lustre/llite/llite_lib.c

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