *req = NULL;
1783         int rc, flags = 0;
1784         ENTRY;
1785
1786         switch(cmd) {
1787         case EXT3_IOC_GETFLAGS: {
1788                 struct mdt_body *body;
1789                 struct obd_capa *oc;
1790
1791                 oc = ll_mdscapa_get(inode);
1792                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1793                                 OBD_MD_FLFLAGS, 0, &req);
1794                 capa_put(oc);
1795                 if (rc) {
1796                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1797                         RETURN(-abs(rc));
1798                 }
1799
1800                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1801
1802                 flags = body->flags;
1803
1804                 ptlrpc_req_finished(req);
1805
1806                 RETURN(put_user(flags, (int *)arg));
1807         }
1808         case EXT3_IOC_SETFLAGS: {
1809                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1810                 struct obd_info oinfo = { { { 0 } } };
1811                 struct md_op_data *op_data;
1812
1813                 if (get_user(flags, (int *)arg))
1814                         RETURN(-EFAULT);
1815
1816                 oinfo.oi_md = lsm;
1817                 OBDO_ALLOC(oinfo.oi_oa);
1818                 if (!oinfo.oi_oa)
1819                         RETURN(-ENOMEM);
1820
1821                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1822                                              LUSTRE_OPC_ANY, NULL);
1823                 if (IS_ERR(op_data))
1824                         RETURN(PTR_ERR(op_data));
1825
1826                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1827                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1828                 rc = md_setattr(sbi->ll_md_exp, op_data,
1829                                 NULL, 0, NULL, 0, &req, NULL);
1830                 ll_finish_md_op_data(op_data);
1831                 ptlrpc_req_finished(req);
1832                 if (rc || lsm == NULL) {
1833                         OBDO_FREE(oinfo.oi_oa);
1834                         RETURN(rc);
1835                 }
1836
1837                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1838                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1839                 oinfo.oi_oa->o_flags = flags;
1840                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1841                                        OBD_MD_FLGROUP;
1842                 oinfo.oi_capa = ll_mdscapa_get(inode);
1843
1844                 obdo_from_inode(oinfo.oi_oa, inode,
1845                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1846                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1847                 capa_put(oinfo.oi_capa);
1848                 OBDO_FREE(oinfo.oi_oa);
1849                 if (rc) {
1850                         if (rc != -EPERM && rc != -EACCES)
1851                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
1852                         RETURN(rc);
1853                 }
1854
1855                 inode->i_flags = ll_ext_to_inode_flags(flags |
1856                                                        MDS_BFLAG_EXT_FLAGS);
1857                 RETURN(0);
1858         }
1859         default:
1860                 RETURN(-ENOSYS);
1861         }
1862
1863         RETURN(0);
1864 }
1865
1866 int ll_flush_ctx(struct inode *inode)
1867 {
1868         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1869
1870         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
1871
1872         obd_set_info_async(sbi->ll_md_exp,
1873                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1874                            0, NULL, NULL);
1875         obd_set_info_async(sbi->ll_dt_exp,
1876                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1877                            0, NULL, NULL);
1878         return 0;
1879 }
1880
1881 /* umount -f client means force down, don't save state */
1882 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1883 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
1884 {
1885         struct super_block *sb = vfsmnt->mnt_sb;
1886 #else
1887 void ll_umount_begin(struct super_block *sb)
1888 {
1889 #endif
1890         struct lustre_sb_info *lsi = s2lsi(sb);
1891         struct ll_sb_info *sbi = ll_s2sbi(sb);
1892         struct obd_device *obd;
1893         struct obd_ioctl_data ioc_data = { 0 };
1894         ENTRY;
1895
1896 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1897         if (!(flags & MNT_FORCE)) {
1898                 EXIT;
1899                 return;
1900         }
1901 #endif
1902
1903         /* Tell the MGC we got umount -f */
1904         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1905
1906         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1907                sb->s_count, atomic_read(&sb->s_active));
1908
1909         obd = class_exp2obd(sbi->ll_md_exp);
1910         if (obd == NULL) {
1911                 CERROR("Invalid MDC connection handle "LPX64"\n",
1912                        sbi->ll_md_exp->exp_handle.h_cookie);
1913                 EXIT;
1914                 return;
1915         }
1916         obd->obd_force = 1;
1917         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
1918                       &ioc_data, NULL);
1919
1920         obd = class_exp2obd(sbi->ll_dt_exp);
1921         if (obd == NULL) {
1922                 CERROR("Invalid LOV connection handle "LPX64"\n",
1923                        sbi->ll_dt_exp->exp_handle.h_cookie);
1924                 EXIT;
1925                 return;
1926         }
1927
1928         obd->obd_force = 1;
1929         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
1930                       &ioc_data, NULL);
1931
1932         /* Really, we'd like to wait until there are no requests outstanding,
1933          * and then continue.  For now, we just invalidate the requests,
1934          * schedule() and sleep one second if needed, and hope.
1935          */
1936         schedule();
1937 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1938         if (atomic_read(&vfsmnt->mnt_count) > 2) {
1939                 cfs_schedule_timeout(CFS_TASK_INTERRUPTIBLE,
1940                                      cfs_time_seconds(1));
1941                 if (atomic_read(&vfsmnt->mnt_count) > 2)
1942                         LCONSOLE_WARN("Mount still busy with %d refs! You "
1943                                       "may try to umount it a bit later\n",
1944                                       atomic_read(&vfsmnt->mnt_count));
1945         }
1946 #endif
1947
1948         EXIT;
1949 }
1950
1951 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
1952 {
1953         struct ll_sb_info *sbi = ll_s2sbi(sb);
1954         int err;
1955         __u32 read_only;
1956
1957         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
1958                 read_only = *flags & MS_RDONLY;
1959                 err = obd_set_info_async(sbi->ll_md_exp,
1960                                          sizeof(KEY_READ_ONLY),
1961                                          KEY_READ_ONLY, sizeof(read_only),
1962                                          &read_only, NULL);
1963                 if (err) {
1964                         CERROR("Failed to change the read-only flag during "
1965                                "remount: %d\n", err);
1966                         return err;
1967                 }
1968
1969                 if (read_only)
1970                         sb->s_flags |= MS_RDONLY;
1971                 else
1972                         sb->s_flags &= ~MS_RDONLY;
1973         }
1974         return 0;
1975 }
1976
1977 int ll_prep_inode(struct inode **inode,
1978                   struct ptlrpc_request *req,
1979                   struct super_block *sb)
1980 {
1981         struct ll_sb_info *sbi = NULL;
1982         struct lustre_md md;
1983         int rc;
1984         ENTRY;
1985
1986         LASSERT(*inode || sb);
1987         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
1988         prune_deathrow(sbi, 1);
1989         memset(&md, 0, sizeof(struct lustre_md));
1990
1991         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
1992                               sbi->ll_md_exp, &md);
1993         if (rc)
1994                 RETURN(rc);
1995
1996         if (*inode) {
1997                 ll_update_inode(*inode, &md);
1998         } else {
1999                 LASSERT(sb != NULL);
2000
2001                 /*
2002                  * At this point server returns to client's same fid as client
2003                  * generated for creating. So using ->fid1 is okay here.
2004                  */
2005                 LASSERT(fid_is_sane(&md.body->fid1));
2006
2007                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1),&md);
2008                 if (*inode == NULL || IS_ERR(*inode)) {
2009                         if (md.lsm)
2010                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2011 #ifdef CONFIG_FS_POSIX_ACL
2012                         if (md.posix_acl) {
2013                                 posix_acl_release(md.posix_acl);
2014                                 md.posix_acl = NULL;
2015                         }
2016 #endif
2017                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2018                         *inode = NULL;
2019                         CERROR("new_inode -fatal: rc %d\n", rc);
2020                         GOTO(out, rc);
2021                 }
2022         }
2023
2024         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2025                          ll_i2info(*inode)->lli_smd);
2026 out:
2027         md_free_lustre_md(sbi->ll_md_exp, &md);
2028         RETURN(rc);
2029 }
2030
2031 int ll_obd_statfs(struct inode *inode, void *arg)
2032 {
2033         struct ll_sb_info *sbi = NULL;
2034         struct obd_export *exp;
2035         char *buf = NULL;
2036         struct obd_ioctl_data *data = NULL;
2037         __u32 type;
2038         int len = 0, rc;
2039
2040         if (!inode || !(sbi = ll_i2sbi(inode)))
2041                 GOTO(out_statfs, rc = -EINVAL);
2042
2043         rc = obd_ioctl_getdata(&buf, &len, arg);
2044         if (rc)
2045                 GOTO(out_statfs, rc);
2046
2047         data = (void*)buf;
2048         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2049             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2050                 GOTO(out_statfs, rc = -EINVAL);
2051
2052         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2053         if (type == LL_STATFS_MDC)
2054                 exp = sbi->ll_md_exp;
2055         else if (type == LL_STATFS_LOV)
2056                 exp = sbi->ll_dt_exp;
2057         else
2058                 GOTO(out_statfs, rc = -ENODEV);
2059
2060         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2061         if (rc)
2062                 GOTO(out_statfs, rc);
2063 out_statfs:
2064         if (buf)
2065                 obd_ioctl_freedata(buf, len);
2066         return rc;
2067 }
2068
2069 int ll_process_config(struct lustre_cfg *lcfg)
2070 {
2071         char *ptr;
2072         void *sb;
2073         struct lprocfs_static_vars lvars;
2074         unsigned long x;
2075         int rc = 0;
2076
2077         lprocfs_llite_init_vars(&lvars);
2078
2079         /* The instance name contains the sb: lustre-client-aacfe000 */
2080         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2081         if (!ptr || !*(++ptr))
2082                 return -EINVAL;
2083         if (sscanf(ptr, "%lx", &x) != 1)
2084                 return -EINVAL;
2085         sb = (void *)x;
2086         /* This better be a real Lustre superblock! */
2087         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2088
2089         /* Note we have not called client_common_fill_super yet, so
2090            proc fns must be able to handle that! */
2091         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2092                                       lcfg, sb);
2093         if (rc > 0)
2094                 rc = 0;
2095         return(rc);
2096 }
2097
2098 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2099 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2100                                        struct inode *i1, struct inode *i2,
2101                                        const char *name, int namelen,
2102                                        int mode, __u32 opc, void *data)
2103 {
2104         LASSERT(i1 != NULL);
2105
2106         if (namelen > ll_i2sbi(i1)->ll_namelen)
2107                 return ERR_PTR(-ENAMETOOLONG);
2108
2109         if (op_data == NULL)
2110                 OBD_ALLOC_PTR(op_data);
2111
2112         if (op_data == NULL)
2113                 return ERR_PTR(-ENOMEM);
2114
2115         ll_i2gids(op_data->op_suppgids, i1, i2);
2116         op_data->op_fid1 = *ll_inode2fid(i1);
2117         op_data->op_capa1 = ll_mdscapa_get(i1);
2118
2119         if (i2) {
2120                 op_data->op_fid2 = *ll_inode2fid(i2);
2121                 op_data->op_capa2 = ll_mdscapa_get(i2);
2122         } else {
2123                 fid_zero(&op_data->op_fid2);
2124                 op_data->op_capa2 = NULL;
2125         }
2126
2127         op_data->op_name = name;
2128         op_data->op_namelen = namelen;
2129         op_data->op_mode = mode;
2130         op_data->op_mod_time = cfs_time_current_sec();
2131         op_data->op_fsuid = current->fsuid;
2132         op_data->op_fsgid = current->fsgid;
2133         op_data->op_cap = cfs_curproc_cap_pack();
2134         op_data->op_bias = MDS_CHECK_SPLIT;
2135         op_data->op_opc = opc;
2136         op_data->op_mds = 0;
2137         op_data->op_data = data;
2138
2139         return op_data;
2140 }
2141
2142 void ll_finish_md_op_data(struct md_op_data *op_data)
2143 {
2144         capa_put(op_data->op_capa1);
2145         capa_put(op_data->op_capa2);
2146         OBD_FREE_PTR(op_data);
2147 }
2148
2149 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2150 {
2151         struct ll_sb_info *sbi;
2152
2153         LASSERT((seq != NULL) && (vfs != NULL));
2154         sbi = ll_s2sbi(vfs->mnt_sb);
2155
2156         if (sbi->ll_flags & LL_SBI_NOLCK)
2157                 seq_puts(seq, ",nolock");
2158
2159         if (sbi->ll_flags & LL_SBI_FLOCK)
2160                 seq_puts(seq, ",flock");
2161
2162         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2163                 seq_puts(seq, ",localflock");
2164
2165         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2166                 seq_puts(seq, ",user_xattr");
2167
2168         if (sbi->ll_flags & LL_SBI_ACL)
2169                 seq_puts(seq, ",acl");
2170
2171         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2172                 seq_puts(seq, ",lazystatfs");
2173
2174         RETURN(0);
2175 }