lustre/llite/llite_lib.c

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